Coasting along to a wider range: niche conservatism in the recent range expansion of the Tawny Coster, Acraea terpsicore (Lepidoptera: Nymphalidae)

The Tawny Coster Acraea terpsicore is a highly mobile butterfly that has recently expanded its spatial distribution from South Asia to South‐East Asia and Australia. Here, we determine if the realized climatic niche has changed during the expansion and analyse the geographic pattern of spread in Australia.


| INTRODUC TI ON
The distribution of many species is changing, in response to climate change and habitat deterioration (Chown et al., 2007;Lenoir et al., 2010;Nieto-Lugilde et al., 2015;Parmesan, 2006;Walther et al., 2009). Such changes have been observed in many taxonomic groups, including insects (Forister & Shapiro, 2003;Furlong & Zalucki, 2017;Parmesan et al., 1999), mammals (Moritz et al., 2008), birds (Maclean et al., 2008) and amphibians (Davies et al., 2019;Urban et al., 2008). The climatic niche is the range of environmental conditions in which a species can sustain itself (Broennimann et al., 2007;Guisan et al., 2014), and it can be maintained in the face of a changing climate (niche conservatism), or it might adjust to the new conditions at a location (niche shift) (Broennimann et al., 2007;Guisan et al., 2014;Peterson, 2011;Wiens et al., 2010;Wiens & Graham, 2005). Spatial models predicting species' potential range and changes in species' geographic distribution in response to climate change or habitat alteration often assume that when a species encounters suitable conditions it will be able to persist (Fleishman et al., 2001;Peterson et al., 2001;Sykes et al., 1996), although it is known that when only a small fraction of a landscape is habitable, species might colonize very slowly (Collingham & Huntley, 2000), or extinctions due to Allee effects might occur (Blackburn et al., 2015;Liebhold et al., 2016;Liebhold & Tobin, 2008;Taylor & Hastings, 2005;Tobin et al., 2011).
Many analyses assume niche conservatism when modelling range expansion, especially in the case of non-native species which can be detrimental to biodiversity, industry and human health (Kriticos et al., 2015;Pearman et al., 2008;Wiens et al., 2009;Clark et al., unpublished data). Yet, many invasive animal species are able to occupy climatic conditions in their newly invaded range that differ markedly from those in their original range (Araújo et al., 2013;Broennimann et al., 2007;Guisan et al., 2014;Hill et al., 2013;Tingley et al., 2014), although such climatic niche shifts appear to be rare among terrestrial plants (Petitpierre et al., 2012). Distinguishing niche shift from niche conservatism is an important aspect of understanding the responses that species make to changing conditions, such as climate, and can help predict future speciation and biological invasion events (Lockwood et al., 2005;Fridley et al., 2007;Sax et al., 2007;Kearney et al., 2008;van Klinken et al., 2009;McGeoch et al., 2010;Early & Sax, 2014;Ricciardi et al., 2017;Sutherland et al., 2018). In addition, invasive species are often considered habitat generalists due to their ability to thrive in a wide range of habitats (Marvier et al., 2004;Richardson et al., 2000). In that sense, habitat specialists appear more prone to extirpation, although Franco et al. (2006) showed that both specialists and generalists can be vulnerable to sudden environmental change. It is often hard to predict the future distribution of species with large geographic ranges, especially when the range is expanding. Yet, we have limited information on whether non-invasive expanding species show niche conservatism or niche shift.
It is not entirely clear how A. terpsicore became established in South-East Asia (Indo-China). Braby, Bertelsmeier, et al. (2014) outlined four main hypotheses: (a) the species naturally expanded its range out of India/Bangladesh and colonized Thailand via Myanmar (Burma); (b) the species was accidentally introduced into Thailand from the Indian subcontinent; (c) the species was intentionally introduced into Thailand for the commercial butterfly house industry from which it escaped; and (d) the species always existed in Thailand and Vietnam, but has since become more abundant and widespread as a result of habitat modification.
Given the sudden and massive range expansion in South-East Asia, we suspect the first hypothesis is more likely and this may have been facilitated by substantial habitat loss and modification in Indo-China because the butterfly exploits ephemeral or disturbed habitats. Regardless of the mechanism of initial colonization, the species is now on the move and coasting along to a wider range.
We have three aims: (a) to study the range expansion of A. terpsicore to determine whether it is colonizing a new climatic niche space (niche shift) or maintaining the same niche (niche conservatism); (b) to describe the rate and geographic pattern of its recent spread in Australia; and (c) to establish whether migration is associated with the leading edge of the range expansion.

