Evaluation of Tubulointerstitial Lesions’ Severity in Patients with
Glomerulonephritides: An NMR-Based Metabonomic Study
Nikolaos G. Psihogios,† Rigas G. Kalaitzidis,‡ Sofia Dimou,§ Konstantin I. Seferiadis,†
Kostas C. Siamopoulos,‡ and Eleni T. Bairaktari*,†
Laboratory of Clinical Chemistry, Department of Nephrology, and Department of Histopathology,
Medical School, University of Ioannina, GR-451 10, Ioannina, Greece
Received March 27, 2007
An 1H NMR-based metabonomic approach was used to investigate the correlation of histopathologically
assessed tubulointerstitial lesions with the urinary metabolite profile in 77 patients with glomerulonephritides submitted to renal biopsy. The presence of renal damage was predicted with a sensitivity
of 96% and a specificity of 99%. Patients with mild, moderate, and severe tubulointerstitial lesions
were progressively differentiated from the healthy individuals in the Orthogonal Signal Correction Partial
Least-Squares-Discriminant Analysis (OSC/PLS-DA) models with a statistically significant separation
between those with mild and with severe lesions. The onset of the tubulointerstitial lesions is
characterized by decreased excretion of citrate, hippurate, glycine, and creatinine, whereas further
deterioration is followed by glycosuria, selective aminoaciduria, total depletion of citrate and hippurate,
and gradual increase in the excretion of lactate, acetate, and trimethylamine-N-oxide. NMR-based
metabonomic urinalysis could contribute to the early evaluation of the severity of the renal damage
and possibly to the monitoring of kidney function.
Keywords: glomerulonephritis • tubulointerstitial • urine • 1H NMR spectroscopy • metabonomics • kidney
Introduction
Glomerulonephritis (GN) is a group of disorders characterized by inflammation in the filtering unit of the kidney, the
glomerulus, which along with a long tubule comprises the
anatomical and functional unit of the kidney, the nephron. The
unit is surrounded by a functionally important tissue called
interstitial tissue. GN may be primary; secondary to drugs,
infections, or tumors; or the presenting feature of systemic
disease.1 GN causes significant morbidity and mortality, and
is a potentially preventable cause of renal failure and cardiovascular risk. Although the glomerulus is the primary site of
damage, subsequent injury to the tubulointerstitium plays a
major role in the overall outcome of glomerular disease.1,2
The diagnosis of GN can be suspected by clinical and
laboratory findings, such as proteinuria and abnormal urine
microscopy.1 However, renal biopsy is considered as the main
tool for the evaluation of the type and degree of renal injury in
almost all cases of glomerulonephritides.3 In addition, careful
pathological analysis reveals the extent of tubulointerstitial
damage and degree of renal tubular fibrosis, findings that in a
number of cases are correlated better with the deterioration
of renal function than the degree of glomerular damage itself.4
* Author for correspondence. Eleni T. Bairaktari, Ph.D., EurClinChem,
Assistant Professor, Laboratory of Clinical Chemistry, Medical School
University of Ioannina, 451 10 Ioannina, Greece. Phone/fax, +30-26510
97620/97871; e-mail, [email protected].
†
Laboratory of Clinical Chemistry, Medical School, University of Ioannina.
‡
Department of Nephrology, Medical School, University of Ioannina.
§
Department of Histopathology, Medical School, University of Ioannina.
3760
Journal of Proteome Research 2007, 6, 3760-3770
Published on Web 08/18/2007
1
H NMR spectroscopy of urine provides overall profiles of
low molecular weight (LMW) metabolites that alter characteristically in response to changes in physiological status, toxic
insult, or disease processes.5-8 In situations where renal damage
is present in humans or experimental animals, the LMW
metabolite profile of urine is significantly altered, and this is
closely reflected in the 1H NMR spectral fingerprint. Furthermore, in studies with experimental animals exposed to regionspecific nephrotoxins, the NMR-generated metabolite profiles
were characteristically changed according to the exact site and
mechanism of the lesion (glomeruli, lower or upper regions of
the proximal tubules, renal medulla).8,9 In the clinical field,
NMR urinalysis has contributed to the assessment of renal
transplant dysfunction,10,11 to the early detection of latent
tubulointerstitial distortions in glomerulonephritis,12,13 and to
the detection of renal dysfunction in several pathological
states.14-17
The exploitation of the NMR-generated metabolic data sets
can be increased by the application of multivariate statistical
analysis including pattern recognition (PR) methods that allow
sample classification and effective interpretation.18-20 This
relatively new approach, known as metabonomics,21 has had
major applications in clinical and biomedical topics such as
drug toxicity assessment, identification of biomarkers of toxicity
and disease, and the understanding of the mechanisms of
metabolic responses.8,22-24
This prospective study investigates the correlation of tubulointerstitial lesions found in renal biopsies with the metabolite
10.1021/pr070172w CCC: $37.00
 2007 American Chemical Society
Tubulointerstitial Lesions’ Severity in Patients with GN
profile of urine analyzed by NMR-based metabonomics in
patients with glomerulonephritides.
Materials and Methods
Subjects. The study initially included 80 consequently
admitted patients to the Department of Nephrology of the
University Hospital of Ioannina to be submitted to renal biopsy
due to renal function abnormalities such as increased serum
creatinine and/or proteinuria and chronic renal disease stages
1-3.25 Three patients were excluded during the analysis of the
spectroscopic data, as described in the Results. The inclusion
criteria were moderate proteinuria (<2 g/24 h) and serum
creatinine <3 mg/dL. Eighty-five sex- and age-matched healthy
individuals who did not require regular medication other than
oral contraception or over-the-counter drugs constituted the
control group. All study participants gave informed consent for
the investigation, which was approved by the Ethical Committee of the University Hospital of Ioannina.
Histopathology. Patients were submitted to renal biopsy,
and the renal tissues were examined by light microscopy and
in certain cases with immunofluorescence and/or electron
microscopy. Immunostainings were performed on formalinfixed, paraffin-embedded tissue sections by the labeled streptavidin biotin (LSAB) method. All biopsies were reviewed by one
pathologist who was blind to the NMR data. The renal biopsy
diagnoses included focal segmental glomerulosclerosis in 23
patients, membranous nephropathy in 13 patients, IgA nephropathy in 8 patients, mesangioproliferative glomerulopathy in
5 patients, systemic lupus erythematosus in 8 patients, vasculitis in 5 patients, diabetic nephropathy in 9 patients, minimal
change disease in 4 patients, and other causes in 5 patients.
