2.4. Analyses using SymPortal

Demultiplexed paired fastq.gz sequencing files were submitted directly to the remote instance of SymPortal for analysis (symportal.org; Hume et al., 2019). The SymPortal analytical framework was used to predict ITS2 profiles as proxies of Symbiodiniaceae genotypes and to generate between‐sample dissimilarity metrics, based on ITS2 sequence assemblages. The framework leverages information encoded in the intragenomic sequence diversity, harbored within the Symbiodiniaceae genome, to offer fine‐scale delineations. To plot ITS2 type profile relative abundances, we made use of the ITS2 type profile absolute count table output by SymPortal. As part of the built‐in quality control pipeline of SymPortal, Minimum Entropy Decomposition (MED; Eren et al., 2015; Hume et al., 2019) is performed on a per‐sample basis before the sequence abundances are used to predict ITS2 type profiles. To plot ITS2 relative sequence abundances, we used the post‐MED absolute abundance sequence count tables (as opposed to the pre‐MED sequence count tables). To produce a heat map of between‐site average sample dissimilarities, and two PCoA ordinations, we used the Symbiodinium, Bray–Curtis‐derived, between‐sample dissimilarity matrix (including a square root transformation) that is output by default by SymPortal. The between‐site average dissimilarities were computed using a bespoke Python script and plotted using matplotlib's implementation of imshow. Specifically, for a given pairwise site comparison, for every sample in the first site, the distance to every sample in the second site was collected. A mean average and standard deviation were then computed from these distances. For self‐site comparisons (e.g., Duba‐Duba), average within‐site distances were calculated. PCoAs were computed using Python and the scikit implementation of pcoa and plotted using matplotlib's implementation of scatter. To assess for statistical difference between sites across samples based on Symbiodinium ITS2 sequence assemblages, we ran a one‐factor PERMANOVA (Anderson, 2001) using scikit‐bio's function PERMANOVA and the Symbiodinium Bray–Curtis square root transformed distances. PERMANOVA is resilient to heteroscedasticity across factor groups when factor groups have equal numbers of samples (i.e., a balanced experimental design; Anderson & Walsh, 2013). Given the balanced design in this study, a PERMDISP2 analysis (Anderson, 2006) was not conducted. To further investigate the sequences driving structure in the Bray–Curtis‐derived distances, we conducted a SIMPER analysis (Clarke, 1993) on the post‐MED Symbiodinium sequences with site as the grouping factor. The abundances were square‐root transformed (to match the transformation of the abundance matrix used in the Bray–Curtis calculation), and the analysis was conducted in Python using ecopy's simper function.