The relationship between microhabitat use, allometry and functional variation in the eyes of Hawaiian Megalagrion damselflies

Study Species

Adult male individuals of 13 species of Megalagrion damselflies were collected from the wild using hand nets from the islands of Molokai, Maui, Oahu and Kauai (Table 1) from 2010 to October of 2012. All individuals within a species were collected from the same island. We focused on adult males in this study for two reasons. First, females are difficult to collect (often located only when in tandem with a male). Secondly, male odonates use visual signalling in establishing territories and finding mates (Corodoba-Aguilar 2008) which may result in a tighter link between visual characteristics and microhabitat. Only for M. koelense were adult females included in eye morphology data (the two males and two females did not show any differences). All morphological measurements were conducted within 3 days of capture.

Table 1. Megalagrion species means, standard errors and sample sizes (N) for body size (thorax length), eye size (eye height) and eye morphology variables: maximum facet diameter (), minimum interommatidial angle () and regionalization in facet diameter (regD), and interommatidial angle (regΔϕ). Regionalization variables are reported as fraction of the eye based on a species mean. Island and breeding microhabitats also given

Species	Body size (mm)	N	Eye size (mm)	N	m)	N		N	regD	regΔϕ	Island	Microhabitat
blackburni	6·25 ± 0·05	25	2·43 ± 0·03	20	32·7  ± 0·98	5	0·627 ± 0·044	5	0·32	0·36	Molokai	Stream
heterogamias	6·05 ± 0·05	26	2·31 ± 0·02	26	31·4  ± 0·85	5	0·734 ± 0·016	5	0·27	0·41	Kauai	Stream
oceanicum	5·54 ± 0·11	3	2·06 ± 0·06	3	31·5  ± 0·31	3	0·704 ± 0·048	3	0·26	0·34	Oahu	Stream
hawaiiense	4·82 ± 0·04	33	1·81 ± 0·01	29	28·8  ± 0·83	5	0·783 ± 0·043	6	0·25	0·28	Molokai	Seep
koelense	4·56 ± 0·09	7	1·77 ± 0·02	6	28·6  ± 0·50	4	0·782 ± 0·48	4	0·22	0·35	Molokai	Plant
vagabundum	4·49 ± 0·04	53	1·78 ± 0·01	48	33·8  ± 0·32	6	0·762 ± 0·055	6	0·28	0·27	Kauai	Seep
oahuense	4·46 ± 0·10	6	1·89 ± 0·04	6	36·7  ± 0·35	4	0·932 ± 0·030	4	0·26	0·37	Oahu	Terrestrial
n. nigrohamatum	4·39 ± 0·02	21	1·80 ± 0·01	20	35·1  ± 0·31	6	0·574 ± 0·022	6	0·14	0·26	Maui	Pool
calliphya	4·01 ± 0·03	69	1·54 ± 0·01	54	32·9  ± 0·90	6	0·765 ± 0·045	6	0·18	0·29	Molokai	Pool
n. nigrolineatum	3·92 ± 0·06	25	1·67 ± 0·01	25	34·2  ± 0·87	6	0·658 ± 0·030	6	0·16	0·27	Oahu	Pool
xanthomelas	3·89 ± 0·03	32	1·5 ± 0·016	32	34·2  ± 0·42	5	0·823 ± 0·036	5	0·17	0·28	Oahu	Pool
leptodemas	3·76 ± 0·06	6	1·48 ± 0·03	6	32·3  ± 0·35	6	0·798 ± 0·039	6	0·18	0·27	Oahu	Pool
oresitrophum	3·43 ± 0·06	34	1·50 ± 0·01	31	32·4  ± 0·49	6	0·830 ± 0·022	6	0·17	0·27	Kauai	Pool

Microhabitat Types

The five microhabitat types used in this study (stream, pool, seep, plant and terrestrial) are those recognized by field workers and taxonomists (Polhemus & Asquith 1996, Table 1). Males of this genus are generally territorial, guarding preferred oviposition sites at specific structural microhabitats where visual performance may be important. We use ‘microhabitat’ here to indicate small locations within a larger habitat that are preferred by a particular species. Microhabitats may define the ‘niche’ of species within a community, not only in terms of where it is found, but also correlating with strategies for feeding, locomotion or other aspects of functional biology (Elton 1927; Gause 1934; Hutchinson 1957; MacArthur 1958; Williams 1983).

