1H NMR Based Metabonomics of Human Amniotic Fluid for the

May 19, 2009 - Metabolic Characterization of Fetus Malformations ... Keywords: amniotic fluid • fetus malformations • NMR spectroscopy • chemome...
0 downloads 0 Views 776KB Size
1

H NMR Based Metabonomics of Human Amniotic Fluid for the Metabolic Characterization of Fetus Malformations

Gonc¸alo Grac¸a,† Iola F. Duarte,† Anto ´ nio S. Barros,‡ Brian J. Goodfellow,† Sı´lvia Diaz,† § Isabel M. Carreira, Ana Bela Couceiro,| Eula´lia Galhano,| and Ana M. Gil*,† CICECO, Department of Chemistry, Campus Universita´rio de Santiago, University of Aveiro, 3810-193 Aveiro, Portugal, QOPNAA, Department of Chemistry, Campus Universita´rio de Santiago, University of Aveiro, 3810-193 Aveiro, Portugal, Cytogenetics Laboratory and Center of Neurosciences and Cellular Biology, Faculty of Medicine, University of Coimbra, 3001-401 Coimbra, Portugal, and Maternidade Bissaya Barreto, Centro Hospitalar de Coimbra, 3000 Coimbra, Portugal Received March 18, 2009

An NMR-metabonomic study of malformed fetuses was carried out through human amniotic fluid (HAF) analysis. Over 70 compounds were detected in control HAF by NMR. Possible confounding variables (fetus gender and gestational and maternal ages) were shown not to induce detectable compositional trends in the control group considered. Malformed fetuses showed variations in glucose, some amino acids and organic acids and proteins. In tandem with enzymatic assays, these NMR results suggest that changes in gycolysis and gluconeogenesis as well as kidney underdevelopment occur in the malformed fetuses studied here. Keywords: amniotic fluid • fetus malformations • NMR spectroscopy • chemometrics • metabonomics

Introduction Metabonomics is now a well-established field in systems biology,1 referring to the study of individual metabolic profiles and their changes over time due to disease, toxicity, nutritional or other effects.2-4 Nuclear Magnetic Resonance spectroscopy (NMR) and Liquid Chromatography coupled with Mass Spectrometry (LC-MS) have been used to study such variations, allowing screening of metabolites to be performed over a wide concentration range.1,5 The application of these methods to large numbers of samples generates complex data sets which require the use of pattern recognition methods for interpretation. Such methods include Principal Components Analysis (PCA) and other more specific methods such as Soft Independent Modeling of Class Analogy (SIMCA), Partial Least SquaresDiscriminant Analysis (PLS-DA) and its orthogonal variant OPLS-DA.6 An NMR-based metabonomics approach captures information on tens to hundreds of metabolites of different types, in one single analytical run, enabling rapid metabolite identification and quantitation to be performed. The increasing use of high field magnets and cryo-probes, often coupled with MS, gives significant increases in sensitivity and resolution.7,8 Biofluids and biological tissues are well-suited for individual metabolome evaluation, and blood plasma and urine are the biofluids most extensively studied by NMR metabonomics.9-14 The composition of human amniotic fluid (HAF) reflects both * To whom correspondence should be addressed. Phone, +351 234 370707; fax, +351 234 370084; e-mail, [email protected]. † CICECO, Universidade de Aveiro. ‡ QOPNAA, Universidade de Aveiro. § University of Coimbra. | Centro Hospitalar de Coimbra.

4144 Journal of Proteome Research 2009, 8, 4144–4150 Published on Web 05/19/2009

the mother’s health and fetal status and development. This has been recognized in previous NMR studies of HAF composition in order to establish correlations with disorders such as open spina-bifida,15,16 diabetes mellitus,17 Down syndrome,15 cystic fibrosis,18 fetal lung19,20 and kidney maturation.15 However, besides the limited number of samples considered in these studies, univariate analysis was the main spectral evaluation method while considering only particular resonances in the 1D NMR spectrum.15,16,19,21 This leaves out potentially valuable spectral information, unknown a priori, an aspect which is tackled when a metabonomics approach is employed. Recently, the compositional characterization of HAF has been carried out by high field 1H NMR and LC-NMR/MS, indentifying a total of 75 different compounds in HAF.22,23 This work presents the application of NMR-based metabonomics for the metabolic characterization of fetus malformations, through HAF analysis. Congenital malformations may result from genetic abnormalities, the intrauterine environment, errors of morphogenesis or chromosomal abnormality24 and most can be detected before birth with ultrasonography. In addition, chromosomal analysis and biochemical testing (e.g., for alpha-fetoprotein) performed on amniotic fluid allow the diagnosis of several disorders (including several trisomies) to be carried out. However, these methods are still associated with a nonzero percentage of false results and some of the procedures involved are often lengthy. A metabonomics approach, based on the rapid analysis of HAF, can provide metabolic knowledge which may support faster diagnostic methods, as well as providing fundamental metabolic insight into fetal malformation disorders. 10.1021/pr900386f CCC: $40.75