Tubulointerstitial lesions included tubular atrophy, interstitial
fibrosis, and mononuclear cell infiltration. The extent of these
lesions was graded as follows: mild (n ) 25), moderate (n )
27), and severe (n ) 25).
Samples. All subjects were requested to fast overnight and
abstain from any medication (including over-the-counter
drugs), alcohol, and fish consumption, known to significantly
affect the urinary metabolite profile,26 24 h before sampling.
Blood and urine samples were obtained before patients were
submitted to renal biopsy. Serum was separated by centrifugation at 1500g for 15 min. First void urine samples were
centrifuged at 1000g for 10 min, and an aliquot was taken for
clinical chemistry tests. Sodium azide (1 g/L, 100 µL) was added
to the remaining urine sample to prevent bacterial contamination and stored at -80 °C until NMR analysis.
Clinical Chemistry. Analysis of clinical chemistry parameters
of serum and urine was carried out on an Olympus AU600
Clinical Chemistry analyzer (Olympus Diagnostica, Hamburg,
Germany) by standard procedures. GFR was calculated by the
MDRD equation.27
1
H NMR Spectroscopy. Four hundred microliters of urine
was mixed with 200 µL of phosphate buffer (0.2 M Na2HPO4/
0.2 M NaH2PO4, pH 7.4) in order to minimize pH variations,
and then a solution of 0.075% sodium 3-trimethylsilyl-(2,2,3,32
H4)-1-propionate (TSP) in D2O was added.
1
H NMR spectra were measured at 300 K on a 500 MHz
Bruker DRX NMR instrument operating at 500.13 MHz and
running on XWINNMR V.2.6 software. For the suppression of
the water signal, the standard 1D pulse sequence NOESYPRESAT (RD-90°-t1-90°-tm-90°-FID acquisition) was used.28 RD was
a 3 s relaxation delay to ensure T1-relaxation between successive scans and during which the water peak was selectively
research articles
irradiated; t1 represented the first increment in the NOESY
experiment and was set to 3 µs; tm was the mixing time of 150
ms, during which the water resonance was again selectively
irradiated. For each spectrum, 128 scans were collected into
64K computer data points with a spectral width of 6009.6 Hz.
The FIDs were multiplied with an exponential line broadening
function of 0.3 Hz prior to Fourier transformation. The acquired
NMR spectra were manually corrected for phase and baseline
distortions by applying a simple polynomial curve fit with
TopSpin 1.2 (Bruker Biospin Ltd.) and referenced to TSP (δ1H
0.0). The metabolites were assigned according to published
literature and 2D experiments (Supplementary Figure 1 in
Supporting Information).
Two-dimensional (2D) NMR spectra were carried out on
selected samples for identification of urine metabolites. 1H-1H
TOCSY spectra were acquired using the MLEV17 spin-lock
scheme (mlevesgpph). Fifty-six transients per increment for 800
increments were collected into 2048 data points in the F2
dimension using a spectral width of 12.02 ppm in both
frequency axes and a relaxation delay of 1.2 s. A sine-bell
squared function was applied to the data prior to Fourier
transformation.
Statistical Analysis. Statistical analysis was performed with
Statistica Ver. 6.0 (StatSoft, Inc., Tulsa, OK). Values were
expressed as mean value ( standard deviation (SD) and
compared by using t test. Significance levels were set at 0.05.
NMR Data Reduction and Pattern Recognition (PR). The
1
H NMR spectra were automatically reduced using AMIX
(Analysis of MIXtures) software package (version 3.2.4, Bruker
Analytik, Rheinstetten, Germany) to 244 continuous integral
segments (variables or bins) of equal width of 0.04 ppm
corresponding to the chemical shift range δ1H, 0.2-10.0. The
area between 4.38 and 6.30 ppm was excluded to remove any
effect of variation from the suppression of the water resonance
and from any cross-relaxation effect on the urea signal via
solvent exchanging protons. The integral regions of the citrate
(2.50-2.58 and 2.66-2.74) and the creatinine (3.02-3.06 and
4.02-4.06) resonances were merged to take into account the
pH-dependent peak shifts and formed the “superbins” 2.54 and
2.7 for citrate and 3.04 and 4.04 for creatinine, respectively. All
data was normalized by dividing each integral segment by the
total area of the spectrum in order to compensate for the
differences in overall concentration between individual urine
samples. The resulting data matrix, consisting of 194 NMR
integral segments, was exported to the SIMCA-P software
package (version 10.5, UMETRICS AB, Box 7960, SE 90719,
Umeå, Sweden) for the PR analysis. Prior to the analysis, the
NMR data were centered and Pareto scaled (scaling factor
1xSD).
PCA was used for the overview of the metabonomic data set
and the spotting of outliers, and then for the detection of any
grouping or separation trend.20 The PCA scores plot was used
to reveal observations lying outside the 0.95 Hotelling’s T2
ellipse (strong outliers) and the loadings plot to interpret the
patterns seen in the scores plot. The model residuals plot,
DModX, was used to detect observations that exceeded the
critical distance of significance <0.05 (moderate outliers).20
With Partial Least-Squares Discriminant Analysis (PLS-DA)
a relationship was sought between the matrix of variables X
(NMR spectral bins) and a matrix of dependent variables Y
(dummy variables encoding the class membership, i.e., patient
or control). The method was used to find the best possible
discriminant function (model) that separates renal patients
Journal of Proteome Research • Vol. 6, No. 9, 2007 3761
research articles
Psihogios et al.
Table 1. Demographic and Clinical Chemistry Characteristics of the Populations Studied
Subgrouping of GN patientsa
controls
Number
Age
Sex (males/females)
S Creatinine (mg/dL)
S Albumin (g/dL)
U Total Proteins (g/24 h)
GFR (mL/min/1.73 m2)
85
52.0 ( 8.7
42/43
0.9 ( 0.2
4.4 ( 0.3
0.015 ( 0.012
85.2 ( 11.1
patients
p
mild
moderate
severe
<0.001
<0.001
<0.001
<0.001
25
49.2 ( 17.8
12/13
1.1 ( 0.4
3.6 ( 0.7
0.150 ( 0.209
68.8 ( 23.9
27
56.9 ( 9.7
19/8
1.5 ( 0.4c,##
3.6 ( 0.7
0.427 ( 0.595c,#
49.4 ( 15.5c,##
25
55.4 ( 16.8
15/10
1.6 ( 0.5c,###
3.2 ( 0.7
0.622 ( 0.710c,##
40.4 ( 14.4c,###
77b
55.5 ( 15.4
45/32
2.0 ( 1.8
3.5 ( 0.7
0.370 ( 0.540
48.6 ( 25.2
NS
a
Severity of the tubulointerstitial lesions in renal biopsy estimation. b Three out of the 80 patients were excluded (see Results). c #, p < 0.05; ##, p < 0.01;
###, p < 0.001 t test compared to mild.
from controls as well as the three defined histopathology groups
on the basis of their X variables.20 For the interpretation of the
scores plot, the regression coefficients plot was used, which
shows all spectral regions that contribute to the separation
between the studied groups.
