Variation of salivary gene expression profiles and identification of mRNA biomarkers

Schematic of the study design and demographic information of all subjects used for the discovery and pre-validation phases are shown in Figure 1 and Table 1, respectively. Transcriptomic profiling identified 1402 genes exhibiting >2 fold up-regulation, and 2247 genes exhibiting >2 fold down-regulation, in the saliva of breast cancer patients, relative to the matched controls (n = 20, P<0.05). These transcriptomic changes were unlikely to be due to chance alone (χ² test, P<0.0001), considering the false positive rate with P<0.05. Using a predefined criterion of a change in regulation >2-fold, and a more stringent cutoff of P<0.01, 358 up-regulated and 943 down-regulated transcripts were identified in the saliva of breast cancer samples. RT-qPCR was performed to verify the microarray results on the discovery sample set (n = 20). The top 27 up-regulated candidates (Table S1) were selected based on p-value and fold-change (P<0.01, and >10-fold). The RT-qPCR results confirmed that the relative RNA expression levels of 11 up-regulated transcripts were consistent with the microarray. These verified transcriptomic biomarker candidates were then subjected to independent pre-validation by RT-qPCR using a cohort of 30 breast cancer patients and 63 controls (Figure 1). Eight up-regulated genes were pre-validated, showing significant differences between breast cancer and healthy controls (n = 93, Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Schematic of the study design following the STARD reporting guideline.

https://doi.org/10.1371/journal.pone.0015573.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Demographic Information of All Subjects Used for the Discovery and Pre-validation Phases.

https://doi.org/10.1371/journal.pone.0015573.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Validated biomarkers for breast cancer detection and effect of confounding factors (Pre-validation sample set n = 93).

https://doi.org/10.1371/journal.pone.0015573.t002

Variation of salivary proteomic profiles and identification of protein biomarkers

Proteomic profiling by 2D-DIGE revealed 35 up-regulated proteins/spots and 32 down-regulated proteins/spots in the saliva of breast cancer patients, relative to the matched controls (n = 20). Twenty spots, 14 up-regulated (>1.5 fold) and 6 down-regulated (>1.5 fold), were selected for protein identification, resulting in the identification of 10 up-regulated and 4 down-regulated proteins (Table S1). Four proteins (carbonic anhydrase VI (CA6), psoriasin, transthyretin, and cyclophilin A) with available antibodies were subjected to verification using immunoblot on the discovery sample set. The levels of CA6 and psoriasin between cancer and control samples showed significant differences (p = 0.012 and 0.014, respectively). These verified proteomic biomarker candidates were then independently validated by protein immunoblotting using the pre-validation cohort (30 breast cancer patients versus 63 controls). The level of CA6 showed a significant difference between breast cancer and healthy controls (n = 93, Table 2).

Evaluation of the validated transcriptomic and proteomic biomarkers

Using logistic regression, the accuracy, sensitivity and specificity of 9-validated-biomarker combination on the pre-validation sample set (n = 93) were 92% (86 of 93), 83% (25 of 30) and 97% (61 of 63), respectively (Figure 2A). Principle component analysis (PCA) of this 9-biomarker combination could separate the breast cancer patients from the controls along the first principal component (t-test, P-value = 2.7E-15) (Figure 2B). None of the confounding factors (age, ethnicity, smoking status, menopausal status, and HRT treatment) significantly affected the validated biomarkers (Table 2). These indicate that cancer onset is a major source of variation in the expression of the validated biomarker. Furthermore, cross-disease comparisons showed that none of the validated mRNA biomarkers' expression was significantly altered in other salivary transcriptomic profiling studies, indicating their specificity for breast cancer detection (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Clinical utility of the validated biomarkers.

