The invasive cactus Opuntia stricta creates fertility islands in African savannas and benefits from those created by native trees

The patchy distribution of trees typical of savannas often results in a discontinuous distribution of water, nutrient resources, and microbial communities in soil, commonly referred to as “islands of fertility”. We assessed how this phenomenon may affect the establishment and impact of invasive plants, using the invasion of Opuntia stricta in South Africa’s Kruger National Park as case study. We established uninvaded and O. stricta-invaded plots under the most common woody tree species in the study area (Vachellia nilotica subsp. kraussiana and Spirostachys africana) and in open patches with no tree cover. We then compared soil characteristics, diversity and composition of the soil bacterial communities, and germination performance of O. stricta and native trees between soils collected in each of the established plots. We found that the presence of native trees and invasive O. stricta increases soil water content and nutrients, and the abundance and diversity of bacterial communities, and alters soil bacterial composition. Moreover, the percentage and speed of germination of O. stricta were higher in soils conditioned by native trees compared to soils collected from open patches. Finally, while S. africana and V. nilotica trees appear to germinate equally well in invaded and uninvaded soils, O. stricta had lower and slower germination in invaded soils, suggesting the potential release of phytochemicals by O. stricta to avoid intraspecific competition. These results suggest that the presence of any tree or shrub in savanna ecosystems, regardless of origin (i.e. native or alien), can create favourable conditions for the establishment and growth of other plants.

Two-way ANOVAs indicated that the presence of native trees increased the diversity of bacterial communities, especially in uninvaded soils. Furthermore, all diversity metrics were significantly affected by invasion status, being higher in invaded than in uninvaded soils ( Table 2, Fig. 2). The presence of V. nilotica, S. africana, and O. stricta also significantly altered soil bacterial community structure and composition based on Horn distances (Table 3, Fig. 3).
Analysis of multivariate homogeneity of group dispersions indicated that uninvaded soils were significantly more over dispersed (BETADISPER F 1,16 = 8.76, p = 0.009), suggesting that invasive O. stricta plants have a homogenizing effect on soil bacterial communities, i.e. areas invaded by O. stricta had more similar bacterial communities than uninvaded areas. The effect of tree cover did not, however, have a significant homogenizing effect (BETADISPER F 2,15 = 0.15, p = 0.865).
Our linear discriminant analysis effect size (LEfSe) highlighted numerous OTUs (n = 94) that were significantly more abundant in invaded areas, irrespective of cover type (Fig. 4). This indicates that O. stricta invasion leads to a significant increase in the abundance of these OTUs. On the other hand, only three OTUs (two belonging to Rubrobacter and one to Blastococcus) were significantly more abundant in uninvaded soils.
Germination. The percentage and speed of germination of O. stricta were generally higher in soils collected under V. nilotica canopies and in soils from uninvaded areas ( Table 4). The percentage and speed of germination of V. nilotica were significantly lower under uninvaded soils collected from underneath its own canopy. We found no significant differences in the percentage and speed of germination of S. africana.

