Bmi1 marks distinct castration-resistant luminal progenitor cells competent for prostate regeneration and tumour initiation

Identification of defined cell populations with stem/progenitor properties is key for understanding prostate development and tumorigenesis. Here we show that the polycomb repressor protein Bmi1 marks a population of castration-resistant luminal epithelial cells enriched in the mouse proximal prostate. We employ lineage tracing to show that these castration-resistant Bmi1-expressing cells (or CARBs) are capable of tissue regeneration and self-renewal. Notably, CARBs are distinct from the previously described luminal castration-resistant Nkx3.1-expressing cells (CARNs). CARBs can serve as a prostate cancer cell-of-origin upon Pten deletion, yielding luminal prostate tumours. Clonal analysis using the R26R-confetti allele indicates preferential tumour initiation from CARBs localized to the proximal prostate. These studies identify Bmi1 as a marker for a distinct population of castration-resistant luminal epithelial cells enriched in the proximal prostate that can serve as a cell of origin for prostate cancer.

6. The conclusion that tumors predominantly arise in the proximal prostate should be supported by low-power H&E images showing the proximal-distal distribution of neoplastic regions. 7. The authors interpret the increased size of marked cell clusters in the proximal region versus the distal prostate as indicating that proximal Bmi1-expressing cells are more efficient at cancer initiation. Can the authors exclude the possibility that clones arising in the proximal region remain coherent, whereas clones arising more distally tend to disperse, thereby giving the appearance of single marked cells? The authors could perhaps address this issue in a different way by performing grafts using proximal versus distal regions. 8. Supp. Fig. 2 summarizes data from flow-sorting experiments, but no FACS plots are presented to show primary data, and the methods used are not described.

Reviewer #2 (Remarks to the Author):
Bmi-1 is a component in the PRC-1 chromatin silencing complex that plays a critical role in regulating the activities of normal and cancer stem cells. A previous study, published in Cell Stem Cell (Lukacs et al., 2010) and using tissue recombination assays, has shown that Bmi-1 is important for murine prostate stem/progenitor cell activity and prostate regeneration upon experimental castration. What's unclear is whether some endogenous Bmi-1 expressing cells might be unique and be able to function as a cell of origin for prostate cancer. This study developed a lineage-tracing model to address this critical outstanding question in the field and is thus significant. Also noteworthy is that most experiments were meticulously quantified.
Major points: 1. Supplementary Fig. 1: Most Bmi1+ cells are luminal, which is inconsistent with a previous report by Lukacs et al (Cell Stem Cell, 2010) that suggested that the majority of Bmi-1+ cells coexpressed CK5. Unfortunately, their results with the use of Bmi1-GFP KI prostates (Supplementary Figure 2) did not provide the definitive support to their immunostaining results, as the GFP+ cells seemed to be much lower than IF staining. Perhaps use of different antibodies and appropriate negative controls might clarify these inconsistent reports. 2. The lineage-marking efficiency is extremely low with only 0.3-0.4% total epithelial cells labeled with YFP, compared to ~60% luminal (CK8+) and ~22% basal (CK5+) cells expressing endogenous Bmi1 protein in the AP. Why so low efficiency of labeling from the endogenous Bmi-1 locus? What's the difference (e.g., cell-cycle profile, clonogenic properties, regenerative capacities, etc) between the minor population of traced Bmi+ cells vs. the majority of non-traced Bmi-1+ cells? 3. It is interesting that upon castration, the % of Bmi1+ cells, or CARBs, increased to ~0.7% of the total epithelial cells. Since Bmi1+ cells seem to be largely quiescent in intact prostate, where did this doubling of the Bmi1+ cells come from? From initially weakly/negatively marked YFP+ cells or from the duplication of YFP+ cells? 4. The results about the cell cycle entry and exit ( Figure 5) are potentially interesting, but the overall data is relatively weak considering the low GFP+% and subtle differences in quantification data (Fig. 5d, 5e). 5. After castration the majority (>95%) of Bmi1+ cells are luminal, but the overall Bmi1-GFP+ cells are few (0.7%). The authors assumed that the Cre-mediated PTEN deletion primarily occurred in luminal cells, but at least theoretically Cre could also be activated in basal cells, which could lead to the same observations. 6. Supplementary Fig. 1i: It's particularly interesting that CARBs are distinct from CARNs. Since NKX3.1 is exclusively expressed in luminal cells, and the Bmi-1 are also mainly localized in the luminal layer, are they co-localized before castration? If yes, why do some cells express NKX3.1 while others express Bmi1 in regressed prostate? 7. What is the difference in tumor growth kinetics of Bmi1-PTEN tumor versus published NKX3.1-PTEN tumor?
Minor points: 1. Supplementary Fig. 2d, 2e: Data presented and interpreted are confusing and may be inconsistent. Authors interpreted S2d as "GFP in both luminal and basal cell fractions". Based on the data, about 42% GFP+ cells are luminal (Sca1-CD49f-lo), and >30% GFP+ cells are basal (Sca1+CD49f-hi). However, in the S2e, authors stated that "94% of all GFP+ (i.e. Bmi1+) prostate epithelial cells were luminal ( Supplementary Fig. 2e)." Are 94% GFP+ cells in luminal fraction or 94% of total GFP+ cells are luminal? 2. Their finding that CARBs are distinct from CARNs in regressed mouse prostates is interesting, which raises the possibility that many subpopulations of cells in both the luminal and basal cells can survive castration. The main difference is that CARNs, also ranging from 0.3-0.7%, are purely luminal whereas CARBs are predominantly luminal but may also be basal. It will be very interesting to compare CARBs and CARNs in future studies. 3. Some statements are inaccurate. For example, on page 7 (first paragraph), the authors wrote that "Lesions were distinctly labeled with one of the 4 confetti colors.....' but in Supplementary

