Letter | Published:

CAR T cell–induced cytokine release syndrome is mediated by macrophages and abated by IL-1 blockade

Nature Medicinevolume 24pages731738 (2018) | Download Citation

Subjects

Abstract

Chimeric antigen receptor (CAR) therapy targeting CD19 is an effective treatment for refractory B cell malignancies, especially acute lymphoblastic leukemia (ALL)1. Although a majority of patients will achieve a complete response following a single infusion of CD19-targeted CAR-modified T cells (CD19 CAR T cells)2,3,4, the broad applicability of this treatment is hampered by severe cytokine release syndrome (CRS), which is characterized by fever, hypotension and respiratory insufficiency associated with elevated serum cytokines, including interleukin-6 (IL-6)2,5. CRS usually occurs within days of T cell infusion at the peak of CAR T cell expansion. In ALL, it is most frequent and more severe in patients with high tumor burden2,3,4. CRS may respond to IL-6 receptor blockade but can require further treatment with high dose corticosteroids to curb potentially lethal severity2,3,4,5,6,7,8,9. Improved therapeutic and preventive treatments require a better understanding of CRS physiopathology, which has so far remained elusive. Here we report a murine model of CRS that develops within 2–3 d of CAR T cell infusion and that is potentially lethal and responsive to IL-6 receptor blockade. We show that its severity is mediated not by CAR T cell–derived cytokines, but by IL-6, IL-1 and nitric oxide (NO) produced by recipient macrophages, which enables new therapeutic interventions.

Main

To model CAR T cell–induced CRS in mice, we aimed to establish conditions whereby CD19 CAR T cells would engage a high tumor burden and initiate CRS within a few days, as is commonly observed in the clinical setting2,3,9,10. Whereas CRS could not be induced in mice with medullary disease, intraperitoneal tumor growth allowed for a sufficient tumor burden to accumulate and for severe CRS to develop in SCID-beige mice within 2–3 d of CAR T cell administration (Fig. 1a). Human 1928z CAR T cells reproducibly elicited an acute inflammatory response associated with reduced activity, general presentation of malaise, piloerection, weight loss (Fig. 1b) and eventual mortality (Fig. 1c). Remarkably, the serum cytokine profile elicited in these mice was highly similar to that reported in clinical studies2,11,12 (matching 18 out of 19 reported cytokines; Supplementary Table 1). Serum levels of the murine equivalent of C-reactive protein (CRP), serum amyloid A3 (SAA3)13,14 (Fig. 1d), and IL-6 (Fig. 1e and Supplementary Fig. 1a) increased as was observed in the clinic2,3,10, as did several other proinflammatory cytokines (Fig. 1e and Supplementary Fig. 1a)13,14. The overall levels of these cytokines, including mouse IL-6 (mIL-6), mouse chemokine (C–C motif) ligand 2 (mCCL2), mouse granulocyte colony–stimulating factor (mG-CSF), human interferon-γ (hIFN-γ), hIL-3, human granulocyte–macrophage-CSF (hGM-CSF) and hIL-2, correlated strongly with CRS severity and survival (Fig. 1e). Taking advantage of the xenogeneic nature of this model to discern the T cell or host cell origin of these cytokines and chemokines, we demonstrated that some cytokines, such as IFN-γ and GM-CSF, were products of human CAR T cells, whereas others, such as IL-6, were produced by endogenous mouse cells (Fig. 1f and Supplementary Fig. 1b). This finding establishes that the CRS cytokine signature is the result of a multicellular network and not merely a binary tumor–CAR T cell interaction. Furthermore, the lack of activity of hIFN-γ and hGM-CSF on the mouse cognate receptor (Supplementary Table 2) indicates that other CAR T cell–derived cytokines and/or CAR T cell activities account for CRS. Although dispensable in this model, T cell–derived IFN-γ and GM-CSF may yet contribute to CRS in other settings. Consistent with clinical CRS11, IL-15 was not differentially elevated upon CAR transfer (Supplementary Fig. 1c). In accordance with clinical experience2,3,9,10, treating mice with a mIL-6R-blocking antibody prevented CRS-associated mortality (Fig. 1g and Supplementary Fig. 1d).

Fig. 1: A mouse model of CRS recapitulates clinical CAR T cell–induced CRS.
Fig. 1

a, Schematic of the mouse model. Raji tumor cells were intraperitoneally injected in mice and allowed to grow for 3 weeks, and they eventually grew into vascularized solid tumor masses. Thirty million CAR T cells were transferred, and mice were monitored over the days following transfer. Mice were euthanized, and cells were obtained for analysis through peritoneal lavage and tissue harvesting. b, Weight change of tumor-bearing mice after 1928z CAR T cell transfer. Weight per mouse is normalized to starting weight before CAR transfer (Tumor only, n = 12 mice; Tumor + CAR, n = 18 mice). At least 20 independent experiments monitoring weight under CRS-eliciting conditions were performed. A two-way ANOVA was used for statistical analysis. c, Percentage of survival of mice after 1928z CAR T cell transfer (Tumor only, n = 12 mice, Tumor + CAR, n = 18 mice). At least 20 independent experiments monitoring survival under CRS-eliciting conditions were performed. A log-rank Mantel–Cox test was used for statistical analysis. d, Serum levels of mSAA3 at 42 h after 1928z CAR T cell transfer as measured using ELISA (No tumor + No CAR, n = 5 mice; Tumor only, n = 5 mice; CAR only, n = 5 mice; Tumor + CAR, n = 7 mice). SAA3 levels under CRS-eliciting conditions were measured in at least in two independent experiments. A two-tailed unpaired two-sample t-test was used for statistical analysis. e, Serum cytokine levels 4.5 h before (pre-CAR) and 24 h after 1928z CAR T cell transfer. Mice that eventually died from CRS were grouped under Severe CRS, whereas mice that survived but lost more than 10% of their initial body weight were grouped under CRS (Severe CRS, n = 5 mice; CRS, n = 10 mice; pre-CAR, n = 16 mice). Stratification of cytokines on the basis of survival was performed in at least three independent experiments. A two-tailed unpaired two-sample t-test was used for statistical analysis. f, Species of origin of proinflammatory cytokines (human cytokines from n = 15 mice; mouse cytokines from n = 15 mice). g, Percentage of survival of tumor-bearing mice treated with 1928z CAR T cells that received anti-mIL-6R blocking antibody or isotype (vehicle) (Vehicle, n = 6 mice; Anti-mIL-6R, n = 5 mice). Experiments measuring survival after blocking mIL-6R were performed once. A log-rank Mantel–Cox test was used for statistical analysis. All data are shown as mean ± s.e.m. Numbers on the graphs are P values.

