Factor analysis reveals differences in brain metabolism in macaques with SIV/AIDS and those with SIV-induced encephalitis

Abstract

MRS has often been used to study metabolic processes in the HIV-infected brain. However, it remains unclear how changes in individual metabolites are related to one another in this context of virus-induced central nervous system dysfunction. We used factor analysis (FA) to identify patterns of metabolite distributions from an MRS study of healthy macaques and those infected with simian immunodeficiency virus (SIV) which were moribund with AIDS. FA summarized the correlations from nine metabolites into three main factors. Factor 3 identified patterns that discern healthy animals from those with SIV/AIDS. Factor 2 was able to differentiate between animals that had encephalitis and those moribund with AIDS but lacking encephalitis. Specifically, Factor 2 was able to distinguish animals with moderate to severe encephalitis from animals with mild or no encephalitis as well as uninfected controls. FA not only confirmed the involvement of neuronal metabolites (N-acetylaspartate and glutamate) in disease severity, but also detected changes in creatine and myo-inositol that have not been observed in the SIV macaque model previously. These results suggest that the divergent pathways of N-acetylaspartate and creatine in this disease may enable the commonly reported ratio N-acetylaspartate/creatine to be a more sensitive marker of disease severity.

INTRODUCTION

Infection with human immunodeficiency virus (HIV) can often lead to encephalitis and dementia induced by central nervous system penetration by the virus and not as an opportunistic infection (1,2). Proposed mechanisms for this injury include neurotoxic factors and inflammatory cytokines that can produce neuronal injury and gliosis (3). MRS has been used to investigate HIV-induced brain disease, which has documented metabolic changes in patients at various stages of infection (4-7). Such studies have commonly reported decreases in the neuronal marker N-acetylaspartate (NAA) and increases in myo-inositol (MI) and/or choline-containing compounds (Cho). These metabolites are often reported as ratios with respect to total creatine (Cr), which is used as an internal standard because stable energy metabolism is assumed. However, the belief that creatine is unvarying in neurodegenerative disease is debated, and changes in this marker have been reported in neuroAIDS (8).

In the simian immunodeficiency virus (SIV) macaque model, NAA/Cr changes have been reported and have, in some instances, been found to be reversible at different stages of SIV infection (9-12). However, significant Cho/ Cr and MI/Cr changes have only been observed during acute infection. Recently, we reported that NAA/Cr could serve as a biomarker that indicates the severity of encephalitis in SIV-infected macaques (13). We also noted changes in other neurotransmitter ratios, such as glutamate (Glu) and g-aminobutyric acid (GABA) relative to Cr, but these were found to have little bearing on the severity of encephalitis when analyzed individually, even though both correlated very well with NAA.

Although MRS allows quantification of several brain metabolites, it remains unclear how the various changes in these markers relate to one another in the context of HIV/SIV neuropathogenesis. Factor analysis (FA) is a tool for uncovering relationships in a multivariate dataset. This statistical technique (of which principal components analysis is the first step) summarizes the variance in the data by generating linear combinations of variables, reducing the dimensionality down to a few orthogonal factors, and thus eliminating “noise” caused by other irrelevant variables measured. This has been particularly helpful in research that utilizes MRS, where the identification of predominant metabolic profiles has been pivotal to the field of metabonomics (14-17). As more metabolites become quantifiable from in vivo spectra, FA could be effective in reducing error from individual tests as well as uncovering broader patterns in the metabolic processes underlying different stages in the pathogenesis of neuroAIDS. As yet, this method has seldom been applied to HIV MRS studies. AIDS-related leukoencephalopathy has been compared with profiles of other neurological diseases (18,19), and another study examined pediatric cases of HIV-related encephalopathy (20). More recently, the technique has been applied to meta-bolite ratios of HIV-infected subjects across three other brain regions to distinguish between subjects with AIDS dementia complex and those who were asymptomatic (21), and a similar study has correlated MRS-derived factors with monocyte chemoattractant protein-1 (22).

