We assessed the autoantibody repertoire of a mouse model engineered to develop breast cancer and the repertoire of autoantibodies in human plasmas collected at a preclinical time point and at the time of clinical diagnosis of breast cancer. In seeking to identify common pathways, networks, and protein families associated with the humoral response, we elucidated the dynamic nature of tumor antigens and autoantibody interactions. Lysate proteins from an immortalized cell line from a MMTV-neu mouse model and from MCF7 human breast cancers were spotted onto nitrocellulose microarrays and hybridized with mouse and human plasma samples, respectively. Immunoglobulin-based plasma immunoreactivity against glycolysis and spliceosome proteins was a predominant feature observed both in tumor-bearing mice and in prediagnostic human samples. Interestingly, autoantibody reactivity was more pronounced further away than closer to diagnosis. We provide evidence for dynamic changes in autoantibody reactivity with tumor development and progression that may depend, in part, on the extent of antigen–antibody interactions. Cancer Res; 73(5); 1502–13. ©2012 AACR.
The dynamic interactions between circulating tumor antigens and autoantibodies during breast cancer development and progression have not been well characterized. Most studies of autoantibodies in cancer have relied on samples obtained after clinical diagnosis. Analysis of preclinical samples provides an opportunity to evaluate changes in the humoral response with tumor development and progression.
Several studies have yielded circulating autoantibodies against specific antigens at the time of diagnosis (1–3). High-throughput screening of proteins for autoantibody response is facilitated by the use of microarray technology. Discovery of novel autoantigens has been made through arrays composed of recombinant proteins (4–6), tumor homogenates (7), and phage-display libraries (8, 9). Peptide arrays have also been used to identify cancer-associated autoantibodies (10). Arrays of lysate-derived proteins allow delineation of immunogenic signatures involving natural proteins that may be subject to posttranslational modification, as previously applied to the studies of lung (11–13), colon (14, 15), prostate (16), and pancreatic (17) cancers. We recently showed that global profiling of the plasma proteome allows identification of sets of proteins progressively released into circulation at a preclinical stage of breast tumor development that may be grouped into biologic pathways (18). There is similarly a need to delineate sets of proteins associated with biologic pathways and networks that elicit autoantibodies with breast tumor development. To this effect, we assessed the autoantibody repertoire of a mouse model engineered to develop breast cancer and of both preclinical human samples from a longitudinal cohort and clinical human samples obtained at the time of clinical diagnosis of breast cancer.
Materials and Methods
Mouse model samples.
Plasma samples from MMTV-neu mouse model were serially collected at the University of Washington Tumor Vaccine Group, Specific Pathogen-Free Facility, and Institutional Animal Care and Use Committee protocol #2878–01, from a baseline of 8 weeks until the animals were euthanized because of excessive tumor volume. Baseline samples and 2 blood collections just before the palpable tumors were used for 23 tumor-bearing mice. Samples were collected retro-orbitally at approximately 100–200 μL of whole blood. Analyzed blood samples were collected on average 121 days and 144 days after baseline sample.
Prediagnostic EDTA plasma samples were collected as part of the Women's Health Initiative (WHI) observational study (Table 1). Autoantibody analysis was conducted using plasmas from 48 postmenopausal women having no history of hormone therapy use who were later diagnosed with estrogen receptor–positive (ER+)/progesterone receptor positive (PR+) breast cancer and 65 healthy controls with similar distributions of age, time of blood collection (±6 months), and hormone therapy use. Newly diagnosed plasma samples from 61 postmenopausal women diagnosed with stage I/II ER+/PR+ breast cancer and 61 matched healthy controls were also investigated (Table 1). Assays of pyruvate kinase isozyme M1/M2 (PKM2) were conducted on plasma samples from an additional 118 postmenopausal WHI participants who were later diagnosed with ER+ breast cancer and 118 healthy controls matched on age and ethnicity. These samples were not matched on hormone therapy usage.
