UvA-DARE (Digital Academic Repository) Beyond the Last Glacial Maximum: Island endemism is best explained by long-lasting archipelago configurations

Aim : To quantify the influence of past archipelago configuration on present‐day insu‐ lar biodiversity patterns, and to compare the role of long‐lasting archipelago configu‐ rations over the Pleistocene to configurations of short duration such as at the Last Glacial Maximum (LGM) and the present‐day.


General rights
It is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), other than for strictly personal, individual use, unless the work is under an open content license (like Creative Commons).
Disclaimer/Complaints regulations If you believe that digital publication of certain material infringes any of your rights or (privacy) interests, please let the Library know, stating your reasons.In case of a legitimate complaint, the Library will make the material inaccessible and/or remove it from the website.Please Ask the Library: https://uba.uva.nl/en/contact, or a letter to: Library of the University of Amsterdam, Secretariat, Singel 425, 1012 WP Amsterdam, The Netherlands.You will be contacted as soon as possible.

| INTRODUC TI ON
Oceanic islands are among the most dynamic systems in the world: they emerge and submerge; they shrink and expand; and they split and merge.Changes in island geography and archipelago configuration are shaped by geological processes (e.g., plate tectonics, island ontogeny and within-island volcanism) and sea-level fluctuations (Fernández-Palacios et al., 2016).These sea-level fluctuations, driven by glacial-interglacial cycles over the Pleistocene, have influenced all archipelagos and their constituent islands simultaneously.During glacial periods, sea levels were low and archipelago configurations were often very different: islands had larger surface areas than at present, with some islands fused into larger islands.
For example, the Hawaiian islands of Maui, Molokai and Lanai have repeatedly merged to form a single landmass (the Maui Nui complex; Price, 2004).Islands were also less isolated, with their larger areas reducing inter-island distance and with emerging sea mounts forming stepping stones for dispersal (Ali & Aitchison, 2014;Pinheiro et al., 2017;Rijsdijk et al., 2014).In contrast to these glacial periods, during interglacial high sea-level stands islands were smaller and further apart, as some islands were submerged and palaeo-islands fragmented.Glacial-interglacial cycles have followed a recurrent pattern over the Pleistocene with glacial periods spanning a much longer duration than interglacials (Figure 1).Therefore, for most of the Pleistocene, sea levels were lower than today, corresponding to larger and less isolated islands.
It has long been suggested that past archipelago configurations during lower sea levels have influenced the distribution and evolution of insular biota (Heaney, 1985;Mayr, 1944).Recently, attempts have been made to quantify this relationship (Ali & Aitchison, 2014;Fernández-Palacios, 2016;Heaney, Walsh, & Peterson, 2005;Papadopoulou & Knowles, 2017;Rijsdijk, Hengl, Norder, Ávila, & Fernández-Palacios, 2013;Rijsdijk et al., 2014;Weigelt, Steinbauer, Cabral, & Kreft, 2016).However, as Heaney, Balete, and Rickart (2013) noted, emphasis has been on the relatively short-lasting configuration prevailing during the Last Glacial Maximum (LGM) (e.g., Weigelt et al., 2016).The LGM refers to an exceptional and extreme situation (at approximately 21 ka) of maximum sea-level fall within only the most recent glacial-interglacial cycle of the nine cycles occurring over the last 800 kyr.Even when summing up the duration 11 Island Ecology and Biogeography Group, Instituto Universitario de Enfermedades Tropicales y Salud Pública de Canarias (IUETSPC), Universidad de La Laguna (ULL), Canary Islands, Spain of all glacial maxima over the last 800 kyr, together they would constitute only about 2% of the time elapsed (Figure 1).The duration of these glacial maxima may have been insufficient to shape the assembly of island biotas and especially their endemic component.
Similarly, Porter (1989) asked whether short-lasting extremes such as the LGM and the current interglacial might have received undue attention, and recognized the important role of average Quaternary conditions in landscape evolution and geomorphology.The cyclic nature of Pleistocene sea-level fluctuations leads to alternating periods of island separation and connection, and of shrinking and expanding island areas.Some of these periods lasted longer than others, and some sea-level stands occurred repeatedly, while others were more exceptional (Figure 1).Although Porter (1989) suggested the potential relevance of intermediate Pleistocene conditions for understanding biogeographical patterns, we are unaware of any study so far that has quantitatively analysed their relationship.Given that both LGM and current interglacial situations are exceptional and short lasting, it seems important to explore the extent to which present-day distributions of insular biota reflect past environmental conditions that are more representative of the Pleistocene.
