Retrotransposon derepression leads to activation of the unfolded protein response and apoptosis in pro-B cells

Epigenetic mechanisms regulate developmental transitions by mediating the activation or stable repression of lineage-appropriate or lineage-inappropriate genes, respectively. Dysregulation of epigenetic machineries therefore has adverse consequences for development. In the context of hematopoiesis, impairment of repressive chromatin marks, such as DNA methylation, or Polycomb silencing result in compromised stem and progenitor cell differentiation and may even lead to the development of malignancies (Beguelin et al., 2013; Su et al., 2003). Another major repression mechanism involves the heterochromatin modification H3K9me3. This modification is broadly enriched at pericentric heterochromatin, various classes of retrotransposons, imprinted loci and repressed developmental genes. The histone methyltransferase Setdb1 mainly controls H3K9me3 outside of pericentric heterochromatin and plays crucial roles in development (Karimi et al., 2011; Regha et al., 2007). Setdb1 mutant embryos die even before implantation (Dodge et al., 2004), and conditional inactivation of Setdb1 during neurogenesis or in mesenchymal cells coincides with severe developmental phenotypes (Lawson et al., 2013; Tan et al., 2012; Yang et al., 2013). Thus, Setdb1-mediated H3K9me3 appears to regulate developmental transitions.

The functions of Setdb1 have been well characterized in mouse embryonic stem cells (ESCs). In these cells, Setdb1 depletion leads to the derepression of lineage-specifying genes and loss of pluripotency. Transcriptional dysregulation in Setdb1-deficient ESCs is due to the loss of H3K9me3 from promoter regions of lineage genes (Bilodeau et al., 2009; Yuan et al., 2009). In addition, Setdb1-deficient ESCs are characterized by enhanced transcription of retrotransposons. How retrotransposon derepression might have phenotypic consequences is unclear. Two hypotheses are currently discussed. First, that derepression of retrotransposons leads to elevated transcription of neighboring genes through enhancer action of chimeric transcripts (Karimi et al., 2011; Matsui et al., 2010). This, in turn, might impair the transcriptional stability in cells. Second, that enhanced activity of functional retrotransposon copies, which are still able to jump, may lead to mutations (Lee et al., 2012) or genomic instability (Bourc'his and Bestor, 2004).

Here, we show that conditional inactivation of Setdb1 in pro-B cells leads to a block in B cell development. We found that Setdb1-deficient pro-B cells show derepression of specific classes of retrotransposons, among which the endogenous murine leukemia virus (MLV) elements exhibited the highest transcriptional activation. Notably, forced MLV expression coincides with massive production of MLV-derived proteins which, in turn, triggers activation of the unfolded protein response (UPR) and subsequent apoptosis of pro-B cells. Expression of pro-survival Bcl2 antagonizes UPR-mediated apoptosis and leads to a partial rescue of B cell development, indicating that retrotransposon silencing is the primary role of Setdb1 during early B cell development. In summary, our data provide a novel molecular explanation for the adverse consequences of impaired retrotransposon silencing during development.

Setdb1 is constitutively expressed during B cell development (Heng et al., 2008). To delete Setdb1 specifically during early stages of B cell development, we generated a mouse strain in which we combined Mb1 (Cd79a)-Cre with a conditional Setdb1 allele that has the crucial exon 4 flanked by loxP sites. These Mb1-Cre; Setdb1^flox/− mice are referred to hereafter as Setdb1^Mb1, and Mb1-Cre; Setdb1^flox/+ or +/+; Setdb1^flox/+ mice provide the controls. Mb1-Cre initiates deletion of loxP-flanked regions at the transition from pre-pro-B cells to committed pro-B cells (Hobeika et al., 2006).

Spleen size was severely reduced in Setdb1^Mb1 mice (Fig. 1A). Histological examination revealed reduced number and size of follicles (Fig. 1B). Total cell numbers of spleen and bone marrow were strongly reduced in Setdb1^Mb1 mice (Fig. 1C). Consistent with the lack of Mb1-Cre activity in T cell development, no changes in thymus cell number were detected (Fig. 1C). FACS analysis of B cell markers revealed almost complete absence of B cells [B220 (Ptprc)⁺ CD19⁺] in Setdb1^Mb1 spleen and severely reduced B cell numbers in bone marrow (Fig. 1D,E). The lack of peripheral B cells prompted us to identify the stage at which B cell development is impaired in Setdb1^Mb1 mice. Analysis of the B cell compartment in bone marrow revealed that the percentages of pro-B cells [CD19⁺ IgM⁻ IgD⁻ CD25 (Il2ra)⁻ Kit⁺] are comparable between control and Setdb1^Mb1 mice; however, pre-B cells (CD19⁺ IgM⁻ IgD⁻ CD25⁺ Kit⁻) were severely reduced and neither immature (B220⁺ IgM⁺ IgD⁻) nor recirculating (B220⁺ IgM⁻ IgD⁺) B cells could be detected (Fig. 1F,G). Analysis of hematopoietic stem cell and progenitor populations revealed no difference between control and Setdb1^Mb1 mice, as expected (Fig. S1).

