Early Diagnosis of Sepsis: Is an Integrated Omics Approach the Way Forward?

Abstract

Sepsis remains one of the leading causes of death in the United States and it is expected to get worse as the population ages. Moreover, the standard of care, which recommends aggressive treatment with appropriate antibiotics, has led to an increase in multiple drug resistant organisms (MDROs). There is a dire need for the development of new antibiotics, improved antibiotic stewardship and therapies that treat the host response. Development of new sepsis therapeutics has been a disappointment as no drugs are currently approved to treat the various complications from sepsis. Much of the failure has been blamed on animal models that do not accurately reflect the course of the disease. However, recent improvements in metabolomic, transcriptomic, genomic and proteomic platforms have allowed for a broad spectrum look at molecular changes in the host response using clinical samples. Integration of these multi-omic data sets allows researchers to perform systems biology approaches to identify novel pathophysiology of the disease. In this review, we will highlight what is currently known about sepsis and how integrative omics has identified new diagnostic and predictive models of sepsis as well as novel mechanisms. These changes may improve patient care as well as guide future preclinical analysis of sepsis.

1. Introduction

Sepsis is now defined as a patient presenting with or developing a life-threatening organ dysfunction caused by a dysregulated host response to infection [1]. Globally, 20 million people will be treated in hospitals for sepsis and as many as five million of those patients will die each year [2]. It is a complex, heterogeneous syndrome with high mortality. Outcomes are influenced by the site of infection, causative organisms, acute organ dysfunctions and co-morbidities [3, 4].

Precision medicine is founded on the concept that therapies can be customized based on the unique phenotypic and molecular features of each patient, and relies on enrichment strategies that can that can disentangle the patient heterogeneity and determine which interventions will be most effective [5]. However, these strategies require clear understanding of the biological mechanisms that contribute to the disease process. Recent studies integrating large molecular datasets from clinical studies has led to improved understanding of the mechanisms that lead to poor outcomes due to sepsis. In this review, we will highlight the recent clinical and research findings and how these new signatures may improve patient diagnosis, outcome prediction and therapy in critically ill patients.

2. Improved Patient Management versus 30+ Years of Therapeutic Failures

While it is common for critical care doctors to quickly triage both high-risk and low-risk cases at presentation, it is often much more difficult to manage moderate risk patients. An often-described scenario is two patients with nearly the same symptoms at presentation can have dramatically different outcomes. One patient responds to antibiotics and fluid resuscitation and is sent home with no complications. The other patient, despite similar therapeutic strategies, may suddenly develop critical illness.

Surviving septic shock can also have long-term consequences. Most sepsis clinical trials have used 28–30d outcomes as a primary endpoint. However, mortality remains high up to 2-years after presentation and quality-of-life (QOL) is significantly reduced in survivors [6, 7]. This reduction in QOL is worse in severe sepsis and septic shock survivors than other admissions to the ICU [8] and has been related to muscle wasting – 25% of skeletal muscle loss has been reported within a one-week stay in the ICU [9]. In many cases, the loss of muscle tissue never fully recovers. The survivors are reported to have permanent mitochondrial loss in muscle tissue [10]. It was hypothesized that early mobilization in the ICU would reduce length-of-stay, improve long-term outcomes and QOL [11]. However, implementation of physical therapy in ICU patients that require intubation did not significantly improve physical functioning or QOL [12].

There are currently no biomarkers that can predict outcomes with high accuracy. However, increased blood lactate concentration has been useful in triaging patients. Hypotension or blood lactates >4mmol/l often trigger early aggressive protocols which involve appropriate broad spectrum, antimicrobial administration, optimization of cardiac function and oxygen delivery within the early hours after a patient presents with septic shock [13]. However, these definitions lacked sensitivity in moderate risk patients. Mortality in patients with moderate lactates (2.0 to 3.9 mmol/l) have been reported as high as 16.4% [14] and delayed therapy in patients that do not initially meet full criteria for early therapies accounts for 56% of in-hospital sepsis deaths [15]. In the hope of identifying more moderate risk patients that would benefit from aggressive treatment, septic shock was recently redefined as a patient requiring vasopressors, or a mean arterial pressure (MAP) <65 mmHG or lactate >2.0 mmol/l after adequate fluid resuscitation [16, 1]. The primary weakness of lactate is that it does not specifically relate to specific cellular dysfunction [16]. However, the authors for the new definitions of shock argue that lactate remains useful since there are no superior and readily available alternatives.

