Co-option of Retrotransposons Promotes Antibody Diversification

Discovery of LINE-1 Retrotransposons in the Immunoglobulin Locus

We discovered the existence of LINE-1 (L1) retrotransposons found upstream to many human V_H genes (Figure S1A) and observed several patterns: i) many V_H genes have more than one nearby upstream L1, with some containing up to five tandem L1s; ii) of the ~50 human V_H genes, ~66% possess at least one L1 within a 2.5 kb upstream window from the V_H transcription start site (TSS); iii) the first V_H upstream L1s are often found 0.6–1.0 kb away from their nearby V_H TSS; iv) the first V_H upstream L1 is often in the same orientation as their nearby V_H gene, with subsequent L1s sometimes in the opposite orientation (Figure S1A-B); v) all V_H upstream L1s belong to evolutionarily “old” L1 subfamilies, incapable of autonomous retrotransposition.^{44– 48} These observations were similar for mouse V_H genes (Figure S1C-D). Looking upstream to all V_H TSSs, we observed a significant enrichment of L1s, relative to all protein-coding genes and known AID off-target genes implicated in aberrant SHM during lymphomagenesis^21,54, with similar findings for mouse V_H upstream L1s (Figure 1A-B; Table S1-S3).

We observed no human V_H upstream enrichment of other transposable elements (TEs) including SINEs and DNA transposons, except for a farther upstream density of LTRs (Figure S1E-G), and L1 enrichment downstream to all V_H genes (Figure S1H), which would be permanently deleted during V_H(D)J_H recombination. We observed similar patterns for mouse V_H genes, except for the lack of V_H upstream LTR enrichment (Figure S1I-L, Table S3).

V_H Upstream L1s Promote B Cell Somatic Hypermutation

The discovery of these V_H upstream L1s surprised us. Given their proximity to V_H gene, we wondered if they influenced B cell function and decided to investigate V_H(D)J_H SHM.

To test this, we used a human-derived B cell line, Ramos, constitutively expressing mouse AID fused to the ligand-binding domain of estrogen receptor (AID-ERT2). This reporter line contains an mCherry sequence, with several AID hotspot motifs, fused in frame to the 5’ end of exon 2 of the endogenous, recombined V_H4–34.^55–58 Upon 4-hydroxytamoxifen (4-OHT) treatment, AID-ERT2 translocates into the nucleus and initiates hypermutation, where loss of mCherry expression via AID mutagenesis approximates SHM, validated by more accurate targeted V_H(D)J_H sequencing. Because V_H4–34 is one of the few V_H genes lacking a nearby upstream L1, we inserted an L1 element at ~600 bp upstream to the recombined V_H4–34 (Figure 1C); this L1 element is the same one found in front of many other human V_H genes (Table S4). Reporter cells with the nearby upstream L1 (“yesL1”) showed increased SHM after 7 days of 4-OHT treatment compared to cells lacking an upstream L1 (“noL1”) (Figure 1D-E). Targeted Nanopore and next-generation sequencing of recombined V_H(D)J_H sequences from sorted mCherry^low cells revealed increased mutation frequency across the entire V_H(D)J_H, particularly at AID hotspots (i.e. RGYW/WRCY) (Figure 1F-G, Figure S2A-B). YesL1 cells exhibited no defects in proliferation, surface IgM, cell viability, or AID protein expression (Figure S2C-G).

To better understand these observations, we cultured noL1 and yesL1 cells together in 4-OHT for weeks and observed mCherry^high and mCherry^medium populations, which comprised predominantly noL1 cells and yesL1 cells, respectively (Figure S2H-I), further supporting that V_H upstream L1s increase SHM.

We wanted to determine whether AID mutated the V_H4–34 upstream L1 element itself. After inducing SHM in our reporter cell lines, we PCR amplified: the L1-lacking region upstream to the recombined V_H4–34 in noL1 cells, the L1-lacking region upstream to the non-recombined V_H4–34 in yesL1 cells, and the L1-containing region upstream to the recombined V_H4–34 in yesL1 cells (Figure 1H). Surprisingly, we detected SHM covering the upstream L1 region in yesL1 cells, with correspondingly sparse SHM in noL1 cells (Figure 1H-I). When comparing both V_H4–34 alleles within the same yesL1 cells, we observed essentially no SHM on the L1-lacking, non-recombined allele, but detectable SHM on the L1-containing, recombined allele (Figure 1J). Finally, mouse spleen germinal center B cells showed hypermutation of endogenous V_H upstream L1-rich regions, which was absent in matched tail DNA from the same mice (Figure S2J-K). These results suggest that AID targets V_H upstream regions (with L1s) in addition to recombined V_H genes themselves.

We performed AID ChIP-qPCR and found greater AID occupancy at the recombined V_H(D)J_H in yesL1 cells compared to noL1 cells (Figure 1K), suggesting that nearby upstream L1s could somehow attract AID to the locus. Introducing a different V_H upstream L1 (“yesL1#2”) similarly increased AID occupancy, suggesting that the phenotype was not dependent on a specific L1 sequence (Table S4).