The technique of Orthogonal Signal Correction (OSC) was
applied to remove linear combinations of variables X that were
orthogonal to the Y vector of the dependent variables, to
eliminate the intersubject variability and to describe maximum
separation based on class.29
The default method of 7-fold internal cross-validation (CV)
of SIMCA-P software was applied, and the extracted parameter
Q2 was used to provide an estimation of the predictive
capability of the PLS-DA models with Q2 > 0.5 considered
‘good’ and Q2 > 0.9 ‘excellent’.20 The parameter R2 describes
the explained variation and how well the data can be mathematically reproduced by the training model.
In addition, validation was performed using both held-back
and external data procedures. In held-back data validation, as
test set serves a portion of the training set used for the
construction of the model, whereas in external data validation,
as test set (prediction set) serves a new set of data not used
when the model was built.
Held-back data validation for the patients-controls model
was performed using 81% of the data as the training set and
the remaining 19% as the test set, whereas for the model
between patient groups, 68% of the data defined the training
set and the remaining 32% the test set. External validation for
the patients-controls model was performed using 70% of the
data as the training set and the remaining 30% as the prediction
set, whereas for the model between patient groups, 68% of the
data defined the training set and the remaining 32% the
prediction set. All observations were assigned with a classspecific numerical value to form a response Y matrix. Correct
classification was based on a predicted Y cutoff of 0.5 with a
95% confidence level.
Correct and incorrect assignments were used to define
True Positives (TP), True Negatives (TN), False Positives
(FP), and False Negatives (FN) classification rates and then to
estimate as percent sensitivity [TP/(TP + FN) × 100] and
specificity [TN/(TN + FP) × 100].
Results
Clinical Chemistry. In Table 1, the main demographic and
clinical chemistry parameters of the populations studied are
shown. Patients with GN presented statistically significant
higher levels of serum creatinine and urine total proteins and
lower levels of serum albumin and GFR than the control
population. There were also statistically significant differences
3762
Journal of Proteome Research • Vol. 6, No. 9, 2007
among the three groups of patients defined by the severity of
the tubulointerstitial lesions in renal biopsy estimation. Patients
with moderate and severe lesions had significantly different
levels of creatinine, GFR, and urine protein from those with
mild lesions. However, no significant differences were observed
between those with moderate and severe lesions.
1H
NMR Spectroscopy. Four typical 1H NMR 500 MHz
spectra of urine from a healthy individual and patients with
mild, moderate, and severe renal damage indicating the
different excretive profile of the LMW metabolites in each case
are shown in Figure 1. The main constituents of the urine
spectrum from the healthy individual are creatinine, hippurate,
citrate, glycine, trimethylamine-N-oxide (TMAO), dimethylamine (DMA), and small amounts of lactate, 3-hydroxybutyrate
(3-HB), N-acetyl groups from glycoproteins (N-Acs), and amino
acids such as alanine, phenylalanine, and valine.30 The spectrum from the patient with mild renal damage indicates partial
inhibition in the excretion of hippurate, citrate, and glycine,
whereas the spectrum from the patient with moderate renal
damage reflects further decrease in the excretion of hippurate,
citrate, and glycine followed by an increase in the levels of
lactate, alanine, and phenylalanine. The spectrum from the
patient with severe renal damage indicates significant to
complete inhibition in the excretion of hippurate, citrate, and
glycine; increased levels of glucose, lactate, alanine, phenylalanine, and histidine; and a slight elevation of the spectrum’s
baseline in the region (1.8-0.5 ppm) due to the resonance of
the aliphatic moieties of proteins excreted in urine. A significant
number of patients, apart from elevated TMAO levels, excreted
one or more choline headgroup containing metabolites (between 3.20 and 3.30 ppm), but not in accordance to the severity
of the renal damage.
The metabonomic data set initially consisted of 165 urine
NMR spectra: 80 from patients that underwent renal biopsy
and 85 from healthy individuals. The urine spectra were visually
inspected, and 3 from the patient group were excluded: the
first one showed 2 intense unidentified peaks at 2.16 and 2.18
ppm (probably metabolites from paracetamol ingestion) and
the other two showed intense peaks within the region 3.5-3.8
ppm probably due to metabolites from drugs that the patients
had received before sampling.
Pattern Recognition Analysis. In this data set (from 77
patients and 85 healthy individuals), PCA was applied, and the
scores plot (Figure 2a) showed a separation trend between the
two groups with healthy individuals clustering to the right
section and patients spreading mainly to the left side of the
plot. The PCA plot also revealed 2 spectra with significant
alterations from patients mainly characterized by high peaks
of glucose and decreased excretion of creatinine. In the relative
Tubulointerstitial Lesions’ Severity in Patients with GN
research articles
Figure 1. 1H NMR 500 MHz spectra of urine (δ 0.3-4.6 and 6.8-8.7 ppm) from one healthy subject and patients with mild, moderate,
and severe renal damage. Abbreviations: 3-HB, 3-hydroxybutyrate; Ac, acetate; Ala, alanine; Chl, choline headgroup containing
metabolites; Cit, citrate; Crn, creatinine; DMA, dimethylamine; Fm, formate; Gly, glycine; Glc, glucose; Hip, hippurate; His, histidine;
Lac, lactate; N-Acs, N-acetyl groups from glycoproteins; Phe, phenylalanine; TMAO, trimethylamine-N-oxide; Val, valine.
loadings plot (Figure 2b), variables (i.e., bins) contributing
similar information are grouped together, and hence, metabolites located at the left part (glucose) are positively correlated
with patients, whereas those located at the right part (hippurate, citrate, and creatinine) are positively correlated with
healthy subjects.
With PLS-DA, an improved separation was achieved still containing, however, a degree of overlapping between the two classes (Figure 2c). The model parameters for the explained variation R2 and the predictive capability Q2 were significantly high,
0.67 and 0.61, respectively (Table 2). The relative regression
coefficients plot (Figure 2d) showed that, in addition to the
Journal of Proteome Research • Vol. 6, No. 9, 2007 3763
research articles
Psihogios et al.
