Background

Insect compound eyes exhibit remarkable variation in size as a result of differences in the number and diameter of the individual subunits, known as ommatidia. For example, the silverfish Tricholepidion gertschi has ~ 40 ommatidia [1] whereas dragon- and damselflies (Odonata) sport up to 30,000 ommatidia [2]. Differences in eye size as well as the number, size, and angles between facets allow different visual behaviours, lifestyles, and adaptation to a large range of environments [3,4,5,6,7,8,9,10,11,12,13]. How these diverse eyes of insects have evolved and to what extent even small changes in the optics affect vision is still not well understood. Investigating and comparing natural variation in eye size and composition, and its impact on optical capacity within and between closely related species can provide valuable insight into the functional evolution of the insect eye. Generally, wider ommatidia can harvest more light, allowing greater sensitivity, while more ommatidia and narrower interommatidial angles (IOs) can provide greater acuity [4, 14, 15]. Ommatidia diameter and number therefore represent a trade-off that is optimised for the specific visual needs of each insect species, strain, sex, or morph.

Several studies have reported the extensive variation in ommatidia number, ommatidia diameter, and overall eye size within and between species of Drosophila [16,17,18,19,20,21]. There is evidence that some of this is the result of trade-off between eye size and antennal size and, by extension, the visual and olfactory systems, as well as overall head capsule size [19, 20, 22]. The genetic basis of these differences in eye size is complex but, in some cases, the underlying genes and developmental mechanisms have been characterised [17, 20, 23,24,25,26]. We previously found that one D. mauritiana strain has larger eyes than its sibling species D. simulans as a result of wider ommatidia, potentially caused by differential expression of the transcription factor Orthodenticle (Otd) during eye development [17, 21, 26]. The larger eyes of D. mauritiana with respect to D. simulans are also associated with reciprocal changes in the distance between the eyes (face width), but the antennae were not examined [17].

Optical parameters can also vary within eyes: killer flies, including Coenosia attenuata, have evolved specialised wide frontal ommatidia with small IOs for diurnal aerial hunting [8], and males of many families of dipterans have enlarged ommatidia in the dorso-frontal region of the eye that allows them to detect and pursue females in flight [27]. D. mauritiana and D. simulans also exhibit structural variation across the eye, with significantly wider anterior frontal ommatidia than central and posterior ommatidia in both males and females [26].

However, to fully understand the functional impact of eye size variation within and between Drosophila species, it is crucial to test predictions based on eye morphology in vivo. Here we further explored the variation in ommatidia number and diameter that contribute to eye size differences in males and females among different D. mauritiana and D. simulans strains. We used synchrotron radiation microtomography to obtain 3D information of optical parameters in focal strains of these two species and tested predicted differences in their vision via optomotor responses in a virtual reality flight simulator in vivo.

Results

Ommatidia number and size vary within and between closely related Drosophila species

The overall eye size of compound eyes is determined by ommatidia number and size (here reflected by facet area). While D. mauritiana generally have larger eyes than the closely related species D. simulans [16, 21, 28], it remains unclear if this difference is always caused by one or both parameters. We analysed total eye area, central facet area, and total ommatidia number from scanning electron microscopy images in multiple strains of both species (Additional File 1: Table S1) and found a negative correlation between central ommatidia facet size and number in D. simulans (females R =  − 0.38 p = 0.00021; males R =  − 0.3 p = 0.0036) suggesting a potential trade-off between these characteristics (Fig. 1, Additional File 2: Fig. S1). In contrast, D. mauritiana had generally wider and more numerous ommatidia and consequently overall larger eyes than D. simulans and the trade-off seen in D. simulans was absent in females. Interestingly, D. mauritiana males showed a positive correlation between ommatidia number and facet size (R = 0.34, p = 0.0087; Fig. 2, Additional File 2: Fig. S1).

Fig. 1
figure 1

Variation in eye size, ommatidia number, and ommatidia size across closely related D. mauritiana and D. simulans. Average eye size (mm2, circle area) of D. simulans (blue) and D. mauritiana (red) strains (circle labels) plotted against total ommatidia number and ommatidia facet area (in μm.2) for females (a) and males (b). D. mauritiana generally have larger eyes with more and larger ommatidia compared to D. simulans. Eye size was measured from side view scanning electron micrographs of single eyes. n = 11 for MS17 females and n = 15 for males and females of all other strains. Measurements provided in Fig. 1 morphological measurements.xlsx on figshare [29]

