Platelet Factor-4 Is an Indicator of Blood Count Recovery in Acute Myeloid Leukemia Patients in Complete Remission*S

To investigate whether serum biomarkers can be used to indicate the responsiveness of acute myeloid leukemia to remission induction chemotherapy, we performed MALDI-TOF protein profile analysis of patient sera. The resulting spectra revealed a protein (or peptide) peak at m/z 7764 that varied in intensity; its intensity was much higher in samples from patients in complete remission than in those from patients with resistant disease or in samples taken prior to treatment (at the time of diagnosis). Using fractionation, trypsin digestion, MS/MS, and protein molecular weight analyses, we identified the m/z 7764 protein as platelet factor-4 (PF4). This identification was confirmed by a magnetic bead-based MALDI immunoassay. Statistical comparison of PF4 levels and platelet counts in patient sera revealed a significant positive correlation between the two variables. This study demonstrates that PF4 protein levels are a good indicator for the recovery of blood count in the complete remission of acute myeloid leukemia. The linear positive correlation curve indicates that blood count recovery of platelets to >100,000/mm3 is equivalent to a serum PF4 recovery level of >2.492 μg/ml.

To investigate whether serum biomarkers can be used to indicate the responsiveness of acute myeloid leukemia to remission induction chemotherapy, we performed MALDI-TOF protein profile analysis of patient sera. The resulting spectra revealed a protein (or peptide) peak at m/z 7764 that varied in intensity; its intensity was much higher in samples from patients in complete remission than in those from patients with resistant disease or in samples taken prior to treatment (at the time of diagnosis). Using fractionation, trypsin digestion, MS/MS, and protein molecular weight analyses, we identified the m/z 7764 protein as platelet factor-4 (PF4). This identification was confirmed by a magnetic bead-based MALDI immunoassay. Statistical comparison of PF4 levels and platelet counts in patient sera revealed a significant positive correlation between the two variables. This study demonstrates that PF4 protein levels are a good indicator for the recovery of blood count in the complete remission of acute myeloid leukemia. The linear positive correlation curve indicates that blood count recovery of platelets to >100,000/mm 3 is equivalent to a serum PF4 recovery level of >2.492 g/ml. Molecular & Cellular Proteomics 7: 431-441, 2008.
The acute myeloid leukemias (AMLs) 1 comprise a heterogeneous group of diseases characterized by infiltration of the blood, bone marrow, and other tissues by neoplastic myeloid cells of the hematopoietic system. The traditional "acute" designation of these diseases is a result of the rapidly fatal nature of the untreated diseases. The incidence of AML, which averages ϳ3.6 per 100,000 people per year, increases with age, so that it is 1.7 in individuals aged Ͻ65 years and 16.2 in those aged Ͼ65 years (1). Heredity, radiation, chemical and other occupational exposures, and drugs have been implicated in the development of AML, and the incidence of AML has increased significantly over the past 10 years.
The curability of AML is influenced by many pretreatment factors, including patient age, chromosome abnormality, previous hematological disorder, presenting leukocyte count, and leukemic cell characteristics. In addition to these variables, several treatment factors correlate with prognosis in AML including, most importantly, achievement of complete remission (CR) (1). Combination chemotherapy with N 4 -behenoyl-1-␤-D-arabinofuranosylcytosine (BH-AC) and idarubicin is one of the most commonly used CR induction regimens. BH-AC is usually administered as a continuous intravenous infusion at 200 -300 mg/m 2 /day for 7 days; idarubicin is also administered intravenously, usually at 12 mg/m 2 /day on days 1, 2, and 3 (the "7 and 3" regimen). All-trans-retinoic acid (ATRA) is also used only for remission induction of a subtype (M3) of AML. After induction chemotherapy, both bone marrow and blood are examined to determine whether the leukemia has been eliminated.
Major improvements in the treatment of AML have been achieved over the last 30 years. This improvement has been achieved mainly by an enormous intensification of chemotherapy (2). However, for many AML patients, even the most intense therapy does not prevent a fatal outcome. One of the greatest challenges in the treatment of AML is that of the design of risk-adapted therapies. To achieve this goal, it is essential to identify residual leukemic cells that can cause a relapse and hence necessitate further treatment for the patient. In hematological malignancies, monitoring of minimal residual disease and the defined indicators of CR has become one of the most important diagnostic tools for risk-adapted therapy. However, sensitive and specific markers for CR detection have yet to be identified.
The objective of this study was to investigate whether AML patient serum contains a biochemical marker that reflects the responsiveness of the disease to remission induction chemotherapy. Identification of a biomolecule whose levels are indicative of either CR or treatment-resistant disease (RD) after chemotherapy would provide an important clue to the biological mechanism of the human body's recovery from AML. Our results demonstrate that MALDI-TOF protein profile analysis of bone marrow sera of AML patients after CR induction chemotherapy can differentiate between CR and RD states. Moreover a high intensity peak of m/z 7764 is present in the MS spectra of CR patient sera but absent from the sera of RD patients and from sera previously taken at the time of the AML diagnosis.

