Metabolomics pilot study identifies circadian desynchronization and 1 distinct intra-patient variability patterns in critical illness

25 Background: Synchronized circadian rhythms play a key role in coordinating 26 physiologic health. Desynchronized circadian rhythms may predispose 27 individuals to disease or be indicative of underlying disease. Intensive care unit 28 (ICU) patients likely experience desynchronized circadian rhythms due to 29 disruptive environmental conditions in the ICU and underlying pathophysiology. 30 This observational pilot study was undertaken to determine if circadian rhythms 31 are altered in ICU patients relative to healthy controls by profiling circadian 32 rhythms in vital signs and plasma metabolites. 33 Methods: We monitored circadian rhythms in 5 healthy controls and 5 ICU 34 patients for 24 hours. Heart rate and blood pressure were measured every 30 35 minutes, temperature was measured every hour, and blood was sampled for 36 mass spectrometry-based plasma metabolomics every 4 hours. Bedside sound 37 levels were measured every minute. Circadian rhythms were evaluated in vitals 38 and plasma metabolites individually and in each group using the cosinor method. 39 Results: ICU patient rooms were significantly louder than healthy controls’ 40 rooms and average noise levels were above EPA recommendations. Healthy 41 controls generally had significant circadian rhythms individually and as a group. 42 While a few ICU patients had significant circadian rhythms in isolated variables, 43 no significant rhythms were identified in ICU patients as a group, except in 44 cortisol. This indicates a lack of coherence in circadian phases and amplitudes 45 among ICU patients. Finally, principal component analysis of metabolic profiles 46 metabolomics data from 25,000 features to those that differ between ICU patients and healthy controls at the level of q<0.05 only reduced the data set to 15,000 features. These results indicate that ICU patients have highly disordered metabolism relative to healthy controls.

This observational pilot study was undertaken to determine if circadian rhythms 31 are altered in ICU patients relative to healthy controls by profiling circadian 32 rhythms in vital signs and plasma metabolites. 33 Methods: We monitored circadian rhythms in 5 healthy controls and 5 ICU 34 patients for 24 hours. Heart rate and blood pressure were measured every 30 35 minutes, temperature was measured every hour, and blood was sampled for 36 mass spectrometry-based plasma metabolomics every 4 hours. Bedside sound 37 levels were measured every minute. Circadian rhythms were evaluated in vitals 38 and plasma metabolites individually and in each group using the cosinor method. 39 Results: ICU patient rooms were significantly louder than healthy controls' 40 rooms and average noise levels were above EPA recommendations. Healthy 41 controls generally had significant circadian rhythms individually and as a group. 42 While a few ICU patients had significant circadian rhythms in isolated variables, 43 no significant rhythms were identified in ICU patients as a group, except in 44 cortisol. This indicates a lack of coherence in circadian phases and amplitudes 45 among ICU patients. Finally, principal component analysis of metabolic profiles 46 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint Introduction 58 59 Circadian rhythms span multiple levels of hierarchical organization in biological 60 systems, through molecular processes (peripheral oscillators) to whole-body 61 rhythms such as core body temperature (1). Synchronized, intact rhythms are 62 fundamental to human health and their desynchronization is now known to 63 contribute to morbidity and pathophysiology in otherwise healthy individuals (2-64 4). Critical care physicians increasingly recognize the influence of circadian 65 desynchronization on physiology and outcomes in critically ill patients (5, 6). 66 There is clear evidence that supporting circadian rhythms in cancer patients 67 contributes to higher quality of life (7). Likewise, it may be beneficial to support 68 circadian rhythms in intensive care unit (ICU) patients (8,9). 69 While there is evidence that circadian rhythms are disrupted in the 70 critically ill, the etiology is complex and multifaceted (10). A host of 71 desynchronizing influences are present in critical illness, including poor sleep 72 (11), continuously-administered parenteral nutrition (12-14), unregulated 73 light/dark cycles and sound levels (15-17), as well as the influence of 74 pharmacological treatments (18). 75 Due to the hierarchical and widely distributed organization of the circadian 76 system, multiple data types should be collected to determine the extent of 77 circadian desynchronization in critical illness regardless of the underlying cause. 78 High-throughput -omics data, particularly metabolomics, is suited to this 79 endeavor, since metabolism is tightly coupled to the core clock mechanism (19-80 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint 21). Multiple circadian metabolites have been reported in the literature, identified 81 with mass spectrometry-based metabolomics (22-24) and alterations to circadian 82 rhythms in the metabolome have been documented in shift work (25) and sleep 83 disruption protocols (26). 