Histopathological analyses performed 2 and 5 d after CAR T cell infusion did not reveal any evidence of graft-versus-host disease (GVHD) or tissue destruction (Supplementary Fig. 2), consistent with the initiation of this inflammatory response following tumor recognition by CAR T cells as well as the full recovery of mice surviving CRS. Histopathological examination of the central nervous system (CNS) and meninges at 1, 2 and 5 d after CAR T cell transfer did not reveal morphological evidence of acute damage or toxicity (Supplementary Fig. 3), consistent with the absence of overt neurological symptoms (seizures, limb dyskinesia or paralysis). None of the reported pathologic findings indicative of neuropathology15 or associated with neurotoxicity (cortical laminar necrosis, hemorrhages, disseminated intravascular coagulation (DIC), gliosis or vasogenic, neurotoxic or interstitial edema) in human patients16 were observed in any of the mice examined in the present study. The occurrence of subclinical functional alterations or ultrastructural morphological changes cannot be excluded. Mice surviving CRS rapidly returned to a highly active state, akin to healthy, tumor-free mice. No clinical neurological anomalies were noted until mice were euthanized because of tumor progression.

The high serum levels of mIL-6, a predominantly myeloid-derived cytokine, together with the presence of tumor-infiltrating myeloid cells (Fig. 2a) prior to CAR T cell transfer and more so thereafter led us to hypothesize that tumor-associated myeloid cells are closely linked with the induction of CRS. Only after infusion of CAR T cells in the presence of tumor was toxicity observed (Supplementary Figs. 1a and 4a), concurrent with a brisk accumulation of peritoneal neutrophils, eosinophils, dendritic cells (DCs), monocytes (Fig. 2b and Supplementary Fig. 5a) and F4/80int-loLy6Cint-hi macrophages (hereafter referred to as CRS-associated macrophages), which differ from the typical F4/80hiLy6Cneg-lo resident peritoneal phenotype (Fig. 2c). The rapid elevation of myeloid cell numbers, which was already noticeable 18 h after CAR T cell administration (Fig. 2d), suggested that recruitment was a major contributor to myeloid accumulation. RNA-seq analyses at 18 h showed increased expression of genes promoting the cell cycle, suggesting that cell proliferation may play a secondary role in local myeloid cell accumulation.

Fig. 2: Tumor–CAR T cell interactions trigger myeloid cell recruitment and activation.
Fig. 2

a, Immunohistochemical staining for MAC2 in sections from 3-week Raji tumor explants. Mice had not received CAR T cells. Immunohistochemical staining of tumor explants for MAC2 was performed once. b, Absolute counts of myeloid cell populations obtained by peritoneal lavage 68 h after 1928z CAR T cell transfer (No tumor + No CAR, n = 5 mice; CAR only, n = 5 mice; Tumor only, n = 6 mice; Tumor + CAR, n = 7 mice). Enumeration of myeloid cells under these conditions was performed at least 10 times. A two-tailed unpaired two-sample t-test was used for statistical analysis. c, Representative flow cytometric plot showing total peritoneal macrophages within the gated population (resident peritoneal macrophages, F4/80hiLy6Cneg-lo; CRS-associated macrophages F4/80int-loLy6Cint-hi). Cells were obtained from peritoneal lavage. Analysis of peritoneal macrophages for levels of F4/80 and Ly6C was performed at least 10 times. d, Absolute counts of myeloid cell populations obtained from multiple organs 18 h after 1928z CAR T cell transfer (Tumor only, n = 4 mice; Tumor + CAR n = 4 mice). Enumeration of cells retrieved from peritoneum, spleen and bone marrow under CRS conditions was performed at least three times. A two-tailed unpaired two-sample t-test was used for statistical analysis. e,f, Fold change of proinflammatory cytokine gene expression in myeloid populations as determined by RNA-seq analysis. Fold change was determined by comparing each population under Tumor only and Tumor + CAR conditions. Green bars, significant downregulation; red bars, significant upregulation; gray bars, no significant change. Gene expression levels were determined from cells obtained from n = 9 Tumor only mice and n = 9 Tumor + CAR mice. Cells from three mice of the same group were randomly pooled to create one biological replicate group for the purpose of obtaining three biological replicate groups of Tumor only mice and three biological replicate groups of Tumor + CAR mice. A binomial test was used for statistical analysis, and P values were adjusted for false discovery rate (FDR). All data are shown as mean ± s.e.m. except for that in eg, in which data are shown as mean ± s.d. Numbers on the graphs are P values.

To address whether these alterations were regional or systemic, we enumerated neutrophils, eosinophils, DCs, monocytes and macrophages in spleen, bone marrow, lungs, liver and peripheral blood (Fig. 2d and Supplementary Fig. 4b). Whereas neutrophils, DCs and macrophages accumulated at the tumor site, systemic perturbations were limited to increased counts of macrophages in spleen (Fig. 2d) and of neutrophils in peripheral blood, coinciding with neutrophil depletion in bone marrow (Fig. 2d and Supplementary Fig. 4b). The major alterations in myeloid cell distribution were thus confined to the tumor vicinity and to the spleen.

As IL-6 is a signature cytokine of CRS, we hypothesized that tracking down IL-6-producing cells would identify the main physiopathological sites. We therefore purified DC, macrophage and monocytic populations from peritoneum and spleen (IL-6 is not typically produced by neutrophils17) (Supplementary Fig. 5a,b) and performed RNA-seq analysis. Remarkably, only peritoneal, but not splenic, DCs, monocytes and macrophages showed upregulated IL-6 transcripts (Fig. 2e,f). The induction of other major proinflammatory cytokines was likewise confined to the site of CAR T cell–tumor tissue colocalization (Fig. 2e,f). Combined with cell-enumeration data, macrophages emerged as the main overall source of IL-6. As CAR T cells were found in the peritoneum, but not the spleen or other organs (Supplementary Fig. 4c), these findings further suggest that IL-6 induction and myeloid activation require proximity of CAR T cells and myeloid cells and possibly their direct interaction.

