Rhodopsin-cyclases for photocontrol of cGMP/cAMP and 2.3 Å structure of the adenylyl cyclase domain

The cyclic nucleotides cAMP and cGMP are important second messengers that orchestrate fundamental cellular responses. Here, we present the characterization of the rhodopsin-guanylyl cyclase from Catenaria anguillulae (CaRhGC), which produces cGMP in response to green light with a light to dark activity ratio >1000. After light excitation the putative signaling state forms with τ = 31 ms and decays with τ = 570 ms. Mutations (up to 6) within the nucleotide binding site generate rhodopsin-adenylyl cyclases (CaRhACs) of which the double mutated YFP-CaRhAC (E497K/C566D) is the most suitable for rapid cAMP production in neurons. Furthermore, the crystal structure of the ligand-bound AC domain (2.25 Å) reveals detailed information about the nucleotide binding mode within this recently discovered class of enzyme rhodopsin. Both YFP-CaRhGC and YFP-CaRhAC are favorable optogenetic tools for non-invasive, cell-selective, and spatio-temporally precise modulation of cAMP/cGMP with light.

This manuscript presents an in-depth analysis of the Catenaria anguillulae rhodopsin-guanylyl cyclase, which is a light-activated enzyme producing guanylyl cyclase with a very large ratio of the activities in the light and in the dark. Moreover, they authors successfully mutated this enzyme into a light-activated adenylyl cyclase, though with a smaller light-to-dark ratio. The authors also present a crystal structure of the newly developed adenylyl cyclase domain in complex with a substrate analog.
The enzymological work is interesting, the crystallography is sound and the structure shows for the first time a nucleobase bound tightly in its binding pocket on such an enzyme. However, since this is the predicted binding mode, and the rest of the nucleotide clearly binds in a nonproductive manner, the structure does not add a great deal of novelty.
I therefore recommend not to publish the paper in Nature Communications but suggest the authors submit this manuscript to a more specialized journal.
In addition to this there are several issues that would need to be addressed as well. I will list them below: *My main concern is with the proposed mechanism of activation. The authors argue that since the cyclase domain alone is active, whereas it is inactive in the dark when coupled to the rhodopsin domain, the rhodopsin acts as a clamp in the dark state, only to "release" the activity upon light activation of the rhodopsin. This may well be part of the story, but probably isn't all that is going on here. Looking at table 1, the light-activated rhodopsin-coupled cyclase is three times more active than the cyclase alone. A possible explanation for this would be that the rhodopsin in its light-activated state affects the dynamics of the cyclase. The authors should have addressed the possibility of other models that are more complex than their current proposal. *Indeed, the authors write in line 225 that they chose a particular dimer interface to analyse because it has the smallest overall B-factor. Does this mean that the B-factors of the others, and therefore the density of the other interfaces was too high for interpretation? That would be not be a problem in and of itself, but more likely a finding pointing to high structural dynamics in itself.
*Then there is the chemical mechanism. In line 251 the authors write about a "pseudobinuclear nucleophilic substitution SN2 mechanism"… This error was taken almost literally from the review cited. First of all binuclear is a term used for metal complexes. What is meant is probably "bimolecular". This, in turn, is wrong, too. There are first-and second order reactions, and under certain circumstances pseudo-first-and pseudo-second order reactions, but the order of the reaction is something other than its molecularity. Moreover SN2 already means "nucleophilic substitution", so the way this is phrased is double. But more errors are copied from the same review without critical evaluation: l.252 "resulting in a negatively charged Pa during the transition state". In a pentacovalent transition state such as could be proposed for this reaction, indeed an extra negative charge develops on one of the oxygens of the alpha phosphate. But the way this is written implies that the phosphorous gets a negative charge, which is nonsense. Again, this wording was taken from the review cited, but is, depending on how it was meant, at best highly confusing, and at worst plain wrong.
