EXPERIMENTAL MODELS OF SEPSIS AND THEIR CLINICAL RELEVANCE

Abstract

Sepsis remains a major cause of morbidity and mortality mainly because of sepsis-induced multiple organ dysfunction. In contrast to preclinical studies, most clinical trials of promising new treatment strategies for sepsis have failed to demonstrate efficacy. Although many reasons could account for this discrepancy, the misinterpretation of preclinical data obtained from experimental studies and especially the use of animal models that do not adequately mimic human sepsis may have been contributing factors. In this review, the potentials and limitations of various animal models of sepsis are discussed to clarify to which extent these findings are relevant to human sepsis. Such models include intravascular infusion of endotoxin or live bacteria, bacterial peritonitis, cecal ligation and perforation, soft tissue infection, pneumonia or meningitis models using different animal species including rats, mice, rabbits, dogs, pigs, sheep, and nonhuman primates. Despite several limitations, animal models remain essential in the development of all new therapies for sepsis and septic shock because they provide fundamental information about the pharmacokinetics, toxicity, and mechanism of drug action that cannot be replaced by other methods. New therapeutic agents should be studied in infection models, even after the initiation of the septic process. Furthermore, debility conditions need to be reproduced to avoid the exclusive use of healthy animals, which often do not represent the human septic patient.

INTRODUCTION

Sepsis remains a major cause of morbidity and mortality worldwide despite developments in monitoring devices, diagnostic tools, and new therapeutic options (^{1, 2}). It is a clinical syndrome resulting from a complex interaction between host and infectious agents, characterized by a systemic activation of multiple inflammatory pathways, including cytokine network and coagulation (³). The main cause of death is multiple organ failure, which is the final pathway for sepsis-induced systemic and regional hemodynamic changes, widespread microcirculatory disturbances, and cellular alterations, leading to an uncoupling between blood flow and metabolic requirements (^{4, 5}).

Extensive clinical and animal research, with substantial expenses, have been undertaken to address the pathophysiology and treatment of severe sepsis and septic shock (^{6, 7}). In contrast to many preclinical studies, most clinical trials of promising new treatment strategies for sepsis have failed to demonstrate efficacy (^{8, 9}). Although many reasons could account for this discrepancy, the misinterpretation of preclinical data obtained from experimental studies and especially the use of animal models that do not adequately mimic human sepsis may have been contributing factors.

We reviewed the experimental models of sepsis, addressing their limitations and benefits, to clarify the extent to which their findings are relevant to human sepsis.

LIMITATIONS

The animal model most frequently used at the beginning of preclinical studies is rodents (⁶). However, they are quite resistant to endotoxin, have distinct hemodynamic profiles, and limited blood volume in comparison with humans (¹⁰). Endotoxemia and bacteremia represent models without an infectious focus. They may reproduce many characteristics of sepsis and are highly controlled and standardized. However, they reflect a primarily systemic challenge without an infectious focus and the sepsis-induced immune reaction that characterizes human sepsis. Therefore, experimental models with an infectious focus are more clinically relevant (¹¹).

In human sepsis, gram-positive organisms and fungi have exceeded gram-negative organisms as a cause of sepsis, but they are very uncommon in animal studies (⁸), in contrary to gram-negative bacteria. This does not reflect the diversity of infectious agents, sites of infection, and progress of the infection encountered clinically. There is an increasing concern about possible important differences in host inflammatory responses to sepsis caused by gram-positive versus gram-negative bacteria (^{8, 12}). Experimental evidences suggest that the efficacy of mediator-specific anti-inflammatory agents in sepsis may be altered by the bacteria type of underlying infection with significant differences between gram-negative and gram-positive strains (⁸).

I.v. bacteria infusion, caused by the relatively large inocula that are required, probably constitutes a model of endotoxin intoxication rather than evolving infection. Intraperitoneal challenges typically require 100- to 1,000-fold fewer bacteria (¹³). Thus, the models used most extensively do not precisely replicate many important clinical parameters and do not duplicate the dynamic interactions among investigational drugs, microbial pathogens, and host defenses that occur in patients with sepsis.

