Host specificity pattern and chemical deception in a social parasite of ants

In natural ecosystems, relationships between organisms are often characterised by high levels of complexity, where vulnerabilities in multi-trophic systems are difficult to identify, yet variation in specific community modules can be traceable. Within the complex community interactions, we can shed new light on dynamics by which co-evolutionary outcomes can inform science-led conservation. Here we assessed host-ant use in six populations of the butterfly Phengaris (=Maculinea) rebeli, an obligate social parasite of Myrmica ants and a model system in evolutionary and conservation ecology. Starting from the initial distribution of eggs, we estimated the survival of the parasite in the wild in nests of seven Myrmica ant species, and analysed the chemical cues evolved by the parasites to subvert its host defences. We found local variations in host specificity that are consistent with similarities found in the chemical profiles of hosts and parasites on different sites. At some sites, only one ant species is successfully exploited; at others, multiple-host populations are used. Understanding how stable or adaptable these associations are is essential knowledge when devising conservation measures to maintain keystone species of ant and locally adapted populations of Phengaris butterfly species, which are rare, threatened and a high priority for conservation worldwide.

Inside host ant nests, Phengaris larvae are fed mainly by the regurgitations of nurse ants, gaining about the 98% of their final body mass 4 over a period of either 11 or 23 months 4 , the result of a developmental polymorphism that exists in some populations. Full-grown butterfly larvae pupate in the upper chambers of the ant nest and emerge unharmed as adults via the colony's galleries.
More recent field studies have shown that an early finding, that each Phengaris species was an obligate parasite of a single and different Myrmica species, was oversimplified at a continental scale 22 . P. rebeli is now recorded from eight different Myrmica species in separate parts of its range in Europe 23 and references therein . However, while the existence of distinct host races within Phengaris species is well established 15 , it has been shown that P. alcon and P. rebeli, in particular, tend to be specific to a single Myrmica host species at population and usually regional scales, even though several potential Myrmica species co-exist in abundance on the same sites and largely overlap in their distributions 21 . More recently, "multiple-host using populations" have been reported for certain P. alcon and P. teleius populations in Denmark 24 , Poland 25 and Hungary 26 . Because non-host Myrmica colonies are known to tolerate Phengaris parasites if the colony is well provisioned, it remains unclear whether these comparatively rare instances represent random occasions driven by processes not dependent on the chemical mimicry the parasites have evolved or whether host associations differ fundamentally in these populations 21 .
To distinguish the two possibilities, we estimated the survival rates, rather than study the occurrence, of Phengaris rebeli caterpillars in Myrmica host colonies at six sites in Italy, and analysed similarities in the cuticular hydrocarbon signatures among host ants and parasites on which the chemical mimicry is based. We suggest that where the estimated survival rates do not differ between multiple Myrmica host species, the generalist host use is based on host-parasite interactions rather than environmental factors. How tight the relationships between the host ant and parasite are is central not only to understanding the evolution of widespread associations between butterflies and ants, but also for the survival and conservation of Phengaris spp., especially when faced with additional threats such as rapid climate change that necessitate more intrusive land management 27 .

Results
Myrmica communities. We located and excavated 186 Myrmica nests within 2 m (=foraging range) of G. cruciata plants in the 6 sites ( Fig. 1). Each site supported 2 to 5 species of Myrmica ( Fig. 1; Table S1), with M. schencki present at all sites and overall the commonest species present (43% of all colonies found). M. scabrinodis was found at all sites except for Col di Tenda. In contrast, M. lobulicornis and M. sulcinodis were found only at Col   1). Since previous studies showed no differential mortality of the eggs or early larval instars on G. cruciata growing in the niches of different Myrmica species, nor any bias in the discovery and retrieval of wild final-instar larvae from beneath food plants, P. rebeli egg distributions may be taken as equating to the proportion of young larvae entering nests of the different ant species 28,29 see the appendix .