Discussion
Woody species in savannas accumulate moisture and soil nutrients, and harbor unique bacterial communities under their canopies, creating "fertility islands" 16 . Accordingly, and in support of our hypothesis, our results show higher soil humidity, nutrient contents, and phosphatase activity under the canopies of V. nilotica and S. africana trees in KNP compared to open patches with no tree cover. Interestingly, the same soil physicochemical properties were also elevated as a result of O. stricta invasion. These results are surprising, since previous studies Table 1. Mean (± SE) of pH, humidity (%), organic matter (%), nitrogen (mg/Kg), phosphorus (mg/Kg), phosphatase activity (µmol/g h), β-glucosidase activity (µmol/g h) and urease activity (µmol/g h) in patches invaded and uninvaded by O. stricta under the canopies of V. nilotica and S. africana and in open patches with no tree cover. Letters indicate significant differences between soils (P < 0.05). www.nature.com/scientificreports/ have shown that cacti in arid ecosystems generally provide little shade and limited amounts of aboveground litter, and therefore, that soils beneath them have similar humidity and nutrient levels than those in bare patches 37,38 . Differences in morphology, physiology, or root symbiosis of woody species can cause differences in the characteristics of the fertility islands they create 39 . Accordingly, we found higher concentrations of soil nutrients and phosphatase activity under V. nilotica canopies than under S. africana canopies. This can be explained by the fact that acacias have rapid growth rates and the capacity to fix atmospheric nitrogen via rhizobium symbioses, and legumes in general have been repeatedly shown to induce changes in soil nutrient levels and cycles [40][41][42][43][44] . Moreover, V. nilotica has previously been found to increase soil nutrient concentrations due to high above-and belowground organic matter input 45 . In particular, the presence of V. nilotica trees in our uninvaded areas caused a significant increase in soil nitrogen content, which is a particularly limiting macronutrient in the study area 10 .
It has also been suggested that, due to leaf litter input, V. nilotica trees increase nutrient cycling 45 . However, we found significantly lower levels of β-glucosidase and urease activities under V. nilotica canopies. These results are unexpected, and more research is needed to unravel the mechanisms underlying these observations.
In agreement with our findings, previous studies have reported the phyla Actinobacteria and Firmicutes as abundant bacterial groups in semi-arid savannas 46 . Actinobacteria taxa are generally drought and heat resistant 47 , while many Firmicutes are spore-forming, an adaptation to harsh and unpredictable environmental conditions 48 .
Our results showed that the presence of V. nilotica, S. africana, and O. stricta altered soil bacterial composition and increased relative bacterial abundance and diversity compared to open patches with no tree cover. This likely reflects the increases in soil humidity and nutrient contents we observed under these plants [49][50][51] . However, Figure 1. Distribution of OTUs (operational taxonomic units) between invasion (invaded and uninvaded; top left) and tree cover (V. nilotica, S. africana, none; top right) categories, together with relative abundances of soil bacterial taxa (Class | Phylum). Class-level relative abundances were calculated using the number of sequences for each taxon as a percentage of the total sequences for each invasion/cover combination. The "Other" category includes taxa that were unclassified at Class level and classes representing less than 0.5% of the total number of sequences. www.nature.com/scientificreports/ increases in bacterial abundance and diversity were more pronounced in O. stricta-invaded soils and O. stricta invasion had a significant homogenizing effect on soil bacterial communities. These results suggest that fertility islands created by O. stricta might have stronger effects on the diversity and composition of bacterial communities of savanna soils than those created by the native woody species. Seeds have various mechanisms to detect suitable conditions for establishment and to adjust their timing of germination accordingly 52,53 . Our results suggest that the germination of O. stricta in the study area might be facilitated by the presence of V. nilotica trees (also see 32,33 ). Such enhanced germination kinetics may be the result of the higher soil nutrient levels present under the canopy of these trees compared with those under the canopy of S. africana or in open patches with no tree cover, which might trigger the germination of O. stricta seeds.
Timing of germination can also be adjusted as a response to chemicals released by conspecific or other plant species 54 . Such mechanisms can help plants to avoid intra-or inter-specific competition or to detect the presence of facilitating or nursing species, maximizing establishment potential 55 . Accordingly, while V. nilotica and S. africana trees appear to germinate equally well in invaded and uninvaded soils, we found O. stricta to have reduced germination performance (lower and slower) in invaded soils. These results suggest the potential release of phytochemicals by O. stricta, which may retard its own germination, i.e. the creation of negative soilfeedbacks, aiming to avoid potential intraspecific competition. Similarly, we also found V. nilotica to have lower   to adjust their timing of germination as a response to chemically-induced signals released by adult plants 56 . Moreover, V. nilotica is known to produce several chemicals, including tannins, flavonoids, and phenolic acids, capable of stimulating or inhibiting seed germination 57 .
Overall, our results showed that native tree cover and invasive species can create fertility islands in savanna ecosystems, causing changes to the abiotic and biotic conditions of soils. This suggests that the presence of any tree or shrub in savanna ecosystems, regardless of origin (i.e. native or alien) or growth from (e.g. woody tree or succulent shrub), can result in the formation of fertility islands, which usually create favourable conditions for the establishment and growth of other plants 20 . We also found the presence of V. nilotica trees to be linked to higher increases in soil nutrient contents than the presence of S. africana trees and that these increases might favour the germination of invasive O. stricta seeds. These results provide a mechanistic basis to previous studies suggesting that establishment of O. stricta in KNP might be facilitated by the microenvironment created by V. nilotica trees 32 . Finally, invasive O. stricta did not affect the germination of native trees, but affected soil bacterial communities more strongly than native trees. Since the eradication of O. stricta in KNP is no longer feasible 30 , the consequences of these effects deserve further investigation.

Methods
All the local and national guidelines were followed in this study. A contract between the researchers and SAN-Parks (permit NOVOA1292) provided all formal permission needed to conduct the research (which fulfils the NEMBA: Protected Areas Act).

Study site.
Our study site was located in the south of KNP (Fig. 5), approximately 1 km south-west of Skukuza Rest Camp (− 25.0049, 31.5852), in a typical savanna vegetation with sparse grass cover, open patches with no tree cover, and a patchy distribution of trees (Fig. 6). Two woody species, V. nilotica and S. africana, make up more than 90% of all trees in the study site, and their individuals are mixed in the landscape. The study site has low grass cover, minimising herbivory by native ungulates and due to the open patchy landscape fire in the immediate area is minimal. Moreover, the study site presents granitic geological substrate and thus has nutrient-poor soils 58 , suggesting a high importance of fertility islands for the establishment of plants in the area.