‘Stream’ species are those that commonly use open areas in large, swift streams. ‘Pool’ species frequently occur along stream margins, pools and ponds. Pool species use water that is slower moving and often associated with vegetation. ‘Seep’ species are associated with the wet vertical surfaces of stream valleys and bedrock walls in deep forests, as well as bogs. The naiads of seep species are not found in streams, but under wet moss covering the walls with permanent water seepage. ‘Plant’ species breed in water pockets in the leaf axils of plants with a bromeliad growth form and are found in forests or bogs, typically away from streams. Finally, the single ‘terrestrial’ species breeds in the leaf litter of the native Uluhe fern and can be found in the forest, along trails and ridge tops (Williams 1939). Megalagrion oahuense is one of the very few damselfly species in the world known to be fully terrestrial. These five microhabitats are present on all islands and appear to be similar across islands however, terrestrial species are found only on Oahu.

Morphological Measurements

Gross morphology measurements were comprised of two measurements: Thorax length (THORAXL), the proxy for body size, was the distance from the junction of the second leg and the thorax to the base of the hindwing. Eye height (EYEHT), the proxy for eye size, was the diameter of the eye along its vertical axis and measured at the tallest part of the eye. All measurements were performed with ImageJ (Rasband 2012) on lateral photographs of individual damselflies including a size standard (taken with a Canon EOS 5D digital camera, Canon, Melville, NY, USA). Each morphological variable was checked for deviations from normality using qq-plots in the R computing environment (R Core Team 2013). We log-transformed body size and eye size to facilitate interpretation of scaling relationships. As the data were statistically well-behaved (met the assumptions for parametric statistical tests), the remaining variables were analysed on their natural scales (i.e. untransformed) for ease of interpretation.

Interommatidial Angle and Facet Diameter Measurements

We mapped variation in facet diameter and interommatidial angle across the eye as correlates of visual performance. Whereas spatial resolution is determined by interommatidial angle, several parameters contribute to light sensitivity (Land & Nilsson 2002) which can vary between species. Along with facet diameter, differences in focal length, rhabdom diameter and the ratio between the two may all play a significant role in determining light sensitivity. However, because the species included in our study are very closely related (separated by between 0·5–10 Myr Jordan, Simon & Polhemus 2003), it is reasonable to assume that a primary contributor to variation in light sensitivity will be differences in facet diameter (with the caveat that other traits contribute to light sensitivity and to the extent that they vary may also influence visual abilities), an assumption made by previous comparative studies of compound eyes (Jander & Jander 2002; Rutowski, Gislen & Warrant 2009).

