Mouse rod cells communicate with cone cells through direct connections

The kinetics protocol was delivered in rods as a control (Figure 2A1,A2). As expected, the first dim G flash evoked a large response, the bright_a G flash evoked a saturating response consisting of a fast peak and plateau (for an analysis of the currents involved, see Della Santina et al., 2012), and dim G flash sensitivity recovered slowly after the saturating flash. Rod responses run down in kinetics during patch recordings (Cangiano et al., 2012), a process that also alters their time course of recovery from saturating flashes (Figure 2A1, cf. black and gray records). Thus, the kinetics protocol was also delivered in a number of rods recorded in loose seal mode, as with this technique, light response kinetics can be stable for several hours (Figure 2A2). The time course of the response to the kinetics protocol of rods in the initial minutes of patch recordings (n = 19, Figure 2A1, black records, taken before a significant amount of rundown occurred) was similar to that in loose seal recordings (n = 17), as shown in a graph of the normalized dim flash response amplitudes (Figure 2B). In cones, the kinetics protocol evoked a spectrum of response types depending on the cone and on the time from seal (see below). Some cones responded only to the bright_a G flash (Figure 2C, top), the expected behavior for uncoupled cones (Figure 1). However, the large majority of cones displayed, similar to rods, responses to the first dim G flash, a plateau after the bright_a G flash, and a slow recovery of the dim flash response (Figure 2C, middle/bottom). This similarity in the time course of recovery after the bright flash emerges from comparing graphs of normalized dim flash response amplitudes during the kinetics protocol in rods (Figure 2B) and cones (Figure 2D, n = 11). This is strong evidence for the presence of rod input in cones.

We observed in cones a progressive increase in dim and bright flash peak response amplitudes and in the plateau. The net change in their response to the kinetics protocol was a scaled version of the typical rod response (Figure 3A). The rapid time course of this process implied that it began with the recording itself. It did not depend on the repeated exposure of the photoreceptor to light, since it occurred also when single deliveries of the kinetics protocol were separated by several minutes of uninterrupted darkness (Figure 3A). A response amplitude vs flash strength graph, obtained in two time ranges in the same cone (Figure 3B), highlights how cones acquired light sensitivity at intensities normally covered by rods. Note that collecting each full flash response curve of Figure 3B required ∼10 min, a time comparable to the time course of the spontaneous increase; therefore, different flash strengths were unavoidably delivered at different levels of progress of the phenomenon. Figure 4 shows, in simplified form, the evolution of dim G flash response amplitude in 50 cones. Its rate of increase varied greatly among cones. When possible we estimated the dim G flash response amplitude prior to patching, by extrapolating to time zero the values observed in the first minutes of recording (Figure 4, short red horizontal bars). Based on these estimates, it appears that most cones responded to dim G flashes already before patching, although response amplitudes were generally modest when compared to those expressed during the recordings. If the dim flash sensitivity of cones is due to rod input, then these observations imply that rod–cone junctional conductance is not hardwired, but it is modifiable in a wide range.

We attempted to isolate the pure cone component by saturating rods with a continuous light background of 6100 photons·µm⁻²·s⁻¹ at 520 nm (for 5 min) based on ex vivo ERG data in mouse (Heikkinen et al., 2011) and our own rod data obtained with patch clamp and in loose seal. Control recordings confirmed that this particular background was, indeed, rod saturating (Figure 5A1) and showed that, upon returning to darkness, rod responses to the kinetics protocol recovered to near control levels after ∼2 min (n = 3). This is presented in Figure 5A2, where dim G flash, bright_a G flash, and slow plateau response amplitudes are plotted normalizing them to their control values before exposure to the background. The same background was delivered in cones expressing prominent dim G flash responses. In contrast to rods, a sharp response to the bright_a G flash persisted during the background, while the slow plateau disappeared as one would expect if it was generated by rods (Figure 5B1). The sharp peak did not represent a residual rod response, since increasing flash strength led to a marked increase in its amplitude (Figure 5B1, arrow). Importantly, when darkness was restored, the dim and bright flash responses recovered to near control levels in ∼2 min (n = 3), in agreement with what was observed in rods (cf. Figure 5 B2 with A2). This strongly suggests that the peculiar features displayed by cones—a high light sensitivity and slow kinetics—may be entirely explained by rod input fed into the cone pathway through GJs.