To test whether blocked B cell development was due to cell-intrinsic defects, we performed competitive transplantation experiments by injecting a 1:1 mixture of wild-type CD45.1 bone marrow progenitors together with control or Setdb1^Mb1 bone marrow cells carrying the congenic marker CD45.2 into lethally irradiated recipients. Whereas the control bone marrow cells contributed to the B cell lineage comparably to the co-injected wild-type cells, Setdb1^Mb1 B cells were hardly detectable in the periphery (Fig. 2A). Analysis of the B cell compartment in the bone marrow of transplanted mice revealed that Setdb1^Mb1 B cells were severely compromised at the pre-B cell stage (Fig. S2).

We then performed in vitro B cell differentiation assays. Control bone marrow cells readily formed colonies in IL7-supplemented MethoCult M3630 medium, which supports B cell differentiation. Conversely, almost no colonies were obtained from Setdb1^Mb1 bone marrow (Fig. 2B). Furthermore, in vitro differentiation of lineage-depleted bone marrow cells on OP9 stromal cells in the presence of IL7 revealed severely impaired differentiation of Setdb1^Mb1 progenitors into the B cell lineage (Fig. 2C). Thus, our data demonstrate an essential cell-autonomous function of Setdb1 in the early stages of B cell development.

To identify the molecular mechanisms leading to impaired B cell development in the absence of Setdb1, we performed transcriptional profiling by high-throughput RNA sequencing (RNA-seq). RNA was isolated from FACS-sorted pro-B cells (CD19⁺ IgM⁻ IgD⁻ CD25⁻ Kit⁺) from control and Setdb1^Mb1 mice. Setdb1 was completely deleted in Setdb1^Mb1 pro-B cells as no band could be detected upon PCR amplification of the floxed exon 4 (Fig. 3B). Through RNA-seq analysis we found 130 upregulated and 136 downregulated (>2-fold) genes in Setdb1^Mb1 pro-B cells (Fig. 3A, Table S1). GO term analysis of the regulated genes revealed enrichment for pathways implicated in immune system development (Fig. 3C). However, closer examination of the dataset indicated that none of the known essential factors for B cell development was dysregulated.

Rather, we found transcriptional changes in genes associated with the pro-B to pre-B cell transition. One example is Aiolos (Ikzf3), which is normally upregulated in late pro-B cells. Aiolos expression was detected in control pro-B cells, but it failed to be activated in Setdb1^Mb1 pro-B cells. This suggested that the pro-B cell population isolated from Setdb1^Mb1 bone marrow resided in a more immature stage. To substantiate this finding we performed gene set enrichment analysis using gene sets that are representative of early pro-B and late pro-B/early pre-B cells, respectively (Jojic et al., 2013). Notably, the early pro-B signature was enriched in Setdb1^Mb1 pro-B cells, whereas the late pro-B/early pre-B signature was over-represented in control cells (Fig. 3D). These analyses suggest a bias toward having reduced numbers of late pro-B/early pre-B cells in Setdb1^Mb1 bone marrow. In order to test this hypothesis we analyzed B cell populations using a different marker set (Hardy et al., 1991). We found significantly reduced late pro-B/early pre-B cells [B220⁺ CD43 (Spn)⁺ HSA (CD24a)⁺ BP-1 (Enpep)^+/−] in Setdb1^Mb1 mice (Fig. S3). These data suggest that the majority of the transcriptional changes that we detected by RNA-seq simply reflect the block in pro-B to pre-B cell transition in Setdb1^Mb1 mice.

To identify genomic regions that may be controlled by Setdb1 in pro-B cells, we performed chromatin immunoprecipitation for Setdb1, H3K9me3 and H3K9ac in short-term cultured Rag2^−/− pro-B cells followed by high-throughput sequencing (ChIP-seq). This analysis revealed 5368 Setdb1 binding sites in pro-B cells, which are shared in two independent datasets (see Fig. S4 for analysis of the individual datasets).

