Research ArticleGraft-Versus-Host Disease

Graft-versus-host disease, but not graft-versus-leukemia immunity, is mediated by GM-CSF–licensed myeloid cells

See allHide authors and affiliations

Science Translational Medicine  28 Nov 2018:
Vol. 10, Issue 469, eaat8410
DOI: 10.1126/scitranslmed.aat8410

Keeping the baby but not the bathwater

The reconstituted immune system of an allogeneic hematopoietic cell transplant patient is responsible for preventing leukemia relapse by means of graft-versus-leukemia (GvL) activity. This same reconstituted immune system can turn on the recipient in the form of graft-versus-host disease (GvHD). Tugues et al. used multiple mouse transplant models to decipher the mechanisms driving GvHD and GvL. They discovered that GM-CSF–activated myeloid cells mediated GvHD, but did not contribute to GvL, which is largely carried out by T cells. These results suggest that blocking GM-CSF in hematopoietic cell transplant recipients may be able to specifically target cells that promote GvHD, without hampering those that enact GvL.

Abstract

Allogeneic hematopoietic cell transplantation (allo-HCT) not only is an effective treatment for several hematologic malignancies but can also result in potentially life-threatening graft-versus-host disease (GvHD). GvHD is caused by T cells within the allograft attacking nonmalignant host tissues; however, these same T cells mediate the therapeutic graft-versus-leukemia (GvL) response. Thus, there is an urgent need to understand how to mechanistically uncouple GvL from GvHD. Using preclinical models of full and partial MHC-mismatched HCT, we here show that the granulocyte-macrophage colony-stimulating factor (GM-CSF) produced by allogeneic T cells distinguishes between the two processes. GM-CSF drives GvHD pathology by licensing donor-derived phagocytes to produce inflammatory mediators such as interleukin-1β and reactive oxygen species. In contrast, GM-CSF did not affect allogeneic T cells or their capacity to eliminate leukemic cells, retaining undiminished GvL responses. Last, tissue biopsies and peripheral blood mononuclear cells from patients with grade IV GvHD showed an elevation of GM-CSF–producing T cells, suggesting that GM-CSF neutralization has translational potential in allo-HCT.

INTRODUCTION

For patients suffering from hematological malignancies, allogeneic hematopoietic cell transplantation (allo-HCT) is a potentially curative and life-saving intervention. However, between 40 and 60% of all patients will develop clinically acute or chronic graft-versus-host disease (GvHD), which together carry a mortality rate of about 50% (1, 2). Donor-derived alloreactive T cells attack host malignant cells, producing the beneficial graft-versus-leukemia (GvL) effect. However, this same alloreactivity can be targeted toward healthy tissues (typically the skin, gut, and liver), leading to GvHD. Although depleting T cells from the donor material before allo-HCT can prevent or reduce GvHD, this comes at the cost of decreased GvL activity and increased relapse rates (3). Therefore, there is an urgent need to understand how the mechanisms of GvL and GvHD can be separated at the T cell level and modulated for clinical benefit.

Much work has attempted to define key T cell subsets and cytokines that underpin GvHD in murine models, but to date, few consistent conclusions have been drawn. Although GvHD was originally proposed to be a T helper 1 (TH1)–mediated pathology (4), studies in mice showed that donor T cells deficient in the TH1 cytokine interferon-γ (IFN-γ) can exacerbate the disease (5). TH2 cells can both suppress experimental GvHD (6, 7) and induce GvHD affecting the liver and skin (8, 9). Similarly, although there is evidence that interleukin-17A (IL-17A)–producing T cells can mediate immunopathology in inflammatory diseases (10), their role in GvHD remains controversial, as in murine models IL-17 seems able to either promote (11, 12) or ameliorate GvHD (9), depending on the experimental conditions. Together, polarized TH cells are implicated in the emergence and perpetuation of GvHD, but so far, it has not been possible to identify any specific soluble mediator that has a reproducible and nonredundant function in the pathogenesis of the disease in both murine models and in humans.

One cytokine with an emerging role across a range of inflammatory disorders is granulocyte-macrophage colony-stimulating factor (GM-CSF also known as Csf2) (13, 14). Despite its original designation as a hematopoietic growth factor, mice lacking GM-CSF or its receptor develop normally and have a functional myeloid cell compartment, while exhibiting specific defects in pulmonary immunophysiology (15) and in the homeostasis of CD103+ and CD11b+ dendritic cells (DCs) in nonlymphoid tissues (16). However, GM-CSF produced by activated TH cells has a profound impact on the differentiation and activation of several myeloid cell subsets during pathologic tissue inflammation (13, 17, 18) and can drive myeloid cell cytokine production (19), phagocytosis (20), and oxidative burst (21). We therefore asked whether GM-CSF plays a role in the inflammatory tissue damage of GvHD after allo-HCT.

Using murine models of major histocompatibility complex (MHC)–mismatched HCT, we here show that T cell–derived GM-CSF drives GvHD without compromising alloreactive T cell control of tumors (GvL), thus uncoupling phagocyte-mediated immunopathology from lymphocyte-mediated control of cancer cells. Therefore, we propose GM-CSF as a potential therapeutic target to attenuate GvHD while maintaining GvL in patients receiving allo-HCT.