Our study uses a multivariate factor analysis (FA) approach to examine distributions of metabolites from a MRS study of SIV-induced encephalitis (SIVE) in rhesus macaques found to have terminal AIDS (13). The purpose of this analysis was to identify metabolite patterns associated with SIV/AIDS and severity of virus-induced encephalitis. By moving beyond analysis of individual metabolites, this method may (i) improve our understanding of how MRS markers of metabolic processes are related in this disease and (ii) help to resolve apparent inconsistencies in the literature that have arisen from traditional univariate methods involving ratios with respect to Cr or H₂O.

EXPERIMENTAL

This is a further analysis of data from a recent study on the predictive power of NAA/Cr in SIVE (13). Detailed descriptions of the animals in this study have been previously reported (13). Briefly, 23 rhesus macaques (Macaca mulatta) were infected with SIV (SIVmac251 or 239) and euthanized when moribund with AIDS. Their brains were classified as non-encephalitic (n = 6) or encephalitic (n = 17) on the basis of the presence of multinucleated giant cells (the hallmark of HIV/SIV encephalitis) in many regions of the brain, including the frontal cortex. These encephalitic animals were further classified as having mild (n = 6), moderate (n = 4), or severe (n = 7) encephalitis on the basis of the quantity of multinucleated giant cells observed. Six non-infected macaques were sacrificed as controls. The age of the animals ranged from 1.7 to 10.3 years. Previous results indicated no age effect between cohorts (13). Nine animals were female and 20 were male. The study was approved by both Massachusetts General Hospital's Subcommittee on Research Animal Care and the Institutional Animal Care and Use Committee of Harvard University.

Frozen frontal cortex tissues from these 29 animals were extracted with chloroform/methanol, as previously described (13). Extracted metabolites were analyzed with high-resolution ¹H MRS on a Bruker Avance 600 MHz 54 mm vertical-bore spectrometer using the XWIN-NMR 3.5 software package (Bruker Instruments Inc., Billerica, MA, USA). A CP TXI 5 mm H-C/N-D Z-gradient cryo-probe was used for the one-pulse experiments with spectral acquisition parameters including a 90° pulse width of 9 μs, 20 s recycle time, 7.2 kHz spectral width, 32000 complex points, and 64 scans that were averaged (13). Spectral peaks were referenced to TMS and fitted using PERCH NMR software version 1/2005 (PERCH Solutions Ltd, Kuopio, Finland). The metabolites included NAA, Cr, total choline-containing compounds (Cho; 3.19−3.22 ppm), MI, GABA, Glu, glutamine (Gln), N-acetylaspartylglutamate (NAAG), and glycine (Gly) (Fig. 1). As NAA breaks down to acetate and aspartate in even the most carefully harvested and stored samples, NAA reported herein represents the sum of the acetate resonance at 1.91 ppm to the NAA resonance at 2.01 ppm (23).

Figure 1.

600 MHz ¹H MR spectrum of frontal cortex metabolites from a healthy rhesus macaque. Signals arising from the proton resonances used in this analysis are labeled: choline-containing compounds (Cho), creatine (Cr), myo-inositol (MI), glycine (Gly), γ-aminobutyric acid (GABA), glutamine (Gln), glutamate (Glu), N-acetylaspartylglutamate (NAAG), N-acetylaspartate (NAA), and acetate (Ace). Resonances marked with † contribute to Factor 2, and those marked with ‡ contribute to Factor 3. The main Cr y and NAA resonances were truncated to allow a clearer view of the resonances with smaller peaks.

The relative concentration of each metabolite was calculated by dividing its area by the total area of all metabolites of interest (Met/Σ), and patterns of metabolite concentrations were examined by FA. This method differs from another approach used in metabonomic spectroscopy studies in which the variables examined with FA are regions of adjacent spectral points grouped into ‘bins’ or ‘buckets,’ and thus FA produces linear combinations of spectral regions, rather than combinations of metabolites (24). Here, deconvolution of peaks for metabolite quantitation results in factors that represent linear combinations of metabolite concentrations while obviating the need for inter-sample alignment and reducing the complications from overlapping signals.