While the new definitions may identify more at risk patients, it comes with a potential weakness – more patients aggressively treated for sepsis may lead to an increase in treatment with unneeded antimicrobials. This is a major concern as it is estimated that ~30% of antibiotic prescriptions given to the general population were likely unnecessary [17]. Whereas in critical care, prophylactic usage of antimicrobials to prevent surgical site infections is common, potentially inappropriate and linked to excessive duration [18]. The National Action Plan for Combating Antibiotic-Resistant Bacteria aims to reduce unnecessary antibiotic usage by 50% by 2020 [19].

Recent studies have also questioned the effectiveness of early antibiotic administration in patients that do not have confirmed bacterial infections [20]. Adverse neonatal outcomes are increased in very-low-birthweight infants treated with antibiotics that did not have culture-proven sepsis [21]. While in adults, all-cause mortality, appropriate therapy and shorter mean duration of therapy was improved in surgical ICU patients that followed conservative antimicrobial treatment as compared to aggressive antimicrobial administration [22, 23].

Clearly new antibiotics are not going to solve the problem of antimicrobial resistance unless appropriate stewardship is also implemented. Biomarkers that can more accurately diagnosis sepsis, as well as therapeutics that improve the host’s response to fight infections and prevent multi-organ dysfunction are needed to make the action plan most effective.

Traditionally, new pharmacological therapies are developed in preclinical animal models. Unfortunately, most animal models utilized do not appropriately mimic sepsis in humans [24]. The current “Gold” standard animal model is the cecal-ligation and puncture (CLP) method that leads to polymicrobial sepsis (reviewed in Kingsley et al., [25]). The strength of the model is that it mimics many of the common features of polymicrobial sepsis found in humans (e.g., cytokine profile and acute phase reactant production, necrosis and gut bacterial infection related to perforation of the colon). However, the model has also been criticized as it rarely takes into consideration standard care such as antibiotic treatment, fluid resuscitation, vasopressors, or mechanical ventilation. These models also often fail to consider age and comorbidities. It is well established that the elderly are at greater risk for poor outcomes. Cardiac disease, respiratory disease, liver disease and renal failure all have increased risks for mortality.

Because of the described weakness in the animal models, therapies that are successful in the CLP animal model have poorly translated in clinical trials [24]. One criticism of the animal model is that it leads to a “cytokine storm” and a hyper-inflammatory response [26]. However, attenuation of the host inflammatory response has been relatively ineffective in clinical trials [27]. One hypothesis for this failure is that a hypo-inflammatory response in the later stages of the disease is correlated with poor outcomes [28]. Recent effort has been made to tailor treatments that target both pro- and anti-inflammatory pharmacological therapies based upon the phase of the disease [29].

Despite these weaknesses, animal models remain the most effective means for preclinical evaluation of potential therapeutics. Moreover, while the animals do not exactly replicate the human disease they still provide significant insight into how the immune system responds to an infection. Therefore, a better understanding of the pathophysiology of the disease may allow for the selection of clinically precise endpoints that can be monitored in these animal models. For example, pro-immunomodulatory drugs are being developed that improve the host’s innate and adaptive immune response [30–32]. Testing these drugs in older animals that are immunosuppressed may provide more relevant information as to how the drugs will perform clinically.

Clearly, new pharmacologic and therapeutic strategies are needed to improve acute and long-term outcomes as well as improve QOL in survivors. Moreover, biomarkers that inform patient management will allow for individualized therapies. A better understanding of the pathophysiology of the disease will help in the selection of these biomarkers as well as identify new targets for therapy.