 2009 American Chemical Society

NMR-Metabonomics Characterization of Fetus Malformations

research articles

Figure 1. 1H NMR 500 MHz spectra of HAF from a representative healthy control sample: (a) standard 1D (noesy 1D) spectrum; (b) 1D T2 relaxation-edited (CPMG) spectrum; (c) 1D diffusion-edited spectrum. Peak assignments are based on previous work.22 1, leucine; 2, valine; 3, ethanol; 4, lactate; 5, alanine; 6, lysine; 7, acetate; 8, glutamate; 9, pyruvate; 10, glutamine; 11, citrate, 12, creatine; 13, betaine; 14, glucose; 15, threonine; 16, fumarate; 17, tyrosine; 18, histidine; 19, phenylalanine; 20, formate; 21, glycoprotein (P1); 22, protein (P2) and 23, glycoprotein (P3, probably albumin).

Experimental Section Samples. Amniotic fluid samples were obtained from pregnant women aged from 13 to 42, who underwent amniocentesis at the second trimester of pregnancy (15-24 weeks of gestation) for cytogenetic-based diagnosis. The use of the samples for research was carried out under ethical committee approval. Information about occurrence of fetal and maternal health problems, intake of medication, chronic maternal disease and maternal life style was obtained from the obstetrical and neonatal medical records, as well as from an individual questionnaire. For this work, HAF samples fell into two groups: controls (healthy pregnancies, n ) 51) and fetal malformations (n ) 12). After collection of HAF (15-20 mL), samples were centrifuged (177.4g, 22 °C, 10 min) and the supernatants collected and stored at -70 °C for up to 6 months until NMR analysis. For NMR, the samples were thawed at room temperature and 100 µL of 0.5 M sodium phosphate buffer pH 7.2 (in 100% D2O) were added to 500 µL of HAF. The mixture was then centrifuged (4500g, 25 °C, 5 min) and transferred into 5 mm NMR tubes. NMR Spectroscopy. NMR spectra were recorded on a Bruker Avance DRX 500 spectrometer equipped with an actively shielded gradient unit with a maximum gradient strength output of 53.5 G/cm. Sample temperature was set at 300 K and continuously controlled throughout the experiments. For each sample, three 1D spectra were obtained: a standard 1H NMR spectrum, a Carr-Purcell-Meiboom-Gill (CPMG) or T2weighted spectrum and a diffusion-edited spectrum. Standard 1D 1H NMR spectra were acquired using a noesy 1D pulse sequence (1D version of noesyphpr), with tm of 100 ms, a fixed 3 ms t1 delay and water suppression during relaxation delay and mixing time. 1D T2 relaxation edited 1H NMR spectra were

