The microanatomic segregation of selection by apoptosis in the germinal center

Light- and dark-zone death dynamics Germinal centers (GCs) are areas within lymphoid organs where mature B cells expand and differentiate during normal immune responses. GCs are separated into two anatomic compartments: the dark zone, where B cells divide and undergo somatic hypermutation, and the light zone, where they are selected for affinity-enhancing mutations after interacting with T follicular helper cells. Mayer et al. studied apoptosis reporter mice and found that both GC zones experience very high rates of apoptosis (see the Perspective by Bryant and Hodgkin). However, the underlying mechanisms were distinct and microanatomically segregated. Light-zo ne B cells underwent apoptosis by default unless they were rescued by positive selection. In contrast, apoptotic dark-zone B cells were highly enriched among cells with genes damaged by random antibody-gene mutations. Science, this issue p. eaao2602; see also p. 171 The selection of germinal center B cells by apoptosis is regulated by microanatomically distinct mechanisms. INTRODUCTION Germinal centers (GCs) are transient microanatomic structures that form in lymphoid organs during an immune response. They are the sites of B cell clonal expansion and affinity maturation, a process that leads to the production of high-affinity antibodies. GCs are highly dynamic and contain activated B cells, specialized T follicular helper (TFH) cells, and antigen-trapping follicular dendritic cells. GCs are organized into two functionally distinct compartments: a dark zone (DZ) and a light zone (LZ). The DZ is the site of rapid cell division and random antibody-gene mutation, which is initiated by activation-induced cytidine deaminase (AID). The mutation process leads to the accumulation of a large number of closely related B cells that carry receptors with distinct antigen-binding properties. Once they stop dividing, DZ B cells migrate to the LZ, where their newly generated B cell receptors (BCRs) are tested: GC B cells with relatively higher-affinity receptors capture and process more antigen, leading to positive selection by interaction with TFH cells. The positively selected LZ B cells return to the DZ, where they undergo further cycles of division and mutation. Concomitantly, small numbers of memory B cells and antibody-secreting plasma cells exit the GC. Together, these processes provide the mechanistic basis for affinity maturation, which is essential for effective vaccination and protection from infections. RATIONALE In addition to producing antibody variants, AID expression is also a threat to the genome. AID can produce double-strand breaks that are substrates for chromosome translocations. It can also produce immunoglobulin (Ig) gene missense mutations and deletions or create self-reactive antibodies. These deleterious mutations need to be selected against. Indeed, histologists have long appreciated large numbers of apoptotic nuclei in the specialized tingible body macrophages found in GCs. However, beyond histology, little is known about the exact rate of GC B cell apoptosis and whether it differs in the DZ and LZ of the GC. Moreover, the mechanisms that cause apoptosis, their relative importance in each GC compartment, and their role in GC B cell selection have not been defined. To examine these questions, we created fluorescent apoptosis-indicator mice and used them to enumerate, isolate, and characterize dying cells in the GC. RESULTS We found that apoptosis is prevalent in both the DZ and LZ compartments of GCs throughout the immune response: up to 50% of GC B cells undergo programmed cell death every 6 hours. Single dying GC B cells were isolated, and their antibody genes were cloned, expressed by transient transfection, and tested for antigen binding and other properties. Apoptotic DZ cells were highly enriched for Ig genes damaged by AID, including missense mutations and deletions. By contrast, dying LZ cells primarily expressed intact antibodies with a range of affinities indistinguishable from GC B cells in the live LZ compartment. By experimentally blocking positive selection and by using reporter mice for Myc, a proto-oncogene, as an indicator of positive selection, we found that apoptosis is the default fate for LZ GC B cells that are not actively positively selected. Thus, LZ GC B cells carrying