Defects in perforin and related genes lead to abnormal T-cell activation and are associated with HLH.
The physiological mechanism by which perforin protects from HLH involves CD8+ T-cell elimination of rare antigen-presenting dendritic cells.
Humans and mice with impaired perforin-dependent cytotoxic function may develop excessive T-cell activation and the fatal disorder hemophagocytic lymphohistiocytosis (HLH) after infection. Though cytotoxic lymphocytes can kill antigen-presenting cells, the physiological mechanism of perforin-mediated immune regulation has never been demonstrated in a disease-relevant context. We used a murine model of HLH to examine how perforin controls immune activation, and we have defined a feedback loop that is critical for immune homeostasis. This endogenous feedback loop involves perforin-dependent elimination of rare, antigen-presenting dendritic cells (DCs) by CD8+ T cells and has a dominant influence on the magnitude of T-cell activation after viral infection. Antigen presentation by a minor fraction of DCs persisted in T-cell– or perforin–deficient animals and continued to drive T-cell activation well beyond initial priming in the latter animals. Depletion of DCs or transfer of perforin-sufficient T cells dampened endogenous DC antigen presentation and T-cell activation, demonstrating a reciprocal relationship between perforin in CD8+ T cells and DC function. Thus, selective cytotoxic “pruning” of DC populations by CD8+ T cells limits T-cell activation and protects against the development of HLH and potentially other immunopathological conditions.
Precise regulation of the immune response is essential for defense against pathogens and for avoiding damaging immune-mediated pathologies. Primary human immune deficiencies have demonstrated the importance of multiple immunoregulatory pathways for maintaining this critical balance. For example, disorders due to inborn genetic errors, such as autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy or immunodysregulation, polyendocrinopathy, enteropathy, X-linked syndrome lead to the development of potentially severe autoimmunity. Perhaps most unexpectedly, mutations in PRF1 (perforin) and related genes affecting the perforin-dependent pathway of lymphocyte cytotoxicity lead to a fatal inflammatory disorder known as hemophagocytic lymphohistiocytosis (HLH).1 Patients with HLH experience discrete episodes of extreme immune activation and widespread organ damage that are often (though not always) triggered by initial infection with a variety of pathogens or, rarely, vaccination. However, unlike patients with autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy or immunodysregulation, polyendocrinopathy, enteropathy, X-linked syndrome, patients with HLH do not have evidence of autoimmunity. Thus, perforin-dependent cytotoxicity appears to have a distinctive and critical immune regulatory function.
Perforin-deficient (prf−/−) mice and other mice with defects in this pathway develop a severe HLH-like syndrome after infection with lymphocytic choriomeningitis virus (LCMV).2-10 In this context, prf−/− mice develop a striking increase of CD8+ and CD4+ T-cell activation, associated with increased antigen presentation by as yet undefined cells.5 While the possibility of negative feedback from cytotoxic lymphocytes (both T and natural killer [NK]) to antigen-presenting cells (APCs) has been recognized for many years,11,12 the details of this putative immune regulatory loop remain undefined or in dispute. Several reports have demonstrated that dendritic cells (DCs) may be eliminated in vivo, as in vitro, by cytotoxic lymphocytes.13-16 However, these reports relied on exogenously administered DCs, cultured with synthetic antigen, and thus did not clarify which endogenous cell types may participate in a physiological feedback loop. Studies examining endogenous DCs have not shown an effect of perforin in primary herpes simplex virus or influenza infection (though an effect was seen with memory rechallenge in the latter).17,18 Therefore, how perforin protects from HLH, which is most often triggered by a primary infection (not rechallenge or reactivation), remains unclear. Furthermore, depending on the experimental context, various lymphocytes have been found to influence immune responses via potentially cytotoxic mechanisms: CD8+ T cells,19 NK cells,20,21 and regulatory CD4+ T cells.22,23 However, it is not clear which cell type(s) is most important to human disease because none of these experimental contexts is clearly relevant to HLH. Thus, while circumstantial evidence supports a role for perforin in the feedback control of immune activation, the principle mechanisms of this effect, the components of a putative feedback loop, and how this may relate to disease development in the context of deficiency, remain unclear.
