The early steps in vertebrate vision require fast interactions between Rh (rhodopsin) and Gt (transducin), which are classically described by a collisional coupling mechanism driven by the free diffusion of monomeric proteins on the disc membranes of rod and cone cells. Recent findings, however, point to a very low mobility for Rh and support a substantially different supramolecular organization. Moreover, Rh–Gt interactions seem to possibly occur even prior to light stimuli, which is also difficult to reconcile with the classical scenario. We investigated the kinetics of interaction between native Rh and Gt in different conditions by surface plasmon resonance and analysed the results in the general physiological context by employing a holistic systems modelling approach. The results from the present study point to a mechanism that is intermediate between pure collisional coupling and physical scaffolding. Such a ‘dynamic scaffolding’, in which prevalently dimeric Rh and Gt interact in the dark by forming transient complexes (~25% of Gt is precoupled to Rh), does not slow down the phototransduction cascade, but is compatible with the observed photoresponses on a broad scale of light stimuli. We conclude that Rh molecules and Rh–Gt complexes can both absorb photons and trigger the visual cascade.
Phototransduction in vertebrates represents a paradigm for GPCR (G-protein-coupled receptor)-mediated signalling pathways. The cascade is initiated by the absorption of as few as one photon by the membrane receptor Rh (rhodopsin) and the subsequent catalytic activation of the cognate G-protein Gt (transducin) .
The molecular events underlying the rapid activation/deactivation of the cascade in response to the broad variety of light stimuli require fast interactions between all the components. Traditionally, such rapid interactions have been explained by a collisional coupling mechanism driven by the free diffusion of monomeric proteins on the disc membranes of rod and cone cells supported by early observations of the lateral diffusion of Rh on the disc surface [2,3].
Further studies using high-resolution structural biology, however, appear substantially incompatible with the traditional mechanistic model based on free diffusion. In particular, some lines of evidence seem to be at odds with the classical scenario. First, AFM (atomic force microscopy) and electron microscopy indicate that Rh in native mouse discs is organized in highly dense arrays of paracrystalline rafts [4,5]. Secondly, recent experiments on Rh diffusion performed by high-speed dichroic microspectrophotometry in native amphibian and gecko rods  point to a heterogeneous mobility of Rh in disc membranes. The actual diffusion coefficient of Rh eventually results as 3–5-fold lower compared with the classical value, consistent with the existence of a substantial immobile fraction of Rh, whose size can range from virtually 0 to 100% of the disc area, probably reflecting massive oligomeric states of the receptor . Finally, recent AFM determinations  are consistent with Rh being loosely packed in the central zone of the disc membrane, again offering a picture that significantly differs from the classical one.
On the other hand, early studies , previous plasmon resonance technology-based findings [9,10], computational structural analysis  and molecular dynamics simulations  strongly suggest that Rh and Gt may also interact in the dark, thus further complicating the scenario and possibly pointing to some other mechanism of Rh–Gt interaction. Precoupling between the G-protein and the receptor prior to agonist stimulation was predicted as a possible state in theoretical kinetic analysis of GPCRs  and it was experimentally demonstrated for systems such as α2A adrenergic receptors, muscarinic M4 receptors  and the β2-adrenoceptor  by in vivo studies. Interestingly, phototransduction in Drosophila, one of the fastest known signalling systems, was shown to require a scaffolding protein [INAD (inactivation no after potential D)] that dynamically preassembles parts of the molecular machinery, hence ensuring fast and co-ordinated visual signalling . Nevertheless, it is arguable that specific features of the phototransduction cascade in vertebrates – the low concentration of Gt compared with Rh and yet the very fast interactions required – make it hardly compatible with a purely precoupled Rh–Gt state. Such a mechanism, if existing, should be a dynamic phenomenon. To properly address the physiological relevance of such a dynamic precoupled Rh–Gt state in the dark, it is fundamental to obtain kinetic rather than equilibrium information, but this information was missing until now.