V_H Upstream L1s Drive V_H(D)J_H Somatic Hypermutation In Vivo

To see if we could observe L1-driven SHM in vivo, we generated four mouse lines, each containing one designated V_H gene with its upstream L1s homozygously deleted (Figure 2A/D/G/J; Table S5). The four V_H genes we chose were V_H1–11, V_H1–64, V_H1–47, and V_H9–4, as they were found to be used or hypermutated more in mouse intestinal Peyer’s patch (PP) germinal center (GC) B cells.⁵⁹ Spleen GC B cells with recombined V_H1–11(D)J_H genes, isolated from immunized V_H1–11 L1 KO mice, showed marked reductions in V_H1–11(D)J_H SHM, often within or near AID hotspot motifs (Figure 2B-C). As a control, V_H1–11(D)J_H genes from immunized AID KO mouse GC B cells showed no SHM (Figure 2B-C). We also observed decreased SHM in V_H1–64(D)J_H, V_H1–47(D)J_H, and V_H9–4(D)J_H sequences from our V_H1–64 L1 KO mice, V_H1–47 L1 KO mice, and V_H9–4 L1 KO mice, respectively (Figure 2D-L). Finally, we observed decreased SHM in intestinal PP GC B cells from all of our L1 KO mouse lines (Figure S3A-D), consistent with the results seen in spleen GC B cells.

While overall mutation frequency decreased across the tested V_H(D)J_H sequences from our L1 KO mice, no particular regions showed repeatedly decreased SHM peaks, consistent with the hypothesis that these nearby upstream L1s are involved in the initial recruitment of AID to the IGH locus, rather than informing specifically where AID mutagenesis occurs within the recombined V_H(D)J_H.

We next wanted to determine in vivo whether V_H upstream L1s acted in cis on their nearby V_H gene or in trans on other V_H genes. Using genomic DNA of spleen GC B cells from immunized WT and V_H1–11 L1 KO mice, we examined SHM levels of the flanking V_H genes, V_H1–12 and V_H1–9, corresponding to GC B cells specifically expressing recombined V_H1–12(D)J_H and V_H1–9(D)J_H (Figure 2M). Indeed, we observed similar levels of flanking V_H gene SHM between WT and V_H1–11 L1 KO mice. Likewise, we observed the same phenomenon in our other three mouse lines and their respective flanking V_H genes (Figure 2N-P). Thus, deletion of V_H upstream L1s only affects SHM activity of the associated downstream recombined V_H gene, without affecting SHM activity of other V_H genes.

We found no differences in CSR in vivo from immunized WT vs L1 KO mice or CSR ex vivo from WT vs L1 KO mice (Figure S3E-J). Because SHM is inherently linked to transcription of the recombined V_H(D)J_H^60,61, we verified that the presence or absence of upstream L1s had no effect on recombined V_H(D)J_H steady-state transcription in ex vivo-stimulated spleen B cells (Figure S3K).

Lastly, we wanted to rule out the possibility of nearby upstream L1s influencing V_H gene selection, which occurs early in B cell development, long before the occurrence of SHM in secondary lymphoid organs. Next-generation sequencing of naïve spleen B cells from WT and L1 KO mice showed no differences in V_H family usage, specific V_H gene usage, or specific V_H gene expression (Figure S3L-N). Furthermore, homozygous deletion of V_H upstream L1s did not affect the usage or expression of flanking V_H genes (Figure S3O-P). For instance, V_H1–11 L1 KO mice showed no skewed usage or expression of upstream V_H1–12, downstream V_H1–9, or V_H1–11 itself, compared to WT mice. We conclude that AID-mediated L1-driven SHM occurs in vivo and acts in cis on the recombined V_H(D)J_H.

HUSH Complex Links V_H Upstream L1s with AID-Mediated SHM

We sought to understand the relationship between AID and L1 biology. AID belongs to the APOBEC/AID superfamily of enzymes, sharing a common catalytic domain that deaminates C to U on DNA and/or RNA.^62,63 Some APOBEC members recognize and mutate nucleotides in viral and retrotransposon sequences. Another defense mechanism involves the human silencing hub (HUSH) complex, comprising the scaffold protein TASOR and its binding partners Periphilin (PPHLN1) and MPP8.^64–66 L1 transcription triggers the recruitment of HUSH to that specific site, where HUSH will recruit chromatin modifier SETDB1 to deposit H3K9me3, preventing further transcription of the L1.^67–70 Importantly, HUSH silences non-self genetic elements that generate intronless RNAs, primarily L1s, retrogenes, and retroviral DNA sequences, and therefore serves as an important component of the genome’s innate immune system.^67,71,72

To assess whether HUSH could explain the missing link between AID and L1-driven SHM, we generated multiple versions of our noL1 and yesL1 reporter cells, lacking or possessing each HUSH subunit (Figure 3A). In unedited cells, we observed L1-driven SHM. However, loss of any HUSH subunit – TASOR, MPP8, PPHLN1 – blunted the effect of L1 on SHM. Importantly, HUSH WT and KO noL1 cells showed identical levels of SHM (Figure 3A, Data S1). These results suggest that the HUSH complex plays a role in L1-driven, AID-mediated SHM in B cells.