Full size image
Fig. 2
figure 2

3D analysis of ommatidia size and shape in male and female D. simulans M3 and D. mauritiana RED3. a Synchrotron radiation microtomography analyses of males and females show a gradient from smaller to larger ommatidia from dorsal to anterior-ventral. D. simulans M3 females and especially males show an overall shift to smaller ommatidia compared to D. mauritiana RED3. b Analysis of ommatidia numbers show similar ommatidia numbers in females of both species and fewer ommatidia in males in line with overall smaller eye size. c 3D reconstruction of the optic lobe (lamina, medulla, lobula, and lobula plate) of a male D. mauritiana RED3. d Volume analysis of optic lobe neuropil sizes show a similar pattern to ommatidia number differences (b) between sexes and strains: males of D. simulans M3 and D. mauritiana RED3 have generally smaller neuropils, most evident in lamina and medulla (n (a–d) = 3). e, f Shape analysis of frontal (yellow) and central (green) ommatidia reveals separate clustering of frontal and central ommatidia for both species. n = 30. Data points are provided in Fig. 2 synchrotron analysis.xlsx and Fig. 2 morphometric analysis.xlsx on figshare [29]

Full size image

To test whether larger eyes of D. mauritiana were an effect of overall larger body size, we also measured second-leg tibia length, which have been previously used as a proxy for body size [16, 30, 31] and the length of the L3 wing vein as an estimate of overall wing size [32]. The tibia of D. mauritiana were not generally larger than the tibia of D. simulans (Additional File 3: Fig. S2), suggesting the increase in eye size has evolved independently of body size [16, 17, 28]. Consistent with this, tibia size was only positively correlated with eye size in a subset of strains in both species (Additional File 4: Fig. S3). Interestingly, wing size is generally smaller in D. mauritiana strains, and we found strain-specific positive, negative, or no correlation with eye size (Additional File 5: Fig. S4).

While some D. simulans and D. mauritiana strains overlap in either ommatidia area or number, none of the strains overlapped in both parameters, leading to the clear separation of the species in eye composition (Fig. 1). Previously, a large-effect quantitative trait locus has been identified that explains about 30% of the eye size difference between D. simulans and D. mauritiana [17, 26] due to differences in ommatidia area. However, the functional consequences for vision in these flies remain unknown. To test this, we selected strains D. simulans M3 and D. mauritiana RED3 which have very similar ommatidia numbers but significantly different mean ommatidia (facet) areas (Fig. 1). We first performed detailed 2D and 3D morphological analysis of optical parameters of these focal strains to model their vision, and subsequently tested our predictions with behavioural experiments.

Facet size and shape change in a dorsal to ventral-anterior gradient across species and sexes

Drosophila compound eyes are 3D structures that are roughly shaped like a hemisphere. To analyse optical parameters across the entire eye, we used synchrotron radiation microtomography to collect high-resolution 3D image data of entire eyes and associated brain structures in D. simulans M3 and D. mauritiana RED3 (Fig. 2). Automated segmentation and measurement of individual facets for three individuals of each species and sex revealed a size gradient from smaller dorsal-posterior to larger anterior-ventral facets in both focal strains and both sexes (Fig. 2a). Facet size was overall smaller in D. simulans M3 and the size difference between females and males was more pronounced in D. simulans M3 compared to D. mauritiana RED3 (Fig. 2a). Comparison of ommatidia number differences and neuropil volumes indicates similar patterns between sexes and strains. Lamina and medulla in males are generally smaller in line with lower ommatidia numbers (Fig. 2b–d).

Additionally, we used geometric morphometric analysis of facet shapes to compare central ommatidia to frontal ommatidia in both sexes of D. simulans M3 and D. mauritiana RED3 (Fig. 2e, f): the six corners of each facet were landmarked and analysed via principal component and hierarchical clustering analysis. We recovered three clusters, which can be interpreted as three distinct facet shapes. Clusters 1 and 2 contained only frontal lenses, and cluster 3 contained only central lenses indicating that the position of the facet within the eye influences lens shape. Frontal lenses (clusters 1 + 2) were defined by longer dorsal and ventral edges (PC1 = 87.3% variation) than central lenses (cluster 3). Within the frontal lenses, PC2 (4.5% variation) and PC9 (< 1% variation) separated clusters 1 and 2, with cluster 1 being slightly elongated along the antero-posterior axis. There were no differences in sex (chi-square = 0.07, df = 2, p = 0.967) or strain (chi-square = 2.74, df = 2, p = 0.254) between clusters, implying that these factors do not influence facet shape.

D. mauritiana RED3 has greater optical sensitivity than D. simulans M3, especially in the frontal and ventral visual field

To compare the optical capacity of both fly strains and their variation across the visual field, we implemented the open-source Python-based automated pipeline ODA [33], which estimates the location and approximate orientation of each lens with high resolution across the eye. This generated eye maps of the volume (Additional File 6: Fig S5a), diameter (Additional File 6: Fig S5b), cross-sectional area (Additional File 6: Fig S5c), and length (Additional File 7: Fig S6a) of the corneal lenses of the ommatidia and the mean IO angle of each lens with its nearest neighbours (Additional File 7: Fig S6b). Three male and three female eyes were scanned from both D. simulans M3 and D. mauritiana RED3. The coordinates were rotated manually to align the eye equators horizontally, visible as a horizontal band of smaller ommatidia in Fig. 3a (and Additional File 7: Fig. S6a, b, and c). This area projects roughly onto the visual horizon during flight [34] and marks the region of the eye where rows of ommatidia initiated and grew during eye development, establishing a line of mirror symmetry about which rhabdomere arrangements flip vertically [35]. To compare the change in these parameters from the posterior to the anterior eye, we used ordinary least squares to fit an affine function of azimuth and compared the resulting slope parameters for each subject.

