Environmental stress influences Malesian Lamiaceae distributions

Abstract Dual effects of spatial distance and environment shape archipelagic floras. In Malesia, there are multiple environmental stressors associated with increasing uplands, drought, and metal‐rich ultramafic soils. Here, we examine the contrasting impacts of multifactorial environmental stress and spatial distance upon Lamiaceae species distributions. We used a phylogenetic generalized mixed effects model of species occurrence across Malesia's taxonomic database working group areas from Peninsular Malaysia to New Guinea. Predictor variables were environmental stress, spatial distance between areas and two trait principal component axes responsible for increasing fruit and leaf size and a negative correlation between flower size and plant height. We found that Lamiaceae species with smaller fruits and leaves are more likely to tolerate environmental stress and become widely distributed across megadiverse Malesian islands. How global species distribution and diversification are shaped by multifactorial environmental stress requires further examination.

Here, we examine whether multifactorial environmental stress influences Malesian plant distributions. A recent definition states that stressors are "any deviation in the value of an external environmental … variable from the range of values that is favorable for … an entity" (Love and Wagner, 2022). Here our "entity" is plants, and the "deviation" in question are differences from a seasonal wet tropical lowlands either with altitude, drought, or metal-rich ultramafic soils.
The "favorable" lowland wet tropics enable plant communities to achieve both high biomass and their greatest levels of productivity (Cleveland et al., 2011;Shenkin et al., 2019). Crucially, this simple definition can be applied to large-scale studies. This differs from the definition used in fine-scale studies of populations that focuses upon differences from optimum conditions in stressful environments (Love and Wagner, 2022;MacLean et al., 2013), an approach that is less tractable across large scales and many taxa (McGill, 2019). The three stressors focused upon here all have documented examples of how they damage plant function (Zandalinas et al., 2021). First, drought causes hydraulic failure, carbon starvation, and increased pathogen attack and herbivory (Anderegg et al., 2015;Anderegg et al., 2012;Choat et al., 2018;Fensham et al., 2009;McDowell et al., 2008;Powers et al., 2020). Second, ultramafic soils with toxic high metal content damage enzymes, DNA, and cell membranes (Küpper and Andresen, 2016;Singh et al., 2013). Ultramafic soils also have low P, K, and Ca -all key nutrients for plant growth (Proctor, 2003).
Traits should influence how plants overcome dispersal distances and environmental stress (Crayn et al., 2015;Grime, 1977;Ottaviani et al., 2020;Schrader et al., 2021;van Steenis, 1962;Yap et al., 2018). There has been evidence from island systems that traits may vary consistently dependent upon the traits of close relatives on the mainland. Known as the "island rule"; traits of insular species with large relatives on the mainland show lower trait values, whereas species with small mainland relatives show increases (Biddick et al., 2019). These changes depend upon the trait in question but also upon the environmental conditions of islands (Biddick et al., 2019;García-Verdugo et al., 2019). By focusing upon the environmental drivers of species traits, we can hypothesize how traits influence both inter-island dispersal and toleration of environmental stress. Leaf size and height in tropical ecosystems generally declines when species are better adapted toward stressful environments (Fajardo et al., 2019;Wright et al., 2017). Smaller leaves are less at risk of extreme water loss via transpiration (Wright et al., 2017) and shorter species have smaller conduit size that reduces chance of embolism-linked death (Olson et al., 2018). Fruit size likely follows a similar pattern whereby low-productivity high-stress environments limit the production of large, high-energy cost fruit (McConkey et al., 2022;Moles et al., 2007). Smaller fruit, however, could also promote dispersal because they can be consumed by both small and large frugivores meaning dispersal is possible via a greater number of agents (Chen and Moles, 2015;Green et al., 2022). Smaller flowers may be advantageous for drought stress tolerance because they lose less water via transpiration (Galen, 1999). Alternatively, because small flowers are more likely to be resource cheap and short-lived than larger flowers, they could promote dispersal by enabling fast reproduction in newly occupied habitat (Roddy et al., 2021). Small flowers can also be high cost and therefore long-lived, attracting a greater range of pollinators, increasing the chance of successful pollination in a new area (Roddy et al., 2021). Some traits, therefore, could support both the tolerance of environmental stress and longdistance dispersal. To address this, we compare how these traits improve the chances of overcoming either stress or the distances between islands. The results will help determine how these traits shape plant distributions at large scales.