In the present study, we show that it is possible to reconcile recent findings on Rh diffusion and assembly with the established activation kinetics of phototransduction by complementing robust biophysical investigation with systems-level analysis of the cascade dynamics. A SPR (surface plasmon resonance)-based methodology developed previously in our laboratory [17,18] to functionally immobilize native bovine Rh on a biosensor chip was used to perform interaction kinetic analyses focusing on different states of the receptor. We especially focused on the currently missing kinetic characterization of Rh–Gt dark binding. In order to investigate the extent to which our findings are compatible with the overall physiological process of phototransduction, the experimental information was integrated into a comprehensive quantitative model of phototransduction in rod cells that explicitly includes most of the molecular components of the cascade . The dynamic pre-assembly of Rh and Gt was found to be fully compatible with the experimentally observed photoresponses over a broad range of light stimuli and does not slow down the photoresponse kinetics. We conclude that an isolated Rh molecule or a Rh–Gt complex involving 25% of the available Gt in the dark can both absorb a photon and normally trigger the cascade.
Preparation of Rh and Gt
Bovine ROSs (rod outer segments) were prepared from fresh bovine retinae and stored at −80°C as described previously . Bovine retinae were purchased from a local slaughterhouse. Rh was extracted from hypotonically stripped disc membranes under dim red light as described previously . Gt was obtained as described previously  and the buffer was exchanged to SPR running buffer (50 mM Mops, pH 7.5, 50 mM NaCl, 3 mM MgSO4, 10 μM CaCl2 and 10 μM MnCl2) by PD10 chromatographic columns (Amersham Biosciences). Protein concentration was determined by Bradford assay  and the purity was checked by standard SDS/PAGE (12.5% gels). The presence of GDP as well as the absence of GTP was confirmed by HPLC. Freshly prepared aliquots of Gt were frozen and stored at −80°C until use.
Rh–Gt binding SPR experiments were performed with a Biacore 2000 apparatus (GE Healthcare) using CM5 sensor chips, on which Rh was immobilized via ConA (concanavalin A) and 1D4 monoclonal antibody as described previously [17,18]. The instrument was modified previously by the addition of a Diode Pumped Nd:Yag solid-state laser emitting at λ=532 nm connected to the sensor chip surface through an optical fibre . ConA (Serva) and anti-Rh 1D4 monoclonal antibody (Chemicon Europe) were immobilized at similar levels, whereas Rh was obtained by solubilization in 25 mM CHAPS at 4°C for 30 min (gentle shaking) and centrifugation (18000 g) for 15 min at 4°C to separate it from insoluble material  (all these and the following steps were performed in the dark). Rh was then immobilized on the sensor chip after 5-fold dilution in 50 mM Mes buffer, pH 6.0, 1 mM CaCl2 and 1 mM MnCl2, to a final concentration of approximately 0.4 mg/ml and injected at a flow rate of 2 μl·min−1 on the flow cell for 25 min in the dark. In order to remove unbound Rh and detergent in excess, the system was washed 5 times with running buffer (50 mM Mops, pH 7.5, 50 mM NaCl, 3 mM MgSO4, 10 μM CaCl2 and 10 μM MnCl2) and then the signal was detected for another 25–40 min, until a stable baseline was reached.
In each Rh–Gt binding experiment, one flow cell on the sensor chip was used as a blank. This surface contained either ConA or 1D4 antibody at similar levels, but was not coated with Rh. The signal from the blank was subtracted upon each parallel injection of Gt in binding experiments. Kinetics was detected following 60 s of injection of Gt at 10 μl·min−1 (association) and then running buffer only was injected for a time varying from 60 s to 20 min (dissociation). Experiments were repeated several times at different Gt concentrations and Rh immobilization levels (Table 1). After a series of illumination experiments, the pigment was regenerated by flowing 10 μM 9-cis retinal for 3.5 h (5 μl·min−1) followed by repetition of several washes and a ~30 min stabilization of the baseline.
|kon (M−1·s−1)||koff (s−1)||KD (nM)|
|kon (M−1·s−1)||koff (s−1)||KD (nM)|
Results from 19 replicate experiments.
