Somatic Mutation and Light Chain Rearrangement Generate Autoimmunity in Anti–Single-Stranded DNA Transgenic Mrl/lpr Mice

The construction of site-directed transgenic (sd-tg) mice expressing the H and/or L chain genes coding for well-defined anti-DNA Abs has been described previously 22,23. BALB/c 3H9Vκ8 sd-tg mice (3H9Vκ8/BALB/c) were crossed onto the MRL-lpr/lpr background and backcrossed three times to generate 3H9Vκ8 MRL-lpr/lpr mice (3H9Vκ8/lpr). Animals homozygous for the lpr gene were identified by two PCR assays using tail DNA. In brief, 1–2 mm of tail was snipped off and placed into 80 μl tail digestion buffer (50 mM Tris-HCl, pH 8.0, 50 mM KCl, 2.5 mM EDTA, 0.45% NP-40, and 0.45% Tween 20) containing 2 μl proteinase K (20 mg/ml; Boehringer Mannheim). After overnight incubation at 55°C, tail samples were boiled for 10 min and kept on ice. 1 μl of tail DNA was used for both PCR assays. The oligonucleotides used for these PCRs have been described previously 24. They were Fas-I2F (forward: 5′ agcatagattccatttgct 3′), with Fas-Z8 (reverse: 5′-caaattttattgttgcgaca-3′) to identify the lpr allele, or Fas-I2B (reverse: 5′-agtaatgggctcagtgca-3′) to identify the wild-type fas allele. PCR amplifications were set up in 50-μl vol containing 1 U of AmpliTaq Gold™ (Perkin Elmer Corp.), 1× buffer II, 250 μM of each dNTP, 2.5 mM MgCl₂, and 40 pmol of each primer. Amplifications were carried out in an OmniGene Hybaid thermocycler (Hybaid) under the following conditions: denaturation/enzyme activation for 9 min at 92°C; then 35 cycles of 30 s each at 94, 56, and 72°C; final elongation at 72°C for 7 min.

B cell hybridomas were generated from unmanipulated spleen cells from a 2-mo-old 3H9Vκ8/lpr mouse using SP2/0 myeloma cells 25 as the fusion partner. Spleen cells from 3H9Vκ8/BALB/c mice were stimulated in vitro for 3 d with 20 μg/ml LPS (Sigma Chemical Co.) and fused as previously described 13. Hybridomas were plated at limiting dilution, and only wells bearing single colonies on 96-well plates containing <30 hybrids per plate were expanded for analysis.

Isotypes were determined using an indirect solid-phase ELISA as previously described 26. Plates were coated with anti-Ig Ab and developed with alkaline phosphatase–labeled anti-IgM, anti-IgG, anti-IgA, anti-κ, or anti-λ. The enzyme activity was revealed by the substrate p-nitrophenyl phosphate (Sigma Chemical Co.) and optical density was read at 405 nm. All commercial Abs were from Southern Biotechnology Associates, Inc. Samples were titrated, and concentrations were estimated by comparison to a standard curve.

Binding to ssDNA was measured by solid phase ELISA. Immunolon^® 4 plates (Dynex Technologies, Inc.) were coated with 60 μl of histone-free protamine (Sigma Chemical Co.) at 500 μg/ml. Then, 60 μl of boiled and quickly chilled salmon sperm DNA (90 μg/ml) was allowed to bind to protamine-coated plates overnight at 4°C in a humid chamber. The next day, the plates were blocked with PBS containing 1% BSA. After washing with PBS-Tween, serum samples or hybridoma supernatants at various dilutions were added and incubated for 2 h at 37°C. To detect DNA binders, alkaline phosphatase–labeled mouse isotype-specific antibody was used (Southern Biotechnology Associates, Inc.), and the enzyme activity was revealed by the specific substrate p-nitrophenyl phosphate (Sigma Chemical Co.). Optical density was read at 405 nm.

Binding to dsDNA was measured by a two-step solution phase ELISA as previously described 27. In brief, antibody concentrations in the supernatants were first standardized to 2 μg/ml. Then, supernatants and protein-free calf thymus biotinylated DNA (1 μg/ml) were mixed, incubated at 37°C for 60 min, and transferred to microtiter plates (Immunolon^® 4; Dynex Technologies, Inc.) coated with pure avidin (10 μg/ml, 50 μl/well; Avidin D; Vector Labs., Inc.). After a 60-min incubation at room temperature, plates were washed, and DNA–Ab complexes were detected with alkaline phosphatase–conjugated goat anti–mouse isotype-specific Ab (Southern Biotechnology Associates, Inc.).