Reviewer #1 Concern 1. The authors report that Bmi1-expressing cells are mostly localized to the proximal region of the prostate, but do not explicitly define "proximal". In particular, the authors should show lowmagnification immunostaining of hormonally-intact and regressed prostate sections to show the distribution of Bmi1-expressing cells along the proximal-distal axis.
Response 1: We divided the prostate gland into proximal, intermediate and distal thirds. This has been clarified in the text. Low-magnification images for the distribution of Bmi1 or YFP of hormonally-intact and regressed prostate sections are also now included in the new Fig. 1d, Supplementary Fig. 1b, and Supplementary Fig. 3a-c to clearly show this.
Concern 2. The lineage-marking efficiency for the Bmi1-CreER driver appears to be extremely low in hormonally-intact mice. In wild-type mice, the authors report that 60% and 21.6% of luminal and basal cells express Bmi1, respectively, yet only 0.3-0.4% of epithelial cells are marked by YFP after tamoxifen induction of the BY mice. How do the authors interpret this highly inefficient marking? Is this marking representative? In contrast, the efficiency of lineage-marking appears to be much higher in the regressed prostate. What accounts for this large difference in marking efficiency between the experiments? Response 2: We do make note of the low marking efficiency of the Bmi1-CreER driver in the prostate. Our first concern was with Bmi1 antibody specificity, which led us to evaluate the percentage of Bmi1+ cells using two different antibodies against Bmi1 (Supplementary Table 4) obtaining similar results. We also evaluated tissues with known Bmi1 expression patterns (duodenum, pancreas) as additional controls. The low recombination efficiency in Bmi1-Cre ER transgenic mice might due to Bmi1 promoter activity. Nevertheless, Bmi1+ and YFP+ cells showed many similarities in cell lineage composition and other characteristics. For example, the ratio of Bmi1+ luminal to basal cells was similar to that of YFP+ luminal to basal cells in the adult intact and regressed prostates (Supplementary Table 1 and 2). Moreover, we present new data showing that both Bmi1+ and YFP+ cells are more resistant to apoptosis after castration ( Fig. 2i-n). Finally, using either Bmi1 immunostaining or YFP expression to mark CARBs showed them to be distinct from CARNs in either castrated Bmi1-Cre ER ;R26R-YFP mice or castrated Nkx3.1Cre ERT2 ;R26R-YFP mice (Fig.  2h, i and Supplementary Fig. 3d-h). Altogether, these observations suggest that although lineage marked YFP+ cells were detectable at low efficiency, this marking is most likely representative (page9, line4). Fig. 1f Figure 1o, why does the Ki67 immunostaining appear to be cytoplasmic? Response 4: Statistical analysis showed that this difference is significant (p<0.01 data from3 mice). A clearer image for Ki67 immunostaining in Figure 1o is now shown.