In this study, we have defined the components of a dominant perforin-dependent immune regulatory feedback loop in LCMV-infected prf−/− mice, the context that most closely mimics human HLH. Surprisingly, we found that T-cell hyperactivation in these mice is largely limited to lymphoid tissues, though LCMV infection is known to be systemic. Underlying this tissue specificity, we found that rare endogenous DCs containing and presenting viral antigen to T cells were increased in prf−/− mice. Depletion or genetic elimination of T cells, and in vivo blockade of caspases, all increased wild-type (WT) DC antigen presentation to levels seen with prf−/− DCs, while transfer of prf+/+ T cells into prf−/− mice suppressed antigen presentation and endogenous T-cell activation. Finally, depletion of DCs suppressed T-cell activation in WT and prf−/− mice, demonstrating a novel role for this cell type in the control of effector T-cell activation. Thus, a reciprocal relationship exists between DCs and CD8+ T cells involving antigen presentation and perforin-dependent “pruning” of DC populations. Defects in this feedback loop appear to be the principle mechanism of immune dysregulation underlying HLH.
Mice and in vivo treatments
C57BL/6, prf−/−, CD11c-DTR (prf+/+ and prf−/−), P14, OT1, and RAG−/− mice were bred in our animal facility. For LCMV infection, 200 plaque-forming units of LCMV-WE was injected intraperitoneally (IP). Viral stocks were tittered using a standard plaque-forming assay.24 DC depletion (in mice reconstituted with CD11c-DTR marrow) was accomplished by injection of diphtheria toxin (DT) (5 ng/g, IV; List Biological) 2 days in a row. CD8 T-cell depletion was produced by injecting 1 mg YTS-169 1 day before LCMV infection. For adoptive transfer, 20-30e6 purified naive, polyclonal CD8+ T cells from B6.SJL/BoyJ mice (prf+/+ or prf−/−) were transferred 2 days after administering cyclophosphamide (50 mg/kg IP). Animals were then allowed to recover/ reequilibrate for 3 weeks before the LCMV challenge. These transfers resulted in a normal number of CD8 T cells, CD4 T cells, and B cells in treated mice by week 3 (with 10% to 30% donor chimerism within the CD8+ compartment). For in vivo inhibition of caspases, 0.5 mg Q-VD-OPH (Biomedical) was administered twice, IP, 12 hours apart on day 5 after LCMV infection. All studies were conducted on an Institutional Animal Care and Use Committee–approved protocol.
Purification of DCs
DCs were purified from spleens treated with collagenase/DNase (Liberase CI; Roche) using anti-CD11c MicroBeads (Miltenyi) and were typically 80%+ CD11c+, major histocompatibility class IIhi. Briefly, spleens were minced in Click's media, incubated at 37°C with gentle agitation for 25 minutes, treated with EDTA, centrifuged over a ficoll gradient, and purified using limiting numbers of magnetic beads (one-third of the manufacturer’s protocol). For purification of non-DC APCs (Figure 2A), spleen cells were incubated with biotinylated anti-T cell receptor (H57-597) and anti-CD11c, then negatively sorted using Anti-Biotin MicroBeads.
T-cell purification and in vitro generation of effector T cells
Naive polyclonal CD8+ T cells were purified using a negative bead sorting kit (Miltenyi) per the manufacturer’s protocol. Effector T cells were generated in vitro based on the method of Manjunath et al,25 in which spleen cells from T-cell transgenic mice (P14 or OT1) were first stimulated with cognate peptide antigen for 48 hours, then washed and cultured in IL-2 for 3 to 4 days, and then “rested” in interleukin (IL)-7 (0.5 ng per 107 cells) for 2 to 3 days.
In vitro antigen presentation assays
For DC/T-cell stimulation assays, DCs were plated in decreasing numbers with effector T cells, where the concentration of T cells remained constant at 3 × 105 per well. Interferon-γ (IFN-γ) production was determined by enzyme-linked immunosorbent assay (ELISA). DCs or T cells cultured in separate wells did not produce measurable IFN-γ. For limiting dilution stimulation assays (Figure 4), 5 × 104 effector T cells were plated in round-bottomed plates with the indicated number of DCs along with 0.5 ng/mL IL-7 and 0.5 mcg/mL biotinylated anti-IFN-γ (R46A2). Plates were then centrifuged at 300 × g for 10 seconds. Supernatants were assayed at 18 to 24 hours for IFN-γ, using the cytokine capture ELISA assay previously described in Finkelman and Morris.26 The limit of detection for these assays represents the background signal derived from wells in which effector T cells were cultured without DCs. Peptide-loaded DCs were generated by incubating DCs with Gp33-41 peptide (KAVYNFATM) at 1 ng/mL for 1 hour at 37°C and washing extensively.
