Higher-order neural processing tunes motion neurons to visual ecology in three species of hawkmoths

Abstract

To sample information optimally, sensory systems must adapt to the ecological demands of each animal species. These adaptations can occur peripherally, in the anatomical structures of sensory organs and their receptors; and centrally, as higher-order neural processing in the brain. While a rich body of investigations has focused on peripheral adaptations, our understanding is sparse when it comes to central mechanisms. We quantified how peripheral adaptations in the eyes, and central adaptations in the wide-field motion vision system, set the trade-off between resolution and sensitivity in three species of hawkmoths active at very different light levels: nocturnal Deilephila elpenor, crepuscular Manduca sexta, and diurnal Macroglossum stellatarum. Using optical measurements and physiological recordings from the photoreceptors and wide-field motion neurons in the lobula complex, we demonstrate that all three species use spatial and temporal summation to improve visual performance in dim light. The diurnal Macroglossum relies least on summation, but can only see at brighter intensities. Manduca, with large sensitive eyes, relies less on neural summation than the smaller eyed Deilephila, but both species attain similar visual performance at nocturnal light levels. Our results reveal how the visual systems of these three hawkmoth species are intimately matched to their visual ecologies.

1. Introduction

Sensory systems provide the crucial information that allow animals to live and procreate successfully, and in order to support an animal optimally, the senses need to be adapted to an animal's ecological demands [1]. Insects, despite their incredible diversity in terms of habitats and lifestyles [2], have successfully been used to study how the senses have adapted to ecological requirements. Much of the reason for this lies in the conserved nervous systems of these animals [3] which allow studies of homologous structures relatively easily. Adaptations to ecology can occur at different stages of sensory processing: peripherally in the anatomical structures of the sensory organ and its receptors, and centrally in higher-order neural processing circuits in the brain. Together, they tune the senses to an animal's specific needs [1]. While peripheral adaptations shape the entire sensory system of the animal, central adaptations can act specifically on different sensory sub-systems in parallel, allowing them to fine-tune responses more precisely without having to comply with the requirements of all other sub-systems as well.

In the sensory periphery, several studies on insects have successfully unravelled ecological adaptations. In the insect visual system, in particular, many striking examples of peripheral sensory adaptations have been documented at optical and morphological levels. The eyes of male bibionid flies, which have greatly enlarged dorsal regions with highly increased visual acuity for spotting females against the sky [4], are an extreme example of how a sensory organ has become adapted to a specific ecological need. Similar adaptations, tuned to a range of specific ecologies, can also be found in the eyes of many other invertebrate species [5]. Additional physiological adaptations in the visual periphery of insects have also been well described [6–10]. For example, diurnal insects have both higher spatial acuity [7,10] and faster photoreceptors [9,11] than their nocturnal relatives.

While the contributions of the visual periphery to specific sensory tuning are well understood in insects, less is known about the role of higher-order processing in shaping visual responses to specific ecological needs. One exception is the wide-field motion vision system, which forms the basis of the insect flight control mechanism (see [12] for review). The response properties of its main output neurons are matched to flight speed [13,14], to the spatial structure of their habitat [15], and to the light intensity they are active in [13,14,16]. However, it has so far not been possible to determine whether the tuning of these neurons is exclusively shaped by well-known peripheral adaptations, or whether higher-order processing plays an additional role, because the neurons carrying out the motion computations have been difficult to access physiologically. We recently overcame this problem by using an indirect technique to quantify higher-order processing in the insect motion vision pathway [17,18], which involved recording from both the inputs (photoreceptors) and the outputs (wide-field motion-sensitive neurons) of the system, and using well-established computational models [17,19,20] to calculate the amount of higher-order neural processing that occurred prior to the output.

In this study, we extended this approach to investigate the relative contributions of the visual periphery and higher-order processing on the spatial and temporal tuning of motion-sensitive neurons in different insect species. We focused our investigation on three hawkmoth species that forage at radically different light intensities (figure 1a): in bright daylight (Macroglossum stellatarum), exclusively at night (Deilephila elpenor), or at a diverse range of crepuscular and nocturnal light levels (Manduca sexta). All three species are nectivorous hover-feeders [21]. A major advantage of constraining our analysis to a taxonomically close group [22] with very similar behaviours is that we can record from homologous neural systems, and be reasonably certain that the main selection pressure that shaped their neural responses was the light intensity window in which each species is active, rather than their foraging mode or flight kinetics.

