Selection of individual V_H genes occurs at the pro-B to pre-B cell transition

Abstract

B cells are subjected to selection at multiple checkpoints during their development. The selection of antibody heavy chains is difficult to study because of the large diversity of the CDR3. In order to study the selection of individual antibody heavy chain variable region genes (V_H), we performed CDR3 spectratyping of 75–300 rearrangements per individual V_H in C57BL6/J mice. We measured the fraction of rearrangements that were in-frame (IF fraction) in B cell DNA. We demonstrate that individual V_H genes have different IF fractions, ranging from 10% to 90%, and that these IF fractions are reproducible in different mice. For most V_Hs, the IF fraction in pro-B cells approximated 33% and then shifted to the nearly final (mature) B cell value by the cycling pre-B cell stage. The frequency of high IF V_H usage increased in cycling pre-B cells compared to pro-B cells, whereas this did not occur for low IF V_Hs. The IF fraction did not shift as much in BCR-expressing B cells and was minimally affected by light chain usage for most V_H. High IF clan II/III V_Hs share more positively charged CDR2 sequences, whereas high IF clan I J558 CDR2 sequences are diverse. These data indicate that individual V_Hs are subjected to differential selection, that V_H IF fraction is mainly established through pre-BCR mediated selection, that it may operate differently in clan I vs. II/III V_Hs, and that it has a lasting influence on the antibody repertoire.

Introduction

Immunoglobulin loci of developing B cells undergo a series of DNA rearrangements (V(D)J recombination) that culminate in the assembly of antibody heavy and light chain variable regions (1–2). A large and diverse repertoire of antibodies is created by the recombination of multiple V, D and J segments; variability at the junctions between these gene segments; and numerous combinatorial possibilities for heavy + light chain pairing. But this diversity comes at a cost: non-productive (out of frame) rearrangements occur frequently and, among the rearrangements that are productive (in-frame, IF), many of the resulting antibodies are autoreactive (3).

The manner in which the primary antibody repertoire is purged of autoreactive receptors is fundamental to the understanding of self-tolerance. It is typically assumed that the major stage of bone marrow B cell development where censoring of the primary antibody repertoire takes place is at the Pre-B to naive B cell transition. At this stage, B cells with autoreactive IgM antibodies can edit their antibody receptor specificity by undergoing further light chain rearrangement (reviewed in (4)) or undergo clonal deletion. It seems logical for editing of autoreactivity to occur after the full antibody (heavy + light chain) has been formed, but several lines of evidence suggest that antibody heavy chains are also subjected to specificity-based selection during early B cell development.

For example, antibody heavy chains are thought to undergo “structural selection” at the pro-B to pre-B cell transition. Consistent with this model, heavy chains that pair well with the surrogate light chain result in a pre-BCR that is capable of signaling via associated Igα and Igβ domains (5), resulting in down-regulation of the V(D)J recombinase and IL-7-dependent proliferation in large, cycling pre-B cells (Hardy bone marrow (BM) fraction C’, hereafter Fr. C’ (6–8)). Ten Boekel, Rolink and Melchers showed that approximately half of antibody heavy chains were able to pair well with the surrogate light chain and promote B cell development (9). But why some heavy chains pair well with surrogate light chain and others pair poorly is incompletely understood, in part because early experiments did not distinguish between selection based on CDR3 (which generally derives a minority of its sequence from the V_H segment) vs. selection elsewhere in the V_H. Furthermore, the mechanism of pre-BCR signaling is unclear, e.g., is it ligand-dependent? Compelling data indicate that galectin 1, secreted by bone marrow stromal cells, binds to the λ5 unique region of the pre-BCR and influences pre-B cell differentiation and proliferation (10–11). On the other hand, the crystal structure of the human pre-BCR Fab, along with other biochemical data, suggest that the long tails of the surrogate light chains could mediate the self-assembly of multimers (5, 12–13). Furthermore, recent data implicate a role for N-linked glycosylation within the H chain constant region in surrogate light chain binding and pre-BCR crosslinking (14).

Other recent evidence suggests that antibody heavy chains in developing pre-B cells are subjected to distinctive selection, particularly with regard to autoreactivity: surrogate light chain knock-out mice develop spontaneous autoimmunity, including the production of anti-nuclear antibodies (15). The heavy chains of antibodies from surrogate light chain knock-out mice are enriched for positively charged arginine residues in the CDR3, resembling the canonical sequences of anti-DNA antibodies (15–16). In the absence of λ5, these heavy chains were expressed on the cell surface, bound to nuclear antigens such as DNA, and permitted B cell proliferation. In contrast, in the presence of surrogate light chain, freshly isolated anti-DNA pro-B cells were reduced in frequency after a few days in culture, suggestive of negative selection. Additionally, two groups have demonstrated that heavy chains with CDR3s that mimic or are derived from known autoantibodies promote surrogate light chain independent B cell maturation and proliferation (17–18).

Work from other laboratories supports the selection of D_H reading frames, CDR3 length and amino acid content during early B cell development (3, 9, 18–23). Earlier studies of antibody heavy chain selection also revealed a progressive shift in the frequency of V_H usage, particularly within the 7183 family, from 3’ (D-proximal) in the earliest stage B cells to more 5’ V_H usage in more mature B cell subsets. This shift in V_H usage is recapitulated during ontogeny, with fetal B cell repertoires exhibiting increased 3’ V_H gene usage (24–27). However, the analysis of selection based upon V_H frequency is problematic because multiple factors, including the frequency of rearrangement, can independently influence the observed V_H gene usage (28). Furthermore, the frequency of V_H rearrangement is not necessarily proportional to whether the V_H is positively or negatively selected. A case in point is the most 3’ V_H, 81X, which is frequently but often non-productively rearranged (29). The analysis of CDR3 sequences of certain V_H genes, including 81X, V_H12 and Q52 (9, 22–23, 30–31), clearly documents specific changes in the repertoire that occur during the pre-BCR selection step. These and other studies have led to the prevailing view that the CDR3, which contributes the lion’s share of H chain diversity (32), is the critical region for H chain selection (for example, (33–34)). In point of fact, however, the selective consequences of V_H sequences outside the CDR3 are almost entirely unknown. Only a handful of studies, focused on specific V_Hs such as 81X, V_H11 and V_H12, have documented selection of amino acids outside of the CDR3 (23, 30, 35).

