Spt5 accumulation at variable genes distinguishes somatic hypermutation in germinal center B cells from ex vivo–activated cells

To study mutation in V regions on the Igh locus, we used two independent knock-in mice that contained a rearranged V–diversity (D)–joining (J) gene on both alleles: the V_H186.2 gene from the J558 V_H family rearranged to D and J_H2 segments, and cloned into the J_H4 intron (B1-8^hi mice; Shih et al., 2002); and the VGK7 gene from the VGAM3.8 V_H family rearranged to D and J_H2 segments, and cloned into the J_H4 intron (8D10-GL mice, this work). For germinal center cells, mice were immunized with phycoerythrin, an antigen which has broad specificity for many V genes—including V_H186.2 (Pape et al., 2011)—and GL7⁺ splenic B cells were isolated on day 7. For ex vivo activation, naive spleen cells from B1-8^hi mice were stimulated with LPS and IL-4 for 2–5 d in culture. We first determined the level of expression of AID in cells under both conditions of activation. AID mRNA was measured by qPCR relative to 18S ribosomal RNA; there was fivefold more AID expressed after ex vivo activation compared with germinal center activation (Fig. 1 A). Thus, the lack of mutation in cultured cells is not due to deficient AID expression.

We then measured mutation frequencies in V regions by sequencing ∼600 bp around the VDJ exon (Fig. 1 B) in cells from both types of activation. For the B1-8 gene, we used B1-8^hi Ung^−/− cells, which are compromised for base excision repair due to the lack of uracil DNA glycosylase (UNG), to increase the mutation frequency. In germinal center cells 7 d after immunization, both B1-8^hi Ung^−/− and 8D10-GL mice had significant levels of mutation compared with B1-8^hi Aid^−/− control mice (Fig. 1 C). In ex vivo cells 5 d after LPS and IL-4 activation, B1-8^hi Ung^−/− cells had no significant increase in mutation compared with B1-8^hi Aid^−/− cells (Fig. 1 D), which confirms previous reports (Manser, 1987; Reina-San-Martin et al., 2003). Thus, hypermutation of V genes is robust in germinal center cells but debilitated in cells stimulated in culture. Other attempts to establish ex vivo systems to study somatic hypermutation have been elusive as well (McHeyzer-Williams et al., 1991; Decker et al., 1995; Källberg et al., 1996; Nojima et al., 2011; Matthews et al., 2014).

We then analyzed RNA pol II activity by nuclear run-on experiments—which measure polymerase activity in specific regions—in B1-8^hi cells activated ex vivo for 2 d. Transcription in isolated nuclei was initiated in vitro, and transcripts were hybridized to 500-bp DNA probes spanning a distance of 1.5 kb from the TATA box to the intron enhancer (Eμ; Fig. 2 A, green probes V1-5), and a control probe in Cμ1. Data were corrected for dTTP content between the different probes because radiolabeled UTP was incorporated. Compared with the C probe, polymerase loading was significantly higher over the V2, V3, and V4 probes located in the VDJ exon and J_H intron in day 2 cells relative to day 0 (Fig. 2 B). To examine if the increase is a common feature of highly transcribed genes, the γ-actin gene was also analyzed with probes similarly located downstream of the TATA box (Fig. 2 A, green probes G1-5). In contrast to the V region, there was no increase in the γ-actin gene on day 2 and no difference in levels between the different probes (Fig. 2 B). For a negative control, a probe located in the promoter region of the nontranscribed CD3 gene showed no transcription. Thus, pol II uniquely accumulated over V regions in ex vivo–activated B cells.

To determine whether pol II also accumulated in V regions from germinal center cells, we performed chromatin immunoprecipitation (ChIP) assays. For B1-8^hi and 8D10-GL mice, V and I primers were located 0.6 and 1.2 kb downstream of the TATA box, and E and C primers were placed 1.6 and 6.4 kb downstream in the Eμ and Cμ regions, respectively (Fig. 2 A, red primers). For the γ-actin locus, primer a was located in the TATA box, and primers b and c were placed 0.6 and 1.6 kb downstream of the TATA box. For a negative control, a primer was located over the TATA box of the nontranscribed β-globin gene. As shown in Fig. 2 C, total pol II levels from germinal center and ex vivo cells were significantly higher over the V and I primers compared with primers for E, C, γ-actin, and β-globin from both B1-8^hi and 8D10-GL mice. Pol II levels differed slightly at probes and primers when comparing run-on and ChIP data, which is likely due to the different techniques. For run-on, signals were received directly at the probe location. For ChIP, signals were obtained from large shear sizes of DNA, which could occur ∼500 bp on either side of the primer location. In summary, both germinal center and ex vivo activation produced significant accumulation of pol II for an extended distance of 1.2 kb from the TATA box.