Although promising agents are studied later in larger animals, including primates, and attempts are made to mimic various aspects of human septic shock, the experimental conditions encountered in human sepsis trials are more complicated than simulated even in large-animal models. Animals are carefully selected to have no preexisting diseases and to have a similar genetic background, age, weight, sex, and nutritional status. These animals are then challenged with a single well-defined precipitating event, whereas patients with sepsis patients are heterogeneous with respect to age, preexisting conditions, sources of infection, types of infecting microorganism, and many of them have experienced trauma or major surgery. Adequacy, rapidity and quality of care, including antibiotics, and timing of surgical intervention among others are critical for survival in human sepsis (¹⁰).

The natural history of severe sepsis in laboratory animals is generally distinct from human sepsis, with animals more often having a rapid onset of hypodynamic circulatory collapse and a more rapid resolution or decline to mortality. In clinical sepsis, the mortality is most commonly caused by the development of multiple organ failure days to weeks after initial presentation. For this reason, animal models that lead to significant mortality within the first 6 to 12 h may not describe an outcome that is relevant to humans (⁶). Agents under investigation are often administered to animals before or immediately after the induction of sepsis, conditions that can rarely be achieved in clinical trials.

EXPERIMENTAL MODELS

Signs and laboratory findings seen in human sepsis can be observed in a variety of animal models including intravascular infusion of endotoxin (^{9, 14-17}) or live bacteria (^{14, 18-21}), bacterial peritonitis (^{8, 22-25}), cecal ligation and perforation (^26-30), soft tissue infection (³¹), pneumonia model (^{32, 33}), and meningitis model (³⁴). Different animal species have been used including rats, mice, rabbits, dogs, pigs, sheep, and nonhuman primates (²⁶). Even though those models replicate many of the features of sepsis, it is important to critically evaluate the extent to which they mimic the septic picture.

ENDOTOXIN

Endotoxin is commonly used in animal models of sepsis. However, there is controversy over its relevance to our understanding of human sepsis. When administered to human subjects, endotoxin may mimic many of the features of sepsis (²⁶). In critically ill patients, increased concentrations of serum endotoxin have been associated with the development of sepsis, disease severity, and mortality (^{6, 26}). Detectable levels of endotoxin were identified in up to 75% in intensive care unit (ICU) sepsis patients in intensive care settings (³⁵). Endotoxin levels often remain undetectable in serum in more indolent forms of uncomplicated sepsis, with the recorded levels being of no prognostic significance (⁶). Occasionally, very high levels of endotoxin can be detected in patients with meningococcemia and with the beginning of the antibiotic therapy, killing bacteria (³⁶). The hypothesis that endotoxin plays a significant role in sepsis is supported by many studies that show that antibiotic administration may lead to a sudden release of massive amounts of endotoxin from dead bacteria and an acute hemodynamic deterioration (^{6, 12, 15, 29, 36}).

Endotoxin or LPS, as the principal component of the gram-negative bacterial cell wall, stimulates the release of inflammatory mediators from various cell types, responsible for initiating the process of sepsis (²⁶). LPS is a stable relatively pure compound that can be stored in lyophilized form. An accurate dose can be measured and may be administered as a bolus or infusion (²⁶). This has formed the basis for the simplest sepsis model and many endotoxicosis models (³⁵).

The sensitivity to endotoxin shows considerable differences between species. Rodents, cats, and dogs are relatively endotoxin resistant, whereas humans, rabbits, sheep, and nonhuman primates show an enhanced response (^{6, 10, 26}). In insensitive animals, presensitization with killed organism or D-galactosamine reduces the dose of LPS needed to produce an inflammatory response (³⁵). Despite lower doses being more physiological, most studies have continued to use high endotoxin doses in nonsensitized animals. The duration and route of administration have varied between studies as well.

