MACHINE LEARNING AND MURINE MODELS EXPLAIN FAILURES OF CLINICAL SEPSIS TRIALS

Abstract

Background:

Multiple clinical trials failed to demonstrate the efficacy of Hydrocortisone, Ascorbic Acid, and Thiamine (HAT) in sepsis. These trials were dominated by patients with pulmonary sepsis and have not accounted for differences in the inflammatory responses across varying etiologies of injury/illness. HAT has previously revealed tremendous benefits in animal peritonitis sepsis models (cecal ligation and puncture, CLP) in contradiction to the various clinical trials. The impact of HAT remains unclear in pulmonary sepsis. Our objective was to investigate the impact of HAT in pneumonia, consistent with the predominate etiology in the discordant clinical trials. We hypothesized that in a pulmonary sepsis model, HAT would act synergistically to reduce end organ dysfunction by the altering the inflammatory response, in a unique manner compared to CLP.

Methods:

Using Pseudomonas aeruginosa pneumonia, a pulmonary sepsis model (PNA) was compared directly to previously investigated intra-abdominal sepsis models. Machine-learning applied to early vital signs stratified animals into those predicted to die (pDie) vs. predicted to live (pLive). Animals were then randomized to receive antibiotics and fluids (vehicle, VEH) vs. HAT. Vitals, cytokines, vitamin-C, and markers of liver and kidney function were assessed in the blood, bronchoalveolar lavage (BAL), and organ homogenates.

Results:

PNA was induced in 119 outbred wild-type-ICR mice (predicted mortality approximately 50%) similar to CLP. In PNA, IL-1RA in 72-hour BAL was lower with HAT (2.36 ng/ml) compared to VEH (4.88 ng/ml); p=0.04. The remaining inflammatory cytokines and markers of liver/renal function showed no significant difference with HAT in PNA. PNA vitamin-C levels were 0.62 mg/dL (pDie HAT), lower than vitamin-C levels after CLP (1.195 mg/dL). Unlike CLP, PNA mice did not develop acute kidney injury (BUN pDie 33.5 mg/dL vs. pLive 27.6 mg/dL; p=0.17). Furthermore following PNA, HAT did not significantly reduce microscopic renal oxidative-stress (mean gray area of pDie 16.64 vs. pLive 6.88; p=0.93). Unlike CLP where HAT demonstrated a survival benefit, HAT had no impact on survival in PNA.

Conclusion:

HAT therapy has minimal benefits in pneumonia. The inflammatory response induced by pulmonary sepsis is unique compared to the response during intra-abdominal sepsis. Consequently, different etiologies of sepsis respond differently to HAT therapy.

INTRODUCTION

It has been estimated that over 31 million patients are diagnosed each year with sepsis, nearly 20 million of these are estimated to be characterized as severe sepsis.(1) Sepsis carries an estimated mortality of 26.7%, which increases to nearly 50% in patients requiring ICU level of care.(2) Despite advances elsewhere in the management of critically ill patients in the past decade, there has been no significant decrease in mortality in septic patients.(3, 4) We speculate that the broad definition of sepsis and failure to understand differences in the inflammatory responses account for the lack of progress in sepsis care.

In 2016, investigators found that the use of three relatively benign medications significantly improved outcomes. In particular these medications lowered the incidence of acute kidney injury for patients with sepsis.(5) Since this index study, multiple clinical trials have failed to demonstrate the efficacy of hydrocortisone, ascorbic acid, and thiamine (HAT) in sepsis.(6–10) However, these trials have not accounted for potential differences in the inflammatory responses across patients as well as the varying etiologies of injury/illness. The predominate etiology of sepsis in the completed clinical trials was pneumonia. The goal of this study was to investigate the role of HAT as an adjunctive therapy using an animal sepsis model that resembles the vast majority of cases represented in the multiple clinical trials, pulmonary sepsis. (6–10).

