Chromophore switch from 11-cis-dehydroretinal (A2) to 11-cis-retinal (A1) decreases dark noise in salamander red rods


Corresponding author P. Ala-Laurila: Department of Physiology and Biophysics, Boston University School of Medicine, Boston, MA, USA. Email:


Dark noise, light-induced noise and responses to brief flashes of light were recorded in the membrane current of isolated rods from larval tiger salamander retina before and after bleaching most of the native visual pigment, which mainly has the 11-cis-3,4-dehydroretinal (A2) chromophore, and regenerating with the 11-cis-retinal (A1) chromophore in the same isolated rods. The purpose was to test the hypothesis that blue-shifting the pigment by switching from A2 to A1 will decrease the rate of spontaneous thermal activations and thus intrinsic light-like noise in the rod. Complete recordings were obtained in five cells (21°C). Based on the wavelength of maximum absorbance, λmax,A1= 502 nm and λmax,A2= 528 nm, the average A2 : A1 ratio determined from rod spectral sensitivities and absorbances was ∼0.74 : 0.26 in the native state and ∼0.09 : 0.91 in the final state. In the native (A2) state, the single-quantum response (SQR) had an amplitude of 0.41 ± 0.03 pA and an integration time of 3.16 ± 0.15 s (mean ±s.e.m.). The low-frequency branch of the dark noise power spectrum was consistent with discrete SQR-like events occurring at a rate of 0.238 ± 0.026 rod−1 s−1. The corresponding values in the final state were 0.57 ± 0.07 pA (SQR amplitude), 3.47 ± 0.26 s (SQR integration time), and 0.030 ± 0.006 rod−1 s−1 (rate of dark events). Thus the rate of dark events per rod and the fraction of A2 pigment both changed by ca 8-fold between the native and final states, indicating that the dark events originated mainly in A2 molecules even in the final state. By extrapolating the linear relation between event rates and A2 fraction to 0% A2 (100% A1) and 100% A2 (0% A1), we estimated that the A1 pigment is at least 36 times more stable than the A2 pigment. The noise component attributed to discrete dark events accounted for 73% of the total dark current variance in the native (A2) state and 46% in the final state. The power spectrum of the remaining ‘continuous’ noise component did not differ between the two states. The smaller and faster SQR in the native (A2) state is consistent with the idea that the rod behaves as if light-adapted by dark events that occur at a rate of nearly one per integration time. Both the decreased level of dark noise and the increased SQR amplitude must significantly improve the reliability of photon detection in dim light in the presence of the A1 chromophore compared to the native (A2) state in salamander rods.

Thermal activations of visual pigment molecules in retinal rods trigger discrete electrical ‘dark’ events that are identical to responses to single photons (Baylor et al. 1980, 1984). Such randomly occurring events constitute an irreducible background noise that sets an ultimate limit to visual sensitivity (Barlow, 1956; Aho et al. 1988, 1993). Between-species comparisons indicate that long-wavelength sensitivity of visual pigments (i.e. the capacity to be activated by low-energy photons) is, on average, associated with high rates of thermal activations (Ala-Laurila et al. 2004a,b). Spectral tuning of rod pigments for good performance in dim light must then be understood not only in terms of quantum catch, but also in terms of noise control. Indeed, rod pigments tend to be significantly blue-shifted (‘Purkinje shift’) from the position that would maximize quantum catch (Barlow, 1957; Lythgoe, 1988). Present evidence for this, however, is wholly based on correlation across different photoreceptors.

Our purpose was to study, in single isolated salamander rods, how dark noise changes when the visual pigment is spectrally shifted by exchanging the chromophore within these same rods. Vertebrate visual pigments use either 11-cis-retinal (A1) or 11-cis-3,4-dehydroretinal (A2). In all pigments in the Rh1 pigment family, the wavelength of maximum absorption is shorter when the opsin is bound to the A1 chromophore rather than the A2 form (Dartnall & Lythgoe, 1965). In order to exchange the chromophore, we bleached most of the native A2 visual pigment and then regenerated with the A1 chromophore. This allowed us to evaluate flash sensitivity and membrane current noise in darkness in a single cell with a single opsin in two different conditions. We find that the thermal stability of the 502 nm A1 pigment is over 36-fold greater than the thermal stability of the 528 nm A2 pigment. This suggests that the A1 pigment can support a higher absolute sensitivity of vision than the A2 pigment, since fewer photons will suffice to produce a criterion signal-to-noise ratio.


Animals and preparation

The experiments were done on isolated red rods of larval tiger salamander Ambystoma tigrinum (Charles D. Sullivan Company, Nashville, TN, USA). All procedures were performed according to protocols approved by the Animal Care and Use Committee of Boston University School of Medicine and in accordance with the standards set forth in the Guide for the Care and Use of Laboratory Animals (ISBN: 978-0-309-05377-8) and the Animal Welfare Act (

Salamanders were kept at ca 10°C in aquarium tanks on a 12 h–12 h light–dark cycle. Animals were dark-adapted overnight before each experiment. The animal to be used in an experiment was decapitated and double-pithed under dim red light. The eyes were enucleated and hemisected and the retinas were isolated in Ringer solution. All subsequent procedures were carried out under infrared illumination with the aid of infrared image converters (Find-R-Scope). Rods were isolated by teasing pieces of retina apart in ca 1 ml of Ringer solution. The sample was placed on the stage of an inverted microscope (Invertascope D; Carl Zeiss, Inc.) in a light-tight Faraday cage for electrical recordings. Cells were viewed via a high sensitivity CCD camera (LCL-902HS; Watec) connected to a TV monitor. The video signal was also fed to a computer equipped with a frame grabber card (WinTV-PVR-250, Hauppauge) to capture and store bright-field images of the cells. Cells were superfused by standard Ringer solution (pH 7.8) of the following composition: (mm): 110 NaCl, 2.5 KCl, 1.6 MgCl2, 1.0 CaCl2, 10 dextrose, 10 Hepes; plus 1.5 μm bovine serum albumin. The room temperature was maintained at 21 ± 1°C and monitored throughout the experiment.

Light stimulation

Unpolarized light stimuli were provided from a dual-beam optical stimulator equipped with a tungsten–halogen light source. Computer-controlled shutters (Vincent Associates) were used to deliver 20 ms flashes as well as continuous background illumination and bleaching exposures. The light intensity was controlled by a set of calibrated neutral density filters. The wavelength of light stimuli over the range 400–760 nm was set by interference filters placed in the light beam. Individual filters had a nominal half-bandwidth of 10 nm (Chroma Technology, Rockingham, VT, USA). The unattenuated intensity of a uniform circular spot (diameter, 900 μm) of the stimulus light (540 nm) focused on the plane of the specimen holder was adjusted at the beginning of each experiment to be 1.21 × 108 photons μm−2 s−1. This intensity was checked again at the end of each experiment. Light intensities were measured for different combinations of interference filters and neutral density filters used in the experiment by a calibrated radiometer (Model 350; UDT Instruments) placed on the microscope stage at the same level as the preparation.

Light intensities were expressed as rates or numbers of photoisomerizations per rod based on two calibration principles described below in the section on single-photon response analysis. One is based on the response ensemble variance-to-mean ratio obtained from responses to a sequence of very dim flashes, the other on estimations made from OS geometry, pigment density and photon flux. We denote numbers of isomerizations by Rh*. When referring to isomerizations per rod, Rh* is used without specification; when referring to isomerizations per molecule of visual pigment, this is specified (Rh* molecule−1).

Visual pigment bleaching and regeneration

After the stable dark-adapted ‘native’ state of a rod had been thoroughly characterized, the cell was bleached by a 30 s exposure to very bright light (520 nm, 1.1 × 107 photons μm−2 s−1).

The wavelength used for bleaching, 520 nm, was originally chosen just because it is quite close to the λmax of the native pigment and our bleaching protocols had previously been optimized for this wavelength. Beyond that, it has no special meaning, as opposed to the stimulus wavelength, 540 nm, which was chosen because it is close to the isosbestic point of salamander A1 and A2 pigments and thus stimulates both with the same probability (see further below). As there would have been no specific advantage in bleaching equal fractions of native A2 and A1 pigment, however, we found no compelling reason to switch to 540 nm for the bleaching light.