Results from 7 replicate experiments.
Results from 9 replicate experiments.
To study the direct interaction of Gt with a lipid bilayer we preceded as follows. Liposomes were prepared as described previously . Briefly, a mixture of 4 mg of lipids with ROS membrane composition [40% (v/v) phosphatidylethanolamine, 40% (v/v) phosphatidylcholine, 15% (v/v) phosphatidylserine and 5% (v/v) cholesterol] was dried by vacuum in a SpeedVac concentrator. The sample was resuspended in 2 ml of degassed buffer (20 mM Hepes, pH 7.5, 150 mM KCl and 3 mM EGTA) and sonicated for 2×15 min using a Bandelin Sonorex. Liposomes were produced using the extrusion technique. The suspension was soaked for 15–20 min and extruded through polycarbonate filter with a pore diameter of 100 nm. A sensor chip L1 (GE Healthcare) was used to immobilize the liposomes by injecting 150 μl with a flow rate of 5 μl/min, and the stability of the formed bilayer was checked for approximately 1 h before the injections of Gt.
Investigation of the oligomeric state of Rh immobilized on the sensor chip
Experiments aimed at assessing the oligomeric state of Rh immobilized on the sensor chip via ConA or 1D4 antibody were performed after direct immobilization of Rh on the sensor chip via amine coupling (5-fold dilution in 10 mM sodium acetate, pH 4.3; running buffer: 10 mM Hepes, pH 7.5, 150 mM NaCl, 3 mM EDTA and 0.005% Tween 20). Rh was covalently bound to the dextran matrix via free amines located either at its N-terminus or at the side chain of one of the ten lysine residues. These latter are situated mostly at the intracellular region, but not in the transmembrane domain, hence allowing for effective interactions with other Rh molecules through the transmembrane moiety. Reactive carboxyl groups in excess on the dextran matrix were blocked with 1 M ethanolamine, pH 8.5. One flow cell was similarly activated and deactivated, without any immobilization of Rh, and used as a blank. In order to study Rh–Rh interactions, the initial dilution step after detergent solubilization was performed in Mops running buffer (as described above), in order to minimize bulk effects upon injection of Rh. Rh injections over 3 min were performed at different concentrations in the range 1.5–7.6 μM after specific dilutions. The concentration of injected Rh was determined by comparing the absorption at 500 nm in the dark and after complete bleaching, using a molar extinction coefficient equal to 40600 M−1·cm−1 . The Rh–Rh binding experiments were performed in the dark after removing non-covalently bound Rh from the surface of the sensor chip via a short pulse of 0.25% SDS (5 μl, 30 s). Binding experiments were repeated with different immobilization levels and concentrations of Rh in the flowing phase and the results were averaged. In each binding experiment, a significant amount of Rh remained bound to the chip, indicative of the formation of a very stable Rh–Rh complex in the tested conditions.
Determination of the delipidation profile of immobilized Rh
By keeping track of the volume of running buffer flowing over the flow cell in which Rh was immobilized during the five washing steps (approximately 342 μl, RINSE protocol in the Biacore protocols, repeated 5 times) and adding the subsequent volume flowed during the stabilization procedure (40 min with high flow of 20 μl/min), the Rh-containing FCV (flow cell volume) (~0.02 μl) was, overall, flow-dialysed against a 57000-fold excess volume of running buffer. The delipidation profile was assessed by collecting ten 10 μl fractions of sample dissociating from the flow cell immediately after the immobilization of Rh via ConA, which occurred after washing the recovery cup with running buffer. Each fraction represents the result of a consecutive 500-fold FCV dilution and allows a determination of the course of delipidation of Rh bound to the sensor chip surface. This can be assessed by monitoring the SPR signal in the same time frame in which the fractions are collected. Overall, the ten fractions collected represent a 5000-fold flow dialysis of the FCV. Each fraction was analysed separately to determine the lipid content using the Stewart assay . The lipids were also determined in the sample initially used for Rh immobilization and the concentration of lipids in each fraction was related to that initial value (100% of initial lipids) to assess the percentage of initial lipid concentration lost in each fraction with the progress of the flow dialysis. The same applied to the amount of Rh lost with each fraction, hence referring to the initial level of Rh immobilized (100% of initial Rh). The delipidation and Rh loss profiles could then be both followed quantitatively at each 500-fold FCV washing step. The experiments were repeated twice.