To see if the increased pol II density exposed more single-strand DNA, which is the substrate for AID deamination, we measured its frequency in germinal center cells. To eliminate AID-dependent background deaminations, spleens from B1-8^hi Aid^−/− mice were used 7 d after immunization with phycoerythrin. Nuclei were isolated and treated with sodium bisulfite, which deaminates single-stranded cytosine to uracil, similar to AID. The products were sequenced for C:G to T:A transitions to distinguish deamination on the nontranscribed (C to T) and transcribed (G to A) strands in three 700-bp segments of DNA (Fig. 3 A). As seen in Fig. 3 B (top half), activated cells had substantially increased single-strand DNA compared with naive cells (dotted lines). Because pol II could accumulate on the transcribed strand, single-strand cytosines might be more frequent on the nontranscribed strand. However, the frequency was similar on both DNA strands, which indicates that both strands are accessible during transcription for deamination by bisulfite, and presumably AID. It has been suggested that the RNA exosome removes nascent RNA from the transcribed strand in S regions (Basu et al., 2011), and this may occur in V regions as well (Milstein et al., 1998; Rada et al., 2004; Ronai et al., 2007; Parsa et al., 2012). To compare frequencies, the deamination events were grouped into V, I, and E segments and corrected for the nucleotide composition of C:G bases. The single-strand DNA frequency was significantly higher in the V and I segments compared with the E segment (Fig. 3 C). A similar analysis of single-strand DNA from ex vivo–activated cells also revealed a higher frequency in V and I regions compared with E (unpublished data).

To confirm that somatic hypermutation does indeed mirror RNA pol II accumulation and single-strand DNA density, we collected extensive data from sequencing the B1-8^hi region from germinal center B cells from Peyer’s patches. As shown in Fig. 3 B (bottom half) and Fig. 3 D, the frequency of somatic hypermutation was significantly higher in the V and I segments compared with E. Although the E segment was not associated with many mutations or RNA pol II accumulation, the single-strand DNA detected there may be due to activity of the intronic promoter (Lennon and Perry, 1985). Collectively, elevated pol II levels correlate with increased single-strand DNA and mutations in the VDJ and J_H intron regions in germinal center cells. However, the lack of mutations after ex vivo activation indicates that pol II density and single-strand DNA alone are not sufficient to generate mutations.

To identify what is missing in cultured cells, we first characterized the type of RNA pol II accumulation. Initiating polymerases are marked by having the C-terminal domain phosphorylated at Ser5 (Ser5P; Buratowski, 2009). Using an antibody for Ser5P, ChIP analysis showed that this form of pol II was abundant at the V and I primers in both B1-8^hi and 8D10-GL germinal center B cells but not in ex vivo–activated cells (Fig. 4, A and B). Furthermore, Spt5, which is a member of the DSIF complex associated with the initiating form of pol II, was associated with the V and I primers in germinal center cells but was missing in ex vivo cells (Fig. 4 C). Concomitantly, AID was abundant at the V and I locations in germinal center cells but was nonexistent in ex vivo cells (Fig. 4 D). For controls, the transcribed γ-actin gene had significant amounts of the Ser5P form of pol II and Spt5 at primer a over the promoter in both germinal center and ex vivo cells, whereas the promoter of the nontranscribed β-globin gene was negative for these two proteins. AID was not present in the off-target γ-actin and β-globin genes. Thus, there is good correlation between the Ser5P form of RNA pol II, Spt5, and AID proteins bound to V and I sequences after in vivo immunization. These proteins are totally absent in ex vivo–stimulated cells.