Figure 2. (a) PCA scores plot (PC1 vs PC2) of the 1H NMR urinary spectroscopic data from 162 subjects. The samples are colored for
the 77 patients (red squares) and for the 85 controls (blue triangles). (b) The corresponding loadings plot. (c) PLS-DA scores plot (PC1
vs PC2). (d) The corresponding regression coefficients for PC1. (e) OSC/PLS-DA scores plot (t1 vs t2): 85 healthy subjects (blue triangles),
25 patients with mild (red squares), 27 patients with moderate (green diamonds), and 25 patients with severe renal damage (purple
circles). The ellipses surrounding the samples denote each group. (f) The corresponding regression coefficients plot for t1. Abbreviations
as in Figure 1; Prot: proteins.
above-mentioned metabolites, glycine was found in relatively
higher levels in controls (positive coefficients), whereas lactate,
3764
Journal of Proteome Research • Vol. 6, No. 9, 2007
acetate, TMAO, and aliphatic moieties of proteins were found
in relatively higher levels in patients (negative coefficients).
research articles
Tubulointerstitial Lesions’ Severity in Patients with GN
Table 2. The PLS-DA and OSC/PLS-DA Parameters for the Patients-Controls Models
All subjects: 85 Controls - 77 Patients (25 mild - 27 moderate - 25 severe)
Groups
Total (Patients-Controls)
Mild-Control
Moderate-Control
Severe-Control
parameters
R2
Q2
R2
Q2
R2
Q2
R2
Q2
PLS-DA
OSC/PLS-DA
0.67
0.80
0.61
0.73
0.63
0.75
0.39
0.58
0.73
0.81
0.62
0.70
0.85
0.89
0.79
0.84
To determine the ability of the 1H NMR-based metabonomic
analysis to distinguish the severity of the tubulointerstitial
injury, PLS-DA was also applied to compare the healthy
subjects with each one of the 3 patient groups characterized
by mild, moderate, and severe damage (Table 2). From the
corresponding R2 and Q2 parameters, it can be seen that the
more severe the renal damage is assessed, the more the values
and therefore the predictive capability of the model increase.
To minimize the possible intrinsic contribution of intersubject variability, the method of OSC filtering was applied from
which two orthogonal components were removed and PLSDA was repeated. Inspection of the two excluded orthogonal
components revealed the metabolites possessing the greatest
intersubject variability, that is, hippurate, creatinine, citrate,
and TMAO, that are known to exhibit a large biological
variation.26,31 The scores plot of the first two PCs from the
resulting OSC/PLS-DA model revealed a clear separation
between patients and controls, which was more distinct as the
renal damage deteriorated from mild to severe (Figure 2e). The
regression coefficients plot for the first component of the OSC/
PLS-DA model indicated that the spectroscopic regions contributing to the clustering were almost the same as in the
unfiltered data set, but with enhanced importance of the
coefficients, mainly for creatinine, hippurate, citrate, and
glucose (Figure 2f). The values of the corresponding R2 and Q2
parameters were improved to 0.80 and 0.73, respectively, as
well as those between each patient group and the controls
(Table 2). In Supplementary Figure 2 (Supporting Information),
the scores plots between each patient group and the controls
show clearly that the spectra from patients are placed away
from the control region following the severity of the disease.
On the basis of the values of the OSC/PLS-DA regression
coefficients, controls mainly excreted higher levels of citrate,
hippurate, and creatinine, whereas patients mainly excreted
higher levels of glucose and a group of unidentified metabolites
(3.70-3.74 ppm), choline headgroup containing metabolites,
proteins, and acetate (Table 3).
Additional models were developed to compare the three
patient groups together as well as pairwise (Table 4, Figure 3).
As it is indicated by the R2 and Q2 parameters of the PLS-DA
models in Table 4, the separation between the three patient
groups was of low significance. In the pairwise comparison,
moderate damage group was partially separated from both mild
and severe groups, whereas a more distinct separation was
observed between those of mild and severe damage (R2 ) 0.73
and Q2 ) 0.46). The OSC/PLS-DA models were of similar
discriminating power (Table 4), except for the mild-severe
model, in which a higher and significant predictive capability
(0.46 vs 0.55) was noted, seen also in the corresponding scores
plot (Figure 3d). On the basis of the values of the regression
coefficients, the metabolites that predominantly contributed
to the separation of the moderate from the mild damage group
were citrate, creatine, phenylalanine, glucose, and a group of
unidentified metabolites (3.70-3.74 ppm) and from the severe
damage group were creatinine, acetate, hippurate, glucose, a
Table 3. The OSC/PLS-DA Regression Coefficients for the
Patients-Controls Model
Patients-Controls
a
loadings
metabolites
C
Pa
Coefficients
0.94-0.98
1.34
1.94
2.54, 2.70
3.04, 4.04
3.22-3.30
3.58
3.34-3.90
3.70-3.74
3.94
3.98, 7.82
7.38, 7.46
Proteins
Lactate
Acetate
Citrate
Creatinine
Choline metabolites
Glycine
GlucoseUnidentifiedb
Creatine
Hippurate
Phenylalanine
V
V
V
v
v
V
v
V
v
v
v
V
V
v
V
v
0.22
0.16
0.20
0.67
0.44
0.30
0.16
0.31
v
v
V
V
V
v
0.06
0.55
0.1
a
Abbreviations: C, controls; P, patients. b A group of unidentified metabolites (3.70-3.74 ppm) probably from overlapping resonances from
glucose, other sugars, and R-protons of amino acids.42
group of unidentified metabolites (3.70-3.74 ppm), and the
aliphatic moieties of proteins (Table 5). The metabolites that
mainly contributed to the separation of the mild from the
severe damage group were citrate, creatinine, hippurate, creatine, glucose, a group of unidentified metabolites (3.70-3.74
ppm), the aliphatic moieties of proteins, and phenylalanine
(Table 5).
Exclusion of the 0.2-1.82 ppm Region. Since proteinuria
often characterizes renal patients,25 the spectral region containing bulk signals from the aliphatic moieties of proteins (0.21.82 ppm) was excluded, and OSC/PLS-DA was repeated with
the residual spectral data. The new models were still able to
distinguish the patient groups with a similar predictive capability (Table 4), whereas the spectral region attributed to proteinuria was not able on its own to distinguish the three groups
(Table 4 and Supplementary Figures 3 and 4 in Supporting
Information).
Prediction of Class Membership. To test the reliability of
the OSC/PLS-DA models between patients and controls and
between patients with mild and severe renal damage, validation
with both held-back and external data was carried out.