Protein fractionation and array construction
One hundred and fifty milligrams of protein derived from MMTV-neu and MCF7 cell lysates were each subjected to orthogonal two-dimensional high-performance liquid chromatography (2D-HPLC) fractionation in an automated system (Shimadzu Corporation; Fig. 1A; ref. 19). An excess of protein from each cell line was fractioned to ensure adequate protein content in arrayed spots and availability of protein fractions for further investigation and validation. Fractionation was based on anion exchange (SAX/10 column, 7.5 mm ID×150 mm, Column Technology Inc.) using a 40 step-elution, followed by a second dimension reversed-phase separation (RP/5D column, 4.6 mm ID×150 mm, Column Technology Inc.). A total of 2,430 fractions were collected from the 2D separation. Fr_X_Y denotes the Yth fraction from the reverse-phase liquid chromatography of the Xth fraction from the anion exchange separation. The first dimension anion-exchange chromatography mobile-phase A was 20 mmol/L Tris, pH 8.5, and mobile-phase B was 20 mmol/L Tris, 1 mol/L NaCl, and pH 8.5. The second dimension reversed-phase chromatography mobile-phase A was 95% water, 5% acetonitrile, and 0.1% trifluoroacetic acid (TFA), and mobile-phase B was 90% acetonitrile, 10% water, and 0.1% TFA.
Three hundred microliters of each fraction was lyophilized and resuspended in 30 μL of printing buffer (250 mmol/L of Tris-HCl, pH 6.8, 0.5% SDS, 25% glycerol, 0.05% TritonX-100, and 62.5 mmol/L of dithiothreitol). A total of 1,950 MCF7 fractions or 2,808 MMTV-neu fractions, together with printing buffer as negative controls and purified human or mouse immunoglobulin G (IgG) as positive controls were printed onto nitrocellulose-coated slides using a contact printer as previously described (11, 20). Approximately 500 fractions were excluded from arraying due to low UV absorbance observed during fractionation. Plasma samples were hybridized with an individual microarray at a dilution of 1:150. Reactivity was quantified using an indirect immunofluorescence protocol, as previously described (13). Local background-subtracted median spot intensities were generated using GenePix Pro 6.1 and used for downstream statistical analysis using R 2.9.0. Spot intensities were log (base 2) transformed before statistical analysis. P-values were calculated using a student's t-test.
Western blot analysis
One hundred microliters of individual fractions was lyophilized and resuspended in 40 μL of loading buffer. Fractions were run in separate lanes of a 4% to 12% Bis-Tris Criterion XT Precast Gel. Gels were transferred to polyvinylidene difluoride membranes for 1.5 hours at 80 V. Membranes were blocked in 3% bovine serum albumin (BSA) at room temperature for 1 hour. Plasma samples were diluted 1:500 in 3% BSA and incubated with the membrane at 4°C overnight. Samples were removed and membranes were washed with 0.1% PBS with tween (PBST) 5 times for 5 minutes each. Horseradish peroxidase–labeled anti-mouse or anti-human IgG at a 1:2,000 dilution was incubated with the membrane at room temperature for 1 hour. Solutions were removed and membranes were washed with 0.1% PBST 5 times for 5 minutes each. Membranes were exposed to enhanced chemiluminescence (ECL) for 1 minute and exposed to ECL hyperfilm for varied lengths of time. Films were developed and scanned for qualitative analysis.
Mass spectrometry analysis
On the basis of the protein microarray analysis, 50 μL of each interesting fraction from the 2D-HPLC was lyophilized using a freeze-drying system (Labconco). The lyophilized protein samples were dissolved in 100 mmol/L NH4HCO3 (pH 8.5) followed by overnight in-solution digestion with trypsin at 37°C. The digestion was quenched by adding 5 μL of 1.0% formic acid solution before liquid chromatography/tandem mass spectrometry (LC/MS-MS) analysis as described previously (21). Briefly, peptides were separated by reversed-phase chromatography using a nano HPLC system (Eksigent) coupled online with a LTQ-FT mass spectrometer (Thermo Fisher Scientific, Inc.). Mass spectrometer parameters were spray voltage 2.5 kV, capillary temperature 200°C, FT resolution 100,000, FT target value 8 × 105, LTQ target value 104, 1 FT microscan with 850 ms injection time, and 1 LTQ microscan with 100 ms injection time. Mass spectra were acquired in a data-dependent mode with the m/z range of 400 to 2000. The full mass spectrum (MS scan) was acquired by the FT, and tandem mass spectrum (MS/MS scan) was acquired by the LTQ with 35% normalized collision energy. Acquisition of each full mass spectrum was followed by the acquisition of tandem mass spectrometry spectra for the 5 most intense +2 or +3 ions within a one second duty cycle. The minimum signal threshold (counts) for a precursor occurring during a mass spectrometry scan was set at 1,000 for triggering a MS/MS scan.
