Chemometric Strategy for Modeling Metabolic ... - ACS Publications

Oct 29, 2010 - Biological Space along the Gastrointestinal Tract and Assessing Microbial Influences. Franc¸ ois-Pierre J. Martin,* Ivan Montoliu, Sun...
0 downloads 0 Views 2MB Size
Anal. Chem. 2010, 82, 9803–9811

Chemometric Strategy for Modeling Metabolic Biological Space along the Gastrointestinal Tract and Assessing Microbial Influences Franc¸ois-Pierre J. Martin,* Ivan Montoliu, Sunil Kochhar, and Serge Rezzi Nestle´ Research Center, P.O. Box 44, Vers-chez-les-Blanc, CH-1000 Lausanne 26, Switzerland Over the past decade, the analysis of metabolic data with advanced chemometric techniques has offered the potential to explore functional relationships among biological compartments in relation to the structure and function of the intestine. However, the employed methodologies, generally based on regression modeling techniques, have given emphasis to region-specific metabolic patterns, while providing only limited insights into the spatiotemporal metabolic features of the complex gastrointestinal system. Hence, novel approaches are needed to analyze metabolic data to reconstruct the metabolic biological space associated with the evolving structures and functions of an organ such as the gastrointestinal tract. Here, we report the application of multivariate curve resolution (MCR) methodology to model metabolic relationships along the gastrointestinal compartments in relation to its structure and function using data from our previous metabonomic analysis. The method simultaneously summarizes metabolite occurrence and contribution to continuous metabolic signatures of the different biological compartments of the gut tract. This methodology sheds new light onto the complex web of metabolic interactions with gut symbionts that modulate host cell metabolism in surrounding gut tissues. In the future, such an approach will be key to provide new insights into the dynamic onset of metabolic deregulations involved in region-specific gastrointestinal disorders, such as Crohn’s disease or ulcerative colitis.

state.1 Both methods are comprehensively used to generate high density data, from which meaningful biological information is recovered using advanced statistical tools.2 Overall, the application of currently available approaches to metabolic pathway modeling faces the challenge of multicompartmental cellular and tissue metabolic relationships, including those of organs, systemic fluids, and microbial symbionts. Over the past decade, the analysis of the metabolic data generated from biofluids and tissues with advanced chemometric techniques has offered the potential to explore correlations of metabolic pathways across various biological compartments.2,3 For instance, we have previously investigated region-specific metabolic signatures reflecting the structure and function of the intestine via spectroscopic profiling of intestinal biopsies from humans,4 rats,5 and gnotobiotic mice,6,7 which has provided a set of reference metabolic profiles that can be used to assess gut microbial impact at the tissue level.6 Nevertheless, the employed methodologies, generally based on regression modeling techniques,8,9 give emphasis to region-specific metabolic patterns, while providing only limited insights into the metabolic biological space and spatiotemporal metabolic changes along the complex gastrointestinal system. The gastrointestinal system has indeed a histologically diverse epithelial structure superimposed on a complex longitudinal and cross-sectional ordering of multiple connective and muscular cell types. The primary function of this complex biological system is to recover nutrients from ingested aliments via absorption by the gut epithelial cells followed by transport to all the other cells of the whole organism via the circulatory system. The different

Modern nutrition research has promoted the use of metabonomics to gain deeper understanding of the interactions between nutrition and physiological processes. Metabolic profiles provide new insights into physiological regulatory processes of complex organisms which are expressed as a result of interactions between genes and environmental factors, including lifestyle and diet. Complex metabolic profiles of biofluids and tissue biopsies can be generated using mass spectrometry and nuclear magnetic resonance (NMR) spectroscopy. In particular, high resolution magic angle spinning NMR (HR-MAS NMR) spectroscopy is a well-suited approach to provide a snapshot of the metabolism of intact tissue samples through its metabolic profiling at a semisolid