Kinetic data analysis
Kinetic data were analysed by starting from the dissociation phase as follows:
Dark Rh–Gt binding: The dissociation process was successfully described as a first order exponential, directly yielding koffdark, whose value was fixed in a standard pseudo-first-order reaction to yield the fitting of the association SPR response according to:
in which kdarkon is the association rate constant, R0 is the baseline value and Rmax indicates the maximum response expected if all the injected Gt bound Rh (saturation).
Rh* (photoactivated rhodopsin)–Gt binding: In order to account for the concomitant processes of Gt binding to Rh and Rh*, the dissociation phase was fitted as the sum of two exponential decays:
in which koffdark was fixed to the average value determined in dark binding experiments and klightoff was determined by the fitting procedure (other parameters are for the baseline and the number of Rh* following the stimulus). Accordingly, the association phase was fitted by a sum of two pseudo-first-order terms:
in which kdarkon, koffdark and klightoff were fixed according to the values determined above in independent fittings.
Opsin–Gt binding: The dissociation phase and the association phase were successfully fitted according to a single exponential and a pseudo-first-order exponential respectively, directly yielding koffopsin and konopsin, as performed for the case of dark Rh.
The kinetic constants obtained in each experiment were averaged and the mean and its standard error were reported (Table 1). The equilibrium dissociation constant KD for each Rh variant was computed as the ratio between the koff and the kon values (Table 1). In order to transform the volume-related constants into more convenient surface density (2D) units of physiological relevance as referred to the situation on rod discs surface, we used the relation elucidated by Heck and Hofmann :
in which KD is the volumetric dissociation constant obtained as described above, [R]2Ddisk is the average concentration of Rh on the disc surface (48000 μm−2 ) whereas [R]3Dchip is the volumetric concentration of the Rh immobilized on the sensor chip (on average 12.4 μM in all the experiments considering a 0.02 μl volume for the flow cell).
The mathematical model of rod phototransduction including 91 reactions, 71 molecular components and 63 parameters developed previously  (BioModels ID: BIOMD0000000326) and validated over a broad range of experimental data on normal and genetically manipulated rods was modified by inserting a reversible precoupling reaction with the kinetic constraints resulting from SPR experiments, that is:
in which the kinetic rate constants were constrained to be kdarkon=1.6×klighton and koffdark=315×klightoff, according to SPR data (Table 1).
Similarly to previous studies , the stimulus was set to define the velocity of the activation reactions in the cascade, and was defined as a flash of specific duration and intensity:
where flashMag is the potential number of photoisomerizations led by the flash, proportional to its intensity, and flashDur (24 ms) is the duration of each flash. Based on such definition, the number of photoactivations of either Rh or Rh–Gt preformed complexes will be proportional to the stimulus and the relative abundance of the pigment form that defines the probability of the process. In the case of the preformed complexes the following rate expression stands:
where d is part of the time derivative d/dt, and the number of molecules rather than their molar concentration are considered, and RhTOT represents the pigment overall available in the cell, independent of its binding status.
Parameter estimation for klightoff, klighton and kRKo (the rate constant of unphosphorylated Rh to Rh kinase binding) was necessary and sufficient to reproduce the benchmark experiments described previously . The estimated parameters were within a factor of two compared with the prior values reported previously , and none of the remaining 60 parameters needed to be newly estimated, indicative of robustness of the proposed network.