The acquired LC/MS-MS data were processed by the Computational Proteomics Analysis System (22–24). Briefly, LC/MS-MS data were first converted to mzXML format using ReAdW software (version 1.2) to generate the peak list for protein database searching. The X!Tandem search engine (version 2005.12.01) parameters included cysteine (Cys) alkylated with iodoacetamide (57.02146@C) as a fixed modification and methionine (Met) oxidation (15.99491@M) as a variable modification. Data were searched against the International Protein Index human protein knowledge base (version v3.57), which contained entries for 76,542 proteins. The minimum criterion for peptide matching was a Peptide Prophet Score of 0.2 or more. Peptides meeting this criterion were grouped to protein sequences using the Protein Prophet algorithm at an error rate of 5% or less to maximize protein discovery and identification. Total peptide count in each fraction was used as a measure of protein concentration within that fraction.
Identification of immunogenic proteins
Significantly elevated fractions and neighboring fractions were grouped into “fraction clusters” based on microarray reactivity (Fig. 1B). Clusters were subjected to analysis by Western blot and mass spectrometry analyses to determine the immunogenic proteins within each cluster. Western blot analysis was used to determine the molecular weight of proteins with autoantibody reactivity in plasma samples (Fig. 1C). Blots of fraction clusters were probed with plasma samples seen to be reactive from the microarray analysis. Observed bands that matched microarray reactivity data were counted as positive hits. When multiple reactive bands were observed within a single fraction cluster, multiple proteins were identified as reactive for that cluster. Mass spectrometry analysis was used to identify proteins present in each fraction cluster (Fig. 1D). Total peptide count from individual fractions was matched to microarray reactivity data to determine protein identifications. When no reactive bands were observed in Western blot analyses, protein identification was based solely on results from the mass spectrometry analysis of fractions. While most proteins had more than 1 peptide identified in each analyzed fraction, no minimum peptide count criterion was applied.
Enzyme linked immunosorbent assay
PKM2 (Schebo) measurements were conducted on prediagnostic plasma samples according to the manufacturer's suggested protocol. Absorbance was measured using a SpectraMax Plus 384 and results calculated with SoftMax Pro v4.7.1 (Molecular Devices). Sample optical density (OD) values were log2 transformed and median normalized across plates. Normalized values were further standardized such that the mean of the control samples is 0 and the standard deviation (SD) is equal to 1. P values were computed using a Mann–Whitney–Wilcoxon test.
Autoantibody profiles in neu-transgenic mouse models before the occurrence of palpable tumor
Three plasma samples, consisting of a baseline blood draw and 2 draws before palpable tumor, from 23 individual mice were hybridized in singlet with mouse breast cancer cell lysate protein arrays. Of the approximately 2,800 fractions from the MMTV-neu mouse cell line that were arrayed, 120 fractions displayed significantly elevated IgG reactivity (P < 0.05) with a case-to-control ratio of at least 1.2 in an assay of the initial plasmas collected at an average of 25 weeks of age and before palpable tumor. Analysis of plasmas from a second blood draw before palpable tumor from the same mice also yielded significant reactivity for a subset of 38 of the 120 reactive fractions. A pattern of reactivity among neighboring fractions was observed, consistent with elution of reactive proteins over sequential fractions. Reactivity profiles across reversed-phase fractions were used to select clusters around statistically significant fractions that formed a peak pattern (Fig. 1B). Twenty-seven such fraction clusters were subjected to mass spectrometry and Western blot analysis that yielded identification of 25 reactive proteins (Table 2) from the MMTV-neu model.
Reactivity against anti-IgG controls printed on the arrays indicated that there was no difference in IgG amount between case and control samples. More than 90% of reactive proteins identified were annotated as intracellular with enrichment in the nuclear proteins (Fig. 2A) notably in spliceosome C complex proteins (e.g., Hnrnpa2b1, Sfrs3, and Sfrs7; ref. 25). Identified proteins were subjected to analysis of gene sets represented in the Kyoto Encyclopedia of Genes and Genomes (KEGG) biologic pathway database (26). Four of the 27 identified proteins (Aldoa, Aldoc, Eno1, and Pkm2) were identified in the glycolysis gene set (Fig. 2C) with an estimated false discovery rate (FDR = 0.0037). Interestingly some of the immunogenic proteins identified (Hist1h1d, Eno1, Hnrnpa2b1, Sfrs3, Nme2, and Krt18) were previously associated with autoimmune disease (25, 27–31) indicative of an overlapping set of antigens between autoimmune disease and the humoral response observed in tumor-bearing mice.
