Spatial analyses of two color polymorphisms in an alpine grasshopper reveal a role of small‐scale heterogeneity

Abstract Discrete color polymorphisms represent a fascinating aspect of intraspecific diversity. Color morph ratios often vary clinally, but in some cases, there are no marked clines and mixes of different morphs occur at appreciable frequencies in most populations. This poses the questions of how polymorphisms are maintained. We here study the spatial and temporal distribution of a very conspicuous color polymorphism in the club‐legged grasshopper Gomphocerus sibiricus. The species occurs in a green and a nongreen (predominately brown) morph, a green–brown polymorphism that is common among Orthopteran insects. We sampled color morph ratios at 42 sites across the alpine range of the species and related color morph ratios to local habitat parameters and climatic conditions. Green morphs occurred in both sexes, and their morph ratios were highly correlated among sites, suggesting shared control of the polymorphism in females and males. We found that in at least 40 of 42 sites green and brown morphs co‐occurred with proportions of green ranging from 0% to 70% with significant spatial heterogeneity. The proportion of green individuals tended to increase with decreasing summer and winter precipitations. Nongreen individuals can be further distinguished into brown and pied individuals, and again, this polymorphism is shared with other grasshopper species. We found pied individuals at all sites with proportions ranging from 3% to 75%, with slight, but significant variation between years. Pied morphs show a clinal increase in frequency from east to west and decreased with altitude and lower temperatures and were more common on grazed sites. The results suggest that both small‐scale and large‐scale spatial heterogeneity affects color morph ratios. The almost universal co‐occurrence of all three color morphs argues against strong effects of genetic drift. Instead, the data suggest that small‐scale migration–selection balance and/or local balancing selection maintain populations polymorphic.


