Offspring Metabolomic Response to Maternal Protein Restriction in a

May 24, 2011 - Angelica Dessì , Flamina Cesare Marincola , Vassilios Fanos. Best Practice & Research Clinical Obstetrics & Gynaecology 2015 29, 156-1...
0 downloads 0 Views 3MB Size
TECHNICAL NOTE pubs.acs.org/jpr

Offspring Metabolomic Response to Maternal Protein Restriction in a Rat Model of Intrauterine Growth Restriction (IUGR) Marie-Cecile Alexandre-Gouabau,*,† Frederique Courant,‡ Gwena€elle Le Gall,§ Thomas Moyon,† Dominique Darmaun,† Patricia Parnet,† Berengere Coupe,† and Jean-Philippe Antignac‡ †

INRA and University of Nantes, UMR-1280 Physiologie des Adaptations Nutritionnelles CHU H^otel Dieu, 44093 Nantes cedex 1, France ‡ ONIRIS, USC 2013, Laboratoire d’Etude des Residus et Contaminants dans les Aliments (LABERCA), Nantes, France § Institute of Food Research, Norwich Research Park, Colney, Norwich, NR4 7UA, United Kingdom ABSTRACT: Intrauterine growth restriction (IUGR), along with postnatal growth trajectory, is closely linked with metabolic diseases and obesity at adulthood. The present study reports the time-dependent metabolomic response of male offspring of rat dams exposed to maternal adequate protein diet during pregnancy and lactation (CC) or protein deprivation during pregnancy only (IUGR with rapid catch-up growth, RC) or through pregnancy and lactation (IUGR with slow postnatal growth, RR). Plasma LC-HRMS metabolomic fingerprints for 8 male rats per group, combined with multivariate statistical analysis (PLS-DA and HCA), were used to study the impact of IUGR and postnatal growth velocity on the offspring metabolism in early life (until weaning) and once they reached adulthood (8 months). Compared with CC rats, RR pups had clear-cut alterations in plasma metabolome during suckling, but none at adulthood; in contrast, in RC pups, alterations in metabolome were minimal in early life but more pronounced in the long run. In particular, our results pinpoint transient alterations in proline, arginine, and histidine in RR rats, compared to CC rats, and persistent differences in tyrosine and carnitine, compared to RC rats at adulthood. These findings suggest that the long-term deregulation in feeding behavior and fatty acid metabolism in IUGR rats depends on postnatal growth velocity. KEYWORDS: intrauterine growth restriction, growth velocity, metabolomics, carnitine, catch up growth, amino acids

’ INTRODUCTION The concept that fetal nutrition influences the risk of developing metabolic and cardiovascular disease in adulthood is commonly referred to as the Barker hypothesis.1 Infants born with intrauterine growth restriction (IUGR) due to undernutrition during fetal life may be “reprogrammed” toward a thrifty phenotype, enhancing the chances of fetus survival at the cost of permanent alterations that may have a negative impact on longterm health. In addition, the growth trajectory over the first few months of neonatal life following IUGR can further impact the risk of developing metabolic or cardiovascular disease: subjects born with a birth weight below the 10th percentile for gestational age and experiencing rapid weight gain during early childhood (“catch-up growth”) indeed are exposed to a higher risk of cardiovascular disease as adults.2,3 Decreased uteroplacental blood flow, which limits the supply of nutrients critical for fetal growth, is the most common cause of IUGR. A large number of studies described specific alterations in the expression and activity of placental transporters mediating the transfer of amino acids (system A),46 ions7 and lipids8 in human IUGR. Similarly, placental system A activity is altered in a model of IUGR produced by a low-protein maternal diet in rat.9,10 In addition, in such IUGR model, various alterations in islet cells,11 hypothalamo-pituitaryadrenal axis,12 insulin secretion,13 insulin-stimulated glucose uptake r 2011 American Chemical Society

