Genetic variability in microbial eukaryotes reshapes marine biodiversity assessment in the age of amplicon sequencing

PLOS One | https://doi.org/10.1371/journal.pone.0326053 June 20, 2025 23 / 29 Table 3 tracks the mean percent of reads that passed through each stage of the DADA2 pipeline for each order. It is presumed that the relatively low passing rates at the quality and length filtering step is primarily due to length rather than quality. The sequencing facility reported that 100% of our barcoded reads had Phred quality scores ≥Q20. Thus, PCR artefacts or off-target amplification are the most likely explanations for our passing rates. The mean read length of the raw demultiplexed data was 1252 bp; however, based on known genomic characteristics of the target taxa, we expected on-target reads to be approximately 1400–1500 bp long. We thus required that reads passing our length filter be between 1300 and 1600 bp long, which alone would have removed approximately half the raw reads for being too short. These rel- atively short reads may have originated from artefacts generated during PCR, or from contaminating taxa that happened to have shorter regions of the 18S gene and were unexpectedly amplified by our primers. ASV data were then imported into QIIME2 [72] for taxonomic assignment and filtering. In QIIME2, naïve bayes classi- fiers were trained on the PR2 database with the custom primer pairs (Table 2). After taxonomic classifications were made (using minimum confidence = 0.3), any sequences that did not belong to polycystine Radiolaria or had < 3 total reads were filtered out. In an additional effort to remove potential sources of environmental contamination from each specimen, any

ASV that was not assigned to the same family as the morphospecies was excluded from analyses. An exception to this protocol was made for naked Collodaria and putative juvenile specimens, which lacked the skeletons needed for confi- dent morphospecies identifications. For these specimens, ASVs with any polycystine family assignment were permitted. Finally, a per-specimen read abundance threshold was applied, which required each ASV to have at least 3 reads in a given specimen before it was counted as present in that specimen. After these vetting procedures, any specimens lacking a minimum of 5 taxonomically-consistent ASV reads were removed from all analyses. A maximum-likelihood phylogenetic tree (constructed with RAxML-NG [73]) was used to detect patterns of suspi- cious topology that may indicate hidden chimeric sequences. Input data included 258 polycystine radiolarian reference sequences, 18 Acantharia outgroup sequences downloaded from the PR2 database (S1 Table), and the most abundant sequence from each of the specimens in this study (available in S1 Data). Sequences were aligned using MAFFT [74] with linsi parameters and then trimmed with a 5% gap threshold using trimAL [75]. The tree utilized the GTR + G evolution- ary model with 100 bootstrap replicates and 1853 aligned positions. After visual inspection, ASVs that formed long- branching clades without any reference sequences and which had relatively low read numbers (i.e., < 50) were manually checked for chimeras by querying NCBI–BLAST (https://blast.ncbi.nlm.nih.gov/Blast.cgi) with the first ~300 bases and last

~300 bases of each sequence separately. This test led to the discovery of 14 chimeric ASVs that had not been removed by the DADA2 algorithm and were removed manually. All these chimeric sequences had been taxonomically assigned to family Pterocorythidae (Nassellaria) in PR2, although NCBI–BLAST identified their sequence fragments as belonging to various groups, including Rhizaria, stramenopiles, and alveolates. Following all quality checking procedures, 173 out of 232 individual specimens passed through all process- ing steps (75%), comprising 1,224 unique ASVs among them. This included 836 ASVs assigned to Collodaria (95 specimens), 187 ASVs assigned to Nassellaria (36 specimens), and 201 ASVs assigned to Spumellaria (42 specimens). This made up the dataset used for all analyses in this study (S1 Data). There were 59 spec- imens that successfully amplified during PCR but did not pass bioinformatic processing and filtering steps, Table 3. Mean percent of total input CCS reads that passed through each step of the DADA2 pipeline per polycystine radiolarian order. Order Primers trimmed (% input) Quality & length filtered (% input) Denoised (% input) Non-chimeric (% input) Nassellaria 92 37 33 30 Spumellaria 89 35 30 24 Collodaria 90 59 56 53 https://doi.org/10.1371/journal.pone.0326053.t003