EXPERIMENTAL PROCEDURES
Patients and Specimens-The serum specimens used in this study were only from the Genome Research Center for Hematopoietic Diseases of Chonnam National University Hospital, a tertiary care center. Patients who were diagnosed as having AML and underwent chemotherapy for a 1-month period from July 2003 to March 2005 were eligible for collection of bone marrow serum to be archived in the Genome Research Center for Hematopoietic Diseases. Clinical information was obtained by chart review, and all diagnoses were confirmed histologically in the bone marrow biopsy section. At the time of bone marrow and blood collection, all cases provided the informed consent of allowing their postexamination specimen to be used for any further research-only purpose, and the Institutional Review Board on Clinical Investigation reviewed and approved this study. Serum samples were immediately placed on ice for transport to the laboratory where they were centrifuged, aliquoted, and immediately frozen at Ϫ80°C until use.
Postchemotherapy Response Criteria-A CR designation applies to appropriately treated patients who survived at least 7 days after completion of the final dose of the initial course of treatment and who achieved a morphologic leukemia-free state (Յ5% blasts in an aspirate sample containing marrow spicules and Ն200 nucleated cells) and had an absolute neutrophil count of Ͼ1.000/l and a platelet count of Ն100,000/l. Blasts with Auer rods or persistence of extramedullary disease should not be present (supplemental Fig. 1I). An RD designation applies to appropriately treated patients who survived at least 7 days after completion of the final dose of the initial course of treatment but whose final posttreatment peripheral blood smear and/or bone marrow sample showed persistent AML (supplemental Fig. 1II) (3). In this retrospective study, the patients were divided into two groups, CR and RD, according to their response to idarubicin/ BH-AC and ATRA (only for M3 AML) remission induction chemotherapy (Table I). The collection of postchemotherapy bone marrow serum was made at the same time after 1-month chemotherapy for all the AML patients. The evaluation of chemotherapy response (i.e. CR or RD) for AML patients was done at least 7 days later. Of the archived specimens, 26 were from patients in the CR state of AML, and five were from patients in the RD state of AML. Two samples were taken from each patient; the first sampling occurred at the time of diagnosis of AML (PreCR and PreRD samples), and the second occurred immediately after the 1-month remission induction chemotherapy (CR and RD samples).
AnchorChip MALDI Measurement of Serum-Serum samples were processed using a magnetic bead (MB)-based fractionation method (Bruker Daltonics, Bremen, Germany) according to the manufacturer's protocol. Serum peptides and proteins were fractionated using C 8 beads after evaluation for serum profiling. Five microliters of bone marrow serum and 10 l of MB-HIC binding solution (Bruker Dalton-ics) were transferred to a PCR tube, and 5 l of MBs were added and mixed by pipetting up and down five times. After 5 min, the tube was placed in an MB separator (Bruker Daltonics) for 20 s to separate the beads from the supernatant. The supernatant was removed with a pipette; care was taken not to touch the pipette tip to the beads. The tube was then transferred to a PCR tube rack, and 100 l of wash solution were added. The tube was placed in an MB separator and moved back and forth 20 times between the adjacent wells. The tube was allowed to rest for 20 s to allow collection of the beads on the tube wall, and the supernatant was carefully removed using a pipette. This washing step was performed twice. To elute peptides and proteins from the beads, 10 l of 50% acetonitrile (Merck) were added to the tube, which was mixed thoroughly and incubated for 1 min. Thereafter the tube was incubated for 30 s in an MB separator to separate the elution buffer from the beads at the wall of the tube. Finally the eluted solution containing purified peptides and proteins was transferred into a fresh tube. One microliter of eluted sample was diluted in 10 l of ␣-cyano-4-hydroxycinnamic acid (0.3 mg/ml in 2:1 ethanol:acetone; Bruker Daltonics), and 1 l of this diluted sample was spotted onto a 384-spot AnchorChip TM target (Bruker Daltonics). The sample was air-dried for further MALDI protein profiling. All procedures were carried out by the same person who performed the reproducibility study (see below). The HIC-C 8 beads and tube strips came from a single batch, and the same MALDI platform was used to measure the protein in all of the sera.
Protein profiles were acquired with an Ultraflex MALDI-TOF mass spectrometer (Bruker Daltonics) in linear mode with external calibration to the protein calibration standard 1, which contained insulin (m/z 5734.  ). For each target, 30 laser shots at each of seven locations (a total of 210) were accumulated at 16% laser power with 15 matrix blaster shots at each location at 40% laser power. Spectra were collected using peak evaluation values of signal-to-noise ratio ϭ 8, resolution Ͼ50, and no smoothing or base-line subtraction.
Spectral Analysis-To detect and locate spectral peaks, preprocessing steps, including smoothing, base-line correction, peak detection, and peak alignment, were performed using an in-house program. The spectral mass range from 1000 to 20,000 Da was selected for analysis because this range contained the majority of the resolved protein/peptides. The molecular masses from 0 to 1000 Da were eliminated from analysis because the data in this range contain adducts and artifacts of the energy-absorbing molecule and may contain other chemical contaminants. For data smoothing, adjacent data points were averaged to reduce noise. The noise level was relatively high at low mass range, and it decreased as the mass value increased. Therefore, the number of data points comprising each average was selected based on criteria that do not affect the peak intensity; 10 points were used at the initial mass (ϳ1000 Da), and the number of points decreased linearly as mass increased; five points were used at the terminal mass (ϳ20,000 Da).