The light intensity used for bleaching was ca 10 times lower than the maximum available from our light source, because we noticed that the probability of successful regeneration was higher with the lower intensity and longer exposure time than when the same fractional bleach was produced with the maximum intensity and shorter exposure time. We can only speculate on the reasons for this difference. Very intense bleaching might generate a large amount of reactive all-trans-retinal, which may compromise the physiological condition of the cell. It is also possible that it generates a larger fraction of long-lasting photoproducts, e.g. metarhodopsin III, where the chromophore binding site is occupied and does not allow the regeneration of the visual pigment with 11-cis-retinal. The amount of metarhodopsin III may depend on the intensity of the light used for bleaching as well as its wavelength (Ritter et al. 2007). The light exposure was set to bleach a fraction F≈ 0.9 of the native pigment according to the relation: F= 1 − exp(−IPλt), where I is the bleaching light intensity (photons μm−2 s−1), Pλ is the pigment's photosensitivity at the wavelength of stimulus light (μm2), and t the duration of the exposure (s). For Pλ (520 nm), we used the value 6.2 × 10−9μm2 (Jones, 1995). More accurate estimates of the pigment composition before and after bleaching and regeneration were based on spectral sensitivity measurements as explained below.

In the microspectrophotometry, we tested regeneration after even stronger bleaches, exposing the cells for 20 min to a lower light intensity. The combination of relatively low intensity and long exposure time enabled practically complete regeneration with the A1 chromophore even after having bleached 99% of the native pigment (see Fig. 1B). However, such very long and strong bleaches were not tried in the electrophysiological experiments.

Figure 1.

Determination of pigment content in rods before and after bleaching and regeneration with A1
A, log values of relative spectral sensitivities determined electrophysiologically in one rod (cell 3 in Table 1). Red, native state; blue, final state after ca 90% bleach and regeneration with A1. The curves fitted to the data are sums of Govardovskii et al. (2000) templates for A1 (λmax= 502 nm) and A2 pigments (λmax= 528 nm). In the native state, a fair fit (red curve) is obtained for a sum of templates corresponding to molecular proportions 30.0% A1 and 70.0% A2; in the final state (blue curve) the proportions are 90.4% A1 and 9.6% A2. The peak value of the ‘final’ spectrum has been normalized to unity and all sensitivities are expressed relative to this. The crossing point of curves lies at ca 546 nm and the ratio of the peak values (final/native) is 1.46. B, absorbance spectra recorded by MSP. The ordinate gives optical densities (O.D.) at each wavelength. Red: native state, mean ±s.e.m. of 10 rods from the same retina as the cell in A. Blue: final state, means ±s.e.m. of a sample of 14 cells from a different retina after a complete pigment bleach and regeneration with A1. The red curve is a sum of Govardovskii et al. (2000) templates for A1 and A2 corresponding to molecular proportions 19% A1 and 81% A2. The blue curve is a pure A1 (λmax= 502 nm) template. The crossing point lies at ca 539 nm and the ratio of the peak O.D. values (A1/native) is 1.43. Error bars show s.e.m.

We used 11-cis-retinal (chromophore A1) to regenerate the visual pigment of the bleached rods. The retinoids were synthesized and purified in the Department of Ophthalmology at the Medical University of South Carolina (for review, see Crouch et al. 2002). The retinoid solution was prepared as described earlier (Estevez et al. 2006) by adding 3 μl ethanol into 50 μg 11-cis-retinal (24 900 m−1 cm−1 at 380 nm) and diluting this solution into ca 1 ml of Ringer solution. The final concentration of the retinoid solution was measured in a spectrophotometer (U-3010; Hitachi) to be in the range 40–50 μm in Ringer solution with ca 0.3% ethanol. When the cell had settled into a new bleach-adapted steady state ca 80 min after the bleach (cf. Fig. 2 below), the gravity-controlled perfusion was stopped and the retinoid solution was added to the specimen chamber. After 20 min of incubation, perfusion was restarted and the retinoid was washed out of the chamber.

Figure 2.

The time course of changes in the light-sensitive current and fractional sensitivity of one rod throughout a complete experiment
The vertical dashed lines mark the time points of: (1) the bleaching exposure (t= 0 min, middle point of 30 s light exposure); (2) introduction of ca 40 μm 11-cis-retinal (t= 89 min). Bath perfusion was halted for 20 min after the addition of 11-cis-retinal to incubate cells in the Ringer solution containing 11-cis-retinal. A, response families recorded at the times indicated by arrows in B. Flashes were 540 nm, 20 ms square-wave pulses. Each trace is the average of 1–30 responses to the same nominal flash intensity; intensities (photons μm−2 flash−1) increased in 0.5 log unit steps starting at the following values (response families from left to right, n= number of flashes averaged): 0.242 (n= 5); 271 (n= 1; the intensity step from the first to the second response was 1 log unit in this case to avoid crowding); 23.0 (n= 5); 0.242 (n= 20). The light-sensitive current and fractional sensitivity values plotted in B and C were determined by fitting Michaelis functions to flash intensity–response amplitude data of this type recorded throughout the experiment (see Methods). B, time course of changes in the light-sensitive current. C, time course of changes in log fractional sensitivity. Fractional sensitivity values have been normalized relative to the peak value. (Cell 1 in Table 1.)

Suction electrode recordings in single rods

The membrane current of isolated rods was recorded by drawing the inner segment into a tight-fitting glass micropipette filled with Ringer solution, connected to a patch-clamp amplifier (EPC-7, List Associates) (Baylor et al. 1979a; Cornwall et al. 2000). Pipettes were fire-polished to an internal diameter of ca 12 μm and silanized. The resistance was typically ca 1 MΩ (open pipette) and ca 3 MΩ with a cell drawn into the pipette. Signals were low-pass filtered at 20 Hz with an 8-pole active Bessel filter, digitized at 100 Hz (Digidata 1322A; Axon Instruments) and stored on a computer hard-drive. pCLAMP software components (version 9.2, Axon Instruments) were used for data acquisition and analysis. The recordings were conducted in complete darkness save for the epochs of light flashes, dim backgrounds and saturating light that were part of the experimental protocol. The position of the cell within the micropipette was monitored between recording periods throughout the experiment by capturing bright-field images of the cell under infrared illumination.

Fractional sensitivity and light-sensitive current

The photosensitivity and the maximum response amplitude of the rod were probed at regular intervals throughout the experiments by measuring the response–intensity relation of the rod. The stimulus wavelength, 540 nm, was chosen since it is close to the isosbestic point (537 nm), where the mean number of photoisomerizations produced is independent of the A1/A2 pigment ratio (Cornwall et al. 2000). We measured photoresponses to a series of 20 ms flashes at five to six intensities increasing in ca 0.5 log unit steps. At each intensity, 1–30 responses were recorded and averaged. The amplitudes of the averaged responses were plotted against flash intensity and fitted by the Michaelis equation:

display math(1)

where i is the response amplitude (pA), imax is the maximum dark current (pA), measured as the amplitude of the response to a saturating flash (closing all light-sensitive channels), I is flash intensity (photons μm−2) and Is is half-saturating flash intensity. The half-saturating intensity is inversely proportional to fractional sensitivity, i.e. the fraction of the light-sensitive current that is turned off by one photoisomerization. Changes in imax and Is were followed throughout each experiment (for an example, see Fig. 2).

Determination of pigment composition: measurements of spectral sensitivity and spectral absorbance

Spectral sensitivity measurements The spectral sensitivity of each cell was determined both in the native and the final state, i.e. before bleaching and after regeneration of the visual pigment with 11-cis-retinal. The half-saturating intensity at seven test wavelengths in the range 440–740 nm was determined relative to that at the reference wavelength 540 nm (λref). A series of 5–15 dim-flash responses at each test wavelength was interleaved between complete intensity–response families recorded at 540 nm. The flash intensity at each test wavelength was selected to elicit ca 10–20% of the saturated response amplitude. Equation (1) was fitted to the intensity–amplitude data recorded at each test wavelength (λ) assuming that imax was the same as in the immediately preceding full 540 nm intensity–response family, thus yielding the half-saturating intensity parameter Is(λ) for the test wavelength. Observing that fractional sensitivity is inversely proportional to Is(λ), the relative fractional sensitivity, S(λ), at each test wavelength is

display math(2)

Microspectrophotometry (MSP) A single-beam microspectrophotometer (MacNichol, 1978) was used to record visual pigment absorbance spectra from rod outer segments under dark-adapted conditions and following bleaching and regeneration of visual pigment. Spectra for the native state were recorded in rods taken from the same cell suspension as used for electrophysiological recordings. In each experiment, 10 intact rods or isolated outer segments containing the native visual pigment were studied. Another set of absorbance spectra were recorded after total bleaching and regeneration of the visual pigment with 11-cis-retinal. The rods in these latter recordings, however, were from retinas different to those used for the electrophysiology, as it was not essential that the initial A2/A1 ratio was exactly the same as in any single electrophysiological experiment. For convenience we therefore chose to do the MSP regeneration measurements separately from electrophysiology experiments.