Also (line 323): an in-line orientation of the reactants does not by itself mean that an associative mechanism, i.e. with a pentavalent phosphorous in the transition state, is formed. In fact, as the authors have little or no data that really touches upon the details of the transition state, such as a transition state analog or better kinetic isotope effects, I would not talk about it. *Then, the way the paper is written seems to be aiming at an audience of experts in this field, as a lot of jargon is used. "light-to-dark ratio", for instance, and "after light-off". Such things should be explained. And in line 55: "the enzyme's activity declines with a delay of 300 ms.". Does it wait for 300 ms and then decay? Or does it decay with a 300 ms time constant? Please define such things clearly.
* Another case of a vague use of terms in the legends of fig 2 and 6: electroporation cannot be done with proteins. It was done with DNA coding for these proteins. * l.267 how close is the ribose ring oxygen to S576? Why is this interaction called a "polar bond" (whatever that may be!), and not a hydrogen bond? * fig.7. Although the 2Fo-Fc density (which is probably actually 2mFo-DFc) for the ligand is shown at a relatively high contour level (2 sigma) it would have been more convincing to use an omit map for the ligand, given that its B-factor is even higher than the average of the waters according to the crystallographic data table.
* Then there are some linguistic/typographic problems: l.51. "putative coiled coil linker" -is the linker putative? Or it's coiled coil structure? l. 74 . "extra helices"? on top of what? the rhodopsin? that becomes clear only afterwards l.71. 2,6 x higher should be 2.6x higher l.350 "Hereby"? Whereby? The authors mean that the role of the sulfur in determining ligand conformation is unclear. l.393 N-terminal YFP tagged --> N-terminally YFP-tagged (adjective, not adverb!) Table 1. I am pretty sure the Km in kcat/Km should not be in subscript… *As a final note, the structure report from the PDB asserts that the R-epimer of the ligand has been built in. This is a mistake by the PDB, the authors have correctly built in the S-epimer, as shown in their figures. There is probably something wrong with the ligand database; if I were the authors I would ask the PDB to change this, otherwise people who download the structure might be confused.
Reviewer #3 (Remarks to the Author): Scheib and coauthors described new optogenetic tools, green-light activated rhodopsin-guanylyl cyclase (GC) and adenylyl cyclase (AC). Because of the breadth of cGMP and cAMP signaling in animals, improved optogenetic tools are undoubtedly welcome. Specifically, (i) the researchers identified and characterized the rhodopsin GC from fungus Catenaria (CaRhGC) --in vitro, in oocytes and in neuronal culture. They showed that CaRhGC is more stable in high light than the related rhodopsin GC from fungus Blastocladiella (BeRhGC = CyclOp). (ii) They converted the GC into the photoactivated adenylyl cyclase (AC) by mutagenizing residues in the substrate binding pocket. (iii) They also determined the X-ray structure of the catalytic domain of the designed AC. Each of these accomplishments is a step forward, but an incremental step. Together, these advances amount to a good, solid paper, yet it is unclear whether they amount to a high-profile paper.
(i) When expressed in neurons, currents evoked by brief irradiation of neurons expressing CaRhGC were similar to the currents in neurons expressing BeRhGC ( Fig. 2c-2e). There were some differences in kinetics. Were these differences significant to suggest that CaRhGC outperforms BeRhGC? Disturbingly (page 5, lines 106-8), photocurrents were evoked in ~70% of neurons expressing BeRhGC but only in ~30% of neurons expressing CaRhGC. Is retinal incorporation the same in both enzymes? Isn't this a major problem for CaRhGC that outweighs its photostability and somewhat higher activity advantage?
