Multivariate modelling with ¹H NMR of pleural effusion in murine cerebral malaria

All the animals were treated in accordance with the guidelines set forth by the local animal ethics committee of TIFR.

25 female C57BL/6 mice of age 6 to 8 weeks and weight (24-25 gm) were housed in five cages (5 mice in each cage). They had free access to standard food pallets and water. The temperature was maintained at 22 ± 2°C and they were kept under 12 hour day night cycle. Twenty of these mice were injected inrtaperitonially with 10⁷ RBCs each infected with P. berghei ANKA. The remaining five mice served as uninfected controls. These parasites were maintained in Swiss female mice of 6 to 8 weeks. Rectal temperature of all the mice was measured twice a day with a digital thermometer.

Parasitaemia was monitored daily by counting number of infected cells per 1,000 RBCs in Giemsa-stained [30] slides of blood smear from tail bleeds. The mice were considered to have CM if they displayed one or more of the neurological symptoms such as ataxia, paralysis, convulsions, deviation of head etc or had body temperature < 34°C [31, 32]. When the prominent symptoms of the CM were visible on day 9 post-infection CM mice were anesthetized with ether and blood was collected by retro-orbital bleeding. The mice were dissected after this and the pleural fluid was collected by aspiration. The collection of the two biofluids (PE and sera) from two different regions of the body of CM mice ensures no mixing of the fluids. The PE collected from the mice was colorless. Among 10 mice that showed symptoms of CM, one mouse died during the course of the experiment and one exhibited too small a volume of PE compared to other CM mice, so their PE is not used for the experiment. The remaining eight animals were used for the study. The amount of fluid varied from 200 to 400 μl. The mice that did not exhibit neurological symptoms or a high body temperature > 34°C were considered as the non-CM (NCM) mice. The parasitaemia of the CM mice rose to 18 ± 2% during the time of sacrifice. The parasitaemia of NCM mice were comparable. In the control and NCM mice pleural fluid was collected by protocol mentioned in [33]. Briefly, the mouse was anaesthetized, a 30 μl gel loading tip was inserted through an inter-coastal space and the pleural fluid was withdrawn. In some animals 100 μl of 0.075 M phosphate buffer pH 7.4 was injected into the pleural cavity of the animal and after waiting for a minute the fluid was withdrawn.

The blood samples as well as the PE and pleural fluid were incubated in 37°C for 10 mins and were centrifuged for 10 mins at 13100 g. The supernatant was collected, frozen in liquid N₂ and stored at -80°C. For NMR experiments, 180 μl of sample (sera or pleural effusion) was mixed with 400 μl of the pH 7.4 buffer containing 0.075 M Na₂HPO_4.7H₂0, 4% NaN₃ and 0.02% DSS, 20%D₂O (the buffer recipes were provided by Bruker Biospin Metabonomic unit). In case of pleural fluid of NCM the final volume of the sample was maintained to 580 μl by adding 0.075 M phosphate buffer pH 7.4. The samples were placed in 5 mm NMR tube (Norrell Inc.).

One dimensional ¹H NMR spectra were acquired on AVANCE 700 MHz Bruker spectrometer with triple resonance probe using D₂O as the frequency lock at 310 K. The pulse sequence used has the form -RD-90°-(t-180°-t)_n-ACQ with presaturation at the water peak. Here RD implies a relaxation delay of 4 s; no of loop (n) being 128 and CPMG spin echo time (t used was 300 μs). The purpose of using CPMG spin echo is to attenuate the broad signals from the macromolecules that may be present in the sample. A total of 32 transients were collected into 88640 data points using a spectral width of 20.059 ppm. For pleural fluid of NCM the number of transients was increased to 512. The FID was subjected to exponential multiplication leading to an additional line broadening of 1 Hz, a Gaussian multiplication of 0.01 Hz and a sine squared bell apodization function prior to Fourier transformation. Spectra were phased and baseline corrected using the protocols provided by Bruker Biospin. The assignments of metabolites were based on 2D COSY and TOCSY spectra. In addition, human metabolome database [34] was also used for this purpose. For COSY experiments, 64 transients per increment and 256 increments were collected in direct and indirect dimension respectively. A QSINE function with 2048 and 1024 digital points were used for processing. Exponential multiplication of 0.20 and 0.30 Hz in the direct and indirect dimension respectively was used. TOCSY was processed using SINE function. Exponential multiplication of 0.20 and 0.30 Hz in the direct and indirect dimension was used respectively. In addition, a Gaussian multiplication of 0.1 Hz in the indirect dimension was applied for the TOCSY.