Proteins bound to the Sμ region have been well-characterized in ex vivo–stimulated cells, but little is known about their participation in germinal center cells. Because the Sμ region undergoes somatic hypermutation after both types of activation, we determined if recruitment of transcription and AID proteins was identical by ChIP. As shown in Fig. 5 A, there is an intronic promoter located downstream of the Eμ enhancer that is constitutively activated and initiates transcription through the Sμ region. We used E, S, and C primers located before the promoter, in the 5′ Sμ region, and in Cμ, respectively, to measure proteins in germinal center– and ex vivo–activated cells from B1-8^hi Ung^−/− mice. In Fig. 5 B, pol II was significantly accumulated at the S primer, and it was the initiating form characterized by Ser5P. Furthermore, Spt5 and AID were significantly associated with the stalled polymerases in both germinal center and ex vivo cells. These results confirm previous reports of ex vivo transcription, which is defined by stalled polymerases (Rajagopal et al., 2009; Wang et al., 2009), Ser5P (Wang et al., 2009), Spt5 (Pavri et al., 2010), and AID (Pavri et al., 2010) and extend the findings to germinal center transcription. Because these proteins are down-regulated in Eμ, the results suggest that they are brought in during de novo transcription from the intronic promoter and become amassed during the stalled migration of pol II through R-loop structures. All these markers of transcription and mutation are made manifest by inducing somatic hypermutation in the Sμ region from germinal center cells (Fig. 5 C) and ex vivo cells (Fig. 5 D). Therefore, the Sμ region is accessible for mutation under both conditions of cell activation, perhaps because of the constitutively active intronic promoter and unique DNA sequence.

The 3′RR on the heavy chain locus is implicated in controlling transcription, somatic hypermutation, and class switch recombination in mature B cells (Vincent-Fabert et al., 2010; Rouaud et al., 2013). To determine if it affects the recruitment of Spt5 and AID to the V region in germinal center cells, we examined mice which have a 30-kb deletion of the 3′RR (Vincent-Fabert et al., 2010). Mice were immunized with phycoerythrin for 7 d, and GL7⁺ B cells were isolated from spleens for ChIP assays. Because the mice contain a heterogeneous repertoire of VDJ rearrangements, we focused on the J_H4 intron, with the caveat that only 25% of the cells may contain rearrangements to J_H4. To see if transcription could be measured in these heterogeneous cells, RNA pol II ChIPs were performed using J_H4 and E primers (Fig. 6 A). The results in Fig. 6 B showed that in WT cells, pol II was reduced fourfold compared with similar regions in the knock-in B1-8 mice (Fig. 2 C), confirming that WT mice had fewer rearrangements to J_H4. In 3′RR cells, pol II molecules were drastically reduced 80% in both J_H4 and E regions, compared with WT cells. Pol II binding in the mutant mice was unaffected in primers located at the TATA box in the γ-actin gene, and the β-globin gene was a negative control. In accord with the ChIPs from knock-in mice, Spt5 was present in J_H4 from WT cells, albeit fourfold lower than from B1-8^hi cells, but not in the E region, suggesting that the initiating form of pol II is retained over J_H4. However, Spt5 was absent over J_H4 in 3′RR cells. As controls, Spt5 was bound to primers located in the promoter of the γ-actin gene from both types of mice. In confirmation of a previous study (Rouaud et al., 2013), the AID protein was also missing in the J_H4 region from 3′RR mice. Therefore, in the absence of the 3′RR, transcription was markedly reduced, and both Spt5 and AID proteins were not recruited to the J_H4 region.

V genes have long been enigmatic in efforts to understand how they are substrates for AID activity during transcription. In this first report of AID targeting in germinal center B cells, we examined features of transcription that are unique to V genes, which provide insight into their ability to sustain high levels of somatic hypermutation. Two distinct events occur after transcription.

