Statistical analyses
Community data frames were constructed using individuals as “sites”
against ASV, family or order identifications. Incidence and
Hellinger-transformed relative read abundances were used as values.
Incidence data alone outweighs rare taxa and can significantly alter
detected ecological patterns. While read abundances are not directly
translatable to taxon abundances in a community, there is utility in
transformed abundances to detect otherwise obscured ecological patterns.
Hill numbers (Alberdi & Gilbert 2019) were calculated to quantify prey
diversity for all three community data frames. Welch’s t-test was used
to test the hypothesis of no differences in dietary prey diversity
between spiders in ginger and native forest. Additionally, Welch’s
t-test was used to test the hypothesis of no differences in read
abundances between both groups.
To assess overall compositional differences between the diets of spiders
in ginger sites versus native forests, community matrices were
constructed using block number combined with transect number as the
site, against ASV or order. An order-level food web grid was
constructed. Family, genus or species was not used because of overall
low BLAST matches; these taxonomic levels provided less compositional
information as well as the data was biased towards taxa well-represented
in GenBank. Order-level provided the coarsest view of diets while ASV
provided the finest grain view. Beta diversity was calculated based on
Jaccard distances on both Hellinger-transformed reads and incidence
data. Produced values were plotted against distance between sites to
assess level of spatial autocorrelation (see S5 in Supplemental
Information). Because of very weak correlation between distance and beta
diversity and our interest in only a single explanatory variable
distinctly split across sites, spatial autocorrelation was not
identified as a concern. Non-metric dimensional scaling (NMDS) was
performed using beta diversity values, with k = 2 and using a maximum of
1,000 random starts to achieve convergence. Permutational multivariate
analysis of variance (PERMANOVA) was performed based on beta diversity
values to test the hypothesis that the center and spread of dietary
communities across ginger and native forest sites are equivalent.
To assess compositional differences within the two most prevalent orders
(Hemiptera and Lepidoptera), 16s sequences were aligned using MUSCLE and
a neighbor joining phylogenetic tree using the Jukes-Cantor model of
evolution was constructed in Geneious (See S8 and S9 in Supplemental
Information). This tree was then used to calculated phylogenetic beta
diversity. Phylogenetic beta diversity was used here to allow assessment
of relationships between taxa within each order, as family, genus or
species identifications were not obtainable using BLAST. Again, NMDS was
used to visually assess community similarity in ordination space and
differences tested using PERMANOVA.
The lack of confident BLAST identifications resulted in a lower number
of identifiable native/non-native sequences; more identifiable sequences
were detected in the spiders in ginger because of the higher proportion
of introduced taxa. Parasites were identifiable across ginger and native
forest sites using BLAST. Parasite frequency was assessed using both the
number of ASVs within ginger and native forest sites that were
identified as parasitic, and the relative number number of parasitic
reads in individual spiders. Welch’s t-test was used to test the
hypothesis of no differences in parasitic load between ginger and native
forest sites.
BLAST assignment, alignment and phylogenetic reconstruction was
completed in Geneious Prime v. 2022.0.2. Analyses were conducted in R
version 4.1.2. Analyses and figures were produced using the following
packages in R: vegan, BAT, ape, tidyverse, reshape2, ggplot2, ggpubr,
ggvenn, formattable, gapminder, bipartite, and ComplexHeatmap. Code for
analysis and data is available on GitHub and Dryad (see Data
Accessibility).
Results