in skeletal muscle,14 epipidymal and intra-abdominal fat mass and the antilipolytic action of insulin,15 oxidative stress,16 and mitochondrial function in islets17 and liver, contributing to increased hepatic gluconeogenesis,18 combined with a decrease in glycolysis,19 have been reported to occur at various ages in postnatal life. Taken collectively, literature evidence therefore suggests maternal protein restriction may be a relevant model to assess the effect of IUGR on offspring metabolism. In our laboratory, we recently demonstrated that IUGR altered neuronal hypothalamic circuity and feeding behavior20 and resulted in an early deregulation of enzymes involved in ketogenesis, fatty acid β-oxidation, and mitochondrial oxidative system in the male suckling pups’ hypothalamic proteome [Alexandre et al, JNB, on line], along with permanent alterations in several genes of the insulin and leptin signaling pathways, and of the family of nuclear receptors involved in lipid metabolism.21 Consistent with these findings, once they reached adulthood, rats born with IUGR had a higher fat mass, elevated plasma triglycerides, and a resistance to leptin that was evident after a rapid catch-up growth or the consumption of a high fat diet.22 We therefore hypothesized that reduced nutrient availability during a critical period of development may program tissue sensitivity to energy Received: April 11, 2011 Published: May 24, 2011 3292

dx.doi.org/10.1021/pr2003193 | J. Proteome Res. 2011, 10, 3292–3302

Journal of Proteome Research

TECHNICAL NOTE

Table 1. Composition of the Maternal Experimental Diets Administered during Gestation and/or Lactation 8% protein diet

20% protein diet

(g/kg food)

(g/kg food)

Proteins and amino acids casein Methionin

90 0.8

220 2.0

Carbohydrates Dextrose Starch of corn

681.7

551.5

80.0

80.0

43.0

43.0

Lipids Soya oil

Fibers Cellulose

50.0

50.0

Vitamins and minerals mixture

50.5

49.5

Cholin Energy (kCal/kg diet)

4.0

4.0

3674.0

3722.7

intake and mitochondrial activity, and thus impact metabolism. These observations led us to question (i) whether these alterations translate into meaningful alterations in the neonatal metabolome, defined as the complete set of the small molecules that contribute to metabolism,23 (ii) if so, whether such changes are modulated by the velocity of growth in the first few weeks of life, and (iii) whether these changes are sustained and still detectable in adult rats. Metabolomics has emerged over the past decade and is used to characterize the metabolic phenotype of individuals integrating genetic polymorphism, metabolic interactions with commensal and symbiotic partners such as gut microbiota, as well as environmental and behavioral factors including food preferences.24,25 Therefore, metabolomics stands as an innovative tool in the field of (nutritional) pediatric research.26 It has been used to study inborn errors of metabolism, such as propionate metabolism27 or to assess the nutritional phenotypes of 8-year old children receiving a diet rich in milk or meat protein28 or the impact of maternal high-fat nutrition in a fetal primate model.29 The latter studies encouraged us to use metabolomics as a relevant approach to assess the metabolic adaptations to either rapid catch-up growth, or slow postnatal growth, of the offspring in a model of IUGR induced by maternal dietary protein restriction both in early life (before weaning) and at adulthood (at 8 months of age). The multivariate statistical analysis of plasma metabolomic fingerprints, acquired using liquid chromatography coupled to high resolution mass spectrometry (LCHRMS), revealed major differences in plasma metabolome between IUGR and control rats. The prominent potential biomarkers we identified were related to fatty acid beta-oxidation (carnitine and its derivatives), and amino acid metabolism (histidine, proline, arginine and tyrosine, ...).

Local Animal Care Committee of the Pays de Loire Region (CREEA). Ten-week old female and male SpragueDawley rats (Janvier, Le Genest Saint Isle, France) were maintained under controlled conditions (22 °C, 12-/12-h dark/light cycle) with free access to regular food containing 16 g protein per 100 g pellets (A04, Safe, Augy, France) and tap water. After 10 days of habituation, female rats were mated overnight with a male, and copulation was verified the next morning by the presence of spermatozoa in vaginal smears. From the day of conception, 26 pregnant dams were housed individually and randomize to receive either an adequate protein diet (Control diet C, containing 20% protein; n = 14), or an isocaloric, low protein diet (Restricted diet, R, containing 8% protein; n = 12) as described30 (Table 1). Diets were purchased from Arie Block BV (Woerden, The Netherlands). At delivery, female pups were discarded and male pups born from undernourished (Restricted, R) mothers or from normally fed mothers (Control, C) were randomly adopted either by six control foster mothers, or by three R foster mothers so as to obtain litters of eight male pups in each of the three experimental groups: CC, RC and RR, where the first letter depicts the maternal diet administered during gestation, and the second letter the diet received by dams during lactation. Plasma Collection

