RUNX1, a transcription factor mutated in breast cancer, controls the fate of ER-positive mammary luminal cells

The reviewers remain unconvinced by the transcriptional mechanisms provided, and do not feel that the authors have provided a compelling case for Runx1 being a master regulator of the luminal lineage. Although carrying out ChIP-seq analyses on in vivo material seems beyond the scope of current technology for the mammary epithelial field, RNA-seq or microarray on purified populations of in vivo material still seems within the grasp of what can be reasonable to request. The authors argue that they can't do this because the residual YFP-marked MECs in the ML gate are not truly Runx1-null MLs. And the population of LPs is a mixed population of progenitors for either alveolar luminal or ductal luminal cells. Therefore, they don't have evidence that the YFP-marked LPs are progenitors upstream of MLs in the luminal differentiation hierarchy and are blocked in differentiation to ER+ MLs. I agree that this poses a hurdle but it still does not address the caveat that at present the data are restricted to culture studies with a cell line. The authors are encouraged to explore other possible Cre lines in an effort to bolster the in vivo expression data. The inducible KO line does not have to be lineage-specific if coupled with cell surface markers to be used soon after the induction. The authors should expect to see the genes downregulated in the LP cells before the numbers of ML cells actually change. The data as they presently stand could be misleading.

We agree that an inducible Cre line [e.g., K8-CreER for luminal mammary epithelial cells (MECs)] combined with the Runx1 conditional knockout allele (Runx1^L) and a conditional Cre-reporter [e.g., Rosa26-Stop-YFP (R26Y)] may allow us to characterize early molecular changes upon induced Runx1 disruption in ER⁺ luminal MECs. However, this would require extensive mouse breeding in order to put multiple alleles together in the same mouse; therefore it was not feasible for us to use this approach to obtain new in vivo data within two months or an even longer time window. We therefore decided to still focus on our existing MMTV-Cre-based mouse models and performed extensive in vivo expression analysis using sorted MEC subsets from these animals.

If the authors fail to rectify this point, one alternative would be to remove or shorten and tone down the mechanistic data on how Runx1 regulates other transcription, and then look a bit more closely on how RB/p53 and Runx1 interact. Another possibility would be to look in the p53null Runx1nullYFP+ cells and explore their defects, if any. If these cells appear as normal ML cells, this would suggest that the Runx1 null defect is in proliferation/survival of the ML precursors. RB and p53 may not be directly related to the differentiation phenotype and the targets shown, but rather to survival with loss of function mutants. Even with only a small fraction of ML cells being knocked out efficiently in the p53null Runx1null or RBnull Runx1 null double mutant, if the authors can see Runx1 down in this mixed fraction, then they should be able to see other genes changed that came up in their in vitro data. They could at least try qRT-PCR, instead of microarray, on the genes they think are essential from in vitro data, and then see whether these genes are changed in the double mutant. This would strengthen their conclusions.

We have mainly followed this suggestion to obtain our in vivo expression data for genes that have already been tested in our initial cell culture model. Our main strategy was to monitor levels of Runx1 reduction in different subsets of MECs sorted from our MMTV-Cre-based conditional knockout mice and then to determine whether there was any correlation of changes in expression of other genes to reduced Runx1 expression. Took advantage of the rescue of Runx1-null ER⁺ luminal MECs by Rb1-loss, we performed expression analysis in the rescued ER⁺ LPs and ER⁺ MLs from the Rb1/Runx1-null mice (we did not perform this experiment in p53/Runx1-null mice as they exhibited early lethality, which prohibited us from obtaining enough animals for expression analysis). In addition, we also performed expression analysis for these genes in the ER⁺ LP subset from the Runx1 conditional knockout mice (MMTV-Cre;Runx1^L/L;R26Y), as ER⁺ LPs may represent the precursor population for ER⁺ MLs and our new Runx1 expression analysis data shows that this population has partial Runx1 reduction (Figure 4–figure supplement 1G). These in vivo expression analysis data are presented in several new figures, including Figure 7, Figure 7–figure supplement 1 and Figure 7–figure supplement 2, and are described in a new sub-section in the main text.