Climatic and environmental fluctuations over the Pleistocene affected the extinction, speciation, fragmentation, merging and population size of biota (Blois, Zarnetske, Fitzpatrick, & Finnegan, 2013;Dynesius & Jansson, 2002;Hofreiter & Stewart, 2009).However, the rates of biogeographical processes shaping island biota during Pleistocene climatic fluctuations varied over time and across taxa (Knowles, 2001a(Knowles, , 2001b;;Shepard & Burbrink, 2009;Willis & Niklas, 2004).To complicate matters further, the patterns of island endemism that we observe today might have been shaped by past biotic and abiotic regimes that are difficult to distinguish in retrospect (Figure 2).Despite the difficulty in making generalized statements about the prime response of biota to glacial-interglacial cycles, it is unambiguous that present-day biota consist of a large proportion of species (native and endemic) that have survived several glacial and interglacial periods (Hewitt, 2000;Webb & Bartlein, 1992).Presentday insular species richness might therefore be considered to be the cumulative outcome of varying biogeographical processes operating during different periods in the past (cf.Waldron, 2010;Dynesius & Jansson, 2014).
Volcanic oceanic islands are isolated from continental landmasses and therefore receive few colonists.The corresponding low rate of genetic exchange results in a relatively large number of endemic species confined to a single island (single-island endemics, SIE), or to several islands within the same archipelago (multiple-island endemics, MIE).This contrasts with (non-endemic) native species (N), which also occur outside the archipelago.There are two reasons to suggest that the effect of palaeo-configuration on extant species will be stronger for endemics (especially SIE) than natives (N).First, endemics differ from natives (non-endemic) in their adaptation to the insular environment and archipelagic setting; endemic species have frequently undergone a longer duration of in situ insular evolution (Warren et al., 2015;Whittaker & Fernández-Palacios, 2007;Whittaker, Fernández-Palacios, Matthews, Borregaard, & Triantis, 2017) than non-endemic natives (N) and have therefore experienced glacial-interglacial cycles for a longer period (in the case of palaeo-endemics insular evolution was not the only process, but the idea of experiencing glacial-interglacial cycles for longer periods still applies).Second, taxa frequently exhibit high levels of endemism as a consequence of low levels of gene flow with neighbouring landmasses (Kisel & Barraclough, 2010).Further, low gene flow is often a consequence of a low dispersal capacity.Such poordispersing taxa may be expected to be impacted more profoundly by changes in the geographical configuration of archipelagos than good dispersers (cf.Borges & Hortal, 2009).Such enhanced impact is due to the lower chances of successful colonization of another island and the narrower habitat availability within their dispersal range.However, the degree to which archipelago configuration influences patterns of endemic species richness probably differs among taxa [e.g., resulting from differences in dispersal capabilities (Claramunt, Derryberry, Remsen, & Brumfield, 2012), number of life cycles (Comes & Kadereit, 1998), population sizes and ecological requirements].Because of their isolated nature and high levels of endemism, volcanic oceanic islands are excellent study systems for understanding the role of long-term geographical processes on speciation and species richness, such as glacial-interglacial changes in archipelago configuration (Warren et al., 2015).
Here, we explore to what extent the persistence and recurrence of different archipelago configurations have left an imprint on present-day species richness on oceanic islands of volcanic origin.Archipelago configuration refers to any combination of area and connectedness (or its antonym: isolation) of islands within the same archipelago (palaeo-configuration refers to an archipelago configuration in the past).We focus here exclusively on changes in archipelago configuration driven by sea-level fluctuations, which have affected all islands globally (Norder et al., 2018).Although island bathymetry is also shaped by geological processes (such as volcanic eruptions, uplift, subsidence and erosion), these are not the main focus of our analysis because they are highly island and archipelago specific (Triantis, Whittaker, Fernandez-Palacios, & Geist, 2016;Whittaker, Triantis, & Ladle, 2008).We restrict the analysis to volcanic oceanic islands to avoid the confounding effect of different abiotic conditions and archipelago configuration dynamics among other island types, such as archipelagos of atolls, land-bridge continental shelf islands and continental fragments (Ali, 2017;Fernández-Palacios et al., 2016;Warren, Strasberg, Bruggemann, Prys-Jones, & Thébaud, 2010;Whittaker & Fernández-Palacios, 2007).We focus on two contrasting taxa with generally good availability of data, land snails and angiosperms, because they differ in terms of dispersal capabilities, ecological requirements and endemism level (which, on volcanic oceanic islands, is much higher for land snails than for angiosperms; Groombridge, 1992;Whittaker & Fernández-Palacios, 2007).Specifically, we test three hypotheses, that: (a) the signal of palaeo-configuration is stronger for SIE than for those that have wider distributions (i.e., MIE and N); (b) for SIE, palaeo-configurations that are more representative of the Pleistocene, associated with intermediate sea levels, will have left a stronger signal than extreme configurations of a short duration (such as the LGM); and (c) land snails will be more affected by past archipelago configurations than angiosperms because they have more restricted distributions and often have lower dispersal capabilities.We test all hypotheses against the classical expectation that present-day richness is best explained by current archipelago configuration.