Study species.
Vachellia nilotica (previously Acacia nilotica 59 ), is a semi-deciduous tree in the Fabaceae (legume) family. It can grow up to 10 m in height and its branches are armed with straight, paired spines. Vachellia nilotica is widely distributed in southern Africa, from Tanzania to South Africa 60 . Spirostachys africana is a spineless deciduous tree from the Euphorbiaceae (spurge) family. It can grow up to 18 m in height, although it usually only reaches about 10 m. Spirostachys africana is also widely distributed in southern Africa 61 .
Opuntia stricta is a perennial, succulent, shrubby plant in the Cactaceae (cactus) family and is native to Cuba, Mexico, and the USA 34 . Commonly introduced around the world as an ornamental plant, O. stricta is currently considered invasive in 21 countries 31 . In its invasive ranges, O. stricta causes multiple ecological and    www.nature.com/scientificreports/ socioeconomic impacts. It reduces agricultural production and biodiversity, causes loss of grazing potential, transforms habitats, and causes injuries to animals and people via its spines 62 .
Soil and seed collection. We collected samples of O. stricta-invaded and uninvaded soils under the canopies of V. nilotica and S. africana and in open patches with no tree cover. The identification of the plant species was undertaken by AN, LCF and JJLR. No voucher specimens were collected or deposited in a publicly available herbarium. In each of these six soil types, we randomly established five plots of 0.5 × 0.5 m. We did not establish more than one plot under the canopy of a single tree. In each plot, we took five subsamples from the top 10 cm of soil using a shovel. The subsamples within each plot were then sieved through a 2.0 mm mesh and homogenized into a single sample, resulting in 30 samples in total (2 invasion per plot), to measure the diversity and composition of soil bacterial communities. Soils used for pH, humidity, nutrients, and germination analyses were kept at room temperature. Soils used for enzymatic activity analysis were refrigerated at 4 °C and analysed within 3 days of collection. Soils used for DNA extraction (i.e. for next-generation sequencing analyses of bacterial communities) were kept on ice during transport and stored at −80 °C as soon as was possible.
Opuntia stricta, V. nilotica and S. africana seeds were also collected in the study area and stored in the dark at 4 °C. O. stricta is declared as invasive in the park, and therefore no permission was needed to collect its seeds. Seeds of the native trees were collected through the Skukuza plant nursery, which has permission to collect and store such seeds, as well as germinate and sell them.
Soil pH, humidity, nutrients and enzymatic activities analysis. Soil pH was determined by dilution with water (1:2.5; soil: distilled water) 63 , using a CRISON GLP 22 + pH & Ion-Meter. To determine soil humidity, we first weighed the fresh soil samples, and then dried (70 °C for 48 h) and reweighed them, calculating soil humidity as: (Fresh soil weight-Dry soil weight)/(Fresh soil weight) × 100. Soil nutrients (nitrogen, organic matter, and phosphorous) were measured by Labserve Laboratories (Nelspruit, South Africa).
We also analysed three enzymes that play key roles in soil nutrient cycling: β-glucosidase (E.C. 3.2.1.21), involved in carbon metabolism through the release of glucose from cellulose; urease (E.C 3.5.1.5), involved in the release of nitrogen by degrading urea to ammonium; and phosphatase (E.C. 3.1.3.1), involved in the release of phosphate from organic matter by hydrolyzing phosphate ester bonds 52 . We used the methods described by Tabatabai and Bremner 64 , Kandeler and Gerber 65 and Allison and Vitousek 66 , for the β-glucosidase, urease, and phosphatase assays, respectively. Following the recommendation of German et al. 67 , we conducted the enzyme assays at environmental pH conditions and within 48 h of soil collection. DNA extraction and next generation sequencing. We extracted whole genomic DNA from 0.25 g of soil using the PowerSoil ® DNA extraction kit (MO BIO laboratories Inc., Carlsbad, CA, USA), following the manufacturer's protocol. We assessed DNA quality using the NanoDrop ND-1000 UV-Vis Spectrophotometer (Nanodrop Technologies, Wilmington, DE, USA). Part of the 16S rRNA gene (consisting of nine hypervariable regions: V1-V9) was targeted for amplification, since it is frequently used for the identification of bacterial taxa 68,69 . The more variable regions are useful for genus-or species-level identifications 70 . We targeted the V5-V7 hypervariable regions using primers 799F (5'-AAC MGG ATT AGA TAC CCK G-3') and 1391R (5'-GAC GGG CGG TGW GTR CA-3'). These primers are known for low non-specificity, and can accurately and reproducibly differentiate species [71][72][73] . Amplification was done with sample-specific barcodes in the forward primer, using a 30 cycle PCR and the HotStarTaq Plus Master Mix Kit (Qiagen, Valencia, CA, USA) under the following PCR conditions: 94 °C for 3 min, followed by 30 cycles of 94 °C for 30 s, 53 °C for 40 s and 72 °C for 1 min, followed by a final elongation at 72 °C for 5 min. We checked the PCR products on a 2% agarose gel to determine the success of amplification and the relative intensity of bands. Multiple PCR samples were then pooled together in equal proportions based on their molecular weight and DNA concentrations. Pooled samples were purified using calibrated Ampure XP beads (Agencourt Bioscience Corporation, Beverly, MA, USA) and were used to prepare DNA libraries following the Illumina TruSeq DNA library preparation protocol. We sequenced the samples using the Molecular Research LP next generation sequencing service (https:// www. mrdna lab. com, Shallowater, TX, USA) on an Illumina MiSeq instrument (Illumina, San Diego, CA, USA) following the manufacturer's guidelines.
