Metabolic Phenotyping of Traumatized Patients ... - ACS Publications

Article pubs.acs.org/ac

Metabolic Phenotyping of Traumatized Patients Reveals a Susceptibility to Sepsis Benjamin J. Blaise,*,†,‡ Aurélie Gouel-Chéron,† Bernard Floccard,† Guillaume Monneret,§ and Bernard Allaouchiche† †

Hospices Civils de Lyon, Service de réanimation, Hôpital Edouard Herriot, 5 place d’Arsonval, 69437 Lyon Cedex 03, France Hospices Civils de Lyon, Service de néonatalogie et réanimation néonatale, Hôpital Femme Mère Enfant, 56 boulevard Pinel, 69500 Bron, France § Hospices Civils de Lyon, Laboratoire d’immunologie cellulaire, Hôpital Edouard Herriot, 5 place d’Arsonval, 69437 Lyon Cedex 03, France ‡

ABSTRACT: Sepsis is one of the leading causes of morbidity and mortality in patients admitted in intensive care units (ICU) for trauma. The identification of biochemical mechanisms and prediction of patients at risk of early sepsis remain unsolved. Metabolic phenotyping allows the recovery of coordinated metabolic concentration variations. There are no predictive metabolic phenotyping studies based on noninvasive human samples to identify the later development of sepsis in traumatized patients. The aim of this study was to investigate whether the metabolic phenotype could help in the discrimination of patients according to the later development of sepsis. Plasma samples were taken from severely injured patients in the hours following their admission in the ICU. Nuclear magnetic resonance (NMR) based metabolic phenotyping was performed on this prospective cohort. Statistical analyses were run on NMR spectra to discriminate patients according to the later development of sepsis. Twenty-two patients were included. One was excluded because of aberrant metabolic phenotype. Orthogonal partial least-squares analysis allowed the recovery of a predictive metabolic phenotype identifying patients with a later development of sepsis (1 + 4 component model, R2 = 0.855, Q2 = 0.384). A cross-validated receiver operator characteristic curve showed a remarkable prediction capacity (AUC = 0.778). Eight metabolic hotspots were identified. NMR-based metabolic phenotyping allows the prediction of patients at high risk of early sepsis after ICU admission for trauma. A larger cohort is necessary to validate and complete this study, understand biochemical mechanisms promoting sepsis development, and identify patients at risk.

I

sciences, might prove to be of interest to decipher biochemical mechanisms occurring after initial injury. Metabolic phenotyping is one of the youngest postgenomic sciences, aiming at the identification of variations in low molecular weight compounds, in response to pathophysiological events, drug treatments, or genetic modifications.4,5 It is a top-down hypothesis-free approach based on analytical technologies, such as nuclear magnetic resonance (NMR), quantifying metabolites in parallel and relying on multivariate and univariate statistical analyses to discriminate samples according to subtle and coordinated metabolic changes, called metabolic phenotype. Furthermore various samples can be used such as biofluids (plasma, serum, urine, or cerebro-spinal fluid), or semi solid samples such as biopsies. Statistical analyses can then identify significant molecular perturbations within the comprehensive metabolic network and give insights of the reorganization in response to different stimuli to restore the homeostasis, at the systems biology level.6−8 Metabolic

njury is the major cause of death worldwide in patients under 40 years of age, according to the World Health Organization and national public health institutes.1 For those surviving initial resuscitation, mortality and morbidity may occur during follow-up in the intensive care unit (ICU) because of various complications, especially sepsis. Injuries or stresses are risk factors of sepsis, likely based on the immunomodulation they induce. The initial systemic inflammatory response syndrome (SIRS) is often countered by an antiinflammatory response, which maintains inflammatory immune homeostasis albeit, exposes the organism to communityacquired and nosocomial pathogens by lowering the host defenses.2 Despite progresses achieved in understanding this immunosuppression phenomenon, particularly the diminished expression of human leukocyte antigen DR on circulating monocytes,3 the identification of predictive biomarkers and comprehensive description of this susceptibility state for infections remain unsolved. The early identification of risk factors for sepsis development after the admission in ICUs for trauma is thus a challenge. New approaches, based on analytical © 2013 American Chemical Society

