Self-reactive immunoglobulins (Ig) are generated at high frequency during normal B-cell ontogeny in the bone marrow. Functional Ig genes result from the combinatorial joining of gene segments from two (light chain) or three (heavy chain) classes; V (variable), D (diversity), and J (joining) [1, 2]. This process, known as V(D)J recombination, generates tremendous receptor diversity through the pairing of various gene segments, the selection of the recombination sites at which the segments are joined, and the addition of non-templated nucleotides (n-nucleotides) between adjoining gene segments. These processes are stochastic and can produce 10¹⁴ or so different protein specificities to be generated from the fewer than 100 gene segments at the immunoglobulin heavy chain locus [3]. In response to antigenic stimuli, mature B-cells undergo further diversification through somatic hypermutation (SHM), whereby mutations are introduced into the rearranged immunoglobulin gene at a rate approximately 10⁶ times higher than the normal background rate [4]. The adaptive immune system generates diversity and adapts its antigen receptors via stochastic somatic processes as microbes themselves diversify and adapt through Darwinian evolution. With such randomness in the formation of antigen receptor genes, it is inevitable that autoreactivity will arise. Indeed, 55–75% of early bone-marrow B-cells express polyreactive and self-reactive surface Ig [5]. Three primary B-cell-specific mechanisms for avoiding Ig-mediated autoimmunity have been identified: selective deletion, anergy and receptor editing [6–10].

Experiments using site-directed mutagenesis and CDR3 replacement have shown that the heavy chain CDR3 provides the primary structural correlates of polyreactivity [11, 12]. Shiokawa et al [13] found that CDR3 length differs between neonates and adults and provided evidence for somatic selection on CDR3 length during B-cell development. Aguilera et al [14] examined 8 autoreactive monoclonal antibodies, and found no clear relationship between polyreactivity and heavy chain CDR3 length. Wardemann et al [5] pooled immature and bone marrow "new emigrant" B-cells to show that cells reactive to HEp-2 cell exctract are biased toward longer CDR3 compared to those that are not so reactive.

Autoantibodies are commonly present in healthy human serum [15]; in murine models, positive selection for some autoreactive antibodies has been observed [16]. It is possible that these autoantibodies result largely from "leakiness" of negative selection in the bone marrow, in which case one expects autoantibodies in the periphery to bear the same genetic signatures as autoantibodies in immature B cells. But they may also arise in the periphery via somatic mutation after central tolerance has acted. In this case, we might expect to find a genetic signature of autoreactivity, which may differ from that common to pre-selection autoreactive cells in the bone marrow.

To determine the genetic signature of peripheral autoreactivity, we performed a comprehensive and comparative study of over 7,300 Ig collected from Genbank. We assembled four sets of human heavy chain genes for comparison: a set of productive Ig genes for which evidence of autoreactivity does not exist (P); a heterogeneous set of genes annotated by their discoverers as associated with autoreactivity, but exlcuding rheumatoid arthritis (A); a set genes annotated as associated with rheumatoid arthritis (RA); and a set of genes (NP) that appear to have been rearranged out-of-frame and thus nonproductively.

We performed a detailed analysis of each gene set in terms of gene segment composition, CDR3 length, n-nucleotide addition, and mutation frequency, and employed statistical methods to detect biases that may exist among these sets. We find differential biases in gene segment usage and n-nucleotide tract length, but not in CDR3 length between autoreactive and non-autoreactive productive genes. We do, however, find a striking increase in the proportion of n-nucleotides in CDR3 from autoreactive genes. This fact suggests to us that germline gene segments have evolved under selective pressure, not only to avoid autoreactivity, but to avoid acquiring autoroactivity through somatic mutation. Consequently, autoreactivity in the periphery arises more frequently by somatic mutation in those genes in which the CDR3s are encoded more extensively by n-nucleotides rather than germline gene segments.

In spite of selection against B cells bearing autoreactive antigen receptors during ontogeny in the bone marrow, substantial frequencies of autoreactive B cells in the periphery are observed. There are two hypotheses, not mutually exclusive, to account for this fact. One such hypothesis holds that such selection is intrinsically leaky, and that autoreactivity in the periphery is due largely to original autoreactivity that has slipped through the selective filter. The other hypothesis does not depend on leakiness of central B cell tolerance, but instead relies on the generation of autoreactivity by somatic mutations incorporated after primary B cell development.