low-affinity BCRs do not preferentially undergo apoptosis. Instead, apoptosis occurs irrespective of BCR-affinity, and LZ B cells carrying high-affinity BCRs are simply more likely to be positively selected. CONCLUSION Apoptosis is a major feature of GC B cell biology and is required to counterbalance the high rate of proliferation and purge B cells that carry deleterious mutations. Although apoptosis occurs in both the DZ and LZ, the underlying mechanisms of apoptosis in each zone are distinct and microanatomically segregated. These insights into GC B biology are relevant for vaccine design, particularly for pathogens that normally evade effective antibody responses. Germinal center B cells expressing an apoptosis indicator. Intravital two-photon microscopy of GC B cells in popliteal lymph nodes of immunized mice (GC B cells, yellow and green; follicular dendritic cell networks, red). The fluorescence resonance energy transfer–based INDIA reporter was used to visualize and purify dying GC B cells. B cells undergo rapid cell division and affinity maturation in anatomically distinct sites in lymphoid organs called germinal centers (GCs). Homeostasis is maintained in part by B cell apoptosis. However, the precise contribution of apoptosis to GC biology and selection is not well defined. We developed apoptosis-indicator mice and used them to visualize, purify, and characterize dying GC B cells. Apoptosis is prevalent in the GC, with up to half of all GC B cells dying every 6 hours. Moreover, programmed cell death is differentially regulated in the light zone and the dark zone: Light-zone B cells die by default if they are not positively selected, whereas dark-zone cells die when their antigen receptors are damaged by activation-induced cytidine deaminase.

LZ, where their newly generated B cell receptors (BCRs) are tested: GC B cells with relatively higher-affinity receptors capture and process more antigen, leading to positive selection by interaction with T FH cells. The positively selected LZ B cells return to the DZ, where they undergo further cycles of division and mutation. Concomitantly, small numbers of memory B cells and antibody-secreting plasma cells exit the GC. Together, these processes provide the mechanistic basis for affinity maturation, which is essential for effective vaccination and protection from infections. RATIONALE: In addition to producing antibody variants, AID expression is also a threat to the genome. AID can produce double-strand breaks that are substrates for chromosome translocations. It can also produce immuno-globulin (Ig) gene missense mutations and deletions or create self-reactive antibodies. These deleterious mutations need to be selected against. Indeed, histologists have long appreciated large numbers of apoptotic nuclei in the specialized tingible body macrophages found in GCs. However, beyond histology, little is known about the exact rate of GC B cell apoptosis and whether it differs in the DZ and LZ of the GC. Moreover, the mechanisms that cause apoptosis, their relative importance in each GC compartment, and their role in GC B cell selection have not been defined. To examine these questions, we created fluorescent apoptosis-indicator mice and used them to enumerate, isolate, and characterize dying cells in the GC. RESULTS: We found that apoptosis is prevalent in both the DZ and LZ compartments of GCs throughout the immune response: up to 50% of GC B cells undergo programmed cell death every 6 hours. Single dying GC B cells were isolated, and their antibody genes were cloned, expressed by transient transfection, and tested for antigen binding and other properties. Apoptotic DZ cells were highly enriched for Ig genes damaged by AID, including missense mutations and deletions. By contrast, dying LZ cells primarily expressed intact antibodies with a range of affinities indistinguishable from GC B cells in the live LZ compartment. By experimentally blocking positive selection and by using reporter mice for Myc, a proto-oncogene, as an indicator of positive selection, we found that apoptosis is the default fate for LZ GC B cells that are not actively positively selected. Thus, LZ GC B cells carrying low-affinity BCRs do not preferentially undergo apoptosis. Instead, apoptosis occurs irrespective of BCR-affinity, and LZ B cells carrying high-affinity BCRs are simply more likely to be positively selected.
