Focused stochastic sampling of a FOXO expression dichotomy in 3D breast epithelial cultures. (A) Heterogeneous expression and localization of FOXO1 and FOXO3 proteins at day 10 of morphogenesis. Frozen sections were stained with the indicated antibodies and imaged by wide-field immunofluorescence. (Scale bar: 25 μm.) (B) Time-dependent induction of a FOXO gene panel during morphogenesis. MCF10A-5E cells were placed under morphogenesis conditions for the indicated times, and population-level mRNA measurements were performed by qPCR. Validated and reported FOXO target genes are shown in red and orange, respectively. Loading controls for qPCR normalization are shown in gray, and genes not previously implicated as FOXO targets are shown in black. Data are shown as the mean ± SEM of triplicate biological samples. (C) Stochastic sampling and qPCR profiling of the FOXO gene panel in matrix-attached cells at day 10 of morphogenesis. qPCR cycle thresholds for the indicated genes were mean-centered, scaled by their amplification efficiency, and clustered by Euclidean distance with average linkage. Samplings are numbered according to their organization in E. (D) Identification of candidate heterogeneities by stochastic sampling. Local gene neighborhoods were compared against a log-normal distribution with 30% coefficient of variation at a false discovery rate equal to 0.05 (yellow) as described (2). Note the clear distinction between genes predicted to be heterogeneously expressed (purple) and those genes whose fluctuations are consistent with background biological variation (gray). (E) Candidate coexpression groups organized by stochastic sampling. Sampling data from the candidate heterogeneities identified in D were scaled to unit log-normal variance and clustered by Euclidean distance with Ward's linkage.

Three-color RNA FISH reveals two groups of FOXO genes with decoupled single-cell expression patterns. (A–C) Multicolor RNA FISH provides a semiquantitative means for characterizing (A) strong, (B) weak, and (C) no coregulation of gene pairs within individual cells. Haptenylated riboprobes complementary to the indicated genes were hybridized to day 10 frozen sections of MCF10-5E acini and imaged by wide-field immunofluorescence. Fluorescence intensities for the individual genes were pseudocolored (Upper), and a mixture of three housekeeping genes was used as a loading control (Lower Left). (Scale bar: 25 μm.) The loading-normalized fluorescence intensities of each gene pair were correlated for single cells of three independent acini to quantify the extent of coregulation (R_{single cell}; Lower Right). (D) FOXO genes are coregulated at the single-cell level within groups 1 and 2 but not between groups. Pairwise single-cell correlations (R_sc) measured by RNA FISH were analyzed as in A–C and compared with the dendrogram derived from stochastic sampling (Fig. 1E). The complete RNA FISH dataset is shown in SI Appendix, Fig. S3. (E and F) Single-cell distributions of RNA FISH intensities from FOXO groups 1 and 2 are heavy tailed, indicating two expression dichotomies. The data from SI Appendix, Fig. S3 are shown as a quantile–quantile plot displaying the observed RNA FISH quantiles (after log transformation) vs. the quantiles of a standard normal distribution. If the data are consistent with a log-normal distribution, the quantiles will plot as a straight line, which was observed for the RNA FISH intensities for the homogeneously expressed gene UBC (SI Appendix, Fig. S3T). Both FOXO groups deviated significantly from a log-normal distribution because of more frequent than expected observations of very high and very low RNA FISH fluorescence (P < 0.001 by Jarque–Berra test).

Heterogeneous RUNX1 signaling decouples FOXO groups 1 and 2. (A) MEME analysis identifies a discriminatory motif for group 1 genes that contains a consensus binding site for RUNX family transcription factors; 5 kb upstream and 1 kb downstream of the transcription start site for the genes in the two FOXO groups were analyzed for discriminative motif discovery by MEME (50). The TRANSFAC database was then searched for aligning consensus sites by TOMTOM, with the motif P value shown for RUNX (89). A similarly informative motif was not identified for FOXO group 2. (B) Relative mRNA copy numbers of RUNX family members in MCF10A-5E cells. Levels of endogenous RUNX1, RUNX2, and RUNX3 were compared with a universal genomic DNA standard and normalized to measured RUNX2 expression. HeLa cells, which express all RUNX family members, were used as a positive control. Data are shown as the mean ± SEM of triplicate biological samples. n.d., not detected. (C and D) RUNX1 protein is homogeneously expressed but heterogeneously phosphorylated at day 10 of morphogenesis. Frozen sections were stained with the indicated antibodies and imaged by wide-field immunofluorescence. (Scale bar: 25 μm.) (E) ChIP-based survey of FOXO and RUNX1 promoter occupancy for the genes in the two FOXO groups. FOXO and RUNX1 consensus sites within 5 kb on either side of the transcription start site are shown in green and yellow, respectively; 150- to 200-bp qPCR amplicons used in the study are shown flanked by 800-bp whiskers indicating the extent of chromatin shearing and thus, the region conceivably detected by each amplicon. (F) Summary of FOXO and RUNX binding sites validated by ChIP. Binding sites were considered validated if their median enrichment over a control IgG ChIP was twofold or more across four independent ChIPs. The number of binding sites between groups was compared by Wilcoxon rank sum test. Similar results were obtained when a more redundant RUNX consensus site, YGYGGTY (91), was used. (G) Knockdown of endogenous RUNX1 without compensatory changes in RUNX2. MCF10A-5E cells were infected with the indicated viral hairpins, and RUNX levels were determined by immunoblotting. β-tubulin was used as a loading control. (H and I) RUNX1 knockdown causes the two FOXO groups to become coregulated. Representative genes from FOXO groups 1 (BCL2L13) and 2 (CDKN1C) were analyzed by RNA FISH as described in Fig. 2 A–C. Single-cell correlations (R_{single cell}) across three independent acini are shown with 90% confidence intervals in parenthesis. (Scale bar: 20 μm.)