The following summer we found a total of 137 Phengaris individuals in their final stages of growth (full-grown larvae or pupae) in 52 (28%) of the 186 Myrmica ant nests excavated, with site infestation levels ranging from 10% (S. Agostino, Collelongo) to 43% (Campitello). In contrast to the races of P. rebeli studied in the Pyrenees and French Hautes-Alps 30 , only one biennial P. rebeli larva was found (within one M. schencki colony at Col di Tenda) indicating that the incidence of polymorphic larvae in Italy is extremely low.
The mean (±SE) number of Phengaris full-grown larvae or pupae per infested nest was 2.6 ± 1.8, ranging from 1 to 8 individuals per colony. P. rebeli full-grown larvae or pupae were found only in M. schencki nests at Bardineto and S. Agostino (Northern Apennines), while at Col di Tenda and Oulx, we found the parasite inside the nests of both M. schencki and M. lobicornis. In the southernmost sites of Collelongo and Campitello, both M. sabuleti and M. schencki colonies were exploited ( Fig. 1; Table S1).
Results of the Two-Proportion Z tests showed that at Bardineto, Col di Tenda and S. Agostino P. rebeli survived significantly better in M. schencki colonies (Fig. 2). At Campitello, we found that estimated survival rates in nests of M. sabuleti and M. schencki were similar, but P. rebeli experienced significantly higher mortality when living with M. scabrinodis (Fig. 2). A similar pattern was found at Collelongo, although we found only 3 infested nests. P. rebeli showed slightly higher estimated survival in M. schencki nests than was expected from the distribution of eggs at Oulx, but the change in proportions was not significant with any of the Myrmica species present.
Patterns in chemical profiles. MDS ordinations of the 75 samples based on the Euclidean distances calculated on relative abundances of 46 CHCs produced a good discrimination between the five ant species and P. rebeli larvae (ANOSIM: R = 0.948, p = 0.001; 2D STRESS value = 0.07).
Pooling the samples from all localities indicated that the averaged chemical profiles of P. rebeli larvae were closer to those of M. schencki (Mean ± SE = 44.86 ± 0.47) than to other ant species (Mann-Whitney U test: p < 0.001 for all comparisons;  Within each species of Myrmica, workers taken from the same site were chemically more similar to each other than to specimens of the same species collected at different sites. This difference was statistically significant for M. schencki (ANOSIM: R = 0.974, p = 0.001), M. sabuleti (ANOSIM: R = 1, p = 0.029) and M. lobicornis (ANOSIM: R = 1, p = 0.029). In the case of M. scabrinodis, the global ANOSIM test was also significant (ANOSIM: R = 0.209, p = 0.038), but none of the pairwise comparisons between the CHC profiles of M. scabrinodis workers from different sites were significant (ANOSIM: p > 0.057). Even the P. rebeli caterpillars from different sites could be distinguished by their CHC profiles (ANOSIM: R = 0.715, p = 0.001), apart from those larvae collected in Bardineto and S. Agostino, which were chemically similar (ANOSIM: R = 0.490, p = 0.057). In general, the average chemical distance between the CHC profiles of M. rebeli caterpillars and Myrmica workers explained the estimated survival of P. rebeli full-grown larvae or pupae within Myrmica colonies (GLMM: estimate ± SE = −0.078 ± 0.029, z = −2.733, p = 0.006; R 2 m = 0.287; R 2 c = 0.420), meaning that when the chemical profiles of P. rebeli pre-adoption larvae were more similar to a certain Myrmica species, the likelihood of finding successful full-grown larvae or pupae within ant colonies was higher.   Table S3).