Received: July 19, 2013 Accepted: October 17, 2013 Published: October 21, 2013 10850

dx.doi.org/10.1021/ac402235q | Anal. Chem. 2013, 85, 10850−10855

Analytical Chemistry

Article

were observed during NMR analysis. For metabolic phenotyping, samples were prepared as previously described by Beckonert et al.22 Serum samples were thawed at room temperature before use. 200 μL of each was diluted with 400 μL of a 0.9% saline solution (NaCl 0.9%, D2O 10%) in a microtube, then centrifuged for 5 min at 4 °C at 12 000 g. Finally, 550 μL of supernatant was transferred into 5 mm NMR tubes. Samples were kept at 4 °C until analysis. 1 H NMR Spectroscopy. All NMR experiments were carried out on a Bruker Avance III spectrometer operating at 800.14 MHz (proton resonance frequency) equipped with a 5 mm TXI probe, and high throughput sample changer that maintained the samples temperature at 4 °C until actual NMR acquisition. The temperature was then regulated at 300K throughout the NMR experiments. Standard 1H 1D NOESY NMR pulse sequence,23 with water presaturation, was applied on each sample to obtain corresponding metabolic profiles. One hundred twenty-eight transient free induction decays (FID) were collected for each experiment with a spectral width of 20 ppm, for an acquisition time of 1.36s. The relaxation delay was set to 2s. The NOESY mixing time was set to 100 ms. The 90° pulse length was automatically calibrated for each sample at around 9.25 μs. In addition, 2D NMR experiments (1H−13C HSQC, 1H−1H TOCSY, and J-Resolved) were recorded for structural assignment of NMR metabolic signals. Data Processing. All FIDs were multiplied by an exponential function corresponding to a 0.3 Hz line broadening factor, prior Fourier transformation. 1H NMR spectra were automatically phased and referenced to the α-glucose anomeric proton signal (δ = 5.23 ppm) using Topspin 2.1 (Bruker GmbH, Rheinstetten, Germany). Residual water signal (4.68 to 4.86 ppm) was excluded. Spectra were divided into 0.001 ppmwide buckets over the chemical shift range [0; 10 ppm] using the AMIX software (Bruker GmbH). Spectra were normalized to their total intensity and Pareto-scaled or mean-centered prior to analysis. Statistical Analysis. Patients were classified into 2 groups according to the later development of sepsis. Groups were compared using the Mann−Whitney U test for continuous nonparametric variables and the chi2 test for categorical data. Data were exported to SIMCA−P 12 (Umetrics, Umea, Sweden) and MATLAB (Mathworks, Natick, MA) for statistical analysis. Principal component analysis (PCA) was performed to derive the main sources of variance within the data set, check population homogeneity and eventually identify outliers. Data were visualized as score plots, where each point represents a single sample on the main principal components and as loading plots, which represent the coordinated variations of NMR spectral regions. Supervised regression methods such as Orthogonal Partial Least-Squares24 (O-PLS) were performed to establish a robust sample classification model. The O-PLS analysis was run to discriminate populations by regressing a supplementary data matrix Y, containing information about the development of sepsis, on the X NMR data set matrix. Model performances were assessed by goodness-of-fit parameters R2 and Q2, related respectively to the explained and predicted variance. We performed model validation by resampling the model 1000 times under the null hypothesis. Model classification performance was characterized by a receiver operating characteristic (ROC) curve and its area under the curve (AUC). Statistical recoupling of variables25,26 (SRV) algorithm was used to identify metabolic NMR variables.