A small volume of cell suspension in Ringer solution (50 μl) was placed between two cover slips that were sealed together with double-sided tape. The optical density of rod outer segments was recorded over the wavelength range 380–740 nm. The rectangular measuring beam was aligned parallel to the long axis of the outer segment (OS), covering ca 1/4 of its width. The axis of polarization of the measuring light was orthogonal to the long axis of the OS. Before each scan, a baseline recording was measured in a cell-free space adjacent to the OS. Typically, 10 scans were averaged in each case. The mean optical density (O.D.) between 400 and 740 nm was calculated by the data acquisition software as O.D. = log(I0/It), where I0 is the incident light intensity measured in the cell-free space and It the transmitted light through the OS. The spectra were baseline-corrected by subtracting from each data point a value given by a straight line least-squares-fitted to the data points over the interval 650–700 nm, where the absorbance of the visual pigment is negligible (for further details, see Govardovskii et al. 2000).

Estimation of visual pigment composition Spectral sensitivity and absorbance data were fitted with a sum of A1 and A2 visual pigment templates of Govardovskii et al. (2000). These templates are uniquely defined by a single variable (λmax), whereby λmax,A1 and λmax,A2 are coupled according to the rule: λmax,A2= 1.575λmax,A1− 263 (nm) (Hárosi, 1994). Estimates of the molecular ratios of A1 and A2 pigments were deduced from the weights of the two templates in the sum that provided the best fit to normalized data as judged by a least-squares criterion in the wavelength range 440–700 nm. The fitting was performed on a linear sensitivity scale in the case of absorbance spectra recorded by MSP, where the range of reliable absorbance values covers less than 2 log units of absorbance. In the case of relative spectral sensitivity determined electrophysiologically the fitting was done on a log sensitivity scale (see Fig. 1). The reason is that the range of reliable spectral sensitivity values spans as much as ca 6 log units. Furthermore, the data points in the long-wavelength region having the lowest absolute sensitivity values depend strongly on differences in the A1/A2 pigment ratio. Thus, these points will be most efficiently taken into account, while fitting the spectral sensitivity data on a logarithmic scale. In the fitting routine, we assumed λmax,A1= 502 nm (Makino et al. 1999), which provided an excellent fit to absorbance spectra recorded after a complete bleach and regeneration with A1 chromophore (see Fig. 1B). Furthermore, we assumed λmax,A2= 528 nm as predicted by the Hárosi (1994) relation for λmax,A1= 502 nm (see above). The ratio of peak photosensitivities of A1 and A2 pigments was taken to be 1.42 (Dartnall, 1968). Thus, our estimates of A1 and A2 percentages (see Table 1) refer directly to the relative numbers of A1 and A2 visual pigment molecules.

Table 1.  Summary of the main results from the five rods on which complete experiments were achieved
CellStateA1 (%)A2 (%) i max (pA)SQR (pA) t i (s)σ2dark (pA2) σ2D (pA2) σ2c (pA2)Event freq. (Rh* s−1)Rod OS vol. (μm3)Molecular rate constant (10−11 Rh* molecule−1 s−1)
  1. Pigment composition (percentage A1 and A2), maximum current (imax), the single quantum response (SQR) amplitude and integration time (ti), total variance of dark noise 2dark), its discrete 2D) and continuous components 2c) as well as the estimates for the spontaneous event rates per cell and per molecule of visual pigment are given. The primary estimates for pigment composition were obtained from electrophysiological measurements of spectral sensitivity in the same cell (see Methods). For the native state, these were supplemented by MSP measurements on 10 cells obtained from the same cell suspension as that used in the experiment. The estimate of pigment composition obtained from these absorbance measurements are given in parentheses.

Final94.35.730.10.533.640.0080.0050.0040.0154 0.433
Final87.412.632.20.372.960.0330.0070.0250.0519 1.20
Final90.49.637.80.783.660.0630.0370.0230.0314 0.782
Final92.37.748.80.674.260.0500.0210.0260.0183 0.472
Final88.911.140.80.512.810.0190.0100.0080.0322 0.641
Mean ±s.e.m.
 Native26.3 ±73.7 ±37.4 ±0.41 ±3.16 ±0.075 ±0.055 ±0.020 ±0.238 ±2110 ±5.694 ±
(19.9 ± 2.4)(80.1 ± 2.4) 
Final90.7 ±9.3 ±37.9 ±0.57 ±3.47 ±0.035 ±0.016 ±0.017 ±0.030 ± 0.706 ± 0.138

Single-photon response analysis

The membrane current response to isomerization of a single rhodopsin molecule (the single-quantum response, abbreviated SQR) was estimated by measuring the mean and the time-dependent ensemble variance of a sequence of dim-flash responses (see, e.g. Rieke, 2000). This method assumes that the variability in response amplitude can be accounted for primarily by variability in the number of photoisomerizations Rh* per flash. For a Poisson distribution in Rh* per flash, the square of the mean response inline image will be proportional to the ensemble variance of the response σ2r(t) according to inline image, where the scalar inline image is the expected number of Rh* per flash and t is time. We estimated inline image by scaling σ2r(t) to match inline image. The SQR size and waveform, inline image, can be estimated as: inline image. This method does not take into account variation in the amplitudes of individual SQRs. In rods, the variance due to this is ca 20% of the SQR amplitude (Baylor et al. 1979b, 1984), which limits the accuracy of individual estimates obtained by this method.

Sequences of 30–50 dim-flash responses were recorded regularly throughout the experiment. We used two different flash intensities, nominally producing 2.1 and 4.3 Rh* on average. Each response was baseline-corrected to remove DC drift by subtracting from each data point of the response the ensemble average of the membrane current recorded over the initial 200 ms time window preceding response onset.

An independent estimate for SQR amplitude was based on photometric calibration of the stimulus light and measurement of the cell's collection area (Ac). The collection area was estimated by the following approximation (Baylor et al. 1979b; Cornwall et al. 2000):

display math(3)

where d is OS diameter, l is OS length, ɛ is the specific optical density at the stimulus wavelength 540 nm (ɛ= 0.0118 μm−1, Cornwall et al. 2000), γ is the quantum efficiency for bleaching (γ= 0.67; Dartnall, 1972), and f is an ‘orientation factor’= 0.5 for unpolarized light. Cell dimensions d and l were estimated from bright-field images of the cell. If the incident flash intensity based on photometric calibration is I, then inline image=AcI and inline image. The SQR amplitude estimated in this way for each individual dim flash sequence was generally within 20% of the estimates obtained from ensemble mean and variance.

We fitted dim-flash responses with a modified version of a function used by Schneeweis & Schnapf (2000) for macaque rods:

display math(4)

where w(t) = (1 + (τ3/t)m)−1, r1=Atnexp(−t1), r2=B(exp(−t2) − exp(−t3)) and A and B are constants. This function has no mechanistic meaning and was used as a purely empirical equation for smoothing and zero-padding responses. Zero-padding implies that extra zero values are added after the end of the response (comprising in our case 10.23 s, 1024 points) to extend the full stretch used for analysis (in our case to 41 s, 4096 points). The purpose is to extend the spectral frequency range (see e.g. Vu et al. 1997) when power spectral densities are calculated as explained below. These model fits were also used to estimate the integration time of dim-flash responses (Baylor & Hodgkin, 1973):

display math(5)

where ti is the integration time, r(t) is the response (in our case eqn (4) fitted to the averaged SQR) and rpeak is the response amplitude (at peak).

Analysis of dark noise power spectra and the estimation of dark event rates per cell

The rate of ‘dark’ isomerizations per rod was estimated on the basis of power spectral densities of membrane current measured in darkness and in bright light. The analysis of power spectral densities has been widely used in earlier studies on dark event rates in both rod and cone photoreceptors (Lamb & Simon, 1977; Schwartz, 1977; Baylor et al. 1980; Bodoia & Detwiler, 1985; Leibrock et al. 1994; Rieke & Baylor, 1996, 2000; Schneeweis & Schnapf, 2000; Sampath & Baylor, 2002; Kefalov et al. 2003; Holcman & Korenbrot, 2005). One-sided power spectral densities of membrane current traces recorded in darkness, in bright light and in dim background light were calculated using the algorithms of the Clampfit 9.2 program (Axon Instruments), which rely on a fast Fourier transform of the input signals. Power spectra were determined from current sweeps each of which was 81.92 s in length. The total durations of recordings in each experiment in the native and final states were typically as follows: ca 30 min (complete darkness), ca 7 min (under each dim background light) and ca 20 min in bright light. Linear drift was first removed by subtraction of the best-fitting straight line from the data. Power spectral densities of current traces were then calculated over 40.96 s current segments with 50% overlap using cosine windowing. Power spectra of quantal events were calculated similarly using the zero-padded SQR (40.96 s) estimates obtained by fitting eqn (4) to the original SQRs (10.24 s) (see above). We estimated the instrumental Johnson noise level using the Nyquist equation: S(f) = 4 kT/R, where k is the Boltzmann constant, T is the absolute temperature and R is the resistance of the pipette with the cell in place. In all of the studied cells the estimated instrumental noise level nicely matched the plateau level of the noise power spectral densities between ca 3 and 20 Hz (as an example, see the dashed line in Fig. 4B).