RUNX1 knockdown causes proliferative and morphogenetic defects that are blocked by homogenization of FOXO activity. (A) Add back of RUNX1 levels in shRUNX1 cells with murine RNAi-resistant Runx1. MCF10A-5E cells stably expressing murine Runx1 (lane 3) or vector control (lanes 1 and 2) were infected with shGFP (lane 1) or shRUNX1 (lanes 2 and 3), and RUNX1/Runx1 levels were determined by immunoblotting. β-tubulin was used as a loading control. (B) RUNX1 knockdown delays growth arrest and causes abnormally shaped acinar structures. MCF10A-5E acini were fixed at day 14, stained for pRb and HA-tagged murine Runx1 (Runx1), and analyzed by confocal immunofluorescence. (C) Delayed growth arrest caused by RUNX1 knockdown is blocked by the CDK inhibitor roscovitine. MCF10A-5E acini were treated with 5 μM roscovitine on day 10, fixed at day 14, and stained for pRb. The percentage of acini with five or greater pRb-positive cells was quantified in >200 acini per replicate. (D) pRb quantification of the RUNX1 knockdown and add-back experiment described in B. (E) Overexpression of a DN-FOXO1 in the context of RUNX1 knockdown. MCF10A-5E cells stably expressing an HA-tagged DN-FOXO1 (lanes 3 and 4) or vector control (lanes 1 and 2) were infected with shRUNX1 (lanes 2 and 4) or shGFP control (lanes 1 and 3), and RUNX1 levels were determined by immunoblotting. Expression of the DN-FOXO1 was verified by HA expression, and β-tubulin was used as a loading control. (F and G) DN-FOXO1 restores normal acinar shape and growth arrest in acini derived from shRUNX1 cells. Acini from the MCF10A-5E lines described in E were fixed at day 14, stained for pRb and HA-tagged DN-FOXO1, and analyzed by confocal immunofluorescence as described in B and C. Note that acini lacking DN-FOXO1 (orange) remain hyperproliferative and are misshapen. For C, D, and G, data are shown as the mean ± SEM of quadruplicate 3D cultures at day 14. Significance of the interaction between shRUNX1 and DN-FOXO1 in F was determined by two-way ANOVA. (Scale bar: B and F, 25 μm.)

Dual inhibition of RUNX1–FOXO function causes oxidative stress that leads to proliferative suppression. (A and B) RUNX1 knockdown plus DN-FOXO1 synergistically increases global ROS levels in MCF10A-5E acini at day 6. Cells expressing shRUNX1 or an RFP-tagged DN-FOXO1 (Inset, red) were labeled with DCFDA (green) as described in Methods and counterstained with DRAQ-5 (blue) to label nuclei. The percentage of acini staining heterogeneously or homogeneously for DCFDA on a cell by cell basis is shown in B. The remaining acini were negative for DCFDA staining. (C–E) Treatment with the antioxidant Trolox restores hyperproliferation in shRUNX1 + DN-FOXO1 cells. Acini from the MCF10A-5E lines described in Fig. 4E were fixed at day 14, stained for pRb and HA-tagged DN-FOXO1, and analyzed by confocal immunofluorescence as described in Fig. 4 B and C. For B and C, data are shown as the mean ± SEM of quadruplicate 3D cultures at (B) day 6 or (C) day 14. Significance of the interaction between shRUNX1 and DN-FOXO1 was determined by two-way ANOVA. (Scale bar: A, 20 μm; D and E, 25 μm.)

Low RUNX1 expression is associated with high FOXO1 expression in triple-negative breast cancer specimens. (A) RUNX1 and FOXO1 mRNA expression is anticorrelated in triple-negative breast cancers. Correlated probe sets from the Gene Expression Omnibus accession no. GSE6861 dataset were standardized as z scores and averaged to give the estimated relative levels of RUNX1 and FOXO1. The 10th and 90th percentiles of RUNX1 expression (dashed boxes) were further analyzed in B. (B) Tumors with the lowest RUNX1 mRNA expression have the highest FOXO1 mRNA expression. Box and whisker plots from the samples in the dashed boxes in A are shown with notches to indicate 90% nonparametric confidence intervals. Significance for the increase in FOXO1 expression was determined by the Wilcoxon rank sum test.

Stochastic profiling of FOXO target genes reveals cross-talk with the transcription factor RUNX1. (A) Stochastic sampling strategy used to examine single-cell heterogeneities in FOXO target genes (red) for matrix-attached cells in 3D breast epithelial cultures. Housekeeping genes that exhibit more modest biological variations (blue) will not fluctuate as substantially between random 10-cell samplings. (B) Functional relationships between FOXO and RUNX1 expression in breast epithelia. RUNX1 loss or down-regulation causes a hyperproliferative phenotype in the presence of functional FOXO signaling, but it leads to oxidative stress-induced proliferative suppression when FOXO signaling is lost or inhibited.