Discussion
Our results, based on the estimated survival of parasitic larvae or pupae, reveal that the M. schencki adapted race of P. rebeli inhabits most sites across the Italian peninsula, and that the parasite suffered disproportionately higher or total mortalities when adopted into nests of other Myrmica species, which can be classed, sensu Thomas et al. 21 , respectively as secondary or non-host Myrmica species. M. schencki was the commonest species and was used as host at all sites, but in Campitello and Collelongo, the parasite was reared with equally success in nests of M. sabuleti, suggesting the existence of communities with two primary hosts (multiple-host populations). Communities with more than two primary host species were never observed. At Bardineto and S. Agostino, P. rebeli survived exclusively with M. schencki and can be ascribed to single-host populations. But if, as in most studies, we had sampled only for the presence of the parasite as full-grown larvae or pupae, the populations at Col di Tenda and Oulx would have been reported as exploiting multiple hosts; instead, after factoring in the initial distribution of eggs on each site, our results indicate the presence of a primary host, M. schencki and a secondary host (sensu Thomas et al. 21 ), M. lobicornis, at least at Col di Tenda. The pattern of survival rates at Oulx must be interpreted with caution because the differences are not statistically significant. However, the large number (28) of P. rebeli full-grown larvae or pupae found in the nests of M. lobicornis, together with the results found in the other alpine site (Col di Tenda), suggests that this species might be considered a sub-optimal "secondary" host (sensu Thomas et al. 21 ).
It is notable that no surviving full-grown larvae or pupae were recorded from M. scabrinodis colonies, even though at Collelongo and San Agostino 53% and 45% of P. rebeli eggs were respectively laid beside its nests, and a similar proportion of larvae can be assumed 4,28,29 to have been adopted into its colonies. In contrast, M. scabrinodis is the sole host exploited by P. rebeli's sibling species, P. alcon, across most of Europe.
We also confirmed and quantified distinct differences in the optimum turf structure inhabited by the various Myrmica species at these latitudes and altitudes, providing essential knowledge for the future grazing management and conservation of this threatened butterfly. Only at Oulx was there an apparent absence of turf height preference by different Myrmica species. This is possibly explained by the relatively homogeneous micro-topography at Oulx and by the fact that the grassland was rarely grazed in recent years, and is at a transitional stage towards taller turf. We suspect that M. schencki is persisting sub-optimally before being succeeded by M. lobicornis, which is adapted to colder microclimatic niches 31 and towards which P. rebeli may already be adapting its mimetic semio-chemicals.  To our knowledge, this is the first field-study measuring a direct correlation between the CHC similarity of caterpillars and workers and the estimated survival rate of a social parasite within the ant colonies in populations where multiple host species are exploited. The chemical similarity between pre-adoption P. rebeli larvae and Myrmica ants explained a significant proportion (around the 30%) of the variation in the estimated parasite survival (R 2 m = 0.287). Random factors (ant species nested within sampling sites) increased the explanatory power (R 2 c = 0.420) suggesting that the parasite survival within the nest can be linked to other ant species traits and butterfly adaptations (e.g. vibroacoustic emissions). Overall, P. rebeli pre-adoption CHC profiles from all available populations were most similar to M. schencki workers compared to those of other Myrmica species 10 .
Only at Oulx did P. rebeli pre-adoption caterpillars show an analogous level of chemical similarity to M. lobicornis compared to M. schencki, explaining the high number of full-grown larvae or pupae found within M. lobicornis colonies. Nevertheless P. rebeli larvae may lack perfect adaptation to this host during the "full integration" phase when additional mimetic chemicals are secreted once larvae are underground 11,15 , resulting in a lower observed proportion of full-grown larvae or pupae than expected compared with a fully-adapted local race. This is consistent with laboratory experiments in which P. rebeli caterpillars were successfully adopted by various Myrmica species but subsequently experienced severe mortality in all except host-species' colonies 15 , especially when stressed 20 .