| INTRODUC TI ON
Balanced intraspecific color polymorphisms have fascinated researchers for a long time, because they demand eco-evolutionary explanations for how polymorphisms are maintained in the long run (Fisher, 1930;Ford, 1965;Huxley, 1955;Svensson, 2017). The mechanisms that maintain populations polymorphic are, however, often unexplored and likely differ between species. The insect order Orthoptera-the crickets, bush-crickets and grasshoppersis particularly remarkable because it represents large, phylogenetically old clade in which various polymorphisms, including a conspicuous green-brown polymorphism, are shared among species (Dearn, 1990;Rowell, 1971). Among the Central and Western European Orthoptera, for example, about 20% of the species are known to be green-brown polymorphic (Bellmann, 2006;Bellmann & Luquet, 2009) and this color polymorphism occurs in both suborders, Ensifera and Caelifera, that have diverged about 200 Mya (Misof et al., 2014). Even some stick insects, Phasmatodea, and matids, Mantodea, two groups that have diverged from the Orthoptera about 250 Mya (Misof et al., 2014), are green-brown polymorphic.
The old age of the clade and the immense number of green-brown polymorphic species, interspersed with some exclusively brown and some exclusively green taxa, makes Orthopterans particularly suitable for studying the mechanisms that maintain balanced color polymorphisms.
Color polymorphisms often have a genetic basis, and the rarest morph appears too frequent as to be explained solely by recurrent mutation (Fisher, 1930;Huxley, 1955). Genetically based phenotypic polymorphisms are necessarily labile to loss by directional selection favoring the most adapted morph and to random loss by genetic drift.
So which mechanisms maintain populations polymorphic? Polygenic traits can remain polymorphic in mutation-selection-drift balance when selection at individual loci is weak and the mutational target is large (Bürger & Lande, 1994;Bürger, Wagner, & Stettinger, 1989), but, as we argue below, we consider it unlikely that the green-brown color polymorphisms is a highly polygenic trait. Temporally fluctuating selection may help to maintain some polymorphisms, even though fluctuating selection is susceptible to overshooting (Bell, 2010;Sasaki & Ellner, 1997). More efficient may be spatially heterogeneous selections with gene flow that can maintain populations polymorphic in migration-selection balance (Yeaman & Whitlock, 2011). Finally, selection itself may favor polymorphisms if either different fitness components are involved in selective trade-off and lead to net disruptive selection or if selection is negative frequencydependent (Fitzpatrick, Feder, Rowe, & Sokolowski, 2007;Gigord, Macnair, & Smithson, 2001). Not all cases of discrete phenotypes need to be genetically controlled, and some may be due to phenotypic plasticity (West Eberhard, 2003). This, however, only shifts the question to how selection favors the maintenance of phenotypic plasticity in the light of the added costs that are likely to occur when two alternative options need to be maintained.
The green color in Orthoptera is formed by tetrapyrroles, in particular biliverdin (Fuzeau-Braesch, 1972). The genetic pathway to the production of a green skin thus needs to include functional metabolism, transport, and deposition of the green pigments in the developing skin prior to regular molting between nymphal instars and finally into the adult insect (Shamim, Ranjan, Pandey, & Ramani, 2014). As we consider green-brown polymorphisms that are discrete by state rather than gradual variation along a continuum, the pathway(s) seem(s) to be switched off in brown individuals that lack green colors entirely. Loss-of-function mutations are likely to occur more frequently than gain-of-function mutation that restore the genetic pathway (Behe, 2010). We therefore argue that the discrete nature of the polymorphism, the apparently simple genetic inheritance in green-brown polymorphic stick insects (Comeault, Carvalho, Dennis, Soria-Carrasco, & Nosil, 2016) and the presumably relatively simple biochemical pathway, makes polygenic inheritance and thus mutation-selection-drift balance less likely in the case of the greenbrown polymorphism in Orthopterans. Formal genetic studies are, however, required to verify this interpretation.
Temporal and spatially variable selections are possible mechanisms even for discrete traits that are controlled by few genes.
Fluctuating selection has been claimed to be common in natural populations (Siepielski, DiBattista, & Carlson, 2009), although this conclusion has been challenged (Morrissey & Hadfield, 2012). But even if selection is fluctuating, the system needs to be very fine-tuned with sufficiently strong selection to counter genetic drift effects and regular switches in the sign of selection in order to protect populations against accidental loss of one of the morphs. While such systems may occur, it is not trivial to find a situation in which selection varies not only in magnitude, but also in direction on a regular basis.
More promising for the long-term maintenance of polymorphisms seems spatially heterogeneous selection with gene flow or systematic temporal changes in selection induced by negative frequency dependency.
In some species of Orthoptera, the green-brown polymorphism seems to be under environmental control with background color of the habitat and moisture being the main determinants during nymphal development (Dearn, 1990;Rowell, 1971). However, this is not universally true and, in some species, including the species that we study here, background coloration does not affect color morph development (Valverde & Schielzeth, 2015). Notably the green-brown polymorphism that we address here is different from the phase polymorphism in some locusts (Song, 2011). The phase polymorphism is triggered by density and is mediated via juvenile hormone that coordinates changes in color along with changes in morphology, physiology, and behavior (Tanaka, Harano, Nishide, & Sugahara, 2016). In phase polymorphism systems, it is difficult to distinguish those different components. Our color polymorphism system, in contrast, suggests that color itself is under selection.
We here study the club-legged grasshopper Gomphocerus sibiricus (Linnaeus 1767), an alpine ground-dwelling species that displays a striking green-brown color polymorphism in both sexes. In most populations in the European Alps, the brown morphs are more abundant than the green morphs (see below). We combine large-scale with small-scale spatial and temporal sampling that allows us to address the evolutionary ecology of the color polymorphism in this species. We predict that color morph ratios are spatially variable and predictable by environmental conditions if selection for local adaption contributes to the maintenance of color polymorphisms in this species. Besides the striking green-brown polymorphism, the species also features a distinctive pied morph that, like brown morphs, lacks any green pigments, but is characterized by a black-and-white contrast on head and pronotum ( Figure 1). Pied morphs occur at lower numbers, but are also widespread and the questions about the maintenance of the polymorphism applies equally to pied morphs.

