Abstract
Early diagnosis of the onset of osteoporosis is key to the delivery of effective therapy. Biochemical markers of bone turnover provide a means of evaluating skeletal dynamics that complements static measurements of BMD by DXA. Conventional clinical measurements of bone turnover, primarily the estimation of collagen and its breakdown products in the blood or urine, lack both sensitivity and specificity as a reliable diagnostic tool. As a result, improved tests are needed to augment the use of BMD measurements as the principle diagnostic modality. In this study, the serum proteome of 58 postmenopausal women with high or low/normal bone turnover (training set) was analyzed by surface enhanced laser-desorption/ionization time-of-flight mass spectrometry, and a diagnostic fingerprint was identified using a variety of statistical and machine learning tools. The diagnostic fingerprint was validated in a separate distinct test set, consisting of serum samples from an additional 59 postmenopausal women obtained from the same Mayo cohort, with a gap of 2 yr. Specific protein peaks that discriminate between postmenopausal patients with high or low/normal bone turnover were identified and validated. Multiple supervised learning approaches were able to classify the level of bone turnover in the training set with 80% sensitivity and 100% specificity. In addition, the individual protein peaks were also significantly correlated with BMD measurements in these patients. Four of the major discriminatory peaks in the diagnostic profile were identified as fragments of interalpha-trypsin-inhibitor heavy chain H4 precursor (ITIH4), a plasma kallikrein-sensitive glycoprotein that is a component of the host response system. These data suggest that these serum protein fragments are the serum-borne reflection of the increased osteoclast activity, leading to the increased bone turnover that is associated with decreasing BMD and presumably an increased risk of fracture. In conjunction with the identification of the individual proteins, this protein fingerprint may provide a novel approach to evaluate high bone turnover states.
Key words: surface enhanced laser desorption/ionization time of flight mass spectrometry, biomarkers, fracture risk, bone turnover
INTRODUCTION
Osteoporosis is a skeletal disease characterized by low bone mass and structural deterioration of bone leading to bone fragility.(1) The increased bone fragility directly increases susceptibility to fractures, especially of the hip, spine, and wrist, although any bone can be affected.(2,3) In the United States, ∼10 million individuals are estimated to have the disease already, whereas ∼34 million more are estimated to have low bone mass (osteopenia), placing them at increased risk for osteoporosis. What is of major concern is that the onset of bone loss is asymptomatic. Currently, the single best method available for confirming the diagnosis of osteoporosis and assessing future risks is measuring BMD. A low BMD is known to contribute to increased fracture risk, which is a major source of the associated morbidity and mortality.(3–5)
However, although the level of BMD has been shown to be a strong and independent predictor of fracture risk in postmenopausal women, about one half of the patients with incident fractures have BMD values above the operating diagnostic threshold for osteoporosis (T-score ≥ −2.5).(6) Several recent prospective studies have shown that increased bone resorption, as determined by the measurement of biochemical markers, is strongly associated with increased risk of fractures independent of BMD.(7–10) When above-normal levels of bone resorption are combined with low BMD, the risk of fracture in otherwise healthy premenopausal women increases, further supporting the independent contributions of BMD and high bone turnover to fracture risk.(6)
Thus far, the majority of bone turnover biomarker discovery studies have focused on the identification and/or measurement of single protein fragments or individual proteins, all of which have an inherently high degree of variability.(7) This variability has a major impact on clinical interpretation, especially in osteoporosis, where subtle differences are found that seem to be related to the actual measurement of the individual biomarkers.(9) As such, there is an urgent need to identify clinically useful and sensitive biomarkers for early disease detection.
The advent of proteomic technologies such as surface enhanced laser desorption ionization (SELDI) time-of-flight mass spectrometry (TOF-MS) has provided the means to analyze directly a broad array of proteins of different physical properties in patient samples.(11–16) This technology has the capability to identify serum biomarker patterns without the need for a priori knowledge of their existence or relevance to a particular disease state.(16–19) We and others have used SELDI for the discovery of biomarkers important for the diagnosis of a variety of human cancers, including ovarian,(18,19) prostate,(15,20) pancreas,(12,21) breast,(22,23) and colon.(24) However, currently, the use of SELDI for the identification of biomarkers in nonneoplastic diseases is quite limited.
In this study, 58 archival serum samples from the well-characterized Rochester Mayo Clinic cohort(25,26) were used to identify a pattern of specific protein biomarker peaks that collectively discriminated between postmenopausal patients with high or low/normal bone turnover. In addition, to assess the true performance of the detected biomarkers and the reproducibility of the analysis, we performed independent validation on a separate sample set of 59 postmenopausal women that were analyzed with a gap of 2 yr. Four of the significant discriminatory peaks were purified, sequenced, and identified as proteolytic fragments of the interalpha-trypsin-inhibitor heavy chain H4 precursor (ITIH4) protein. We hypothesize that the identification of the ITIH4 fragments that were diminished in the high turnover patients reflects the specific signature fragmentation pattern generated by the increased activity of osteoclasts. In agreement with the notion that the biomarker signature reflects specific skeletal changes, the individual components of the identified fingerprint are also significantly correlated with clinical BMD measurements.
MATERIALS AND METHODS
Sample collection and preparation
Archival serum samples of 58 women (training set) and an additional 59 samples (validation set; obtained with a gap of 2 yr) from the well-characterized Rochester Minnesota cohort were analyzed.(25,26) Existing data, including measurements of BMD and serum and urinary markers of bone turnover, had already been obtained.(26)
Subjects were recruited from an age-stratified random sample of Rochester, MN, men and women that were selected using the medical records linkage system of the Rochester Epidemiology Project (http://mayoresearch.mayo.edu/mayo/research/rep/).(27) The enumerated population approximates the underlying population of the community, including both free-living and institutionalized individuals. A subset (the training set; see the flow chart of data analysis; Fig. 1) of 58 serum specimens from the cohort were classified into two groups (high versus low/normal turnover) based on their urine N-telopeptide of type I collagen (NTX) scores and analyzed by SELDI (see below). The women ranged in age from 60.6 to 88.8 yr, with a median age of 80.5 yr in the high turnover group (> +1 SD NTX T-score) and 70.8 yr in the low/normal turnover group (−2 to +1 SD NTX T-score). For the NTX measurements, T-scores were based on premenopausal women 30–39 yr of age. All postmenopausal women were not on estrogen replacement thearpy (ERT) or any other treatment known to effect BMD or turnover. The archival serum samples were stored in small aliquots at or below −80°C until processing, and no samples underwent more than one freeze-thaw cycle before analysis.
