Short and long-term effects of CDK4/6 inhibition on early stage breast cancer

Abstract

CDK4/6 inhibitors are used in the treatment of advanced estrogen receptor (ER)(+) breast cancer (BC). Their efficacy in ER(−) and early stage BC is currently under investigation. Here, we show that palbociclib, a CDK4/6 inhibitor, can inhibit both progression of ductal carcinoma in situ (DCIS) and growth of invasive disease in both an ER(−) basal BC model (MCFDCIS) and an ER(+) luminal model (MCF7 intraductal injection). In MCFDCIS cells palbociclib repressed cell cycle gene expression, inhibited proliferation, induced senescence and normalized tumorspheres formed in Matrigel whilst the formation of acini by normal mammary epithelial cells (MCF10A) was not affected. Palbociclib treatment of mice with MCFDCIS tumors inhibited their malignant progression and reduced proliferation of invasive lesions. Transcriptomic analysis of the tumor and stromal cell compartments showed that cell cycle and senescence genes, and MUC16, an ovarian cancer biomarker gene, were repressed during treatment. Knockdown of MUC16 in MCFDCIS cells inhibited proliferation of invasive lesions but not progression of DCIS. After cessation of palbociclib treatment genes associated with differentiation, e.g. p63, inflammation, IFNγ response and antigen processing and presentation remained suppressed in the tumor and surrounding stroma. We conclude that palbociclib can prevent progression of DCIS and is anti-proliferative in ER(−) invasive disease mediated in part via MUC16. Lasting effects of CDK4/6 inhibition after drug withdrawal on differentiation and the immune response could impact the approach to treatment of early stage ER(−) breast cancer.

Introduction

Entry into the cell cycle is mediated by a restriction point controlled by the CDK4/6-RB axis, which protects against aberrant proliferation (1-3). Cancer cells frequently subvert this restriction point using a variety of signaling cascades that converge on enhanced CDK4/6 activity and inhibition of RB to facilitate cell cycle entry (4-6). Consequently, the ability to take therapeutic control of the restriction point to block proliferation has been long sought after.

Palbociclib is a second-generation CDK inhibitor, and unlike previous inhibitors in this class, palbociclib demonstrates high affinity to CDK4/6 with little influence on other CDK family members and low toxicity (7). CDK4/6 inhibitors, when used in conjunction with standard-of-care hormone therapy, provide substantial improvements in progression-free survival compared to hormone therapy alone in cases of advanced ER(+)/HER2(−) breast cancer (8-10). While approved use is currently limited to advanced ER(+) breast cancer, these drugs have the potential for widespread application due to the central nature of cell cycle escape in carcinogenesis (11). Clinical trials in a variety of other cancer types are underway such as squamous cell lung cancer (), pancreatic neuroendocrine tumors (), Oligodendroglioma and Oligoastrocytoma (), along with others.

Though efficacious in ER(+) breast cancer, the use of palbociclib in earlier stages of breast cancer and in ER(−) disease is currently under investigation. One lesion type that is frequently diagnosed alongside invasive disease is Ductal Carcinoma in Situ (DCIS). DCIS is considered an in-obligate precursor to Invasive Ductal Carcinoma (IDC) in that it has the potential to become aggressive and potentially life-threatening but will not progress to IDC with 100% penetrance (12-16). Here we examine the effects of inhibition of CDK4/6 on DCIS and early stage IDC. A challenge in these type of studies is that there are few models of early stage breast cancer progression (17,18). The best model of DCIS progression was derived from the MCF10A series of basal-like triple negative (ER(−), PR(−), HER2 non-amplified) cell lines known as MCFDCIS (19,20). These cells exhibit bi-polar progenitor properties and are able to give rise to luminal and myoepithelial cell populations in xenografts, forming mammary acinar structures that will progress through DCIS to IDC with a predictable time course (21,22).

Here we demonstrate the palbociclib is able to inhibit proliferation and induce senescence in both normal MCF10A and MCFDCIS cells mediated through the downregulation of cell cycle driving genes and senescence genes. In 3D culture systems palbociclib was able to normalize the architecture of MCFDCIS spheres but interestingly has no effect on normal MCF10A mammary acini. In vivo treatment of MCFDCIS tumors inhibited DCIS progression and reduced proliferation of invasive lesions. Palbociclib was also able to delay the invasive transition of a luminal model of DCIS, the Mouse Mammary Intraductal (MIND) model, in which MCF7 cells are injected intraductally. The inhibition of proliferation of invasive lesions was in part mediated by MUC16, a protein that also serves as an ovarian cancer biomarker. Drug withdrawal experiments demonstrate that palbociclib is predominantly cytostatic in MCFDCIS although lasting effects of CDK4/6 inhibition on differentiation and the immune response after drug withdrawal is of potential interest for rational design of therapeutic strategies targeting early stage ER(−) breast cancer.

Materials and Methods

Cell Culture

MCF10A and MCFDCIS cell lines were obtained from Dr. Susette Mueller (Georgetown University) and were cultured as previously described (22,23). HEK293 cells were obtained from Dr. Rabindra Roy (Georgetown University) and were cultured in Dulbecco’s modified Eagle’s medium (DMEM) (GIBCO/ Invitrogen, Carlsbad, CA, USA) with 10% fetal bovine serum. All cell lines were fingerprinted to confirm identity and Mycoplasma tested regularly (most recent test 6/2018). Cells were used between passages 4 and 15. The FOXM1-12D phosphor-mimic plasmid was obtained from Dr. Peter Sicinski (Dana Farber) (24) and was used to transfect HEK293 cells alongside an empty-vector control using the Fugene 6 reagent as recommended by the manufacturer (Promega, Fitchburg, Wisconsin, USA). 48 hours after transfection, G418 selection agent was added to cells at 800ug/mL and maintained for 14 days generate stable HEK-12D and HEK-EV cell lines. shMUC16 MCFDCIS lines were established using unique shRNA and non-silencing control sequences inserted into the pGIPz (Dharmacon) as previously described (25).