3. System Biology Analysis of Clinical Samples Provides Unique Insight to Molecular Changes Due to Sepsis.

To identify new treatment strategies it is important to understand the host response in large, heterogeneous cohorts throughout the time course of the disease [33]. Systems biology analysis can offer tremendous insight about the pathophysiology of the disease in well designed, and powered retrospective and prospective observational studies [34–41]. The results can identify clinical endpoints that may improve outcomes as well as new interventions that can be appropriately tested in preclinical models. Systems biology relies on identifying molecular differences due to sepsis from broad spectrum, unbiased analysis of transcriptomic, metabolomic, proteomic and genomic changes. While observational, these pathways and biomarker changes can be utilized to predict outcome, diagnose infection as well as identify hypothetical mechanisms that may point to novel pathophysiology. Currently there are four major omic analyses that have been performed in patient blood (serum or plasma) and white blood cell (WBC) samples.

Transcriptomic changes, or gene expression analysis, can be performed using microarrays or digital gene expression analysis (also known as RNAseq analysis [42]). Microarray analysis provides semi-quantitative results which utilizes anti-sense hybridization assays of isolated messenger RNAs (mRNAs) to determine gene expression changes. RNAseq is a “shot-gun” sequencing method that sequences isolated fragments of mRNAs [32]. The sequenced fragments are aligned to a reference genome and counted to determine the change in expression. The results are highly quantitative, has excellent sensitivity and nearly unlimited dynamic range. RNAseq analysis also has the added benefit of identifying gene fusions, splice variants and differential expression of known disease causing variants [32]. The method can also be utilized to determine expression of microRNAs (mirRNA), non-coding RNAs (ncRNA) and metatranscriptomic analysis (i.e., viral, bacterial or parasitic RNAs). However, RNAseq is more expensive and the associated bioinformatics can be difficult.

Genomic analysis is performed similarly to transcriptomic analysis, except that changes in the DNA sequence are identified to determine single nucleotide polymorphisms (SNPs) that may be predictive of genetic risks for disease progression [43]. Hybridization microarray assays identify common variants throughout the genome for genome-wide association studies (GWAS) [44, 45]. Because the variants are common, the studies often fail to identify the causal variants but rather can focus research to an area of the genome that may harbor the disease-causing variant. DNAseq is more specific and can identify millions of variants throughout the genome [46, 47]. Whole genome sequencing examines nearly every single base in the human genome with high accuracy. To reduce costs and data analysis, exome sequencing (i.e., genomic sequencing of the exonic coding regions of the genome) is often performed. This method accounts for <2% of the total genome and can identify causal variants that lead to non-functional proteins or nonsynonymous amino acid substitutions that may affect protein function [43, 48]. Despite the high cost and bioinformatics infrastructure required to analyze these datasets [49], the results are very precise and offer the potential to identify causal genetic factors that increase patients risks for poor outcomes.

Metabolomic changes involve small molecular biochemicals that are the substrates and products of metabolism. These changes can be determined by targeted quantitative analysis of specific molecules using nuclear magnetic resonance (NMR) spectroscopy [50], ultrahigh-performance liquid chromatography mass spectrometry (UHPLC MS), matrix-assisted laser desorption ionization time-of-flight (MALDI TOF) MS, broad spectrum semi-quantitative analysis of UHPLC MS and that can identify a broad spectrum of molecular changes without a priori knowledge (reviewed in Johnson et al., [51]).

Proteomic changes can be identified by multiplexed immunoassays, 2D-gel electrophoresis, aptamer analysis, MALDI or by UHPLC MS [52–55]. While the most biologically relevant of the single omic analyses, as it can identify the specific proteins (enzymes) that may be responsible for disease, the studies are limited by non-specific binding, low specificity and accuracy, and limited dynamic range. Further, aptamers and immunoassays need a priori knowledge of specific proteins of interest and therefore lacks broad-spectrum identification of novel proteomic changes that may not previously have been implicated in sepsis outcomes. UHPLC-based methods can identify between 100 to 3000 unique proteins in patient plasma [55]. However, cost can be prohibitive to perform in appropriately powered patient studies and because of high risk for false positives require validation.

