Not flying blind: a comparative study of photoreceptor function in flying and non-flying cockroaches

Visual systems of fast-flying insects generally outperform those of slow flyers and crawlers. In several species of fly, photoreceptor performance was linked to specific visual ecological aspects of lifestyle and behaviour (Niven et al., 2007; Weckström and Laughlin, 1995). In general, airborne predators possess large compound eyes with photoreceptors capable of resolving contrast changes with outstanding precision and speed. Temporal resolution of photoreceptors of the top diurnal fliers is extremely high, with membrane corner frequencies exceeding 100 Hz (Laughlin and Weckström, 1993). Such photoreceptors are also characterized by relatively low absolute sensitivity due to diurnal lifestyle, fast light adaptation, low input resistance and low membrane gain. The two latter properties are caused by large, rapidly activating, sustained voltage-activated potassium (K_v) conductances (Laughlin and Weckström, 1993; Weckström and Laughlin, 1995). However, the fitness costs of adaptations that underlie superior performance, especially that of leaky membranes, can be high (Niven and Laughlin, 2008), preventing widespread occurrence of such features. As a consequence, it has been shown that photoreceptors of slow fliers tend to express a more inactivating K_v conductance with a small sustained component, and therefore have relatively high input resistance, high membrane gain and low membrane corner frequency.

Although similar adaptations can be expected in other flying insects, this problem was until recently not addressed beyond Diptera. We recently examined K_v conductances in 15 phylogenetically diverse species from several orders (Frolov et al., 2016). We found that rapid diurnal flyers indeed express large non-inactivating K_v conductances, but their nocturnal and less visually guided relatives do not generally exhibit strongly inactivating K_v conductances: a positive correlation was found between the extent to which a species relies on vision and the density of sustained K_v conductance within the physiological voltage range. This has raised the questions of whether the pattern is specific to flies or whether other insect orders evolved their own suites of electrophysiological features.

Although the majority of insects possess wings, only a fraction of these fly habitually. It should therefore be interesting to test how flying affects photoreceptor function when it is not the main mode of locomotion. One way to investigate this is to compare electrophysiological properties of related species with overall similar lifestyles but different flying habits (Niven et al., 2007). We have previously characterized photoreceptors in the American cockroach Periplaneta americana (Blattidae) in detail (Heimonen et al., 2012, 2006; Immonen et al., 2014b). Periplaneta is able to fly but in laboratory conditions it flies rarely and only at high ambient temperatures. In comparison to Periplaneta, many cockroaches from families Blaberidae and Blattellidae fly more habitually. One such species is the tropical green Cuban cockroach Panchlora nivea (Blaberidae); it shares many morphological and behavioural features with Periplaneta, including nocturnal lifestyle, but flies easily when disturbed even at room temperature.

This study aims to: (1) describe anatomical and electrophysiological properties of photoreceptors from the compound eyes of Panchlora using microscopy and whole-cell patch-clamp recordings; (2) investigate the main differences between photoreceptors of Panchlora and Periplaneta; and (3) examine them in light of differing evolutionary histories and behaviours. What differences in photoreceptor properties could be expected a priori between Periplaneta and Panchlora if vision is assumed to be important for in-flight navigation of Panchlora? Based on the studies in dipterans, Panchlora photoreceptors should have a relatively small membrane time constant, a high density of the sustained K_v conductance, a high membrane corner frequency, a low membrane gain, and a high signal-to-noise ratio and information rate (Frolov et al., 2016; Laughlin and Weckström, 1993). As shown here, some of these predictions were not validated, suggesting a possibility of distinct evolutionary strategies for dealing with visual ecological challenges in Blattodea.

Cuban cockroaches, Panchlora nivea (Linnaeus 1758), of both sexes were purchased from Virginia Cheeseman (High Wycombe, Buckinghamshire, UK). American cockroaches, Periplaneta Americana (Linnaeus 1758) (males only) were purchased from Blades Biological (Edenbridge, Kent, UK), and were also given by Prof. Hiroshi Nishino of Hokkaido University, Japan from his stock culture.