Fig. 3
figure 3

3D analysis of optic parameters in D. simulans and D. mauritiana eyes. a–d Ommatidial lens volumes from 3 males and 3 females from each of the two species. Eye maps show the smallest and largest eye of each of the two species, which also happen to be one male and female of each. Each dot of the scatter plot represents the location of an individual ommatidium in polar coordinates coloured by its 3D volume according to the colour scale indicated in the x- and y-axes of b and c. Line colours in b and c and dot colours in d indicate the fly’s rank in order of eye size per species, such that the darkest one is the largest eye of that species. The volume data is divided into 20 evenly spaced bins of elevation (b) and azimuth (c) with error bars indicating 3 times the standard error of the mean. d Ordinary least squares were used to regress lens volume on azimuthal position to estimate and compare the azimuthal slope of lens volume between the two species. The resulting slope coefficients from those models are plotted. e–h IO angle from three males and females from each of the two species, plotted as in a–d except for the elevation plot in f. The IO angle value used for each lens represents the average IO angle between that lens and all immediate neighbours. f The same IO angle data from e but sampling ommatidia from a narrow vertical band between 0 ± 15° azimuth. Note that this is different from b, c, and g because plotting the binned averages obfuscates the horizontal band of high acuity along the equator, likely due to the large range of IO angles along azimuth. h Ordinary least squares was used to regress IO angle on azimuthal position as in d. All the eyes demonstrated negative azimuthal slopes, with no significant difference between species. i–m Scatterplots of total lens count (i), mean lens diameter (j), median IO angle (k), median equatorial IO angle (l), and IO angle interquartile range (m) plotted along the y-axes and their allometric relationship to the surface area of their eye along the x-axis. Lines in the 2D scatter plots represent the predicted mean and the bands represent the 95% CI of the mean based on ordinary least squares regression of each y variable on surface area. Note that simple group differences based on ANOVA are indicated in the left margins and group differences after accounting for surface area using linear regression are indicated at the top of each scatterplot with the following key: * = p < .05, ** = p < .01, and *** = p < .001. Data provided in Fig. 3_share.zip on figshare [29]

Full size image

Both D. simulans M3 and D. mauritiana RED3 eyes have the largest lenses in the frontal visual field just below the eye equator (Fig. 3a). For both species, lens volume increases with elevation, peaks just below the eye equator, and then decreases steadily (Fig. 3b). In D. mauritiana RED3 this increase is greater, starting at similar volumes at the dorsal and ventral extremes but increasing to larger maxima near the equator than D. simulans M3. Lens volume for both species decreases along elevation until a minimum around − 45° and then increases, peaking at the anterior extreme (Fig. 3c). Moreover, in 5 of the 6 size-ordered pairs, D. mauritiana RED3 have significantly greater lens volumes than D. simulans M3 for every azimuthal bin (Additional File 6: Fig S5a). Lens volume for all eyes has a positive azimuthal slope, but the slope for D. mauritiana RED3 was significantly greater than D. simulans M3 (t(10) = 2.3, d = 1.5, p = 0.043). This is consistent with measurements of lens diameter (Additional File 6: Fig S5b), cross-sectional area (Additional File 6: Fig S5c), and length (Additional File 7: Fig S6a), except that the azimuthal slope only differed for lens diameter (t(10) = 2.3, d = 1.4, p = 0.047) but not for either cross-sectional area (t(10) = 1.4, d = 0.9, p = 0.18) or length (t(10) = 1.9, d = 1.2, p = 0.08). Overall, this means that D. mauritiana RED3 have larger, broader, longer, and wider-spread ommatidial lenses than D. simulans M3, which could improve sensitivity in general, and especially in the frontal visual field below the eye equator. This increase in ventral optical sensitivity is also greater in D. mauritiana RED3 than D. simulans M3 and is predicted to improve the detection of low-contrast objects below the visual horizon, such as rotting fruit or other oviposition sites.