FIG. 1.
Flowchart depicting sample processing to spectra collection by SELDI-TOF MS methodology. Serum is fractionated according to pH and processed and analyzed. A SELDI protein profile is generated for every fraction of every patient sample (in duplicate).
Bone resorption was evaluated by measurement of serum levels of NTX in 24-h urine collections by an ELISA kit (Osteomark NTx Serum; Ostex, Seattle, WA, USA; interassay CV < 17%). BMD (g/cm2) was determined for the lumbar spine (L2–L4), total hip, and mid-distal radius and ulna using DXA with the QDR2000 instrument (Hologic, Waltham, MA, USA) using software version 5.40. The CVs for the anterior-posterior (AP) lumbar spine, total hip, and radius were 2.1%, 1.8%, and 1.7%, respectively.
Serum processing and fractionation
To increase the sensitivity of peak detection and to alleviate suppression of signal from low-abundance species by major components such as albumin, all serum samples were fractionated into six fractions containing proteins separated on the basis of their isoelectric point (Fig. 1). Serum samples were loaded into single wells of a 96-well filter plate prefilled with an anion exchange sorbent (Serum Fractionation kit; Ciphergen Biosystems, Fremont, CA, USA) and eluted in a stepwise pH gradient using a BIOMEK 2000 (Beckman Coulter) liquid-handling robot according to the manufacturer's protocol. The six fractions obtained in this stepwise fashion, designated F1 through F6, contained flow-through plus proteins eluted with buffers of pH 9, pH 7, pH 5, pH 4, pH 3, and organic solvent, respectively. Each serum sample was diluted ∼10-fold during fractionation in 50 mM Tris-HCl + 0.1% nonionic detergent with the pH adjusted as above for the different fractions.
Protein chip SELDI TOF-MS analysis
Three different chip chemistries (metal binding, anionic, and cationic; Ciphergen) were initially evaluated in a pilot study to determine which type provided the best spectra profiles in terms of peak number and resolution (data not shown). The IMAC30-Cu metal binding chip consistently captured the most peaks in the majority of the fractions and was selected for use in the analysis. The serum samples from each fraction were diluted 1:5-fold in PBS and applied in duplicate to wells of a 96-well bioprocessor containing 8-spot IMAC-30 protein chips, previously activated with 100 mM CuSO4, as described by the manufacturer. The bioprocessor was sealed and incubated with the samples for an hour with vigorous agitation on a Micromix 5 platform shaker. Excess sera was discarded, and the chips were washed three times with PBS and twice with deionized water before being removed from the bioprocessor and air dried for 20 min. A saturated solution of sinapinic acid in 50% acetonitrile and 0.5% trifluoroacetic acid (0.5 μl) was applied to each spot of the protein chip arrays. Each spot surface was allowed to dry for 10 min before another application of 0.5 μl of the sinapinic acid solution. All sample handling procedures were carried out using the BIOMEK 2000 robotic system, minimizing errors caused by human intervention.(17)
Protein chips were placed in the Protein Biological System II C mass spectrometer reader (Ciphergen), and the time-of- flight spectra were collected in two mass/charge ranges of 1.5–10 (low) and 7–30 (mid) kDa. All data acquisition parameters were optimized to detect peaks in the two mass/charge ranges. Spectra were collected by averaging 156 laser shots in the positive mode at laser intensities and detector sensitivities of 160 and 6, respectively, for the low mass range and 170 and 6, respectively, for the mid mass range. Mass accuracy was calibrated using the All-in-one peptide and All-in-one protein molecular weight standards (Ciphergen). Each chip generated included a randomly assigned commercially available quality control (QC) sample (pooled serum from normal healthy individuals) to assess reproducibility of the SELDI spectra. Five peaks with a signal-to-noise ratio of at least 5 were chosen randomly from each of the two mass ranges, and their CVs were calculated. These were combined to calculate a pooled CV of each mass range.
Acquisition and preprocessing of all spectral data were performed using Ciphergen ProteinChip software version 3.1. All peaks were baseline-corrected, and their intensities were normalized to the total ion current in each of the low and mid mass ranges separately. Spectra were compiled, and peaks that were consistently present across a minimum of 10% of the spectra with a signal-to-noise ratio of ≥2.0 were selected. Selected peaks were clustered using a second-pass peak selection with 0.3% of the mass window. Peak intensities from spectra compiled across all the six fractions were transformed to their base-2 logarithms and centered and scaled to mean = 0 and SD = 1.
Protein purification and identification
Protein purification was performed in two steps. First, the serum sample containing the protein peak of interest was denatured in 9 M urea, 2% CHAPS, and 50 mM Tris-HCl, pH 9, and subjected to fractionation by loading onto Q Ceramic HyperD F Anion Exchange columns (Ciphergen Biosystems) that were pre-equilibrated in the same buffer. Flow-through fractions were collected at pH 9, pH 7, pH 5, pH 4, pH 3, and organic solvent washes. The Q column fractions (fractions 1–6) were profiled onto IMAC30_Cu chips, and the spectra were collected.
In the second step, the pH 9 fraction was further purified on reverse-phase beads (Ciphergen Biosystems) pre-equilibrated with 10% Acentonitrile/0.1% TFA. The sample was adjusted to a final concentration of 10% Acentonitrile/0.1% trifluoro acetic acid (TFA) before mixing with the beads. The mixture was centrifuged, the supernatant was collected, and the bound proteins were eluted with increasing concentrations of Acetonitrile + TFA (1:0.001, vol/vol). The supernatant and the eluates were profiled onto IMAC30_Cu ProteinChip arrays, and spectra were collected.
Four discriminating peaks from the SELDI profile of mass/charge ratios 3961, 3978, 4287, and 4304 Da were purified from the flow through of the reverse-phase beads. Accurate and sensitive mass detection was first performed using a prOTOF 2000 (PerkinElmerSciEx), which is a matrix-assisted laser desorption ionization (MALDI) orthogonal time-of-flight mass spectrometer, with higher mass accuracy than the SELDI-TOF.(28) Identification of the selected biomarker peptide peaks was performed by sequencing the purified proteins with tandem mass spectrometry on a Q-TOFII system (MicroMass) equipped with a PCI 1000 ProteinChip Tandem MS interface.
