Background: Marked heterogeneity exists among patients with sepsis, both in terms of distribution of organ dysfunction and its severity. Such heterogeneity could be explained by the presence of multiple subtypes of sepsis that may have important implications for treatment. Methods: Narrative review of published literature involving endotoxin from 1970 to 2022. Results: In humans, endotoxemia is most consistently associated with a specific pattern of organ failure including shock, endothelial dysfunction, acute kidney injury, and hepatic dysfunction. This pattern is consistent with complement activation and uncontrolled inflammation, two features of endotoxemia. Unbiased discovery using artificial intelligence also identifies a subtype of sepsis which features these same organ failures. Conclusion: Endotoxin appears to represent an important molecular phenotype of sepsis with unique clinical features and high mortality.
The concept of endophenotype has been advanced for its ability to differentiate between potential diagnoses that present with similar symptoms. For example, in psychiatric genetics (e.g., bipolar disorder and schizophrenia), it can be used to bridge the gap between symptoms and genetic variability . More recently, endophenotypes have triggered interest among researchers studying addiction, Alzheimer’s disease, obesity, cystic fibrosis, and asthma. Sepsis is defined by the Sepsis-3 definition as life-threatening organ dysfunction caused by a dysregulated host response to infection . Marked heterogeneity may exist from patient to patient, both in terms of distribution of organ dysfunction and its severity. Such heterogeneity could be explained by the presence of multiple subtypes (endophenotypes) of sepsis that may have important implications for treatment.
Sepsis Phenotypes Using Artificial Intelligence
Seymour et al.  analyzed data from over 60,000 patients and used machine learning to derive 4 novel sepsis phenotypes (α, β, γ, and δ) with different demographics, laboratory values, and patterns of organ dysfunction. In the simulations using data from 3 randomized clinical trials involving 4,737 patients, the outcomes related to the treatments were sensitive to changes in the distribution of these phenotypes. Compared to other phenotypes, the δ-phenotype, occurring in up to 15% of patients, is characterized by greater rates of acute kidney injury (AKI), hepatic dysfunction, and endothelial dysfunction. Mortality rates are much greater with the δ-phenotype – 32% at hospital discharge compared to 2% for the α-phenotype .
These phenotypes are not related to the site of infection or the organism but are correlated with inflammation (γ and δ being much higher than α and β). The distribution of patients exhibiting each phenotype is very consistent over time – 30–35% α, 25% each β and γ, and 10–15% δ. However, mortality corresponding to these phenotypes has all improved dramatically over the last 2 decades, with exception of the δ phenotype, which remains near 40%. For example, in patients enrolled in a trial of drotrecogin alpha reported in 2001 , patients with the α phenotype experienced a 15% 28-day mortality . In a trial of early goal-directed therapy reported in 2014 , the α phenotype resulted in a mortality of only 6% at 28 days. Conversely, mortality was approximately 40% for the δ phenotype in both trials .
Sepsis Phenotypes Based on Clinical Syndromes
Several rare diseases can be triggered by certain types of infection, especially in patients with genetic susceptibility. Two such cases are macrophage activation syndrome (MAS) and atypical hemolytic uremic syndrome (aHUS). Presentations of both are sepsis-like. Organ failure clusters that are common in sepsis (Fig. 1) include AKI together with endothelial dysfunction (thrombocytopenia with anemia but without coagulopathy) reminiscent of aHUS; and liver injury together with endothelial dysfunction (thrombocytopenia plus coagulopathy) reminiscent of MAS. Phenotypes resembling both of these two syndromes have been described in both adults and children with sepsis and genetic markers may be present for both [6‒9]. For example, aHUS encompasses a group of disorders that results from dysregulation of the alternative complement pathway. Up to 60% of patients with aHUS have an identifiable pathogenic gene variant in the complement pathway . In an analysis of more than a thousand adult patients with sepsis, serum ferritin was used to select patients with severe inflammation, and six were subjected to whole exome sequencing. All six exhibited one or more pathologic or potentially pathologic gene variants and half had variants associated with aHUS (the other half, interestingly, associated with MAS) . However, aHUS is known to have incomplete genetic penetrance, with only 40–50% of carriers with known pathogenic variants going on to develop the clinical syndrome. It is therefore proposed that aHUS occurs following a trigger event, such as infection or other event that activates complement. Unregulated complement activation ultimately results in the formation of the C5b-C9 membrane attack complex, leading to more endothelial injury and the formation of microvascular thrombi .