A, Combination of nine validated biomarkers achieved a sensitivity of 83% (25 of 30 cancer subjects) with only a 3% false-positive rate (2 of the 63 control subjects). The shading of the contingency table boxes reflects the fraction of each samples type in each quadrant. ‘Cancer’ and ‘Non’ headings indicate subjects with and without cancer, respectively. SB+ and SB−, salivary biomarker test positive or negative, respectively; NPV, negative predictive value; PPV, positive predictive value; Sen, sensitivity; Spec, specificity. B, Score plot of principle component analysis (PCA). Combining the nine validated biomarkers, the control subjects (green dots) separate from breast cancer patients (red dots), with cumulative proportions of 66.9% for PC1 and 21.6% for PC2.

https://doi.org/10.1371/journal.pone.0015573.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Cross-disease comparisons of 8 validated salivary mRNA biomarkers.

https://doi.org/10.1371/journal.pone.0015573.t003

Subject information and study design

This study, which was approved by the UCLA and Cedars-Sinai Medical Center Institutional Review Boards (#06-07-043 and #3870, respectively), began sample collection in February 2007. Written informed consents and questionnaire data sheets were obtained from all patients who agreed to serve as saliva donors. The saliva bank for breast cancer project at the UCLA Dental Research Institute, in collaboration with the Cedars-Sinai Medical Center, has collected 178 saliva samples from subjects recruited from the Saul and Joyce Brandman Breast Cancer Center. Of these, 113 samples, including 40 breast cancer patients and 73 healthy control individuals (Table 1), were used for the discovery and pre-validation phases of this study. Inclusion criteria of cancer patients consisted of a confirmed diagnosis of breast cancer. Exclusion criteria of cancer patients included therapy/surgery and/or a diagnosis of other malignancies within 5 years prior to saliva collection. Exclusion criteria of control patients included a diagnosis of any malignancies within 5 years prior to saliva collection (Figure 1). The information on patient characteristics, such as age, ethnicity, smoking history, menopausal status, and hormone replacement therapy (HRT), is presented in Table 1. Unstimulated saliva samples were consistently collected, stabilized, and preserved as previously described [34] (Figure S3). The sample supernatants were reserved at −80°C prior to assay.

This study consisted of a discovery phase, followed by an independent preclinical validation phase. Of the 113 samples, 10 breast cancer samples and 10 matched control samples were used for the discovery phase. All breast cancer cases were invasive ductal carcinoma (IDC), the most common type of breast cancer. Biomarkers identified from the discovery studies were first verified using the discovery sample set. An independent sample set, including 30 breast cancer patients and 63 controls, was used for the biomarker pre-validation phase (Figure 1).

Salivary transcriptomic profiling and data analysis

RNA was isolated from 330 µl of saliva supernatant using MagMax™ Viral RNA Isolation Kit (Ambion, Austin, TX). This process was automated using KingFisher mL technology (Thermo Fisher Scientific, Waltham, MA), followed by TURBO™ DNase treatment (Ambion, Austin, TX). Extracted RNA was linearly amplified using the RiboAmp RNA Amplification kit (Molecular Devices, Sunnyvale, CA). After purification, cDNA was transcribed and biotinylated using GeneChip Expression 3′-Amplification Reagents for in vitro transcription labeling (Affymetrix, Santa Clara, CA). Chip hybridization and scanning were performed at the UCLA microarray core facility. Using the MIAME criteria [47], all Affymetrix Human Genome U133 Plus 2.0 Array data generated in this study were uploaded to the GEO database [48], accession number GSE20266.

The analysis was performed using R 2.7.0 with samr and ROC packages [49]. The Probe Logarithmic Intensity Error Estimation (PLIER) expression measures were computed after background correction and quantile normalization for each microarray dataset. Probeset-level quantile normalization was performed across all samples to make the effect sizes similar among all datasets. Finally, for every probeset, significance analysis of microarray (SAM) was applied to identify differential expression between the cancer and healthy control samples. The probesets were then ranked by the false discovery rate (FDR) corrected p-values.

