The role of biotic interactions in determining metal hyperaccumulation in plants

1. Heavy metal hyperaccumulation (MH) is a rare trait found in plant species that inhabit metal-contaminated soils. Two main hypotheses proposed to explain the selective advantage of MH are the elemental defence hypothesis and elemental allelopathy hypothesis. The elemental defence hypothesis suggests that MH functions as de fence against herbivores while the elemental allelopathy hypothesis suggests that MH acts to inhibit the growth of neighbours. Nevertheless, these hypotheses are not likely to be mutually exclusive. Here, we present the first study to test both hypotheses simultaneously. We examined these hypotheses with the Cd hyperaccu -mulator, Arabidopsis halleri , which inhabits both metalliferous and non-metalliferous soils, thus providing an opportunity to test the hypotheses of both habitats. 2. Arabidopsis halleri plants originating from several populations in both metalliferous and non-metalliferous soils were grown in a greenhouse in


| INTRODUC TI ON
Heavy metal hyperaccumulation (MH) is a rare trait found in some plant species that inhabit metalliferous soils, that is soils with high metal content. Metal hyperaccumulating plants can accumulate heavy metals, such as cadmium (Cd) and zinc (Zn) at concentrations 100-1,000-fold higher than those found in other species, which are far beyond lethal doses for most other plants in their above-ground organs (Boyd, 2007;Boyd & Jhee, 2005;Mohtadi, Ghaderian, & Schat, 2012). It is thus not surprising that this trait has attracted many studies in plant physiology and ecology alike. Interestingly, while some metal hyperaccumulating species are restricted to contaminated soils (Boyd, 2007;Boyd & Martens, 1998), few species can also be found in non-metalliferous soils, where they have also been shown to hyperaccumulate heavy metals (Rascio & Navari-Izzo, 2011). In some cases, MH by plants from non-contaminated soils may even be higher than that of plants from contaminated soils when exposed to metalliferous soils (Bert, Bonnin, Saumitou-Laprade, Laguerie, & Petit, 2002;Bert, Macnair, Laguerie, Saumitou-Laprade, & Petit, 2000;Stein et al., 2016).
Another hypothesis suggested to explain MH is the 'elemental allelopathy hypothesis', which proposes that hyperaccumulation evolved as a strategy to reduce performance of competing neighbour species via release of heavy metals (Boyd, 2004).
Allelopathy has usually been studied in the context of organic compounds (Morris, Grossl, & Call, 2009), but the concept can also apply to the release of inorganic elements by metal hyperaccumulators (El Mehdawi, Quinn, & Pilon-Smits, 2011a, 2011b. Such elemental allelopathy can be achieved either by the decomposition of leaf litter or by the extraction of elements from leaves through rain water, both of which can result in enrichment of the soil in toxic compounds (El Mehdawi, et al., 2011a). Unlike the elemental defence hypothesis, there are very few studies that tested the elemental allelopathy hypothesis, and those that did so had contradictory results (El Mehdawi, Cappa, Fakra, Self, & Pilon-Smits, 2012;El Mehdawi et al., 2011a;Zhang, Angle, & Chaney, 2007). For example, El Mehdawi et al. (2011a) found that the soil around the selenium (Se) hyperaccumulators, Astragalus bisulcatus and Stanleya pinnata, was highly enriched with Se, suppressing the germination and growth of the metal-intolerant plant Arabidopsis thaliana. However, cause and effect of this Se enrichment were not tested, that is the study did not clarify if Se enrichment was indeed produced by the plants or if the plants grew more in patches with high Se. In contrast, Zhang et al. (2007) showed that the nickel (Ni) hyperaccumulator, Alyssum morale, can increase Ni concentration in its surrounding soil, but this increase had no effect on neighbouring plant germination. Yet, this study was conducted on metalliferous soils, where neighbouring plants are likely to be metal-tolerant. However, MH is likely to offer a much greater selective advantage in non-metalliferous soils, where neighbour plants were not exposed to heavy metals. Unfortunately, no previous study has compared elemental allelopathic effects between populations from metalliferous and non-metalliferous soils.
The elemental defence hypothesis and the elemental allelopathy hypothesis have been examined separately in varying hyperaccumulating species. However, these hypotheses are not likely to be mutually exclusive: both the need for herbivore defence and neighbour inhibition could jointly select for the hyperaccumulation of metals in plants, especially in plants from non-metalliferous soils. Therefore, a comprehensive test of the two hypotheses should include populations from metalliferous versus non-metalliferous soils as well as the response of neighbours from these origins. Our study was designed to fill these research gaps.
We compared the predictions of the two hypotheses across populations originating from both metalliferous and non-metalliferous soils using several interrelated experiments and observations with the metal hyperaccumulating plant Arabidopsis halleri.
First, we compared the capacity of MH between metalliferous and non-metalliferous populations by cultivating them in soils with or without Cd. Using leaf extracts from the same plants, we then studied both the elemental defence and elemental allelopathy hypothesis by exposing herbivores and co-occurring neighbours from natural populations. Additionally, we conducted a field survey to compare natural herbivore loads between A. halleri and neighbours from metalliferous versus non-metalliferous soils.