84 We collected preliminary data to evaluate the extent of circadian 85 desynchronization in ICU patients using mass spectrometry-based metabolomics 86 in this observational study. Secondary objectives were to evaluate circadian 87 desynchronization in vital signs and to measure sound levels in patient rooms. 88 We hypothesized that disruptions to circadian rhythms in ICU patients relative to 89 healthy controls would be identified in multiple metabolites and in vital signs. For the metabolomics portion of the study, blood samples (5 mL) were 141 drawn for metabolomics analysis every 4 hours beginning at 9:00 AM in each 142 participant. Plasma was extracted and aliquots were stored at -80°C until 143 preparation for analysis with mass spectrometry. Reagents were obtained from 144 Fisher Chemical Co., and were of LC-MS grade or better. Samples were 145 prepared according to published protocols (27, 28). Internal standards (see S1 146 Table) were added to thawed plasma samples followed by the addition of 4 147 volumes of cold (-80oC) 10% acetone 90% methanol (i.e. added 400µL solvent to 148 100µl sample). Samples were vortexed and incubated at -20oC for 15 minutes. 149 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint Samples were spun down at 13,000g for 10 minutes at 4oC and the supernatant 150 transferred to a clean tube. The incubation and centrifugation were repeated and 151 the final supernatant transferred to a clean tube, which was then dried under 152 nitrogen. Preparation for reverse-phase LC-MS analysis was completed by 153 adding a starting buffer (100 µl of 5% acetonitrile, 95% water, 0.1% formic acid). 154 Sample pH was adjusted to approximately 2 by adding 10-20 µl of 10% formic 155 acid. 156 Samples were analyzed via chromatographic separation in-line with mass 157 spectrometry. Ultra-high performance liquid chromatography (UHPLC) was 158 performed using the Thermo Scientific Ultimate 3000 UHPLC platform. For 159 reverse-phase analysis the instrument was fitted with a Waters Acquity BEH C-160 18 column (2.1x100 mm, 1.7 µm particle size). Flow rate was 0.4 mL/minute and 161 the column compartment was set to 40oC. Ten microliters were injected onto the 162 column. Elution solutions were A) water with 0.1% formic acid and B) acetonitrile 163 with 0.1% formic acid. The elution gradient was as follows: 2% B for 0.5 minutes; 164 increase to 25% B over the next 0.5 minutes; increase to 80% B over the next 7 165 minutes; increase to 100% B for 2 minutes; decrease to 2% B over the next 0.5 166 minutes; hold at 2% B for 2.5 minutes. 167 The Q Exactive™ Quadrupole-Orbitrap mass spectrometer (Thermo 168 Fisher Scientific, Waltham, MA) was employed for mass analysis. The 169 instrument profiles for both high mass accuracy (<2 ppm) and spectral resolution 170 (typically at 70,000 resolution in full scan mode). Analysis was performed in 171 positive mode over a mass range of 70-1050 m/z. 172 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint   Sound levels were significantly higher in ICU patients' rooms than in 247 healthy controls' rooms, particularly at night (Fig 1). Average minimum equivalent 248 continuous noise levels (LAeq) were 48 dB (ICU) vs. 33 dB (controls; p<0.0001). 249 Sound levels in ICU patient rooms were, on average, above recommended World 250 Health Organization limits of 35 dB (36). In two of the patients' rooms, the 251 minimum sound levels were quite loud: 43.5 and 54.9 dB, respectively. Mean 252 noise levels overlapped in ICU patient rooms and healthy controls' rooms from 253 9:00 to 13:30 and again at 9:00 the following morning. Noise levels remained 254 elevated in ICU patients' rooms even during nighttime hours while healthy 255 controls had a clear decrease in sound levels at night, particularly around 256 midnight. We noted a clear lack of a 24-hour pattern in ICU noise levels relative 257 to healthy controls. 258 259 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint  As a group, healthy controls had a significant (PMC p<0.007) 24-hour 285 rhythm in heart rate and diastolic blood pressure, and near-significant (PMC 286 p<0.091) rhythms in temperature and systolic blood pressure (Fig 2 and Table 2). 287 As a group, ICU patients had no significant 24-hour rhythms in the same 288 variables, even though some patients had a significant rhythm at the individual 289 level. This indicates a lack of coherence in the acrophases and amplitudes of 290 ICU patients in these variables. This can also be seen in Table 3, (Table 3 and S3 and S4 Figs) and none had a significant rhythm in the ICU 318 group. For brevity, we present information only on the 10 significant metabolites. 319 Significant 24-hour rhythms were identified in cortisol both individually and 320 at the group level in healthy controls and in ICU patients (Laboratory value, Table  321 2). Even though PMC showed that both the ICU group and the healthy control 322 group had a significant 24-hour rhythm in cortisol, patients' cortisol rhythms had 323 some abnormal features: they had a higher mean (20.5 vs. 6.9 g/dL, p=0.01) 324 than controls as well as scattered acrophases (Fig 3). Healthy controls had tightly 325 clustered acrophases (9:24 (8:36, 10:44)) while ICU patients did not (10:08 326 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity. is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint (6:04, 13:52); data are reported as acrophase (95% confidence interval)). The 327 scatter of phases in the ICU patient group leads to a