We hypothesized that arming the CAR T cells with a cell-surface ligand capable of activating myeloid cells would determine whether such a direct cellular contact can take place. To address this question, we again took advantage of our xenogeneic model to probe the potential for CD40 ligand (CD40L)–CD40 interactions. As CD40L is mainly expressed by T cells, whereas DCs, monocytes and macrophages express the CD40 receptor18, and as human CD40L does not functionally interact with the mCD40 receptor19, we constitutively expressed mCD40L in human CAR T cells (Supplementary Fig. 6a,b) and infused these in mice. This modification resulted in more severe and sustained weight loss in mice (Fig. 3a) and markedly increased mortality (Fig. 3b). The similar overall number of recruited myeloid cells in the CAR and CAR–mCD40L treatment groups (Supplementary Fig. 6c) suggested that the increased severity of CRS was due to qualitative and not quantitative changes in the myeloid compartment. Indeed, mice receiving 1928z-mCD40L CAR T cells showed an increased proportion of CRS-associated macrophages (Fig. 3c). Notably, although cell-surface expression of CD40 was exclusive to macrophages and DCs in peritoneal myeloid cells (Supplementary Fig. 6e), only macrophages downregulated CD40 expression (Fig. 3d and Supplementary Fig. 6d,f), an expected consequence of its ligation by CD40L20,21,22. In line with the increased severity of CRS, levels of murine inflammatory cytokines were significantly increased, including IL-6, which is known to be directly induced by CD40L signaling23 (Fig. 3e). These findings establish that CAR T cell–macrophage interactions can take place at tumor sites, and furthermore, that such interactions, albeit not obligatory, have the potential to aggravate CRS severity. Interestingly, expression of endogenous CD40L positively correlates with CAR expression in human CD4 T cells (Supplementary Fig. 6e), which suggests that such interactions potentially take place in the clinical setting. The genetic manipulation of receptor–ligand pairs in our model further serves as a proof of concept that species barriers can be exploited to investigate immunological interactions between adaptive and innate immune responses in xenogeneic tumor models. Altogether, our enumeration and expression profiling of tumor and splenic myeloid cells, compounded by the exacerbation of CRS via engineered CD40–CD40L interaction, designate CRS-associated macrophages as a major mediator of CRS severity.

Fig. 3: Modulation of macrophage function determines CRS severity.
Fig. 3

a, Weight change of tumor bearing mice after 1928z CAR T cell transfer. Weight per mouse was normalized to starting weight pre-CAR transfer (Tumor only, n = 8 mice; 1928z-LNGFR, n = 7 mice; 1928z-mCD40L, n = 5 mice). Results are representative from one experiment. A two-way ANOVA was used for statistical analysis. b, Percentage of survival of tumor-bearing mice treated with 1928z-LNGFR or 1928z-mCD40L CAR T cells (1928z-LNGFR, n = 16 mice; 1928z-mCD40L, n = 13 mice. Results were pooled from two independent experiments. A log-rank Mantel–Cox test was used for statistical analysis. c, Representative flow cytometric plot showing total peritoneal macrophages within the gated population (resident peritoneal macrophages and CRS-associated macrophages). Data from analysis of peritoneal macrophages for levels of F4/80 and Ly6C in the presence of the 1928z-mCD40L CAR are representative of one experiment of cells obtained from n = 8 Tumor only mice, n = 7 1928z-LNGFR mice and n = 5 1928z-mCD40L n = 5 mice. The experiment was performed once. d, Percentage of CD40+ total peritoneal macrophages, obtained by peritoneal lavage at 61 h following 1928z-LNGFR or 1928z-mCD40L CAR T cell transfer, as analyzed by flow cytometry (Tumor only, n = 8 mice; 1928z-LNGFR, n = 7 mice; 1928z-mCD40L, n = 5 mice). Analysis of peritoneal macrophages for surface expression of CD40 in the presence of the 1928z-mCD40L CAR was performed once. A two-tailed unpaired two-sample t-test was used for statistical analysis. e, Serum levels of mouse cytokines at 18 h following CAR T cell transfer. (Tumor only, n = 8 mice; 1928z-LNGFR, n = 7 mice; 1928z-mCD40L, n = 7 mice). Serum cytokine measurements when using the 1928z-mCD40L CAR were performed once. A two-tailed unpaired two-sample t-test was used for statistical analysis. f, Percentage of myeloid populations from peritoneum, spleen and bone marrow expressing iNOS protein in Tumor only mice and Tumor + CAR mice (for peritoneum, n = 14 mice per group; for spleen and bone marrow, n = 10 mice per group). Peritoneum results are shown from three pooled independent experiments. Spleen and bone marrow results are shown from two pooled independent experiments. A two-tailed unpaired two-sample t-test was used for statistical analysis. g, Weight change of tumor-bearing mice after 1928z CAR T cell transfer. Weight per mouse is normalized to starting weight pre-CAR transfer. Mice were treated with L-NIL or vehicle (PBS) (Tumor only, n = 7 mice; CAR + L-NIL, n = 7 mice; CAR + Vehicle, n = 8 mice). Weight measurement under conditions of CRS was performed once with L-NIL and once with 1400 W (Supplementary Fig. 7b). A two-way ANOVA was used for statistical analysis. h, Percentage of survival of tumor-bearing mice after 1928z CAR T cell transfer receiving 1400 W or Vehicle (PBS) (Vehicle, n = 20 mice; 1400 W, n = 19 mice). Results were pooled from three independent experiments. A log-rank Mantel–Cox test was used for statistical analysis. All data are shown as mean ± s.e.m. Numbers on the graphs are P values.

To further define the contribution of macrophages to CRS, we investigated the role of inducible nitric oxide synthase (iNOS), an enzyme predominantly expressed by macrophages upon their activation24. In line with a requirement for myeloid cell proximity to CAR T cells, only peritoneal, but not splenic or bone marrow, myeloid populations significantly increased iNOS production in CRS (Fig. 3f). Macrophages showed the highest induction (Fig. 3f) and were numerically the most abundant iNOS-expressing population (Supplementary Fig. 7a). Aberrant NO production is known to cause vasodilation and hypotension25,26, which are common features of clinical CRS that require vasopressor administration9. We treated mice with one of two iNOS inhibitors: L-NIL27 or 1400W28. Both alleviated clinical toxicity (Fig. 3g and Supplementary Fig. 7b), including mortality under conditions of severe CRS (Fig. 3h). These findings further support the direct role of macrophage-derived products in CRS and may provide a new means to mitigate CRS severity.