Spectral region of 0.5 to 4.14 ppm was bucketed into frequency window of 0.003 ppm. The region corresponding to water (4.5 to 5.5 ppm) was excluded during binning to avoid artifacts due to pre-saturation of water. The aromatic region was also excluded because the signal to noise ratio in this region was poorer compared to that of aliphatic region. The peaks (1.15(t), 3.35(q)) corresponding to ether were removed before binning. The resulting integrals were normalized to the working region (0.5-4.14) ppm of the spectrum to correct for inter-sample differences in dilution. The binning and the normalizations were achieved using AMIX 2.0 software. The matrix obtained in AMIX was imported to SIMCA P+ 12.0 (Umetrics AB, Sweden) and Solo 6.0 (eigenvector research incorporated software) for further multivariate data analyses.

MPCA helps us to understand the metabolic correlations between different organs and compartments of the body [41]. Since the two biofluids, serum and PE are present in two different compartments, MPCA would give metabolic correlations that could happen across these two compartments. MPCA will give us information of how the PE is correlated to the changes in the sera and vice versa during CM. MPCA is an extension of PCA and is used for higher dimension of data matrix. Here the N order data are decomposed into scores (t) and loadings (p) and the residual matrix (E). The N order of the data corresponds to number of compartments, which is two in this case; PE and serum. The residuals are considered to the non-deterministic part of the data matrix. MPCA models were prepared, to study the metabolic correlation of PE and serum. A 3D matrix was prepared in Matlab 7.0.1 where the first axis corresponds to spectral variables, the second axis corresponds to compartments (serum, PE) and the third corresponds to individual animals. The loading matrix contains the information of second and third mode [38]. The analysis of the MPCA is done using Solo 6.0 (eigenvector research incorporated software). The peaks in MPCA loading plot were assigned in accordance to the resonances in real ¹H CPMG spectrum of corresponding compartment.

Glucose quantification in all the sera and PE samples of CM mice and crude pleural fluid of uninfected C57BL/6 were done by glucose oxidase/peroxidase method using glucose quantification kit (Siemens) [42]. Briefly, 10 μL of the sample (serum or PE) was mixed with 1 ml of the "working solution" (95 mmol phosphate buffer, 0.3 mmol 4-aminoantipyrine, 5.9 mmol p-hydroxy benzoic acid, glucose oxidase ≥(5000 U/L), peroxidase ≥ (50000 U/L). The samples were incubated for 15 mins at 37°C. The intensity of the color was measured at 505 nm wavelength in UV-VIS spectrophotometer (Specord, Analytikjena). The concentration of glucose was calculated for all the samples and the significance was checked by Mann Whitney U test.

Relative levels of creatine (3.03 ppm; singlet) and glutamine (2.44 ppm; multiplet) and lipids (1.27 ppm; broad hump) were estimated for both sera and PE from the integrals of the corresponding resonances of the normalized ¹H NMR spectra. The significance of levels of metabolites in the two biofluids (sera and pleural effusion) was checked by Mann Whitney U Test.

