ClinCirc identifies alterations of the circadian peripheral oscillator in critical care patients

Abstract

BackgroundAssessing circadian rhythmicity from infrequently sampled data is challenging; however, these types of data are often encountered when measuring circadian transcripts in hospitalized patients.MethodsWe present ClinCirc. This method combines 2 existing mathematical methods (Lomb-Scargle periodogram and cosinor) sequentially and is designed to measure circadian oscillations from infrequently sampled clinical data. The accuracy of this method was compared against 9 other methods using simulated and frequently sampled biological data. ClinCirc was then evaluated in 13 intensive care unit (ICU) patients as well as in a separate cohort of 29 kidney-transplant recipients. Finally, the consequences of circadian alterations were investigated in a retrospective cohort of 726 kidney-transplant recipients.ResultsClinCirc had comparable performance to existing methods for analyzing simulated data or clock transcript expression of healthy volunteers. It had improved accuracy compared with the cosinor method in evaluating circadian parameters in PER2:luc cell lines. In ICU patients, it was the only method investigated to suggest that loss of circadian oscillations in the peripheral oscillator was associated with inflammation, a feature widely reported in animal models. Additionally, ClinCirc was able to detect other circadian alterations, including a phase shift following kidney transplantation that was associated with the administration of glucocorticoids. This phase shift could explain why a significant complication of kidney transplantation (delayed graft dysfunction) oscillates according to the time of day kidney transplantation is performed.ConclusionClinCirc analysis of the peripheral oscillator reveals important clinical associations in hospitalized patients.FundingUK Research and Innovation (UKRI), National Institute of Health Research (NIHR), Engineering and Physical Sciences Research Council (EPSRC), National Institute on Academic Anaesthesia (NIAA), Asthma+Lung UK, Kidneys for Life.

Keywords: Anesthesiology; Inflammation; Organ transplantation; Translation; Transplantation.

Figure 1. Performance of ClinCirc in detecting circadian rhythmicity.

(A) Flow diagram depicting how cosinor analysis was combined with L-SP to create ClinCirc. (B) Sampling periods of 24 hours and (C) 48 hours were used to characterize the sensitivity and specificity of 10 mathematical methods on infrequently (every 4 hours) sampled data. Sensitivity was calculated using a waveform created following the addition of 40% noise (Supplemental Methods) to a sinusoidal wave. Specificity was calculated from a straight line (5,000 simulations). Data are represented as mean ± SD. (D) Two data sets measuring clock-gene expression in healthy human volunteers were reanalyzed using the same 10 mathematical methods. The average number of clock genes per volunteer that each method detected as having a circadian rhythm is shown. Data are represented as mean ± SEM. (E) ClinCirc was used to evaluate circadian rhythmicity of clock genes in ICU patients (n = 13) as well as in healthy volunteers (n = 23). Bar chart shows the proportion of subjects in which ClinCirc detected a circadian oscillation per clock gene. Heatmap shows which subjects had a detected circadian oscillation. Each subject is represented by a row, and a filled square represents a detected circadian oscillation. (F) ClinCirc was used to evaluate circadian rhythmicity of clock genes in kidney-transplant recipients at 0 to 24 hours after transplantation (n = 22; 24 hours after transplant) or 48 to 72 hours after transplantation (n = 7; 72 hours after transplant) as well as in healthy volunteers (n = 23). Bar chart shows the proportion of subjects that had a circadian oscillation for the clock gene. Heatmap shows which subjects had the circadian oscillation. Each subject is represented by a row in which the filled squares represent a detected circadian oscillation. *P < 0.05, χ² test.

Figure 2. Inflammation is associated with reduced detection of circadian oscillations in the ICU.

ICU patients were split into 2 groups based on whether ClinCirc detected unchanged or a reduced number of circadian oscillations in the peripheral blood molecular oscillator. (A) Heatmap displaying the proportion of patients in each group in whom ClinCirc detected a circadian oscillation in the measured inflammatory mediator. *P < 0.05, χ² test. (B) Forty-eight–hour expression profiles for the 3 inflammatory mediators (MMP2, MMP3, and TSLP) that showed differential circadian oscillations between ICU patients who had standard or reduced detection of circadian oscillations. Data are represented as mean ± SEM. Traces were acrophase aligned. (C) Volcano plot showing differences in mean expression of 37 inflammatory mediators between ICU patients with standard or reduced detection of circadian oscillations. Positive fold change reveals that the mediator was elevated in patients in whom detection of circadian oscillations was reduced. Dotted line shows P = 0.05. Significant cytokines are labeled. Inflammatory mediators that were differentially regulated are IFN-α2, IFN-γ, IL-2, IL-8, IL-10, IL-11, IL-12 (p40), IL-19, IL-20, IL-26, IL-27 (p28), IL-28A, IL-29, IL-35, LIGHT, Pentraxin-3, and TSLP. (D) Difference in CRP expression between ICU patients grouped according to the detection of circadian oscillations and those who underwent kidney transplantation. **P < 0.01, ANOVA post hoc Tukey’s. (E) CRP expression in ICU patients was also plotted against the number of clock genes for that participant in which ClinCirc detected the presence of a circadian oscillation. r² = 0.49 linear correlation. n = 13. For circadian rhythm analysis, data were only plotted if a circadian rhythm was detected.

Figure 3. Kidney transplantation induces a phase shift that is associated with altered clinical outcomes.

(A) The relative amplitude of each clock gene’s circadian oscillation 0 to 24 hours after transplantation (24 hours, n = 22) and 48 to 72 hours (72 hours, n = 7) after kidney transplantation was compared with the amplitude from healthy volunteers. Data are represented as mean ± SEM. **q < 0.01, 2-way ANOVA, post hoc Dunnett’s. Each circle indicates patient. (B) Acrophase plot of PER3’s circadian oscillation 0 to 24 hours and 48 to 72 hours after kidney transplantation was compared with that of healthy volunteers. Circles indicate individual patients. Data are represented as median ± IQR (color band). (C) PER3 transcript expression plotted against either time of day or time after allograft reperfusion for the first 24 hours following transplantation (n = 22). Circles indicate individual patients. Regression line is shown. (D) The acrophase of PER3’s circadian oscillation was compared against time of organ reperfusion immediately after kidney transplantation (24 hours). Circles indicate individual patients. r² = 0.87 linear regression. (E) The prevalence of DGF after kidney transplantation from brain-dead donors was calculated in a 10-year retrospective cohort (n = 536). The probability density for DGF was then plotted against allograft reperfusion time. Data are represented as mean ± 95%CI. Gaussian smoothing with bootstrap. (F) The probability density of DGF was also plotted against allograft harvest time from the donor. Data are represented as mean ± 95%CI. Gaussian smoothing with bootstrap. For circadian rhythm analysis, data were only plotted if a circadian rhythm was detected. Black dotted lines show uniform lines.