Preclinical validation of mRNA biomarkers using reverse transcription quantitative PCR (RT-qPCR)

The identified mRNA biomarkers were first verified by RT-qPCR using the discovery sample set (10 cancer versus 10 controls) as described previously [34]. RT-qPCR primers were designed using Primer Express 3.0 software (Applied Biosystems, Foster City, CA) (Table S2). All primers were synthesized by Sigma-Genosys (Woodlands, TX), and the amplicons were intron spanning whenever possible. RT-qPCR was carried out in duplicate. Verified biomarkers were then assayed by RT-qPCR in the set of 93 independent samples (30 breast cancer patients versus 63 controls). Raw data were normalized by subtracting GAPDH Ct values from the biomarker Ct values to generate ΔCt. The Mann-Whitney rank sum test was used for between-group biomarker comparisons.

Salivary proteomic profiling and data analysis

Two-dimensional difference gel electrophoresis (2D-DIGE) was performed by Applied Biomics (Hayward, CA). Briefly, by taking equal amounts of protein from each sample, 10 cancer samples and 10 control samples were pooled separately, with each pool containing 250 µg of proteins. The proteins in each pool were precipitated by methanol and labeled with Cy3 and Cy5, respectively, and then combined for 2D-DIGE. After loading the labeled samples, the isoelectric focusing (IEF, pH 3–10) was run following the protocol provided by Amersham BioSciences (Piscataway, NJ). The immobilized pH gradient (IPG) strips were rinsed in the SDS-gel running buffer before transferring onto 13.5% SDS gels. The fold change of the protein expression levels was obtained from in-gel DeCyder analysis (Amersham BioSciences). Spots with a fold-change larger than 1.5 on the gel were subjected to in-gel trypsin digestion. The digested tryptic peptides were then mixed with CHCA matrix (alpha-cyano-4-hydroxycinnamic acid) and spotted into wells of a MALDI plate for MALDI-TOF MS identification (ABI4800, Applied Biosystems, Foster City, CA).

Preclinical validation of protein biomarkers by immunoblotting

Protein immunoblotting was used to verify and validate the proteomic biomarker candidates. Reduced protein samples (15 µg total protein per lane) were loaded onto a 10% Bis-Tris gel and run at 150 Volt for one hour. Pre-stained protein standard (Invitrogen, USA) was used to track protein migration. The proteins were transferred to a nitrocellulose membrane and blocked for one hour in 5% non-fat dry milk. After further washes in TBST wash buffer, the membrane was incubated with the primary antibody (Lifespan bioscience, Seattle, WA) at room temperature for two hours. The membrane was then washed in TBST wash buffer before applying the secondary antibody (Anti-mouse IgG, peroxidase-linked species-specific whole antibody from sheep, GE healthcare, Piscataway, NJ) for one hour at room temperature. Finally, the membrane was washed in TBST wash buffer and visualized using the ECL Plus detection kit (GE Healthcare, Piscataway, NJ). The signal intensity of the bands was measured using Image J software (NIH, Bethesda, MD, USA). The intensity of a band representing the protein of interest was divided by the intensity of its corresponding β-actin expression on the same blot for normalization.

Statistical analysis

Leave-one-out cross-validation was applied to assess the true accuracy of the model. In this procedure, each observation is iteratively taken out and the model is trained using all other observations. A prediction is then made on the left-out observation. The overall accuracy rate for each model is then the proportion of left out observations that are correctly predicted. To evaluate possible confounders for the markers versus cancer relationship, we examined factors such as age, ethnicity, smoking status, menopausal status, and HRT treatment. Linear regression model was constructed for each marker and used the factors cancer/normal and one of the above confounders.

The pre-validated breast cancer mRNA biomarkers were checked in other microarray studies that have been conducted in our laboratory on different diseases, including oral cancer [31], primary Sjögren's Syndrome (pSS) [30], pancreatic cancer [34], lung cancer, ovarian cancer, and type 2 diabetes. Briefly, P-value derived from Wilcoxon rank sum test were calculated for all breast-cancer-study-validated genes in other microarray datasets to check whether significant variation between breast cancer and controls also appeared in those disease datasets. After Bonferonni correction, variation was considered significant with p-values less than 0.006.