acquired using the CPMG pulse sequence (RD-90°-{τ-180°-τ}nacquire),25 with simple presaturation of the water peak and a total spin-spin relaxation time (2nτ) of 210 ms. 1D diffusionedited spectra were recorded using the bipolar pulse longitudinal eddy current delay (BPPLED) pulse sequence.26 To attenuate the signals from low molecular weight compounds, square gradients with duration of 2 ms and strength of 50.8 G/cm were used, with 200 ms diffusion time. All 1D spectra were acquired with 256 transients, 32k complex data points, 8012.82 Hz spectral width (SW) and 3 s relaxation delay with 2 s of acquisition time. Each FID was zero-filled to 64k points, multiplied by a 0.3 Hz (noesy 1D and CPMG) or 1 Hz (diffusion edited) exponential line-broadening function prior to Fourier transform. Spectra were baseline corrected using fifth degree polynomial functions and chemical shifts referenced internally to the R-glucose H1 resonance at 5.23 ppm. Enzymatic Assays. Spectrophotometric enzyme based assay kits (Megazyme) were used to quantify L-lactate, ammonia and urea. Paired samples of controls and malformations (n ) 12 for each group) were assayed carrying out 2 replicas for ammonia and urea and 3 replicates for L-lactate. Chemometric Analysis. Each set of spectra (noesy 1D, CPMG and diffusion-edited) was used to construct data matrices for the analysis, either using the full spectra or spectral regions. In the latter case, three spectral regions were considered for separate analysis: aliphatic (-1.5 to 2.80 ppm); sugar (2.80-5.50 ppm) and aromatic (5.50-12 ppm). In all cases, the water resonance region (4.50-5.00 ppm) was excluded and the urea resonance (5.52-6.00 ppm) was also removed from the noesy 1D spectra. Each full spectrum or spectral region was divided into “buckets” of fixed (diffusion edited spectra, 0.005 ppm) or variable sizes (noesy 1D and CPMG spectra), and integrated. Journal of Proteome Research • Vol. 8, No. 8, 2009 4145

research articles

Grac¸a et al.

Figure 2. 1H NMR CPMG average spectra from (a) control and (b) malformation HAF samples. Some differences are indicated by arrows.

All PCA was carried out using SIMCA-P 11.5 software and OPLSDA calculations were made using in-house software. This supervised method removes changes in the data by identifying orthogonal (unrelated) variations not related to the a priori established groups (e.g., random diet and lifestyle effects fall into this category). For OPLS-DA calculations, 1 latent variable was used and the corresponding models were validated using the cross-validation “leave-one-out” method. The full control group (n ) 51) was used for fetal gender, maternal and gestational ages evaluation and a reduced control group (n ) 34) was used in the malformations (n ) 12) study, to diminish discrepancy between group sizes. Spectral integrations were performed using the Amix 3.8.10 (BrukerBioSpin, Rheinstetten, Germany) software.

Results and Discussion 1

H NMR Spectra of Human Amniotic Fluid. For each HAF sample, three 1D 1H NMR experiments were acquired: a 1D standard experiment (noesy 1D), a T2-weighted experiment (CPMG) and a diffusion edited experiment. In Figure 1, representative spectra are shown with some assignments based on previous work.22 The standard spectrum (Figure 1a) shows both sharp and underlying broad peaks, arising from lower Mw metabolites (amino acids, sugars and organic acids) and higher Mw molecules (mostly proteins), respectively. In the CPMG spectrum (Figure 1b), only sharp peaks are observed, resulting from lower Mw metabolites as the experiment filters out signals arising from molecules with short T2’s. Conversely, the diffusion edited spectrum shows mainly broad signals (Figure 1c) arising from compounds with lower diffusion rates, that is, proteins, in the case of HAF. The narrower peaks also observed in Figure 1c arise from glucose, lactate and citrate possibly in close interaction with the protein fraction and, hence, with restricted diffusivities. Such effects have been observed in plasma, where citric and lactic acid can bind to glycoproteins to some extent.27,28 Effect of Gestational Age, Maternal Age, and Fetal Gender on HAF Metabolic Profile. Since the subjects will necessarily differ in gestational age, maternal age and fetal gender, it is important to know if these factors lead to any visible changes in the composition of HAF, in order to be able to recognize and evaluate any disorder-related changes. Some amino acids, glucose, choline, urea, proteins, enzymes and several ions are 4146