Factor analysis

FA is an exploratory method of data reduction that explains many variables with just a few components or factors by grouping correlated variables together. First, the principal components are identified. The variables are standardized (subtract the mean and divide by the standard deviation for each variable in each case), and new orthogonal axes are successively assigned along the directions of greatest variation. These are the eigenvectors of the correlation matrix. Most of the variation in the sample is captured in the first few components. According to the Kaiser criterion, only the components with eigenvalues greater than 1 are kept, as these contain a greater proportion of the variance than any of the original variables individually (25). The remaining variation in the dataset is considered to be noise. When the data are plotted using these new axes, in the principal components space correlated variables will cluster together.

Varimax rotation was applied to these components to aid their interpretation, generating rotated factors that align more closely with the directions of these clusters. The simple structure obtained from this orthogonal transformation aids in the interpretation of factors by making clusters of correlated variables more prominent, while keeping the factors uncorrelated with each other (25). This resulting factor loading matrix is reported, from which investigators attempt to infer the latent structure (referred to as loadings) of patterns or causes underlying the observable variables. In addition, for each case, values for each factor can be calculated from the original variables using a scoring matrix. The scoring matrix is obtained by multiplying the loading matrix by the inverse of the correlation matrix. The resulting scores can be used for further analysis of variance (ANOVA) across groups or for correlations with additional measures.

Analysis of disease severity

For this study, each animal was assigned a score for each factor based on the respective loadings of its nine metabolite concentrations. The association between the resulting components and the severity of encephalitis was assessed by ANOVA. If the ANOVA was significant, specific differences between disease classifications were isolated using two-tailed least squares means t tests. Statistical tests were performed using JMP 5 software (SAS Institute, Cary, NC, USA).

RESULTS

FA identified three main factors that jointly accounted for 67% of the variance in the data (Table 1). Loadings for each metabolite found to have a strong (0.6+) or moderate (0.4−0.6) involvement appear in bold in Table 1, and these metabolites are noted in Fig. 1. Each animal was assigned a score for each factor based on its loadings (Table 2). The scores for Factors 2 and 3 were significantly different across cohorts (ANOVA: P = 0.005, P < 0.05, respectively), whereas scores for Factor 1 were not (ANOVA: P = 0.82).

Table 1.

Factor loadings

	Factor 1	Factor 2	Factor 3
Variance explained
Eigenvalue	1.615	2.304	2.093
Percentage	17.949	25.598	23.250
Loadingsa
Cr	−0.285	−0.479	−0.637
NAA	−0.130	0.884	0.013
Glu	0.246	0.689	0.564
Gln	−0.195	0.209	−0.697
GABA	−0.412	0.229	0.679
Gly	−0.240	−0.030	0.633
NAAG	−0.879	−0.062	0.078
Cho	0.596	−0.248	0.021
MI	0.252	−0.810	0.118

^aStrong (0.6+) or moderate (0.4−0.6) loadings are shown in bold to indicate the metabolites involved in each factor. Concentrations of each metabolite were calculated by dividing its area by the total area of all metabolites of interest (Met/Σ).

Table 2.