Figure 4.

Noise recordings and power spectra
A, representative noise recordings. The two top traces show noise under a dim steady background light (2.4 Rh* s−1) in the final (blue) and native (red) state; the two middle traces show dark noise in the final (blue) and native (red) state; the bottom trace shows instrumental noise (black). All traces have been low pass filtered at 1 Hz. Same cell as in Fig. 3. B, total noise power spectra based on many recording sequences similar to those shown in A. Spectra from top to bottom (according to the zero-frequency asymptote): blue squares, noise under the dim background in the final state; red squares, noise under the dim background in the native state; red circles, dark noise in the native state; blue circles, dark noise in the final state; black circles, instrumental noise. Power spectra were smoothed at higher frequencies by averaging over 5 neighbouring points in the frequency range 2–10 Hz and over 10 points in the frequency range 10–40 Hz. The dashed line shows the Johnson noise level (0.0043 pA2 Hz−1) calculated from the resistance with the cell in the pipette (3.8 MΩ).

Noise not associated with the light-sensitive current of the rod was recorded after the strong bleaching exposure (see Fig. 2) and in bright (saturating) light at the end of the experiment. The noise due to discrete and continuous membrane current fluctuations of the rod was then isolated as the difference between the power spectral densities measured in complete darkness minus the power spectral densities of noise measured in saturating light conditions. In order to separate continuous from discrete noise, the plateau of the difference spectra in the frequency range 0.3–0.5 Hz was fitted by a straight horizontal line representing the zero-frequency asymptote of the continuous noise component. The difference spectrum in the frequency range 0.018–0.21 Hz was then fitted with the sum of the continuous noise asymptote plus the SQR power spectrum multiplied by a scaling factor that gave the optimal fit (see Fig. 5). This factor expresses how many dark events of the given shape and size must occur within the duration of the recording on which the power spectrum is based to produce the dark current variance observed. Thus the event frequency per cell was obtained by dividing the scaling factor by the record duration (40.96 s) used for construction of each power spectrum. The method is identical to that of Kefalov et al. (2003), with the exception that we corrected for the contribution of the continuous noise component as mentioned above. The advantage of the analysis described above is that the goodness of the fit can be judged immediately from the match of the recorded low-frequency noise spectral densities and the theoretical spectrum (the sum of the scaled SQR power spectrum and the continuous noise plateau).

Figure 5.

Difference spectra: dark noise – instrumental noise
Dark noise difference spectra calculated from Fig. 4. The instrumental noise has been subtracted from the total dark noise: red circles, native state; blue circles, final state. The low-frequency noise component has been fitted with the sum of the power spectrum of the continuous noise plateau level (averaged over 0.3–0.5 Hz) and the scaled power spectrum of the respective SQR, calculated from the responses shown in Fig. 3C (see Methods). These fitted curves are shown by dashed lines: red, native state; blue, final state. The difference in this component representing discrete noise indicates a ca 10-fold difference in dark event rate between the native and final states. (Note that the corresponding noise power in the final state is relatively increased by the larger SQR.) The continuous noise component has been fitted with a straight line in the plateau region. There is no significant difference in this component between the native and the final state.

Estimation of dark event rates per molecule of visual pigment

Molecular rate constants (k) for spontaneous visual pigment activations were calculated as:

display math(6)

where f is the mean frequency of dark events per cell estimated from the low-frequency component of the power difference spectrum and P is the total number of visual pigment molecules in the part of the outer segment ‘seen’ by the electrophysiological recording, estimated as P=cV. Here, c is the visual pigment concentration in the OS (3.27 mm in red rods of the larval salamander: Hárosi, 1975) and V the effective volume of the part of the OS outside the constriction of the suction pipette (see Table 1). We estimated V by measuring the OS diameter (d) and length outside the pipette (l) from bright-field images of the cell (Vd2l/4).

Error estimates throughout are given as standard error of means (s.e.m.s) unless otherwise stated.

Estimating the total dark current variance and decomposing it into continuous and discrete noise components

The total dark noise variance arising from the phototransduction cascade was estimated by fitting the difference spectra (dark – bright light) by a linear combination of two functions, one representing continuous noise and the other discrete noise, and then integrating the fitted functions in the domain 0–20 Hz.

The discrete noise component was fitted by the scaled power spectrum of the SQR used for the analysis of dark event rates and obtained as explained above (eqn (4)). The continuous noise component in most cells was fitted with a function of the form:

display math(7)

where S(0) is the zero-frequency asymptote of the continuous noise component, f is the frequency and n and α are model parameters (for further details, see Lamb & Simon, 1977; Baylor et al. 1979b). Equation (7) gives the power spectrum of a ‘Poisson’-type impulse response, but it is used here strictly as a phenomenological description of the continuous noise spectrum (with n= 5, cf. Vu et al. 1997). However, in a few cells with even steeper high-frequency roll-off than given by eqn (7), a power function was fitted instead to the high-frequency edge (see Table 1 for results).


Visual pigment content in the native state and after bleaching and regeneration

Pigment densities and the relative proportions of chromophores A1 and A2 were assessed by measuring the spectral properties of cells in the beginning (native state) and at the end (final state) of each experiment. The spectral sensitivity of the investigated rod was measured electrophysiologically in both states. For the native state, this was supplemented by MSP recordings of spectral absorbance in samples of rods from the same retina. Further, the general success of regeneration with the A1 chromophore after bleaching was assessed by MSP recording in samples of rods from other retinas. The pigment composition (A1 : A2 ratio) was determined by fitting the recorded spectra with sums of Govardovskii et al. (2000) templates for A1 and A2 visual pigments using a least-squares criterion (see Methods for further details).

Figure 1 shows spectra from one experiment. Red symbols and curves refer to the native state, blue symbols and curves to the final state after pigment bleaching and regeneration. In Fig. 1A, spectral sensitivities of the rod (cell 3 in Table 1) are plotted on a logarithmic ordinate relative to the peak of the final-state (blue) spectrum, which has been normalized to unity (log relative sensitivity = 0). The fitted curves correspond to A1 : A2 = 30.0% : 70.0% in the native state and 90.4% : 9.6% in the final state. Figure 1B compares two MSP spectra, one representing the native state (averaged from 10 rods from the same retina as that in Fig. 1A), the other representing the final state after exposure to a light that was calculated to have bleached > 99% of the pigment and pigment regeneration with the A1 chromophore (averaged from 14 rods from a different retina). The fit to the native spectrum indicates an average A2 fraction of 81% in this rod sample, slightly higher than derived from spectral sensitivities for the single cell in Fig. 1A. The pigment spectrum following regeneration with the A1 chromophore after the nearly complete bleach is well-fitted by a pure A1 (502 nm) template (blue curve) with peak optical density 0.151, indicating satisfactory regeneration. The ratio of the peak absorbances of the two spectra is 1.426, consistent with strong A2 dominance in the native spectrum (cf. Dartnall, 1972).

In both panels A and B of Fig. 1, the native and final spectra cross near 540 nm, previously found to be the isosbestic point for A2 and A1 in salamander rod pigment (Cornwall et al. 2000). Light of this wavelength (that of the isosbestic point) will produce the same number of photoisomerizations in a rod regardless of the chromophore proportions, and it was therefore a convenient choice for stimulus wavelength in our experiments.

The mean values for the A1 : A2 ratio as estimated from spectral sensitivities in the five rods where complete experiments were achieved were 26% : 74% in the native state and 91% : 9% in the final state (see Table 1). The mean native-state estimate based on MSP spectra of rod samples from the same retinas was 20% : 80%. The presence of ca 10% A2 in the final state is consistent with the nominal 90% bleach (see Methods).

Time course of changes in light-sensitive current and fractional sensitivity before and following bleaching and regeneration

Figure 2 illustrates the time course of changes in light-sensitive current (saturating response amplitude) and fractional sensitivity that occurred before and following exhaustive bleaching of the native visual pigment and then during regeneration of visual pigment with the A1 chromophore.