The direct involvement of iNOS in CRS pathophysiology prompted us to further examine the role of IL-6 and IL-1, both of which are inducers of iNOS production29,30. Our RNA-seq data in myeloid cell types harvested at the onset of CRS showed that the type 1 IL-1 receptor (IL1R1), which is required for functional IL-1 signaling, was exclusively upregulated in tumor-associated myeloid cells, but not splenic cells (Fig. 4a,b). Conversely, splenic myeloid cells upregulated only the type 2 IL-1 receptor (IL1R2), which does not functionally signal and serves as a decoy receptor to blunt proinflammatory IL-1 signaling. We also observed upregulation of IL-1 receptor antagonist (IL1RN/IL-1Ra) in splenic myeloid cells (Fig. 4b), which suggested an adaptive response to inhibit IL-1 signaling outside of the tumor bed31. We hypothesized that endogenously mounted IL-1 suppression was insufficient to inhibit the proinflammatory effects of IL-1 but that a pharmacological intervention could mitigate CRS symptoms. Indeed, IL-1 blockade by anakinra, an IL-1 receptor antagonist, abrogated CRS-related mortality (Fig. 4c). In order to obtain more insight in the protective mechanism underlying IL-1 blocking and to assess how it compares to IL-6 blockade, we assessed the impact of anakinra on expression levels of iNOS in macrophages. Both blockades resulted in similarly reduced iNOS+ macrophage fractions (Fig. 4d and Supplementary Fig. 7c), establishing downregulation of iNOS as one unifying mechanism through which IL-6 and IL-1 blockades can abate severe CRS. Combined IL-1 and IL-6 blockade, however, did not further decrease the fraction of iNOS+ macrophages (Fig. 4d), suggesting that the inhibition afforded by these two blockades operates through a common pathway.

Fig. 4: IL-1Ra protects from severe CRS without compromising antitumor efficacy.
Fig. 4

a,b, Fold change of IL-1 signaling component gene expression in myeloid populations as determined by RNA-seq analysis. Fold change was determined by comparing each population under Tumor only and Tumor + CAR conditions. Green bars, significant downregulation; red bars, significant upregulation; gray bars, no significant change. Gene expression levels were determined from cells obtained from n= 9 Tumor only mice and n= 9 Tumor + CAR mice. Cells from three mice of the same group were randomly pooled to create one biological replicate group to the end of obtaining three biological replicate groups of Tumor only mice and three biological replicate groups of Tumor + CAR mice. A binomial test was used for statistical analysis, and P values were adjusted for FDR. c, Percentage of survival of tumor-bearing mice after 1928z CAR T cell transfer receiving anakinra or vehicle (PBS) (Anakinra, n = 11 mice; Vehicle, n = 10 mice). Results were pooled from two independent experiments. A log-rank Mantel–Cox test was used for statistical analysis. d, Percentage of peritoneal macrophages expressing iNOS at 18 h following CAR T cell transfer. Mice were treated with isotype, mIL-6-blocking antibody, anakinra or mIL-6-blocking antibody + anakinra (Tumor only, n = 4 mice; Isotype, n = 3 mice; Anti-mIL-6, n = 3 mice; Anakinra, n = 3 mice; Anti-mIL-6 + Anakinra, n = 4 mice). Measurement of the percentage of iNOS+ macrophages with isotype or anti-mIL-6 blockade was performed in two independent experiments. Measurement of the percentage of iNOS+ macrophages with isotype, anakinra or anakinra + anti-mIL-6-blocking antibody was performed once. A one-way ANOVA was used for statistical analysis. e, Levels of mIL-1Ra in supernatants of 1928z-LNGFR and 1928z mIL1Ra transduced CAR T cells after 48 h in culture as determined by ELISA. One sample was analyzed in duplicate for each CAR construct. Measurement of IL-1Ra in supernatants of 1928z-mIL1Ra CAR T cells was performed once. f, Percentage survival of tumor-bearing mice after 1928z-LNGFR or 1928zmIL1Ra CAR T cell transfer (1928z-LNGFR, n = 22 mice; 1928z-mIL1Ra, n = 18 mice). Results are from three pooled independent experiments. A log-rank Mantel–Cox test was used for statistical analysis. g, Serum levels of murine cytokines from mice surviving CRS at 18 h following CAR T cell transfer. Tumor-bearing mice received 1928z-LNGFR or 1928zmIL1Ra CAR T cells (1928z-LNGFR, n = 3 mice; 1928z-mIL1Ra, n = 3 mice). Results are representative of three independent experiments. A two-tailed unpaired two-sample t-test was used for statistical analysis. h, Percentage of survival of NSG mice receiving 0.2 × 106 or 0.5 × 106 1928z- LNGFR or 1928z-mIL1Ra CAR T cells. Tumors were injected intravenously on Day –4, and CAR T cells were injected on Day 0 (Tumor only, n = 4 mice; 0.2 × 1061928z- LNGFR, n = 7 mice; 0.2 × 106 1928z-mIL-1Ra, n = 7 mice; 0.5 × 106 1928z-LNGFR, n = 11 mice; 0.5 × 106 1928z-mIL-1Ra, n = 11 mice). The experiment was performed once. A log-rank Mantel–Cox test was used for statistical analysis. All data are shown as mean ± s.e.m. Numbers on the graphs are P values.

To prevent CRS mortality without exogenous intervention, we engineered CAR T cells to constitutively produce IL-1 receptor antagonist (IL1RN/IL-1Ra) (Fig. 4e and Supplementary Fig. 8a). First, we demonstrated that this new construct protects against CRS-associated mortality (Fig. 4f and Supplementary Fig. 8b). Importantly, CAR T cell–derived serum cytokine levels were unaffected (Fig. 4g), consistent with the blockade of endogenous CRS pathways independent of CAR T cell activation. Next, we assessed whether antitumor activity might be affected by evaluating 1928z-mIL1Ra T cells in the CAR ‘stress test’ at limiting CAR T cell doses in NSG mice32. At two different dose levels, 1928z-mIL1Ra matched the therapeutic efficacy of control 1928z-LNGFR CAR T cells (Fig. 4h and Supplementary Fig. 8c,d). Therefore, we not only identified a new actionable target for CRS, but we also designed a CAR construct that autonomously prevents CRS-associated mortality in mice without compromising antitumor efficacy.