Autoantibody signatures in human breast cancer plasmas before clinical diagnosis
Prediagnostic plasmas from 48 women with ER+/PR+ breast cancer and 65 healthy controls, all participants in the WHI cohort study, were individually hybridized with MCF7-derived protein arrays. Of the approximately 1,960 printed fractions, 285 individual fractions yielded a case-to-control ratio of 1.2 or more and P value <0.05 or less using a Student t test. Analysis of a subset of these fraction clusters yielded 90 protein identifications (Table 3). Thirty-five percent of identified proteins were of nuclear origin (Fig. 2B), concordant with the mouse model data.
Analysis of gene sets represented in the human KEGG biologic pathway database revealed significant enrichment of proteins in the glycolysis gene set (FDR = 7.5E-6) and spliceosome gene set (FDR = 0.0011). Nine proteins were identified in the glycolysis gene set (Fig. 2C): ALDH7A1, ALDOA, DLD, ENO1, FBP1, GAPDH, GPI, PKM2, and TPI. Three of these proteins, ALDOA, ENO1, and PKM2 were also identified in plasma from tumor-bearing mice. A set of 9 proteins was associated with the spliceosome by KEGG analysis: EFTUD2, HNRNPA1, HNRNPK, HSPA8, SF3A1, SFRS1, SFRS3, SFRS6, and U2AF1. Additional proteins, namely HNRNPA2B1, PTBP1, RALY, SAP18, and SYNCRIP, not included in the spliceosome signature by KEGG analysis, are known to be part of the spliceosome complex (25, 32, 33). SFRS3 and HNRNPA2B1 overlap with antigenic proteins identified in the mouse. As with the mouse model, some of the identified proteins have been associated with autoimmune diseases. Thirteen proteins (AHNAK, CALR, ENO1, GAPDH, HADH, HIST1H1D, HIST1H1E, HNRNPA1, HNRNPA2B1, HNRNPK, NCL, and SFRS3) were previously described as autoantigens in systemic lupus erythematosus (SLE; refs. 25, 27, 28, 31, 32, 34–39), whereas others (EZR, GPI, and TXN) were associated with rheumatoid arthritis (27, 31, 36) and other autoimmune diseases (40–45).
To speculate on the use as of these biologic signatures as potential biomarker panels, receiver operator characteristic (ROC) analysis was conducted on the basis of the set of proteins in the glycolysis (9 proteins) and spliceosome (14 proteins) signatures using a linear regression model based on a least squares estimation (Supplementary Table S1). The glycolysis and spliceosome signatures gave areas under the ROC curve (AUC) of 0.68 [95% confidence interval (CI): 0.59–0.78] and 0.73 (95% CI: 0.63–0.82), respectively. Combining these 2 signatures yielded an AUC of 0.77 (95% CI: 0.68–0.86), with 35% sensitivity at 95% specificity (Fig. 3A). This combination, while not statistically better than the spliceosome signature alone, shows the additive potential of autoantibody signatures.
Temporal patterns of circulating protein and autoantibody levels preceding a diagnosis of breast cancer
Plasmas from cases were divided on the basis of the time of blood collection in relation to diagnosis of breast cancer. Interestingly, autoantibody reactivity among case plasmas collected further from clinical diagnosis (more than 150 days prior) was more than reactivity among more proximal case plasmas, compared with controls. Plasmas from cases collected closer to clinical diagnosis (less than 150 days) exhibited less significant elevation of autoantibody response with only 4 identified proteins found to be significantly elevated. Further support for a temporal pattern of reactivity in relation to time of diagnosis was derived from analysis of plasmas from newly diagnosed postmenopausal women, which did not exhibit significantly increased reactivity among cases relative to controls (Fig. 3D).
Using the same linear regression model previously established, ROC analysis yielded increased performance for samples collected further from diagnosis for the spliceosome signature (AUC = 0.83 further from diagnosis and 0.63 closer to diagnosis), whereas that of the glycolysis signature remained relatively constant (AUC = 0.69 and 0.70, respectively; Fig. 3B and C).