D. mauritiana RED3 and D. simulans M3 have higher spatial acuity along the eye equator

IO angles are largest at the posterior and peripheral extremes, reaching a minimum around 45° azimuth along the eye equator (Fig. 3e, f). For both species, IO angle stays relatively constant—remaining between 4° and 6° from about − 45° to 45° elevation—except for dramatic increases at the ventral and dorsal extremes and a region of smaller IO angles around the eye equator (Fig. 3f). D. mauritiana RED3 ranges less in IO angle than D. simulans M3 reaching smaller maxima in the top and bottom of the eye (≤ 15° versus ≤ 25°). For both species, IO angle decreases along azimuth from a maximum in the posterior extreme (≤ 15°) to a minimum around 45° azimuth (≥ 4°; Fig. 3g). We found no significant difference between species in the azimuthal profile or slope (Fig. 3g, h). Because spatial resolution is limited inversely by IO angle, maximum spatial resolution in both species is highest around 45° azimuth and 0° elevation, along the eye equator. This increase in equatorial spatial resolution might be an adaptation to terrain statistics of different habitats [33, 36], and due to the horizontal band of smaller ommatidial diameters at the eye equator formed during eye development [37, 38]. Regardless, this would improve the resolution of small objects near visual horizon, a feature that would help in avoiding predators and locating oviposition sites.

Eye allometry in D. mauritiana RED3 prioritises contrast sensitivity more than D. simulans M3

In holometabolous insects, body size and the size of organs derived from imaginal discs depend on, and are proportional to, environmental factors like temperature and food availability during larval development [39, 40]. In flies, larval feeding has been shown to affect eye size, ommatidia size, and ommatidia count [41]. As a result, variation in eye size and composition may reflect rearing differences. To address this, we modelled the scaling relationships between eye surface area and the following measurements: total lens count, mean lens diameter, median IO angle, median equatorial (elevation = 0 ± 15°) IO angle, and IO angle interquartile range (Fig. 3i–m).

Eye surface area (SA) is an ideal reference for allometric scaling because it is proportional to the rate of light absorption of the entire eye. Also, because the ommatidial lenses almost completely cover the surface of the eye, the mean lens area (A) is approximately SA divided by the number of lenses (N), A ≈ SA / N, implying that SA ≈ N × A. This equation is approximate because we estimated A by assuming circular facets even though facet shapes vary. Because the number of discernible brightness levels is proportional to lens area, SA is also proportional to the total number of images the eye can resolve, its spatial information capacity [42]. Using ordinary least squares, we regressed each measurement on the sum of eye area and a dummy-coded species variable (Additional File 8: Table S2). We performed post hoc pairwise t-tests to compare means between species, defining the interspecific difference as D. mauritiana RED3–D. simulans M3, such that significant positive values mean that D. mauritania RED3 was greater than D. simulans M3 and vice versa for negative values. All models were a good fit, explaining a substantial proportion of the variance in SA plus the species variable (R2 = 0.65–0.97, F = 8–159, P ≤ 0.01).

Lens count and size had significant positive slope coefficients, such that larger flies have more and larger ommatidia in both species. However, lens count and lens diameter had significant interspecific differences after accounting for SA, but lens count was greater for D. simulans M3 and lens diameter was greater for D. mauritiana RED3. Therefore, D. mauritiana RED3 have lower ommatidial density than D. simulans M3. Conversely, the interquartile range (IQR) of lens diameters has a significant negative slope and a significant positive interspecies difference, implying that D. mauritiana RED3 lens diameters are more variable than D. simulans M3 after accounting for eye size. This is consistent with the lens volume eye maps discussed above, which found a greater range of lens sizes in D. mauritiana RED3 than D. simulans M3 along elevation, generally larger ommatidia for every azimuthal bin, and a greater azimuthal slope.

For both general and equatorial IO angles and comparison across both species, the slope coefficient was significant and negative, meaning that median angles scale inversely with eye size. However, the interspecific difference was only significant for median equatorial IO angles, such that D. mauritiana RED3 has significantly greater equatorial IO angles than D. simulans M3 after accounting for SA. Because spatial resolution is inversely proportional to IO angle, D. simulans M3 has greater spatial resolution at the eye equator but similar resolution elsewhere. The IQR of IO angles had an insignificant slope coefficient and a significant but negative interspecific difference, meaning that D. simulans M3 have a greater range of IO angles. This is consistent with the IO angle eye maps above, which found a greater range in the elevation profiles of IO angle in D. simulans M3 (Fig. 3f). The increase in IO angles near the boundaries of the eye should effectively increase the field of view (FOV) of the eye. Overall, these allometric relations suggest that D. mauritiana RED3 prioritise optical sensitivity more than D. simulans M3, which instead prioritise spatial resolution along the visual horizon and FOV at the peripheral extremes.

D. mauritiana RED3 and D. simulans M3 optomotor responses trade off contrast sensitivity and spatiotemporal resolution

Our morphological analysis suggested that D. simulans M3 have higher spatial acuity due to smaller IO angles for equatorial ommatidia and D. mauritiana RED3 have higher optical sensitivity due to larger facet apertures, particularly in the central visual field just below the horizon. However, neural summation can recover sensitivity loss due to suboptimal optics by effectively sacrificing temporal or spatial resolution [41]. To measure the ethological implications of these optical differences, we used the flies’ optomotor response in a virtual reality flight simulator that allowed the presentation of different sinusoidal gratings moving to the left or right (Additional File 9: Fig. S7). Using a wingbeat analyser, we measured the flies’ steering effort in response to gratings of various contrasts, spatial frequencies, and temporal frequencies sorted randomly. Contrast sensitivity is defined here as the reciprocal of the lowest discernible contrast, and both spatial and temporal acuity are defined by the maximum discernible frequency. Assuming that the IO angle limits the maximal spatial sampling or resolution of the eye according to the Nyquist limit, such that the highest possible discernible frequency, fs, for a hexagonal lattice is given by the following equation: fs = 1/√3 * Δɸ−1. So, for every fs, there is a corresponding ideal IO angle, Δɸ = 1/√3 * fs −1.