phenotyping has been successfully used in a wide range of medical applications.9−11 A special focus has recently been given on anesthesiology and intensive care medicine,12−15 where these approaches could help physicians monitoring patients undergoing surgery. Several metabolic phenotyping studies, based on NMR spectroscopy or mass spectrometry, have investigated sepsis or trauma. However none of them include human samples collected in an ICU, but on the contrary deal with animal models and experimentally induced sepsis or trauma.16−19 Mao et al. showed the ability of 1H NMR to discriminate control patients from patients with SIRS or multiple organ dysfunction syndrome.20 The objectives of our study were to validate the feasibility of 1 H NMR spectroscopy in ICU patients and to investigate whether the metabolic phenotype could help in the discrimination of patients according to the later development of sepsis.

■

METHODS Patient Inclusion. This work belongs to a global study on ICU-induced immune dysfunctions. This prospective observational study was carried out over a 15-month period. The protocol was reviewed by the institutional ethics committee, which waived the need for informed consent because the study was observational and involved sampling of very small quantities of blood. Ten samples belonging to the later development of sepsis group were randomly selected from this cohort and matched with 12 control patients. Samples were collected from residual blood after completion of routine follow-up. Inclusion criteria were an Injury Severity Score (ISS) of more than 25 and admission to the ICU. Clinical exclusion criteria were age of less than 18 years, ISS of less than 25, inhalation pneumonia or gut perforation during trauma, chronic corticosteroid therapy, and death in the first 48 h after admission. All patients included were followed up prospectively until day 14 by daily clinical examination and blood tests. Clinical data included demographic, infection, therapeutics, duration of hospitalization, and clinical scores (ISS, Simplified Acute Physiology Score or SAPS II). Sepsis Definition. The American College of Chest Physicians/Society of Critical Care Medicine Consensus Conference definition of sepsis was used for this study,21 namely, the presence of an identifiable site of infection and evidence of SIRS on the basis of at least two of the following criteria: (a) body temperature of more than 38 °C or of less than 36 °C, (b) heart rate of more than 90 beats per minute, (c) respiratory rate of more than 20 breaths per minute or hyperventilation as indicated by an arterial partial pressure of carbon dioxide (PaCO2) of less than 32 mmHg (less than 4.3 kPa), and (d) a white blood cell count of more than 12 000 cells/mm3 or of less than 4000 cells/mm3 or the presence of more than 10% immature neutrophils. The onset of sepsis was defined, as recommended by the Consensus Conference, as the day on which the site of infection was identified. The final diagnosis of sepsis was retrospectively established by two experts assessing the complete medical data and not involved in case management. Blood Sampling and Preparation. Ethylenediaminetetraacetic acid (EDTA)-anticoagulated blood samples were collected at 8 a.m. every 2 days after injury. Blood samples were immediately stored at 4 °C and then at −80 °C after plasma extraction. Samples were collected on EDTA because of the need for complementary analysis. No additional signals 10851



Article

Data-Driven Sample Size Determination. We used a recently developed approach allowing the determination of sample size to identify statistically significant variations of metabolic NMR signals. Methodology is described elsewhere in Analytical Chemistry.27 This approach is based on the simulation of NMR spectra based on the evaluation of each SRV variable distribution using kernel density estimator. The cumulative probability is then inversed. Random numbers are finally selected in the ]0,1[ interval to build simulated NMR spectra. It is then possible to consider simulated NMR data sets of variable sizes and to measure the number of statistically significant variables for each data set size. The sample size determination is then based on the identification of the smallest data set size allowing the identification of a maximum of statistically significant NMR variables (metabolic exploration) or at least one statistically significant variable (biomarker discovery).