In Vitro Drug Sensitivity Assays

Effects on cell phenotype were assessed after cells were treated for 72 hours with various concentrations of palbociclib (PALBOCICLIB-0332991, Taizhou Crene Biotechnology, Taizhou, Zhejiang, China—NMR was performed by Dr. Milton Brown at Georgetown University to confirm compound structure prior to use). Treatment was initiated 24 hours after cells were seeded. Changes in cell morphology were captured with an Olympus IX-71 inverted epifluorescence microscope. Cell proliferation was assessed via crystal violet as previously described (25). Total protein and RNA were extracted from cells for analysis via Western Blot and qRT-PCR (described below). Cell cycle analysis was done on one million cells per condition using the Vindelov method (26). Apoptosis was assessed in an aliquot of the cell cycle cell sample suspension via an Annexin V staining and flow cytometry.

Western Blot

Cells were cold lysed in an NP40-based lysis buffer while frozen tumor samples were homogenized in lysis buffer using the Roche MagNA Lyser instrument (Roche). Following protein extraction, cellular and tumor lysates were prepared and separated on an SDS-PAGE gel prior transfer and immunoblotting as described previously (25). Information on primary and secondary antibodies used can be found in Supplemental File 1, Table 1.

qRT-PCR

Total RNA was extracted from cells using the RNeasy kit (Qiagen). RNA was extracted from tissues using the RNeasy kit as well but following a homogenization step using the MagNA Lyser. qRT-PCR was performed on extracted RNA as previously described (27). Primers were purchased from Integrated DNA technologies (Coralville, IA, USA), sequences can be found in Supplemental File 1, Table 2.

Senescence Assays

To assess the senescence related phenotype of increased cell size, cells were treated for 72 hours with different concentrations of palbociclib prior to collection and measurement of cell diameter utilizing a Multisizer3 Coulter Counter. After 72 hours of treatment with either vehicle or a dose of palbociclib, cells were stained for the senescence marker β-Galactosidase (Cell Signaling Technologies: 9860) following manufacturer’s instructions. Stain was allowed to develop for 48 hours. Images were acquired using the Olympus IX-71 microscope as described above, and the number of positive cells per field was quantified using the cell counting feature in ImageJ.

3D Culture

Matrigel sphere formation assay was conducted as previously described (23,28) with a modification to the protocol to add 100nm palbociclib or sterile water control to the assay media in which the tumor spheres were seeded. Spheres were photographed at days 7 and 10 for MCF10A and MCFDCIS respectively, using an Olympus IX-71 Inverted epifluorescence microscope. Sphere sizes were quantified using Fiji software. Sphere isolation and immunofluorescent staining were also performed as previously described (23) using Laminin 5 (details in supplemental file one, table one) and mounted with ProLong Gold Antifade Mountant with DAPI (Thermo Fisher, Waltham, MA, USA). Images were acquired using the Leica SP8 Confocal microscope and fluorescent intensity was quantified using Fiji software. To isolate RNA from spheres, Matrigel was dissolved by adding dispase 5U/mL (Stem Cell Technologies, Vancouver, Canada) to chamber slides and incubating at 37°C for 45 minutes. Solubilized tumorspheres were collected, spun and washed with sterile PBS prior to RNA extraction using the Qiagen RNeasy Kit (Qiagen).

Animal Models

Subcutaneous models: Six-week-old Athymic nude mice were purchased from Envigo. MIND Model: six-week-old female NOD-SCID mice purchased from Jackson Labs Animals were housed in a pathogen-free environment with controlled temperature and humidity. All animal experiments were conducted in accordance with protocols approved by the Georgetown University IACUC.

Xenograft Experiments

For subcutaneous xenograft experiments, MCFDCIS cells were injected as previously described (22). Animals received sterile water or 50mg/kg palbociclib diluted in a water-based vehicle containing 0.5% methylcellulose and 0.2% Tween 80 for one or two weeks total depending on cohort, as outlined in figure 3A and 5A. Upon tumor collection, samples were divided for Hematoxylin and Eosin (H&E) and Immunohistochemistry (IHC) (below), qRT-PCR (Methods, above) and western blot (Methods, above). Blood was collected to assess the impact of treatment on white blood cell counts using a Giemsa Stain Solution (Sigma-Aldrich, St. Louis, MO, USA) according to manufacturer’s instructions. shMUC16 MCFDCIS cells were injected as previously described and tracked as outlined in supplemental figure 5A. Serum was taken from animals for circulating Muc16 expression using an ELISA for detection of the human protein following manufacturer’s instructions (R&D Systems). For the Mammary Intraductal Method (MIND model) MCF7-luc2-DSred cells (kindly donated by Dr. Cathryn Briskin—Swiss Federal Institute of Technology in Lausanne) were prepared and injected into 6-week-old female NOD-SCID mice purchased from Jackson Labs as described previously (29), outlined in supplemental figure 4A. Mammary glands were collected and H&E stained.

Figure 3. Palbociclib (PD) Treatment in vivo. PD delays growth and progression of MCFDCIS xenograft tumors in nude mice.