In the flight arena, D. simulans M3 and D. mauritiana RED3 traded off between higher contrast sensitivity and spatiotemporal tuning (Fig. 4). In accord with their larger ommatidia, D. mauritiana RED3 demonstrated higher contrast sensitivity (0.14−1 = 7.4) than M3 (0.27−1 = 3.7). Conversely, the spatial tuning curves demonstrate that D. simulans M3 has a higher spatial acuity (0.1 cpd) than D. mauritiana RED3 (0.08 cpd), implying smaller IO angles (~ 5.8° versus ~ 7.2°) consistent with smaller measured IO angles in the eye equator of D. simulans M3. D. simulans M3 also responded with greater strength around 0.04 cpd, likely supported by their wider peripheral IO angles and greater IQR. Lastly, D. simulans M3 demonstrated higher temporal acuity, 50 Hz, than D. mauritiana RED3, 20 Hz. Overall, this demonstrates sharper (higher spatial acuity) and faster vision (higher temporal acuity) in D. simulans M3 but a greater ability to compare brightness values (higher contrast sensitivity) in D. mauritiana RED3.

Fig. 4
figure 4

Behavioural measurement of D. simulans and D. mauritiana contrast sensitivity, spatial resolution, and temporal resolution. Gratings of various contrasts (a–c), spatial frequencies (d–f), and temporal frequencies (g–i) were presented to 3 males and 3 females from each of the two species in a rigid tether flight simulator equipped with a wingbeat analyser. The gratings were filtered through a Gaussian window and remained still for .2 s before moving to the left or right, indicated by the dotted line. For each subject, responses to leftward moving gratings were averaged with responses to the same grating moving rightward so that positive values represent mean steering in the direction of the grating (red or blue) and negative represents counter steering (grey). Mean normalised responses taken between .5 and 1.25 s were baseline corrected, subtracting the mean response during the .1 s before the onset of motion. Two of these ranges are indicated by annotations in a and b. connected by dashed arrows to their mean in c. Sample sizes are indicated in the bottom left corner of the colourmaps. The images of gratings in the bottom of c, f, and i are meant to give a sense of the change in the stimulus along the x-axis. Green arrows indicate the change in speed of the grating, ft/fs, which remains constant in the contrast experiment, decreases in the spatial frequency experiment, and increases in the temporal frequency one. a–c As contrast increases, RED3 begins responding significantly at .14 (red arrow in c) and M3 at .27 (blue arrow in c). d–f As spatial frequency increases and therefore rotational velocity decreases, mean responses decrease gradually until the Nyquist limit determined theoretically by the IO angle, reducing the contrast for higher frequencies as a result of aliasing. This limit differed between the two species, with RED3 responding significantly to spatial frequencies as high as .08 CPD (red arrow in f) and M3 as high as .1 CPD (blue arrow in f). g–i As temporal frequency and therefore rotational velocity increases, mean responses increase until they reach the Nyquist limit determined by the temporal resolution of the optomotor response, reducing the contrast for higher frequencies. M3 demonstrated higher temporal acuity, responding significantly to frequencies as high as 50 Hz (blue arrow) while RED3 stopped at 20 Hz (blue arrow). Data provided in Fig. 4_share.zip on figshare [29]

Full size image

Discussion

The evolution of variation in overall eye size, ommatidia number and facet area has been shown within and between Drosophila species by several research groups and differences in vision have been proposed based on these optical parameters [16,17,18,19,20,21]. D. mauritiana has evolved generally larger eyes composed of wider ommatidia than its sibling species D. simulans and D. sechellia since their divergence approximately 240,000 years ago [16, 17, 21, 26, 43, 44]. The eye size of their common ancestor was likely similar to D. simulans because their sister species D. melanogaster has more similar sized eyes to this species rather than D. mauritiana [21]. In this study, the demonstrated allometries and regional specialisations of D. mauritiana and D. simulans were found to differ in quantity but not quality: (1) maximum sensitivity in the central visual field below the horizon, with very similar elevation and azimuthal profiles; (2) maximum acuity along the visual horizon of the eye; and (3) improvements in optical sensitivity and spatial resolution for larger conspecifics. Future investigations of the developmental origins of these gradients and regional specialisations in spatial resolution and optical sensitivity and how they may differ between flies will aid our understanding of how the astonishing diversity in insect eyes has evolved.