| Sampling sites
We sampled 42 sites from across large parts of the alpine range of the club-legged grasshopper from eastern Austria to southwestern Switzerland, covering most of the altitudinal range of the species (Table S1). We identified target regions (17 regions in total), operationally defined as mountain areas without division by major valleys and inhabitable mountain ridges, and aimed to sample two to three sites per region (with some variation) that differed in habitat, in particular altitude, aspect, and/or vegetation cover. Nearest-neighbor distances were 8-92 km (mean 51 km) among regions and 0.2-23 km (mean 5 km) among sites within regions. From each site, we aimed to sample 50-100 individuals and to score them for their color morph.
A priori power analysis shows that with 50 (100) individuals, the expected standard error is about ±0.067 (±0.047) for estimated proportions in the range of minor morph frequencies around 0.33.
At one of the sites, a large SW-facing slope in Central Austria (site Albitzen I/Heiligenblut), we sampled at six locations in two consecutive years in order to estimate the local spatial and temporal repeatability of color morph ratios. This site was chosen, because the habitat was visually rather homogenous without obvious barriers to dispersal and we hence expect stable morph ratios among the six nonoverlapping sampling locations. Locations were separated by about 50 m. Any remaining heterogeneity beyond sampling variance may be attributed to fine-scale microhabitat heterogeneity.
We refer to variation among regions as large-scale variation, as it involves a scale of multiple kilometers and populations that are separated by dispersal barriers. We refer to variation among sites within regions as small-scale variation, as it involves a scale of a few hundred meters to a few kilometers (sometimes a few dozen kilometers) among sites that are not separated by major dispersal barriers, but that differ in habitat structure. Finally, we refer to variation among locations within sites to fine-scale variation, since it involves a scale of a few dozen meters with no dispersal barriers and no habitat differences.

| Field sampling and scoring
Grasshoppers were sampled in July and August in 2014 and 2015 (Table S1), and in August 2016, we resampled the six locations at the site Albitzen I/Heiligenblut (Table S1). Sampling was carried out between late morning and late afternoon under warm conditions when grasshoppers are mostly active. Grasshoppers were detected by slowly walking across the low-vegetation habitats based on their escape movements. Detected individuals were caught by hand or using small nets. As grasshoppers were detected by motion rather than when sitting still on their natural background, detection was independent of color, albeit not necessarily independent of activity levels. Individuals were temporarily retained in small vials and when sufficient numbers of individuals were caught, they were all scored for their color morphs. This procedure ensured that no individual was recorded twice. Individuals were released on spot after scoring.
All individuals were sexed, and their color morph was recorded.
We scored three morphs that we call green, brown, and pied ( Figure 1). Green individuals are readily identified by their green F I G U R E 1 Typical examples of the three color morphs in males and females. Only green morphs show green color. Pied morphs are distinguished from brown morphs by the characteristic pale transverse stripe across the sides of the head and the pronotum. Brown morphs are very variable in darkness but are overall rather uniformly colored. This particular green male has an ecoparasitic mite attached near the mandibles pronotal lobes (and sometimes dorsal parts of the pronotum) as well as green frontal part of the head. Individuals darken within a few days after imaginal molting (Valverde & Schielzeth, 2015), and while green females are always immediately recognized, green males can become so dark that the green parts are barely visible in side views. However, the diagnostic green front always reveals even the darkest green males. Importantly, only green morph individuals display green colors. Brown individuals are highly variable greyish, brownish, or blackish, but always lack green tones, and they also lack the characteristic pattern of pied morphs. Pied morphs show distinct black-and-white stripes across the sides of the head and the pronotum and typically (particularly in females) show a variable black face mask. Pied males, while being still recognizable as those in the last instar nymphal stages, darken so much as imagoes that they are often indistinguishable from brown morphs. We therefore analyze the prevalence of green morphs using data from both sexes (green vs. brown/pied) and the prevalence of pied morphs for the female sample only (pied vs. brown/ green). Along with sex and morph identities, we recorded the extent of the face mask ( Figure S1) as well as marked deformations and mite infections.