Antinuclear specificities of mAbs and serum Igs were determined by indirect immunofluorescence staining of HEp-2 cells using an ANA test kit (The Binding Site, Inc.) according to the manufacturer's directions. Substrate cells were incubated with culture supernatants or serum at different dilutions for 20 min, washed in cold PBS/BSA 1%, and stained with Alexa™ 488 goat anti–mouse IgG (Molecular Probes, Inc.). Slides were viewed under a Zeiss LSM510 confocal microscope using a C-apochromat 40×/1.2 NA water immersion objective. Positive ANAs were further tested for their capacity to bind dsDNA of the trypanosoma Crithidia luciliae using an anti-nDNA antibody test kit (CrithiDNA; Antibodies Inc.) following the manufacturer's directions. The procedure is the same as that described for the ANA test.

Genomic DNA was purified from individual hybrids as previously described 26. Primers and conditions used for H and κ chain PCR assays have already been detailed (references 22, 23, 26, and Fig. 1). A LD/JHCH PCR was designed (see below) and used as a first approach to distinguish between 3H9V_H replacement 22 and somatic mutation because somatic mutation in the CDR3 of the 3H9 sd-tg was shown to prevent amplification in the LD/CDR3 PCR. All hybrids negative in the LD/CDR3 PCR were tested with the LD/JHCH PCR. This PCR is not as specific as the LD/CDR3 PCR, because the JHCH primer is not sd-tg specific, and the 3H9 leader primer binds V genes with similar leader sequences. Hence, 3H9 sd-tgs with mutations in CDR3 as well as V_H replacements where the invading V_H gene uses a 3H9-like leader sequence will be detected in this assay. All PCR amplifications were carried out in a TouchDown thermocycler (Hybaid). All primers were made by the Princeton University Synthesizing/Sequencing Facility.

Genomic DNAs from individual hybridomas were used as templates from which to amplify V_H and Vκ genes. PCR amplifications were set up in a 30 μl vol containing 1 U of AmpliTaq Gold™ (Perkin Elmer) 1× buffer II, 200 μM of each dNTP (Boehringer Mannheim), 50 pmol of each primer, and 1.5 mM MgCl₂ (Perkin Elmer). The 5′ primers used to amplify V_H regions were either LD3H9 (5′-ctgtcaggaactgcaggtaagg-3′) or VH5.3 (5′-(g/c)aggt(g/t)cagctgcag(g/c) agtctgg-3′) in combination with a primer located in the JH–CH intron (JHCH: 5′-cttctctcagccggctccctc-3′). For PCR detection of Vκ genes, the forward primers were either specific for Vκ8 (MW133: 5′-ggtacctgtgggacattgtg-3′) or degenerate Vs (5′-ggctgcag(c/g)ttcagtggcagtgg(a/g)tc(a/t)gg(a/g)ac-3′) 28 and the reverse primer was a primer specific for the Jκ–Cκ intron (MW176: 5′-tgccacgtcaactgataatgagccctctc-3′). Amplifications were carried out with a thermal reactor (subambient TouchDown; Hybaid) as follows: V_H: 92°C for 9 min, followed by 38 cycles of denaturation for 45 s at 94°C, annealing for 50 s at 56°C, extension for 1 min and 40 s at 72°C, and a final extension of 7 min at 72°C. Vκ8: 92°C for 9 min, followed by 38 cycles of denaturation for 30 s at 94°C, annealing for 30 s at 56°C, extension for 30 s at 72°C, and a final extension of 7 min at 72°C. Vs: 38 cycles of denaturation for 30 s at 94°C, annealing for 40 s at 65°C, extension for 1 min and 40 s at 72°C, and a final extension of 7 min at 72°C. Each fragment was size selected on a 1.5% agarose gel (Ultra Pure Agarose; GIBCO BRL) and purified by QIAquick gel extraction (Qiagen Inc.). Nucleotide sequencing was performed using the ABI Prism™ Big Dye™ Terminator Cycle Sequencing Ready Reaction Kit with AmpliTaq^® DNA polymerase, FS, according to the manufacturer's directions (Applied Biotechnology, Inc.). Reactions were run on the Applied Biosystems 377 PRISM automated DNA sequencer (PE Applied Biosystems).