RESULTS

Tissue-infiltrating donor T cells produce GM-CSF and IFN-γ after MHC-mismatched HCT

To understand the contribution of T cell–derived GM-CSF to acute GvHD, we first used a model of MHC-mismatched HCT. We lethally irradiated CD45.2+ wild-type (WT) BALB/c mice (MHC haplotype H2d) and then intravenously injected them with T cell–depleted (TCD) bone marrow (BM) cells from CD45.1+ WT C57BL/6 (B6) mice (MHC haplotype H2b), with or without CD45.1+ B6 splenocytes, which served as a source of mature T cells. Recipients of BM plus splenocytes developed fatal GvHD between days 3 and 6 after allo-HCT (Fig. 1A), which coincided with the reconstitution and expansion of donor CD45.1+ cells in the spleen, liver, and skin (Fig. 1B). Within the CD45.1+ donor compartment, T cell populations contained high frequencies of GM-CSF– and IFN-γ–producing cells, seen first in the liver (Fig. 1C) and spleen (Fig. 1D) from day 3 after allo-HCT and then in the skin from day 6 (Fig. 1D). IL-17A was barely detectable in these populations (Fig. 1, C and D), but about half of GM-CSF–producing T cells coexpressed high amounts of IFN-γ (Fig. 1E). Accordingly, we found substantially higher amounts of GM-CSF and IFN-γ in sera from mice receiving BM plus splenocytes compared to BM alone (Fig. 1F). We confirmed the production of GM-CSF and IFN-γ during the allogeneic response in vitro using allogeneic mixed lymphocyte reactions (MLRs) (Fig. 1G). Furthermore, T cells from B6 mice lacking GM-CSF (Csf2−/−), IFN-γ (Ifng−/−), or IL-17A (Il17a−/−) proliferated equally after coculture with allogeneic DCs, excluding a priori differences in the proliferative potential of these cells (Fig. 1H). Together, GM-CSF and IFN-γ, but not IL-17A, are produced by T cells infiltrating target organs during GvHD.

Fig. 1 Donor T cells secrete GM-CSF and IFN-γ during allogeneic responses.

(A) Survival of lethally irradiated CD45.2+ BALB/c mice after allo-HCT with CD45.1+ WT C57BL/6 TCD-BM alone or combined with 10 × 106 CD45.1+ WT C57BL/6 splenocytes. Data were pooled from four individual experiments, each with n = 5 per group. (B) Frequency of CD45.1+ cells within live singlets in the spleen, liver, and skin at 3 and 6 days after allo-HCT. Data were pooled from two experiments to obtain n = 8 to 10 per group. (C) Frequency of IFN-γ–, GM-CSF–, and IL-17A–producing CD4+ and CD8+ T cells within the CD45.1+ population from spleens of mice 3 days after allo-HCT. Representative plots are shown from a total of three independent experiments. (D) Frequencies of IFN-γ–, GM-CSF–, and IL-17A–producing CD4+ and CD8+ T cells within the CD45.1+ populations from liver, spleen, and skin 3 and 6 days after allo-HCT. Data were pooled from three individual experiments, total n = 8 to 10 per group. (E) Frequencies of IFN-γ– and GM-CSF–producing CD4+ and CD8+ T cells within the CD45.1+ populations from liver and spleen 3 and 6 days after allo-HCT. Representative data from one of three experiment are shown, each with n = 3 to 4 per group. (F) Serum IFN-γ and GM-CSF in mice 6 days after allo-HCT. Data were pooled from three individual experiments, each with n = 2 to 5 per group. (G) IFN-γ, GM-CSF, and IL-17A in supernatants from cocultures of T cells from WT C57BL/6 mice with either syngeneic (C57BL/6) or allogeneic (BALB/c) splenic CD11c+ DCs. Data were pooled from three individual experiments, each with n = 3 to 5 per group. (H) Tritiated thymidine incorporation by T cells from C57BL/6 WT, Ifng−/−, Csf2−/−, or Il17a/− mice cocultured with syngeneic (C57BL/6) or allogeneic (BALB/c) splenic CD11c+ DCs. Data were pooled from three individual experiments, each with n = 2 to 5 per group. For comparison of survival curves, a log-rank (Mantel-Cox) test was used in (A). For comparison of the means, an unpaired two-tailed t test with Welch’s correction was used in (B), (F), and (G), and one-way analysis of variance (ANOVA) with Bonferroni posttest was used in (D) and (H). *P < 0.05, **P < 0.01, ***P < 0.001. Data are shown as means ± SEM. cpm, counts per minute; ns, not significant.

GM-CSF production by allogeneic T cells is essential for GvHD pathology

Given the production of GM-CSF and IFN-γ by alloreactive TH cells, we assessed the relevance of these cytokines in determining the severity of acute GvHD. As described above, we lethally irradiated WT BALB/c mice and then compared the outcomes of injecting TCD-BM cells from WT B6 mice, with or without splenocytes from WT B6 mice or Csf2−/−, Ifng−/−, or Il17a−/− mice. Mice receiving BM plus Csf2−/− splenocytes were protected from lethal GvHD up to 20 days after allo-HCT, whereas all mice treated with BM plus WT splenocytes had died or reached final termination criteria by day 15 after HCT (Fig. 2A). Injecting mice with BM plus Ifng−/− or Il17a−/− splenocytes also significantly affected the timing of fatal GvHD, but still, almost 100% of these mice died by day 20 (Fig. 2A).

Fig. 2 GM-CSF is crucial for acute GvHD after fully MHC-mismatched allo-HCT.