(1) Nicholson, J. K.; Wilson, I. D. Nat. Rev. Drug Discovery 2003, 2 (8), 668– 676. (2) Montoliu, I.; Martin, F. P.; Collino, S.; Rezzi, S.; Kochhar, S. J. Proteome Res. 2009, 8 (5), 2397–2406. (3) Martin, F. P.; Sprenger, N.; Yap, I. K.; Wang, Y.; Bibiloni, R.; Rochat, F.; Rezzi, S.; Cherbut, C.; Kochhar, S.; Lindon, J. C.; Holmes, E.; Nicholson, J. K. J. Proteome Res. 2009, 8 (4), 2090–2105. (4) Wang, Y.; Holmes, E.; Comelli, E. M. Z.; Fotopoulos, G.; Dorta, G.; Tang, H.; Rantalainen, M. J.; Lindon, J. C.; Corthesy-Theulaz, I. E.; Fay, L. B.; Kochhar, S.; Nicholson, J. K. J. Proteome Res. 2007, 6 (10), 3944–3951. (5) Wang, Y.; Tang, H.; Holmes, E.; Lindon, J. C.; Turini, M. E.; Sprenger, N.; Bergonzelli, G.; Fay, L. B.; Kochhar, S.; Nicholson, J. K. J. Proteome Res. 2005, 4 (4), 1324–1329. (6) Martin, F. P.; Wang, Y.; Sprenger, N.; Holmes, E.; Lindon, J. C.; Kochhar, S.; Nicholson, J. K. J. Proteome Res. 2007, 6 (4), 1471–1481. (7) Martin, F. P.; Wang, Y.; Yap, I. K.; Sprenger, N.; Lindon, J. C.; Rezzi, S.; Kochhar, S.; Holmes, E.; Nicholson, J. K. J. Proteome Res. 2009, 8 (7), 3464–3474. (8) Trygg, J. J. Chemom. 2002, 16 (6), 283–293. (9) Wold, S.; Geladi, P.; Esbensen, K.; Ohman, J. J. Chemom. 1977, 1, 41–46.

* Corresponding author. Phone: +41 (21) 785 8771; fax: +41 (21) 785 9486; e-mail: [email protected]. 10.1021/ac102015n  2010 American Chemical Society Published on Web 10/29/2010

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9803

functional compartments are all part of a continuous organ that runs from mouth to anus and that must be coordinated so that functions occur in the proper sequence and at appropriate rates. Starting from the stomach, the small intestine is the major site for digestion and the absorption of nutrients, water, and electrolytes.10 It is divided along its length into three unequally sized portions: the duodenum, the jejunum, and the ileum. Any undigested matter moves to the large intestine, divided into the cecum, colon, and rectum, where the main activity arises from bacterial metabolic processes.10,11 Giving the biological complexity of the gastrointestinal tract, novel approaches are needed to capture interdependence of this topographical-specific metabolism. We previously described the application of multivariate curve resolution (MCR) to model the metabolic dependencies between biological matrixes (blood plasma and tissues).2 The major benefit of applying such an approach lies in the simplified visualization of biochemical relationships across the biological matrices. The method simultaneously provides matrix-specific fingerprints that summarize metabolite occurrence and contribution to the metabolic profile in a meaningful way. In the current contribution, we adapted the MCR methodology to model metabolic relationships along the gastrointestinal space in relation to its structure and function using data from our previous metabonomic analysis.7 The current work involves the characterization of continuous topographical metabolic signatures along the gut tract and the visualization of similarities and differences induced by variable gut microbial ecology. MATERIALS AND METHODS Animal Handling Procedure. This study was conducted under the appropriate national guidelines at the Nestle´ Research Center (Lausanne, Switzerland) and is part of a larger study previously published.7,12 A total of 36 C3H female mice (9 conventional and 27 germ-free animals, Charles River, France), aged 6 weeks, were fed with a standard semisynthetic irradiated rodent diet13 and received saline solution ad libitum. While the conventional mice were kept as controls (n ) 9), one group of germ-free mice was conventionalized by exposure to a normal environment for a period of 4 weeks (n ) 10). The remaining germ-free mice were inoculated with a simplified model of human baby microbiota (HBM), as previously described.7 At 8 weeks of age, HBM mice were given L. paracasei probiotic bacteria in Man, Rogosa, and Sharpe culture medium (108 CFU/ mL) daily for a period of 2 weeks and were separated into two groups. From 8 to 10 weeks of age, conventional, conventionalized, and one group of HBM mice supplemented with probiotics (n ) 7, defined as probiotic treatment) were fed a basal mix diet containing in composition 2.5% of a glucose-lactose mix (1.25% each). Conventional and conventionalized animals (10) DeSesso, J. M.; Jacobson, C. F. Food Chem. Toxicol. 2001, 39 (3), 209– 228. (11) Van, D. S., Sr.; Reeds, P. J.; Stoll, B.; Henry, J. F.; Rosenberger, J. R.; Burrin, D. G.; Van Goudoever, J. B. Gastroenterology 2002, 123 (6), 1931–1940. (12) Martin, F. P.; Dumas, M. E.; Wang, Y.; Legido-Quigley, C.; Yap, I. K.; Tang, H.; Zirah, S.; Murphy, G. M.; Cloarec, O.; Lindon, J. C.; Sprenger, N.; Fay, L. B.; Kochhar, S.; van, B. P.; Holmes, E.; Nicholson, J. K. Mol. Syst. Biol. 2007, 3, 112. (13) Reeves, P. G.; Nielsen, F. H.; Fahey, G. C., Jr. J. Nutr. 1993, 123 (11), 1939–1951.