Animal peritonitis sepsis models investigating the role of HAT have revealed tremendous benefits which stand in contradiction to the various clinical trials. Unlike the clinical trials, (6–10), animal trials offer the ability to fine tune the model, to assess the response to therapy for a given etiology and clinical severity. (11) The peritonitis animal models stratified septic animals by degree of illness and risk of death prior to randomization into treatment regimens to further identify a discrete population that may benefit from HAT. In animals with peritonitis, HAT led to a significant survival benefit, improved cardiovascular status, and reduced oxidative stress in the sickest animals predicted to die as compared to antibiotics and fluids alone.(12) More recent studies from our lab have demonstrated significantly downregulated inflammatory markers (IL-6, IL-1RA, KC, and MIP2), reduced acute kidney injury, and decreased oxidative stress following treatment with HAT in the setting of intra-abdominal sepsis. Additionally, mice with intra-abdominal sepsis that were treated with HAT had significantly higher vitamin-C levels compared to VEH. (Publication pending). While the previously described findings from murine models of intra-abdominal sepsis are discordant to the various human trials, they do not represent the study populations as the murine model focused on intra-abdominal sepsis which was an uncommon etiology of disease in the human trials. As such, we propose the use of a murine model of pulmonary sepsis to more closely represent the most common cause of sepsis.

It is naive to assume that patients with sepsis due to a pulmonary source would have the same inflammatory response as patients with peritonitis. As a consequence of a different inflammatory cascade, the impact of sepsis on end organ injury varies with etiology. As such, different inflammatory processes may respond to different treatment regimens. Our objective in this study was to investigate the impact of HAT in pneumonia, consistent with the predominant etiology in the discordant clinical trials, while stratifying animals for degree of illness and risk of death prior to randomization. We hypothesized that in a pulmonary sepsis model, hydrocortisone, ascorbic acid, and thiamine would act synergistically to reduce end organ dysfunction by the altering the inflammatory response in a unique manner compared to the response observed in peritonitis sepsis models.

METHODS

Murine Models and Chemicals

Adult (22–25 gram) outbred wildtype ICR mice (Envigo, Indianapolis, IN) of both sexes, (to increase rigor and general applicability), were used. All experiments were performed in accordance with the ARRIVE guidelines and approved by the Institutional Animal Care and Use Committee (IACUC); (protocol # 14857).

Pulmonary sepsis was induced with Pseudomonas aeruginosa via hypopharyngeal instillation; (PNA). (13–15) Pseudomonas aeruginosa (Strain: Boston 41501) was streaked from a frozen stock onto a blood agar plate for incubation at 37-degrees Celsius overnight. Several identical appearing colonies were subsequently isolated and added to incubate in Todd-Hewitt Broth until an optical density was obtained that correlates with approximately 2.0–5.0 ×10⁹ colony forming units per ml (CFU/ml), previously determined to achieve approximately 50% mortality in three days. The final concentration was verified after streaking serial dilutions on a blood agar plate.

Machine Learning

The introduction of stratification by severity of disease in the randomization scheme provides improved granularity on which population (if any) may benefit from HAT as an adjunctive therapy.(16) As such a validated machine learning algorithm was developed using non-invasive vital signs collected early after infection/injury to risk stratify animals prior to randomization into therapy. (Publication under review). (Supplemental 1). Septic animals were divided into two populations using just 6- and 24-hour vital signs with Lasso regression for variable selection (17) and 10-fold cross validation to define the optimal shrinkage parameter. High-risk sepsis responders (predicted to Die, pDie) were defined as animals expected to die within 72-hours after infection/injury. Low-risk sepsis responders (pLive) were defined as animals expected to live beyond 72-hours after infection and as previously published by our lab.(12, 16, 18–22) Stratified mice (pDie vs. pLive) were then randomized into HAT or vehicle (VEH) treatment groups.

Treatment and Intervention

All animals in the pneumonia model were treated in an identical manner to the peritonitis model which used cecal ligation and puncture (CLP), with the exception of buprenorphine, which was not administered to PNA animals. Starting 2-hours after PNA, animals were treated with broad spectrum antibiotics (imipenem, 25 mg/kg; Merck, West Point, PA) and fluid resuscitation (5% Dextrose-Lactated Ringers Solution). Fluid and antibiotics were administered every 12-hours.

Hydrocortisone, ascorbic acid, and thiamine (HAT) were obtained from Sigma-Aldrich (St. Louis, MO). The treatment regimen was based on the original study in humans.(5) A combination of hydrocortisone (1.5 mg/kg), vitamin-C (45 mg/kg), and thiamine (3 mg/kg) were administered in a 5% Dextrose-Lactated Ringers solution (total volume of 0.5ml for the first dose and 0.25ml for all subsequent doses) every 12 hours with imipenem (25 mg/kg) for HAT treated animals. Control animals were treated with vehicle (VEH), identical volumes of crystalloid plus imipenem every 12 hours.