Journal of Proteome Research • Vol. 8, No. 8, 2009

known to change throughout the pregnancy as a response to fetal organ maturation;21,29,30 however, little is known about specific changes within periods as short as that considered in this study (15-24 weeks of gestation). Figure S1a (Supporting Information) shows the PCA scores plot obtained using the CPMG spectra for three different gestational age subranges within the 15-24 week period under study. The plot shows that no measurable trend is visible and this was confirmed when a supervised method such as PLS-DA was employed (data not shown). A similar analysis on the diffusion-edited spectra to investigate changes in the protein fraction (data not shown) again revealed a lack of consistent differences within the gestational ages considered. Although significant compositional changes may indeed be absent, this observation may also be due to the effects of variables such as different rates of fetal development for different subjects, which will lead to sample dispersion within the control group. In relation to maternal age, the PCA scores plot in Figure S1b also indicates the absence of measurable trends in HAF CPMG spectra, as a function of maternal age, with large spreads noted for each of the age subranges considered. An identical result was obtained for the diffusion-edited spectra (not shown). In addition, OPLS analysis was also performed in order to cancel out random effects due to uncontrolled variables such as diet or lifestyle, which might be masking possible compositional trends due to gestational or maternal age. Again, no trends were observed. To probe the effect of fetal gender on HAF composition, a similar approach was taken (data not shown), showing that HAF composition, as viewed by NMR spectroscopy under the conditions employed in this work, is not influenced by fetal gender. The Effect of Fetus Malformations on HAF Metabolic Profile. In Figure 2, the two average CPMG spectra of the control group and the group with fetal malformations are shown. The spectral profiles are generally similar but some relative intensity differences can be observed, as indicated by the arrows in the figure. PCA was carried out on each of the three different 1D 1 H NMR experiments and Figure 3 shows the results obtained for the CPMG and the diffusion-edited data sets. The results obtained with the standard spectra (not shown) were similar to those obtained with the CPMG spectra. The malformation cases (filled symbols) overlap with control samples, with a very slight grouping tendency of malformations cases toward nega-

NMR-Metabonomics Characterization of Fetus Malformations

Figure 3. PCA scores scatter plots of controls (0) and malformations (9) samples, as viewed by the 1H NMR spectra of HAF using (a) the CPMG experiment and (b) the diffusion-edited experiment.

tive PC2 in the CPMG spectra analysis (Figure 3a) and toward positive PC2 in the diffusion-edited spectra (Figure 3b). Such effects are, nevertheless, too weak to be significant. To minimize or remove variations not related to group distribution (orthogonal variations), OPLS-DA analysis was carried out using the CPMG and the diffusion-edited spectra (Figure 4). In the one-dimensional scores plot obtained, the first latent variable (LV1) separates the two groups rather satisfactorily with the malformations group being characterized by negative LV1 values (Figure 4a,b). The resulting models are characterized by “lack-of-fit” (or error of classification) values of 11% for CPMG model and 7% for diffusion-edited model, obtained from crossvalidation by the “leave-one-out” method. Models were also obtained for different spectral subregions (not shown), confirming the separation between the two groups and improving the identification of the varying NMR peaks. The b-vector loadings plots obtained for the CPMG and diffusion-edited spectra models (Figure 4c,d) enable the identification of the resonances responsible for the separation between groups visible in the scores plots (Figure 4a,b). In the CPMG analysis (Figure 4c), control samples are suggested to have relative increases in leucine (1), valine (2), ethanol (3), alanine (5), proline (6), glutamate (9) and glucose (14). On the other hand, the same group seems to show decreases in methionine (7), succinate (10), glutamine (11), citrate (12) and glycine (13). In the diffusion-edited analysis (Figure 4d), negative peaks 15 and 16, arising from glycoprotein P1 (Figure 1),22 indicate enhanced protein contents in malformation HAF samples. The remaining narrower signals observed in Figure 4d should correspond to low Mw metabolites with restricted diffusion, possibly due to interaction with proteins. With respect to lactate, a negative tendency is suggested by the higher field peak (peak 4, at 1.32 ppm) in the CPMG model (Figure 4c), although a strong first derivative effect hinders any definite interpretation. This would, however, suggest higher