9804

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

received a saline drink ad libitum containing MRS culture medium as a placebo for probiotic intervention. The second group of HBM mice supplemented with probiotics was fed a diet containing 3 g per 100 g diet of an in-house preparation of galacto-oligosaccharides7 for a period of 2 weeks (n ) 10, defined as synbiotic treatment). Sample Collection. Animals were euthanized at 10 weeks age, and the intestine was removed from each animal. A 3 cm sample was excised from the middle of each section of the intestine, namely duodenum, jejunum, ileum, proximal, and distal colon. Each sample was flushed using 1 mL of an iso-osmotic phosphate buffer solution (0.2 M Na2HPO4/0.04 M NaH2PO4, pH ) 7.4) using a sterile syringe and then preserved at -80 °C until analysis. Sample Preparation and 1H NMR Spectroscopic Analysis. Intact intestinal tissue samples were bathed in 0.9% saline D2O solution. A portion of each tissue (approximately 15 mg) was packed individually into a 4 mm diameter zirconia rotor and a drop of deuteriated isotonic saline solution was added to provide a field-frequency lock for the NMR spectrometer. 1 H NMR spectra were acquired on a Bruker DRX 600 NMR spectrometer (Bruker Biospin, Rheinstetten, Germany) operating at 600.11 MHz using a standard Bruker 4 mm high resolution MAS probe under magic-angle-spinning conditions at a spin rate of 5000 Hz.14 To minimize biochemical degradation during acquisition, the temperature was set to 283 K using cold N2 gas. For each sample, a spectrum was acquired using a Carr-Purcell-Meiboom-Gill (CPMG) spin-echo pulse sequence. 1H NMR spectra were acquired and processed according to previously published parameters.7 Briefly, CPMG spin-echo spectra were measured using a spin-echo loop time (2nτ) of 200 ms and a relaxation delay D1 of 2.0 s. A total of 128 transients were collected into 32 k data points for each spectrum with a spectral width of 20 ppm. Data Processing. Free induction decays were multiplied by an exponential function equivalent to a 0.3 Hz line-broadening factor prior to Fourier transformation. 1H MAS NMR spectra of tissues were manually phased and baseline-corrected using XwinNMR 3.5 (Bruker Analytik, Rheinstetten, Germany). The 1 H NMR spectra were referenced to the chemical shift of the methyl resonance of alanine at δ 1.47. The spectra over the range of δ -1.0 to 10.0 were imported into Matlab (Version 7.0, The Mathworks Inc., Natwick, MA) and reduced to 2057 variables by integrating spectral intensity in segments (width in chemical shift δ 0.005). The water resonance signal (δ 4.50-5.21) was removed, and normalization of each spectrum to a constant sum was carried out prior to statistical analysis. Multivariate Statistical Analyses. Multivariate curve resolution-alternating least squares (MCR-ALS) provides a bilinear decomposition of the response present at the initial data matrix X in a set of factors,15,16 which can be defined as a combination of pure contributions C (concentration profiles) and S (spectral profiles). This decomposition is expressed as: (14) Waters, N. J.; Garrod, S.; Farrant, R. D.; Haselden, J. N.; Connor, S. C.; Connelly, J.; Lindon, J. C.; Holmes, E.; Nicholson, J. K. Anal. Biochem. 2000, 282, 16–23. (15) Tauler, R.; Smilde, A.; Kowalski, B. J. Chemom. 1995, 9, 31–58. (16) Tauler, R.; Kowalski, B.; Fleming, S. Anal. Chem. 1993, 65 (15), 2040– 2047.

X ) CST + E

(1)

where X corresponds to the set of 1H NMR spectra of the different regions of the gut tract. Concentration profiles C reflect the changes in the contribution of the different regions to each factor, and S is the matrix containing the spectral features that codefine each factor. While data can be structured in different ways before the algorithm application, we opted for a column-wise augmentation scheme onto a set of five matrixes corresponding to the five sections of the gut tract. According to this ordering, the MCR model is then described as per eq 2.

