Opposing Effects of Pigment Epithelium-Derived Factor on Breast Cancer Cell versus Neuronal Survival: Implication for Brain Metastasis and Metastasis-Induced Brain Damage

Cell lines

The human MDA-MB-231 brain tropic cell line (231-BR) and murine 4T1-BR5 brain tropic line were described previously (32, 33). The 231-BR line was obtained from Toshiyuki Yoneda, University of Texas, San Antonio, and was authenticated by our laboratory to the ATCC MDA-MB-231 line by short tandem repeat profiling. The 4T1-BR5 line was obtained from Suyun Huang, MD Anderson Cancer Center. R28 retinal progenitor cells were obtained from Gail Seigel, University of Buffalo, NY (34).

Subcloning and production of stable cell lines

The coding region for the human PEDF gene was a gift from the laboratory of Marta Garcia-Diez, CIEMAT, Madrid, Spain. The coding region, with two tandem FLAG tags added to its C-terminus, was subcloned into the lentiviral expression vectors pCDH-CMV-MCS-EF1-Puro and pCDH-EF1-MCS-T2A-Puro (SBI). PEDF was also subcloned into the pJC13-1 vector (35), a gift from the laboratory of G. Felsenfeld, NIDDK, Bethesda, MD. Additional details of DNA subcloning can be found in Supplementary Methods. Stable cell lines were generated in 231-BR cells by selection in 500 μg/ml Neomycin, and in 231-BRins and 4T1-BR5 cells by selection in 2–30 μg/ml of Puromycin, for at least 2 weeks. Cells were subsequently maintained in the absence of selective agent, and showed no loss of transgene expression.

Purification of PEDF, real-time-label-free cellular analysis and western blot analysis

Recombinant human PEDF (or PEDF-FLAG) was purified as previously published (36). Cell impedance was measured using a real-time microelectronic sensor system (ACEA/Roche, CITY, ST) as described previously (37). Cell viability was measured using a CellTiter-Glo^™ viability assay kit (Promega) following the manufacturer’s instructions. Data from two or more replicates were averaged. Western blots were performed according to standard protocols. The following antibodies were used: anti-FLAG (SIGMA); anti-GAPDH (Cell Signaling Technology, CS); mouse anti-human PEDF (Millipore); rabbit anti-PEDF (BioProducts Maryland); anti-α-Tubulin (Calbiochem), diluted 1:1000–1:5000. Secondary antibodies (SantaCruz) were diluted 1:2000–1:5000.

Hematogenous mouse models of brain metastasis

Experiments were conducted under an approved Animal Use Agreement with the NCI. Under isoflurane/O₂ anesthesia, 5- to 7-week-old female NCr nu/nu mice (Charles River Laboratories) were inoculated in the left cardiac ventricle with 1.75×10⁵ human 231-BR cells; immunocompetent BALB/c mice were injected with 5×10⁴ murine 4T1-BR5 cells. Mice were euthanized under CO₂ asphyxiation at time points correlated with the onset of weight loss, ataxia, and/or paralysis. Brains, bisected along the sagittal plane, were processed for cryosectioning as previously described (38). Ten micron cryosections were prepared and processed for histology and immunofluorescence. To quantify the number of metastases, ten sagittal sections, spaced every 300 microns, were cut through one hemisphere and H&E stained. Metastases were counted at 50x magnification using an ocular grid. Any metastasis that measured >300 μM in any direction was considered a large metastasis, as previously published (38).

Intracranial model of brain metastasis

Six- to eight- week old female, NCr nu/nu mice (Charles River Laboratories) were used. Mice, divided into 3 groups to receive intracranial injections of PBS, vector control 231-BRins cells, or PEDF-231-BRins cells, were anesthetized with ketamine (83 mg/kg) and xylazine (8.3 mg/kg). Cells were injected into the right caudate nucleus as previously described (39). Mice, anesthetized as above, were perfusion fixed with ~10 ml of PBS followed by ~10 ml of 4% paraformaldehyde prior to extirpation of the brain 0, 1, or 3 days post-implantation. Whole brain was immediately frozen in OCT. Serial 10 μM cryosections were prepared along the horizontal plane starting from the dorsal surface of the cerebral cortex and progressing to a depth of ≥3 mm. H&E sections were prepared every 100 μM.

Histological and immunofluorescence analysis

For Bielschowsky’s silver impregnation, 10 μM frozen sections were fixed in 10% phosphate-buffered formalin for 24 hours at 37°C or 48 hours at room temperature, and stained using Sigma reagents. Fluorojade-B (Histochem) staining was performed similar to the published method (40), omitting sodium hydroxide treatment and staining in a final dye concentration of 0.001%. Immunofluorescence analyses were performed as previously described (41). The following antibodies were used: anti-CD31 (BD Pharmingen); anti-Cleaved Caspase 3 (CS); anti-FLAG (SIGMA); anti-human mitochondria (Thermo Scientific); anti-Ki67 (Vector Laboratories); anti-human PEDF (Millipore). Secondary antibodies were conjugated to Alexa488 and Alexa568 (Molecular Probes). Nuclei were visualized with of 200 μg/mL DAPI (4′, 6-diamidino-2-phenylindole, dilactate) stain (Molecular Probes).