Total RNA was isolated from hybridomas using the RNAeasy kit (Qiagen Inc.), and 10 μg RNA was converted into cDNA using avian reverse transcriptase (Promega Corp.) and an oligo-dT/random hexamer primer mixture. 1 μl of cDNA was then amplified by PCR using Vs 28 as a forward primer and Cκ-specific primer as a reverse primer (Cκ: 5′-tggatggtgggaagatg-3′). The PCR program was as follows: 35 cycles consisting of 20 s at 94°C, 40 s at 55°C, and 90 s at 72°C. A Taq enzyme activation step of 9 min at 92°C was performed before the first cycle, and a final extension step of 7 min at 72°C ended the program. PCR products were purified on a gel with the QIAquick gel extraction kit (Qiagen Inc.) and were sequenced with the ABI prism system. For each sequence identified as Vκ23, cDNA was also amplified using a Vκ23 primer specific for the framework (FW)1 (Vκ23: 5′-gatattgtgctaactcagtctccagccac-3′) and the same Cκ-specific primer. PCR products were sequenced as described above.

Genomic DNA was extracted from the tails of two MRL-lpr/lpr mice as described above. 1 μl of DNA was used in a PCR containing 50 pmol of a Vκ23 forward primer specific for the FW1 region (Vκ23F: 5′-gatattgtgctaactcagtctccagccac-3′), 50 pmol of Vκ23 reverse primer complementary to the FW3 region (Vκ23R: 5′-gaggccagctgttactctgttg-3′), 1.5 mM MgCl₂, 200 μM dNTPs, 3 μl 10× buffer II, and 1 U AmpliTaq Gold™ (Perkin Elmer). The PCR program was as follows: 92°C for 9 min, followed by 35 cycles of denaturation for 30 s at 94°C, annealing for 30 s at 60°C, and extension for 40 s at 72°C. A final extension step at 72°C was carried out for 7 min. PCR products were size selected on a 1.5% agarose gel (Ultra Pure Agarose; GIBCO BRL) and purified with the QIAquick gel extraction kit (Qiagen Inc.). Purified DNA fragments were cloned into the pGEM^®-T easy vector (Promega Corp.). JM109 competent cells were transformed by electroporation, and transformants were directly tested for the nature of the insert by PCR, using two different sets of Vκ23-specific primers. One set was made up of the Vκ23 primers used in the initial amplification (see above) and another set was made up of Vκ23 primers (Vκ23′F: 5′-gtgactccaggagatagc-3′; Vκ23′R: 5′-gttgataccagtgtaggtt-3′) specific for the new Vκ23 gene (Vκ23GL2) similar to our Vκ23 sequences (see text). Plasmid DNA from positive colonies of each group was sequenced as described above using a plasmid-specific T7 primer. Two Vκ23 germline sequences were identified. One was identical to the DP12 mAb sequence 4 and was not amplified by the second set of primers. The other one differed from the DP12 sequence by two nucleotides located in the FW1 and CDR2 (see Fig. 5) and was amplified with the Vκ23′ primers.

Sequences were analyzed for homology to the original 3H9 and Vκ8 transgenes as well as to published Ig gene segments using EMBL/GenBank databases and Kabat et al. 29. The sequences of 3H9 V_H gene and the Vκ8 gene can be found in EMBL/GenBank/DDBJ databases under the accession numbers M18234 (3H9) and M34742 (Vκ8). The Vκ23 sequences described in this paper are available from EMBL/GenBank/DDBJ under the following accession numbers: AF139842–AF139849. The accession numbers for the V_H sequences presented in Fig. 3 B are AF145959–AF145963.

We tested the hypothesis that mutations accumulated at the same rate and for equal duration at three loci. Under the assumption that all three loci mutate at the same rate, we reasoned that if the three genes had mutated for equal lengths of time, they should have equal numbers of mutations per base. We used the genealogical trees to determine the number of independent mutations (the total sum of the branch lengths in each tree) that occurred at each of the three loci. By this measure, 3H9, Vκ8, and Vκ23 loci accumulated M_3H9 = 34, M_Vκ8 = 33, and M_Vκ23 = 18 mutations (see Fig. 6), respectively, in sequences of length of L_3H9 = 363 bases, L_Vκ8 = 339 bases, and L_Vκ23 = 324 bases (see Fig. 3 A, 4, and 5). We used only mutations in the coding sequences for this analysis.