| Plant and soil collection
We focused on the model metal hyperaccumulating plant A. halleri. This clonal species occurs at a number of metal-contaminated and non-contaminated sites, mostly across Europe (Bert et al., 2002), and both ecotypes have the property of MH (Bert et al., 2002). A. halleri individuals for all experiments were collected in January 2014 from four metalliferous and four non-metalliferous sites within Germany (Table 1). Twenty individuals were collected per site in a haphazard manner with a minimum distance of 2 m and a maximum distance of 150 m between individuals, to ensure they belonged to different genets. The collected individuals were planted in 1 L pots filled with standard potting soil (Topferde, Einheitserde, Gebr. Patzer GmbH & Co. KG) and placed in a greenhouse in Tübingen University, Germany. In order to avoid maternal effects due to metal remains in plant tissues, the plants were clonally propagated for four generations until the beginning of the experiment for which new cuttings were obtained from the propagated clones.
The soil used in the experiments was collected from the same metalliferous and non-metalliferous sites where A. halleri was sampled (Table 1), at a depth of 30 cm from three different locations within each site. All soils from the same habitats (metalliferous or non-metalliferous) were mixed, sieved (2 mm mesh size) and steam-sterilized for 2.5 hr at 80°C to destroy the seed bank and remove potential pathogens in the soil. Metal concentration analyses conducted in a parallel study (Gruntman et al., 2016) confirmed our assumption that Cd concentration was markedly greater for metalliferous soils compared to non-metalliferous soils (3.04 vs. 0.71 µg/g dry soil, respectively). Therefore, these soils are hereafter referred to as high-Cd or low-Cd soils respectively.

| Cd accumulation experiment
This experiment was performed in order to learn whether A. halleri ecotypes from metalliferous versus non-metalliferous soils differ in their Cd accumulation. Two newly grown ramets of A.
halleri were selected and severed from each of the 10 randomly selected mother plants per population. The ramets were grown in water-filled containers in the greenhouse to induce root formation. After 2 weeks, they were transplanted into pots with either high-Cd or low-Cd soils to obtain contrasting Cd leaf concentra- After 6 months, six to eight leaves per plant were harvested and analysed for their Cd concentration. The leaf extracts were prepared with the same methodology described in Gruntman et al. (2016) and analysed with ICP-OES technique for Cd quantification (Stein et al., 2016). The same plant individuals used in this experiment were also used in the following herbivore feeding experiment, seed germination experiments and seedling growth experiments.

| Herbivore feeding experiment
In order to investigate the elemental defence hypothesis, a nonchoice feeding experiment was performed to test whether Cd accumulation deters a leaf herbivore. Caterpillars of Pieris brassicae, a model specialist herbivore of Brassicaceae (Pollard & Baker, 1997), were used in this experiment. Eggs of P. brassicae were obtained from the laboratory of entomology at Wageningen University. The caterpillars were reared on cabbage Brassica oleracea, at 20°C and a 16:8 hr, light:dark cycle.
In October 2016, one randomly selected leaf from each A. halleri ramet grown in the different Cd treatments (see above in Cd accumulation) was cut to a 2 cm 2 piece and placed in a Petri dish (5 cm diameter) on moistened filter paper. One 4-to 7-day-old P. brassicae larva (third instar) was placed in each of these Petri dishes for a period of 48 hr. The caterpillars were not starved before the experiment, as starvation could decrease food preference (Bernays & Chapman, 1978). These experiments were conducted in a greenhouse (24°C, 16:8 hr, light:dark) with 10 replicates for each treatment resulting in 160 Petri dishes (2 Soil types [high-Cd, low-Cd] × 2 A. halleri ecotypes [metalliferous, non-metalliferous] × 4 populations × 10 individuals). At the end of the experiment, the remaining leaves were photographed, and the percentage of leaf area consumed was quantified using the software Adobe Photoshop (CC 14.0). If the leaves were damaged along the edges, approximate leaf edges were added to the image.