Figure 1.

Quantifying the effects of peripheral and central ecological adaptations in the wide-field motion vision system of hawkmoths. (a) We investigated the motion vision pathway in three hawkmoth species: the diurnal Macroglossum stellatarum, the crepuscular Manduca sexta, and the nocturnal Deilephila elpenor. M. stellatarum is active in bright light and has little need for adaptations that increase visual sensitivity. However, the two species active at lower light levels have a greater need for increased visual sensitivity. (b) We quantified the extent to which peripheral (eyes and retina) and central adaptations (higher-order neural processing) contribute to the processing of visual motion in the three hawkmoth species. We recorded from photoreceptors in the retina and from wide-field motion-sensitive neurons in the lobula complex. Moths viewed moving sinusoidally modulated patterns of black-and-white stripes (gratings) generated on an LCD screen, with the intensity controlled across six orders of magnitude via neutral density filters. We stimulated neurons with gratings across a range of spatial and temporal frequencies (by varying the width and speed of the moving stripes). This allowed us to quantify the spatial and temporal properties of photoreceptors and wide-field neurons, which in turn made it possible to extract the additional spatial and temporal processing of visual signals that occurred in the motion pathway between the retina and the lobula complex. (Online version in colour.)

These three hawkmoth species face different challenges with respect to their natural light environments, because light intensities at night can be more than 6–8 orders of magnitude lower than during the day [23]. In addition, photon shot noise (the variance in the number of photons arriving per visual integration time) is relatively greater in dim than in bright light (because it grows as the square root of the signal) [24,25], leading to low signal-to-noise ratios for photon detection in dim light. To cope, the visual systems of nocturnal insects generally trade-off spatial and/or temporal resolution for improved sensitivity [26,27], both peripherally (via adaptations in the eye that result in photoreceptors having wider receptive fields [28–30] and increased integration times [9,11,30]), and centrally (by integrating signals in space and time [18,31]). We therefore investigated to what degree the three different hawkmoth species used peripheral versus central adaptations to shape their visual motion sensitivity in space and time. Our results show that in addition to peripheral tuning, higher-order processing substantially shaped the spatial and temporal characteristics of the motion vision system in all three hawkmoth species, uniquely matching each species to its ecological needs.

2. Results and discussion

(a). The sensitivity and spatio-temporal profiles of motion-sensitive neurons are tuned to visual ecology

Before attempting to unravel the underlying mechanisms, we first established that the wide-field motion system, which processes optic flow information, was indeed tuned to the ecological requirements of each species. To achieve this, we recorded from wide-field motion-sensitive neurons in the lobula plate region of the optic lobe of the brain using moving, sinusoidal grating patterns as the visual stimulus and a neutral density filter to adjust image brightness (figure 1b; see the electronic supplementary material). We obtained the contrast sensitivities of motion neurons (quantified as the contrast C₅₀ at which the neurons reached 50% of their maximum response, figure 2b) over a large range of spatial and temporal frequencies, and at light intensities ranging from sunset to starlight levels (100–0.0001 cd m⁻²). The responses we obtained were characteristic for hawkmoths [13,14,16,32]: good spatial acuity and maximum sensitivity to relatively slow-moving patterns compared with many species of flies, bees, or butterflies [13,14]. The temporal responses showed an inhibition at temporal frequencies above 15 Hz (blue regions in each surface plot in figure 2a), which has also been previously described for hawkmoth wide-field motion-sensitive neurons [13,14].

Figure 2.