A more systematic study of V_H selection would enhance our understanding of tolerance mechanisms that apply specifically to antibody heavy chains. For example, V_H replacement is a mechanism of antibody heavy chain editing that exchanges a V_H gene in an existing VDJ rearrangement with a new V_H gene by invasion into a cryptic heptamer near the 3’ end of the V_H (36–38). As such, V_H replacement does not remove the existing CDR3, but rather adds to it and swaps out the V_H gene. If the CDR3 is the dominant factor in heavy chain selection, this receptor editing mechanism should be ineffective. Understanding the developmental timing of V_H selection would also provide better insights into how the pre-BCR vs. the BCR are involved in autoantibody regulation (15).

Accordingly, we have applied a more general approach to the issue of H chain selection by measuring the fraction of in-frame rearrangements (IF fractions) of twenty different V_H genes in genomic DNA of B cells from C57BL/6J (hereafter B6) mice. In the B6 spleen, we found that different V_H genes had different IF fractions, ranging from 10% to 90%. Most of the clan I V_H genes, including all but one member of the J558 family that were evaluated, had a high IF fraction. In a more detailed analysis of BM B cell subsets, most V_Hs exhibited an IF fraction of approximately 33% at the pro-B cell stage and the IF fraction increased substantially for high IF V_Hs by the large cycling pre-B cell stage. These findings suggest that pre-BCR driven selection influences the V_H IF fraction. In support of this hypothesis, the IF fraction appears to be positively correlated with an increase in rearrangement frequency as cells progress from the pro-B to the pre-B stage. In contrast to the large changes in the V_H IF fraction at the pro-B to pre-B cell transition, changes in BCR-expressing B cell subsets tended to be more subtle and, for most heavy chains, light chain usage did not significantly alter the IF fraction. Taken together, these findings suggest that the major shift in the IF fraction occurs at the pro-B to pre-B cell transition and involves pre-BCR rather than BCR-based selection.

Materials and Methods

Mice

B6 mice were purchased from Jackson Laboratories (Bar Harbor, ME) and maintained at the University of Pennsylvania, School of Medicine under an Institutional Animal Care and Use Committee-approved protocol. Vκ8 light chain knock-in mice were maintained at the University of Chicago under an IACUC-approved protocol.

Flow cytometry

Cell suspensions from 3–4 month old B6 mice were prepared from femurs, tibias, and spleens in FACS buffer (PBS, 0.5% BSA, 0.01% NaN₃, 1 mM EDTA) after red blood cell lysis (ACK Lysing Buffer; BioWhittaker, Walkersville, MD). All antibodies were purchased from eBioscience (San Diego, CA), BD Biosciences (San Jose, CA) and BioLegend (San Diego, CA). The following antibody-fluorophore combinations were used to resolve BM subsets: FITC anti–heat-stable antigen (30F1), PE anti-CD43 (S7), PE-Cy7 anti-IgM (II/41), PE-Cy5.5 anti-CD19 (1D3), APC anti-CD93 (AA4.1), APC-AF750 anti-B220 (RA3-6B2), Pacific Blue anti-IgD (11–26). BM B cell fractions are defined as follows, based upon the scheme by Hardy (7): Fr.B-C (B220⁺CD19⁺CD43⁺IgM⁻IgD⁻CD24^dimCD93⁺), Fr.C’ (B220⁺CD19⁺CD43⁺IgM⁻IgD⁻CD24^briCD93⁺), Fr.D (B220⁺CD43⁻IgM⁻IgD⁻CD93⁺), Fr.E is B220⁺CD43⁻IgM⁺IgD⁻CD93⁺ and Fr.F is B220⁺CD43⁻IgM^dim/+IgD⁺CD93⁻. The following antibodies were used to resolve splenic B cell subsets: FITC anti-IgM (II/41), PE anti-CD21 (7G6), APC anti-CD93 (AA4.1) and APC-AF750 anti-B220 (RA3-6B2). Splenic B cell subsets were defined as follows: Transitional (B220⁺CD93⁺), Follicular (B220⁺CD93⁻IgM^dimCD21^dim) and Marginal zone (B220⁺CD93⁻IgM^briCD21^bri). For all sorts, dead cells were first eliminated by DAPI staining and cell doublets/aggregates by pulse width gating. The purities of the Fr. B-C sorted cell populations were 98–99%. FACS analysis and sorting were performed on the LSR II and FACSAria cytometers, respectively (BD Bioscience, San Jose, CA) in the University of Pennsylvania Flow Cytometry core facility. Flow cytometry data were analyzed using FlowJo software (version7.5.5, Treestar Inc., Ashland, OR). For sorting of λ+ B cells, splenic B cells were stained with APC anti-λ (RML-42), FITC anti-κ (187.1), APC-AF750 anti-B220 and PE-Cy5.5 anti-CD19.

CDR3 Spectratyping

CDR3 spectratyping was performed to evaluate the V_H IF fractions. Genomic DNA was isolated from either spleens or sorted cells, purified according to the manufacturer’s directions using PureGene (Qiagen, Valencia CA), and amplified using the V_H primers in Supplemental Table 1 and a labeled reverse primer in J_H2 (5’-FAM-CTG TGA GAG TGG TGC CTT G-3’). All primers were synthesized by Integrated DNA Technologies (Coralville, IN). Each 20 μL PCR contained 5-50 ng genomic DNA, 1x PCR buffer (AmpliTaq Gold^TM 10x PCR buffer I with 15 mM MgCl₂ (Roche Applied Sciences, Indianapolis, IN)), 0.2 mM of each dNTP (Promega, Madison, WI), 0.6 μM of each primer, and 1 U of DNA polymerase (AmpliTaq Gold (Roche Applied Sciences)). The cycling conditions were: 10 min at 94°C, followed by 40 cycles of 94°C for 30 s, T_a (Supplemental Table 1) for 30 s, and 72°C for 30 s, followed by 20 min at 72°C. To minimize notching artifacts due to non-uniform adenosine addition by the Taq polymerase, some V_H primers were modified by “PIGtailing” (incorporating the sequence GTTTCTT at the 5’ end, see italicized sequences in Supplemental Table 1). This modification resulted in nearly 100% adenylation of the 3’ end of the FAM-labeled opposite strand, consistent with what has been described previously (39). 2 μL of PCR products were resolved by capillary electrophoresis on an ABI 3100 analyzer (Applied Biosciences Inc., Foster City, CA).