Chemical noise is one of the major obstacles in obtaining useful MALDI-MS and MS/MS spectra from small amounts of peptides. This noise is a result of the desorption and ionization of peptides, matrix, and impurities in the sample (4). Base-line correction is performed on a spectrum in isolation to eliminate any base-line signal that can be caused mostly by chemical noise. The optimum base-line curve was made by selecting the minimum level within a selected interval of the spectrum as the base line; as the range of mass values in the intervals increased, the following intervals for selecting a base line were increased serially by 1.2-fold. The initial interval began at m/z 120.
Peaks and valleys of the spectra were detected by applying a slope-based method at 50-data point intervals. Of the detected peaks, those with areas less than the selected threshold value were considered to be noise peaks and were eliminated. Peak areas were calculated by detecting the valleys on both sides of the peak followed by integrating the signal intensity according to a straight line connecting the valleys. Threshold values were applied proportionately with the mass range in one MS spectrum because the peak areas in the high mass range are much larger than those in the low mass range.
Optimal threshold values were determined by testing a small set of spectra using different settings. The threshold value of peak area for the initial mass range was set at 1; it increased linearly to 20 at the final mass range. The number of peaks detected per spectrum using the above process was ϳ110. The intensities of the selected peaks were obtained from the raw spectra to quantify the data.
All of the labeled peaks identified in this process were then aligned using an algorithm that grouped peaks with similar molecular weights across all the spectra into a peak cluster while allowing for slight variations in mass. Each cluster was then taken to represent a particular protein. The masses of peaks within each cluster differed by Յ5 Da, and the entire cluster was considered to represent the same peak. Missing peaks in each peak cluster were filled with the corresponding spectral value.
We used the intensity of the peaks to measure the amount of protein in the sera. Reproducibility can be defined in both spatial (i.e. alignment) and quantitative (i.e. intensity) terms. It is worth noting that the coefficient of variance (CV) for the peak position data using the Ultraflex MALDI-TOF mass spectrometer and similar methods was within 0.05% (Bruker Daltonics). The quantitative reproducibility of the MB-based MALDI spectra (i.e. the variation in peak intensity between measurements) was determined using bone marrow serum from a healthy donor. The reproducibility study was performed by a single person. Moreover the HIC-C 8 beads used were from one batch as were the tube strips, and the same MALDI platform was used to spot the eluted samples. Fifteen protein peaks with the highest amplitude in the range of 1000 -20,000 Da, observed on the 96 spectra over the 3-day course of the analysis, were used to calculate the CV. The CV values for peak intensity in the crude data were between 14.33 and 44.12% and averaged 25.19% (data not shown).
Only the peak or peaks for which the spectral intensities indicated a relatively large difference between the two groups were selected by univariate and multivariate statistical analyses (see "Statistical Analysis") as a classifier or biomarker. The peaks for which the spectral intensities indicated a relatively small difference between the two groups were not chosen because the peak intensities of the two groups overlapped due to the similar abundance distribution or the relatively high CV.
Serum Protein Fractionation and Purification-High abundance proteins such as albumin and IgG were removed from 1 ml of a bone marrow serum (30 l ϫ 34) using a multiple affinity removal system (MARS; Agilent Technologies, San Diego, CA) according to the manufacturer's protocol. The flow-through fraction was concentrated in a 5000 molecular weight cutoff concentrator and then fractionated by elution from a copper-MB preparation with an MB-IMAC copper kit (Bruker Daltonics). The concentrated copper-purified fractions of protein were then loaded onto a C 8 column of inner diameter 6 mm, length 150 mm, and particle size 5 m (Shiseido, Tokyo, Japan). The component proteins/peptides were eluted at a flow rate of 0.75 ml/ min with 20% acetonitrile for 2 min followed by a gradient of 20 -60% acetonitrile over 60 min. Every C 8 fraction was analyzed on an An-chorChip target using an Ultraflex MALDI-TOF mass spectrometer (Bruker Daltonics), and those fractions confirmed by MALDI-TOF-MS to contain a significant majority of the 7.8-kDa peak were pooled. The pool was then split into two identical samples, which were dried by centrifugal evaporation.

Nano-LC/ESI-MS/MS and MS Analysis for 7.8-kDa Protein Identi-
fication-One sample from the duplicate pair was dissolved in digestion buffer containing 100 mM Tris (pH 8.0), 5 mM DTT, 5 mM CaCl 2 , and 1 g of trypsin and incubated at 37°C for 12-16 h. The resulting peptides were lyophilized and dissolved in 0.1% formic acid for LC-MS/MS. MS/MS experiments for peptide identification were performed using a nano-LC-MS system consisting of an Ultimate HPLC system (Dionex) and a Q-TOF mass spectrometer (Waters) equipped with a nano-ESI source. An autosampler was used to load 10-l aliquots of the peptide solutions onto a C 18 trap column of inner diameter 300 m, length 5 mm, and particle size 5 m (Dionex). The peptides were desalted and concentrated on the column at a flow rate of 20 l/min. Then the trapped peptides were back-flushed and separated on a 100-mm homemade microcapillary column composed of C 18 (Aqua; particle size, 5 m) packed into 75-m silica tubing with an orifice inner diameter of 8 m.