later development of sepsis or the absence of later sepsis. The O-PLS score plot shows a clear discrimination between the two groups, as presented in figure 2a, assessed by elevated values of goodness-of-fit parameters, R2 = 0.855 and Q2 = 0.384, referring to the explanation and prediction capacities of the established model. Coordinated metabolic variations sustaining this discrimination are represented by the O-PLS regression coefficient plot (Figure 2b) colored according to the correlation between metabolic NMR signals and the class information matrix. Metabolic NMR signals were identified based on the SRV algorithm, defined as an automated variable size bucketing procedure using the covariance/correlation relationship between consecutive NMR variables. The robustness of the model was validated by resampling under the null hypothesis, showing an overall decrease of goodness-of-fit parameters R2 and Q2 with the correlation between the original and permuted class information matrices necessary for supervised multivariate analyses (Figure 2c). However this tendency is not fully validated, since statistical significance was out of reach for this proof-of concept study. A few R2 and Q2, associated with random class information matrices, overpass the model goodness-of-fit parameters showing that these results could not have been achieved with a reduced population. Finally we characterized the model classification performances with a cross-validated receiver operating characteristic curve, showing an AUC of 0.778 (Figure 2d). Assignment of Metabolic Phenotype Associated with the Risk of Sepsis Development in Traumatized Patients. On the basis of correlation coefficients between the NMR data set and the class information matrix Y, one can identify hot spots on the metabolic phenotype. Eight NMR signals present high level of correlation (>0.6): 2.27 (multiplet = valine), 2.31 (multiplet = beta-hydroxybutyrate), 2.59 (multiplet = citrate), 2.66 (doublet = aspartate), 2.81 (doublet = aspartate), 5.42 (singlet = allantoin), 5.54, and 5.67 ppm (multiplets = unknowns). Sample Size Determination. A simulated data set of 400 spectra was generated. A sample size of 200 spectra was identified as sufficient to identify at least one statistically significant NMR signal variation (biomarker discovery). A cohort of this size is being recruited in our department to identify prospective candidate biomarkers among traumatized patients admitted to the intensive care unit to discriminate those with a later development of sepsis from those that will not develop sepsis.

■

RESULTS Patient Characteristics. We included 22 patients from the global cohort, based on chronological sample availability and median delay of first sample after trauma (to obtain a homogeneous cohort). Ten of them developed sepsis during ICU management (median delay = 4.8 days; 8 pulmonary, 1 abdominal, and 1 urinary initial site of infection). No significant differences were detected among all recorded characteristics (demographic, infection, therapeutics, length of stay in the ICU, clinical scores), except for age and duration of hospitalization (longer stay in the sepsis group) as shown in Table 1. Supervised analyses of NMR data were not able to discriminate our samples according to the age factor, demonstrating its absence of influence on the results. Table 1. Major Clinical Characteristics of Patientsa global (n = 21) age, years weight, kg ISS SAPS II length of stay in ICU, days

42 (26−52) 75 (64−78) 40 (32−50) 41 (30−59) 9 (4−14)

sepsis (n = 9) 25 75 50 41 18

(24−34) (61−76) (34−58) (30−62) (12−32)

no sepsis (n = 12)

p-value

50 (39−55) 72 (65−81) 34 (30−39) 40 (29−58) 6 (3−8)

0.006 0.5 0.09 0.8 0.002

a

SAPS II: Simplified acute physiology score. ISS: Injury severity score. ICU: Intensive care unit. Values are expressed as median (interquartile range).

■

DISCUSSION Although it is known that injuries are a major risk factor for sepsis development, the pathophysiological mechanisms remain unsolved and the identification of early biomarkers a challenge. Here we investigated a prospective and observational human cohort aiming at the identification of early biomarkers of sepsis in traumatized patients after their admission in the ICU. 1H NMR spectroscopy of blood plasma, collected in the hours following admission and before the identification of any routine clinical or biological signs of sepsis, was used as a metabolic probe. We used metabolic phenotyping and established a statistical model discriminating traumatized patients admitted in the ICU according to the later development of sepsis. We identified important variations concerning 8 NMR metabolic signals (valine, citrate, aspartate, allantoin and hydroxybutyrate notably) associated with the later development of sepsis.