Spatial and temporal tuning of motion-sensitive neurons was matched to hawkmoth visual ecology. (a) Contrast sensitivity (threshold) surfaces were calculated from the contrast C₅₀ at which the motion neurons reached 50% of their maximum response at each combination of spatial and temporal frequency (contrast sensitivity = 1/C₅₀). Each surface shown here was obtained using the light intensity at which neurons from each hawkmoth species showed their peak sensitivity. The spatial and temporal frequencies at which sensitivity peaked (white solid lines) are indicated. The colour scale for contrast sensitivity is shown to the right of each panel. Warmer colours indicate excitation, cooler colours indicate inhibition. The number of neurons recorded for each species is given in (a). (b) Peak contrast sensitivity of motion-sensitive neurons at different light intensities (M. stellatarum did not respond below 0.01 cd m⁻²). (c,d) Spatial (c) and temporal (d) frequencies at which sensitivity peaked (solid lines) and dropped to 50% of its maximum (dashed lines) of motion-sensitive neurons at different light intensities. (b–d): Median (solid and dashed lines) and inter-quartile ranges (shaded regions). (Online version in colour.)

The contrast sensitivity of these neurons clearly matched the diel activity pattern of the animals (figure 2b). At sunset light intensities (100 cd m⁻²), all species had peak contrast sensitivities of around 15–20 (thus being able to see contrasts as low as 5–7.5%): 14.9 ± 5.6 (±standard deviation for all presented data if not stated otherwise) in the diurnal M. stellatarum, 17.4 ± 4.1 in the crepuscular M. sexta, and 15.3 ± 5.4 in the nocturnal D. elpenor. As light intensities decreased, the sensitivity of the diurnal moth dropped rapidly, with no measurable response once light intensities fell below moonlight levels (0.01 cd m⁻²). By comparison, contrast sensitivity in the crepuscular and nocturnal moths actually rose as intensities initially fell below sunset levels (figure 2b) and continued to remain high at moonlight levels (0.01 cd m⁻²: 14.3 ± 5.6 in M. sexta and 15.9 ± 4.1 in D. elpenor). Sensitivity only decreased below sunset values once the light intensity decreased below moonlight intensities. No detectable responses could be measured in any species at intensities below starlight levels (0.0001 cd m⁻²).

Consistent with earlier work on hawkmoths, the basic spatial and temporal tuning of the motion responses were clearly adapted to the light levels that the species are active within (figure 2c,d) [13,14,16]. In terms of temporal tuning (figure 2d), the diurnal M. stellatarum had the fastest overall temporal responses (at 100 cd m⁻²: 50% cut-off at 12.1 ± 2.9 Hz), followed by the crepuscular M. sexta (at 1 cd m⁻²: 50% cut-off at 7.3 ± 0.9 Hz), while the nocturnal D. elpenor had the slowest temporal tuning (at 1 cd m⁻²: 50% cut-off at 6.1 ± 1.2 Hz). A similar picture arose for the spatial resolution of the motion pathway. Spatial resolution (figure 2c) was greatest in the diurnal hawkmoth at high intensity (at 100 cd m⁻²: 50% cut-off at 0.26 ± 0.05 cyc/°) but this decreased nearly linearly with decreasing light intensity. In the crepuscular and nocturnal species, spatial resolution remained similar to that at the brightest intensity until moonlight levels (at 0.01 cd m⁻² : 50% cut-off at 0.15 ± 0.02 cyc/° and 0.13 ± 0.02 cyc/°, respectively). Below moonlight levels, spatial resolution decreased steeply in the nocturnal hawkmoth, but remained somewhat higher in the crepuscular hawkmoth. In conclusion, the tuning of the motion neurons in these hawkmoth species matched their visual environment, with the highest absolute sensitivity and lowest spatio-temporal resolution in the nocturnal hawkmoth, and vice versa in the diurnal hawkmoth.