Flow cytometric assays
All labeled antibodies for flow cytometry were obtained from eBioscience or BioLegend, except anti-LCMV antibody. Staining for LCMV antigen was with Alexa 647–labeled, protein-G purified, polyclonal mouse IgG obtained from LCMV convalescent mice. Cells were permeabilized with saponin and stained for LCMV in the presence of 10% normal mouse serum. Direct ex vivo intracellular cytokine staining was performed by injecting brefeldin A (100 mcg IP) and removing spleens 12 to 18 hours later.5 Spleens were crushed immediately in 1% paraformaldehyde/phosphate-buffered saline, then permeabilized and stained. This technique measures the number of cells that are producing cytokine in vivo at the moment cells are harvested, unlike conventional intracellular cytokine staining methods, which measure the number of cells that are capable of producing cytokines after in vitro stimulation.
All studies were repeated at least twice with consistent results and with a minimum of 4 mice per group, though typically more. All P values were calculated using a 2-tailed Student t test unless otherwise noted. Data are presented +/− standard error of the mean.
T-cell activation is selectively increased in lymphoid tissues of prf−/− mice after LCMV infection
We previously reported that CD8+ and CD4+ T-cell activation is abnormally increased in prf−/− mice after LCMV infection.5 We assessed activation of T cells by a variety of means, including cell surface markers, cytokine production, ex vivo degranulation, and intracellular signaling. We found that the most dynamic and sensitive direct measure was in vivo IFN-γ production, as assessed by fixing T cells directly and immediately upon harvest, then staining intracellularly. In the current study, acute CD8+ T-cell activation was assessed via this method in a variety of tissues from WT and prf−/− mice (Figure 1). We found that the previously reported increase in T-cell activation was, in fact, specific to lymphoid tissues. In both percentage and absolute terms, the number of activated T cells was increased three- to fourfold in prf−/− mice, compared with WT mice, 1 week after LCMV infection. This abnormal activation correlated roughly with the peak of T-cell expansion, preceding viral clearance and ultimately subsiding with delayed kinetics, compared with that of WT animals.5 Moreover, this increased activation occurred despite similar concurrent viral loads in WT and prf−/− mice, though the latter ultimately failed to clear the virus.5 Finally, while abnormal activation of prf−/− T cells was seen in vivo, it was not seen with uniform stimulation in vitro.5 These features and the tissue specificity of this abnormality suggested that it was due to extrinsic/environmental factors, such as the strength of antigen presentation.
DCs are the principle cell type presenting antigen and driving T-cell activation 1 week after LCMV infection
While the initial priming of T cells has been demonstrated to be dependent on DCs,27 it is possible that many cell types in addition to DCs are presenting antigen to primed CD8+ T cells at later time points relevant to the development of immunopathology. Moreover, LCMV produces a systemic infection with broad cellular tropism.28 In order to define the cell type(s) that was driving excessive T-cell activation in prf−/− mice, we tested various splenic cell populations for their ability to present endogenously derived viral antigen in an ex vivo assay with LCMV-specific transgenic T cells. When we assayed spleen populations from prf−/− mice 6 days after LCMV infection, we found that essentially all detectable antigen presentation was by CD11c+ DCs, while other splenic cell populations were unable to stimulate antigen-specific effector T cells (Figure 2A). We saw a similar pattern in WT mice, though overall antigen presentation was weaker (data not shown). To confirm these findings, we transplanted WT and prf−/− mice with prf+/+ or prf−/− CD11c-DTR transgenic bone marrow (respectively), which allows selective in vivo depletion of DCs with DT. After engraftment, chimeric mice were challenged with LCMV infection and given DT on days 5 and 6. This time point is well after initial T-cell priming and around the time that developing effector T cells begin to acquire cytotoxic function.29-32 When we examined in vivo IFN-γ production by splenic T cells on day 7, we found that there was substantial suppression with this late depletion of DCs (Figure 2B). Thus, DCs continue to be the primary cell type driving T-cell activation in vivo (at least in lymphoid tissues) after initial priming and up to the peak of an antiviral response.
