We characterize the integrated response of a rat host to the liver fluke Fasciola hepatica using a combination of 1H nuclear magnetic resonance spectroscopic profiles (liver, kidney, intestine, brain, spleen, plasma, urine, feces) and multiplex cytokine markers of systemic inflammation. Multivariate mathematical models were built to describe the main features of the infection at the systems level. In addition to the expected modulation of hepatic choline and energy metabolism, we found significant perturbations of the nucleotide balance in the brain, together with increased plasma IL‐13, suggesting a shift toward modulation of immune reactions to minimize inflammatory damage, which may favor the co‐existence of the parasite in the host. Subsequent analysis of brain extracts from other trematode infection models (i.e. Schistosoma mansoni, and Echinostoma caproni) did not elicit a change in neural nucleotide levels, indicating that the neural effects of F. hepatica infection are specific. We propose that the topographically extended response to invasion of the host as characterized by the modulated global metabolic phenotype is stratified across several bio‐organizational levels and reflects the direct manipulation of host–nucleotide balance.
Fascioliasis is a re‐emerging zoonotic disease caused by two liver fluke species, Fasciola hepatica and F. gigantica. As many as 17 million people might be infected with Fasciola spp. (Keiser and Utzinger, 2009) and the disease is of considerable economic and public health importance. At present, triclabendazole is the only drug available for treatment of infected human beings and livestock and worryingly, parasite resistance to triclabendazole is already widespread in farm animals (Fairweather, 2009). Hence, a deeper understanding of the biology of this parasite is warranted to uncover novel therapeutic targets. Fundamental investigations in well‐controlled host–parasite models hold promise for recovery of metabolic biomarkers, at different bio‐organizational levels. This may promote the mechanistic understanding of the host response to infection at the systems level with a view to identifying drug targets.
The regulation of any mammalian system is such that events originating in a discrete tissue or organ can trigger a cascade of consequent events throughout the host system as the host attempts to maintain homeostatic equilibrium (Lederberg, 2000; Nicholson et al, 2002). Thus, the effects of a parasitic infection are rarely confined to a single target tissue; instead a network of molecular events can generally be detected throughout the host system. Top‐down systems biology driven by metabolic phenotyping and metabonomics (Nicholson et al, 1999; Holmes et al, 2008a; Li et al, 2008a, 2008b; Tsang et al, 2009) has been shown to be a useful tool for studying host–parasite (Saric et al, 2008; Wang et al, 2008; Li et al, 2009) and host–symbiont interactions (Nicholson et al, 2005; Martin et al, 2007; Holmes et al, 2008b), and it is clear that any transgenomic interaction can only be studied in vivo, as the effects of microbial or parasitic modulations are panorganismal (Wang et al, 2008; Saric et al, 2009).
In this study, we characterize and interpret the system‐wide effect of F. hepatica in the rat, using a metabolic profiling strategy to develop mechanistic hypotheses. A multilevel statistical approach was applied to the analysis of 1H nuclear magnetic resonance (NMR) spectroscopy‐generated data to obtain an integrative profile of the global response of the host based on differential metabolic, immunological, and biometric responses across multiple tissues. We report novel findings regarding the effect of F. hepatica on the neural metabolite profile of the rat and show metabolic connections between the liver, gut, and brain, which we compare with the metabolic effects elicited by two separately assessed trematode infections, namely Schistosoma mansoni and Echinostoma caproni in a mouse model. Although F. hepatica‐induced liver damage has been described in detail before (Lim et al, 2007; Marcos et al, 2008), no direct association with the central nervous system has been found in the extant literature. The findings from this study expose a potential mechanism of parasite‐induced immune modulation and exemplify the value of top‐down systems approaches based on metabolic profiling for recovering mechanistic information from a system without a priori knowledge.
Results and discussion
Parasite burden, physiological monitoring, and histology
Patent infection of rats with F. hepatica resulted in a mean parasite burden of 5.5 (range=2–10; s.d.=2.6) on day 77 postinfection and a significant reduction of the mean packed cell volume (PCV) at day 71 postinfection (uninfected control rats: mean=50.8%, s.d.=2.8%; infected: mean=42.0%, s.d.=3.8%). The mean bodyweight did not significantly differ between infected and uninfected rats at any of the investigated time points. Clear evidence of hepatic necrosis was found in F. hepatica‐infected rats (Supplementary Figure S1) together with follicular hyperplasia of splenic white pulp and interstitial lymphoplasmocytic inflammatory infiltrates in the kidney.