Data analysis
Spectrum pairs from each sample were used to assess intrasample correlations by determining the Pearson correlation among pairs for each peak. The Pearson correlation coefficients had a median value of 0.76, with the interquartile range lying between 0.60 and 0.89. Even though samples were applied robotically in a randomized fashion on the protein chip surfaces, high median intrasample correlations were observed because of high spot-to-spot reproducibility of the SELDI system. Hence, the spectrum pairs from each patient were averaged together on a peak-by-peak basis for all subsequent data analysis.
Because the long-term objective was to identify a subset of peaks that are able to discriminate between high versus low/normal bone turnover in a clinical and/or diagnostic setting, SELDI peaks were first selected for their potential use if they showed at least a 1.5-fold change in the median intensity between groups and then analyzed for the statistical significance of the fold change. Statistical significance was assessed using both the Wilcoxon rank-sum test with t-approximation and multiple-comparison adjustments(29) using 100,000 random permutations of class labels. All statistical analyses were performed using SAS version 9.0 (SAS institute) and S-plus version 6.2 (Insightful Corp.) statistical software. Classification and regression tree analysis (CART) was performed using BioMarker pattern software (Ciphergen).
Principal components analysis
Principal components analysis (PCA) was performed on the original explanatory peak variables using the correlation matrix. t-tests were performed on the principal components with the sample group as the class variable, and the peak variables with the strongest loadings on the most significant principal component were selected.
Partial least squares-discriminant analysis
The partial least squares-discriminant analysis (PLS-DA) component generated from the PLS-DA model was determined by cross-validation, and the peaks with strongest weights on the PLS-DA component were selected. PLS-DA was also performed on the dataset containing only the first (the lowest) and the third (the highest) tertiles based on the NTX score and the peaks with strongest weights on the PLS component were selected.
Decision tree analysis with bootstrap resampling (bagging)
Biomarker Pattern software (Ciphergen) implements the CART statistical procedure as described(30) to build a decision tree. Classification trees were developed using the 27 peaks identified with the required fold change between groups as input predictors to build trees. Tree building was repeated to yield the best prediction success with the lowest error cost. Because the dataset is relatively small, the original training data set (58 samples) was divided into a training set (48 samples) and a test set (10 samples). A set of trees were generated from each of the slightly “perturbed” versions of the original training set that were generated, in turn, by resampling with replacement (100 times), a method known as bootstrap aggregation or bagging.(30) The trees were combined by an unweighted plurality-voting scheme, and the important peaks were selected based on their median importance scores.
ANOVA analysis of SELDI-TOF MS peaks with osteoporosis classification of the sample set based on BMD measurements
After correlation of the identified SELDI-TOF MS peaks with patient bone turnover status, the correlation of SELDI-TOF MS peaks with BMD was determined. In the analysis of the sample set based on BMD, the sample set was reclassified in two ways: one using only femoral neck BMD and the other using BMD of the femoral neck, spine, and mid-radius.
For femoral neck BMD correlations, each sample was classified as osteoporotic if the BMD measurement of the subject providing it was <2.5 SD below the average of a normal premenopausal adult woman. For the second classification, each sample was classified as osteoporotic if the subject's BMD measurements of femoral neck, spine, and mid-radius had a minimum value <2.5 SD below the average of a normal premenopausal adult woman. Multivariate regression analysis (general linear model) was performed with each SELDI peak with or without adjustments for the variables, age, and NTX. A p value <0.05 was considered significant.
RESULTS
Protein chip profiling of the training dataset
The training dataset (Fig. 1) of 58 serum specimens from the Mayo Clinic cohort that was classified into two groups (high [30 samples] versus low/normal [28 samples] turnover) based on their urine NTX scores was analyzed. A subset of the complied peaks after log transformation and standardization is shown (Fig. 2). A total of 242 individual peaks were resolved from all six fractions in the low and mid mass ranges covering 1.5–30 kDa. A total of 27 peaks met the potential use criterion of having a median fold change of 1.5 or more between the two groups; 11 of these were found by the Wilcoxon rank-sum test to be statistically significant, with adjusted p < 0.05. Several of the representative SELDI-TOF–generated peaks in the molecular weight ranges of 3.8–4.7 kDa are shown in the representative SELDI spectra (Fig. 3). The pooled CV of the low mass range was between 17.54–22.58% and 18.24–26.78% in the mid mass range across all the six fractions. The pooled CVs of the low and mid mass ranges were well within acceptable ranges for SELDI data.(31,32) The statistically significant peaks (seven in total) that discriminated the two sample groups in all algorithms were identified (Table 1).
FIG. 2.
Transformation and standardization of the raw peak intensities. (A) Raw peak intensities from a section of the compiled spectra. (B) The same peaks after log2 transformation. (C) After centering and scaling, the same peaks showed increased uniformity in their distribution.
FIG. 3.
Representative SELDI spectra showing the differences in normalized peak intensities of high bone turnover (High TO) vs. low/normal bone turnover (Low/normal TO) groups. The top six spectra depict the differences in expression levels in TraceView mode, whereas the bottom six depict the difference in GelView mode for the univariately significant peaks indicated by the arrows. Peaks not identified by arrows are not significantly different.
Table 1.
Seven Most Significant Peaks Identified From the Analysis Using Various Statistical and Data Mining Tools
| Peaks | Fold change (osp-effect) | Raw t-test p value | Raw WRS* p value | Adjusted p value† |
Peaks with top weights in |
Spearman correlations |
|||
| CART‡ | PCA§ | PLS-DA¶ | NTX | Femoral BMD | |||||
| F1lowM04287 | −2.7 | 0.0027 | 0.0029 | 0.0059 | √ | √ | √ | −0.48 | +0.22 |
| F1lowM03961 | −2.5 | 0.0037 | 0.0035 | 0.0059 | √ | √ | √ | −0.49 | +0.24 |
| F1lowM03978 | −3.2 | 0.0014 | 0.0071 | 0.0059 | √ | √ | √ | −0.49 | +0.20 |
| F1lowM07165 | +1.7 | 0.0160 | 0.0028 | 0.0181 | √ | +0.24 | −0.20 | ||
| F1lowM04304 | −2.7 | 0.0034 | 0.0037 | 0.0059 | √ | √ | −0.50 | +0.19 | |
| F1lowM04648 | −2.3 | 0.0039 | 0.0046 | 0.0059 | √ | √ | √ | −0.49 | +0.23 |
| F3lowM03192 | −2.2 | 0.0093 | 0.0024 | 0.0139 | √ | −0.44 | +0.28 | ||
* Wilcoxon rank-sum test with t-approximation.