RESULTS AND DISCUSSION
Gt interacts with every Rh state with specific kinetics
Native bovine Rh and Gt were extracted from dark-adapted retinae. Two different coupling strategies, which have been investigated previously  were used to immobilize detergent-solubilized Rh on the surface of a biosensor chip at sufficient concentration (average density of 0.2 pmol·mm−2, corresponding to approximately 12 μM concentration), namely: (i) immobilization via high affinity binding of the carbohydrate moiety of Rh to the lectin ConA ; and (ii) immobilization via 1D4 monoclonal antibody, which recognizes the last 8 amino acids of the C-terminal tail of Rh . Our recent results employing the same immobilization strategies pointed to no significant difference in the kinetics and in the affinity of Rh*–Gt interaction as detected by the two coupling techniques . This unexpected result suggests that Rh is dimeric on the surface of the sensor chip, and whereas one monomer is involved in the coupling with the chip through either of its terminals, the other one is free to interact with Gt with an ideal stoichiometry. In order to test such a hypothesis, we performed Rh–Rh binding experiments after immobilizing Rh on the sensor chip via direct amine coupling. The coupling was heterogeneous, as any amine group, at the N-terminus or at lysine residue side chains, might couple to the dextran matrix. However, the spatial distribution of such groups is mostly at the intracellular domain (apart from Lys16, which is in the intradiscal region), thus keeping the transmembrane region free for interaction with another Rh molecule. After the direct immobilization of Rh, a short pulse of detergent (0.25% SDS, 30 s) was injected in order to dissociate potential oligomeric forms of Rh formed on the chip, resulting in a significant reduction of the SPR signal (>80%), indicative of a substantial dissociation process. After checking for baseline stability, injections of Rh in the range 1–8 μM were performed, resulting in a slow but stable binding, in which a significant amount of injected Rh remained bound to Rh previously immobilized on the chip (Figure 1). Short pulses (30 s) of regeneration buffer (Mops buffer containing 100 mM methyl-α-D-mannopyranoside and 30 mM octyl-β-D-glucopyranoside) following the binding experiments allowed a significant dissociation of high-affinity Rh–Rh complexes. Taking into account the bulk effect due to the presence of detergent in the injected sample, the data were effectively fitted with a simple 1:1 binding model, leading to a submicromolar affinity (on average 17 nM for four repeated experiments). Therefore, we conclude that most, if not all, of the Rh immobilized via ConA or 1D4 antibody and used in the Gt binding experiment was dimeric, as hypothesized.
SPR study of Rh–Rh interaction
For the Rh–Gt interaction experiments, the immobilization and binding steps were performed in the dark, except for those concerning the illumination of the sensor chip by pulses of green light (λ=532 nm) delivered via an optical fibre connected with a laser source as described previously . In every case, bleached pigments could be regenerated with 9-cis-retinal after any cycle of repeated illumination, yielding 80–90% of the former response. The purity of heterotrimeric Gt samples used for binding experiments was confirmed by SDS/PAGE (12.5% gels) (Figure 2a).
Gt purification and SPR study of Rh–Gt binding
Figure 2(b) shows an overlay of representative sensorgrams obtained after a 60 s injection of 0.33 μM Gt over 0.22 pmol·mm−2 Rh immobilized on the sensor chip. Further flowing of running buffer only for 60 s monitored the dissociation of the protein–protein complex. The observed kinetics was found to strongly depend on the receptor state. Very fast associations and dissociations were observed when the binding experiments were performed in the dark (broken black line in Figure 2b). The measured rate constants are reported in Table 1. The high association rate (kdarkon=4.2×105 M−1·s−1) is paralleled by an extremely fast dissociation (koffdark=0.148 s−1), leading to a relatively high 0.36 μM affinity, which is in line with the values reported in previous equilibrium studies (64 nM–10 μM) [9,10,28]. When the same Rh sample was illuminated for 4 s, sensorgrams showed a significantly higher response amplitude (Figures 2b and 2c, continuous black line) with a slightly slower association and a significantly slower dissociation, in agreement with our previous studies [17,18] and the known high stability of the Rh*–Gt complex .