Rats were rapidly euthanized between 09:00 and 11:00 a.m. by CO2 inhalation, and less than two hours after birth for the pups at postnatal day 0 (PND0). Blood was collected by decapitation into heparinized tubes (Laboratories Leo SA, St Quentin en Yvelines, France) and centrifuged at 2500 g for 15 min at 4 °C, and the entire serum of each animal was frozen immediately at 80 °C until subsequent LC-HRMS analysis. We chose to keep plasma preparation to a minimum in order to preserve the wide diversity of potential metabolites in terms of chemical structure and concentrations. Plasma samples were filtered on PALL centrifugal devices (NANOSEP OMEGA, cut of at 10 kDa) at 6000 rpm for 30 min at 10 °C to remove high molecular weight proteins. Metabolite Standards

Amino acids, carnitine and its derivatives were obtained from Sigma-Aldrich (St Quentin-Fallavier, France). Standards were dissolved in 98% Ethanol to obtain solutions of 0.1 mg/mL before LCMS or LCMS/MS analysis. The chemicals used were of analytical or reagent grade. Plasma Hormone Concentrations

Plasma leptin and insulin concentrations were determined with specific ELISA kits (rat/mouse insulin ELISA kit and rat/ mouse leptin ELISA kit; Linco Research, St. Charles, MO). Body Weights and Statistical Analysis

Data are expressed and presented, in Table 2, as mean ( SEM, and differences among groups were determined using Mann Whitney U test, by using SAS version 9.1 (SAS Institute Inc., Cary, NC). In all tests, P < 0.05 was considered significant. Metabolomic LC-HRMS Fingerprinting

’ MATERIALS AND METHODS Animals and Diet

All experiments were performed in accordance with the European Communities Council Directive of November 24, 1986 (86/609/ECC) regarding the care and use of animals for experimental procedures. All protocols were approved by the

Reverse phase HPLC using positive electrospray mode was used, because it is known to have a large application range and has been used with success under the same conditions in a previous metabolomic study in our laboratory.31 Briefly, the instrument used was a Finningan Surveyor Plus HPLC system, coupled with an LTQ-Orbitrap hybrid mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). All solvents 3293

dx.doi.org/10.1021/pr2003193 |J. Proteome Res. 2011, 10, 3292–3302

Journal of Proteome Research

TECHNICAL NOTE

Table 2. Physiological and Biochemical Outcomes of the 3 Groups of Rats According to Their Postnatal Age in Days (PND) PND PND0 PND5

PND12

PND16

PND22

PND260

perinatal nutritional status

plasma leptin (ng/mL)a

body weight (g)

