Peripheral blood and cord blood samples were collected from 38 HD of different ages (Supplementary Table 1 in Data Sheet 2). From all donors, peripheral blood or cord blood was obtained with informed consent and according to the guidelines of the Medical Ethics Committees of the Erasmus MC.

PBMC’s were isolated from peripheral blood or cord blood samples using Ficoll separation. mRNA was isolated using the GenElute Mammalian total RNA miniprep kit from Sigma Aldrich (St. Louis, MO, USA). cDNA was created from 2 μg RNA using the Superscript II reverse transcriptase kit from Invitrogen (Paisley, UK) in a 40 μl reaction. IGH transcripts were amplified from 5 μl cDNA per reaction in a multiplex PCR using the forward VH1-6 FR1 (BIOMED-2) primers (26) and either the CgCH1 (27) or the IGHA (28) reverse primer. These PCR products were purified and sequenced using Roche 454 sequencing as previously described (29). In short, PCR products were purified by gel extraction (Qiagen, Valencia, CA, USA) and using Agencourt AMPure XP beads (Beckman Coulter, Fullerton, CA, USA). Subsequently, the concentration of the PCR product was measured using the Quant-it Picogreen dsDNA assay (Invitrogen, Carlsbad, CA, USA). The purified PCR products were sequenced on the 454 GS junior instrument using the Lib-A V2 kit according the manufacturer’s recommendations. Data processing was performed using the long-amplicon pipeline 1 from Roche.¹ This pipeline is similar but more stringent to the standard Amplicon pipeline,² resulting in the lowest throughput, but the highest quality filtering.

Sequences were demultiplexed based on their multiplex identifier sequence and 40 nucleotides trimmed from both sides to remove the primer sequence using the IGGalaxy tool (25). Fasta files were uploaded in IMGT/High-V-Quest (selection of IMGT reference directory set: F + ORF + in-frame P with all alleles, search for insertions and deletions: yes, parameters for IMGT/Junction Analysis: default) (30), and subsequently, the IMGT output files were analyzed in an extended version of our IGGalaxy tool (ARGalaxy) (25). Information on junction characteristics, CDR3 length and composition, CSR, the number of mutations, and SHM patterns were obtained. Only productive sequences, that were complete, without ambiguous bases, present twice or more, with a C subclass that could be defined, were included in the analysis as a single sequence. All information on the FR1 region was excluded from the analysis since the forward primers used to amplify the transcripts were located in FR1. The age, gender, and number of sequences after each filtering step are listed in Supplementary Table 1 in Data Sheet 2, and detailed information on the mutations used to calculate the SHM patterns (transition tables) are listed in Supplementary Table 2A (IGG) and 2B (IGA) in Data Sheet 2. In addition, the immunoglobulin analysis tool (IgAT) (31) was used to determine the percentage of antigen-selected sequences. To determine the selection strength of the CDR and FR regions, the BASELINe tool was used (selection statistics: focused, SHM targeting model: human Tri-nucleotide, custom boundaries: 25:26:38:55:65:104:-) (32, 33).

Clonal relation between sequences was determined using Change-O using the nucleotide hamming distance substitution model with a complete distance of maximal three. For clonal assignment, the first genes were used, and the distances were not normalized. In case of asymmetric distances, the minimal distance was used (34).

Lineage trees were created based on a minimal substitution model. The clonal evolution of the sequences was determined by identifying mutations that overlap between all sequences or groups of sequences. The tree structure was based on the model in which the least number of mutations was needed to create all different transcripts present in the clone, for details see Supplementary Table 4 in Data Sheet 2.

For the naive repertoire data, previously published data were used for analysis (20). In short, the demultiplex and trimmed fasta files were uploaded into IMGT/High-V-Quest and, subsequently, the IMGT output files were analyzed in an extended version of our IGGalaxy tool (ARGalaxy) (25, 30). Only productive sequences with a unique combination of their VH, JH, and amino acid sequenced of the CDR3 were used for the analysis.

FASTQ files of the raw and filtered data of the naive B cell repertoire and IGG and IGA transcripts are available from the European Nucleotide Archive (ENA) project number PRJEB15348.

To study the antigen-experienced repertoire of HD of different ages, IGG and IGA transcripts from peripheral blood of 36 HD (age: 1–74 years; 18 males and 18 females), and two female cord blood samples were sequenced using next generation sequencing (Supplementary Table 1 in Data Sheet 2). We used Roche 454 sequencing because of its ability to handle the long IGG and IGA amplicons (700–800 bp).