Light and electron microscopy was performed as described previously (Matsushita et al., 2012). In brief, isolated eyes were pre-fixed in 1% paraformaldehyde and 4% glutaraldehyde in 0.1 mol l⁻¹ sodium cacodylate buffer (CB; pH 7.3) overnight at 4°C. After a brief wash with CB, the eyes were post-fixed in 2% osmium tetroxide in CB for 2 h at 20–25°C. Following en bloc staining with 2% uranyl acetate in 50% ethanol, dehydration in an ethanol series and infiltration with propylene oxide, eyes were embedded in Quetol 812 (Nisshin EM, Tokyo). For light microscopy (LM), 5 µm sections were examined under a light microscope (BX51, Olympus, Tokyo) equipped with a digital camera (DP71, Olympus). For electron microscopy (EM), ultrathin sections double-stained with uranyl acetate and lead citrate were examined with a transmission electron microscope (H7650, Hitachi, Tokyo).

For estimating volumes of ommatidium and rhabdom, we first measured the areas of individual ommatidia and rhabdoms in the 5 µm serial LM sections using Image J software (NIH, Bethesda, MD, USA). We then multiplied the areas by the section thickness to obtain the unit volumes, which were summed over the entire length of the ommatidium. We also measured the cross-sectional areas of ommatidia and rhabdoms in EM sections cut immediately below the crystalline cone using iTEM software (Soft Imaging System, Riverside, CA, USA), and used these values to adjust the measurement in LM sections.

Ommatidia were dissociated and whole-cell recordings were performed as described previously (Frolov, 2015). In brief, an Axopatch 1D patch-clamp amplifier and pClamp software (Axon Instruments/Molecular Devices, CA, USA) were used for data acquisition. Patch electrodes were fabricated from thin-walled borosilicate glass (World Precision Instruments, Sarasota, FL, USA). Electrodes had a resistance of 5.0–10.0 MΩ. Bath solution contained (in mmol l⁻¹): 120 NaCl, 5 KCl, 4 MgCl₂, 1.5 CaCl₂, 10 mol l⁻¹ Tris-(hydroxymethyl)-methyl-2-amino-ethanesulfoncic acid (TES), 25 proline and 5 alanine (pH 7.15). Patch pipette solution contained (in mmol l⁻¹): 120 potassium glutamate plus 20 KCl (or 140 KCl for Periplaneta recordings), 10 TES, 2 MgCl₂, 4 Mg-ATP, 0.4 Na-GTP and 1 NAD, pH 7.15. The differences in composition of intracellular solutions did not affect photoreceptor properties with the exception of inward hyperpolarization-activated Cl⁻ current (Salmela et al., 2012), which was suppressed in the presence of potassium glutamate. The liquid junction potential (LJP) between bath and intracellular solution was −12 mV in Panchlora experiments and −4 mV in Periplaneta experiments. All voltage values cited in the text were corrected for the LJP. The series resistance was compensated by 80%. Membrane capacitance was calculated from the total charge flowing during capacitive transients for voltage steps from −112 to −92 or −82 mV.

Ten monochromatic LEDs (355, 405, 435, 470, 490, 525, 535, 591, 625 and 639 nm) combined with a series of neutral density filters (ranging from 0 to 8 log units of attenuation) were used for light stimulation. To determine the spectral class of photoreceptors, a stimulus consisting of 20 ms isoquantal flashes of light from all 10 LEDs was used. Recordings were performed at room temperature (21–22°C).

To evaluate photoreceptor performance, two contrast-modulated light stimuli, a naturalistic and Gaussian white-noise, were used as described previously (Frolov, 2015). The white-noise stimulus was made of 30 repetitions of a 2 s Gaussian randomly modulated sequence preceded by an adapting 1 s steady light to accommodate the initial depolarizing transient. The stimulus had a contrast of 0.36 and a 3 dB cut off (corner) frequency of 50 Hz. The analysis of responses to white-noise stimulus was performed using Matlab 7.5 (MathWorks, Natick, MA, USA), first by averaging 30 repeats in the time domain to obtain the 2 s signal, and then subtracting the signal from each individual response to get noise traces, spectra of which were subsequently averaged. The signal-to-noise ratio (SNR) was calculated in the frequency domain and information rate (IR) was obtained according to the Shannon equation IR=∫(log₂[|S(f)|/|N(f)|+1])df (Shannon, 1948), where f is frequency, S(f) is the signal power and N(f) is the noise power, within a frequency range from 0.5 to 50.0 Hz (Frolov, 2015). In the analysis of responses to the naturalistic stimulus, the SNR was obtained from the coherence function. The contrast gain of voltage responses were calculated by dividing the cross-spectrum of photoreceptor input and output by the autospectrum of the input and taking the absolute value of the resulting frequency response function.