A) Schematic of experimental design and time points. B) Tumor growth curves for vehicle- or PD-treated MCFDCIS tumors following subcutaneous injection of cells, represented as the change (delta) in tumor size normalized to initial tumor size (Two-Way ANOVA and Bonferroni post-tests, * = P<0.05, ** = P<0.01, *** = P<0.001, **** = P<0.0001 relative to vehicle). C) Cohort 2 tumor sections stained with hematoxylin and eosin as well as IHC for phosphorylated RB (S807/811), total RB, Ki67, and p63, treated with PD as indicated and associated quantifications. (Chi-Square Tests, **** = P<0.0001, 20x mag., scale = 100μm). D) qPCR analysis of cell cycle regulated mRNA targets in MCFDCIS tumors treated with PD as indicated. Mean fold change ± S.E.M. (Student’s T-test, ** = P<0.01, *** = P<0.001, **** = P<0.0001 relative to vehicle). E) Western blot for expression of FOXM1 and actin in tumor lysates at both one and two weeks of treatment with PD as indicated. F) qPCR analysis of senescence regulated mRNA targets in MCFDCIS tumors treated with PD as indicated. Mean fold change ± S.E.M. (Student’s T-test, **** = P<0.0001 relative to vehicle).

Figure 5. Palbociclib (PD) treatment in vivo is reversible. Most, but not all of PD’s effects on tumor growth and progression are reversible upon discontinuation of drug treatment.

A) Schematic of experimental design, time points and comparisons made using RNA isolated from tumors. B) Tumor growth curves for vehicle and PD-treated MCFDCIS tumors, represented as the change (delta) in tumor size normalized to initial day of treatment (Two-Way ANOVA and Bonferroni post-tests, * = P<0.05, ** = P<0.01, *** = P<0.001, **** = P<0.0001 relative to vehicle). C) Cohort 4 tumor sections stained with hematoxylin and eosin as well as IHC for phosphorylated RB (S807/811), total RB, Ki67, and p63, treated with PD as indicated and associated quantifications. (Chi-Squared Tests, * = P<0.05, *** = P<0.001, **** = P<0.0001, 20x mag., scale =100μm). D) Fold change values of the cell cycle gene panel in palbociclib-treated (cohort 3) and -recovered (cohort 4) tumors relative to a vehicle baseline, represented by the dotted line. E) Fold change values of genes significantly upregulated during treatment (palbo) and their expression following recovery (palbo-recovered) relative to a vehicle baseline indicated by the dotted line. F) Fold change values of genes significantly downregulated during treatment (palbo) and their expression following recovery (palbo-recovered) relative to a vehicle baseline indicated by the dotted line. D-F are based on values derived from RNA-seq. All values were filtered for minimum expression (CPM>2) and significance (P<0.05). G) ELISA for circulating MCFDCIS tumor-derived MUC16 in serum taken from cohort 4 mice.

Histological Analysis

H&E and IHC analyses were performed on paraffin-embedded 5-μm sections using standard protocols described elsewhere (30). Information on primary antibodies (pRB, RB, Ki67, p63, Muc16 and cleaved Caspase 3) for IHC can be found in Supplemental File 1, Table 1. Images were captured using an Olympus BX40 microscope and quantification of histological areas and staining positivity were performed using ImageJ.

cDNA Array and RNAseq Analysis

Total tumor RNA was extracted in triplicate, as described above. RNA with an Integrity Number greater than 8.7 were sent to the UCLA Neuroscience Genomics Core (UNGC). Microarray analysis which was carried out by the UNGC using Illumina HumanHT-12 v4 Expression BeadChip representing 47,000 well-annotated genes according to the manufacturer’s instructions on RNA isolated from the xenograft described in figure 3A (Illumina, Inc., San Diego, CA). Data were analyzed in R using the Bioconductor software. Greater than two-fold difference in expression and a P-value <0.05 were set as the threshold for significance. Differential gene expression lists were generated by comparing palbociclib-treated tumor expression to vehicle-treated tumor expression and ranking genes by log2 fold change. RNA from the xenograft described in figure 5A was used for RNA library construction and sequencing (RNAseq). Mouse genome GRCm38.p6 and human genome GRCh38.p10 were concatenated into a meta-genome. A STAR index was generated using GTF annotation from GENCODE for mouse and human genome features (version v18 and v27, respectively). Samples were aligned using single sample two-pass alignment, limiting multi-mapping reads to not more than one unique location. Unique alignment rates were 80-85%, with 10-15% reads mapping to more than once location. Only unique alignments were used to quantify expression at mouse and human genome features. Feature counts were used in downstream comparisons. Data available on GEO: GSE130903.

Statistics

RStudio and Prism 7 (Graphpad Inc) were used for statistical analysis and graphing. Analysis of Variance was used for multiple comparisons and t-tests or chi-square tests were used for paired comparisons, with P<0.05 as the threshold for statistical significance in all tests.

Results

Effects of palbociclib on normal breast and DCIS cells in vitro

We compared the effects of palbociclib on MCF10A, an immortalized normal breast epithelial cell line, and MCFDCIS cells, a DCIS model cell line (20). Palbociclib, and a related CDK4/6 inhibitor abemaciclib, inhibited cellular proliferation in a concentration-dependent manner with similar IC₅₀ in MCF10A and MCFDCIS (Supplemental Figure 1A & B), accompanied by diminished phosphorylation of the canonical CDK4/6 target RB (Supplemental Figure 1C) and reduced expression of known cell cycle target genes (Figure 1A) (20,31,32). Palbociclib-treated cells accumulated in G1 with a reduction in both G2 and S phase (Supplemental Figure 1D).

Figure 1. Palbociclib (PD) treatment in 2D. PD elicits canonical anti-proliferation and senescence responses in both MCF10A and MCFDCIS that are reversible upon discontinuation of treatment.