Assuming the hypothesis is true and an underlying Poisson distribution of mutations, we determined the mean number of mutations per base (λ₀) from the maximum likelihood estimate, in this case, the average:

Under the null hypothesis (H₀: mutation for equal duration at three loci), the number of mutations in each sequence (M_3H9, etc.) would be determined by the same value of λ₀ mutations per base estimated by λ̂₀ above. Our hypothesis amounts to equating the three individual mutation frequencies λ_3H9, λ_Vκ8, and λ_Vκ23 to a common frequency λ₀. Thus, the total number of mutations in the three sequence sets would be governed by Poisson distributions with means M_3H9=L_3H9λ₀, M_vκ8=L_vκ8λ₀, and M_vκ23=L_vκ23λ₀. Put formally: H₀: λ_3H9=λ_vκ8=λ_vκ23=λ₀, i.e., all have the same number of mutations per base, meaning the same duration of mutation. We tested the alternative: H_A: λ_vκ23<λ_3H9=λ_vκ8 using the following statistic:

The next section describes how Z₀ was derived. The alternative hypothesis includes the statement that λ_3H9=λ_vκ8, so we estimated their common value by:

λ̂_vκ23 is estimated by M_vκ23L_vκ23. Under H₀, these two formulas estimate the same quantity. Thus under H₀, the expectation of their difference:

is zero. To obtain a standardized (mean = zero, variance = one) statistic, namely Z₀, under H₀, we divided the difference by the square root of its estimated variance, which we calculated as follows. The variance of a difference is the sum of the terms' separate variances, which is:

The variance of a Poisson deviate is equal to its mean, thus,

and similar formulas hold for Var(M_vκ8) and Var(M_vκ23). Thus, the variance of the difference is:

This quantity is estimated by substituting the above estimates of the λ's for their true values. For these data, Z₀ = 2.063.

We sampled the Poisson distribution with parameters M_i, i = 3H9, Vκ8, Vκ23 to obtain deviated M₃H₉, M_vκ8, and M_vκ23, rigorously under H₀. From these, a new λ̂₀ was obtained as above, and finally, a new value of Z was derived. If the new Z was at least Z₀, the event was tallied. The procedure was repeated 100,000 times. In these trials, Z₀ was equalled or exceeded only 1,780 times, requiring H₀ to be rejected at the 1.78% level.

In nonautoimmune mice, 3H9Vκ8 B cells are anergized. Consequently, 3H9Vκ8 transgenic mice have no detectable DNA binding activity in their serum 12. To assess whether such anergic B cells escape tolerance in autoimmune MRL/lpr mice, sera from 3H9Vκ8/lpr mice were tested for the presence of anti-DNA Abs. Anti-DNA Abs were readily detected by ELISA as well as by indirect immunofluorescence on Crithidia luciliae (data not shown). Furthermore, ANAs were detected in the sera of 3H9Vκ8/lpr as early as 2 mo of age. None of these Ab specificities was detected in sera from 3H9Vκ8/BALB/c mice (Fig. 2 and data not shown).

Hybridomas from 3H9Vκ8/lpr mice have many features that indicate 3H9Vκ8/lpr B cells are activated rather than anergic, as they are in the BALB/c background. Unmanipulated 3H9Vκ8/lpr B cells yield high frequencies of hybridomas, unlike 3H9Vκ8/BALB/c B cells, which require LPS stimulation or immunization for hybridoma formation. In this regard, 3H9Vκ8/lpr B cells are similar to those of diseased, nontransgenic, MRL/lpr mice that generally have high spontaneous fusion efficiencies 30. A high percentage of spontaneous hybridomas from diseased mice produce autoantibodies such as anti-DNA and RF suggesting that self-antigens are driving fusable B cells 31. Hybridoma panels from 3H9Vκ8/lpr mice also have high frequencies of anti-DNA-secreting B cells. For example, in one hybridoma panel derived from a 2-mo-old 3H9Vκ8/lpr mouse, 34 (71%) out of 48 hybridomas secreted anti-DNA Abs (Table). This is not surprising given that the tg Ab binds ssDNA, but of these 34 hybridomas, 9 bound dsDNA in ELISA, and of these, 5 displayed a homogeneous nuclear staining pattern on HEp-2 cells (Fig. 2). Anti-dsDNA and ANAs are rarely found in hybridoma panels from LPS-activated B cells from 3H9Vκ8/BALB/c mice; instead, most of these hybridomas produce anti-ssDNA Abs (reference 13 and Table). Finally, the presence of isotype-switched anti-DNA Abs in the serum of 3H9Vκ8/lpr and the secretion of IgG by 40% of the hybridomas (Table and Table) are other indications of antigen activation. Strikingly, all the dsDNA/nuclear binders are IgG. Together, these results provide strong evidence that these mice are undergoing a (self-)antigen-driven immune response.