CONCLUSION: Apoptosis is a major feature of GC B cell biology and is required to counterbalance the high rate of proliferation and purge B cells that carry deleterious mutations. Although apoptosis occurs in both the DZ and LZ, the underlying mechanisms of apoptosis in each zone are distinct and microanatomically segregated. These insights into GC B biology are relevant for vaccine design, particularly for pathogens that normally evade effective antibody responses. ▪ G erminal centers (GCs) are divided into two anatomic compartments: the light zone (LZ) and the dark zone (DZ). B cells divide and undergo somatic hypermutation (SHM) in the DZ and are positively selected for affinity-enhancing mutations by interacting with T follicular helper (T FH ) cells in the LZ (1)(2)(3). Cell division is a dominant feature of the GC, with rapid cell division rates of 4 to 6 hours and up to 30% of cells in cycle at any time (1,3,4). Despite extensive cell division, the size of the GC compartment can be relatively constant for weeks or months (5). Equilibrium is attributed to a combination of cell death by apoptosis (negative selection) and emigration of memory and plasma cells from the GC. Emigration rates are estimated to be relatively low [<0.1% for plasma cells (6) and <2% for memory cells (7)]. By contrast, cell loss by apoptosis is reported to be high (8,9), but the precise rate and causes of apoptosis, its contribution to GC B cell selection, and whether it is differentially regulated in the LZ and DZ of the GC has not been determined.
Analysis of apoptosis among DZ and LZ GC B cells revealed that 3.7 to 5.7% of the DZ and 2.6 to 5.6% of the LZ were aCasp3 + at all time points analyzed (Fig. 1, C and D). Similar results were also obtained for Peyer's patch GC B cells (fig. S1E). Thus, the frequency of apoptotic GC B cells is relatively constant over time and nearly equivalent in LZ and DZ compartments.
The size of GCs in Peyer's patches is relatively constant over time in mice housed under specific pathogen-free conditions, and thus, the number of dividing cells should equal the number of dying cells plus a small number that leave the GC to become memory B or plasma cells (6,7,11). To estimate the proportion of dividing cells in GCs, we performed kinetic labeling experiments with the nucleoside analog 5-ethynyl-2′-deoxyuridine (EdU), which is incorporated into DNA during the S phase of the cell cycle. About 50% of all Peyer's patch GC B cells were labeled by EdU in 5.3 hours (Fig. 1, E and F   . This reporter consists of an optimized fluorescence (Förster) resonance energy transfer (FRET) pair, mNeonGreen and mRuby2, linked by a peptide containing an aCasp3-cleavage site (DEVDG, Fig. 2A; D, aspartic acid; E, glutamic acid; V, valine; G, glycine). When aCasp3 cleaves the linker, mNeonGreen and mRuby2 should be separated, resulting in FRET loss and increased emission from mNeonGreen ( Fig. 2A) S2C). To validate the reporter, transgenic B cells were activated in vitro and induced to undergo apoptosis by incubation with staurosporine (12). Flow cytometry revealed two distinct populations based on the mNeonGreen/FRET ratio (termed "FRET loss"; Fig. 2B, left). Whereas FRET + B cells were alive, FRET -B cells were apoptotic, as confirmed by aCasp3, TUNEL, or Annexin V-DAPI staining (TUNEL, terminal deoxynucleotidyl transferasemediated deoxyuridine triphosphate nick end labeling; DAPI, 4′,6-diamidino-2-phenylindole) ( Fig. 2B and fig. S2D).