Concern 3. Although Figures 1 and 2 show images from each prostate lobe, it is unclear which lobe(s) is being quantitated in
Concern 5. The authors conclude that the Bmi1-expressing cells in the hormonally-intact prostate are castration-resistant based on their percentage increase from 0.3% to 0.64% after castration. However, this quantitation lacks a statistical analysis to determine whether 0.64% is indeed different from 0.3%. Furthermore, the authors should directly assess whether lineage-marked cells undergo apoptosis during regression.
Response 5: To directly examine whether YFP+ cells identified in intact mice by treating BY mice with tamoxifen are castration resistant, we first treated BY mice by tamoxifen to label the cells then castrated the mice (Fig. 3j). After castration, the fraction of YFP+ cells increased from 0.3% to 0.64% (n=253 YFP+ cells counted from a total of 39281 cells from 4 mice; p<0.001; Fig. 3k). These results suggest that tamoxifen treatment of BY mice before castration labels castration-resistant cells. Since we observed an increased fraction of YFP+ cells after castration compared to that in intact mice, we further investigated whether YFP+ cells are resistant to apoptosis after castration. As it had been reported that the peak of epithelial apoptosis in the mouse prostate occurred by 3-4 days after castration, the fraction of apoptotic cells were evaluated at 3 days after castration of BY mice following tamoxifen treatment using cleaved caspase-3 immunodetection (Fig. 3j). Interestingly, no YFP+ or Bmi1+ cells co-expressing cleaved caspase-3 were observed (Fig. 3l-n; P<0.001, n= 242 YFP+ and 862 Bmi+ cells counted from 4 mice), demonstrating the relative castration resistance of YFP+ cells prospectively marked in intact adult animals.

Concern 6. The conclusion that tumors predominantly arise in the proximal prostate should be supported by low-power H&E images showing the proximal-distal distribution of neoplastic regions.
Response 6: Low-magnification H&E images showing the proximal-distal distribution of neoplastic lesions is now included in Figure 6c.

Concern 7. The authors interpret the increased size of marked cell clusters in the proximal region versus the distal prostate as indicating that proximal Bmi1-expressing cells are more efficient at cancer initiation. Can the authors exclude the possibility that clones arising in the proximal region remain coherent, whereas clones arising more distally tend to disperse, thereby giving the appearance of single marked cells? The authors could perhaps address this issue in a different way by performing grafts using proximal versus distal regions.
Response 7: We only observed singly labelled confetti cells in both distal and proximal prostates of Bmi1-Cre ER ;R26R-confetti intact mice early after tamoxifen induction, making it unlikely that the differences in clone size observed are due to existing coherent clones in the proximal prostate. These data are now shown in Supplementary Fig. 4c, d. Concern 8. Supp. Fig. 2 summarizes data from flow-sorting experiments, but no FACS plots are presented to show primary data, and the methods used are not described. Response 8: FACS plots and methods are now included in Supplementary Fig. 2b-d. We have also added new RT-PCR data from sorted basal and luminal cells showing that CD49f+Sca1-luminal cells robustly express Bmi1.