(b). Characteristics of the optics and photoreceptors match the visual ecologies of hawkmoths

What role do the optics and photoreceptors of hawkmoth eyes play in shaping this match between diel activity phase and the spatio-temporal properties of hawkmoth motion neurons? Hawkmoths have superposition eyes, in which hundreds of facets collect and focus light onto a single rhabdom in the retina (which processes one pixel of the image). The pool of facets collecting the light for this image functions as a pupil that sets the effective light intensity on the retina. During light adaptation, pigment granules migrate proximally into the eye, thus progressively cutting out light and reducing pupil size [33]. We quantified the effective pupil of all three species over the same range of intensities as used in the physiological experiments (figure 3a). The diurnal hawkmoth showed no change in pupil aperture under our experimental light conditions and only closes its pupil under extremely bright conditions [34]. The other two species had closed pupils at the highest intensity tested (100 cd m⁻²), but had a nearly fully open pupil by 1 cd m⁻², equivalent to early dusk light levels (figure 3a). The fully open pupil of the diurnal hawkmoth had the smallest effective aperture (an average of 0.086 mm², figure 3b; electronic supplementary material, table S1), while the nocturnal species had an aperture more than five times larger (0.44 mm²). In the crepuscular hawkmoth, the aperture was a further three and half times larger (1.56 mm²).

Figure 3.

The visual periphery reflects the visual ecology of hawkmoths. (a) Eye glow, the light reflected from the open superposition pupil (see inset), was used to measure the relative openness of the superposition aperture at light intensities ranging from 100 to 0.0001 cd m⁻². Averages and standard errors are shown. Colours refer to the three hawkmoth species depicted in (b). (b) The diameter of the fully open superposition aperture. Averages and standard errors are presented. (c,d) Best-fit Gaussian models of photoreceptor responses to sinusoidal gratings of varying spatial frequency at 100 cd m⁻² (c) and at the lowest light intensity photoreceptors responded to ((d), M. stellatarum: 1 cd m⁻²; D. elpenor and M. sexta: 0.01 cd m⁻²). The photoreceptor acceptance angle Δρ was quantified as the half-width of the Gaussian response function (see arrows in (d)). (e,f) Best-fit lognormal models of photoreceptor responses to sinusoidal gratings of varying temporal frequency at 100 cd m⁻², (e) and impulse responses of photoreceptors in the dark adapted state (f). In D. elpenor and M. sexta, responses were corrected for the discrepancy in temporal tuning obtained using impulse and grating stimuli (see the electronic supplementary material and also figure S1) (DA = dark adapted). (c–f) Lines represent averages, and shaded areas standard errors. (Online version in colour.)

The close match between intensity-dependent changes in pupil aperture (figure 3a) and the contrast sensitivity changes we observed at the higher luminance ranges tested (figure 2b) suggest that the pupil plays a major role in shaping contrast sensitivity in the nocturnal hawkmoth, at least down to dusk light levels (0.1 cd m⁻²). In particular, the contrast sensitivity of the two species active in dim light was maximal at 1 cd m⁻², the highest intensity at which the pupil fully opened (figure 3a), despite the overall decrease in light intensity from the maximum of 100 cd m⁻², (figure 2b). This observation can be explained by the larger pupil aperture at 1 cd m⁻², which results in a higher effective light intensity on the retina than the closed pupil at 100 cd m⁻². Correspondingly, in the diurnal hawkmoth, with the pupil already fully open at the brightest intensity, contrast sensitivity fell continuously with decreasing light intensity.

In addition to differences in the superposition pupil, we also observed clear tuning of the spatial characteristics of photoreceptors to the different diel activity phases (figure 3c,d; electronic supplementary material, table S2). The physiological limit to spatial resolution in a compound eye is set by the shape and width of the spatial receptive fields (or ‘angular sensitivity functions’) of the photoreceptors. In particular, the half-width of the angular sensitivity function—the ‘acceptance angle’ Δρ—is a convenient measure of physiological spatial resolution, with larger values indicating poorer resolution. The diurnal M. stellatarum had the highest spatial resolution, indicated by the smallest acceptance angle (Δρ = 1.75° ± 0.40° at 100 cd m⁻²; figure 3c,d), while the nocturnal D. elpenor had the lowest spatial resolution (Δρ = 4.04° ± 0.48° with a fully open pupil at 0.01 cd m⁻²). The crepuscular M. sexta was intermediate (Δρ = 3.26° ± 0.34° at 0.01 cd m⁻²). Thus, the general pattern of spatial resolution in the retina matched that of the wide-field motion neurons across species.