These two hypotheses make different predictions regarding the relationships one will find between the genetic signatures of autoreactive Ig in the bone marrow and those in the periphery. If peripheral autoreactivity is due primarily to the escape of original autoreactivity from central tolerance, we expect to find that autoreactive peripheral B cells will be similar to pre-selection and autoreactive B cells in the bone marrow. The main distinguishing feature of such cells is a long CDR3. We find no statistically significant difference between CDR3 lengths in autoreactive and non-autoreactive Ig, though CDR3 is much longer in NP sequences than in any other class, confirming the observation by others that selection during primary ontogeny favors shorter CDR3.

No such correspondence is necessary under the second hypothesis. What one does expect to see under the second hypothesis are differences in somatic mutation, and we do indeed find that the mutation frequency is about 50% higher in autoreactivity-associated genes than in non-autoreactive genes. This result is consistent with many previous studies of the role of somatic mutation in autoimmunity [17–22]. Beyond the simple observation of differential mutation frequencies, however, we do see additional genetic biases that distinguish the autoreactives from the normals.

We observe, for example, differential usage in both VH and JH gene segments, with an increase in VH1 family usage in genes from both the general autoimmune (A) and rheumatoid arthritis (RA) datasets, as well as a corresponding reduction in family VH4 usage. Family VH4 is typically second-highest in usage frequency and has been reported to provide some of the most frequently used segments in the adult Ig repertoire, particularly V4-34 [23–26].

If peripheral autoreactivity were due largely to the leakiness of central tolerance, we would expect gene segment usage in autoantibodies to trend toward the pre-selection repertoire represented by the non-productively rearranged genes. This is not what we find. In fact, the VH segment use in the A and RA genes is not intermediate between the NP and normal genes, it is less like that of the NP genes than the normals are. VH1, for example, has a higher frequency in the normal genes than in the NP genes, and higher frequency still in the A and AR genes, while VH4 is lower in normals than in NP and lower still in A and AR.

The selective enhancement of usage for VH1 and VH5 segments in the A and RA genes may be due partially to the presence of charged residues in the framework region of these gene segments. Studies of autoimmune anti-erythrocyte human protein antibodies and non-autoimmune anti-staphylococcal antibodies show that framework region 1 may play a critical role in effective binding to these antigens [27, 28]. The VH1 and VH5 family genes contain a lysine-lysine amino acid motif in framework 1, which would produce a positively charged region in the folded protein. No other VH segment contains a contiguous pair of positively charged residues. We did also observe an enhancement of VH5 segment usage in the RA genes. Thus, it may be possible that part of the framework region 1 of the VH1 family genes promotes binding to some self-antigens.

We did observe, however, significant differences in the proportions of N-nucleotides among CDR3 nucleotides between either autoreactive set and the normals. This result is consistent with those from studies showing that TdT-deficient mice crossed with autoimmune-prone mice, which make B- and T-cells with nearly no N-nucleotides, have lower incidence of autoimmune disease and longer life spans than non-TdT deficient autoimmune-prone controls [29, 30].

These analyses suggest to us that the premutation genetic signatures observed in the autoreactive sets are indicative of autoreactive potential, and mark those genes that are more likely to give rise to autoreactivity under somatic mutation. B-cell central tolerance can only counter-select V(D)J rearrangements that are already autoreactive, not those with a tendency to become autoreactive. Germline evolution, however, can and does select gene segments for their potential repertoire under somatic mutation [31, 32].

The idea is that germline VH, DH, and JH gene segments have been shaped by evolution to avoid becoming autoreactive too easily under somatic hypermutation. When much of CDR3 is encoded by N-nucleotides rather than by these time-tested gene segments, autoreactivity may be much closer in genotype space than it would otherwise be.

We have examined the genetic correlates of B cell autoreactivity by compiling a large set of Ig genes associated with autoimmunity and comparing these genes statistically to normal and non-productive Ig gene sequences. We do not find a difference in CDR3 length between the autoimmunity-associated genes and the normals, as one might expect. There are, however, marked differences in mutation frequency between these sets, differences in gene-segment usage frequency, and, most strikingly, an increased use of N-nucleotides in the autoimmune genes. The hypothesis that ties these results together is that the premutation genetic signature of peripheral B-cell autoreactivity is not related to autoreactivity per se (because autoreactivity is largely eliminated by counter-selection in the bone marrow) but to autoreactive potential, the proclivity to become autoreactive under somatic hypermutation.