Sponge collection from Autonomous Reef Monitoring Structures and field surveys:
Autonomous Reef Monitoring Structures (ARMS) are standardized sampling devices that mimic reef interstices, attracting cryptobiota colonization (Brainard et al. 2009, Knowlton et al. 2010). Standard ARMS units are comprised of an eight-tiered stack of gray Type I PVC plates (22.9 x 22.9 centimeters (cm)), arranged in four open and four semi-closed layers (Figure S1a; Leray and Knowlton 2015), while modified ARMS units are composed of a two-tiered stack of one open and one semi-closed layer (Figure S1b; Timmers et al. 2020). Sponges were sampled from six standard ARMS deployed along the reef slope of Moku o Lo'e (Coconut Island) and from six modified ARMS (Figure S1c) hovering in the water column attached to a Moku o Lo'e intake pipe, along an adjacent reef slope. An additional 24 modified units were placed within mesocosm tanks on Moku o Lo'e that received unfiltered seawater from the same intake pipe but were exposed to future climate conditions as described in Bahr et al. (2020) (Figure S1a). The ARMS in mesocosm tanks and those at the intake pipe were retrieved for sponge subsampling every two months for two years, and sponges from the full ARMS on the reef slope were collected once upon recovery, in July 2018. The ARMS units in ensemble provided a total combined sampling surface area of 15 square meters (m²) at each period of collection. At each collection period, ARMS units were disassembled for high-resolution plate imagery and carefully examined for newly settled sponge recruits. Sponges showing unique morphological features on each plate were individually photographed, carefully subsampled, and fixed in 95% ethanol for DNA extraction. If enough tissue was available, sponges were additionally fixed in two solutions: one containing 4% paraformaldehyde in seawater, and the other containing 4% glutaraldehyde in 0.1M sodium cacodylate with 0.35M sucrose for future histological evaluation. A total of 439 sponge samples were collected from the ARMS units.

Sponges were also collected on reef substrates along a 50-meter (m) transect line from 34 sites in Kāne'ohe Bay and one site on the Makai Pier in Waimanalo (see Table S2 for GPS coordinates and collection depth) throughout the two-year mesocosm experimental period. Collection on Kāne'ohe Bay reef sites included samples taken by global taxonomic experts during the Smithsonian-led MarineGEO biodiversity surveys in 2017. 163 marine sponges showing unique morphologies were haphazardly collected in the photic zone of the reef at a depth range of 1-16 m from within crevices, beneath coral rubble, fouling upon structures and under overhangs. Presence/absence of sponge OTUs were recorded at each surveyed site (Table S3), were photographed, and fixed in 95% ethanol for DNA extraction. Additional sponge metadata pertaining to specimen morphology, such as color, consistency, surface, oscules, exudates, and odors was also recorded.

All samples were vouchered with the Florida Museum of Natural History at the University of Florida (UF Porifera) and the Hawaiʻi Institute of Marine Biology ("KB" or "KBOA") (Table S4). Images and associated metadata of each sponge sample are publicly available at https://www.invertebase.org/portal/ and http://specifyportal.flmnh.ufl.edu/iz/. All samples were collected under special activities collection permit (SAP) nos. 2018-03 and 2019-16 (covering the period of 13 Jan 2017 through 10 Apr 2019) issued by the State of Hawai'i Division of Aquatic Resources.

Sponge DNA extraction and sequencing:
Vouchered specimens were subsampled for DNA extraction using the E-Z 96 Tissue DNA Kit (Promega Bio-Tek, Norcross, GA, USA) following the manufacturer protocols. Care was taken to subsample only sponge material free of other organisms, which would contaminate the sponge DNA extract. Multiple primers were used in a stepwise fashion to successfully amplify partial fragments of both 28S rRNA and COI genes using polymerase chain reaction (PCR) (Table S1 in Timmers et al., 2020). Fragments of the COI were initially attempted with primer pairs LCO1490 / COXR1 (~1400 bp fragment) (Folmer et al. 1994), followed by primers jgLCO1490 / jgHCO2198 (Geller et al. 2013) (~648 bp fragment) within the previous PCR fragment region and a final attempt with subsequent internal primers mlCOIint / jgHCO2198 (~313 bp) (Leray et al. 2013). Similar to the approach used for the COI, amplification of 28S rRNA fragments were first attempted with primers F63mod / 1072RV (~1050 bp) (Medina et al. 2001), followed by internal primers (28S-C2-fwd / 28S-D2-rev) (450 bp) (Chombard et al. 1998) within the previous fragment, and a final attempt with 28SMycF / 1072RV. PCR reactions were carried out in a total volume of 40 microliters (µL) including the following: 14.4 µL of H₂O, 20 µL of BioMix Red (Bioline, Taunton, MA, USA) PCR Mastermix, 0.8 µL of each primer (10 mM), 3.2 µL of bovine serum albumin (BSA) (100 mg/mL), and 0.8 µL of template DNA (1 to 30 nanograms per microliter (ng/µL)). PCR products were examined on a 1% agarose gel stained with GelRed and purified using ExoFAP (Exonuclease I and FastAP - Life Technologies, Carlsbad, CA) prior to sequencing. When products showed multiple bands above the 100 bp ladder mark, products were purified using gel excision by loading 40 µL of the PCR product onto a 2% agarose gel made with 1x modified (no EDTA) TAE running them at 50 millivolts (mV). After 90 minutes, bands were excised with a sterile scalpel, loaded onto a column filter fitted inside a 1.5 milliliter (mL) centrifuge tube, and centrifuged for 10 minutes at 5,000 rpm. Sequencing reactions were performed in both directions using the Big Dye TM terminator v. 3.1, and sequencing was done on an ABI Prism 3730 XL automated sequencer at the University of Hawai'i Advanced Studies of Genomics, Proteomics and Bioinformatics sequencing facility.