| Islands and archipelagos
In total, 53 volcanic oceanic islands representing 12 archipelagos (Azores, Canary Islands, Cook Islands, Galápagos, Gulf of Guinea, Hawaii, Madeira, Mascarenes, Pitcairn, Revillagigedo, Samoan Islands and Tristan da Cunha) were included.Our criteria for inclusion of an island were: (a) species data were available for both land snails and angiosperms; (b) islands are oceanic and of volcanic origin.

| Species richness data
Species richness data for land snails for each island of the dataset were compiled from existing literature and species checklists (references in Supporting Information Table S1).Infraspecific entities were grouped into their respective specific taxonomic rank.Species status was standardized based on MolluscaBase (2017; https:// www.molluscabase.org).We considered only islands for which complete lists were available.Recorded extinct species were included in the dataset, while species presumed to be introduced were excluded (Cameron et al., 2013;Triantis, Rigal, et al., 2016).Land snails were classified according to chorotype (a group of species with their distribution restricted to a certain region; see Table 1 for an overview of chorotype acronyms) as: native non-endemic (N S ), multiple-island endemic (MIE S ) and single-island endemic (SIE S ).
Angiosperm richness data were obtained from Weigelt et al. (2016) for native non-endemics (N P ) and single-island endemics (SIE P ).Weigelt et al. (2016) also included angiosperm species endemic to past island units at a sea level of −122 m (PIE P ).In a similar way, land snail species endemic to past island units (PIE S ) at various sea levels (see below) were initially calculated.However, for both land snails and angiosperms, the correlation between SIE and palaeo-island endemics (PIE) was r > 0.99, suggesting that for oceanic islands of volcanic origin, this distinction does not provide additional insights (Supporting Information Tables S2 and S3).Therefore, we only consider present-day chorotypes (i.e., N, MIE and SIE classes, but not PIE) for further analysis.

| Palaeo-configuration data
We considered three archipelago configurations (Figure 3; To represent long-term palaeo-configuration at intermediate sea levels (SLI), we calculated two alternative summary measures: SLI FREQ for the most recurrent, and SLI MED for the most persistent sea level.
Both were calculated for the last nine full glacial-interglacial cycles using the estimated duration of interglacials from Tzedakis, Channell, Hodell, Kleiven, and Skinner ( 2012) and sea-level data from Bintanja, van de Wal, and Oerlemans (2005).Over these nine glacial-interglacial cycles (between 787.9 and 11.2 ka), sea levels between −90 m mean sea level (MSL) and −80 m MSL occurred most frequently (16% of the time sea levels were within this interval; Figure 1).We used sea level (although for 1.5% of the last ~800 kyr sea levels were higher; Figure 1).
We obtained the palaeo-configuration of all islands at the respective sea level stands (SLI MED , SLI FREQ , SLL -122 and SLL GM ) from the Palaeo-Islands and Archipelago Configuration (PIAC) database (Norder et al., 2018).For each palaeo-configuration, we calculated delta area (dA; km 2 ) per reference sea level as the log-transformed absolute difference between current area and palaeo-area.Palaeo-connectedness (PC) was calculated for each respective sea level stand as the number of present-day islands that were connected within a single palaeo-island at a lower sea level.For the highest sea level we used the current area (CA) from the Database of Global Administrative Areas (GADM; https://www.gadm.org/version1),as reported in Weigelt et al. (2016).Current isolation (CI) was calculated as the distance to the nearest other island for which species data were available.As island age is known to influence endemism patterns on individual islands and archipelagos (Peck, 1990;Whittaker et al., 2008), we tested for correlation (herein Pearson's correlation) between each of the aforementioned archipelago configuration variables and island age (island ages and sources in Supporting Information Table S4).All correlations were low and non-significant (Supporting Information Table S5 and Figure S1a-h).This is unsurprising; although volcanic and erosional processes show some age-progressive trends (Whittaker et al., 2008), it is problematic to stereotype the consequences of such trends for area and connectivity through sea-level fluctuations.The aim of the current study is to assess biotic responses to sea level driven changes in archipelago configuration, which is a necessary in-between step towards an integrated understanding of the role of archipelago dynamics and complex island geologies in shaping island biodiversity (Borregaard et al., 2017).