Interestingly, all species had a similar and surprisingly fine spatial sampling baseline as indicated by the interommatidial angle, Δϕ, which determines the sampling frequency for underlying visual processing. This averaged 0.91° ± 0.08° in M. sexta, 1.12° ± 0.11° in D. elpenor, and 1.3° in M. stellatarum (electronic supplementary material, table S1). When anatomical and physiological resolution are matched to sample optimally, the ratio of Δρ to Δϕ is around 2 [35]. For ratios less than around 2, the eye is said to undersample while for ratios greater than this the eye oversamples. In the two species active in dim light, a high anatomical sampling resolution (due to a narrow Δϕ) results in oversampling of the visual image (due to a wide Δρ). In both species, Δρ was around 3.6 times coarser than the limiting anatomical resolution (Δρ/Δϕ = 3.6). However, the physiological resolution of the diurnal hawkmoth indicates undersampling (Δρ/Δϕ = 1.35), suggesting that spatial resolution has been prioritized over sensitivity (electronic supplementary material, table S1). Visual oversampling is well known from other nocturnal species with selective pressure to increase sensitivity, such as nocturnal sweat bees [36,37].

The close agreement between sampling and diel activity is also seen in the temporal properties of the photoreceptors (as given by the shape and duration of the temporal impulse response). Photoreceptor temporal resolution at higher light intensities was greatest in the diurnal hawkmoth, and lowest in the nocturnal species, while the crepuscular hawkmoth was intermediate (figure 3e). Interestingly, temporal resolution became similar across species in the dark-adapted state (figure 3f), suggesting that all photoreceptors might share a base level ‘slow’ temporal resolution in darkness that is actively sped up to different extents in the species active at brighter light intensities, as seen in diurnal flies [9].

Taken together, the anatomical and physiological characteristics of the visual periphery in the three hawkmoth species showed a clear match to their visual ecologies, and to the properties of their wide-field motion neurons, with spatial and temporal resolution being highest in the diurnal species, and lowest in the nocturnal species, while visual sensitivity scaled opposite.

(c). Higher-order neural processing contributes to the tuning of motion neurons

Given similar trends in the spatial and temporal resolution of photoreceptors and wide-field motion neurons across species, do the photoreceptor properties completely define the tuning of motion neurons, or does additional higher-order processing play a role? Knowing the spatial and temporal characteristics of the photoreceptors, we were able to quantify higher-order spatial and temporal processing in the motion vision pathways of the three species using a well-developed computational model for insect wide-field motion processing ([18], electronic supplementary material, figure S2). This model computes motion responses to sinusoidal input signals with the same spatial and temporal frequency range as used for physiological measurements, using standard elementary motion detector (EMD) units implementing a Reichardt-type correlator [38]. Prior to the EMDs, the signals were filtered with the spatial and temporal characteristics of the photoreceptors—thus defining the properties of the visual input—before being passed to an additional stage of spatial and temporal low-pass filtering. This additional low-pass filtering was adjusted to fit the responses of motion neurons, and represented spatial and temporal summation [39].

Spatial summation was modelled as a simple Gaussian filter implemented between the peripheral filters and the EMDs (electronic supplementary material, figure S2). The modelled angular half-width Δρ_s of this spatial summation filter, quantifying the extent of spatial summation, exceeded the anatomical sampling base for all three species at all light intensities (figure 4a), thus suggesting some degree of spatial summation between neighbouring visual units beyond that provided by the angular sensitivity function alone. In all three species, the extent of this additional spatial summation was very low at the highest effective retinal light intensity (100 cd m⁻² in the diurnal hawkmoth, 1 cd m⁻² in the crepuscular and nocturnal hawkmoth for fully open pupils) but then increased with decreasing effective intensity (figure 4a). The extent of spatial summation in the nocturnal D. elpenor at the lowest light intensities exceeded that of the crepuscular and diurnal hawkmoth by a factor of 2.

Figure 4.