CDR3 length analysis

Capillary electropherograms were generated and analyzed using ABI Genotyper 3.7 or ABI Peak Scanner Software, version 1.0. Peak sizes were interpolated using a ROX ladder “DS30” (Applied Biosystems) alone or combined with our homemade ladder (sizes of the homemade ladder are: 195, 198, 201, 204, 207, 210, 213, 216, 219 bp). The IF fraction was determined by counting the number of peaks that are IF and dividing by the total number of peaks (IF+OF). The CDR3 lengths were calculated, beginning with the conserved cysteine (Cys) residue at the 3’ end of the V_H (position 96) and ending with the last conserved glycine (Gly) in the middle of J_H2.

CDR3 peak count metadata

In Fig. 2B, the numbers of peaks and IF fraction (the latter in percent) for each V_H are: J558.88 n=92 76%, J558.85 n=86 84%, J558.72 n=124 85%, 3609.11 n=152 81%, J558.47 n=55 42%, V_H15 n=84 29%, V_H10 n=91 69%, 3609.1 n=109 42%, J606.1 n=178 81%, V_H12 n=218 21%, 3609N.2 n=371 36%, S107.3/4 n=107 82%, V_H11 n=109 27%, SM7.1/2/4 n=77 88%, S107.1 n=78 15%, Q52.8/5 n=63 38%, 7183.9 n=68 75%, 7183.4 n=100 60%, Q52.2 n=86 45%, 7183.2 n=93 13%. In Fig. 3A, the numbers of peaks and IF fraction analyzed for each subset are: J606.1, Fr. B-C n=83 39%, Fr. C’ n=231 87%; V_H10, Fr.B-C n=84 30%, Fr.C’ n=69 57%; V_HJ558.47 Fr. B-C n=56 38%, Fr. C’ n=22 64%; V_H3609N.2 Fr. B-C n=48 35%, Fr. C’ n=31 35%; V_H7183.2 Fr. B-C n= 69 32%, Fr. C’ n=55 29%. In Fig. 3B the numbers of peaks analyzed for each subset are: J606.1, Fr. B-C n=83, Fr. C’ n=231; V_HJ558.85 Fr. B-C n=42, Fr. C’ n=88; V_H3609N.2 Fr. B-C n=47, Fr. C’ n=21; V_HJ558.47 Fr. B-C n=18, Fr. C’ n=4; V_H7183.2 Fr. B-C n= 69, Fr. C’ n=55. The number of peaks in Fig. 3B are lower for some of the V_H than in Fig. 3A because in Fig. 3B all of the measurements were performed on highly diluted genomic DNA samples for optimal quantification of rearrangement frequency. In Fig. 4A, the numbers of rearrangements and the IF fractions analyzed for each subset and each V_H are: J606.1, Fr.B-C n=83 39%, Fr.C’ n=231 87%, Fr.D n=72 89%, Fr.E n=74 88%, Fr.F n=43 91%, Trans n=118 86%, FO n=123 82%, MZ n=273 87%; 3609N.2, Fr.B-C n=48 35%, Fr.C’ n=31 35%, Fr.D n=92 46%, Fr.E n=76 39%, Fr.F n=61 46%, Trans n=47 33%, FO n=68 31%, MZ n=61 38%; V_H10, Fr.B-C n=84 30%, Fr.C’ n=69 57%, Fr.D n=53 59%, Fr.E n=50 56%, Fr.F n=54 54%, Trans n=29 66%, FO n=19 68%, MZ n=75 68%. In Fig. 4B the numbers of rearrangements and their IF fractions are 7183.2-Vκ8 n=64 14%, V_H15-Vκ8 n=61 25%, 3609N.2-Vκ8 n=129 25%, J558.47-Vκ8 n=64 39%, 3609.1-Vκ8 n=109 32%, V_H10-Vκ8 n=35 69%, J558.88-Vκ8 n=49 84%, 3609.11-Vκ8 n=51 78%, J606.1-Vκ8 n=100 77%, J558.72-Vκ8 n=77 82%, SM7.1/2/4-Vκ8 n=50 90%; 7183.2-λ n=40 13%, V_H12-λ n=27 15%, 3609N.2-λ n=29 28%, V_H10-λ n=98 82%, 3609.11-λ n=55 89%, J606.1-λ n=31 77%, J558.85-λ n=77 91%, J558.72-λ n=79 85%, SM7.1/2/4-λ n=17 94%. For the V_Hs in this comparative analysis, the κ peak numbers and IF fractions are the same as in Fig. 2.

Figure 2. V_Hs differ in their IF fractions.

A. Mouse-to-mouse reproducibility of the V_H IF fraction. V_HIF fraction analysis was performed on spleen DNA from 3-month-old B6 mice (n=2) for V_H J606.1 (white bars) and 3609N.2 (grey bars). The numbers of J606.1- J_H2 rearrangements and their corresponding IF fractions (the latter in percent) were: mouse #1, 124 peaks, 80%; mouse #2, 194 peaks, 81%. For 3609N.2-J_H2: mouse #1, 290 peaks, 36%; mouse #2, 293 peaks, 33%. The IF fractions calculated for individual mice are plotted for each V_H. The difference in the average IF fraction between the two V_Hs is significant (p≪1 x 10⁻¹⁰ by cumulative binomial probability). B. Survey of the IF fraction in 20 different V_Hs. V_H IF fractions were measured in genomic DNA from 3-month-old B6 mice spleen DNA (n=2). V_H IF fractions are arranged in the order of their location in the germline locus, with 5’ V_Hs on the left. The numbers of rearrangements and the IF fractions analyzed for each V_H are given in the Materials and Methods section. Some of the primers bind to more than one V_H (see Supplemental Table 1 for primer information).