Non-Invasive Physiologic Parameters

Non-invasive physiologic parameters (vitals) including: heart rate (HR), respiratory rate (RR), O₂ saturation (SPO2), and pulse distention (PD, a measure of local blood flow at the neck) were collected using a cervical collar (MouseOx Plus, Starr Life Science, Oakmont, PA). In addition, weight and temperature (Model Temp1, TempIR) were collected. Vitals were recorded at baseline prior to infection/injury then again at 2-hours, 6-hours, and every 12-hours thereafter following infection.

Assays of Organ Dysfunction and Inflammatory Markers

Various cytokines, vitamin-C, and markers of liver and kidney function (aspartate aminotransferase (AST), bilirubin, and blood urea nitrogen (BUN)) were assessed in the blood, bronchoalveolar lavage (BAL), and organ homogenates (kidney, liver, lung, and spleen). Cytokines of interest, interleukin-6 (IL-6), interleukin-1-receptor antagonist (IL-1RA), Macrophage inflammatory protein-2 (MIP2), and Keratinocyte derived Chemokine (KC) were assessed by enzyme-linked immunosorbent assay (ELISA) as previously reported.(12, 16, 19, 20) These four inflammatory markers are commonly assessed in animal models of sepsis as they behave similarly in humans, and provide a broad perspective of the inflammatory cascade at a given point of time. The degree of oxidative stress was assessed in histology slides of kidneys that were stained with Hypoxyprobe following the manufacturer’s instructions and subsequently quantified using Image-J software, (National Institutes of Health, Bethesda, MD).

Statistics

Statistical analysis was performed using commercially available software, Prism Version 9.1.2 (Graphpad, San Diego, CA). Differences in assays of organ dysfunction, inflammatory markers, vitamin C, and oxidative stress were determined using Kolmogorov-Smirnov test. Log-Rank survival analysis was used to assess differences in survival. A p-value of less than 0.05 was used to determine significance. The machine learning algorithm, as described above, was developed using R-Studio (Boston, MA).

Sample size analysis using pilot data revealed that at least 85-animals were needed to achieve a power of 80% to identify a 15% reduction in mortality (baseline mortality 50%) for subjects treated with HAT compared to VEH. Mortality data was used as a surrogate to estimate sample size as we hypothesized the rate of end organ injury would vary between models but presumed that mortality would correlate with the rate of end organ injury observed.

RESULTS

Stratification and Mortality of Pulmonary Sepsis Model

Pulmonary sepsis was induced in 119 outbred wild-type ICR mice (Envigo, Indianapolis, IN). Using a validated machine learning algorithm (Lasso regression with 10-fold cross validation), mice were stratified into pDie or pLive after obtaining 6- and 24-hour vitals. Animals were consequently randomized into HAT or VEH treatment regimens as outlined in Figure 1. The mean predicted mortality of all PNA mice was 62% with the mean probability of death for pDie PNA animals of 75% and pLive PNA animals of 35%. After stratification, the population was randomized into well balanced groups with equivalent probabilities of death; 62% for both treatment groups. (Supplemental 2).

Figure 1: Stratification and Randomization Methodology.

pLive animals are defined as having a probability of death <50%. pDie animals are defined as having a probability of death >50%. HAT = hydrocortisone, ascorbic acid, and thiamine. Vehicle = imipenem and crystalloid fluid resuscitation only. All animals received imipenem and weight based crystalloid fluid resuscitation.

Inflammatory Response

Our first goal was to investigate if the inflammatory and organ injury response in PNA is fundamentally different as compared to that induced by CLP, even when animals have an equivalent chance of death. To explore if HAT altered the local inflammatory response in a pneumonia sepsis model, the BAL and lung homogenate were studied. Analysis of the BAL revealed that the addition of HAT had no impact on BAL levels of IL-1RA compared to VEH for both pLive and pDie mice. However, there were significantly lower levels of IL-1RA in 72-hour BAL between all mice that received HAT compared to VEH (2.36 ng/ml vs. 4.88 ng/ml; p=0.04). (Figure 2). Additionally, there was an associated increase of IL-1RA in lung homogenate of pDie mice compared to pLive at 72-hours, however it did not reach statistical significance; (3.9 ng/ml/g-protein vs. 1.48 ng/ml/g-protein; p=0.06). The remaining inflammatory markers of liver and renal function showed no significant difference in plasma, BAL, or other organ homogenates.