Association and dissociation rates decreased even more when the binding experiment was repeated on a preilluminated Rh-coated sensor surface (Figure 2c, broken black line). The highly stable Rh*–Gt complex rapidly dissociated after injection of 100 μM GTP for 60 s. In general, the kinetics of Rh*–Gt dissociation were followed for up to 20 min (see an example in Figure 2d). The association kinetics was found to depend strongly on the amount of Rh* (i.e. illumination time) as observed previously . This is apparent when comparing the sensorgrams obtained after 2 s and 4 s illumination (Figure 2c). However, the dissociation process also varied with the illumination protocol. The initially fast phase in the dissociation process was due to the formation of dark Rh–Gt complexes that rapidly dissociated, and it significantly reduced when the illumination time was prolonged from 2 (continuous grey line) to 4 s (continuous black line).
By taking the concerted phenomena of Gt binding to either Rh or Rh* into account in the fitting procedure, the kinetic parameters for Rh*, reported in Table 1, were found to be independent on the illumination time. Interestingly, the association of Gt with dark Rh appeared to be approximately 1.6-fold faster compared with Rh*, whereas the dissociation of dark Rh–Gt complexes was found to be approximately 315-fold faster compared with the photoactivated complexes (Table 1). The rate constants used for comparison were the result of kinetic analyses performed on several replicates of the same binding experiment with different Gt concentrations and Rh immobilization techniques, as well as different amounts of immobilized Rh. The fitting procedure was performed according to the respective kinetic models (see the Experimental section) and gave satisfying results as shown in Figure 3.
Examples of SPR kinetic analysis of dark and photoactivated Rh binding to Gt
Bleaching the pigment completely by applying 30–40 min of continuous illumination with the same light source allowed the study of the the binding kinetics of Gt to opsin, the chromophore (ligand)-free form of Rh that is able to constitutively activate the phototransduction cascade with lower efficiency [29,30]. The binding experiments with opsin showed a greater variability compared with the other receptor states, probably due to some intrinsic lower stability. Figure 4 shows a series of four replicates of the same 60 s injections of 0.33 μM Gt over opsin. Greater variability concerns mostly the maximum amplitude and to a lower extent the estimated rate constants. The kinetic analysis showed some interesting features (Table 1). Although the association occurred approximately 1.8-fold faster for opsin compared with Rh*, the dissociation was approximately 3 times faster (but still 106-fold slower than with Rh, Table 1), hence resulting in a slightly lower affinity (1.8±0.7 nM for Rh*, compared with 2.7±1.5 nM for opsin). However, we point out the great instability observed in all the opsin experiments, which are reflected by the high relative error.
Examples of four repeated 60 s injections of 0.33 μM Gt over a sensor chip coated with opsin after full bleaching of the formerly active visual pigment
This, on the first glance, surprising and yet reproducible result appears to be at odds with the known catalytic potency of opsin toward Gt activation, which was shown to be 30-fold  to 106-fold  lower than that of Rh*. However, SPR measurements only allow the evaluation of binding constants, not that of catalytic (enzymatic) rates. The data therefore reflect the difference between binding to, and activation by, opsin to Gt respectively.
The interaction between Rh and Gt in the dark is a specific protein–protein interaction
In order to probe whether the highly transient interaction detected between Rh and Gt in the dark could reflect unspecific binding of the Gt lipophilic anchors to potential lipids remained bound to solubilized receptors, we proceeded in two different ways. First, we employed liposomes containing the same lipid composition of bovine ROS discs by immobilizing on a sensor chip a phospholipid bilayer resembling that in discs, as performed previously . We then studied the direct interaction between Gt and the phospholipid bilayer in the same conditions as performed in the presence of Rh (Figure 5). Injections of the same amounts of Gt led to different kinetics of interaction in the absence of Rh. The fast association was followed by a significantly slower dissociation in which 25% of the Gt remained bound to the bilayer surface. In contrast, in all the experiments performed in the dark, a very fast and complete dissociation between Rh and Gt was observed (Figure 2 and Table 1). Secondly, we proved that after Rh immobilization, a set of 5 washing steps and further high flow stabilization leading to a ~57000-fold dilution in running buffer of the FCV (see the Experimental section) resulted in a highly efficient flow dialysis system. Indeed, the profiles shown in Figure 6 suggest that, although in 5000-fold FCV dilutions the amount of Rh was reduced by only 2% with respect to the initial immobilized level, phospholipids continuously dissociated from the flow cell upon further washing. The lipid concentration reached the detection limit in this dilution frame, 11-fold lower than the one used in our binding experiments (Figure 6), thus ensuring a substantial delipidation of Rh on the chip.