| Field herbivory survey
In order to learn if A. halleri plants are more susceptible to herbivory in metalliferous versus non-metalliferous soils, a field herbivory survey of A. halleri and its neighbouring species was performed. The survey was carried out between August and September 2016 at the same four metalliferous and four non-metalliferous sites used

| Seed germination experiments
In order to investigate the elemental allelopathy hypothesis, two germination experiments were performed. In August 2016, fresh leaves from each A. halleri ramet grown in the high or low-Cd treatments (see above) were harvested. Leachates were prepared by soaking the crushed leaves in water for 72 hr (a tissue-to-volume ratio of 0.1 g/ml) and subsequently filtering the liquid through a vacuum pump to remove any solid particles. The leachates were stored in 4°C and analysed with ICP-OES technique for Cd quantification.
The first experiment used commercial seeds of five species, which co-occur with A. halleri, particularly in non-metalliferous soils: Knautia arvensis (Dipsacaceae), Trifolium repens and Lotus corniculatus (Fabaceae), Potentilla erecta (Rosaceae) and Pimpinella saxifraga (Umbelliferae; Rieger-Hofmann GmbH, Blaufelden). In September 2016, the seeds were sown in 5-cm diameter Petri dishes with filter paper (16 seeds of the same species per dish). The seeds were treated with either high or low-Cd A. halleri leachates (10 ml per watering) from both ecotypes. The Petri dishes were placed in the greenhouse with a temperature of 23°C. Germination success of the seeds was estimated by recording the germination fraction after 2 weeks.
Two experiments were performed. The first experiment used commercial seeds (see above in the seed germination experiment).
In May 2017, 40 seedlings per species were transplanted into a germination-tray cell (24 cm 3 volume). Once in every week, seedlings were treated with leachates from leaves of A. halleri that grew in either high or low-Cd soil (10 ml per watering). The trays were placed in the greenhouse with 25-35°C. After 28 days, seedling performance was measured as above-ground biomass, which was determined by harvesting and drying the plants at 70°C for 48 hr. This experimental set-up resulted in a total of 200 seedlings (2 A. halleri ecotypes × 10 individuals × 2 soil types × 5 species).
The second germination experiment used seeds of neighbouring species collected in the field (see above in the seed germination experiment). The experiment took place in June 2017 using the same experimental setup as in the first seedling experiment, which resulted in 800 seedlings (2 A. halleri ecotypes × 2 soil types × 2 ecotypes of neighbour species × 5 species × 5 neighbour individuals).
During the experiment, 54 individuals died within 3 days and were therefore excluded from the analyses. However, survival did not differ between ecotypes or treatments (Table S2).

| Data analysis
All analyses were GLMMs with a normal distribution and an identity link function. For all analyses, differences between treatment groups were analysed using post hoc pairwise comparisons using the false-discovery rate correction (Benjamini & Hochberg, 1995). IBM SPSS Statistics 22 was used for all the statistical analyses.

| Cd accumulation
When growing in high-Cd pots, A. halleri ramets accumulated Cd in their leaves to concentrations exceeding the threshold for Cd hyperaccumulation (100 µg/g; Table 2, soil type; Figure 1A). However, there was no difference in leaf Cd accumulation between A. halleri ramets from metalliferous and non-metalliferous soils ( Table 2, ecotype; Figure 1A).

| Herbivore feeding experiment and field herbivory survey
In the feeding experiment, P. brassicae caterpillars consumed a higher percentage of leaves from A. halleri ramets that grew in low versus high-Cd pots (Table 3, soil type; Figure 2A). This effect was similar for leaves of A. halleri ramets originating from metalliferous versus non-metalliferous soils (

(B)
In the field survey, in both the metalliferous and non-metalliferous sites, individuals of neighbouring plant species incurred higher herbivore damage compared to A. halleri individuals (Table 3, plant type; Figure 2B). Moreover, herbivore damage was lower in A. halleri individuals from the metalliferous compared to the non-metalliferous sites ( Figure 2B), but there was no difference in herbivore damage between neighbours from metalliferous and non-metalliferous soils (Table 3, site × plant type; Figure 2B).