The mobile phases, A and B, were composed of 0 and 80% acetonitrile, respectively, each containing 0.1% formic acid. The gradient began with 5% B for 15 min; ramped to 20% B over 3 min, to 60% over 45 min, and to 95% over 2 min; and remained at 95% B for another 7 min. The column was equilibrated with 5% B for 10 min before the next run. The voltage applied to produce an electrospray was 2.5 kV, and the cone voltage was 30 eV. Argon was introduced as a collision gas at a pressure of 10 p.s.i. The three most abundant MS ions were selected by data-dependent peak selection, and the collision energy was stepped up to 30 eV.
For each sample, the MS/MS spectra were processed using Mass-Lynx version 3.5 (Waters); they were smoothed one time using the Savitzky-Golay method set at Ϯ3 channels and centered using the top half of each peak. Peak lists were generated using MassLynx version 3.5 and automatically combined into a single .pkl file.
To identify the peptides, MASCOT (version 2.0, Matrix Science, London, UK), operated on a local server, was used to search the human sequences within the International Protein Index (IPI) database (version 3.32). The human database contains 67,524 protein sequences. MASCOT was used with monoisotopic mass selected, a precursor mass error of 1.5 Da, and a fragment ion mass error of 0.8 Da. Trypsin was selected as the enzyme with one potential missed cleavage. ESI-Q-TOF was selected as the instrument type, and oxidized methionine was chosen as variable modification. Only proteins that were identified by one or more high scoring peptides were considered to be true matches. The high scoring peptides corresponded to peptides that were above the threshold in our MASCOT search (expected, Ͻ0.05; peptide score, Ͼ38). For peptides that matched with multiple members of a protein family, we selected the protein whose predicted molecular weight was closest to the actual molecular weight measured by MALDI-MS or ESI-MS.
The exact molecular weight of the second sample from the duplicate pair was determined to unambiguously identify the proteins. This second sample was dissolved in water containing 0.1% formic acid without additional preparation. ESI-Q-TOF-MS analysis was performed using this protein sample under the same LC-MS conditions, and the MS spectra were processed using the parameters described above without data-dependent MS/MS acquisition. The MS spectrum was deconvoluted using MassLynx version 3.5.
Magnetic Bead-based MALDI Immunoassay-Protein G-immobilized MBs (MB-IAC Pro G kit, Bruker Daltonics) can bind the Fc regions of various antibodies. An aliquot (7.2 g) of purified mouse monoclonal antibody specific for the human platelet factor-4 (PF4) protein (R&D Systems, Minneapolis, MN) in 30 l of incubation buffer (Bruker Daltonics) was applied to the preactivated protein G beads in a thin walled PCR tube and incubated for 1 h at room temperature. The beads were washed twice with 100 l of incubation buffer by placing the tube in an MB separator and moving it back and forth 20 times between adjacent wells. The beads were incubated for 2 h on a shaker with 5 l of serum diluted with 10 l of binding buffer (Bruker Daltonics), washed two times as above with 100 l of binding buffer, and finally washed two times with 100 l of wash solution (Bruker Daltonics).
The protein antigen was eluted by adding 10 l of elution solution (Bruker Daltonics) to the tube. After thorough mixing, the eluted protein was transferred to a fresh tube. A 1-l aliquot of the eluted protein was mixed with 10 l of a matrix solution composed of 0.3 g/liter ␣-cyano-4-hydroxycinnamic acid in 2:1 ethanol:acetone (Bruker Daltonics). One microliter of the diluted sample was spotted onto a 600-m-diameter AnchorChip target. Protein profiles were acquired using an Ultraflex MALDI-TOF mass spectrometer (Bruker Daltonics) exactly as described above.
ELISA Determination of PF4 Levels-To confirm that the m/z 7764.8 peak (i.e. PF4), which was discovered by MB-based MALDI protein profiling, is a determinant or biomarker capable of grouping the sera of AML patients, we directly measured the PF4 levels in the serum samples using conventional ELISA with anti-PF4 antibody. Serum levels of PF4 were determined using a sandwich immunoassay according to the manufacturer's recommendations (R&D Systems) with all steps carried out at room temperature. Briefly each well of a 96-well microtiter plate was coated with 100 l of mouse monoclonal antibody specific for human PF4 (2.0 g/ml in PBS) and incubated overnight. After three washes with 300 l of wash buffer (0.05% Tween 20 in PBS, pH 7.4), the well was blocked with 300 l of reagent diluent (1% BSA in PBS, pH 7.4) for 1 h. Each serum sample was diluted 1:20,000 with reagent diluent, 100 l of diluted sample were added to each well, and the plate was incubated for 2 h. The wells were washed as described above with wash buffer and then incubated with 100 l of a biotinylated polyclonal goat antibody specific for human PF4 (100 ng/ml in reagent diluent) for 2 h. The plates were washed again as described above, incubated with 100 l of horseradish peroxidase-conjugated streptavidin (diluted 1:200 with reagent diluent) for 20 min, and washed three times.
For detection, 100 l of substrate solution (1:1 H 2 O 2 :tetramethylbenzidine) were added to each well, and the plate was incubated for 20 min. The color reaction was stopped by addition of 50 l of stop solution (2 N H 2 SO 4 ) to each well, and the plates were read in a VERSA max microplate reader (Molecular Devices, Sunnyvale, CA) at 450 nm. Data analysis was performed using SoftMax Pro version 5 software (Molecular Devices). All of the measurements were collected twice, and the detection values (i.e. the PF4 concentrations) are given as the mean.