(A) Survival of lethally irradiated BALB/c mice after allo-HCT with WT C57BL/6 TCD-BM alone or with 10 × 106 splenocytes from C57BL/6 WT Csf2−/−, Ifng−/−, or Il17a−/− mice. Data were pooled from four individual experiments, each with n = 5 per group. (B) Survival of lethally irradiated BALB/c mice after allo-HCT with WT C57BL/6 TCD-BM alone or combined with 3 × 106 T cells purified from spleens of C57BL/6 WT or Csf2−/− mice. Data were pooled from three individual experiments, each with n = 5 per group. (C) Composite histopathological score for liver, small intestine, and skin from BALB/c mice 6 days after allo-HCT with WT C57BL/6 TCD-BM combined with 10 × 106 splenocytes from either WT or Csf2−/− C57BL/6 mice. Data were pooled from three individual experiments, each with n = 5 per group. (D) Representative images of hematoxylin and eosin–stained sections from skin and small intestine of mice 6 days after allo-HCT described in (A), each with n = 4 to 5 per group. Scale bars, 100 μm. (E) Colon length in centimeters from BALB/c mice 6 days after allo-HCT, as described in (C). Data were pooled from two individual experiments. (F) Representative images and quantification of apoptotic cells [TUNEL (terminal deoxynucleotidyl transferase–mediated deoxyuridine triphosphate nick end labeling) staining] in the colon from BALB/c mice 6 days after allo-HCT, as described in (C). Data were pooled from three individual experiments. Scale bars, 50 μm (top) and 20 μm (bottom). (G) Concentrations of AP, ALT, BUN, and albumin in serum from BALB/c mice 6 days after allo-HCT, as described in (C). Data are representative of two individual experiments, each with n = 3 to 5 per group. (H) Representative images of p22phox labeling in sections from liver, small intestine, and skin of mice 6 days after allo-HCT, as described in (C) (left). Scale bars, 100 μm. Data were pooled from four independent experiments, each with n = 5 to 7 per group. Quantification of mean percentage of total area labeled positively for p22phox per visual field (right). (I) Survival of lethally irradiated BALB/c mice after allo-HCT with WT C57BL/6 TCD-BM alone or combined with 10 × 106 splenocytes from C57BL/6 WT mice. Mice were treated with phosphate-buffered saline (PBS), 300 μg of isotype control antibody, or 300 μg of anti–GM-CSF antibody (a–GM-CSF) injected intraperioneally three times per week for the duration of the experiment, starting 2 days before HCT. Data were pooled from two individual experiments, each with n = 5 per group. For comparison of survival curves, a log-rank (Mantel-Cox) test was used in (A) (WT versus other groups), (B) (WT to Csf2−/−), and (I) (a–GM-CSF versus isotype). For comparison of the means (WT versus Csf2−/−), an unpaired two-tailed t test was used in (C), (E), (F), and (H). *P < 0.05, **P < 0.01, ***P < 0.001. Data are shown as means ± SEM.

Because donor splenocyte preparations contain multiple different immune cell types, we then asked whether T cells were the biologically relevant source of GM-CSF in our GvHD model. We injected lethally irradiated BALB/c recipients with TCD-BM cells from WT B6 donors, with or without purified T cells isolated from spleens of WT or Csf2−/− B6 mice. This approach confirmed that mice receiving Csf2−/− T cells were protected from lethal GvHD up to 70 days after allo-HCT, in stark contrast to those receiving WT T cells (Fig. 2B). There was no evidence that this protective effect was related to intrinsic differences between splenic T cells from WT and Csf2−/− mice because both CD4 and CD8 populations exhibited comparable frequencies (fig. S1A), activation status (fig. S1B), and suppressive capacity of regulatory T cells (Tregs) (fig. S1C). Also after HCT, the capacity to produce GM-CSF did not affect the ability of T cells to infiltrate GvHD target organs (fig. S1, D and E), the frequencies of activated (CD44high) T cells and Tregs in the spleen or peripheral lymph nodes (LNs; fig. S1F), their proliferative capacity and granzyme B production (fig. S1G), or their ability to produce proinflammatory cytokines such as IFN-γ, tumor necrosis factor–α (TNFα), or IL-17A (fig. S1H). The transfer of nonlethal doses of T cells confirmed the protective effect in the absence of GM-CSF, leading to only a mild GvHD but a significant difference in the clinical score between the groups receiving WT or Csf2−/− T cells (fig. S2, A and B).

We then examined the relative contribution of GM-CSF from CD4 and CD8 T cells to GvHD pathology. For this purpose, we performed an experiment using four combinations of CD4 and CD8 T cells transferred at a 1:1 ratio: CD4 WT T cells + CD8 WT T cells, CD4 WT T cells + CD8 Csf2−/− T cells, CD4 Csf2−/− T cells + CD8 WT T cells, and CD4 Csf2−/− T cells + CD8 Csf2−/− T cells. Only the combinations in which Csf2 was missing from the CD4 T cells conferred protection against lethal GvHD (fig. S2C), supporting the notion that alloreactive TH cells are a prominent source of GM-CSF in GvHD pathogenesis.

The survival advantage conferred to the recipients of allo-HCT of T cells lacking GM-CSF was also evident at the level of individual GvHD target tissues. Histopathological analysis at 6 days after allo-HCT revealed that transfer of Csf2−/− splenocytes caused less tissue damage in skin and small intestine compared to transfer of WT splenocytes (Fig. 2, C and D). The skin of mice receiving Csf2−/− cells showed a decrease in cellular apoptosis and cellular infiltration around and within hair follicles and epidermis, whereas the small intestine showed a decrease in outright crypt destruction (Fig. 2D and fig. S2, D and E). We also observed reduced colonic pathology upon Csf2−/− cell transfer, as depicted by increased colon length and a decreased apoptotic index (Fig. 2, E and F). A screen for subclinical signs of liver dysfunction showed decreased serum concentrations of alanine aminotransferase (ALT), alkaline phosphatase (AP), and blood urea nitrogen (BUN) in mice transferred with Csf2−/− splenocytes in comparison to the WT counterparts (Fig. 2G). In addition, the systemic amounts of TNFα were decreased in the absence of GM-CSF, whereas the amounts of IL-6 did not change (fig.S2F).

Because reactive oxygen species (ROS) have been implicated in GvHD tissue injury (22), we asked whether the decreased tissue damage observed in the absence of GM-CSF was associated with lower ROS production. We found that sections from liver and small intestine of mice 6 days after transfer of Csf2−/− splenocytes contained significantly fewer cells expressing p22phox, a critical component of the phagocyte ROS production machinery, than did sections from mice receiving WT splenocytes (Fig. 2H).

To further confirm the functional role of GM-CSF in GvHD, we treated recipient mice with anti–GM-CSF antibodies before MHC-mismatched HCT of BALB/c mice with WT B6 TCD-BM alone or with BM plus WT splenocytes. Systemically blocking the action of GM-CSF successfully replicated the survival advantage of transferring Csf2−/− splenocytes into these mice (Fig. 2I).