Figure 2: IL-1RA Levels in Bronchoalveolar Lavage Over Time After Pneumonia.

Interleukin 1 receptor antagonist (IL-1RA) was assessed in pulmonary sepsis model of animals treated with hydrocortisone, ascorbic acid and thiamine (HAT) or vehicle; fluid resuscitation and antibiotics alone (VEH). A. p-Die vs. p-Live comparison 48-hours after infection. B. p-Die vs. p-Live comparison 72-hours after infection; (HAT 2.36 ng/ml vs. VEH 4.88 ng/ml; p=0.04). C. HAT vs. VEH comparison 48-hours after infection. D. HAT vs. VEH comparison 72-hours after infection.

Unlike in peritonitis, there was no significant change in plasma cytokines with HAT at 24-hours or 48-hours after pulmonary infection with gram-negative pneumonia. (Supplemental 3). The lack of difference in PNA animals persisted when comparing HAT vs. VEH across all subjects, pDie subjects, and pLive subjects. At 72-hours after infection with pulmonary sepsis, the plasma cytokine levels of surviving animals descended below detectable levels.

Prepared slides of bronchoalveolar lavage collected in a subset of animals sacrificed at 48 hours revealed no difference in inflammatory cells (% neutrophils, % macrophages, or % lymphocytes) between HAT vs. VEH and pDie vs. pLive. (Supplemental 4).

Vitamin-C Levels in Pulmonary Sepsis

Given the underwhelming response of HAT on the inflammatory response in PNA, vitamin-C levels were assessed for differences between the two treatment groups. The mean vitamin-C level in the plasma of naïve animals was 0.59 mg/dL. In pulmonary sepsis, vitamin-C levels are highest in pDie animals 24-hours after infection (HAT pDie= 0.62 mg/dL, VEH pDie=0.60) (Figure 3/Supplemental 5), but remain lower than the vitamin-C levels after injury by CLP (CLP pDie HAT=1.195 mg/dL). (Publication pending). Of PNA animals that were alive 72-hours after infection, the pLive mice treated with HAT had significantly higher plasma vitamin-C levels compared to their VEH treated counterparts (0.30 mg/dL vs. 0.00; p=0.04). (Figure 3).

Figure 3: Vitamin-C Levels in Plasma Over Time.

Vitamin-C levels (mg/dL) assessed in the plasma of mice with pneumonia (PNA) at 24-, 48-, and 72-hours. The only significant difference was significantly higher plasma Vitamin-C levels between mice treated with hydrocortisone, ascorbic acid, and thiamine (HAT) vs. vehicle (VEH) at 72 hours (0.30 mg/dL vs. 0.00; p=0.04).

Assessment of End Organ Injury

It has been previously demonstrated that in the setting of intra-abdominal sepsis, pDie animals have significantly greater severity of acute kidney injury than pLive mice with BUN elevations nearly two-times the upper limit of normal at approximately 50mg/dL; (normal BUN range previously determined to be 12.0–26.7 mg/dL for ICR mice). (16) In the pneumonia model of sepsis, that produces similar mortality to CLP, there was no evidence of acute kidney injury. The BUN of all pDie mice peaked 24-hours after PNA at an average of 33.5 mg/dL compared 27.6 mg/dL for all pLive mice at the same time after PNA; p=0.17. (Figure 4).

Figure 4: No Kidney Injury in Pulmonary Sepsis Model as Measured by Blood Urea Nitrogen Levels in 24-hour Plasma.

BUN = Blood Urea Nitrogen. Assessed in plasma at 24- , 48- , and 72-hours after infection with Pseudomonas aeruginosa pneumonia. At 24-hours, the mean BUN of: HAT treated pDie animals was 35.92 mg/dL; VEH treated pDie animals was 31.14 mg/dL; HAT treated pLive animals was 28.24 mg/dL; VEH treated pLive animals was 26.41 mg/dL, and naïve mice was 28.36 ng/ml.