| Statistical analyses
All the statistical analyses conducted in this study were implemented within the R statistical programming environment (R Core Team, 2016).To test our first hypothesis ("H1: palaeo-configuration either intermediate or lowest sea levels.We adopted linear mixed models with archipelago identity as random effect to account for non-independence of data due to the underlying archipelagic structure (Borregaard et al., 2017;Bunnefeld & Phillimore, 2012;Triantis, Economo, Guilhaumon, & Ricklefs, 2015).For example, subsidence rates and erosion regimes (which are mainly climate-driven) vary greatly between archipelagos (Triantis, Whittaker, et al., 2016).We fitted the models with the lmerTest R package, which is a wrapper around lme4 (Bates, Mächler, Bolker, & Walker, 2015).
To To make an informed decision about which archipelago configurations were most relevant in ecological terms, we adopted two complementary approaches to indicate which configuration had strongest statistical support: (a) we assessed for each model the total variance explained, and the proportion of variance explained by archipelago configuration; (b) we ranked significant models based on Akaike's information criterion corrected for sample size (AICc).For the first approach, we calculated the marginal and conditional R 2 (Nakagawa & Schielzeth, 2013) for each model per chorotype.We choose these metrics because they are appropriate within a linear mixed model framework (Nakagawa & Schielzeth, 2013).The conditional R-squared (R 2 C ) provides a measure of the variance explained by the full model (fixed and random effects).
The marginal R-squared (R 2 M ) indicates the variance explained by archipelago configuration (fixed effects).The difference between R 2 M and R 2 C was calculated to reflect the variance explained by archipelago identity (see Ibanez et al., 2018 for a similar approach).
For the second approach, we started by selecting those models for which all individual predictors were significant at p < 0.05 to arrive at a set of "suggestive, but inconclusive" models (Murtaugh, 2014).
The remaining models were ranked based on AICc.Although a cut-off rule of Δ AIC > 2 relative to the best model is often used, it is an arbitrary rule and models with a Δ AICc value between 2 and 7 should not be neglected (Burnham, Anderson, & Huyvaert, 2011).It should be noted that AICc is not an absolute measure of fit (Symonds & Moussalli, 2011) but is a metric that balances model complexity and model fit (Mundry, 2011).Therefore, Burnham et al. (2011) recommend inclusion of a metric to quantify how well models perform (we choose R 2 C and R 2 M ).To summarize, we calculated R 2 C to assess goodness-of-fit of the full model, R 2 M to assess the variance explained by archipelago configuration and Δ AICc to assess model parsimony.

| Description of the data
While all islands were larger than today during lowest and intermediate sea levels, each island has a unique area change pattern in response to sea level fluctuations (Supporting Information Figure S2a-c).Consider, for example, these four islands, which today have a similar area of roughly 140 km 2 : Socorro (Revillagigedo), Flores (Azores), Tutuila (Samoan Islands) and Príncipe (Gulf of Guinea).At SLI MED their sizes were respectively 179%, 154%, 224% and 516% of the present day.At SLL GM , the respective values were 219%, 240%, 308% and 951% (cf.Norder et al., 2018).Also PC responded very differently across islands following the same amount of sea-level change.At the median and most frequent sea levels (SLI MED , SLI FREQ ) seven and eight of the 53 present-day islands were connected to another island within their archipelago, respectively.At a sea level of −122 m MSL (SLL -122 ) and during the LGM (SLL GM ), 12 and 13 islands were connected, respectively.Pearson's correlations of the predictor variables PC and palaeo-area range from r = 0.28 to r = 0.55, with the highest values for palaeo-area and PC at the same sea-level stand (Supporting Information Table S6).PC values at different sea levels are strongly correlated, with lowest correlations between SLI MED and SLL GM (r = 0.67) and highest correlations between SLL -122 and SLL GM (r = 0.97).The correlations between palaeo-area at different sea levels show the same pattern: palaeo-areas at SLI MED and SLL GM are least correlated (r = 0.87), while the palaeo-areas at SLL GM and SLL -122 can be considered identical (r > 0.99) for our dataset of 53 volcanic oceanic islands.
For land snails, our data represented 1,903 species, consisting of 1,430 SIE S , 302 MIE S and 171 native species N S .In total, 1,627 SIE P were included in our dataset.Native species richness for angiosperms could not be calculated from the available data because we only had data on species richness per island but no species identities (see Weigelt et al., 2016).For land snails, the mean proportion of each chorotype across islands was 28.7% for N S , 34% for MIE S and 37.2% for SIE S .Mean inter-island chorotype proportion for angiosperms was 92.9% for N P and 7.1% for SIE P .