First, RNA polymerases were paused for a long distance downstream of the TATA box. Using both nuclear run-on and ChIP techniques, pausing was detected over 1.2 kb encompassing the VDJ exon and J_H intron in B cells activated either ex vivo or in vivo. In contrast, there was no pausing in the transcribed γ-actin or CD19 genes (unpublished data) from these cells. Prolonged pausing in the V region could be induced during co-transcriptional splicing of the leader and V exons. For example, the transcriptional machinery may be sluggish because of multiple splicing reactions at the leader and rearranged V exons on the heavy and light chain loci, and slow-moving transcription bubbles would increase pol II density. It is notable that CTNNBL1 and PTBP2 factors associated with the spliceosome also bind to AID (Conticello et al., 2008; Nowak et al., 2011). Pol II accumulation could also be due to cis DNA sequences in the J_H introns, which are shared by all V gene rearrangements. Intron sequences have been suggested to form secondary structures during transcription (Golding et al., 1987; Wright et al., 2008), which may impact histone assembly and polymerase migration. Both these predictions can be tested by modifying DNA sequences of rearranged genes in cell lines to see how they affect pol II accumulation. Furthermore, when pol II progression was blocked in DT40 and Ramos cell lines, mutations were increased (Kodgire et al., 2013; Wang et al., 2014). In summary, pausing for more than 1 kb after the promoter is an intrinsic feature of rearranged V genes and can occur regardless of the type of cell stimulation. Paused polymerases then establish the boundary of mutation by exposing single-strand DNA as the substrate for AID in cells from either germinal centers or ex vivo culture. However, although pol II density and single-strand DNA are necessary for somatic hypermutation, they are not sufficient because the V and I segments from ex vivo activation have virtually no mutations.

Second, although RNA pol II density profiles are similar between germinal center and ex vivo cells, the initiating form of pol II characterized by Ser5 phosphorylation in the C-terminal domain is retained only in germinal center cells. Typically, the conversion to transcriptional elongation occurs within 100 bp after the start of transcription, followed by a rapid loss of Ser5P (Brookes et al., 2012), which we detected in the control γ-actin gene. Thus, it is notable that the Ser5P form of pol II was elevated in the V and I primers located 0.6 and 1.2 kb, respectively, downstream from the TATA box preceding the VDJ gene. It is not known what causes the delay from initiation to elongation, but the process catalyzes the recruitment of Spt5 and AID to V genes. Our ChIP data conclusively show that Spt5 and AID are present at the V and I primers in germinal center cells but are totally absent in ex vivo cells. Thus, there is a strong correlation between Ser5P, Spt5, AID, and mutations, as summarized in Fig. 7. The data also indicate that pol II pausing occurs independently of Spt5 recruitment in ex vivo cells and may precede it in germinal center cells. We then examined whether the 3′ RR enhancers affect Spt5 and AID accumulation using mice with a 30-kb deletion of the region. Pol II density in germinal center cells was decreased 80% in the J_H4 and Eμ regions, and Spt5 and AID were nonexistent, explaining the lack of somatic hypermutation in these mice (Rouaud et al., 2013). However, a definitive role for the 3′RR is elusive because transcription was greatly reduced in the knockout mice. Perhaps sub-enhancers—such as hs3a, hs1-2, hs3b, and hs4—within this region may be differentially used by WT cells during activation in vivo compared with ex vivo.

The other targets of somatic hypermutation are the S regions on the Igh locus. These regions are a sponge for AID due to the draconian R-loop structures which produce a pile-up of RNA pol II molecules. Wang et al. (2009) have reported that the Ser5P form is also enriched in S regions located 3 kb from transcription start sites in introns. This suggests that retention of the initiating form for long distances in the V and S regions is quite different from the promoter-proximal pausing described in other genes (Kwak and Lis, 2013). Paused initiating polymerases can then more efficiently recruit Spt5, AID, and other factors involved in hypermutation (Willmann et al., 2012; Aida et al., 2013; Singh et al., 2013).

Genome-wide studies of ex vivo–stimulated cells have shown that AID localization mirrors RNA pol II and Spt5 density (Pavri et al., 2010; Yamane et al., 2011). Because RNA pol II and Spt5 are present at most promoters, as many as 6,000 genes could be potential targets of AID (Yamane et al., 2011). We demonstrate here that transcribed V genes are not a target for Spt5 or AID in ex vivo cells and suggest that the number of potential targets may be very different in germinal center B cells. Furthermore, not all genes are mutated equally (Liu et al., 2008; Yamane et al., 2011), which indicates that additional factors regulate AID activity. For example, the mutated c-myc gene has a specific RNA pol II pause site in close proximity to the area of mutations (Krumm et al., 1992), and the murine Bcl6 gene has a significant number of nucleotide repeats near the site of mutations (Liu et al., 2008). Therefore, the sequences of these genes may cause transcriptional perturbation, followed by subsequent pol II accumulation and AID activity. In contrast, we show that the highly transcribed γ-actin gene has high levels of the Ser5P form of pol II and Spt5 protein at the promoter region, but no increased accumulation of pol II in the gene body and no detectable AID localization. Thus, a model evolves where AID may be present at open promoters, but the ability to access downstream DNA is regulated by the combined effects of RNA pol II density, DNA sequence, and AID cofactors.