† p values adjusted for multiple comparisons using false discovery rate.
‡ CART.
§ PCA.
¶ PLS-DA.
PCA
PCA was performed on the correlation matrix of the 27 peaks that met the potential use criterion. The first principal component explained 32.7% of the variance, the first three principal components explained 69.3% of the variance, and the first nine principal components explained 90.7% of the variance in the data. A plot of the samples against the first three principal components (Fig. 4) clearly indicated the efficient differentiation between the samples with high versus low/normal bone turnover. Clear differences between the sample groups are indicated along the planes of the first versus the second principal components and also the first versus the third principal components as seen from the 2D plots. t-tests were performed on the principal components, and the first principal component showed a highly significant (p < 0.0001) group difference. The peak variables with the strongest loadings on the first principal component were as follows: fraction 1, 4.287, 4.648, 3.978, 4.304, and 3.961 kDa; fraction 4, 9.192 kDa; fraction 3, 9.398 kDa; fraction 5, 9.193 kDa.
FIG. 4.
PCA. The first three principal components explained 69.3% of the variance. The plot of the samples against the first three principal components clearly indicates the efficient differentiation between the samples with or without high bone turnover. Clear differences between the sample groups are indicated along the planes of the first vs. the second principal components and also the first vs. the third principal components as shown in the 2D plots.
PLS-DA
PLS-DA performed on the first versus third tertile of the dataset also had a single significant PLS component (determined by cross-validation). The analysis showed that 10.3% of the variation in the predictor variables accounted for by this component could explain 58.2% of variation in the response variable indicating class membership. A 2D plot (data not shown) of the PLS x-scores plotted against the response variable y showed distinct separation in the two patient groups (high versus low/normal turnover) despite some overlap.
Decision tree analysis with bootstrap resampling (bagging)
The 27 input peak variables' importance scores generated from the 100 trees were compiled, and the top contributory peaks were selected based on their median importance scores. The major peaks from CART analysis were 3.961, 3.978, 4.287, and 7.165 kDa from fraction F1, 7,168 kDa from fraction F2, and 3.156 and 3.192 kDa from fraction F3 in the low mass ranges.
The median fold changes of the identified peaks in the two groups of patients are shown as the diagnostic fingerprint (Fig. 5). Peak levels are different by >2-fold, suggesting potential use for the development of putative diagnostic serum assays. Several of the representative SELDI-TOF mass spectrometer–generated peaks in the molecular weight ranges of 3.8–4.7 kDa are shown in the representative SELDI spectra in Fig. 3. Decision tree analysis with bootstrap resampling on the training dataset resulted in a sensitivity of 100% and a specificity of 80%.
FIG. 5.
Diagnostic fingerprint of postmenopausal women with high (striped bars) and low/normal (white bars) bone turnover. The median fold changes in the SELDI peak intensities are shown.
Validation of the diagnostic fingerprint on an independent dataset
The validation set was made up of serum samples from 59 postmenopausal women, of which 22 were high turnover and 37 were low/normal turnover patients based on their respective NTX T-scores (Table 2). When the test samples were analyzed 2 yr later, the spectral profiles were not identical (data not shown). However, once the SELDI-TOF MS data acquisition parameters were adjusted to match those of the training set, and all spectral intensities normalized and standardized (as described above), the protein ions of interest were found to be identical to those in the initial training data set (Fig. 6). The peak intensity distribution of the SELDI spectra of the validation set was compared with that of the training set by PCA, and no significant difference in the data distribution along the first three principal components was observed between the two datasets (data not shown), confirming the high degree of correlation and overlap between the two independent datasets.
Table 2.
Summary of Training and Validation Sets
| Sample set |
Classification (NTX based) |
Classification (BMD based) |
|||||
| Class | n | NTX T-score (SD)* | F neck† | T-score (SD) | At any site‡ | T-score (SD) | |
| Training | High turnover | 28 | +1 to +2 = 12 | N = 8 | > −1.0 | N = 8 | > −1.0 |
| > +2 = 16 | Opn = 21 | −1 to −2.5 | Opn = 12 | −1 to −2.5 | |||
| Low/normal turnover | 30 | −1 to +1 = 29 | Osp = 28 | < −2.5 | Osp = 38 | < −2.5 | |
| −2 to −1 = 1 | Unkn = 1 | ||||||
| Validation | High turnover | 22 | +1 to +2 = 6 | N = 7 | > −1.0 | N = 4 | > −1.0 |
| > +2 = 16 | Opn = 31 | −1 to −2.5 | Opn = 17 | −1 to −2.5 | |||
| Low/normal turnover | 37 | −1 to +1 = 36 | Osp = 21 | < −2.5 | Osp = 38 | < −2.5 | |
| −2 to −1 = 1 | |||||||
* SDs from the mean of premenopausal women 30–39 yr of age.
† Classification based on femoral neck BMD: N, normal; Opn, osteopenic; Osp, osteoporotic.
‡ Classification based on minimum BMD T-score at any of the three sites: femoral neck, lumbar spine, or mid-radius.
TO, turnover, F neck, femoral neck.
FIG. 6.
Mass spectra of 9 × 9 representative patient samples from each of the low/normal vs. high turnover groups depicted in trace view. Mass peaks indicated by the arrows indicate several of the significant peaks (p < 0.05) that discriminate the two sample groups that were present in the samples analyzed with a gap of 2 yr.
The SELDI spectra from nine different patients representing each of the high and low/normal turnover groups in the mass range of 3750–4750 Da from the training set were compared side-by-side to the validation set (Fig. 6). As shown, the significantly different peaks are observed to be discriminatory between the high and low/normal turnover groups in both the training and validation sets. The peak intensity levels were slightly lower in the validation set compared with the training set, perhaps because of technical issues such as sample processing at different times or slight differences in spectra acquisition parameters. Also, whereas the cluster of peaks containing 3961, 3978, and 3994 Da biomarkers is observed in the training set, the same cluster appears to contain only the peaks 3961 and 3978 Da in the validation set. Overall, these data show the high degree of reproducibility in the SELDI-based analysis and in the detection of the identified biomarker peaks.