| Habitat characterization
Local habitat characteristics were compiled from three sources.
First, we estimated vegetation cover on spot. Specifically, we scored the percentage cover of (i) bare ground, (ii) gravel and stones, (iii) herbal and grassy vegetation, (iv) low-growing shrubs (Vaccinium myrtillus, V. ulingosum, V. vitis-idaea, Arctostaphylos uvaursi), and (v) junipers. Furthermore, we noted whether the habitats were grazed, indicated by presence or feces of cattle. Second, we derived topographical characteristics, that is, elevation and aspect, from a digital elevation model (DEM; spatial resolution of 30 arc-seconds) (USGS/GLCF 2000) for each site and transformed the aspect data into north-south and east-west component by arcsine and cosine transformations (Leyer & Wesche, 2007). Sampling dates were recorded along with geographic positions (latitude and longitude). Third, we downloaded all 19 BIOCLIM variables from the CHELSA (Climatologies at high resolution for the earth's land surface areas) data (Karger et al., 2017a,b) to characterize the general predominant climate at our study sites. The CHELSA data depict downscaled model output temperature and precipitation estimates of the ERA-Interim climatic reanalysis (spatial resolution of 30 arc-seconds) and refer to the period 1979-2013 (Karger et al., 2017a,b). As BIOCLIM variables are highly correlated among each other, we used principle component analysis to summarize climatic variation in independent dimensions.
As color morphs may differ in their thermal activity, which could lead to differences in catching probabilities, we gathered sunshine duration for the sampling days (estimated in hours) and cloud cover following the Okta classification by estimating how many eighths of the sky were covered by clouds (World Meterological Organization 2008).

| Statistical analyses
We fitted generalized linear mixed models (GLMM) with a binomial error distribution and logit link to color morph identities with sex and environmental predictor variables fitted as fixed effects and site and region as random effects in the model. The response was coded as binary (yes/no) rather than aggregated to proportions. The analysis of pied morphs was limited to females, and we therefore omitted sex as a predictor. Variance components are summarized as intraclass correlations (repeatabilities) R, which are variance components standardized by the total phenotypic variance on the observed scale (Nakagawa & Schielzeth, 2010). Uncertainties in the estimates of intraclass correlations were quantified by parametric bootstrapping.
We explored specific environmental predictors that may explain spatial variability in morph ratios. Predictors were grouped in three classes: (i) geographic predictors, in particular longitude, latitude, altitude, and aspect; (ii) general climatic conditions based on BIOCLIM data; and (iii) local habitat characteristics recorded in the field.
Within each class, we fitted predictors one at a time as well as in combination. When fitting predictors in isolation, we used Benjamini and Hochberg (1995) family-wise false-discovery rate control for adjusting p values for multiple testing. In the case of the multiple regression, we apply full model test within classes to control for multiple testing as recommended (Forstmeier & Schielzeth, 2011), which in our case involves keeping the random effects and sex as a fixed effect in both the full and the reduced models. We also fitted a model that included all predictors from all classes simultaneously, but this model did not converge due to the unfavorable ratio of predictors to populations sampled. The full model test across all variables should therefore be treated with caution.
All models were fitted in R 3.4.3 using the package lme4 1.1-12 for fitting mixed-effects models (Bates, Mächler, Bolker, & Walker, 2015) and the package rptR 0.9.2 for quantifying ratios of variance components (Stoffel, Nakagawa, & Schielzeth, 2017). Furthermore, we used the p.adjust and the prcomp functions form the basic stats package (R Core Team, 2016) for false-discovery rate control and principle component analysis, respectively. A total of 1,093 individuals were green (24% in females, 27% in males), 2,479 individuals were brown (52% in females, 64% in males), and 709 individuals were pied (25% in females, 9% in males). Besides the obvious deficiency of pied males (due to blurring of the patterns by darkening), there was a slight, but significant excess of green individuals in males (GLMM: b = 0.19 ± 0.07, z = 2.55, p = .011). The proportion of green morphs was highly correlated between the two sexes across sites (r = .83, t 40 = 9.32, p < 10 −10 , Figure 2).

| Fine-scale spatial and temporal heterogeneity within sites
Fine-scale sampling across six locations on a homogenous mountain slope in Central Austria (i.e., within a single site) resulted in fractions of green morph individuals between 28% and 36% in 2015 and 32% and 45% at the very same locations in 2016 (Figure 3).