A large i.v. dose of LPS in rats results in a sudden cardiovascular collapse and death (³⁷), whereas a lower dose promotes a hyperdynamic response, with an early increase in cardiac output (³⁸). Similarly, rabbits challenged with high doses of LPS (5 mg/kg) show low cardiac output and high systemic vascular resistance, but when challenged with a much lower dose (1-3 μg/kg), they manifest a hyperdynamic state (²⁶). A low dose of LPS (0.75 μg/kg) in sheep promotes a biphasic response characterized by an early reduction and a later increase in cardiac output, whereas a prolonged extremely low dose of LPS (9, 12, or 24 ng/kg per h for 24 h) promotes a delayed hyperdynamic state, with high cardiac output and vasodilatation (¹⁵). As in human sepsis, a higher dosage of endotoxin in sheep has been associated with a more profound myocardial depression (⁹). Endotoxin administration in dogs (2 mg/kg) provokes a severe hypodynamic state with an abrupt decrease in arterial pressure, cardiac output, hepatic blood flow, and an increase in systemic vascular resistance and blood lactate levels (³⁹). Most nonhuman primate endotoxicosis models have used massive i.v. doses, resulting in rapid circulatory collapse and early death, whereas lower doses more closely resembled human sepsis with coagulopathy and progressive multiorgan dysfunction (¹⁶).

Several authors argued that the endotoxin model is not a suitable one to study sepsis (^{6, 26}), despite the fact that endotoxin may play an important role in the pathogenesis of sepsis. The use of high doses of endotoxin in animals that are resistant to endotoxin has toxic effects that are not seen when low doses are administered to endotoxin-sensitive species, such as man (⁶). Although released by gram-negative sepsis, endotoxin is not released in gram-positive bacteria, but mortality is similar (³⁸). The use of corticosteroids and anti-TNF-α has been effective in animal models of endotoxemia, but has failed in clinical trials (^{6, 40}). In addition, killed Escherichia coli are more lethal than endotoxin. As the endotoxin is only one component of gram-negative bacteria, it is suggested that the other cell wall components may contribute to systemic inflammatory response (²⁶).

Thus, caution must be exercised whenever assessing clinical efficacy of novel therapeutic agents in animal models of endotoxemia (⁶). There is general agreement among researchers that LPS injection may serve as a model for endotoxic shock but not for sepsis (²⁶).

INTRAVASCULAR INFUSION OF LIVE BACTERIA

Because the rate of positive blood cultures is associated with increasing sepsis severity, (sepsis [17%], severe sepsis [25%], septic shock [69%]), it has been suggested that bacteremia plays an important role in the outcome of sepsis (²⁶).

Different aerobic bacterial species have been investigated to induce sepsis and septic shock (²⁶). Escherichia coli is the most common one. There is a wide variability in the dose of and duration of infusion, as seen with endotoxicosis models.

In small animals, low doses of E. coli, administered over several hours, have been associated with minimal early physiological changes, whereas higher doses have often produced a biphasic response, with an early rise and late fall in cardiac output (³⁵). In baboons, a bolus i.v. injection of LD₁₀₀ dose of E. coli induced an exaggerated TNF-α response with cardiovascular collapse and early death. In this model, pretreatment with a TNF-α inhibitor showed hemodynamic improvement and better survival, highlighting the role of TNF-α in cardiovascular collapse and resultant organ dysfunction and death (⁴¹). A severe disseminated intravascular coagulation induced by both sublethal and lethal doses of live E. coli and endotoxin has been described in baboons (¹⁴), resulting in a complex inflammatory and hemostatic response that involves the microvascular endothelium and its regulatory anticoagulant networks, thereby contributing to the multiorgan dysfunction development (¹⁴).

In a porcine model, in which both gram-positive and gram-negative bacteria were used at a similar dose, the hemodynamic and pulmonary changes depended on the bacterial species used. Whereas Staphylococcus aureus induced minimal changes, both E. coli and Pseudomonas aeruginosa resulted in shock and acute respiratory failure (²¹).