| Seed germination experiments
Leaf leachates of A. halleri ramets that grew in high-Cd pots had a greater Cd concentration compared to low-Cd pots (Table 2 soil type; Figure 1B). Additionally, Cd concentration was higher in leachates of A. halleri originating from non-metalliferous compared to metalliferous soils (Table 2, ecotype × soil type; Figure 1B).
The germination of both commercial and field-collected seeds was similar when treated with leaf leachates of A. halleri ramets that grew in high versus low-Cd pots (Table 4, soil type; Figure 3). However, there was a greater negative effect of A. halleri from non-metalliferous soil on the commercial seed germination (Table 4, A. halleri ecotype; Figure 3A) but not on the germination of field-collected seeds (Table 4; Figure 3B).

| Seedling growth experiments
Leaf leachates of A. halleri plants from high-Cd pots had a greater negative effect on seedling biomass compared to plants from low-Cd pots for both commercial (Table 4, soil type; Figure 4A) and field-collected species (Table 4, soil type; Figure 4B). Moreover, this negative effect was higher for A. halleri from non-metalliferous compared to metalliferous soils, for both the commercial species (Table 4, ecotype × soil type; Figure 4A) and the field-collected species (Table 4; Figure 4B). At the same time, for seedlings of the field-collected species, neighbours originating from non-metalliferous soils were more negatively affected by leachates from high-Cd pots compared to neighbours from metalliferous soils (Table 4, neighbour ecotype × soil type; Figure 4B).