A) qPCR analysis of cell cycle regulated mRNA targets in MCF10A (left panel) and MCFDCIS cells (right panel), treated with PD as indicated. Mean fold change ± S.E.M. (one-way ANOVA followed by Dunnett’s multiple comparisons test, relative to vehicle). B) Quantification of Beta Galactosidase-stained MCF10A (left panel) and MCFDCIS (right panel) cells. Mean ± S.E.M. (one-way ANOVA followed by Dunnett’s multiple comparisons test, relative to vehicle, 72-hour treatment). C) qPCR analysis of Lamin B1 (LMNB1) expression in MCF10A (left panel) and MCFDCIS (right panel) treated as indicated. Mean fold change ± S.E.M. (one-way ANOVA followed by Dunnett’s multiple comparisons test, relative to vehicle, 72h treatment). D) qPCR analysis of FOXM1 expression in MCF10A (left panel) and MCFDCIS (right panel) treated as indicated. Mean fold change ± S.E.M. (one-way ANOVA followed by Dunnett’s multiple comparisons test, relative to vehicle, 72h treatment). E) Crystal violet proliferation curves for MCFDCIS cells in vitro treated for 6 days with vehicle (constant vehicle), 6 days with 0.1μM palbociclib (constant palbociclib) or 3 days with 0.1μM palbociclib followed by 3 days with vehicle (Reversal). F) qPCR analysis of cell cycle-regulated mRNA targets in MCFDCIS cells treated with PD as indicated. Mean fold change ± S.E.M. (Student’s T-test, relative to vehicle). G) Cell cycle analysis of MCFDCIS cells treated with PD as indicated. H) Annexin-5 apoptosis assay on MCFDCIS cells treated as indicated; all apoptotic death is represented as a fold change from the normalized vehicle-treated baseline. I) Quantification of Beta Galactosidase-stained MCFDCIS (left panel) cells vitro treated with PD as indicated. Mean ± S.E.M. (Student’s T-test, relative to vehicle). J) qPCR analysis of LMNB1 (left panel) and FOXM1 (right panel) expression in MCFDCIS treated as indicated. Mean fold change ± S.E.M. (Student’s T-test, relative to vehicle). K) Quantification of PD senescence challenge in HEK293 cells transfected with constitutively active FOXM1-12D or empty vector. HEK293-EV and HEK293-12D were treated with indicated concentrations of PD for 72 hours prior to staining for Beta Galactosidase activity. Mean ± S.E.M. (Student’s T-test, relative to vehicle). For all: * = P<0.05, ** = P<0.01, *** = P<0.001, **** = P<0.0001.

CDK4/6 inhibition can also induce cellular senescence (Supplemental Figure 1E), which likely contributes to the observed increased cell size in dose-response assays and the measured concentration-dependent increases in median cell diameter (Supplemental Figure 1F). We also observed a dose-dependent accumulation of β-galactosidase positivity in both MCF10A and MCFDCIS cells with >20% stained for the senescence marker at IC₅₀ concentrations of palbociclib (Figure 1B & Supplemental Figure 1G) and significantly reduced expression of an early senescence response gene, nuclear envelope factor Lamin B1 (LMNB1) (33) in MCFDCIS cells (Figure 1C). Notably, palbociclib was able to significantly reduce phosphorylation and expression of FOXM1 a CDK4/6 target involved in the senescence response (24) (Supplemental Figure 1H, Figure 1D). These data indicate that palbociclib inhibits proliferation by regulating cell cycle machinery and also induces senescence in both normal mammary cells and early stage breast cancer.

To determine the durability of palbociclib response, we designed reversal experiments in which MCFDCIS cells were treated with palbociclib for 72 hours followed by a recovery period of an additional 72 hours. In reversal experiments, cell proliferation rate did not recover to vehicle baseline following discontinuation of treatment despite the recovery of cell cycle genes to pre-treatment expression levels, with the exception of MCM2 which remained significantly downregulated (Figure 1E & F). Consistent with this, we found that G1 arrest is transient and dependent on the continued presence of drug (Figure 1G). However, recovered cells experienced 2.5-fold higher levels of apoptosis than control cells which could explain the lag in proliferative rate recovery (Figure 1E & H). In terms of senescence during recovery, β-Galactosidase staining returned to baseline, as did expression of LMNB1 although there was a significant induction of FOXM1 following recovery (Figure 1I & J). This suggested that palbociclib might influence both the proliferative and senescent responses through effects on FOXM1. To investigate this, we over-expressed, in HEK293 cells, a constitutively active FOXM1 “phospho-mimic” in which 12 CDK4/6 consensus sites have been altered to aspartic acid residues (24) (Supplemental Figure 1I). Expression of the FOXM1-12D did not alter the palbociclib IC₅₀ for proliferation (Supplemental Figure 1J) but did significantly inhibit palbociclib-induced senescence as measured by β-Galactosidase staining (Figure 1K). FOXM1-12D also attenuated the repression of several cell cycle genes including CCNB1 and KIF20A (Supplemental Figure 1K). This suggests that FOXM1 may play an important role in the senescence response to CDK4/6 inhibition in the context of early breast cancer.

Effects of palbociclib on MCFDCIS cells in 3D culture

The effects of palbociclib on normal mammary cells and early breast cancer could be modified by their stromal environment. To investigate this, we examined palbociclib effects on both MCF10A and MCFDCIS cells grown on top of Matrigel and found that MCF10A cells form normal mammospheres whereas MCFDCIS cells form abnormal tumorspheres (22,23,28). The MCF10A mammosphere phenotype is largely unchanged over 7-10 days of treatment in the presence of palbociclib whereas MCFDCIS tumorspheres are significantly smaller with a more normalized phenotype when compared to vehicle controls (Supplemental Figure 2, Figure 2A & B). MCFDCIS but not MCF10A spheres showed significant reductions in the cell cycle gene genes as well as the senescence marker, LMNB1 (Figure 2C). Laminin 5, a basement membrane marker (23), is deposited at the periphery of MCF10A mammospheres and remains unaltered after palbociclib treatment (Figure 2D & E). In contrast, Laminin 5 shows a disorganized expression across MCFDCIS tumorspheres, but reverted to a normal mammosphere pattern after palbociclib treatment (Figure 2D & E). Together these results indicate potential selectivity for highly proliferative transformed cells in the 3D context as differentiated MCF10A mammospheres were unaffected by palbociclib.