Multivariate statistical modeling of F. hepatica infection
Assignment of metabolic components in the tissues and biofluid profiles for both uninfected and infected rats were made on the basis of extant literature (Coen et al, 2003; Holmes et al, 2006; Beckonert et al, 2007; Li et al, 2009; Saric et al, 2009; Tsang et al, 2009), and in the case of the brain extract spectra, from addition of authentic standards (assignments provided in Supplementary Figure S2; Supplementary Table SI).
The metabolic changes in the rat caused by patent infection with F. hepatica were modeled separately for each biological matrix through principal component analysis (PCA) and projection to latent structure discriminant analysis (PLS‐DA). In brief, the metabolic profiles of F. hepatica‐infected rats could be differentiated from those of uninfected control rats through PCA for all assessed biological compartments except the ileum and the renal medulla. A list of important infection‐discriminatory metabolites is given in Table I together with the P‐values, validated by permutation testing, Figure 1.
Consistent changes in lipid metabolism, including elevated choline and/or choline derivatives such as betaine, phosphocholine, and glycerophosphocholine (GPC) occurred throughout most tissues (Figure 1; Table I). This indicates a generally increased usage of choline‐derived metabolites that can be further converted to phosphocholine (Zeisel et al, 2003; Li and Vance, 2008), a basic component of membrane anabolism, or used for the production of polyunsaturated fatty acids such as arachidonic acid, which gives rise to pro‐inflammatory eicosanoic mediators (Calder, 2008, 2009). Prostaglandins, for instance, exert multiple pro‐inflammatory effects, such as chemotaxis of neutrophils, which are typically the first infiltrating cells at sites of tissue damage. Evidence of inflammatory infiltrates was detected in the histological and metabolic profiles of the liver, spleen, and kidney (Supplementary Figure S1) and inflammation was implicated biochemically in the colon and plasma. Increased plasma intensities of O‐ and N‐acetyl glycoprotein signals were noted in F. hepatica‐infected rats. Increased concentrations of acetylated glycoproteins have been earlier identified in Trichinella spiralis‐ and Trypanosoma brucei brucei‐infected mice (Martin et al, 2006; Wang et al, 2008) as inflammatory markers and include acute phase reaction proteins, such as α1 acid glycoprotein, haptoglobin, and transferrin, which are produced in the liver and which show markedly increased serum levels during such systemic responses to inflammation (Schreiber et al, 1982; Bell et al, 1987a, 1987b) (Figure 1; Table I).
Other features of F. hepatica infection were specific to a single biological compartment; for example hippurate, which is only observed in the urine, was found to be decreased in infected rats, reflecting a parasite‐induced disturbance of the gut‐microbial composition or activity (Nicholls et al, 2003; Li et al, 2008a, 2008b). The liver, spleen and plasma showed the greatest changes at the low molecular weight level, consistent with the direct damage caused by the migrating larvae of F. hepatica during the acute stage of an infection in the hepatic tissue and the immunological activity in the spleen (Supplementary Figure S1). The proliferation of B cells in the spleen primary follicles as response to blood‐borne antigens uses amino acids and lipid fractions for the new cellular membrane bilayers and organelles and may account for the major changes in the metabolic profile of the spleen, and perhaps in other tissue compartments. The depletion in hepatic glucose is more likely to be related to the direct damage of the energy stores and is consistent with elevated glucose levels in plasma.
One of the strongest effects of infection with F. hepatica manifested in the neurochemical profiles, further highlighting the ability of the parasite to impact on remote tissue. Clear visual separation of the infected and uninfected animals was achieved in the PCA and PLS‐DA scores plots (Figure 2A and B) of the neural profiles, with a model predictivity (Q2Y) of 0.46. The pattern of differentiating metabolites in the neurochemical profiles as extracted from the orthogonal PLS‐DA (O‐PLS‐DA) coefficients was composed of a significant increase in the relative levels of inosine, tyrosine, and phenylalanine. Conversely, the relative tissue concentrations of GPC, succinate, inosine mono‐, di‐, and triphosphate, adenosine, and adenosine mono‐, di‐, and triphosphate were lower in the brains of infected animals (Figure 2C and D).