During statistical analysis, the Shapiro–Wilk normality test was first applied to data samples to determine whether parametric or non-parametric statistical tests need to be used. The samples that did not pass the normality test were described with medians and interquartile ranges (25% quartile:75% quartile), and tested using the Mann–Whitney U-test (MWUT). The samples that passed the normality test were analysed with parametric statistical methods, as indicated. Such data are presented as mean±s.d. and compared using a two-tailed unpaired t-test with unequal variances. Spearman's rank order correlation coefficient (SROCC, ρ) was used in the analysis of correlations. Spearman's ρ was considered significantly different from zero when P<0.05. Throughout the text, n indicates sample size.

Fig. 1 shows photographs of the heads of female and male Panchlora cockroaches. There is obvious sexual dimorphism in the head size and eye morphology. In most males, the distance between eyes was very small, up to the point of confluence: 0.096±0.03 mm versus 0.272±0.068 mm in females (n=12 in each sample, P<10⁻³, t-test). This dorsal region faces forwards and can be used for binocular vision. Having more facets in this region might be linked to better stereopsis in males. As no similarly prominent sexual differences were found in Periplaneta (P. Saari, personal communication), this development may be related to flight, which Panchlora males undertake more readily than females, according to our observations. However, this remains to be studied through detailed behavioural experiments.

Light and electron microscopy (LM and EM, respectively) were performed to study the anatomy of the Panchlora eye, and the differences between Panchlora and Periplaneta (Fig. 2). Both species share the apposition eye design. Although we could not detect any clear differences in the optical apparatuses (Fig. 2B,D), we saw considerable differences in the photoreceptor layer. Fig. 2A shows schematic drawings of ommatidia. In both species, the ommatidium contains eight photoreceptors (R1–R8), but the arrangements of photoreceptors are dissimilar. The rhabdom of Panchlora is two-tiered where different sets of photoreceptor contribute in the distal and the proximal tiers, respectively. This is in contrast to Periplaneta, where all photoreceptors contribute their rhabdomeres along the entire axis of the ommatidium (Butler, 1973).

We used 5 µm LM sections to estimate the rhabdom length and the volume of the ommatidium (Fig. 2B–E). The average rhabdom length was 182 µm in Periplaneta and 114 µm in Panchlora (Table 1). The average ommatidial volume was 143,505 µm³ in Periplaneta but only 54,063 µm³ in Panchlora. Rhabdom volumes were 11,910 and 6478 µm³ for Periplaneta and Panchlora, respectively (Table 1).

Electron micrographs of the retinal sections made immediately below the crystalline cone (140 and 75 µm from the corneal surface for Periplaneta and Panchlora, respectively) show that four large photoreceptors contribute to X-shaped rhabdoms in Panchlora (Fig. 2G). In contrast, rhabdoms in Periplaneta are characterized by irregular shapes and surrounded by much bigger pigment granules than in Panchlora (Fig. 2F). On average, the rhabdom cross-sectional area was 42±16 µm² (n=8) in Periplaneta and 81±12 µm² (n=11) in Panchlora.

In order to estimate the number of microvilli per rhabdom, we measured microvillar lengths and diameters using high magnification EM images. Microvillar lengths were similar in both species: 2 µm on average. Likewise, microvillar diameters were 69.8 nm in Periplaneta (Fig. 2H) and 68.2 nm in Panchlora (Fig. 2I). The numbers of microvilli per rhabdom were obtained by dividing the rhabdom volume value with the average microvillus volume value, yielding 1,338,400 microvilli for Periplaneta and 762,480 for Panchlora (Table 1).

Dissociated Panchlora ommatidia were notably smaller both in length and diameter, and had less screening pigment than ommatidia of Periplaneta. The two-tiered organization of ommatidia in Panchlora was often clearly identifiable by eye. Out of more than 30 Panchlora photoreceptors, from which light responses were obtained, all except three showed broad spectral sensitivity, responding most strongly to the green LED (520 nm). The other three cells responded exclusively to UV and violet LEDs (355 and 405 nm). As photoreceptors in cockroach ommatidia form a fused rhabdom and not an open type as in flies, all photoreceptor cells are probably physically accessible for patch clamping, suggesting that UV-sensitive cells are present in the majority of ommatidia.