[][] []

X1 C1 E1 XCW ) l ) l ST + l XI CI EI

(2)

The application of the ALS algorithm requires the initialization of one set of profiles. Several methods can be used with this purpose to initialize C or S. In our case, contribution profiles were initialized by using evolving factor analysis (EFA).17 The fitting of the model was performed using non-negativity constraints in both modes and assuming the nontrilinearity of the data set. Even if MCR is an extremely flexible exploratory method, it is recommended to have some estimative value on the number of real factors present in the data. Assuming that most of the data variance expressed deterministic information, as opposed to noise, principal component analysis (PCA) using random subset crossvalidation was used in parallel to EFA to achieve a suitable estimation of the number of factors. Under these conditions, the number of relevant factors for each MCR-ALS model was determined to be four or five as obtained from the interpretation on EFA and PCA results of the column-wise augmented matrix. Non-negativity constraints were applied by non-negative leastsquares to both the contribution and the spectral profiles. Spectral profiles were normalized to have equal length. Stopping criterion of the algorithm was set to a difference of 0.1% between the standard deviations of the residuals of two consecutive iterations. The significance of the differences between the concentration profiles of one compartment with that of another was assessed for each factor using the Wilcoxon rank sum test (Supporting Information). Similarly to PCA, the generated MCR factors can be ordered according to the amount of information they explain, which can be calculated for each factor as per the percentage of explained variance. Explained variances for each model are reported for each factor in Table 1. This explained variance can be calculated by: Table 1. Percentage of Explained Variances per MCR Factor in Each Model model

factor 1

factor 2

factor 3

factor 4

factor 5

global conventional conventionalized HBM + probiotics HBM + synbiotics

15.2344 45.6616 43.2693 30.9763 25.3486

27.8551 12.7176 12.8330 31.2527 37.2217

21.9350 10.6590 30.0296 12.3557 11.3022

39.4161 21.6743 15.0513 22.1641 24.8525

29.7857 13.2188 12.4848

[

% explained variance (f) ) 1 -

m

(

n

∑ ∑ (x

ij

- xˆij)2)

i)1 j)1 n m

(

∑ ∑ (x ) ) 2

ij

i)1 j)1

]

(3)

where xˆij corresponds to the outer product between the f th pure concentration profile and its associated spectrotype and xij represents original NMR data. Interpretation and Assignment of the MCR Spectrotypes. For the analysis of tissues, the CPMG NMR pulse sequence was used to attenuate the spectroscopic contributions of large macromolecular species and to favor observation of sharp signals arising from low molecular weight metabolites. The metabolite identification was achieved using an in-house reference compound database and literature data12,18,19 and confirmed by 2D homoand heteronuclear NMR spectroscopy experiments, as reported previously.7 The modeling of 1H NMR profiles using MCR-ALS approaches enables the characterization of numerical spectroscopic constructs, hereafter called spectrotypes,20 that describe fingerprints, which summarize the metabolite presence. Indeed, since the MCR-ALS biochemical profiles are numerical constructs from 1H NMR spectra, the interpretation and assignment are similar to those of real 1H NMR spectra of gut tissues, as previously reported.7 Within each factor, the interpretation of the spectrotypes is based on the presence and relative proportion of metabolite signals. RESULTS MCR-ALS Application to the Modeling of Compartment Metabolic Relationships across the Gastrointestinal Tract of Mice. Initial data modeling was performed using 1H HRMASbased metabolic profiles of tissues from the different compartments of the gastrointestinal tract (duodenum, jejunum, ileum, proximal and distal colon) obtained from all the microbiome mouse models. To obtain a general overview, no previous data scaling was applied. Under the conditions described in Material and Methods, a five-factor MCR-ALS model was obtained, explaining 97.81% of the variance expressed as the sum of squares. Explained variances for each factor are reported in Table 1. For each MCR-ALS factor, pure contribution profiles showed the distribution of individual metabolic profiles according to their inherent biochemical composition (Figure 1 A). By displaying the median values for each gut functional compartment (Table 2), Figure 1A shows how region-specific patterns can be captured across the different compartments of the gastrointestinal tract in each MCR-ALS factor. Simultaneously, the metabolic changes responsible for the differences between samples can be extracted from the pure spectral profiles (Figure 2), or spectrotypes.20 For each factor, the spectrotypes summarized both metabolite occurrence and relative proportion in a meaningful way. Moreover, the (17) Keller, H. R.; Massart, D. L. Chemom. Intell. Lab. Syst. 1992, 12, 209–224. (18) Fan, T. W. Prog. Nucl. Magn. Reson. Spectrosc. 1996, 28 (2), 161–219. (19) Nicholson, J. K.; Foxall, P. J.; Spraul, M.; Farrant, R. D.; Lindon, J. C. Anal. Chem. 1995, 67 (5), 793–811. (20) Richards, S. E.; Wang, Y.; Lawler, D.; Kochhar, S.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Anal. Chem. 2008, 80 (13), 4876–4885.