Secretion of nucleotide‐degrading enzymes occurs in a variety of ecto‐ and endoparasitic organisms, such as ticks, blood sucking insects, and helminths and serves the primary purpose of minimizing immune reactions in the host organism to prolong their survival (Parshad and Guraya, 1977; Ribeiro and Francischetti, 2003). For example, adenosine deaminase and 5′‐nucleotidase have been found in secretory channels of T. spiralis and F. gigantica (Gounaris, 2002; Ali, 2008). The former catalyzes the intracellular conversion of adenosine to inosine, whereas the latter hydrolyzes inosine monophosphate (Figure 2E).
We observed a substantial shift of adenosine and phosphorylated nucleotides to inosine in the host brain, which indicates an attenuated inflammatory response and may be suggestive that F. hepatica, similarly to F. gigantica, secretes nucleotide‐degrading enzymes. Both inosine and adenosine induce mast cell degranulation, an event that has a central function in localized inflammatory response, an important mediator of type I hypersensitivity reactions, with adenosine being the more potent activator (Marquardt et al, 1978; Tilley et al, 2000). Furthermore, inosine, which was believed to be inert for a long time, has been proven to effectively suppress pro‐inflammatory cytokines such as IFN‐γ, TNF‐α, and IL‐12 (Hasko et al, 2000, 2004; Mabley et al, 2003) in vitro and in vivo.
We assessed the concentrations of selected plasma cytokines and found the levels of both IL‐13 and IL‐5 to be significantly higher in F. hepatica‐infected rats compared with the control group on day 22 postinfection. IL‐13 maintained a significant increase on day 43 postinfection (Figure 2E). An enhanced Th2 immune response such as that suggested by the increase in both IL‐5 and IL‐13 would serve to counteract mechanisms of inflammation. IL‐13 is also an important direct negative regulator of inflammatory cytokines in macrophages (Minty et al, 1993). Minimization of such an intense immune reaction, and hence prolongation of the period in which the parasite can remain undetected by the host at earlier stages of infection, is clearly beneficial for the survival of the worms.
Comparison of the systemic effects of F. hepatica with S. mansoni and E. caproni
To ascertain the specificity of the global metabolic response of the rat to F. hepatica across different tissue compartments, particularly with respect to the observed modulation of the neural nucleotide balance, two further rodent‐trematode models were compared, namely S. mansoni and E. caproni, in a murine host. Both trematodes induced a biochemical response across a range of biological compartments, but whereas S. mansoni caused pathology in the spleen and liver (Li et al, 2009), no overt necrosis was found in any tissue after E. caproni infection (Saric et al, 2009). At a global level, E. caproni infection was predominantly characterized by malabsorption of amino acids, which manifested predominantly in the biochemical profiles of the small intestine and liver. Altered amino‐acid levels and energy metabolism were also a core signature of S. mansoni infection with disturbed levels of choline‐containing membrane components in the liver and intestinal tissues.
As the altered neural nucleotide signature was one of the strongest metabolic responses in rats infected with F. hepatica, we directly compared the metabolic profiles of brain extracts with those from S. mansoni and E. caproni in the mouse. S. mansoni infection induced obvious changes in the metabolic composition of the host brain, as indicated by the PCA and PLS‐DA scores plots (Supplementary Figure S3), with the PLS‐DA model indicating a predictive value (Q2Y) of 0.61. Infection with S. mansoni was characterized by higher levels of glutamine, but lower concentrations of γ‐aminobutyric acid (GABA), choline, phosphocholine, and scyllo‐inositol (Supplementary Figure S3C). The E. caproni‐infected animals could not be statistically differentiated from the corresponding control group based on the brain extract spectra (Q2Y=−0.21, data not shown), indicating that the infection did not systematically alter the brain biochemistry of the host.
Although both S. mansoni and F. hepatica induced a marked perturbation of the neurochemical profiles in their host animals, the patterns of differentiating metabolites were distinct for these two trematode‐rodent models, with the exception of the common effect of decreasing concentrations of lipid degradation products, such as choline, phosphocholine, and GPC, all of which have a function in cell membrane formation. In contrast to the perturbed immunological function suggested by the altered nucleotide levels in the brains of F. hepatica‐infected animals, modulation of the neural metabolite profile by S. mansoni infection suggested the dominant effect related to the neurotransmitters, as supported by depleted levels of GABA and increased tissue concentrations of glutamine, which is a substrate for both excitatory and inhibitory neurotransmitters, including GABA. However, as neural GABA decreased in infected animals, the increased concentrations of glutamine are more likely to be indicative of a detoxification process. Compensatory locations for ammonium detoxification are muscle cells and the brain, combining ammonia and glutamate to glutamine (Kreis et al, 1991; Olde Damink et al, 2002), which could also contribute to the relatively higher levels of glutamine in the brain extracts of S. mansoni‐infected mice documented in the current investigation.