Central adaptations in the hawkmoth motion vision system. (a) Model estimates of the angular half-width Δρ_s of the best-fitted spatial low-pass filters performing spatial summation for each species at different light levels. (b) Model estimates of the time constants Δt_s of the best-fitted temporal low-pass filters performing temporal summation in each species at different light levels. Shaded regions around the mean lines in (a,b) represent the inter-quartile range. See the electronic supplementary material, figure S2 for model details. (c,d) The ratio between the peak spatial and temporal frequencies of the motion-sensitive neurons and the peak obtained from models of motion vision excluding higher-order spatial and temporal processing (based on the spatial and temporal properties of the photoreceptors at the respective light intensities). The ratios were calculated for all three species, both at high effective retinal light intensities (100 cd m⁻² M. stellatarum, 1 cd m⁻² D. elpenor, M. sexta) and at low intensities (0.01 cd m⁻² M. stellatarum, 0.001 cd m⁻² D. elpenor, M. sexta). A ratio of 1 means that the spatial and temporal tuning of the motion neurons was set by the photoreceptors, while a ratio less than 1 implies the presence of higher-order summation (and decreased spatial and temporal resolution). (Online version in colour.)

Temporal summation was modelled as an exponential low-pass filter with a time constant Δt_s (electronic supplementary material, figure S2). The best-fit values of this temporal summation filter also followed the pattern observed in the photoreceptor and motion responses (figure 4b). The diurnal M. stellatarum had the smallest extent of temporal summation, the nocturnal D. elpenor the greatest extent, while the crepuscular M. sexta was intermediate. In the crepuscular and nocturnal hawkmoth, temporal summation increased as light intensities decreased below 0.01 cd m⁻², while it remained rather stable down to 0.01 cd m⁻² in the diurnal hawkmoth.

In summary, our modelling showed substantial additional higher-order processing (neural pooling) in the motion vision pathway of all three hawkmoth species, reflecting their diel activity patterns. The extents of spatial and temporal summation were greatest in the nocturnal D. elpenor, reflecting its need to maximize sensitivity to see well in very dim light. While the crepuscular M. sexta and the diurnal M. stellatarum reached similar maximum levels of spatial summation, there was a general correlation between temporal summation and light intensity in all species: the lower the effective light intensity on the retina, the greater the extent of temporal summation.

(d). Both peripheral and central adaptations shaped motion vision in bright and dim light

Photoreceptor responses and higher-order visual processing were both matched to the visual requirements of the three hawkmoth species, but what effect did the additional filtering contribute to the responses of wide-field motion neurons? In order to quantify this, we calculated the ratio between the peak spatio-temporal tuning of the actual neurons (i.e. including the spatial and temporal summation within the motion pathway) and modelled motion responses from a simplified model based only on the spatio-temporal filtering as measured in the photoreceptors (i.e. in the absence of neural summation). We then compared the peak spatial and temporal frequencies of those two responses: a ratio of 1 indicates that the responses of the visual motion pathway are entirely shaped by the periphery, a ratio higher than 1 would indicate central contributions that increase resolution and a ratio lower than 1 indicates central contributions that decrease resolution (and increase sensitivity).

In the spatial domain, this ratio was close to 1 at high effective retinal light intensities for all three species (at 100 cd m⁻² for M. stellatarum, and at 1 cd m⁻² for M. sexta and D. elpenor). In other words, at high luminance the spatial resolution of the motion vision pathway was dominated by the tuning of the visual periphery (figure 4c). This finding does not contradict our finding that there are still low levels of spatial summation present at these light intensities, especially in the crepuscular and nocturnal hawkmoth at these light intensities (figure 4a). This is due to the high degree of redundancy resulting from retinal oversampling discussed above. Spatial summation that does not exceed the photoreceptor acceptance angle has little impact on the spatial acuity of the system as a whole, thereby increasing sensitivity without sacrificing spatial resolution. However, at low effective retinal light intensities (0.01 cd m⁻² for M. stellatarum, and 0.001 cd m⁻² for M. sexta and D. elpenor), this ratio falls, indicating that the spatial tuning of the visual motion pathway is strongly influenced by additional processing (figure 4d). In the diurnal and nocturnal hawkmoth, the spatial peak was only 50% of that set by the photoreceptors (figure 4d). The crepuscular species retained more of its original spatial acuity, due to consistently lower levels of spatial summation, as discussed earlier. These results suggest that spatial summation is used in all three species to increase the sensitivity of the wide-field motion system at low light intensities (thus compromising spatial acuity), while it remains low in bright light, optimizing spatial acuity to a value close to that set by the visual periphery.