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9805

Figure 1. Global MCR-ALS analysis. (A) Pure concentration profiles for each of the five factors. The median is displayed using a black line. (B) Radar plot displaying the median of the concentration profiles along the different gut compartments for each of the five factors. Table 2. Median Values of the Pure Concentrations Profiles for Each Compartment and Factor compartment

factor 1

factor 2

factor 3

factor 4

factor 5

duodenum jejunum ileum proximal colon distal colon

5.72 2.65 2.63 0.94 1.18

4.59 6.48 1.79 2.94 2.14

1.84 3.15 8.19 1.83 1.23

1.90 1.94 0.88 13.93 11.82

3.39 4.87 2.22 4.58 5.78

information encapsulated in the spectrotype described the biochemical relationships across the different biological compartments as per matrix-specific fingerprints that are directly associated with the region-specific patterns observed in Figure 1A. 9806

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

MCR-ALS pure compartment profiles pinpointed the contributions of pure spectrotypes in different biological compartments (Figures 1 and 2). In order to simplify the visualization of the data (i.e., the biochemical variations along the gut resulting from the overall model), we summarized the trajectories along the gut tract, for each factor, by radar plots using the median values of the pure contributions within each gut compartment (Figure 1B). In order to preserve the proportion of the contribution profiles for the different factors and for clear data display, we plotted cumulative median values from the first to the fifth factors for each gut compartment (Table 3). Such a plot provides at the same time a simplified overview of the

Figure 2. Global MCR-ALS analysis. Spectrotypes for five factors derived from 1H NMR spectra of intact intestinal necropsies obtained from all the mice. Table 3. Cumulative Median Values of the Pure Concentrations Profiles for Each Compartment compartment

factor 1

factor 2

factor 3

factor 4

factor 5

duodenum jejunum ileum proximal colon distal colon

5.72 2.65 2.63 0.94 1.18

10.31 9.13 4.42 3.87 3.32

12.16 12.28 12.61 5.70 4.55

14.06 14.23 13.49 19.63 16.36

17.45 19.10 15.71 24.21 22.14

topographical biochemical distribution along the intestinal tract for each factor and the overall model performance. The modeling of 1H NMR profiles using MCR-ALS approach enables the characterization of numerical spectroscopic constructs, which summarize the metabolite presence and the proportion within each factor. Indeed, since the MCR-ALS biochemical profiles provide numerical constructs from 1H NMR spectra, the interpretation and assignment are similar to those of real 1H NMR spectra of gut tissues, as we previously reported.7 For instance, the first factor modeled a metabolic feature that was dominant in the first gut section, the duodenum, and gradually decreased along the small intestine until

reaching a lower proportion across the large intestinal section. The corresponding 1H NMR spectrotype was dominated by the signatures of several amino acids, lactate, choline, phosphocholine (PC), and glycerophosphocholine (GPC) (Figure 2, factor 1). The second factor described a metabolic pattern that was characteristic of the duodenal and jejunal sections of the tract, and with a low proportion to the remaining of the gastrointestinal system (Figure 1). The spectrotype was mainly dominated by the metabolic signature of lactate, alanine, creatine, choline, and taurine (Figure 2, factor 2). Moreover, the third pure contribution profile showed a metabolic pattern with a contribution gradually increasing along the small intestine, and with a minor contribution to the large intestine (Figure 1). This third spectrotype summarizes mainly greater concentrations of glutathione, glutamate, choline, betaine, PC, GPC, and taurine (Figure 2, factor 3). While the fourth factor captured a colon-specific metabolic signature associated with a higher content in lipids, the fifth factor modeled a metabolic fingerprint with similar contribution to the entire gastrointestinal tract, except in the ileum where the contribution is lower (Figures 1 and 2). The corresponding fifth spectrotype contains signals from lactate, creatine, PC, and taurine. Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9807

Figure 3. Microbiome specific MCR-ALS analysis. Radar plots displaying the median of the concentration profiles along the different gut compartment for each of the factors generated from MCR-ALS analysis together with selected spectrotypes. MCR-ALS models generated from (A) conventional, (B) conventionalized, (C) HBM and probiotics, and (D) HBM and synbiotics mice.