It is also important to note that CHC profiles for each Myrmica species, with the exception of M. sabuleti, resulted in single clusters in Fig. 3A. In contrast, CHCs of M. sabuleti from Col di Tenda and Campitello differed significantly, and were more similar to P. rebeli CHC profiles at Campitello where M. sabuleti was identified as one of the primary host species. The differences in the CHC profiles among the M. sabuleti workers from Campitello and Col di Tenda could possibly also result from the two ant populations occupying different niches 32 or belonging to an as yet unidentified cryptic species (a regular occurrence in Myrmica) 31 . Data collected on the turf height indicate that at the cool alpine site (Col di Tenda), M. sabuleti colonies occur in areas where the turf is taller than the patches occupied by the southernmost populations. If there was a single temperature optimum, we would expect them to occupy higher turf (=cooler soil surface temperatures) at warmer sites and lower turf areas at cooler sites. It is also worth noting that the turf height surrounding M. sabuleti nests at the southern sites (Campitello and Collelongo) was not significantly different from that surrounding M. schencki nests, so that they share similar temperature niches and probable diets, which might also affect the CHC profiles and the differences between the M. sabuleti populations 33 . Evidence of regional switches in host specificity in certain Phengaris species is increasing 15 . In Europe, P. rebeli was described as having two forms, one adapted exclusively to exploit M. schencki in Western Europe (Pyrenees, Haute-Alpes), the other dependent on M. sabuleti in Central-Northern Europe (primarily Poland) 4,15,21 , even though the majority of caterpillars are adopted by M. sabuleti in Spain, and vice versa in Poland, due to the major severe depression of host colonies inflicted by the rapacious caterpillars on occupied sites 34 . In each region, the CHC profiles synthesised (as opposed to acquired) by P. rebeli from Polish and Spanish populations differed prior to adoption and, especially, after 4-6 days with ants, and explained the different survival with different Myrmica species 15 . The difference in the CHC profile of the two populations of M. sabuleti may reflect the different niche utilisation and probably diet affecting the ant profiles or the existence of separate ant clades, or cryptic species, as reported by Ueda and colleagues 35 for M. kotokui ants hosting P. teleius and P. arionides in Japan. Further genetic analyses on Italian ant populations are needed.
Rather than being a "sabuleti type" (as exists in Poland 15 ), the high estimated survival of P. rebeli in nests of both M. schencki and M. sabuleti suggests that the population at Campitello and Collelongo are genuinely more generalist [24][25][26] . Multiple-host using P. rebeli at Campitello and Collelongo suggests that, while rare, there are some populations where these social parasites are less specialised 36-38 than reported from the vast majority of populations studied in the past 22 . An alternative explanation could be that we failed to distinguish between two differentiated, specialist host races that live sympatrically on rare occasions but are otherwise cryptic, comparable to the co-existence on many sites of the non-cryptic Phengaris species, P. teleius and P. nausithous, which share the same food plant but generally exploit different species of Myrmica.
How different host races in Phengaris could have arisen is poorly understood. It has been suggested that European populations became isolated during glaciations in refuges in the Southern Alps, Western Hungary and South-Eastern Europe 39 , where different host associations may have evolved in isolation 40 . Re-colonising the continent, the different host types could have given rise to the specialised populations in Western 22,31 and East-Central Europe [36][37][38] . Within such a scenario, the Alps can be regarded as a geographical and genetic barrier for butterflies, separating Italian and French populations. This could explain the lack of a developmental polymorphism on the Italian side in contrast to the strong evidence of this phenomenon in French populations 30,41 .
Alternatively, populations might experience rare host switches during the history of the species, which were too recent to be detected in their mitochondrial DNA 42 . The distribution of the host types over large, non-overlapping regions of Central and Northern Europe suggest that these are rare events, because frequent host shifts would be expected to produce geographic mosaics 5 . The fact that we do not observe multiple-host use, i.e. transient populations, often, suggests transitions are quick. A mathematical model based on the Phengaris system also suggests that multiple-host use can arise in both a transient or stable state, while single host use is the more likely outcome across a wide range of the parameter space 43 . A separate modelling approach also suggested that multiple-host use is more likely on sites where the similarity between the chemical profiles of distinct host ant species is high. This scenario enables the successful exploitation and deception of the hosts, without requiring a super-specialisation of the parasite that instead can evolve an intermediate CHC profile 44,45 .