We previously showed progressively increased release of glycolysis proteins into the circulation in relation to time to diagnosis (18). We therefore examined the patterns of PKM2 levels in relation to time to diagnosis following blood collection and in relation to PKM2 autoantibodies in a separate set of samples from the WHI cohort. Circulating PKM2 levels were significantly elevated in WHI samples collected within 150 days before diagnosis compared with controls, but not in samples collected further from diagnosis (Fig. 4A). In contrast, autoantibody response to PKM2 exhibited an opposite trend, with significant elevation further from diagnosis (Fig. 4B). Seropositivity, defined as 2 SDs above the mean of the controls for that marker, ranged for autoantibodies to individual glycolysis proteins from 6.3% (ALDOA, GPI) to 14.6% (TPI1) and for spliceosome proteins from 2% (HSPA8, SFRS3) to 12.5% (HNRNPA1; Fig. 4C and D), consistent with the range of biomarker positivity reported in other autoantibody studies (46). Seropositivity for both sets of proteins, based on time of blood draw before diagnosis, ranged from approximately 2 months to 8.5 months. Multiple positive markers among glycolysis or spliceosome proteins were often observed within an individual sample, indicating a broad immune response across proteins in these pathways.
Immune complex formation with increasing levels of antigen is one possible explanation for the observed decrease in autoantibody signal closer to diagnosis. Mass spectrometry analysis of affinity-purified immunoglobulin fractions from newly diagnosed and prediagnostic samples yielded evidence of circulating immune complexes for a number of proteins identified by microarray analysis. Of the 9 identified glycolytic enzymes, 5 (ALDOA, ENO1, GAPDH, PKM2, and TPI1) were observed as part of immune complexes. The most highly significant of these, GAPDH, exhibited an increase in the immunoglobulin bound fraction in cancer samples compared with controls in plasmas from newly diagnosed cases, but not in prediagnostic plasmas consistent with increasing amount of antigen bound to immunoglobulin with tumor development and progression (Supplementary Table S2).
We have used natural protein arrays for comprehensive profiling of autoantigens and autoantibody signatures in a mouse model of breast cancer and in ER+/PR+ breast cancer. Far more proteins were identified in human samples than from the mouse model. Humans are genetically heterogenic and diverse with many more class II alleles represented than the mouse. All mouse models are genetically inbred; therefore, one would expect a more restrictive repertoire. Despite this immunogenic difference, 2 autoantibody signatures consisting of glycolysis and splicesome proteins observed in the mouse model were also observed in human breast cancer plasmas. A strong similarity with antigens associated with SLE and other autoimmune diseases, was noted among identified antigenic proteins in our breast cancer study. We have recently provided evidence of release into circulation of proteins associated with the glycolysis pathway in ER+ breast cancer plasmas before clinical diagnosis (18). We showed increased release of glycolysis proteins with decreasing time-to-diagnosis after blood collection indicative of a positive correlation between circulating glycolysis protein levels and tumor growth. In this study, we provide evidence of stronger reactivity of circulating immunoglobulins with glycolysis proteins arrayed further from diagnosis in relation to blood collection. Analysis of circulating plasma PKM2 levels and associated autoantibodies yielded increased circulating protein levels closer to diagnosis, whereas measurable autoantibodies to PKM2 exhibited an inverse relationship, with increased levels observed further from diagnosis. Because of limited availability of ELISAs, only PKM2 was available for testing of the proteins identified in the glycolysis and spliceosome signatures. The formation of circulating immune complexes depleting plasma of free autoantibodies could explain these observed phenomena. Mass spectrometry analysis of proteins bound to immunoglobulin fractions provided evidence for an increase in immune complexes to the glycolytic enzyme GAPDH in newly diagnosed samples.
An immune response to spliceosome proteins has been associated with autoimmune disease (25). A recent study of fine-needle aspirates from breast cancers and benign lesions yielded evidence of differential expression of spliceosome assembly proteins in tumors compared with benign lesions (47). Reported responses in autoimmune diseases have been limited to Sm and nRNP proteins that are part of the spliceosome complex. In this work, multiple nRNP proteins were identified as autoantigens in prediagnostic breast cancer plasmas. HNRNPs play central roles in DNA repair, cell signaling, and regulation of gene expression at transcriptional and translational levels often through spliceosome complexes. Some splicesome proteins have been implicated in cancer through their regulation of downstream targets. Increased mRNA levels of HNRNPA2B1 were previously reported in melanoma and associated with increased levels of the protein in circulation (48). Expression of HNRNPQ has been shown to be affected by siRNA targeting of ER-α in MCF7 (49). Other HNRNPs have been observed as overexpressed in cancer plasma compared with control (50).