Systems impact of the modulation of the neural metabolic phenotype of F. hepatica infection: multivariate modeling and integration of physiological compartments
Metabolic correlates of the six important discriminatory spectral regions containing inosine, GPC, succinate, tyrosine, phenylalanine, and the overlapped regions containing adenosine and the phosphorylated nucleotides in the brain tissue of F. hepatica‐infected rats were identified in each of the other tissue or biofluid matrices. Sequential correlations between the integrated‐discriminatory signals in the brain spectra tissue were made with the whole spectra from each tissue and with the spectra from biofluids obtained at day 71 postinfection (Figure 3; Supplementary Table SII). Here, univariate correlation of single signals was preferred over a multivariate approach to maximize information recovery and to infer direct inter‐compartmental links between the metabolites. A comprehensive list of compartment‐specific correlations with the neural metabolites is provided in Supplementary Table SII. The liver, urine, and spleen showed the highest number of components correlated with the selected brain metabolites. The high degree of correlation between the metabolic markers in the brain and the liver spectra across the tissue compartments (Figure 3A) may simply relate to the magnitude of spectral changes in both organs and does not necessarily imply any causal relationships between metabolites. The gross changes in the hepatic metabolite profile are reflective of direct mechanical damage of the liver tissue by the migration and feeding of the juvenile fluke, whereas the effect of the parasite on remote organs such as the brain is more difficult to rationalize. Hepatic dysfunction in F. hepatica infection is known to result in increased circulating toxins, such as ammonia, thiols, and phenols (Zaki et al, 1983), which are typically released after hepatic failure, because of reduced ability to degrade aromatic amino acids and ammonia. Animals infected with F. hepatica showed increased neural concentrations of phenylalanine and tyrosine. Indeed, it has been shown that hepatic failure induces an increased passive permeability of the blood–brain barrier for several substances, among which phenylalanine and tyrosine were found to increase up to 30% in the host brain (Zaki et al, 1984).
A recurrent theme across multiple levels of this host–parasite system was the inflammatory response, which manifested at both the level of structural damage and that of the metabolite signature. The increased signal intensities of plasma acetyl glycoprotein fragments were statistically associated with all six cerebral markers of infection underscoring the presence and global effect of a strong infection‐induced inflammatory response (Figure 3B; Supplementary Table SII). Interestingly, a shift has been observed in the brain nucleotides inosine and adenosine in favor of inosine, which has been shown to have anti‐inflammatory effects at the posttranscriptional level (Hasko et al, 2000), including suppression of pro‐inflammatory cytokines (e.g. IFN‐γ and TNF‐α) and minimizing macrophage‐mediated mechanisms of inflammation (Hasko et al, 2000, 2004; Liaudet et al, 2002). This shift may be directly induced by nucleotide‐degrading secretory enzymes of F. hepatica. Significant direct correlations between the selected neural indices of F. hepatica infection and compounds involved in lipid metabolism were detected.
In addition to the correlations driven by the liver pathology, direct metabolic interactions such as plasma glucose being anti‐correlated with brain succinate and 2‐ketoglutarate, suggestive of an interaction through glycolysis and the tricarboxylic acid cycle, were found. Another example of such co‐variation is the inverse correlation between cerebral tyrosine and urinary hippurate, deriving from gut‐microbial/mammalian co‐metabolism. Direct alteration of urinary metabolite signatures through gut‐microbial species seems to be a common feature of many host–parasite models (Martin et al, 2006; Li et al, 2008a; Saric et al, 2008; Wang et al, 2008). Here, the negative correlations of cerebral adenosine with urinary dimethylglycine and the bile acids, which are co‐metabolized by gut microbiota, strengthens the notion of a bidirectional communication between gut and brain (gut–brain axis). This leads to the hypothesis that nucleotide‐initiated immunoactivity has a function in the changed gut‐microbial dynamics observed in parasitic infections.