C57BL/6 mice infected with malarial parasite P. berghei ANKA is a well established mouse model of experimental CM [43, 44]. The fractions of infected mice that transit into CM exhibit variation [16, 43, 45]. In our experiments we monitored this combination of mice and parasite over the time of disease progression. During several such trials a consistent observation was that 40-60% of the animals exhibited symptoms of CM. The grouping of a mouse as having CM, versus non-cerebral malaria (NCM) was based on daily measurement of rectal temperature and parasitaemia post-infection (p.i.) (Figure 1). In addition, clinical symptoms including ataxia, paralysis, coma and death were also recorded. Upon dissection on day 9 p.i. we observed that all mice of CM category showed collection of fluid in the pleural cavity. None of either the NCM or uninfected mice showed any pleural effusion (PE). It was therefore, decided to explore this phenomenon further and compared the constituents of serum and the PE of the CM mice.

The ¹H NMR spectra of the PE and serum of mice that transited into CM are shown in Figure 2 while that of the pleural fluid of the NCM mouse is shown in Figure 3. The assignment of the resonances was achieved by a combination of database search [34] and 2D-Correlated spectroscopy (COSY) and 2D-Total Correlated spectroscopy (TOCSY) experiments, and it was apparent that the constituents of the two biofluids were similar. A comparison of the SDS-PAGE of the serum and PE were similar as well (Additional file 1). The high similarity in SDS-PAGE profile and ¹H NMR profile of the PE with that of the sera indicates that there is transfer of the serum from the blood vessels resulting in the accumulation of PE. In this context, it is also important to note that there are differences in the ¹H NMR profile of pleural fluid of NCM and pleural effusion of CM as seen in Figure 4. The prominent differences appear in the absence of lipid peaks in pleural fluid. This is also evident from the multivariate analysis in which we see a high concentration of lipids in CM PE and high concentration of lactate in NCM pleural fluid (Table 1).

To begin with, the serum and PE of the CM mice were compared using PCA and it was expected that equilibrium would prevail between these two biofluids since the constituents of the two biofluids exhibited by ¹H NMR spectra and the SDS-PAGE are similar. The PCA scores plot (Additional file 2) represents a NMR spectral profile (sera and PE). Here PC1 is able to segregate between the two classes of biofluids. The corresponding loadings plot shows the metabolites responsible for the discrimination (Additional file 3). Since the ¹H NMR spectra of the serum and PE were similar (Figure 2), a Discriminant analysis, namely OPLS-DA, was performed in order to tease out the possible subtle differences in terms of concentration of metabolite(s). The OPLS-DA scores plot (Figure 5A) of the serum and PE showed a clear segregation. The R² X (cum) of this model was 0.76 and Q² (cum) is 0.85. This signifies considerable distinction between these two classes of fluids. The complementary OPLS loadings of the NMR variables; p(1) and their correlation towards class segregation are presented in Table 2. The OPLS coefficient plot (Figure 5B) showed a marked difference in glucose followed by creatine, glutamine and lipids concentration in serum with respect to PE. The glucose, creatine, glutamine concentration in the PE is higher than that in the serum of the CM mice while lipid concentration is higher in the sera (Figure 5B). Glucose quantification test by the GOD/POD also indicated a significant difference in the glucose levels (p < 0.001; Mann Whitney test) in the serum and PE (Figure 6D). A negative correlation of glucose in the PE with that of serum from MPCA loadings plot suggested that the source of comparatively high amounts of glucose in the PE may derive from the serum (Figure 7B). The concentration of the glucose in the pleural fluid of control mice was found to be 8.9 mM (Figure 6E), consistent with earlier observations [33]. However, the pleural fluid is no more than 5-10 μL. In CM mouse, although the final concentration of glucose in the PE was 7.5 ± 0.7 mM, the volume of the fluid was 200 to 400 μl. Therefore, the entire glucose in PE comes from the serum. Therefore, it appears that the glucose from the serum moves into PE during PE formation in CM mice.

