A) Representative p16 immunohistochemistry.
B) High p16 staining fails to stratify women with DCIS that develop subsequent disease. Recurrence-free survival plots demonstrate that women with DCIS staining high or low for p16 develop subsequent disease at the same rate.
C) Ki67 index labelling stratifies recurrence-free survival in women with DCIS.
D) DCIS exhibiting high p16 immunostaining and elevated Ki67 identifies women that have a reduced recurrence-free survival.
E) Ki67 does not differentiate risk in DCIS lesions with low p16. Box plots and corresponding p-values were determined using Wilcoxon/Kruskal-Wallis rank of sums test. Survival plots were generated using Kaplan-Meyer analysis.

A) Unsupervised hierarchical clustering of 130 invasive tumors using 589 unique genes that are intrinsically variable (see Supplemental File 1) identifies luminal A, luminal B, normal-like, HER2 positive and basal-like clusters. Molecular subtypes, additionally classified by a nearest centroid approach are indicated by coloured bars (light blue: Luminal B; dark blue: Luminal B; green: Normal-like; orange: Basal-like; pink: HER2 positive tumors; grey: Unclassified).
B) Expression levels of p16, Rb, E2F3, cyclin E1, cyclin D1 and COX-2. The level of expression of each gene is relative to the median across all samples as represented by color saturation.

mRNA expression for p16, Rb, cyclin D1 and COX-2 in the indicated four publicly available datasets utilizing different microarray platforms. Subtype classifications are as determined previously.

A) p16 mRNA and protein levels show concordance using Wilcoxon rank of sums test.
B) p16 mRNA and protein concordance across molecular tumor subtypes. The log median centered mRNA level of p16 is plotted (red triangles) in comparison with the corresponding protein levels (black squares).
C) COX-2 mRNA and protein levels demonstrate poor concordance using Wilcoxon rank of sums test.
D) Discordance between COX-2 mRNA and protein levels is enriched in HER2 positive tumors. The mRNA level of COX-2 for each sample is plotted (red triangles) in comparison with corresponding protein levels (black squares).

A) Representative COX-2 immunohistochemistry.
B) COX-2 measured as a single variable fails to stratify risk of recurrent disease or recurrence-free survival.
C) DCIS lesions high for COX-2 and Ki67 identify women that develop subsequent breast cancer and have reduced recurrence-free survival.
D) Ki67 does not differentiate risk in DCIS lesions with low COX-2. Box plots and corresponding p-values were determined using Wilcoxon/Kruskal-Wallis rank of sums test. Survival plots were generated using Kaplan-Meyer analysis.

A) Cells of varying malignant potential were retrovirally infected with a constitutive expressing COX-2 construct or an empty vector control (LXSP). Cells were pulsed for 4 h with BrdU and analyzed by flow cytometry following propidium iodide staining. The S-phase fraction is expressed as fold increase of COX-2 expressing cells compared to vector control cells.
B) Phase contrast micrographs of HMEC, vHMEC and HMEC expressing HPV16-E7. Cells either retrovirally infected with a constitutive COX-2 expressing construct or an empty vector control (LXSP).
C) Western blot analysis of COX-2, p16, p53 and p21 in cell lysates from COX-2 overexpressing and vector control cells.
D and E) HMEC were infected with retrovirus expressing a short hairpin RNA targeting p16 (shp16), cyclin D1 (cycD1), a short hairpin RNA targeting Rb (shRb), HPV16-E7, or their corresponding empty vector controls. (D) COX-2, p16, Rb, cyclin D1, E2F1 and actin were detected by western blot. (E) COX-2 protein expression was determined by immunoblotting following 24 h exposure to 1ng/ml TGF-β.
F) Diagram representing the connection between p16/Rb pathway and COX-2.