Predicting cytokine storms: it's about density

Matthew J. Frigault and Carl H. June

Römer and colleagues in this issue of Blood report that the density of T cells during culture increases sensitivity to the CD28 cosignaling agent TGN1412, providing an improved means of predicting cytokine release syndrome (CRS).1

CRS is a common clinical event with antibody therapies such as rituximab and anti-CD3 antibody. After the infusion of the CD28 superagonist TGN1412 all 6 healthy recipients unexpectedly suffered a severe form of CRS, often referred to as a cytokine storm.2 The massive release of proinflammatory cytokines rapidly progressed to severe multiorgan failure requiring advanced medical resuscitation. The events of the March 2006 first-in-human trial of TGN1412 have since sparked a fundamental discussion regarding the failure of preclinical models to predict toxicities and how new biologics are assessed before first-in-human testing (see figure).

Density-dependent effects of CD28 cosignaling in vivo. In the peripheral blood T-cell compartment, T cells are at low density, and while they express CD28 the receptor is not competent to transmit signals to trigger cytokine release after ligation by “superagonist” TGN1412. In contrast, in secondary lymphoid organs, T cells reside at higher density and CD28 is competent to release large amounts of TNF-α and other cytokines on ligation. A mechanism suggested for this density-dependent effect may be signals transmitted by MHC class I and II to T-cell receptors.

These events were surprising given the benign preclinical safety testing. An extensive investigation concluded that the adverse events involved no contamination or errors in the manufacturing, formulation, dilution, or administration of TGN1412.3 Theories explaining the disparity between the initial preclinical studies and life-threatening outcomes observed in these 6 patients began to emerge. Findlay et al suggested that the initial in vitro studies missed the potential proinflammatory outcome because of functional differences between soluble and bound antibody.4 The initial in vitro studies of peripheral blood mononuclear cells (PBMCs) were done in a fashion similar to stimulation using soluble OKT3. Under these conditions, no inflammatory response was elicited; however, once artificially bound to a cell-culture surface, TGN1412 reproduced a similar inflammatory response in human PBMCs. In fact, the introduction of the soluble form actually inhibited cytokine production, presumably via epitope blockade.

Initial in vivo studies also failed to predict the induction of CRS. TGN1412 showed promising results in the treatment of autoimmune disease, activating and expanding polyclonal Tregs in immunecompetent mice.5 These mouse studies most likely failed to reproduce the dramatic inflammatory response because of the differences of host immune system and cumulative antigen exposure history between young laboratory mice and humans. In mice, natural Tregs quickly act as IL-2 sinks and given a relative paucity of memory cells are able to quench the immune response induced by TGN1412. The much larger proportion of CD4+ effector memory cells present in an antigen experience human immune system most likely overwhelmed this same immunosuppressive safeguard.3.

Initial dose selection may explain the rapid onset of symptoms observed in all 6 patients. For the first-in-human trial the starting dose was calculated from the no observed adverse effect level (NOAEL) determined in preclinical animal models. Simply put, the maximum dose at which no statistically significant or biologically relevant adverse event was determined, and following various correction factors, a dose of 0.1 mg/kg/d, 16 times less than the maximum safe starting dose (MSD) and 500 times less than the dose administered to macaques, was chosen. However, these calculations depend on the presence of a similar biologic effects between test animal and humans. With no recorded adverse events described, even with doses up to 50 mg/kg, the macaque model poorly represented the activity of TGN1412 in humans. Later analysis would show that human subjects were actually given a near maximum dose with in silico calculations predicting 86% to 90% CD28 receptor occupancy.6 Because of these findings, most trials with biologics now select first-in-human dosing using the minimum anticipated biologic effect level (MABEL), which calculates initial administration based on the dose at the lowest end of the dose-response curve.

This failure of primates to respond to TGN1412 was later traced back to differences in immune regulation between humans and macaques. Eastwood et al demonstrated that unlike other antibodies bound to the surface of lymphocytes such as rituximab or alemtuzumab, which also induce a similar but significantly less dramatic cytokine release of TNF-α and IL-8, TGN1412 is unique in that it also causes robust secretion of IL-2 and IFN-γ.7 This difference was attributed to the unique effect of TGN1412 acting directly on CD4+CD45RO+ effector memory T cells, the predominant source of these inflammatory cytokines. Furthermore, they demonstrated that the macaque equivalent of human CD4+ effector memory T cells actually lacks CD28 expression, thereby preventing CD28 superagonist effects. Other species specific effects of CD4+ T-cell differentiation have been uncovered in nonhuman primates, which explains in part the relative pathogenicity of HIV and SIV infections.8

One last question remained as to how TGN1412 led to cytokine storm given that its effects on whole blood and PBMCs could not be reproduced in vitro unless antibody was artificially immobilized on a plastic surface. Römer et al may now have an explanation. They have shown that monocytes and T cells up-regulate functional activity during high-density culture mediated by adhesion-mediated up-regulation of TCR signal sensitivity as well as TCR priming by surface scanning for MHC class I and II molecules.1 Tissue-resident CD4+ effector memory T cells, which comprise upwards of 90% of the body's lymphocytes, are therefore able to immediately respond to TGN1412 with cytokine release. This is supported by their work showing that lymph node cells, but not PBMCs, are able to mount a response to soluble TGN1412. This response was further attributed to CD4+CD45RO+ cells and declined if cells were allowed to dissociate. Collectively, these data demonstrate the importance of appropriate preclinical modeling that adequately predicts lymphocyte biology.

There are several other implications from this work by Römer and colleagues. First, most studies of lymphocyte signal transduction are conducted with cells at low density, and thus may not represent in vivo signaling, particularly in the case of cosignaling through CD28. One of the “dirty little secrets” of immunology is that lymphocyte cloning is often superior in round-bottom rather than flat-bottom vessels.9 Studies with large-scale lymphocyte cultures indicate improved growth at high density rather than low density.10 Finally, these studies imply that the peripheral blood T-cell compartments is likely a “safe haven” in that the T cells are hyporesponsive compared with their tissue-bound counterparts, and this may provide some protection to superantigen-induced CRS, which is dependent on CD28.11


  • Conflict-of-interest disclosure: The authors declare no competing financial interests. ■