Figure 3. Establishment of the V_H IF fraction occurs early during B cell development.

A. V_H IF analysis in bone marrow pro-B (Fr. B-C) and cycling pre-B (Fr. C’) cells. IF fractions (given in percent) for J606.1-J_H2, V_H10-J_H2, J558.47-J_H2, 3609N.2-J_H2 and 7183.2-J_H2 rearrangements are plotted for each B cell subset. White bars denote Fr. B-C and gray bars represent Fr. C’. Asterisks denote IF fractions that differ significantly between Fr. B-C and Fr. C’ (p<0.05 by Binomial probability). B. V_H peak number analysis in Fr. B-C and Fr. C’. Plotted are the frequencies of V_HJ606.1-J_H2, J558.85-J_H2, 3609N.2-J_H2, J558.47-J_H2 and 7183.2-J_H2 rearrangements per 10,000 sorted B cells from Fr. B-C versus Fr. C’. The total raw peak numbers are given in the Materials and Methods section. Asterisks denote peak numbers that differ significantly between Fr. B-C and C’ (p<0.05 by Chi square test). Data for both panels in the figure are pooled from sorted cells of 3-month-old B6 (n=4–8 mice).

Figure 4. Minimal changes in the V_H IF fraction in BCR-expressing B cells.

A. V_H IF analysis in B cell subsets in bone marrow and spleen. IF fractions (given in percent) for J606.1-J_H2, 3609N.2-J_H2 and V_H10-J_H2 rearrangements are plotted for each B cell subset. The same data for Fr. B-C and C’ from Fig. 3A are re-plotted here for comparison. The numbers of rearrangements and the IF fractions analyzed for each subset and each V_H are given in Materials and Methods. Asterisks denote IF fractions that differ significantly between the indicated subsets (p<0.05 by Binomial probability). B. The V_H IF fraction is not significantly altered by light chain usage for most V_Hs. The V_H IF fraction is compared among splenocytes (mostly κ+, filled circles, same data as in Fig. 2B), λ+ B cells (open squares, n=2 mice) and Vκ8+ splenocytes (open triangles, n= 2 mice). The IF fractions (percentages) of different V_H-J_H2 rearrangements are plotted in order of increasing V_H IF fraction, based upon the unsorted splenocyte data. Only V_H10-J_H2 rearrangements display a statistically significant difference in the IF fraction in the different light chain expressing populations (black arrow, p<0.05 by the Mann-Whitney test). The numbers of rearrangements and their IF fractions are given in Materials and Methods.

Cloning of PCR products and sequencing

PCR products were cloned into pCR2.1 TOPO per the manufacturer’s directions (Invitrogen, Carlsbad, CA). Randomly selected clones were sequenced at the University of Pennsylvania DNA sequencing facility, and analyzed using Ig-BLAST (40).

Protein data analysis

The following crystal structure data were obtained from the protein database (PDB): 2H32 (human pre-BCR Fab, (12)), 3GHB (human V_H3 anti-HIV Fab, (41)) and 1KC5 (anti-hepatitis virus peptide Fab, (42)). All of the amino acid sequence alignments, 3-dimensional structure visualization and isoelectric point (pI) calculations were performed using Geneious software (Geneious v4.5, available from http://www.geneious.com/.) B6 germline V_H sequences (43) were analyzed for the CDR2 pI using Geneious software.

Statistical analysis and modeling

The overall IF fraction can be estimated using a simple model of rearrangement, as described in Fig. 1. Other mathematical models are described in the legend to Supplemental Fig. 2. In order to determine if two different V_Hs have a statistically significant difference in their IF fractions in spleen DNA, we used binomial probabilities. Statistical comparisons of median CDR3 lengths and CDR2 pI correlations with the IF fraction were performed with a 2-tailed Mann-Whitney test. Statistical significance of the rearrangement frequencies of low (IF fraction <50%) vs. high (>70%) IF V_Hs in pro-B vs. pre-B cells was analyzed using a Chi square test. P-values of less than 0.05 were considered statistically significant.

Figure 1. Measurement and theories of the V_H IF fraction.

A. Schematic of spectratyping PCR. Shown is a simplified H chain locus with two V_Hs (J606.1 and 7183.4), two D_Hs, two J_H gene segments and one C_H. The primers (arrows) specifically amplify rearrangements that involve the J606.1 V_H, either D_H and J_H2. The reverse primer in J_H2 has a fluorescent tag (star) that is used to analyze product sizes by capillary electrophoresis. Boxes denote exons, lines represent introns, and dashed lines show which gene segments participated in a V_HJ606.1-D_H-J_H2 rearrangement. B. Spectratypes of V_HJ606.1 - J_H2 rearrangements. Spleen DNA from 3-month-old B6 mice was amplified using a V_H-specific J606.1 primer and a FAM-labeled J_H2 primer. PCR products from five individual reactions are displayed as spectratypes. In each spectratype, the y-axis represents the fluorescence intensity and x-axis gives the product sizes in nucleotides (nt). Out of frame peaks (OF) are marked with asterisks. C. Calculation of the V_H IF fraction. The in-frame (IF) fraction of a single V_H amongst a population of B cells can be envisioned as a ratio of the number of productive (V_HDJ_H⁺) rearrangements to the total number of productive and non-productive rearrangements (V_HDJ_H⁺ + V_HDJ_H⁻) of that particular V_H in a population of B cells. B cells expressing a given V_H can have one of two genotypes, V_HDJ_H⁺/DJ_H or V_HDJ_H⁺/V_HDJ_H⁻. D. Model of V_H IF fraction under conditions of uniform selection. According to a simplified model of H chain gene rearrangement, B cells undergo a maximum of two rearrangement attempts per cell and rearrangement promptly ceases once a productive rearrangement has occurred (27). With estimated B cell genotype frequencies of 0.6 for VDJ⁺/DJ and 0.4 for VDJ+/VDJ-, the overall IF fraction (VDJ⁺/(VDJ⁺ + VDJ⁻), according to this model, is (0.6 + 0.4)/(0.6 + 0.4 + 0.4) = 0.71. E. Model of V_H IF fraction under conditions of differential V_H selection. Under conditions of differential V_H selection, V_Hs will have different IF fractions. According to this model, if a V_H has a high IF fraction, then B cells that express a functional H chain rearrangement (green cells) will be more frequent than B cells that harbor the same H chain as a nonfunctional rearrangement (red cells). The converse occurs if a V_H has a low IF fraction.