Anti-dsDNA/ANAs in 3H9Vκ8/lpr mice could be generated either by somatic mutation or by gene replacement. To distinguish mutation from replacement, we examined the status of the H and L chain sd-tgs. Those hybridomas that retain both the 3H9 H and Vκ8 genes but express unique specificities are likely to be mutated. Those that have lost 3H9 and/or Vκ8 gene(s) would most likely have specificities generated de novo. We tested for the presence of V_H and Vκ transgenes in hybridomas from the 3H9Vκ8/lpr mouse using a series of PCR assays (Fig. 1). Out of 48 hybridomas studied, 38 (79%) were positive for both the 3H9 and Vκ8 genes. Of these double-positive hybridomas, some did not bind DNA, whereas others have acquired specificity for anti-dsDNA (Table), suggesting that the 3H9 and/or Vκ8 transgenes have undergone somatic mutation. Similar results were obtained from a hybridoma panel of a 4-mo-old 3H9Vκ8/lpr mouse (Brard, F., and M. Weigert, unpublished data).

The V_H genes of 13 IgG-secreting hybridomas with changed specificity (anti-dsDNA/ANA⁺ or non-anti-DNA) were amplified and sequenced. Of these, eight were 3H9⁺Vκ8⁺ mutants and five were 3H9⁻Vκ8⁺. Their nucleotide and deduced amino-acid sequences are shown in Fig. 3a and Fig. b, and are summarized in Table. Mutations in the 3H9 sd-tg are located both in the coding and in the noncoding regions. The average number of mutations per V region is high, 12.5, with the most mutated V region having 22 mutations (hybridoma 28) and the least mutated having 6 (hybridoma 16) (Fig. 3 A). In the JHCH intron, the average number of mutations within the 60 bp immediately 3′ of J_H4 is 2.4, with the most mutated sequence having 5 mutations. Of note, hybridoma 46 has a deletion of three codons in CDR2. In the same region, hybridoma 28 has a stretch of five nucleotides that are different from their germline counterparts. We think this most likely results from some form of insertion and deletion. Wilson et al. 32 described similar insertion/deletion mutants in Ig hypervariable loops. In their study, deletions occurred at tandem repeat sequences and inevitably destroyed one of the tandem repeats. Insertions usually involved duplications of the immediately adjacent sequence. In 3H9's CDR2, we have identified such a repeat sequence, AGA GAT GGA GAT, at the site of the insertion/deletion mutations. The deletion in hybridoma 46 removes one full repeat element plus one adjacent codon. The sequence of the insertion in hybridoma 28, GAGTC, is similar to the repeat motif GAGAT. Interestingly, the GAGAT motif contains the putative hot spot sequence RGYW 33, and it is known that deletion/insertion events usually occur in the vicinity of mutational hot spots 32.

Mutations were also found in 10 of the Vκ8 sd-tgs (Fig. 4). Here again, mutations were found throughout the sequence. The average number of mutations per V region was 9.2, ranging from 18 mutations (hybridoma 50) to 3 mutations (hybridoma 16). All together, many of the V_H and V_L genes sequenced from the hybridomas presented in Table have replacement mutations in CDRs. Thus, it is possible that the novel specificities of these antibodies are the result of mutation in the 3H9 and/or Vκ8 genes.

Hybridomas 50, 79, 15, 83, 28, and 38 share eight mutations in their 3H9 sequences, therefore it is likely that they come from an expanded B cell clone. Ordinarily, B cells are defined as members of a clone based not only on the presence of shared mutations but also on sequence identity of V_HCDR3. However, sequence identity of V_HCDR3 is not an applicable criterion by which to judge clonality in transgenic models. Nevertheless, the following observations indicate that these six hybridomas are indeed derived from a single B cell clone: first, these hybridomas all express the same isotype, IgG2a. Second, these hybridomas also share 15 mutations in their Vκ8 and 3H9 sd-tgs, collectively. Third, each of these hybridomas maintains a germline H chain allele and has identical endogenous κ chain rearrangements (see below). Furthermore, it is unlikely that the shared mutations are entirely the result of positive selection and/or hotspot mutation, because only two of these shared mutations are found in the sequences of hybridomas that we know to be unrelated on the basis of rearrangement status (hybridoma 46, Fig. 3 A, and hybridoma 2, Fig. 4); moreover, 7 out of the 15 shared mutations in both H and L transgenes are nonselectable, either because they are silent or because they occur in introns.