Statistical Analysis-All of the statistical computations were performed using statistical analysis system software release 8.2, and all p values reported are two-sided. The Mann-Whitney U test was used to compare the change of MALDI protein peak intensities or ⌬P in sera of AML patients (n ϭ 31) who had achieved CR or RD after chemotherapy (␣ ϭ 0.05). For a total of 110 ⌬P values per spectrum, ⌬P values were differences in serum protein peak intensities calculated from the intensities after chemotherapy (P CR or P RD ) and at diagnosis (P PreCR or P PreRD ) (i.e. ⌬P CR ϭ P CR Ϫ P PreCR and ⌬P RD ϭ P RD Ϫ P PreCR ).
The major statistical end point in the ⌬P analysis was the determination of factors associated with the diagnosis of achieving CR. Factors associated with this outcome were selected based on multivariate logistic regression modeling (5). For the multivariate analysis, a forward selection procedure with the predictor variables left continuous was used to obtain a final model (␣ ϭ 0.10).
After determining PF4 levels in the individual CR, RD, PreCR, and PreRD samples using ELISA, we performed two-sample comparisons between the PreCR and CR samples and between the PreRD and RD samples. A nonparametric alternative to the paired t test, the Wilcoxon signed ranks test, was used to test the null hypotheses of no median differences between paired observations, PreRD and RD or PreCR and CR.
Linear regression analysis was used to investigate the relationship between platelet count (PLT) and PF4 levels. A graphical residual plot was used to examine the adequacy of both the linear regression model and the homoscedasticity (data not shown). Due to the nonconstancy of the error variance, logarithmic transformations of the two variables were used for a better fit. (Table I) was assayed with C 8 MBs. Of these samples, 26 (designated CR) were from AML patients who had achieved the CR state as a result of 1-month remission induction chemotherapy with idarubicin/BH-AC or ATRA. Another 26 serum samples (designated PreCR) were from the same patients but had been taken previously at the time of diagnosis (1 month prior to the chemotherapy); five serum samples (designated RD) were from AML patients who had not achieved the CR state after remission induction chemotherapy, exhibiting instead the RD state; and the remaining five samples (designated PreRD) were from the RD patients at diagnosis (supplemental Fig. 1).

Magnetic Bead-based MALDI Analysis of Serum Proteins-A panel of 62 bone marrow serum samples
We obtained MS spectra of these 62 samples using the AnchorChip MALDI technique for detection of serum proteins/ peptides with ␣-cyano-4-hydroxycinnamic acid as the matrix. We used an in-house algorithm to detect and align protein peaks. Up to 110 aligned peaks (or peak clusters) per spectrum were detected between M r 1000 and 20,000.
Using a significance level of 5%, the Mann-Whitney U test selected 10 ⌬P values (i.e. the MALDI protein peak intensity difference: after-chemo intensity minus before-chemo intensity) variables (⌬3882.6/p ϭ 0.0056, ⌬4208.8/p ϭ 0.0168, ⌬4458.9/p ϭ 0.0467, ⌬4646.7/p ϭ 0.0099, ⌬4708.1/p ϭ 0.0386, ⌬6631.9/p ϭ 0.0362, ⌬7764.8/p ϭ 0.0014, ⌬8814.9/ p ϭ 0.0440, ⌬8921.0/p ϭ 0.0499, and ⌬9290.6/p ϭ 0.0473) displaying the highly significant difference in its distribution in the CR group compared with the RD group. Then multivariate logistic regression analysis with the 10 selected variables revealed that the only significant predictor was ⌬7764.8 (Table II). In the multivariate analysis, the odds ratio (OR) computed with the predictor variable being left continuous for the risk of being a CR responder was 11.688 (95% confidence interval, 0.773-176.775). In fact, the OR 11.688 was computed for an increase of 1 unit of the ⌬7764.8 level. Because this may not be easily interpretable, we again computed the OR for a change from the 10th percentile (ϭϪ0.2) to the 60th percentile (ϭ 5.1) of the predictor variable as a way to compare subjects in the middle of the 20th percentile of the distribution (considering the smaller RD group; n ϭ 5) with those in the middle of the remaining 80th percentile of the distribution (considering the larger CR group; n ϭ 26). Thus, the computed OR for this change was 456,151 (Table II). The odds of a subject being a CR was 456,151 times higher with a value of 5.1 than with a value of Ϫ0.2. (i.e. with an increase of ⌬7764.8 from Ϫ0.2 to 5.1, the odds of a subject being a CR increased by 456,151 times).
The composite (or averaged) MALDI spectra of the four sample groups comprising sera from PreCR, PreRD, CR, and RD patients are shown in Fig. 1, A and B, and individual contour plot views of the peak at m/z 7764.8 are shown in Fig.  1C. The protein represented by this peak was overexpressed in the CR sera but did not change in the other groups.
Identification of PF4 by Fractionation, MS, and MS/MS Analysis-To identify the protein (or peptide) giving rise to the m/z 7764.8 peak, proteins corresponding in size were partially purified from serum by MARS depletion, copper-bead preparation, and C 8 reverse-phase chromatography (supplemental Fig. 2). After the partially purified protein was digested with trypsin, the tryptic digest was subjected to nano-LC-ESI/MS/ MS, and the resulting peptide peaks were searched against human protein databases using MASCOT. This search iden-

FIG. 1. Composite MALDI spectra of the four sample groups (A and B) and their individual gel views (C) of a peak at m/z 7764.8.