Both peritonitis and pneumonia-based sepsis resulted in microscopic evidence of oxidative stress using hypoxyprobe staining and histological examination of kidneys, again, to a greater degree in the setting of intra-abdominal sepsis. (16) The lack of AKI as measured by plasma BUN (Figure 4) was mirrored by a lack of oxidative stress in the kidney. Unlike peritonitis, HAT was not beneficial at significantly reducing visible oxidative stress for pDie mice after PNA (mean gray area of 16.64 vs. 6.88; p=0.93). (Figure 5).

Figure 5: Hypoxyprobe Staining of Kidneys from Animals with Pulmonary Sepsis.

A. Kidney from HAT treated pDie mouse (80.3% probability of death) harvested 48-hours after infection with gram-negative pneumonia; mean gray area=122.17. B. Kidney from VEH treated pDie mouse (72.7% probability of death) harvested 48-hours after infection with gram-negative pneumonia; mean gray area=53.30. C. Kidney from VEH treated pLive mouse (48.7% probability of death) harvested 48-hours after infection with gram-negative pneumonia; mean gray area=2.79. D. Kidney from VEH treated pLive mouse (23.2% probability of death) harvested 48-hours after infection with gram-negative pneumonia; mean gray area=4.14.

In addition to the outlined differences in the inflammatory response between CLP and PNA models as well as the impact of HAT on each etiology of sepsis, we found there were also differences in clinical parameters. Unlike CLP where HAT demonstrated a tremendous survival benefit for pDie animals (12, 19), HAT had no impact on percent survival of pDie animals with pulmonary sepsis, nor did it improve survival for either pLive animals or the entire population. (Figure 6).

Figure 6: Survival of Sepsis Due to Peritonitis and Pneumonia.

CLP = Cecal ligation and puncture; peritonitis induced sepsis using cecal ligation and puncture, n=91. (CLP mortality data previously published, (19)). PNA = Pseudomonas aeruginosa pneumonia induced sepsis, n=119.

Altered Physiology

It has previously been documented that CLP results in notable differences in 6- and 24-hour heart rate (HR) and pulse distension (PD) as well as 6-hr respiratory rate (RR) between pDie and pLive mice after injury.(12) Similarly, following infection with gram negative pneumonia, there were significant differences between the pDie and pLive in the 24-hour HR (354 vs. 462 beats/min; p<0.0001), 24-hour RR (114 vs. 146 breaths/min; p=0.0037), 24-hour SPO2 (88.4% vs. 92.8%; p=0.0078), 36-hour SPO2 (82.9% vs. 91.26%; p=0.0033), and 24-hour temperature (31.9 vs. 35.2 degrees Celsius; p=0.0002). Treatment with HAT resolved significant differences appreciated between all pDie and pLive mice in the physiologic parameters listed above (24-hour HR, 24-hr RR, 24-hr SPO2, 36-hour SPO2, and 24-hour temperature). (Supplemental 6).

DISCUSSION

In a murine model of pulmonary sepsis, we failed to uncover any significant benefit of hydrocortisone, ascorbic acid, and thiamine, contrary to the initial HAT study,(5) yet consistent with the discordance demonstrated in recent clinical trials.(6–10) This study highlights three key factors: The first is that pulmonary sepsis does not invoke the same inflammatory response and associated end organ damage as intra-abdominal sepsis; second, the supplementation of HAT in addition to fluid resuscitation and broad-spectrum antibiotics does not result in the same improvements in pulmonary sepsis as it does in peritonitis; and third, HAT has the greatest benefits in the sickest animals with intra-abdominal septic shock and shows minimal benefits in those with more mild illness. This investigation more closely modeled the failed clinical trials that enrolled predominately pulmonary sepsis patients however went further, to investigate discrete populations that may benefit from the adjunctive use of HAT by stratifying by both etiology and degree of illness.

One of the greatest weaknesses of the previously performed clinical trials was the failure to introduce stratification into the randomization methodology. As a consequence, a sweeping dismissal of the proposed therapy has been implemented for all septic patients. But as outlined, all sepsis is not the same and differs by etiology and severity. Previous investigators,(5–10), have failed to account for variations in the inflammatory mechanisms between different causes of sepsis as well as possible variations in the innate immune responses between patients, i.e. sepsis phenotypes.(23) We introduced risk of death/illness severity stratification into the randomization scheme which allowed direct comparisons between the most sick (pDie) with the least sick (pLive). Additionally, two separate studies were performed focusing on unique, but clinically relevant etiologies of sepsis.