NMR Spectroscopy of Plasma Samples. A typical 1H 800 MHz 1D NOESY spectrum is presented in Figure 1. Narrow peaks belong to small molecular weight compounds (metabolites) and are superimposed on a protein background. A tentative assignment based on 2D (1H−1H and 1H−13C) correlation experiments has been achieved. One sample from the sepsis group appeared as an outlier in the PCA analysis. This sample showed coordinated variations of glucose (decrease) and lactate (increase) concentrations, compared to the other samples. This pattern is often observed when a microbial degradation occurs due to improper storage or handling of the samples. The corresponding sample was eventually excluded of all analyses. Metabolic Phenotype Assessing a Risk of Sepsis in Traumatized Patients. A 1 + 4 supervised O-PLS analysis was carried out to discriminate the 21 samples according to the 10852



Article

Figure 1. Typical 1H NMR NOESY plasma spectrum acquired on a Bruker Avance III spectrometer operating at 800.14 MHz (proton resonance frequency), equipped with a fully automated and cooling sample changer and an automated TXI probe (tuning, matching, locking, presaturation calibration and shimming). Experimental setup includes a temperature regulated at 300 K, an acquisition time set at 10 min.

each group) with homogenized characteristics. We observed that the patients were younger in the septic group without clear explanation given the fact that immunity usually decreases with age. Once again, this effect was not significant and is the result of the chronological inclusion of patients in our cohort. This will be canceled out in our future confirmation cohort. From a medical point of view, the interpretation of these encouraging results is not straightforward. The first limit to discuss is the small size of the cohort, which prevents the identification of candidate biomarkers. This limitation will be canceled out by the analysis of a larger cohort. A second limitation is the biological meaning of the observed discrimination. Although we were able to discriminate traumatized patients who will later develop a sepsis from those who will not, it is difficult to state if these metabolic variations are connected to an early stage of sepsis, the syndrome of systemic inflammatory response observed in traumatized patients or an immunomodulation caused by the trauma that will lead to sepsis. These considerations are beyond the scope of the present study, however the metabolic exploration based on a larger cohort will hopefully improve our understanding of the pathophysiology of trauma-induced sepsis and maybe help physicians in the management of traumatized patients in the ICU (early antibiotic prescription, targeted immunotherapy, intensive biological follow-up...)

Interestingly Kinross et al. already highlighted these metabolites in their experimental laparotomy model of surgical trauma,15 as well as Lin et al.17 regarding the hydroxybutyrate, both studies being based on NMR spectroscopy. It is not possible to determine if these variations are connected to a pre septical status that will later lead to sepsis or linked to a pro inflammatory process driving to a lowering of the host immune defense and thus favoring the sepsis development. The tendency observed in this study, is not fully validated, since statistical significance was out of reach for this proof-of concept study with reduced sample availability. Nevertheless, the crossvalidated ROC curve shows a remarkable prediction capacity (AUC = 0.778). We have thus established a statistical model discriminating traumatized patients admitted in the ICU according to the later development of sepsis, based on the quantification of metabolite concentrations in blood plasma samples. Highfield 1H NMR spectroscopy allows the parallel quantification of plasmatic metabolites, whose coordinated variations identified by multivariate statistical analyses, such as O-PLS, lead to a metabolic phenotype associated with the later development of sepsis. These are the first results concerning trauma-induced sepsis in human patients admitted in the ICU, based on a cohort recruited in our department. Compared to previous animal model studies, we included a similar number of samples (10 in 10853



Article

Figure 2. Metabolic phenotyping of sepsis susceptibility in traumatized patients. (a) O-PLS score plot discriminating plasma samples, taken at the admission in the intensive care unit, from traumatized patient who will later develop sepsis (red) from those who will remain away from septic complications (black). (b) O-PLS regression coefficient plot colored according to the correlation between the metabolic NMR data and the class information matrix encoded the membership of each sample to the 2 groups under study for the purpose of supervised multivariate statistical analyses. (c) Corresponding model validation by resampling under the null hypothesis. (d) Cross-validated receiver operating characteristic (ROC) curve showing the prediction capacity of the model, with an area under the curve of 0.778.