Though total DC numbers are not increased in prf−/− mice after LCMV infection, antigen presentation by these cells is increased, compared with WT mice
Because DCs appeared to play a critical role in driving T-cell activation in prf−/− mice and because work by other groups has suggested cytotoxic destruction of APCs, we predicted that we would observe a quantifiable loss of DCs in WT mice, compared with prf−/− mice. However, when we enumerated splenic DCs on day 6 after LCMV infection, we found no difference between WT and prf−/− mice (Figure 3A). Similarly, analysis of DC subsets (defined by CD11b, CD8a, SIRPα, and B220) revealed no consistent differences (data not shown). In order to compare the function of DC populations from WT and prf−/− mice, we sorted splenic DCs 6 days after LCMV infection and cultured them overnight with LCMV-specific (P14) or OT1 effector CD8+ T cells and measured IFN-γ production across a range of DC/T-cell ratios (Figure 3B). We found that DCs from prf−/− mice were superior stimulators of LCMV-specific T cells, approximately threefold more potent in terms of their ability to stimulate cytokine production. This difference appeared to be due to increased presentation of endogenously acquired viral antigen, because no IFN-γ was produced when DCs were cultured with T cells of irrelevant specificity. Furthermore, when loaded with the relevant peptides (either LCMV-GP33 or Ova-8), DCs from WT and prf−/− mice stimulated T cells in an identical fashion (data not shown). Next, we sorted DCs at a range of time points after LCMV infection and compared antigen presentation. We found that early after infection, DCs from WT and prf−/− mice had similar function. However, by day 6, antigen presentation by DCs from WT mice was significantly suppressed (Figure 3C). Notably, this occurred despite the fact that WT and prf−/− mice harbor similar amounts of virus at this time point.5 However, this suppression of DC function correlates with the acquisition of perforin-dependent cytotoxic function by CD8+ T cells in WT hosts.29,30
Prf−/− mice harbor increased numbers of DCs that contain viral antigen and present it to T cells after LCMV infection
Because bulk DC populations from infected prf−/− mice presented antigen more potently than those from WT mice, and while total DC numbers were unchanged in vivo, we reasoned that prf−/− DC populations must contain an increased number of viral antigen-containing cells. Indeed, when we assessed splenic DCs using polyclonal anti-LCMV sera, we found a substantial increase in the number of LCMV antigen+ cells (Figure 4A-B). However, the presence of antigen does not equal functional processing and presentation of that antigen. In order to enumerate the number of DCs presenting antigen, we developed a sensitive limiting dilution assay. Briefly, limiting numbers of sorted splenic DCs were plated with an excess of previously primed effector T cells in the presence of biotinylated anti–IFN-γ capture antibody.26 Using this approach, we were able to detect antigen presentation by as few as 3 peptide-loaded DCs in a single well, though approximately 15 DCs were needed to detect antigen presentation in 50% of the wells assayed (Figure 4C and data not shown). When we assessed splenic DCs (without peptide loading) from day 6 LCMV-infected WT and prf−/− mice, we found that prf−/− mice harbored an increased number of DCs that were presenting viral antigen (Figure 4C). Antigen presentation was detectable in 50% of wells when plating approximately 15 peptide-loaded DCs, or 100 prf−/− DCs (not loaded with peptide), or 500 WT DCs. Thus, we estimate that 1 in 7, or 1 in 35, splenic DCs obtained from prf−/− or WT mice continue to present viral antigen 6 days after LCMV infection. It is clear why perforin-dependent pruning of endogenous DCs is not readily detectable with conventional assays; the stimulatory DC population represents less than 15% of DCs or <0.5% of the entire spleen. A caveat of our methods is that we were only able to measure antigen presentation by DCs that survived tissue disruption and sorting. However, it is notable that the activation of endogenous and adoptively transferred T cells in prf−/− mice is roughly proportional to the increased numbers of stimulatory DCs we detected in these mice (Figure 1 and Lykens et al5 ). Additionally, though these methods are not directly comparable, we were able to detect threefold more LCMV antigen+ DCs in prf−/− mice. Of note, we measured 2 to 3 times more antigen-containing DCs than antigen-presenting DCs in prf−/− mice. The specific identity of antigen-presenting DC subsets remains under investigation.