Histopathological classifications and factor scores

Animal	Central nervous system pathologya	Factor 1	Factor 2	Factor 3
C1	Healthy seronegative control	0.818	1.533	0.930
C2	Healthy seronegative control	−0.108	1.588	1.380
C3	Healthy seronegative control	−0.420	1.386	0.045
C4	Healthy seronegative control	0.521	0.847	0.805
C5	Healthy seronegative control	−0.116	0.253	1.706
C6	Healthy seronegative control	0.723	−0.235	0.274
M1	Severe encephalitis	−0.574	−0.265	−0.079
M2	Severe encephalitis	0.434	−1.106	0.532
M3	Severe encephalitis	1.295	−1.131	0.728
M4	Severe encephalitis	0.572	−1.907	0.389
M5	Severe encephalitis	−0.364	0.553	−0.765
M6	Severe encephalitis	−0.005	−1.299	−1.882
M7	Atypical severe encephalitis	1.228	−1.370	0.044
M8	Moderate encephalitis	0.301	−1.792	0.487
M9	Moderate encephalitis	−0.422	0.484	0.989
M10	Moderate encephalitis	−0.732	−0.511	−0.319
M11	Moderate encephalitis	1.197	−0.842	0.351
M12	Mild encephalitis	−4.029	−0.848	0.307
M13	Mild encephalitis	−1.053	−0.166	0.784
M14	Mild encephalitis	−0.412	0.153	−2.453
M15	Mild encephalitis	0.415	1.661	−1.326
M16	Mild encephalitis	0.182	0.566	0.273
M17	Mild encephalitis	0.795	0.072	−0.668
M18	No encephalitis	0.362	0.381	−0.513
M19	No encephalitis	0.142	0.633	−0.259
M20	No encephalitis	−0.286	0.322	0.449
M21	No encephalitis	0.761	1.046	−1.709
M22	No encephalitis	−0.499	0.373	0.862
M23	No encephalitis	−0.724	−0.377	−1.363

^aNote: M1-M23 were all inoculated with either SIVmac251 or SIVmac239, and all were sacrificed because they were moribund with AIDS.

Analysis of healthy and SIV-infected macaques

Scores for Factor 3 were able to discern between animals that had SIV/AIDS and those that did not. Lower scores were associated with SIV infection (Fig. 2). Factor 3 was able to distinguish control animals from infected animals with SIVE (P < 0.03), as well as from infected animals without encephalitis (P = 0.02). However, no distinction between the two SIV-infected groups could be found using this factor (P = 0.55).

Figure 2.

Factor 3 as a predictor of SIV/AIDS. Factor 3 represents a linear combination that summarizes metabolic changes in Cr/Σ, GABA/Σ, Gln/Σ, Glu/Σ, and Gly/Σ in the frontal cortex in SIV-infected animals. Animals with SIV/AIDS had significantly lower scores for Factor 3 than healthy controls (left), but this factor was not able to distinguish between classifications of SIVE severity (right). Error bars represent standard error of the mean.

Analysis of encephalitis and SIVE severity among macaques

Factor 2 was able to discriminate animals with SIVE from uninfected controls (P = 0.002), and from animals moribund with AIDS without SIVE (P = 0.04) (Fig. 3, left). This factor was, however, unable to discern between control animals and infected animals that lack encephalitis (P = 0.32). Further analysis indicated that Factor 2 could distinguish between disease classifications (ANOVA: P < 0.002), with lower scores associated with encephalitis severity (Fig. 3, right). Factor 2 was able to differentiate animals with severe encephalitis from control animals (P < 0.0003), as well as from those found moribund with AIDS but lacking signs of encephalitis (P < 0.005). Factor 2 was also able to distinguish animals with moderate encephalitis from control animals (P = 0.004), as well as from animals found to be moribund with AIDS without encephalitis (P = 0.04). Factor 2 was significantly different between severely and mildly encephalitic animals (P = 0.01), but was not significantly different between moderately and mildly encephalitic animals (P < 0.08). Loadings from Factor 2 represent a linear combination of changes in both primarily neuronal (NAA/Σ, Glu/Σ) and primarily glial (Cr/Σ, MI/Σ) metabolites during SIV infection. Graphical representations of the mean values of these four metabolites for each cohort are shown in order to clarify the ways in which these metabolites are changing (Fig. 4). This figure illustrates that the changes in Factor 2 arise from the decreases in NAA/Σ and Glu/Σ and increases in Cr/Σ and MI/Σ that occur with disease severity.

Figure 3.

Factor 2 as a predictor of encephalitis severity. Factor 2 represents a linear combination that summarized major metabolic changes in both neuronal (NAA/Σ, Glu/Σ) and glial (Cr/Σ, MI/Σ) cells during SIV infection. Animals with SIVE had significantly lower scores for Factor 2 than those without encephalitis (left). Factor 2 was able to distinguish animals with moderate to severe encephalitis from animals with mild or no encephalitis as well as uninfected controls (right). Error bars represent standard error of the mean.