1. Held-Back Data Validation. For each model, the corresponding training and test sets were randomly selected, and
validation was repeated 3 times with a new random selection
of equally numbered sets each time.
For the patients-controls model, 132 samples (62 patients/
70 controls) from the data set were selected as the training set,
and the remaining 30 samples (15 patients/15 controls) served
as the test set (Supplementary Figure 5 in Supporting Information). As seen in Table 6, the R2 value of the models for the
three repeats was 0.80 in all cases, and the Q2 values ranged
from 0.71 to 0.72. The classification rate for both patients and
controls was 100% (15 out of 15) in all cases. Sensitivity and
specificity that were calculated on the basis of a predicted Y
cutoff of 0.5 were both 100%
Journal of Proteome Research • Vol. 6, No. 9, 2007 3765
research articles
Psihogios et al.
Figure 3. (a-d) OSC/PLS-DA scores plots (PC1 vs PC2) of the urinary spectroscopic data from the 77 patients: (a) 25 with mild (red
squares), 27 with moderate (green diamonds), and 25 with severe renal damage (purple circles); (b) 25 with mild (red squares) and 27
with moderate renal damage (green diamonds); (c) 27 with moderate (green diamonds) and 25 with severe renal damage (purple
circles); (d) 25 with mild (red squares) and 25 with severe renal damage (purple circles).
Table 4. Parameters of the PLS-DA and OSC/PLS-DA Models for the Three Patient Groups Together and Pairwise for the Full
Spectrum, after the Exclusion of the Bulk Signals from the Aliphatic Moieties of Proteins (0.2-1.82 ppm) and for the Aliphatic
Region Alone
All patients: 25 mild - 27 moderate - 25 severe
Groups
Mild-Moderate-Severe
R2
Q2
Full spectrum
0.39
0.12
Full spectrum
0.2-1.82 ppm excluded
Region 0.2-1.82 ppm
0.43
0.36
0.19
0.16
0.10
0.04
parameters
Mild-Moderate
R2
PLS-DA
0.66
OSC/PLS-DA
0.72
0.64
0.37
The training set for the mild-severe model consisted of 34
patients (17 with mild and 17 with severe renal damage) and
the test set of 16 patients (8 with mild and 8 with severe renal
damage) (Supplementary Figure 6 in Supporting Information).
As seen in Table 6, R2 ranged from 0.81 to 0.84 and Q2 from
0.41 to 0.43, whereas the classification rate was 100% for the
mild renal damage sets (8 out of 8) and 88% for the severe renal
damage sets (7 out of 8) in all cases.
2. External Validation. For each model, the corresponding
training and prediction sets were randomly selected, and
3766
Journal of Proteome Research • Vol. 6, No. 9, 2007
Q2
Moderate-Severe
Mild-Severe
R2
Q2
R2
Q2
0.24
0.59
0.17
0.73
0.46
0.31
0.28
<0
0.60
0.59
0.39
0.23
0.14
0.15
0.79
0.80
0.45
0.55
0.61
0.16
validation was repeated 3 times with equally numbered sets
of different samples each time, which were not used when the
models were built.
For the patients-controls model, 114 samples (53 patients/
61 controls) from the data set defined the training set, and 48
samples (24 patients/24 controls) served as the prediction set.
In Figure 4a, the Y-predicted scatter plot of the second repeat
is shown with the Y cutoff of 0.5 for classification, whereas the
other two plots are shown in Supplementary Figure 7 (Supporting Information). As seen in Table 6, R2 ranged from 0.81
research articles
Tubulointerstitial Lesions’ Severity in Patients with GN
Table 5. The OSC/PLS-DA Regression Coefficients for the Three Patient Models Shown in Pairwise Comparison
Patient models
Mild-Moderate
Moa
Moderate-Severe
Mild-Severe
loadings
metabolites
Mia
coefficients
Mo
Sea
coefficients
Mi
Se
coefficients
0.94-0.98
1.34
1.94
2.54, 2.70
3.04, 4.04
3.22-3.30
3.58
3.34-3.90
3.70-3.74
3.94
3.98, 7.82
7.38, 7.46
Proteins
Lactate
Acetate
Citrate
Creatinine
Choline metabolites
Glycine
GlucoseUnidentifiedb
Creatine
Hippurate
Phenylalanine
V
V
V
v
V
V
v
V
v
v
v
V
v
v
V
v
0.26
0.29
0.20
0.46
0.29
0.33
0.33
0.34
V
v
v
v
v
v
v
V
v
V
V
V
V
V
V
v
0.32
0.16
0.42
0.27
0.51
0.23
0.07
0.33
V
V
v
v
v
v
v
V
v
v
V
V
V
V
V
v
0.34
0.03
0.24
0.49
0.36
0.23
0.24
0.34
v
v
v
V
V
V
0.35
0.09
0.34
v
v
v
V
V
V
0.28
0.35
0.06
v
v
V
V
V
v
0.34
0.35
0.25
a
Abbreviations: Mi, mild; Mo, moderate; Se, severe. b A group of unidentified metabolites (3.70-3.74 ppm) probably from overlapping resonances from
glucose, other sugars, and R-protons of amino acids.42
Figure 4. Y-predicted scatter plots of the OSC/PLS-DA models validated with external data. (a) Second repeat of the patients-controls
model. The samples are colored for the training set: 61 controls (blue triangles) and 53 patients (red squares); for the prediction set:
24 controls ([C) and 24 patients ([P). (b) Second repeat of the mild-severe model. The samples are colored for the training set: 17
patients with mild (red squares) and 17 with severe renal damage (purple circles); for the prediction set: 8 with mild ([g1) and 8 with
severe renal damage ([g3).
to 0.90 and Q2 from 0.73 to 0.88. The average classification rate
was 100% for controls (24 out of 24) and 85% for patients (20.33
out of 24). The calculated sensitivity and specificity were 96%
and 99%, respectively.
Similarly for the mild-severe patient group, the training set
comprised 34 patients (17 with mild and 17 with severe renal
damage), and the prediction set comprised 16 patients (8 with
mild and 8 with severe renal damage). In Figure 4b, the
Y-predicted scatter plot of the second repeat is shown with the
Y cutoff of 0.5 for classification, whereas the other two plots
are shown in Supplementary Figure 8 (Supporting Information).
As seen in Table 6, R2 ranged from 0.84 to 0.87 and Q2 from
0.29 to 0.44. The average classification rate was 83% for the
mild renal damage sets (6.67 out of 8) and 79% for the severe
renal damage set (6.33 out of 8).