Recordings such as are illustrated were successfully completed on five rods where the full experimental protocol lasted 10–14 h. These data were used for the further analysis. The characteristics of these cells, which had been selected for a high A2 content in the native state, are summarized in Table 1. The data illustrated are from cell 1 in Table 1. Figure 2A shows four examples of response families, recorded at the time points indicated by arrows in (Fig. 2B). The light-sensitive current and fractional sensitivity, plotted in panels B and C, were determined by fitting Michaelis functions (eqn (1) in Methods) to flash intensity–response amplitude data extracted from response families such as those shown in Fig. 2A. Note that the maximum light-sensitive current (imax) and the half-saturating flash intensity (IS) (proportional to fractional sensitivity) are independent parameters in eqn (1), and although both change strongly after a bleach and during regeneration, the changes can be followed independently. Thus, panels B and C in Fig. 2 trace two independent processes (i.e. in fractional sensitivity, changes in imax have been factored out).

In the native dark-adapted state (from −2.5 h to time = 0 in Fig. 2B and C), the light-sensitive current in this cell was ca 25 pA. At the end of the experiment, following regeneration, the current had reached ca 30 pA, suggesting that there had been no general physiological deterioration over the course of the intervening measurements. At time 0 (first vertical dashed line), the cell was exposed to a 30 s bleaching light calculated to have bleached ca 90% of the pigment. This light exposure resulted in complete suppression of the current for ca 20 min. Following this initial phase, the current recovered back to a level that was about 60% of the original dark current. This partial recovery was interrupted and even slightly reversed by the application of 40 μm 11-cis-retinal in the bath (second vertical dashed line at 85 min = 1.4 h), as has been previously reported (Kefalov et al. 2001). Further recovery was delayed for almost 100 min (1.7 h) in this experiment, and in no experiment for less than 60 min. The plateau phase was followed by a final exponential return of the light-sensitive current, which was complete in 4–5 h after the introduction of the 11-cis chromophore. The major difference in the recovery of fractional sensitivity (Fig. 2B) compared with the light-sensitive current is that the introduction of the chromophore immediately triggered a monotonic rise that could be described by a single exponential (time constant ca 90 min for the cell in Fig. 2) all the way to the final dark-adapted state.

These data are qualitatively consistent with those of Cornwall et al. (2000), who found that no response could be elicited immediately after bleaching 50–75% of the pigment, that the new steady state of the dark current (40% of the initial dark current) was reached within 50 min after the bleach, and that the recovery of sensitivity upon the addition of 11-cis-retinal (5 μm) took ca 2 h. It should be noted, however, that the recovery time courses after bleaches depend critically on the experimental protocol. Kefalov et al. (2001) found significantly faster recovery than either Cornwall et al. (2000) or the present study, which is probably due to the following three main differences in the experimental parameters. (1) Kefalov et al. (2001) used lipid vesicles and a high concentration (100 μm) of 11-cis-retinal for regeneration. This gives significantly faster kinetics of recovery: in our own control experiments both dark current and fractional sensitivity recovered more than 2 times faster when 11-cis-retinal was added in lipid vesicles (180 μm) than when ethanol solution (40 μm) was used in otherwise similar experimental protocols, including in both cases 20 min exposure of the cell to the retinoid solution. (The rates of recovery were measured by the times needed for 70% recovery upon addition of 11-cis-retinal.) (2) Kefalov et al. (2001) used 20% bleaches compared with 50–75% in Cornwall et al. (2000) and 90% here. Stronger bleaches will suppress the dark current for longer times, as demonstrated in salamander red rods by Tsina et al. (2004). (3) The duration of the plateau phase where no dark current recovery occurs (although sensitivity does recover) may vary with the time of incubation in 11-cis-retinal. The plateau might be related to the non-covalent interaction between 11-cis-retinal and opsin (Kefalov et al. 2001), which would be relieved only when the exogenous retinoid is washed out. Our retinoid exposure (20 min) was much longer than that of Kefalov and colleagues. However, the exact mechanisms and time courses of the recovery of dark current and fractional sensitivity are beyond the scope of this study.

The noise analyses and SQR determinations (see Figs 3–5 below) are based on measurements made during epochs of darkness and dim backgrounds as well as dim-flash responses recorded in the longer intervals where panels B and C show no data points (see, for instance Fig. 2, ca 1.5 h before the bleach and then during the last 2.5 h of the experiment, when sensitivity had essentially recovered). Instrumental noise was recorded during the post-bleach period when the light-sensitive current was zero and again at the end of each experiment in continuous bright light that closed all the light-sensitive channels.

Figure 3.

The single-photon response
A, response variation exemplified in a sequence of 30 responses to the same nominal flash intensity (2.0 Rh* flash−1; Ac= 19 μm2) in the native state. Cell 4 in Table 1. B, square of the mean response (black) and scaled time-dependent ensemble variance (green) of dim flash responses at two different nominal light intensities (2.0 Rh* flash−1; Ac= 19 μm2) and (4.1 Rh* flash−1; Ac= 19 μm2) in the native state, same cell as in A. At both light intensities the results are based on three separate sequences of 30 responses each. The scaling factor of variance provides an estimate of the mean number of photoisomerizations produced by the flash: 4.0 Rh* at the higher intensity (cf. nominal value 4.1) and 2.6 Rh* at the lower intensity (cf. nominal value 2.0). The estimates for single-quantum amplitude are obtained simply by dividing the mean response by the scaling factor of the variance. The estimates for single quantum response (SQR) amplitude derived from the two examples in the figure are 0.50 pA (based on the responses to the brighter flash light intensity) and 0.47 pA (based on the responses to the lower flash intensity). C, the mean SQR in the native state (red) and the final state (blue); same cell as in A and B. The responses are averages of 7 SQR estimates obtained at different times of the experiment from the same cell. Each of these seven individual SQR estimates used for the calculation of the mean SQR were obtained by comparison of the squared mean response and the variance of a sequence of responses to 30–50 dim flashes such as shown in A. By estimating the SQR size at different times throughout the experiment, we could also check whether any systematic changes in the SQR amplitude occurred during the experiment. The black curves are model fits according to eqn (4).

The quantal response

The size and shape of the SQR was determined by analysing response variance in sequences of 30–50 presentations of dim flashes at two different intensities nominally delivering 2.0 and 4.1 Rh* (see Methods). In Fig. 3A, the trace shows one recording sequence including 30 flashes (nominal intensity 2.0 Rh*; the recording is from cell 4 in Table 1). It is possible to discern variation in response amplitude as well as response failures. If the variation is due to Poisson variation in the actual numbers of photoisomerizations produced on successive presentations of the same nominal flash intensity, the time-dependent ensemble variance should have the same waveform as the square of the mean response and the scaling factor required to make the traces coincide is then equal to the mean number of photoisomerizations per flash actually delivered in the series (see Methods). Figure 3B shows a comparison of variance (green traces) and mean squared (black traces) for response series to two nominal flash intensities, each comprising 90 presentations (i.e. 3 similar series as shown in panel A). The traces with smaller amplitude are based on responses to nominal flash intensity 2.0 Rh* (including the series shown in panel A), the traces with higher amplitude on responses to nominal flash intensity 4.1 Rh*. The waveforms show a satisfactory agreement. The scaling factors used to produce the best coincidence were 4.0 for the larger responses and 2.6 for the smaller responses. The SQR amplitudes (obtained by dividing the mean response by the scaling factor at both light intensities) were in both cases in the range 0.4–0.5 pA. Similar response series comprising 30–50 trials were recorded at different times during the experiments in order to study possible systematic changes in response amplitude. No significant changes were found.

Figure 3C shows the SQRs in the native state (red) and in the final state (blue) of the same cell as in Fig. 3A and B. Each was obtained by averaging responses obtained from seven sequences of 30 flash presentations such as shown in Fig. 3A, and scaling the amplitude of the averaged response by the numbers of photoisomerizations actually produced, estimated as exemplified in Fig. 3B. The SQR waveform could not be well described by filter models with a single basic time parameter, as often used for linear-range responses in rods or cones (e.g. Poisson or independent activation models as defined by Baylor et al. 1974, 1979a). Moreover, the waveform varied somewhat between cells and changed with the state of adaptation. Thus the kinetics of the response could not be meaningfully captured by one simple measure, such as the time to peak. We chose to express it in terms of integration time (ti), defined as the integral of the response divided by its peak amplitude (eqn (5) in Methods).

