The goal of precision medicine in the context of critical care is for treatment strategies to be developed that take into account the variability seen between individuals.
Heterogeneity in the groups of patients with conditions seen most often in the intensive care unit (sepsis and acute respiratory distress syndrome) may explain the lack of significant progress in treatment of such syndromes.
Studies related to genetics, transcriptomics, and plasma biomarkers have been used to begin to understand the heterogeneity in patients with sepsis or acute respiratory distress syndrome and to identify subgroups with common characteristics.
Thus far, both studies using transcriptomic approaches and those using novel analytic approaches to serum biomarkers suggest that patients with septic shock and acute respiratory distress syndrome can be separated into at least two subgroups distinguished in part by characteristics of their immune response.
In 2015, the National Institutes of Health launched a Precision Medicine Initiative with the goal that both prevention and treatment strategies in medicine move toward an approach that takes into account the variability observed between individuals. To date, this has been used most successfully in the field of cancer, which has used genetics and transcriptomics to examine heterogeneity within cancer types and to identify subtypes based on specific molecular signatures. Such studies have resulted in targeted treatments based on the underlying pathophysiology seen within a cancer subtype. Although there is significant heterogeneity in many of the conditions seen in both pediatric and adult intensive care units (ICUs), studies are just beginning to address this heterogeneity by identifying subgroups at highest risk of poor outcome (prognostic enrichment) or subgroups with common underlying pathophysiology that may be more likely to benefit from a specific treatment (predictive enrichment).
Two of the most common causes of admissions to ICUs are sepsis and acute respiratory distress syndrome (ARDS), two conditions known to have significant heterogeneity both in underlying triggers and in response to treatment. Why one individual responds to treatment and another progresses to severe illness and even death is unclear. This variability in response to treatment is thought to be due, at least in part, to the heterogeneity within these conditions. In addition, the failure of many of the treatments tested in the last 20 years has been attributed, in part, to heterogeneity among these patients. Simulation models indicate that heterogeneity in cohorts of patients with acute respiratory failure can significantly impact clinical trial results by showing no benefit for the entire cohort, although a high-risk subgroup of patients may actually benefit, or by showing benefit for the entire cohort, though a subgroup of patients may incur harm. The recent consensus in the field is that strategies aimed at prognostic and/or predictive enrichment should be leveraged in research related to critical illnesses to identify subgroups at highest risk and/or distinguished by differences in the underlying pathophysiology to identify targeted therapies with a higher likelihood of reducing morbidity and mortality. This chapter will discuss results of studies related to genetic variation, gene expression, and plasma biomarkers designed to begin to address the heterogeneity seen in patients with sepsis and ARDS.
Genetic variation and critical illness
When contemplating the great diversity among humans, it is somewhat surprising to realize that the deoxyribonucleic acid (DNA) of two unrelated humans is more than 99.9% identical. Although the vast majority of nuclear DNA is identical from one person to the next, a small fraction of DNA sequence (∼0.1%) varies among individuals and is responsible for the genetically determined variation in our physical characteristics and physiology. Genetic variability also appears to be involved with susceptibility to some diseases, therapeutic responses to treatment, and disease outcomes.
The sequencing of the human genome and advent of high-throughput sequencing and genotyping technologies have revolutionized the understanding of gene structure and genetic variation. Many genes are polymorphic; that is, there are small differences (variations) in DNA sequence between individuals. Single-nucleotide substitutions in DNA are called single-nucleotide polymorphisms (SNPs) or single-nucleotide variants (SNVs) and are the most common type of polymorphism, thought to account for approximately 90% of human variation. One SNP is believed to occur in every 100 to 300 bases. Although such substitutions may occur spontaneously and represent a new mutation, the vast majority of the observed substitutions are stable variations in the human gene pool.
Variations in DNA sequence within genes may also be due to insertions or deletions of sections of DNA or to the presence of a variable number of tandem repeats of short, repetitive DNA sequences. Some of these insertions or deletions, although submicroscopic, can be relatively large, resulting in gene copy number variations (CNVs). It has become clear that CNVs are common in human genomes and contribute significantly to human genetic variation. , It is thought that alterations in phenotypes due to CNVs are due to differences in gene dosage or to gene disruption, which may be caused during duplication or deletion.