In a different set of experiment sera of the CM mice were compared to the sera of uninfected controls by OPLS-DA where glucose seems to be an important molecule for discrimination and was decreased in CM sera (Table 3). This experiment was followed by another experiment where the serum of individual mouse with CM was mixed the corresponding PE of that mouse and then compared to the sera of controls by OPLSDA. The role of glucose as a discriminatory metabolite between these groups decreases i.e pcorr value of glucose in this model decreases (Table 3). This also indicates that pleural effusion causes hypoglycaemia in the host. Plasmodium infected RBCs consume glucose at higher rate than the healthy RBCs and hypoglycaemic condition is common amongst severe malaria patients [46, 47]. Such a loss of glucose would make the situation more serious. The relative concentration of creatine and glutamine in PE is higher than that of serum as is evident from the OPLS coefficient plot (Figure 5B) and from Figure 6A and 6B respectively, (p < 0.05 for creatine and p < 0.05 for glutamine). On the other hand concentration of lipids in the sera is higher compared to that of pleural effusion (p < 0.05) (Figure 6C). However unlike glucose their correlation towards segregation of PE and serum is not very high (Table 2). Creatine is known to decrease in the serum during malaria due to the increase of creatine phosphokinase (CPK) [48]. Reduced glutamine levels in plasma during malaria are associated with sepsis and mortality rate [49].

This study indicates that cerebral complication of the disease leads to additional 'hypoglycaemia', 'hypocreatine' and 'hypoglutamine' conditions in the blood because of the movement of these components into the PE. This observation also suggests that depletion of glucose, creatine and glutamine in the blood occurs in addition to lactic acidosis, increase of CPK and glutaminolysis, which would add further stress in the patient during CM.

In order to further investigate the correlations of metabolites in the two biological fluids, sera and PE, an MPCA model was generated. PC 1 and 2 captured 59.60 and 24.95% data variation respectively (Figure 7). The eigenvalue (representing a measure of the amount of variance of the score vector) of PC3 onwards was too low, therefore, only the first two PCs were of interest. Since PC2 exhibits homogeneous dispersion of the scores (Figure 7A), the loadings corresponding to this PC were used for further analysis of metabolic inter-compartmental correlations. PC1, on the other hand, mainly shows inter animal variations not related to the disease as it separates one animal from the rest. These loading values of the MPCA of metabolites in serum bear a significant meaning in context to their correlation to form PE. For example, negative loading value of a specific metabolite in serum and its corresponding positive loading value in PE would mean that a decreased concentration of this metabolite in serum is correlated to its increase in PE. This is also valid for the loadings of different metabolites in one particular compartment. From Figure 7B, we see that glucose in the sera is negatively correlated to lactate which indicate glycolysis. Apart from lactate there are positive loadings of pyruvate and alanine in sera. The positive loadings of pyruvate, lactate and alanine with respect to that of glucose in sera indicate the presence of high lactic acidosis in malaria and glucose-alanine cycle operating during CM. The lipids in the sera are negatively correlated to that of lipids in PE; which indicate that lipids in the PE increase with their corresponding decrease from sera. Possibly for lipids/lipoproteins the movement is slow owing to the larger size and so there was a positive loading of lipids in OPLS coefficient plot in sera; indicative of its higher concentration in sera despite its negative loading of MPCA in serum.

The loss of glucose in serum is correlated to increase of glucose in PE, indicating a risk of secondary infection in pleura. However the mechanism behind such transport is not understood. The fate of lactate is just the opposite, i.e. a lower lactate level in PE is correlated to its enhanced level in sera (Figure 7B). A plausible reason could be the need of fast gluconeogenesis in fatal stage of CM.

In order to find out whether the two compartments have contributions to each other at the terminal stage of CM, MCR-ALS was performed over these two compartments. Figure 8A shows the pure contribution profile of the two biological fluids. It is apparent that certain changes in the sera co-vary with certain changes in PE and vice versa. Thus the first and the second factor, representing blood serum and PE respectively, also explain a part of PE and sera profile respectively. The pure spectrotype of sera (Figure 8B) is dominated by lactate whereas glucose and creatine are not significant. On the contrary the spectrotype of PE is dominated mainly by glucose followed by creatine and the other metabolites.