To determine the binding sites at which Setdb1 could induce H3K9me3, we calculated the ChIP-seq read coverage for Setdb1, H3K9me3 and H3K9ac in 1500 bp windows across all Setdb1 binding sites. These data were then clustered according to H3K9me3 and H3K9ac density (Fig. 4A). This analysis revealed that only a subset of Setdb1 binding sites associates with H3K9me3 (clusters A, B). A large number of Setdb1 peaks are found together with H3K9ac (clusters D, E, F) or do not display prominent enrichment for either H3K9me3 or H3K9ac (clusters C, G). We then analyzed the structural features underlying Setdb1 binding sites. H3K9me3-associated regions are clearly enriched for repeat elements (clusters A, B). By contrast, H3K9ac-rich Setdb1 peaks are mainly present at promoter regions (clusters D, E, F). We then asked whether genes near to Setdb1 binding sites are transcriptionally regulated in Setdb1^Mb1 pro-B cells. For each cluster, genes with a transcriptional start site within 5 kb of the Setdb1 binding site were extracted. However, we could not detect any significant enrichment of regulated genes in any of the Setdb1 binding site clusters. These data demonstrate that in pro-B cells Setdb1 regulates only a small number of genes in the vicinity of its binding sites.

The ChIP-seq analysis showed that Setdb1 binding sites that are enriched for H3K9me3 (clusters A, B) mainly reside within repeat elements. Therefore, we investigated whether distinct classes of repeat elements are dysregulated in Setdb1^Mb1 pro-B cells. Analysis of our RNA-seq data uncovered three classes of retrotransposons (MLV, MMVL30 and MMTV) to be upregulated in Setdb1^Mb1 pro-B cells (Table S2, Fig. S5A). Re-examination of RNA-seq reads with unique mapping in the genome allowed to distinguish individual copies of these retrotransposon classes. The most strongly upregulated retrotransposons were four copies of the MLV class, with >100-fold overexpression (Fig. 4B). Notably, these four copies showed the strongest enrichment for Setdb1 and H3K9me3 (Fig. 4C). Intriguingly, IAP-Ez retrotransposons, which are major targets of Setdb1 in ESCs, were also enriched for Setdb1 and H3K9me3 in pro-B cells, but did not show significantly enhanced expression in Setdb1^Mb1 pro-B cells (Fig. 4B,C).

In summary, our data suggest that although Setdb1 occupies many binding sites in pro-B cells, only a subset of these regions correlates with changes in gene expression when Setdb1 is absent. Our results also show that in pro-B cells the major repressive function of Setdb1 is exerted on specific classes of retrotransposons.

Our ChIP-seq data revealed strongest enrichment of Setdb1 and H3K9me3 at distinct MLV elements. Close inspection of the genomic regions around the derepressed MLV retrotransposons revealed that genes in close proximity were also highly upregulated. An example is the MLV retrotransposon on chromosome 8 that lies in the vicinity of Tubb3 and Def8 (MLV8, Fig. 5A). Tubb3 encodes a neuron-specific tubulin involved in axon guidance; Def8 (differentially expressed in FDCP 8) is a largely uncharacterized gene of unknown function. RNA-seq data show strong derepression of MLV8 and upregulation of both Tubb3 and Def8. Notably, upregulation of both genes is not mediated by read-through transcripts originating from MLV8, which is oriented tail-to-tail with Tubb3 and head-to-head with Def8. ChIP-seq data revealed enrichment of Setdb1 and H3K9me3 across MLV8, but only background levels were detected across Tubb3 and Def8. These data suggest that loss of Setdb1 leads to derepression of MLV8, which then exerts enhancer effects on the neighboring genes leading to their upregulation. A similar example is MLV1, which lies in proximity to the Fcgr2b gene (Fig. S5B). In Setdb1^Mb1 pro-B cells, both MLV1 and Fcgr2b are strongly derepressed (Fig. 5B) and Fcgr2b protein is even highly incorporated into the cell membrane (Fig. S5C,D). Interestingly, Fcgr2b is also upregulated in Setdb1-deficient T cells (Martin et al., 2015), although it remains to be determined if derepression of MLV1 can be observed in this context. RT-qPCR analyses for another MLV element (MLV5) similarly revealed strong derepression of the corresponding MLV transcript and upregulation of the neighboring gene (Fig. 5B).

To test whether Setdb1-dependent chromatin changes lead to dysregulation of MLV elements, we isolated control and Setdb1^Mb1 CD43⁺ CD19⁺ pro-B cells by FACS sorting and performed ChIP-qPCR analyses. H3K9me3 was prominently enriched on MLV elements; however, loss of Setdb1 only resulted in a small reduction of this modification (Fig. 5C). It is possible that other histone methyltransferases, such as Suv39h, contribute to H3K9me3 establishment (Bulut-Karslioglu et al., 2014). H3K4me3, a mark of active chromatin, was strongly elevated on MLV elements and on the promoters of Tubb3 and Def8 in Setdb1-deficient pro-B cells (Fig. 5D). No enrichment of H3K4me3 could be detected on IAP-Ez elements, which showed no transcriptional changes in Setdb1^Mb1 pro-B cells. We also tested whether any changes in DNA methylation might contribute to transcriptional activation of these MLV elements. Very high levels of DNA methylation were detected in the strongly upregulated MLV8, demonstrating that loss of DNA methylation is not required for transcriptional activation (Fig. 5E,Fig. S6). IAP elements and a transcriptionally unchanged MLV retrotransposon did not show any changes in DNA methylation (Fig. 5E, Fig. S6). Our data suggest that the presence of Setdb1 inhibits the binding and/or activity of specific transcription factors that would otherwise establish an active chromatin structure. Complete loss of H3K9me3 or DNA methylation is apparently not necessary to allow the establishment of H3K4me3 and productive transcription.