We measured facet diameters (D) and interommatidial angles (Δϕ) using the pseudopupil technique (Horridge 1978; Stavenga 1979; Land 1997; Rutowski & Warrant 2002). Heads were detached from the body and placed on a goniometer stage under a dissecting microscope (Zeiss Stemi DV4 Spot, Zeiss, Jena, Germany). The eye was viewed from the same angle as the incident light using orthodromic illumination (i.e. with the light source emanating through the lens). The black area apparent from this view defines the ‘pseudo pupil’ (e.g. Fig. 1e). The ommatidia in line with the view absorb all incident light giving these ommatidia a black appearance. Heads were positioned face upward on the goniometer stage at the centre of rotation of one eye so that the pseudopupil moved vertically or medially across the eye. Anatomical references were set for the 0 longitude by centring both pupils of the eyes so that they were equidistant from the centre of the clypeus. The 0 latitude reference was set to the point where the centre of the pseudopupil was in line with the top of the clypeus with the face perpendicular to the scope (Fig. 1). The eye was dusted with chalk powder to allow identification of individual facets across photographs. We obtained a 180 dorsal–ventral transect by photographing the head at 10 steps from positive 90 (top eye) to negative 90 (bottom eye) using a Canon Powershot A650 IS digital camera mounted to the ocular of the microscope. Visual inspection of the eye revealed a single (Movie S1, Supporting information), frontally located acute zone. Thus, we focused on the front dorsoventral transect because this transect encompassed the greatest variation in the traits of interest (Fig. S1). Additionally, behavioural observations indicated that male Megalagrion damselflies orient themselves to maximize visual cues in the forward direction when perching or flying to explore objects (R. Schroeder & M. A. Butler, unpublished data), suggesting that the forward visual field is especially important in these damselflies.

Pseudopupil measurements began with importing the photographs into Image J and drawing a circle around the perimeter of the pseudopupil to identify its vertical centre in each photograph. As the eye is rotated, the pseudopupil moves across the eye. The average Δϕ was measured as 10 (the angular rotation between pictures) divided by the number of facets crossed by the centre of the pseudopupil along the diagonal frontoventral facet rows.

Facet diameter measurements were based on the average width of three adjacent facets in a row within each pseudopupil. The hexagonal geometry of the lens facets results in three possible axes for ommatidial rows (roughly at 60 angles to one another). As each of these directions results in a similar, but slightly different facet diameter, we took the mean D for each of the three axes through a central ommatidia and averaged these together. Facet diameters were measured every 5 between −40 and 40 and every 10 thereafter to  ± 90 along the same transect as Δϕ were measured.

The raw data were smoothed using a cubic spline in the r computing environment (R Core Team 2013). The minimum interommatidial angle () and maximum facet diameter () were obtained from the smoothed data. We chose to analyse minimum Δϕ and maximum D because they should represent maximum visual abilities. If the visual demands of microhabitats are limiting to species distributions, we expect peak abilities at the most sensitive part of the eye to be more closely related to habitat use than average abilities.

Eye Regionalization

As there is no established standard measure of eye regionalization, we used the following index for the degree to which D and Δϕ vary regionally across the eye. We first calculated mean eye transects for each species. We defined the mid-point D size as halfway between the maximum and minimum D obtained for each species mean transect (Fig. 5a). Regionalization in D was therefore the proportion of the transect with facet diameters (regD) above the mid-point D size for the species. Essentially, regD is the width of the transect at half-max, so that highly regionalized eyes will be low in regD.

Using the transects of D as in Fig. 5a, regD, is calculated as follows. Let R be a subrange within the degree range (0, 180). Then, is the degree range of the eye with facet diameters that are greater than half-max and is defined as follows:

(eqn 1)

Then:

(eqn 2)

A similar procedure was used to obtain the index of regionalization in interommatidial angle, regΔϕ:

(eqn 3)

and:

(eqn 4)

Therefore, regD and regΔϕ can range in value from nearly 0 to nearly 1, with small values indicating greater regionalization.

Statistical Analyses

Linear regressions of log-eye size with log-body size were used to test hypotheses of allometry. Isometry implies that each linear measurement of morphology will scale one to one with body length. Therefore, if morphological variation scales isometrically with size, we expect that a log–log regression will result in a slope of 1·0 (Fig. 2a). A slope greater than one indicates hyperallometry, and slope < 1·0 indicates hypoallometry.

We also examined whether eye morphology was explained by scaling or by association with ecological variation using anova. For eye size, we tested models including body size, microhabitat type and their interaction. For each functional variable (, , regD and regΔϕ), we tested models with eye size, microhabitat type and their interaction as dependent variables, dropping non-significant interactions. All statistical analyses were performed in the r statistical computing environment (R Core Team 2013).