Below is a summary of key findings from our new expression analysis experiments:

A) Consistent with our in vitro data, we also obtained strong in vivo data to demonstrate that Elf5 is a key target gene repressed by RUNX1. In addition to its repression by RUNX1 in ER⁺ luminal MECs, we found Elf5 is repressed by RUNX1 in almost all other MEC subsets (including basal MECs) in which Runx1 is expressed. The de-repressed expression of Elf5 in basal cells also allows us to provide an explanation for the nursing defects observed in our Runx1 conditional knockout mice (via a potential Runx1-Elf5-Snail2 link in basal/myoepithelial cells). Since the main focus of this paper is to determine the role of RUNX1 in ER⁺ luminal MECs and luminal breast cancer, we did not pursue this further (but this certainly opens a new avenue for a future study).

B) Unexpectedly, from our in vivo expression analysis, we found that Esr1 is upregulated in ER⁺ luminal MECs. This is very different from our previous in vitro data in T47D cells in which we observed a reduction in ERα level upon RUNX1 knockdown. Although we showed previously that RUNX1 binds to a RUNX1-binding motif ∼1.4kb upstream of the ESR1 transcription start site, this binding is relatively weak (∼3-fold enrichment). We repeated the ChIP analysis for this site, as well as for all the other sites we tested previously for ELF5, FOXA1 and CITED1. By applying statistical analysis, we found that while RUNX1 binding to the control regions of ELF5, FOXA1 and CITED1 is all statistically significant, its binding to the -1.4kb site of ESR1 is not significant. Although we cannot rule out RUNX1 binding to other sites in the ESR1 locus, all these data suggest that RUNX1 may not regulate transcription of ESR1 directly. The downregulation of ERα in T47D cells upon RUNX1 knockdown is most likely indirect [e.g., due to RUNX1-loss-induced upregulation of ELF5, as overexpression of ELF5 in T47D cells can also suppress ER expression (Kalyuga et al., 2012)]. To explain why Esr1 appears upregulated in vivo upon Runx1 reduction, we discussed several possibilities and in particular, we provided evidence to support that the observed Esr1 upregulation may be in part due to hyperproliferation of the rescued Runx1-null ER⁺ luminal MECs (i.e., more Esr1-expressing cells present in a sorted MEC subpopulation compared to the same subpopulation from control mice).

C) We provided evidence to support that Foxa1 and Cited1 are target genes of RUNX1. Runx1-loss reduces their expression levels but does not abolish their expression entirely. Their downregulation is more profound in the rescued Runx1-null ER⁺ MLs (new Figure 7B-C), the subpopulation of luminal MECs that is affected by Runx1-loss the most. To strengthen this conclusion, we used multiple normalization approaches (i.e., normalized to Rb1-null single mutant control mice, to wild type control mice, and to younger mice with the same genotype) to control for gene expression changes introduced by differences in cell populations and/or genetic backgrounds.

The authors either need to molecularly link this to differentiation or overcome the hurdles for bolstering the in vitro data with in vivo gene expression data. It would seem that one of these two avenues might be successful, which is needed to clear the path to publishing in eLife.

Our new in vivo expression data, combined with our previous observations, allow us to propose the following revised model: In the ER⁺ luminal lineage, RUNX1 is mainly required during differentiation from ER⁺ LPs to ER⁺ MLs (this is also the stage during which Runx1 expression is notably elevated, Figure 7–figure supplement 2E, new data). Loss of Runx1 leads to abnormally differentiated ER⁺ MLs (i.e., Elf5⁺Esr1^highFoxa1^lowCited1^low ML-like cells, due to de-repression of Elf5 and insufficient upregulation of Foxa1 and Cited1, all of which may impair the sensitivity and/or output of the ER program). This molecular defect may cause cellular stress and subsequently activation of the p53 pathway in these abnormal cells, leading to cell cycle arrest and/or apoptosis; as a result, abnormally differentiated Runx1-null ER⁺ luminal cells are outcompeted by their wild type neighbors in vivo. However, loss of p53 or Rb1 would relieve the cell cycle arrest or positively activate cell cycle, respectively, and/or rescue apoptosis in them, leading to rescue of these abnormal ER⁺ ML-like cells. Upon acquisition of additional mutations, these Runx1/p53-mutant or Runx1/Rb1-mutant ER⁺ premalignant luminal cells eventually progress to ER⁺ luminal breast tumors. We believe this revised model has provided important in vivo mechanistic advances to better explain why and how loss-of-function of RUNX1 leads to development of ER⁺ luminal breast cancer specifically.