As a result of these data, Ziegler and colleagues [4] did a randomized, double-blind, placebo-controlled clinical study in 136 patients to assess whether treatment with antiserum to the inner core of the Escherichia coli O111:B4, Rc (J5) mutant would decrease the mortality from gram-negative bacillary sepsis. This study identified two patient subgroups likely to derive benefit from this antiserum: those with severe sepsis and gram-negative bacteremia (39% mortality in the placebo group compared with 22% in the J5 antiserum group, P = 0.011) and those with septic shock requiring vasopressors for more than 6 hours (77% mortality compared with 44%, P = 0.003). They were unable, however, to show that the ameliorative effect of the antiserum was due to the J5-specific antibody.

With the advent of monoclonal antibody technology, it was believed possible to directly and unambiguously assess the validity of the deceptively simple concept that antibody to lipopolysaccharide core glycolipid was beneficial in the treatment of gram-negative bacterial sepsis. Monoclonal antibodies directed at the lipid A region of the lipopolysaccharide were generated from human (HA-1A) and murine (E5) splenic tissue for use in large, multicenter, placebo-controlled clinical studies that used similar entry and exclusion criteria [5, 6]. Although no beneficial effect was observed with either preparation for the entire septic-appearing population, both preparations showed benefit for specific patient subgroups. Receipt of HA-1A antibody decreased the mortality from 49% in patients with gram-negative bacteremia to 30% in patients given placebo (P = 0.014) and from 57% to 33% in patients with shock at entry (P = 0.017). The E5 monoclonal antibody, however, reduced mortality only in patients who had gram-negative bacterial infection but who were not in shock (43% to 30%, P = 0.01).

These studies were extensively analyzed in the literature (well reviewed by Warren and colleagues [7]) and were critically reviewed in 1991 by a Food and Drug Administration advisory panel. In view of the questions posed as well as the possibility that the benefit for one patient subgroup may have been attained at the expense of increased mortality in another subgroup, follow-up studies were recommended for both monoclonal antibodies.

In this issue, McCloskey and colleagues [8] present their findings in a follow-up study of the HA-1A monoclonal antibody in a patient subpopulation that, in the first study, appeared to derive benefit from HA-1A: those with shock and documented gram-negative bacteremia. No difference was noted between the treatment and placebo groups in the primary end point, 14-day all-cause mortality; however, in patients without gram-negative bacteremia, increased mortality was noted among those receiving the monoclonal antibody (41% compared with 37% for placebo). This finding satisfied the safety stopping rule for the study. The investigators are to be commended for presenting data from this essentially negative study that until now was widely discussed only in the press. This study raises several important methodologic and scientific issues.

The clinical design of this study [8], described as a large, simple randomized trial, attempted to determine the safety and efficacy of treatment in the "real world". Before conducting a "confirmatory" study with HA-1A in the "real world," in the absence of more convincing data in the first study, it would have been better to unequivocally establish its efficacy in a more complete follow-up study. In order to give a "yes/no" answer in a short amount of time with as many patients as possible in the present study, the investigators collected a minimum of data. Once patients met entry criteria, only the mortality at day 14 and the occurrence of adverse events within 3 hours of infusion were recorded. Unlike the previous HA-1A study [5], there was no assessment of underlying disease severity, no determination of serum cytokine or endotoxin level or appropriateness of antimicrobial therapy, or no assessment of organ failure at entry. One hazard of such a clinical design is the inability to analyze unexpected data. Because the mortality in the placebo group was 32%, far below the projected mortality of 49%, we cannot determine whether statistically significant imbalances in the two groups may have affected the lack of treatment effect or whether the characteristics of patients in community hospitals that comprised nearly 50% of the enrollees differed from those of patients enrolled from the academic centers, who comprised most of the enrollees in the initial study.

Other difficult issues of clinical design are relevant. First, the choice of all-cause mortality in an intent-to-treat analysis, as opposed to a reversal in a relevant, endotoxin-induced physiologic variable(s), may be too severe a primary end point for determining the efficacy of a potentially useful antisepsis product. Second, because the septic response is a complex, multifactorial process, it may be unrealistic to expect that any one therapeutic intervention directed at only one stage can show a statistically significant treatment effect. Rather, it might be that combinations of antisepsis therapy are required to show a convincing clinical effect, as is the case for many anticancer drugs.

Shortly after the disappointing results of this and the second E5 [9] studies were reported in the lay press, the validity of the hypothesis of treating sepsis with anti-endotoxin antibodies was re-examined. In retrospect, it is remarkable that the development of E5 and HA-1A has reached this stage. There were few encouraging data from the many studies with polyclonal anti-lipid A antiserum to suggest that antibodies directed toward this epitope might be protective [10]. The monoclonal antibodies were hastened into long, expensive clinical trials without an adequate published scientific record or, in the absence of a clinically relevant in vitro or in vivo assay, a clearly formulated and demonstrable mechanism of action. Thus, the lack of data sufficient for licensure of these monoclonal antibodies for the treatment of sepsis should not have been unexpected. Although an obvious antiendotoxin antibody candidate does not appear to be waiting in the wings, data support the idea that antibody to the inner-core region of lipopolysaccharide merits further investigation as one component of an effective antisepsis strategy. In addition to the positive results in the original clinical trial with the J5 antiserum, several retrospective serologic surveys [11, 12] correlated patient survival during gram-negative bacterial sepsis with levels of anti-core glycolipid antibody. In addition, recent experimental studies [13, 14] found protection from lethal sepsis after treatment with antibody directed against core glycolipid epitopes. In contrast, a recently reported clinical study [15] in 100 patients using an intravenous immunoglobulin prepared from plasma of volunteers immunized with a boiled, whole J5 bacterial vaccine failed to show efficacy after a single 200 mg/kg infusion. In another study [16], treatment with post-J5 immune plasma was without effect in children with severe infectious purpura. These conflicting data must be resolved to yield a successful antiendotoxin antibody. Finally, although the increased mortality observed in patients treated with monoclonal antibody who did not have gram-negative bacteremia did not attain statistical significance, such observations were also noted in some patient subpopulations reported in the initial E5 and HA-1A studies. Might these data augur a potentially substantial adverse effect with antiendotoxin antibodies in general? Strategies directed at modulating host defenses in sepsis may be "two-edged" swords, and the indications for their use must be clearly defined through carefully conducted clinical studies with sufficient power to detect these complications. If a normal cytokine response is an important part of host defenses against some infections, then its interruption with cytokine modulators actually may accelerate the infectious process [17] as suggested by studies treating patients with sepsis using interleukin-1 receptor antagonist or using monoclonal antibody to tumor necrosis factor. Rather, modulators of cytokines may only improve outcome in those patients whose cytokine regulatory network is already dysfunctional such that its interruption, on balance, would be beneficial. Thus, the increased mortality with HA-1A treatment is fascinating. Might some aspect of the lipopolysaccharide response of humans be beneficial, such that its modulation or elimination may be similarly deleterious?

Before we continue our frontal assault on endotoxin and endotoxin-induced responses, we should ponder the words that Ivan Bennett [18] offered 30 years ago, "It has been said in jest that endotoxins will probably turn out to be the cause of most of the human diseases now classified as idiopathic and that they may also prove to be the cause of human health.these are possibilities that have yet to be excluded".