Phylogenetic Analyses

Because species share evolutionary history, they are not independent data points for standard statistical analyses (Felsenstein 1985). To account for phylogeny in our analyses, we used a model-based approach to analyse the evolution of quantitative traits, comparing the fit of Brownian motion and Hansen models. Brownian motion models describe the evolution of traits as a purely stochastic process (Eq. eqn 6; Martins & Hansen 1997; Butler & King 2004). Written as a stochastic differential equation, the Brownian motion model describes evolutionary change in a phenotypic trait (X) through time (t) with dB(t) an increment of a standard Brownian motion (BM) process. One parameter is fit in this model, (σ) which is proportional to the strength of stochastic changes or macroevolutionary ‘drift’:

(eqn 5)

Hansen models are Ornstein–Uhlenbeck processes developed for phylogenetic comparative analysis (Eq. eqn 6; Martins & Hansen 1997; Butler & King 2004), which describe the evolution of quantitative traits under a model of stabilizing selection. It is important to note that these are used here to represent macroevolutionary models that describe evolution over very long time-scales (millions of years) and cannot be extrapolated down to natural selection within populations, which occur on very short time-scales. Written as a stochastic differential equation, the Hansen model describes evolutionary change in a phenotypic trait (X) through time (t) with dB(t) an increment of a standard Brownian motion (BM) process, with additional terms for stabilizing selection (α), drift (σ) and an optimal trait value (θ):

(eqn 6)

Importantly, Hansen models allow for the specification of different optima on different branches of a phylogeny to indicate transitions to different selective regimes, as when species rapidly evolve in morphology in response to a new microhabitat type.

We used the Megalagrion phylogeny (topology and branch lengths) published by Jordan, Simon & Polhemus (2003, Fig. 3). We pruned the Jordan, Simon & Polhemus (2003) tree to include only the species for which we have phenotypic data using the ‘ape’ package in r (Paradis, Claude & Strimmer 2004), and scaled the time depth of the tree to 1. All evolutionary analyses were conducted on log-transformed variables.

Size and Scaling

Means with standard errors and sample sizes for all measurements are given in Table 1. We found that body size in Megalagrion varies over twofold in length and is strongly associated with microhabitat type (Fig. 4a, Table S1). Stream species are significantly larger than seep (P = 0·01, Tukey HSD test), plant (P = 0·03), terrestrial (P = 0·02) and pool species (P < 0·001). We found support for hypoallometry (i.e. a trend of reduced eye size with increasing body size), along with an allometric microhabitat effect (Figs 2b and 4b, Table S1).

Functional Morphology of the Eyes

We found extensive variation in facet diameter, interommatidial angle and their regionalization, and strong associations with microhabitat type (Fig. 5, Table 2). Despite being the smallest in size, pool species have some of the largest maximum facet diameters (light sensitivity, Fig. 5a,c) and smallest interommatidal angle (resolution capabilities, Fig. 5b,d) among Megalagrion damselflies. and are able to be simultaneously maximized via a high degree of regionalization across the eye so that the largest D and smallest Δϕ occur jointly in a small acute zone (Fig. 5a,b). This arrangement has allowed small species to have larger than species which are much larger in size (sharp high peak in D for pool, compared to lower peak for stream Fig. 5a; also compare in Fig. 5c). In contrast, the large stream species have less regionalized eyes. While not best in either sensitivity or resolution compared to other Megalagrion, they had relatively high resolution (small Δϕ) and sensitivity (large ) capabilities over more of their eye when compared to species in other microhabitats (note broader trough in Δϕ Fig. 5b and in Figs S2–S5).