| The role of archipelago configuration differs between chorotypes and taxa
We found that the variance in species richness that was explained by palaeo-configuration was larger for SIE than for species with a wider distribution, supporting H1 (palaeo-configuration per chorotype).The variance explained by palaeo-configuration (R 2 M of SLL GM , SLL -122 , SLI FREQ , SLI MED ) was 30%-47% for SIE S and 33%-41% for SIE P (Figure 4).For the more widespread chorotypes, the corresponding values were generally much lower: only 2%-3% for N S , 13%-20% for MIE S and 22%-27% for N P .Comparing models in terms of AICc revealed a similar pattern.For SIE S and SIE P some palaeo-configuration models were within ∆ AICc < 7, while for MIE S , N S and N P , there were no significant palaeo-configuration models within this range.These p-values were rather unrestrictive because consistent overdispersion was present across models.As a result, the subset of significant models initially included before AICc ranking was relatively broad.
The largest part of the variance in SIE richness for both taxa could be explained by palaeo-configuration at intermediate sea levels, supporting H2 (intermediate configuration and SIE).Although for SIE S , the model SLI FREQ had the lowest AICc, the largest part of the variance (73%) was explained by SLI MED .Despite this model being outside ∆ AICc < 7, it is the only model in which the variance explained by palaeo-configuration was larger than that explained by archipelago identity (47% and 26%, respectively; Figure 4).Also for SIE P , the model SLI MED explained the largest part of the variance (86%, of which 41% was explained by archipelago configuration and 45% by archipelago identity; Figure 4).In addition, this model also had the lowest AICc.As expected, palaeo-configurations at intermediate sea levels were able to explain a larger part of the variance than extreme configurations of a short duration.The performance of models for palaeo-configuration at lowest sea levels was generally poorer.For SIE S , the variance explained by palaeo-configuration at lowest sea levels (32% for SLL -122 , 30% SLL GM ) was similar to SLI FREQ (33%) but lower than SLI MED (47%); the palaeo-configuration models at lowest sea level were within ∆ AICc < 7 (Figure 4).Just as for SIE S , the variance in SIE P explained by palaeo-configuration at lowest sea level (34% for SLL -122 , 33% SLL GM ) was similar to SLI FREQ (34%), but lower than SLI MED (41%).However, for SIE P , none of the models for palaeo-configuration at lowest sea level was entirely significant.
The directionality of the relationships between the predictors in palaeo-configuration models (current area, CA; delta area, dA; palaeo-connectedness, PC) is consistent across taxa (Figure 5): richness of SIE S and SIE P increase with CA and dA, but decrease with PC.
However, the effect size of CA and dA show opposing patterns for SIE angiosperms and land snails: CA has a larger effect on SIE P , while F I G U R E 4 Performance of different archipelago configuration models for 53 islands in 12 archipelagos for land snails and angiosperms.To explain species richness in both taxa, we considered archipelago configuration models at the following sea levels: lowest [palaeoconfiguration at −122 m mean sea level (MSL), SLL -122 ; and at −134 m MSL, SLL GM ], intermediate (palaeo-configuration at the most frequent sea level, SLI FREQ ; and at the median sea level, SLI MED ) and highest (current area at present-day sea level, SLH CA ; and current area and isolation at present-day sea level, SLH CACI ).The size of each bar indicates the explained variance by archipelago configuration (R 2 M , darker shades) and archipelago identity (R 2 C -R 2 M , lighter shades).The difference in Akaike's information criterion corrected for sample size (∆ AICc) is provided for those models for which all predictors are significant (p < 0.05), or marked "n.s." for models for which not all predictors are significant (the bars of these non-significant models are greyed out) [Colour figure can be viewed at wileyonlinelibrary.com] dA has a larger effect on SIE S (Figure 5).This is partly in line with our expectation that land snails would be more affected by palaeo-configurations than angiosperms ("H3: palaeo-configuration per taxon").
The contrast in effect sizes of CA and dA was consistent for other palaeo-configurations (Supporting Information Table S7).