plasma insulin (ng/mL)a

C

6.64 ( 0.14, n = 8

1.25 ( 0.22, n = 8

0.44 ( 0.06, n = 8

R

5.80 ( 0.35b, n = 6

0.00 ( 0.00, n = 6

0.46 ( 0.09, n = 6 1.69 ( 0.14, n = 7

CC

15.51 ( 0.15, n = 8

6.54 ( 0.51, n = 8

RC

15.23 ( 0.55, n = 7

10.87 ( 1.18b, n = 7

1.35 ( 0.24, n = 8

RR

13.36 ( 0.61d,h, n = 8

4.20 ( 0.33e,f, n = 8

0.80 ( 0.17e, n = 8

CC

32.17 ( 0.53, n = 7

8.30 ( 0.71, n = 7

0.84 ( 0.14, n = 8

RC

30.61 ( 1.52, n = 8

10.28 ( 1.02, n = 7

0.71 ( 0.13, n = 8

RR CC

21.55 ( 1.19f,h, n = 8 37.15 ( 0.64, n = 8

5.49 ( 0.63c,d, n = 8 6.40 ( 0.35, n = 8

0.44 ( 0.08c, n = 8 0.78 ( 0.10, n = 8

RC

40.90 ( 1.17b, n = 8

10.93 ( 0.64g, n = 8

1.14 ( 0.29, n = 8

RR

23.91 ( 0.73 , n = 7

2.69 ( 0.24h,i, n = 7

0.19 ( 0.07e,f, n = 7

h,i

CC

50.43 ( 2.36, n = 7

2.00 ( 0.26, n = 8

0.30 ( 0.05, n = 8

RC

50.93 ( 1.37, n = 7

1.88 ( 0.13, n = 8

0.25 ( 0.05, n = 8

RR

31.56 ( 1.35h,i, n = 8

1.66 ( 0.30, n = 8

0.10 ( 0.02e,f, n = 8

CC

665.83 ( 42.54, n = 6

11.89 ( 1.71, n = 6

0.70 ( 0.20, n = 6

RC RR

645.22 ( 38.87, n = 6 631.14 ( 11.19, n = 7

10.61 ( 1.54, n = 6 7.16 ( 1.22c, n = 6

0.70 ( 0.15, n = 6 0.27 ( 0.05c,d, n = 6

a

Plasma leptin and insulin values were measured in fasted rats for PND260. Significant variations between means were tested by MannWhitney U test. Results are expressed as mean ( SEM. b P < 0.05: RC vs CC. c P < 0.05: RR vs CC. d P < 0.05: RR vs RC. e P < 0.005: RR vs CC. f P < 0.005: RR vs RC. g P < 0.001: RC vs CC. h P < 0.001: RR vs CC. i P < 0.001: RR vs RC.

and reagents were of analytical or HPLC grade quality and purchased from Solvent Documentation Synthesis (SDS, Peypin, France). Fifteen milliliters of filtered plasma samples were injected onto a 150  2.1 mm Uptisphere HDO-C18 3 μm column (Interchim, Montluc-on, France). Elution was performed with a mobile phase consisting of water containing 0.1% acetic acid (A), and acetonitrile containing 0.1% acetic acid (B) and with the same MS parameters previously described.31 Data were acquired in centroid and full scan mode (m/z 50 to 800) using the Orbitrap mass analyzer operating at a 30 000 mass resolution (fwhm as defined at m/z 400). Quality Controls

Two types of quality control (QC) samples were used, which were injected four to five times in randomized order in each batch of injection. QC1 was Milli-Q water sample and QC2 was a standard mixture solution 1 ng/μL, previously used in a metabolomic study in our laboratory31 and consisting of methyltestosterone, stanozolol, medroxyprogesterone acetate, triamcinolone and ponasterone A. Metabolomic Fingerprint Preprocessing

The open-source xcms software32 was used for non linear alignment of the data and automatic integration and extraction of the peak intensities for each mass-retention time ([m/z; rt]) features (ions). XCMS parameters for the R language were implemented in an automation script with the algorithm “match-filter” by default except the interval of m/z value for peak picking that was set to 0.1, the noise threshold set to 6, the group bandwidth set to 20 and the minimum fraction set to 0.95. Univariate Analysis of LC-HRMS Data

After preprocessing, XCMS compares through a Student t-test the signal abundances observed for identical ions in two groups of samples, to identify metabolites presenting intensities significantly different in the two groups and ranked these metabolites according to the associated statistical confidence level (p-value). A p-value lower than 0.05 as well as a fold change higher than 1.5

were used as selection criteria to identify the potentially most relevant metabolites permitting to discriminate our various subgroups. Multivariate Analysis of LC-HRMS Data

All multivariate data analyses and modeling were performed on the compiled table of the selected variables using SIMCA-Pþ  software (version 12, Umetrics Inc., Umea, Sweden). To take into account the systematic variation in X matrix (i.e., metabolomic fingerprints) and to identify variables x (i.e., ions) which explained the best the Y variable (in our case the perinatal nutritional regimen) through the development postnatal period and at adulthood, we applied a supervised method, the Partial Least Squares-Discriminant Analysis (PLS-DA) for each PND data set analyzed separately. When we analyzed the three nutrional groups (CC, RC and RR) during the lactating period (for PND5, PND12, PND16 and PND22 together), we used PLS-DA combined with a multivariate preprocessing filter called Orthogonal Signal Correction (OSC). By removing with-in class variability and confounders that may interfere with chemometric analysis such as LCMS technical variability, environment, OSC can significantly improve PLS-DA performance, yielding a better discrimination of the clusters.33 Because of the wide dynamic range typically observed in MS-based metabolomic, these analyses were classically performed on Log-transformed data,34 submitted to a Pareto scaling procedure also of common use in the field of MS-based metabolomic.35 The quality of the generated OSC-PLS-DA and PLS-DA models was evaluated by several goodness-of-fit parameters: R2 (X), the proportion of the total variance of the dependent variables that is explained by the model, R2 (Y), defining the proportion of the total variance of the response variable explained by the model (i.e., the class of the samples), and the predictive ability parameter Q2(Y), which was calculated by a seven-round internal cross-validation of the data. In addition, a permutation test (n = 100) was carried out to validate and to test the degree of over fitting for OSC-PLS-DA and PLS-DA models.33 The correlation coefficient between the 3294

