1. Head and Neck Cancer: microRNA analysis
Amy Li
Monti Lab Rotation
Boston University
11/25/13

2. Dataset
Head and Neck Cancer Dataset from The Cancer Genome Atlas (TCGA)
Contains large and well-documented data in many cancer subtypes
DNA-methylation, SNP Array, RNA-seq, miRNA-seq, low pass DNA-seq, Reverse Phase Protein Array
Normalized miRNAseq data
463 samples
39 patients: tumor tissue and adjacent normal tissue
385 patients: tumor tissue only
1046 miRNA
Clinical information
360 samples, 71 clinical attributes (ie. anatomic subdivision, gender, grade, race, stage)

3. Goals Identify miRNA markers of:
Cancer status:
Normal vs. tumor
Cancer progression:
Differentially expressed in each stage or grade
Integrate miRNA expression with gene expression data
Gene set enrichment in miRNA targets
mRNA data (Vinay)

4. Tumor Progression Classification
Tumor Grade:
Assigned based on how abnormal the tumor cells looks under a microscope
Ranges from G1 (well-differentiated) to G4 (undifferentiated)
Well differentiated tumor cells from a lower grade resemble normal cells, tend to spread slowly, and is generally indicative of better prognosis
Tumor Stage:
Based on size or extent (reach) of the primary tumor

5. Analysis Overview Exploratory data analysis Unsupervised clustering
Mean vs. standard deviation
Boxplots of miRNA expression
Data filtering
Clinical demographics
Unsupervised clustering
Heatmaps
Fisher test for association between clusters and sample attribute assignment (tumor status, grade, stage, etc)
Tests for confounders
Association between grade and other attributes, ie: ethnicity, gender, smoking history, age, alcohol consumption
Differential Analysis
Look for differentially expressed genes with respect to grade or stage
miRNA targets and Gene Set Enrichment Analysis
Identify sets of miRNA targets and see whether such gene sets are enriched with respect to disease phenotype

6. Exploratory Data Analysis: Mean vs. standard deviation
For each gene: plot mean vs standard deviation
Linear relationship

7. Exploratory Data Analysis: Boxplot of Expression
For each gene: made boxplot of expression across the samples
Reordered based on median

8. Exploratory Data Analysis: Data filtering
Sample filtering:
Samples without clinical labels
Gene filtering:
Lowly expressed genes
Row maximum
Genes with constant expression
Standard deviation
Full Matrix
1046 × 463
Filtered Matrix
692 × 393

9. Exploratory Data Analysis: Clinical Demographics
Grade:
G1: well-differentiated
G4: Undifferentiated

10. Exploratory Data Analysis: Clinical Demographics
Oral: oral cavity, oral tongue,, bucal mucosa

11. Exploratory Data Analysis: Clinical Demographics

12. Unsupervised Clustering: Paired samples: Tumor vs. Adjacent Normal38 1
Tumor 2 37
Fisher Test:
Test for association between Cluster Assignment and Actual Class Label
P-val ~ 0
Row: pearson, ward
Col: pearson, ward

13. Unsupervised Clustering: Grades: G1, G2, G3, G4
Fisher Test:
Tested for association between grades and cluster assignments for total number of clusters ranging from 2 to 5
P-vals not significant in all cases
Row: pearson, ward
Col: euclidean, ward

14. Tests for confounders
Tested for association between grade and the putative confounding variable using Fisher test (discrete variables) or ANOVA (continuous variables)
Ethnicity (p=0.57), race (p=0.84), gender (p=0.09), age (p=0.55), alcohol consumption (p=0.63)
Correct gene expression for gender using a linear regression model prior to performing differential analysis

15. Data Processing for Differential Analysis
Sample Filtering
Removed samples without clinical labels
Removed samples sequenced on IlluminaGA (kept IlluminaHiseq samples)
Removed samples with minority races (kept “white”)
Gene filtering
Removed miRNAs with low expression (90% quantile < 100)
Removed miRNAs with constant expression (sd < 0.1)
Attribute Filtering
Grade:
Removed GX and “Not Available”
Kept G1, G2, G3, G4
Stage:
Removed “Not Available”
Kept S1, S2, S3, S4A, S4B

16. Differential Analysis: Grade
diffAnal.R
Performs permutation tests to identify significant genes differentially regulated in one of two classes
Normalized expression matrix corrected for gender
Class label: grade attribute binarized to “low” vs. “high”
Run diffAnal for each high vs. low cutoff:
G0 (adjacent normal) vs. G1-G4 (tumor)
G1 vs. G2-G4
G1-G2 vs. G3-G4
G1-G3 vs. G4

17. Differential Analysis: Grade : G0 vs. G1-G4

18. Differential Analysis: Grade : G1 vs. G2-G4

19. Differential Analysis: Grade : G1-G2 vs. G3-G4

20. Differential Analysis: Grade : G1-G3 vs. G4

21. Differential Analysis: Trends
Found more significant markers for tumors vs. normal than for distinguishing between low and high grades
Performed same analysis for stage, significant markers for stage are weaker than that of grade
For both grade and stage, most significant markers found by diffAnal show upregulation in the later disease state.

22. Differential Analysis: Tumor Classification Marker
148 total genes (90% quantile > 100) used for diffAnal
65 significant genes upregulated in tumors
37 significant genes downregulated in tumors
Cutoff: FDR < 0.01

23. Differential Analysis: Cancer Progression Marker for Grades
Cancer progression marker will satisfy ALL of:
Tumor classification marker
Significant FDR in 2/3 runs of diffAnal
Monotonous increase or decrease across grades
4 miRNA markers identified (all are upregulated with increasing grade)
G1-_vs_G2+_fdr
G2-_vs_G3+_fdr
G3-_vs_G4+_fdr
hsa-mir-106b
0.03
0.01
0.04
hsa-mir-15b
0.05
0.81
hsa-mir-582
hsa-mir-151
0.61
-0.25
hsa-mir-196b
0.18
hsa-mir-10a
0.32
-0.96
hsa-mir-374a
-0.6
hsa-mir-128-2
0.26
hsa-mir-25
0.02
hsa-mir-128-1
0.47
0.17
hsa-mir-28
0.44

24. Differential Analysis: Cancer Progression Marker for Grades

25. Finding miRNA Targets
miRWalk
Targetscan
mirBase

26. Finding miRNA Targets miRNA AhR Targets Targets (54) (162)
miRWalk
“Validated targets” module
Targets for differentially expressed miRNAs: 162 targets
Intersect targets found by miRWalk with AhR targets
6 matches: NQO1, NFE2, IL1B, TNF, TGFB1, MYC
miRNA
Targets
(162)
AhR Targets
(54)
miRNA markers
(4)
(6)

27. Work in Progress Gene set enrichment analysis:
Consider targets of strong miRNA markers as a gene set
Is there an enrichment of this defined gene set in certain disease phenotypes, ie. high grade?
Pathway analysis:
Which pathways are these miRNA markers involved in?
Modeling tumor progression:
Explore other definitions of tumor progression markers