Of the top seven discriminatory peaks observed in the training set, five were also statistically significantly (p < 0.05) different in the test set. These peaks were 4287, 3961, 3978, and 4648 Da from fraction F1 and 3192 Da from F3 in the low ranges. The peak 4304 Da that was observed in F1 was not statistically significantly different between the two groups (although observed). Similarly, the peak at 7165 Da, observed in F1 in the training set, was not significantly different in the validation set.
After the correlation of the identified SELDI peaks with patient bone turnover status, the entire dataset of SELDI peaks was correlated with patient BMD (Table 3). This analysis was performed to determine whether the serum proteome contained information that correlated with individual BMD status. The SELDI peaks were reclassified based on femoral neck BMD alone or using either femoral neck, spinal, or mid-radial BMD, based on the standard definitions of a T-score <2.5 SD. The SELDI-TOF MS peaks that showed statistically significant differences based on the NTX-based classifications were also significant in the BMD-based classifications but lost significance when adjusted for age and NTX. Another set of SELDI-TOF MS peaks that were not shown to be significant in NTX-based classifications were statistically significant (p < 0.05) in the BMD-based classifications after adjustment for age and NTX. Finally, a third set of SELDI peaks had statistical significance in either or both of the BMD-based classifications, with or without NTX and age adjustments.
Table 3.
ANOVA Analysis of SELDI Peaks With Osteoporosis Classifications Based on (A) Femoral Neck BMD or (B) Min of Fem_Neck/Spine/Mid_Rad BMD
| SELDI peak |
Classification FEM_NECK |
Classification MIN_T-SCORE |
||
| Significant unadjusted | Significant adjusted (for age, NTX) | Significant unadjusted | Significant adjusted (for age, NTX) | |
| F1lowM04287 | Yes (p < 0.02) | No | Yes (p < 0.02) | No |
| F1lowM03961 | Yes (p < 0.03) | No | Yes (p < 0.03) | No |
| F1lowM03978 | Yes (p < 0.04) | No | Yes (p < 0.03) | No |
| F1lowM04304 | Yes (p < 0.04) | No | Yes (p < 0.04) | No |
| F1lowM04648 | Yes (p < 0.04) | No | ||
| F3lowM04646 | No | Yes (p < 0.02) | ||
| F1midM08678 | No | Yes (p < 0.02) | ||
| F3midM07469 | No | Yes (p < 0.03) | ||
| F3midM11727 | No | Yes (p < 0.05) | ||
| F3midM11939 | No | Yes (p < 0.02) | ||
| F3midM15134 | No | Yes (p < 0.02) | No | Yes (p < 0.02) |
| F3midM15335 | No | Yes (p < 0.05) | ||
| F3midM15995 | No | Yes (p < 0.04) | No | Yes (p < 0.01) |
| F4lowM03885 | No | Yes (p < 0.02) | No | Yes (p < 0.01) |
| F5lowM04471 | No | Yes (p < 0.04) | ||
| F5lowM04635 | No | Yes (p < 0.03) | ||
| F5midM07809 | No | Yes (p < 0.03) | ||
| F5midM13912 | No | Yes (p < 0.02) | ||
| F1midM23555 | Yes (p < 0.02) | Yes (p < 0.04) | ||
| F2lowM04795 | Yes (p < 0.00) | Yes (p < 0.02) | ||
| F2midM09074 | Yes (p < 0.01) | Yes (p < 0.02) | ||
| F3lowM05066 | Yes (p < 0.01) | Yes (p < 0.03) | ||
| F3midM07808 | Yes (p < 0.03) | Yes (p < 0.01) | No | Yes (p < 0.04) |
| F3midM07923 | Yes (p < 0.01) | Yes (p < 0.05) | ||
| F3midM07970 | Yes (p < 0.02) | Yes (p < 0.03) | Yes (p < 0.03) | Yes (p < 0.03) |
| F3midM08141 | Yes (p < 0.01) | Yes (p < 0.01) | Yes (p < 0.02) | Yes (p < 0.03) |
| F3midM09493 | Yes (p < 0.04) | Yes (p < 0.04) | ||
| F4lowM05067 | Yes (p < 0.04) | Yes (p < 0.04) | ||
| F5midM08143 | Yes (p < 0.03) | Yes (p < 0.04) | ||
| F5midM08698 | Yes (p < 0.01) | Yes (p < 0.01) | ||
| F5midM08864 | Yes (p < 0.00) | Yes (p < 0.02) | ||
| F6midM11731 | Yes (p < 0.02) | Yes (p < 0.01) | ||
Bold, peaks significant in NTX-based classification; italic, peaks significant in classifications A and/or B when adjusted for covariates, age, and NTX; underline, peaks significant both in adjusted and unadjusted classifications A and/or B.
Identification of individual protein peaks
Four discriminating proteins of mass/charge ratios 3961, 3978, 4287, and 4304 were purified from the flow through of the reverse-phase beads. Accurate masses of two of the discriminatory peptides were determined using MALDI prOTOF and Q-TOF mass spectrometers.
The monoisotopic masses of the peptides whose median SELDI masses were 3961 and 3978 Da were found to be 3953.86 and 3969.86 Da, respectively. The accurate mass of the peptide at 4287 Da was found to be 4280.02 Da, and the mass of the peptide at 4303 Da was determined to be 4296.08 Da. All peptides were subsequently fragmented and sequenced by tandem MS/MS and identified by a MASCOT database search. The peptide with the higher mass of each pair that differed by 16 Da had an MS/MS spectrum virtually identical to the lower MW peptide. All four identified peptides represent proteolytic cleavage products of the same protein.
The peptides were fragmented and sequenced by tandem MS/MS and identified by a MASCOT database search as ITIH4. ITIH4 is a protein a plasma kallikrein-sensitive glycoprotein (120 kDa) that is a component of the host response system.(33,34) The MS/MS spectra of the representative 3953.68 Da peptide and the MASCOT database search are shown (Fig. 7).
FIG. 7.
MS/MS spectrum of the 3953.86 peptide (monoisotopic mass) to confirm identification of the SELDI peak at 3961 Da. The MASCOT search algorithm identified the peptide as a fragment of human Interα trypsin inhibitor heavy chain H4 or ITIH4.