For analysis, only transcripts assigned as functional by IMGT/High-V-Quest were selected (30). To exclude low quality reads, sequences containing uncalled “N” bases in the CDR1-CDR3 sequence were excluded and reads missing a CDR1, FR2, CDR2, or FR3 sequence were discarded. To allow analysis of SHM, an additional filtering strategy was developed, which allows better distinguishing between sequencing errors and true mutations. Yaari et al. proposed a strategy in which only unique reads of which the exact CDR1-CDR3 nucleotide sequence occurs two or more times are used for further analysis (35). HD in which this filtering resulted in less than 45 unique transcripts were excluded from further analysis (IGG n = 5 and IGA n = 1). This filtering resulted in a unique reference data set of 31,380 IGG and IGA transcripts derived from 38 individuals, which is deposited, so publically available for other studies.

During early childhood, the peripheral B-cell compartment changes significantly, with an increase in the absolute numbers of B cells and the percentage of memory B cells (36). However, SHM has not been studied extensively in children and adults. We found that the percentage of mutations in both IGA and IGG transcripts increased during early childhood and remained around 7% (range 5.6–9.4%) from the age of about 6 years onward (Figure 1).

Somatic hypermutation is initiated by deamination of a C to a U by AID, resulting in a U:G mismatch (Figure 2A). AID has been shown to preferentially target DNA at RGYW and WRCY motifs. In both IGG and IGA transcripts, approximately 35% of the mutations were located in RGYW/WRCY motifs (Figure 2B), and although the variation is small, the frequency negatively correlates with age (Supplementary Figure 1A in Data Sheet 1).

Dependent on the DNA repair pathway used to solve U:G mismatches, transversion and transition mutations at G/C and A/T locations are introduced (Figure 2A). In the HD, more than half (52.8–62.6%) of the mutations occurred at G/C locations (Figures 2C,E, Supplementary Figure 1B in Data Sheet 1), and approximately 50% (45–59.2%) of the mutations at G/C locations were transitions (Figures 2C,E; Supplementary Figure 1B in Data Sheet 1). Mutations at A/T locations are introduced via the MMR pathway in which polη has been shown to preferably mutate TW and WA motifs. The frequency and absolute number of mutations at A/T locations (Figures 2C–E), as well as the percentage of mutations in WA/TW motifs slightly correlated with age (Figure 2A; Supplementary Figure 1C in Data Sheet 1). Overall, the repair of the U:G mismatches was very consistent in all HD, except for the cord blood samples in which the targeting and repair patterns were different. This might be related to the very low number of mutations present in these samples and the high level of clonal relation within these samples.

It has been previously described that antigen-experienced B cells are selected against long CDR3 regions and IGHV4-34 usage, qualities associated with autoimmunity (37). To look at the effect of selection on the CDR3, our data were compared with our previously published next generation sequencing data on productive IGH rearrangements amplified from sorted naive B cells (20). As previously described, a reduced CDR3 length was found in the antigen-experienced repertoire, as well as an increased number of deletions in the junctions (Figures 3A,B). The number of N-nucleotides was similar between the naive and antigen-experienced repertoire (Figures 3A,B).

As previously shown, IGHV4-34 gene usage is significantly lower in the IGG and IGA transcripts compared to naive B cells (Figure 3C) (37). In addition, an increase in IGHJ4 gene usage was found in the antigen-experienced repertoire, while IGHJ6 gene usage was reduced (Figure 3D) (24). Looking at the amino acid usage in the CDR3, the tyrosine (Y) usage was reduced in the antigen-selected repertoire (Figure 3E). These changes were found in all 38 HD and both in IGG and IGA transcripts, indicating that selection for those parameters are uniform and irrespective of age.