Our transcriptional profiling of Setdb1^Mb1 pro-B cells did not reveal dysregulation of important B cell-related transcription factors, which could have explained a developmental block. Another possible explanation for impaired B cell development is that pro-B cells die from apoptosis. To test if apoptosis is elevated in Setdb1^Mb1 pro-B cells we performed annexin V (annexin A5) staining on control and Setdb1^Mb1 bone marrow cells. In control mice, ∼30% of pro-B cells enter apoptosis, which is likely to be due to non-productive VDJ recombination. Interestingly, the apoptosis rate was strongly increased in Setdb1^Mb1 pro-B cells (Fig. 6A).

If enhanced apoptosis of pro-B cells is a major reason why B cell development is blocked in Setdb1^Mb1 mice then we would predict that introducing a pro-survival protein might compensate for Setdb1 deficiency in B cell development. To test this hypothesis we introduced an allele that overexpresses anti-apoptotic Bcl2 in all hematopoietic cells [Vav-Bcl2 (Egle et al., 2004)] into Setdb1^Mb1 (Setdb1^Mb1; Bcl2) or control (Bcl2) mice. Apoptosis in both Bcl2 and Setdb1^Mb1; Bcl2 pro-B cells was comparably low (Fig. 6B). We then examined B cell development in the bone marrow of Setdb1^Mb1; Bcl2 mice. Compared with Setdb1^Mb1 (Fig. 1F,G) there was an increase in the percentages of pre-B, immature B and mature B cells in the Setdb1^Mb1; Bcl2 mice (Fig. S7). Significant numbers of Setdb1-deficient mature B cells were observed in the spleen of Setdb1^Mb1; Bcl2 mice (Fig. 6C-E), which were virtually absent in Setdb1^Mb1 mice (Fig. 1D,E). These results were further supported by in vitro differentiation experiments with lineage-depleted bone marrow cells from Bcl2 and Setdb1^Mb1; Bcl2 mice, which showed almost comparable growth of pre-B cell colonies (Fig. 6F).

An important process during pro-B cell development is VDJ recombination of the Igh locus to produce a functional pre-B cell receptor. Consistent with the block in pro-B to pre-B transition in Setdb1^Mb1 mice, we detected lower efficiency of VDJ rearrangement (Fig. S8). By contrast, VDJ recombination was unaffected in Setdb1^Mb1; Bcl2 pro-B cells (Fig. S8). Thus, our data demonstrate that counteracting apoptosis can partially rescue the developmental phenotype of Setdb1^Mb1 mice.

We then examined whether apoptosis is linked with the transcriptional changes that we observe in Setdb1^Mb1 pro-B cells. As we only detected moderate regulation of protein-coding genes (Fig. 3), we hypothesized that upregulation of retrotransposons might lead to apoptosis. Upregulation of ERVs has been connected with enhanced DNA damage that may lead to apoptosis. However, we did not detect increased DNA breaks in Setdb1^Mb1 B cells (Fig. S9). We then attempted to establish whether the transcription of retrotransposons leads to the production of retroviral proteins. The top upregulated retrotransposon class in Setdb1^Mb1 pro-B cells is MLV. MLV transcripts have coding potential and can lead to the production of retroviral proteins. For example, the retroviral glycosylated envelope (Env) protein can be detected on cells with elevated expression of MLV transcripts (Evans et al., 1990; Young et al., 2012). We tested MLV Env protein production by FACS analysis and could detect high levels on the surface of Setdb1^Mb1 pro-B cells as compared with littermate controls (Fig. 7A). Western blot analyses confirmed high expression of MLV Env protein in Setdb1-deficient pro-B cells (Fig. S10A).

Could apoptosis in Setdb1^Mb1 pro-B cells be linked to the excessive production of MLV proteins? Unlike plasma cells, for example, which are specialized for the mass production of secreted proteins (Brewer and Hendershot, 2005; Gass et al., 2004), pro-B cells are relatively small, with a limited capacity to produce such proteins. The strongly elevated translation of proteins into the endoplasmic reticulum may lead to an accumulation of improperly folded proteins, which triggers a cellular stress pathway known as the unfolded protein response (UPR). We tested for activation of the UPR pathway in Setdb1^Mb1 pro-B cells by expression analysis of key UPR genes.