| Smale-scale and large-scale spatial heterogeneity among sites and regions
Green morph ratios varied between 0% at two nearby sites in Eastern Austria (with samples sizes of N = 10 and 58 individuals, respectively) and three sites with a majority of green morphs (53%, 58%, and 70%, all N ≥ 86, Figure 4). There was no significant heterogeneity among regions (R = .033 ± .028, χ 2 1 = 1.14, p = .14), but significant heterogeneity among sites within regions (R = .115 ± .036, χ 2 1 = 191.0, p < 10 −40 ), so in total about 14% of the variation was among sites.  The univariate block shows estimates from GLMMs controlled for site and region as random effects and, for models on green morphs, also for sex as a fixed factor, while other parameters were fitted one at a time. p* shows p value corrected for multiple testing. The multivariate block shows fits from models that included all predictors from each block simultaneously. Full model tests (FMTs) refer to full model tests within blocks. FMT across all variables was significant for both the occurrence of green morphs (χ 2 15 = 30.44, p = .010) and the occurrence of pied morphs (χ 2 15 = 30.00, p = .012), albeit model convergence was impaired. P-values <0.05 are shown in bold.

| Geographic and climatic predictors
The proportion of green individuals did not depend on altitude, latitude, longitude, or aspect (Table 1). The proportion of green individuals tended to increase with PC1 of the BIOCLIM variables, although this effect was not robust to multiple testing control and thus should be treated with caution (Table 1). Precipitation loaded most strongly and negatively on PC1, and the trend was thus toward an increase in green individuals with decreasing precipitation ( Table 2).
The proportion of pied individuals significantly decreased with altitude and increased with longitude, although these effects were not significant after multiple testing control (Table 1). Latitude and aspect did not significantly affect the proportion of pied individuals (Table 1). The proportion of pied individuals decreased with higher values of PC2 of the BIOCLIM variables (Table 1), and PC2 loadings were negatively with temperature and variable with precipitation tending to load negatively ( Table 2), suggesting that the pied morphs decreased in frequency with lower temperatures and with lower precipitation during summer and the wettest month.

| Habitat characteristics
The proportion of green individuals was not significantly influenced by shrub, stone, juniper, herbal and grassy vegetation cover, and grazing (Table 1). Bare ground cover tended to have a positive effect on the occurrence of green morphs, but the effect was not significant after multiple testing correction and also not in a multiple regression (Table 1). The proportion of pied individuals was not significantly influenced by bare ground, shrub, stone, juniper, and herbal and grassy vegetation cover (Table 1), but grazing had a significant positive effect on the proportion pied and this effect was significant after multiple testing correction and also in the multiple regression (Table 1).

| Color variants
Front pattern was highly variable from being plain or plain with small black marks to being boldly marked black. Green individuals were typically largely unmarked, while brown individual were more variable. However, it was mostly pied females that showed strong, bold marks at their fronts ( Figure S1). A total of 176 individuals (7.9% of females, 0.5% of males) showed a clear pink color tinge (GLM for sex difference: b = −2.75 ± 0.41, z = −6.67, p < 10 −10 ) that was most common among brown (5.4%) and pied (5.2%) individuals and almost absent among greens (0.5%) (GLM for morph differences green vs. brown/pied: b = −2.48 ± 0.62, z = −3.97, p < 10 −4 ). There was no evidence for an altitudinal trend in the occurrence of pinkish individuals (GLM: b = 0.062 ± 0.040, z = 1.58, p = .11).

| Deformations and infections
While scoring color morphs, we also recorded occasional deforma-  Figure S2).