Results

Using CDR3 spectratyping, it is possible to survey the diversity of a single set of V_H and J_H gene rearrangements that include any of the D_H genes (Fig. 1A). When many antibody gene rearrangements are amplified from a peripheral B cell population, the CDR3 spectratype typically assumes a Gaussian distribution with peaks that appear at 3 nucleotide intervals, due to the over-representation of in-frame (IF) rearrangements. But when single V_H-J_H primer sets are used at limiting amounts of input DNA, it is evident that some of the rearrangements are out of frame (Fig. 1B, asterisks). To define a rearrangement as having an IF length, the number of nucleotides in the CDR3 that spans the 3’ end of the V_H, through the D_H gene segment and into the 5’ end of the J_H gene segment must be an integral multiple of three (Fig. 1C). All rearrangements that fail this test of size are, by definition, out of frame (OF). However, not all sequences that are of the correct IF length are productive because they could harbor stop codons. Thus, the measured IF fraction, [IF/(IF+OF)] x 100%, is a maximal estimate of the true IF fraction.

To obtain an accurate measurement of the frequencies of individual rearrangements by CDR3 spectratyping, B cell genomic DNA was diluted to very low concentrations so that it was unlikely that more than one rearrangement corresponded to a peak of a given size. The strategy was to perform several independent amplifications of the same DNA sample and count the numbers of IF vs. OF sized peaks to determine the overall IF fraction. To prove that single PCR amplifications were detecting only one rearrangement per peak, amplicons of a given size were cloned and sequenced from single reactions. In each case only one sequence of a given size was repeatedly recovered per reaction, as expected if each peak in a single reaction corresponds to one and only one rearrangement (Supplemental Fig. 1). We also performed computer simulations to estimate the probability of obtaining more than one rearrangement per peak (Supplemental Fig. 2A). This problem is analogous to estimating the likelihood of two people in a group having the same birthday, and the “collision probability” can be modeled using different underlying distributions of rearrangement data. Assuming a normal distribution with a standard deviation of 10 peaks, a range of 60 rearrangement lengths, and 3 peaks per PCR (which resemble our typical reaction conditions), the collision probability is approximately 10%.

The IF fraction differs in different V_H genes throughout the H chain locus

If V_H genes are not subjected to distinctive selective pressures, the expected IF fraction is approximately 70% (Fig. 1D). This assumes a simple model with instant and complete feedback inhibition of rearrangement once an IF rearrangement is produced and a maximum of two rearrangement attempts (27). Alternatively, if a particular V_H were favored, one would expect more of the B cells to express that V_H as an IF rearrangement than as an OF rearrangement (Fig. 1E). Conversely, if a V_H were counter-selected, more B cells should harbor that V_H as an OF rearrangement than as an IF rearrangement (Fig. 1E).

To determine if different V_Hs have different IF fractions, we first analyzed the IF fractions of two V_Hs (J606.1 and 3609N.2) which are closely positioned in the middle of the germline H chain locus in B6 mice. The measured IF fractions of J606.1 and 3609N.2 rearrangements were 80% and 34%, respectively. These findings were reproducible in two individual B6 mice (Fig. 2A). Differences in the IF fraction can be modeled using binomial probabilities (see Materials and Methods), and the difference between J606.1 and 3609N.2 is highly significant (n>200 rearrangements per group, p≪1x10⁻¹⁰). To further validate these findings, we performed DNA sequence analysis. Nine of eleven (~80%) sequences cloned from J606.1-J_H2 and 3 of 8 (~40%) from 3609N-J_H2 had IF V_H rearrangements, consistent with the spectratyping results (data not shown).

In order to determine if other V_Hs also differ significantly in their IF fractions, we next examined the IF fractions amongst other V_H families, including the most D-proximal V_H (7183.2, equivalent to 81X in BALB/c), the most D-distal V_H (J558.88), some B-1 lineage-associated V_Hs (V_H11, V_H12, Q52, S107), some that are overused in anti-nuclear antibodies (ANAs) (J558.85 (VMU3.2), J558.72 (V_H124), 7183.9) and one that is overused in non-ANAs (SM7 (V130), (44)). (The V_H nomenclature for B6 is applied to all V_H names (43) and D_H names (45) in this paper.) The measured IF fractions of different V_Hs are displayed in Fig. 2B. At 75 peaks per V_H, a difference of 10% or more in the IF fraction is statistically significant, based upon cumulative binomial probabilities (p<0.05). By this criterion, several of the V_Hs varied significantly in their IF fractions from one another, with a range of IF fractions of 10% to 90% (Fig. 2B). Hereafter we refer to the V_H IF fraction as being “high” if it is >70%, “intermediate” (50–70%) or “low” (<50%), based upon measurements in splenic B cells. 70% is the cut-off predicted by the standard rearrangement model (Fig. 1D), whereas any IF fraction below 50% requires some form of counter-selection. (Even if all B cells underwent complete V(D)J rearrangement on both alleles, the minimum V_H IF fraction in the absence of counter-selection would be 50%.) To determine if the V_H IF fraction is reproducible in different B6 mice, the data for each V_H were analyzed separately for each mouse and compared between mice (for all V_Hs having more than 20 peaks per mouse). This analysis revealed that the average difference in the IF fraction between different mice was 6% which was not statistically significant (data not shown).