| D ISCUSS I ON
Our results are consistent with the hypothesis that palaeo-configurations at intermediate sea levels -which are representative of the Pleistocene-have left a stronger imprint on SIE S and SIE P richness patterns on volcanic oceanic islands than extreme archipelago configurations.This suggests that the relatively short-lasting configurations that have prevailed during the LGM are not sufficient to explain endemism patterns on volcanic oceanic islands.

| Palaeo-configuration at different spatial scales of endemism
Our results conform to our first hypothesis that the signal of palaeoconfiguration is stronger for SIE than for MIE and N. The proportion of the variance that could be explained by palaeo-configuration (R 2 M at intermediate and lowest sea level) was indeed larger for SIE than it was for MIE and N. Ranking significant models based on AICc shows that for SIE, both models with palaeo-configuration and current configuration are within the set of most parsimonious models.
In contrast, for MIE S , N S and N P the set of most parsimonious models only contains configurations at present-day (highest) sea level.
However, for these more widespread chorotypes, the variance explained by archipelago configuration was generally low (especially for N S and MIE S ).For all archipelago configuration models across chorotypes (except SLI MED for SIE S ), the largest part of the variation is explained by the identity of the archipelago (random effects in the model), suggesting that other factors besides archipelago configuration (e.g., climate, geological dynamics, distance from the mainland, island age, human impact, etc.) probably play an important role in shaping current diversity patterns on oceanic islands.These findings suggest that archipelago configuration is an important factor related to patterns of single-island endemics but less so for (non-endemic) natives.Furthermore, for SIE, those palaeo-configurations that are representative of the Pleistocene are more relevant than short-lasting configurations (Figure 6).

| Persistence and recurrence of palaeoconfigurations
Palaeo-configuration at lowest (glacial maxima) sea level had a weaker explanatory power on SIE S than intermediate configurations (noting the small ∆ AICc between models for SIE S ), and a weak non-significant effect on SIE P .This supports our second hypothesis that palaeo-configurations that are more representative of the Pleistocene, associated with intermediate sea levels, have left a stronger signal than extreme configurations of a short duration.Our findings contrast with those of Weigelt et al. (2016), who reported that the number of SIE P could be explained by palaeo-configuration at a sea level of −122 m below today.The difference might be explained by the fact that we selected a subset of angiosperm data exclusively from volcanic oceanic islands, thus preventing confounding geological or genetic effects derived from mixing with islands of other geological origins (Ali, 2017;Whittaker & Fernández-Palacios, 2007).On continental fragments in particular, differences in bathymetry lead to dissimilar responses to cyclic sea-level fluctuations.
The granitic Seychelles are a case in point illustrating the drastic area change and PC of continental fragments (Warren et al., 2010).intermediate sea levels for understanding present-day patterns of SIE richness, they do not provide conclusive evidence regarding the best choice of approach.We anticipate that future studies focussed on specific archipelagos (and other taxa) will shed more light on which sea levels are most relevant in a particular archipelagic context.We hypothesize that the answer will depend on the sea-level thresholds at which island area and connectedness change significantly.

| Contrasting roles of palaeo-area and palaeoconnectedness across taxa
Sea-level fluctuations in the past have modified island area and isolation simultaneously.However, when considering these elements of archipelago configuration separately, it becomes clear that SIE richness of both land snails and angiosperms increases with CA and dA, but decreases with PC.Although CA and dA both hold a positive relationship with SIE S and SIE P richness, dA has a larger statistical effect on SIE S , while CA is most important for SIE P (cf.Kreft, Jetz, Mutke, Kier, & Barthlott, 2008).This finding is consistent with our third hypothesis that land snails will be more affected by palaeo-configuration than angiosperms.As already mentioned in the Introduction, this pattern may be explicable in relation to general differences in speciation and dispersal between the two taxa; land snails tend to be able to speciate at smaller spatial scales than most angiosperms (Kisel & Barraclough, 2010).
Hence land snails can produce more SIE in any given island area, and show stronger effects of island area being formerly larger than do angiosperms.Conversely, a greater mobility of plants could connect "would-be endemics" or replace them with fresh colonists and as such reduce the effect of area change.This reasoning seems in line with the chorotype proportions for SIE in our dataset, which are high for land snails and low for angiosperms.
An interesting avenue for future research would be therefore to further explore the underlying mechanisms that might explain the differential response of both taxa to palaeo-area.Our results indicate a negative relationship between PC and SIE richness of both taxa.The decreasing number of SIE S and SIE P with PC might result from higher levels of gene flow, hindering diversification into distinct lineages (cf.Heaney et al., 2005).This agrees with a recent study on the Puerto Rico Bank where repeated connectedness and fragmentation impeded divergence and speciation of ground crickets (Papadopoulou & Knowles, 2017).Alternatively, elevated biotic interchange following climatic fluctuations and geographical rearrangements over the Pleistocene might have resulted in local extinctions (Vermeij, 1991).Weigelt et al. (2016) also found a negative relationship between palaeo-connectedness and SIE P and concluded that this result falsifies the species pump hypothesis, that is, that repeated separation and connectedness drive speciation (Gillespie & Roderick, 2014;Qian & Ricklefs, 2000).However, it may also be explained by the fragmentation of a population of a SIE species on a palaeo-island into subpopulations, changing the Archipelago configuration models containing PC as a predictor performed better for SIE than those containing current isolation.
This suggests that the actual fusion and splitting of islands may be more important as a moderating factor reducing numbers of SIE than the proximity of islands within an archipelago.