(ii) The GC-to-AC conversion by mutagenesis of substrate-binding residues has been known for two decades (as pointed out in the paper), so the conversion of CaRhGC to CaRhAC in itself is not particularly impressive. CaRhGC has a somewhat higher dynamic range (Fig. 5b) compared to the similarly designed BeRhAC (reference 14), which is an improvement. However, it is unclear how it compares to the blue-light ACs referenced in the manuscript or to the OptoXR and its improved variants (not referenced here).
(iii) The 2.25A-crystal structure of the catalytic dimer of the engineered AC domains is a welcome addition to previously determined structures of type III nucleotidyl cyclases. It is unclear that it adds much to the understanding of substrate specificity of ACs and GCs or catalytic mechanism. The lengthy description of the structure seems to be confirmatory of the conclusions reached from the AC structures described previously. The structure of a homodimeric GC (from Chalmydomonas) is unfortunately not even cited. Most disappointing is that no new insights emerged into the photoactivation mechanism.
Reviewer #4 (Remarks to the Author): This is a very informative, well-designed and well-executed study which characterized the properties of rhodopsin-guanylylcyclase (RhoGC) from Catenaria anguilulae(CaCylclOp). Earlier studies had previously explored the properties of rhodopsin guanylylcylase from Blastocladiella emersonii (BeCyclOP) which has a 77% sequence homolog with CaCylOP.
Importantly, the CaAC structure elucidation provided a basis and confirmation for much earlier work to understanding how the E497K and C566D mutations resulted in the switch in specificity from GC to AC. It also revealed difference compared to other adenylyl cyclase structures. Overall, the work provides evidence that the rhodopsin structure acts to "clamp" the CaGC domain (or CaAC domain) that is relieved during the photocycle, although it does not reveal the molecular mechanism of the clamp.
The paper should be published after the following revisions: 1) Several experiments indicate that CaCylclOp has superior properties as an optogenetic tool compared to BeCyclOP. For example, the onset of photocurrent measured in neurons was much shorter for CaRhGC compared to BeRhGC (23 ms vs. 120 ms) allowing faster response times when used as a light trigger for GMP production. However, even though the photocurrents for CaRhGC and BeRhGC were similar, it was also found that CaRhoGC photocurrents were evoked in only 30% of neurons compared to 70% for BeRhGC. No explanation was offered for this observation. It would therefore be very helpful for the reader if possible causes were listed and how experiments might be designed to test these possibilities.
2) The photocycle characterization of the rhodopsin domain revealed that CaRhGC has a very slow rise time (31 ms) for the M intermediate and even longer than previously reported time for BeRhGC (8 ms). This is an unusual feature for a microbial rhodopsin (e.g. bacteriorhodopsin is only 40 microseconds) and would be worth some discussion.
3) While M is proposed to be the active state for GC activity, there is no direct evidence to support this claim. Furthermore, no data is presented to support the claim that CaRhGC has a higher photo-stability compared to BeRhGC. Finally, it should be noted that these measurements are made in detergent and not a lipid bilayer membbrane where the kinetics can be considerably different. Table I compares the enzymatic parameters for full-length CaRhGC and truncated GC which in some cases causes dramatic change such as for vmax and kcal. While the BeGC (truncated) is also listed, the full-length BeRhBC is not. This would be very useful, especially since an important theme of the paper is to compare the properties of CaRhGC and BeRhGC. I recommend publication provided that the following points can be addressed.