A hallmark of the UPR response is the accumulation of a specific splice form of the mRNA encoding the transcription factor Xbp1 (Calfon et al., 2002). Importantly, compared with control pro-B cells, we detected an increased level of spliced Xbp1 (Xbp1s) in Setdb1-deficient pro-B cells (Fig. 7B). Furthermore, expression of pro-apoptotic Bcl2l11 was elevated (Fig. 7B). Other key UPR genes, such as those encoding the chaperones Hspa5 (Grp78 or BiP) and Pdia6 (Groenendyk et al., 2014; Lee, 2005), were also upregulated in Setdb1-deficient pro-B cells (Fig. 7B). As elevated apoptosis in Setdb1^Mb1 pro-B cells might prevent the accumulation of UPR transcripts, we also tested the UPR pathway in Setdb1^Mb1; Bcl2 pro-B cells. Notably, we detected even stronger upregulation of key UPR genes in Setdb1^Mb1; Bcl2 pro-B cells (Fig. 7B). In addition, gene set enrichment analysis using the hallmark gene sets from MySigDB revealed significant enrichment of the UPR gene signature in Setdb1^Mb1 pro-B cells (Fig. 7C). Thus, the increased expression of UPR components in both Setdb1^Mb1 and Setdb1^Mb1; Bcl2 pro-B cells clearly demonstrates activation of the UPR pathway.

The key inducer of apoptosis in the context of UPR is Bcl2l11 (Puthalakath et al., 2007). To test if apoptosis in Setdb1^Mb1 pro-B cells is mediated through Bcl2l11, we performed knockdown experiments in in vitro B cell differentiation assays. Control cells with scrambled shRNAs could readily form colonies in MethoCult M3630, whereas Setdb1^Mb1 cells did not form colonies (Fig. 7D). Knockdown of Bcl2l11 in Setdb1^Mb1 cells resulted in significantly elevated colony numbers (Fig. 7D), demonstrating that Bcl2l11 is mainly responsible for apoptosis induction in Setdb1^Mb1 cells.

Finally, we examined whether UPR-mediated apoptosis is mainly caused by overexpression of MLV proteins. We performed in vitro differentiation assays of wild-type cells in which we overexpressed GFP or MLV Env protein of the endogenous MLV1 retrovirus. Wild-type CD19⁺ B cells, in which we did not overexpress additional proteins, show a low rate of apoptosis (Fig. 7E). GFP overexpression in B cells did not reveal elevated apoptosis (Fig. 7E,Fig. S10B). By contrast, overexpression of MLV Env protein led to strongly increased apoptosis in B cells (Fig. 7E, Fig. S10B). Importantly, RT-qPCR analyses of in vitro differentiated B cells revealed specific activation of UPR genes upon Env overexpression (Fig. 7F).

In summary, our data show that elevated expression of MLV transcripts leads to excessive production of MLV Env protein, which triggers the UPR pathway, leading to apoptosis in Setdb1^Mb1 pro-B cells and resulting in a block to B cell development.

Our study demonstrates that it is crucial to silence specific retrotransposons during development (Fig. 7G), which is in full agreement with a recent study of Setdb1 function in B cells (Collins et al., 2015). In wild-type cells, Setdb1 binds to MLV retrotransposons and establishes a repressive chromatin structure that prevents access of activating transcription factors. Upon deletion of Setdb1 in pro-B cells, some ‘responsive' MLV elements display strong transcriptional activity. It is possible that, owing to the loss of Setdb1, transcription factors can now access these MLVs leading to the establishment of active H3K4me3 marks. Interestingly, this activation can occur even though H3K9me3 is only minimally reduced and DNA methylation is unchanged. The activation of MLVs leads to two outcomes. First, responsive MLV retrotransposons may act as enhancers to strongly stimulate the transcription of neighboring genes. Second, the excessive production of MLV-derived proteins triggers UPR in pro-B cells, leading to apoptosis.

Genome-wide Setdb1 binding had so far only been characterized in ESCs, where targets include retrotransposons and promoters of developmental genes (Bilodeau et al., 2009; Yuan et al., 2009). In pro-B cells, we also detected Setdb1 on promoter regions; however, we did not observe strong transcriptional changes of these target genes in Setdb1-deficient pro-B cells. Interestingly, in pro-B cells, Setdb1 is mostly unable to establish repressive H3K9me3 marks on promoter binding sites, as we detected strong H3K9ac occupancy at these regions. Transcription of these targets does not seem to be affected by Setdb1 loss and, therefore, it is currently unclear what function Setdb1 exerts on such binding sites. The major target sites at which Setdb1 mediates establishment of H3K9me3 and transcriptional repression in pro-B cells are retrotransposons (Fig. 4). How Setdb1 is recruited to these elements remains to be clarified. In ESCs, targeting is mainly mediated by Trim28; however, Trim28 is not enriched on MLV retrotransposons in B cells (Santoni de Sio et al., 2012). Moreover, Trim28 deletion in the B cell lineage did not result in impaired B cell development, but rather led to transcriptional dysregulation in mature B cells (Santoni de Sio et al., 2012). Thus, Trim28-dependent mechanisms are unlikely to mediate Setdb1 targeting to MLVs.