We demonstrate here that recipient myeloid cells play a critical role in the pathogenesis of CRS. We establish that tumor-activated CAR T cells recruit and activate myeloid cells, and that colocalized myeloid cells are the main source of IL-6. Macrophages contribute to CRS pathophysiology through their production of IL-6 and iNOS. The aggravation of CRS by CAR T cells expressing mCD40L demonstrates that CAR T cells and macrophages can functionally interact within the tumor microenvironment. Whether CAR T cell–macrophage contact is required for CRS remains to be determined, as CRS could develop in our model in the absence of a functional CD40–CD40L axis. CAR T cells may activate myeloid cells through a variety of other pathways including cytokines, other cell contact–dependent pathways and Toll-like receptor stimulation. Our findings on the importance of CAR T cell–macrophage proximity suggest that the incidence and severity of CRS in CAR therapies will at least in part depend on the extent of myeloid infiltration in the targeted tumor, which may be variable among hematological and solid tumors33.

Selectively modulating macrophage activity with CD40L, iNOS inhibitors or anakinra revealed the central role played by macrophages and provides insights for new therapeutic interventions. We identified IL-1 as a new actionable target suitable to treat severe CRS and diminish its severity. Although circulating IL-1 levels were below detection, as observed in CD19 CAR T cell recipients undergoing CRS11, it is noteworthy that other IL-1-driven inflammatory conditions associated with low or undetectable levels of circulating IL-1 have been found to resolve with the administration of anakinra34.

Our findings that both induction of IL-1 signaling and of iNOS expression are critical determinants of severe CRS potentially explain why SCID-beige mice develop more severe CRS. SCID-beige mice and NSG mice differ in their genetic background (C.B17-SCID, BALB/c and NOD/LtSz-SCID, NOD/LT, respectively), and the impaired IL-1 response to IFN-γ priming and lipopolysaccharide (LPS) stimulation is found in the NSG background35. It is plausible that additional developmental and maturation defects of monocytes and macrophages in NOD mice36 contribute to reduced macrophage reactivity and diminished CRS in NSG mice. Moreover, NSG mice lack the common γc chain receptor, which is essential for signaling through IL-2, IL-15 and other cytokines. Peritoneal macrophages lacking γc are defective in NO production primed by IL-15 signaling37, a cytokine that is detected in our mouse model.

Our findings informed the design of an IL-1Ra-secreting CAR construct that can demonstrably prevent CRS-related mortality while maintaining intact antitumor efficacy. The benefits of an IL-1 blockade through IL-1Ra are especially intriguing given the latter’s ability to cross the blood–brain barrier38, unlike tocilizumab9. As human microglia activated by IL-1 may produce iNOS and proinflammatory cytokines39,40, blocking IL-1 could potentially not only protect from severe CRS, but also could reduce the severity of CAR T cell–related neurotoxicity.

In addition to the toxicity in itself, CRS carries a high cost because of the stringent patient monitoring and supportive therapy it may require. CRS toxicity and its management costs presently hamper broad use of CAR therapy. This study demonstrates that CAR T cells may be engineered to mitigate these burdens without requiring exogenous intervention.

Methods

Cell culture

Burkitt Lymphoma Raji cells and NALM-6 pre-B-ALL cells were obtained from ATCC. Raji GFP–firefly luciferase (FLuc) and NALM-6-GFP-FLuc cells were cultured in RPMI (Invitrogen) supplemented with 10% FBS (HyClone), 10 mM HEPES (Invitrogen), l-glutamine 2 mM (Invitrogen), NEAA 1 × (Invitrogen), 0.55 mM β-mercaptoethanol, 1 mM sodium pyruvate (Invitrogen), penicillin–streptomycin 50 U/ml (Invitrogen). Raji and NALM-6 cells were routinely tested for mycoplasma and found to be negative.


Source and handling of T cells

Buffy coats from anonymous healthy donors were purchased from the New York Blood Center (IRB-exempted) and were handled following all required ethical and safety procedures. The researchers were blind to any covariate characteristics.


Isolation and culture of T cells

Primary human T cells were purified from buffy coats of healthy donors by negative magnetic selection (Pan T Cell Isolation Kit, Miltenyi). Purified T cells were cultured in XVIVO 15 (Lonza), supplemented with 5% human serum type AB (Gemini), 10 mM HEPES, 2 mM GlutaMax (Invitrogen), 1 × MEM vitamin solution (Invitrogen), 1 mM sodium pyruvate (Invitrogen), penicillin–streptomycin 50 U/ml (Invitrogen) and 60 U/ml recombinant IL-2.


Mice

Mice were treated under a protocol approved by the Memorial Sloan Kettering Cancer Center (MSKCC) Institutional Animal Care and Use Committee. All relevant animal use guidelines and ethical regulations were followed. For the CRS model, 6- to 8-week-old female C.B.Igh-1b/GbmsTac-PrkdcscidLystbgN7 (SCID-beige) mice (Taconic) were intraperitoneally injected with 3 million Raji-GFP-FLuc cells, and tumors were left to grow for 20 d. Tumor burden was evaluated by in vivo bioluminescent imaging 2 d before CAR T cell transfer. Outlier mice with extreme burdens (either too high or too low compared to most) were excluded from the experiment before CAR T cell infusion. No mice were excluded afterwards at any point. Mice were injected intraperitoneally with 30 million CAR+ T cells in PBS supplemented with 2% human serum. Control mice received PBS supplemented with 2% human serum. In the stress-test model, 6- to 8-week-old male NOD.Cg-PrkdcscidIl2rgtmWjl/SzJ (NSG) mice (Jackson Laboratory) were inoculated with 0.5 × 106 NALM-6-GFP-Fluc cells by tail vein injection followed by 0.2 × 106 or 0.5 × 106 CAR T cells 4 d later. Bioluminescence imaging utilized the Xenogen IVIS Imaging System (Xenogen) with Living Image software (Xenogen) for acquisition of imaging datasets. Tumor burden was assessed as previously described41.