Forward and reverse reads were trimmed and edited by eye using Geneious 10 (Kearse et al. 2012). Assembled and edited sequences were exported as fasta files and checked for contamination by using the BLAST (Altschul et al. 1990) function in GenBank. Sequences showing >85% sequence identity to those belonging to Porifera were kept and used for further analysis. 28S rRNA sequences for 592 samples were produced, but only 340 sequences were deposited in GenBank as many were repetitive sequences with 100% identity. When available, up to three replicate sequences per OTU were deposited and assigned accession numbers MW016037 - MW016376. 98 COI sequences were deposited in GenBank and assigned accession numbers MW059039- MW059109; MW144969-MW144988; MW143251-MW143256; MW349624 (Table S4).

Phylogenetic analysis and taxonomic assignments:
Sequences were aligned with the closest sequence relatives in the GenBank database using the ClustalW algorithm with default parameters in Geneious. Sequence KJ483037.1 Parazoanthus puertoricense was used as an outgroup for all phylogenetic topologies of partial 28S rRNA sequences and AB247348.1 Epizoanthus arenaceus was used as an outgroup for the phylogenetic topology of partial COI sequences. Bayesian inference (BI) using MrBayes version 3.2.1 (Huelsenbeck & Ronquist 2001) and a maximum likelihood (ML) framework using RaxML (Stamatakis 2006) were added to the phylogenetic analysis. The GTR substitution model and GTRGAMMA nucleotide model with 1,000 bootstrap replicates were implemented in the BI and ML analyses, respectively. The BI was run using 5 million generations sampled every 200 generations. The analysis was stopped when the standard deviation (SD) of split frequencies fell below 0.01.

Most sponge OTUs were delineated using a combination of ≥1% COI and 28S rRNA sequence divergence combined with unique morphological features and classified as distinct operational taxonomic units (OTUs). A handful of OTUs that were morphologically clearly differentiable, but had ≤ 1% sequence divergence were also recognized as distinct OTUs (Table S6). This conserved threshold was chosen based on the different rates of evolution that can exist within poriferan families and even genera which make the selection of an accurate threshold for delineating sponge OTUs arbitrary (Erpenbeck et al. 2007, Wang & Lavrov 2008, Redmond et al. 2011, Voigt & Wörheide 2016, Yang et al. 2017).

Preliminary assessments of morphological characters (i.e. color, consistency, surface, oscules, and skeleton composition) were made mostly from OTUs that matched previous vouchered sponge collections in Kāneʻohe Bay (De Laubenfels 1950, Bergquist 1967, 1977, Pons et al. 2017) (Table S5). We assigned OTUs to taxonomic levels based on the placement of each barcode into the lowest well-supported clade (bootstrap support of ≥50) in the COI and 28S rRNA tree topologies. On average, taxonomic identities followed these barcode sequence identity percentages: Order (>90%), Family (>95%), Genus (>98%) and for the species above (100%). Phylogenetic topologies were first generated with only full-length amplicons for COI and 28S rRNA, then repeated with shorter sequences to maximize the inclusion of reference sequences from GenBank. Matches and identification at the species level (17 OTUs) were based on sequences and a preliminary analysis of skeleton and spicule composition, which matched sequences from vouchers in GenBank linked to a publication with a rigorous morphological assessment and description of the voucher. The remaining OTUs (including GenBank accession matches without taxonomic support) were identified as “sp.” since further morphological analysis are needed for accurate classification. In addition, species identification is impossible using molecular methods for polyphyletic groups (such as suborders, families and genera within Haplosclerida) without a complete morphological assessment of OTUs. However, the objective here is to determine species richness mostly based on molecular OTUs rather than a full species description of OTUs.

Diversity assessment:
R v.3.6.3 (R Core Team 2020) was used to visualize and analyze the molecular diversity assessments of sponges recruited on ARMS and reef substrates. Phylogenetic analyses of COI and 28S rRNA sequence data were used to prepare a taxonomy table (Table S5) for OTU classification (OTU) to the lowest level possible. An OTU distribution table (Table S2) specifying OTU presence/absence on either 'ARMS' or 'reef' substrate at each of the 35 sites was used to map sponge OTU richness using the 'ggmap v.3.0.0.901' package (Kahle & Wickham 2013). We used the specaccum function from the 'vegan v.2.5-6' (Oksanen et al. 2013) package to generate OTU richness rarefaction curves for comparison between the two substrates across the most specious sponge groups according to sponge class and order. Number of OTUs as a function of sites were used to generate rarefaction curves for reef substrate sponges and number of OTUs as a function of time points was used for ARMS as these were only present at one site. Venn diagrams were generated using the 'limma v.3.42.2' (Ritchie et al. 2015) package to determine the number of shared OTUs between the survey method types. Calculation of new OTU records were based on species comparisons to previous studies focused on Kāne'ohe Bay sponge collections.

Note: Table and figure numbers refer to Vicente et al., 2022 (doi: 10.1007/s00338-021-02109-7).