Although murine models of fully MHC-mismatched HCT are informative, we next asked whether the role of GM-CSF in GvHD pathology was conserved in the more clinically relevant setting of a haplo-mismatched model of acute GvHD (23). We combined WT B6 TCD-BM with WT, Csf2−/−, or Ifng−/− splenocytes and transferred the cells into lethally irradiated B6D2F1 mice, which are a cross between B6 and DBA mice and so bear the MHC haplotype H2b/d. The transfer of Csf2−/− splenocytes conferred a significant survival advantage compared to mice that received WT splenocytes, protecting mice from fatal GvHD to an almost comparable level as TCD-BM alone (Fig. 3A), which was again replicated upon the transfer of purified Csf2−/− splenic T cells (Fig. 3B). In both the splenocyte and T cell transfer experiments, we also noted a marked increase in mortality after transfer of Ifng−/− cells (Fig. 3, A and B), which is consistent with the known protective, immunoregulatory function of this cytokine in GvHD (5). At day 9 after HCT, the skin was the most affected organ with highest histological scores; here, we saw less tissue damage in mice receiving Csf2−/− compared to WT splenocytes and more tissue damage in recipients of Ifng−/− splenocytes (Fig. 3, C and D). This coincided with changes in sebaceous gland numbers, tissue organization, and cellular infiltrates around and within hair follicles and epidermis (Fig. 3D and fig. S3A). At a later stage (28 days) after allo-HCT, the transfer of Csf2−/− splenocytes also led to decreased liver pathology (Fig. 3E) characterized by changes in the portal tract infiltrates and bile duct disorganization (fig. S3B). On the cellular level, we also observed signs of reduced GvHD in the liver and small intestine of mice receiving Csf2−/− cells, which had significantly fewer p22phox+ cells accompanied by decreased infiltration of F4/80+ myeloid cells (Fig. 3, F and G) but no changes in the T cell infiltrate, similar to what we observed in the full mismatched model (fig. S3C). Given the GvHD protective role of IFN-γ seen in these mice, we then asked whether the absence of GM-CSF was related to amounts of this cytokine: 10 days after haplo-mismatched HCT, we found significantly higher amounts of IFN-γ in the serum of mice that received csf2−/− splenocytes compared to WT splenocytes (Fig. 3H). Intriguingly, we also uncovered evidence of increased GM-CSF in sera from mice receiving Ifng−/− splenocytes (Fig. 3H), suggesting that GM-CSF might additionally be mediating enhanced GvHD induced by the absence of IFN-γ. Together, we conclude that T cell–derived GM-CSF has a vital role in the development of GvHD immunopathology in two murine models of allo-HCT.

Fig. 3 GM-CSF mediates GvHD pathology after partially MHC-mismatched allo-HCT.

(A) Survival of lethally irradiated B6D2F1 mice after partially MHC-mismatched allo-HCT with WT C57BL/6 TCD-BM alone or combined with 20 × 106 splenocytes from C57BL/6 WT, Csf2−/−, or Ifng−/− mice. Data were pooled from five individual experiments, each with n = 5 per group. (B) Survival of lethally irradiated B6D2F1 mice after allo-HCT with WT C57BL/6 TCD-BM alone or combined with 7.5 × 106 T cells purified from spleens of C57BL/6 WT, Csf2−/−, or Ifng−/− mice. Data were pooled from two individual experiments, each with n = 5 per group. (C) Composite histopathological score for liver, small intestine, and skin sections from B6D2F1 mice at 11 days after allo-HCT, described in (A). Data were pooled from two individual experiments, each with n = 5 per group. (D) Representative images of hematoxylin and eosin–stained sections from skin of mice 11 days after allo-HCT with WT C57BL/6 TCD-BM combined with 20 × 106 splenocytes from either WT, Ifng−/−, or Csf2−/− C57BL/6 mice, each with n = 4 to 5 per group. Scale bars, 100 μm. (E) Composite histopathological score for liver sections from B6D2F1 mice at 9 or 28 days after allo-HCT, as described in (A). Data are representative of two independent experiments, each with n = 4 to 6 per group. (F and G) Representative images of p22phox (top panels) and F4/80 labeling in sections from the (F) liver and (G) small intestine of mice 9 days after allo-HCT with WT C57BL/6 TCD-BM combined with 20 × 106 splenocytes from C57BL/6 WT or Csf2−/− mice. Scale bars, 100 μm. Data were pooled from four independent experiments, each with n = 5 to 7 per group. Quantification of percentage of total area labeled positively for p22phox or F4/80 per visual field. (H) Serum IFN-γ and GM-CSF in mice 9 days after allo-HCT with WT C57BL/6 TCD-BM combined with 20 × 106 splenocytes from C57BL/6 WT, Ifng−/−, or Csf2−/− mice. Data were pooled from two to three individual experiments, each with n = 4 to 5 per group. For comparison of survival curves (WT versus other groups), a log-rank (Mantel-Cox) test was used in (A) and (B). For comparison of the means (WT, Csf2−/−, and Ifng−/−), one-way ANOVA with Bonferroni posttest was used in (C). For comparison of the means (WT versus Csf2−/−), an unpaired two-tailed t test was used in (F) and (G). For comparison of the means (WT versus Ifng−/− or WT versus Csf2−/−), an unpaired two-tailed t test with Welch’s correction was used in (H). *P < 0.05, **P < 0.01, ***P < 0.001. Data are shown as means ± SEM.