To examine the kinetics of activated B cell death, we tracked FRET loss in real time in cultured Igk INDIA B cells ( Fig. 2C and fig. S2E). On average, the first morphological signs of apoptosis were observed within 12.5 min of FRET loss, including cell shrinkage, bleb formation, and changes in motility (Fig. 2, C and D, fig. S2E, and movies S1 to S3). Secondary necrosis, as revealed by loss of membrane integrity and leakage (Fig. 2C, fig.  S2E, and movies S1 to S3), was observed an average of 68 min after FRET loss (Fig. 2D). Similar results were obtained in vivo by tracking Rosa26 INDIA knock-in GC B cell death using twophoton laser scanning microscopy. GC B cell frag-mentation occurred on average 20.6 min after FRET loss and was observed in both DZ and LZ compartments (Fig. 2, E to G; Movies 1 to 3; and fig. S3, A and B). Although the precise relationship between caspase staining, FRET loss, and cell fragmentation has not been defined, it is clear that the apoptotic compartment in GCs turns over with rapid kinetics. At an apoptosis rate of 3% every 20.6 min (fig. S1, A and B), 46% of GC B cells in Peyer's patches are estimated to be lost in 5.3 hours, which agrees with our measurements made by EdU labeling (Fig. 1, E and F). Thus, apoptosis is a major feature of the B cell program in the GC.  creating base pair mismatches in DNA. The absence of AID in mice and humans is associated with enlarged GCs (13,14) and reduced GC B cell apoptosis, as measured by aCasp3 [ fig. S4, A to E, and (15)]. To determine whether AID differentially affects cell death in the two GC compartments, we stained AID-deficient DZ and LZ cells for aCasp3. The absence of AID was associated with a clear reduction in apoptosis primarily in the DZ ( fig. S4, F to H). Thus, AID activity is a key component of apoptosis in the DZ, and apoptosis appears to be differentially regulated in the DZ and LZ. AID introduces random mutations in Ig genes that can increase antibody affinity, but can also be deleterious. To determine how Ig mutation impacts apoptosis, we cloned antibodies from single FRET -Igk INDIA GC B cells that had started undergoing apoptosis ( Fig. 2H and fig. S5A). Thus, apoptotic DZ B cells are enriched for nonfunctional Ig transcripts as a consequence of AID activity. We hypothesize that the small number of apoptotic LZ GC B cells with nonfunctional BCRs derive from recent DZ emigrants in which a delay between aberrant Ig gene mutation, loss of BCR expression, and apoptosis has occurred, as has been reported for naïve B cells (18).

Characterization of monoclonal antibodies cloned from dying GC B cells
In addition to compromising the integrity of the BCR, SHM can also alter antibody affinity, produce autoreactive or polyreactive BCRs, or render Ig heavy and light chains incompatible. To measure the contribution of each of these effects to GC B cell apoptosis, we produced IgH V1-72 Igl a-NP antibodies cloned from GC B cells of NP-OVAimmunized Igk INDIA mice. The relative affinity to NP was measured by enzyme-linked immunosorbent assay (ELISA) on NP 4or NP 25 -bovine serum albumin (BSA). High-affinity antibodies    Affinity bind better to NP 4 -BSA than low-affinity antibodies, whereas both bind equally well to NP 25 -BSA [ Fig. 3A; (19)]. All but one of the 167 IgH V1-72 Igl GC-derived antibodies tested bound to NP 25 -BSA (Fig. 3, A and B). As might be expected from the observation that the DZ contains affinityselected B cells, high-affinity antibodies were slightly overrepresented in the DZ over LZ compartment (60 and 46%, respectively; P = 0.089, Fig. 3C). This small increase in high-affinity reactivity in the DZ was confirmed by increased proportions of the affinity-enhancing W33L mutation in the DZ (W, tryptophan; L, leucine) ( fig. S6, A and B). However, high-and low-affinity antibodies were equally distributed among live and apoptotic B cells in both GC compartments (Fig. 3, A and B, and fig. S6A). Thus, the observed differences in affinity between the LZ and DZ are likely due to rapid transit of positively selected cells from the LZ to the DZ (1), as well as a relatively longer dwell time of higheraffinity cells in the DZ (2), and not apoptosis.
To document autoreactivity, we performed HEp-2 ELISAs, which are used clinically to measure antibody binding to cytoplasmic and nuclear self-antigens. Only 8% of the IgH V1-72 Igl GC antibodies cloned from live B cells displayed autoreactivity ( Fig. 3D and fig. S6C). Consistent with the possibility that autoreactive antibodies can be redeemed by continued SHM (20,21), there were similar numbers of autoreactive cells in live and apoptotic compartments (Fig. 3D and fig. S6C). Thus, autoreactivity accounts for only a small fraction of the overall death in IgH V1-72 Igl NPspecific GC B cells, and it is not strongly selected into the apoptotic compartment.