Table 2. ANOVA tables for maximum facet diameter (), minimum interommatidial angle (), regionalization in facet diameter (regD) and regionalization in interommatidal angle (regΔϕ). Each variable was tested in separate anovas as dependent variables explained by variation in eye size, microhabitat type and their interaction as independent variables. Non-significant interactions were dropped from the models. P values in bold indicate significant effects in the model. All models without interactions report type II and models with interaction report type III sums of squares

Dependent	Independent	Sum Sq	d.f.	F	P ( > F)
	Eye size	5·40	1	15·09	0·0116
	Habitat	48·86	4	34·17	0·0008
	Eyeht : habitat	14·76	2	20·65	0·0038
	Residuals	1·79	5
	Eye size	0·034	1	9·31	0·0185
	Habitat	0·065	4	4·47	0·0415
	Residuals	0·025	7
regD	Eye size	0·0008	1	5·36	0·068
	Habitat	0·0074	4	12·11	0·009
	Eye size : habitat	0·0024	2	7·83	0·029
	Residuals	0·0008	5
regΔϕ	Eye size	0·0000	1	0·01	0·908
	Habitat	0·0091	4	5·03	0·031
	Residuals	0·0032	7

Not all microhabitat types maximized both capabilities, as the terrestrial species had the best light sensitivity overall (highest , Fig. 5c) but the worst resolution capabilities (largest , Fig. 5d). In general, resolution capabilities were more conserved, with greater variation in light sensitivity among microhabitat types. For example, the plant and seep species had average resolution () among Megalagrion damselflies, but the plant species had very poor sensitivity (comparatively small , Table 1, Fig. 5a,b).

Ecology or Allometry

For three of the four parameters (, regD and regΔϕ), ecology explained more variation than allometry among species (compare P-values for the anova effects in Table 2). Only for was eye size a stronger organizing principle than microhabitat type (eye size effect: P < 0·02, habitat effect: P < 0·04; Table 2).

Light sensitivity

For , in general, larger eyes had larger (positive regression slope for eye size Table 3), but there is also evidence for strong ecological effects with highly significant effects for microhabitat type and its interaction with size (see anova effects in Table 2). In particular, seep species have a different slope between and eye size from all other microhabitat types, as well as plant species with much smaller than pool species after controlling for the other effects (Table 3, Fig. 5).

Table 3. Parameter estimates for the regression slopes and intercepts (effect sizes) of , , regD and regΔϕ

	Eye size	Plant	Seep	Stream	Terrestrial	Eye size : seep	Eye size : stream
	8·67  ± 2·23*	−6·50  ± 0·76***	334·4  ± 54·2**	5·70  ± 6·23	0·53  ± 0· 93	−188·6  ± 30·2**	−5·84  ± 3·17
	−0·488  ± 0·16*	0·127  ± 0·07	0·130  ± 0·06	0·276  ± 0·12*	0·337  ± 0·08**
regD	−0·107  ± 0·046	0·075  ± 0·016**	1·70  ± 1·12	−0·348  ± 0·129*	0·122  ± 0·019***	−0·883  ± 0·623	0·236  ± 0·065*
regIO	0·007  ± 0·056*	0·076  ± 0·025	0·003  ± 0·021	0·089  ± 0·041	0·097  ± 0·028*

• Regressions for D and regD contained the full model with eye size and microhabitat type as main effects and their interaction. The models for and regΔϕ contained main effects only with no interaction. The microhabitat term was parameterized as contrasts against the pool microhabitat types, so that the intercept for all other microhabitat types refers to values above that of the pool type. Standard errors are given and significance levels are indicated by asterisks: * ≤ 0·05, ** ≤ 0·01, *** ≤ 0·001.

Regionalization in facet diameter exhibited a dichotomous pattern relative to ecology. Variation among the large stream species showed that the fraction of the eye covered by relatively large D increases with larger eye size. The opposite was true for the pool species, where the fraction of the eye with relatively large D actually decreased with larger eye size (compare slopes for regΔϕ regressions with eye size and eye size:stream interaction in Table 3, Fig. 5).