dx.doi.org/10.1021/pr2003193 |J. Proteome Res. 2011, 10, 3292–3302

Journal of Proteome Research original Y and the permuted Y is plotted against the cumulative R2 and Q2 and a regression line is calculated. Generally, the R2and Q2-intercept limits (when the correlation coefficient is zero) for a valid model should be less than 0.4 and 0.05, respectively.36

TECHNICAL NOTE

Table 3. Significant [m/z, rt] Features Revealed by Statistical Analysis for the Three Conditions Tested (t-Test p-Value 1.5) between R and C groups, suggesting that maternal protein restriction during gestation induced significant alterations in the offspring’ metabolome compared to control pups. Then, the proportion of significant [m/z, rt] features between CC and RR and between RC and RR increased similarly from PND5 (both 4%) to PND16 (14 and 11%) then decreased at weaning (8 and 5%). At PND260, the proportion of [m/z, rt] features significantly different between CC and RR continued to fall to 0.07% suggesting that the slow postnatal growth of the offspring during the first few weeks of their postnatal life had a long-lasting effect on plasma metabolome by erasing metabolic differences with control pups at adulthood. We noticed that the proportion of significant [m/z, rt] features, having a retention time beyond 20 min, which corresponded to essentially nonpolar compounds such as lipids, varied from 45% at birth, to 2038% from PND5 to PND22 (data not shown) and dramatically fell to 0% at adulthood. The MVA analysis was performed on LCMS data which had both a reliable criterion from the statistical point of view (p-value 1.5 in RR vs CC, RC vs CC and RR vs RC for each age). As the primary aim of our study was to capture metabolic variations in the offspring specifically associated with perinatal dietary regimen during early postnatal life, a PLS-DA analysis was performed on the 835 selected variables and 92 samples (for 12 classes, 3 nutritional groups and four PND, from PND5 to PND22). This initial model depicted, on the first two components, a stage of development-related mean trajectory of samples from PND5 to PND16 with a break at weaning but failed to clearly separate the three nutritional groups (CC, RC and RR) at each PND, due to the complexity of the data (data not shown). In an attempt to improve this analysis, an OSC filter was carried out before the PLS-DA which removed 2 components representing 78% of the variation in the original data set (eigenvalue 60 and 10 for PC1 3295

dx.doi.org/10.1021/pr2003193 |J. Proteome Res. 2011, 10, 3292–3302

Journal of Proteome Research

Figure 1. OSC-PLS-DA score plot of component 1 vs component 2 showing plasma LC-MS metabolomic fingerprints changes in the three groups of rats (RR, RC, CC) at PND5, PND12, PND16 and PND 22 together (A) or independently of the time period of postnatal development (B), (red b) CC for control diet (20% protein during gestation and lactation); (black 9), RC for restricted diet, containing 8% protein during gestation and control diet during lactation and (blue [) RR for restricted diet during gestation and lactation.