| ME THODS
Australia has a highly variable and mostly semi-arid climate (Head et al., 2014), and during the past 140 years, there have been 12 incursions by nine species of butterfly, of which only two species-Monarch Danaus plexippus (Zalucki & Clarke, 2004) and Cabbage White Pieris rapae (Jones et al., 1980;Peters, 1970)-have become established (Braby, Bertelsmeier, et al., 2014). At least seven butterfly taxa from the Oriental/Pacific regions have reached the continent as rare vagrants/visitors; however, there are no breeding records of those species (Braby, Bertelsmeier, et al., 2014). Acraea terpsicore represents only the third butterfly species to have recently successfully colonized and become established on the Australian continent.
Acraea terpsicore is a medium-sized butterfly with a reddish-orange (males) or orange-brown (females) colouration, although occasionally females are similar in colour to males. Adults generally fly 1-3 m above the ground but are capable of higher and more powerful bouts of flight (Braby, 2016). It completes several generations annually, and in northern Australia, it is most abundant during the wet season and early-dry season (Braby et al., 2018). Larval food plants include Passifloraceae, Violaceae and Cucurbitaceae (Braby, Thistleton, & Neal, 2014;Kehimkar, 2008;Khot & Gaikwad, 2011).
In Australia, the butterfly mainly utilizes Hybanthus enneaspermus and, in the Northern Territory, occasionally the "Top End" form of stinking passionflower, Passiflora foetida (Braby et al., 2018;Braby, Thistleton, & Neal, 2014). In Queensland, it has been more commonly found breeding on Passiflora spp., including P. aurantia and the Queensland form of introduced P. foetida.
We collated occurrence records from across the known geographic distribution of A. terpsicore and then developed ecological niche models for three different parts of its range: (1) pre-expansion, to present (Australia). These models were then used to investigate the geographic spread into Australia, and to assess whether the species is maintaining its original realized climatic niche or is adjusting to a new niche.

| Occurrence data
Multiple sources were used to build a global database of occurrence records of A. terpsicore (see Supporting Information). Initially, we downloaded records from GBIF (Global Biodiversity Information Facility) and ALA (Atlas of Living Australia), and supplemented these with additional records collated from the literature. However, there were relatively few records from the pre-expansion range (India, Bangladesh and Sri Lanka), and so we sought reports from social media posts by amateur observers ("citizen scientists"). Records were excluded where a specific and identifiable locality was not mentioned. Geographic coordinates were determined by searching atlases and gazetteers, and records were discarded in cases where geographic locations could not be confidently assigned. For locations without geo-coordinates, we estimated the coordinates for the locality using Google Map with a spatial precision of up to 50 km.
Finally, we included personal observations from Bangladesh (SC) and Australia (MFB), as well as those derived from correspondence with naturalists and entomologists (see Acknowledgements). Overall, the total number of spatial (i.e. unique occurrence) records for our dataset was 1,203.
To study the geographic spread of A. terpsicore in Australia, all occurrence records from 9 April 2012 (when the butterfly was first detected) to 31 December 2019 were collated. The data consisted of four main sources: (a) observations, (b) photographs, (c) specimens and (d) the scientific literature. The major source of data was from photographs (e.g. in iNaturalist where there is a robust community review process before a record is considered research grade). Observation records were not strictly moderated beyond the processes embedded in each of the underlying sources, so there is potential for some uncertainty in the data. However, given the distinctiveness of A. terpsicore, and the absence of extreme outliers in the dataset, we consider the error rate in underlying data to be very low and not significant.