CD8+ T cells and the action of caspases are necessary for suppression of antigen presentation by endogenous DCs in WT mice
Loss of antigen-presenting DCs in WT (but not prf−/−) mice strongly suggested that DCs were being killed by cytotoxic lymphocytes. In order to more directly test this hypothesis, we administered a potent pan-caspase inhibitor (Q-VD-OPH), suitable for in vivo use,33-35 or a carrier (dimethylsulfoxide) to WT and prf−/− mice for 24 hours before DC purification after LCMV infection (at day 6). We found that this treatment had little effect on antigen presentation by DCs from prf−/− mice, but it greatly enhanced the function of WT DCs such that they became equivalent to those from prf−/− mice (Figure 5A). Because perforin-mediated, granzyme-dependent killing relies significantly upon the activation of caspases,36,37 these data suggest that perforin-mediated suppression of DC function in WT mice is occurring via a conventional cytotoxic mechanism.
The timing of perforin-dependent suppression of antigen presentation after LCMV infection suggested that CD8+ effector T cells were the predominant perforin-expressing cell type affecting DC populations (Figure 3). To directly test this idea, we administered CD8-depleting antibody (or isotype) to WT or RAG−/− mice and assessed antigen presentation by DCs after LCMV infection (Figure 5B). We found that DC antigen presentation was greatly increased by the depletion of CD8+ cells or the genetic elimination of T (and B) cells. While administration of anti-CD8 antibody dramatically increased antigen presentation in WT mice, it had only a very slight effect in RAG−/− animals, suggesting that the depletion of CD8+ T cells was the critical mechanism for this effect. Together, these data suggest that the suppression of viral antigen presentation by DCs in WT mice after LCMV infection is due to the cytotoxic action of CD8+ T cells.
Perforin-expressing CD8+ T cells are sufficient to suppress host T-cell activation in vivo and viral antigen presentation by endogenous DCs
Next, we performed adoptive T-cell transfers to further test the role of CD8+ T cells in a physiological perforin-dependent feedback loop. Transfer of low numbers of naive polyclonal (1e6) or transgenic (1e4) LCMV-specific CD8+ T cells did not affect endogenous T-cell activation or antigen presentation (data not shown). Because transfer of high numbers of monoclonal, antigen-specific T cells has been shown to be nonphysiological and distorts the developing immune response, we avoided this approach.38,39 Instead, we transferred large numbers of (CD45.1-marked) naive, polyclonal, prf+/+ or prf−/−, CD8+ T cells into prf−/− recipients. In order to achieve a relatively high level of donor chimerism (>20% donor T cells in most animals), we first lymphodepleted animals with cyclophosphamide, then transferred purified CD8+ T cells, then waited 3 weeks to allow lymphocyte numbers to return to normal levels, and then infected with LCMV and assessed in vivo T-cell activation. We found that activation of endogenous (and transferred) T cells was suppressed in animals that received prf+/+ T cells but not in those that received prf−/− T cells (Figure 6A and data not shown). Similarly, presentation of viral antigen by DCs was suppressed in recipients of prf+/+ T cells (Figure 6B). For both T-cell activation and antigen presentation by DCs, WT-like levels were reestablished in prf−/− mice with T-cell transfer. Thus, normal perforin expression in a significant fraction of CD8+ T cells is sufficient to reestablish physiological perforin-dependent immune regulation in vivo.