Recently, an effort has been made to move towards the analysis of microbiome and epigenetic factors that may affect patient outcomes. For the microbiome, deep sequencing or quantitative PCR (qPCR) of RNA and/or DNA can identify known and novel viral and bacterial infections with high accuracy [56]. The viral or bacterial genomes found in whole genome sequencing are identified by taking reads that fail to align to the host genome and testing for alignment to references of known viral, parasitic and bacterial reference genomes. This method successfully identified a novel human pegivirus that was found in a patient that died from sepsis from an unknown etiology [57]. While highly accurate, the time to analysis can take days to weeks, whole genome sequencing is costly, and it can also identify commensals that may not be related to the disease. Deep sequencing technologies have also been used to determine epigenetic changes due to sepsis such as DNA methylation [58], histone modification [59], oxidative base damage [60] and mirRNA, ncRNA and mitochondrial RNA expression [61]. These emerging studies could provide unique insight in the disease process.

Using the methods described above, researchers have identified hundreds of novel biomarkers that have been linked to sepsis [33]. However, many of the signatures may or may not be specifically related to the pathophysiology of the disease. Moreover, due to the complexity of the disease, many of these biomarkers fail to validate in prospective trials [33]. Potential reasons these biomarkers fail are related to the fact that many of these studies are limited in the number of patients enrolled, overfitting of predictive models, and the signatures are not independently validated in unique and diverse patient cohorts.

Another potential concern is that the identification of biomarkers is often agnostic to the biology of the disease and therefore may not be relevant to the disease course. Recent advances in systems biology can minimize some weaknesses of the single omics methods by utilizing holistic approaches that compare multiple data streams for correlative molecular pathways in the host response. The integrative approach assumes that a comprehensive, hypothesis-agnostic description of the molecular changes can identify unbiased signatures that maybe lost in one-dimensional analyses [35, 37]. Moreover, the correlation of molecular networks and pathways in complementary datasets can potentially identify and prioritize likely causal molecular mechanisms. For example, metabolomic changes that correlate with transcriptomic or proteomic pathways can identify known or novel substrate-enzyme-product reaction models. These pathways point not only to key biomarkers that predict outcomes but also highlight hypothetical mechanistic models that are not easily found in traditional one-dimensional omic analyses, nor easily tested in typical reductionist methodologies.

The primary strength of integrative omics analysis is that it is hypothesis generating and cab therefore identify novel pathways from precious clinical samples without a priori knowledge. The results can be tested in future targeted hypothesis driven studies using in vivo and in vitro models. Multiple omic’s strategies can also be coupled with reductionist methodology to look at many downstream targets rather than one single change at a time.

4. Transcriptomic Analysis of the Host Response to Diagnose Infections and Predict Patient Outcomes.

As previously mentioned, there are currently no molecular biomarkers that diagnosis infections that are sensitive enough to utilize for antimicrobial stewardship. The “Gold Standard” for identification of infections is blood culture analysis. However, the time-to-results are often too late to make informed decisions and are positive in only a minority of cases (−20%) [62]. Infections can also be identified by PCR analysis of bacterial 16S ribosomal RNA (rRNA) that has been developed for species/genus level identification of bacterial infections. While the time-to-results are considerably faster than blood cultures, the results are only moderately better and are best used as a rule-in diagnostic [63, 64]. A rule-out test that relies on the host response takes into account that the host immune response can effectively control minor infections, while also differentiating between systemic inflammatory responses caused by infectious versus non-infectious etiologies. However, due to ethical as well as clinico-legal concerns, it is suggested that a rule-out test for diagnosis should have an area-under the receiver operator characteristic curve (AUROC) greater than 0.9 with a very low false negative rate [33].

Over the past several years numerous transcriptomic studies have looked at sepsis diagnosis as well as patient outcomes (review in [65]). Unfortunately, these biomarker studies have performed poorly as numerous pathways and signatures were implicated, but rarely validated independently. The one consensus from these studies was that both innate and adaptive immune responses are activated throughout the course of the disease [66].

Recently, significant strides have been made as the sensitivity of the transcriptomic platforms has improved, as well as advances in predictive modeling techniques that reduce over-fitting, clinical study design that includes planned independent validation cohorts, greater patient heterogeneity and diversity, increased patient enrollment numbers and the usage of public data for meta-analysis. Several promising biomarker studies have been published that were well powered, independently validated and have high accuracy in predicting viral versus bacterial infections in pediatric [67, 68], and adult sepsis [41, 69, 70]. The studies vary in number of predictive molecules from as low as two signatures to over 100 signatures. Future large, prospective trials will be needed to validate the models. However, the transcriptomic signatures are promising as new technologies for qPCR analysis are moving toward point-of-care devices that can measure small panels of transcriptomic biomarkers quickly and accurately [71].