We analyzed the same data set with respect to CDR3 length (Supplemental Fig. 3A). To compare CDR3 lengths between different V_Hs, we defined the CDR3 as beginning with the conserved cysteine (Cys) residue at the 3’ end of the V_H (position 96) and ending with the last conserved glycine (Gly) in the middle of J_H2. This definition of the CDR3 length allowed for accurate comparison of V_H sequence lengths when there is extensive nibbling of the J_H. The following V_H have additional nucleotides at the end of framework 3 (CAR): 3609.11 (4 nt), 3609.1 (6 nt), V_H12 (6 nt), V_H11 (2 nt), S107.1 (4 nt), Q52.8/5 (2 nt); therefore their mean CDR3 lengths are longer. There is no statistically significant correlation between the mean CDR3 length and the V_HIF fraction, if one stratifies the data based upon V_Hs (r²=0.03).

The major shift in the V_H IF fraction occurs at the pro-B to pre-B cell transition

We considered two possible explanations for differences in the IF fraction amongst rearrangements of different V_Hs. The first was that successful rearrangement of different V_Hs occurred with different probabilities. The second possibility was that all V_Hs rearranged randomly (e.g. each started off with an IF fraction of 33%) and were subsequently subjected to differential selection. To distinguish between these alternatives, we analyzed the IF fraction at different stages of B cell development, using sorted B cell subsets from the bone marrow of B6 mice (Fig. 3A and see Materials and Methods for sort parameters). We analyzed the IF fraction of a high IF (J606.1), one intermediate IF (V_H10) and three low IF (J558.47, 3609N.2 and 7183.2) V_H in BM pro-B cells. All five pro-B IF fractions are approximately 33%. Therefore, these data disfavor biased recombination as an explanation for the different V_H IF fractions.

To evaluate the role of the pre-BCR in shaping the V_H IF fraction, we next analyzed the IF fraction in cycling pre-B cells (Fr. C’). Consistent with earlier reports of H chain selection being tied to pre-BCR expression (9), the IF fraction increased for the high and intermediate IF V_H (p<0.05). Conversely, none of the low IF V_Hs exhibited a statistically significant increase in the IF fraction at the cycling pre-B cell stage.

The frequency of high IF V_H rearrangements increases in cycling pre-B cells compared to pro-B cells

To determine if the V_H IF fraction is linked to expansion at the cycling pre-B cell stage, we compared the frequency of V_H rearrangements in Fr. B-C vs. Fr. C’ (Fig. 3B). If the V_H IF fraction is linked to proliferation in Fr. C’, then B cells expressing a low IF V_H would be expected to undergo fewer rounds of cell division than B cells expressing a high IF V_H. Consistent with this prediction, the number of J606.1 and J558.85 (both high IF V_H) rearrangements increased in Fr. C’ compared to Fr. B-C (p<0.05), whereas the number of 3609N.2, J558.47 and 7183.2 rearrangements were either lower or equivalent in Fr. B-C and Fr. C’. These data are consistent with the hypothesis that B cells expressing high IF V_Hs undergo more rounds of cell division at the cycling pre-B cell stage than B cells expressing low IF V_Hs.

Minor shifts in the V_H IF fraction occur in BCR-expressing B cell subsets

To determine if BCR mediated selection also influences the V_H IF fraction, we analyzed a high IF V_H (J606.1), an intermediate IF V_H (V_H10) and a low IF V_H (3609N.2) at different stages of development (Fig. 4A). For J606.1 and 3609N.2, there were no statistically significant changes in the IF fraction in BCR-expressing B cell subsets. Inthe case of V_H10, however, there were smaller but statistically significant changes that occurred beyond the cycling pre-B cell stage. To further evaluate the influence of the BCR on the V_H IF fraction, we measured the V_H IF fraction for several different V_H genes in populations of splenic B cells that differed with respect to their BCR light chains (Fig. 4B). Unsorted B6 splenocytes express mostly kappa light chains, whereas lambda B cells and Vκ8 B cells express more restricted light chain repertoires. When the V_H repertoires of these different light chain expressing B cell populations were compared, only V_H10 rearrangements exhibited a statistically significant shift in the IF fraction in the λ-expressing B cells compared to Vκ8 or kappa expressing B cells (Fig. 4B). When the average CDR3 lengths were compared, V_H10, J606.1, J558.85 and J558.72 rearrangements differed significantly between the κ-expressing splenocytes vs. the more light chain restricted B cell populations (Supplemental Fig. 3C). Taken together, these data indicate that the major shift in the IF fraction occurs between Fr. B-C and Fr. C’. Furthermore, the stability of the V_H IF fraction in BCR-expressing developmental stages and populations with different light chains suggests that the V_H IF fraction is relatively insensitive to BCR-mediated selection.

Sequence comparison and structural modeling of V_Hs

The preceding experiments indicate that the V_H IF fraction is determined mainly by selection at the pro-B to pre-B cell transition and may be linked to pre-B cell proliferation in cycling pre-B cells. To search for a mechanism for this selective process, we compared the amino acid sequences of high vs. low IF V_Hs. We began by aligning all of the V_H amino acid sequences and looking for motifs that were shared amongst the high IF V_Hs. Individual V_Hs can be grouped on the basis of their framework sequences into larger families called clans. Clan I contains the large D-distal J558 family, as well as SM7 and V_H15. Clans II and III contain all of the remaining V_H families.