So far, our knowledge of how insects developmentally and evolutionarily balance the trade-off between ommatidia number (resolution) and ommatidia size (sensitivity) is very limited. The genetic basis of evolutionary differences in eye size has been difficult to determine, partly because ommatidia size and number seem to be genetically uncoupled and differences in these features polygenic. While differences in ommatidia size between D. mauritiana and D. simulans have been mapped to orthodenticle [17, 26], and a cis-regulatory region of eyeless has been shown to contribute to differences in eye size within D. melanogaster and between this species and D. pseudoobscura [20], these changes do not explain the full extent of variation. Other genes involved in the regulation of cell proliferation and differentiation in developing eye imaginal discs are the most likely candidates to contribute to the diversification of eye size. For example, phosphoinositides including the Drosophila class I(A) PI 3-kinase Dp110 and its adaptor p60, the gap-junction protein inx2 and the 40 s ribosomal protein S6 kinases, have all been shown to alter ommatidia number and/or size [45,46,47,48].

Very little comparative functional data is available to truly understand the impact of natural variation in eye structure on vision. Here we modelled and tested optical capacity in two Drosophila species—D. mauritiana and D. simulans—between two strains that had similar ommatidia number but significantly different ommatidia facet sizes, to assess whether predicted differences in contrast sensitivity, spatial resolution, and temporal resolution could be observed in behavioural experiments. In principle, larger facets could evolve to provide either better sensitivity (collecting more light in a similar amount of time) or better temporal resolution (collecting similar amounts of light over shorter times), or some combination. Indeed, we confirmed higher spatial and temporal acuity in D. simulans with smaller ommatidia and improved contrast sensitivity in M3, with larger ommatidia. Whether and how these differences reflect meaningful adaptations to ecological differences remains to be explored, but the recapitulation of morphological divergence through behavioural paradigms is compelling.

We also identified substantial intraocular variation in lens volume, interommatidial spherical angles, facet shape, and lens diameter, as has been reported in other dipterans. While the optomotor experiments reported here tested global responses, future behavioural experiments might target different parts of the visual field to see whether this regional variation has a functional significance. Potentially increased spatial resolution at the equator of the eye, and in the anterior-ventral FOV, was predicted from our morphological data. Increased acuity at the horizon, combined with horizontally narrower facets at the centre of the eye, for example might enhance the detection of lateral optic flow. Likewise, the stronger anterior frontal gradient in predicted resolution for females could have implications for the detection of oviposition sites.

While our analysis supports the use of 3D morphological data to predict optical capacity, many other factors are involved in information acquisition and processing in the insect eye that are not as easily accessible. Recent discovery of smooth and saccadic retinal muscle movement to improve perception of moving and stationary objects respectively [49, 50] as well as hyperacute vision via photomechanical photoreceptor contractions (microsaccades) [51,52,53,54] have revealed much more sophisticated mechanisms are employed in Drosophila eyes to sample visual information. In particular, the spatial resolution of compound eyes can exceed the spatial Nyquist limit set by the IO angle due to brief, stereotyped photomechanical contractions (microsaccades) that sharpen and shift rhabdomere receptive fields, affording so-called hyperacuity [52]. These contractions are optimal for processing brief bursts of light followed by periods of darkness to better match the refractory phase of rhabdomere microvilli [52] and generally match the optical flow of forward translation [54]. As a result, these phases of improved acuity apply to specific combinations of motion direction, duration, speed, and visual field region. Moreover, the advantages and magnitude of the photomechanical rhabdomere contractions are limited by IO angle [53], so that the difference in IO angles measured here still confer an important difference in visual capacity.

The complexity of the visual system overall, incorporating mechanisms of neural summation and hyperacuity, further highlights the importance of using behavioural measurements of acuity and sensitivity and reinforces the conceptual distinction between optical and contrast sensitivity. Neural summation could have reversed these differences as it did for D. mojavensis due to darkness adaptation [55] or facultatively within D. melanogaster individuals in response to forward optical flow [56]. An assessment of the optics alone would have ignored the difference in temporal acuity and overestimated the difference in contrast sensitivity between D. mauritiana and D. simulans based on differences in optical sensitivity.

Aside from the functional aspect, the maintenance of eyes and the underlying complex neurocircuits are a metabolically expensive investment [57]. For example, comparison between photoreceptor information rates of larger and more active flies like the blowfly Calliphora with the smaller D. melanogaster showed a five times higher performance in Calliphora but at a ten times higher energetic cost [58]. The evolution of overall larger eyes with more and wider ommatidia and resulting increase in contrast sensitivity in D. mauritiana must therefore represent an economically viable investment aligned to their specific optical needs. The balance between sensory system requirements and energy efficiency has been observed in other fly species: The male housefly (Musca domestica) has a 60% higher bandwidth (measure of speed of response) in their contrast-coding R1-6 compared to females, allowing them to track these females in flight, whereas bandwidth decreases in male blowflies by 20% towards the back of the retina [59, 60]. It is therefore conceivable that absolute eye size is under stronger selection than ommatidia number or ommatidia size on their own, and at least to some extent independent of body size and other functional traits [61,62,63]. Evidence from Drosophila wings suggests that compensatory mechanisms guarantee a certain wing size if overall size deviates too much [64, 65]. A similar mechanism could be at play in D. simulans where ommatidia number and size seem to be coordinated to maintain similar eye size across strains.