Another component of the spliceosome not previously reported as autoantigenic is the SFRS family of proteins, which we found to be associated with autoantibodies in breast cancer. While most reports localize the SFRS proteins to the cytoplasm and nucleus, one recent study provided evidence of release of SFRS1 into the media of pancreatic cancer cell lines (51). Cell surface localization of SFRS proteins has also been shown in lung cancer (52). Surface-localized SFRS proteins bound fucosylated oligosaccharides through the same mechanism as RNA binding. Our extensive proteomic analysis of the MCF7 breast cancer cell line revealed 10 SFRS proteins that occurred both on the cell surface and in the conditioned media, and 2 additional SFRS proteins in the media. SFRS1, SFRS3, and SFRS6 were all identified on the cell surface and in the media of MCF7.
There is prior evidence of occurrence of autoantibodies to nuclear proteins in cancer (53); however, identification of spliceosome-related autoantibodies is a novel finding in our study. A prior study using recombinant proteins did not yield splicesome autoantibodies in cancer (54). Given the known occurrence of posttranslational modifications in spliceosome proteins, it is plausible that the use of natural proteins in our study may account for the autoantibody reactivity we have observed for splicesome proteins.
Our study shows the use of microarrays of fractionated tumor cell lysate proteins for uncovering immunogenic pathways and autoantibody signatures in breast cancer plasmas. Signatures discovered in a breast cancer mouse model were recapitulated in human, indicative of a similar process of immunogenicity to breast tumor antigens in both mouse and humans. Postmenopausal women with newly diagnosed breast cancer exhibited reduced plasma immunoglobulin binding to arrayed proteins compared with prediagnostic plasmas. This was consistent with the trend observed in prediagnostic, postmenopausal women, in that autoantibody response decreased as time of blood collection before diagnosis decreased. Many recent publications support the idea of using a combination of complementary autoantibody markers for diagnosis. In our study, combining the glycolysis and spliceosome signatures yielded an AUC of 0.77 with 35% sensitivity at 95% specificity. The AUCs obtained may be optimistic, as ROC curves were based on the same data used in generating the pathways of interest. Because these signatures were chosen based on their biologic significance, their use and additivity as biomarker panels may be limited. The sensitivity of biomarkers in prediagnostic samples can also be limited depending on disease progression (55). Because of the limited availability of prediagnostic samples, validation of the described classifier was not possible in this study. Proteomic analysis showed an increase in cancer-related immune complexes to glycolytic enzymes in newly diagnosed postmenopausal patients. The concurrent increase in circulating glycolysis protein levels (18) suggests formation of these complexes may explain the observed immune suppression. With a relatively low sensitivity, but high specificity, one could envision application of this signature to determine women that may be at higher risk of developing breast cancer within a year, thus compelling them to seek screening. Another potential application is for identifying or monitoring masses observed during a routine mammogram. A prospective study involving samples collected at the time of mammography would be necessary to test this application. Blood-based biomarkers for early cancer detection would have great use, especially for women with dense breasts or chronically abnormal mammograms. However, there are data to suggest that general screening may be useful. In a study of nearly 600 reduction mammoplasties, 20% of women, ages 30 to 49 years, had evidence of a proliferative lesion, atypical hyperplasia, or carcinoma in situ in the pathologic specimen (56). These lesions all increase the risk of the development of subsequent invasive malignancy. An inexpensive blood test may assist in identifying cancers early in these patients.
Disclosure of Potential Conflicts of Interest
N.L. Disis has an ownership interest (including patents) in the University of Washington and is a consultant/advisory board member of VentiRx, Roche, BMS, EMD Serono, and Immunovaccine. No potential conflicts of interest were disclosed by the other authors.
Conception and design: J.J. Ladd, T. Chao, J. Qiu, C.I. Li, S.M. Hanash
Development of methodology: J.J. Ladd, T. Chao, J. Qiu, M.W. McIntosh, S.M. Hanash
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.J. Ladd, T. Chao, M.M. Johnson, J. Qiu, A. Chin, J. Mao, M. Wu, C.I. Li, R. Prentice
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.J. Ladd, T. Chao, M.M. Johnson, J. Qiu, R. Israel, S.J. Pitteri, L.M. Amon, M.W. McIntosh, R. Prentice, N.L. Disis, S.M. Hanash
Writing, review, and/or revision of the manuscript: J.J. Ladd, T. Chao, S.J. Pitteri, C.I. Li, N.L. Disis, S.M. Hanash
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Chao, R. Israel, N.L. Disis, S.M. Hanash
Study supervision: S.M. Hanash
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received July 2, 2012.
- Revision received November 3, 2012.
- Accepted December 7, 2012.
- ©2012 American Association for Cancer Research.