Derepression of retrotransposons in Setdb1^Mb1 pro-B cells implies that the absence of Setdb1 allows binding of specific transcription factors to these elements. There is emerging evidence that transposable elements contain transcription factor binding sites (Xie et al., 2010) and may even physiologically act as tissue-specific enhancers (Xie et al., 2013). A recent study showed that the master B cell transcription factor Pax5 may mediate forced expression of the same MLV elements that we found to be upregulated in Setdb1-deficient pro-B cells (Collins et al., 2015) (Table S3). However, additional cell type-specific transcription factors are likely to contribute to MLV transcription, as Setdb1-deficient mouse embryonic fibroblasts (Matsui et al., 2010) or myeloid cells (data not shown), which do not express Pax5, also display derepression of MLV retrotransposons. It is interesting to note that in pro-B cells only very few retrotransposon classes are derepressed, whereas Setdb1-deficient ESCs show strong derepression of many retrotransposon classes, including IAP-Ez, ETn, MusD and others (Karimi et al., 2011). In particular, in ESCs the major Setdb1 targets are IAP-Ez elements, which, although bound by Setdb1 and enriched for H3K9me3, display no transcriptional changes in Setdb1^Mb1 pro-B cells. This might be explained by the lack of IAP-specific transcription factors in pro-B cells. Alternatively, redundant repression mechanisms, such as DNA methylation, might ensure silencing of these repeats in the absence of Setdb1. Consistent with this hypothesis, we did not detect significant changes in DNA methylation at these elements (Fig. 5).

Our data demonstrate that transcripts from endogenous MLV retroviruses have coding potential and lead to the strong production of MLV proteins. In pro-B cells, this triggers activation of the UPR pathway, and cells enter apoptosis. To our knowledge, this is the first demonstration that retrotransposon activation is linked with this specific cellular stress pathway. Several lines of evidence support our notion that UPR-mediated apoptosis is a major cause of impaired B cell development in Setdb1^Mb1 mice. First, no crucial B cell transcription factors are dysregulated in Setdb1^Mb1 pro-B cells (Fig. 3, Table S1). The only hematopoiesis-related gene that shows strong upregulation is Fcgr2b (Fig. S5). However, overexpression of Fcgr2b does not lead to defective B cell development (Brownlie et al., 2008). Second, the apoptosis phenotype is mainly due to elevated MLV protein expression, as overexpression of MLV Env protein is sufficient to induce apoptosis in B cells (Fig. 6D). Interestingly, expression of MLV Env protein from exogenous MLV retroviruses also results in increased apoptosis (Zhao and Yoshimura, 2008) and infection with MLV retroviruses results in impaired B cell development in vivo (Finstad et al., 2007). Furthermore, it is possible that transcripts/proteins produced from other upregulated repeat elements contribute to the phenotype. Third, blocking apoptotic pathways in Setdb1^Mb1; Bcl2 mice resulted in a partial rescue of B cell development. Importantly, detection of mature B cells in the spleen of Setdb1^Mb1; Bcl2 mice demonstrates that expression of key developmental genes is unlikely to be affected by Setdb1. The rescue might be partial because preventing apoptosis by Bcl2 expression does not cease MLV Env production in Setdb1^Mb1; Bcl2 pro-B cells. Therefore, those B cells are still compromised and may die through apoptosis-independent pathways. Lastly, apoptosis is mainly due to activation of the UPR pathway. We detect upregulation of key UPR genes in both Setdb1^Mb1 and Setdb1^Mb1; Bcl2 pro-B cells (Fig. 6B). Knockdown of the key UPR apoptosis inducer Bcl2l11 leads to rescue of B cell development (Fig. 6C). We also attempted rescue experiments by knocking down other UPR components. However, owing to the redundancy in UPR signaling pathways, knockdown of individual UPR genes does not compromise UPR activation (Puthalakath et al., 2007). Furthermore, important UPR genes are essential for B cell development (Zhang et al., 2005) and cannot be knocked down. We cannot fully exclude additional causes for apoptosis, such as DNA damage or large-scale changes in chromatin architecture. However, as overexpression of MLV Env protein is already sufficient to induce apoptosis (Fig. 6D) and to trigger UPR (Zhao and Yoshimura, 2008), our data strongly suggest that UPR activation is a major cause for apoptosis in Setdb1^Mb1 pro-B cells. Unfortunately, owing to the limited number of pro-B cells and the unavailability of good antibodies, protein expression levels of UPR-related genes and protein-protein interactions between MLV-derived and UPR proteins could not be further assessed.