Conclusions

Insects play vital roles in various ecosystems, including pollination and decomposition. Climate change and the disappearance of ecological niches around the world highlights the need to understand how they perceive and interact with their environment and vision is a primary sensory modality for many insects, shaping their behaviour, foraging strategies, and reproductive patterns. Our study demonstrates that even subtle differences in ommatidia size between closely related species can have a measurable effect on their vision. Therefore, comparative studies of natural variation in eye morphology and the consequences for vision across dipterans and beyond are needed to fully understand how the diversification of eye size, shape, and function allowed insects to adapt to the vast range of ecological niches around the world.

Methods

Fly strains and husbandry

Multiple strains of D. simulans and D. mauritiana were used in this study [21] (Additional File 1: Table S1). All stocks used were kept on standard yeast extract-sucrose medium at 25 °C under a 12:12-h dark/light cycle. For experiments, flies were reared at controlled, low density, achieved by transferring set numbers of males and females (typically between 10 and 20 of each sex) into fresh food containers to lay offspring. Adult offspring were removed soon after eclosion for experiments.

Scanning electron microscopy

Fly heads were prepared and imaged as previously described [21]. Briefly, heads were fixed in Bouin’s solution (Sigma-Aldrich) and dehydrated to 100% ethanol. For SEM imaging heads were critical point dried in a Tousimis 931.GL Critical Point Dryer and mounted onto sticky carbon tabs on SEM stubs, gold coated (10 nm) and imaged in a Hitachi S-3400N SEM with secondary electrons at 5 kV.

Morphological measurements

SEM images of eyes were analysed using FIJI/ImageJ [66]. For each strain, 15 males and 15 females (except MS17 females: n = 11) were analysed. Ommatidia number was counted manually by using multi-point tool for one compound eye per individual (from side views of compound eyes). Ommatidia size and overall eye area were measured manually with the polygon selection tool. Frontal and central ommatidia area were measured for each eye with the polygon selection tool. The area of six central ommatidia was average to determine mean ommatidia (facet) size. Wing and tibia of the second leg of each fly were dissected in 70% ethanol and mounted in Hoyer’s solution and cured overnight at 60 °C. Wings and tibia were imaged at × 5 (× 1.25) magnification using a Zeiss Axioplan microscope equipped with a ProgRes MF cool camera (Jenaoptik). Wing and tibia size were measured using the line tool in Fiji/ImageJ. Plots and statistical analysis were carried out in RStudio Version 2023.03.0 + 386 using the Tidyverse suite of packages [67]. Where analysis required comparison between a length and an area, the length measurement was squared. Linear lines of fit were added to plots using geom_line(stat = "smooth", method = lm). For correlation analysis, data was first checked for normality using Shapiro–Wilk and then tested using either Pearson’s correlation coefficient or Spearman’s rank correlation coefficient using stat_cor(method = "pearson") or stat_cor(method = "spearman"). Standard deviations for the raw measurements were calculated in Excel.

Synchrotron radiation tomography

Fly heads were prepared as described for SEM to 100% ethanol, then stained with 1% iodine and washed in ethanol. Fly heads were mounted in 20-µl pipette tips filled with 100% ethanol for synchrotron radiation X-ray tomography and scanned at the TOMCAT beamline of the Swiss Light Source (Paul Scherrer Institute, Switzerland [68] and Diamond-Manchester Imaging Branchline I13-2 (Diamond Light Source, UK) [69, 70] as previously described [21, 26].

3D segmentation

The IMOD Software package [71] was used to generate cropped mrc stacks for 3D segmentation and analysis of the head tissue, lenses, and optic lobes in Amira v.2019.2 (Thermo Fisher Scientific). Ommatidial lenses were segmented through threshold and separate objects tools. Lens sizes were analysed and colour-coded depending on size with the label analysis and sieve module.

Morphometric analysis

Heads were tilted in the SEM to obtain flat images of frontal and central ommatidia for geometric morphometric analysis. The six corners of each facet were landmarked using the digitize2D function in the R package Geomorph (v.4.0.4) [72, 73]. Data were registered and Procrustes transformed using procSym function in the package Morpho (v.2.10) [74] to account for reflection, before principal component analysis using the Geomorph package. Hierarchical clustering was performed and visualised using the factomineR (v.2.6) and factoextra (v.1.0.7) [75, 76] packages, using the option nb.clusters = -1 to select the optimal number of clusters. The strain, sex, and positional identities of the resulting clusters were analysed by chi-square in base R, and the contribution of the principal components to clustering was extracted from desc.var generated by the HCPC function for clustering. Plots were generated using ggplot2 (v.3.4.2) [77] n = 30.