Elementary current and voltage responses of a microvillar photoreceptor to discrete photons of light are known as current and voltage quantum bumps, respectively. Voltage bumps are complex derivatives of current bump amplitude, momentary membrane resistance and membrane time constant. The latter two factors change dynamically with voltage and time, whereas the former is influenced powerfully by the history of photoreceptor activity.

Current quantum bumps (henceforth quantum bumps) were elicited using a 1 ms flash of light of such intensity as to evoke single bumps with ∼50% probability, using a continuous light of low intensity or after prolonged stimulation with bright light. The first protocol was used to study bump amplitude and latency, the second indicated absolute sensitivity to light and the third indicated bump adaptation. The first two protocols were applied to dark-adapted photoreceptors, which were not exposed to light for over 5 min.

Fig. 3A shows a typical average quantum bump recorded from a Panchlora photoreceptor at a holding potential of −82 mV. The average quantum bump amplitude of Panchlora was 27.0±11.7 pA (n=20 cells), approximately three times higher than in Drosophila (Henderson et al., 2000) but smaller than in Periplaneta, −41.1±14.8 pA (n=26). Fig. 3B shows an example of latency distribution for quantum bumps recorded from one cell (the same cell as in Fig. 3A). The average latency was 50.4±7.3 ms (n=9). In comparison, bumps in Periplaneta had average latency of 62.2±14.1 ms (n=18, P<0.01, t-test).

Absolute sensitivity to light was measured using responses to continuous low intensity light, which on average evoked ≤10 bumps s⁻¹. Similar to other species (Frolov, 2016), a positive correlation was found between sensitivity and capacitance. Fig. 4 shows plots of sensitivity versus capacitance for Panchlora and Periplaneta. Spearman's rank order correlation coefficient was 0.79 in Periplaneta (P<10⁻⁵, n=35) but only 0.37 in Panchlora (P=0.041, n=31). The average capacitance of Panchlora photoreceptors was much smaller than in Periplaneta: 262±72 pF (n=56) versus 408±138 pF (n=70), respectively. However, despite their smaller size, Panchlora photoreceptors were overall as sensitive as photoreceptors in Periplaneta (Fig. 4, coinciding dotted lines indicate sensitivity medians). In Periplaneta, median absolute sensitivity was 0.40 (0.17:1.40) bumps s⁻¹ (for males only, at the same intensity as in Fig. 4). In comparison, in Panchlora, median absolute sensitivity was 0.48 (0.22:1.35) bumps s⁻¹. Our data suggest that Panchlora males have more sensitive photoreceptors than females: absolute sensitivity was 0.65 (0.40:1.92) in males (n=15) versus 0.33 (0.18:0.50) in females (n=16, P=0.01, Mann–Whitney U-test). Correspondingly, capacitance values were 285±107 and 261±65 pF.

How do quantum bumps in Panchlora adapt to bright light? Bump adaptation is an essential mechanism that limits membrane gain and prevents excessive membrane depolarization. In practice, bump adaptation complements the voltage-shunting mechanism that results from a decreased driving force for light-induced current (LIC) during depolarization. To assess the bump amplitude adaptation, we compared the quantum bumps evoked in the dark-adapted photoreceptors by continuous stimulation by very dim light as described above (control) with bumps evoked in similarly dim light immediately after 60 s stimulation by progressively brighter stimuli (pre-pulses).

Fig. 5A depicts averaged quantum bumps obtained in control using a light stimulus that evoked an average of 3.4 bumps s⁻¹ in this photoreceptor, and shows what happens after exposure to adapting pre-pulses in tenfold intensity increments. Fig. 5B shows the dependence of quantum bump amplitude on pre-pulse intensity for three photoreceptors. On average, the bump amplitude decreased eightfold from the dark-adapted photoreceptor to the recording after the brightest pre-pulse. However, these results underestimate bump amplitude reduction, because after high intensity pre-pulses only the largest bumps could be extracted from the background noise for analysis.