Figure 2. Palbociclib (PD) treatment in 3D. PD selectively influences MCFDCIS spheres in 3D culture, normalizing their architecture.

A) MCF10A (top panel) and MCFDCIS (bottom panel) spheres grown in 3D in the presence of vehicle or 0.1μM PD. 20x mag., scale = 200μm. B) Quantification of A; sphere areas calculated using Fiji software. Mean ± S.E.M. (Student’s T-Test, **** = P<0.0001 relative to vehicle). C) qPCR analysis of cell cycle/senescence-regulated mRNA targets in MCF10A (upper panel) and MCFDCIS (lower panel) spheres treated with PD as indicated. Mean fold change ± S.E.M. (Student’s T-test, ** = P<0.01, *** = P<0.001, **** = P<0.0001 relative to vehicle). D) Vehicle-treated MCF10A (top panel) and MCFDCIS (bottom panel) spheres stained for Laminin 5 (red) and DAPI (blue) along with quantification of the distribution of Laminin 5 signal intensity measured along the yellow line indicated in the photos. 63x mag., scale = 25μm. E) PD-treated MCF10A (top panel) and MCFDCIS (bottom panel) spheres stained for Laminin 5 (red) and DAPI (blue) along with quantification of the distribution of Laminin 5 signal intensity measured along the yellow line indicated. 63x mag., scale = 25μm.

Palbociclib treatment reduces MCFDCIS xenograft growth and malignant progression

The selective “normalization” effects of palbociclib on MCFDCIS cells in Matrigel suggested that this CDK4/6 inhibitor could have beneficial effects on DCIS lesions in vivo. Therefore, we tested palbociclib efficacy over 2 weeks in vivo on xenograft tumors in athymic nude mice (Figure 3A). We used a low dose of 50mg/kg relative to previous preclinical studies (34-39) to minimize potential toxicity such as neutropenia, which is a common adverse event associated with palbociclib (8). We saw a small, non-significant reduction in white blood cell counts following two weeks of treatment relative to control (Supplemental Figure 3A) and observed no difference in animal weight between groups over the course of our study indicating that this dose of palbociclib was well tolerated (Supplemental Figure 3B). Tumors showed a significantly reduced growth rate after palbociclib leading to stasis in some cases but not to regression (Figure 3B).

The MCFDCIS model is unique in that xenograft tumors transition from in situ carcinoma to invasive cancer between 4 to 6 weeks after injection (21). Thus we were able to monitor the impact of palbociclib on malignant progression of MCFDCIS lesions and found a significant delay of progression in palbociclib-treated tumors. At 4 weeks, the tumor area of about half of the control group showed invasive lesions, whereas about 90% in the palbociclib-treated group remained non-invasive DCIS (Figure 3C, first panel). Palbociclib treatment resulted in significantly diminished expression of phosphorylated RB without significantly impacting total RB staining by IHC, though western blot demonstrates a reduction in treated tumor lysates (Figure 3C, Supplemental Figure 3C). Palbociclib treatment also significantly reduced Ki67 proliferative index from >25% of nuclei stained to <10% within the tumor (Figure 3C). p63 expression, which marks the invasive transition in the MCFDCIS model (21), was also significantly reduced, (Figure 3C) corroborating the above conclusion from the surgical pathology analysis. Despite the smaller size of tumors and reduced proliferation, Caspase 3 cleavage was unaltered within the tumors, which suggests that palbociclib does not induce apoptosis in this context (Supplemental Figure 3D). As expected, palbociclib-responsive cell cycle genes also showed significant mRNA downregulation in tumors (Figure 3D) and FOXM1 expression was reduced at both the protein and mRNA levels, mirroring in vitro assays (Figure 3E & F). This observation, along with significant loss of LMNB1 suggests that the DCIS tumors are undergoing both proliferative arrest and senescence induction following two weeks of palbociclib treatment (Figure 3E & F).

Effect of palbociclib in the Mammary Intraductal (MIND) model of DCIS

To determine if palbociclib effects on DCIS tumors were dependent on the tumor microenvironment we utilized the MIND model, in which tumor cells are directly injected into the mammary duct. In this model, early stage DCIS-like breast cancer lesions develop after injection of MCF7 luminal ER(+) breast cancer cells (29) (Supplemental Figure 4A). Following injection of MCF7-luc2-DSred cells (kindly donated by Dr Cathryn Briskin—Swiss Federal Institute of Technology in Lausanne) into mammary ducts, successful injection was verified by IVIS imaging (Supplemental Figure 4B). DCIS lesions were established over 2 months prior to treatment with palbociclib for 28 days by oral gavage. IVIS imaging and histology of the mammary duct lesions revealed that palbociclib-treated mice had reduced tumor burden and bore fewer invasive lesions than their vehicle-treated counterparts (Supplemental Figure 4C-E). Palbociclib is therefore effective in postponing the invasive transition not only in basal-like DCIS but also in hormone-dependent luminal DCIS.

