The primacy of the gastrointestinal tract as a target organ of acute graft-versus-host disease: rationale for the use of cytokine shields in allogeneic bone marrow transplantation

Geoffrey R. Hill and James L. M. Ferrara


Acute graft-versus-host disease (GVHD), the major complication of allogeneic bone marrow transplantation (BMT), limits the application of this curative but toxic therapy. Studies of inflammatory pathways involved in GVHD in animals have shown that the gastrointestinal (GI) tract plays a major role in the amplification of systemic disease. Damage to the GI tract increases the translocation of inflammatory stimuli such as endotoxin, which promotes further inflammation and additional GI tract damage. The GI tract is therefore critical to the propagation of the “cytokine storm” characteristic of acute GVHD. Experimental approaches to the prevention of GVHD include reducing the damage to the GI tract by fortification of the GI mucosal barrier through novel “cytokine shields” such as IL-11 or keratinocyte growth factor. Such strategies have reduced GVHD while preserving a graft-versus-leukemia effect in animal models, and they now deserve formal testing in carefully designed clinical trials.

Allogeneic bone marrow transplantation (BMT) remains the treatment of choice for a number of malignant conditions. Graft-versus-host disease (GVHD), the primary complication of allogeneic BMT, remains the major limitation to this therapeutic approach. Donor T cells are critical in the induction of acute GVHD because depletion of T cells from the bone marrow graft effectively prevents GVHD but also results in an increase in leukemic relapse.1 Although there is a fundamental requirement for donor T cells in GVHD, the process still occurs in the absence of the T-cell cytotoxicity pathways (perforin, FasL, granzyme).2 3This anomaly has forced investigators to reconsider the traditional T-cell–restricted paradigms of GVHD pathophysiology.

Primacy of the gastrointestinal tract as a GVHD target organ

Increasing evidence in experimental and clinical BMT settings suggests that damage to the gastrointestinal (GI) tract during acute GVHD plays a major pathophysiologic role in the amplification of systemic disease. Endotoxin or lipopolysaccharide (LPS) is a constituent of normal bowel flora known to play an important role in GVHD pathogenesis. Early animal studies showed that death after BMT was prevented if mice were given antibiotics to decontaminate the gut; normalization of the gut flora at or before day 20 abrogated this effect.4 In the clinical setting, gram-negative gut decontamination has also been shown to reduce GVHD.5 6Furthermore, the intensity of this decontamination has recently been demonstrated to be an important predictor of GVHD severity.7 8

Lipopolysaccharide is a potent stimulator of inflammatory cytokine production, such as tumor necrosis factor (TNF)-α, IL-1, and IL-12, which are important mediators of clinical9-11 and experimental GVHD.12-14 The production of TNF-α by monocytes and macrophages is transduced by 2 signals.15 The first is a priming signal that may be provided by interferon (IFN)-γ15 and radiation.13 The second is a triggering signal provided by bacterial products such as LPS.15 Other bacterial products, such as CpG DNA repeats, in the GI tract lumen have also been shown to have potent immuno-stimulatory properties and to induce strong Th1 responses.16 Furthermore, microbial superantigens may activate B cells by direct stimulation of major histocompatibility complex (MHC) class II molecules.17 During GVHD, the IFN-γ produced by donor T cells renders monocytes and macrophages extremely sensitive to endogenous LPS.18 Further support for the role of LPS in GVHD comes from a recent study comparing donor monocytes and macrophages that were genetically sensitive (ie, normal) or resistant to the effects of LPS for their ability to induce GVHD.19 The donor mouse strains were otherwise identical, and, in particular, their T-cell responses to host alloantigens were equivalent. Allogeneic BMT recipients of LPS-resistant bone marrow demonstrated significantly less TNF-α production and reduced GVHD than recipients of LPS-sensitive bone marrow. Significantly, the reduced production of TNF-α in recipients of LPS-resistant marrow was associated with reduced GVHD of the GI tract. The causality of the association was demonstrated by the systemic neutralization of TNF-α, which reduced GVHD of the GI tract and lessened the severity of overall disease.