We inserted various sequences upstream to the recombined V_H4–34 gene in our noL1 reporter cell lines to see if there was an effect on SHM comparable to that of L1. Insertion of the ~1.7 kb human beta-globin gene HBB (containing two exons and one intron), in either orientation, had no effect on SHM (Figure 3B). Insertion of other transposon sequences of varying orientations and lengths also had no effect on SHM (Figure 3C), consistent with HUSH’s preference to silence L1-derived sequences.⁶⁷

Given HUSH’s ability to target transcribed L1s, we wanted to see if the V_H nearby upstream L1 could recruit HUSH. We performed CUT&RUN with our reporter cells, with yesL1 cells exhibiting greater AID occupancy, greater HUSH occupancy via MPP8 and PPHLN1, and greater H3K9me3 deposition, a mark of HUSH activity, over the recombined V_H(D)J_H and upstream L1 itself (Figure 3D). Interestingly, even in noL1 cells, HUSH appears to be present at low levels along the recombined V_H(D)J_H, including over the short ~80 bp intron (Figure 3D). This suggests that while the presence of V_H upstream L1s greatly increases HUSH recruitment, perhaps the V_H(D)J_H itself has properties of a HUSH substrate. Thus, the summation of V_H upstream L1s and recombined V_H(D)J_H can be viewed as a long 3–4 kb length of DNA that potently recruits HUSH and AID.

To determine whether HUSH influences L1-driven SHM in vivo, we utilized two different mouse models lacking HUSH subunit MPP8: a B cell-specific conditional KO (CD19^Cre/WT;MPP8^f/f) and a constitutive KO (MPP8 Δexon2) lacking physiological levels of MPP8 protein expression due to homozygous deletion of Mphosph8 (MPP8) exon 2 (Figure S4A-C).⁷³ First, we ruled out any effect on CSR ex vivo between WT and MPP8 Δexon2 mice (Figure S4D-F). Corroborating this, two recent CRISPR screens for factors affecting B cell CSR did not identify HUSH subunit genes.^74,75 We also verified there were no differences in spleen or PP GC B cell frequency between WT and MPP8 KO mice (Figure S4G).

We immunized WT, CD19^Cre/WT;MPP8^f/f, and MPP8 Δexon2 mice and measured SHM frequency for the V_H genes studied in our V_H upstream L1 KO mouse models – V_H1–11, V_H1–64, V_H1–47, V_H9–4 (Figure 2). GC B cells from immunized CD19^Cre/WT;MPP8^f/f and MPP8 Δexon2 KO mice showed marked reductions in SHM across recombined V_H1–11(D)J_H, V_H1–64(D)J_H, V_H1–47(D)J_H, and V_H9–4(D)J_H sequences (Figure 3E-H, S4H-K). Overall mutation frequency decreased when considering either all bases within the V_H(D)J_H or only those specifically within AID hotspot motifs (Figure 3I-L, S4M-P). Finally, we observed decreased SHM in recombined V_H1–64(D)J_H sequences from MPP8 Δexon2 KO PP GC B cells (Figure S4L). The SHM defect in both constitutive and B cell-conditional KOs confirms a B cell-intrinsic role for HUSH. Altogether, these data highlight an essential role for HUSH in L1-driven SHM.

HUSH and AID Functional Interaction

We performed proximity ligation assay (PLA) in our yesL1 cells and found that AID interacted with all three HUSH subunits (Figure 4A-C). We also detected AID:MPP8 interactions in mouse spleen GC B cells from immunized WT, but not AID KO control, mice (Figure 4D). Co-immunoprecipitation assays from HEK293T cells showed interactions between AID and both MPP8 and PPHLN1 (Figure 4E-F). CUT&RUN comparing PPHLN1 WT yesL1 and PPHLN1 KO yesL1 cells showed significantly decreased AID occupancy over the recombined V_H(D)J_H in absence of HUSH (i.e. PPHLN1 KO), with concomitant decreases in MPP8, PPHLN1 (as expected), and H3K9me3 occupancy (Figure 4G-H). In HUSH KO samples, we noted residual, maintenance levels of H3K9me3, a heterochromatic IgH signature of B cells.⁷⁶ Overall, AID occupancy at the recombined V_H(D)J_H depends, in part, on its interaction with HUSH.