In an ovine model, a nonlethal dose of E. coli promoted a hyperdynamic cardiovascular response with hypotension, increased output, tachycardia, fever, oliguria, tachypnea, and hyperlactatemia (⁴). In a porcine model, P. aeruginosa infused over a week resulted in biphasic changes to the cardiac output, with late systemic hypotension and pulmonary hypertension (⁴²).

In dogs, both sublethal and lethal i.v. dose of live E. coli promoted early profound cardiovascular deterioration with hypotension, very low cardiac output, splanchnic hypoperfusion, and severe metabolic changes (^{18-20, 43-45}). Animals challenged by a lethal dose of E. coli (1.2 × 10¹⁰ colony-forming units/kg) presented only partial and transient improvements in systemic and regional blood flows during fluid resuscitation, but the progressive cardiovascular collapse was unavoidable (Fig. 1). In this model, the magnitude of changes at the splanchnic region was greater than the systemic ones. Moreover, the transient benefits after fluid replacement were much less evident within the splanchnic region, particularly at the microcirculatory level, as demonstrated by the PCO₂ gastric mucosal-arterial gradient (Fig. 1). This discrepancy between systemic and regional parameters has been well demonstrated in experimental and clinical studies (^{18, 43-47}). The intense compromise of splanchnic perfusion, particularly at the gut mucosa, has been implicated in the genesis, amplification, and perpetuation of the systemic inflammatory response and the progression of multiple organ dysfunction (⁴⁷). The pathophysiological basis that has been used to explain this phenomenon is that the gut hypoxia and/or ischemia contribute to gastrointestinal tract barrier dysfunction and translocation of cytokines, bacteria, and their products (⁴⁷).

Fig. 1: Experimental model of septic shock induced by lethal i.v. dose of live E. coli in dogs. Changes in MAP, cardiac index, portal vein blood flow index, and PCO₂ gastric mucosal-arterial gradient (PCO₂ gap) during the experimental protocol (mean ± SEM). BI-bacterial infusion (1.2 × 10¹⁰ colony-forming units/kg over 30 min); FR-fluid resuscitation period; CT-controls, no fluid (n = 7); RL-Ringer's lactate solution 32 mL/kg over 30 min (n = 7). Modified from Garrido (⁴³).

At the time of autopsy, those animals receiving live E. coli showed evidence of major inflammatory alterations and injury in lungs and liver (⁴³). An illustrative example of the impact of this model of septic shock is presented in Figure 2. Tissue inflammatory recruitment, presence of small-vessel thrombosis, vascular congestion, focal hemorrhage, and microabscess were observed in lung and liver. These alterations have been described in experimental and clinical sepsis (⁴⁸), and reflect the complex host systemic response with activation of multiple inflammatory pathways and the coagulation system contributing to widespread microcirculatory disturbances.

Fig. 2: Histological examination of the lung (A and B) and liver (C and D) in an experimental model of septic shock induced by lethal i.v. dose of live E. coli in dogs. A, Two recent fibrin thrombi in pulmonary small vessels. B, Intense inflammatory process in pulmonary parenchyma forming a microabscess. C, Liver parenchyma and portal triad display an intense mixed inflammatory infiltration. D, Microabscess in liver parenchyma. From Garrido (⁴³).

The acute i.v. live bacteria injection results in immediate cardiovascular collapse and early death, behavior rarely seen in human sepsis. However, within these limitations, this model may mimic extreme clinical sepsis such as seen in meningococcemia, pneumococcal bacteremia in splenectomized individuals, and gram-negative bacteremia in the setting of profound granulocytopenia (²⁶), which are increasingly seen nowadays. This model also allows the study of the acute effects of early interventions in short periods, such as fluid replacement. Large animals allow the use of systemic and regional monitoring similar to the ones used in ICUs, in addition to regional blood flow measurements and blood sampling only feasible in experimental studies.