Fig. 6 shows typical macroscopic voltage (Fig. 6A) and LIC (Fig. 6B) responses of Panchlora photoreceptors to a 60 s white-noise modulated stimulus over a physiological range of light intensities, from ∼500 to 500,000 effective photons per second (eff. photons s⁻¹). However, responses tended to saturate over the course of recordings in the brightest light, which manifested in a gradual decrease in depolarization/LIC amplitude, diminishing contrast resolution and disappearance of rapid components of the response. Such saturation is likely to be due to unnatural side-on illumination in the absence of the normal optical adaptation mechanisms, and was observed in moderately bright light in virtually all photoreceptors.

Fig. 6C shows voltage-light intensity relationships for peak (blue traces) and mean sustained (red traces) depolarization for the entire experimental sample (18 cells). There is a notable variability from cell to cell in the degree of depolarization. However, unlike in Periplaneta, this variability depended neither on the photoreceptor size, nor on the sensitivity to light (Immonen et al., 2014b).

Fig. 6D–F shows membrane gain, SNR functions and IR values obtained from the voltage responses shown in Fig. 6A. With increasing light intensity, shot noise decreased, while gain, SNR and IR increased up to the level of 50,000 eff. photons s⁻¹. In still brighter light, IR decreased as a result of saturation. The maximal IR (IR_max) was on average 23±9 bits s⁻¹ when the entire 60 s (30 repeats of a 2 s white-noise sequence) stimulus was used for analysis. However, when only the first 10 repeats were used, the average IR_max was 33±12 bits s⁻¹, indicating a drastic decrease in information capacity for prolonged responses in moderately bright light (however, the residual noise level is 1.73 times higher when 10 repetitions are used for signal estimation instead of 30). Similar to several other species (Frolov et al., 2012; Frolov and Weckström, 2014; Frolov, 2015; Immonen et al., 2014a), a positive correlation was found between photoreceptor capacitance, a proxy for cell size, and maximal information rate.

The membrane corner frequency was obtained by fitting the frequency dependence of gain with a first-order Lorentzian function. For the photoreceptor responses shown in Fig. 6A, the corner frequency values changed from 6 Hz at the two lowest intensities to 4.7 and 3.6 Hz at the two brightest levels. This decrease in corner frequency illustrates the effects of saturation on the relative transfer of high frequencies.

As in many other insect species, at least two types of voltage-activated outward currents can be electrophysiologically and pharmacologically distinguished in Panchlora photoreceptors (Fig. 7). Fig. 7A shows the total K_v current elicited from a holding potential of −82 mV by 400 ms pulses between −82 and +28 mV in 10 mV increments, and Fig. 7B demonstrates the non-inactivating fraction of the current (I_DR). The transient A-type current (I_A) with hallmarks of the Shaker-like K_v current can usually be completely removed using long inactivating voltage pulses lasting several hundred milliseconds between −50 and −40 mV (Fig. 7C). I_A was measured by subtracting current traces evoked after a depolarizing pre-pulse from the currents recorded after a hyperpolarizing pre-pulse to −102 mV, which was used to recover I_A channels from inactivation (Fig. 7C). This current can also be almost completely blocked by application of 2 mmol l⁻¹ 4-AP (4-aminopyridine, data not shown).

The functional availability of I_A channels was tested using a steady-state inactivation protocol consisting of 500 ms pre-pulses between −112 and −12 mV followed by a testing pulse to +28 mV (Fig. 7D,E, black trace). The ‘window’ for I_A current had a peak at −50 mV, indicating that I_A can be involved in modulation of short depolarizing responses over most of the physiological voltage range. The half-activation potential was −34.6±7.4 mV for I_A and −34.5±7.1 mV for I_DR. As previously reported for several other insect species, larger photoreceptors expressed larger I_DR than smaller photoreceptors: a moderate positive correlation was found between capacitance and average I_DR in the physiological voltage range between −52 and −32 mV (ρ=0.46, P=0.024, n=24, Fig. 7F). No significant correlation was detected between capacitance and I_A (Fig. 7F).

In contrast to Periplaneta, the expression patterns of I_A and I_DR in Panchlora varied little from cell to cell. In Periplaneta, the majority of photoreceptors expressed a large slowly activating I_DR with a minuscule I_A, while a minor fraction of photoreceptors displays a prominent I_A with a smaller I_DR (Salmela et al., 2012). Photoreceptors of the first type were usually characterized by relatively high capacitance and ‘normal’ sustained depolarization, whereas cells of the second kind were comparatively small and demonstrated low depolarization (hence dubbed ‘hyperadapting’). In Panchlora, no such variability could be detected, with all cells showing a distinct I_A that usually exceeded I_DR in terms of peak conductance. Out of more than 30 cells, only one showed light responses resembling the hyperadapting photoreceptors of Periplaneta.