| Island-and archipelago-specific factors
Glacial-interglacial cycles over the Pleistocene have simultaneously influenced the geography of all islands globally.However, there are many regional factors shaping differences in insular biodiversity patterns among and within archipelagos.Islands commonly occur in archipelagos that exhibit biogeographical coherence, that is, similar patterns, in species diversity as a result of shared climate, distance from the potential species pool, intra-archipelagic isolation and geological history (Ali, 2017;Heaney et al., 2013;Triantis et al., 2015).In our analyses archipelago identity explained a large part of the variance (random effect in linear mixed models), highlighting the importance of accounting for among-archipelago variation (Bunnefeld & Phillimore, 2012;Cameron et al., 2013).
Regarding the within-archipelago differences, geological dynamics arguably have a large role in shaping island geography and archipelago configuration.For example, geological processes of plate tectonics, volcanism, subsidence and erosion may drive major changes in island geography and archipelago configuration (Borregaard et al., 2017;Carracedo, 2014;Gillespie & Clague, 2009;Gillespie & Roderick, 2002;Price, Clague, Bay, Road, & Landing, 2002;Stuessy, 2007;Whittaker et al., 2017Whittaker et al., , 2008 ) ).While general developmental trends may be identified for particular classes of oceanic islands (Whittaker et al., 2017(Whittaker et al., , 2008 ) ), in practice, island ontogeny and volcanic activity are island specific.For example, the eight main islands of the Hawaiian archipelago show linear age progression from east to west and range in age between 0.5 Ma (Hawaii Island) and 5.1 Ma (Kauai).Hawaii itself is the only island that is volcanically active, all others being disconnected from the hotspot and inactive for at least 0.75 Myr.This contrasts with the complex geological setting of the Azores, with a western group of two islands located on the North American plate, and a central and eastern group (of five and two islands, respectively) located at the junction between the Eurasian and Nubian lithospheric plate (Ramalho et al., 2017), and no linear age progression from one side of the archipelago to the other (Ávila et al., 2016).In addition, some islands in our dataset are younger than the last nine glacial-interglacial cycles (~800 ka) we used to calculate the most frequent and median sea levels.However, due to the recurrent character of sea-level oscillations, later stages of these cycles will nonetheless have affected younger islands.Incorporating glacialinterglacial driven changes in island geography becomes challenging as longer time-scales are considered because they overlap and interact with geological dynamics.For future studies it will be important to include greater detail on regional geological dynamics that have shaped archipelago configuration in the past.

| CON CLUS ION
To our knowledge this is the first time that the effects of long-and short-lasting archipelago configurations on species richness pat-

F
I G U R E 1 Characterization of sea-level fluctuations over the last ~800 kyr at highest, intermediate and lowest sea levels.(a) Sea-level fluctuations over the nine most recent glacial-interglacial cycles covering the period of the last ~800 kyr.(b) The percentage of time over the last ~800 kyr that the sea level was below a certain sea level.(c) The most frequently occurring sea levels are quantified as the percentage of time over the last ~800 kyr that sea levels were within a certain interval [in 10-m bins, e.g., between −90 m mean sea level (MSL) and −80 m MSL].The period of ~800 kyr was chosen because it spans nine full glacial-interglacial cycles (estimated duration of interglacials from Tzedakis et al., 2012).All figures are based on data from Bintanja et al. (2005) [Colour figure can be viewed at wileyonlinelibrary.com]