Heikkinen et al. (2011) interpreted the effects of non rod-saturating light backgrounds on the cone-driven ERG flash response as evidence that light uncouples rod–cone GJs, with a time course of minutes. Given our observation that the rod component in cones recovers from a prolonged rod-saturating background with a similar time course as the rods themselves, we could not confirm the conclusions of that study. It is possible, however, that the process leading to a progressive increase in rod–cone coupling during patch recordings might interfere with the relevant signaling pathways.

In the presence of rod input, one should observe a shift in the spectral preference of S- and S/M-cones toward that of rods. The spectral protocol (Figure 1D) enabled us to rapidly determine the apparent spectral preference of cones for dim and bright_b flashes, as well as their intrinsic spectral preference by removing any rod contribution with a rod-saturating pre-flash. We delivered the spectral protocol in rods and cones and quantified spectral preference by the ratio of the peak response amplitudes to G over UV light (for the same flash strength in photons·µm⁻²). As expected, rods were more sensitive to dim G than to dim UV flashes (Figure 6A1,A2, arrows; same rod), with an estimated dim G/UV ratio of 2.7 (SEM 0.2; n = 8) (all rhodopsins have a prominent secondary absorption peak in the ultraviolet [Rodieck, 1973], including mouse rhodopsin [Lyubarsky et al., 1999]). For both G and UV bright_b flashes, rods expressed saturating responses (ratio of 1; Figure 6A2, filled box), while they did not respond to bright_b flashes delivered after a rod-saturating pre-flash (Figure 6A1, empty box). Figure 6A1,A2 shows a cone exposed to the same spectral protocol delivered in rods. Surprisingly, while for dim flashes we observed a larger response to G than to UV light (Figure 6A1,A2, arrowheads; same cone), the response to bright_b flashes delivered after the rod-saturating pre-flash showed that this cone had an intrinsic UV preference (Figure 6A1, empty circle) and was therefore a mixed S/M cone. We found that all cones stimulated with dim flashes displayed larger responses to G than to UV light (n = 20). This behavior was independent of their intrinsic spectral preference in a wide range of G/UV ratios (Figure 6B) (corresponding to a wide range of M- vs S-opsin expression levels), again supporting our hypothesis that dim flash responses in cones are driven by rods.

An important question is whether the cones’ maximal strength of coupling to rods varies with their dominant opsin type. We examined this by plotting the maximum dim G flash response amplitude observed in each cone vs its intrinsic spectral preference (Figure 6C). The graph shows that both M- and S-dominant cones could possess or acquire dim flash sensitivity during recordings. Performing a finer analysis under our experimental conditions would be complicated by the fact that the maximum level of coupling displayed by cones is strongly dependent upon recording duration, which varied widely. We also recorded from three putatively pure S-cones (i.e. blue cones; Figure 6C, leftmost circles), an important sample since the immunohistochemically derived frequency of these neurons is very low (Haverkamp et al., 2005). In one of these cones (Figures 4, 6C, 7C, black triangle), the seal was maintained for >85 min, but no coupling was detected. The remaining two cones were recorded for only 4 and 7 min and thus (i) little averaging of their responses to the spectral protocol could be performed to improve signal over noise, and (ii) the possible development of coupling could not be monitored. With these limitations in mind, in one cone, no coupling could be detected, while in the other, a dim G flash response of ∼0.3 mV may have been present although superimposed on a noisy baseline.

Figure 7 also serves the purpose of illustrating in greater detail three cones shown in previous plots (Figure 4 and Figure 6C, five-pointed stars and black triangle) exhibiting different relative opsin expression levels: an M-dominant cone initially uncoupled, which then develops strong coupling (panel A); a weakly coupled S/M cone, with approximately equal sensitivity to G and UV light (panel B); an uncoupled presumably pure S-cone (panel C).

If, as we hypothesized, the cone dim flash response originates in coupled rods, while the bright flash response originates in both photoreceptors, they should have a different apparent reversal potential (assuming the actual reversal potential E_OS in rods and cones to be the same; E_OS ≈ 0 mV; Luo et al., 2008). The reason is that the cone membrane potential at the pipette tip (V_Cone) would differ from the membrane potential of its coupled rods (V_Rod) when a current flows through the finite resistance of the GJs (i_GJ), a condition occurring when the cone is held depolarized via the pipette (Figure 8A, equivalent circuit). In essence, any coupled rods would represent electrotonically distant compartments from the patch pipette. The prediction is that there should be a range of depolarized cone membrane potentials (beyond E_OS) within which V_Rod(DARK) < E_OSand dim and bright flash responses display opposite polarities (Figure 8A, expected responses: bottom row) (note that, since homotypic Cx36 GJs decrease their conductance in the presence of large voltage differentials (Bukauskas, 2012), it is entirely possible that all one would observe when V_Cone > E_OS is an uncoupling, that is disappearance of dim flash responses). If, on the other hand, dim and bright flash responses originated entirely in cones, their polarity should always be concordant.