In contrast to spatial tuning, temporal tuning even at the highest effective light intensity was clearly influenced by temporal summation in all three species (figure 4c). Particularly in the nocturnal species, the peak of temporal tuning was only about 25% of that set by the photoreceptors (figure 4c). The contribution of temporal summation increased even further at lower light intensities, especially in the diurnal and crepuscular hawkmoth (figure 4d): peak acuity in the diurnal hawkmoth decreased to 60% of the acuity set by the photoreceptors, and in the crepuscular species it decreased to less than 30%, while it remained at around 25% in the nocturnal species. The presence of temporal summation even at brighter light intensities suggests that higher-order temporal processing in the motion vision system of hawkmoths is not only used to increase sensitivity. If this were the case, it should be minimal in bright light, to optimize temporal acuity. Rather, we suggest that such filtering helps to tune the temporal properties of the motion vision system in all three hawkmoth species to a certain range of temporal frequencies. Previous studies have suggested that hovering flight, as performed by hawkmoths during flower feeding, might bias tuning to lower temporal frequencies to improve coding of low image speeds [13,14].

Taken together, higher-order processing shaped the responses of motion neurons in all three hawkmoth species. While the effects of higher-order processing on spatial tuning emerged only in dim light, additional filtering limits temporal resolution at all light intensities.

(e). The balance between peripheral and central adaptations also depends on physical constraints

While we have so far analysed the balance between the contribution of peripheral and central visual adaptations to the tuning of motion neurons, our three model species also allowed us to investigate how differences in the size and structure of their eyes relate to the peripheral and central strategies employed by their visual systems. Of the three hawkmoth species, M. sexta is by far the biggest, and has the largest eyes [32], with twice the facets of the nocturnal D. elpenor, and more than four times the facets of the diurnal M. stellatarum, despite relatively similar facet sizes in all three species (electronic supplementary material, table S1). Larger eyes also allow for a larger pupil aperture in Manduca (electronic supplementary material, table S1), suggesting that their eyes might have an intrinsically higher sensitivity than those of the other two species. In order to compare the contribution of peripheral and central adaptations on sensitivity across species, we calculated the proportion of photons absorbed by each visual unit (in this case, a rhabdom of an isolated ommatidium, representing a single pixel of the image) per visual integration time from an extended source of broad-spectrum light (in units of photons µm⁻² s⁻¹ sr⁻¹, see the electronic supplementary material, equation S1) [40]. These calculations do not account for photoreceptor spectral sensitivities, and assume a fully dark-adapted eye with a fully open superposition pupil. Such calculations reveal that an isolated rhabdom in M. sexta would absorb by far the most photons (34.2 photons), about three times more than the nocturnal D. elpenor (11.3 photons) and 171 times more than the diurnal M. stellatarum (0.2 photons).

Thus, the very large eyes of the crepuscular hawkmoth allow for a much greater superposition aperture (electronic supplementary material, table S1), giving it a clear advantage in visual sensitivity. More specifically, the ratio of the focal length of the eye f to the aperture diameter A—the so-called F-number—is substantially lower in Manduca than in the other two species (electronic supplementary material, table S1). Eyes of lower F-number produce brighter images, suggesting that Manduca requires a less substantial contribution from central processing to achieve its sensitivity. Conversely, the much smaller nocturnal hawkmoth—which achieves a contrast sensitivity similar to that of M. sexta (figure 2b)—would be expected to rely on a stronger contribution from central processing. Indeed, M. sexta's extent of spatial summation was rather low (figure 4a), which is also reflected in the relatively low contribution of central processing to the spatial tuning of their motion neurons, both in bright and dim light (figure 4c). The nocturnal D. elpenor, on the other hand, showed high levels of spatial summation in dim light (figure 4a), which contributed strongly to the spatial characteristics of its motion neurons (figure 4c). We quantified the contribution of neural pooling to the overall increase in signal by extending the previous calculation with the spatial and temporal summation parameters estimated at the lowest light intensity at which each species responded (electronic supplementary material, equation S2; figure 4a,b). Since spatial and temporal summation in the nervous system operate on the neural signals resulting from single photon detections, and because these circuits do not actually capture photons, we termed the resulting quantity ‘equivalent photon catch’. This equivalent photon catch was 221 equivalent photons per visual processing unit and integration time for D. elpenor, 239 equivalent photons for M. sexta, and three equivalent photons for M. stellatarum. We conclude that the additional neural filtering in the nocturnal hawkmoth indeed boosted its visual sensitivity, while the crepuscular species obtained similar sensitivities with much lower levels of spatial summation, albeit at the expense of a larger eye. The equivalent photon catches of the two species active in dim light were around two orders of magnitude greater than that of the diurnal hawkmoth, a difference which was well matched with their physiological contrast sensitivities (which extended to light intensities at least two orders of magnitude dimmer than that of the diurnal hawkmoth (figure 2b)).