GM-CSF mediates GvHD lethality through donor-derived myeloid cells

After lethal conditioning, radiosensitive host antigen-presenting cells (APCs) are lost within the first days after HCT and substituted by APCs from the donor. Whereas host APCs are required for the priming phase of GvHD, donor APCs play less of a role in the induction of the disease but may be involved in perpetuating tissue injury (24). To delineate whether the pathologically relevant GM-CSF–responsive cell type(s) are of donor or host origin, we used Csf2rb−/− mice lacking the β subunit of the GM-CSF receptor (GM-CSFR) as donors or hosts of HCT. GM-CSFR deficiency in the recipient compartment did not influence GvHD survival (Fig. 4A). In contrast, the transfer of BM from Csf2rb−/− mice resulted in substantially delayed mortality (Fig. 4B), phenocopying the transfer of Csf2−/− splenocytes (Fig. 2A). We used unsupervised nonlinear dimensionality reduction [t-distributed stochastic neighbor embedding (t-SNE)] (25) to identify and visualize the GM-CSF–responding cells within the donor graft. With this approach, we found that monocytes and neutrophils were particularly sensitive to GM-CSF, which we assessed by single-cell analysis of phosphorylated STAT5 (signal transducer and activator of transcription 5), the transcription factor downstream of the GM-CSFR complex (20). Csf2rb−/− mice were used as a source of GM-CSF–unresponsive cells for comparison (Fig. 4, C and D, and fig. S4, A and B).

Fig. 4 GM-CSF drives GvHD through donor-derived myeloid cells.

(A) Survival of lethally irradiated WT C57BL/6 and Csf2rb−/− mice after MHC-mismatched allo-HCT with WT BALB/c TCD-BM alone or combined with 20 × 106 splenocytes from BALB/c WT mice. Data were pooled from two individual experiments, each with n = 5 per group. (B) Survival of lethally irradiated BALB/c WT mice following MHC-mismatched allo-HCT with C57BL/6 WT or and Csf2rb−/− mice TCD-BM alone or combined with 10 × 106 splenocytes from C57BL/6 WT mice. Data were pooled from two individual experiments, each with n = 5 per group. (C) Annotated t-SNE map displaying 200,000 randomly sampled cells from the BM of WT C57BL/6 and Csf2rb−/− mice showing STAT5 phosphorylation (pSTAT5; black to yellow gradient) upon GM-CSF stimulation, analyzed by flow cytometric analysis. Data represent two independent experiments, each with n = 3 per group. (D) Frequencies of GM-CSF–induced pSTAT5 up-regulation in monocytes and neutrophils from WT C57BL/6 and Csf2rb−/− mice as shown overlaid in (B). (E) Flow cytometric analysis of different myeloid cell populations (DCs, neutrophils, and monocytes and MDCs) after HCT. Example gating for the liver is shown. (F) Quantification of myeloid cell populations [from (E)] in the spleen (top row) and liver (bottom row). One representative from three individual experiments is shown, each with n = 4 to 5 per group. (G) Frequency of pro–IL-1β–producing neutrophils and monocytes within the H2Db+ CD45+ population from spleens of mice 6 days after allo-HCT. Representative plots are shown for neutrophils, monocytes, and MDCs [gated as in (E)]. Data represent two independent experiments, each with n = 5 per group. (H) Flow cytometric analysis of ROS (CellROX reagent) in mice 6 days after allo-HCT. Representative histograms of median fluorescence intensity (MFI) for neutrophils, monocytes, and MDCs [gated as in (E)]. Data represent one experiment with n = 5 to 7 per group. For comparison of survival curves (WT versus Csf2rb−/− recipients/donors), a log-rank (Mantel-Cox) test was used in (A) and (B). For comparison of the means, an unpaired two-tailed t test was used in (D) and (F) to (H). ***P < 0.001, **P < 0.01, *P < 0.05. Data are shown as means ± SEM.

Having confirmed that the cells responding to GM-CSF are limited to the myeloid compartment, we performed a thorough characterization of myeloid cell subsets found in the inflamed target organs 5 days after allo-HCT. The transfer of Csf2−/− splenocytes resulted in decreased frequencies of several myeloid subsets including DCs, neutrophils, monocytes, and monocyte-derived cells (MDCs) (Fig. 4, E and F, and fig. S4C). A decrease in neutrophils and monocytes/MDCs was also found in target organs upon anti–GM-CSF treatment, whereas the systemic numbers of these myeloid subsets remained unaltered (fig. S4D). We observed that donor monocytes/MDCs produced less pro–IL-1β in the absence of GM-CSF (Fig. 4G). IL-1β is a signature cytokine of the pathogenic program elicited by GM-CSF (20) and was shown to play a critical role in GvHD development (26). We also investigated the expression of ROS by donor myeloid cells after allo-HCT, given their low amounts in tissue sections of mice receiving allogeneic Csf2−/− T cells and their reported role in driving tissue damage (27). We found that the transfer of Csf2−/− splenocytes led to a decreased ROS production by both the monocytes/MDCs and neutrophil subsets (Fig. 4H). Collectively, these findings demonstrate that GM-CSF–licensed donor-derived myeloid cells are crucial for GvHD pathology.