DISCUSSION
A major challenge confronting the management of postmenopausal osteoporosis is the prediction of patients at increased risk of fracture. The serum proteome analysis described here accurately and reliably distinguished patients with high bone turnover from patients with low/normal bone turnover. The development of the diagnostic fingerprint (Fig. 5) depended on the mass spectrometry–based proteomic technology and thorough postacquisition analyses of the obtained spectra using statistical and data mining tools. Several unrelated analytical techniques were used to mine the data, ensure analytical robustness, and improve the prediction accuracy of the models obtained.
A set of seven peaks was identified that correlated with the measured bone resorption marker NTX (Table 1). Six peaks were downregulated in patients with high bone turnover and a single peak was upregulated. These peaks were determined to be significant by each of the different algorithms used, suggesting that these serum-derived peptide peaks reflect the cellular activity of the skeleton. As such, the peptides may have the potential for the clinical assessment of the status of skeletal homeostasis. In other studies correlating serum biomarkers with clinical BMD status, we detected a unique panel of serum protein biomarkers that identify patients with low BMD, independent of the clinically measured site (Table 3). Such biomarkers, when validated in larger patient populations, may have use as surrogate endpoints that augment the clinical measurement of BMD.
It has been suggested that SELDI tends to identify protein patterns that may not be related to the pathology of interest but rather represent the result of the technical limitations of the method.(35,36) These issues are associated with the time period of data collection; the particular instrument on which the data were collected; and sample handling and processing, which maybe specific to that dataset only. To counteract these concerns, independent validation in a separate data set, collected over a different time period, was performed to determine the true performance and reproducibility of the biomarkers identified. In this study, patient serum was analyzed using strict quality control (QC) criteria and optimized data acquisition and preprocessing parameters. Importantly, the technique identified five of the seven diagnostic peaks to be reproducibly and consistently discriminatory between the two groups of samples in the validation set, a subset of which were subsequently sequenced and identified as ITIH4, an acute phase host response protein.
In all, four ITIH4 fragments were identified and three were sequenced. The 4287-Da peptide is derived from an alternatively spliced variant of ITIH4 (Swiss Prot entry Q14264-2), and the other peptide (4304 Da) is the same peptide containing an oxidized methionine. The 3961/3978 Da fragment includes amino acids R649-R688 in the proline-rich region (PRR; adjacent to the carboxy-terminal) region of ITIH4 isoform 1, whereas the 4287-Da fragment covers residues R616 to R658 in the PRR of isoform 2. Interestingly, both peptides end with a classical Kallikrein cleavage site (Phe-Arg-Xaa).(32–34) Such ITIH4 peptide fragments have been reported as biomarkers in different types of cancer.(34)
The observed ITIH4 peptide fragments were downregulated in patients with high bone turnover compared with the low/normal turnover controls. Villanueva et al.(35) reported the presence of the 3970-Da fragment in bladder cancer samples and similar to the data described here, Koomen et al.(36) reported the downregulation of the 4287-Da species in pancreatic cancer patients compared with controls. It is tempting to speculate that, in the case of high bone turnover, the observed decrease in the specific ITIH4 fragments is related to the increased proteolysis that is associated with increased osteoclastic activity. We hypothesize that the identification of the ITIH4 fragments reflects the specific fragmentation pattern generated by the ongoing activity of osteoclasts. In cases of increased bone turnover (and presumably increased bone resorption), increased proteolytic cleavage and the associated decrease in abundance of specific peptide fragments are associated with the further elevation in osteoclast activity, known to be correlated with the measurement of the proteolytic fragment of collagen, NTX. As such, it is likely that further examination of the role of ITIH4 in osteoclast function may provide new insight into the pathophysiology of postmenopausal osteoporosis. We are currently evaluating the expression of ITIH4 in bone and whether it is stored within the bone matrix and is a substrate for enzymatic degradation by osteoclasts.
Despite the statistically significant discriminatory power in classifying high bone turnover patients from the normal/low turnover controls, questions do arise as to whether the ITIH4 fragments identified here have diagnostic or predictive value. These fragments have been identified in a number of other diseases including cancer(37–39) and inflammatory diseases.(40,41) However, they have not been associated previously with increased bone resorption, although both cancer and inflammatory diseases often share a skeletal component. Interestingly, several papers have suggested that proteolytic processing patterns of acute phase serum proteins such as ITIH4 may represent clinically surrogate markers for the detection and classification of different disease types.(33,34,37)
Based on the identification of the SELDI biomarker profile and the identification of ITIH4, our current working hypothesis (Fig. 8) schematically presents the combined contribution of the 13 SELDI peaks correlating with both BMD and NTX, which were the first two bone quality measures examined in this study. The total “biomarker space” is the combination of the SELDI peaks correlated with NTX alone (5 peaks) or BMD (14 peaks) alone or both parameters (13 peaks). Correlations with other bone quality parameters, such as trabecular architecture, volumetric BMD, etc., would likely extend the number of correlating peaks and expand the potential biomarker space.
FIG. 8.
Schematic of the investigated biomarker space. Specific SELDI peaks (5) correlate with serum NTx and 14 peaks correlate with BMD T-score, representing the individual components examined in this study. The interface between these two parameters (bone physiology) contains 13 SELDI peaks that may correlate with other parameters of skeletal metabolism and/or bone quality.
In summary, these data have identified ITIH4 as a possible circulating product of increased bone resorption. The identified serum biomarker profile allows the accurate discrimination of postmenopausal patients with high and low/normal bone turnover and also low BMD. The specific serum protein fingerprint identified here, including ITIH4, is the reflection in the serum of the increased osteoclast activity associated with elevated bone turnover, and presumably, decreased BMD. Further interrogation and subsequent independent validation of this unique biomarker profile represents the first step in the discovery of multiple biomarkers to improve the reliability of clinical measures of bone turnover. As such, the role of ITIH4 and the discriminating serum biomarker profile in bone resorption, as well as the identification and management of patients with osteoporosis, is intriguing. The evaluation of such diagnostic paradigms awaits further interrogation of the biomarker profile and additional and ongoing studies in expanded patient cohorts.
ACKNOWLEDGMENTS
These studies were supported by the Carl L. Nelson Chair of Orthopaedic Surgery (LJS) and a grant from the National Institutes of Health (AG04875 to SK).
Footnotes
The authors state that they have no conflicts of interest.
