Enabling high-accuracy long-read amplicon sequences using unique molecular identifiers with Nanopore or PacBio sequencing

Key results

1. ShownOn a ZymoBIOMICS mock community, UMI consensus of ~4400 bp rRNA operons exceeded Q40 accuracy on both platforms, with PacBio UMI the most accurate and needing the least read coverage per molecule.

   One shared library: ONT UMI (one MinION R10 flowcell) gave 23,365 consensus sequences at ≥25x, mean 4,381 bp, mean residual error 0.0049%; PB UMI (Sequel II 8M, CCS mode) gave 39,678 at ≥3x, 4,376 bp, 0.0006%; raw PB CCS reads without UMIs gave 135,823 at ≥40x with 0.0080%. Coverage cutoffs were chosen to reach mean error <0.01% (Q40). · Main

2. ShownDual-UMI filtering nearly eliminated PCR chimeras, and naive phasing of consensus sequences recovered all 43 mock operon variants, with a smaller share of consensus data falling into error-containing or chimeric variants for UMI data than for plain PB CCS.

   Chimera rate 0.017% (ONT UMI) and 0.022% (PB UMI) vs 1.9% for raw PB CCS reads without UMI clustering. All 43 reference variants were found in every data type, with errors only in 2 ONT UMI variants (1 homopolymer error each); consensus data assigned to variants with systematic errors: 1.00%/6.99%/0.18%, to chimeric variants: 0%/0.46%/0% (ONT UMI/PB CCS/PB UMI). · Main; Phasing of consensus sequences and variant…

3. ShownResidual errors in ONT UMI and PB CCS consensus came largely from homopolymer deletions (mostly in >4 bp C/G runs), plus non-homopolymer G-insertions (the majority for ONT UMI) and random mismatches (PB CCS), and were partly position-specific; PB UMI consensus kept too few errors to reveal a pattern.

   Error types other than homopolymer deletions levelled off above 20x for ONT UMI and PB CCS; deletions sat mostly in >4 bp C/G homopolymers; most ONT UMI residuals were G-insertions outside homopolymers, some recurring in >50% of sequences; PB UMI had 1109 errors in 39678 sequences. Random mismatches (likely PCR errors) were the other main PB CCS error. · Main; Error profiling

4. ShownIn 70 American Gut Project fecal samples, PB UMI produced 253,089 full-length bacterial rRNA operons, and combining 16S with 23S raised strain- and species-level database assignment over full-length 16S alone.

   BLAST against the Web of Life database (86,200 genomes): 16S alone assigned 11.3% at strain and 38.4% at species level; 16S+23S assigned 22% and 72.2% (chi-squared 124,086, p<2.23e-308). Genus abundances from operon 16S correlated with Earth Microbiome Project 16S V4 data from the same extracts (Spearman r=0.527, Pearson r=0.553; 82 shared genera). · Main; Taxonomic consistency between 16S V4 fragments…; Taxonomic specificity of operons from real…; Figure legends

5. SuggestedA ligation-based variant, with UMI adapters on sheared genomic DNA instead of a tagging PCR, gave E. coli consensus sequences up to 10,578 bp with error and chimera rates in line with the amplicon results.

   E. coli K-12 MG1655 DNA (mean fragment ~8 kbp), ONT only: 3,658 UMI consensus sequences at ≥30x, mean length 4,476 bp (min 2000, max 10578), mean error 0.009% and chimera rate 0.024%, profiled against the U00096.3 genome. · Main; Data analysis; Figure legends