Mouse treatment

Anakinra was administered intraperitoneally at a dose of 30 mg per kg body weight once per day for 5 d, starting 5 h before CAR T cell transfer. Control mice received PBS. Anti-mIL-6 (clone MP5-20F3, BioXcell) or isotype (Rat IgG1, clone HRPN, BioXcell), anti-mIL-6R (clone 15A7, BioXcell) or isotype (Rat IgG2b, clone LTF-2, BioXcell) and anti-mIL-1b (clone B122, BioXcell) or isotype (Armenian Hamster IgG, clone Armenian Hamster IgG, BioXcell) were administered intraperitoneally once per day at 25 mg per kg body weight for the first dose and 12.5 mg per kg body weight for subsequent doses for 5 d starting 5 h before CAR T cell transfer. L-NIL (Enzo Life Sciences) or 1400 W (Cayman Chemical) were administered intraperitoneally at 5 mg per kg body weight once per day for 5 d starting 5 h before CAR T cell transfer or before PBS (vehicle) treatment for control mice.


Flow cytometry

Antibodies were titrated for optimal staining. The following fluorophore-conjugated antibodies were used (‘h’ prefix denotes anti-human, ‘m’ prefix denotes anti-mouse): hCD4 BUV395 (clone RPA-T4, BD, cat. no. 564724), hCD8 PE-Cy7 (clone SK1, eBioscience, cat. no. 25-0087-42), hCD3 PerCP-eFluor710 (clone OKT3, eBioscience, cat. no. 46-0037-42), hCD19 BUV737 (clone SJ25C1, BD, cat. no. 564303), hLNGFR BB515 (clone C40-1457, BD,cat. no. 564580), hCD40L BV421 (clone TRAP1, BD, cat. no. 563886), mF4/80 BV421 and BV711 (clone T45-2342, BD, cat. nos. 565411 and 565612, respectively), mLy6C Alexa Fluor 647 and BV786 (clone ER-MP20, AbdSerotec and clone HK1.4, BioLegend respectively, cat. nos. MCA2389A647 and 128041, respectively), mGR-1-APC-Cy7 (clone RB6-8C5, BD, cat. no. 557661), mMHCII BB515 (clone 2G9, BD, cat. no. 565254), mCD11c BV650 (clone N418, BioLegend, cat. no. 117339), mLy6G APC-Fire750 (clone 1A8, BioLegend, cat. no. 127652), mSIGLEC-F PE-CF594 (clone E50-2440, BD, cat. no. 562757), mCD40 BV786 (clone 3/23, BD, cat. no. 740891), mCD40L PE (clone MR1, BD, cat. no. 106506), mCD11b BUV395 (clone M1/70, BD, cat. no. 563553) and mNOS2 PE-Cy7 (clone CXNFT, eBioscience, cat. no. 25-5920-82). For flow cytometry with live cells 7-AAD (BD) was used as a viability dye. For flow cytometry with fixed cells eFluor 506 fixable viability dye (eBioscience) was used. Fc receptors were blocked using Fc Receptor Binding Inhibitor Antibody Human (eBioscience) and Fc Block Mouse (Miltenyi). Cells were fixed using the Intracellular Fixation and Permeabilization Buffer Set (eBioscience) according to the manufacturer’s instructions. For CAR staining, an Alexa Fluor 647–conjugated goat anti-mouse antibody was used (Jackson Immunoresearch). For cell counting, Countbrite beads were used (Invitrogen) according to the manufacturer’s instructions.


Retroviral vector constructs and retroviral production

The 1928z-LNGFR construct has been previously described42. 1928z-mCD40L and 1928z-mIL1RN were prepared using standard molecular biology techniques. To obtain the 1928z-mCD40L construct, the cDNA for mCD40L was inserted in the place of LNGFR. To obtain the 1928z-mIL-1Ra construct, the cDNA for mIL-1Ra was inserted in the place of LNGFR. Plasmids encoding the SFG γ-retroviral (RV) vector43 were prepared as previously described42. gpg29 (H29) cells were transfected to produce VSV-G pseudotyped retroviral supernatants. These supernatants were used to transduce stable retroviral-producing 293 T cell lines as previously described44. Retroviral supernatants carrying the various CAR genetic constructs harvested from 293 T cell lines were used to transduce human T cells. T cells were activated with CD3/CD28 T cell Activator Dynabeads (Invitrogen) immediately after purification at a 1:1 bead-to-cell ratio. After 48 h of bead activation, T cells were transduced with retroviral supernatants by centrifugation on Retronectin (Takara)–coated plates in order to obtain 1928z-LNGFR, 1928z-mCD40L or 1928z-mIL-1Ra CAR T cells. Transduction efficiency was verified 3 d later by flow cytometry. CAR T cells were injected in mice 7 d after the first T cell activation.


Serum collection

Blood was collected from mice by tail-clip or by retro-orbital bleeding. After collection blood was left to clot for 30 min at room temperature. Following clotting blood was centrifuged at 6,000 g for 10 min at 4 °C. Serum was aliquoted in tubes in order to prevent multiple freeze–thaw cycles and was immediately stored at –80 °C until analysis.


Cytokine measurements

Serum and plasma cytokines were measured using cytometric bead arrays (BD) or ELISA kits for mouse IL-1Ra (Thermo-Fisher), mouse SAA3 (Millipore) and mIL-15–IL-15R complex (Thermo-Fisher), per the manufacturer’s instructions.


Animal pathology

Mice were transferred to the pathology core facility of Memorial Sloan Kettering, where they were euthanized by cardiac puncture. Tissues obtained were fixed in 10% buffered formalin and were further processed for H&E staining and immunohistochemistry.


RNA extraction and transcriptome sequencing

Cells were sorted directly into 750 µl of TRIzol LS (Invitrogen). The volume was adjusted to 1 ml with PBS, and extraction was performed according to instructions provided by the manufacturer. After ribogreen quantification and quality control with an Agilent BioAnalyzer, total RNA underwent amplification using the SMART-seq V4 (Clonetech) Ultra Low Input RNA kit for sequencing. For 2–10 ng of total RNA, 12 cycles of amplification were performed. For a lesser amount (0.13–2 ng), 13 cycles of amplification were performed. Subsequently, 10 ng of amplified cDNA was used to prepare Illumina HiSeq Libraries with the Kapa DNA library preparation chemistry (Kapa Biosystems) using eight cycles of PCR. Samples were barcoded and run on HiSeq 2500 1 T, in a 50bp–50 bp paired-end run, using the TruSeq SBS Kit v3 (Illumina). An average of 38.5 million paired reads were generated per sample, and the percent of mRNA bases was over 77% on average.