| Ecological niche modelling
Ecological niche modelling is a technique that is used to analyse and predict the probable distribution of species in terms of various explanatory variables (Elith et al., 2006(Elith et al., , 2011Guisan & Zimmermann, 2000), and has become a key tool in macroecological studies Hanson et al., 2019;Norberg et al., 2019). Such models have been widely used to predict the distributions of widespread species, as well as species that have been accidentally or deliberately introduced into new geographic areas (Elith et al., 2006(Elith et al., , 2011Guisan & Zimmermann, 2000;van Klinken et al., 2009;Thuiller et al., 2005Thuiller et al., , 2006. We used MaxEnt  in R (version 3.5.3) (R Core Team, 2013) to model the global distribution of A. terpsicore. Prior to model fitting, records were removed from the database using the "CoordinateCleaner" package (Zizka et al., 2019) in cases where: (a) they duplicated an existing record; (b) they had imprecise or invalid coordinates; or (c) they were assigned to a country inconsistent with the position of the coordinates. We used 19 bioclimatic variables concerning rainfall and temperature with a spatial resolution of 2.5 min (~21.63 km 2 ) which were downloaded from the WorldClim database (https://world clim.org/data/world clim21.html).
The model was tuned using three feature class combinations: "L," "LQ" and "LQP"  and two regularization multipliers (1, 2) using the ENMeval package (Muscarella et al., 2014). The regularization multiplier imposes a penalty for including additional parameters, while feature classes control the response curve shape . We selected the best predictive model by calculating the AUC value (area under a receiver operating characteristic-ROC-curve), which shows how well the model fits the data (Elith et al., 2011;Jiménez-Valverde, 2012;Phillips & Dudík, 2008).
If the AUC value lies between 0.7 and 1.0, it indicates a well-fitted model that is substantially better than random (Phillips & Dudík, 2008). The contribution of each variable to the overall model was estimated, and the species distribution model was thresholded by maximizing the sum of the sensitivity and specificity statistics (Liu et al., 2005).

| Climatic niche overlap
As noted above, occurrence records were grouped into three phases, reflecting the major spatio-temporal stages of the range expansion: and calculated the niche overlap in R (version 3.5.3). We built a principal component analysis from the values of the climate variables at each cell (~22 km 2 ) with one or more occurrence points. We removed one variable (BIO 16: Precipitation of Wettest Quarter) from the analyses as there were no values for more than 50% of the cells.
To calculate the percentages of niche overlap between the pre-expansion, early-expansion and late-expansion ranges, we overlapped the ecological niche map from each spatio-temporal region and used the "nicheOverlap" function in the "dismo" package in R (Hijmans et al., 2017).

| Range expansion in Australia
To determine how the spatial distribution of A. terpsicore has changed since it was first detected in Australia near Darwin in 2012, we constructed convex hulls for each yearly accumulated distribution using ArcMap (version 10.7.1). We then calculated the spatial range by accumulating records up to and including each year-thus, the spatial map for 2013 included all the data points for 2012 and 2013; the spatial map for 2014 included all the data points for 2012, 2013 and 2014, and so forth; see Gupta et al. (2020) for the rationale. These polygons included unsuitable habitats because they encompass all the known, inferred or projected sites of present occurrence of the taxon. After developing the convex polygons for each year, we modified the convex hull boundaries using the "Editor" tool of ArcMap according to the criteria developed by Braby et al. (2018) to estimate the geographic range. Areas in the polygon were included if they were located within a specified distance or had intervening larval food plant records. Thus, the distribution was considered to be continuous when spatial records within the geographic range were up to 200 km apart, with or without intervening larval food plant records, or 200-500 km apart, but only with intervening larval food plant records.
Conversely, distribution records were considered to be disjunct or isolated when the closest points were separated by 200 km or more and there were no intervening larval food plant records, or when the closest points were separated by more than 500 km and there were intervening larval food plant records. Different distance rule sets and thresholds were applied to the lower rainfall areas of the semi-arid zone (350-700 mm: 750 km) and arid zone (≤350 mm: 1,000 km) to account for the low sampling effort in these areas. Areas in the ocean were excluded, and areas near the coast (i.e. within 150 km of the coastline) and nearby small islands that fell outside the line joining two distribution records were included in the geographic range, but only if the larval food plant was present or if the butterfly would be expected to occur in the intervening area based on expert opinion.
To calculate the rate of range expansion, we used two different approaches. First, we calculated the centroid of the range polygon for each year (time slice) and then measured the distance from the first record to each successive centroid. Second, we used the accumulated records-the same procedure that we used to construct the convex hull-and then calculated the mean distance from the first record point to the convex hull edges. We then considered the average of these two methods as an estimate of the rate of expansion.