Figure 4.

Ratio of metabolites/total spectrum. Mean values and standard error of the mean are presented to demonstrate individual metabolites across classifications of encephalitis severity. Cr/Σ appears to be increased in all animals infected with SIV and moribund with AIDS. Glu/Σ shows the opposite trend. For the severest disease classifications, MI/Σ increases while NAA/Σ decreases. Cho/Σ does not differ significantly across the cohorts. FA captures these effects in combination, and is more powerful than traditional univariate analyses, as revealed by post hoc ANOVA (Cr/Σ: P = 0.07; NAA/Σ: P = 0.0006; Glu/Σ: P = 0.05; MI/Σ: P = 0.005; Cho/Σ: P = 0.29) uncorrected for multiple comparisons.

DISCUSSION

Currently, there are many ways to evaluate MRS data, including ratios with respect to Cr (13,21), H₂O (sometimes referred to as ‘absolute concentrations’) (7,22), or the total area of the spectrum (19,20), as well as the use of different statistical analyses. Various methods have been tried in human and macaque studies, which have resulted in some apparent contradictions. It is unclear whether these occasional inconsistencies are the result of differences in measurement or data analysis. Here, we present FA as an alternative means of examining spectroscopy data in HIV/SIV studies and have obtained results that shed new light on some inconsistencies reported from traditional univariate methods involving ratios with respect to Cr or H₂O.

Application of FA

FA seeks to find the latent structure of patterns underlying the observable variables. In psychological studies that involve surveys, FA may be used to identify the opinions and attitudes (unobservable factors) that generate the answers given in a questionnaire (observable variables). In our study, we seek relationships (unobservable factors) between metabolites that can explain the changes in individual metabolite concentrations (observable variables). The loading matrix extracts the factors from the original variables. Just as the answers to a variety of questions may all stem from a few basic beliefs held by an individual, so the changes in these metabolite concentrations may derive from a few core mechanisms or physiological responses. The scoring matrix gives the reverse conversion, combining observable variables to create values for the unobservable factors. In the example of a survey, an individual's answers are used to generate a score for his or her opinions. In our study, each animal's metabolite concentrations are converted into a score for these physiological responses. Assuming that the factors cause the variable changes, by convention it is the loading matrix that is used for interpretation (Table 1).

Various types of FA (including principal components analysis) have been applied in studies of psychological and psychiatric assessment (26-30), in measures of animal behavior (31,32), in face recognition and data-processing models (33,34), in analyses of microarrays and drug libraries (35,36), as well as in clinical investigations and meta-analyses to define disease profiles using all the available variables (37-40). Exploratory FA can create composite measures for multidimensional phenomena, but the resultant structure pattern and its interpretation require confirmation from further studies. The validity of the factors may depend on the consistency and robustness of the procedure used, especially in deciding the number of components to retain (25). To keep the procedure simple, in this study we have followed a standard method of performing FA: identifying the principal components, retaining those that met the Kaiser criterion, and applying a varimax rotation. Beyond this traditional procedure, there are other methods of FA which are discussed in detail in the literature. These include various criteria for selecting the number of components to retain, such as the scree test, broken stick rule, and parallel analysis, as well as several different rotation methods (25).

FA was performed on the dataset as a whole, producing factors that describe the variance across all 29 animals, without consideration of grouping. Only afterwards were the factors evaluated to see if they correspond to differences between cohorts. As this type of analysis is based on the correlation matrix between the original variables, at minimum it requires more cases than variables. However, most practitioners prefer 3−5 times as many cases as variables for reliability. As we performed FA on nine variables (metabolite measurements) using 29 cases (animals), this can be considered adequate in terms of the general standard.