We then identified the top MPP8-occupied and PPHLN1-occupied loci by subtracting HUSH WT and HUSH KO datasets, the intersection of which revealed 77 HUSH complex-occupied loci, including V_H4–34 (Figure 4I; Table S2). Of the 52 AID off-target loci, none overlapped with HUSH-occupied loci, suggesting that the interaction between HUSH and AID occurs preferentially at the recombined V_H(D)J_H rather than genome-wide (Figure 4I). Global MPP8 occupancy was much higher across the 77 HUSH-occupied loci than across AID off-target loci, and disappeared in the HUSH KO condition, with no significant occupancy changes over AID off-target loci (Figure 4J). Promoters and first introns of AID off-target genes, where pathological SHM is occasionally detected^20,21, showed very low levels of AID occupancy, with no difference between HUSH WT and HUSH KO conditions, again suggesting no role for HUSH in AID targeting genome-wide (Figure 4K). Finally, AID occupancy at all young, full-length L1s genome-wide and at old, truncated L1s upstream to AID off-target genes was negligible compared to AID occupancy at the recombined V_H4–34 upstream L1 (Figure 4L). We conclude that AID forms a physical nexus with the HUSH complex, largely at the recombined V_H(D)J_H, and that global HUSH occupancy at other L1s and protein-coding genes does not guarantee corresponding AID recruitment.

V_H Antisense Transcription Encoding Upstream L1s Triggers SHM Recruitment to V_H(D)J_H

We sought to understand the mechanism by which V_H upstream L1s recruited HUSH and AID. Because transcription of intronless L1s triggers HUSH recruitment, we asked if these V_H upstream old L1s could initiate transcription themselves.⁶⁷ We cloned several human V_H upstream L1s, in both orientations, into promoterless IRES-EGFP reporter constructs (Figure 5A). For some constructs, we deleted the 3’ ends of L1s to avoid potential transcription termination signals. As expected, these old L1s produced no EGFP⁺ cells, even after treatment with 5-AZA-dC, which promotes DNA demethylation and de-repression of L1s.^77–80 However, young, full-length, human L1 produced GFP⁺ cells, independent of orientation, indicating the ability to recruit RNAPII and initiate bidirectional transcription; the young L1 was also sensitive to 5-AZA-dC treatment (Figure 5A).

We wondered if transcription factor (TF) binding sites near V_H upstream L1s could act as potential sources of transcription. We scanned the upstream 2 kb regions of all human V_H genes and AID off-target genes for TF binding motifs and found that V_H upstream regions are thoroughly depleted of TF binding motifs, when considering just the antisense DNA strand (with respect to V_H sense transcription) or both DNA strands (Figure S5A-B).

We reconsidered other potential sources of L1 transcription and wondered if V_H promoters themselves transcribed bidirectionally. Prior in vitro work focusing on single V_H genes seemed to support this idea.^81,82 Indeed, our noL1 and yesL1 reporter cells expressed V_H4–34 sense RNA, but also V_H4–34 antisense RNA, which appeared to cover the upstream L1, with similar levels between HUSH subunit MPP8 WT and KO conditions (Figure 5B-C). As controls, we inserted the same L1 sequence upstream to V_H4–34 in HEK293T cells and SUDHL10 cells, another B lymphoma line, and detected no V_H4–34 sense or antisense RNA from these cell lines; this was expected, given that these cells are not known to express recombined V_H4–34, and explains the large fold change magnitude observed (Figure 5B-C). GRO-seq from Ramos cells further confirmed the existence of nascent antisense transcription originating from the recombined V_H gene.⁸³ The nascent V_H antisense RNA is long enough to cover nearby upstream L1s and is stronger than all other antisense RNAs from AID off-target genes (Figure 5D). Bulk RNA-seq from WT mouse spleen GC B cells demonstrated that recombined V_H promoters are bidirectional and generate antisense RNAs long enough to cover V_H upstream L1s (Figure 5E).

We hypothesized that the long, strong, recombined V_H antisense L1-encoding RNA initiates HUSH/AID-mediated SHM activity. We deleted the recombined V_H4–34 promoter in noL1 and yesL1 reporter cells in an attempt to completely kill antisense transcription while preserving the V_H(D)J_H sense transcription essential for AID accessibility and mutagenesis^4,5,8 (Figure 5F). We generated several V_H4–34 promoter-dead clones and compared their levels of sense and antisense transcription. Unedited noL1 and yesL1 cells exhibited normal, baseline levels of bidirectional transcription, while some noL1 and yesL1 clones exhibited nearly dead sense transcription with dead antisense transcription. We also retrieved rare noL1 and yesL1 clones with near normal levels of sense transcription and dead antisense transcription (Figure 5G). After 4-OHT treatment, we observed the usual L1-driven SHM between unedited noL1 and yesL1 cells (Figure 5H). The next two cell lines showed a severe, relative loss of SHM, likely due to the near absence of V_H(D)J_H sense transcription. The last two cell lines showed levels of SHM similar to unedited noL1 cells, despite the presence of a nearby upstream L1 (Figure 5H, last bar). These results implicate the L1-encoding, V_H antisense RNA as the trigger for L1-driven SHM at the recombined V_H(D)J_H.