4)
1. In the last part of the discussion, the activation of the new AC domain is compared to that of bPAC. The related OaPAC structure (also cited) has been solved to high resolution in the light and dark states, suggesting the allosteric mechanism involves a strong dynamic element. The authors of this paper apparently prefer the more rigid mechanical model suggested on the basis of bPAC, but perhaps half a sentence more could inform the reader that a role for flexibility is by no means ruled out. This is a good point, indeed there may well be a dynamic element in the RhACs (and GCs). We now include some discussion of the OaPAC activation mechanism and make clear that we do not rule out a role for flexibility (see line 379-381).

Please indicate H-bonds in the legend of Figs. 7D, E, Ffor example "black dashed lines indicate hydrogen bonds between 2.4 Å -3.5 Å in length."
H-bonds are now indicated Table 3.

Show the Wilson B-factors and highest-resolution shell limits in
We now show the Wilson B-factors and the limits of the highest-resolution shell (2.33 -2.249 Å) in table 3.

Reviewer #2 (Remarks to the Author):
This manuscript presents an in-depth analysis of the Catenaria anguillulae rhodopsin-guanylyl cyclase, which is a light-activated enzyme producing guanylyl cyclase with a very large ratio of the activities in the light and in the dark. Moreover, they authors successfully mutated this enzyme into a lightactivated adenylyl cyclase, though with a smaller light-to-dark ratio. The authors also present a crystal structure of the newly developed adenylyl cyclase domain in complex with a substrate analog.
The enzymological work is interesting, the crystallography is sound and the structure shows for the first time a nucleobase bound tightly in its binding pocket on such an enzyme. However, since this is the predicted binding mode, and the rest of the nucleotide clearly binds in a nonproductive manner, the structure does not add a great deal of novelty.
I therefore recommend not to publish the paper in Nature Communications but suggest the authors submit this manuscript to a more specialized journal.
In addition to this there are several issues that would need to be addressed as well. I will list them below: 1. My main concern is with the proposed mechanism of activation. The authors argue that since the cyclase domain alone is active, whereas it is inactive in the dark when coupled to the rhodopsin domain, the rhodopsin acts as a clamp in the dark state, only to "release" the activity upon light activation of the rhodopsin. This may well be part of the story, but probably isn't all that is going on here. Looking at table 1, the light-activated rhodopsin-coupled cyclase is three times more active than the cyclase alone. A possible explanation for this would be that the rhodopsin in its light-activated state affects the dynamics of the cyclase. The authors should have addressed the possibility of other models that are more complex than their current proposal.
We fully agree with the reviewer that in addition to acting as an inactivation clamp in darkness, illumination of the full-length Ca/BeRhGC further increases v max compared to the isolated cyclase domains. We modified the discussion to make this clear and we extended the activation mechanism model, see line 289 ff.
2. Indeed, the authors write in line 225 that they chose a particular dimer interface to analyse because it has the smallest overall B-factor. Does this mean that the B-factors of the others, and therefore the density of the other interfaces was too high for interpretation? That would be not be a problem in and of itself, but more likely a finding pointing to high structural dynamics in itself.
As stated in line 217 we obtained well interpretable electron density for all 8 dimers and dimer interfaces with the average B-factor for the ligand ranging from ~41 A 2 (chain A) to ~76 A 2 (chain J). In all 8 dimer interfaces the electron density shows the same conformation of the ligand and its protein surrounding (except for the β4/5 loop). Some particular residues e.g. the sidechain of R577 or some water molecules are not equally well resolved for all dimers, therefore we specify the dimer composed of chain A/B as the basis of our description.
These differences most likely result from the overall packing of the dimer within the unit cell.
3. Then there is the chemical mechanism. In line 251 the authors write about a "pseudobinuclear nucleophilic substitution SN2 mechanism"… This error was taken almost literally from the review cited. First of all binuclear is a term used for metal complexes. What is meant is probably "bimolecular". This, in turn, is wrong, too. There are first-and second order reactions, and under certain circumstances pseudo-first-and pseudo-second order reactions, but the order of the reaction is something other than its molecularity. Moreover SN2 already means "nucleophilic substitution", so the way this is phrased is double.
We have removed the term "pseudobinuclear" which we originally understood to refer to a reaction catalyzed by two metal centers (ion A and B), albeit only one (ion A) being actively involved (therefore "pseudo").
We re-worded the sentence to "ATP cleavage and cyclization is considered to follow an intramolecular nucleophilic substitution (S N 2), which is initiated through the attack of ribose 3'-OH oxygen at Pα, resulting in a negatively charged αphosphate during the transition state.", line 244. about the current understanding of the mechanism is intended to facilitate the discussion regarding the nucleotide conformation in CaAC, the position of crucial residues defining the ATP vs. GTP selectivity (e.g. R577), the overall conformation of the dimer and the possible mechanism of photoregulation.
*Then, the way the paper is written seems to be aiming at an audience of experts in this field, as a lot of jargon is used. "light-to-dark ratio", for instance, Light-to-dark ratio refers to the ratio of the activity in light to the activity in darkness, we hope this is clearer when we write "light/dark activity ratio" (l. 23, 178) and we replaced "L/D" by light/dark in l. 182. Light-off means: after the light has been switches off, which should be clear.

