2bRAD Sequencing:
Raw 2bRAD reads were deduplicated and trimmed using the FASTX toolkit (http://hannonlab.cshl.edu/fastx_toolkit). Reads under 25 base pairs (bp) in length or with quality scores <15 were discarded. Following Rippe et al. (2021), photobiont reads were removed by discarding reads that mapped to concatenated Symbiodiniaceae genomes (Symbiodinium (Aranda et al., 2016), Breviolum (Shoguchi et al., 2013), Cladocopium (Dougan, 2020), and Durusdinium with Bowtie2 v2.4.2 (Langmead & Salzberg, 2012). The remaining host reads were then mapped to the Porites lobata genome (Noel et al., 2023). Genotyping was performed using ANGSD v0.923 (Korneliussen et al., 2014). Filters that were used across all analyses included loci that were present in ≥ 80% of individuals, and a minimum read depth of 6 across all samples. Triallelic sites were removed. Reads had a minimum quality of 25, minimum mapping quality of 20 with a strand bias p-value of 1e-5 and a heterozygosity bias p-value of 1e-5. Clones were detected using hierarchical clustering based on pairwise identity by state (IBS) with an additional minor allele frequency (MAF) filter of 0.05. Technical replicates provided the clone detection threshold, and one member of each clone pair was removed for downstream analyses.

A total of 75 samples remained after quality control and technical replicate removal. These libraries were selected for further population genomic analyses due to a higher proportion of the genome covered compared to reduced RAD samples. For all population genomic analyses an additional MAF filter of 0.05 was added, with the exception of site frequency spectrum (SFS) based analyses. Admixture was estimated using NGSadmix; admixture plots were then created using a custom R script (https://github.com/z0on/2bRAD_denovo/blob/master/admixturePlotting_v5.R). Principal Component Analysis (PCA) was conducted using a covariance matrix based on single-read resampling calculated in ANGSD. Admixture results were visualized using the K with the least cross validation error reported from ADMIXTURE. These analyses demonstrated the presence of three distinct lineages amongst our six sampling sites. F_ST was estimated between pairs of lineages using ANGSD before and after outlier loci were removed using Bayescan (https://doi.org/10.1534/genetics.108.092221, Foll & Gaggiotti, 2008).

Microbial community sequencing:
Photobiont communities were characterized through sequencing of the internal transcribed spacer region 2 (ITS2) region using SYM_VAR_5.8S2 and SYM_VAR_REV primers (Hume et al., 2015, 2018). The PCR profile included 26 cycles of 95 °C for 40 seconds, 59 °C for 2 minutes, 72 °C for 1 minute and a final extension of 72 °C for 7 minutes. A negative control was included in the initial amplification but failed to amplify, so it was not included in downstream library preparations. Successful amplifications were cleaned using the GeneJET PCR Purification kit (ThermoFisher Scientific) and a second PCR was conducted to attach Illumina MiSeq dual barcodes to the PCR product before samples were pooled. Volumes for pooling were based on the visualization of barcoded sample band intensity on a 1% agarose gel. This pool was cleaned using the GeneJET PCR Purification kit, gel extracted, and submitted for sequencing as described below.

To characterize bacterial communities, the V4 region of the 16S rRNA gene was amplified from the same samples via PCR using Hyb515f (Parada et al., 2016) and Hyb806R (Apprill et al., 2015) primers and the following PCR profile: 35 cycles of 95 °C for 40 seconds, 65 °C for 2 minutes, 72 °C for 1 minute and a final extension of 7 minutes. Subsequent PCR amplification, cleaning, dual-barcoding, and gel extraction followed the same protocol described for ITS2 with the inclusion of three negative controls, which were also submitted for sequencing. ITS2 and 16S pools were quantified and combined in a 1:3 ratio, respectively. Libraries were sequenced together on Illumina MiSeq (paired-end 250 bp) at Tufts University Core Facility (TUCF) Genomics.

Sequences with adaptor contamination were removed and raw 16S and ITS-2 sequences were separated based on primer sequences using bbduk following Bove et al. (2023). Raw ITS-2 reads were processed by Symportal (Hume et al., 2019) to produce defining intragenomic sequence variant (DIV) profiles for each coral colony. Two samples with <1,000 reads were removed, as well as one outlier sample with >1 million reads; the remaining samples (n = 73) had an average of ~5,500 reads per sample (min: 1,116; max: 16,931). All samples were dominated by one of ten Cladocopium C15 types, and four samples hosted low abundances of Symbiodinium A3 sequences.