Although polymorphic sites within a gene do not necessarily affect the expression or the function of the gene product (as one might suspect given the huge number of polymorphic sites in the human genome), it is clear that many genetic variants do alter protein function or level of expression. An alteration in protein function may occur after a change in a single base in the coding region of the gene, which alters the corresponding amino acid in the protein and, hence, its function. Polymorphisms can also have significant effects without altering protein function. A polymorphism occurring in a promoter region, which controls gene expression through controlling messenger ribonucleic acid (mRNA) synthesis, may lead to increased or decreased synthesis of that protein, which may have significant effects on cell function. Although polymorphisms are still being mapped and the function of most polymorphisms is still being defined, it is clear that these genetic variations account for many inherited human phenotypes, from differences in hair color to differences in response to medications.
Recent studies in both critically ill adults and children have suggested that the clinical presentation, treatment, and outcome from critical illnesses are influenced in part by the genetic makeup of the patient. Unlike genetic diseases that have a mendelian inheritance pattern and are caused by a mutation in a specific gene, diseases and syndromes seen in critically ill patients, such as sepsis and ARDS, are complex disorders in which genetic predisposition to the development of the disease or to disease outcome is much more complicated and likely involves multiple genes in addition to other factors, such as environmental exposures.
There are two hypotheses addressing the involvement of genetic variants in non-mendelian inherited diseases or predispositions. One hypothesis is that the genetic component of these diseases or predispositions is due to the combination of many common genetic variants (15–50) acting together. Early studies aimed at identifying such variants used a candidate gene approach in which the association of variants in the genes known to be involved in the pathophysiology of sepsis or ARDS was examined for association with development or outcome of disease. In the last 15 to 20 years, many genetic association studies have been reported; early studies often failed to replicate previously reported findings. There are many reasons for this lack of consistent results in early studies, including inadequate power (resulting from small cohorts and/or examination of multiple polymorphisms), population stratification (resulting from inclusion of multiple races and/or ethnicities in the study group), different study populations, and different definitions of response. Currently, journals require authors to demonstrate that their study does not suffer from such weaknesses. Journals also often require such findings to be supported by biological evidence in animal or in vitro models showing the potential effect of the variant on the function or level of the protein, thereby demonstrating its potential role in the syndrome or disease.
As technology has advanced genome-wide association (GWA) studies, at 1 to 2 million variant sites across the genome have been used to identify common variants associated with disease. In a GWA study, genetic variants across the genome are studied with no preconceived notion as to whether the gene of interest encodes for a protein involved in disease. The advantage of such an unbiased approach is that it has the capability to identify variants in genes not yet known to be involved in the disease of interest and thus can increase the understanding of the underlying pathogenesis. GWA studies generally require many hundreds to thousands of patients while candidate gene association studies can be done with hundreds of patients. Consequently, most genetic association studies done in patients with sepsis or ARDS have used a candidate gene approach.
Thus far, many studies examining the impact of common genetic variants on disease have explained only a small component of the heritability of such disease. Consequently, more recently, a second hypothesis proposed is that these multifactorial diseases and predispositions are due to the combination of a smaller number of more rare polymorphisms. Such rare variants are generally identified by either exome sequencing (sequencing only the DNA that codes for coding regions in proteins) or whole-genome sequencing. Exome and whole-genome sequencing costs have decreased substantially. The next 5 to 10 years will likely demonstrate whether rarer polymorphisms are also involved in non-mendelian inherited diseases and predispositions.
The roles of genetic variations in sepsis and ARDS have been two areas of intense research. The primary objective in both areas is to investigate whether new or previously identified genetic variations are associated with the susceptibility to or outcome from disease. These studies have the potential to impact clinical practice because they may help identify patient subgroups who have an increased susceptibility to or have increased risk of worse outcome from critical illnesses, thereby identifying those in most need of treatment. In addition, they may discover subgroups of patients with common underlying pathophysiology identified by sharing common variants in genes or pathways involved, thereby generating the potential to tailor treatment to the specific subgroup. This section will briefly highlight some of the studies supporting the concept that genetic variations influence the susceptibility to or outcome from critical illnesses. These studies are generally designed to determine whether specific genetic variants are associated with susceptibility or outcome of critical illness.
Genetic variation and sepsis
A genetic component to outcome from infection is supported by a classic study in adoptees that demonstrated that the early death of a biological parent from infection is associated with a much greater risk of death of the adopted child from infection than is the death of the adoptive parent from infection. At this time, the association of genetic variants with susceptibility to or outcome from sepsis has been examined using candidate gene association studies and, less frequently, GWA studies and exome sequencing. Many genetic variants have been reported to be associated with sepsis or outcome from sepsis. Genetic variants that most consistently demonstrate association with susceptibility to sepsis include variants in genes involved with recognition of and response to pathogens, including variants in mannose-binding lectin (MBL), Fc γ-receptors (FcγRIIIb and FcγRIIa receptors), Toll-like receptor-1 (TLR-1) and TLR-2, and tumor necrosis factor-α (TNF-α). Many fewer studies have been conducted in children; currently, only MBL and TLR-2 have strong data supporting a potential role in susceptibility to sepsis. Studies in children have reported an association of MBL variants with an increased risk for meningococcal infections, , with pneumonia and sepsis in neonates, with acute respiratory infections in children, and with hospitalizations for infections in children. Several studies have indicated that a variant in TLR-2 is associated with infection in children. , These findings have been supported by a recent meta-analysis.