ODA and allometry

To approximate the optical performance of the two species, we processed CT stacks of six flies (three males and three females) from the RED3 strain of D. mauritiana and the M3 strain of D. simulans. This allowed us to apply the 3D ommatidia detecting algorithm (ODA-3D; Additional File 9: Figure Fig. S7a), a pipeline for automatically measuring a number of visual parameters for compound eyes [33]. Each dataset was manually cleaned to generate binary images of only the corneal lenses. Then, the programme fitted a cross-sectional surface through the coordinates of the lens cluster and projected these coordinates onto the cross-section, allowing a custom clustering algorithm to find ommatidia-like objects in the 2D projected images. Finally, the volume, diameter, cross-sectional area, length, and average IO angle of each lens were measured. Eye surface area was estimated as the sum of the lens areas based on the ODA-derived lens diameters. Allometric scaling relations were derived by regressing each of the measured visual parameters on eye surface area plus a dummy-coded species variable. Post hoc pairwise t-tests were used to compare means between species. The interspecies difference was defined as D. mauritiana RED3–D. simulans M3, such that a significant positive difference implies that D. mauritiana RED3 was greater than D. simulans M3 after effectively accounting for differences in eye surface area (SA). The resultant parameters of these models are found in Additional File 8: Table S2.

Flight arena

To measure the optomotor performance of D. mauritiana RED3 and D. simulans M3, we performed psychophysics in a rigid tether flight simulator equipped with a wingbeat analyser (Additional File 9: Fig. S7b). Flies from 3 to 6 days post-eclosion were cold-anesthetised for about 5 to 30 min, glued to a 2-mm-diameter tungsten rod, and left to recover for about an hour with a piece of paper placed on their feet to prevent wing beating. They were then centred within an acrylic cube lined with rear-projection material (with 1/6 of the panels left open), immersing them in the projection surrounding 5/6 of their FOV (Additional File 9: Fig. S7c). Stimuli were generated by a computer using a custom open-source graphics library and projected onto the front panel of the arena at 120 Hz by a high-speed projector (technical information can be found in Currea, Smith and Theobald [41] and Currea et al. [55]. An IR light cast the shadow of each wing onto photodiodes below the fly designed to output the amplitude of each wingbeat shadow as a 1000-Hz voltage signal. The difference between the left and right wingbeat amplitudes (ΔWBA) is proportional to yaw torque and indicates the fly’s steering effort. For instance, in Additional File 9: Fig. S7d, we plot the ΔWBA time series for an exemplary fly in response to 9 gratings of different contrast (corresponding to the line’s saturation) moving to the left or right (warm vs. cool hue). Note that the strength of the response is affected by contrast while the direction corresponds generally to the direction of motion. These responses were taken from Currea et al. [55], which used the same methods.

Psychophysics

In the flight arena, flies viewed sinusoidal moving gratings of various Michelson contrasts (0, 0.01, 0.02, 0.04, 0.08, 0.16, 0.32, 0.64; Fig. 4a–c, with examples at the bottom of c), spatial frequencies (1.2, 1.4, 1.8, 2.3, 2.9, 3.8, 4.8, 5.7, 7.2, 9.6 cycles/°; Fig. 4d–f), and temporal frequencies (0.1, 0.2, 0.5, 1, 2, 5, 10, 20, 50, 100 Hz; Fig. 4g–i) to measure the functional consequences of their optical differences. The gratings were filtered through a Gaussian window and remained still for 0.2 s before moving to the left or right, indicated by the dotted lines in Fig. 4. For each subject, responses to leftward moving gratings were (1) averaged with responses to the same grating moving rightward, (2) baseline corrected, subtracting the mean response during the 0.1 s before the onset of motion, and (3) normalised to the maximum mean response per fly so that positive values represent mean steering in the direction of the grating, with a maximum of 1 (fully saturated red or blue) and negative represents countersteering (grey). These baseline-corrected normalised responses were averaged across each group to make the colormaps in Fig. 4. For each fly, an average of these normalised responses was taken from 0.5 to 1.25 s and used for plotting and comparing means in the bottom row of subplots in Fig. 4.

Bootstrapping was used to test for a grating’s discernibility by estimating the standard error of the mean and 90% C.I.s for the mean response. We bootstrapped the means taken between 0.5 and 1.25 s 10,000 times at the subject level to generate empirical sampling distributions of the mean for each parameter value accounting for repeated measures. The 68% C.I. of each distribution was used as an approximation of the standard error (error bars in the bottom row of Fig. 4) and the lower bound of the 90% C.I. was used to test for positive significance with a two-tailed alpha of 0.1 or one-tailed alpha of 0.05. Contrast sensitivity was defined as the reciprocal of the lowest discernible contrast and spatial and temporal acuity were defined by the highest discernible frequency.