A full analysis of the voltage-dependence of the photoresponses could not be carried out due to the instability of recording at positive membrane potentials and the need of acquiring several responses for averaging. We chose instead a more limited approach of delivering the spectral protocol in cones sensitive to dim flashes, while they are depolarized well beyond the reversal potential of their photoresponse. This was done both in current clamp (CC) and voltage clamp (VC). In CC, we imposed V_Cone(DARK) > E_OS by constant current injection. In all five cones tested, we observed that the bright_b flash responses reversed in polarity (i.e., depolarizing) and actually became larger than in control conditions. In contrast, four of these cones had their dim flash responses reduced in amplitude to the point of being no longer distinguishable from noise, while the remaining one retained a small hyperpolarizing response (Figure 8B). This result suggests that in these conditions, V_Rod(DARK) was slightly below or anyway near E_OS (Figure 8A, middle and bottom rows). In VC, the photocurrent responses of a cone to the spectral protocol were recorded at holding potentials (V_Cone(HOLD)) of −40 mV and +60 mV. At the depolarized potential, bright flash photoresponses were reversed and of much larger amplitude, while the dim flash responses became undetectable (Figure 8C). This suggests that also in this case (VC recording) V_Rod(DARK) was close to E_OS (Figure 8A, middle row). Again, it is possible that the absence of dim flash responses was partly due to a voltage-dependent reduction of the junctional conductance.

As a further confirmation of the results obtained thus far, we tested the specific GJ blocker meclofenamic acid (MFA), previously known to be effective at GJs containing Cx36 in AII amacrine cells (Pan et al., 2007; Veruki and Hartveit, 2009). In three control rod recordings, 100 µM MFA was superfused with essentially no effect, except for a marginal reduction in light sensitivity that we attribute to the normal rundown observed during long rod recordings. In four cones tested, on the other hand, superfusion with 100 µM MFA markedly reduced both dim flash responses and the slow plateau after bright flashes. An example is shown in Figure 9A1, where a prominent level of coupling was reached about 30 min from seal formation, after which MFA was delivered. The slow pharmacodynamics of the blocker observed in our recordings (Figure 9A1, lower plot) confirmed a previous report in AII amacrine cells (Veruki and Hartveit, 2009). The significant rundown of the cone response kinetics, particularly evident toward the end of the experiment, was first described in Cangiano et al. (2012) and is unrelated to the effect of the blocker, since fast cone responses were observed when the seal was made with MFA already present in the bath (n = 2). Comparing graphs of response amplitude vs flash strength before and during perfusion with MFA in the same cone highlights how junctional coupling transforms the sensitivity profile of the cone making it responsive to dim flashes, thereby widening its dynamic range (Figure 9A2). Although control data for this graph were obtained before the cone reached its full coupling potential, a remarkable effect of the blocker on dim flash sensitivity was already visible.

Uncoupled cones, including those recorded during superfusion with MFA (Figure 9—figure supplement 1), displayed markedly decreased flash sensitivities compared to those found in Cangiano et al. (2012), and approached those predicted using published estimates from functionally rodless retinas (Figure 1A,B, simulated curves on the right; ‘Materials and methods’).

As the experiments shown so far were performed near room temperature, we verified the occurrence of rod–cone coupling also near body temperature. The response to the kinetics protocol of a rod recorded in loose seal (Figure 10A) showed, as expected, faster kinetics near body temperature (cf. Figure 2A2). Nevertheless, the rod was still unable to respond to the dim G flash delivered at the earliest delay following the bright_a G flash. In all five cones recorded near body temperature, we observed dim G flash responses (0.28, 0.32, 0.43, 1.47, and 1.70 mV) and plateaus after bright_a G flashes (the response of 1.70 mV is shown in Figure 10B). In one of these cones, we tested the effect of the GJ blocker MFA (100 µM): both dim flash responses and plateaus following the bright_a flash were abolished, leaving behind a fast response to the bright_a flash (Figure 10B). The flash response profile of this cone, acquired during superfusion with MFA (Figure 10B), revealed that it was significantly more sensitive to G light than the dorsalmost (i.e., the ‘greenest’) simulated cones in Figure 1A. This was not necessarily unexpected, given that the simulated profiles: (i) are based on models which predict the average ratio of S- to M-opsin expression over a large cone population, and (ii) assume that all cones express the same amount of opsin (‘Materials and methods’).