Genes with the strongest evidence for association with outcome from sepsis in adults include interleukin-1 (IL-1) receptor–associated kinase-1, TLR-1, , and proprotein convertase subtilisin/kexin type 9 (PCSK9). In children, there are many fewer studies; however, there is some evidence that TLR-1 also impacts sepsis outcomes in children. The strongest support for the association of a genetic variant with outcome in children with infection is seen with a variant in the promoter of plasminogen activator inhibitor-1 (PAI-1). This variant is associated with differences in plasma PAI-1 levels and with outcome in children with meningococcal disease.
Several GWA studies have been performed in adults and two in children. A variant in the Fps/Fes-related tyrosine kinase has been shown to be associated with death in patients with sepsis due to pneumonia and a variant in the vacuolar protein sorting 13 homolog A gene with death in a more general cohort of adults with sepsis. Future studies will be needed to determine whether disease models support a role for these two proteins in the susceptibility to or outcome from sepsis. GWA studies in children indicate that there are variants in the complement H gene that are associated with risk of meningococcal disease. , The only whole-exome sequencing study published to date in individuals with sepsis suggests that a combination of rare deleterious variants can also be associated with sepsis disease course.
Genetic variation and acute respiratory distress syndrome
The influence of genetic variation on ARDS is another area of intense research. Thus far, variants in ∼20 genes have been reported to be associated with the development of ARDS or the outcome from ARDS in multiple studies or in studies with replication cohorts. , Most of these genes also have data derived in model systems that support a possible mechanism for how the variant may be associated with ARDS risk or outcome. These genes include the genes encoding TLR-1, , surfactant protein B (SP-B), angiotensin-converting enzyme, IL-6, and angiopoietin-2. In children, there is a paucity of studies examining association of variants with pediatric ARDS (PARDS). Thus far, only variants in SP-B, IL-1 receptor antagonist, IL-10, TNF-α, and the cystic fibrosis conductance transmembrane regulator have been reported to be associated with lung injury in children.
Recently, several GWA studies have identified novel variants associated with ARDS. In African Americans with ARDS, a variant that changed an amino acid in the P-selectin ligand gene, a protein not previously known to be involved in the pathophysiology of ARDS, was associated with risk of ARDS. Animal model studies were used to confirm that P-selectin is involved in ARDS pathophysiology. Further studies will be required to determine whether the P-selectin ligand is involved in the pathophysiology of ARDS in humans. Last, an exome-sequencing study in which all exons encoding for proteins were sequenced has been completed in ARDS patients. The study identified variants in two genes not previously implicated as playing a role in ARDS pathophysiology: the X-linked gene arylsulfatase and the XK Kell blood group complex member 3 gene. Future studies will be needed to determine whether this finding can be replicated and whether these genes are indeed involved in ARDS pathophysiology.
Although considerable progress has been made in identifying genetic variants associated with sepsis and ARDS, further studies are warranted. At this time, it seems clear that future studies will identify additional genetic variants that are associated with sepsis and ARDS and that future studies will need to address the impact of genetic variants in multiple genes concurrently on the susceptibility to and outcome from critical illness. Genetic variants have not yet been used to attempt to prospectively identify subgroups of patients in part because of the lack of technology that enables a real-time determination of genotype and the lack of a clear understanding of which variants are most important to consider.
Transcriptomics and critical illness
The transcriptome refers to all RNA produced in the cell, including mRNA, ribosomal RNA (rRNA), and transfer RNA (tRNA) in addition to noncoding RNAs, such as microRNAs (miRNAs). The majority of research on the transcriptome has focused on either examining gene expression profiles in different disease states, such as sepsis or ARDS, or following a specific treatment. Changes in gene expression are determined by assaying the amount of mRNA that is present for a gene of interest. When this is done for all genes expressed in a cell, this is called transcriptomics . In most cases, the level of mRNA expression is an indicator of the amount of protein present in the cell. Transcriptomic studies examining the level of expression of all genes in a sample can be done using genome-wide gene expression arrays or, more recently, by sequencing all mRNA in the sample (RNAseq).