REFERENCES
- 1.U.S. Surgeon General. McLean, VA, USA: International Medical Publishing; 2004. Surgeon General Report: Bone Health and Osteoporosis. [Google Scholar]
- 2.National Institutes of Health. Osteoporosis prevention, diagnosis and therapy. JAMA. 2001;285:785–795. doi: 10.1001/jama.285.6.785. [DOI] [PubMed] [Google Scholar]
- 3.Kanis JA, Glüer CC. An update on the diagnosis and assessment of osteoporosis with densitometry. Osteoporos Int. 2001;11:192–202. doi: 10.1007/s001980050281. [DOI] [PubMed] [Google Scholar]
- 4.McGarry KA, Kiel DP. Postmenopausal osteoporosis: Strategies for preventing bone loss, avoiding fracture. Postgrad Med. 2000;108:79–91. doi: 10.3810/pgm.2000.09.1.1206. [DOI] [PubMed] [Google Scholar]
- 5.Kanis JA. Diagnosis of osteoporosis and assessment of fracture risk. Lancet. 2000;359:1929–1936. doi: 10.1016/S0140-6736(02)08761-5. [DOI] [PubMed] [Google Scholar]
- 6.Garnero P, Delmas PD. Contribution of bone mineral density and bone turnover markers to the estimation of risk of osteoporotic fracture in postmenopausal women. J Musculoskel Neuron Interact. 2004;4:50–63. [PubMed] [Google Scholar]
- 7.Delmas PD. Treatment of postmenopausal osteoporosis. Lancet. 2002;359:2018–2026. doi: 10.1016/S0140-6736(02)08827-X. [DOI] [PubMed] [Google Scholar]
- 8.Dogan E, Posaci C. Monitoring hormone replacement therapy by biochemical markers of bone metabolism in menopausal women. Postgrad Med J. 2002;78:727–731. doi: 10.1136/pmj.78.926.727. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Garnero P, Delmas PD. Contribution of bone mineral density and bone turnover markers to the estimation of riskof osteoporotic fracture in postmenopausal women. J Musculoskel Neuron Interact. 2004;4:50–63. [PubMed] [Google Scholar]
- 10.Garnero P, Sornay-Rendu E, Claustrat B, Delmas PD. Biochemical markers of bone turnover, endogenous hormones and the risk of fractures in postmenopausal women: The OFELY study. J Bone Miner Res. 2000;15:1526–1536. doi: 10.1359/jbmr.2000.15.8.1526. [DOI] [PubMed] [Google Scholar]
- 11.Merchant M, Weinberger SR. Recent advancements in surface enhanced laser desorption/ionization-time of flight-mass spectrometry. Electrophoresis. 2000;21:1164–1167. doi: 10.1002/(SICI)1522-2683(20000401)21:6<1164::AID-ELPS1164>3.0.CO;2-0. [DOI] [PubMed] [Google Scholar]
- 12.Bhattacharyya S, Siegel ER, Petersen GM, Chari ST, Suva LJ, Haun RS. Diagnosis of pancreatic cancer using serum proteomic profiling. Neoplasia. 2004;6:674–686. doi: 10.1593/neo.04262. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Bhattacharyya S, Epstein J, Suva LJ. Biomarkers that discriminate Multiple Myeloma Patients With or Without Skeletal Involvement Detected Using SELDI-TOF Mass Spectrometry and Statistical and Machine Learning Tools. Dis Markers. 2005;22:245–255. doi: 10.1155/2006/728296. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Marguiles AG, Klimberg VS, Bhattacharyya S, Gaddy D, Suva LJ. Genomics and proteomics of bone cancer. Clin Cancer Res. 2006;12:6217s–6221s. doi: 10.1158/1078-0432.CCR-06-1070. [DOI] [PubMed] [Google Scholar]
- 15.Kohli M, Siegel ER, Bhattacharyya S, Khan MA, Shah R, Suva LJ. Surface enhanced laser desorption/ionization time of flight-mass spectrometry (SELDI-TOF MS) for determining prognosis in advanced stage hormone relapsing prostate cancer. Cancer Biomarkers. 2006;2:249–258. doi: 10.3233/cbm-2006-2603. [DOI] [PubMed] [Google Scholar]
- 16.Bendre MS, Margulies AG, Walser B, Akel NS, Bhattacharyya S, Skinner RA, Swain F, Ramani V, Mohammad KS, Wessner LL, Martinez A, Guise TA, Chirgwin JM, Gaddy D, Suva LJ. Tumor-derived interleukin-8 stimulates osteolysis independent of the receptor activator of nuclear factor-κB ligand pathway. Cancer Res. 2005;65:11001–11009. doi: 10.1158/0008-5472.CAN-05-2630. [DOI] [PubMed] [Google Scholar]
- 17.Liotta LA, Petricoin EF. Serum peptidome for cancer detection: spinning biologic trash into diagnostic gold. J Clin Invest. 2006;116:26–30. doi: 10.1172/JCI27467. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Petricoin EF, Ardekani AM, Hitt BA, Levine PJ, Fusaro VA, Steinberg SM, Mills GB, Simone C, Fishman DA, Kohn EC, Liotta LA. Use of proteomic patterns in serum to identify ovarian cancer. Lancet. 2002;359:572–577. doi: 10.1016/S0140-6736(02)07746-2. [DOI] [PubMed] [Google Scholar]
- 19.Kozak KR, Amneus MW, Pusey SM, Su F, Luong MN, Luong SA, Reddy ST, Farias-Eisner R. Identification of biomarkers for ovarian cancer using strong anion-exchange ProteinChips: Potential use in diagnosis and prognosis. Proc Natl Acad Sci USA. 2003;100:12343–12348. doi: 10.1073/pnas.2033602100. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Qu Y, Adam BL, Yasui Y, Ward MD, Cazares LH, Schellhammer PF, Feng Z, Semmes OJ, Wright GL., Jr Boosted decision tree analysis of surface-enhanced laser desorption/ionization mass spectral serum profiles discriminates prostate cancer from noncancer patients. Clin Chem. 2002;48:1835–1843. [PubMed] [Google Scholar]
- 21.Koopmann J, Zhang Z, White N, Rosenzweig J, Fedarko N, Jagannath S, Canto MI, Yeo CJ, Chan DW, Goggins M. Serum diagnosis of pancreatic adenocarcinoma using surface-enhanced laser desorption and ionization mass spectrometry. Clin Cancer Res. 2004;10:860–868. doi: 10.1158/1078-0432.ccr-1167-3. [DOI] [PubMed] [Google Scholar]