Additional evidence for the importance of GI tract integrity during GVHD comes from studies of the effect of BMT conditioning on GVHD severity after allogeneic BMT. Clinical studies first suggested a correlation between GVHD severity and radiation dose (less than 1200 versus more than 1200 cGy)20 21 and more severe GVHD after conditioning regimens that included radiation therapy than those that included only chemotherapy.22 The increased incidence of conditioning-related toxicity has also been associated with GVHD incidence.23 However, these studies demonstrated an inverse correlation between conditioning intensity and compliance with immunosuppression prophylaxis, which might partially account for the increase in GVHD. Indeed, only 1 of these studies identified increased radiation intensity as an independent risk factor for GVHD,20 and there was no demonstrable association with severe disease (grades 2 to 4). By contrast, the association of increased GVHD with dose reductions in cyclosporine was maintained throughout all GVHD grades. Using clinical data alone, it has thus been difficult to separate the influence of conditioning intensity and suppression of T-cell function on subsequent GVHD severity.

We have recently examined this issue in murine BMT models in which these variables could be tightly controlled. These studies demonstrated an increase in GVHD severity in several donor–recipient strain combinations, after intensification of the conditioning regimen, by increasing the total body irradiation (TBI) dose from 900 cGy to 1300 cGy. Synergistic damage to the GI tract was caused by increased TBI and allogeneic donor cells, permitting increased translocation of LPS to the systemic circulation. In vitro, LPS triggered the most TNF-α secretion in macrophages taken from animals with the worst GI tract damage. Neutralization of TNF-α eliminated deaths from GVHD. Thus, the higher TBI dose increased gut damage after allogeneic BMT, causing higher systemic levels of inflammatory cytokines and more severe GVHD.13 High levels of inflammatory cytokines may perpetuate GVHD because TNF-α has been shown to be directly toxic to the GI tract during GVHD in other experimental systems.19 24 25

The addition of cyclophosphamide (Cy) to TBI also increased damage to the GI tract and the subsequent incidences of mortality and morbidity related to GVHD.14 After Cy/TBI conditioning, neutralization of IL-1 with a hamster antibody to the IL-1 receptor significantly reduced serum LPS levels and GVHD deaths, whereas neutralization of TNF-α did not. Although the expansion of donor T cells on day 13 after BMT was increased after Cy/TBI than with Cy or TBI alone, cytotoxic T lymphocyte (CTL) function was not different between groups. Taken together, these studies confirm that increasing the dose of TBI or combining TBI with cyclophosphamide increases the severity of GVHD. Importantly, the pathophysiologic mechanisms involved in this increase in GVHD appear to be focused on GI tract damage and subsequent amplification of the inflammatory effectors of GVHD rather than on the effects on donor T cells.

Rationale for shielding the gastrointestinal tract

These studies suggest that the GI tract is not only a major target organ of GVHD but is a critical amplifier of systemic GVHD severity. We have previously suggested that acute GVHD pathophysiology can be conceptualized in 3 sequential phases (Figure1).26 In phase 1, the conditioning regimen (irradiation, chemotherapy, or both) leads to damage to and activation of host tissue by release of the inflammatory cytokines TNF-α and IL-1. These cytokines can increase the expression of MHC antigens and adhesion molecules on host antigen-presenting cells, enhancing the recognition of host MHC and minor histocompatibility antigens by mature donor T cells. Donor T-cell activation in phase 2 is characterized by the proliferation of Th1 T cells and the secretion of IL-2 and IFN-γ. IL-2 and IFN-γ induce further T-cell expansion, CTL- and NK-cell responses, and prime additional mononuclear phagocytes to produce IL-1 and TNF-α. Effector functions of mononuclear phagocytes (phase 3) are triggered by the secondary signal provided by bacterial products such as LPS. Damage to the intestinal mucosa in phase 1 and by cytolytic effectors activated in phase 2 allows translocation of LPS from the intestinal lumen to the circulation. Subsequently, LPS may stimulate additional cytokine production by gut-associated lymphocytes and macrophages in the GI tract and by keratinocytes, dermal fibroblasts, and macrophages within the skin. This mechanism may amplify local tissue injury and further promote an inflammatory response that, together with the CTL and NK component, leads to target tissue destruction in the BMT host. Damage to the GI tract in phase 3 increases LPS release, stimulating further cytokine production and causing additional GI tract damage. Thus the GI tract is critical to propagating the “cytokine storm” characteristic of acute GVHD.