Gene expression studies in the pediatric ICU setting have primarily been used to examine expression profiles in children with sepsis and septic shock. Wong and colleagues have pioneered this field and have studied the largest cohort of children with sepsis in whom gene expression profiles have been generated. Genome-wide expression profiling in children with sepsis has been instrumental in providing a better understanding of the molecular biology of sepsis and septic shock and in identifying subgroups in children with septic shock. Analysis of genome-wide gene expression profiling from whole blood indicates that children with septic shock fall into two different endotypes (subgroups distinguished by differences in the underlying pathophysiology) based on their gene expression signature. Children with endotype A have decreased expression of genes of the adaptive immune system and glucocorticoid receptor signaling pathway compared with children with endotype B. This finding indicates that there are differences in the pathophysiology between these two groups. Interestingly, children with endotype A have a higher mortality rate compared with children with endotype B. Perhaps more important, additional multivariable analyses revealed that patients with endotype A exhibited an independent association of corticosteroid treatment with increased risk of death. If validated, septic shock endotype assignment might be useful for guiding treatment in pediatric septic shock patients.
The septic shock endotypes described by Wong and colleagues can be distinguished by the expression of 100 genes. Using only these endotype-defining genes, gene expression mosaics were generated that could be easily differentiated by clinicians who lack formal training, allowing them to identify the two septic shock endotypes. , Interestingly, very few of these genes yield proteins that are measurable in blood, indicating that these transcriptomic endotypes will likely complement and enhance the understanding of heterogeneity provided by plasma biomarkers.
Interestingly, these 100 pediatric shock endotype-discriminating genes can also be used to identify pediatric septic shock endotypes in adults with sepsis. Evaluation of the relationship between the pediatric septic shock endotypes A and B (defined using the 100-gene targeted gene expression data) and the two sepsis response signatures identified in adults with sepsis (SRS1 and SRS2) , resulted in identification of four groups with the endotype/SRS assignments—A/SRS1, A/SRS2, B/SRS1, and B/SRS2—indicating that the endotypes identified in children and adults may provide complementary information. Interestingly, both endotype A and the SRS1 endotype were independently associated with mortality in the adult cohort.
These studies clearly demonstrate that transcriptomics can identify subgroups of children with septic shock. Similar approaches are just beginning to be used in children with acute respiratory failure and, in the future, this approach is likely to provide data that will allow the identification of subgroups of PARDS patients with differing underlying pathophysiology. In addition, such studies will begin to provide information related to the similarities and differences between children with septic shock and PARDS. To be useful to the clinician and for clinical trials, identification of subgroups using transcriptomics will require a platform that allows the rapid identification of endotypes/subgroups. That this capability is approaching rapidly is indicated by the recent study indicating that endotype assignment in children with septic shock can be determined prospectively within ∼12 hours.
Plasma biomarkers and critical illness
A biomarker is a biologically derived indicator of the presence or progression of a disease, of the predicted outcome of the disease, or of the presence of subgroups within a disease. These markers have the potential to enable physicians to better identify the presence and progression of various conditions or illnesses, to help in targeting the optimal therapy, or to identify the optimal time point for delivery of therapy during the course of an illness. More recently, biomarkers have also been used to identify subgroups of patients within heterogeneous diseases so that the groups of patients who are most likely to benefit from a given therapy are treated accordingly.
Although genetic variants and transcriptomic signatures are both examples of types of biomarkers, a vast amount of research has been done examining plasma proteins as potential biomarkers. In general, the biomarkers examined are usually proteins that are known to be elevated during the disease process. Studies were performed to determine whether the protein could be used as a prognostic marker of disease or of outcome. Plasma biomarkers have been studied extensively in children and adults with relationship to both sepsis and ARDS. Recent studies have demonstrated that plasma biomarkers can be used by themselves or in combination with clinical and demographic data to distinguish subgroups in patients with sepsis and ARDS.
Plasma biomarkers and sepsis
In pediatrics, plasma biomarkers have been used to identify subgroups with higher risk of mortality in patients with sepsis and septic shock. Wong and colleagues have led this effort using data from their transcriptomic analyses to identify potential plasma biomarkers that might be involved in increased risk of mortality. Using Classification and Regression Tree (CART) analysis, they developed a risk stratification tool for septic shock called the pediatric sepsis biomarker risk model (PERSEVERE), which uses age and the level of five plasma biomarkers on day 1 of septic shock to stratify pediatric shock patients into groups with variable risk of mortality. When tested in a cohort of 182 children with septic shock, the patients could be grouped into those with low (<2.5%), medium (18%–27%) and high (47%–63%) predicted risks of mortality. Interestingly, CART analysis in adults with septic shock indicates that some of the biomarkers that predict risk of mortality are the same as those in children while others are different, suggesting that there may be differences in the underlying pathophysiology in children and adults.