Implications of Factor 1

Factor 1 represents the contrast of Cho/Σ with GABA/Σ and NAAG/Σ. It distinguishes animals with high Cho/Σ, low GABA/Σ, and low NAAG/Σ from animals with low Cho/Σ, high GABA/Σ, and high NAAG/Σ. Animals with moderate values of Factor 1 may have high values of all three, low values of all three, or moderate levels of all three. Factor 1 identified the correlations between these metabolites, but the ANOVA showed that the variance summarized in this factor did not differ across cohorts (Fig. 5). Therefore, it is not diagnostic of SIV/AIDS or encephalitis severity. This factor may reflect a different source of variation, but one that we cannot speculate on at this point because of the lack of data to formulate a hypothesis. Although in vivo MRS studies on subjects with HIV-associated dementia as well as SIV-infected macaques have shown a major role for choline, the same changes have rarely been observed ex vivo. Further investigation is needed to clarify whether the metabolite extraction process changes the MRS signal of choline-containing compounds (41-43).

Figure 5.

Factor 1 is not diagnostic of SIV/AIDS or encephalitis severity. Factor 1 represents a linear combination that summarizes metabolic changes in Cho/Σ, GABA/Σ, and NAAG/Σ. Scores for Factor 1 did not significantly differ among cohorts. Error bars represent standard error of the mean.

Implications of Factor 2

The MRS patterns found in Factor 2 were associated with increasing severity of SIVE. An examination of the loadings for Factor 2 (Table 1) shows a significant positive contribution from NAA/Σ and Glu/Σ, indicating that lower NAA/Σ and Glu/Σ will contribute to a lower score for Factor 2, which is associated with worse disease severity. We recently reported significantly lower NAA/Cr and Glu/Cr in animals with more severe encephalitis (13). This correlation is confirmed and summarized here in their combined loading in Factor 2. Both NAA and NAA/Cr are often cited as markers of neuronal integrity, and decreases in NAA and NAA/Cr have been found to correlate with severity of neuropsychological impairment in patients with HIV (4).

Cho, MI, and Cr are known to be present at higher concentrations (∼3 times) in glial cells than in neurons (44,45). Increases in these metabolite concentrations have been reported in subjects with neuroAIDS (8). Although we did not discover a major contribution from Cho in Factor 2, a significant negative contribution from MI/Σ was observed, as well as a moderate negative contribution from Cr/Σ. Thus, higher MI/Σ and Cr/Σ will contribute to a lower score for Factor 2, which is associated with worse disease severity. Cr, a resonance arising from creatine and the high-energy phosphate phosphocreatine, is often assumed to be constant, and MRS studies often report metabolites normalized with respect to Cr. Increased MI is typically considered to be indicative of increased glial activity, and has been reported in studies of patients with HIV (4). Increases in MI/Cr that correlate with disease severity were expected for this dataset, but no statistically significant differences in MI/Cr across classifications of disease severity were found (13). The Cr/Σ loading in our FA offers a possible explanation of why this was not seen: the higher concentrations of MI in animals with worse encephalitis may have been masked by the accompanying higher concentrations of Cr.

Individually, NAA/Σ, Glu/Σ, Cr/Σ, and MI/Σ are changing, but only slightly, except for NAA/Σ (Fig. 4). In assessing changes in brain metabolism of animals infected with SIV and animals with encephalitis, the analysis is more powerful when all four metabolites are examined together. Post hoc ANOVAs on individual metabolites show that a traditional univariate analysis is not as sensitive, and would have reduced statistical power if corrections for multiple comparisons were performed.

