Circadian Gene Expression Rhythms during Critical Illness

Abstract

Objective:

Core clock genes regulate tissue-specific transcriptome oscillations that synchronize physiologic processes throughout the body, held in phase by the central circadian rhythm. The central circadian rhythm rapidly dampens with onset of critical illness, but the effect of critical illness on gene expression oscillations is unknown. The objective of this study was to characterize the rhythmicity and phase coherence of core clock genes and the broader transcriptome after onset of critical illness.

Design:

Cross-sectional study.

Setting:

Intensive care units and hospital clinical research unit.

Patients:

Critically ill patients within the first day of presenting from the community and healthy volunteers.

Interventions:

Usual care (critically ill patients) and modified constant routine (healthy volunteers).

Measurements and Main Results:

We studied 15 critically ill patients, including 10 with sepsis and 5 with ICH, and 11 healthy controls. The central circadian rhythm and rest-activity rhythms were profiled by continuous wrist actigraphy and serum melatonin sampled every two hours along with whole blood for RNA isolation over 24 hours. The gene expression transcriptome was obtained by RNA-Seq. Core clock genes were analyzed for rhythmicity by cosinor fit. Significant circadian rhythmicity was identified in 5 of 6 core clock genes in healthy controls, but none in critically ill patients. TimeSignature, a validated algorithm based on 41 genes, was applied to assess overall transcriptome phase coherence. Median absolute error of TimeSignature was higher in individual critically ill patients than healthy patients (4.90 vs 1.48 hours) and was correlated with encephalopathy severity by Glasgow Coma Scale in critically ill patients (rho −0.54, p = 0.036).

Conclusions:

Gene expression rhythms rapidly become abnormal during critical illness. The association between disrupted transcriptome rhythms and encephalopathy suggests a path for future work to elucidate the underlying pathophysiology.

Introduction

Brain arousal, sympathetic tone, cardiovascular function, coagulation, immune system activity, glycemic control and metabolism all exhibit circadian variability.(1) The circadian system synchronizes tissue-level functions with sleep, activity, feeding and the environment, and dyssynchronization creates acute and cumulative deleterious effects on health.(2) The suprachiasmatic nucleus (SCN) functions as a central pacemaker for the circadian rhythm and communicates the central rhythm to peripheral tissues by autonomic signals, rhythmic melatonin secretion from the pineal gland and other efferents to maintain peripheral clocks in phase.(3, 4) Peripheral cellular clocks regulate gene transcription to change tissue function. Approximately 43% of mammalian protein-encoding genes demonstrate circadian transcription rhythms, mostly in an organ-specific pattern.(5) Therefore, it is the rhythmic activity of the transcriptome that directly effectuates much of the physiology of the circadian rhythm.

The central circadian rhythm rapidly becomes disrupted during critical illness, with encephalopathy as a major risk factor.(6, 7) Although the central circadian rhythm has an entraining influence on peripheral oscillators, transcription rhythms are an inherent cellular function. External inputs are required for individual cells to maintain alignment in phase (timing), but not to oscillate gene expression. The extent to which circadian oscillations in body tissues may persist intact during critical illness, with or without an intact central circadian rhythm, is not known. Moreover, biomarkers of the circadian rhythm (e.g. melatonin, cortisol, temperature) that are reliable in medically stable individuals are deranged by severe illness, so alternative methods to characterize circadian rhythm integrity and timing, like measurement of peripheral gene expression rhythms, are of interest. The objective of this study was to characterize the rhythmicity and phase coherence of core clock genes and broader transcriptome after onset of critical illness.

Materials and Methods

Study Patients and Measurements

Patients admitted to intensive care units at Northwestern Memorial Hospital between April 2014 and December 2018 were prospectively enrolled in an observational cohort study. The study was approved by the Northwestern University Institutional Review Board (STU00200711). Written informed consent was obtained from the patient or their legally authorized representative. Given the role of the brain in efferent expression of the central circadian rhythm, we had prespecified separate consideration for critical illness from brain disease and multisystem critical illness. We enrolled patients ≥18 years old with either spontaneous intracerebral hemorrhage (ICH) as a representative neurologic critical illness, or acute sepsis as a representative systemic critical illness. The methods of this study, patient characteristics and exposures, melatonin rhythm profiles and rest-activity rhythms have been reported elsewhere in detail.(6, 7) In brief, measurements relevant to this study were initiated within 24 hours of emergency department presentation and within two hours of enrollment. Blood was sampled every two hours for serum melatonin and 2.5 mL of whole blood were collected into RNA isolation tubes (PAXgene Blood RNA Tube, BD Diagnostics, Franklin Lakes, NJ, USA). RNA extraction was performed, followed by total RNA sequencing. Transcripts were aligned and quantified, and gene expression measurements were renormalized within subjects as we have previously described.(8) The transcriptome characterization study reported here was added as an aim to the primary referenced study and included all subsequently enrolled patients until accrual was complete. Healthy adult volunteers were studied in a clinical research unit under modified constant routine with the same blood sampling frequency, as previously reported.(8)