Memory B cells show increased polyreactivity relative to naïve B cells (22)(23)(24)(25). In agreement with these observations, 21% of all live GC B cells were polyreactive, with a slightly higher abundance in the DZ (27%) over the LZ (16%) (Fig. 3E  and fig. S6D). Moreover, polyreactive cells were equally represented in the live and apoptotic compartments. Thus, polyreactivity is neither positively nor negatively selected into the apoptotic compartment in the GC.
Nearly all Igs cloned from live GC B cells produced secreted antibodies in transient transfection experiments. By contrast, a significant fraction of the antibodies derived from apoptotic IgH V1-72 Igl GC B cells did not, despite functional Ig genes (Fig. 3F). This phenomenon was particularly prominent among apoptotic DZ cells, where approximately half of the clones did not produce secreted antibodies in transient transfection experiments. Despite the lack of secreted Ig, immunoblot analysis of transfected cell pellets showed normal levels of IgH and Igl expression, suggesting that structural problems and/or defective Ig pairing interfered with normal antibody secretion ( fig. S7). Thus, a significant fraction of the apoptotic cells in NP-specific GCs express antibodies that are structurally compromised (28%), and this phenomenon is exclusive to DZ GC B cells. In summary, selection against SHMs that introduce nucleotide insertions or deletions, produce stop codons, change the reading frame, or otherwise compromise IgH V1-72 Igl expression or stability account for 75% of all the apoptotic cells in the DZ of NP-specific GCs (Fig. 3G).  live compartments (Fig. 3H and fig. S8B). Similar results were also obtained in Peyer's patches and in lymph nodes of NP-OVA-immunized AIDsufficient C57BL/6J mice ( fig. S8, C and D). Thus, IgG1 + GC B cells are more prone to apoptosis than IgM-expressing cells irrespective of AID expression or SHM. This effect may be due to altered IgG1 BCR signaling, as indicated by lower levels of nuclear receptor Nur77-GFP induction in IgG1 + when compared to IgM + GC B cells (22).

Apoptosis in the LZ due to the absence of positive selection
Our data indicate that in contrast to the DZ, 80% of apoptotic LZ cells express intact, nonautoreactive BCRs whose antigen-binding properties are similar to those found in the live compartment (Fig. 3, A, B, and G). The transition from the DZ to the LZ is associated with increased expression of genes that regulate apoptosis (1,16). To examine the possibility that apoptosis may be the default fate for cells that are not positively selected in the LZ, we inhibited positive selection by blocking CD40-CD40L interactions with a-CD40L antibody ( fig. S9A). Although GC size and the number of positively selected cells undergoing proliferation decreased, the rate of LZ B cell apoptosis remained similar [ Fig. 4A and (26)]. Absence of a measurable increase in apoptosis (26) in a-CD40L antibody-treated mice is consistent with the relatively small number of LZ B cells (5 to 10%) undergoing positive selection at any time (27).
To gain further insights into the mechanisms responsible for cell death in the LZ, we measured apoptosis in cells undergoing BCR signaling and positive selection using Nur77-GFP and Myc-GFP (myelocytomatosis oncogene) reporters, respectively (27)(28)(29). A fraction of LZ B cells expressed Nur77, but only a subset of these cells were protected from apoptosis (Fig. 4, B and C, and fig. S9, B and C). By contrast, positive selection, as indicated by Myc-GFP expression, was associated with nearly complete protection from apoptosis (Fig. 4, B and C, and fig. S9, B and C). Thus, positive selection, as measured by Myc expression, protects B cells from apoptosis, but BCR cross-linking alone, as measured by Nur77, appears to be insufficient.