Discussion
The application of pattern recognition techniques in NMRbased urinalysis for the evaluation of renal damage has been
predominantly focused on experimental studies.9 In the current
study, an 1H NMR-based metabonomic approach was used for
the first time to investigate the correlation of histopathologically
assessed tubulointerstitial lesions with the urinary metabolite
profile in patients with glomerulonephritides.
The urinary metabolite profiles of the patients with glomerulonephritis at any disease stage presented distinct alterations
from those recorded from healthy individuals. These alterations
were more obvious in patients with severe renal disease and
reflected the extent of the damage in the proximal tubules and/
or the tubulointerstitial tissue as it was assessed by the
histopathological analysis. Similar alterations have been seen
in previously published experimental and human studies.9,12,13,32
The onset of the tubulointerstitial lesions is characterized by
decreased excretion of citrate, hippurate, glycine, and creatinine, whereas further deterioration is followed by glycosuria,
selective aminoaciduria, total depletion of citrate and hippurate, and gradual increase in the excretion of lactate, acetate,
and TMAO.
Journal of Proteome Research • Vol. 6, No. 9, 2007 3767
research articles
Psihogios et al.
Table 6. OSC/PLS-DA Parameters and Classification Scores of the Patients-Controls and the Mild-Severe Models Validated with
Held-Back and External Dataa
OSC/PLS-DA
Patients-Controls
Training set
held-back data
132 (62P/70C)
Test set
30 (15P/15C)
First model
Second model
Third model
external data
First model
Second model
Third model
a
Training set
Prediction set
114 (53P/61C)
48 (24P/24C)
Parameters
Mild-Severe renal damage
Classification
R2
Q2
controls
patients
0.80
0.80
0.80
0.72
0.71
0.71
15/15b
15/15
15/15
15/15
15/15
15/15
Parameters
Classification
R2
Q2
controls
patients
0.81
0.90
0.90
0.73
0.88
0.87
24/24
24/24
24/24
19/24
20/24
22/24
Training set
34 (17Mi/17Se)
Test set
16 (8Mi/8Se)
Training set
Prediction set
34 (17Mi/17Se)
16 (8Mi/8Se)
Parameters
Classification
R2
Q2
mild
severe
0.81
0.82
0.84
0.41
0.43
0.42
8/8
8/8
8/8
7/8
7/8
7/8
Parameters
Classification
R2
Q2
mild
severe
0.84
0.87
0.84
0.29
0.44
0.42
6/8
7/8
7/8
6/8
7/8
6/8
Abbreviations: C, controls; P, patients; Mi, mild; Se, severe. b Fifteen out of 15 correctly classified.
Depletion of urinary citrate has been attributed to either an
impairment of the tricarboxylic acid cycle or to renal tubular
acidosis, which typically appears as part of a generalized
proximal tubule dysfunction.9,33,34 A significant decrease of
hippurate in urine may be indicative of a metabolic alteration
and, even more importantly, of the efficacy of tubular secretion,35 whereas increased renal hippurate synthesis would
require over-utilization of glycine that could account for the
low levels of glycine detected in patients with moderate and
severe renal damage.26 The urinary levels of acetate can be
affected by the metabolic status of the organism,36 and increased excretion has been reported in proximal tubular
necrosis after exposure to HgCl2.9
Lactic aciduria has been related to increased activity of
anaerobic metabolic pathways, to decreased proximal tubular
reabsorption, and also appears to be a general marker of renal
cortical necrosis.9,37 The pattern of selective aminoaciduria,
lactic aciduria, and glycosuria that were detected in the present
study indicate impairment of the reabsorption mechanisms in
the proximal tubular epithelial cells.38
Leakage of methylamines in urine, mainly TMAO and DMA,39
has been reported in medullary damage9 and in acute graft
rejection following renal transplantation.11 In the present study,
elevated excretion of TMAO was detected mainly in the
moderate and severe damage groups, but it was not followed
by increased excretion of DMA indicating tubulointerstitial
distortions rather than papillary necrosis.13
The application of PR methods allowed the extraction of the
most discriminate information from the multivariate NMR data
and an effective sample classification. PCA along with visual
inspection of the raw data was important for the detection of
outliers in order to assess a consistent metabonomic approach.
Through PLS-DA, a strong separation trend between patient
and control groups was detected, whereas the application of
OSC-filtering led to the elimination of the intersubject variation
and enabled a clear separation in the resulting models. These
models were able to predict the presence of renal damage with
a sensitivity of 96% and a specificity of 99% based on a 95%
confidence limit for class membership. Patients with mild,
moderate, and severe tubulointerstitial lesions were progressively differentiated from the healthy individuals. The comparison between groups showed a statistically significant
separation between patients with mild and severe lesions and
a high predictive ability of the corresponding OSC/PLS-DA
models. Concerning the comparisons between moderate and
3768
Journal of Proteome Research • Vol. 6, No. 9, 2007
mild and between moderate and severe damage groups, the
separation was not statistically significant, but quite evident
in the OSC/PLS-DA models.
It is of interest that similar findings were also observed in
conventional clinical chemistry analysis. However, it is wellknown that the degree of proteinuria is not always correlated
to the severity of the interstitial damage. Proteinuria usually
reflects an increase in glomerular permeability that allows the
filtration of normally nonfiltered macromolecules such as
albumin, and thus, tubular proteinuria may be masked by an
overt glomerular hyper-filtration.40 On the other hand, serum
creatinine and GFR estimation are affected by factors such as
blood pressure levels and hydration of the patients.41 Therefore,
NMR findings further support the existence of the tubulointerstitial lesions and could be a useful tool to the global
assessment of kidney damage and contribute to the attenuation
of the confounding factors mentioned above.
In conclusion, since the coexistence of tubular and interstitial
lesions in glomerulonephritides is of crucial prognostic importance for the progress of renal glomerular function, NMR-based
metabonomic urinalysis, as a rapid and noninvasive technique,
could contribute to the early evaluation of the severity of the
renal damage and possibly to the monitoring of the kidney
function. This last indication is currently being evaluated in
our laboratory.
Abbreviations: LMW, low molecular weight; 1H NMR, proton
nuclear magnetic resonance; PR, pattern recognition; MDRD,
modification of diet in renal disease; TSP, 3-trimethylsilyl(2,2,3,3-2H4)-1-propionate; FID, free induction decay; PCA,
principal component analysis; PC, principal component; PLSDA, partial least-squares discriminant analysis; OSC, orthogonal
signal correction; 3-HB, 3-hydroxybutyrate; TMAO, trimethylamine-N-oxide; DMA, dimethylamine.