F
I G U R E 2 Conceptual figure illustrating how observed present-day endemism patterns might result from fragmentation and/or dispersal.The chorotype of a species might change from single-island endemic (SIE) to multiple-island endemic (MIE) as a result of either of these processes (or a combination thereof).(a) SIE on palaeo-island A becomes MIE by fragmentation.(b) MIE shared by palaeo-islands A and B continues to be a MIE (but as a result of fragmentation and dispersal).(c) SIE on palaeo-island B becomes MIE by dispersal [Colour figure can be viewed at wileyonlinelibrary.com] the central value (−85 m) of this interval to represent the most frequent long-term sea level stand (SLI FREQ ).For 32.5% of the time, MSL was below −85 m.Over the same nine glacial-interglacial cycles, the median sea level (SLI MED ) was −65 m MSL (the mean is −64 m MSL).For 50% of the time, MSL was below −65 m; for 9.5% of the time sea levels were between −70 and −60 m MSL (Figure 1).To reconstruct palaeo-configuration during the lowest sea levels (SLL) we used two sea-level stands: (a) the most recent estimate for the LGM low stand (SLL GM ) of −134 m MSL (Lambeck, Rouby, Purcell, Sun, & Sambridge, 2014); and (b) the sea-level low stand of −122 m MSL (SLL -122 ) selected by Weigelt et al. (2016).To represent archipelago configuration at highest sea levels (SLH), we used the present-day Conceptual figure showing three archipelago configurations (highest, intermediate and lowest sea level) that were used in this study, illustrated for one hypothetical sea-level cycle.(a) Conceptual illustration of how sea-level change affects archipelago configuration.(b) The three panels show how sea level (top), area (middle) and connectedness (bottom) change for one specific island.Connectedness is quantified as the number of present-day islands connected in a palaeo-island.Archipelago configurations at intermediate sea level are more representative of this cycle than the highest and lowest sea levels, which both represent an extreme configuration of short duration [Colour figure can be viewed at wileyonlinelibrary.com]per chorotype")-that the role of palaeo-configuration is stronger for SIE than for MIE and N-we fitted linear mixed models separately for each chorotype.In each model, we used archipelago identity as a random effect(Bunnefeld & Phillimore, 2012;Cameron et al., 2013) and used a Poisson error structure.Each of the models for palaeo-configuration consisted of the following fixed effects: current area (CA), delta area (dA) and palaeo-connectedness (PC) at However, the most important reason for the poor performance of models based on palaeo-configuration at lowest sea levels compared to intermediate configurations is probably related to the short lasting and interruptive character of glacial maxima.Intermediate palaeo-configurations were reconstructed at the SLI MED and the SLI FREQ .Palaeo-configuration at SLI MED explained most of the variance (R 2 M ) for both SIE S and SIE P .Also in terms of AICc intermediate palaeo-configurations performed best for SIE; however, for SIE S , SLI FREQ had the lowest AICc, while for SIE P , SLI MED had the lowest AICc.Although both R 2 M and AICc suggest that it is worthwhile to consider palaeo-configurations atF I G U R E 5 Bars indicate effect size of parameters in the best archipelago configuration models for 53 islands in 12 archipelagos for land snails and angiosperms.Standardized effect size of significant parameters (p < 0.05) in the model with lowest difference in corrected Akaike's information criterion (∆ AICc).The colours of the bars represent archipelago configurations at intermediate (dark grey) and highest (light grey) sea levels.Each model contains a selection of the following predictors: current area (CA), delta area (dA), palaeo-connectedness (PC) and current isolation (CI)

F
I G U R E 6 Conceptual figure illustrating the influence of sea-level driven changes in archipelago configuration on species richness patterns of single-island endemics and non-endemic natives.(a) The percentage of time the sea level was either above or below a certain level.(b) The duration of an archipelago configuration as shaped by sea-level fluctuations.(c, d) The width of the bars indicates the importance of an archipelago configuration (b) in shaping single-island endemic and non-endemic native richness [Colour figure can be viewed at wileyonlinelibrary.com] to MIE as sea levels rose towards the current interglacial high sea level (Figure 2).
figurations.Our results suggest that for understanding evolutionary dynamics of insular biota it is relevant to look beyond extreme palaeo-configurations that persisted for only a few thousand years (such as the LGM) and to test for biological legacies of alternative palaeo-configurations.

Table 1 )
: (a) current configuration at the present-day high interglacial sea level (hereafter: "highest") with small island area and small connectedness (i.e., large isolation); (b) palaeo-configuration at intermediate sea levels (hereafter: "intermediate") with intermediate island area and connectedness; (c) palaeo-configuration at minimum sea level (hereafter: "lowest") with largest island area and largest connectedness.
Our findings cor-