Notes

In conclusion, we have isolated an early metabolic phenotype associated with the later development of sepsis in traumatized human patients at their admission in the ICU. The crossvalidated ROC curve shows a remarkable prediction capacity. The model and hot spots could hopefully be confirmed and strengthened by a future larger cohort aiming at the establishment of a robust predictive metabolic phenotype. This step could be followed by the recovery of statistically significant candidate biomarkers and a perturbed metabolic network investigating the metabolic changes induced by sepsis development at the systems biology level. These results might help physician identifying traumatized patients at risk of sepsis and improving sepsis management in traumatized ICU patients.

■

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We would like to thank the TGIR−RMN and the Centre de RMN à Très Hauts Champs for granting access to the 800 MHz spectrometer, and more particularly Anne Fages and Dr Bénédicte Elena-Herrmann for NMR scientific support. We also thank Anne Portier and Catherine Jouvene-Faure for technical assistance.

■

REFERENCES

(1) Belanger, F.; Ung, A.-B.; Jougla, E.; Thélot, B.; Bene, M.; Bruzzone, S.; Denissov, G.; England, K.; Gjersten, F.; Nectoux, M.; Steiner, M.; Frimodt-Moller, B. Analysis of Injury Related Mortality in Europe, The ANAMORT project Final implementation report; Institute deVeille Sanitare: Saint-Maurice, France, 2008. (2) Bone, R. Ann. Intern. Med. 1996, 125, 680−687. (3) Cheron, A.; Floccard, B.; Allaouchiche, B.; Guignant, C.; Poitevin, F.; Malcus, C.; Crozon, J.; Faure, A.; Guillaume, C.; Marcotte, G.; Vulliez, A.; Monneuse, O.; Monneret, G. Crit. Care 2010, 14, R208. (4) Nicholson, J. K.; Lindon, J. C.; Holmes, E. Xenobiot. Fate Foreign Compd. Biol. Syst. 1999, 29, 1181−1189.

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Author Contributions

B.J.B. and A.G.-C. contributed equally to this work. B.J.B., A.G.C., B.F., G.M., and BA designed the study. A.G.-C. collected the samples. B.J.B. performed NMR analysis. B.J.B. and A.G.-C. performed statistical analysis. B.J.B., A.G.-C., B.F., G.M., and B.A. wrote the paper. 10854