| Climatic niche space
The ecological niche modelling of environmental suitability indicated that the extant distribution of A. terpsicore closely matched each of the three phases of the geographic range ( Figure 1). Models had an AUC of 0.96 for the pre-expansion range, 0.95 for the pre-expansion plus early-expansion range and 0.93 for the overall distribution, indicating excellent performance with low rates of commission error.
Qualitative comparison of the models indicated that the overall model (i.e. based on all occurrence records for the three geographic range phases) best fitted the observed data (Figure 1d). In contrast, the models generated using records from the pre-expansion range only (Figure 1b), or from the pre-expansion and early-expansion range combined (Figure 1c), did not predict the late-expansion distribution as accurately. However, all models (Figure 1b-d) generally under-predicted the range in South-East Asia and Australia, only partially covering the "known" distribution (Figure 1a), suggesting that the species occurs in environmental space that is partially novel in some parts of South-East Asia and Australia, at least in comparison with its pre-expansion distribution in South Asia. These results suggest that the environmental conditions occupied by A. terpsicore in Australia (late-expansion range) are rather similar to those in India, Bangladesh and Sri Lanka (pre-expansion range), but slightly dissimilar to those in South and South-East Asia (early-expansion range).
The principal component analysis showed that a broad range of environmental conditions was occupied across the species' geographic range (Figure 2). The environmental conditions for data points within the early-expansion and late-expansion ranges were nested within the overall conditions for the pre-expansion range.
The niche overlap analysis showed that 83% of the early-expansion and late-expansion distribution fits within the pre-expansion range.

| Range expansion in Australia
Since first being discovered in Australia near Darwin in April 2012, A. terpsicore has expanded its range extensively (Figure 3a).

| Migration
There were 20 records of migration (Table 1) range far beyond its predicted climatic envelope (Figure 1d).
[Correction added on 7 December 2020, after first online publication: the figure citation has been updated.]