A temporal model of PERSEVERE (tPERSEVERE) using the plasma biomarker levels on days 1 and 3 has also been developed. CART analysis suggests that there are two subgroups, one characterized predominantly by excessive inflammation and the other by immunosuppression. A modification of the PERSEVERE model that includes platelet count, called PERSEVERE-II, results in broader applicability of the model across septic shock phenotypes, including thrombocytopenia-associated multiple-organ failure. PERSEVERE has also been revised in a model that includes both the five plasma biomarkers and mRNA levels of genes previously identified as associated with mortality that were not related to the plasma biomarkers already evaluated in the PERSEVERE model (PERSEVERE-XP). PERSEVERE-XP results in significant improvement in the model for predicting mortality risk compared with PERSEVERE.
Plasma biomarkers and acute respiratory distress syndrome
Recently, plasma biomarkers have been used in different types of clustering analyses to identify subgroups in adults with ARDS. Latent class analysis (LCA) is one such approach. Its advantage is that it defines subgroups by considering multiple variables concurrently without considering outcome. LCA in adult studies of ARDS used demographic and clinical variables and biomarker levels in patients at the time of enrollment to determine whether ARDS subgroups could be identified. These studies used data derived from multiple large funded adult ARDS trials and, for the first time, demonstrated that endotypes exist in adult ARDS patients and that biomarkers can distinguish these endotypes. The two ARDS subgroups identified have different clinical and biological characteristics, clinical outcomes, and—remarkably—three of the four studies have shown a differential response to treatment, both nonpharmacologic , and pharmacologic. Both subgroups are also found across all ARDS severity levels. The two endotypes differ in part by the levels of biomarkers, with the “hyperinflammatory” endotype having higher levels of inflammatory biomarkers (IL-6, IL-8, soluble TNF receptor-1), lung vascular and epithelial cell markers (angiopoietin-2, receptor for advanced glycation end products), and the coagulation marker, PAI-1, compared with the “hypoinflammatory” endotype, in which inflammatory biomarkers are also elevated but not to the same degree. The hyperinflammatory endotype, which includes about one-third of the patients, is associated with higher mortality and fewer ventilator-free days. These adult endotypes appear to be stable for up to at least 3 days after meeting ARDS criteria. The two subgroups can be identified by a parsimonious model using three biomarkers, which will be advantageous if it is to be used clinically or to stratify patients in clinical trials.
A second clustering approach using only biomarkers also identified two subgroups, which were described as a reactive subgroup and an uninflamed subgroup. The reactive subgroup was characterized by elevated levels of inflammatory biomarkers and could be distinguished from the uninflamed subgroup using just 4 of the 20 biomarkers examined in the analysis. A transcriptomic analysis of leukocytes from the same patients indicated significant differences in expression between the reactive and uninflamed subgroups, supporting the idea that there are differences in underlying pathophysiology between the two groups.
No studies have examined whether there are endotypes in children with PARDS. However, many of the same biomarkers used in the adult ARDS clustering analyses described earlier have been shown to be associated with PARDS or PARDS outcomes in children. , Future studies will be required to determine whether hyperinflammatory and hypoinflammatory subgroups exist in children with PARDS.
Over the last 10 to 15 years, multiple approaches have been used to address the heterogeneity seen in patients with sepsis and ARDS. These studies have been designed to identify subgroups enriched in individuals that are at increased risk for poor outcome (prognostic enrichment) and subgroups enriched in individuals that have common underlying pathophysiology (predictive enrichment). Such studies have examined genetic variation, transcriptomics, and novel analytic approaches to evaluating plasma biomarkers to identify subgroups. Identification of subgroups at greatest risk of poor outcome or greater likelihood of benefiting from a particular treatment has the potential to revolutionize treatment of critical illness. Thus far, results from studies suggest that patients with septic shock and ARDS can be separated into at least two subgroups distinguished in part by characteristics of their immune response. However, to benefit from this information will require development of platforms that will allow the rapid measurement of the markers that distinguish these subgroups. There has been some progress toward such platforms, but more research will be required before these strategies can be used to stratify patients for clinical trials or precision treatments. Last, studies will also be required to better understand the relationship between the subgroups identified using these different, but potentially complementary, strategies and to determine whether such strategies can be used together to identify additional subgroups.