| Study area
Malesia (10° S -19° N, 94° E -151° E) is the region spanning the countries of Malaysia, the Philippines, Indonesia, Timor Leste, and Papua New Guinea. Other than mainland Peninsular Malaysia, the region consists of islands separated by seas. Climate varies from the wet tropics to drought-prone seasonally dry tropics. Altitude shapes habitats from mangroves at sea level to alpine mountain peaks. The region is also home to the tropics' largest area of ultramafic soil/rock which outcrops across most islands (Galey et al., 2017;Garnica-Díaz et al., 2022).

| Malesian Lamiaceae traits
The dataset (Table S1) Figure S2). For simplicity, herein we refer to the two axes as (1) leaf and fruit size and (2) flower size vs height.

| Environmental stress
Environmental stress was a single PC axis responsible for covariation in increasing ultramafic soil area (PC axis loading = 0.52) and Malaysia, Sumatra, Borneo, Philippines, Java, Sulawesi, the Lesser Sundas, Moluccas, and New Guinea (Brummitt, 2001). This axis represented 53% variation in the environmental variables ( Figure S3).
Lowland area was the percentage area below 400 m. A 400 m cutoff was chosen because at these altitudes there are noticeable shifts in plant traits and in certain locations in Malesia, montane flora is observed (Holthuis and Lam, 1942;Trethowan, 2021;Umaña and Swenson, 2019). Ultramafic soil area was estimated as the percentage covering tdwg areas from the map presented in Galey et al. (2017). There is currently not an ultramafic soil layer available.
Minimum monthly rainfall values were taken from WorldClim, these were values recorded from 1970-2000 (Exposito-Alonso, 2017) (Table S2). All these variables were scaled prior to the PC analysis.
The environmental stress values peak at the archipelago's center, in Sulawesi, Moluccas, Lesser Sundas, and the Philippines, where there is least lowland, most ultramafic soils, and the strongest dry season ( Figure 1b).

| Environmental stress and Malesian Lamiaceae distributions
To test the effect of environmental stress upon species occurrence, we built the following phylogenetic generalized mixed effects model (Li et al., 2020): Greek letters above refer to fixed effects and latin to mixed effects (Gelman and Hill, 2006;. Here, Y i represents the observations i of presence/absence n across Malesia's nine tdwg areas m according to Bramley et al. (2019). The logit-transformed probability of species presence p i was modeled as a function of leaf and fruit size and flower size vs height and their interaction with environmental stress and a spatial eigenvector . The spatial eigenvector was the first selected Moran's eigenvector of spatial distance between tdwg centroids (Dray et al., 2012;Griffith, 1996) ( Figure 1b). This involved calculation of a Gabriel neighbor graph between tdwg centroids and subsequent unweighted orthogonalization of the resulting distance matrix (Dray et al., 2012). The intercept estimates species average presence in tdwg areas.
We included two random effects for species identity: one without a spp [i] and one with b spp [i] the covariance in species effects decided by phylogenetic distance between them 2 b Σ spp (Ives, 2018;. spp i connects observations to species. Σ spp represents the n x n phylogenetic distance matrix that assumed a Brownian motion model of evolution and was calculated from a phylogeny built for all species in the dataset (Li et al., 2020). The species random effect without phylogenetic covariance was drawn from a Gaussian distribution with mean 0 and variance 2 . Phylogenetic data were derived from the latest Lamiaceae backbone (Zhao et al., 2021).
Genera missing from the backbone were manually placed, using phytools (Revell, 2012), based upon more finescale phylogenetic studies (Li et al., 2016;Steane et al., 2004). Species not in the backbone phylogeny were randomly imputed alongside congeners to produce a bifurcating tree using pez (Pearse et al., 2015). This is not expected to affect overall variability in phylogenetic distances between species used in the random effect b spp [i] (Li et al., 2019).
To account for more species-rich areas sharing more species simply because of sample size, we took 25 randomly sampled communities the size of the least species-rich area and repeated the model for each of these (Cardoso et al., 2009;Nash, 1950;Stier et al., 2016).
The averaged effects and Wald test p values from these models were used to identify significant effects of predictor variables upon species occurrence.
To further explore model behavior, we extracted predicted occurrences of species. This allowed us to examine species-predicted occurrence across the phylogeny, tdwg areas, and the environmental stress gradient.
All analyses were carried out in R version 4.0.2.