Bioinformatics and taxonomic identification. We processed all raw MiSeq DNA sequence data following standard procedures as described in Schloss et al. 74 using the mothur version 1.37.1 75 . First, we removed low quality sequences and optimized the sequence lengths (to between 385 and 395 bp). We then aligned unique sequences to the SILVA-ARB reference database (release 123) to the same region of the 16S rRNA gene we sequenced and removed those columns that contained gaps only. Furthermore, independent of a reference database, we removed all the chimeric sequences using the uchime algorithm 76 and the template as self, i.e. de novo removal. Subsequently, we clustered sequences into operational taxonomic units (OTUs) at the 97% DNA sequence similarity level. Representative sequences for OTUs were chosen as those that were most abundant in each cluster. We determined the taxonomic identity of each OTU with the ribosomal database project (RDP) Classifier 77 , and all sequences classified as chloroplast, mitochondria, and Archaea were removed. In order to standardize the number of reads across all samples, we subsampled (i.e. rarefied) equivalent reads from each www.nature.com/scientificreports/ sample. Rarefaction is believed to increase the false discovery rate 78 , but this is not true 79 and instead, rarefying can lower sensitivity (false negatives) as a result of data discarding. Thus, recommendations are to rarefy to the highest depth possible 80 , which is what we did. Rarefaction is still considered a useful normalization technique, especially for uneven library sizes between groups, like here, and results in a higher PERMANOVA R 2 for studied biological effects. However, it should still be noted that some OTUs can potentially be lost during rarefaction.  82 . OTU accumulation curves were generated with the function specaccum to determine whether sampling was adequate to detect all OTUs present. Four diversity metrics were calculated from the sample x OTU matrix: OTU richness, the exponent of Shannon diversity, inverse Simpson diversity, and Pielou's evenness (OTU abundance equality) 83,84 . The exponent of Shannon and Inverse Simpson diversities were chosen since these metrics represent true diversities (i.e. "effective species"), unlike other diversity indices/entropies 83,84 . We calculated these various metrics with the function renyi. In order to investigate the influence of invasion (invaded vs. uninvaded) and tree cover (V. nilotica, S. africana, none) on the various diversity metrics, we performed two-way ANOVAs, including interaction effects. Significant differences between means were assessed using Tukey HSD post hoc tests.
For visualizing soil bacterial community composition, we performed Non-Metric Multidimensional Scaling (NMDS) using function metaMDS based on Horn similarity values 85 86 . To test for significant differences in soil bacterial community composition between different soils, we performed a Permutation Multivariate Analysis of Variance (PERMANOVA) 87 with 9999 permutations using the function adonis.
We were interested whether the presence of native trees or invasive O. stricta homogenized soil bacterial communities. We tested this with the betadisper function (using 9999 permutations), which calculates differences in multivariate homogeneity of group variances between tree covers, and invaded and uninvaded soils.
To identify shifts in the abundance of OTUs due to the presence of native trees or the invasive O. stricta, we performed a linear discriminant analysis (LDA) effect size (i.e. LEfSe) 88 using mothur 75 at OTU level. LEfSe taxa are those that have shifted in abundance (identified using an alpha value of 0.05 and LDA score of > 2). Finally, we investigated which OTUs were broadly present in all soils as these are presumably not affected by invasion or tree cover.
Finally, in order to investigate the influence of invasion (invaded vs. uninvaded) and tree cover (V. nilotica, S. africana, none) on the soil pH, humidity, nutrients and enzymatic activities datasets and the germination indices, we performed two-way ANOVAs, including interaction effects. Significant differences between means were assessed using Tukey HSD post hoc tests.

Data availability
All data generated or analysed during this study are included in this published article (and its Supplementary  Information files).