RNA-seq analysis

The output FASTQ data files were mapped (two-pass method) to the target genome (MM10 assembly) using the STAR RNA aligner, resolving reads across splice junctions (ENSEMBL assembly). The first mapping pass uses a list of known annotated junctions from ENSEMBL. New junctions found in the first pass are then added to the known junctions, after which a second mapping pass is performed using the RemoveNoncanoncial flag. After mapping, the output SAM files were postprocessed using the Picard Tools utility AddOrReplaceReadGroups to add read groups. The files were then sorted and coverted to BAM format. The expression count matrix for the mapped reads was then computed using HTSeq. Finally, DESeq was used to normalize the full dataset and analyze differential expression between sample groups. The program version was: HTSEQ, htseq/HTSeq-0.5.3; Picard, picard/picard-tools-1.124; R, R/R-3.2.0; STAR, star/STAR-STAR_2.5.0a; SAMtools samtools/samtools-0.1.19.


Statistical analysis

Statistical analyses were performed using GraphPad Prism V7 for Mac. Statistical tests included the two-tailed unpaired two-sample t-test; the one-way ANOVA; the two-way ANOVA; the log-rank Mantel–Cox test and the binomial test with FDR-adjusted P values. The statistical test used for each figure is described in the corresponding figure legend.


Reporting Summary

Further information on experimental design is available in the Nature Research Reporting Summary.


Materials availability

Materials are available upon request to the corresponding author.


Data availability

Data presented in Figs. 2e–f and 4a, b can be accessed under accession number GSE111236.

Additional information

Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  1. 1.

    Sadelain, M., Rivière, I. & Riddell, S. Therapeutic T cell engineering. Nature 545, 423–431 (2017).

  2. 2.

    Davila, M. L. et al. Efficacy and toxicity management of 19-28z CAR T cell therapy in B cell acute lymphoblastic leukemia. Sci. Transl. Med. 6, 224ra25 (2014).

  3. 3.

    Maude, S. L. et al. Chimeric antigen receptor T cells for sustained remissions in leukemia. N. Engl. J. Med. 371, 1507–1517 (2014).

  4. 4.

    Park, J. H. et al. Long-term follow-up of CD19 CAR therapy in acute lymphoblastic leukemia. N. Engl. J. Med. 378, 449–459 (2018).

  5. 5.

    Barrett, D. M., Teachey, D. T. & Grupp, S. A. Toxicity management for patients receiving novel T-cell engaging therapies. Curr. Opin. Pediatr. 26, 43–49 (2014).

  6. 6.

    Bonifant, C. L., Jackson, H. J., Brentjens, R. J. & Curran, K. J. Toxicity and management in CAR T-cell therapy. Mol. Ther. Oncolytics 3, 16011 (2016).

  7. 7.

    Brudno, J. N. & Kochenderfer, J. N. Toxicities of chimeric antigen receptor T cells: recognition and management. Blood 127, 3321–3330 (2016).

  8. 8.

    Oluwole, O. O. & Davila, M. L. At the bedside: clinical review of chimeric antigen receptor (CAR) T cell therapy for B cell malignancies. J. Leukoc. Biol. 100, 1265–1272 (2016).

  9. 9.

    Neelapu, S. S. et al. Chimeric antigen receptor T-cell therapy—assessment and management of toxicities. Nat. Rev. Clin. Oncol. 15, 47–62 (2018).

  10. 10.

    Lee, D. W. et al. T cells expressing CD19 chimeric antigen receptors for acute lymphoblastic leukaemia in children and young adults: a phase 1 dose-escalation trial. Lancet 385, 517–528 (2015).

  11. 11.

    Teachey, D. T. et al. Identification of predictive biomarkers for cytokine release syndrome after chimeric antigen receptor T-cell therapy for acute lymphoblastic leukemia. Cancer Discov. 6, 664–679 (2016).

  12. 12.

    Hay, K. A. et al. Kinetics and biomarkers of severe cytokine release syndrome after CD19 chimeric antigen receptor-modified T-cell therapy. Blood 130, 2295–2306 (2017).

  13. 13.

    Cray, C., Zaias, J. & Altman, N. H. Acute phase response in animals: a review. Comp. Med. 59, 517–526 (2009).

  14. 14.

    Meek, R. L., Eriksen, N. & Benditt, E. P. Murine serum amyloid A3 is a high density apolipoprotein and is secreted by macrophages. Proc. Natl Acad. Sci. USA 89, 7949–7952 (1992).

  15. 15.

    Vandevelde, M., Higgins, R. & Overmann A. Veterinary Neuropathology: Essentials of Theory and Practice. in Veterinary Neuropathology: Essentials of Theory and Practice 29 (Wiley, Hoboken, NJ, USA, 2012).

  16. 16.

    Gust, J. et al. Endothelial activation and blood–brain barrier disruption in neurotoxicity after adoptive immunotherapy with CD19 CAR-T cells. Cancer Discov. 7, 1404–1419 (2017).

  17. 17.

    Rincon, M. Interleukin-6: from an inflammatory marker to a target for inflammatory diseases. Trends Immunol. 33, 571–577 (2012).

  18. 18.

    Quezada, S. A., Jarvinen, L. Z., Lind, E. F. & Noelle, R. J. CD40/CD154 interactions at the interface of tolerance and immunity. Annu. Rev. Immunol. 22, 307–328 (2004).

  19. 19.

    Bossen, C. et al. Interactions of tumor necrosis factor (TNF) and TNF receptor family members in the mouse and human. J. Biol. Chem. 281, 13964–13971 (2006).

  20. 20.

    Tucker, T. A. & Schwiebert, L. M. CD40 ligation decreases its protein half-life at the cell surface. Eur. J. Immunol. 38, 864–869 (2008).

  21. 21.

    Wang, H. et al. A novel role of the scaffolding protein JLP in tuning CD40-induced activation of dendritic cells. Immunobiology 218, 835–843 (2013).

  22. 22.

    Wang, H. M. et al. Scaffold protein JLP is critical for CD40 signaling in B lymphocytes. J. Biol. Chem. 290, 5256–5266 (2015).

  23. 23.

    Elgueta, R. et al. Molecular mechanism and function of CD40/CD40L engagement in the immune system. Immunol. Rev. 229, 152–172 (2009).

  24. 24.

    Murray, P. J. & Wynn, T. A. Protective and pathogenic functions of macrophage subsets. Nat. Rev. Immunol. 11, 723–737 (2011).

  25. 25.

    Ohashi, Y. et al. Hypotension and reduced nitric oxide–elicited vasorelaxation in transgenic mice overexpressing endothelial nitric oxide synthase. J. Clin. Invest. 102, 2061–2071 (1998).