and PC2). The resulting OSC-PLS-DA score plot (Figure.1A) showed a two significant components model, characterized by R2X, R2Y and Q2 values of 0.51, 0.17 and 0.16, respectively, and with no outlier, but with a higher RR separation from both RC and CC groups, especially from PND12 to PND22, with satisfactory permutation test values of R2 and Q2 intercepts for RR classes (respectively, 0.007 and 0.077 for RR class at PND16 for example). This result was confirmed in a new OSC-PLS-DA model, constructed solely with the three perinatal nutritional status classes independently of PND. The latter model was characterized by R2X, R2Y and Q2 values of 0.45, 0.28, and 0.25, respectively, and its score plot (Figure 1B) showed that the RR group was clearly separated from both CC and RC groups (permutation testing led to R2 and Q2 intercepts of 0.045 and 0.165, respectively). In a second step, the susceptibility of the metabolic phenotypes of the offspring to maternal diet was assessed by PLS-DA analysis without OSC filter of data set for each PND separately from PND0 to PND260. We chose to report only PLS-DA score plots at birth (Figure 2A), at PND16 (Figure 2B) and at adulthood (Figure 2C). The PLS-DA analysis allowed a clear discrimination of the IUGR pups at birth and of the three CC, RC and RR groups for all the PND studied with satisfactory cumulative R2X, R2Y, Q2 values and permutation testing results as shown in Table 4. During suckling period, the RR group was clearly discriminated on the first component (40% of variance) on the PND16-PLS-DA score plot (Figure 2B) whereas, the RC and CC groups were separated on the second component, albeit

TECHNICAL NOTE

more weakly (7% of variance) although the R2X, R2Y and Q2 values (0.46, 0.80 and 0.42), R2 (0.25) and Q2 (0.25) intercepts for RR class were satisfactory. In contrast, the PND260PLS-DA score plot (Figure 2C) showed that, at adulthood, it was the RC group that was clearly discriminated on the first component (61% of variance) with the RR and CC groups weakly separated on the second component (12% of variance) (R2X, R2Y and Q2 values of 0.88, 0.92 and 0.64, respectively) and with good R2 (0.65) and Q2 (0.43) intercepts for RC class. The HCA processing of the PND0 data set (PND03D heat map reported in Figure 2D) showed a cluster of 71 [m/z, rt] features, the abundances of which are high in all the RR pups and in 2 or 3 control rats. From PND5 to PND16, the abundance of the metabolites were similar in CC and RC pups and the RR group is clearly discriminated and characterized by a cluster of ions highly expressed in plasma metabolome. For example, at PND16, the HCA processing (PND163D heat map reported in Figure 2E) highlighted a cluster of 177 features that were highly abundant in RR group and low in both CC and RC groups. At weaning (data not shown), we observed a shift of some groups of ions which have a high abundance in the RC pups and a low similar abundance in CC and RR pups suggesting changes in plasma metabolome in the groups of pups which were more pronounced in adulthood (PND2603D heat map reported in Figure 2F), as a cluster of 121 [m/z, rt] features showed strong abundances in the group RR, similar to those in control group where as the RC group was characterized by 58 features which high abundances differed from RR and CC groups. Mass Spectrometry Reveals that Perinatal Nutrition Had a Widespread Impact on Pups Plasma Biochemistry

Metabolite identification is the most time-consuming task of a metabolomic experiment. To expedite this procedure, we used clusters of [m/z, rt] features characterized by quasi-molecular ions, and nonquasi-molecular ions, including in-source fragments (related to a loss of water and/or fragment coming from the loss of carboxyl moiety (46 Da), or even NH4 (17 Da) or adducts (Na (þ22 Da) adduct)) for the database research to hypothesize related structures. Although, this process remains a work in progress, the identification of discriminating features produced to date (summary in Table 5 and illustrated in Figure 3) suggest that observed differences between the three groups of pups during development and at adulthood were explained at least in part by the alterations in amino acid metabolism. However, structures of a high proportion of nonpolar compounds statistically significant remains to be elucidated. The statistical parameters of the identified biomarkers (t-test p-value, fold) are shown in Table 6. At PND0, we found that IUGR induced a significant decrease in the three branched-chain amino acids (BCAAs), leucine (71%), isoleucine (54%), and valine (49%), two cationic amino acids, arginine (24%) and histidine (13%), and proline (46%), a non essential amino acid, in IUGR pups, compared to control pups, whereas other amino acids (alanine, serine, glycine, glutamine, tryptophan, methionine, threonine) were unaltered. During early postnatal life, particularly at PND16, the decrease in valine (39%), leucine (33%), proline (29%) and arginine (46%) was still significant in RC pups but not in RR pups. The same feature was observed for two aromatic amino acids, (i) tyrosine (38%), and (ii) tryptophan (64%), an essential amino acid and a precursor of serotonin and niacin, and (iii) methionine (40%), the sole essential sulfur amino acid and methyl donor, a precursor of 3296

