Page 162 - Read Online

and miRNA data are derived from the de-identified less than 10 counts and missing data exceeded 10%
publically available TCGA dataset, it is not possible to of all subjects. MiRNAs with less than 10 counts per
[15]
link to any individual. Therefore, no Institutional Review million may be due to sequencing errors. A low
Board (IRB) approval was required. missing value (< 10%) provides the most reliable
and consistent result without the need for further
MiRNA sequencing data from 8 other solid tumors normalization. A total of 153 miRNAs passed the
[16]
in TCGA dataset filtering criteria and data were log2-transformed for
MiRNA sequencing and clinical data from other solid final statistical analysis in HCC.
cancers were also downloaded from TCGA data portal.
Eight solid tumors with available miRNA and clinical Paired t-test with Bonferroni correction for multiple
data in over 40 paired tumor and non-tumor tissues comparisons was used to identify miRNAs that were
were considered in the final statistical analyses, significant different (P < 0.0001) with at least a 2-fold
including female breast invasive carcinoma (BRCA), expression change between the 48 paired HCC tumor
head and neck squamous cell carcinoma (HNSC), kidney and adjacent non-tumor tissues. The volcano plot and
renal cell carcinoma (KIRC), lung adenocarcinoma hierarchical clustering were performed using the panel
(LUAD), lung squamous cell carcinoma (LUSC), prostate of significant miRNAs to describe the distribution of
adenocarcinoma (PRAD), stomach adenocarcinoma miRNAs and tumor classification, respectively. The
(STAD), and thyroid carcinoma (THCA). The samples same miRNA panel was used to construct a heat-map
sizes (pairs) were 102 for BRCA, 71 for KIRC, 59 for and classify the 302 unpaired tumor tissues. The general
THCA, 52 for PRAD, 46 for LUAD, 43 for HNSC, and 41 linear model was used to compare miRNAs expression
for both STAD and LUSC. levels between unpaired HCC tumor and non-tumor
tissues adjusted for covariates significantly different
HCC patients and miRNA data used as the between groups. Prediction analysis of microarrays
validation set using the nearest shrunken centroid methodology
For the first set of validation, we used 32 HCC frozen was used to separately evaluate the classification of
tumor and adjacent non-tumor tissues (16 pairs) that tissues (tumor vs. non-tumor) for paired and unpaired
were collected by the Center for Liver Disease and tumors by those significantly altered miRNAs, and
Transplantation, and stored in the Molecular Pathology estimate prediction error, sensitivity, specificity,
Shared Resource of the Herbert Irving Comprehensive positive predictive value and negative predictive
Cancer Center, Columbia University Medical Center value via cross-validation. Two-sample t-tests were
[17]
(CUMC). This study has been approved by the IRB of applied to identify significant miRNAs (P < 0.0001)
CUMC. Total RNA, including miRNAs was isolated from with over 2-fold changes by age group (< 60 vs. ≥ 60
HCC tissues by RNeasy Microarray Tissue Mini Kits years), gender (male vs. female), BMI (≥25 vs. <25),
(Qiagen, Frederick, MA) according to the manufacturer’s etiologies [alcohol vs. hepatitis B surface antigen
protocol. TaqMan Low Density Arrays (TLDA, Applied (HBsAg) positive vs. anti-HCV positive], AFP (≥ 400 vs.
Biosystems, Foster City, CA), covering 733 miRNAs < 400 ng/mL), and other clinicopathological covariates
(670 unique human mature miRNAs), were used to described above. Subgroups analyses were further
generate miRNA profiles thatwere deposited in NCBI’s conducted among HCC tumor and non-tumor tissues
Gene Expression Omnibus database (accession number carrying one specific risk factor (alcohol, HBsAg or
GSE54751). TaqMan MicroRNA assays were used to anti-HCV) to identify etiologic-specific miRNA panels.
[13]
further evaluate the consistence of candidate miRNA
expression patterns in 66 paired HCC tumor and non- Similar stringent filtering criteria and statistical analysis
tumor tissues from CUMC. U6 snRNA stable in liver strategies were used to identify aberrantly expressed
tumor/adjacent tissues (Ct: 21.19 vs. 21.08, P = 0.398) miRNA profiles from the other 8 different solid tumors.
was used as an endogenous control to normalize the The identified miRNA panels from different tumors
expression of miRNAs using the 2 (-ΔΔCt) approach. [14] were compared to each other to discover “tumor type
specific” or “tumor common” miRNA panels. We define
Statistical analysis “tumor common” miRNAs as those significant for at least
We applied stringent criteria to filter available miRNA 5 tumor types, including HCC, and with fold-changes in
sequencing data before performing any statistical the same direction. “Tumor type specific” miRNAs are
analysis to ensure the reliability and abundance of defined as only significant for one type of tumor among
candidate miRNAs in the target tissues. MiRNAs were the 9 investigated tumors. If miRNAs are significant
excluded from further data analyses if the RPM was for several different tumor types, but the direction in

Hepatoma Research | Volume 2 | June 1, 2016 153