In Fig. 3C, the SQR recorded in the final (A1) state was both larger and slightly more protracted than in the native (A2) state. This was true of three cells, whereas in two of the cells there was no significant difference in native and final SQR (Table 1). The smooth curves fitted to the responses in Fig. 3C are phenomenological descriptions according to eqn (4). The mean (±s.e.m.) values are: amplitude, 0.41 ± 0.03 pA (native) and 0.57 ± 0.07 pA (final); integration time, 3.16 ± 0.15 s (native) and 3.47 ± 0.26 s (final). If it is accepted that responses initiated by A1 and A2 pigment molecules in a single rod in a single state are indistinguishable, as has been shown to be the case in bullfrog (Firsov et al. 1994), differences in SQR reflect differences in the state of the cell's transduction machinery. One interesting possibility is that dark events occurring at a rate of about one per integration time in the native state could ‘light-adapt’ the rod causing a smaller and shorter SQR (see Discussion). Alternatively, the difference could be due to a general slowing-down of the response shut-off mechanisms over the course of the long experiment, causing a larger and longer SQR in the final state.

Dark noise and photon noise

Figure 4A shows representative samples of membrane current noise recorded under five different conditions, each comprising 164 s. Figure 4B shows noise power spectra corresponding to each of these conditions (but based on much longer recordings). The data are from the same cell as in Fig. 3. The traces in Fig. 4A show (from top to bottom): (i) noise under a dim (2.4 Rh* s−1) background light, recorded in the final state (blue trace); (ii) noise under the same background light recorded in the native state (red trace); (iii) dark noise recorded in the final state (blue trace); (iv) dark noise recorded in the native state (red trace); (v) instrumental noise recorded when the light-sensitive current had been abolished (black trace). The corresponding power spectra in panel B have been computed from a large number of noise epochs similar to those shown in panel A. For example, the data sets underlying the native- and final-state dark noise spectra in Fig. 4B comprise a total of 22 min recordings each and those under background light a total of 4 min of recordings each. Power spectral densities were calculated over 41 s segments from 82 s traces with 50% overlap (see Table 1 and Methods). The colour code of the spectra in Fig. 4B corresponds to that of the noise samples in Fig. 4A. ‘Background’ noise spectral densities are identified by squares and ‘dark’ noise spectral densities by circles.

It is immediately obvious from Fig. 4 that dark noise power at low frequencies is higher in the native (A2) than in the final (A1) state. By contrast, under a background light where photon fluctuations are strong enough to dominate over the random occurrence of thermal events, low-frequency dark noise is similar in the two states. At the two lowest frequencies (the two leftmost pairs of data points) there is a discrepancy between the two, evidently due to the limited recording time under background light (only 4 min in this cell), making the data susceptible to low-frequency artefacts. The ‘background’ power spectra cannot be perfectly fitted by scaling the power spectrum of the dark-adapted SQR, since the 2.4 Rh* s−1 background used light-adapts the cell, making the SQR smaller and faster. In the frequency domain, this corresponds to lowering the zero-frequency asymptote (S0) and extending the power spectrum towards higher frequencies. These phenomena have been analysed for native salamander rods by Jones (1998), who predicts that the total discrete noise variance arising from a background light delivering 2.3 Rh* s−1 (i.e. practically the same as used here; the corresponding power spectrum is shown in Fig. 7 of Jones, 1998) is about 1.6-fold lower than would arise from the same rate of events having the amplitude and time course of the dark-adapted SQR (eqn (2) in Jones, 1998).

Figure 7.

Dark event rates per molecule of visual pigment in 11 species of rods
Molecular dark event rates are plotted against the reciprocal of the wavelength of peak absorbance (1/λmax). The Briggsian logarithm of the rate constant (k) is plotted as a function of 1/λmax. The molecular rate constants have been corrected for differences in cell dimensions and temperature (all have been recalculated to 21°C; see details in Ala-Laurila et al. 2004a). The black symbols have been re-plotted from Ala-Laurila et al. (2004a). Red circles, native-state and final-state estimates for salamander rods from the present work; blue diamonds, present estimates for pure A1 and pure A2 pigment of salamander. The open squares show estimates for rods from the A2-dominated (dorsal) and A1-dominated (ventral) fields of bullfrog retina (Donner et al. 1990). Error bars are s.e.m. The black continuous line shows the original fit of the multimodal model of Ala-Laurila et al. (2004a) to the black data points.

The instrumental noise (shown by black circles in Fig. 4B), on the other hand, has much less power in the low-frequency range than the dark noise of the rod. The ‘biological’ noise components that interest us here are those that are not present in the instrumental noise.

Figure 5 shows difference spectra isolating these biological components of the dark noise by subtraction of the instrumental noise (red, native state; blue, final state). The low-frequency branch of the difference spectra can be fairly well fitted by appropriately scaling the power spectrum of the respective SQR (responses shown in Fig. 3B). This suggests that it originates in such quantal events. The rate of these events that would account for the observed variance can be calculated from the fit. The dashed curves in Fig. 5 correspond to the best-fitting function obtained as a sum of the scaled SQR power spectrum and the plateau level of the remaining noise component, read approximately over the range 0.3–0.5 Hz (see Methods). In Fig. 5, the difference in the low-frequency branch between the native and final states can be explained by a ca 10-fold difference in dark event rates. (Note that for a given event rate, the noise power in the final (A1) state is relatively higher due to the larger SQR.) In the remaining noise component, the ‘continuous’ noise, there is no significant difference between the native and the final state, as is evident from the approximate coincidence of the spectra beyond ca 0.3 Hz in Fig. 5. The dark current variance associated with the two noise components was calculated by integrating the two theoretical curves fitted to the difference spectrum (dark – bright light) separately over the range 0–20 Hz (see Methods). The estimates for the variance arising from discrete events were 0.055 ± 0.014 pA2 in the native state and 0.016 ± 0.006 pA2 in the final state (mean ±s.e.m., see Table 1). By contrast, the variance estimates for the continuous noise component did not differ significantly between the two states, being 0.020 ± 0.005 pA2 (native) and 0.017 ± 0.005 pA2 (final). Thus, in the native state 73% of the total dark-current variance was associated with the discrete component, whereas in the final state both noise components contributed equally. The noise characteristics of the individual cells are given in Table 1.

Rates of thermal activation of A2 and A1 pigment molecules

The mean estimates for dark event rates in the native and final states are 0.238 ± 0.026 and 0.030 ± 0.006 Rh* s−1, respectively. The underlying change in pigment composition entailed a decrease in the mean fraction of A2 pigment from 73.7 ± 2.6% (native) to 9.3 ± 1.2% (final). Thus the dark event rate decreased by 0.238/0.030 = 7.93-fold as the A2 fraction decreased by 73.7/9.3 = 7.92-fold. To a first approximation the event rate appears to be directly proportional to the amount of A2 pigment, which thus would seem to dominate the generation of spontaneous events even in the final state.

The relation between event rate and A2 fraction can be extrapolated to 100% and 0% to provide estimates for the event rates of pure A2 and pure A1 pigment, respectively. Assuming (1) response univariance, i.e. that SQRs initiated in A1 and A2 pigment molecules under the same conditions are identical (Firsov et al. 1994), and (2) complete pigment regeneration, i.e. that the total number of functional pigment molecules in the cell is the same in the native and final states, event rates per rod (f) can be expressed as follows:

display math(8)

where fA1 and fA2 are the event rates per rod due to A1 and A2 pigment fractions (xA1, xA2). The measured event rate (ftotal) is the sum of fA1 and fA2, and k1 and k2 are constants. It can be seen from eqn (8) that the constants k1 and k2 correspond to the event rates of pure A1 and A2 pigment, respectively: at xA2= 0, ftotal=k1(100% A1) and at xA2= 1, ftotal=k2(100% A2).

We plotted ftotal for each cell as a function of the A2 pigment content (corrected for volume differences) and fitted the data by linear regression according to eqn (8). The data and linear regression line are shown in Fig. 6 on both linear (Fig. 6A) and logarithmic (Fig. 6B) coordinates. Each cell is represented by two data points, one for the native and one for the final state. The advantage of the logarithmic plot (Fig. 6B) is that it expands the scales at small values of the variables, revealing a clear correlation between event rates and the small fractions of A2 remaining in the final state after bleaching and regeneration with A1 (lower left-hand group of data points). In the linear plot (Fig. 6A), the final-state data points are crowded near the origin and the correlation cannot be resolved.

Figure 6.

Dark event rates per rod as a function of the fraction of A2 pigment
A, the estimates of dark event rates of all 5 rods in Table 1 in the native and final states plotted against the fraction of A2 pigment. The red line shows the linear regression of event rate (y) on A2 content (x): y= 0.0009 + 0.318x. All event rates have here been normalized to the mean outer segment (OS) volume recorded in our experiments (2100 μm3), thus correcting for volume differences in the individual recordings. B, the same data as in A shown on logarithmic scales.