Fig. 1.

The immunopathophysiology of GVHD.

Schematic representation of central role of GI tract damage during GVHD. In phase 1, the conditioning regimen (irradiation, chemotherapy, or both) leads to the damage and activation of host tissues, especially the intestinal mucosa. This allows the translocation of LPS from the intestinal lumen to the circulation, stimulating the secretion of the inflammatory cytokines TNF-α and IL-1 from host tissues, particularly macrophages. These cytokines increase the expression of MHC antigens and adhesion molecules on host tissues, enhancing the recognition of MHC and minor histocompatibility antigens by mature donor T cells. Donor T-cell activation in phase 2 is characterized by the proliferation of Th1 T cells in the presence of IL-12 and the secretion of IL-2 and IFN-γ. IL-2 and IFN-γ induce further T-cell expansion and CTL and NK cell responses, and they activate mononuclear phagocytes. The CTL and NK effectors damage tissue by perforin/granzyme, FasL, and TNF-α. In phase 3, effector functions of activated mononuclear phagocytes are triggered by the secondary signal provided by LPS and other immuno-stimulatory molecules that leak through the intestinal mucosa damaged during phases 1 and 2. This damage results in the amplification of local tissue injury, and it further promotes an inflammatory response. Damage to the GI tract in this phase, principally by inflammatory cytokines, amplifies LPS release and leads to the “cytokine storm” characteristic of severe acute GVHD. Double lines show the points at which KGF and IL-11 both interrupt this process, whereas single lines reflect interruption by IL-11 only. IL-11, but not KGF, promotes type 2 donor T-cell differentiation and inhibits IFN-γ secretion. Neither IL-11 nor KGF impairs CTL or NK function, thereby preserving GVL effects. The reduction in GI tract damage by these “cytokine shields” prevents systemic LPS translocation and reduces inflammatory cytokine production, culminating in reduced GI tract damage and subsequent death from GVHD.

Although cytokines clearly play important roles in incidences of morbidity and mortality related to systemic GVHD, their relative importance as mediators of damage in GVHD target organs is less well established. The unusual cluster of GVHD target organs (skin, gut, and liver) is inadequately explained by the systemic release of cytokines. For example, intravenous infusion of TNF-α and IL-1 does not cause the lymphomononuclear cell infiltration of liver and skin observed in GVHD. Furthermore, the absence of GVHD toxicity in other visceral organs, such as the kidney, argues against circulating cytokines as the sole cause of tissue-specific damage. The infiltrates seen in GVHD target organs are generally thought to contain T cells responding to alloantigens on host tissues. As mentioned above, LPS leakage through skin or mucosa may act as an adjuvant to the antigens expressed in these tissues, attracting and activating alloreactive donor T cells. A second possibility is that tissue-specific neoantigens are expressed at these sites as the result of ongoing inflammation. Such inflammation may alter ligands for homing receptors on T cells (eg, selectins) that enables them to traffic into specific tissues.

It is clear in murine systems that cytolytic T-lymphocyte effectors contribute to GVHD. The application of knock-out technology and the identification of pertinent mutant mice have provided important tools for dissecting effector mechanisms. Three cytolytic pathways have been identified as important to GVHD: the perforin/granzyme B pathway, the Fas/FasL pathway, and direct cytokine-mediated injury. Donor cells from perforin-deficient and granzyme B-deficient knock-out mice can mediate lethal GVHD, but the onset of clinical manifestations is significantly delayed.27 28 Similarly, when mutant mice deficient in FasL (gld) are used as donors, GVHD occurs in an attenuated fashion.3 When FasL-deficient mice are crossed with those having perforin or granzyme B knock-outs, the use of lymphocytes from these donors further diminishes but does not abrogate GVHD.27 Interestingly, the Fas pathway is important in the development of hepatic and cutaneous GVHD.24