Apart from microbial parasites, there are rather few host-parasite systems where the variation in the parasite phenotype can be related directly to survival with one or more hosts. Another example where this is possible and has been exploited in numerous studies involves avian brood parasites such as cuckoos 46 . Yet there is enormous interest in a better understanding of species interactions and their degree of specificity, not least because range changes under climate change may cause some species interaction pairs break, while new ones might form 47 , which clearly would be difficult for extreme specialists 48 . Generalism is seen as a costly parasite trait, and apparently generalist species are increasingly found to represent a complex of more specialised cryptic species 49,50 . Yet our study suggests even in species that are specialist throughout most of their ranges, more generalist populations exist 47,51 ; although they may be difficult to identify. If this was true in other specialist parasite species as well, it suggests that they are more resilient to range changes at species level than previously thought and might possess more adaptive potential if selection for generalism predominates. More immediately, however, understanding the host-associations and how stable or adaptable they are, makes a major contribution to the conservation effort of these high priority butterfly species 52,53 . Identifying multiple-host-use populations could provide a potentially important genetic resource, but more importantly an opportunity to study what triggers host switches and what happens during transition periods.

Materials and Methods
Myrmica communities sampling. Egg counts were performed to estimate the relative number of P. rebeli caterpillars that each Myrmica ant colony and species received 22,28,29 . At each site, by the end of July, 30 individuals of the host plant Gentiana cruciata were selected at random, marked and the number of P. rebeli eggs was recorded on each plant. Ant baits were placed at the base of each selected gentian to verify if the plant was visited by Myrmica foragers. A visual search of Myrmica nests was also performed in a 2 m radius from the food plant, which approximates to the foraging area for workers of these Myrmica species 31 . In most cases, only one nest was found in the area around each gentian, therefore all the eggs laid on the plant were referred to a Myrmica sp. nest. Six plants grew within the foraging range of two distinct colonies. When this occurred, they belonged to the same species and we halved the number of counted eggs and attributed them to each nest. All nests were georeferenced and marked in order to identify them the following spring. Since all Myrmica species present at the sites adopt P. rebeli caterpillars with equal alacrity 4 , and because initial larval survival on gentians does not differ on plants growing in different ant niches 28,29 , the number of eggs on plants reflects the frequency at which P. rebeli caterpillars are carried into nests of different Myrmica species in late summer 54 . The following April to June, we pinpointed again all Myrmica nests within a 2 m radius around each of the 30 G. cruciata plants at every site. No ant species substitution occurred, and each nest was identified and located. Nests were excavated and examined for the presence and abundance of P. rebeli full-grown larvae or pupae. After the excavation, the ground and vegetation were restored to as close to the original conditions as possible.
Myrmica species were identified first in the field and a sample of ten workers was collected and preserved for further inspection in the laboratory using keys by Czechowski et al. 55 . The Myrmica species identified at each site are listed in Table S1. Observed egg numbers per plant were compared with the expected values calculated as the average number of eggs per plant multiplied by the number of gentians overlapping with various Myrmica species, using Chi square test.
We used the proportion of eggs associated with nests of each Myrmica species and compared them to the proportion of P. rebeli full-grown larvae or pupae found inside nests of those species in the following spring, providing an estimate of relative survival with each Myrmica species present at each site. We adopt the term "estimated survival" throughout the text to clarify that the survival was not directly measured as in laboratory experiments. Under a Null model that the parasite survives equally well with all Myrmica species, the proportion of eggs associated with each species provides an expectation for the same proportions to be preserved for the full-grown larvae or pupae the following summer. Differences between expected and observed proportions of P. rebeli full-grown larvae or pupae were analysed using Two-Proportion Z tests. Where observed proportions at the full-grown larvae or pupae were smaller than expected, the respective Myrmica species was regarded as "non host", i.e. significantly higher mortality had occurred inside the nest. Similarly, Myrmica species were regarded as hosts when observed proportions of P. rebeli full-grown larvae or pupae were greater than expected. When the estimated survival was higher than expected in more than one species we distinguish two scenarios: (i) if this difference was statistically significant only in one of the two species, the latter was considered as the "primary host" and the other as the "secondary host" sensu Thomas and colleagues 21 . Overall these populations are defined as single-host populations; (ii) if the observed proportions of P. rebeli full-grown larvae or pupae did not differ between the two hosts, both were considered "primary" and the population was described as "multiple-host". Chemical analysis. Phengaris caterpillars infiltrate Myrmica colonies by mimicking the cuticular hydrocarbons (CHCs) used by the ants for nestmate recognition 6,10,11,15,56 . Here, we analysed the CHCs of preadoption caterpillars from five study sites and compared them with the chemical signatures of the Myrmica worker ants available for exploitation.