What causes RNA pol II to retain its Ser5P initiating form for a long distance over the V gene and downstream intronic DNA is not known. Conversion to the elongating form in germinal center cells could be blocked by differential usage of transcription factors from either the enhancers or promoter after activation. It has been proposed that the same enhancer may have different factors bound at various times (Pinaud et al., 2011), and the same gene may be controlled by different enhancers during various activation states (Kieffer-Kwon et al., 2013). Another question is what role Spt5 is playing during V gene transcription. The protein is generally associated with the DSIF complex which coordinates pol II promoter release (Kwak and Lis, 2013), or it could potentially be taking a larger role in regulating V region transcription similar to recent results for the A20 gene (Diamant et al., 2012).

In summary, transcription sets up the V locus as a single-strand substrate by augmenting paused RNA polymerases, perhaps due to the DNA sequence of the region. Proteins associated with hypermutation, such as Spt5 and AID, are recruited to the initiating form of pol II specifically in cells activated in germinal centers. Differential activation programs in B cells induced in germinal centers versus ex vivo may produce an altered landscape of transcription factors that favors initiating polymerases and the cascade of mutation proteins.

B1-8^hi knock-in mice (Shih et al., 2002) were obtained from Rafael Casellas (National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health). B1-8^hi Ung^−/− and B1-8^hi Aid^−/− mice were generated at the National Institute on Aging. 8D10-GL mice were made at Brandeis and Harvard Universities. 8D10-GL expresses the unmutated version of the mutated heavy chain (VGAM3.8 family VGK7 germline gene rearranged to D and J_H2) from 8D10, a hybridoma derived from the BALB.B memory response to (T,G)-A-L antigen (Press and Giorgetti, 1993). The 8D10 heavy chain rearrangement was cloned; a fragment containing VDJ_H2 but lacking J_H3 and J_H4 was obtained; and V_H codon #92 was changed from TGT to TGC to eliminate a cryptic heptamer site. To construct 8D10-GL, a 123-bp BglII–NdeI fragment (containing the mutated V_H region) was removed from 8D10 VDJ_H2 and replaced with the corresponding piece from the VGK7 germline gene using a genomic phage clone (VFM1.5) from M. Taussig (Babraham Institute, Cambridge, UK; Sims et al., 1992). The pIVHL2neo^r vector was provided by K. Rajewsky (Harvard University, Cambridge, MA) and contained an HSV-tk cassette (∼1.8 kb), a 5′ homology region (∼9 kb of BALB/c germline DNA, a BamHI–XhoI fragment 5′ of DQ52), a neomycin resistance gene (neo^r, ∼1.3 kb) flanked by two loxP sites, and a 3′ homology region (∼0.8 kb of germline DNA from 129/Ola, the NaeI-PvuII fragment upstream of the Eμ enhancer). The 8D10-GL fragment, containing ∼1.5 kb of flank, was cloned into this vector after extending the vector’s 3′ homology region from 0.8 to 2.5 kb (Sonoda et al., 1997). TC1 embryo stem (ES) cells electroporated with recombinant vector were selected with both G418 and gancyclovir. To screen for recombinant clones, Southern blots of restriction enzyme–digested genomic DNA from ES clones were hybridized with a XbaI-BglII piece from an ∼870-bp XbaI–BamHI genomic fragment beginning 5′ of Cμ (Bottaro et al., 1998; Sakai et al., 1999). Recombinant clones, still containing the neo^r cassette, were injected by Laurie Davidson (Harvard University) into C57BL/6 blastocysts to produce chimeric mice. Male chimeras were mated to C57BL/6 or 129SvEv (129S6) females (Taconic), and the F1 progeny were screened by Southern blot analysis of BamHI-digested tail DNA for the presence of the knock-in allele and correct targeting of the locus. Mice containing the knock-in allele were backcrossed onto the 129S6 background and were bred to EIIa-Cre transgenic mice on the 129S6 background (Lakso et al., 1996) to delete the loxP-flanked neo^r cassette in vivo. The 8D10-GL heterozygous strain was backcrossed 9–10 generations onto C57BL/6, and intercrossed to produce 8D10-GL homozygous mice. 3′RR-deficient mice were previously described (Vincent-Fabert et al., 2010). WT C57BL/6 mice were purchased from Charles River.