| D ISCUSS I ON
Acraea terpsicore has rapidly expanded its range well beyond its original distribution in the Indian subcontinent to colonize South-East Asia and the Australian continent over the past four decades.
Initially, we hypothesized that this butterfly may have shifted or "WA" is "Western Australia," "NT" is "Northern Territory" and "QLD" is "Queensland." (b) Summary statistics of the rate of spread. "Edge distance" is the distance between first record in Australia and each edge of the convex hull, "Mean centre distance." is the distance between the first record and each year's mean centre of the polygon(s), and "Average distance" is the average of "Edge distance" and "Mean centre distance" that A. terpsicore colonizes the Channel Country periodically, breeding temporarily along watercourses but only after good wet seasons.
When predicting the extent of an expansion or invasion, "climate-matching" is a common approach, which is predicated on the species maintaining its original ecological niche when colonizing a new habitat (Peterson, 2003). For example, using museum records and climate data, Peterson et al. (1999) developed climatic niche modelling for 37 sister taxon pairs of birds, butterflies and mammals and found that many species show niche conservatism. More recently, Cardador and Blackburn (2020) showed that alien birds tend to maintain their original (native) niche in their new (alien) ranges. In contrast, Fitzpatrick et al. (2007) showed that when fire ants (Solenopsis invicta) are first introduced and become established in a new area, they invade areas that are similar to their original niches, but then they expand into a new niche, which is colder and drier and mostly dissimilar to their original niche. These limited studies suggest that there may be differences between range-shifting species and invasive species, with niche conservatism predominating in range shifters like A. terpsicore and niche shifting occurring in invasive species. Note we follow the definition of Wallingford et al. (2020) for invasive species, which is "an introduced species (i.e. a non-native species transported to a new ecosystem by humans intentionally or unintentionally) that causes negative ecological, economic or environmental impacts".
From the principal component analysis, most of the bioclimatic variables were strongly correlated, but the variables with the highest correlations were bio7 (temperature annual range, r = −0.96) and bio-4 (temperature seasonality, r = −0.91) for PC1 and bio1 (annual mean temperature, r = −0.95) and bio-10 (mean temperature of the warmest quarter, r = −0.85) for PC2. Both the early-expansion range (South and South-East Asia) and late-expansion range (Australia) overlapped 83% of the pre-expansion (South Asia) climatic niche.
The partial niche shift for A. terpsicore suggests that there might be some implications in terms of changes to ecological process (e.g. competition with Acraea andromacha, herbivory of the native larval food plant and possibly pollination of some plants), which could lead to changes in the realized niche over ecological and evolutionary time (Broennimann et al., 2007).
Considering the extensive logging of tropical forests in South-East Asia, Braby, Bertelsmeier, et al. (2014) hypothesized that deforestation led this species to colonize initially into new areas (disturbed habitats). In the Sundaland biodiversity hotspot (which includes the Malay Peninsula, Sumatra, and Java, as well as Borneo), more than 92% of the original extent of primary vegetation has been lost (Myers et al., 2000). Such land-use changes have led to profound losses of, and threats to, South-East Asia's unique tropical biodiversity. In contrast, for habitat generalists or pioneer species like A. terpsicore, extensive deforestation and agricultural expansion and to a lesser extent, urbanization, during the past few decades may have provided conditions favourable for colonization and ultimately range expansion (Braby, Bertelsmeier, et al., 2014; see also Halsch et al., 2020 regarding the Gulf Fritillary butterfly Agraulis vanillae).
However, we cannot test this hypothesis with our data and analysis.
Breeding records in several different parts of the range in Australia (Kimberly, Top End and northern Queensland) suggests that A. terpsicore is not only well established, but probably breeds throughout most of its geographic range on the continent (Figure 4a) based on the co-occurrence of its major larval food plant Hybanthus enneaspermus.
[Correction added on 7 December 2020, after first online publication: F I G U R E 4 Spatial, breeding and migration records of A. terpsicore in Australia. (a) total occurrence records, 2012-2019, with red triangles representing confirmed breeding records (mating or immature stages), "WA" is "Western Australia," "NT" is "Northern Territory" and "QLD" is "Queensland"; and (b) migration records in 2017, with colours representing month and arrows representing the direction of migration    (Braby, Bertelsmeier, et al., 2014 Acraea terpsicore is a highly mobile species, and migration has been recorded both in the pre-expansion range (India, Bangladesh and Sri Lanka) and late-expansion range (Australia) by several authors (Dunn & Petrie, 2017;Field, 2017;Larsen, 1988;Williams, 1927Williams, , 1930Chowdhury et al. unpublished data). In Australia, available records of migration in 2017 showed that most individuals were moving in a south-easterly direction, aligning with the overall direction of the range expansion. This suggests that migration is involved in the range expansion, perhaps especially at the leading edge of the colonization front, and was not related to Tropical Cyclone Debbie, which occurred well before the migration records in April-June 2017. Most of the migration records of A. terpsicore were female-biased, and on many occasions, only females were observed migrating.
These butterflies did not stop to refuel, search for the larval food plant or exhibit mating behaviour, thus meeting the classic criteria of migration (Kennedy, 1985). To our knowledge, this is the first documentation of a female-biased migration in butterflies. Sex-biased dispersal is not uncommon, with many birds showing female-biased dispersal and mammals showing male-biased dispersal (Handley & Perrin, 2007;Moussy et al., 2013;Trochet et al., 2016). However, sex-biased migration is quite rare, having been recorded among some salmonids (Kitanishi et al., 2012), but not among butterflies.
In summary, we found no evidence of a niche shift in A. terpsicore. Although some partial niche shifting has occurred, this appears to be restricted to the early-expansion range in South-East Asia. While there is much overlap between the original and newly

ACK N OWLED G EM ENTS
We thank Don Franklin for providing comments on the manuscript.
We are most grateful to the following people for providing records of the Tawny

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1111/ddi.13200.

DATA AVA I L A B I L I T Y S TAT E M E N T
We have attached all the spatial records in the Supporting Information, and all the other related datasets that we collected or generated during the analysis will be provided upon request.