Article

(5) Blaise, B. J.; Giacomotto, J.; Elena, B.; Dumas, M.-E.; Toulhoat, P.; Segalat, L.; Emsley, L. Proc. Natl. Acad. Sci. 2007, 104, 19808− 19812. (6) Blaise, B. J.; Navratil, V.; Domange, C.; Shintu, L.; Dumas, M.-E.; Elena-Herrmann, B.; Emsley, L.; Toulhoat, P. J. Proteome Res. 2010, 9, 4513−4520. (7) Blaise, B. J.; Navratil, V.; Emsley, L.; Toulhoat, P. J. Proteome Res. 2011, 10, 4342−4348. (8) Cazier, J.-B.; Kaisaki, P. J.; Argoud, K.; Blaise, B. J.; Veselkov, K.; Ebbels, T. M. D.; Tsang, T.; Wang, Y.; Bihoreau, M.-T.; Mitchell, S. C.; Holmes, E. C.; Lindon, J. C.; Scott, J.; Nicholson, J. K.; Dumas, M.E.; Gauguier, D. J. Proteome Res. 2012, 11, 631−642. (9) Holmes, E.; Loo, R. L.; Stamler, J.; Bictash, M.; Yap, I. K. S.; Chan, Q.; Ebbels, T.; De Iorio, M.; Brown, I. J.; Veselkov, K. A.; Daviglus, M. L.; Kesteloot, H.; Ueshima, H.; Zhao, L.; Nicholson, J. K.; Elliott, P. Nature 2008, 453, 396−400. (10) Nicholson, J. K.; Holmes, E.; Kinross, J.; Burcelin, R.; Gibson, G.; Jia, W.; Pettersson, S. Science 2012, 336, 1262−1267. (11) Lindon, J. C.; Holmes, E.; Nicholson, J. K. Pharm. Res. 2006, 23, 1075−1088. (12) Muirhead, L. J.; Kinross, J.; FitzMaurice, T. S.; Takats, Z.; Darzi, A.; Nicholson, J. K. Pers. Med. 2012, 9, 593−608. (13) Kinross, J. M.; Holmes, E.; Darzi, A. W.; Nicholson, J. K. Lancet 2011, 377, 1817−1819. (14) Mirnezami, R.; Kinross, J. M.; Vorkas, P. A.; Goldin, R.; Holmes, E.; Nicholson, J.; Darzi, A. Ann. Surg. 2012, 255, 881−889. (15) Nicholson, J. K.; Holmes, E.; Kinross, J. M.; Darzi, A. W.; Takats, Z.; Lindon, J. C. Nature 2012, 491, 384−392. (16) Liu, X.-R.; Zheng, X.-F.; Ji, S.-Z.; Lv, Y.; Zheng, D.-Y.; Xia, Z.-F.; Zhang, W.-D. Burns 2010, 36, 992−998. (17) Kinross, J. M.; Alkhamesi, N.; Barton, R. H.; Silk, D. B.; Yap, I. K. S.; Darzi, A. W.; Holmes, E.; Nicholson, J. K. J. Proteome Res. 2011, 10, 277−287. (18) Xu, P.; Lin, Z.; Meng, H.; Yan, S.; Yang, Y.; Liu, X.; Li, J.; Deng, X.; Zhang, W. J. Infect. 2008, 56, 474−481. (19) Lexcen, D. R.; Lusczek, E. R.; Witowski, N. E.; Mulier, K. E.; Beilman, G. J. J. Trauma Acute Care Surg. 2012, 73, S147−S155. (20) Mao, H.; Wang, H.; Wang, B.; Liu, X.; Gao, H.; Xu, M.; Zhao, H.; Deng, X.; Lin, D. J. Proteome Res. 2009, 8, 5423−5430. (21) Bone, R.; Balk, R.; Cerra, F.; Dellinger, R.; Fein, A.; Knaus, W.; Schein, R.; Sibbald, W.; Abrams, J.; Bernard, G.; Biondi, J.; Calvin, J.; Demling, R.; Fahey, PJ; Fisher, C.; Franklin, C.; Gorelick, K.; Kelley, M.; Maki, D.; Marshall, J.; Merrill, W.; Pribble, J.; Rackow, E.; Rodell, T.; Sheagren, J.; Silver, M.; Sprung, C.; Straube, R.; Tobin, M.; Trenholme, G.; Wagner, D.; Webb, C.; Wherry, J.; Wiedemann, H.; Wortel, C. Crit. Care Med. 1992, 20, 864−874. (22) Beckonert, O.; Keun, H. C.; Ebbels, T. M. D.; Bundy, J.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Nat. Protoc. 2007, 2, 2692− 2703. (23) Jeener, J.; Meier, B. H.; Bachmann, P.; Ernst, R. R. J. Chem. Phys. 1979, 71, 4546. (24) Trygg, J.; Wold, S. J. Chemom. 2002, 16, 119−128. (25) Blaise, B. J.; Shintu, L.; Elena, B.; Emsley, L.; Dumas, M.-E.; Toulhoat, P. Anal. Chem. 2009, 81, 6242−6251. (26) Navratil, V.; Pontoizeau, C.; Billoir, E.; Blaise, B. J. Bioinformatics 2013, 29, 1348−1349. (27) Blaise, B. J. Anal. Chem. 2013, 85, 8943−8950.

10855


Metabolic Phenotyping of Traumatized Patients ... - ACS Publications

Recommend Documents