| D ISCUSS I ON
Our overall results support both the elemental defence and elemental allelopathy hypothesis and suggest that both the need for herbivore TA B L E 3 Results of GLMMs used to test the effects of Arabidopsis halleri ecotype (metalliferous vs. non-metalliferous soils) and soil type (low vs. high-Cd pots) on the percentage of leaf eaten by Pieris brassicae in the herbivore feeding experiment, as well as the effects of site (metalliferous vs. non-metalliferous soils) and plant type (A. halleri vs. neighbours) on leaf herbivore damage in the field herbivory survey. Population was used as a random factor. Significant values are indicated in bold. F is for the fixed effects and Wald Z for the random factor Metal Non-metal defence and neighbour inhibition may jointly select for hyperaccumulation of heavy metals in plants. Interestingly, our results also suggest that these selection pressures may differ between populations growing in metalliferous and non-metalliferous soils, as neighbouring plants from metalliferous soils were more tolerant to the allelopathic effects of Cd leachates compared to neighbours from non-metalliferous soils.
Our findings showed clear evidence that Cd accumulation in the leaves of A. halleri can deter feeding by a specialist herbivore. Although it cannot be excluded that other toxic anti-herbivore compounds, such as glucosinolates, could have affected P. brassicae's feeding, these compounds were not likely to be affected by Cd accumulation in A. halleri leaves. This is because, as shown by Kazemi-Dinan, Sauer, Stein, Krämer, and Müller (2015), no trade-off or correlation exists between Cd accumulation and glucosinolate production in the leaves. In addition to our feeding experiment, the results of our field survey also provide support for the role of Cd in herbivore defence. These results show that at both metalliferous and non-metalliferous soils, A. halleri individuals incurred lower herbivore damage compared to neighbouring species. It should be noted however that the standing levels of herbivory in the field might not truly reflect defence levels in plants as some herbivores might have coevolved tolerance mechanisms to them. In addition, Cd concentrations in the leaves of A. halleri and its neighbour species were not quantified in this field study. Despite these limitations, the greater herbivore protection in A. halleri leaves might be attributed to hyperaccumulation of Cd at both origins. Interestingly, former studies have also shown that A. halleri growing in non-metalliferous habitats can accumulate Cd at concentrations above the threshold for Cd hyperaccumulators (Kazemi-Dinan, Sauer, et al., 2015;Stein et al., 2016;Stolpe, Krämer, & Müller, 2017). Our results might therefore provide support for the notion that MH could be selected for as herbivore protection also in uncontaminated soils. In fact, selection for this trait may actually be stronger in such habitats because on the one hand, soil metal contents are low and thus higher accumulation rates are needed to obtain high-leaf metal contents. On the other hand, one may also expect that herbivores are less adapted, and are thus more susceptible to heavy metals in leaves. These results require further field studies that will correlate the quantity of heavy metals accumulated by plants with the herbivore damage they experience in both metalliferous and non-metalliferous soils.
Our results also provide evidence in support of the elemental allelopathy hypothesis, which suggests that next to defence, the adaptive value of MH is the inhibition of neighbours via release of inorganic elements such as Cd and Zn El Mehdawi et al., 2011a;Zhang et al., 2007;Zhang, Angle, Delorme, & Chaney, 2005). Namely, leaf leachates from A. halleri ramets that grew in high-Cd pots had a higher inhibitory effect on the growth of heterospecific neighbours, irrespective of their origin. As for the herbivory results, it cannot be excluded that secondary compounds produced by A. halleri had allelopathic effects in addition to Cd, but these effects were not likely to interact with that of Cd (Kazemi-Dinan, Sauer, et al., 2015). Interestingly, the allelopathic effect of Cd was higher for A. halleri ramets originating from non-metalliferous than metalliferous soils, particularly when tested with commercial seedlings. To the best TA B L E 4 Results of GLMMs used to test for the effects of Arabidopsis halleri ecotype (metalliferous vs. non-metalliferous soils), neighbour ecotype (metalliferous vs. non-metalliferous soils) and soil type (low vs. high-Cd pots) on germination percentage and seedling biomass of commercial and field-collected species in the seed germination and seedling growth experiments. Species and population were used as random factors. Significant values are indicated in bold (na indicates values not applicable for the particular model) of our knowledge, this is the first study to demonstrate the differences in elemental allelopathy between populations inhabiting metalliferous and non-metalliferous soils. Interestingly, the difference in allelopathic effect between the two ecotypes of A. halleri was due to higher concentration of Cd in the leachates but not in the leaves of plants from non-metalliferous compared to metalliferous soils. These differences could be attributed to different mechanisms of Cd sequestration that might be employed by these ecotypes. For example, in a previous study with A. halleri from the same populations, metalliferous populations had a higher Cd tolerance compared to non-metalliferous populations (Gruntman et al., 2016), suggesting that Cd sequestration in the cells might be more efficient in metalliferous populations. Similarly, Meyer et al. (2015) found that in non-metalliferous populations of A. halleri, drastic modifications of the shoot cell wall occur due to high-Cd toxicity, and suggested that in these populations, Cd might not be sequestered in specific compartments such as vacuoles but stored in spaces outside the plasma membrane (apoplast; Isaure et al., 2015;Meyer et al., 2015).
Here, we suggest that perhaps due to different detoxification strate-  (Dechamps et al., 2007(Dechamps et al., , 2008. Our results therefore highlight the importance of incorporating both the origin of the focal species as well as its neighbours when studying ecotypic differences in the evolution of allelopathic potential. The allelopathic effects could only be detected on seedling growth but not on seed germination. Seed germination and seedling establishment are the main plant phenological stages usually affected by allelochemicals (Fernandez et al., 2013;Linhart et al., 2015;Vivanco et al., 2004). The lack of inhibition of seed germination in our study could be due to a potential protection provided by the seed coats of the studied species (Mohamed-Yasseen, Barringer, Splittstoesser, & Costanza, 1994). This protection may also explain why among the very few studies that have investigated the elemental allelopathy hypothesis (Zhang et al., 2005;2007) (2000) showed that a non-metalliferous population of A. halleri exhibited higher Zn transport rate than a metalliferous population, suggesting that in long-term experiments, differences in metal accumulation between ecotypes might diminish, as shown in our study with Cd. This notion could also be supported by a former study whose duration was 14 weeks and revealed no differences in Zn accumulation between A. halleri populations from different ecotypes (Macnair, 2002).
In summary, this study is the first to show that both the need for herbivore defence and neighbour inhibition could jointly select for MH in plants, and potentially to a greater extent in non-metalliferous soils, where neighbouring plants have not developed adaptations to heavy metals. Interestingly, plants from non-metalliferous soils release more Cd in their leachates providing it with greater selective advantage against neighbours that are not tolerant to heavy metals. These results emphasize the importance of including different origins and populations of both the target species and its neighbouring plant species when studying the ecological role of MH. Our results call for additional studies that will simultaneously examine other hypothesized non-mutually exclusive roles of MH, such as enhanced drought tolerance or cation uptake (Boyd & Martens, 1998). Moreover, generalization of our results and their relevance should be further studied with other metal hyperaccumulators, particularly as the vast majority of these plants accumulate Ni rather than Cd (Pollard, Reeves, & Baker, 2014). Nevertheless, the results of this study indicate that some chemicals or secondary compounds can have multiple roles in plants under varying stresses, which may increase the selection pressure for their production or uptake.

ACK N OWLED G EM ENTS
We are grateful to U. Correspondence and requests for materials should be addressed to A.M. (Anubhav.mohiley@uni-tuebingen.de).

DATA AVA I L A B I L I T Y S TAT E M E N T
The source data for this manuscript are deposited in the Dryad Digital