L1-Encoding V_H Antisense Transcription in CSR vs. SHM

Puzzlingly, B cells expressing AID do not seem to undergo SHM and CSR simultaneously.^84–88 We therefore examined whether there was a difference in the strength of V_H antisense transcription relative to V_H sense transcription in B cells undergoing SHM vs. CSR. RNA-seq from WT mouse spleen GC B cells exhibited VH promoter bidirectional transcription (Figure 5E), as did spleen B cells stimulated ex vivo for IgG1 CSR.^26,89 Moreover, we observed stronger relative V_H antisense transcription in GC B cells undergoing SHM than in CSR-activated B cells (Figure S5C-D). Within GC B cells, antisense transcription is stronger at V_H promoters than at IgM or IgG1 promoters (Figure S5E-F). Thus, weak V_H antisense transcription in B cells undergoing CSR could explain why these B cells do not appear to simultaneously hypermutate their V_H genes, despite AID expression.

Insertion of V_H Upstream L1s Increases Off-Target SHM

We wanted to see if strong transcription of L1s was sufficient to promote SHM at non-V_H(D)J_H off-target sites. We chose three candidate loci – hBCL6, hMYC, and hMKL1 – to insert ~1.7 kb of tandem old L1s upstream to these genic promoters in our noL1 reporter line (Table S4); in this experiment, mCherry signal is irrelevant to off-target hypermutation. We targeted BCL6 and MYC because they are well-characterized AID off-target genes^{11,14–17,20,21} and MKL1 because it was reported to be located within a “cold” TAD, resistant to AID-mediated SHM in Ramos cells.⁹⁰ Our insert included the core portion of human EF1α promoter alone (control cassette) or followed by three tandem old V_H upstream L1s in both transcriptional orientations (Figure 6A). We included a small human promoter for several reasons. First, we had established that these L1s could not initiate transcription (Figure 5A) and that transcription of these L1s was essential in recruiting SHM activity. Second, in physiological B cell V_H(D)J_H SHM, several IGH-specific super-enhancers greatly increase recombined V_H promoter activity in both directions (Figure 5B-C). Therefore, a source of off-target L1 transcription, albeit much weaker, in the form of the core EF1α promoter, attempts to replicate this phenomenon, given the lack of comparable super-enhancer activity at these off-target regions. Third, we inserted this EF1α-old3xL1 cassette upstream to each off-target gene, such that EF1α promoter sense transcription and off-target sense transcription would travel in opposite directions (Figure 6A). This setup prevents EF1α-driven transcription from “spilling” into off-target gene bodies. Finally, we wanted to insert the cassette at a distance upstream to off-target promoters similar to that between V_H upstream L1s and V_H promoters.

In generating these cell lines, we produced many heterozygous clones, some containing the EF1α-old3xL1 cassette upstream to BCL6 on one allele and wildtype on the other allele, and some containing the EF1α cassette upstream to BCL6 on one allele and wildtype on the other allele (Figure 6A). Insertion of EF1α-old3xL1 significantly increased overall mutation frequency within the first 1.8 kb of BCL6 compared to EF1α alone (Figure 6B). Using analogous heterozygous cell lines, we observed similar findings for MYC and MKL1, where nearby upstream transcription of old L1s promoted off-target hypermutation (Figure 6C-F). Of note, SHM at off-target sites can be 10–1000 folds lower relative to V_H(D)J_H SHM, making detection of transient AID occupancy challenging.^11–13 Off-target SHM, which requires AID activity, is a permanent and more reliable readout of AID recruitment and is what we measured at these three off-target loci.

Old and Young V_H Upstream L1s Increase V_H(D)J_H SHM

We inserted the old3xL1 cassette from Figure 6A-F, without the core EF1α promoter, upstream to the recombined V_H4–34 in our reporter cells. We also generated cell lines where we inserted the L1 from Figure 1C in the flipped orientation and a 1.5 kb fragment of human L1 ORF2 in the opposite orientation relative to V_H(D)J_H sense transcription (Figure 6H). L1 ORF2 and ORF2 fragments were previously shown to act as bona fide HUSH targets.⁶⁷ In all cell lines, expression of the inserted sequence relies on strong, endogenous antisense transcription from the recombined V_H4–34 promoter. All L1 knock-in cell lines exhibited higher SHM via mCherry loss compared to noL1 cells, confirmed by more accurate targeted V_H(D)J_H amplicon sequencing from sorted mCherry^low cells (Figure 6G-K). We conclude that strong expression of both old and young intronless L1(s) is capable of enhancing SHM.