5. Metabolomic Changes in Sepsis Suggest an Energy Crisis in Nonsurvivors

Lactate remains one of the most useful biomarker for the diagnosis of septic shock and outcome prediction [1, 16]. Increased blood lactate concentrations was considered to be caused by poor blood perfusion that can lead to local and systemic hypoxia. Early Goal Directed Therapy (EGDT) was based on this premise and the primary aim is to maximize perfusion and oxygen delivery [13, 72]. These protocols dramatically improved patient outcomes in the early hours after a patient developed septic shock. However, some patients that didn’t respond to EGDT had high lactate concentration despite appropriate cardiac function and blood oxygenation. The increased concentration of lactate was believed to be due to cytopathic hypoxia [73]. These changes suggested that mitochondrial dysfunction could play a role in sepsis-induced multi-organ failure [74].

Since the lactate concentrations can often be mitigated by improved blood oxygenation and perfusion in the early hours after a patient develops septic shock, it is not a good marker for the identification of mitochondrial dysfunction. To identify new biomarkers that can predict poor patient outcomes, a multi-omic analysis of WBC and plasma isolated from patients with suspected community-acquired infections in the emergency department was performed [75, 35, 36]. A broad spectrum analysis of metabolomic and proteomic changes found that a number of biochemical pathways were significantly altered at presentation in the plasma of patients who died. The biochemical changes were primarily related to six biochemical pathways: 1) decreased acyl-glycerophosphocholines (acyl-GPCs); 2) increased substrates for the key redox coenzyme nicotinamide adenine dinucleotide (NAD); 3) increased tricarboxylic acid (TCA) intermediates; 4) increased short-chain, medium-chain fatty acids and branched-chain amino acids (BCAA) bound to carnitine; 5) sulfated steroids; 6) increased bile acids. The changes in TCA, NAD+ and carnitine esters pointed toward an energy crisis in nonsurvivors. Most of these biochemical changes were validated in an independent cohort of sepsis patients enrolled in the ICU. Using predictive modeling approaches, the metabolites could predict sepsis patient outcomes in the ED and ICU better than lactate, SOFA and APACHEII [76, 36, 37]. Many of these metabolites have been independently identified in a number of recent clinical ICU studies pointing to a common mechanism of metabolic dysfunction [77, 40, 78, 79].

The most notable altered biochemical pathway was increased concentrations of small-chain, medium-chain fatty acids and branched-chain amino acids (BCAA) bound to carnitine which from hereon will be referred to as carnitine esters [40, 37, 35, 36]. The increases are similar to plasma changes found in patients with acyl-CoA dehydrogenase deficiencies which is characterized by the accumulation of fatty acids bound to carnitine [80].

A similar systems biology analysis of sepsis was performed in a non-human primate (NHP) model and confirmed that carnitine esters were increased in septic animals [37]. RNAseq analysis was performed on the whole lungs taken from these animals. Along with expected changes in the inflammatory response, the primary pathways affected were related to mitochondria, peroxisome, fatty acid metabolism and BCAA degradation. Cross correlation analysis identified known substrate-enzyme-product reaction models. A strong proof of concept of this hypothesis was the correlation found between NAD+ related substrate metabolites including increased kynurenine to the enzyme kynureninase. Transcriptomic expression of the TCA checkpoint proteins pyruvate dehydrogenase kinase 2 (PDK2) and pyruvate dehydrogenase phosphatase (PDP2) were decreased and had a strong negative correlation with TCA-associated metabolites. Acyl-GPCs and acyl-carnitines were strongly correlated with decreased expression of β-oxidation and peroxisome related enzymes. Many of the metabolites also correlated with transcription factors that are known to regulate mitochondrial biogenesis and fatty acid oxidation such as PPARG co-activator 1A (PPARGC1A). The results suggested there was a coordinated genomic response that leads to dysregulated β-oxidation and TCA energy production. Over the past several years, several drug therapies using PPAR agonists have demonstrated improved survival in experimental models of sepsis [81–86]. Most of these studies have focused on the anti-inflammatory effects of PPAR activation. Therefore, direct analysis of the TCA cycle, β-oxidation, carnitine esters and immune function are needed to determine if these therapies improve survival by targeting these clinical endpoints.