UPR-induced cell death mediated by retrotransposons might not be limited to pro-B cells. Derepression of other retrotransposon classes in different cell types may involve similar mechanisms. For example, apoptosis in Setdb1-deficient ESCs (Matsui et al., 2010) or neurons (Tan et al., 2012), which involve strong derepression of IAP retrotransposons, may be linked with UPR stress. However, additional cellular stress pathways exist that may detect the overexpression of endogenous retroviruses (ERVs) in other systems. For example, production of double-stranded RNA from ERVs could be recognized by specific pattern-recognition receptors leading to activation of the interferon response pathway (Roulois et al., 2015). Future experiments will reveal which cell types are particularly sensitive to overexpression of ERVs and which cellular pathway(s) can be triggered by distinct ERV classes.

Mice carrying the floxed Setdb1 allele were purchased from the EUCOMM project [Setdb1^{tm1a(EUCOMM)Wtsi}]. Mb1-Cre and Vav-Bcl2 transgenes have been described previously (Egle et al., 2004; Hobeika et al., 2006). Setdb1 genotyping is described in the supplementary Materials and Methods. Animals were housed in ventilated cages in the mouse facility at the Adolf Butenandt Institute. All procedures involving animals were in accordance with EU regulations. For the experiments, 5- to 10-week-old animals were used.

HEK 293T, OP9 stroma cells and progenitor cells were cultivated in DMEM (Gibco), IMDM and RPMI (Gibco), respectively. Media were supplemented with 10% FCS, 1% non-essential amino acids, 1% penicillin/streptomycin and 0.2% β-mercaptoethanol (Sigma). For progenitor cell short-term culture, RPMI was supplemented with 10 ng/ml IL7 (PeproTech).

Single-cell suspensions from bone marrow and spleen were stained for 20 min at 4°C using combinations of antibodies (see Table S4) conjugated with fluorochromes detectable in the following channels: FITC, PE, PE-Cy5, Pe-Cy7, APC, APC-Cy7. All samples were pre-incubated for 20 min at 4°C with unconjugated CD16/CD32 (Fcgr3) Fc-blocking antibody to avoid nonspecific binding, unless otherwise indicated. Data were acquired using a FACSCanto (BD Biosciences) and cell sorting was performed using either a MoFlo (Beckman Coulter) or FACSAria III (BD Biosciences). Sorted samples were pretreated with red blood cell lysis buffer (BD Biosciences) or enriched using CD45R (B220) microbeads (Miltenyi). Data from flow cytometry were analyzed using FlowJo software (TreeStar).

Erythrocyte lysis was performed using BD Pharm Lyse (BD Pharmingen). 10× RBC lysis buffer (BD Biosciences) was diluted using distilled water and kept at room temperature. 3 ml 1× RBC lysis buffer was used to treat bone marrow cells derived from one mouse. Bone marrow cells were washed with 1× PBS and centrifuged at room temperature for 10 min at 1300 rpm (300 g). Pellets were incubated for 15 min at room temperature with 3 ml lysis buffer and then centrifuged at room temperature for 10 min at 1300 rpm. Cell pellets were then washed with 1× PBS to remove traces of lysis buffer and broken erythrocytes.

Hematopoietic cell types were defined according to specific surface markers, as detailed in the supplementary Materials and Methods and Table S5.

Bone marrow cells were harvested from CD45.1 wild-type and CD45.2 donor mice 3-4 days after 5-fluorouracil (5-FU; Sigma) injection. To perform competitive bone marrow transplantation, 1×10⁶ cells from CD45.1 wild-type mice were mixed 1:1 with cells from either control (Mb1-Cre;+/+) or Mb1-Cre; Setdb1^flox/− (Setdb1^Mb1) mice. The mixture was transplanted into lethally irradiated (9 Gy) wild-type mice through tail vein injection. CD45.1 and CD45.2 surface markers were used to discriminate between the donors. Bone marrow and spleen from recipients were analyzed by flow cytometry 7 to 9 weeks after transplantation.

Spleen from 6- to 8-week-old animals was fixed overnight in 4% formaldehyde and embedded in paraffin. Spleen sections were then stained with Hematoxylin and Eosin.

To test B cell differentiation, lineage-depleted or whole bone marrow cells were treated with RBC lysis buffer and were seeded (1×10⁵ cells/ml) in duplicate in 60 mm dishes on MethoCult M3630 (STEMCELL Technologies) containing IL7 according to the manufacturer's protocol. Cells were grown at 37°C in 5% CO₂ and checked each day to monitor colony formation. After 10-12 days of culture, colonies were scored and subsequently analyzed by flow cytometry using B220 and CD19 markers in combination with the viability dye 7AAD (eBioscience).