Upstream L1-V_H Gene Pairs are Evolutionary Units Across Species

To see if the L1-driven SHM phenotype was observable in humans, we re-analyzed germinal center BCR sequencing data from patients who received COVID-19 vaccine boosters.⁹¹ We grouped V_H genes by the proximity of their closest upstream L1, with one group representing V_H genes with nearby upstream L1s and the other group representing V_H genes without nearby upstream L1s. For three of six donors, we found significantly higher levels of SHM in B cells that used recombined V_H genes with nearby upstream L1s compared to those that used V_H genes without nearby upstream L1s (Figure S6A). Commonly hypermutated V_H genes belonged to IGHV3 (e.g. V_H3–21, V_H3–48) and IGHV1 (e.g. V_H1–2, V_H1–69) families, while V_H4–34 – lacking upstream L1s – was consistently lowly mutated. Notably, our reporter cells use recombined V_H4–34, and introducing L1s upstream to V_H4–34 enhanced its SHM (Figure 1D-G). In doing so, we converted a mutationally “cold” V_H gene to a “hot” one, suggesting that V_H gene hypermutation potential is partly encoded within the V_H upstream region, and not entirely within the V_H sequence itself.

Beyond humans and mice, we wanted to see if V_H upstream L1s existed in ~20 other species with identifiable L1s and V_H genes (Figure 7A). Surprisingly, we observed significant enrichment of V_H nearby upstream L1s for all species (Figure 7B). Chickens were a unique case, explained further in the Discussion. We focused on this L1-V_H gene relationship in closely related primate species by categorizing all V_H genes into their respective V_H families. We found that in humans, L1MCb or L1PREC2 sequences were often found upstream to V_H1 family genes, while L1ME1 or L1ME2 sequences were often found upstream to V_H3 family genes. This pattern held true for all primate species (Figure 7C). Moreover, the distance between V_H genes from the same family and upstream L1 was consistent across species. For example, human and primate V_H3 family genes possess upstream L1ME1 sequences located ~600 bp away (Figure S6B). We constructed a heatmap comparing the first closest L1ME elements upstream to V_H3 genes across primate species (n = 86) and found many clusters where human V_H3 upstream L1ME sequences were nearly identical across species (Figure 7D). For example, the L1ME element upstream to human V_H3–21 is ~85% similar to the L1ME element upstream to human V_H3–7 (heatmap row 9, column 10), but almost 100% homologous to the L1ME element upstream to a gorilla, bonobo, and chimp V_H3 gene (heatmap row 9, columns 8,11,12). We then created a second heatmap by anchoring the positions from the first heatmap (based on upstream L1ME elements), but instead compared the L1ME-associated V_H3 gene sequences themselves across species (Figure 7E). We observed similar clusters between our two heatmaps, suggesting that several L1-V_H3 units likely existed in a primate universal common ancestor. For example, we observed that human V_H3–53 and V_H3–66 cluster together with analogous chimp and gorilla V_H3 genes, and that human V_H3–53 and V_H3–66 upstream L1ME elements cluster together with analogous chimp and gorilla V_H3 upstream L1ME elements. We generated similar heatmaps comparing V_H1 family upstream L1s – L1MCb and L1PREC2 – and associated V_H1 gene sequences themselves, across species (Figure S6C-D). Likewise, we observed ‘hot’ clusters shared between the two heatmaps. In fact, many of the human L1-containing V_H3 and V_H1 family genes that clustered with analogous primate V_H genes were often those most hypermutated in SARS-CoV-2 mRNA immunized donors (Figure S6A). Together, these findings reveal a deeper evolutionary connection between V_H genes with upstream L1s and their proclivity for hypermutation during antigen exposure.

HUSH-AID Partnership

HUSH distinguishes self from non-self nucleic acids and acts as a central hub to recruit other partners to mitigate retroelement mobility.⁷² For instance, HUSH recruits effectors for histone and DNA methylation and members of RNA decay pathways.^66,92,93 This study implicates AID as another HUSH binding partner. AID belongs to the AID/APOBEC family of Zn-dependent cytidine deaminases that convert cytidine of RNA or ssDNA into uridine. APOBEC members have been reported to limit viral and retroelement mobility, through editing or non-editing activity.^63,94 While AID/APOBECs mutate retroelement DNA, HUSH silences retroelements at the transcription level by recruiting SETDB1 to deposit repressive H3K9me3 marks.^67,69 In mouse embryonic stem cells, HUSH cooperates with the nuclear exosome targeting (NEXT) complex, which guides post-transcriptional 3’ decay of retrotransposon RNAs at HUSH-resident loci.^92,95 Our work in hypermutating B cells is analogous, wherein strongly transcribed V_H upstream L1-encoding noncoding RNAs trigger HUSH recruitment, along with AID. Of note, HUSH recruitment does not always guarantee HUSH silencing.⁷² However, we found AID, MPP8, PPHLN1, and H3K9me3 occupancy over the V_H upstream L1 (Figure 3D) and AID-mediated mutations within V_H upstream L1s themselves (Figure 1H-J, Figure S2J-K). While these old L1s do not encode the machinery necessary for retrotransposition, the cell still likely interprets these transcribed sequences as intrinsically non-self. Therefore, this HUSH-AID partnership in B cells ensures that V_H upstream L1s are mutated and also transcriptionally silenced. As a result, AID mutagenesis of V_H-proximal retroelements was extended to the V_H gene itself, resulting in enhanced humoral immunity.