| D ISCUSS I ON
We here analyse the spatial distribution of one distinctive and one subtler color polymorphism in the club-legged grasshopper across most of its range in the European Alps. All populations were greenbrown polymorphic with the possible exception of two sites in the east of the range that possibly lack green morphs (although they may occur in low numbers). Pied morphs were also present at all sites alongside typical brown morphs. We found significant spatial heterogeneity in both polymorphisms, particularly among sites within regions, and relative temporal stability across two consecutive years in color morphs ratios (with some changes in pied morphs). Weather conditions and population sizes of G. sibiricus vary substantially across years (Illich & Windig, 1999), so that stronger fluctuations of morphs ratios might have been expected. Although longer timeseries are desired, the data are more suggestive of spatial rather than temporal variability.
We found evidence that the occurrence of green and pied morphs depends on identifiable climatic conditions, while the abundance of pied morphs was also influenced by habitat characteristics.
These findings argue for a role of selection in shaping local morph ratios. Furthermore, we found substantial unexplained variability among sites in morph ratios, suggesting a role of spatially heterogeneous selection caused by unexplored environmental conditions and/or genetic drift. In combination, the rather high frequency of all three color morphs, the evidence for environment dependence of color morph ratios and the reasoning that genetic drift alone is unlikely to maintain populations polymorphic in the long run (Gray & McKinnon, 2007;Svensson & Abbott, 2005) suggest that the most likely explanation for the maintenance of the color polymorphisms involves small-scale migration-selection balance or local balancing (in particular negative frequency-dependent) selection.
Which mechanism may ensure small-scale heterogeneous selection to maintain populations polymorphic in selection-migration balance?
Mountains are characterized by pronounced gradients with respect to altitude and aspect that, in combination with geomorphological and topological heterogeneity, harbor large variety of habitats even at a small spatial scale (Nagy & Grabherr, 2009 Chorthippus parallelus) species. It is currently unknown whether the color morph composition of the grasshopper community is related to color morph variation in individual species, but by the above reasoning, differences in community composition may contribute to spatial variability in morph ratios.
Migration-selection balance and frequency-dependent selection ultimately require the existence of multiple fitness peaks either in space (as for migration-selection balance) or in time (as for negative frequency-dependent selection). The two mechanisms are not mutually exclusive and may act in combination with other forms of multimodal fitness peaks also contributing to the maintenance of the polymorphism. Thermoregulation, predator avoidance behavior, parasite infestation, but also social interactions and mate choice can induce selective trade-offs across contexts that can create multimodality in the fitness landscape with discrete alternative peaks (Ahnesjö & Forsman, 2006;Forsman, 2000;. Multimodal fitness peaks may be reinforced when individuals seek out matching (micro)habitats (Edelaar, Siepielski, & Clobert, 2008;, but so far we have no evidence that green, brown, and pied individuals choose different microhabitats. Beyond the three discrete color morphs brown, pied, and green, we recorded occasional pink variants. Such variants are known to occur in other species of grasshoppers as well, usually at low frequencies, and the pink color is thought to be caused by reduced forms of insectorubin (Uvarov, 1966). Variation in "pinkishness" is gradual ranging from weak tinges of pink to intense purple, as opposed to the discrete nature of the three main color morphs. Pink and purple colors occur regularly among brown and pied individuals, but are rare among green morphs. Own observations from the laboratory suggest that once developed, pink/purple colors persist for life. Whether there is a genetic predisposition for the development of pink/purple colors is currently unclear.
In the present study, we aimed to characterize spatial variation in color morphs the in club-legged grasshopper G. sibiricius to shed light on which mechanisms may be involved in maintaining populations polymorphic (Gosden, Stoks, & Svensson, 2011). Among 42 populations across the alpine range of the species, we found almost universal coexistence of three color morphs with significant spatial variation mostly on a small (hundreds of meters to kilometers), but not on a very fine spatial scale (of a few dozens of meters). This indicates a role for local adaption in shaping morph ratios, in particular as we could identify environmental variables that co-vary with local color morph rations. Local adaptation in combination with gene flow offers a likely contribute to the maintenance of the polymorphism.
This does not exclude a role, even a predominant role, of local balancing selection for example by negative frequency dependency of color morph-specific fitness. Fitness assays are required to address the mode and role of selection in maintaining color polymorphisms in this species.

ACK N OWLED G M ENTS
We thank Anasuya Chakrabarty, Amy Backhouse, Inge Illich, Dennis Rödder, and Nikolaos Toliopoulos for various technical help and insightful discussions. The project was funded by an Emmy Noether grant from the German Research Foundation (DFG; SCHI 1188/1-1). A field trip was supported by a faculty grant by the Faculty of Biology at Bielefeld University.

CO N FLI C T O F I NTE R E S T
None declared.

AUTH O R CO NTR I B UTI O N S
HS conceived the study. All authors contributed to study design and data collection. PD and HS analyzed the data. PD led the writing of the manuscript. All authors contributed to manuscript revision.

DATA ACCE SS I B I LIT Y
Data are available from the Dryad Digital Repository: https://doi. org/10.5061/dryad.676c7g5.