GVHD histopathology in the GI tract has been described in 3 phases.29 The early phases of GI tract changes have been described in animal models that do not use chemotherapy or radiation to condition the host; therefore, direct comparisons to clinical GVHD after BMT are not possible. This initial proliferative phase results in increased crypt cell mitotic activity, crypt lengthening, and intraepithelial lymphocytes. In experimental systems, this phase seems to be linked to IFN-γ production,30 which increases MHC class II expression and gut permeability by altering tight junction integrity31 and may modulate crypt stem-cell turnover.29 The histologic features of the GI tract in clinical GVHD and experimental GVHD after myeloablative conditioning are consistent with the destructive and atrophic phases, characterized by villus blunting, lamina propria inflammation, crypt destruction (with crypt stem-cell loss), and mucosal atrophy. These features can be induced in animals by the administration of exogenous cytokines, including TNF-α32 and IL-1.33 Furthermore, the inhibition of IFN-γ,34 TNF-α,25IL-1,33 or nitric oxide35 can reduce GI tract histopathology in animals with GVHD. In contrast, CTL effectors do not appear to play a dominant role in experimental GVHD of the GI tract,3 13 14 19 24 36 despite the ability of intraepithelial lymphocytes to induce Fas-mediated apoptosis of host type tumor cells.37 It is clear, when these findings are considered in aggregate, that cytokines and cellular effectors combine to produce specific target organ damage and systemic toxicity of acute GVHD.

The experimental studies referenced above used novel cytokine antagonists and transgenic/knock-out mice. As yet, it has not been possible to confirm the extent to which the principals derived from these studies apply to humans. The histologic features of end-stage intestinal GVHD are similar in animals and humans.29 The diarrhea characteristic of GVHD may occur because of a number of mechanisms, including enterocyte damage and epithelial disaccharidase deficiency with an excess of luminal sugars and osmotic water loss, an increase in the proportion of immature enterocytes with subsequent enzyme deficiency and impaired water transport, and protein and water exudation through a hyperpermeable epithelium. In the large intestine, damage to colonic enterocytes also impairs water reabsorption.29

Increasing the intensity of the conditioning regimen in murine models of GVHD to the levels used clinically amplifies GVHD of the GI tract. This increase in gut injury and LPS leak is a common proximal pathway for the dysregulation of inflammatory cytokines after allogeneic BMT. As discussed above, recent work has confirmed that GVHD of the GI tract is principally mediated by inflammatory cytokines. From this perspective, it should be possible to interrupt the process of GI tract damage at a number of steps. First, attempts to eradicate or significantly reduce the load of gram-negative organisms from the GI tract are current practice in a number of transplant centers. Unfortunately, endotoxemia has been noted to be common after BMT even with gut decontamination; it occurs in association with biochemical parameters of gut damage.38 Although gut decontamination can reduce clinical GVHD,5 7 the effect is at best partial. It has not been shown to benefit the recipients of unrelated donor BMT, who are in the greatest need for improved GVHD prophylaxis.8 Inhibition of systemic LPS with neutralizing proteins is under investigation in experimental GVHD models. However, the success of this approach assumes that LPS is the only or primary immuno-stimulatory constituent of the GI tract lumen that amplifies GVHD. There are numerous additional putative immuno-stimulatory molecules, such as soluble peptidoglycan, lipoteichoic acid,39 bacterial DNA,40 and heat-shock proteins,41 suggesting that neutralization of LPS alone may be only partially successful in preventing GVHD.

Reductions in the doses of chemoradiotherapy to condition BMT recipients should also reduce GVHD, as demonstrated in experimental models.13 14 This reduction is the result of reduced priming of mononuclear cells by lower TBI doses and of subsequent reductions in TNF-α production.13 It is also likely that low TBI doses fail to sensitize the GI tract to secondary damage by inflammatory cytokines. The use of non-myeloablative conditioning should therefore also reduce GI tract damage after allogeneic BMT. The activation of donor T cells is enhanced by inflammatory cytokines,42 and a temporal separation of the inflammatory milieu induced by conditioning (which is thought to be self-limited) and donor lymphocyte infusions may interrupt the cytokine cascade that damages the GI tract and increases systemic GVHD.12 43

Much of the therapeutic potential of allogeneic BMT relates to the graft-versus-leukemia (GVL) effect, which is mediated by the cytotoxic pathways of donor T and NK cells, including perforin/granzyme and Fas/FasL.44 In experimental GVHD models, the absence of perforin in donor T cells results in an almost complete loss of GVL,45-47 whereas the absence of FasL does not diminish GVL effects.47 Inhibition of inflammatory cytokine production after BMT might offer an approach to separate GVHD and GVL. However, TNF-α may not be an ideal target because the p55 TNF-α receptor is critical for donor CTL activity after BMT and contributes to the GVL effect.14 GVL effects are also diminished when TNF-α is neutralized in experimental models of GVHD.47 IL-1 may be a more attractive target for neutralization because the inhibition of IL-1 does not inhibit CTL generation or the GVL effect.14Neutralization of inflammatory cytokines must therefore be approached on an individual basis with respect to the potential to separate GVHD and GVL.