Old, Fossilized L1s Can Still Influence Cellular Function Independent of Retrotransposition

HUSH has been thought to primarily silence transcribed, intronless, young retroelements.^67,69,96,97 Evidence from HUSH ChIP has also shown HUSH occupancy at old, truncated, retrotransposition-dead L1s.^70,92 Within the context of our study in hypermutating B cells, transcription of a strong, long, L1-encoding V_H antisense RNA offers one plausible mechanism by which an old L1-derived sequence can create a HUSH-targetable substrate.

What is Unique About Hypermutating B Cells?

This begets the question: if old L1s and other retroelements are scattered throughout genome, then why isn’t the HUSH/AID partnership found genome-wide? What is special about B cells? First, AID expression is limited to B cells specifically undergoing SHM in germinal centers or CSR.^88,98 Second, young full-length L1s are silenced in most somatic cells.^53,99–102 Third, old L1s incapable of initiating transcription themselves won’t be transcribed if located in intergenic regions or too far upstream to genic promoters (Figure 1A-B). Intronic L1s would be spliced out and degraded during pre-mRNA processing, and would likely be resistant to HUSH targeting, given that introns strongly deter HUSH silencing.⁶⁷ Fourth, not all old L1s upstream to genic promoters will be transcribed, as not all genic promoters are strongly bidirectional. IGH is a uniquely privileged locus, located within a tight-knit TAD full of IGH-specific super-enhancers.^{25,28,103–107} V_H upstream L1s are close to their nearby V_H promoters, whose transcriptional output is fortified by the super-enhancer-rich TAD (i.e. IgH TAD). The IgH TAD has two super-enhancers, Eμ and 3’ regulatory region (3’RR), which both drive SHM by activating the V_H promoter.^108,109 Should a genic promoter experience this “perfect storm” – a nearby upstream L1 coupled with enhancer-boosted bidirectional transcription – then this could possibly explain how HUSH/AID contribute to low levels of off-target SHM. Alternatively, AID off-target genes could recruit AID at very low levels through HUSH-independent mechanisms such as transcription stalling and secondary structure formation.¹¹⁰ By and large, however, HUSH does not seem to recruit AID heavily to other genes or non-Ig L1s genome-wide, evidenced by our AID CUT&RUN data (Figure 4I-L).

The strong transcription of V_H upstream L1s, a phenomenon largely unique to IGH, likely explains initial AID recruitment to recombined V_H genes in hypermutating B cells, grounded in the evolutionarily conserved defense against non-self sequences. This can also explain why the 3’RR super-enhancer has a remarkably strong effect on SHM.

How Does Recruited AID Ultimately Reach the Recombined V_H Gene?

Following transcription of the V_H upstream L1, HUSH recruits SETDB1 for H3K9me3 deposition. However, HUSH recruitment does not always guarantee constitutive heterochromatinization. In this case, it is likely that HUSH induces facultative heterochromatinization and does not repress the V_H promoter due to the constant activation from two IGH super-enhancers fortified by the IGH TAD. Heterochromatinization then tightens the genomic DNA (flanking the HUSH/SETDB1/AID-bound L1) around histones, contracting this region and bringing everything closer in proximity. Thus, AID need not “spread” or “translocate” hundreds of basepairs into the downstream V_H gene; rather, the V_H gene comes closer to the AID-localized L1.

SHM at Off-Target Loci vs. Recombined V_H Genes

We introduced a cassette in front of non-Ig genes – BCL6, MYC, MKL1 – to see if we could induce SHM at regions normally blind to AID mutagenesis (Figure 6A-F). The cassette contained a strong EF1α core promoter driving transcription of old L1s that were cloned from upstream regions of human V_H genes. We observed increased SHM at all three off-target sites, highlighting the versatility of this L1-driven SHM mechanism. In theory, these findings suggest that oncogenes with L1s near their promoters could face elevated risk of AID-mediated genome rearrangements.

AID, MPP8, PPHLN1, and H3K9me3 CUT&RUN revealed that noL1 cells had low levels of AID and HUSH along the recombined V_H(D)J_H, with correspondingly low levels of SHM (Figure 1D-G). AID and HUSH recruitment were significantly enhanced after insertion of V_H upstream L1s (Figure 3D), with correspondingly high levels of SHM, particularly observed after insertion of an upstream 1.5 kb L1 ORF2 fragment (Figure 6G-K). Both full-length (~4 kb) and truncated L1 ORF2 are bona fide HUSH binding substrates.⁶⁷ Our work suggests that old L1s, when incorporated into a 2–3 kb, strongly expressed transcript, can also serve as HUSH substrates. Moreover, our HUSH CUT&RUN data suggest that the strongly transcribed V_H(D)J_H RNA may intrinsically possess properties of a HUSH substrate (Figure 3D). Overall, HUSH recruitment is influenced by the strong expression of a 2–3 kb, intronless, L1-encoding V_H antisense RNA coupled with the strong expression of a very small intron-containing, 0.5–1 kb nascent V_H(D)J_H sense RNA originating from stalled/terminated RNAPII.