Role of MUC16 in the palbociclib response

To assess global gene expression changes in DCIS following two weeks of palbociclib treatment, mRNA isolated from MCFDCIS tumors was analyzed using Illumina bead chip array. Overall, 1515 genes were found to be significantly (P<0.05) up- or down-regulated by at least 1.5-fold at two weeks on treatment with the majority of highly regulated genes playing a role in the cell cycle (Figure 4A & B). MUC16 (CA-125), a cancer antigen that has been used as a biomarker and therapy response marker in ovarian cancer (40-43), stood out in this analysis as it is not a cell cycle regulatory gene. We verified by PCR that significant reductions in MUC16 mRNA occurred after one and two weeks of treatment and this was validated at the protein level by IHC (Figure 4C & D). MUC16 expression appears to increase with MCFDCIS tumor progression as 2-week vehicle-treated tumor sections stained more intensely than 1-week treated (Figure 4D). To evaluate a functional role of MUC16 in MCFDCIS tumor growth and progression we generated three independent shMUC16 MCFDCIS cell lines along with a non-silencing control. Subcutaneous tumor growth of the shMUC16 cell lines (Supplemental Figure 5A) was reduced relative to non-silenced control tumors with both shMUC16_2 and shMUC16_3 tumors demonstrating significantly reduced growth (Figure 4E). Despite some differences in growth rate, all shMUC16 MCFDCIS tumors had significantly less MUC16 expression than non-silenced controls (Figure 4F & G). Tumor-derived circulating MUC16 in serum was easily detectable in the controls but below detection threshold in the majority of shMUC16 tumors, with blood-borne expression correlating tightly with tumor growth rate across independent shMUC16 lines (Figure 4H). Though shMUC16 tumors grew more slowly than their control counterparts, loss of MUC16 did not delay the transition from DCIS to invasive lesions and tumor histology was indistinguishable between the groups (Figure 4I). This suggests that MUC16 does not drive MCFDCIS to invasive transition but does contribute to overall tumor growth. Of note is that shMUC16 tumors demonstrate more widespread staining for cleaved Caspase 3 suggesting that the reduction in tumor size relative to control tumors is due to an increased rate of cell death (Supplemental Figure 5B).

Figure 4. MUC16’s role in Palbociclib (PD) response. MUC16 can serve as a biomarker of palbociclib treatment in MCFDCIS tumors and may play a functional role in tumor growth.

A) Cigar plot of Log2 expression of cDNA array data generated from RNA isolated from vehicle- and palbociclib-treated tumors of cohort two. Points highlighted in red are up-regulated by palbociclib while those in blue are downregulated by drug. B) Top 15 most PD-regulated genes from the cDNA array at two weeks of treatment, ranked by Log2 fold change and significance of P<0.05. Up-regulated genes are depicted in red, down-regulated genes in blue. C) qPCR analysis of MUC16 expression in MCFDCIS tumors treated as indicated. Mean fold change ± S.E.M. (Student’s T-test, *** = P<0.001, **** = P<0.0001 relative to vehicle). D) Representative cohort 1 and 2 tumor sections stained by IHC for MUC16, treated with PD as indicated. 10x mag., scale = 200μm. E) Tumor growth curves for non-silenced (NS) or shMUC16 MCFDCIS tumors following subcutaneous injection of cells, represented as the delta in tumor size normalized to initial tumor size (Two-Way ANOVA and Bonferroni post-tests, * = P<0.05, ** = P<0.01 relative to vehicle). F) Representative NS and shMUC16 MCFDCIS tumor sections stained by IHC for MUC16. 10x mag., scale = 200μm. G) Quantification of MUC16 IHC staining as a percent of tumor area. Mean ± S.E.M. (one-way ANOVA followed by Dunnett’s multiple comparisons test, *** = P<0.001, relative to vehicle). H) ELISA for circulating MCFDCIS tumor-derived MUC16 in serum taken at day 28 from mice bearing NS or shMUC16 MCFDCIS tumors. I) Representative NS and shMUC16 MCFDCIS tumor sections stained using the PAS method. 10x mag., scale = 200μm.

Residual effects after cessation of treatment with palbociclib

Palbociclib’s effects in vitro are largely but not entirely reversible as shown above (see Figure 1E & F). Thus, we sought to determine if similar residual effects of CDK4/6 inhibition could be detected in vivo in either the tumor cells or stroma after discontinuation of therapy. To study this, we allowed mice an 11-day drug-recovery period following 2 weeks of palbociclib treatment (Figure 5A). Upon treatment termination, palbociclib-treated tumors resumed growth but final tumor sizes in the palbociclib recovery group remained significantly smaller than vehicle-treated tumors (Figure 5B). Growth rates of tumors during the 11 days of recovery paralleled those of the vehicle-treated tumors (Figure 5B); notably, in tumors recovering from palbociclib treatment, almost 30% of tumor area remained DCIS while vehicle-treated tumors progressed to invasive disease with only 8.5% of tumor area remaining DCIS (Figure 5C). We observed partial re-expression of phosphorylated RB, Ki67 and p63 upon palbociclib withdrawal (Figure 5C). p63 expression remained lower in palbociclib-treated and recovered tumors than in vehicle-treated tumors (Figures 5C & 3C), suggesting that palbociclib has some lasting effects on tumor differentiation status and progression to invasive lesions. Interestingly, the panel of cell cycle genes completely returned to baseline as did the senescence markers LMNB1 and FOXM1 (Figure 5D).