As in other models, the hemodynamic response to the challenge with live bacteria also depends on fluid resuscitation and antibiotic treatment. The use of fluid resuscitation has varied between studies, but in its absence, early death is frequent. There is considerable evidence that antibiotic therapy may, through the destruction of bacteria and release of substantial amounts of cell wall components by gram-negative, gram-positive, and fungi, promote an intense inflammatory response with excessively high levels of TNF-α, causing hemodynamic deterioration, especially in the absence of fluid resuscitation (^{12, 29, 36}).

Not all models using an intravascular infusion of live bacteria are acute preparations characterized by abrupt cardiovascular deterioration secondary to overwhelming bacteremia. Shaw and Wolfe (⁴⁹) described a chronically instrumented unanesthetized canine model, wherein animals were infused intra-arterially with viable E. coli and were studied 24 h later. The animals were aggressively resuscitated at the time sepsis was induced, the dose of bacteria being invariably lethal in the absence of adequate restoration of intravascular volume. However, with fluid resuscitation, 85% of the animals survived the protocol and, at the time of study, were hyperdynamic and hypermetabolic (⁴⁹). In addition, many of the hormonal alterations typical in humans with sepsis were observed. Although not widely used, this model mimics many of the features of clinical sepsis and avoids the confounding effects of anesthesia and surgical preparations.

Because most patients are not challenged with a massive bacterial load at any time, but rather harbor a septic focus that is intermittently and persistently showering the body with bacteria, several authors have questioned the relevance of models using a bolus infusion of viable bacteria (²⁶). In addition, concerns over the use of an appropriate strain of an infective organism extend also to studies of endotoxemia, where the most commonly used strain of endotoxin is uncommonly seen in human bacteremia. There are a number of features of either the host or host-bacterium interactions that are species specific (⁶). For example, Salmonella typhi does not cause systemic infection in laboratory rodents but is responsible for typhoid fever in humans. In mice, a related bacterium, Salmonella typhimurium, causes a systemic infection and it is commonly used as a model of human typhoid infection, despite its low virulence in humans (⁶). Individual bacteria may also cause a broad spectrum of disease processes, depending on the expression of virulence genes. These findings suggest that it may be difficult to make conclusions regarding the efficacy of a therapy based on a study looking at infection with a single bacterial strain, particularly if it is a pathogen not commonly seen in critically ill patients (⁶). Despite these criticisms, numerous laboratories continue to use intravascular infusions of viable bacteria to induce sepsis in animals, and many of these models remain very useful, provided that certain inherent limitations are recognized (²⁶).

PERITONITIS MODELS

Peritonitis may be induced in animals in several ways. Bowel can be perforated, allowing contamination with gastrointestinal contents, or inocula of fecal material or pure bacterial cultures can be instilled into the peritoneal cavity (³⁵). In early models, segments of intact bowel were isolated, and the development of peritonitis was expected (³⁵). The disadvantage of this model was that the onset of peritonitis was uncontrolled and depended on the timing of gastrointestinal perforation. To overcome this limitation, the cecal ligation and puncture (CLP) model was developed as a simple and reproducible model that has been used widely in sepsis research (^{24, 25}). The cecum is ligated distal to the ileocecal valve and perforated using two needle punctures (Fig. 3). Needle size can be used to manipulate CLP to give a lethal and nonlethal sepsis (^{8, 26}).

Fig. 3: Aspect of the necrotic cecum after ligation and puncture for the evaluation of leukocyte-endothelial interactions displayed in Figure 5 (28).

The principal advantage of CLP models is their simplicity. Because sepsis is induced by a straightforward surgical procedure, there is no need to grow and quantify bacteria or in other ways prepare the inoculum. Furthermore, these are models of sepsis caused by peritoneal contamination with mixed flora in the presence of devitalized tissue and thus establish a clear resemblance to clinical problems such as a perforated appendicitis or diverticulitis (³⁵). This technique, without fluid resuscitation, promotes rapid onset of shock. After fluid resuscitation, mortality rate may be reduced with pathophysiological responses resembling those noted in human sepsis (²⁹). The use of intravital microscopy in the vascular mesentery (Fig. 4) allows in vivo observation of altered mesenteric leukocyte-endothelial interactions after cecal ligation/puncture and their modulation by different fluid regimens, and reversion after surgical sepsis source control (Fig. 5) by removing the necrotic cecum (^{27, 28}).