To estimate the event rates for pure A2 and pure A1 pigment, we extrapolated the regression line to 100% and 0% A2 (xA2= 1 and 0, respectively). (Alternatively, we could have determined the constants k1 and k2: see above.) Thus, the estimate for fA2 is 0.324 ± 0.030 Rh* s−1; that for fA1 is 0.0009 ± 0.0033 Rh* s−1. (The estimation was done on linear coordinates, where it is possible to extrapolate to xA2= 0.) The s.e.m. values are based on the standard deviation of estimates obtained in this way from each cell separately. The relatively large s.e.m. for A1 reflects the fact that we have no measurements for very small A2 fractions (< 6%), and the estimate is sensitive even to small changes in the slope of the fitted line. Recalculated to event rates per molecule (denoted fmol) using cell volume and estimated pigment concentration (see Methods), the estimates are fmol,A2= (7.66 ± 0.71) × 10−11 Rh* molecule−1 s−1 and fmol,A1= (2.1 ± 7.8) × 10−13 Rh* molecule−1 s−1, respectively. The ratio of the means is 365, but obviously the error limits are wide. We calculated a lower-limit estimate for the factor by which A2 and A1 pigments differ in stability in two different ways: (1) by calculating in the linear domain the lower 95% confidence limit for fmol,A2 and the upper 95% confidence limit for fmol,A1; (2) by calculating in the logarithmic domain the lower 95% confidence limit for the ratio of the geometric means (fmol,A2/fmol,A1). The two calculations give essentially consistent results:

(1) The 90% confidence limits of the (linear) estimates are mean ± 2.132 s.e.m. (Student's t distribution, d.f. = 4). Thus there is a one-tailed probability P < 0.05 for fmol,A2 to be smaller than 6.15 × 10−11 Rh* molecule−1 s−1 or for fmol,A1 to be larger than 0.19 × 10−11 Rh* molecule−1 s−1. The ratio of these, 6.15/0.19 = 32, represents a lower limit for the ratio fmol,A2/fmol,A1 at the confidence level P < (0.05 × 0.05 =) 0.0025.

(2) The log-transformed data allow statistical comparisons between the ratios of the mean estimates: the difference of means in the log-domain equals the logarithm of their ratio: logfmol,A2− logfmol,A1= log(fmol,A2/fmol,A1). We calculated 95% confidence limits for the difference of logarithms (Student's t distribution) and then converted these limits back to the linear domain. The lower-limit estimate for the ratio of the geometric means thus obtained is 36, i.e. at the confidence level P < 0.05 the event rate of A2 pigment is at least 36-fold higher than that of the A1 pigment.


Steady-state dark noise has here been measured for the first time in the same rod before and after bleaching most of the visual pigment and regenerating with a different chromophore. This made it possible to study the effect of the A2 → A1 switch, while avoiding possible differences in several other factors, notably in the opsin and other transduction proteins. The results are discussed in relation to earlier work on dark noise in photoreceptor cells and the thermal stability of visual pigments differing with respect to chromophore and spectral sensitivity.

Comparison with earlier studies on salamander rods in the native state

The power spectra of dark noise measured in native rods were qualitatively similar to those measured earlier in red rods of the same and other vertebrate species. Spectra could be construed as the sum of a low-frequency component, attributed to spontaneous activation of the visual pigment, and a component extending to somewhat higher frequencies, attributed to spontaneous activation of phosphodiesterase (PDE) (Baylor et al. 1980, 1984; Rieke & Baylor, 1996). The cut-off frequencies of these noise components (i.e. the frequencies where noise power has dropped by 50% from the maximum given by the zero-frequency asymptote, ca 0.1 and 1 Hz, respectively) as well as their relative contributions to total dark current variance were similar to values reported earlier for salamander rods (Vu et al. 1997; Jones, 1998). We attribute 73% of the dark variance to the discrete component, where Vu et al. (1997) give 67% and Jones (1998) 62%. Since these earlier studies do not give the A1 : A2 ratios of the rods, such minor differences may be due to differences in the A2 pigment content. We specifically selected a batch of salamanders with a high A2 fraction for our experiments.

By contrast, the dark event rate reported by Vu et al. (1997), 0.030 ± 0.005 Rh* s−1, is significantly lower than our estimate for native rods (0.238 ± 0.026 Rh* s−1) and close to the value we find for rods with ca 90% A1 pigment (0.030 ± 0.006 Rh* s−1). In view of the similarity of the noise power spectra in our study and theirs, this difference must arise in the analysis. Vu et al. (1997) based their estimate on visual counting of discrete events in matched-filtered records of dark current. In rods with distinct quantal bumps (e.g. toad), this method does give results consistent with those based on power spectrum analysis (Baylor et al. 1980). However, we have found that the method is unreliable in salamander rods with their relatively small SQRs and significantly higher event frequency as compared with e.g. toad rods. The method relying on dark noise power spectra is superior to alternative methods when the SQR amplitude is small and the frequency of events high, as is the case especially in native salamander rods compared with e.g. toad rods (Baylor et al. 1980). We re-analysed the dark noise power spectra of Vu et al. (1997; their Figs 1–3) based on their own data: asymptote of the discrete noise component S0= 0.35 pA2 Hz−1; SQR amplitude ca 0.7 pA, response waveform according to their model. Our analysis suggests that the dark event rate in their rods was roughly 0.16 Rh* s−1. Although somewhat lower than the estimates for our native-state rods, this value would be our prediction for a 50 : 50 ratio of A1 : A2 (see Fig. 6), which according to our experience is not unusual in larval salamanders.

The effect of the chromophore switch on dark noise

The chromophore switch (A2 → A1) was associated with a significant drop in dark noise power at the low frequencies where the scaled power spectrum of the SQR provided a good fit, whereas the ‘continuous’ noise component showed no systematic change. The latter observation is not surprising, as the effect of pigment dark activity on the average number of PDE molecules available for spontaneous activation is expected to be negligible.

The drop in the discrete noise component indicates that the A1 chromophore and the opsin form a pigment that is less susceptible to thermal isomerization than that formed by the A2 chromophore with the same opsin. Qualitatively, this is in line with several earlier studies. Williams & Milby (1968) found that the thermal decomposition of bullfrog A2 rod pigment in solution was 10 times faster than that of the corresponding A1 pigment. However, the molecular mechanism of decomposition in solution may be quite different from that underlying pigment activation in situ. More directly comparable are the results of Donner et al. (1990), who measured dark noise in rods isolated from A2- or A1-dominated fields of the bullfrog retina. They estimated that the thermal activation rate of the A2 pigment was at least 10-fold higher than that of A1 pigment per molecule. Kefalov et al. (2005) found an 11-fold decrease in the rate of spontaneous activations in salamander L-cones upon substituting the A2 by the A1 chromophore (see also Acknowledgements).

As such, our results are consistent with these (lower-limit) estimates, but our experiments have a higher resolving power, allowing us to set the lower limit for the ratio of thermal activation probabilities in A2 versus A1 pigment as high as 36-fold. This is of considerable significance for the interpretation of molecular mechanisms as well as for considerations of visual sensitivity.

We finally wish to comment on a hypothetical alternative explanation of our main result. Could the decrease in discrete noise be a consequence of bleaching and regenerating per se rather than the A2 → A1 chromophore switch? We think this is extremely unlikely for (at least) four reasons. Firstly, we performed control experiments on toad (Bufo marinus) rods, bleaching 90% of their pure A1 pigment and regenerating with the same (A1) chromophore. The estimated dark event rates in these experiments fell in the range 0.015–0.030 Rh* s−1 (21°C) both before and after bleaching and regeneration, consistent with the mean value (0.021 Rh* s−1 at 20°C) reported by Baylor et al. (1980) for the same species. Secondly, bleaching small fractions (0.02–3%) of the rhodopsin in toad rods has been found to cause a transient increase in the rate of photon-like dark events, apparently due to bleaching products of rhodopsin (e.g. metarhodopsin III) being converted back to Rh* (metarhodopsin II) (Lamb, 1980; Leibrock et al. 1994, 1998). This (fully reversible) effect of bleaching per se is obviously opposite to the change observed here after bleaching native salamander rods and regenerating with the A1 chromophore. On the other hand, any hypothetical after-effect specific to strong bleaches, acting to decrease discrete noise in the final state after regeneration (through some change in the physiology of the cell) would be expected to affect fractional sensitivity or the shape of the SQR as well. Thirdly, we find a clear correlation of final-state event rate with the fraction of A2 pigment remaining (see Fig. 6), which would be enigmatic if the noise did not depend on the chromophore. Fourthly, the change we find upon bleaching A2 and regenerating with the A1 chromophore is qualitatively consistent with the difference found by Donner et al. (1990) when comparing dark event rates of bullfrog A2 and A1 rods that had not been bleached.