At Bardineto, Campitello, Col di Tenda, Oulx and S. Agostino we obtained P. rebeli caterpillars by collecting gentians bearing visible eggs. Samples collected in Collelongo were lost before chemical analysis. A plant was only collected if >5 eggs were visible and if there was at least one other plant with a larger number of eggs within a 50 cm range, a strategy that limited the number of samples but minimise the impact on this endangered butterfly. We also collected colonies of every Myrmica species found.
Surface hydrocarbons were extracted from five pre-adoption caterpillars per sample within six hours of leaving their food plant and before any contact with ants. Caterpillars were transferred into a clean glass vial and extracted by submerging them under 200 μl hexane for 20 min. The hexane was decanted and evaporated under N 2 stream until analysis. Similarly, five ant workers from each colony were extracted. Samples were analysed by GC/MS using equipment and methods described by Schönrogge et al. 11 .
Chromatograms were analysed using MSD ChemStation E.02.01.117 (Agilent Technologies) and the area under each peak identified by its ECL (Equivalent Chain Length) index was expressed as the proportion of the sum over the area of all peaks in the chromatogram while mass spectra were analysed by comparing fragmentation patterns 57 . P. rebeli caterpillars and ant species differences within species between sites were compared using Euclidean distances. Samples from all sites were compared using multivariate and nonparametric multidimensional scaling (NMDS) on Euclidean distances between samples. These were analysed further by Hierarchical Cluster Analysis (CA; with group average cluster mode) and the results were combined with the NMDS plots to illustrate sample groups of particular similarity. For each study site Hierarchical Cluster analyses were performed on Euclidean distances using unweighted pair-group average (UPGMA) algorithms and dendrograms obtained. Pairwise differences between species and treatments were assessed using an analysis of similarities (ANOSIM) 58 and the significance of group separation according to their Euclidean distances was tested by Kruskal-Wallis one-way ANOVA and Mann-Whitney U test for pairwise comparisons, having previously confirmed normality and homogeneity of variance of the data. Benjamini-Hochberg procedure was used to control for false discovery rate (FDR = 7.5%) in multiple tests. To examine the effect of P. rebeli chemical mimicry on the estimated survival of parasitic larvae, a Generalized Linear Mixed Model was computed (binomial error term, log-link function) using the glmer function in the R package lme4 59 . The response variable, the estimated survival in ant nests, was calculated as the proportion between the number of P. rebeli full-grown larvae or pupae found in spring and the number of eggs counted the previous summer. Fixed explanatory terms were the average chemical distance between the CHC profiles of P. rebeli caterpillars and each Myrmica species found. Ant species nested within sampling sites were considered random factors. We computed marginal and conditional R-squared values 60 , describing variance explained by fixed effect only, and by fixed and random effects combined. Both values were calculated using the R package 'MuMIn' 61 . All statistics were carried out with PRIMER 6 β (PRIMER-E, Plymouth, UK), SPSS ® package ver. 24 and the software R version 3.4.3.
Myrmica niches on Phengaris rebeli sites. Turf height is correlated with soil temperature and Myrmica ants tend to occupy specific microclimatic niches which can vary depending on the latitude 31,62 . Vegetation height was measured at the entrance of each excavated nest. Kruskal-Wallis one-way ANOVA followed by Mann-Whitney U test were performed with SPSS ® package ver. 24 to determine significant differences among turf height in the surroundings of each of the Myrmica species. Benjamini-Hochberg procedure was used to control for false discovery rate (FDR = 7.5%) in multiple tests. Test for normality and homogeneity of variance showed that the data were appropriate for non-parametric statistics.

Data Availability
The datasets generated and analysed during the current study are available from the corresponding author on request.