significant lowering of 328 rhythm amplitude in the ICU patient group relative to the healthy control group as 329 quantified by PMC. For acetylcarnitine, aminoadipate, tyramine, and 2-345 hydroxylauroylcarnitine, neither patients nor controls have significant rhythms 346 when evaluated individually. We interpret this to mean that patients did not differ 347 much from healthy controls individually in these variables. However, PMC 348 showed the healthy control group had significant rhythms while ICU patient group 349 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint  (Fig 4) of the features identified by Progenesis QI shows that each 358 ICU patient's plasma samples (P07-P12) are strongly clustered together away 359 from healthy controls (P01-P06). Of the 5 ICU patients studied, 4 had ICU stays 360 of >20 days and 3 were discharged to long-term acute care facilities (Table 1). 361 The 5th patient had an 8-day stay and was discharged home. This patient's 362 cortisol rhythm was closest to the rhythms of healthy controls (Fig 3) and their 363 metabolic profiles (red circles) closest to the cluster of healthy controls (Fig 4).  , 4 hr, 8hr, 12hr, 16hr, 20hr). 372 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint injury and abnormalities, and cell death and survival (S7 Table). 389 390

391
In this pilot study of 5 ICU patients and 5 healthy controls, we postulated and 392 confirmed that patients' 24-hour rhythms differ from those of healthy controls. 393 Rhythms were monitored in heart rate, blood pressure, temperature, and 394 circulating plasma metabolites including cortisol. We hypothesized that fewer of 395 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint also been shown to intersect with the circadian system (52). 481 This pilot study had several limitations. We did not use a circadian 482 protocol with healthy controls or ICU patients. As such, we are unable to clearly 483 state whether the observed differences in 24-hour rhythms are due to 484 endogenous factors, exogenous factors, or both. This warrants further study. 485 Only 5 individuals per group were studied. Future work must include larger 486 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint groups. All 5 ICU patients had different underlying conditions. A more focused 487 study of patient groups with distinct underlying conditions (e.g. sepsis vs. stroke) 488 should be done to evaluate how individual diseases contribute to circadian 489 rhythm desynchronization and to observed patterns in the metabolomics data. 490 Core body temperature should be measured in all participants. The lack of 491 significant rhythm in temperature observed in healthy controls (p=0.091) may be 492 due to this. 493 Due to budgetary and logistical constraints, only positive mode reverse 494 phase MS1 data were collected. Future work should include negative mode 495 reverse phase and HILIC columns as well as MS2 data to aid in more complete 496 feature identification. The feature identification reported here is preliminary and 497 should be confirmed by targeted analyses in future work. Melatonin was not 498 conclusively identified in our mass spectrometry data and should be quantified by 499 mass spectrometry and/or radio immunoassay in future work. The IPA results 500 should be considered preliminary as well. Progenesis QI can assign more than 501 one HMDB ID to a feature, and IPA resolves duplicates by choosing the feature 502 with the lowest q-value. This may introduce biasing. 503 More frequent sampling of blood for metabolomics data would improve 504 profiling of circadian rhythms (53). The lack of statistical significance in individual 505 rhythms of metabolites, particularly in healthy controls, may be alleviated by 506 more data. Plasma samples could be collected at a higher frequency, or for a 507 longer period of time, or both. This supports our case that a larger study is 508 needed. 509 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint Finally, we recorded bedside light levels for all study participants. 510 However, our light meter had to be replaced halfway through the study and we 511 did not feel that the data from the two separate meters could be meaningfully 512 compared. We have opted to not present these data in the manuscript. Despite Author contributions:

537
• ERL was study PI. She contributed to study design, data analysis and 538 interpretation, and prepared the manuscript. 539 • JE and LP performed data analysis and assisted with manuscript 540 preparation and approved the final manuscript. 541 • SH obtained mass spectrometry data and contributed to manuscript 542 preparation and approved the final manuscript. 543 • MS and SM identified patients and collected data, and approved the final 544 manuscript. 545 • GB contributed to study design and data interpretation, and approved the 546 final manuscript. 547 • GC-G contributed to study design, data analysis and interpretation, 548 manuscript preparation, and approved the final manuscript. 549 550 551 References 552 . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint  . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint . https://doi.org/10.1101/2019.12.09.19014225 doi: medRxiv preprint . CC-BY-NC 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
is the (which was not peer-reviewed) The copyright holder for this preprint     Principal components-Circadian Rhythms Pilot All MS Features