Gt interaction with a model bilayer
Rh and lipid concentration after flow dialysis
Systems-level analysis of the phototransduction cascade in the presence of a dynamic scaffolding interaction between dark Rh and Gt
The interaction between Rh and Gt in the dark was found to be highly transient and occurred through a fast association and an extremely fast dissociation (Table 1 and Figure 2b). How does this interaction process relate to the actual situation in photoreceptor cells? Two specific conditions of our on-chip experiments need particular attention: Rh is solubilized and delipidated, and the interaction with Gt occurred in solution, in the volume of the sensor chip flow cell. Therefore, it is necessary to transform the volume-affinity constant measured in the present study into one which refers to the surface densities of Rh and Gt on rod discs . By doing so (see the Experimental section), the seemingly low KDdark converts into 673 μm−2, leading to approximately 97% of Gt prebound with moderate affinity to Rh at equilibrium (taking approximately 3000 μm−2 surface density of Gt). Nevertheless, this estimated value as well as the overall kinetic properties measured by SPR need to be considered in the context of the whole molecular machinery to assess the compatibility with the known dynamics of the phototransduction pathway.
The peculiar nature of Rh, and specifically its ability to detect even single photons, prevents any investigation at the cell level of the interactions between Rh and Gt in the dark by means of spectroscopic techniques such as FRET (fluorescence resonance energy transfer) or BRET (bioluminescence resonance energy transfer). However, the availability of a comprehensive dynamic model of phototransduction in rod cells, which was able to describe quantitatively the photoresponses starting from the underlying biochemistry  prompted us to test our findings in their general physiological context. The mathematical model, validated over a broad set of experimental conditions involving different species and illumination ranges includes all the main components of the cascade (71 molecular species including proteins, ions and nucleotides) and the reactions between them that give rise to complex systems properties such as light adaptation. Such a detailed holistic description of the cascade activation/deactivation imposes strict constraints on the kinetic parameters in the model  and represents an ideal benchmark where to probe the physiological compatibility of our experimental finding at a whole network level.
In order to investigate Rh–Gt interaction in the dark, we hence included a further reaction (see the Experimental section) in the phototransduction network model described previously  to describe a transient dark binding as observed by SPR. The relative kinetic constants instead of the absolute values of our kinetic analysis would be significant for this purpose, as they intrinsically account for the molecular determinants responsible for the different timing observed in the interaction between Gt and Rh or Rh*, independent of the environment. The new reaction thus followed the same relative kinetic relationships found in SPR experiments, that is kdarkon=1.6×klighton and koffdark=315×klightoff (Table 1).
Very interestingly, setting the above kinetic constraints for the precoupling reactions and simulating the system in the dark led to a very rapid equilibration, in which ~25% of the Gt was found to be dynamically precoupled to Rh in the dark, a value that significantly differs from the 97% estimated above, by neglecting the constraints imposed by the other components of the cascade and their dynamics. Hence, the numerical simulation at a systems level suggests that, at any time in the dark, there is approximately a 38-fold excess of Rh uncoupled to Gt, compared with Rh–Gt able to capture a photon (or 97–98% of the overall available Rh). The binding/dissociation between the two proteins in the dark demonstrated experimentally is very fast, and the simulations suggest that it leads to a rapid establishment of these equilibrium values. The continuous fast association and dissociation between Rh and to Gt in the absence of light stimulus is thus compatible with a ‘transient scaffolding’ mechanism that is highly dynamic in Nature.