Efficient GvL is retained in the absence of GM-CSF

The immune responses leading to tissue damage in GvHD and tumor elimination in GvL are widely considered to be mediated through the same mechanism (28). Because donor T cell–derived GM-CSF was essential for lethal GvHD, we investigated whether it also mediated GvL in the MHC-mismatched HCT model. To test this, we intravenously injected A20 B cell lymphoma cells (of BALB/c origin) coexpressing green fluorescent protein (GFP) and luciferase (A20-GFP-Luc) into lethally irradiated BALB/c recipients, together with WT B6 TCD-BM cells either alone or with purified B6 WT or Csf2−/− splenic T cells. Tumor growth was monitored by bioluminescent imaging (BLI). When A20 cells were infused with TCD-BM alone, all recipients died or reached the final termination criteria from the growing tumor within 35 days (Fig. 5, A to C), consistent with a critical role for T cells in the GvL effect. Accordingly, adding WT or Csf2−/− T cells to the BM transfer resulted in efficient control of tumor growth (Fig. 5, A to C), although only mice receiving Csf2−/− T cells showed a significant improvement in survival (Fig. 5C). Whereas lymphoma-bearing mice receiving TCD-BM plus WT T cells mainly died from severe GvHD, mice receiving TCD-BM plus Csf2−/− T cells exhibited an undiminished GvL effect but were at the same time protected from GvHD (Fig. 5, C and D). This improvement of survival in mice receiving Csf2−/− T cells was also observed in an alternative GvL model using the monomyelocytic cell line WEHI-3 (fig. S5A). Also, when we increased the number of inoculated A20 lymphoma cells, Csf2−/− T cells executed GvL even more efficiently than WT T cells, leading to a significant increase in overall survival (fig. S5, B and C). Consistently, Csf2−/− and WT T cells isolated from naïve or mice receiving allo-HCT were equally capable of killing A20 lymphoma and WEHI-3 cells ex vivo (fig. S5, D to F), expanded equally in the spleen and LN (fig. S5G) and showed similar frequencies of activated (CD44high), proliferative (Ki67+), and granzyme B–producing cells (fig. S5G). To test the translatability of our findings into a clinical setting, we used neutralizing monoclonal antibodies against GM-CSF. Notably, anti–GM-CSF–treated leukemic mice receiving TCD-BM plus T cells showed a potent GvL effect and had significantly better survival than the isotype-treated control mice (Fig. 5, E to H). Better survival was associated with reduced GvHD incidence (Fig. 5H). Notably, neutralization of GM-CSF did not have any impact on tumor growth in mice which did not undergo HCT. Thus, in the absence of GM-CSF, donor T cells are effective mediators of the therapeutic GvL response in mice, without inducing lethal GvHD, even in the highly immunogenic context of an MHC-mismatched allo-HCT.

Fig. 5 GM-CSF is dispensable for antitumor activity after allo-HCT.

(A to D) Lethally irradiated BALB/c mice were intravenously injected with 250,000 A20 tumor cells expressing GFP and luciferase at the same time as MHC-mismatched allo-HCT with WT C57BL/6 TCD-BM alone or combined with 1 × 105 T cells purified from spleens of C57BL/6 WT or Csf2−/− mice. Mice treated with TCD-BM alone were used as controls. (A) Tumor growth was monitored by in vivo BLI. Images from one representative experiment of three are shown. (B) Signal intensity in the region of interest (ROI) was monitored over time. (C) Survival over time and (D) Incidence of GvHD shown as percentage of mice in each treatment group which developed lethal GvHD over time (gray) or survivors (white). Data were pooled from three individual experiments, each with n = 5 per group. (E to H) Lethally irradiated BALB/c mice were intravenously injected with A20 tumor cells expressing GFP and luciferase at the same time as MHC-mismatched allo-HCT with WT C57BL/6 TCD-BM alone or combined with 1 × 105 T cells purified from spleens of C57BL/6 WT mice. Mice treated with TCD-BM alone were used as controls. Mice were treated with isotype control or anti–GM-CSF antibody three times per week for the duration of the experiment, starting 2 days before HCT. (E) Tumor growth was monitored by in vivo BLI. Images from one experiment with n = 6 to 7 per group are shown. (F) Signal intensity in the region of interest was monitored over time. (G) Survival over time and (H) incidence of GvHD shown as percentage of mice in each treatment group which developed lethal GvHD over time (gray) or survivors (white). For comparison of survival curves [WT versus other groups in (C) and WT versus csf2−/− in (G)], a log-rank (Mantel-Cox) test was used. *P < 0.05 and **P < 0.01. Data are shown as means ± SEM.

GM-CSF+ T cells are elevated in patients with GvHD

We next asked whether there was evidence that GM-CSF plays a parallel role in GvHD in human HCT recipients. We assessed GM-CSF gene expression and protein abundance in gastrointestinal biopsies from patients with different grades of clinically documented acute GvHD after HCT for a range of different conditions (tables S1 to S3). At the transcriptional level, we detected significantly higher expression of GM-CSF in gastrointestinal biopsies from patients with grade IV compared to grade I GvHD (Fig. 6A). Moreover, in samples from patients with grade IV GvHD, we observed strong immunoreactivity for GM-CSF in the stromal compartment, although we did not detect GM-CSF in biopsies from patients without GvHD (Fig. 6B). Most of the cells producing GM-CSF in these samples were T cells, as indicated by the colocalization with CD3, in contrast to the minimal colocalization observed with the macrophage marker CD68 (Fig. 6C). We also analyzed the expression of GM-CSF in T cells from peripheral blood of patients with GvHD (table S4 and gating strategy is shown in fig. S6) and found that the frequencies of GM-CSF–producing CD4+ T cells were significantly increased in patients with GvHD in comparison to healthy donors (Fig. 6D), supporting that activated GM-CSF–producing T cells are pathogenically relevant in the human disease. We also observed a marked increase in the numbers of IFN-γ–producing T cells but found no difference in IL-17–producing T cell frequencies (Fig. 6D). Together, these results implicate GM-CSF in GvHD after allo-HCT in human patients.

Fig. 6 GM-CSF+ T cells are elevated in patients with GvHD.

(A) Relative expression of GM-CSF at the mRNA level in gastrointestinal biopsies from patients with different GvHD grades (see table S1). Data are shown as means ± SEM. (B) Images of GM-CSF labeling in control (patients with no pathological findings) and GvHD grade IV biopsies from the small intestine of allo-HCT patients (see tables S2 and S3). Brown, anti-human GM-CSF; blue, hematoxylin. Scale bars, 100 and 20 μm in zoom images. Representative images of three individual control and patient samples are shown (tables S2 and S3). (C) Immunofluorescence staining for CD3 (pink), CD68 (green), and GM-CSF (red) of gastrointestinal biopsies from patients with grade IV GvHD (table S2). A representative picture of three individual patient samples is shown. Nuclei are depicted in blue (4ʹ,6-diamidino-2-phenylindole). Scale bars, 50 μm (top) and 20 μm (bottom). (D) Flow cytometric analysis of peripheral blood mononuclear cells (PBMCs) collected from healthy donors (HD) and patients with GvHD (GP) (table S4), stimulated for 4 hours with phorbol 12-myristate 13-acetate/ionomycin. Left: Representative plots of cytokine expressing T cells are shown. Right: GM-CSF–, IFN-γ–, and IL-17–producing CD4+ and CD8+ T cells are presented as individual frequencies of CD4 or CD8 T cells, n = 9 to 10 per group. For comparison of the means, one-way ANOVA with Bonferroni posttest was used in (A), and an unpaired two-tailed Mann-Whitney test was used in (D). *P < 0.05 and **P < 0.01.