The progressive increase in rod–cone coupling observed in our recordings was not evoked by slicing of the retina but rather by some interaction between the recording pipette and the cone and/or its nearby rods: cones were patched not earlier than 1 hr after slicing and in some cases, after several hours, yet in most cones, initial coupling upon sealing was weak, developing rapidly thereafter (Figure 4). To shed light on the nature of this interaction, we examined whether the mechanism underlying the coupling process in photoreceptors is related to phenomenologically similar ones found in other systems expressing Cx36 (Zoidl et al., 2002; Del Corsso et al., 2012; Veruki et al., 2008). Two possible drivers of the spontaneous coupling increase emerging from these studies were (i) dialysis of the cell by the pipette, and (ii) an increase in intracellular calcium, both occurring as a consequence of whole cell recordings (‘Discussion’). However, the fact that we observed this phenomenon in perforated patch clamp recordings, a technique known to cause minimal disturbance to the intracellular environment, makes it unlikely that in mouse cones, it is caused by pipette-cell dialysis (we could exclude an unintended rupture of the patch membrane, since the amphotericin-B present in our pipette solution would have perforated the cell membrane, depolarizing it close to 0 mV and shunting the photovoltages—effects that we indeed observed when going whole cell at the end of recordings to stain the neurons with Lucifer Yellow, LY). Nonetheless, we examined the remote possibility that amphotericin-B (nominally 400 µM) formed Ca²⁺-permeable pores (Romero et al., 2009) and that the diffusion of calcium ions from the pipette into the cone led to an opening of GJs (Ca²⁺ traces are normally present in the high purity water used for intracellular solutions). In four of five cones recorded with 1 mM EGTA in the pipette (EGTA zero Ca²⁺perforated patch solution), we observed the same progressive increase in coupling (Figure 4, blue lines).

These results effectively rule out an involvement of pipette Ca²⁺ but leave open the possibility that an increase in free [Ca²⁺]_i in the cone, promoted by the pipette, drives the coupling process. We thus performed whole cell recordings (i.e., without amphotericin-B) with the goal of dialyzing cones with a low Ca²⁺ buffered solution. Seals were made on cone pedicles (as confirmed by LY staining, Figure 11E) to ensure a rapid and effective ‘Ca²⁺ clamp’ on the cone side of the junctional contacts (located on short telodendria that protrude from the pedicles; Tsukamoto et al., 2001; O’Brien et al., 2012). This represented an important step, given what is known about the active compartmentalization of free [Ca²⁺]_i in cones (Wei et al., 2012). We recorded from three cones with an EGTA low Ca²⁺whole cell solution (50 nM free [Ca²⁺]_i; see ‘Materials and methods’), of which only one had weak coupling to rods upon breaking the patch. In two of the three cones, rod coupling increased rapidly in the first minutes of recording (Figure 11A), while in the third cone, coupling did not develop despite a stable recording for more than 1 hr (not shown). Further, we recorded from three cones with a BAPTA low Ca²⁺whole cell solution (50 nM free [Ca²⁺]_i) to rule out an involvement of fast and localized calcium transients. While all three cones were initially uncoupled, they rapidly increased their coupling to rods up to moderate levels (Figure 11B). Since in all these experiments free calcium in the pedicle was buffered at very low levels compared to those normally present in darkness (∼2 µM spatially averaged: Szikra and Krizaj, 2006), we conclude that high calcium (on the side of the cone) is not required for the expression of the progressive increase in coupling. High calcium, however, could favor coupling. To test this, we recorded from six cones with an EGTA high Ca²⁺whole cell solution (approx. 5 µM free [Ca²⁺]_i), all uncoupled or weakly coupled upon breaking the patch. Among five cones that we could monitor for sufficient time, one developed strong coupling over several minutes (Figure 11C), while the remaining four displayed weak coupling, which, however, did not appear to increase over time. Thus, no obvious favoring effect on coupling of high calcium emerged from this group of cones.