Information transfer properties were compared using responses to a 60 s naturalistic light contrast applied over a range of light intensities from moderately dark (eliciting fewer than 100 effective photons per second) to saturating (see typical voltage responses in Fig. 8A). Fig. 8B shows ‘voltage signals’ obtained by averaging IR_max traces with the highest information rate in each photoreceptor. The Panchlora trace clearly demonstrates a higher gain of the light response. Fig. 8C presents averaged gain functions and Fig. 8D shows SNR functions for the IR_max recordings. The average IR_max values were 25.1±7.9 bits s⁻¹ in Panchlora and 16.6±6.5 bits s⁻¹ in Periplaneta (P=0.003, t-test) (Fig. 8E). The corner frequency of IR_max responses was 6.1±2.3 Hz (n=15) in Panchlora and 4.6±1.5 Hz (n=16) in Periplaneta (P=0.042, t-test).

What are the probable causes for the superior performance of Panchlora photoreceptors? First, the LIC of microvillar photoreceptors depends both on the number of available sampling units and the intrinsic noise of the system. Microvilli are the sampling units of the photoreceptor; their functional availability is determined by the total number of microvilli, the fraction of microvilli inactivated by light and the rate of recovery from inactivation (Song and Juusola, 2014). The higher the recovery rate, the more microvilli can be activated at any moment of time. Taking into account bump adaptation (Fig. 5), the momentary amplitude of sustained LIC should be proportional to the number of available sampling units.

Second, membrane filtering can in theory restrict temporal resolution of the voltage response to contrast-modulated light if sustained K_v conductance is low. Therefore, it was necessary to examine both sustained LIC and K_v currents. The maximal sustained LIC values were obtained from LIC responses to stimuli of different intensities up to a saturating level by averaging the entire duration of the current response with exception of the initial transient. Average maximal sustained LIC was notably lower in Panchlora than in Periplaneta: −136±81 pA (n=14) versus −218±135 pA (n=10), respectively (P=0.05, t-test, Fig. 8F). However, owing to lower capacitance, average maximal sustained LIC density in Panchlora was slightly higher than in Periplaneta: −0.59±0.51 pA pF⁻¹ versus −0.48±0.30 pA pF⁻¹, respectively (Fig. 8G). Likewise, the sustained K_v current was much higher in Periplaneta than in Panchlora (Fig. 8J), although K_v densities in the physiological voltage range were comparable (Fig. 8H). Considering the differences in membrane capacitance, it is likely that the permissible membrane bandwidths are very similar for both species.

However, assuming that light adaptation of quantum bumps occurs in both cockroaches to the same extent, and considering that current quantum bumps in Panchlora are much smaller than in Periplaneta, it is likely that the actual number of effective sampling units, i.e. the size of the microvillus pool available for activation by light at the intensity that produces IR_max, may be substantially higher in Panchlora than in Periplaneta, despite the smaller maximal LIC in the former. Although this can explain some of the differences in gain and temporal resolution, further examination of quantum bumps provided additional clues: current bump half-width was 25.4±6.5 ms (n=16) in Panchlora and 35.7±2.6 ms (n=8) in Periplaneta (P<10⁻³, t-test) (Fig. 8I), suggesting that transducer noise is lower in Panchlora than in Periplaneta.

In this work, we compared retinal morphology and electrophysiological properties of photoreceptors in two cockroach species, of which one – Panchlora – flies habitually and readily, whereas the other – Periplaneta – only occasionally. We tested applicability of an evolutionary visual ecological paradigm that was developed for dipterans in 1990s (Laughlin and Weckström, 1993; Weckström and Laughlin, 1995) and recently elaborated for a more diverse range of species (Frolov et al., 2016). Accordingly, it was expected that the flying cockroach would have photoreceptors with a smaller membrane time constant, a lower membrane gain, a higher membrane corner frequency, a higher information capacity, a lower input resistance, a less inactivating K_v current of higher amplitude and a greater density in the physiological voltage range. Although Panchlora photoreceptors had higher membrane corner frequency and information capacity than photoreceptors in Periplaneta, other traits were not consistent with the pattern discovered in flies: Panchlora photoreceptors had higher membrane gain and a stronger inactivating K_v current with a smaller sustained K_v current.