Acknowledgment. The research Project is co-funded by
the European Union - European Social Fund (ESF) & National
Sources, in the framework of the program “HRAKLEITOS” of
the “Operational Program for Education and Initial Vocational
Training” of the 3rd Community Support Framework of the
Hellenic Ministry of Education. The NMR spectra were recorded
on a 500 MHz Bruker DRX NMR instrument, at the NMR
Laboratory of NCSR “Demokritos”, Agia Paraskevi, Greece.
Supporting Information Available: Figures showing
the 1H NMR 2D TOCSY spectra of urine from a patient with
Tubulointerstitial Lesions’ Severity in Patients with GN
severe renal damage; OSC/PLS-DA scores plots of the urinary
spectroscopic data from controls and patients with mild,
moderate, and severe renal damage; OSC/PLS-DA scores plots
of the urinary spectroscopic data, after the removal of the
aliphatic moieties of proteins from patients with mild, moderate, and severe renal damage; OSC/PLS-DA scores plots of the
urinary spectroscopic data of the aliphatic moieties of proteins
from patients with mild, moderate, and severe renal damage;
Y-predicted scatter plots of the 3 repeats of the OSC/PLS-DA
patients-controls and mild-severe models validated with heldback data; and Y-predicted scatter plots of the other 2 OSC/
PLS-DA patients-controls and mild-severe models validated
with external data. This material is available free of charge via
the Internet at http://pubs.acs.org.
References
(1) Feehally, J.; Floege, J.; Savill, J.; Turner, A. N. Glomerular injury
and glomerular response. In Oxford Textbook of Clinical Nephrology, 3rd ed.; Davison, A. M., Cameron, J. S., Grünfeld, J.-P., et
al, Eds; Oxford University Press: New York, 2005; pp 367-387.
(2) Nath, K. A. Tubulointerstitial changes as a major determinant in
the progression of renal damage. Am. J. Kidney Dis. 1992, 20,
1-17.
(3) Howie, A. J. Handbook of Renal Biopsy Pathology; Kluwer
Academic Publishers: Dordrecht, The Netherlands, 2001.
(4) Mackensen-Haen, S.; Bader, R.; Grund, K. E.; Bohle, A. Correlations between renal cortical interstitial fibrosis, atrophy of the
proximal tubules and impairment of the glomerular filtration rate.
Clin. Nephrol. 1981, 15, 167-171.
(5) Nicholson, J. K.; Wilson, I. D. High resolution proton magnetic
resonance spectroscopy of biological fluids. Prog. Nucl. Magn.
Reson. Spectrosc. 1989, 21, 449-501.
(6) Neild, G. H.; Foxall, P. J.; Lindon, J. C.; Holmes, E. C.; Nicholson,
J. K. Uroscopy in the 21st century: high-field NMR spectroscopy.
Nephrol. Dial. Transplant. 1997, 12, 404-417.
(7) Videen, J. S.; Ross, B. D. Proton nuclear magnetic resonance
urinalysis: coming of age. Kidney Int. Suppl. 1994, 47, S122S128.
(8) Nicholson, J. K.; Connelly, J.; Lindon, J. C.; Holmes, E. Metabonomics: A platform for studying drug toxicity and gene function.
Nat. Rev. Drug Discovery 2002, 1, 153-161.
(9) Gartland, K. P.; Bonner, F. W.; Nicholson, J. K. Investigations into
the biochemical effects of region-specific nephrotoxins. Mol,
Pharmacol. 1989, 35, 242-250.
(10) Foxall, P. J.; Mellotte, G. J.; Bending, M. R.; Lindon, J. C.;
Nicholson, J. K. NMR spectroscopy as a novel approach to the
monitoring of renal transplant function. Kidney Int. 1993, 43,
234-245.
(11) Le Moyec, L.; Pruna, A.; Eugene, M.; Bedrossian, J.; Idatte, J. M.;
Huneau, J. F.; Tome, D. Proton nuclear magnetic resonance
spectroscopy of urine and plasma in renal transplantation followup. Nephron 1993, 65, 433-439.
(12) Knubovets, T. L.; Lundina, T. A.; Sibeldina, L. A.; Sedov, K. R. 1H
NMR urinalysis in glomerulonephritis: a new prognostic criterion. Magn. Reson. Imaging 1992, 10, 127-134.
(13) Lundina, T. A.; Knubovets, T. L.; Sedov, K. R.; Markova, S. A.;
Sibelin, L. A. Variability of kidney tubular interstitial distortions
in glomerulonephritis as measured by 1H-NMR urinalysis. Clin.
Chim. Acta 1993, 214, 165-173.
(14) Bell, J. D.; Lee, J. A.; Lee, H. A.; Sadler, P. J.; Wilkie, D. R.;
Woodham, R. H. Nuclear magnetic resonance studies of blood
plasma and urine from subjects with chronic renal failure:
Identification of trimethylamine-N-oxide. Biochim. Biophys. Acta
1991, 1096, 101-107.
(15) Bairaktari, E.; Liamis, G.; Tsolas, O.; Elisaf, M. Partially reversible
renal tubular damage in patients with obstructive jaundice.
Hepatology 2001, 33, 1365-1369.
(16) Bairaktari, E.; Seferiadis, K.; Liamis, G.; Psihogios, N.; Tsolas, O.;
Elisaf, M. Rhabdomyolysis-related renal tubular damage studied
by proton nuclear magnetic resonance spectroscopy of urine.
Clin. Chem. 2002, 48, 1106-1109.
(17) Zuppi, C.; Messana, I.; Tapanainen, P.; Knip, M.; Vincenzoni, F.;
Giardina, B.; Nuutinen. M. Proton nuclear magnetic resonance
spectral profiles of urine from children and adolescents with type
1 diabetes. Clin. Chem. 2002, 48, 660-662.
research articles
(18) Lindon, J. C.; Nicholson, J. K.; Holmes, E.; Everett, J. R. Metabonomics: metabolic processes studied by NMR spectroscopy of
biofluids combined with pattern recognition. Concepts Magn.
Reson. 2000, 12, 289-320.
(19) Trygg, J.; Holmes, E.; Lundstedt, T. Chemometrics in metabonomics. J. Proteome Res. 2007, 6, 469-479.
(20) Eriksson, L.; Johansson, E.; Kettaneh-Wold, N.; Wold, S. Multiand Megavariate Data Analysis, Principles and Applications;
Umetrics AB: Umea, Sweden, 2001.