Reviewer #2
Concern 1. Supplementary Fig. 1 Figure 2) did not provide the definitive support to their immunostaining results, as the GFP+ cells seemed to be much lower than IF staining. Perhaps use of different antibodies and appropriate negative controls might clarify these inconsistent reports. Response 1: We were indeed surprised to find that most Bmi1+ cells are luminal. We therefore used two distinct Bmi1 antibodies in validated immunohistochemical assays, RT-PCR analysis of Bmi1 expression in FACS-sorted prostate epithelial luminal and basal cells, a Bmi1-GFP knockin mouse and lineage tracing with a Bmi1CreER driver to show that Bmi1 is expressed in luminal cells. Furthermore, our functional data on luminal CARBs indicate that Bmi1 marks progenitor luminal cells that can serve as a prostate cancer cell of origin. We do note that although the report by Lukacs et al emphasized the functional role of Bmi1 in basal cells, their study did not rule out a role for Bmi1 expression in luminal cells. Conversely, our study did not exclude a functional role for Bmi1 in prostate basal cells.

: Most Bmi1+ cells are luminal, which is inconsistent with a previous report by Lukacs et al (Cell Stem Cell, 2010) that suggested that the majority of Bmi-1+ cells coexpressed CK5. Unfortunately, their results with the use of Bmi1-GFP KI prostates (Supplementary
Concern 2. The lineage-marking efficiency is extremely low with only 0.3-0.4% total epithelial cells labeled with YFP, compared to ~60% luminal (CK8+) and ~22% basal (CK5+) cells expressing endogenous Bmi1 protein in the AP. Why so low efficiency of labeling from the endogenous Bmi-1 locus? What's the difference (e.g., cell-cycle profile, clonogenic properties, regenerative capacities, etc) between the minor population of traced Bmi+ cells vs. the majority of non-traced Bmi-1+ cells? Response 2: We do make note of the low marking efficiency of the Bmi1-CreER driver in the prostate. Our first concern was with Bmi1 antibody specificity, which led us to evaluate the percentage of Bmi1+ cells using two different antibodies against Bmi1 (Supplementary Table 4) obtaining similar results. We also evaluated control tissues with known Bmi1 expression patterns (duodenum, pancreas) as additional controls. The low recombination efficiency in Bmi1-Cre ER transgenic mice might due to Bmi1 promoter activity. Nevertheless, Bmi1+ and YFP+ cells showed many similarities in cell lineage composition and other characteristics. For example, the ratio of Bmi1+ luminal to basal cells was similar to that of YFP+ luminal to basal cells in the adult intact and regressed prostates (Supplementary Table 1 and 2). Moreover, we present new data showing that both Bmi1+ and YFP+ cells were resistant to apoptosis after castration ( Fig. 2i-n), which means that both Bmi1+ and YFP+ cells observed in the regressed prostate are castration-resistant cells. Finally, using either Bmi1 immunostaining or YFP expression to mark CARBs showed them to be distinct from CARNs in either castrated Bmi1-Cre ER ;R26R-YFP mice or castrated Nkx3.1Cre ERT2 ;R26R-YFP mice (Fig. 2h, i and Supplementary Fig. 3d-h). Altogether, these observations demonstrate that even though lineage marked YFP+ cells were detectable at low efficiency, this marking is most likely representative (page9, line4).