PreCR and PreRD depict the states at diagnosis of patients who achieved a CR or who exhibited RD, respectively. A, composite (averaged) MALDI spectra of PreCR, PreRD, CR, and RD sera. MALDI-MS resolved 110 peaks between m/z 1000 and 10,000. B, composite (averaged) MALDI spectra of PreCR, PreRD, CR, and RD sera between m/z 7500 and 8500. Compared with the PreCR, PreRD, and RD samples, the CR samples overexpressed a protein at m/z 7764.8. C, contour plot of individual spectra between 7500 and 8500. The high intensity band at m/z 7764.8 was found only in CR samples. tified three proteins: isoform 1 of complement factor H precursor (IPI00029739), prothrombin precursor fragment (IPI00019568), and platelet factor-4 precursor (IPI00022446). Each contains at least one specific peptide with a significant hit score (individual ions scored, Ͼ38; see supplemental Table  I).
Among the three proteins, the predicted molecular weight of PF4, except for the signal peptide from platelet factor-4 precursor, was very similar to the value found by MALDI-MS. PF4 was identified from the tryptic peptide ICLDLQAPLYK with a MASCOT search score of 55 and 16% sequence coverage. The MS and MS/MS spectra of the peptide are shown in Fig. 2.
The molecular masses of isoform 1 of complement factor H precursor and prothrombin precursor fragment are 140,000 and 70,000 Da, respectively. In the case of these proteins, partially degraded specific fragments of the original proteins may have been present in the sample fraction. However, considering that the peptides identified by the MS/MS spectra were spread out in terms of sequence and that both proteins are major human plasma proteins, they may be the contaminants introduced during the fractionation steps.
Although PF4 was matched by the tryptic peptide (ICLD-LQAPLYK) with a high search score, the identification was based on a single peptide assignment. To unambiguously identify the protein, we performed LC-ESI/Q-TOF-MS analysis to determine its exact molecular weight. The average molecular mass of the protein was 7764.98 Ϯ 0.13 Da (Fig.  2B), which is about 4 Da lower than the theoretical value (7769.21 Da, estimated from the sequence of PF4). When we consider the two disulfide bonds (at Cys-10 and Cys-36 for one and at Cys-12 and Cys-52 for the other) (ca.expasy.org/ uniprot/P02776) in human PF4, its average molecular mass becomes 7765.18 Da. This is consistent with the molecular weight determined for the intact protein by ESI-MS and MALDI-MS (the m/z 7764.8 ion; supplemental Table II). These results confirm the identity of human PF4.
In this molecular weight data analysis of intact protein, we additionally found a few peaks with small intensities at m/z 7836.63, 7923.00, and 8141.13 (Fig. 2B, panel B). Struyf et al. (7) reported several N-terminally processed isoforms of a platelet factor 4 variant (PF4V, UniProt accession number P10720) and PF4 in thrombin-stimulated human platelets. Based on the report by Struyf et al. (7) and our molecular weight data on intact proteins, the small peaks are thought to be N-terminally processed isoforms of PF4 (supplemental Table III). Interestingly their appearance pattern between tested sample groups is almost identical to that of PF4 (Fig.  1B). All the N-terminally modified peaks including PF4 show the relatively increased signal only in the CR group. However, except for PF4, the intensity change between groups was not very evident. This finding regarding the PF4 isoforms more strongly supports the argument that the 7.8-kDa protein is PF4. MALDI immunoassay with a specific anti-PF4 antibody to analyze four samples; two of the samples were CR sera containing the 7764.8-Da peak, one sample was a RD serum that yielded no 7764.8-Da peak upon C 8 chromatography/ MB, and the fourth sample was a control containing only 1% BSA blocking agent (Fig. 3). Both 7764.8-Da peak-containing CR samples yielded a specific peak of mass 7.8 kDa upon anti-PF4 antibody coated MB (positive control). This peak was detected neither in the RD sample nor in the BSA control sample (negative control).

Confirmation of the
ELISA-based Quantification of PF4 in Serum Specimens from AML Patients-Additional evidence for differential expression of the 7764.8-Da peak (i.e. PF4) in the leukemic serum specimens from CR and RD patients before and after treatment was obtained using ELISA. The samples for this assay included the same 31 serum pairs used in the MBbased MALDI profiling analysis. The PF4 levels in the individual CR, RD, PreCR, and PreRD samples were determined by ELISA (Table III), and the differences in PF4 levels at diagnosis and posttreatment for each individual in the CR and RD sample groups (⌬PF CR and ⌬PF RD , respectively) were compared using statistical analysis.
Four box-and-whisker plots graphically depicting the results of these comparisons are shown in Fig. 4. The Shapiro-Wilk test revealed a normal distribution (p Ͼ 0.05) of ⌬RD and ⌬CR for PF4 levels. The null hypothesis for this test is that the data are normally distributed. Unfortunately normality tests have little power to detect whether a sample comes from a Gaussian population when the sample size is small (Ͻ12). A nonparametric alternative to the paired t test, the Wilcoxon signed ranks test, was used to test the null hypotheses of no median differences between paired observations, PreRD and RD or PreCR and CR.