For immunizations, mice were injected intraperitoneally with 100 µg phycoerythrin (Prozyme) in complete Freund’s adjuvant (Sigma-Aldrich) and used 7 d later. All animal procedures were reviewed and approved by the Animal Care and Use Committees of the National Institute on Aging and Brandeis University.

Naive splenic B cells were isolated by negative selection with anti-CD43 and anti-CD11b magnetic beads (Miltenyi Biotec). Cells were plated at a density of 0.5 million cells/ml and stimulated ex vivo with 5 µg/ml Escherichia coli LPS serotype 0111:B4 (Sigma-Aldrich) and 5 ng/ml recombinant IL-4 (BioLegend). Germinal center B cells from immunized mice were isolated by staining with FITC-conjugated anti-B220 (SouthernBiotech) and Alexa Fluor 647–labeled anti-GL7 (eBioscience), and B220⁺GL7⁺ cells were isolated by flow cytometry.

RNA was extracted from 0.5 million germinal center– or ex vivo–activated B cells using TRIzol reagent (Invitrogen), followed by further purification with RNeasy Mini kit (QIAGEN). cDNA was generated using Superscript III (Invitrogen), followed by qPCR using Power SYBR Green PCR Master Mix (Life Technologies) and the primers listed in Table S1.

B1-8^hi Ung^−/−, B1-8^hi Aid^−/−, and 8D10-GL mice were used to detect mutations. Genomic DNA in cells from day 7 immunized germinal centers, day 5 ex vivo activation, and Peyer’s patches (4 mo) was isolated and amplified using Herculase II enzyme (Agilent Technologies) with primers listed in Table S1. Ex vivo cells were stimulated in 24 independent wells and analyzed separately to minimize clonal amplification. Frequencies were calculated by dividing the number of mutations by the total number of nucleotides sequenced.

Naive splenic B cells were isolated from B1-8^hi Ung^−/− mice and stimulated with LPS and IL-4. Nuclear run-on experiments were performed on days 0 and 2 as described previously (Rajagopal et al., 2009). Primers for 500-bp probes used in hybridization are listed in Table S1.

Experiments on ex vivo– and germinal center–activated B cells were performed using a μChIP protocol (Dahl and Collas, 2008). In brief, 0.25 to 0.5 million cells were cross-linked with 1% formaldehyde for 10 min at room temperature. Cross-linked cells were lysed, and cellular DNA was sheared to a mean size of 500 bp by sonication, followed by incubation with antibody/Dynabead complexes (Invitrogen). Antibodies against pol II (Millipore, clone CTD4H8), pol II CTD-Ser5 (Abcam), SPT5 (clone H-300; Santa Cruz Biotechnology, Inc.), AID (Kohli et al., 2010), and nonspecific rabbit IgG (Millipore) were used at 2.5 µg per reaction. Primers for quantitative PCR are listed in Table S1. The V primer set was positioned at the end of J_H2 to avoid amplification of unrearranged members from the vast J558 V_H family. Calculation of % input was performed by deriving the % input value for the experimental antibody (2^{Ct(input)−Ct(ChIP)} × 100) and subtracting the % input value of the IgG control.

Sodium bisulfite experiments were performed as previously described (Ronai et al., 2007) on 1–2 × 10⁶ naive or germinal center cells from B1-8^hi Aid^−/− mice taken 7 d after immunization with phycoerythrin. DNA was amplified using Taq DNA polymerase (Takara Bio Inc.) with primers listed in Table S1 and sequenced. Analysis of single-strand DNA frequency for individual nucleotides was done by dividing the number of deaminated C:Gs by the total number of C:Gs sequenced.

Table S1 lists sequences of primers used for AID expression, DNA sequencing, nuclear run-on, and ChIP assays.

We thank William Yang for excellent technical support; Fred Alt, Wasif Khan, and Laurie Davidson for assistance in generating the 8D10-GL mice; and R. Wersto, J. Scheers, C. Nguyen, and T. Wallace for flow cytometry analyses.

This research was supported in part by the Intramural Research Program of the National Institutes of Health (NIH), National Institute on Aging, and by grants from the NIH (AI13725 to J.L. Press) and the American Cancer Society (L. Venkataraman).

The authors declare no competing financial interests.