6. Molecular Changes to the Innate and Adaptive Immune Response and the Relationship to Mitochondrial Dysfunction

While sepsis has traditionally been associated with a hyper-innate immune response, growing evidence supports the concept that immunological dysregulation is a central pathogenic response to sepsis [87, 88, 28]. Understanding how the innate and adaptive immune response are affected may lead to improved therapeutics.

Co-morbidities such as cancer, immunosuppression, diabetes, cardiovascular disease, renal function can all influence the immune response. The course of the disease can also influence the innate and adaptive immune response. For example monocytes in sepsis patients switch from an innate response to an adaptive immunosuppressive phenotype. A number of different pathways have been identified that potentially regulate the adaptive immune response in sepsis (reviewed in Bermejo-Martin et al. [87]. These include changes to defective antigen presentation, defective NK cell mediated immunity, increase in T-regulatory cells, increased expression anti-inflammatory molecules such as PD-1, decreased levels of immunoglobulins and alterations in neutrophil function.

Metabolomic changes have also been demonstrated to affect immune function in sepsis and renal dysfunction [38, 39]. It was found that patients that presented to the emergency department with sepsis and chronic renal failure had a pan-transcriptomic decrease in expression in white blood cells isolated from septic patients enrolled in the emergency department. Two metabolites allantoin and 4-pyridone-3-carboxamide (4PY) highly correlated with the majority of genes that were decreased due to chronic renal failure in sepsis patients [39]. It was speculated that 4PY, an endogenous inhibitor of poly(ADP-ribose) polymerase 1 (PARP1), could affect transcription in leukocytes. However, the mechanism has no yet been tested.

Mitochondrial dysfunction related to sepsis outcomes has been implicated for more than 20 years as changes in the mitochondria numbers and size are associated with poor outcomes [74]. Recently, a number of predictive biomarkers directly related to mitochondrial proteins and metabolites has been associated with patient outcomes. These include cardiolipin [89, 90], mitochondrial DNA (mtDNA) damage-associated molecular patterns (DAMPs) [91–93] and the previously described carnitine esters [35, 37]. Interestingly, mtDNA has been reported to induce immune paralysis in septic patients [94]. TLR9-dependent immunosuppression in adaptive T-cell cytotoxicity was induced by a single dose of mtDNA in wildtype mice compared to TLR9 knockout mice.

Over the past five years, excellent strides have been made that have linked the mechanism of mitochondrial dysfunction to immune regulation and outcomes. The landmark Nature paper from Tannahill et al. [95] demonstrated that the TCA cycle is dysregulated in monocytes leading to an increase in succinate, the downstream stabilization of HIF1A and increased cytokine/chemokine production. Treatment of animals and cells with vigabatrim, which inhibits γ-aminobutryic acid (GABA) transaminase, partially prevents the metabolism of GABA into α-ketoglutarate and reduces mortality in mice.

A Warburg-like phenotype (i.e., aerobic glycolysis) was found in monocytes with a decrease in β-oxidation and expression of the related genes, along with increased lactate and NAD+ production due to LPS treatment. In patients with sepsis, isolated monocytes demonstrated immunotolerance with decreased oxygen consumption. However, when these cell were treated with IFN-γ cellular metabolism partially restored function leading to increased lactate production and upregulation of cytokines [96]. A similar finding was noted with the treatment of monocytes isolated from sepsis-patients with a sirtuin 1 (SIRT1) inhibitor EX-527 [97, 98]. Mice treated after 24h with EX-527 also induced a Warburg-like phenotype in monocytes, immune activation and improved survival.

The results demonstrate that the Warburg effect is critical for immune function in monocytes. These results point to promising new therapies that can re-activate the innate immune response in the later stages of sepsis potentially improving survival.