In addition to the changes in the CDR3, selection of the antigen-experienced repertoire can also be measured by analyzing the replacement (R) and silent (S) mutations in the VH gene. B cells with R mutations in the CDRs are positively selected if they increase BR affinity, and B cells with R mutations in the FR regions are negatively selected due to decreased stability of the BR. S mutations do not alter the amino acid sequence and therefore play a limited role in selection. In the IGG and IGA transcripts, the majority of mutations were present in the FR regions (Figure 4A), a consequence of its larger size. Therefore, often the R/S ratio is used to look at antigen selection, as this compensates for the difference in size of the CDR and FR region and for a possible difference in target motives between these regions. In line with selection, the R/S ratio is higher in the CDR compared to FR regions in the HD (Figure 4B). The R/S ratio in the CDR regions was more variable (3.6–8.3), compared to the FR regions (1.6–2.7), but was not correlated to age. In cord blood samples, the R/S ratios were altered. This can most likely be explained by the low mutation rate in these samples and the high clonal relation between transcripts in these samples. Looking at the number of R mutations at the different amino acid locations also confirms an increased frequency of R mutations in the CDR regions (Figure 4C).

In addition to the R/S ratio, several tools have been developed to analyze selection of the repertoire. IgAT looks at the number of R mutations in the CDR in relation to the total number of mutations and uses this to calculate, for each sequence, whether it has undergone antigen selection (31). In this definition of antigen selection, selection is dependent on the number and the location of R mutations but also the frequency of mutations. Applying IgAT to our data showed that the frequency of antigen-selected sequences increased in young children. From 6 years of age onward, the frequency of antigen-selected sequences was around 40% in both IGG and IGA transcripts (Figures 5A,B). The Bayesian estimation of antigen-driven SELectIoN (BASELINe) program uses a similar approach but compares for each transcript the expected mutations based on random distribution of the observed mutations. Subsequently, it calculates the total selection strength for the CDR and FR regions (32, 33). In HD, the selection strength for both the CDR and FR region showed quite some variation; however, the selection strength was always higher in the CDR region, indicating selection (Figures 5C,D).

Besides SHM, AID also initiates CSR by inducing U:G mismatches in switch regions upstream of the constant genes. The IGH locus contains nine constant genes: Cμ, Cδ, Cγ3, Cγ1, Cα1, Cγ2, Cγ4, Cε, and Cα2 (Figure 6A). We chose reverse primers in the constant domain of the IGG and IGA locus, which allowed us to distinguish between IGA1, IGA2, IGG1, IGG2, IGG3, and IGG4 transcripts. Analysis showed that the subclass distribution of both IGA and IGG is variable between all HD (Supplementary Figure 2 in Data Sheet 1). Looking at IGG transcripts in different age categories showed that the relative amount of IGG2 transcripts is higher with an increased age, while the number of IGG1 transcripts is lower (Figure 6B). The percentage of IGG3 transcripts is low in all HD, except for children from 4 to 10 years of age in which 19.9 ± 7.7% of the IGG transcripts is IGG3. IGG4 transcripts are rare in all different age categories. In the IGA transcripts, the contribution of IGA2 transcripts is increased in samples with higher ages (Figure 6B).

Each antibody subclass has a different effector function; however; it is still not completely understood how class switching occurs during an immune response. We used Change-O to identify clonally related sequences within our data set that likely originate from a single precursor B cell (34). For each sample, we have calculated the maximal clone size (the maximal percentage of transcripts that are part of one clone), and the percentage of sequences from unique clones (Supplementary Table 1 in Data Sheet 2). These data represent the clonality of the analyzed data but are not necessarily representative of the clonality of the donor repertoire. We identified several clones in every HD (Supplementary Table 1 in Data Sheet 2). These clones differ in size, and the number of acquired mutations. Assigning clonal relation allows comparison of SHM patterns within a single clone (Figure 6C), and the creation of lineage trees of clonal outgrowth (Figure 6D). Interestingly, we found multiple clones that contained transcripts of different subclasses within the same class (IGA or IGG) (Figure 6C) or both IGA and IGG transcripts (Figure 6D). This shows that within one antigen-driven response, CSR to multiple subclasses can be found. Looking into this further, we also identified sequences with exactly the same CDR1–CDR3 sequence, and thus the same SHM pattern, but different constant genes. The frequency of IGA1 and IGA2 transcripts with exactly the same CDR1–CDR3 increased with age (Figure 6E). In addition, we found more overlap between IGG1 and IGG2 sequences with increased age, while the overlap between IGG1 and IGG3 is deceased with increased age (Figure 6F). This might be at least partly related to the altered subclass distribution within different age categories. We also found sequence overlap between IGA and IGG sequences as summarized in Supplementary Table 3 in Data Sheet 2. The presence of these transcripts suggests that, during the course of an immune response, some B cells change their isotype without acquiring additional mutations, or B cells can directly switch to different subclasses.