The median ⌬PF RD value (PF RD Ϫ PF PreRD ) was 0.07 (IQR of ⌬PF RD ϭ 0.89) with a p value of 0.89, confirming that the null hypothesis of no median difference between PreRD and RD PF4 levels could not be rejected at the 5% level. However, the median ⌬PF CR value (PF CR Ϫ PF PreCR ) was 8.16 (IQR of ⌬CR ϭ 6.49) with a p value of 0.000, indicating that the null hypothesis of no median difference between CR and PreCR PF levels was rejected. These results show that the difference between CR and PreCR PF levels, but not the difference between RD and PreRD PF levels, was statistically significant.
Correlation between PF4 Levels and Platelet Counts in AML Patient Sera-We next sought to determine whether PF4 levels had any statistical relationship with platelet counts (Table  III). To achieve this goal, we compared PF4 levels and platelet counts in the CR and RD samples (Fig. 5), which exhibit a relatively wide range of serum levels, as compared with the PreCR and PreRD samples. Logarithmic transformation of each variable yielded the following linear regression model. This model is indicative of a positive linear correlation between lnPF4 and lnPLT (p ϭ 0.000). Furthermore the value of R 2 , the coefficient of determination, was 0.861, indicating that ϳ86% of the observed variability in the response (lnPF4) was modeled by lnPLT. From the linear correlation curve, we found that the PF4 level corresponding to a platelet count of 100 ϫ 10 3 /mm 3  AML describes a heterogeneous group of clonal hematopoietic progenitor cell disorders with a spectrum of morphologic, immunophenotypic, cytogenetic, and molecular characteristics (3). Although diagnosis may be possible based solely on a peripheral blood examination, a bone marrow aspiration is strongly recommended. Bone marrow trephine biopsy is not routinely indicated, although it may be necessary if the aspirate is dilute, hypocellular, or inaspirable. The  CR1 and CR2). These samples had yielded a ϳ7764.8-Da peak upon AnchorChip MALDI analysis with C 8 beads. A peak corresponding to PF4 was not present in the serum (RD) of a patient whose disease was resistant to the remission induction chemotherapy or when the anti-PF4 antibody was incubated with only 1% BSA (No serum). Intens., intensity; a.u., arbitrary units.
French-American-British (FAB) classification for diagnosis in AML was based initially on morphology, cellularity, blast percentage, and cytochemistry (8). A subsequent update and modification of the FAB diagnostic criteria by the World Health Organization (WHO) (9) instituted a major change in the diagnostic threshold for the AML blast percentage in the peripheral blood or bone marrow aspirate. The WHO recommendations decreased this threshold to 20% from the 30% level recommended by the FAB. Thereafter new recommendations of the International Working Group for diagnosis and standardization criteria in AML were published in 1990 and recently revised (3). This group of investigators used the cytochemical and phenotypic criteria of the WHO (9, 10) and emphasized the prognostic relevance of bone marrow cytogenetics (11)(12)(13) and molecular genetics (14,15). They also recommended new criteria for variable responses to the currently available therapy in AML.
According to the International Working Group criteria, a morphologic CR designation requires that the patient survive for at least 7 days after the final dose of the initial course of treatment and achieve a morphologic leukemia-free state (Յ5% blasts in an aspirate sample containing marrow spicules and Ն200 nucleated cells) and have an absolute neutrophil count of Ͼ1.000/l and a platelet count of Ն100,000/ l. The criteria for treatment failure due to RD includes  4. Box plots of the PF4 protein serum level (g/ml) in sera from patients in various disease states. PreCR and PreRD are samples taken prior to treatment (at the time of AML diagnosis) from patients who had achieved CR or who had RD, respectively.
FIG. 5. Statistical relationship between the logarithms of platelet count (ln_PLT; 10 3 /mm 3 ) and PF4 serum level (ln_PF4; g/ml). We compared platelet counts and PF4 levels in the CR and RD samples, which exhibit a relatively wide range of serum levels compared with the PreCR and PreRD samples. The coefficient of determination was R 2 ϭ 0.861. For a platelet count of 100 ϫ 10 3 /mm 3 , the amount of PF4 derived from the curve of the positive linear correlation was 2.492 g/ml.
appropriately treated patients who survive for at least 7 days after the final dose of the initial course of treatment and whose last posttreatment peripheral blood smear and/or bone marrow sample showed persistent AML. It includes those patients for whom treatment has failed to achieve CR in a phase III trial or failed to achieve at least a partial remission (5-25% blasts, neutrophils Ͼ1.000/l, and platelets Ն100,000/l) in a phase I or II trial. In the present study, CR samples included cases from patients who fulfilled the major requirements of minimal blast count (i.e. Ͻ5% blasts in an aspirate sample) and blood count recovery (i.e. neutrophils of Ն1.000/l and platelets of Ն100,000/l). However, the RD samples included cases from patients who had succeeded in achieving neither minimal blast count nor blood count recovery (i.e. Ն5% blasts in an aspirate sample, neutrophils of Ͻ1.000/l, and platelets of Ͻ100,000/l).