Figure 11D summarizes the time course of coupling for the cones shown in panels A–C, by plotting the dim G flash response amplitude. Also shown is the time course of the amplitude of bright_b G flash responses delivered after the pre-flash. Note that the intrinsic responses of these cones ran down rapidly (cf. lower records in panels A–C, arrows). This run down occurred in all low [Ca²⁺]_i and in two high [Ca²⁺]_i recordings. Interestingly, in the cones resistant to rundown, the distal portion of the neuron (cell body and outer segment) was only weakly stained by LY, as if a barrier to diffusion was present in the axon. Taken together, these experiments appear to exclude any major contribution in the coupling increase of Ca²⁺, at least on the cone side of the GJs.

Except for a small number of pure S-cones (i.e. blue cones), most mouse cones reportedly express both S and M opsins, with their expression ratio varying along the dorsoventral axis of the retina (Applebury et al., 2000; Haverkamp et al., 2005; Daniele et al., 2011; Wang et al., 2011). Moreover, any pure M-cones would respond to UV light since M-opsin has a prominent secondary absorption peak in the near UV region (the β-band; Govardovskii et al., 2000). It was thus not surprising that almost all of our recorded cones were intrinsically sensitive to both G and UV light (Figure 6 and Figure 7, cf. with Figure 1A,B). Our dissection and recording techniques did not allow us to identify their dorsoventral position in the retina, so that we could not directly examine possible correlations between their location and the level of maximal coupling. We could however conclude that both S-dominant cones (known to be mid and ventrally located) and M-dominant cones (dorsally located) are able to couple strongly to rods (Figure 6C). As for pure S-cones, our small recorded sample is in no way sufficient to conclude that they have a reduced propensity to couple to rods, since we found uncoupled cones also within the S/M group. Interestingly, there is some evidence in macaque that blue cones form fewer junctional contacts with rods compared to other cone types (O’Brien et al., 2012).

It is important to note that the G/UV ratios obtained in our sample of cones after rod-saturating pre-flashes (Figure 6C) do not provide a completely unbiased representation of the relative sensitivity to G over UV light within the native cone population. First, while we recorded from both central and peripheral retina, it is unlikely that we obtained a uniform sampling at all eccentricities; this, combined with the dorsoventral gradients in S- and M-opsin expression (see references above), will distort the observed distribution. Second, any differences in size or accessibility among cones would have affected their odds of being recorded (see Cangiano et al. (2012) for rod vs cone recording bias). Third, compared to the classical approach of estimating cone spectral preference by adjusting the strengths of flashes of different wavelengths to match their response amplitudes, our use of the ratio of the peak response amplitude to moderately bright G and UV flashes of equal strength (our G/UV ratios) shifts values somewhat toward unity as cone saturation is approached (we chose this approach because it was quick and avoided exposing rods to very bright light). Fourth, there could be a contribution of other cones coupled to the recorded cone, which would shift the measured G/UV ratio in one or the other direction depending on their relative expression of S- and M-opsins. Finally, we cannot exclude that in some experiments, coupled rods may have retained a small response to the bright flash following the saturating pre-flash, again shifting the G/UV ratio. These last two factors could explain the shifts in G/UV ratio observed in some cones during spontaneous coupling increase and superfusion with MFA (Figure 9—figure supplement 2), as well as in the ‘reversal’ experiment of Figure 8B.

GJs are often documented only anatomically, leaving open the question of their functional impact on neurons. Moreover, since they can be actively regulated, knowledge of their full coupling potential is essential. Our finding that the amplitude of rod signals in cones may attain a significant fraction of their source amplitude is a measure of the importance that rod–cone GJs can have in visual processing by the cone pathway. Is such a strong level of coupling compatible with what is known about rod–cone GJs?