Selection and Microhabitat Type

Evolutionary analyses confirmed that many features of the Hawaiian damselflies studied here were best explained by a model of divergent selection between microhabitats (body size, , regD and regΔϕ). For each of these features, an adaptive model with microhabitat as the selective factor was the best-fitting, strongly supported model (Table 4), over the alternative model of neutral evolution by drift or evolution towards a single global optimum. Parameters for the selection strength were much stronger than drift, and estimated optima that are consistent with the observed data (Table 5). In contrast, we found no evidence of adaptive evolution in maximal resolution, , with no models outperforming any of the others (Table 4).

Table 4. Model fit statistics for eye size, , , regD and regΔϕ. The model with the best fit (Bayesian Information criterion) is listed as 0, with ΔBIC values listed for all other models. Bootstrap model selection frequencies based on 2000 bootstrap replicates are included in parentheses. The microhabitat model best explained variation in all traits except for

	Microhabitat (%)	BM (%)	Global optimum (%)
Size	0 (99)	7·2	10·6
	0 (93·2)	4·5	4·1
	4·0 (13)	0 (41)	1·0 (6)
regD	0 (98)	6·6	11·0
regΔϕ	0 (100)	13·0	14·9

Table 5. Parameters estimated for the microhabitat model for body size and eye morphology. Strength of selection, α, and noise, σ, are shown for all morphological variables for which the microhabitat model performed best. Estimated optimal values (θ) for each microhabitat are shown for body size (mm), (μm), (degrees), regD and regΔϕ. Only σ is displayed for ; it was not best explained by a selection-based model. Bootstrap 95% confidence intervals for parameter estimates are in parentheses

	Size			regD	regΔϕ
α	100 (95·3, 152)	7·7 (2·3, 103)		305 (302, 386)	13 (2·7, 101)
σ	12 (2·95, 16·7)	2e-5 (4e-6, 3e-4)	0·02 (0.01, 0·03)	0·15 (0·03, 0·22)	0·01 (0·00, 0·05)
	6·0 (5·7, 6·2)	31·8 (30·3, 33·3)		28·1 (26·5, 29·9)	36·7 (35·0, 38·7)
	4·6 (4·3, 5·0)	31·3 (29·4, 33·0)		26·4 (24·2, 28·6)	27·7 (23·5, 29·9)
	4·6 (4·1, 5·0)	28·3 (24·9, 31·0)		22·2 (19·2, 25·2)	35·0 (31·5, 38·1)
	4·5 (4·0, 5·0)	37·2 (34·2, 42·2)		25·6 (22·7, 28·6)	37·4 (34·2, 42·6)
	3·9 (3·7, 4·1)	33·6 (32·6, 34·6)		16·6 (15·4, 17·7)	27·3 (26·1, 28·5)

When we fit evolutionary models to our data, the following combinations of parameters (particularly the positions of optima for the various microhabitat types) predict how phenotypic variation may have arisen to produce the variation in species we observe today. In terms of body size, our best OUCH model was a microhabitat model with a large optima for stream species, intermediate optima for seep, plant and terrestrial species, and a small body size optimum for pool species (Table 5). This evolutionary finding is consistent with a scenario where microhabitat use provides strong selection on body size, with correlated variation in eye size (compare Fig. 4b and Fig. 2b).

For light capture ability, the microhabitat model provides a different ordering of ecomorph types for optima in , where terrestrial species have the best light capture ability with the largest by several μm (Table 5). Surprisingly, the smaller pool species have the second largest , overlapping only slightly with the terrestrial species (Table 5). Stream and seep species are evolving towards similar, intermediate optima, while the plant species have very small . These evolutionary results are again consistent with a non-allometric microhabitat effect for , where the optima represent offsets from a common regression line (i.e. different intercepts, Fig. 2c).

The evolutionary optima recovered for regionalization in light capture abilities showed the large-bodied stream species with the lowest regD, with seep and terrestrial species overlapping considerably with the stream optima (Table 5). Pool species are the most regionalized by far, showing no overlap in regD with any of the other microhabitats (Table 5).