Analyses

We selected the following core clock genes for individual gene analyses: CRY1, CRY2, PER1, PER2, PER3, RORA, NR1D1, BMAL1, CLOCK and TIMELESS. Expression profiles for CLOCK, PER2, PER3 AND RORA were not available for the healthy controls. We used melatonin amplitude as a measurement of the central circadian rhythm, activity index amplitude from wrist actigraphy as a measurement of rest-activity rhythm, Sequential Organ Failure Assessment (SOFA) score as a measurement of multiorgan failure severity, and the mean of the hourly Glasgow Coma Scale (GCS) scores as a measurement of encephalopathy severity.(9–11)

Individual subject cosinor fits, including fitted amplitudes, were obtained for each gene along with a population-mean cosinor fit using the methods described by Cornélissen, including a rhythm detection test to reject a null hypothesis of zero-amplitude, as implemented in the Extended Tools for Cosinor Analysis of Rhythms (cosinor2 v0.2.1; Mutak, 2018) package for R.(12) When the rhythm detection test showed significance, we calculated the percent rhythm, which is the proportion of the variance explained by the rhythm, as a measure of the relative strength of the rhythm and tested its significance. Prior work has variably shown a 12 hour harmonic in circadian gene expression, and gene expressions have been fit in previous studies with a 24 hour period, 12 hour period, and with a multiple-component 12 and 24 hour period model.(13–15) We compared the fits of these three cosinor period fits using an established method that included periodogram construction, fit testing and optimization of percent rhythm.(16) We explored for correlations between cosinor-fitted gene amplitudes and the measures of central circadian rhythm strength, rest-activity rhythmicity, multiorgan failure and encephalopathy severity described above using Spearman’s rank correlation, reporting the correlation coefficient (rho; for which 0 means no correlation, 1 perfect positive correlation and −1 perfect negative correlation) and its p-value. Difference in gene amplitudes were tested between patients with ICH and sepsis using Wilcoxon rank sum tests.

Finally, we applied TimeSignature, a validated algorithm based on 41 genes, to evaluate the overall phase coherence of the rhythmic transcriptome. Prior work in healthy subjects has shown that TimeSignature accurately predicts sample time using within-subject normalized gene expression.(8) It is robust across study populations and assay platforms without retraining and renormalizing, and remains accurate even under conditions of dampened gene expression amplitudes. We used TimeSignature accuracy, measured by median absolute error, as a proxy measure for phase coherence across the set of 41 gene used in the algorithm. Statistical analyses were performed in R version 3.5.2 (R Foundation for Statistical Computing, Vienna, Austria). The data that support the findings of this study are available upon reasonable request from the corresponding author [MBM].

Results

We studied 26 subjects: 15 critically ill patients, including 5 with intracerebral hemorrhage and 10 with sepsis, and 11 healthy adults as controls. For the critically ill patients, mean age was 59 (±19), 7 (47%) were female and 12 (80%) were white, the median SOFA score was 4 [interquartile range 3, 11] and median GCS was 14 [10.5, 15]. Acute kidney injury occurred in 5 (33%), 4 (27%) were mechanically ventilated, and 8 (53%) received vasopressors. Comparing patients with sepsis to those with intracerebral hemorrhage, SOFA scores were higher (median 8.5 [4, 12.75] versus 2 [1, 4], p=0.035), age, sex and mechanical ventilation exposure were not significantly different (all p>0.6), and no patients with intracerebral hemorrhage had acute kidney injury or received vasopressors. For the health controls, mean age was 33 (±10) and 6 (55%) were female.

Individual Core Clock Gene Analyses

Gene expression in the healthy controls showed circadian rhythmicity, with significant cosinor rhythm fit in BMAL1, TIMELESS, CRY1, NR1D1 and PER1. Fit was consistently optimal among the core clock genes using a single component cosinor with 24 hour period. The percent rhythm, or percent of the variance in gene expression attributable to rhythmicity according to the cosinor fitted model, ranged from 54% to 88% (all p<0.01) for those five genes. CRY2 showed no clear signal of rhythmicity. In contrast to the findings in the healthy subject sample, no individual core clock gene demonstrated significant rhythmic expression in critically ill patients. There was no difference in gene amplitudes between patients with ICH and sepsis (all p>0.25). Figure 1 shows individual subject gene expression, empiric population-mean expression and population-mean cosinor fit along with cosinor model fit details. Among individual patients, the exploratory search for correlations between core clock gene amplitudes and illness severity (SOFA), encephalopathy (GCS), rest-activity rhythmicity (activity index amplitude) and melatonin rhythm strength (melatonin amplitude) identified no associations.