MCR-ALS Modeling Describes the Differential Impact of Natural Gut Microbial Colonization and Conventionalization on Gastrointestinal Metabolic Space. MCR-ALS was applied to the modeling of metabolic relationships across gut compartments obtained from conventional and conventionalized animals. Anticipating the fact that some of the differences between tissues would be subtle, normalization to unit standard deviation was 9808

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

applied. Under these conditions, a four-factor MCR-ALS model was obtained with the samples obtained from conventional animals, explaining 71.27% of the variance expressed as the sum of squares (Figure 3A; Supporting Information Figures 1 and 2). Explained variances for each factor are reported in Table 1. The first factor captured a metabolic pattern that gradually decreased along the small intestine until reaching a lower proportion in the

colon. The second factor described the ileal specific metabolic signature. The third factor modeled the metabolic features specific to the proximal and distal colon. The fourth factor modeled a metabolic signature constant across the small intestine and increasing gradually from the ileum to the terminal colon. The first factor showed the signatures of several amino acids (e.g., Val, Leu, Ile, Ala, Lys, Arg, Met, Glu, Gln, Gly, Tyr, Phe), lactate, creatine, choline, PC, and GPC. The second spectrotype summarized mainly greater concentrations of lactate, alanine, glutathione, glutamate, choline, GPC, creatine, glycine, and taurine. The third spectrotype described a higher content of lipids. The fourth spectrotype was dominated by high levels of lactate, creatine, taurine, PC, GPC, and myo-inositol. A similar strategy for modeling gut metabolic space in conventionalized mice resulted also in a four-factor model, explaining 74.47% of the variance expressed as the sum of squares (Figure 3 B, Supporting Information Figures 3 and 4). Explained variances for each factor are reported in Table 1. The obtained model showed a number of similarities with the one obtained previously, the greater difference being in the third factor. The first factor captured a metabolic pattern that gradually decreased along the small intestine until reaching a low proportion in the colon. The corresponding spectrotype showed the signatures of most of the amino acids (e.g., Val, Leu, Ile, Ala, Lys, Arg, Met, Glu, Gln, Gly, Tyr, Phe), lactate, creatine, choline, PC, and GPC. The second factor modeled a specific ileum metabolic signature associated with greater levels of glutathione, glutamate, GPC, betaine, and taurine. The fourth component described the specific high content of lipids in proximal and distal colon. The third factor summarized a fingerprint with similar contribution to the entire gastrointestinal tract, except in the jejunum where the contribution was higher. The associated spectrotype showed metabolic signatures associated with lactate, creatine, choline, PC, GPC, taurine, several amino acids, and glutathione. MCR-ALS Modeling Describes How the Interactions between Nutritional Intervention and Gut Microbiome Modulate the Gastrointestinal Metabolic Space. MCR-ALS modeling of gastrointestinal metabolic space in HBM mice supplemented with probiotics or synbiotics resulted in a five-factor model, explaining 75.89 and 77.43% of the variance, respectively (Figure 3C and 3D; Supporting Information Figures 5-8). Explained variances for each factor are reported in Table 1. This observation highlights a major change in the biochemical composition along the gut of these animals when compared to conventional and conventionalized mice. In both cases, the first component described a duodenal metabolic feature marked by high levels of most of the amino acids, lactate, choline, PC, and GPC. For both models, the second component summarized a metabolic pattern present in all the gut compartments but with a higher contribution to the jejunum and ileum. The corresponding spectrotypes showed a contribution of lactate, creatine, choline, PC, GPC, taurine, several amino acids, and glutathione. The third factor was similar in both models and summarized ileal-specific metabolic features, as per greater levels of creatine, glutathione, glycine, glutamate, GPC, betaine, and taurine. The fifth factor modeled a colon-specific phenotype marked by high lipid content.

