Murine responses to endotoxin: another dirty little secret?

For over a century, clinicians and investigators have injected bacterial endotoxins (lipopolysaccharides, LPS) into the veins of animals. Remarkably, some vertebrates (humans, horses, rabbits) are as much as 100,000-fold more likely to succumb to an intravenous dose of LPS than are others (baboons, mice, rats, chickens). With the recent triumph of the laboratory mouse as the major experimental animal model for infectious diseases in humans, the striking difference between human and murine sensitivity to LPS toxicity seems to have become a “dirty little secret” of innate immunology, just as Janeway’s clue to the importance of innate immunity--the potency of adjuvants—was long ignored by students of B and T cells (1). Like many other immunological differences between these species (2), the possibility that results of experiments in mice using LPS and other stimulatory microbial molecules might not apply to humans has been infrequently acknowledged. Now Warren and Cavaillon have called attention to this issue by studying it (3). Their conclusion – that the key factor(s) that account for the striking difference in sensitivity to endotoxin infusion are found in serum, not host cells – should help us think more clearly about many mouse models of human disease. Their results also raise the interesting possibility that these factors might be identified and harnessed to dampen human responses to bacterial diseases.

Remarkably, murine serum also inhibited macrophage responses to microbial molecules other than LPS – peptidoglycan-associated lipoprotein, toxic shock syndrome toxin-1, and others. Whereas the serum factor(s) inhibited LPS-induced production of pro-inflammatory cytokines, there was no diminution in the production of a prominent anti-inflammatory cytokine, IL-10, or the key chemokine, IL-8. Moreover, the serum overlying macrophages could be washed away without removing the inhibitory factor(s). These findings suggest strongly that the serum factor(s), rather than neutralizing LPS per se, act on macrophages to alter their responsiveness to many different agonists. One possibility is that they elevate intracellular cyclic AMP, thus (like epinephrine, prostaglandin E2, and histamine) reprogramming cells to retain IL-10 production in response to agonists while diminishing production of TNF and IL-1β. Mechanism aside, the translational impact of the serum factor(s), when they are identified and purified, could be much broader than would be suggested by the emphasis here on LPS.

As summarized by Bahador and Cross (4), endotoxin infusion experiments have taught us much of what we know about how animals respond to Gram-negative bacteria. It must nonetheless be acknowledged that a bolus intravenous injection of endotoxin does not mimic any Gram-negative bacterial disease. The studies are typically done in healthy experimental animals or volunteers, not in subjects with pre-existing morbidities. The bolus injection does not reproduce natural endotoxemia, which is intermittent and low grade. There is no preceding tissue infection from which endotoxin enters the bloodstream. The initial response to endotoxin infusion is strikingly pro-inflammatory (TNF and IL-1 appear in the blood long before IL-10 does), yet there is little evidence that this pattern occurs in the bloodstream of infected humans. The human infectious disease most closely modeled by endotoxin infusion is fulminant meningococcemia, which may develop rapidly in previously healthy people, yet endotoxin accumulates in the blood of these individuals over several hours and there is suggestive evidence that a prominent anti-inflammatory response to LPS may increase disease severity (5).

Given these reservations about the infusion model, is an animal species’ LD₅₀ for bolus endotoxin infusion a valid measure of its ability to withstand Gram-negative bacterial disease? There is evidence that LPS is the major inflammation-inducing molecule in the Gram-negative bacteria that produce it (6;7). Death from endotoxin challenge could thus reflect an inability to withstand the bacteria that made the LPS. In discussing this issue, Warren et al. take a cue from plant biologists and veterinarians. Botanists found that plants may differ in their ability to withstand infectious agents, not merely by resisting microbial invasion (a.k.a. “resistance”) but also by putting up with some microbes once they have invaded. The term “tolerance” was suggested for the latter, with plants showing different degrees of resistance and tolerance to different microbes. A similar concept has been proposed for animals. In humans, examples suggested to represent tolerance to microbes include persistent parasitemia in individuals immune to malaria and the mild course of malaria in patients with alpha thalassemia or SA hemoglobin (8). In each of these situations, microbes are tolerated for long periods of time with little apparent damage to their hosts.

The term “tolerance” has another time-honored usage in innate immunity (9). Animals that survive an exposure to endotoxin or Gram-negative bacteria often develop tolerance to subsequent challenges with LPS and other microbial agonists. Provided the animal can chemically inactivate the LPS molecules (10), this state of altered responsiveness typically lasts for only a few days. In discussing whether an animal’s ability to withstand Gram-negative bacteria can be measured using endotoxin infusion with death as the end-point, Warren et al. avoid the term “tolerance” and use “resilience” (11) in its stead. Their notion of resilience differs from previous authors’ by not requiring the presence of either living microbes or bioactive microbial molecules for prolonged periods in vivo.