  26. 26.

    Titheradge, M. A. Nitric oxide in septic shock. Biochim. Biophys. Acta 1411, 437–455 (1999).

  27. 27.

    Moore, W. M. et al. L-N6-(1-iminoethyl)lysine: a selective inhibitor of inducible nitric oxide synthase. J. Med. Chem. 37, 3886–3888 (1994).

  28. 28.

    Garvey, E. P. et al. 1400W is a slow, tight binding, and highly selective inhibitor of inducible nitric-oxide synthase in vitro and in vivo. J. Biol. Chem. 272, 4959–4963 (1997).

  29. 29.

    Dinarello, C. A. Immunological and inflammatory functions of the interleukin-1 family. Annu. Rev. Immunol. 27, 519–550 (2009).

  30. 30.

    Saini, A. S., Shenoy, G. N., Rath, S., Bal, V. & George, A. Inducible nitric oxide synthase is a major intermediate in signaling pathways for the survival of plasma cells. Nat. Immunol. 15, 275–282 (2014).

  31. 31.

    Sims, J. E. & Smith, D. E. The IL-1 family: regulators of immunity. Nat. Rev. Immunol. 10, 89–102 (2010).

  32. 32.

    Zhao, Z. et al. Structural design of engineered costimulation determines tumor rejection kinetics and persistence of CAR T cells. Cancer Cell 28, 415–428 (2015).

  33. 33.

    van der Stegen, S. J. et al. Preclinical in vivo modeling of cytokine release syndrome induced by ErbB-retargeted human T cells: identifying a window of therapeutic opportunity? J. Immunol. 191, 4589–4598 (2013).

  34. 34.

    Dinarello, C. A., Simon, A. & van der Meer, J. W. Treating inflammation by blocking interleukin-1 in a broad spectrum of diseases. Nat. Rev. Drug Discov. 11, 633–652 (2012).

  35. 35.

    Shultz, L. D. et al. Multiple defects in innate and adaptive immunologic function in NOD/LtSz-scid mice. J. Immunol. 154, 180–191 (1995).

  36. 36.

    Serreze, D. V., Gaedeke, J. W. & Leiter, E. H. Hematopoietic stem-cell defects underlying abnormal macrophage development and maturation in NOD/Lt mice: defective regulation of cytokine receptors and protein kinase C. Proc. Natl Acad. Sci. USA 90, 9625–9629 (1993).

  37. 37.

    Ohteki, T., Suzue, K., Maki, C., Ota, T. & Koyasu, S. Critical role of IL-15–IL-15R for antigen-presenting cell functions in the innate immune response. Nat. Immunol. 2, 1138–1143 (2001).

  38. 38.

    Gutierrez, E. G., Banks, W. A. & Kastin, A. J. Blood-borne interleukin-1 receptor antagonist crosses the blood–brain barrier. J. Neuroimmunol. 55, 153–160 (1994).

  39. 39.

    Liu, J., Zhao, M. L., Brosnan, C. F. & Lee, S. C. Expression of type II nitric oxide synthase in primary human astrocytes and microglia: role of IL-1β and IL-1 receptor antagonist. J. Immunol. 157, 3569–3576 (1996).

  40. 40.

    Tarassishin, L., Suh, H. S. & Lee, S. C. LPS and IL-1 differentially activate mouse and human astrocytes: role of CD14. Glia 62, 999–1013 (2014).

  41. 41.

    Gade, T. P. et al. Targeted elimination of prostate cancer by genetically directed human T lymphocytes. Cancer Res. 65, 9080–9088 (2005).

  42. 42.

    Maher, J., Brentjens, R. J., Gunset, G., Rivière, I. & Sadelain, M. Human T-lymphocyte cytotoxicity and proliferation directed by a single chimeric TCRζ /CD28 receptor. Nat. Biotechnol. 20, 70–75 (2002).

  43. 43.

    Rivière, I., Brose, K. & Mulligan, R. C. Effects of retroviral vector design on expression of human adenosine deaminase in murine bone marrow transplant recipients engrafted with genetically modified cells. Proc. Natl Acad. Sci. USA 92, 6733–6737 (1995).

  44. 44.

    Gong, M. C. et al. Cancer patient T cells genetically targeted to prostate-specific membrane antigen specifically lyse prostate cancer cells and release cytokines in response to prostate-specific membrane antigen. Neoplasia 1, 123–127 (1999).

Download references

Acknowledgements

We thank the Alexander S. Onassis Public Benefit Foundation for their support (T.G.). We thank the following MSK core facilities for their outstanding support: flow cytometry core facility, laboratory of comparative pathology, animal facility, integrated genomics operation and bioinformatics core. We thank G. Gunset, Z. Zhao, A. Dobrin and P. Lindenbergh for their assistance with some experiments. This study was supported by Juno Therapeutics and the MSK Cancer Center Support Grant/Core Grant (P30 CA008748).

Author information

Affiliations

  1. Center for Cell Engineering and Immunology Program, Sloan Kettering Institute, New York, NY, USA

    • Theodoros Giavridis
    • , Sjoukje J. C. van der Stegen
    • , Justin Eyquem
    • , Mohamad Hamieh
    •  & Michel Sadelain
  2. Laboratory of Comparative Pathology, Rockefeller University, Weill Cornell Medicine and Memorial Sloan Kettering Cancer Center, New York, NY, USA

    • Alessandra Piersigilli

Authors

  1. Search for Theodoros Giavridis in:

  2. Search for Sjoukje J. C. van der Stegen in:

  3. Search for Justin Eyquem in:

  4. Search for Mohamad Hamieh in:

  5. Search for Alessandra Piersigilli in:

  6. Search for Michel Sadelain in:

Contributions

T.G. designed the study, performed experiments, analyzed and interpreted the data and wrote the manuscript. S.J.C.v.d.S. helped design and perform in vivo experiments. J.E. performed in vivo experiments. M.H. performed in vivo experiments. A.P. performed and interpreted histopathological studies. M.S. designed the study, analyzed and interpreted the data and wrote the manuscript.

Competing interests

A patent application on CRS prevention listing T.G. and M.S. as co-inventors has been filed by MSK.

Corresponding author

Correspondence to Michel Sadelain.

Supplementary information

About this article

Publication history

Received

Accepted

Published

DOI

https://doi.org/10.1038/s41591-018-0041-7

Further reading