In conclusion, we have characterized the global metabolic phenotype of a host–parasite system and have shown a clear effect of a trematode infection on the biochemical composition of the host brain using a metabolic profiling strategy to develop and pursue hypotheses relating to the observed neurochemical changes in the rat. F. hepatica induced a focussed response, primarily associated with the hypothesized worm‐induced shift from adenosine toward inosine and the subsequent induction of anti‐inflammatory cytokines. The coherence of the histological, metabolic, and cytokine data further facilitated elucidation of general and specific metabolic events and provided a means of probing inter‐compartmental co‐variation of metabolites. Thus, the application of a top‐down systems approach has been shown to be of value in driving the articulation of novel mechanistic hypotheses relating to parasitic invasion and has wide application in molecular parasitology.
Materials and methods
F. hepatica‐rat model and experimental design
Experimental procedures were carried out at the Swiss Tropical and Public Health Institute (Swiss TPH; Basel, Switzerland), adhering to local and national guidelines of animal welfare (permission no. 2070 and 2081). A total of 24 Wistar female rats were purchased from RCC (Itingen, Switzerland) and kept under environmentally controlled conditions (temperature: 25°C; relative humidity: 60–70%; light/dark cycle: 12/12 h). Rats were acclimatized for 1 week and all animals had free access to commercially available rodent diet obtained from Nafag (Gossau, Switzerland) and community tap water.
Rats were individually marked and group housed with four animals per cage. Twelve rats were orally infected with 20–25 F. hepatica metacercariae each (Cullompton isolate) obtained from Mr G Graham (Addleston, UK). The remaining 12 rats were left uninfected and served as controls.
The body weight of each animal was measured throughout the experiment. On day 1 pre‐infection and days 1, 4, 8, 15, 22, 28, 36, 43, 57, and 71 postinfection, urine and feces were collected into Petri dishes by rubbing the abdomen of the rats gently. Approximately 50 μl of blood was collected from the tail tips collected into hematocrit capillaries (Sodium [Na] heparin coated) and spun at 10 500 g for 4 min. The PCV was calculated and expressed as a percentage of the plasma ratio to red blood cells (Li et al, 2008a).
All animals were euthanized on day 77 postinfection using CO2, and the worm burden was determined in each infected animal by removing adult flukes from the livers and bile ducts on dissection. The whole brains of rats, regardless of their infection status, were removed on dissection and the left hemispheres were transferred into cryo‐tubes, snap frozen in liquid nitrogen and stored at −80°C pending 1H NMR spectroscopic data acquisition. In addition, the left lateral lobe of the liver, the left kidney, spleen, and three parts of the intestine (colon, ileum, and jejunum) were removed and stored in the same manner as the brain for 1H NMR spectroscopic data acquisition. The right brain hemispheres, the right kidney, and parts of the spleen, liver, and intestines were transferred into separate Eppendorf tubes containing 4% buffered formalin for subsequent histological examination.
Methodology relating to the S. mansoni‐ and E. caproni‐mouse models has been described earlier (Saric et al, 2008, 2009; Li et al, 2009) and followed a similar protocol to that described for F. hepatica infection in the rat.
Sample preparation for 1H NMR spectroscopy
Each brain sample was placed in a mortar, mashed using a 1 ml mixture of H2O and acetonitrile (1:1, v:v) and transferred into a glass tube. Another 2 ml of the solvent mixture was used to rinse the mortar twice and transferred into the same glass tube. The brain homogenate was centrifuged for 6 min at 10 000 g. The supernatant was collected into a new glass vial, evaporated overnight and lyophilized. The resultant dry mass was resuspended into 0.55 ml of D2O until completely dissolved and transferred into a 5 mm‐outer‐diameter NMR tube for subsequent analyses. Urine, plasma, and fecal pellets were prepared for conventional high‐resolution 1H NMR data acquisition as described earlier (Saric et al, 2008), whereas tissue metabolic fingerprints were acquired through magic angle spinning (MAS) 1H NMR (Saric et al, 2009).
The extraction protocol for the S. mansoni‐mouse model was slightly extended to gain additional information regarding potential changes in the lipid metabolic profile, as a pharmacological intervention was involved in the original study protocol. Each brain sample obtained from mice infected with S. mansoni and the corresponding uninfected control mice was transferred into a 2 ml Eppendorf tube containing a metal bead, 0.75 ml of water, and 0.75 ml of methanol. The Eppendorf tube was placed in the tissue lyser and shaken for 5 min at the speed of 22 Hz. The resulting brain homogenate was transferred into a glass tube. A further 0.75 ml of each liquid was used to rinse the Eppendorf tube and transferred into the same glass tube. A total of 1.5 ml of chloroform was added into the mixture and centrifuged at 2500 g for 30 min. The aqueous and the chloroform phases were transferred into a new glass tube each and both were left to evaporate overnight and lyophilized. Before 1H NMR analysis, the powder obtained from the aqueous phase was dissolved in 0.55 ml phosphate buffer (D2O:H2O=9:1, v:v, 0.01% of sodium 3‐(trimethylsilyl) propionate‐2,2,3,3‐d4 ([TSP)], pH=7.4), whereas the dry mass of the chloroform fraction was dissolved in deuterated chloroform (CDCl3).