We noticed that nearly all of the clan I V_H genes had a high IF fraction. When we used a degenerate primer to amplify several members of the J558 V_H family on B6 splenocytes, the spectratypes yielded a telltale pattern of peaks oscillating at 3 nucleotide intervals, with virtually no peaks in between, consistent with a very high overall IF fraction for the entire J558 family (data not shown). The only low-IF J558 family member that we identified was J558.47, which had a promoter that lacked a TATA box, unlike other J558 family members (43). The only other clan I member with a low IF fraction was V_H15, which had a very different amino acid sequence from all other clan I (or clan II/III) family members. When we surveyed the CDR2 sequences of different J558 family members, we observed considerable variation in the isoelectric point (pI) and hydrophobicity, but all of them appeared to have high IF fractions when analyzed in individual spectratyping studies with V_H-specific primers (data not shown). Thus nearly all of the J558 family members that were surveyed appear to have high IF V_Hs, despite a diversity of CDR2 sequences. As expected, based upon these data, there was no statistically significant correlation between the CDR2 pI and the V_H IF fraction amongst clan I V_Hs (data not shown). In contrast to clan I, the IF fractions of clan II/III V_Hs were far more variable. Also of note, high IF clan II/III V_Hs tended to have a high frequency of positively charged residues-- specifically, the pIs in the β-strand region within their CDR2s were high (Fig. 5).

Figure 5. Modeling of charged motifs in clan II/III V_H CDR2.

A. Isoelectric point (pI) analysis of the CDR2 in high versus low IF clan II/III V_Hs. Plotted is the mean CDR2 pI +/− standard deviation for each V_H group. Each dot corresponds to a single V_H sequence. Only the non-conserved 5’ portion of the CDR2 sequence is used for this analysis (see boxed regions in Fig. 5B). The high IF group is defined as having an IF fraction of greater than or equal to 50% and low IF is less than 50%. The difference between the high IF vs. low IF clan II/II CDR2 median pI is significant (p<0.01 by Mann-Whitney test). B. Alignment of high IF vs. low IF germline V_H amino acid sequences. The amino acid sequences are based on the germline sequences for B6 (from (43)). The CDR2 is indicated by a horizontal line and the consensus sequence is given on top of each alignment. The motifs used for isoelectric point analysis are enclosed in boxes.

Discussion

The data presented herein demonstrate that individual V_H are subjected to distinctive selection. Each V_H has a characteristic fraction of in-frame rearrangements, and that IF fraction is reproducible in different mice. The IF fractions of twenty different V_Hs range from 10% to 90%. For the V_Hs studied in detail, the IF fraction is determined at the pro-B to pre-B cell transition, implicating the pre-BCR in V_H selection.

How can the range of IF fractions in different V_H be explained? We propose that B cells expressing low IF V_Hs undergo fewer rounds of pre-B cell division than B cells expressing high IF V_Hs. The observed range of cell divisions at the pre-B stage is 4-6 (13), which could readily accommodate the observed range of IF fractions (Supplemental Fig. 2B and 2C). This model predicts that pre-B cells that have undergone more rounds of division will be more likely to express high IF V_Hs and is supported by our analysis of peak number frequencies in Fr. B-C vs. C’, which increase in Fr. C’ for a high IF V_H but not for a low IF V_H. This model also predicts that the IF fractions of high vs. low IF V_Hs will diverge with increasing numbers of pre-B cell divisions (the IF fraction of high IF V_Hs will get higher and low IF V_Hs will get lower).

Individual V_Hs seem to either have a high or a low IF fraction. There were comparatively few V_Hs that had intermediate IF fractions. The simplest explanation for this seemingly bimodal distribution is that high IF V_Hs pair well with surrogate L chain and/or signal well through Igα and Igβ. V_Hs reported to function poorly in pre-BCR signaling or expression, including 7183.2 (81X), V_H11, and V_H12, all had low IF fractions. Furthermore, many of the low IF V_Hs that we surveyed had non-canonical promoter sequences. For example, the only J558 family member with a low IF fraction, J558.47, has a promoter that lacks a TATA box (43). All other members of the J558 family that we studied had high IF fractions and the J558 family is known to exhibit the highest level of transcription of any of the mouse V_H families studied (46). 7183.2 lacks two Pu.1 binding sites and an initiator sequence that are shared by most of the other 7183 family members (43). Similarly, Q52.8, 3609N.2 and 3609.1 have promoter sequences that either lack elements or exhibit altered spacing of elements compared to other V_Hs in their families with higher IF fractions (43). These promoter elements may be weak, potentially resulting in lower RNA transcript abundance and, ultimately perhaps, lower levels of H chain protein expression. If the amount of H chain protein is rate-limiting for pre-BCR assembly, higher levels of H chain proteins could result in increased pre-BCR signaling, driving more rounds of pre-B cell division.

An important unresolved question is whether or not the V_H IF fraction is correlated with autoreactivity. In this connection, it is intriguing that most high IF clan II/III V_Hs have high isoelectric points in CDR2. It is possible that autoreactive or multireactive high pI clan II or III V_Hs are positively selected at the pre-BCR checkpoint. Positive selection of autoreactive H chains has been suggested for B-1 and fetal B cells, which tend to be enriched for the expression of certain clan II/III V_Hs (47–49). The enrichment of autoreactive or multireactive natural (auto)antibody specificities in the B-1 repertoire may represent an innate system of defense against commonly encountered pathogens (48, 50–51). Perhaps these positively charged residues mediate binding to a negatively charged selecting ligand or alter the overall pre-BCR structure in such a way that binding is possible (10, 52–53). Another possibility is that positively charged amino acids in the CDR2 of clan II/III high IF V_H interact with negatively charged residues in the VpreB tail. The murine VpreB tail contains both positively and negatively charged amino acids, it is possible that the VpreB tail forms an amphipathic helix in which one face exhibits mostly positively charged residues whereas the other face exhibits mostly negatively charged residues.