Conclusions
We have investigated the causes of apoptosis and their contribution to cell death in the LZ and DZ of GCs by combining a FRET-indicator of apoptosis, single-cell sorting, and antibody cloning. Only a small fraction of LZ GC B cells are positively selected to return to the DZ and undergo additional rounds of division, and this process is, in part, stochastic (1,30). The element of chance appears to be introduced by random encounters between LZ B cells displaying high levels of peptide-major histocompatibility complex II and cognate T cells. Consistent with this notion, the apoptotic compartment in the LZ contained an arbitrary assortment of B cells, including those with high-affinity BCRs (Fig. 3B). Thus, both low-and high-affinity B cells undergo apoptosis in the LZ, but high-affinity cells are more likely to become positively selected after they encounter cognate T FH cells (1,31,32). By contrast, the DZ is the microanatomic site of antibody quality control by selection against deleterious mutations introduced by AID. B cells expressing Igs damaged by SHM undergo apoptosis. The largest group of apoptotic B cells in the DZ arises by AID-mediated introduction of stop codons, insertions, or deletions into BCR genes (33). Thus, GC B cells resemble developing and naïve B cells in that they require BCR expression for survival (18,(34)(35)(36).
GC B cells are among the most rapidly dividing eukaryotic cells with cell-cycle times as short as 4 to 6 hours. Despite rapid cell division and limited export of memory and plasma cells, the size of the GC is relatively stable over periods of weeks. Our experiments indicate that GC homeostasis is maintained by high rates of cell death. Moreover, whereas positive selection occurs in the LZ, negative selection by apoptosis occurs in both GC zones but is differentially regulated in the LZ and DZ.
To produce Igk INDIA transgenic mice, INDIA was placed under transcriptional control of mouse Igk regulatory elements as previously described (40,41), with the modification that the entire open reading frame was placed after the noncoding Vk exon followed by the Igk polyadenylation signal. After sequence and vector integrity confirmation, the vector backbone was eliminated by NotI/MluI digestion, and the resulting 7.2-kb fragment was injected into the pronuclei of fertilized C57BL/6J oocytes. Transgenic founder mice were identified by PCR (table S1; annealing temperature 55°C) on tail DNA. The transgenic founder line Igk INDIA was selected and maintained by mating to C57BL/6J mice or by intercrossing. Genotyping was performed by flow cytometry on peripheral blood.
To generate Rosa26 INDIA knock-in mice, INDIA was cloned into the AscI site of the CTV targeting vector (42) that was a gift from Klaus Rajewsky (Addgene plasmid #15912). The frt-flanked IRES-eGFP sequence was deleted by L-arabinoseinducible Flp recombineering in SW105 bacteria (43). In brief, Flp expression was induced by 0.09% (w/v) L-arabinose for 1 hour at 32°C, followed by electroporation of 1 ng vector and growth on LB agar plates containing ampicillin over night at 32°C. Vector integrity and deletion of IRES-eGFP in the resulting clones were confirmed by restriction-enzyme digestion and sequencing. CY2.4 albino C57BL/6J-Tyr c-2J -derived ES cells were targeted with the modified CTV vector at the Gene Targeting Resource Center (The Rockefeller University) and homologous recombination was verified by Southern blot and PCR. Chimeric males obtained after blastocyst injection were crossed to B6(Cg)-Tyr c-2J /J females. Rosa26 LSL-INDIA mice exhibiting germline transmission were identified by PCR and crossed to B6.C-Tg(CMV-cre)1Cgn/J mice to induce germline deletion of the LSL cassette. Rosa26 LSL-INDIA CMV-cre + offspring were crossed to C57BL/6J mice. The resulting CMVcre − offspring ubiquitously expressing INDIA were intercrossed to establish Rosa26 INDIA mice. Genotyping was performed by flow cytometry on peripheral blood.