Fig. 4: Intravital microscopy in rat mesentery, allowing the evaluation of leukocyte-endothelial interactions shown in Figure 5.

Fig. 5: Number of rolling leukocytes/10 min, adherent leukocytes/100 μm venule length, and migrated leukocytes/5,000 μm² in rat mesenteric microcirculation. Rats were sham-operated (SHAM, n = 6), submitted to CLP (n = 7), or to CLP + necrotic tissue resection/peritoneal lavage (REL, n = 6). *P < 0.05 among CLP versus SHAM and CLP + REL groups. Removal of the necrotic tissue restored microcirculatory disturbances induced by sepsis. From Nagakawa et al. (²⁸).

However, it is difficult to control the magnitude of the septic challenge in the CPL model. In studies using small animals, the problem of variability is easily overcome by increasing the sample size, whereas variability remains a problem in larger species. The gastrointestinal contents of animals, particularly herbivores, vary between species. Initial attempts to induce peritonitis by intraperitoneal implantation of feces were often disappointing, with animals appearing tolerant to their own fecal flora (³⁵). To overcome those limitations, human feces were used, or barium sulfate, bile salts, or autologous hemoglobin added to the fecal material (⁸).

Both CLP and fecal inoculation models deliver a variable microbiological dose (⁸), so pure bacterial culture peritonitis models have been developed. In 1980, Ahrenholz and Simmons (⁵⁰) showed that 24-h mortality was 100% when viable E. coli suspended in saline was injected intraperitoneally in rats. However, when the same number of bacteria was implanted intraperitoneally in a bovine clot, early mortality was prevented, but the rats developed abscess, and the 10-day mortality rate was 90% (⁵⁰). Thus, fibrin delays the systemic absorption of the entrapped bacteria and promotes the development of chronic intraperitoneal abscess, a more local septic focus (²⁶). This highly reproducible model has been described in small and large mammals, and displays many features of human sepsis including insidious onset, hyperdynamic cardiovascular state (^{8, 25}), reversible compromised myocardial performance (²⁵), and a high mortality rate (^{8, 26}). Moreover, unlike other peritonitis models using fecal implantation or cecal ligation/perforation, the fibrin clot model allows control over the dose of bacteria and the type of organism implanted (²⁶). Mixed cultures more accurately mimic the gastrointestinal flora than single-organism cultures (²³), which can fail to reproduce the synergy between aerobic and anaerobic organisms seen in human peritonitis. The aerobic gram-negative organism seems to be responsible for many of the acute physiological features of sepsis, whereas anaerobes seem to contribute to the development of intraperitoneal abscess (³⁵).

CONCLUSIONS

Many of the problems with the use of animal data in sepsis drug development stem, not only from the animal models per se, but from how those results have been adapted to clinical trial designs. Animal models provide insights about specific components of the septic process but cannot truly mimic the full clinical complexity and intrinsic heterogeneity of patients with sepsis. Despite these limitations, animal models will remain essential in the development of all new therapies for sepsis and septic shock because they provide fundamental information about the pharmacokinetics, toxicity, and mechanism of drug action that cannot be duplicated by other methods.

Nonetheless, there is much to be improved in animal experiments. Examples are the need for long-term studies with ICU-like conditions to simulate the often delayed onset of organ dysfunction in the clinical setting, using sepsis or organ dysfunction criteria to start treatment instead of a fixed time schedule. New therapeutics agents should be studied in infection models even after initiation of the septic process. Furthermore, debility conditions need to be reproduced to avoid using healthy animals, which often do not represent the human sepsis patient.