7. Mitochondrial Dysfunction Can Lead to Permanent Muscle Loss in Septic Shock Survivors

While catabolism of muscle tissue may be beneficial by providing amino acids to the immune system, and the liver for production of acute phase reactants, the loss of muscle tissue can have long-term detrimental effects. The loss appears permanent as physical disabilities can last for up to five years in septic shock survivors [99]. However, the mechanism for persistent muscle loss remains unclear. In a CLP model of sepsis, muscle stem cells were impaired after sepsis with compromised activation, proliferation and expression of myogenic markers [10]. In this study, the authors noted that there was persistent decreased expression of PPARGC1A, mitochondrial activity and mitochondrial mass was reduced. Animals treated with mesenchymal stromal cells (MSCs) 6h after the CLP challenge demonstrated that the majority of mitochondrial parameters returned to normal in the muscle stem cells including improved ATP levels, mitochondrial ATP and mitochondrial biogenesis. The effect was believed to be protective of mitochondrial dysfunction in these cells as the MSC were not detected in the muscle fibers of the surviving animals.

The mechanism for MSC-induced protection is unknown; however, it has been demonstrated that MSCs release exosomes during sepsis that harbor miR-223 which is cardioprotective in the CLP-induced sepsis model [100]. This effect may be related to the anti-inflammatory effects of miR-223 which reduces the expression of proinflammatory cytokines. The fact that MSC therapy remains effective in animal 6h post CLP-challenge after the majority of muscle stem cells have been negatively affected by sepsis suggested that other mitochondrial biogenesis factors may also play a role in the positive effect of MSC-therapy [10].

8. Genetic Variations Relationship to Patient Outcomes

Death from sepsis demonstrates surprising heritability that is higher than any other causes including cancer and cardiovascular disease [101, 102]. Poor sepsis outcomes have been associated in family and twin studies suggesting a common genetic basis [103, 104]. Many opportunistic infections and sepsis outcomes have been associated with rare recessive Mendelian disorders such as complement factor properdin (CFP), toll-like receptor 4 (TLR4), TLR2, CCR5 and solute carrier protein 11 (SLC11A1) [105–110]. In GWAS, variation in TLRs, inflammatory response or coagulation have been implicated [111–114]. However, these studies have had mostly limited success as it is difficult to link common variants to complex polygenic diseases and the studies require extremely large n-values [115–117].

Looking at exome sequences, it is possible that heterozygous rare variants that lead to loss-of-function (LOF) could also predict poor outcomes. The rare variant hypothesis is based on the concept that rare SNPs have functional phenotypic relevance and greater effect size than common SNPs [118]. However, these studies are still very difficult to perform. A tour de force manuscript recently published in Science compared exome sequencing of more than 50,000 patients to 14+ years of electronic health records (HER) to identify rare heterozygous variants that may link to human diseases [119]. The study found that on average each person carries approximately 21 LOF genes and most of these variants were found in less than 1% of the population. The results confirmed that glucose-6 phosphatase (G6PC) is associated with triglyceride levels. However, despite the incredible n-value and design of the study, no new gene variant-based diseases were identified.

However, animal models may still hold promise when comparing the outcomes to rare genetic variants that are integrated with clinical outcomes. Proprotein of subtilisin/kexin type 9 (PCSK9) is a regulatory molecule that inhibits the clearance of endogenous cholesterol from the blood by reducing low-density lipids receptors (LDLRs) expressed on the surface of hepatocytes). It was hypothesized that it would also regulate the metabolism of microbial lipids such as lipopolysaccharide (LPS) [120]. PCSK9 KO mice had decreased cytokine production and pharmacological intervention in the wildtype animals improved survival. The authors then compared the most common missense loss-of-function variant to the most common gain-of-function variants in the Vasopressin and Septic Shock Trial (VASST). The LOF patients had moderate by significant improvement in 28d survival as decreased cytokine production. The reduction in cytokine production was replicated in healthy volunteers that were challenged with LPS. The results suggest that improved metabolism of these microbial lipids could improve patient outcomes.

Future studies will likely move toward determining how polymorphisms and epigenetic factors within promoter regions can affect gene expression. The ENCODE project determined that variants and epigenetic factors within the promoter regions can lead to dysregulation of gene expression and affect common diseases [121]. However, identifying these variants is quite difficult as there are currently no reliable computer programs that predict how variants in the genome may alter transcription factor binding. Indeed, transcription factor binding is still an early developing field and many binding sites that haven’t been directly tested remain mostly hypothetical.