Since a photon could hit either an isolated Rh molecule or one that is transiently coupled to Gt, the extended phototransduction network explicitly included the possibility for a flash light stimulus of fixed duration and intensity to be captured by: (i) a Rh molecule uncoupled to Gt; (ii) a precoupled Rh–Gt complex; or (iii) both. The probability for the stimulus to be captured by Rh or Rh–Gt was set to depend on their relative abundance at the equilibrium in the dark (see the Experimental section). When the light stimulus was assumed to be proportionally absorbed by both forms, the shape and quantitative dynamics resulting from simulations were indistinguishable from those observed experimentally in rod cells (Figure 7a). Thus the dynamic scaffolding hypothesis with the established kinetics is fully able to reproduce the electrophysiological data (see  for direct comparison with experimental data) and is compatible with the kinetic constraints imposed by the whole network of interactions.
Numerical simulation of rod photoresponses to flashes of increasing light intensities and dynamic scaffolding mechanism
This systems level analysis further allows simulating an ideal experiment to assess the possible effects on the timing of the overall cascade if only precoupled Rh–Gt complexes, and not free Rh, ideally captured the same light stimulus. Our simulations clearly show that the effect would be simply a reduction of the overall cascade efficiency, as the shape and the recovery of the photoresponses both appear normal (Figure 7a, red lines). However, not a single biochemical event would be slowed down, including the deactivation steps. Taken together, our simulations suggest that the stimulus might be captured by any of the Rh states, including the one precoupled with Gt, the only discriminator being the probability of such an event, which depends on the abundance of the preformed complexes relative to the available pigment and could be modified, for instance, in pathological conditions.
In conclusion, our findings are consistent with a dynamic scaffolding mechanism for the Rh–Gt interaction, represented schematically in Figure 7(b). Such a dynamic mechanism does not require the presence of specific scaffolding proteins, which for instance is the case in Drosophila photoreceptor cells [16,32]. The structural organization of the molecular machinery in vertebrate discs seems, in fact, sufficient for the purpose. The organization of Rh dimers in supramolecular architectures such as the suggested paracrystalline rafts could facilitate the anisotropic diffusion of Gt in the restricted lipid areas among the receptor rafts despite the high concentration of Rh. This counterintuitive possibility, difficult to probe experimentally, is in fact supported by recent mesoscopic Monte Carlo simulations accounting for geometrical features of the diffusing proteins . However, besides diffusing in the lipid milieu, Gt has a propensity toward protein–protein interactions with Rh in the dark. These two concerted and opposing diffusion/binding phenomena give rise to a dynamic ‘hopping’ of Gt on to dark-adapted Rh, hence building up a convenient scaffold that maintains a dynamic equilibrium in which approximately one out of four Gt proteins are actually bound to Rh at any time independent on the illumination state.
The mechanistic model proposed in the present study seems to answer some of the questions raised on the coupling between Gt and Rh  and finally allows a consideration of the phototransduction cascade as not an exception among the many GPCR-mediated signalling pathways. In particular, the novel concept that interactions between the receptor and the G-protein may occur at any time independent of the stimulus in the case of Rh and Gt is thought to be of relevance for retinal diseases involving photoreceptors. Future work is intended to explore the possible implications.
Daniele Dell'Orco and Karl-Wilhelm Koch conceived this work. They were both involved in the overall design, data analysis and manuscript preparation. Daniele Dell'Orco was also involved in experimental data collection and simulations setting.
We thank Dr Konstantin Komolov for helpful discussions and comments on the paper, and Marco Aquila and Alessandro Colombelli for technical help with preparation of Figure 7(b). Valuable technical assistance by Werner Säftel is gratefully acknowledged.
This work was supported by an A. von Humboldt Research Fellowship (to D.D.O.) and grants from the Deutsche Forschungsgemeinschaft [grant number KO948/7-2 (to K.W.K.)].