| RE SULTS
Phylogenetic generalized mixed effects models of species presenceabsence across tdwg areas showed that increasing fruit and leaf size had a negative effect upon species occurrence across Malesia (mean effect score = −0.17 and mean Wald test p from 25 model iterations < 0.05) ( Figure 1c) and that increasing leaf and fruit size decreased the chance that species occurred in areas of high environmental stress (mean effect score = −0.13 and mean Wald test The contributions of ultramafic soils, minimum monthly rainfall, and lowland area to environmental stress -which is a principal component axis that accounts for 53% variation in these variables. p < .05) (Figure 1c). All other predictor variables had considerably lower effect scores none of which were significant (maximum mean effect score = 0.088 all mean p > .05, Figure 1c). This indicates that species with smaller fruit and leaves are more likely to occur in areas of high environmental stress in Malesia.
Environmental stress effects upon predicted occurrence vary across the phylogeny. The general pattern being species from clades with greater diversity in the tropics tend to have lower predicted occurrence in high-stress environments, except for the genera Vitex and Premna (Figure 2). Species belonging to clades/genera most diverse in temperate and subtropical regions (e.g., Leucas and Salvia) have consistently high predicted occurrence in stressful environments ( Figure 2).

| DISCUSS ION
Species occurrence across island communities is often driven by the spatial distance between them (Ibanez et al., 2018;MacArthur and Wilson, 1963 (Segovia et al., 2020). Similar convergence because of additional stressors may occur (Rillig et al., 2019;Zandalinas et al., 2021).
For instance, in the Neotropics, stressors include altitude in the Andes, nutrient deficiency of white sands, and drought/fire in the seasonal biome -similarities in how they shape biogeography could be sought (Fine et al., 2014;Pérez-Escobar et al., 2017;Segovia et al., 2020;Simon et al., 2009 (Dalling et al., 2012). By gathering trait data from museum specimens to allow sampling of a high percentage of a clade's species, it should be possible to identify whether traits linked to ecological strategies, such as stress tolerance, encourage dispersal events that precede lineage diversification on islands (Cacho and Strauss, 2014;Esquerré et al., 2020;Heberling and Isaac, 2017). The growing understanding of evolutionary relationships for clades that are speciose in Malesia make examination of this achievable (Atkins et al., 2019;Bellot et al., 2020;Kuhnhäuser et al., 2021;Murphy et al., 2020).
In this study, we have analyzed species rather than lineages, and therefore, we are not able to identify insular speciation events that may influence the patterns we observe. For instance, altitude, drought, and ultramafic soils have been implicated as drivers of insular speciation (Garot et al., 2019;Pillon et al., 2014;Steinbauer et al., 2016). To address this, we need greater sampling of Malesian     can increase population phenotypic plasticity (Levis et al., 2020).
Experimental studies also highlight how the effects of stressors are not always consistent between populations or species (Love and Wagner, 2022). Similarly, extratropical clades could experience stress in the warm wet tropics. Therefore, a stress gradient in reverse to that presented in this study could drive diversification of clades with extratropical origins (Baldwin and Wagner, 2010).
This study has not explored human-caused stressors.
Incorporation of these factors alongside climate change predictions will be crucial when modeling future scenarios for the Malesian flora.
How species traits affect toleration of anthropogenic stressors, like we have shown for environmental stressors, may prove useful for predicting change on megadiverse islands. data curation (equal); writing -review and editing (equal).

ACK N OWLED G M ENTS
LAT is supported by NERC/Newton fund grant NE/S007059/1. We thank Richard Olmstead and three anonymous reviewers for comments that much improved the manuscript.

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Rmarkdown document to build the article and all data, including the phylogenetic tree, is available here: https://figsh are.com/proje cts/ Males ian_Mints/ 149389.