Acquisition of 1H NMR spectral data
All 1H NMR spectra from rat and mouse brain extracts, urine, plasma, and fecal water were recorded on a Bruker Avance 600 NMR spectrometer (Bruker; Rheinstetten, Germany), operating at 600.13 MHz for proton frequency. A 5‐mm triple resonance probe with inverse detection for proton frequency was used (Bruker), using a standard one‐dimensional NMR experiment with pulse sequence (recycle delay (RD)‐90°‐t1‐90°‐tm‐90°‐acquisition time). Optimal water suppression was achieved by irradiating the water frequency during the RD set to 2 s and the mixing time of 100 ms. The 90° pulse length was adjusted to 10.62 μs for the brain extracts of both the F. hepatica rat and the E. caproni‐mouse models and 13.5 μs for the S. mansoni‐mouse model. A total of 128 scans were acquired for each sample into 32k data points using a spectral width of 20 p.p.m. All free induction decays were multiplied by an exponential function equivalent to a 0.3 Hz line‐broadening factor before Fourier transformation. Liver, spleen, kidney, and intestinal samples were acquired by an MASm probe, an NMR semi‐solid state method using the parameters described elsewhere (Saric et al, 2009).
Assignments of the spectral peaks were made from the literature (Nicholson et al, 1995; Coen et al, 2003; Li et al, 2008a, 2009; Saric et al, 2008) and confirmed through statistical total correlation spectroscopy (STOCSY) (Cloarec et al, 2005), and through standard two‐dimensional NMR experiments conducted on selected samples, including correlation spectroscopy (Ernst et al, 1986), total correlation spectroscopy (Bax and Davies, 1985), and J‐resolved NMR spectroscopy (Foxall et al, 1993). For selected brain samples, high‐performance liquid chromatography (HPLC) was performed to isolate chemical components relating to unidentified‐discriminatory metabolites with spectral regions at chemical shifts of δ 6.15 (doublet), δ 8.60 (singlet), δ 8.27 (singlet), and δ 4.02 (multiplet). For this purpose, fractions were separated using an Agilent 1100 HPLC with Eclipse XDB‐C18 column, 5 μm, 4.6 × 150 mm (Agilent), attached to an Esquire 6000 mass spectrometer (MS; Bruker; Rheinstetten, Germany). An aliquot of 100 μl of brain extract was injected at a flow rate of 1 ml/min and signals were detected by UV at 214 nm and by MS in positive mode. A gradient of 98% water to 2% acetonitrile was used for the first 25 min and a standard gradient sequence was applied for the remaining 25 min. Fractions were collected over the first 8 min and then lyophilized and resuspended in D2O for NMR analysis. Inosine, inosine monophosphate, inosine diphosphate, inosine triphosphate, adenosine, adenosine monophosphate, adenosine diphosphate, and adenosine triphosphate were confirmed by adding authentic standards purchased from Sigma Aldrich (Gilingham; Dorset, UK).
Data processing and analysis
The raw spectra were processed with an in‐house developed MATLAB script to reduce phase and baseline distortions and manually calibrated to the lactate CH3 signal at δ 1.33 (spectra of brain extracts, plasma, and tissue) or automatically to sodium 3‐(trimethylsilyl) [2,2,3,3‐2H4] propionate (TSP) at a spectral position δ 0.00 (urine and fecal water). The region containing the residual water resonance was removed to avoid any interference with the analysis (i.e. δ 4.55–5.40 in brain extracts, δ 4.2–8.0 in plasma, δ 4.63–5.15 in fecal extracts, δ 4.23–5.32 in urine, and a similar region of ∼δ 4.7–5.2 was removed in each tissue spectral dataset). Before statistical analysis, probabilistic quotient normalization was applied to all spectra and a peak alignment script was implemented in MATLAB to minimize chemical shift variation (Veselkov et al, 2009).