Figure 1:

24 Hour Expression of Core Clock Genes by Whole Blood RNA Sequencing

Transcriptome Rhythmicity Analysis with TimeSignature

Application of TimeSignature analysis to its validated set of 41 rhythmic genes showed that overall transcriptome expression rhythms were less temporally organized in the critically ill group compared to healthy subjects. The median absolute error (MAE) was 4.90 hours for the critically ill group and 1.48 hours for the healthy control group. Within individual critically ill patients, the TimeSignature MAE ranged from 0.93 hours to 8.93 hours, and a significant correlation was identified between encephalopathy severity by GCS and TimeSignature MAE (rho −0.54, p = 0.036). There was no correlation between TimeSignature MAE and type of critical illness (sepsis or intracerebral hemorrhage; p=0.25), SOFA score (rho 0.08, p=0.76), melatonin amplitude (rho −0.19, p=0.49) or activity index amplitude (rho −0.35, p=0.24). Figure 2 shows true versus TimeSignature-predicted sample times and the cumulative error density function for both groups, along with an example of a subject with high and low TimeSignature MAE to illustrate application of TimeSignature to individuals.

Figure 2:

Transcriptome Rhythmicity Phase Coherence by TimeSignature Analysis

Discussion

Within a day of critical illness onset, normal circadian rhythmicity was no longer detectable in expression profiles of individual core clock genes. We found no correlation between normalized core clock gene expression amplitude and disease severity, type of critical illness, melatonin rhythm amplitude or rest-activity rhythm amplitude. TimeSignature analysis of a broader sample of genes showed that overall temporal disorganization of the transcriptome was correlated with encephalopathy severity. These data indicate that peripheral transcriptome rhythms become desynchronized early in the course of severe illness, and encephalopathy severity is associated with desynchronization. These findings in peripheral tissue rhythms are consistent with recent work in which we identified encephalopathy severity as the principal risk factor for disruption of the central circadian rhythm as measured by melatonin secretion rhythms and rest-activity rhythms.(6, 7)

In addition to hormones like melatonin and cortisol, other humoral factors and autonomic signals from sympathetic and parasympathetic outflows communicate timing from the SCN to entrain rhythms in peripheral tissues.(17) Tracing techniques with selective denervation have shown that the SCN and hypothalamus connect with white and brown adipose tissue, the adrenal gland, heart, liver, ovary, kidneys, pancreas and other tissues by parasympathetic and sympathetic neurons, and substantial evidence has accumulated to connect unbalanced autonomic output to chronic diseases.(18) Without ongoing central inputs, individual cells in peripheral tissues drift out of phase with each other but remain individually rhythmic, which manifests in aggregate as a dampening of the transcriptome rhythmicity.(19) In contrast, some entraining signals can rapidly and potently affect the alignment of individual cellular clocks. For example, rat fibroblasts maintained in cell cultures for more than 25 years can be induced to synchronously express multiple core clock genes (e.g. PER1, PER2) for several days by brief exposure to animal serum.(20) The rapid dampening of gene expression rhythmicity we observed quickly after onset of critical illness implicates a direct response to autonomic and/or humoral factors, in contrast to the gradual decaying in phase coherence that is observed when central circadian signals are disconnected from normal signals, which can be simulated by tissue biopsy.(21) We found a correlation between peripheral transcriptome disruption and encephalopathy severity, but not with multisystem organ failure severity, which suggests that acute autonomic disturbances may be more culpable than humoral factors.

These perturbations in gene expression rhythms have immediate physiologic implications, a principle demonstrated in human muscle cells obtain by biopsy from healthy volunteers in which CLOCK function silenced by small interfering RNA quickly deranged lipid metabolism pathways, myokine secretion and insulin responses.(22) Many of the pathologies observed in critical illness, including glycemic dysregulation, muscle wasting and immune dysfunction, are inducible by disruption of circadian rhythms in translational experimental models, and may account for some of the secondary morbidity of critical illness.

There are limitations to these data. We sampled patients with a representative brain critical illness (ICH) and multisystem critical illness (sepsis) to explore two major categories of illness, but these findings may not fully generalize to all critical illnesses. The sample size is small, and therefore possibly insensitive to low amplitude residual rhythmicity in individual genes. Critically ill patients were older than the healthy controls and aging generally dampens absolute circadian rhythm amplitudes. We used gene expression measurements that were within-subject normalized, and significant confounding due to this effect is less likely on a patient-specific relative scale than when using absolute measurements. Limitations in methods that use single genes underscore strengths of the TimeSignature algorithm. First, its use of a broader set of genes makes accuracy less susceptible to natural variability in single gene expression. Second, it has been validated as accurate based on external clock time and by internal circadian time determined from melatonin phase, a crucial feature for populations such as this in whom internal circadian time markers like melatonin, temperature and cortisol are deranged and thus confounded, and clock time is a default surrogate.(23) In validation cohorts, the accuracy of TimeSignature is preserved with as few as two samples per subject, spaced 10–14 hours apart. TimeSignature may be a more feasible method to assess illness-related circadian derangement and recovery than melatonin profiling, which requires frequent blood sampling over 24 hours and can be altered by common exposures like light, caffeine, and beta blockers.

Conclusions

These data provide evidence that gene expression rhythms of core clock genes and the broader transcriptome rapidly become abnormal during critical illness. The association between disrupted transcriptome rhythms and encephalopathy, similar to the associations recently reported between melatonin rhythm disruption and rest-activity rhythm disruption and encephalopathy, suggests a path for future work to elucidate the underlying pathophysiology. TimeSignature analysis of peripheral blood gene expression profiles may be a novel and informative method to study circadian rhythms in critically ill patients.