Spectral tuning and the stability of visual pigments

The chromophore switch (A2 → A1) blue-shifted λmax by 26 nm and decreased the thermal activation rate of the pigment by more than 30-fold. The coupling of these two changes is consistent with a hypothesis proposed by Barlow (1957) stating that the energy (Ea) needed for activation of a pigment is inversely proportional to λmax; hence larger λmax indicates lower Ea and thus a higher probability for activation by thermal energy alone. In his original formulation of this general idea, Barlow (1957) assumed that the fraction of visual pigment molecules on thermal energy levels > Ea is given by the Boltzmann distribution and Ea corresponds to the energy of a light quantum at λmax. An approximately 120-fold increase in thermal activations is predicted when λmax shifts, as here, from 502 (A1) to 528 nm (A2). The prediction falls well within our 95% confidence range for fmol,A2/fmol,A1 (see the last two paragraphs of Results).

By contrast, when comparisons are made across different species of rods (as well as cones) with different λmax, the average dependence between thermal event rate and λmax has been found to be much shallower than predicted by Barlow's simple formulation. Figure 7 reproduces the available rod data, including our present estimates for the pure A1 and pure A2 pigment of salamander (blue diamonds). The general trend of the earlier data was well fitted by a modified theory proposed by Ala-Laurila et al. (2004a; continuous line in figure), which takes into account that the thermal energy of the pigment molecule is distributed on a large number of vibrational modes (cf. Hinshelwood, 1933). While the present estimate for salamander A2 pigment (the blue diamond at 1/λmax≈ 1.9 × 106 m−1) is roughly in line with this model and most other pigments, our A1 estimate is clearly an outlier. This will require further investigation. At this stage, two points should be noted, however. First, the 95% confidence interval of our (extrapolated) rate of thermal events in pure A1 pigment spans two orders of magnitude and thus the data cannot strongly differentiate between models. Second, in studies of photoactivation energies (Ea) of several visual pigments, it has been found that λmax shifts due to differing chromophores are associated with larger Ea shifts (i.e. more in line with Barlow, 1957 prediction) than are λmax shifts due to differing opsins (Ala-Laurila et al. 2004b). One may argue that an opsin that has to form a reasonably noise-free pigment with the relatively unstable A2 chromophore must be evolutionarily tuned to be particularly stabilizing. This would apply to the opsins of salamanders and frogs, but not to those of toads (Muntz & Reuter, 1966), and would become evident as exceptional thermal stability of the A1 form of the pigment (cf. Donner et al. 1990).

We would finally like to point out that the correlation between λmax and dark event rates predicted by theory and empirically exemplified in Fig. 7 relates specifically to the dark-adapted steady state. It does not exclude quite different effects in other conditions. Corson et al. (1990) found greatly increased light-like noise in salamander red rods after bleaching and regenerating the pigment with 4-hydroxy retinal (a chromophore common in insect pigments), although the new pigment was relatively blue-shifted (λmax= 470 nm). However, this noise was observed only transiently after light exposure and is not comparable to discrete noise measured in the dark-adapted steady state.

The chromophore switch and absolute visual sensitivity

Switching between A1 and A2 chromophores in the same opsin occurs naturally in many fish and amphibian species in connection with seasonal changes, migration or metamorphosis (for recent review, see Temple et al. 2006). Visual detection is a signal/noise discrimination task, and functional consequences must therefore be viewed under the double perspective of spectral tuning and noise control. For achromatic rod vision, the pigment that allows the highest quantum catch is always most beneficial in somewhat stronger illumination, but at very low light levels, the thermal dark events may become limiting (Jokela-Määttäet al. 2007).

In amphibians and humans, the threshold intensity for seeing an extended object in darkness has been found to lie roughly where the rate of photoisomerizations from the object coincides with the estimated rate of rod dark events in the retinal area covered by the image of the object. In the toad Bufo bufo at 15°C, threshold corresponds to ca 0.02 Rh* s−1 in each rod covered by the image (Aho et al. 1988, 1993), while the dark event rate in the same species at the same temperature is ca 0.01 Rh* s−1 (Fyhrquist et al. 1998; Firsov et al. 2002). The human threshold (at body temperature, 37°C) in the same detection task as performed by the toads also corresponded to ca 0.01 Rh* s−1 (Aho et al. 1988), close to the rate of dark events per rod estimated from psychophysical data (Barlow, 1956; Donner, 1992) as well as from discrete dark noise in macaque cones (Baylor et al. 1984). Yet it should be noted that in terms of photoisomerizations per unit retinal area (Rh* mm−2 s−1), the human threshold is more than 10 times higher than the toad threshold, since peak rod density in humans is ca 160 000 mm−2, but only ca 12 000 mm−2 in toad (Østerberg, 1935; Aho et al. 1993). Of course the same difference holds for the dark event rates per unit retinal area. At present, this empirical generalization has no straightforward mechanistic explanation in terms of signal-to-noise ratio at threshold. Assuming that it is valid, however, a hypothetical switch from 100% A1 to 100% A2 in a salamander rod, increasing the estimated dark event rate from ≤ 0.01 Rh* s−1 to ca 0.3 Rh* s−1, would pay off only if quantum catch were thereby increased by at least 30-fold. Given the 1.42 times lower photosensitivity of the A2 pigment, retinal illumination would have to increase by 40- to 50-fold to retain the same threshold ratio of photoisomerizations to thermal isomerizations in the presence of pure A2 pigment as in the presence of pure A1 pigment.

Size of the SQR In 3 cells out of 5 we observed an increase in the SQR amplitude and integration time in the final state compared with the native state (cf. Fig. 3C). This might suggest that the thermal events occurring at a rate of ca 1 event per integration time may keep the native (A2) cell in a slightly ‘light-adapted’ state compared with the final state, where the ‘intrinsic background light’ is 8 times dimmer. Baylor et al. (1980) concluded that thermal activation of rhodopsin is unlikely to play a significant role in desensitizing the transduction mechanism in darkness in toad rods, where the rate of discrete dark events is only about one per 20 integration times. Even so, these authors did find an inverse correlation between SQR amplitude and integration time in darkness, suggesting that the sensitivity and time scale of the transduction mechanism are indeed regulated in darkness by the same variable operative in background light. However, we cannot exclude the possibility that the slower response and increased amplitude in the final state may be side-effects due, e.g. to unspecific loss of mechanisms for response cut-off over the course of the long experiments, rather than expressive of normal adaptation.

Whenever randomly occurring events identical to the SQR constitute the dominant noise component, the size of the SQR is irrelevant for the signal-to-noise ratio of vision. In the native state, this may approximately be the case, since the discrete noise constituted 70% of the total noise power. In a pure A1 rod, however, the continuous noise (which was unaffected by chromophore change) is predicted to be the dominant component of dark variance. In such a case light detection would benefit greatly from a larger SQR.

Taken together, we have shown in this paper that the A1 chromophore forms in situ a much more stable pigment with the red rod opsin in the Rh1 pigment family than the A2 chromophore. We estimate that the thermal stability of the 502 nm A1 pigment is over 36-fold greater than its 528 nm A2 pair. This observation is in line with a theory according to which spectral tuning towards shorter wavelengths is coupled with a decreased rate of spontaneous activations (Barlow, 1957; Ala-Laurila et al. 2004a). In addition, the larger and slower single-photon response in the final (A1) state is consistent with the idea that spontaneous activations that occur at a rate of nearly one per integration time in the native (A2) state can light-adapt rods.



We thank Dr Maureen Estevez for skilful help and advice in the preparation of retinoid solutions as well as wonderful encouragement during this study. We thank Mr Howard I. Cohen for upgrading and improving the microspectrophotometer originally designed by Edward MacNichol, and Professor Tom Reuter for commenting on the manuscript. An unpublished study by Y. Fu, V. Ketalov, D.–G Luo, T. Xue and K.–W Yau using transgenic mice has shown that the spontaneous isomerization rate of human red cone pigment with A2 chromphore is ∼ 40 times higher than with A1 chromophore (K.–W. Y., personal communication), thus similar to our finding for salamander rhodopsin. This study was supported by a grant from Helsingin Sanomain 100-vuotissäätiö (to P. Ala-Laurila), from the Academy of Finland (grant 206221) and Societas Scientiarum Fennica (to K. Donner) and NIH grants EY-01157 (to M. C. Cornwall) and EY-04939 (to R. K. Crouch), EY-14793 (MUSC) and C06 grant R015455 from the Extramural Research Facilities Program of the National Center for Research Resources; an unrestricted grant to the Department of Ophthalmology at Medical University of South Carolina from Research to Prevent Blindness (RPB, New York). R. K. Crouch is a RPB Senior Scientific Investigator.