In a first step, PCA was applied to the processed spectral data to gain an overview of the degree of differentiation between infected and uninfected animals or inherent groupings in each tissue and biofluid spectral dataset (Eriksson et al, 2001). PCA reduces the multivariate data to a lower‐dimensionality scores plot, without requiring any earlier class information and delivers a snapshot of the similarity between observations (i.e. brain spectra) based on the linear combination of the spectral components. In a second step, PLS‐DA was applied, as a supervised method, which uses earlier information of class membership, and hence optimizes separation (Eriksson et al, 2001) and extraction of biomarkers associated with the infection.
O‐PLS‐DA is a further development of PLS‐DA applied for optimal recovery of biomarkers (Trygg and Wold, 2002), and removes systematic variation unrelated to infection status through an orthogonal filter. The O‐PLS‐DA model was built by using an X‐matrix containing all spectral information and a binary dummy matrix, as Y‐determining class affiliation, for example infected or controls. Sevenfold cross‐validation was applied (Bro et al, 2008) to validate the models and calculate the goodness of prediction Q2. To validate the statistical power of the discriminatory metabolites, the P‐values were calculated for each data point using 10 000 permutations and back projected to the O‐PLS‐DA plot, whereby regions that significantly differentiate infected from non‐infected animals at a level of P⩽0.05 are coded in red (Pitman, 1938).
For assessing the direct impact of the cerebral metabolic changes caused by F. hepatica infection, a further development of the STOCSY (Cloarec et al, 2005) and the statistical heterospectroscopy (Crockford et al, 2006) method has been applied between two data matrices whereby the selected regions occupied by each of the eight cerebral biomarkers were integrated in MATLAB and represent the Y‐matrix. The spectral information of each other biological matrix (e.g. biofluids and all other tissues) constituted the second matrix (X) for calculating correlation structures between the spectral peaks and the integrated metabolite information. To reduce the display complexity and to avoid including artifacts and noise, spectral peaks were picked by applying a threshold of 100 × the s.d. calculated from the first 500 data points, which consist of noise only. A 10 000‐fold permutation was embedded in the correlation script and only correlations with P⩽0.05 are displayed.
Cytokine multiplex quantification
Plasma from five F. hepatica‐infected rats and five uninfected rats from earlier NMR spectral preparation (e.g. 0.9% saline (NaCl) in D2O:H2O=1:1), over six different time points, were tested by a ‘Mesoscale Multiplex Assay’ (MS6000 Rat Demonstration 7‐Plex Ultra‐Sensitive kit, Meso Scale Discovery). The rat demonstration ‘7‐Plex Ultra‐Sensitivity kit’ was used for this purpose, which includes IFN‐γ, IL‐1β, IL‐4, IL‐5, IL‐13, KC/GRO, and TNF‐α. Each sample was assessed using duplicates and split in 15 μl per well in a 96‐well plate. The method was used according to the manufacturer's specifications (Meso Scale Discovery, Rat cytokine assays: Rat demonstration 7‐Plex Ultra‐Sensitive kit), with the exception of the incubation time, which was extended to 4 h to counteract the 1:1 dilution. An MSD Sector imager was used to read the plates (Meso Scale Discovery, Sector Imager 6000).
We thank Dr K Veselkov, S Robinette, and Dr I Yap for providing in‐house software and G Calvert from Meso Scale Discovery for providing expertise on the cytokine assay. This study received financial support from the Swiss National Science Foundation to JS, JVL, JU (project no. PPOOB‐102883 and PPOOB‐119129), and JK (project no. PPOOA‐114941), and from Chinese Academy of Sciences to YW (project no. KJCX2‐YW‐W11). Additional support was provided for JS by the Sir Henry Wellcome Fellowship (Wellcome Trust award number P23665).
Conflict of Interest
The authors declare that they have no conflict of interest.
Supplementary Tables SI–II [msb201049-sup-0001.doc]
Supplementary Figures S1–3
Supplementary Figures S1–3 [msb201049-sup-0002.pdf]
Dataset 1 [msb201049-sup-0003.xls]
Dataset 2 [msb201049-sup-0004.xls]
Dataset 3 [msb201049-sup-0005.xls]
This is an open‐access article distributed under the terms of the Creative Commons Attribution License, which permits distribution, and reproduction in any medium, provided the original author and source are credited. This license does not permit commercial exploitation without specific permission.
- Copyright © 2010 EMBO and Macmillan Publishers Limited