Mice
B6.C-Tg(CMV-cre)1Cgn/J, B6(Cg)-Tyr c-2J /J, B6.SJL and C57BL/6J mice were purchased from Jackson Laboratories. AID Cre/Cre IgH 96K/96K Rosa26 LSL-YFP , AID -/-, AID-GFP, Nur77-GFP, Myc-GFP and B1-8 hi mice were described previously (10,13,22,28,29,44,45). Bone marrow chimeras were generated as described (46). All animal experiments were approved by the Institutional Review Board and the IACUC at The Rockefeller University. estimate dynamic GC B cell loss by apoptosis, mice were injected intraperitoneally with 1 mg of 5-ethynyl-2′-deoxyuridine (EdU; ThermoFisher Scientific) every 2 hours for up to 10 hours. For identifying GC B cells in S phase, mice were given a single intravenous pulse of 1 mg of EdU 1.0 to 2.5 hours prior to sacrifice. To elicit secondary GCs, mice were immunized intraperitoneally with 100 ml of PBS containing 50 mg of OVA (Grade V, Sigma) precipitated in alum. Two weeks later, mice received 5 × 10 6 B1-8 hi B cells intravenously (~5 × 10 5 Igl + NP-specific B cells; 5% Rosa26 INDIA / 95% Rosa26 WT ) followed one day later by a subcutaneous boost with 25 ml of PBS containing 25 mg of NP 15  B cell isolation and culture B cells were purified from spleens and subcutaneous lymph nodes as previously described (2). Igk INDIA B cells were stimulated for 4 days with 25 mg/ml of LPS (L-2630, Sigma) and 5 ng/ml of IL-4 (I1020, Sigma) in vitro as described (47). Activated B cells were harvested, washed and cultured at 2 × 10 6 cells/ml with LPS/IL-4-free media for 3 hours in the presence of 1 mM staurosporine (1285, Tocris) to induce apoptosis. GC B cells were enriched from immunized Igk INDIA mice by incubating single-cell suspensions with 1.25 mg/ml of biotinylated a-IgD for 10 min on ice followed by incubation with a-Biotinand mouse CD43 MicroBeads (Miltenyi Biotech). Cells were passed through a magnetized LS column (Miltenyi Biotec) and enriched GC B cells were collected in the flow-through.

Live imaging of cultured B cells
On day 4, LPS/IL-4-activated Igk INDIA B cells were washed and placed in a 35-mm m-Dish (Ibidi). After the addition of 1 mM staurosporine, cells were imaged in an environmental chamber set to 37°C using a DeltaVision Inverted Olympus IX-71 Image Restoration Microscope (GE Healthcare) with an Insight SSI 7 color solid state illumination system and a 20× dry objective. Separate excitation and emission filter wheels were employed to collect data for mNeonGreen (488-nm excitation; 525/50 BP), FRET (488-nm excitation; 632/60 BP) and standard brightfield images for each time point. 512 × 512 pixel images were taken every 60 s for 3 hours using Ultimate Focus and a Prior SYZ piezo stage for multiple point visiting. Images were processed and analyzed with ImageJ 1.48q (National Institutes of Health) and videos were generated at a frame rate of 10 fps.

Intravital imaging and image analysis
Intravital imaging of popliteal lymph nodes and image acquisition were essentially performed as described previously (31). Mice were anesthetized by the inhalation of 4% isofluorane in pure oxygen, placed on a stage warmer set to 37°C and maintained on anesthesia by inhalation of 1.25% isofluorane in pure oxygen. Popliteal lymph nodes in shaved hind legs were exposed by microsurgery and animals were placed under the heated Olympus 25× 1.05 NA Plan objective of an Olympus BX61 upright microscope fitted with a Coherent Chameleon Vision II IR laser. The femtosecondpulsed multiphoton laser was tuned to 900 nm. A filter cube containing a 690LP mirror followed by a 495LP mirror was used to split the emission to either 2 GaAsp detectors (with a 500-550-nm filter for mNeonGreen fluorescence and a 575-to 630-nm filter for FRET fluorescence) and a PMT detector (with a 460-to 500-nm filter for CFP/ autofluorescence). Images were acquired every 30 s as 75-mm Z-stacks (5-mm steps) with 1.4× zoom and with 512 × 512 X-Y resolution. Imaris software (Bitplane) was used to process data. Collapsed Z-stacks were exported as TIFF series from Imaris and videos were generated in ImageJ (National Institutes of Health) at a frame rate of 7 fps. The colocalization tool in Imaris was used to detect B1-8 hi Rosa26 INDIA GC B cells that were mNeonGreen + and lacked CFP/autofluorescence. The surface tool in Imaris was used to track mNeonGreen + GC B cells and the mean fluorescence intensities of mNeonGreen and FRET channels over time. The FRET loss ratio was calculated by dividing mNeonGreen and FRET fluorescence.