However, integration linking metabolomic and transcriptomic changes with genetic and epigenetic changes may improve the identification of disease causing mutations [122, 123]. For example, sepsis outcomes have been linked to a single nucleotide polymorphism (SNP) in the dimethylarginine dimethylaminohydrolase 2 (DDAH2) gene. SNP rs805305 is located in the promoter region of the nitric oxide (NO) regulator. DDAH2 regulates the metabolism of asymmetric dimethyl arginase (ADMA), an endogenase inhibitor of nitric oxide synthase (NOS). AMDA and symmetric dimethylarginine (SDMA, the downstream metabolic product of ADMA metabolism) is significantly increased in sepsis survivors [124, 35].

Abstracts presented at the 2016 American Thoracic Society Annual meeting by Andrew Gordon’s group demonstrated that in the Vasopressin versus Noradrenaline as Initial therapy in Septic Shock (VANISH) trial, elevated levels of ADMA, when corrected by SDMA clearance, demonstrated a protective effect in septic shock [125]. When looking at the SNP of the DDAH2 promoter, the group found that the G:G genotype is associated with reduced risk of death, septic shock and SOFA score compared to the C:C and C:G genotype; and the plasma ADMA/SDMA ratio was significantly increased in the G:G genotype [126]. The group previously published results that demonstrated pharmacological intervention of DDAH1 in a sepsis animal model protects the animals from septic shock [127]. These findings suggest pharmacological intervention of ADMA metabolism in patients with the C:C or C:G genotype may improve patient outcomes.

There is clear potential to improve genetic understanding of poor sepsis outcomes. However, as the results presented herein demonstrate, it will likely take large, multicenter trials with detailed transcriptomic, metabolomic, genomic and phenomic (i.e., patient demographics, EHR) data that can link the biomarker changes to in the genomic variants.

9. Conclusion

Sepsis diagnosis has relied on indices of infection and clinical presentation that that identify sepsis and septic shock (Figure 1). Currently these diagnoses govern therapeutic treatment that focuses on appropriate, antimicrobial therapy, improved cardiac function and oxygen delivery. However, these diagnostics and are not very predictive of final patient outcomes, the specific infectious organism or global physiological abnormalities. Nor are they targeted at specific mechanism driving organ dysfunction and long-term outcomes. Here we summarize evidence supporting the concept that integration of clinical and multi-omic data that has the potential to lead to improved diagnosis, antimicrobial stewardship and drug discovery. The ultimate goal of precision medicine will be aimed at therapeutic interventions that improve both patient survival as well as long-term quality of life.

Figure 1: Current and future strategies for sepsis diagnosis and mechanism.

Sepsis and septic shock (blue ellipses) often lead to poor QOL and patient outcomes (red and blue rectangles). Current practice and current molecular understanding of the disease has improved beyond simple clinical indices (green ellipses). However, the ultimate goal is to improve diagnosis, and precision medicine-mediated therapies (orange ellipse). * Septic shock defined as patient requiring vasopressors, or a mean arterial pressure (MAP) <65 mmHG or lactate >2.0 mmol/l after adequate fluid resuscitation.

The integrative omic strategies are the only way to define mechanism in clinical patient samples that can guide future precision medicine therapies. Therapeutic strategies that promote mitochondrial biogenesis and immune activation appear to be strong targets for improved patient outcomes. Large, multicenter trials that combine these multi-omic strategies will likely have improved success at identification of new genetic risk factors. When performed in concert with reductionist methodologies in preclinical animal studies that take into account the clinical variables there is strong potential for the discovery of new therapies that finally improve patient outcomes due to sepsis and multi-organ dysfunction.

Key Points.

1. Sepsis leads to high mortality and patient costs yet there are few biomarkers that can accurately diagnose infections and predict patient outcomes

2. Integration of large molecular datasets can identify novel mechanisms of disease.

3. The integrative analysis has highlighted the importance of mitochondrial function in predicting patient outcomes and regulating the innate and adaptive immune response