Biomarkers in critical care nutrition

The Nutrition Risk in the Critically Ill score (NUTRIC) is a prognostic tool developed by Heyland et al. [10, 32]. The parameters derived, including the inflammatory cytokine interleukin-6 (IL-6), are related to co-morbidity and severity of illness, and thus may be used to enrich study samples for those patients who are more likely to survive if their estimated caloric or protein needs are met. A post hoc analysis of the TOP-UP trial suggested that this may apply in prospective fashion [33], but similar discriminative performance was not replicated in another recent RCT [34], and additional validation is pending [27].

Of the measures that reflect whole-body nutrient absorption and metabolic responses, the most robust are measures of protein synthesis and breakdown [35]. In non-ICU settings, “whole-body protein balance,” as assessed by stable isotope tracer infusion, was a physiological indicator of metabolic response to nutrition or other anabolic interventions (e.g., insulin therapy) [36]. In the critically ill, protein balance was increased by energy intake, and the initiation of enteral feeding promoted anabolism [37, 38]. Although this method is robust, its complexity renders it costly and impractical as a clinical biomarker in large RCTs. Similar limitations appear in the pediatric population [39].

In contrast to whole-body protein balance, the calculation of nitrogen balance estimates the net difference between protein synthesis and breakdown [40, 41]. Nitrogen balance has been used to establish dietary protein requirements in healthy subjects [42], but performs poorly in critically ill adults [43, 44] and children [45, 46]. Sources of variability include rapid changes in total body water, as well as urinary and extra-urinary losses of nitrogen. Nonetheless, higher protein intake appeared to increase [47, 48], or was associated with higher [49], nitrogen balance in the critically ill but did not appear to predict clinical outcomes. Emerging markers of whole-body metabolism include measures of body composition [50] and are discussed further in the next section.

Although changes in whole-body protein, nitrogen balance, or body composition might be useful to monitor metabolic responses in the general population [35, 50], little is known regarding their relationship to clinical outcomes in the ICU. Other putative markers of nutritional status, substrate availability, and metabolism include albumin, pre-albumin, retinal binding protein, transthyretin, and transferrin [51‐54]. Their levels are not, however, acutely altered by nutrition interventions, nor do they reliably predict clinical outcomes.

One hypothesis underpinning the rationale for nutrition support is that the reversal of catabolism might attenuate inflammation or promote tissue regeneration, which are both linked to organ failure and mortality. In fact, the best predictors of mortality in the critically ill (e.g., APACHE, SOFA) include clinical measures of systemic inflammation and organ injury [55], suggesting that attenuation of the latter might improve clinical outcomes. Similarly, the reversal of skeletal muscle catabolism might prevent muscle atrophy during critical illness and improve functional outcomes [56‐58]. A recent review of ICU nutrition RCTs identified a need for studies with pre-planned evaluations of inflammatory mediators as potential biomarkers of nutrition interventions [16].

Little is known regarding how exogenous nutrition affects tissue injury or repair during critical illness, but biomarkers of systemic inflammation and organ injury have been well characterized [59]. These include markers of oxidative and nitrosative stress, complement activation, inflammation (i.e., pro-inflammatory cytokines), and organ dysfunction (e.g., anaerobic metabolism, cell death) [60]. In addition, patients with severe systemic inflammation can exhibit a compensatory anti-inflammatory response syndrome (CARS), which is associated with immune paralysis and reduced survival [61]. Single-parameter (e.g., BAL neutrophils for VAP [62]) or multiplexed (e.g., cytokine panel for ARDS [29, 62]) measures of inflammation can be used as prognostic biomarkers (reviewed in [63]), but their ability to predict responses to nutrition or metabolic heterogeneity in ICU patients has not yet been explored. Emerging studies in non-ICU populations suggest that systemic markers are promising candidates. In patients undergoing cardiac surgery, amino acid supplementation attenuated bypass-related induction of interleukin-6 (IL-6) and tumor necrosis factor-α (TNF-α) [64]. Elevated plasma IL-6 levels were associated with benefit from nutrition interventions in the ICU setting [10], but its prospective use as a monitoring or stratification tool has not been fully explored. These emerging studies suggest that the serial measure of inflammatory markers over time may be useful in determining the timing of nutrition interventions and in identifying patients who will benefit.