In the setting of intra-abdominal sepsis, HAT resulted in a significant decrease in inflammation, reduction of acute kidney injury, improved physiology, and consequently improved survival for the sickest septic animals. (12) There were stark differences in the inflammatory response in pulmonary sepsis and the impact of the HAT treatment regimen in the pneumonia model.

Two of the most impressive impacts of HAT in the peritonitis model of sepsis, are a significant reduction in oxidative stress and reduction of acute kidney injury for the sickest animals.(12) While microscopic evidence of oxidative stress was apparent in many of the histology samples of kidneys harvested from mice after pneumonia, on average it was to a lesser degree than CLP, and there was no difference between degree of sepsis despite treatment with HAT. Concordantly, there was no significant elevation of BUN to suggest evidence of acute kidney injury in the pneumonia model and consequently no alteration of BUN by adding HAT. These findings are consistent with a large retrospective cohort study of 4532 patients that found that patients with intra-abdominal sepsis and genitourinary sepsis had significantly greater risk of developing acute kidney injury compared to pulmonary infections.(24)

One of the few significant differences noted between animals with different severity of illness from pneumonia was a lower level of IL-1RA in the BAL of pDie after the use of HAT. IL-1RA, released promptly following infection from circulating monocytes, macrophages, and neutrophils, and a delayed from the liver, competitively binds the pro-inflammatory interleukin-1 (IL-1) receptor to prevent an unregulated pro-inflammatory cascade.(25, 26) We suspect that the observed decrease in IL-1RA in BAL following HAT implies a concordant decrease in the pro-inflammatory IL-1 indicating an overall down turn in the IL-1 associated pro-inflammatory process occurring in the lung during pneumonia. However, the downregulation of one pro-inflammatory mediator does not imply the downregulation of all inflammatory mediators as demonstrated by the lack of difference found in the other assessed cytokines. This finding is present not only in sepsis, but across other inflammatory processes.(27) From this observation, we suspect that HAT may act uniquely on the macrophages, neutrophils, and monocytes that are recruited to the lung at the early stages following infection and possibly even alter the livers production of IL-1RA at later stages given the that the observed differences occurred 3-days after infection.

In both models of sepsis, (CLP and PNA), animals randomized to receive HAT were dosed in an identical fashion. As previously demonstrated in the peritonitis model, vitamin-C levels were several folds higher for HAT treated animals in the plasma at 24-hours after injury by CLP compared to healthy naive mice. (Publication pending). Following pulmonary sepsis, vitamin-C levels trended higher for animals treated with HAT but never reached significance compared to those who received only antibiotics and fluid resuscitation. However, the peak plasma vitamin-C levels in PNA animals were nearly one-half of that measured in animals with intra-abdominal sepsis. Furthermore, the vitamin-C levels dropped below those measured in naive mice as time progressed after introducing gram-negative bacteria to the lungs, indicating possible consumption of the vitamin-C during the scavenging of free-radicals. This raises the possibility that the weight-based dose of HAT, that promotes improvement in peritonitis may not be sufficient in pulmonary sepsis. This may indicate that a larger dose of at least vitamin-C is required for pulmonary sepsis. This would be contrary to what is implied by lower degree of systemic inflammation and end organ injury observed, but this highlights the complexity of the inflammatory mechanisms at play.

One of the limitations of the study, is that this is a murine model of sepsis. We acknowledge that pre-clinical murine models are several steps away when transitioning from bench to bedside, however they continue to play important roles, particularly in the study of sepsis. There is a strong history of murine models not only foreshadowing the inflammatory response in humans during sepsis, but also in predicting outcomes of proposed novel therapeutics. A paper by Osuchowski et al. highlighted a total of 26 examples, specific to sepsis, where findings in murine models have contributed and correlated to respective human studies.(28) With rigorous and reproducible animal models, and an improved randomization scheme, our series of studies allowed us to expand the scope of the investigation beyond just the impact of HAT on clinical outcomes, and to thoroughly examine the pathobiology while accounting for the etiology of sepsis and degree of illness, which were not accounted for in the clinical trials. (5–10) Additionally, while we argue that not all types of sepsis are the same, we acknowledge that there are numerous infectious pathogens that may induce sepsis throughout the body, all potentially resulting in different inflammatory responses. This investigation focused on one pathogen, P. aeruginosa, highly prevalent in the clinical realm, so that we could demonstrate reproducibility of the presented data across multiple repeated experiments in nearly 120 mice. We suggest future studies consider investigation other infectious pathogens. An additional limitation of this study was the absence of a hydrocortisone only control group. While there is theoretical synergy between the three investigated compounds that drives the observed changes, future studies should include a hydrocortisone only group. A hydrocortisone only group would allow future investigators to determine if the observed benefits are due to a synergistic effect between hydrocortisone, ascorbic acid and thiamine or if they are driven by the glucocorticoid alone.