To better understand lasting versus reversible effects of CDK4/6 inhibition we performed RNA-seq on RNA isolated from treated and recovered MCFDCIS tumors. We confirmed that the majority of regulated genes are involved in cell cycle regulation and that nearly all of them recover to a baseline level of expression following treatment cessation (Supplemental Figure 6A). Interestingly, several genes uninvolved with the cell cycle broke this pattern. Some genes were upregulated >1.5-fold by palbociclib and remained upregulated following drug withdrawal such as the differentiation factor NELL2, which experienced even further upregulation in recovery (Figure 5E). High expression of NELL2 is associated with improved recurrence-free survival in basal breast cancers (Supplemental Figure 6B). Other factors initially upregulated by palbociclib were significantly downregulated following palbociclib withdrawal (Figure 5E) and some genes that experienced significant downregulation during palbociclib treatment did not recover to baseline following discontinuation of treatment, including several RNA processing factors, matrix metalloproteinases, and inflammatory factors (Figure 5F). In the RNA seq analysis we also saw that MUC16 was significantly downregulated by treatment; its mRNA levels increased somewhat following drug withdrawal but did not return to baseline. Consistent with this, MUC16 protein in the circulation remained significantly reduced even after treatment cessation (Figure 5G).

Palbociclib has residual effects on stromal immune and interferon related genes.

To gain insight into the effects of palbociclib on the tumor host, we simultaneously aligned the RNAseq reads from tumors to both the human and mouse genomes (see Methods). Mouse stromal cells contributed 15-25% of the mRNA isolated from tumor samples that were harvested at different points during treatment (Figure 6A & B). The cell cycle and senescence gene panel regulated in tumor cells was also impacted in the mouse stroma, though the downregulation of these genes was smaller than seen in the tumor cell compartment. Following recovery from treatment, several genes exceeded the expression observed at baseline (Figure 6C). Using the same method described above to identify lasting influence of human tumor genes, we identified mouse stromal genes that experienced a lasting impact of palbociclib (Figure 6D & E). Interestingly, the genes that showed significant downregulation during palbociclib treatment that persists after treatment cessation are almost entirely immune-related, specifically those related to interferon signaling with roles in transcriptional regulation and antigen processing and presentation (Figure 6E). Xenografts of human cells do impact stromal gene expression and signaling in the host, but the repression of these immunity-related genes was only noted in palbociclib-treated animals indicating it is a drug-specific effect. Quite strikingly, reduced expression of these genes is associated with a highly significant reduction in recurrence-free survival of patients with basal breast cancers (Supplemental Figure 7A).

Figure 6. Palbociclib (PD)’s influence on the tumor microenvironment. PD impacts gene expression in mouse stromal cells with a residual effect after treatment cessation on immune-related signaling pathways.

A) Percent of RNA-seq reads for each tumor that aligned to the murine or human genome. Plotted as mean ± S.E.M. with individual tumors plotted as separate symbols. B) Schematic of experimental design, time points and comparisons made using mouse RNA isolated from tumors. C) Fold change values of the cell cycle gene panel in palbociclib-treated (cohort 3) and palbociclib-recovered (cohort 4) treated tumor stroma relative to a vehicle baseline, represented by the dotted line. D) Fold change values of genes significantly upregulated during treatment (palbo) and their expression following recovery (palbo-recovered) relative to a vehicle baseline indicated by the dotted line. E) Fold change values of genes significantly downregulated during treatment (palbo) and their expression following recovery (palbo-recovered) relative to a vehicle baseline indicated by the dotted line. C-E are based on values derived from RNA-seq. All values were filtered for minimum expression (CPM>2) and significance (P<0.05). F) Normalized enrichment scores for hallmark pathways identified in both the human tumor and mouse stromal RNA, the dotted line indicates a significance cut off based on P values < 0.05 and FDR Q values <0.25.

Gene Set Enrichment Analysis (GSEA) revealed that similar hallmark pathways were significantly regulated by palbociclib both in the human tumor and the surrounding stroma (Figure 6F). While both the tumor and stroma experience a downregulation of cell cycle driver pathways including E2F Targets, G2M Checkpoint and the Mitotic Spindle, the stroma is less impacted and better able to recover following treatment cessation, as indicated by upregulation of these pathways (Figure 6F). However, unlike the cell cycle pathways, a more dramatic regulation of hallmark inflammatory and interferon gamma signaling pathways is seen in the stroma compared to the tumor cell compartment. These pathways remain significantly downregulated in the tumor microenvironment even after cessation of palbociclib treatment, reflecting a lasting gene signature in the mouse stroma (Figure 6E & F).

Discussion

In this study we show that CDK4/6 inhibition delays malignant progression to invasive disease in both basal MCFDCIS subcutaneous xenograft and luminal MCF7 intraductal injection models of DCIS. Even after cessation of palbociclib treatment, the residual effects of the drug result in less aggressive tumors and reduced overall recurrence. In vitro, palbociclib caused canonical cell cycle arrest in normal and transformed breast cells and in parallel also displayed classic features of senescence with no discernible increase in apoptosis. In 3D assays, which are more faithful predictors of cell behavior and drug responsiveness in vivo (44), there was differential sensitivity of MCF10A and MCFDCIS spheres to palbociclib. This is likely explained by the fact that MCF10A cells grown as spheres undergo a well-described process of differentiation, quickly enter a maintenance phase marked by long term quiescence (23), and are relatively insensitive to the effects of CDK4/6 inhibition. Conversely, MCFDCIS spheres do not experience differentiation or quiescence as they grow in 3D; palbociclib is able to significantly reduce expression of E2F target genes while simultaneously inducing senescence as evidenced by loss of LMNB1. Consistent with these observations, palbociclib treatment had no discernable effects on normal, differentiated mammary gland architecture over the course of therapy in both subcutaneous xenografts and in MIND model experiments. Cell cycle inhibition and G1 arrest following palbociclib is mainly cytostatic as the majority of MCFDCIS tumor cells were able to re-express cell cycle driving genes and reinitiate growth following treatment cessation, similar to other studies (34,37,45).