For regionalization of resolution capabilities, the terrestrial, stream and plant species evolving towards low regΔϕ with substantial overlap in their optimal values (Table 5). On the other hand, the seep and pool species are distinctly more regionalized than the other three microhabitats (Table 5). Thus, microhabitat has a strong evolutionary influence on eye regionalization, with the pool species having the most regionalized eyes overall in both light sensitivity and resolution capabilities.

Finally, in contrast to the other three functional parameters, microhabitat-based selection did not seem to influence . Instead, maximal resolution was equally well explained by a model of neutral evolution (Table 4).

Functional Morphology of the Eyes

The dominance of ecology in shaping eye geometry results in the general conclusion that bigger is not always better. Indeed, the pool species which are the smallest in size have the best visual performance in both sensitivity and resolution, whereas the largest species have merely moderate performance. Among insects, this is a rather unusual pattern. In general, larger-eyed species have improved performance (Jander & Jander 2002; Rutowski, Gislen & Warrant 2009). Furthermore, all of these species studied here are closely related, diurnal, with modest differences in ecology. None have evolved extreme adaptations for nocturnal or crepuscular lifestyles, for example. Rather, they are members of an adaptive radiation that share the same forest, and therefore, these morphological variations may provide sufficient performance advantages that allow for niche partitioning (Arnold 1983).

This unusual pattern in performance is achieved by expanding the dimensionality of visual performance. Rather than scaling all ommatidia equally with body size, the smallest species take advantage of high regionalization, concentrating performance at an acute zone. The largest species do not have the best maximal performance, but have larger D and small Δϕ more evenly distributed over the eye compared to the other Megalagrion species (Figs S2–S5). This more even distribution should confer improved sensitivity and resolution over a larger portion of the eye and visual field.

Regionalization is well known to be associated with prey capture or mate acquisition in insects (reviewed in Land & Nilsson 2002). However, Hawaiian damselflies all share similar predatory and mating behaviours (Polhemus & Asquith 1996). Instead, regionalization here is strongly associated with microhabitat specialization. Thus, as evidenced by the smallest species having the best light-capture abilities, regionalization may provide a mechanism for alleviating the resolution-sensitivity trade-off, allowing greater flexibility in evolutionary adaptation to light levels.

In contrast to the adaptive patterns above, we found that maximum resolution is more conserved across species, with no evidence for the influence of ecology on visual resolution. We therefore conclude that evolution in this parameter is purely stochastic or highly constrained. We found that across all species varied by only 0·36 with the majority of species ranging between 0·70 and 0·83. This suggests that maximum resolution is highly conserved, which would be expected if the tasks of object discrimination are similar across all species (and operating at roughly the same distance). Damselflies and all odonates are aerial predators that use visual cues to locate and capture prey items (Labhart & Nilsson 1995; Olberg, Worthington & Venator 2000; Olberg et al. 2005). Thus, the need for excellent resolution regardless of size or habitat provides a reasonable explanation for the low variation in observed in Hawaiian damselflies as well as the relatively small found in odonates in general (Land 1997).

Potential Selective Pressures on Visual Morphology

In this study, we have taken the first step in trying to understand variation in the visual morphology in a radiation of closely related species. Our evolutionary analyses show that facet diameter, regionalization and body size are evolving adaptively in response to microhabitat type. What selective pressures may be operative to explain these patterns? The different microhabitats occupied by these damselflies may place different demands on the visual system for a variety of reasons. The amount of clutter varies greatly across the microhabitats examined here. For example, the ‘pool’ microhabitat is often heavily vegetated, making it difficult to distinguish prey or mates from the background. Whereas this should primarily be a problem of resolution; light sensitivity, colour, motion and the neural processing required to distinguish ‘object of interest’ from background will all come into play. Conversely, the ‘stream’ and some ‘plant’ microhabitats are open so that prey and mate detection should be relatively easy against a brighter and potentially more visually uniform background.