Each cone in mouse forms junctional contacts with an average of 32 rods (Tsukamoto et al., 2001). To our knowledge, the only measurement in mammals of the transjunctional conductance was made in rod–cone pairs of the cone-dominated ground squirrel retina (legend of supplementary Figure 6 in Li et al., 2010). Assuming that their average value of 121 pS for coupled pairs (corresponding to ∼20 open homotypic Cx36 channels; Moreno et al., 2005) was also representative of the mouse, multiplying it by a rod–cone convergence of 32 gives a summed junctional conductance of ∼3900 pS. This would imply that a 4 mV hyperpolarization in rods relative to a maximally coupled cone would be sufficient to draw from it an overall junctional current of ∼16 pA—a very large value as it is comparable to estimates of the circulating current in mouse cones in darkness (Nikonov et al., 2006). Therefore, our data are compatible with the limited available evidence. Moreover, our results bear direct relevance to other mammals, including primates: in the area centralis of the cat retina, Smith et al. (1986) estimated a convergence of ∼48 rods on each cone, while in the peripheral retina of macaque, ∼25 rods converge on each cone (O’Brien et al., 2012), a remarkably similar value to that in the mouse.

We did not observe cone signals entering rods through GJs, since rods did not respond appreciably to bright cone-stimulating flashes delivered during a rod-saturating background, or following rod-saturating pre-flashes. Two possible explanations for this are: (i) rod–cone divergence in mouse is small (∼1; Tsukamoto et al., 2001); (ii) coupling might not be promoted when patching on rods.

A previous study in macaque reported a twofold speed-up of rod signals as they flow into cones through GJs (Hornstein et al., 2005). We confirmed this in mouse, observing however a much less dramatic effect: in patched cones, the time to peak (TTP) of dim flash responses was 197 ms (SD 35, n = 24), significantly shorter (p<0.0001) than the 258 ms (SD 25, n = 23) of rods recorded in loose seal (rod and cone data were from both the kinetics protocol and the spectral protocol at 24°C) (note that the TTP of dim flash responses in patched rods was 288 ms (SD 51, n = 24), moderately but significantly longer (p=0.02) than that of rods recorded in loose seal; since rods recorded in loose seal did not display a kinetics rundown, their TTP values were assumed to be representative of the unperturbed state of rods in our preparation). The considerable discrepancy between the magnitude of the speedup in macaque and mouse could be explained if macaque rods underwent a rundown of kinetics during recordings (as it occurs in mouse), without this being noticed.

As mentioned in the ‘Results’, a progressive increase in junctional coupling as a specific consequence of patch recording was observed in other studies targeting Cx36-expressing neurons, including heterologous expression systems (Zoidl et al., 2002; Del Corsso et al., 2012) and rat retinal amacrine AII cells (Veruki et al., 2008). Veruki et al., recording in whole cell mode, found that high resistance (i.e., fine-tipped) electrodes prevent the coupling increase, suggesting a dialysis process. Del Corsso et al. (2012) surmised that an increase in intracellular Ca²⁺ triggers the coupling increase during whole cell recordings in neuroblastoma cells in vitro. Moreover, they reported that the coupling increase did not occur when recording in perforated patch clamp mode with amphotericin-B, again implicating pipette-cell dialysis.

However, here we reach a different conclusion from that of the studies cited above. First, in cones, this phenomenon is not dependent on diffusion of Ca²⁺ between the cell and the pipette, as the spontaneous increase in coupling was expressed in perforated patch recordings, even in combination with EGTA in the pipette solution. This, therefore, excludes the possibility of a Ca²⁺ entry through hypothetical Ca²⁺-permeable pores formed by amphotericin-B in the patch membrane (Romero et al., 2009; see also ‘Results’). Second, our recordings in whole cell patch clamp with buffered pipette Ca²⁺ levels, made in close proximity to the GJs, strongly suggest that changes in free [Ca²⁺]_i in cones are not responsible for the expression of this phenomenon. Moreover, since the progressive increase in coupling was observed both when preserving the intracellular environment of the cone (perforated patch recordings) and when intentionally dialyzing it (whole cell recordings), a general tentative conclusion may be advanced: diffusible messenger molecules in cones are not necessarily involved in the process. Thus, any diffusible messengers (including Ca²⁺) could be located in the rods surrounding the recorded cone and act on the connexon hemichannels on the rod side.

Evidence for a primary trigger mechanism can be found in a previous study on macaque (Hornstein et al., 2005), which reported that Neurobiotin injected in cones at the end of perforated patch recordings diffused preferentially to rods located under the recording pipette, leading them to suggest that ‘the electrode might alter the coupling efficiency of the rod–cone junctions by mechanical disturbance’. This observation implies that their cones also expressed a progressive increase in coupling, a process whose electrophysiological counterpart must have gone unnoticed. In this scenario, and taking into account the arguments given above, coupling would be promoted by stretch/deformation of the rods adjacent to the recorded cone. This could explain the large variability in the rate of increase and maximum level of coupling that we observed in our recordings (Figure 4), since the angle of entry and depth of the pipette in the tissue varies from experiment to experiment, likely changing its mechanical interaction with neighboring rods.