Concern 3. It is interesting that upon castration, the % of Bmi1+ cells, or CARBs, increased to ~0.7% of the total epithelial cells. Since Bmi1+ cells seem to be largely quiescent in intact prostate, where did this doubling of the Bmi1+ cells come from? From initially weakly/negatively marked YFP+ cells or from the duplication of YFP+ cells?
Response 3: We believe this observation is due to the higher relative resistance of CARBs to castration. To directly examine whether YFP+ cells identified in intact mice by treating BY mice with tamoxifen are castration resistant, we first treated BY mice by tamoxifen to label the cells then castrated the mice (Fig. 3j). After castration, the fraction of YFP+ cells increased from 0.3% to 0.64% (n=253 YFP+ cells counted from a total of 39281 cells from 4 mice; p<0.001; Fig. 3k). These results suggest that tamoxifen treatment of BY mice before castration labels castration-resistant cells. Since we observed an increased fraction of YFP+ cells after castration compared to that in intact mice, we further investigated whether YFP+ cells are resistant to apoptosis after castration. As it had been reported that the peak of epithelial apoptosis in the mouse prostate occurred by 3-4 days after castration, the fraction of apoptotic cells were evaluated at 3 days after castration of BY mice following tamoxifen treatment using cleaved caspase-3 immunodetection (Fig. 3j). Interestingly, no YFP+ or Bmi1+ cells co-expressing cleaved caspase-3 were observed (Fig. 3l-n; P<0.001, n= 242 YFP+ and 862 Bmi+ cells counted from 4 mice), demonstrating the relative castration resistance of YFP+ cells prospectively marked in intact adult animals.
Concern 4. The results about the cell cycle entry and exit ( Figure 5) are potentially interesting, but the overall data is relatively weak considering the low GFP+% and subtle differences in quantification data (Fig. 5d, 5e). Response 4: We have edited this section to avoid over-interpretation of the data.
Concern 5. After castration the majority (>95%) of Bmi1+ cells are luminal, but the overall Bmi1-GFP+ cells are few (0.7%). The authors assumed that the Cre-mediated PTEN deletion primarily occurred in luminal cells, but at least theoretically Cre could also be activated in basal cells, which could lead to the same observations. Response 5: Although Pten was deleted in some Bmi1-expressing basal cells, the resulting p-Akt+ basal cells were detected specifically only in normal looking glandular structures ( Supplementary Fig.  4h), suggesting Pten deletion in Bmi1+ basal cells does not efficiently initiate transformation. Nonetheless, we agree with the reviewer that alternative possibilities such as rapid luminal differentiation of Pten-deleted basal cells fated to develop PIN/cancer cannot be completely ruled out and it is best not to be dogmatic. We have made note of this caveat in our discussion of the data.
Concern 6. Supplementary Fig. 1i: It's particularly interesting that CARBs are distinct from CARNs. Since NKX3.1 is exclusively expressed in luminal cells, and the Bmi-1 are also mainly localized in the luminal layer, are they co-localized before castration? If yes, why do some cells express NKX3.1 while others express Bmi1 in regressed prostate? Response 6: This is an interesting question which however is difficult to answer at present. This is due to the fact that Nkx3.1 is expressed in virtually all luminal cells in intact adult mouse prostate, and there is no way at present to determine which of these cells are CARNs prior to castration. Thus before castration, Nkx3.1 and Bmi1 maybe co-localized in cells but these is no evidence these cells are stem/progenitor cells. After castration however, Nkx3.1 and Bmi1 are expressed in distinct luminal cells that are functionally prostate progenitor cells susceptible to tumor initiation.

Concern 7. What is the difference in tumor growth kinetics of Bmi1-PTEN tumor versus published NKX3.1-PTEN tumor?
Response 7: The NKX3.1-PTEN tumors are also deficient in Nkx3.1 tumor suppressor protein since the Nkx3.1-CreER allele inactivates the endogenous Nkx3.1 gene. By contrast, the Bmi1-CreER driver was made by insertion into the 3'UTR of the mouse Bmi1 gene and does not affect expression of the endogenous Bmi1 gene. Thus a comparison of the Bmi1-PTEN and NKX3.1-PTEN tumor growth kinetics is not straightforward. We are currently generating and analyzing mice with equivalent genetic changes (i.e. by introducing Nkx3.1 heterozygosity into the Bmi1-PTEN mice) to address this question. Similarly, we are generating similar mice to compare the molecular profiles of nontransformed CARBs and CARNs (i.e. Bmi1CreER;Nkx3.1+/-;R-YFP vs NkxCreER;R-YFP mice).