The models tend to differ in the fourth component. In HBM mice supplemented with probiotics, the fourth component expressed a metabolic pattern present along the whole intestinal tract, with a higher contribution to the jejunum. This metabolic feature showed greater levels of lactate, creatine, glutamateglycine, taurine, choline, and GPC. In HBM mice supplemented with prebiotics, the fourth component described a metabolic pattern associated with the jejunum, ileum, proximal colon, and more relatively the distal colon. This metabolic feature was marked by its complex biochemical composition with many amino acids, taurine, lactate, PC, GPC, and choline. DISCUSSION The aims of the present study were to model the multidimensional metabolic relationships along the intestine to provide an overview of the gastrointestinal metabolic space and variability in the presence of different gut microbiomes using an unsupervised chemometric methodology, i.e., MCR-ALS.2 In our previous publication,7 we extensively described the results generated from the application of conventional PLS-DA approaches and discussed the results in relation to structural and functional features of the gastrointestinal sections. Therefore, we here discuss the relevance of the MCR-ALS outcomes in relation to our previous findings to demonstrate the consistency of the data modeling. Modeling Compartment Metabolic Relationships Using MCR-ALS. Our previous investigations have demonstrated that the topographical variation along the gastrointestinal tract exceeds the variation introduced by nutritional interventions and gut microbiota variations, as per PCA and PLS-DA modeling.7 Therefore, in order to model the intestinal metabolic space, initial data modeling was performed using 1H HRMAS-based metabolic profiles of the combined gut tissues (duodenum, jejunum, ileum, proximal and distal colon) obtained from all the microbiome mouse models. MCR-ALS model generated new insights into metabolic dynamics along the continuous gastrointestinal tract by describing metabolite presence and proportion specific to a single or several compartments. For instance, the model showed the specific metabolic gradient increasing or decreasing from the duodenum to the colon (first and fourth component), but it also provided evidence that specific compartments, such as the ileum, have unique biochemical properties, in agreement with our previous analysis.7 In particular, the third component was able to capture a feature specific to the small intestine that we previously associated with the site-specific digestion of alimentary lipids, mainly converted to PC and GPC in mucosal cells,21 and cell protection against oxidative stress, as per specifically greater levels of the antioxidant glutathione.22 In addition, the first component highlighted that the small intestine also has an important function in amino acid metabolism, since it is involved in the catabolism of more than 50% of the dietary amino acids,23 a feature we have previously observed.7 Moreover, the second factor of the model highlighted interdependence in the biochemical variability of the two first gut sections (duodenum and jejunum). The model also (21) Parthasarathy, S.; Subbaiah, P. V.; Ganguly, J. Biochem. J. 1974, 140 (3), 503–508. (22) Loguercio, C.; Di, P. M. Ital. J. Gastroenterol. Hepatol. 1999, 31 (5), 401– 407. (23) Wu, G. J. Nutr. 1998, 128 (8), 1249–1252.

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9809

captured a feature specific to the large intestine as per its specific content in lipids, that we previously ascribed to colon lipid metabolism.7,24-27 Similarly to other bilinear models, the generated MCR factors express different levels of data variance. The overall analysis of such variance must be done considering simultaneously both the concentration profiles and the pure spectral profiles. However, differences in metabolite contribution within each factor can be assessed, analyzing their relative importance in the spectrotypes. Indeed, similarly to PCA, the generated MCR factors can be ordered according to the amount of information they explain (Table 1). Therefore, the contribution of a metabolite to the overall NMR signal can also be assessed by order of importance. For instance, in the global MCR model (Figure 2), lactate is present in the spectrotype for the factors 1, 2, 3, and 5. In factor 1, the lactate contribution is relatively low compared to the signals of amino acids, while in factors 2 and 5, lactate is a dominant contributor with a few other compounds. In factor 3, lactate is only present as a residual signal. Because of the structure of the MCR model (Table 1), we can comment that lactate is of greater importance to describe the features along the third and fifth components than along the first factor. Indeed, along the first factor, not only the explained variance is lower but also the contribution of lactate to the overall spectrotype is of lesser importance. Finally, the fifth factor highlights a strong relationship between lactate levels along the gastrointestinal tract. More generally, MCR-ALS described an interesting metabolic feature along the fifth factor, which was shared by all the gut compartments (lactate, creatine, PC, and taurine), with a lower contribution to the ileum. This novel observation shows a strong relationship between the levels of these biochemical species along the gastrointestinal tract of one single individual, which may relate to specific structural and functional properties of the gastrointestinal system, and suggests coordinated regulatory mechanisms to ensure osmotic protection and integrity of the continuous tract. Therefore, MCR offers a unique way to capture the different order of biochemical relationships across the continuous tube that is our gastrointestinal system. Gut Microbial Acquisition Impacts Gastrointestinal Metabolic Space. We previously described how the application of supervised multivariate data modeling (e.g., PLS-DA, O-PLS-DA) enabled the assessment of the metabolic impacts of inoculating germ-free mice with different microbiomes. However, this methodology provided the gut compartment specific metabolic signatures but no global overview of the metabolic changes across the intestinal biological space of the intestine. Here, we show how MCR-ALS application can be used as a tool to assess the impact of gut microbiota and nutritional modulation through modeling the whole gut metabolic space. In both conventional and conventionalized animals, the models highlighted three shared metabolic features, as per a biochemical gradient from the small intestine to the colon, a gradient from the ileum to the colon, and an ileal-specific metabolic signature. (24) Robertson, M. D.; Parkes, M.; Warren, B. F.; Ferguson, D. J.; Jackson, K. G.; Jewell, D. P.; Frayn, K. N. Gut 2003, 52 (6), 834–839. (25) Zambell, K. L.; Fitch, M. D.; Fleming, S. E. J. Nutr. 2003, 133 (11), 3509– 3515. (26) Carter, J. R.; Kennedy, E. P. J. Lipid Res. 1966, 7 (5), 678–683. (27) Marie, M. C.; Fred, G. L.; Lauretta, G.; Satchitanandum; George, V. Dig. Dis. Sci. 1985, V30 (5), 468–476.