With regard to the downstream pathways, Del Corsso et al. (2012) found in their heterologous expression system a Ca²⁺-mediated enhancement in the activity of calmodulin-dependent protein kinase II (CaMKII), which would lead to phosphorylation of Cx36 and increased channel conductance. Interestingly, CaMKII was recently found to participate in the physiological regulation of Cx36 in AII amacrines (Kothmann et al., 2012), raising the possibility that this kinase may play a role also in photoreceptors, on the rod side of rod–cone GJs (see above).

At the start of the recordings, we found most cones modestly coupled to rods relative to their full coupling potential, although we observed strong initial coupling in a limited sample (Figure 4). A key question that needs to be addressed is when the retina exploits the strong coupling potential detected in this study. Candidate regulators of rod–cone GJs in mammals are (references in the ‘Introduction’): (i) the circadian synthesis and release of the endogenous retinal neuromodulators, melatonin, dopamine, and adenosine, whereby coupling would be stronger at night when melatonin levels are high; most mouse strains, including the C57BL/6 used here, have deficits in melatonin synthesis (Roseboom et al., 1998) and are thus inadequate to investigate circadian processes; it is therefore possible that the low level of coupling that we estimated for our unperturbed cones may be related to this deficit and that full blown coupling, of the strength emerging during our recordings, is recruited in melatonin-competent mouse strains at night; (ii) acute light exposure, acting on the same neuromodulators and possibly also locally in the photoreceptors, which would inhibit coupling.

Unfortunately, unresolved discrepancies persist in the literature, a major one being the lack of effect of light and dopamine on rod–cone coupling studied with patch clamp in macaque (Schneeweis and Schnapf, 1999). Similar discrepancies (Hartveit and Veruki, 2012) exist for the GJs between AII amacrines, also containing Cx36 and thought of being modulated in a circadian and light-dependent fashion. While a common explanation for these conflicting pieces of evidence may eventually be found, our demonstration that rod–cone coupling can be upregulated so as to have a major impact on cones provides an important framework in which to place circadian/light-dependent modulation of these GJs in mouse. In fact, (i) let’s suppose that the level attained in our experiments by the fully developed coupling process had been quite smaller than what we actually observed; then this would have raised serious doubts about the physiological importance of circadian/light-dependent modulation of rod–cone coupling for mouse visual processing; (ii) given however our demonstrated high level of coupling, should a future direct examination of the role of circadian rhythmicity find that only a small fraction of this coupling potential is utilized, one might then well suspect that yet unidentified physiological factors play a greater role in recruiting rod–cone GJs, or, alternatively, that these may have an important part in response to stress and injury. One should recollect that GJs play a dual role, for example, conduits of electrical signals and of intracellular molecules. Cx36, in particular, is critically involved in neuronal responses to injury and disease (Belousov and Fontes, 2013). Currently, only a few studies have investigated this in the retina, with both positive (Striedinger et al., 2005; Paschon et al., 2012) and negative results (Kranz et al., 2013) depending on the specific models tested.

The existence of a large degree of GJ-mediated anatomical convergence from rods to cones is not peculiar of the mouse, since it has been found in other rod-dominated mammalian retinas including the peripheral retina of macaque. It follows that our findings on the functional impact of rod input in mouse cones may have broad relevance, supporting the possibility that rod–cone coupling plays a significant role in vision and/or in biochemical signaling between photoreceptors.

Full field stimuli of unpolarized light were delivered by a green LED (peak emission at 520 nm; OD520; Optodiode Corp., Newbury Park CA) or an ultraviolet LED (peak emission at 365 nm; APG2C1-365-S; Roithner LaserTechnik, Vienna Austria) mounted beside the objective turret. LEDs were driven by current sources commanded through the analog outputs of a Digidata 1320A (Axon Instruments, Foster City, CA). The power density reaching the recording chamber vs LED drive was measured separately with a calibrated low power detector (1815-C/818-UV; Newport, Irvine, CA) positioned at the recording chamber. Flash duration was in the range of 1–10 ms. Unless specified in the protocol, consecutive bright flashes were delivered at intervals of 12 s or more between each other. The photon flux density reaching the photoreceptors was derived from the measured power density and was likely to be overestimated to varying degrees across recorded cells due to reflection at the air–water interface and absorption by the surrounding tissue, including retinal pigment epithelium. Outer segments could be oriented at a variety of angles with respect to the direction of incident light.