In summary, current biomarkers (Table 2) are either difficult to widely implement or fail to fully reflect the biological mechanisms that predict susceptibility or physiological responses to nutrition interventions [65]. To date, they have not been used as prognostic or predictive tools in contemporary clinical trials. In addition, unidimensional markers (e.g., IL-6, pro-calcitonin) have failed to confer prognostic classification in RCTs evaluating other ICU-related pathologies, such as sepsis or ARDS [63]. There is a pressing need for alternative approaches that can be exploited to better understand and monitor the biological mechanisms related to nutritional risk and metabolic heterogeneity.

Critically ill patients with organ dysfunction are catabolic on admission to the ICU [37]. This results from pathophysiological drivers (e.g., systemic inflammation, stress hormones) as well as nutrient restriction both preceding and after ICU admission [5, 6]. Like the described metabolic adaptations to critical illness [22], cellular adaptations to nutrient restriction can occur rapidly and are mediated by proteins that sense the levels of anabolic substrates or cellular energy. For example, the protein kinase “mechanistic target of rapamycin” (mTOR) coordinates a signaling hub linking nutrient or energy availability (e.g., amino acids) with cellular adaptations to nutrient restriction (e.g., autophagy), gene expression, and epigenetic mechanisms that predict prolonged changes in cellular function [67]. In fact, transporters or intracellular receptors can bind or transport single essential amino acids (e.g., leucine) which control mTOR anabolic activity in pharmacological fashion [68, 69]. Heterogeneity in these responses might arise from genetic variability or from the prolonged epigenetic shifts that can cause sustained dysregulation of bioenergetics and metabolism [67]. Other related cellular nutrient and energy sensors include “general control non-derepressable-2” (GCN2) [70] and the liver kinase B1/5′ AMP-activated protein kinase (LKB1/AMPK) pathway, which sense metabolic stress and control adaptive bioenergetic responses.

Population heterogeneity in the extent and kinetics of nutrient-related metabolic responses can only be defined by detailed multidimensional assays and their repeated measures over time. Several suppositions can nonetheless be made that pertain to biomarker development. First, tissues differ in their responses to metabolic stress, and their responses to nutrient availability are likely to vary. For instance, the profound protein catabolism and atrophy needed for mobilization of amino acids from skeletal muscle could be an adaptive mechanism to support other essential functions (e.g., immunity) required during severe illness [71]. Second, the dysregulated cellular metabolism observed during critical illness may subvert the use of available nutrients for ATP production or protein synthesis, and untimely nutrient delivery may inhibit adaptive responses [22]. Third, genetic and epigenetic variation in mechanisms that mediate energy and substrate availability, sensing, and metabolic responses may account for inter-subject variability. Finally, nutrient restriction or systemic inflammation may induce epigenetic responses leading to sustained alterations in the metabolism that cannot be immediately overcome by nutrition support. Detailed profiling of these changes in patients over time would shed important insights into the range and heterogeneity of responses to nutrient availability, as well as the optimal timing for initiating or withholding nutrition support. Time-sensitive factors include anabolic resistance, microbiome diversity, organ dysfunction, and endocrine stress responses [22].

Several features of intracellular metabolic sensing pathways (e.g., mTOR) render them potentially useful as biomarkers. They can exhibit pharmacological ligand/receptor properties and dose-response characteristics. In some cases, tissue availability of nutrition-derived substrates (e.g., leucine) correlated with dietary intake [72, 73]. Nutrient sensing mechanisms also mediate important organ-specific functions, including skeletal muscle catabolism [74] and immune dysfunction [75‐77], which can be measured and employed as surrogate outcomes. Finally, intracellular sensors of metabolism control the metabolic shifts in response to systemic inflammation or nutrient availability that drive the innate or adaptive immune repertoire [67, 78, 79]. However, the evaluation of enzymatic activity, post-translational modification, or levels of intracellular proteins is impractical in the clinical setting and exhibits suboptimal dynamic range and reproducibility. Alternatively, multiplexed assays for the biological effector mechanisms initiated by nutrient sensing (e.g., patterns of gene expression) can be performed on samples that are easy to obtain at low risk to the patient and that can be potentially adopted as point-of-care clinical tests.