Flow cytometry data were acquired on a BD Fortessa (BD Biosciences) and data were analyzed with Flowjo (Tristar). Intact cells and singlets were identified by their FSC/SSC profiles and in the case of Igk INDIA B cells additionally by mRuby2 expression (561-nm excitation; 582/15 BP). FO B cells were gated CD19 + CD38 + Fas − and GC B cells were gated CD19 + CD38 − Fas + and additionally GL7 + where indicated. Live and apoptotic fractions were discriminated by aCasp3 staining, or by mNeonGreen (488-nm excitation; 505LP and 530/30 BP) and FRET (488-nm excitation; 600LP and 610/20 BP) for Igk INDIA mice. FRET loss was derived as a separate parameter in Flowjo defined as the ratio of mNeonGreen and FRET fluorescence among intact mRuby2 + B cells. GC B cell fractions were differentiated into LZ (CXCR4 lo CD86 hi ) and DZ (CXCR4 hi CD86 lo ) where indicated. Due to lower expression of CD86 in apoptotic compared to live GC B cells (see fig. S1), DZ and LZ were separately gated for live and apoptotic GC compartments. Apoptosis rates in DZ and LZ were calculated as: (% aCasp3 + of GC) × (% DZ or LZ of aCasp3 + GC) / (% DZ or LZ of total GC).

Cell sorting
Cell sorting was carried out on a FACS Aria II (BD Biosciences). For bulk sorting, cultured Igk INDIA B cells were washed and directly re-suspended in PBS containing 1% serum and 2 mM EDTA. (Live: FRET + B220 + ; apoptotic: FRET − B220 lo ). Live and apoptotic GC B cell fractions were further differentiated into LZ (CXCR4 lo CD86 hi ) and DZ (CXCR4 hi CD86 lo ). Single GC B cells of each compartment were sorted into 96-well PCR plates containing 4 ml of lysis buffer (48) and were either directly processed or stored at -80°C.
ELISAs Autoreactivity against nuclear and cytoplasmic self-antigens was determined with QUANTA Lite ANA ELISA (Inova Diagnostics) as described (52). Polyreactivity against ssDNA, dsDNA, Keyhole limpet hemocyanin (KLH), human insulin, and lipopolysaccharide (LPS) was determined as described (22). To measure NP binding, high-binding 96-well plates (Corning) were coated overnight with 50 ml of PBS containing 10 mg/ml of NP 4 -BSA or NP 25 -BSA (Biosearch Technologies). After washing with PBS containing 0.05% of Tween 20 (Sigma), wells were blocked with PBS containing 1% of BSA for 2 hours at room temperature. Monoclonal antibodies were incubated at 4 mg/ml or 7 consecutive 1:4 dilutions in PBS for 2 hours at room temperature. After washing, HRP-conjugated goat a-human IgG (Jackson ImmunoResearch) was added at 0.16 mg/ml for 1 hour at room temperature. After additional washing, HRP was revealed with 1-Step ABTS Substrate Solution (ThermoFisher Scientific). Absorbance was measured at 405 nm after incubation for 20 min at room temperature.

Statistical analyses
Statistical significance was determined with Graphpad Prism Version 6.0 using the tests indicated in each figure.