The extent to which animals withstand bloodborne Gram-negative bacteria and their LPS seems to depend in part upon both the structure of the lipid A moiety of the LPS and the host’s responses to it (the local and systemic reactions triggered by activation of MD-2--TLR4, the signaling receptor for LPS (6)). If one were to measure resilience as the degree of morbidity (say, fever or acute phase protein elevation) per colony-forming bacterium in cultured blood, the most ‘resilient’ humans would probably be those who do not sense Gram-negative bacteria as they grow in vivo, either because the bacteria make LPS that is not detected by MD-2—TLR4 (Yersinia pestis, for example (12)) or perhaps because the early host response is dominated by anti-inflammatory molecules (Neisseria meningitidis (5;13)). In each case, “resilience” would occur during the calm before the storm, as the bacteria are growing to very high density and are about to elicit responses that will help kill the host. It is unlikely that resilience is beneficial in these instances or that it would be reflected in the reactions of volunteers to an intravenous bolus of endotoxin.

An animal species’ threshold tissue sensitivity to a microbial agonist (its ability to sense low concentrations in extravascular tissues) may be a better predictor of an efficacious host defense than is its maximal tolerated intravenous dose. Despite mouse serum’s ability to inhibit macrophage responses to LPS, successful defense from many experimental Gram-negative bacterial infections requires murine cells that express TLR4, the LPS signaling receptor. This apparent paradox may be explained by the important role that LPS recognition by macrophages and other phagocytes plays in killing Gram-negative bacteria in extravascular (subepithelial or submucosal) tissues, before they can enter the bloodstream, and in clearing Gram-negative bacteria from the circulating blood (6). There is much evidence for compartmentalization of responses to LPS and other agonists within animals (14–16), with pro-inflammatory local tissue responses being accompanied by anti-inflammatory changes in the systemic compartment (blood and distant organs) (17). In humans, numerous plasma molecules prevent or modulate cell stimulation by microbial products, either by neutralizing the agonists (bactericidal permeability-increasing protein, lactoferrin, antibodies, plasma lipoproteins, lysozyme, CD14), or by dampening host cellular responses to them (epinephrine, α-melanocyte stimulating hormone, prostaglandin E2, cortisol, IL-10 and many others) (18). Importantly, the concentrations of many of these molecules increase in the blood during the acute phase response to tissue infection and trauma. In fact, serum taken from sick humans inhibits the ability of naive macrophages to respond to LPS and other agonists. It will be interesting indeed if the murine serum molecule(s) that inhibit macrophage responses to LPS and other agonists are ones that increase in human blood and render it more anti-inflammatory in response to infection.

How might acknowledging mouse-human differences influence investigations of bacterial diseases in mice? First, intravenous injections should be avoided if an experiment is intended to test susceptibility to bacterial infection or sensitivity to LPS or other microbial molecules. Intraperitoneal injections may also allow too-rapid entry of bacteria or LPS into the blood. Models should begin with local tissue infection, so that bacterial multiplication and the elaboration of toxins and other virulence factors will occur in vivo in an extravascular compartment and over a period of many hours to days. This should allow the acute phase reaction, an important arm of innate immunity (17), to be expressed prior to the onset of bacteremia. One model that approaches this ideal is cecal ligation and puncture, which has become widely used in recent years; pneumonia models also may meet this criterion. Second, a clearer distinction should be made between Gram-negative commensals (in particular, avirulent E. coli strains and other commensal Enterobacteriaceae) and pathogens (Salmonella, Yersinia, Francisella, etc.) (19). Commensal bacteria rarely cause disease in humans whose innate immune defenses are intact. Murine models to test the virulence or toxicity of these bacteria should include some pre-existing morbidity, such as neutropenia, trauma, surgery, or thermal injury. These insults also allow the acute phase response to occur before the bacterial challenge begins. Such prior morbidities are much less important for modeling pathogens that can cause disease in previously well people. Third, the minimal infectious dose (ID₅₀) should be determined and used whenever possible (20). Most of the bacteria found in a diseased tissue are the progeny of a single organism (21); using large inocula intentionally violates this point and unnaturally overwhelms host defenses. Similarly, low tissue doses of LPS or other microbial agonists are more likely to induce “natural” responses than are large ones. A genetic defect exposed in mice by a large bacterial inoculum may not predict significant human deficiency; mutations in several innate immunity genes have been associated with infection susceptibility in mouse models that has not been observed in humans (22).

These measures may improve odds that an experiment performed in mice will model human disease, but unfortunately they do not address the most important issue raised by Warren’s article. What if the mouse-human difference in responsiveness to TLR agonists exists not only in serum but also in extravascular tissues? Murine and human macrophages have similar sensitivity to LPS and other microbial agonists when they are cultured with fetal calf serum in vitro, but we don’t know that they are equally sensitive in vivo, or even that in vivo tissue responses are reliably predicted by in vitro macrophage sensitivity. It is possible that mice are poor models for both tissue and bloodstream infections in humans. More data on this point are clearly needed.

The provocative article by Warren et al. encourages us to reconsider how we use experiments in mice – whether for understanding bacterial pathogenesis or for evaluating drugs and vaccines intended for use in humans.