The anatomical differences between Panchlora and Periplaneta retinas were striking in two aspects: the organization of ommatidia and the size of screening pigment granules. Although the latter cannot be plausibly explained by visual ecological or behavioural considerations, and probably reflects a large evolutionary distance between the two species, the two-tiered rhabdom in Panchlora might be specifically associated with flying. Such two-tiered rhabdoms usually serve to sharpen up colour or polarization vision (Belusic et al., 2017), and while Panchlora has only green- and UV-sensitive photoreceptors, meaning no proper colour vision, Panchlora might use polarized light for navigation during flight in the dark.

The first issue that arises with regard to relationships between visual ecology and photoreceptor biophysics is whether differences in electrophysiological properties of photoreceptors are related to behaviour at all, considering that the original observations in Periplaneta itself were anything but consistent with the above-mentioned visual ecological theory. Could it be that different insect orders evolved dissimilar suites of visual ecological adaptations? Although proper investigation of such general issues must rely on thorough comparative studies, evidence available at this stage already indicates that cockroaches might indeed have used independent methods to deal with evolutionary challenges to visual systems. The K_v has been recorded from two more cockroach species: a large wingless nocturnal crawler, Gromphadorhina portentosa (Blaberidae), also known as the Madagascar hissing cockroach; and the crepuscular flying ‘dusky’ Lapland cockroach Ectobius lapponicus (Blattellidae) (Frolov et al., 2016). There were notable similarities between Periplaneta and Gromphadorhina, and between Panchlora and Ectobius. Like Periplaneta, Gromphadorhina photoreceptors are large and express a prominent delayed rectifier with relatively little inactivation. I_A current is comparatively small in both species (Frolov et al., 2016). In contrast, photoreceptors in the miniature Ectobius appear to be smaller than in Panchlora, and display a very similar K_v current with a dominant I_A and relatively small I_DR.

There is circumstantial evidence that these results are not a coincidence but an evolutionary trend. Periplaneta demonstrates large morphological and physiological variability in all aspects of organization of its visual system periphery (Butler, 1973; Heimonen, 2008; Heimonen et al., 2006; Ribi, 1977). This variation is not ordered, unlike, for example, the variation in photoreceptor sizes in the ommatidia of the backswimmer Notonecta glauca (Immonen et al., 2014a). Rather, it is intrinsically random, and can be a sign of primitive organization. Variability in Periplaneta photoreceptor size, absolute sensitivity to light, voltage response waveforms, and amplitude and kinetics of light-induced and K_v currents are prominent (Heimonen et al., 2006; Salmela et al., 2012). In contrast, Panchlora photoreceptors demonstrated a comparatively low variability in electrophysiological properties with high functional efficiency. Superior efficiency manifested in improved information processing (see below) and in median absolute sensitivity that matched that of Periplaneta, despite the smaller membrane area of photoreceptors calculated from comparison of average capacitance values (1.6-fold difference) or rhabdom volume estimates (1.8-fold difference) (Table 1). However, more morphological studies are needed to establish whether stochastic variability in visual systems of cockroaches diminishes with increased specialization and natural selection pressure on visual performance.

If increased inactivation of K_v conductance is indeed a feature of more sophisticated Blattodea visual systems, then how can it be reconciled with the observations in Diptera? First, the length of photoreceptor axons differs between cockroaches and flies: in Periplaneta, axon lengths vary between 300 and 1500 μm (Heimonen et al., 2006), whereas in flies axons are much shorter, around 100 µm in Drosophila for R1–6 photoreceptors (Pollock et al., 1990). This indicates that graded signals transferred along the axon attenuate much more strongly in the cockroach, and that action potentials (APs) sporadically recorded in cockroach photoreceptors (Heimonen et al., 2006) may reduce information loss by regenerative amplification of higher frequencies in the response. Expression of a transient K_v conductance (K_v1, Shaker) in the axon initial segment is crucial for regulation of subthreshold membrane potentials, AP initiation and the shape of the propagating APs (Clark et al., 2009). In addition, by reducing the length constant in a voltage- and time-dependent manner, increased expression of I_A can facilitate high-pass filtering of propagated graded signals (Rusanen and Weckström, 2016). Thus, it is possible that flying cockroaches such as Panchlora and Ectobius need higher I_A conductance than the crawlers such as Periplaneta and Gromphadorhina. Alternatively, non-spiking photoreceptors of dipterans (Fuortes and O'Bryan, 1972) may need transient K_v conductances for a different purpose, unrelated to signal transfer along the axon, possibly to avoid unnecessary metabolic expenses associated with large sustained K_v conductances (Niven and Laughlin, 2008).