For short-term culture of B cell progenitors, lineage-negative bone marrow cells were enriched via magnetic sorting using the Lineage Cell Depletion Kit (Miltenyi). 5×10⁵ to 1×10⁶ cells were seeded in 24-well plates together with OP9 stromal cell in the presence of IL7 (10 ng/ml). Every other day, cells were split onto a fresh OP9 cell layer with IL7-supplemented medium as previously described (Holmes and Zuniga-Pflucker, 2009). At day 10 of co-culture, cells were harvested and analyzed by flow cytometry for pre-B cell differentiation using the Hardy scheme (Hardy et al., 1991).

Annexin V staining was performed using the Annexin V Apoptosis Detection Kit (eBioscience). 1-2×10⁶ bone marrow cells were prestained for pro-B cell markers. Next, cells were stained with annexin V according to the kit protocol. To remove unbound annexin V, cells were washed with 1-2 ml annexin V buffer and immediately analyzed by FACS.

Genomic DNA from FACS-sorted bone marrow cells was analyzed for VDJ recombination as described in the supplementary Materials and Methods.

Genomic DNA from control and Setdb1^Mb1 pro-B cells was subjected to bisulfite conversion and sequencing as described in the supplementary Materials and Methods and Table S6.

γH2A.X foci were enumerated in bone marrow cells as described in the supplementary Materials and Methods.

Pro-B cells were sorted with either MoFlo or FACSAria III. For gene expression analyses in Setdb1^Mb1 pro-B cells, RNA was isolated using the RNeasy Plus Kit (Qiagen). Genomic DNA retained by the gDNA columns was purified and used to test the Setdb1 deletion rate.

Alternatively, to detect gene expression changes in Setdb1^Mb1; Bcl2 pro-B cells, mRNA was isolated using Trizol or Direct-zol RNA MiniPrep (Zymo Research). To remove genomic DNA contamination, samples were treated with DNase I (Roche). cDNA was synthesized using random hexamer primers. Ct values were obtained by performing qPCR using SYBR Green dye. Differences in gene expression were calculated either as fold change using the 2^−ΔΔCT algorithm or as relative expression to housekeeping genes.

Pro-B cells were sorted using MoFlo and RNA was isolated using the RNeasy Plus Kit (Qiagen). Library preparation was performed as previously described (Schwickert et al., 2014). For further details, see the supplementary Materials and Methods.

Short-term cultured Rag2^−/− pro-B cells were obtained in vitro by cultivating them together with OP9 stromal cells in the presence of IL7. After 5-6 days of expansion, CD19⁺ B cells were enriched by magnetic sorting and processed to precipitate chromatin using antibodies against Setdb1, H3K9me3 and H3K9ac according to Schwickert et al. (2014). For further details, see the supplementary Materials and Methods, Tables S4 and S7.

Statistical differences between control and mutant groups were determined by unpaired or paired t-tests. P<0.05 was considered significant. Graphs were prepared and statistical tests were performed using R. Bar charts show mean values; error bars denote s.d.

The coding sequence of the envelope protein encoded by MLV1 was PCR amplified using primer pair (5′-3′) ATGGAAGGTCCAGCGTTCT and ACCAAGAACAAACCCCAGCT. The amplicon was gel purified and used as template for a second amplification with primers ATGGAAGGTCCAGCGTTCT and TTATTCACGCGATTCTACTTCT. The resulting fragment was cloned into pLenti6 vector. Detection of MLV Env protein is described in the supplementary Materials and Methods.

A validated Bcl2l11-specific shRNA sequence (TRC shRNA library, Sigma Aldrich) was integrated into the pLKO1 vector using sequence 5′-CGCGTCCGGGACGAGTTCAACGAAACTTACCTCGAGGTAAGTTTCGTTGAACTCGTCTTTTTGGAAATTAC-3′. The sequence of the scrambled shRNA is 5′-CGCGTCCGGCAACAAGATGAAGAGCACCAACTCGAGTTGGTGCTCTTCATCTTGTTGTTTTTGGAAA-3′.

For overexpression and shRNA knockdown, lentiviral particles were produced according to Sadic et al. (2015) and used to transduce 3-4×10⁵ hematopoietic progenitors (lineage negative) prestimulated for 2 days with IL7 (10 ng/ml). Progenitors transduced with viruses carrying shRNA were kept in culture for 24 h before seeding in MethoCult M3630 containing 0.5 µg/ml puromycin. For overexpression assays, transduced cells were kept in culture for 3 days in RPMI supplemented with IL7 (10 ng/ml) before FACS analysis.

For details of the associated lentiviral packaging, overexpression and shRNA plasmid constructs, see Table S8.

We thank Frank Malik for generously providing the 83A25 (Env) antibody and Marc Schmidt-Supprian for sharing the Mb1-Cre mice.