Cytokine shields: interleukin-11 and keratinocyte growth factor

An alternative approach to prevent GI tract damage during allogeneic BMT may permit the exploitation of intensive conditioning as an antileukemic modality without requiring T-cell depletion. This approach involves strengthening the GI mucosal barrier before BMT conditioning to prevent the entry of immunostimulatory molecules from the GI tract lumen into the circulation. Because direct shielding of the GI tract from TBI is not feasible, this effort relies on pharmacologic agents that provide a “cytokine shield” to reduce mucosal sensitivity to radiation, chemotherapy, or both. This approach is attractive because it blocks inflammatory cytokine dysregulation before the initiation of the cascade. In addition, by acting as indirect cytokine antagonists, these shields would not impede the physiological functions of cytokines in cellular differentiation (as might be the case with complete neutralization of TNF-α and IL-1). Two growth factors, IL-11 and KGF, have recently shown particular promise as cytokine shields.


IL-11, a member of the IL-6 cytokine family, is produced by a variety of tissues, including the central nervous system, thymus, lung, bone, skin, and connective tissue, and has pleiotropic effects.48 IL-11 stimulates megakaryopoiesis and accelerates neutrophil recovery after myelosuppressive therapy.49-51 In addition, IL-11 has potent anti-inflammatory effects by virtue of its ability to inhibit nuclear translocation of nuclear factor-κB (NF-κB).52 53 Preclinical studies have demonstrated the efficacy of IL-11 in treating inflammatory disorders, among them oxygen- and radiation-induced lung damage,54 55inflammatory bowel disease,56 and sepsis.57 In addition, IL-11 down-regulates IL-12 production by macrophages,58 which suggests that IL-11 may also modulate T-cell–mediated inflammation. Significantly, IL-11 has direct protective effects on the GI tract epithelium in models of injury by chemotherapy and radiation,59-63 surgery,64 65and ischemia.66 Small bowel crypt recovery is improved through the protection of clonogenic crypt cells,60reductions in apoptosis,62 and increases in cellular mitotic index.59 64 65 These studies confirmed that IL-11 has trophic effects on mucosal epithelium and suggested that continuing therapy with IL-11 after BMT conditioning would not impair mucosal healing. Of interest was the observation that maximum crypt protection occurred if IL-11 treatment was begun before irradiation and was continued for 3 days.60

In experimental studies examining the possible use of IL-11 as a cytokine shield to prevent GVHD, IL-11 administration was begun 2 days before conditioning and was continued for 7 to 14 days after BMT. When used in this fashion, IL-11 almost completely prevented GVHD of the small bowel, and it reduced serum endotoxin levels after BMT by 80%. Treatment with IL-11 also reduced TNF-α serum levels and suppressed TNF-α secretion by macrophages to LPS stimulation in vitro. Donor CTL responses to host antigens were not affected by IL-11. Surprisingly, IL-11 administration polarized the donor T-cell cytokine responses to host antigen after BMT with a 2-fold reduction in IFN-γ and IL-2 secretion and a 10-fold increase in IL-4. This polarization of T-cell responses was associated with reduced IFN-γ serum levels and decreased IL-12 production in mixed lymphocyte cultures. Therefore, IL-11 inhibits GVHD pathophysiology at multiple steps (Figure 1), enabling IL-11 to dramatically reduce GVHD mortality and morbidity rates after allogeneic BMT.67 The 10-day schedule of IL-11 treatment, beginning just before BMT, also provides long-term protection of the GI tract and improved immune reconstitution.67 Further studies showed that IL-11 spared donor CTL/NK function and preserved a GVL effect that was mediated by perforin, improving long-term leukemia-free survival.45These data confirm the ability of IL-11 to separate GVHD and GVL in experimental animal studies.