To our knowledge, this is the first interrogation of early stage breast cancer cells in parallel with the immediate mouse stromal cell transcriptomes during palbociclib treatment and after recovery from treatment. The majority of the 97 genes that recover expression after cessation of palbociclib treatment contribute to hallmark pathways related to the cell cycle. The tight regulation of cell cycle genes on and after treatment highlights the specificity of CDK4/6 inhibition in targeting proliferation. Many genes identified in our signature overlap with those previously identified as responding to CDK4/6 inhibition (35). This work suggests that there is a core set of targets that are influenced by palbociclib treatment across different contexts that may be useful indicators when assessing drug responsiveness in patients. The mouse stroma also experienced similar changes in canonical cell cycle driver genes but to a lesser magnitude, accompanied by a more complete expression recovery following treatment cessation. Stromal tissue is not as proliferative as the tumor, which likely provides some resistance to palbociclib’s primary effects.

The discovery that a non-cell cycle regulatory gene, the cancer specific antigen MUC16 (also known as CA-125), is one of the most significantly down-regulated genes by palbociclib is surprising. MUC16 is well-known in the ovarian cancer field where it has been used clinically as a biomarker of disease in tracking the efficacy of therapy (40-42). Interestingly, it has been reported that ovarian cancer and basal breast cancer have significant genomic similarities, and MUC16 is frequently mutated in breast cancer cases (46). In our model, MUC16 is not only a biomarker of palbociclib responsiveness but also contributes to the growth of invasive lesions. MUC16 has been linked to cancer progression and aggressive behaviors in both ovarian and breast cancers (43) although the mechanism MUC16 utilizes to increase proliferation of cancer cells is not clear. Nevertheless, it appears that CDK4/6-mediated repression of MUC16 is partially responsible for the efficacy of palbociclib in our MCFDCIS model. MUC16 has been previously implicated in triple negative breast cancer as a potential driver of progression and may play a key role in early disease (47). It will be interesting to determine if, in a subset of human basal breast cancers, MUC16 expression could indicate patient response to palbociclib therapy. It is of note that establishment of predictive biomarkers of response to CDK4/6 inhibitors in advanced stages of disease has been challenging with many parameters related to cell cycle such as Cyclin D1 or p16^INK4A expression levels demonstrating no correlation with response (48).

Other genes of interest revealed in the transcriptomic analyses of palbociclib-treated and recovered MCFDCIS tumors were those that did not return to baseline following cessation of treatment. It is notable that in both the human tumor cells and the surrounding mouse stroma the majority of these genes have no known role in cell cycle regulation. In the tumor cells, both NELL2 and KLK11 were upregulated by palbociclib that was further increased following termination of drug treatment, which may suggest that treatment selects for populations of tumor cells expressing high levels of these genes. KLK11 is a serine protease and NELL2 is a glycoprotein with several epidermal growth factor-like domains. While the former has no known role basal breast cancer survival, high expression of the NELL2 is associated with significantly prolonged recurrence-free survival in patients with basal breast cancers (Supplemental Figure 6B). Therefore, treatment with palbociclib could generate lasting benefits. Many genes that were downregulated within the tumor cells by palbociclib did not return to baseline levels after cessation of treatment are involved in processes such as extracellular reorganization, RNA processing and inflammatory signaling. The contributions of these genes to long term CDK4/6 inhibitor response is worthy of further study.

Genes with lasting downregulation in the mouse stroma were more homogeneous and comprised almost entirely of factors involved in interferon-mediated immune signaling. CDK4/6 inhibition and its relationship to immune regulation has been a topic of both interest and controversy, as groups have reported both pro- and anti-immune responses which may impact the overall outcome of treatment with this class of drugs (49-52). One proposed mechanism of anti-tumor immune regulation driven by CDK4/6 inhibition is the induction of an interferon-driven viral mimicry response initiated by cell-intrinsic factors, mediated through repression of the E2F target gene DMNT1 (50). However, in our study we did not observe significant DMNT1 repression in either the human tumor or the mouse stroma in response to palbociclib.

Furthermore, several of the genes involved in antigen processing and presentation such as Tap1, Ifit2, Oas2, Oasl2, and Stat1 that were previously reported to be up-regulated in tumor cells by CDK4/6 inhibition (50) actually showed lasting downregulation in the stromal analysis in our mouse model (Figure 6E). These differences could be due to the cell types examined, treatment duration or the breast cancer models used. Our xenograft experiments were performed in athymic nude mice that lack functional T cell populations and it is difficult to assess the full impact of palbociclib on anti-tumor immunity in this model. The suppression of antigen processing genes is associated with a significant reduction in recurrence-free survival of patients with basal breast cancer and indicates that palbociclib in some contexts may have long-term and potentially detrimental immunosuppressive effects (Supplemental Figure 7). Future studies could address this question in early stage breast cancer models from syngeneic immune-competent animals.

Three CDK4/6 inhibitors are now FDA approved for the treatment of advanced stage breast cancer either alone or in combination with aromatase inhibitor treatment (8-10). Our study demonstrates that short-term treatment with palbociclib is able to slow the growth and progression of both basal DCIS and early invasive tumor models, generating tumor stasis without significant side effects, and that the persistent influence of the drug even after treatment cessation may result in less aggressive tumors and reduced overall recurrence. Based on these findings, palbociclib might be best applied as a neoadjuvant therapy following diagnosis with DCIS and prior to surgical removal of lesions as a means of preventing further disease progression and potential dissemination. An ongoing Phase II clinical trial at Georgetown University titled “Preoperative Palbociclib in Patients with DCIS of the Breast that are Candidates for Surgery (WI223281)” () will investigate this potential. The results of this study and other future studies should address whether different doses, increased duration of treatment, or combination therapies with palbociclib could contribute to a more lasting impact of CDK4/6 in the treatment of earlier stage breast cancer.