The most interesting feature of this clone is that all members share a mutation to the TAG nonsense codon in the Vκ8 sd-tg at the beginning of the CDR1 (Fig. 4). Because these clones secrete an IgG2a/κ Ab, the untargeted κ allele must be rearranged and expressed. To determine the nature of the expressed L chain, Vκ mRNA from these six hybridomas was sequenced (Fig. 5, Table). All express a Vκ23 gene joined to Jκ2′, the MRL counterpart of Jκ2. Analysis of the V–J junction revealed that the first codon of Jκ2′ is deleted and a histidine is found in its place. This unusual junction has previously been observed in other Vκ23Jκ2′ sequences 18,34. The histidine found at the junction may derive from either the end of the germline Vκ23 gene or the nucleotides immediately 3′ of the Vκ23 gene.

To evaluate the number of mutations in the Vκ23 sequences, it was necessary to determine the germline sequence of the Vκ23 gene. Because GenBank did not contain an appropriate sequence, we cloned Vκ23 genes from MRL/lpr tail DNA. Two potential candidates, Vκ23GL1 and Vκ23GL2 (Fig. 5) were identified. Our six hybridoma Vκ23 sequences are most similar to Vκ23GL2. Assuming that Vκ23GL2 is the bona fide germline counterpart, the Vκ23 sequences from our hybridomas had an average of 6.5 mutations per V region, with the most mutated V region having 10 mutations (hybridoma 28) and the least mutated having 5 (hybridomas 38 and 83). These six sequences have three mutations in common, and two other mutations are shared by at least four out of the six clone members.

The shared V_H and V_κ mutations form a hierarchy characteristic of expanded clones. This hierarchy of mutations allowed us to construct a genealogy of the clone members (Fig. 6). The 3H9 and Vκ8 trees have roughly the same height and shape. This similarity illustrates that the onset and rate of mutation are the same in H and L chain genes. Although the Vκ23 tree has a “branch morphology” similar to that of the 3H9 and Vκ8 trees, it appears to have a shorter “trunk.” Furthermore, the average number of mutations is less in Vκ23 than in either 3H9 or Vκ8 (6.5 compared with 14 and 13, respectively). These results indicate that the Vκ23 L chain started to mutate later than the Vκ8 or 3H9 transgenes, but that once initiated, Vκ23 mutation proceeded at a similar rate. Taken together, these mutation patterns and tree shapes imply that the Vκ23 gene was rearranged and expressed after the onset of mutation in the 3H9 and Vκ8 transgenes.

To confirm this conclusion, a statistical analysis testing the hypothesis that mutations accumulated for equal durations at the three loci was performed. Implicit in this calculation is the assumption, based on the absence of clear evidence to the contrary, that the mutation rate was equal at each of these three loci. We also assumed that the number of mutations found in each tree would be Poisson distributed. We estimated mutation frequencies per base from the number of mutations divided by the number of bases in the corresponding sequence set. The frequency estimates for our 3H9 and Vκ8 trees did not differ significantly, and in fact were very close to each other. We combined the 3H9 and Vκ8 data and compared their combined frequency estimate to that obtained for the Vκ23 sequences. To test the significance of this difference, we combined all three data sets for an overall estimate of mutation frequency. We then simulated mutations in the three trees by generating Poisson deviates with means adjusted for each gene's particular sequence length. We tallied the number of times the difference in mutation frequencies observed this way, (i.e., rigorously, under the null hypothesis of identical mutation durations for each sequence), was as large or larger than their actual observed difference. Their observed difference was equalled or exceeded in only 1.78% of 100,000 trials. This allowed us to reject the hypothesis that all three Ig loci mutated for equal duration (p = 0.0178). Therefore, we conclude that the Vκ23 L chain gene was rearranged in the periphery and then started to undergo somatic mutation.

All members of this clone have acquired dsDNA specificity, and five out of six are ANA⁺ (Fig. 2). The acquisition of dsDNA and antinuclear activity could be due to mutation in either 3H9 or Vκ23. In particular, the mutations to asparagine in FW3 and CDR3 in Vκ23 might create anti-DNA specificity 35. Alternatively, the Vκ23 L chain itself may create the ANA specificity (Table), since ANAs and antihistone antibodies derived from autoimmune mice are often associated with Vκ23 L chains 18,19,34,36,37,38.