L1-Encoding V_H Antisense Transcription and V_H Upstream Mutation Window

Our proposed model also addresses other outstanding questions in the field.¹¹¹ The AID mutation window has been reported to span roughly +2 kb from the V_H TSS.^112,113 We showed that transcription of V_H upstream L1s aids in recruiting AID to recombined V_H(D)J_H, supported by AID CUT&RUN signal upstream to the recombined V_H(D)J_H (Figure 3D) and AID mutations in recombined V_H upstream regions containing L1s (Figure 1H-J, S2J-K). Reconsidering AID targeting from the evolutionary perspective of active retrotransposon silencing (Figure 7F) clarifies why AID would initially be recruited upstream to actively transcribed V_H genes, expanding upon the current mutation window. This reinforces our hypothesis that AID first evolved to silence actively transcribed retrotransposons, such as L1s, before being serendipitously co-opted over evolution to enhance immunoglobulin somatic hypermutation.

Human Ig Light Chain, Chicken IGH, and L1-IGH Evolution

There are two human light chain loci, comprising ~44 functional kappa (IGK) Vκ genes and ~37 lambda (IGL) Vλ genes. While recombined light chain V genes undergo AID-mediated SHM, we believe the mechanism of AID targeting differs from that proposed herein for recombined V_H genes. Human light chain V genes possess scattered upstream L1s, but lack a robust, periodic, consistent enrichment of upstream L1s (Figure S7A, Data S2). Surprisingly, we observed an enrichment of LTR retrotransposons upstream to Vκ genes across species (Figure S7B). Furthermore, RNA sequencing reveals that recombined light chain V promoters are bidirectional (Figure S7C). Vκ antisense transcription of upstream LTRs could very well recruit HUSH and other silencing factors, including AID, to the recombined light chain locus, as LTR sequences have been shown to be targeted by HUSH.^70,72,97

Chickens use a vastly different mechanism for antibody diversification known as gene conversion.^114,115 Chickens possess a single functional V_H gene that is hypermutated through homologous recombination with 80–100 upstream pseudo-V_H genes. We detected satellite repeat sequences, not L1s, in front of these pseudo-V_H genes (Figure S7D). Satellite repeats of 10–11 bp were located within ~100 bp upstream to pseudo-V_H TSSs, with high GC content and high single-stranded DNA/RNA forming properties (Figure S7E-H). The potential to form stable GC-rich secondary structures upon V_H antisense transcription could recruit SHM machinery in chickens, a mechanism more analogous to switch region recruitment of AID during CSR.^32–34,111 To a lesser extent, chickens also use pseudogene-independent, non-templated hypermutation mechanisms.^116,117

Not all V_H genes have nearby upstream L1s (Figure S1A-D), and why V_H1 and V_H3 families each possess their own specific upstream L1s remains to be studied. From our L1-IGH evolutionary heatmap analysis (Figure 7C-E, S6B-D), we found that certain L1-V_H1 and L1-V_H3 gene units possibly originated from the primate universal common ancestor. These ancestral L1-containing V_H1 and V_H3 genes were also those most hypermutated among COVID-19 immunized donors. We propose that these L1-associated V_H genes may be more useful at neutralizing antigen; over evolution, we preserved and even duplicated these V_H genes for better adaptive immunity.

Limitations of the study

Future work will aim to identify ‘AID signatures’ over V_H upstream L1s across species, as AID is incredibly homologous across species. Second, HUSH also recruits MORC2, a key effector in chromatin compaction, which might contribute to L1-driven SHM. Third, the relationship between TEs and light chain lambda V genes is less clear, given the complexity of the locus and preference for humans and mice to use light chain kappa V genes more frequently; TE involvement in light chain SHM needs to be experimentally validated. Finally, it is possible that other mechanisms of SHM exist, independent of L1, like those seen at AID off-target sequences or light chain loci. The interplay between L1-driven SHM and these other factors is an area of future study.

Lead Contact

Further information and requests for reagents in this study will be fulfilled by lead contact, Uttiya Basu (ub2121@cumc.columbia.edu)

Materials availability

All the resources generated in this study are available from lead contact with a completed materials transfer agreement.

Data and code availability

• The sequencing data includes CUT&RUN-seq, RNA-seq, Nanopore sequencing, deposited under BioProject ID PRJNA1355250, by the date of publication.

• All original code has been deposited at Zenodo (https://doi.org/10.5281/zenodo) and are publicly available as of the date of publication.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

METHOD DETAILS

QUANTIFICATION AND STATISTICAL ANALYSIS

Quantification and statistical tests were performed using GraphPad Prism (v8.0.2) or R (v3.5.1, http://www.R-project.org) and are explicitly stated in each figure legend. Unless otherwise stated, all data is plotted with individual data points and the error bars represent standard error of the mean, where *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001; ns, not statistically significant. In all plots related to mouse experiments, each data point represents one mouse.