dx.doi.org/10.1021/pr2003193 |J. Proteome Res. 2011, 10, 3292–3302

Journal of Proteome Research

TECHNICAL NOTE

Figure 2. PLS-DA score plot for the first two components indicating discrimination of pups according to perinatal nutritional environment (A) at birth, (B) at PND16 and (C) at adulthood. (red b) CC for control diet (20% protein during gestation and lactation); (black 9), RC for restricted diet, containing 8% protein during gestation and control diet during lactation and (blue [) RR for restricted diet during gestation and lactation. Heat map of Hierarchical Clustering Analysis (HCA) on the variables with a significant p-value for the t-test between groups (D) at birth, (E) at PND16 and (F) at adulthood. Heat maps visualize global metabolic changes [(red 9) increase, (green 9) decrease, (black 9) no relative change] for each variable. The selected variables, defined as a paired variable (m/z: RT), were arranged in rows as a dendogram using a HCA algorithm as an unsupervised method for pattern recognition. Columns represent different pups.

Table 4. PLS-DA Model Summary for the Different Discriminations among Positive ESIMS Spectra of Plasma at Each PND model

components, k variables (%)a R2X (cum)b R2Y (cum)c Q2 (cum)d R2 intercepte Q2 intercepte 0.119

C vs R at PND0 (n = 10 C and 4 R)

2, k = 126 (12%)

0.612

0.816

0.416

0.451

CC v s RC vs RR at PND5 (n = 7 CC, 7 RC and 7 RR)

4, k = 129 (6%)

0.584

0.925

0.705

0.655

0.414

CC vs RC vs RR at PND12 (n = 8 CC, 8 RC and 7 RR)

4, k = 209 (9%)

0.590

0.938

0.780

0.660

0.340

CC vs RC vs RR at PND16 (n = 8 CC, 8 RC and 8 RR)

2, k = 363 (16%)

0.463

0.804

0.424

0.252

0.254

CC vs RC vs RR at PND22 (n = 8 CC, 8 RC and 8 RR)

4, k = 274 (12%)

0.673

0.832

0.636

0.673

0.222

CC vs RC vs RR at PND260 (n = 5 CC, 6 RC and 7 RR)

5, k = 103 (6%)

0.879

0.925

0.635

0.665

0.463

a

Percent of significant selected ions (exhibiting a p-value of the t-test below 0.05 and which were at least 50% more abundant in one group compared to an other group) from 1047 ions at PND0, 2291 ions at each stage (PND5, 12, 16, 22) and 1681 at PND260. b R2X (cum) represents the cumulative Sum of Squares (SS) of all the X’s explained by all extracted components. c R2Y (cum) represents the cumulative Sum of Squares (SS) of all the Y’s explained by all extracted components. d Q2 (cum) is an estimate of how well the model predicts the Y’s and relates to its statistical validity. e R2- ad Q2-intercept are the limits (when correlation coefficient is zero) of the regression line calculated with the original Y and the permuted Y plotted against the cumulative R2 and Q2, after a permutation test (n = 100) for the RR class.

cysteine and carnitine (along with lysine). In contrast, extended maternal protein restriction through lactation was associated with a 4 and 2-fold elevation in plasma histidine at day 5 and 1216, respectively, compared with control pups (the difference was, however, only significant at PND5). In addition, a doubling in plasma carnitine and a 50% increase in two of its acyl-derivatives, hexanoyl- and o-acetylcarnitine were observed at PND16 (with no significant alteration in propionyl- and palmitoylcarnitine) in RR pups, compared with control and RC pups. Weaning was associated

with a restoration of most plasma amino acids and carnitine in IUGR pups toward levels observed in control pups, except for histidine and tyrosine, which remained significantly higher and lower, respectively, and for acetyl-carnitine, which was increased 2-fold in RR rats, compared to CC and RC rats at PND22. Finally, upon adulthood, among the identified biomarkers, only carnitine (but not its acylderivatives) was found to be increased in RR rats compared to RC rats whereas, tyrosine levels were decreased in RC rats compared to CC and RR rats. 3297

dx.doi.org/10.1021/pr2003193 |J. Proteome Res. 2011, 10, 3292–3302

Journal of Proteome Research

TECHNICAL NOTE

Table 5. Summary of Plasma Metabolites Identified that Were Significantly (p value