Most measures of spectroscopy rely on some variation in ratios (with respect to H₂O, Cr, or the sum of all metabolites) for normalization. By their very nature, these variables used for normalization are not exempt from changing with disease. It is possible that the increases in Cr/Σ and MI/Σ can be explained by a decrease in the overall sum of metabolites. However, in the light of increases in absolute concentration of Cr and MI (ratios based on H₂O concentration) found in the frontal cortex of human subjects with HIV infection (7,8), we have interpreted these results on the basis of this precedent. The possibility of changes in Cr is often raised as a note of caution against the use of metabolite ratios in favor of absolute concentrations in MRS studies. Indeed, even variations across different anatomical regions must be considered. Although raised Cr concentrations have been observed in the frontal white and gray matter of patients with HIV, decreases were found in the basal ganglia (8). However, the use of ratios with respect to Cr need not be discarded altogether. We have previously reported that NAA/Cr alone is sensitive in distinguishing between severities of encephalitis (13). Our FA suggests that the combination of lower NAA/Σ and higher Cr/Σ in the frontal cortex in animals with more severe encephalitis actually allows NAA/Cr to be a more sensitive marker of encephalitis. Therefore, although the quantification of absolute concentrations is important, the use of ratios such as NAA/Cr may still be beneficial in the examination of human subjects.

Implications of Factor 3

Many in vivo MRS studies at 1.5 T have tried to quantify the peaks lying between 2.0 and 2.8 ppm (sometimes referred to as ‘Glx’) (8,19,22,46-50). Although Glu and Gln are the largest contributors to this region, at lower field strengths Glx will also include resonances from GABA, NAAG, aspartate, NAA, and succinate. Factor 3 reflects changes occurring primarily in Glu/Σ, Gln/Σ, GABA/Σ, Gly/Σ, and Cr/Σ. Concentrations of these metabolites appear to differentiate between animals that have SIV/AIDS and those that do not. Compared with control animals, animals with AIDS exhibited a pattern of higher Cr/Σ and Gln/Σ, and lower Glu/Σ, GABA/Σ, and Gly/Σ. The changes in Factor 3 show that in vivo measurements of the Glx region must be carefully considered. This region is very difficult to quantify in vivo at low field strengths, because of overlapping resonances that also extend into the base of the main NAA peak. The data suggest that potential variations in Glx could interfere with the accurate quantification of the NAA resonance at lower fields.

Taken together, these results suggest two distinct patterns of injury in SIV/AIDS. The metabolic markers in Factor 3 indicate that a low level of neuronal injury is occurring in all animals with SIV/AIDS whether or not multinucleated giant cell formation is observed. Perhaps the changes summarized in Factor 3 represent the damage caused by the presence of long-lived cells that have already established a viral reservoir in the brain at an earlier stage or a low-level steady-state influx of activated/infected monocytes (51-54). This steady insult would contrast with the type represented by decreases in NAA, which has been reported to demonstrate not only permanent neuronal damage but also reversible injury (9,10,12). Patterns in Factor 2 – which correlate with encephalitis – may signify a more recent shift in viral dynamics. Possible explanations include the recrudescence or re-activation of infected long-lived cells (54), independent viral replication and evolution of existing genetic variants in the central nervous system (55), and/or a higher rate of influx of activated/infected cells from the periphery, especially among specific subsets of blood monocytes/macrophages (51). Such a shift in viral kinetics, immune-cell trafficking, and turnover would produce more viral proteins and pro-inflammatory factors as well as promote the formation of multinucleated giant cells.

In conclusion, the use of FA allowed the identification of two metabolite patterns, one associated with SIV/AIDS and the other with severity of virus-induced encephalitis. These patterns would not have been observed using traditional analytical methods. This technique also allowed us to determine changes in metabolites that had not been reported previously (MI/Σ and Cr/Σ) in this SIV macaque model of neuroAIDS. Further work is needed to clarify the immunological/virological basis of these metabolic patterns.

Abbreviations used

AIDS

acquired immunodeficiency syndrome

ANOVA

analysis of variance

Cho

choline-containing compounds

Cr

creatine

FA

factor analysis

GABA

γ-aminobutyric acid

Gln

glutamine

Glu

glutamate

Gly

glycine

HIV

human immunodeficiency virus

MI

myo-inositol

NAA

N-acetylaspartate

NAAG

N-acetylaspartylglutamate

SIV

simian immunodeficiency virus

SIVE

SIV-induced encephalitis

Σ

sum of all metabolites of interest