In contrast to clan II/III V_Hs, clan I V_Hs exhibit a diversity of CDR2 amino acid sequences with varying pIs. Despite the variations in the CDR2 pI, nearly all clan I V_Hs have high IF fractions. This could mean that sequences outside of CDR2 that are conserved between different J558 family members are conferring a selective advantage upon clan I V_Hs. Or, perhaps J558 V_Hs undergo selection for charge motifs that can at least partially cancel one another out. For example, J558 V_Hs with negatively charged CDR2 sequences could have more positively charged CDR3 sequences and vice versa. Such a model of charge distribution amongst the CDRs is reminiscent of the earlier observation that antibodies with anti-DNA H chains (which usually have high pIs due to the presence of arginine residues (16)) often have L chains with low pIs (54). Similarly, the selection for antibodies that contain both positively charged and negatively charged CDRs typifies what Mohan and colleagues have described as “hardwiring of autoreactive motifs” into the antibody repertoire (55). In particular, the presence of a negatively charged Asp residue at H50 can veto DNA binding by positively charged amino acids in the CDRs. By selecting for H chains with positively charged or negatively charged CDRs or, as we argue here, potentially both, one creates the opportunity for single antibodies to bind to multiple and varied antigens, including autoantigens (56). We focused on CDR2 because it was the most variable germline-encoded sequence in the V_H, but it is entirely possible that sequences outside of CDR2 are important for ligand binding or surrogate light chain interaction or both. For example, within the high IF clan I V_Hs, there are hydrophobic residues at positions 20, 51 and an arginine residue at position 40, which are not found in some of the low IF clan I V_Hs.

Alternatively, rather than positive selection of high IF V_Hs by binding to self- antigen, low IF V_Hs could be subject to negative selection, either by V_H replacement or apoptosis. With regard to V_H replacement, it is interesting that most of the D-distal V_H genes (most members of the J558 family) have high IF fractions. By virtue of their position in the H chain locus, clan I V_Hs would be the ones most likely to replace the more D-proximal low IF clan II/III V_Hs. As discussed above, it is possible that structural features of clan I V_Hs make them more suitable as high IF V_Hs. Alternatively, a high IF fraction of distal V_H genes could be explained by the timing of different V_H rearrangements; D-proximal V_H genes usually rearrange first and by the time a distal V_H gene has rearranged, perhaps the window of opportunity for further rearrangement is more limited, causing fewer B cells with non-productive distal V_H rearrangements to survive. With regard to apoptosis, an analysis in mice with Bcl-xL transgenes revealed an increased frequency of pro-B cells and an accumulation of non-productive rearrangements as well as rearrangements involving D reading frame 2 (57). These data suggest that B cells have a limited time window in which H chain rearrangements are permitted prior to death (they have a high crash factor). In a subsequent study, the same group showed that self-reactive Bcl-xL transgenic B cells were able to escape death but were regulated by other pathways that enforced self-tolerance, namely anergy and receptor editing (58).

Curiously, we were able to repeatedly recover two sequences from independent PCR amplifications V_H J558.85-J_H2 rearrangements in pro-B cells (Supplemental Fig. 1C). This finding is perplexing as pro-B cells are not known to proliferate. These sequences were not observed in water or fibroblast DNA, but we cannot unequivocally rule out the possibility of PCR contamination. We did not observe clonal expansions of any other V_H rearrangement in pro-B cells. The mechanism by which these sequences arise is unclear, but we considered the possibility of homology mediated joining, given the findings of Chukwuocha, Nadel and Feeney, who observed that a V_H in the S107 family exhibited an elevated frequency of IF rearrangements in cytoplasmic immunoglobulin negative pre-B cells from newborn mice (59). However, in the case of the rearrangements observed herein, they were observed in adult mice and there is no obvious sequence homology between J558.85 and D_H DSP2.9 or an alteration in the RSS of J558.85 that readily explains these findings.

Perhaps the most intriguing property of the V_H IF fraction is that it is nearly entirely established by the cycling pre-B cell stage: for most V_Hs the IF fraction did not shift nearly as much beyond the cycling pre-B cell stage as it did before it. Furthermore, the IF fraction of most V_Hs does not differ significantly in λ vs. κ-expressing B cells. The lack of large differences in IF fraction in more mature B cell subsets suggests that the repertoire of H chains is either so dominant and/or so large that many different rearrangements of a given H chain can preserve the IF fraction, even when there is considerable narrowing of the repertoire. Narrowing of the repertoire beyond the naive B cell stage is strongly supported by the reduced level of light chain diversity in mature recirculating compared to naive bone marrow B cells in a J558 heavy chain transgenic mouse model (60). Despite narrowing with regard to particular H+L chain pairs, the antibody repertoire can nevertheless be extremely diverse. The elegant study of Owen and Klinman demonstrated massive clonotypic diversity in the IgM response to phosphorylcholine, despite this response being restricted to only a few V_H (61). Thus, the H chain repertoire is established early during B cell development yielding the IF fraction or selection set-point for each V_H. From a nearly limitless diversity of H chains, antibody specificities are subsequently narrowed to a more “useful” array of protective specificities in more mature B cell stages, based upon H+L chain pairing and further fine-tuning of antigen specificity via somatic mutation.

Overall, our data suggest that the H chain IF fraction can be used as a measure of H chain selection at the pro-B to pre-B cell transition. An altered IF fraction could point to an early B cell tolerance checkpoint defect or to altered early B cell development. B cell tolerance checkpoint defects have been observed in animal models and human subjects with lupus and type 1 diabetes, as early as the L chain rearrangement stage in pre-B cells (62–63). The H chain IF fraction potentially surveys an even earlier stage of B cell development and selection. It is intriguing that certain V_Hs such as V_H4-34, V_H3-23, V_H1-69, V_H3-30 and others are over-represented in humans with autoimmune disease or in polyreactive antibodies (64–69). It will be interesting to determine if the V_H IF fraction for these and other V_Hs is different in individuals with autoimmunity, potentially providing insights into the stage of B cell development where immune tolerance is defective in individual patients with autoimmunity. A measure of early B cell tolerance or development, such as the V_H IF fraction, may be helpful in targeting therapy for autoimmunity to the relevant B cell subset(s). In addition, an abnormal IF fraction of a single germline V_H could reveal a more specific change in the antibody repertoire, such as clonal expansion in response to a particular (auto)antigen (70).

non-standard abbreviations

IF fraction

fraction of rearrangements that are in-frame

IF

in-frame

OF

out of frame

V_H

antibody H chain variable region gene

Fr. B-C

bone marrow pro-B cells

Fr. C’

bone marrow cycling pre-B cells