research articles

lactate contents in malformation samples. However, in the diffusion-edited model (Figure 4d), lower lactate contents are found for the same group. For citrate, a similar behavior is noted. Earlier, it was suggested that two dynamic environments occur for lactate, citrate and glucose: one with higher mobility, thus giving enhanced signals in the CPMG spectrum, and one with restricted mobility due to interaction with proteins, preferentially reflected in the diffusion-edited spectrum. In the case of lactate, the total lactate content was evaluated by enzymatic assays and found not to change significantly between controls and malformations samples (0.70 ( 0.15 and 0.71 ( 0.16 mg/mL, respectively). This suggests that the decrease in protein-interacting lactate noted in Figure 4d for malformation samples is compensated by an increase in free lactate. If so, this would lead to higher (free lactate/interacting lactate) ratios (vide infra). Table 1 lists the metabolite variations found for each sample in the malformations group, obtained by integration and normalization to total spectral area. For each compound, peaks with no or little overlap were selected, in either CPMG or diffusion-edited spectra, for controls and malformation cases. This will not give absolute metabolite concentrations but will enable confirmation of the changes in Figure 4 and evaluation of metabolite variations. The integrals obtained for the malformations group were compared with the average integrals of the controls and a significant deviation (in %) was considered when it exceeded the 95% confidence interval calculated for the controls. It should be noted that the malformation group, still of limited size, comprises malformations of different types: uro-genital, soft tissues, cardiac, polimalformation, abdominal, central nervous system. However, Table 1 shows that many metabolite variations identified by OPLS-DA are shared by most malformation samples. In fact, glutamine, glycine, succinate, glycoprotein P1 and free lactate were found to be generally more abundant in malformation samples with respect to controls. Conversely, alanine, glutamate, leucine, phenylalanine, tyrosine, valine, glucose and protein-interacting lactate tend to have lower contents in malformations group. It is noted that the lactate changes observed indeed translate into higher [free lactate/interacting lactate] ratios in case of malformation. Changes in acetate, histidine, lysine, pyruvate and protein P2 appear to depend on malformation type but the limited sample number hinders any satisfactory conclusion. Metabolic Variations in Fetal Malformation Cases. The above results suggest that some consistent metabolic changes occur in malformation cases, irrespective of their organic nature. For instance, lower glucose and higher free lactate levels suggest that energy production is being conducted preferentially through glycolysis, under anaerobic (hypoxic) conditions. Aerobic metabolic energy production is carried out in the mitochondria through respiratory chain and oxidative phophorylation, and although the necessary enzymatic machinery is complete at 9-17 weeks gestation in the developing fetus, the process is not completely active until birth, when oxygen partial pressure and energetic demands rise.31 Indeed, fetal malformations have been correlated with a deficient function or regulation of the mitochondrial respiratory chain and of oxidative phosphorylation during fetal development.31,32 The higher content of succinate, a substrate of the mitochondrial respiratory chain (complex II), in malformation cases, is also consistent with the “under-use” of this pathway for energy production. In addition, gluconeogenic amino acids (alanine, glutamate, leucine, valine, tyrosine and phenylalanine) are Journal of Proteome Research • Vol. 8, No. 8, 2009 4147

research articles

Grac¸a et al.

Figure 4. OPLS-DA scores and b-vectors profiles for the CPMG spectra (a and c) and diffusion-edited spectra (b and d) 1H NMR spectra of HAF for controls (Ci) and malformation-affected subjects (Mi). Assignments: 1, leucine; 2, valine; 3, ethanol; 4, lactate; 5, alanine; 6, proline; 7, methionine; 8, acetone; 9, glutamate; 10, succinate; 11, glutamine; 12, citrate; 13, glycine; 14, glucose; 15 and 16, glycoprotein P1.

generally decreased in malformation cases, probably due to increased glucose production via gluconeogenesis, active around 16 weeks gestation in the developing liver.33 This is consistent with a need to replenish reduced glucose levels available to other tissues due to hypoxia. To evaluate if the increased consumption of gluconeogenic amino acids leads to increase in the nitrogenated amino acid byproduct ammonia and urea, enzymatic assays were performed for these metabolites. Even though the number of samples is still limited, tendencies for lower contents of urea and ammonia in malformation samples 4148

Journal of Proteome Research • Vol. 8, No. 8, 2009

are noted (Figure 5). Regarding urea, this may be related to fetal inability to synthesize urea until later in gestation.34 Urea can nevertheless diffuse through the placenta into fetal blood, being present in fetal circulation and thus reflecting kidney function in the fetus. Lower urea levels in malformation cases may be indicative of kidney underdevelopment and/or prerenal causes such as deficient blood perfusion thus causing hypoxia, as noted above. The weak tendency for lower ammonia levels (Figure 5), in malformations cases, may relate to the fact that higher

research articles

NMR-Metabonomics Characterization of Fetus Malformations a

Table 1. Variations (in %) of Metabolite Integrals in Malformation Samples Compared to Controls

a Values are indicated by ranges as follows: positive changes: +, 0-50%; ++, 50-100%; and +++, >100% and negative changes: -, 0-50%; - -, 50-100%; and - - -,