Unlike single analytes, multidimensional markers are more likely to delineate biological signaling networks related to intervention and outcome of interest, or to define endotypes of nutritional responsiveness (Fig. 1) [80‐83]. The predictive ability of a surrogate biomarker is driven by the proportion of the treatment effect (PTE) that is mediated by the measured analyte(s) (reviewed in [26]). A biomarker would thus exhibit a high PTE if it was a biological mediator of the response to the intervention and of the clinical outcome of interest. For instance, as a surrogate biomarker, the detection of Streptococcus pneumoniae in the blood and its subsequent eradication would exhibit a high PTE when testing the effect of an appropriate antibiotic on mortality. Unidimensional markers have not captured the biological complexity of critical illness, nor have they effectively classified prognostic groups. Examples include the co-syntropin stimulation test and IL-6, respectively, which were employed unsuccessfully as potential prognostic classifiers in recent RCTs (reviewed in [63]). In contrast, multiplexed assays, or biomarker panels, resemble barcodes that can more effectively classify patients into subgroups according to biological factors that predict therapeutic or prognostic effects (Fig. 1c). Some examples include prognostic panels developed for pediatric and adult sepsis [84, 85]. It remains less clear, however, how well these tools monitor therapeutic interventions.

The development of multidimensional biomarkers is now feasible, in part due to advances in sequencing technologies, detailed annotation of DNA or RNA sequences, and evolving bioinformatic tools including machine learning approaches. Metadata using whole “omics” analyses in the clinical setting can be reduced to construct biomarker panels that exhibit high PTE and that capture patient heterogeneity [85, 86]. Untargeted approaches also provide value by revealing previously unknown (i.e., emergent) mechanisms of disease that can inform the development of novel therapeutics. For example, combined genetic, transcriptomic, and epigenomic analyses have revealed metabolic pathways that drive inflammation in response to microbial products [87], as well as metabolically driven epigenetic mechanisms by which inflammation can be altered in sustained fashion (i.e., trained immunity) [88]. These prolonged epigenetic modifications likely contribute to immune or metabolic dysfunction and might thereby modify the effects of dose and timing of nutrition in the critically ill.

We propose that the most promising assays are those related to nutrient availability and metabolism in easily accessible tissues or fluids and that exploit technologies which can be rapidly adapted to the clinical setting (Table 3).

Investigators and policymakers have outlined requirements for the development and adoption of biomarkers in clinical trial design [25]. For nutrition support, challenges include the complexity of formulations administered and the need to better define the biological mechanisms by which nutrient availability regulates cellular and organ function. Despite these caveats, the pharmacodynamic properties of complex nutritional interventions can be measured (e.g., whole-body metabolism, nutrient-sensing pathways) and can be exploited for biomarker development. Moreover, individual or limited mixtures of macro- and micro-nutrients are increasingly being tested in clinical trials for their immune-modifying effects (e.g., vitamin C, omega-3) [16, 104].

Given the current limitations in the field of critical care nutrition, collaborations between basic, translational, and clinical scientists in well-funded consortia are crucial to better understand the biological mechanisms related to nutrient metabolism in pre-clinical and clinical models. Prospective longitudinal cohorts that interrogate clinical data and banked samples for genetic and molecular analyses can be useful in identifying biomarker candidates. In the absence of existing observational cohorts, phase I and II trials provide an environment in which to prospectively measure the relationship between nutrition dose and the levels of putative biomarkers, to establish pharmacodynamic properties of nutrition support interventions, or to better define metabolic heterogeneity [105]. Effective approaches for the development of biomarkers and their incorporation within phase II and III clinical trials have been reviewed [105, 106]. The primary outcomes in phase II studies are usually measurable disease-related or physiological responses to an intervention (e.g., protein balance) and can be evaluated in small samples; their simultaneous correlation with multidimensional molecular analytes is a promising way to identify multiplex biomarkers of nutritional response. In addition, phase II trials are amenable to adaptive design, in which interim analyses of biomarker levels can be used to change randomization ratios, and thereby enrich for those patients most likely to respond to the intervention. Ultimately, biomarkers would contribute to the creation of core outcome sets that can be used to synthesize research across critical care nutrition trials [107].