- 22.Paweletz CP, Trock B, Pennanen M, Tsangaris T, Magnant C, Liotta LA, Petricoin EF., III Proteomic patterns of nipple aspirate fluids obtained by SELDI-TOF: Potential for new biomarkers to aid in the diagnosis of breast cancer. Dis Markers. 2001;17:301–307. doi: 10.1155/2001/674959. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Li J, Zhang Z, Rosenzweig J, Wang YY, Chan DW. Proteomics and Bioinformatics approaches for identification of serum biomarkers to detect breast cancer. Clin Chem. 2002;48:1296–1304. [PubMed] [Google Scholar]
- 24.Ward DG, Suggett N, Cheng Y, Wei W, Johnson H, Billingham LJ, Ismail T, Wakelam MJO, Johnson PJ, Martin A. Identification of serum biomarkers for colon cancer by proteomic analysis. Br J Cancer. 2006;94:1898–1905. doi: 10.1038/sj.bjc.6603188. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Perrien DS, Achenbach SJ, Bledsoe SE, Walser B, Suva LJ, Khosla S, Gaddy D. Bone turnover across the menopause transition: Correlations with inhibins and FSH. J Clin Endocrinol Metab. 2006;91:1848–1854. doi: 10.1210/jc.2005-2423. [DOI] [PubMed] [Google Scholar]
- 26.Khosla S, Melton LJ, III, Atkinson EJ, O'Fallon WM, Klee GG, Riggs BL. Relationship of serum sex steroid levels and bone turnover markers with bone mineral density in men and women: A key role for bioavailable estrogen. J Clin Endocrinol Metab. 1998;83:2266–2274. doi: 10.1210/jcem.83.7.4924. [DOI] [PubMed] [Google Scholar]
- 27.Melton LJ., III History of the Rochester Epidemiology Project. Mayo Clin Proc. 1996;71:266–274. doi: 10.4065/71.3.266. [DOI] [PubMed] [Google Scholar]
- 28.Collom SL, Jamakhandi AP, Tackett AJ, Radominska-Pandya A, Miller GP. CYP2E1 active site residues in substrate recognition sequence 5 identified by photoaffinity labeling and homology modeling. Arch Biochem Biophys. 2007;459:59–69. doi: 10.1016/j.abb.2006.10.028. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Breiman L, Friedman JH, Olshen RA, Stone CJ. Belmont, CA, USA: Wadsworth International Group; 1984. Classification and Regression Trees. [Google Scholar]
- 30.Breiman L. Bagging predictors. Mach Learn. 1996;24:123–140. [Google Scholar]
- 31.Vermeulen R, Lan Q, Zhang L, Gunn L, McCarthy D, Woodbury RL, McGuire M, Podust VN, Li G, Chatterjee N, Mu R, Yin S, Rothman N, Smith MT. Decreased levels of CXC-chemokines in serum of benzene-exposed workers identified by array-based proteomics. Proc Natl Acad Sci USA. 2005;102:17041–17046. doi: 10.1073/pnas.0508573102. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Suriano R, Lin Y, Ashok BT, Schaefer SD, Schantz SP, Geliebter J, Tiwari RK. Pilot study using SELDI-TOF-MS based proteomic profile for the identification of diagnostic biomarkers of thyroid proliferative diseases. J Proteome Res. 2006;5:856–861. doi: 10.1021/pr050349r. [DOI] [PubMed] [Google Scholar]
- 33.Fung ET, Yip TT, Lomas L, Wang Z, Yip C, Meng XY, Lin S, Zhang F, Zhang Z, Chan DW, Weinberger SR. Classification of cancer types by measuring variants of host response proteins using SELDI serum assays. Int J Cancer. 2005;115:783–789. doi: 10.1002/ijc.20928. [DOI] [PubMed] [Google Scholar]
- 34.Song J, Patel M, Rosenzweig CN, Chan-Li Y, Sokoll LJ, Fung ET, Choi-Miura NH, Goggins M, Chan DW, Zhang Z. Quantification of fragments of human serum Inter-alpha-trypsin Inhibitor heavy chain 4 by a surface-enhanced laser desorption/ionization based immunoassay. Clin Chem. 2006;52:1045–1053. doi: 10.1373/clinchem.2005.065722. [DOI] [PubMed] [Google Scholar]
- 35.Diamandis EP. Mass spectrometry as a diagnostic and a cancer biomarker discovery tool. Mol Cell Proteomics. 2004;3:367–378. doi: 10.1074/mcp.R400007-MCP200. [DOI] [PubMed] [Google Scholar]
- 36.Diamandis EP. Analysis of serum proteomic patterns for early cancer diagnosis: Drawing attention to potential problems. J Natl Cancer Inst. 2004;96:353–356. doi: 10.1093/jnci/djh056. [DOI] [PubMed] [Google Scholar]
- 37.Villanueva J, Shaffer DR, Philip J, Chaparro CA, Erdjument-Bromage H, Olshen AB, Fleisher M, Lilja H, Brogi E, Boyd J, Sanchez-Carbayo M, Holland EC, Cordon-Cardo C, Scher HI, Tempst P. Differential exoprotease activities confer tumor-specific serum peptidome patterns. J Clin Invest. 2006;116:271–284. doi: 10.1172/JCI26022. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Koomen JM, Shih LN, Coombes KR, Li D, Xiao LC, Fidler IJ, Abbruzzese JL, Kobayashi R. Plasma protein profiling for diagnosis of pancreatic cancer reveals the presence of host response proteins. Clin Cancer Res. 2005;11:1110–1118. [PubMed] [Google Scholar]
- 39.Zhang Z. Three biomarkers identified from serum proteomic changes for the detection of early stage ovarian cancer. Cancer Res. 2004;64:5882–5890. doi: 10.1158/0008-5472.CAN-04-0746. [DOI] [PubMed] [Google Scholar]
- 40.Whicher J, Biasucci L, Rifai N. Inflammation, the acute phase response and atherosclerosis. Clin Chem Lab Med. 1999;37:495–503. doi: 10.1515/CCLM.1999.080. [DOI] [PubMed] [Google Scholar]
- 41.Romette J, di Costanzo-Dufetel J, Charrel M. Inflammatory syndrome and changes in plasma proteins. Pathol Biol (Paris) 1986;34:1006–1012. [PubMed] [Google Scholar]