Thus, to analyze the MALDI-TOF protein profiles from the CR and RD groups of serum samples we performed a multiple logistic regression analysis. We found that for the 7764.8-Da protein, the odds of a subject being a CR responder are 456,151 times higher for a change of 5.1 than for a change of Ϫ0.2. Moreover we found that only one variable (i.e. the 7764.8-Da protein) was significantly different between the two groups. Given that the patients have been subjected to chemotherapy in between the sampling times, one would expect many real changes or biomarker candidates. As mentioned under "Spectral Analysis," although the spatial (i.e. alignment) reproducibility of MALDI-TOF protein peaks was excellent (CV values within 0.05%), the quantitative (i.e. intensity) reproducibility of the MALDI-TOF protein peaks was not very good (CV values between 14.33 and 44.12%). In actuality, almost all commercially available MALDI-TOF instruments to date are known to have a similar reproducibility. On the other hand, serum proteins being expressed differentially between groups may be biomarker candidates only if the protein profiling platform can detect the minimal change of protein amount. However, the peaks for which the spectral intensities between the two groups overlapped due to the high CVs, even though their real values are separated statistically, will not be chosen as being significant. Consequently considering the relatively high CV of MALDI protein peak intensities, any protein peaks for which the difference in peak intensity exists but was not so large between the two groups were not selected by the present logistic analysis. In addition, in this study, we enrolled 26 CR but only five RD cases for 21 months (from July 2003 to March 2005) due to the very low incidence of AML. We think that it is necessary to perform another validation step in a new cohort of patients especially with a large sample size of RD patients.
The 7764.8-Da protein was purified by several steps of HPLC/bead separation and identified using nano-LC-MS/MS and MB-based MALDI immunoassay. This protein proved to be a monomer of tetrameric PF4. To investigate the origin of the difference in the calculated values of ⌬PF CR and ⌬PF RD , we determined the PF4 levels in the individual CR, RD, PreCR, and PreRD samples using ELISA. The results of two-sample comparisons between the PreCR and CR samples and between the PreRD and RD samples showed that PF4 was present only at a minimum level in both the PreRD and RD samples, whereas the high ⌬PF CR value originated primarily from the relatively high PF4 level in the CR sample as compared with the same minimum PF4 level found in the PreCR sample.
PF4, also known as CXCL4 (chemokine (CXC motif) ligand 4), was originally described as a platelet-derived heparinneutralizing factor (16). PF4 is synthesized entirely in megakaryocytes and, to a lesser extent, in mature platelets, and is stored in platelet ␣-granules. It accounts for ϳ25% of the protein in the platelet granule. When platelets are activated, PF4 is released along with other protein components of the platelet ␣-granules. PF4 levels in plasma are 2-10 ng/ml, whereas they are 5-10 g/ml in serum prepared by provoking thrombin-mediated platelet aggregation and coagulation (16). The human PF4 monomer is a 7800-Da protein of 70 amino acids (17). NMR studies have shown that PF4 in solution exists in a concentration-dependent equilibrium of monomers, dimers, and tetramers (18).
Unlike other CXC chemokines, such as IL-8, which typically mediate chemotaxis of neutrophils, PF4 is devoid of chemotactic activity for neutrophils (19). It has recently been proposed that some vascular beds in humans express an alternatively spliced form of CXCR3, termed CXCR3B, that binds PF4 (20). PF4 binds with high affinity to heparin and other anionic glycosaminoglycans (21,22), and the PF4 tetramer has an equatorial band of cationic charges that attracts heparin molecules (23). Upon its release from activated platelets, PF4 binds to proteoglycans on the surfaces of endothelial cells (24), platelet membranes (25), and hepatocytes (26). It has been implicated in diverse biological processes; its suggested roles include those of a procoagulant of platelet function and plasmatic coagulation (16), an inhibitor of hematopoiesis (27,28) and angiogenesis (29,30), a promoter of neutrophil adhesion and degranulation (31), an enhancer of oxidized low density lipoprotein binding to the low density lipoprotein receptor (32), and a stimulator of anticoagulantactivated protein C generation (33). These effects have been observed only at high PF4 concentrations (5-25 g/ml), which can be achieved at the site of intense platelet activation (6).
Statistical comparison of the PF4 levels and platelet counts in sera from the 31 AML patients (CR and RD) revealed a significant positive correlation between the PF4 level and the platelet count. The PF4 level corresponding to a platelet count of 100,000/mm 3 , as derived from the positive linear correlation curve, was 2.492 g/ml. Therefore, one of the three requirements for the designation of complete remission, a blood platelet count recovery of Ͼ100,000/mm 3 , might be equivalent to a serum PF4 level recovery of Ͼ2.492 g/ml. The morphology of platelets and other blood cells frequently changes after chemotherapy, and thus even a blood cell counter would not reflect the correct measurement of blood cell counts. In this situation, the serum PF4 assay might be a good alternative for assessing blood cell recovery. In addition, the serum PF4 level can be determined simply by a conventional ELISA method. Considering that a good alternative for assessing platelet count does not exist, serum PF4 measurement seems to provide a cost-effective means of yielding information on the status of hematopoietic recovery in an AML patient after chemotherapy.
In conclusion, we have demonstrated, using an MB-based MALDI profiling technology, that serum PF4 levels in AML patients return to or are elevated above the normal level in patients achieving CR, whereas in RD patients, levels were unchanged or decreased below the normal level. In fact, PF4 was shown to be a good indicator for the recovery of blood count in the complete remission of AML.