Further downstream analysis and visualization were done using Python/matplotlib ( 22). The resulting subsampled files were then analyzed by the complete analysis pipeline. 2D, reads were subsampled to the approximate target levels from the unprocessed fastq file pairs. 6, IgH molecules were grouped into lineages across all samples analyzed using a single linkage clustering approach and a 90% CDR3 similarity cutoff. Data were then converted to the fasta format and annotated using IgBLAST ( 20) with germline data retrieved from IMGT ( 21). 3B, we aligned the raw reads of each molecule group to the assembled molecule using BLAST ( 19).

To analyze molecule coverage distribution in Fig. Control data were processed as previously described ( 7). When multiple extensions are possible, the program always selects the extension that results in the highest quality base being incorporated into the extension. It then reports the completed molecule in fastq format. Finally, the program reduces all of the reads into k-mers and extends one of the ends until the program reaches the other end of the amplicon or the program runs out of extensions. Then the program creates a high-quality consensus of the ends of the amplicon using a combination of quality and abundance of each nucleotide at each position. The program sorts the raw reads into three categories: 1) reads derived from the 5′ end of the amplicon, 2) reads derived from the 3′ end, and 3) reads that are unanchored and likely to come from some place in the middle of the molecule. Third, the AMPssembler algorithm assembles IgH transcripts from each molecular group. Third, reads with highly similar (fewer than two mismatches) molecular identifier 1 (Tn5_B) or molecular identifier 2 (Tn5_A) to the Uncut molecular groups are added into these molecular groups. Second, reads of the Uncut library are grouped into molecular groups when their combined molecular identifiers differ by fewer than five mismatches. For Tn5_B libraries, the first 18 bases of read 1 represent molecular identifier 1. For Tn5_A libraries, the first 18 bases of read 2 represent molecular identifier 2. For Uncut libraries, the first 18 bases of read 1 represent molecular identifier 1 and the first 18 bases of read 2 represent molecular identifier 2.

First, molecular identifiers are extracted from the trimmed fastq files. TMIseq data were further processed according to the following pipeline. For libraries sequenced on the MiSeq 2 × 300, reads were also cropped to 150 bp.

Raw reads in fastq format are trimmed using Trimmomatic ( 18), discarding reads pairs containing adapters. Uncut (aliquot 3), Tn5_A, and Tn5_B libraries were pooled and sequenced according to standard Illumina protocols on an Illumina MiSeq (2 × 300 run) or HiSeq 3000 (2 × 150 run). To create Tn5_B libraries, aliquot 2 is tagmented using Tn5 enzyme ( 17) loaded with Nextera_B adapter and PCR amplified using a universal Nextera_A primer and a Nextera_B primer with an Illumina index not yet present in the library pool and purified and size selected for fragments >380 bp using 2% EX gels (Life Technologies). cDNA was purified and size selected twice with SPRI beads using a 0.7:1 beads/sample ratio corresponding to a cutoff discarding DNA 380 bp using 2% EX gels (Life Technologies). In a two-cycle PCR reaction, second and third cDNA strands were synthesized using Phusion polymerase (Thermo Fisher Scientific) and two modified primer pools complementary to the beginning of the V leader exons and ∼100 bp into CH1 exons of all IgH isotypes and containing molecular identifiers and partial Nextera sequences. RNA (10 μl) was used for SuperScript II (Thermo Fisher Scientific) cDNA first-strand synthesis using a primer pool specific to all exons specific to the secreted isoform of all IgH isotypes (IgM, IgD, IgG1–4, IgA1, IgA2, IgE).

Immune repertoire sequencing