9810

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

The first feature is consistent with our previous observations in conventional animals, and it describes the site-specific functions in processing dietary amino acids and lipids. The second feature also corresponds to a major metabolic characteristic of the large intestine, which shows a compartment-specific content in lipids. Interestingly, the third feature, the ileal specific metabolic fingerprint, corresponds to the differential metabolic profile previously generated using OPLS-DA.7 The specific ileal metabolic signature is composed of glutathione, its precursors in the γ-glutamyl cycle (glutamate, glycine), the osmoprotectants taurine and choline, and GPC, which reflects the major role of this tissue for absorption of emulsified dietary lipids and protective mechanisms against oxidative stress, especially from free radicals generated from lipid metabolism. However, the models diverged from each other in one factor. In conventional animals, factor 4 showed a metabolic signature constant across the small intestine and increasing in the large intestine, including levels of lactate, creatine, taurine, PC, GPC, and myo-inositol. In conventionalized animals, factor 3 modeled a spectrotype with a similar contribution to the entire gastrointestinal tract and a higher contribution in the jejunum. This spectrotype included the metabolites described previously (with the exception of myo-inositol) plus several amino acids and glutathione. This observation indicated that the overall metabolic integrity of the gastrointestinal systems is strongly dependent on colonization by the gut microbiota. While the model highlights the colon-specific osmoregulatory functions and muscular metabolism in conventional animals, MCR-ALS captured two key features in conventionalized animals: (i) the higher content of taurine, PC, GPC, glutathione, and (ii) a relatively high level of free amino acids. The first feature was obtained previously by using OPLS-DA to discriminate both groups of animals, while the second feature was only achieved using PLS-DA to discriminate each gut tissue from the others for the specific microbiome mouse model.7 Nutrition-Gut Microbial Interactions Impact Gastrointestinal Metabolic Space. The fitting of the models for each microbiome mouse model showed a distinct behavior when comparing conventional and conventionalized mice (four factors) with HBM animals (five factors). The overall modeling of the intestinal metabolic space thus suggests that the biochemical variability along the gastrointestinal tract is strongly dependent on the presence and the type of gut bacteria. The modeling of the metabolic variations along the gastrointestinal tract of HBM mice supplemented with probiotics or synbiotics revealed strong similarities across four out of five components. Interestingly, these four factors summarized metabolic features specific to one or two gut compartments, highlighting a greater variability in the metabolic relationships along the gastrointestinal tract of mice inoculated with HBM. The models captured metabolic signatures associated with the duodenum in relation to the processing of dietary lipids and amino acids (factor 1), the ileum for lipid and glutathione metabolism (factor 3), and the large intestine for its content of lipids (factor 5). The models also highlighted a strong relationship between the composition of the jejunum and the ileum (factor 2). Our previous investigations using PLS-DA highlighted strong differences in the profiles of the jejunum, which contains greater levels of many metabolites,

making the biochemical profile similar to that of the ileum.7 In addition, the models showed that the nutritional modulation of HBM impacts the metabolic space of the whole tract. In mice supplemented with the probiotics, a specific metabolic relationship is modeled across the whole intestine with a greater contribution to the jejunum, whereas the animals receiving the prebiotics showed specific metabolic relationships with greater contribution to the jejunum, ileum, and proximal colon (factor 4). These groupspecific observations are strengthened by our previous observations that HBM mice strongly differed from other animals in the metabolic feature of the jejunum, and for HBM mice supplemented with prebiotics in the ileum and the proximal colon as per greater metabolic changes in PC, GPC, and choline.7 Therefore, our work demonstrates that MCR-ALS is a well adapted method to capture divergence across the overall gastrointestinal metabolic space of animals in response to the gut microbial colonization state. This methodology sheds new light onto the complex web of metabolic interactions with gut symbionts that modulate host cell pathways in surrounding gut tissues. In

the future, such an approach will be key to provide new insights into the dynamic onsets of metabolic deregulations involved in region-specific gastrointestinal disorders, such as Crohn’s disease or ulcerative colitis. ACKNOWLEDGMENT The authors acknowledge Sebastiano Collino, Norbert Sprenger, Christine Cherbut, and Florence Rochat for their support and input, John Newell, Monique Julita, Massimo Marchesini, Catherine Schwartz, and Christophe Maubert for their help in animal husbandry. SUPPORTING INFORMATION AVAILABLE Additional information as noted in the text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review July 29, 2010. Accepted October 15, 2010. AC102015N

Analytical Chemistry, Vol. 82, No. 23, December 1, 2010

9811