Earlier studies examined the regulation of the 3H9 H chain tg in lpr/lpr mice 18,19. The 3H9 H chain transgenic without an accompanying L chain transgene is an excellent model for studying tolerance and loss thereof: the V_H gene coding for 3H9 is the most popular V_H among disease-associated anti-DNAs, constituting ∼20% of the >200 sequenced anti-DNAs from MRL/lpr, (NZB × NZW)F₁ and (NZB × SWR)F₁. An unusual but useful feature of 3H9 (and other anti-DNA H chains) is that association with different L chains modifies DNA binding. 3H9/L chain combinations fall into three broad classes: combinations that sustain ss/dsDNA like that of the original 3H9/Vκ4 combination, combinations that modify DNA binding as in the 3H9/Vκ8 combination that only binds ssDNA, and combinations that veto DNA binding as, for example, 3H9/Vκ12/13. A study of 3H9 H chain in combination with a broad variety of endogenous L chains shows that the proportion of the three classes is ∼30% ds/ss : 60% ss : 10% non-DNA binding 39. Moreover, certain L chains in combination with 3H9 yield unique ANA patterns as illustrated in Fig. 2 for the 3H9/Vκ23 combination 21. Thus, 3H9 sd-tg mice yield a spectrum of antibodies, making the 3H9 sd-tg MRL/lpr a multipurpose system for studying the breakdown of tolerance. As might have been predicted, 3H9 tg lpr/lpr mice express the types of anti-DNAs that are ordinarily edited or inactivated. These include anti-ds/ssDNA and antibodies that are ANA⁺. But a limitation of this model is that the precursor to the pathogenic anti-DNAs expressed in this lpr/lpr tg cannot be established; hence, the site(s) at which tolerance is broken is unknown. The advantage of the 3H9Vκ8/lpr mice described here is that the anti-DNA repertoire is limited to one specificity, anti-ssDNA, that in normal mice is known to be regulated by anergy 12,14. That anti-dsDNAs are now expressed in the 3H9Vκ8/lpr means either that anergic cells become activated or that anergy cannot be established in lpr/lpr mice. Based on the evidence for Fas-mediated regulation of peripheral B cells 40,41, we favor the latter interpretation.

The failure to establish or maintain anergy of anti-ssDNA leaves the lpr/lpr mouse with a population of B cells poised for the transition to autoimmune disease. Both in vivo and in vitro studies have shown that anti-DNAs such as 3H9Vκ8 can be substrates for mutation to dsDNA and nuclear antigen binding antibodies 39,42. Here we show that this transition takes place during clonal expansion of 3H9Vκ8 B cells and results in the production of anti-dsDNA and ANAs typical of disease. In this regard, our results recapitulate the nature of MRL/lpr and other autoimmune mice. But, during clonal expansion 3H9Vκ8/lpr B cells are actually subjected to two forms of somatic diversification: first, expanded hybridomas have accumulated a high frequency of mutations. These include mutations to R and N that could account for the shift of 3H9Vκ8 specificity from ssDNA to ss/dsDNA and the acquisition of antinuclear specificity (ANA⁺, Fig. 2). The second form of diversification is L chain editing. This event is an indirect consequence of mutation: all members of the clone share a nonsense mutation in the Vκ8-Jκ5 gene, therefore IgGκ secretion requires L chain expression from the untargeted allele. This novel mechanism rescues a defunct B cell. Alternative explanations for this genotype, such as coexpression of both κ genes throughout the lifetime of the clone, are highly unlikely (p ≤ 0.0178, see Materials and Methods).

V gene rearrangement during clonal expansion is not surprising in view of the evidence for RAG expression in germinal centers 43,44,45. But the relevance of RAG-mediated recombination to the immune response at this stage of B cell development has not been established. RAG-induced DNA nicks 45 may just be another manifestation of programmed cell death. It is unlikely that secondary rearrangement would enhance an ongoing immune response, because changing either H or L usually will produce a new specificity 29. Moreover, these new specificities will rarely, if ever, be propagated because of the lack of antigen and T cell help. Instead, RAG expression in germinal centers simply may be a gratuitous part of impending or actual cell death. This is suggested by the work of Hikida et al., who have shown that most of the B cells in germinal centers that express RAG are undergoing apoptosis 46. In fact, apoptosis may have been the raison d'être for RAG to begin with. The earliest B and/or T cell receptor genes were probably intact V genes, not segmented V, D, and J genes. These genes may have included sequences homologous to recombination signals, and primitive vertebrates may have fixed genes coding for proteins that cut DNA at these sites. Such cuts may have served to inactivate V genes or even to kill cells, thereby enforcing allelic exclusion and/or helping to maintain steady state levels of lymphocytes.