(21) Nicholson, J. K.; Lindon, J. C.; Holmes, E. “Metabonomics”:
understanding the metabolic responses of living systems to
pathophysiological stimuli via multivariate statistical analysis of
biological NMR spectroscopic data. Xenobiotica 1999, 29, 11811189.
(22) Holmes, E.; Nicholls, A. W.; Lindon, J. C.; Connor, S. C.; Connelly,
J. C.; Haselden, J. N.; Damment, S. J.; Spraul, M.; Neidiq, P.;
Nicholson, J. K. Chemometric models for toxicity classification
based on NMR spectra of biofluids. Chem. Res. Toxicol. 2000, 13,
471-478.
(23) Robertson, D. G.; Reily, M. D.; Sigler, R. E.; Wells, D. F.; Paterson,
D. A.; Braden, T. K. Metabonomics: Evaluation of nuclear
magnetic resonance (NMR) and pattern recognition technology
for rapid in vivo screening of liver and kidney toxicants. Toxicol.
Sci. 2000, 57, 326-337.
(24) Robertson, D. G.; Reily, M. D.; Baker, J. D. Metabonomics in
pharmaceutical discovery and development. J. Proteome Res.
2007, 6, 526-539.
(25) National Kidney Foundation. K/DOQI clinical practice guidelines
for chronic kidney disease: evaluation, classification, and stratification. Am. J. Kidney Dis. 2002, 39, S1-S266.
(26) Zuppi, C.; Messana, I.; Forni, F.; Ferrari, F.; Rossi, C.; Giardina,
B. Influence of feeding on metabolite excretion evidenced by
urine 1H NMR spectral profiles: a comparison between subjects
living in Rome and subjects living at arctic latitudes (Svaldbard).
Clin. Chim. Acta 1998, 278, 75-79.
(27) Levey, A. S.; Bosch, J. P.; Lewis, J. B.; Greene, T.; Rogers, N.; Roth,
D. A more accurate method to estimate glomerular filtration rate
from serum creatinine: a new prediction equation. Modification
of Diet in Renal Disease Study Group. Ann. Intern. Med. 1999,
130, 461-470.
(28) Nicholson, J. K.; Foxall, P. J.; Spraul, M.; Farrant, R. D.; Lindon,
J. C. 750 MHz 1H and 1H-13C NMR spectroscopy of human blood
plasma. Anal. Chem. 1995, 67, 793-811.
(29) Gavaghan, C. L.; Wilson, I. D.; Nicholson, J. K. Physiological
variation in metabolic phenotyping and functional genomic
studies: use of orthogonal signal correction and PLS-DA. FEBS
Lett. 2002, 530, 191-196.
(30) Bales, J. R.; Higham, D. P.; Howe, I.; Nicholson, J. K.; Sadler, P. J.
Use of high-resolution proton nuclear magnetic resonance
spectroscopy for rapid multi-component analysis of urine. Clin.
Chem. 1984, 30, 426-432.
(31) Psihogios, N.; Gazi, I.; Elisaf, M.; Seferiadis, K.; Bairaktari, E.
Gender-related and age-related urinalysis of healthy subjects by
NMR-based metabonomics. NMR Biomed. 2007, DOI: 10.1002/
nbm.1176.
(32) Foxall, P. J.; Singer, J. M.; Hartley, J. M.; Neild, G. H.; Lapsley, M.;
Nicholson, J. K.; Souhami, R. L. Urinary proton magnetic resonance studies of early ifosfamide-induced nephrotoxicity and
encephalopathy. Clin. Cancer Res. 1997, 3, 1507-1518.
(33) Simpson, D. P.; Angielski, S. Regulation by bicarbonate ion of
intramitochondrial citrate concentration in kidney mitochondria.
Biochim. Biophys. Acta 1973, 298, 115-123.
(34) Laing, C. M.; Toye, A. M.; Capasso, G.; Unwin, R. J. Renal tubular
acidosis: developments in our understanding of the molecular
basis. Int. J. Biochem. Cell Biol. 2005, 37, 1151-1161.
(35) Zuppi, C.; Messana, I.; Forni, F.; Rossi, C.; Pennacchietti, L.;
Ferrari, F.; Giardina, B. 1H NMR spectra of normal urines:
reference ranges of the major metabolites. Clin. Chim. Acta 1997,
265, 85-97.
(36) Roch-Ramel, F.; Besseghir, K.; Murer, H. Renal excretion and
tubular transport of organic anions and cations. In Handbook of
Renal Physiology; Windhager, E. E., Ed.; Oxford University Press:
New York, 1992; pp 2189-2262.
(37) Holmes, E.; Bonner, F. W.; Sweatman, B. C.; Lindon, J. C.; Beddell,
C. R.; Rahr, E.; Nicholson, J. K. Nuclear magnetic resonance
spectroscopy and pattern recognition analysis of the biochemical
processes associated with the progression of and recovery from
nephrotoxic lesions in the rat induced by mercury(II) chloride
and 2-bromoethanamine. Mol. Pharmacol. 1992, 42, 922-930.
Journal of Proteome Research • Vol. 6, No. 9, 2007 3769
research articles
(38) Anthony, M. L.; Gartland, K. P.; Beddell, C. R.; Lindon, J. C.;
Nicholson, J. K. Cephaloridine-induced nephrotoxicity in the
Fischer 344 rat: proton NMR spectroscopic studies of urine and
plasma in relation to conventional clinical chemical and histopathological assessments of nephronal damage. Arch. Toxicol.
1992, 66, 525-537.
(39) Al-Waiz, M.; Mitchell, S. C.; Idle, J. R.; Smith, R. L. The relative
importance of N-oxidation and N-demethylation in the metabolism of trimethylamine in man. Toxicology 1987, 43, 117-121.
(40) Rose, B. D.; Black, R. M. Manual of Clinical Problems in
Nephrology; Little, Brown: Boston, MA, 1988; p 130.
3770
Journal of Proteome Research • Vol. 6, No. 9, 2007
Psihogios et al.
(41) Klag, M. J.; Whelton, P. K.; Randall, B. L.; Neaton, J. D.; Brancati,
F. L.; Ford, C. E.; Shulman, N. B.; Stamler, J. Blood pressure and
end-stage renal disease in men. N. Engl. J. Med. 1996, 334, 1318.
(42) Ebbels, T. M.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Evaluation
of metabolic variation in normal rat strains from a statistical
analysis of 1H NMR spectra of urine. J. Pharm. Biomed. Anal.
2004, 36, 823-833.
PR070172W