Keratinocyte growth factor

A member of the fibroblast growth factor family FGF-7, KGF shows specificity for epithelial tissues that express its receptors, including gut epithelial cells, hepatocytes,68 skin keratinocytes,69 alveolar type 2 cells,70mammary epithelium,71 and urothelium.72 In early studies, KGF administration before autologous BMT dramatically protected the gut epithelium from injury by lethal chemoradiotherapy.73 This protection appears to result from a potent trophic effect on intestinal epithelium68 and from improved survival of crypt stem cells,74 perhaps through reduced oxidative damage (nonselenium glutathione peroxidase)75 and enhanced DNA repair (DNA polymerase-α, -δ, and -ε).76

In 2 recent studies, investigators have examined the effect of human recombinant KGF on experimental GVHD.77 78 When KGF was given before conditioning only, it reduced the mortality rate and GVHD target organ histopathology.78 When KGF was administered from 3 days before to 7 days after BMT, it abolished GVHD of the GI tract and further improved survival.77 As expected, this improvement was associated with a reduction in serum LPS and TNF-α levels. Therapy with KGF also preserved donor T-cell responses (CTL activity, proliferation, and IL-2 production) to host antigens and significantly improved leukemia-free survival and when a lethal dose of P815 leukemia was given at the time of BMT (42% vs 4%; P < .001).77 KGF has a favorable clinical toxicity profile,79 and its administration, like that of IL-11, thus offers a novel approach to the separation of GVL effects from GVHD (Figure 1).


Insights into the pathophysiology of GVHD have confirmed a central role of GI tract damage in this process and have predicted for the success of novel approaches to preventing GVHD in experimental models. It is important to note that the majority of data regarding our understanding of GVHD pathophysiology derive from murine models that use inbred strains of donors and recipients with highly defined and often limited genetic disparity compared with our own human outbred species. The experimental conditions of these BMT models were highly controlled, and the mice spent their entire lives, from conception to death, in specific pathogen-free conditions. Both circumstances help to maximize the clarity of experimental results, but they may also lead to certain distortions with respect to the more variable clinical reality of human transplantation. It is unlikely that the mechanisms of murine and human GVHD differ completely, but the relative contribution of a given effector pathway to specific organ damage may well differ between humans and mice. Certainly, the use of intensive conditioning schedules and pharmacologic immunosuppression in clinical BMT is in contrast to most published murine systems in which GVHD target organ damage has been mediated primarily by T cells. Strong evidence supports the concept that inflammatory cytokines, GI tract damage, and LPS translocation are important mediators of clinical GVHD.9-11 Reduced GVHD severity seen in patients conditioned with nonmyeloablative conditioning regimens provides further support for this assertion.80 Nevertheless, some initial clinical studies of specific cytokine antagonists have yielded less compelling results than the murine models predict, and the effect of any single cytokine inhibitor may be small because of complex interactions of multiple, potentially redundant cytokines. Clinical testing of novel biologic response modifiers often occurs under high-risk conditions in patients with advanced disease, and considerable time may be required for these novel therapeutic regimens to be introduced under more standard risk situations, in which their true value in the prevention and treatment of GVHD can be discerned.

The preclinical data offer a compelling rationale for the testing of agents that can protect the GI tract and that therefore may modulate the inflammatory amplification of acute GVHD. Success in animal models does not always predict clinical efficacy, however. Given the potential toxicities of these agents and the severity of illness experienced by many patients under allogeneic BMT, the use of such agents should be tested in carefully monitored clinical protocols designed to test the feasibility and safety of this approach and its potential efficacy. If successful, the concept of “cytokine shields” may allow the potent antileukemic effect of high-dose chemoradiotherapy to be used without the induction of life-threatening GVHD and may permit traditional allogeneic BMT to be undertaken in a patient population that would otherwise be ineligible for this therapy.


  • Reprints: James L. M. Ferrara, Departments of Internal Medicine and Pediatrics, University of Michigan Cancer Center, 1500 E Medical Center Drive, Ann Arbor, MI 48109-0560; e-mail:ferrara{at}

  • The publication costs of this article were defrayed in part by page charge payment. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 U.S.C. section 1734.

  • Submitted August 24, 1999.
  • Accepted January 20, 2000.


View Abstract