Fatal inflammation may be triggered in patients with defective perforin-dependent cytotoxicity by a wide array of viral, bacterial, fungal, and parasitic infections,40 suggesting that this pathway is of fundamental importance for regulating immune responses. Though prior studies have suggested a variety of potential mechanisms, precisely how perforin prevents lethal immunopathology in physiological contexts has remained unclear. In the current study, we have defined a physiological negative feedback loop between expanding CD8+ cytotoxic T lymphocyte populations and a minor fraction of DCs, which has a dominant role in preventing pathological T-cell activation in prf−/− mice after LCMV infection. This model system closely mimics human HLH.2 The primary role of DCs is implicated by 4 novel findings of the current study: (1) At relevant time points, DCs are the only cell type presenting antigen with sufficient potency to drive T-cell activation ex vivo. (2) T-cell hyperactivation in prf−/− mice is limited to lymphoid tissues, where DCs reside/accumulate. (3) Pathological T-cell activation in vivo is dependent on the presence of DCs. (4) prf−/− mice harbor increased numbers of DCs that contain and present viral antigen at later time points. Three distinct findings of the current study indicate that the mechanism of DC modulation involves cytotoxic killing of rare DCs: (1) Antigen presentation by bulk DC populations in WT mice is suppressed around the time expanding CD8+ T-cell populations acquire cytotoxic function. (2) This suppression is dependent on CD8+ T cells and the action of caspases. (3) The transfer of prf+/+, CD8+ T cells suppresses DC function (and T-cell activation) in prf−/− mice. Thus, reciprocal interactions between DCs and CD8+ T cells, after initial priming and during the expansion phase of the T-cell response, is the predominant mechanism by which perforin restrains immune responses and prevents immunopathology in this murine model of human HLH. Our previous2,5 and current findings have defined perforin-mediated immunoregulation as a negative feedback loop between the innate (DCs) and adaptive (CD8+ T cells) portions of the immune system (Figure 7).
While the current report suggests a dominant mechanism for perforin-mediated immune regulation in the context of HLH, it does not preclude other immunoregulatory roles in other contexts. NK cells and regulatory CD4+ T cells have also been shown to have immunoregulatory functions that may depend on perforin. Two recent studies have suggested that in the context of chronic LCMV infection, NK cells may play a significant immune regulatory role, with complex and variable effects on immunopathology and control of infection.20,21 How such findings relate to human cytotoxic deficiency is unclear because most patients with HLH do not suffer from chronic or unusual infections.1 Furthermore, in our acute infection model, we did not observe an effect of NK cell depletion in WT mice on DC function or T-cell activation (data not shown).
In addition to perforin-dependent pathways, NK cells and CD8+ T cells may kill target cells by nonperforin-dependent pathways, such as Fas or Trail. This fact suggests additional pathways to be studied. Though humans and mice with abnormalities of the Fas pathway do not develop similar pathology to those with perforin mutations, experimental studies of double knockout mice (Fas and perforin) indicate an important combined role for these pathways in controlling inflammation.41,42 Accordingly, our data (Figure 5 and data not shown) suggest that CD8+ T cells are suppressing antigen presentation by both perforin-dependent and independent mechanisms.
Our finding that a small subpopulation of DCs can drive fatal, T-cell–mediated immunopathology suggests an important but little-studied role for ongoing DC/T-cell interactions, long after initial immune priming. At steady state, DCs have relatively short life spans,43 with additional inflammation-induced populations entering lymphoid tissues after infection. Our findings suggest that a constant influx of antigen-bearing DCs into lymphoid tissues during infection shapes the activation and expansion of reactive T cells. Future studies characterizing these DC populations, and the kinetics of their interactions with T cells, may reveal important aspects of how immune responses are shaped and amplified in both normal and prf−/− hosts. How aberrant DC/T-cell interactions in lymphoid tissues lead to widespread tissue damage will also require further study. Though hormone-like effects of excessive inflammatory cytokines are clearly important, activation of tissue-infiltrating T cells (seen in both patients and mice) is likely to play a role as well.
What do the current studies suggest regarding our understanding of human HLH or the development of improved therapies for patients affected with this disorder? First, our results indicate that DCs and DC precursors deserve to be investigated further in HLH. To date, there have been no studies examining these cells in affected patients. Second, our results suggest that specific targeting of DCs would be a promising therapeutic approach. Such a treatment strategy could replace (at least in a short-term fashion) a core mechanism perforin-mediated immune regulation in vivo. Because DCs turn over rapidly, this approach would be unlikely to cause prolonged immune suppression and may have broad relevance to other immunopathological disorders beyond HLH.
The publication costs of this article were defrayed in part by page charge payment. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.
This work was supported by a grant from the National Institutes of Health (RO1-HL091769) and a grant from the Histiocytosis Association.
Contribution: C.E.T. designed and performed experiments. M.B.J. designed and performed experiments and wrote the manuscript.
Conflict-of-interest disclosure: The authors declare no competing financial interests.
Correspondence: Michael B. Jordan, Division of Cellular and Molecular Immunology, Cincinnati Children’s Hospital, 3333 Burnet Ave, ML 7038, Cincinnati, OH 45229; e-mail: firstname.lastname@example.org.