Rod input in cones was dissected with a kinetics protocol and a spectral protocol. These consisted of a mix of dim and bright flashes at 520 nm (green, G) or 365 nm (ultraviolet, UV). For reference, in mouse, the absorption peaks of rhodopsin, M-opsin, and S-opsin are at 498 nm, 508 nm, and 359 nm, respectively (Sun et al., 1997; Yokoyama et al., 1998). Dim flashes (16.6 photons·µm⁻²) were sufficient to elicit a significant response in rods (Figure 1A,B, curves on the left; our data, see legend) but expected to be too weak to directly stimulate cones (Figure 1A,B, curves on the right; predictions from the literature, see next paragraph). Bright flashes (bright_a = 1570, bright_b = 3140 ph·µm⁻²) were sufficiently strong to saturate rods for 1–2 s (Figure 1A, lower graph, our data) and expected to evoke a measurable response in cones (UV flashes in all cones, while G flashes in most cones; Figure 1A,B, curves on the right; predictions from the literature, see below).

Most mouse cones express both S and M opsins, with their relative proportions varying along the vertical axis of the retina (M dominates dorsally, while S dominates in the mid and ventral retina) (Applebury et al., 2000; Haverkamp et al., 2005; Nikonov et al., 2006; Daniele et al., 2011; Wang et al., 2011). In Figure 1A,B, we show the predicted average flash response profiles of cones at different retinal latitudes (1–4: mixed S/M cones) and that of the small but distinct population of pure S cones sparsely distributed throughout the retina. To generate these profiles, we: (1) used hyperbolic saturation functions (Nikonov et al., 2006); (2) assumed the flash sensitivity of the normalized response of S- and M-cones in G_tα^−/− mice, recorded with suction pipette and without a rod-saturating background (Table 1 in Nikonov et al., 2006: 0.042% photons⁻¹·µm², equivalent to a half-saturating flash strength of 2381 ph·µm⁻²), to be valid for S- and M-cones in wild type mice; moreover, we assumed that all cones express the same total amount of opsin; (3) the typical fraction of M-opsin expressed by cones as a function of their retinal latitude (i.e., the average among cones within each horizontal slice of retina) was taken as the average of the values predicted by two recently published quantitative models (Daniele et al., 2011; Wang et al., 2011) obtained from mice in which rod responses were absent or suppressed; since these models extend to slightly different retinal latitudes (±2.5 mm and ±2.0 mm, respectively), we normalized their ranges prior to averaging; the relevant parameters of the cones shown in Figure 1, given as %M-opsin/position, are 1: 64%/dorsalmost, 2: 24%/dorsal third, 3: 4.7%/ventral third, 4: 1.2%/ventralmost, S: 0%/ubiquitous; (4) the absorbance of M-opsin at 365 nm (peak of the β-band) was taken as 20% of maximum, while for S opsin, its absorbance at 520 nm was taken to be 4 log-units below maximum (based on Govardovskii et al., 2000).

The kinetics protocol (Figure 1C) was designed to detect rod input in cones by exploiting the high sensitivity of rods and their slow recovery after a saturating flash. A dim G test flash was delivered both before a bright_a G flash and at three increasing delays following it (1, 2.5, 4 s). Observing in cones: (i) a response to the first dim flash, (ii) a slow plateau after the bright flash, and (iii) a progressive recovery of the dim flash response, would be evidence for rod coupling. The spectral protocol (Figure 1D) was designed to determine the impact of rod coupling on the spectral preference of cones and the possible relationship between cone opsin expression and coupling strength. The cones’ apparent spectral preference was determined by delivering G and UV flashes of the same strength (equal number of ph·µm⁻²; either dim or bright_b flashes). The cones’ intrinsic preference was determined by delivering G and UV bright_b flashes after a G rod-saturating pre-flash.

Data are reported as mean and SD (standard deviation), SEM (standard error of the mean). Statistical significance was assessed with the Mann–Whitney–Wilcoxon test. In all figures, electrophysiological records were ‘box car’ filtered with a running window of 20 ms.