Here, we show that photoreceptors of Panchlora have higher information rates than photoreceptors of Periplaneta, due to increased gain and membrane corner frequency. What could explain these differences? We hypothesize that one cause might be the differences in transducer noise that apparently favour Panchlora over Periplaneta. There are three sources of intrinsic noise in microvillar photoreceptors: spontaneous activation of phototransduction cascade, thermal noise and transducer noise (Laughlin and Lillywhite, 1982; Lillywhite and Laughlin, 1979). However, contributions of the first two are negligible in comparison with the transducer noise. Transducer noise has three components, which are variations in bump shape (‘bump shape noise’), latency (‘latency noise’) and amplitude (‘shot noise’). Statistical analysis of quantum bumps in the cockroach (Immonen et al., 2014b) implies that all three components of the transducer noise are mutually independent and intrinsically stochastic. Unlike the bump shape and latency noises, shot noise is largely inconsequential for information transfer by the membrane because the effect of differing amplitudes of individual bumps constituting an aggregated response to light will be cancelled out by averaging, except perhaps in very dim light when single bumps dominate the response. Indeed, it has been suggested that shot noise may not affect information capacity at all (Salmela, 2013).

Noise caused by variations in bump latency and shape, however, can affect the kinetics and amplitude of the aggregated macroscopic response. Latency noise adds stochastic desynchronization of bump onset times, leading to a broader response of smaller amplitude than in the absence of such variation, an effect similar to low-pass filtering. Importantly, as latency is considered to be a Poisson process, latency variation equals mean, with an apparent consequence that in order to improve synchronization of bump summation, mean latency needs to be shortened. This is generally consistent with a visual ecological pattern observed among insects: the species that rely heavily on vision have shorter photoreceptor response latencies than do less visually dependent organisms. In addition, as response latency decreases strongly with increasing brightness of the stimulus (about twofold in Panchlora from the background intensity ‘−4’ to ‘−1’ as in Fig. 6C), latency noise should also diminish, with concomitant increase in gain, temporal resolution and SNR. Comparison of mean latencies of Panchlora (50.4 ms) and Periplaneta (62.2 ms) photoreceptors implies that latency noise should indeed be relatively small in Panchlora. Likewise, the bump shape noise is expected to increase the duration and decrease the amplitude of a macroscopic response, thus reducing gain and membrane corner frequency. Although the relative contributions of latency and bump shape noises are not known, both are significantly smaller in Panchlora than in Periplaneta.

One of the shortcomings of this work is that no attempt was made to evaluate possible regional differences in photoreceptor function, which might stem from the well-known regional morphological specializations such as the genuine specialized areas [e.g. the dorsal rim area in crickets (Frolov et al., 2014)] or various high-acuity zones (Land, 1997). Do photoreceptors in such high acuity regions process information better, i.e. faster and with higher SNR, than in the rest of the eye, as would be warranted by visual ecological considerations? However, neither Periplaneta nor Panchlora possess any morphologically distinct regions. The present electrophysiological method, patch-clamp recordings from photoreceptors in dissociated ommatidia, is particularly unsuitable for addressing this question, which requires an intracellular recordings approach. Nevertheless, we can safely assume that any putative variation in photoreceptor performance due to regional differences falls within the variabilities in electrophysiological properties described here.

In conclusion, this work described the biophysical properties of photoreceptors in the cockroach Panchlora nivea and is the second detailed study of a Blattodea species. The findings raise the possibility that, due to different morphologies of visual systems, and, probably, different roles of visual sense in physiology and behaviour, evolutionary adjustment of photoreceptor properties to visually demanding behaviours proceeded in the comparatively ‘non-visual’ Blattodea along different lines than in the highly visual Diptera, a hypothesis that needs to be tested in further comparative studies.

The authors thank Prof. Andrew French for critical reading and valuable suggestions, and Prof. Hiroshi Nishino for providing Periplaneta americana from his stock culture.