MAS is caused by natural killer cell dysfunction and is a life-threatening complication of systemic inflammatory disorders, most commonly systemic juvenile idiopathic arthritis or Still’s disease in adults, as well as systemic lupus erythematosus. Clinical and laboratory features of MAS include sustained fever, hyperferritinemia, pancytopenia, fibrinolytic coagulopathy, and liver dysfunction. Thus, MAS looks very much like sepsis, especially in cases where pancytopenia is absent or only thrombocytopenia is manifest. A definitive diagnosis of MAS can be elusive. Several investigators from around the world have described “MAS features” in association with some severe cases of sepsis in both adults and children. Shakoory et al.  performed a post hoc analysis of a prior sepsis trial and found that anakinra improved mortality in a subset of patients with MAS features defined by disseminated intravascular coagulation (DIC) together with liver dysfunction. The MAS phenotype was present in 5.6% of the sepsis population and treatment with anakinra reduced the 28-day mortality in this subgroup from 64.7% to 34.6% (p < 0.05) . More recently, Anderko et al.  reported that 6% of patients with septic shock in the ProCESS trial  (n = 82/1,341) had this same MAS phenotype, which was an independent risk factor for 90-day mortality (OR = 4.19, 95% CI = 1.13–16.39, p = 0.034). Relative to sepsis controls, the MAS cohort demonstrated increased levels of 21 of the 26 MAS-associated biomarkers (p < 0.05). This panel was highly predictive of both the MAS phenotype (sensitivity = 82%, specificity = 84%) and mortality (sensitivity = 92%, specificity = 90%). The authors concluded that their results suggest that liver dysfunction plus DIC identifies patients with sepsis who might benefit from MAS-directed therapies .
Identification of these syndromes is important because specific therapies exist. Anti-C5 monoclonal Ab (eculizumab) is now standard therapy for aHUS and other drugs targeting complement are being developed. For MAS, the traditional treatments of high-dose steroids, cyclosporin, and even etoposide in refractory cases are being replaced by various specific therapies including IL-1 receptor antagonist (anakinra, canakinumab); anti-IL-6R monoclonal Ab (tocilizumab); IL-18 binding protein; CTLA4-Ig (abatacept); and JAK inhibitor (tofacitinib). Significant experience has been gained with these therapies for the management of COVID-19, though results have been variable, perhaps related to the absence of careful selection of patients with MAS.
Establishing molecular pathogeneses for subtypes of sepsis could lead to breakthrough therapies. Complement activation is an attractive target because of existing therapies (e.g., eculizumab), although complement inhibition in the setting of active infection would be dangerous. The multiple therapies developed for MAS are being reevaluated for sepsis in the wake of experience with COVID-19. The δ-phenotype described by Seymour et al.  appears to have features consistent with both aHUS and MAS and may include both phenotypes.
Interestingly, aHUS and MAS, together forming the bulk of the δ-phenotype, may have a common pathogenesis. Endotoxin is an important molecular target for the δ-phenotype because endotoxin activates both complement and cytokines. Susceptible patients including those with underlying predispositions may develop either MAS or aHUS-like syndrome or both in response to endotoxin. Animal models routinely rely on high-dose endotoxin and in an unusual case of self-injection intravenously of high-dose endotoxin, a patient developed profound shock, AKI, hepatic and endothelial dysfunction with relatively spared pulmonary and neurologic function . Endotoxin triggers inflammation through Toll-like receptor 4 and other pathways and contributes to the pathophysiology of sepsis; a third to a half of patients with septic shock exhibit high levels of endotoxin activity in their plasma . This signal is not related to primary bloodstream infection and appears instead to result from barrier dysfunction in the gastrointestinal track with translocation of endotoxin into the circulation. Evidence that the effects of endotoxemia may be reversible comes from a case report by Akitomi et al. . These authors describe whole blood gene expression profiling in a patient with sepsis treated with polymyxin-B hemoperfusion to remove circulating endotoxin. Comparative gene expression analysis of whole blood from the patient identified 867 upregulated genes and 1,467 downregulated ones. Upregulated genes were found to be involved in oxidative stress, whereas those downregulated were related to neutrophil defensins, tumor necrosis factor-α/nuclear factor-κB, interleukin-8, and -6 signaling cascades, and pyruvate metabolism.
Endotoxin activity can be measured by the FDA-approved assay, EAA, and in the EUPHRATES trial, patients were enrolled who exhibited endotoxin activity >0.6 units . Compared to the ProCESS trial , which enrolled the same clinical phenotype of septic shock but without the EAA, patients in EUPHRATES exhibited much higher rates of AKI, liver dysfunction, and thrombocytopenia with or without DIC (Table 1).
In many parts of the world, especially Japan and Italy, hemoperfusion is used to remove endotoxins from the bloodstream. Polymyxin-B hemoperfusion is highly effective in endotoxin removal and large registries amassed over the last 3 decades show a consistent, albeit small, improvement in survival in unselected patients with septic shock . Careful section of patients by organ failure criteria together with endotoxin activity between 0.6 and 0.89 units has been shown to result in a large improvement in survival in a recent post hoc analysis of data from the EUPHRATES trial  and is the subject of an ongoing randomized trial in the US (NCT03901807).
Endophenotypes of sepsis have been described and may reflect underlying pathology linked to complement activation and natural killer cell dysfunction. Endotoxemia could be the “missing link” in the pathogenesis of these cases.
Conflict of Interest Statement
John Kellum and Debra Foster are full-time employees and hold stock in Spectral Medical. Paul Walker is stockholder and board member in Spectral Medical.
John Kellum drafted the manuscript and Debra Foster and Paul Walker reviewed and revised the text.