The open vs. cluttered axis of habitat variation should also have a strong influence on body size and flight requirements. The wide-open ‘stream’ habitats, in particular, expose flying insects to strong wind currents, requiring much stronger flying abilities. Indeed, the large-bodied ‘stream’ species are much faster fliers, and are almost never found among clutter, as compared to the smaller, slower-flying species (pers. obs; Polhemus & Asquith 1996). It is interesting to note that the evolutionary optima suggested for our body size model indicate three distinct groups of body size: large for the stream species; medium for the seep, plant and terrestrial species; and small for the pool species. As the three medium-sized microhabitats are much more variable in the degree of clutter they contain, this ranking is in general concordance with the degree of clutter that these species experience. We note also that flight speed is known to influence eye geometry in insects, as flight speed should be matched to eye morphology so as to prevent blurring by the flow field that the animal experiences (Howard & Snyder 1983; Land 1997). We have no quantitative data on flight speeds in these damselflies to evaluate this hypothesis; thus, it remains a potential contributing factor to the pressures which may limit stream species in particular from evolving large ommatidia, especially away from the acute zone.

Another possibility is that variation in light levels may be the selective pressure driving visual system evolution. Several studies have suggested that light levels can be limiting in insects (Land 1997; Kapustjanskij et al. 2007; Doring & Spaethe 2009), and we found that the strongest evolutionary responses to microhabitat type are changes in light sensitivity and regionalization of the eye. In these Hawaiian damselflies, varies widely with microhabitat, generally in a manner consistent with the apparent light levels of the habitat. For example, pool-dwelling species often occur in heavily shaded microhabitats that may place a premium on light sensitivity. Our evolutionary analyses show that evolves under an adaptive model with strong selection, and pool damselflies have large D for their eye size. When light is not limiting, for example, over open, well-lighted streams, no additional functional advantage is gained by increasing any further. Rather, greater functional benefit may be gained by increasing spatial or temporal resolution for tracking aerial prey or mates. Interestingly, ‘stream’ species have among the smallest optima for , despite being the largest bodied damselflies (Table 5). To fully explore how variation in light levels drives the evolution of eye morphology, it will be interesting to explore whether other traits that contribute to light sensitivity, such as rhabdom diameter or focal length, show similar evolutionary trends.

Whether regionalization is a direct target of selection or a by-product of selection for light sensitivity at small sizes is an interesting question to consider. If, for example, body size is set by the locomotor requirements of the habitat, then eye size, being highly constrained to covary with body size, will be fixed at a relatively small size for the pool species. Streams in Hawai'i are formed by erosion of volcanic mountains, and thus, smaller streams are frequently high-walled and rather dark, or receive few hours of direct insolation. Pool species are frequently found in these dark cluttered habitats where other ecomorph types, particular the stream species, are rare. Thus, regionalization may be the key innovation that allows pool species to have sufficient light sensitivity by concentrating it in an acute zone so that they can exploit this habitat that is not accessible to the larger species or those with smaller .

Summary

Here, we found that microhabitat use strongly influences eye morphology in Hawaiian damselflies. Although we cannot determine the precise adaptive basis for the eye geometry–microhabitat associations, this study has provided many hypotheses for future study. It suggests that habitat structure can have a strong influence on overall size, and covariation in resource axes in the habitat itself may impose complex selective pressures on vision. To elucidate the adaptive significance behind the ecology–morphology link we have found, future studies should explore variation in microhabitat-scale differences in light levels, degree of clutter, visual performance and locomotor requirements. As adaptation to microhabitat is a common organizing factor in other functional systems such as locomotion and thermoregulation (e.g. Pounds 1988; Adolph 1990; Irschick & Garland 2001; Calvao et al. 2013), this raises the intriguing possibility that microhabitat variation in light levels may be an important factor influencing the design of visual systems (Endler & Thery 1996; Endler & Mielke 2005).