There are distinct differences in the inflammatory responses that result from intra-abdominal sepsis and gram-negative pneumonia models. There are clear benefits of HAT in the sickest animals with intra-abdominal sepsis. These are not similarly demonstrated in the pulmonary sepsis model. It is important to recognize that the predominate etiology of sepsis in recent investigations involving HAT was pneumonia and that clinical trials failed to stratify enrolled subjects by degree of illness which we show is important at discerning for possible benefits of HAT.(5–10) The results of this study provide an explanation for the failure of recent clinical sepsis trials investigating HAT. (6–10) A larger trial incorporating stratification by degree of illness and etiology of sepsis into their randomization methodology is warranted.

CONCLUSION

All sepsis is not the same, consequently, not all cases of sepsis may benefit from the same adjunctive treatments such as hydrocortisone, ascorbic acid, and thiamine. Key factors including the etiology of sepsis and degree of illness have been prematurely neglected in recent clinical trials. But as demonstrated, these factors result in different systemic inflammatory responses that respond differently to HAT. Further investigation is necessary to elucidate the correct patient population that may benefit from HAT before omitting it as an adjunctive therapy.

Supplementary Material

Supplemental Data File (.doc, .tif, pdf, etc.)_2

Supplemental 1: Lasso Regression and 10-Fold Cross Validation Machine Learning Algorithm. Attached .rmd file (R-Studio) with previously developed algorithm utilizing Lasso Regression for variable selection and 10-fold cross validation to optimize shrinkage parameters.

Supplemental Data File 2(.doc, .tif, pdf, etc.)

Supplemental 2: Distribution of Probability of Death as Determined at 24-hours with Machine Learning The scatter plot illustrates the output of the lasso regression/cross-validation model. Resulting output values from the developed model reflected probability of death and ranged from 0: unlikely to die, to 1.0: highly likely to die. The time to death of interest was defined as within 72-hours from infection with PNA. Animals with a high probability of death, (pDie), were defined by model output of greater than or equal to 0.50. While animals with a low probability of death, (pLive), were defined by model output values of less than 0.50. The table below describes the range of severity of illness for the pulmonary sepsis model and respective treatment groups.

Supplemental Data File 3(.doc, .tif, pdf, etc.)

Supplemental 3: Plasma Cytokines 24-hours after Gram-Negative Pneumonia IL-6 = Interleukin 6. IL-1Ra = Interleukin 1 Receptor antagonist. KC = Keratinocyte derived Chemokine. MIP2 = Macrophage Inflammatory Protein 2. There is no difference between pDie vs. pLive for those treated with HAT as well as VEH.

Supplemental Data File 4(.doc, .tif, pdf, etc.)

Supplemental 4: Cellular Differential of Bronchoalveolar Lavage at 48-hours after infection with P. aeruginosa.

Supplemental Data File 5(.doc, .tif, pdf, etc.)

Supplemental 5: Plasma Vitamin-C Levels in Pulmonary Sepsis *p=0.04 for Kolmogorov-Smirnov test for 72-hour pLive mice HAT vs. VEH. HAT=Hydrocortisone, ascorbic acid, and thiamine. VEH=vehicle (imipenem and fluid resuscitation only).

Supplemental Data File 6(.doc, .tif, pdf, etc.)

Supplemental 6: Significant Vital Signs in Pulmonary Sepsis Model Heart Rate (HR) is reported as beats per minute (BPM). Respiratory rate (RR) is reported as breaths per minute. Oxygen saturation (SPO2) is reported as a percent (%). Temperature is reported as degrees Celsius.