DISCUSSION

Although there is a large body of literature discussing the potential contributions of different TH cell subsets and their signature cytokines to GvHD (29), the mechanisms underlying inflammation and subsequent tissue destruction are not fully understood. Here, we reveal that GM-CSF plays a critical role in driving lethal GvHD in mice and is similarly elevated in human allo-HCT patients with severe GvHD. Moreover, GM-CSF appears to be dispensable for the therapeutic GvL effect and so may represent a promising therapeutic target in the separation of GvL from GvHD for the treatment of hematologic malignancies.

We used two different murine models of acute GvHD to show that donor T cell–derived GM-CSF directly mediates severe GvHD, which is associated with abundant myeloid cell infiltration and increased amounts of pro–IL-1β and components of the ROS production machinery, classically used by phagocytes for host defense against pathogens. The same mechanisms are detrimental in the immunopathology of GvHD, where tissue damage and leukocyte infiltration go hand-in-hand (30). The reduction of myeloid-driven oxidative stress in GvHD-susceptible organs observed here in animals receiving alloreactive Csf2−/− T cells supports this notion. In accordance with the reported key role of GM-CSF in chronic intestinal inflammation (31) and the recently described contribution of basic leucine zipper transcription factor, ATF-like–dependent IL-7RhiGM-CSF+ T cells to intestinal GvHD (32), we found GM-CSF to contribute to colonic pathology. The gastrointestinal tract not only is a crucial target organ of GvHD but also propagates the “cytokine storm” characteristic of acute GvHD (33, 34).

We uncovered evidence of a potential reciprocal regulation of GM-CSF and IFN-γ during GvHD in mice. IFN-γ produced by activated donor T cells has previously been shown to both promote and protect against GvHD (35, 36); here, we found that mice undergoing partial MHC-mismatched HCT with IFN-γ–deficient splenocytes were significantly more susceptible to lethal GvHD than those receiving WT splenocytes. Intriguingly, these mice presented elevated serum GM-CSF, whereas mice receiving GM-CSF–deficient splenocytes exhibited elevated IFN-γ in their sera. The ability of IFN-γ to negatively regulate GM-CSF in effector TH cells has been previously reported (17); however, GM-CSF does not directly affect T cells, and thus, the mechanisms by which GM-CSF influences IFN-γ production are currently not known. We also excluded a role for T cell–derived IL-17A in acute GvHD. Previous reports differ in their interpretations regarding the role of IL-17 in GvHD, which was found to be detrimental (11, 12), protective (37), or play no role in the disease (38). Although differences in the conditioning regimens may explain the discordant findings, our data are supported by the generally low frequency of IL-17–producing T cells in preclinical GvHD models. Also in peripheral blood leukocytes from patients with GvHD, IL-17A frequencies were not abundant among memory T cells. Because posttransplant IL-17–producing CD8 T cells have been shown to express GM-CSF (39), we hypothesize that this cytokine might mediate the detrimental effects of CD8 T cells in GvHD pathology.

The prominent role for GM-CSF in the pathogenesis of GvHD is in line with the GM-CSF dependency reported in various T cell–driven models of autoimmunity (13). T cells, lacking the GM-CSFR complex, cannot directly respond to GM-CSF, and myeloid cell subtypes were proposed to be targeted by this cytokine during disease progression. In an animal model for multiple sclerosis, for example, GM-CSF responsiveness by CCR2+ monocytes was shown to be essential for tissue inflammation (20). In the case of GvHD, we identified donor myeloid cells as relevant responders to GM-CSF. Although host APCs are required for the priming of acute GvHD (40), donor APCs may play a role in tissue injury. We propose that donor APCs do not largely contribute to prime alloreactive T cells (24), and that GM-CSF licenses the myeloid population in the donor graft to produce proinflammatory mediators (e.g., IL-1β and ROS) that engage and perpetuate tissue damage. Myeloid cells readily respond to GM-CSF to produce IL-1β (20), and donor-derived APCs have been shown to be the main IL-1β–producing myeloid subset at late GvHD stages (28). Although ROS can directly cause tissue damage, myeloid-derived IL-1β can also potentially act via local stimulation of alloreactive T cells. The cross-talk mechanisms involved in GM-CSF–induced tissue damage in GvHD target organs are not yet fully elucidated.

There has been a long-standing assumption that donor T cell–driven GvHD and the therapeutic GvL effect are mediated via the same mechanisms: Our data require us to question this notion. Treating mice with GM-CSF blocking antibodies before allo-HCT significantly protected from the development of lethal acute GvHD, whereas tumor-bearing mice treated with anti–GM-CSF benefited from effective control of tumor growth and protection from GvHD. Thus, GM-CSF does not control T cell–mediated killing of malignant lymphohematopoietic cells; instead, we propose that, in GvHD, GM-CSF acts as a communication conduit between alloreactive T cells and myeloid cells, licensing the latter to cause tissue destruction. Although GvL likely relies on direct cytotoxicity by T cells, the GM-CSF–driven responses of myeloid cells may be more relevant for GvHD pathology, thus representing a potential mechanism to separate the GvL effect from GvHD.

Current clinical guidelines support the use of CSFs to mobilize peripheral blood progenitors after autologous progenitor cell transplantation. However, the same is not applicable for allogeneic progenitor cell transplantation (41), and the use of GM-CSF treatment in HCT has been associated with more days of fever and prolonged antibiotic therapy and, in addition, did not decrease infection-related mortality (42). We here propose to neutralize GM-CSF in patients with GvHD, given the detrimental role of this cytokine in experimental models of GvHD. We anticipate that blocking GM-CSF in allo-HCT patients could improve clinical outcome and may even permit greater incompatibility between donors and recipients while limiting the risk of fatal GvHD. The marked increase of this cytokine observed in affected tissues from patients with severe GvHD further supports this idea. However, because GM-CSF is crucial for surfactant homeostasis and lung host defense (43), it will be important to closely monitor lung disease parameters when blocking GM-CSF in the clinical setting.