Another case of a vague use of terms in the legends of fig 2 and 6:
electroporation cannot be done with proteins. It was done with DNA coding for these proteins.
This has been corrected.

l.267 how close is the ribose ring oxygen to S576? Why is this interaction called a "polar bond" (whatever that may be!), and not a hydrogen bond?
We changed this to hydrogen bond, line 261.
6. fig.7 Together, these advances amount to a good, solid paper, yet it is unclear whether they amount to a high-profile paper.

(i) When expressed in neurons, currents evoked by brief irradiation of neurons
expressing CaRhGC were similar to the currents in neurons expressing BeRhGC ( Fig. 2c-2e). There were some differences in kinetics. Were these differences significant to suggest that CaRhGC outperforms BeRhGC? Disturbingly (page 5, lines 106-8), photocurrents were evoked in ~70% of neurons expressing BeRhGC but only in ~30% of neurons expressing CaRhGC. Is retinal incorporation the same in both enzymes? Isn't this a major problem for CaRhGC that outweighs its photostability and somewhat higher activity advantage?
We were also disturbed by this observation and therefore had included the 'success' data in the text and supplementary table 1. We have now produced and tested 3 new constructs and found that adding a YFP tag to the N-terminus greatly improved the reliability without changing the other properties. We additionally found that addition of a mycHis tag to the C-terminus of YFP-CaRhGC (l. 185, 321 -323) worsened it but that changing from a human or mouse codon optimization strategy made almost no difference. This is now mentioned in the results section, the overview of all results in Supplementary (ii) The GC-to-AC conversion by mutagenesis of substrate-binding residues has been known for two decades (as pointed out in the paper), so the conversion of CaRhGC to CaRhAC in itself is not particularly impressive. CaRhGC has a somewhat higher dynamic range (Fig. 5b)  We agree completely with the reviewer that the mutations to convert the GC to an AC does not constitute a breakthrough. We do however stand behind the claim that CaRhAC is truly superior to BeRhAC as an optogenetic tool.
In the discussion we now clearly state how we see CaRhAC in comparison to bPAC, the optoXRs and the jellyfish opsins. CaRhAC raises cAMP much faster than bPAC. If a LOT of cAMP is desired than bPAC is still the tool of choice. The OptoXRs and jellyfish opsins require two extra components in the cell of interest to work and the G proteins may well activate other intracellular signaling pathways in addition to raising cAMP -there is no question that they are also important tools but they don't only raise cAMP. The added discussion can be found in lines 328-333. To answer this concern we have now cloned and tested 3 additional versions of CaRhGC and 'cured' the unreliability. The 'cure' was achieved by adding the Nterminal YFP-tag and ensuring that there was no mycHis tag on the C-terminus.
We also had both humanized and murine-optimized codons in our contructs and have directly tested this and could conclude that this made little difference.  We agree that these data were missing. For resubmission of the manuscript we re-cloned BeRhGC into an appropriate vector for insect cell expression and purified it as a detergent-solubilized protein. We found v max of illuminated BeRhGC was ~6x less than for CaRhGC and that BeRhGC had a reduced light-todark activity ratio. We included the data in Supplementary Fig. 4, and Table 1 and described our results in line 136-141. We further discussed these data in lines 309.