For a defunct B cell to be rescued by secondary rearrangement requires truly exceptional circumstances, but anti-DNAs such as 3H9/Vκ8 provide the conditions for rescue. First, L chain replacement does not always destroy DNA binding, because the H chain makes most of the antibody contacts to DNA 42. Hence, specificity for DNA can be maintained or modified even though Vκ8 is replaced (i.e., a kind of impotent receptor editing). Second, the self-specificity of 3H9/Vκ8 (and the L chain alternatives that, with 3H9H chain, sustain DNA binding) ensures that these antibodies will always be exposed to a cognate antigen. That germinal centers are sites of apoptosis is relevant to this point because the germinal centers are considered possible sites where the target lupus autoantigens may be presented. Third, DNA can be thought of as a super-antigen, in the sense that it is associated with a wide variety of both self- and foreign proteins. Thereby, as long as a B cell's receptor binds DNA, that B cell can present a variety of T cell epitopes derived from these DNA-associated proteins. Take, for example, a receptor directed to the complex of DNA–histone. Even though L chain replacement converts the receptor to one that only binds DNA, the receptor will still bind chromatin and ultimately that B cell can present histone peptides. Thus, this kind of autoantibody will rarely be at a loss for T cell help. 3H9/Vκ8 descendents illustrate this strategy: the 3H9/Vκ23 receptor has acquired specificity for dsDNA yet still binds ssDNA, making it so that the revised B cell can still present the same T cell epitope(s) as the parental 3H9/Vκ8 B cell.

These studies show that expression of anti-DNA in Fas-deficient mice is influenced in at least two ways: first, B cells that are anergic in normal mice become activated; second, anti-DNA antibodies can arise in defunct B cells by V gene replacement. Both processes lead to the development of a spectrum of autoantibodies. Activation of anti-ssDNA B cells leads to clonal expansion and mutation and some of these mutations lead to specificity for dsDNA. Thus, anergy is a major tolerance checkpoint in that it prevents low affinity anti-DNAs from mutating to pathogenic types. Secondary V gene rearrangement in B cells generates a new spectrum of specificities that will surely include autospecificities. These might be formed by novel VH/VL combinations as in antibodies associated with Vκ23.

The fact that secondary rearrangements, or editing, create autoreactivity in the autoimmune MRL/lpr is paradoxical, because in normal B cell development each round of editing adds to the number of non-autoreactive, mature B cells an animal can generate from a given set of precursor cells. How can one account for this difference? The answer must lie in a cell's “motivation” for continuing to rearrange. An immature B cell from a normal individual wants to make a functional receptor that does not bind too strongly to surrounding (self-)antigens. A mature cell from an autoimmune individual, on the other hand, is attempting to secure positive selection signals by rearranging until it has a receptor that binds an antigen with high avidity. This notion stems from the work of Hertz et al., who have shown that mature B cells experiencing low affinity interactions with antigen tend to initiate recombination, whereas high affinity antigen interactions abolish recombinase activity in mature B cells 47. They also point out that this type of peripheral regulation of receptor editing would tend to promote autoreactivity. What is not thoroughly considered by their work is the role of T cell help. And this is what may be special in the case of autoreactivity, especially of autoreactivity to DNA. Rescue of a cell that has become autoreactive through editing rearrangement in the periphery also requires T cell help (specific for autoantigens) that is extant in autoimmune mice.

Genes that affect the death of autoreactive cells and the disposal of defunct B cells (such as the fas gene) have a broad influence on the regulation of autoimmunity. In addition to extending the survival (and subsequent activation) of anergic B cells, they may also permit the development of an inappropriate repertoire by secondary rearrangement. The ability to detect peripheral editing in MRL mice is probably greatly enhanced by the lpr mutation that extends the life of B cells which otherwise would die before having had enough time to make a new receptor. Moreover, genes involved in cell death may regulate the concentration or availability of self-antigen. Casciola Rosen et al. have shown that the self-antigens targeted in lupus are found on the surface of apoptotic cells 6. This led them to suggest that failure to kill cells efficiently could influence self-tolerance by limiting the amount of self-antigen. This limitation may be particularly strict in the bone marrow microenvironment where the preimmune repertoire is developed. A shortage of tolerogen could explain why the spectrum of autoantibodies in systemic autoimmunity is biased toward molecules released during cell death and could also resolve the differences between regulation of antibodies directed to facultative and constitutive self-antigens.

The authors thank Sandra Jainandunsing for her excellent technical assistance, Joseph Goodhouse for help with the microscopy, and Daniel Lee for help with 2. We are grateful to all members of the lab for helpful comments on the manuscript. We also acknowledge the CORE Services of Princeton University for oligonucleotide synthesis, DNA sequencing, and animal colony care.

F. Brard was supported in part by the Institut National de la Sante et de la Recherche Medicale and in part by the Fondation pour la Recherche Medicale and the Conseil regional de Haute-Normandie. M. Shannon is supported by the American Heart Association (9704047A) and M. Weigert is the recipient of National Institutes of Health grant (GM20964-24).