This example illustrates that the balance between peripheral and central adaptations depends on the extent to which each can contribute to the specific ecological requirements of a particular insect species. As our example shows, a larger animal having larger eyes, with potentially greater optical sensitivity than achievable by a smaller-bodied animal with smaller eyes, requires a smaller contribution from central processing to achieve the same visual performance. M. sexta can therefore retain a higher spatial and temporal acuity than the smaller D. elpenor, which needs to rely more heavily on neural summation to achieve the required visual sensitivity.

(f). The advantages of central ecological adaptations in the visual system

Irrespective of their diel activity, all three hawkmoth species showed some contribution of central processing to the performance of their wide-field motion vision systems. What are the advantages of tuning the visual system at this higher level of processing, compared to tuning at peripheral levels? The likely answer is that higher-order filtering can extend the dynamic range of visual tuning to a degree that might not be possible in the periphery. In the case of the hawkmoth species investigated here, spatial summation makes it possible to exploit maximum levels of spatial acuity in bright light, and to reduce resolution to increase sensitivity in dim light (figure 4).

Moreover, central adaptations can serve the specific needs of the visual sub-system they are implemented in. In this study, this is the wide-field motion system, but this is only one of many parallel visual pathways that insects possess, all based on information from the same set of photoreceptors in the retina [41]. Other pathways, such as those involved in target detection, pattern recognition, and collision avoidance may experience different constraints on the spatio-temporal resolution of their visual information. The perception of ego-motion is not necessarily dependent on very high temporal acuity. On the contrary, in hovering species such as hawkmoths [13,14] the detection of low image speeds only requires modest temporal resolution. Other visual pathways, however, might benefit from higher spatial or temporal resolution, such as those that are responsible for the recognition of patterns on flowers [42], or the detection of predators via a looming response [43]. Thus, neural summation can adjust the balance between sensitivity and spatio-temporal acuity, thereby matching it to the requirements of each visual pathway.

3. Conclusion

We have shown that both peripheral and central adaptations shape the spatial and temporal characteristics of the wide-field motion pathway in hawkmoths to match the performance of this pathway to the ecological requirements of different species. The relative contribution of peripheral and central adaptations reflects the anatomical constraints of body size in each species, with additional central filtering boosting sensitivity when peripheral adaptations alone are not sufficient. In future investigations of how sensory systems are adapted to animal ecology, we suggest that a complete picture can only be obtained when central adaptations in the brain are also considered, in addition to those occurring peripherally.

Ethics

All experiments were carried out according to the Swedish national guidelines for work with insects.

Data accessibility

The datasets supporting this article have been provided as part of the electronic supplementary material.

Authors' contributions

Conceptualization, methodology, and writing—review and editing: A.S., E.W., and D.O'C.; investigation, visualization, formal analysis, and writing—original draft: A.S.; funding acquisition, resources, and supervision: E.W. and D.O'C.

Competing interests

The authors have no competing interests.

Funding

This research was supported under the Swedish Research Council (VR 621-2012-2205), the Australian Research Council's Discovery Projects funding scheme (project number DP130104561), and the Swedish Foundation for International Cooperation in Research and Higher Education (STINT 2012-2033).