There is evidence in mice that GM-CSF contributes critically to IL-23–mediated immune responses (17, 44) and that anti–IL-23p19 therapy can ameliorate syngeneic GvHD-associated colitis (45). Thus, the direct blockade of GM-CSF might provide a valuable complement to any IL-23–targeted clinical trials. The fact that the absence of GM-CSF can halt the development of GvHD without impairing the GvL response paves the way for testing this therapeutic strategy for the prevention and treatment of GvHD after allo-HCT. Moreover, that GM-CSF was elevated in patient samples with the severest manifestations of GvHD means that this therapeutic strategy holds particular promise for patients with the poorest predicted outcome and highest risk of fatality.

MATERIALS AND METHODS

Study design

The study was initiated to determine whether certain T cell–derived cytokines could separate GvHD from GvL and so may represent promising novel therapeutic targets for the treatment of hematologic malignancies. To achieve this aim, we used two different experimental models of GvHD, an experimental model of GvL, and samples from human subjects. For animal studies, 8- to 12-week-old mice were used. All animal experiments were approved by local authorities (Swiss Cantonal Veterinary Office) and performed under the appropriate experimental licenses (76/2012 and 052/2015). Animals were randomly assigned into the experimental groups, and in-life clinical score was performed in a blinded fashion and image analysis processing on organ sections. Sample size and disease end time points were selected on the basis of previous studies. Flow cytometry, histopathological analysis, MLRs, killing assays, and cytokine analysis were performed to characterize the GvHD/GvL target organs. The effects of the specific GM-CSF blocking antibody on clinical score were assessed by investigators who were blinded to the treatment. To perform reliable statistical analysis, three independent experiments were conducted for each data shown in the manuscript, unless differently indicated in the figure legends.

All human samples were collected after approval by the Ethics Committee of the Albert-Ludwigs University Freiburg, Germany (protocol no. 267/11) following written informed consent. We performed immunohistochemistry and quantitative reverse transcription polymerase chain reaction on the gut biopsies and multiparameter flow cytometry on the PBMCs. Primary data are located in table S5. Also, clinical features of patients can be found in tables S1 to S4. The scoring system used for allo-HCT mice is described in table S6, and the phenotypical analysis of human PBMCs in table S7.

Statistical analysis

Graphs were prepared with GraphPad Prism (GraphPad Software). Survival curves were plotted by the Kaplan-Meier method, and for comparison of survival curves, a log-rank (Mantel-Cox) test was used. Data are shown as individual data points or as means ± SEM as depicted in the figure legends. Comparison of the means was performed using unpaired, two-tailed Student’s t tests (with Welch’s correction where applicable) or one-way and two-way ANOVA with Bonferroni posttest, respectively. No statistical methods were used to predetermine sample size, but our sample sizes were similar to those generally used in the field. No method of randomization was used.

SUPPLEMENTARY MATERIALS

www.sciencetranslationalmedicine.org/cgi/content/full/10/469/eaat8410/DC1

Materials and Methods

Fig. S1. Comparative phenotypic analysis of T cell populations from WT and Csf2−/− mice.

Fig. S2. GM-CSF is crucial for acute GvHD after fully MHC-mismatched allo-HCT.

Fig. S3. GM-CSF mediates GvHD pathology after partially MHC-mismatched allo-HCT.

Fig. S4. GM-CSF drives GvHD through donor-derived myeloid cells.

Fig. S5. GM-CSF is dispensable for antitumor activity after allo-HCT.

Fig. S6. Frequency of cytokine-expressing T cells in PBMCs of patients with GvHD.

Table S1. Clinicopathological features of GvHD patient set I.

Table S2. Clinicopathological features of GvHD patient set II.

Table S3. Characteristics of control subjects.

Table S4. Clinicopathological features of GvHD patient PBMCs.

Table S5. Primary data.

Table S6. GvHD scoring system for allo-HCT mice.

Table S7. Phenotypic analysis of human PBMCs by flow cytometry.

REFERENCES AND NOTES

Acknowledgments: We thank A. Müller for critical input on the manuscript, S. Nemetz, J. Jaberg, and S. Konieczka for technical assistance, and C. Krieg for experimental advice. We also thank L. Robinson of Insight Editing London for assistance with preparation and critical evaluation of the manuscript. Funding: This study was supported by grants from the Swiss National Science Foundation [316030_150768 (to B.B.), 310030_146130 (to B.B.), and CRSII3_136203 (to M.v.d.B. and B.B.)], the University Research Priority Program (URPP) for Translational Cancer Research (to S.T., M.O.H., M.v.d.B., M.G.M., and B.B.), and the European Community FP7 grant no. 602239 (ATECT). Author contributions: S.T., A.A., and S.S. designed and performed experiments, evaluated and interpreted data, and wrote the manuscript. G.M.-B. designed and performed experiments, evaluated data, and edited the manuscript. D.D.F., M.L., N.G.N., and C.H. designed and performed experiments and evaluated data. B.S. and F.G. performed and evaluated histological analysis. R.Z., P.H., and P.A. provided human GvHD samples. M.O.H., M.v.d.B., and M.G.M. interpreted data and edited the manuscript. B.B. supervised and financed the study and wrote the manuscript. Competing interests: B.B. is the inventor on the patent application (EP18194549.4) submitted by Zurich University that covers the use of antibodies specific to GM-CSF or GM-CSFRs in allo-HCT. All other authors declare that they have no competing interests. Data and materials availability: All data associated within this study are present in the paper or Supplementary Materials.
View Abstract

Navigate This Article