Numbers of Cleaved-Caspase-3, FLAG, Fluorojade-B, Ki67, or PEDF-FLAG-positive cells or CD31-positive blood vessels were counted in photographs using DAPI to counterstain all nuclei, and anti-human mitochondria to label human breast cancer cells. Ten sections, ~100 μM apart, were collected from each mouse for Fluorojade analysis. Five sections, ~200 μM apart, were collected for analysis of Ki67 and Cleaved Caspase-3 in the intracranial model.

Statistical Analysis

A variety of analyses of variances (ANOVA) were performed on the raw or transformed data (as indicated by the Box-Cox transformation). Samples from experimental units were averaged and the square root of the sample size was used as a weight in subsequent analyses. One-way ANOVAs were used to analyze simple fixed effect models. Two- or three-factor factorial ANOVAs were used to analyze more complex fixed effect models. Mixed models were used as necessary to model random variation due to repeated experiments or other random effects. Analysis of covariance was used to account for important covariates. Linear regression was used to estimate growth rates of individual mice. Finally, for all ANOVAs, residuals were examined for normality and homogeneity and residuals were partitioned if found to be heterogeneous. The Wilcoxon rank sum test was used to compare distributions if the data were not normally distributed and/or the sample sizes were very small. For pair-wise comparisons, Holm’s method was used to adjust pair-wise p-values.

Effects of PEDF on cultured breast cancer cells and neural retinal precursors

The effects of PEDF on two “triple-negative” breast cancer cell lines were examined in vitro. Cells were analyzed in real time by measuring cell impedance and at endpoint by measuring biomarkers of live cells to confirm cell viability monitoring. At high cell density, 4T1-BR5 cells, a brain-tropic derivative of the 4T1 murine cell line, remained viable for several days in the absence of serum, but addition of increasing concentrations of purified recombinant human PEDF progressively decreased 4T1-BR5 cell viability (Figure 1A; Supplementary Figure 1A, B). Comparison of the 0, 10 and 100 nm doses using the exact Jonckheere-Terpstra test showed a significant dimunition of viability by PEDF (p = 0.022) (Figure 1B); comparison of vehicle versus PEDF for 4T1 cells (P=0.022 by the same test) (Supplemental Figure 1a). Similar trends were also observed using 231-BR cells, a brain tropic derivative of the MDA-MB-231 cell line (Figure 1C–D).

We also tested the pro-survival activity of PEDF on neural progenitor cells. As illustrated in Figure 1E, low density rat retinal progenitor (R28) cells progressively died within a day of serum deprivation, but PEDF significantly increased R28 cell viability in a dose-dependent manner. Comparison of the 0, 10 and 100 nm doses showed a significant protective effect of PEDF (p = 0.022) (Figure 1F). PEDF protein altered to include two-tandem FLAG-epitope tags on the C-terminus of the protein maintained its biological activity, showing a statistical trend in reduced viability using the exact Jonckheere-Terpstra test (Supplementary Figure 1C, D)(P=0.33 for PEDF, 0.067 for PEDF-FLAG). Thus, PEDF appears to have direct but opposing effects on the viability of breast cancer cells and normal neural retinal progenitor cells.

Transient PEDF expression suppressed proliferation of brain metastases in two breast carcinoma models

In agreement with the observation that PEDF expression is down-regulated in human breast cancer brain metastases, neither of the brain metastatic breast cancer cell lines had detectable expression of PEDF protein (Figure 2A, left column and data not shown). Human 231-BR cells were transduced with a human PEDF transgene to produce stable clones. The FLAG-tagged PEDF protein was efficiently secreted from the cancer cells, and was detectable as a single band of 50 kD by Western analysis with both anti-PEDF (Figure 2A) and anti-FLAG antibodies (not shown).

While PEDF exhibited negative effects on breast cancer cell viability in the absence of serum (Figure 1), PEDF expression appeared well tolerated when cells were cultured in complete media, showing no decrease in anchorage-dependent growth when compared to vector control clones or untransfected 231-BR cells (Supplementary Figure 2A). Nor did the PEDF-expressing cells show a decrease in migration across a collagen-coated membrane in a Boyden chamber assay (Supplementary Figure 2B). PEDF protein expression appeared to be relatively uniform across cell populations as visualized by immunofluorescence, and could be maintained in the absence of selection agent for multiple passages without noticeable changes in PEDF protein expression (data not shown).

To analyze the effect of PEDF on the brain metastatic potential of breast cancer cells, three vector control 231-BR clones and three PEDF-231-BR clones were injected into the left cardiac ventricle of immunocompromised mice. The number of brain metastases was quantified by histology of step sections through one brain hemisphere of each mouse 26–28 days later (Figure 2B). PEDF expression had no significant effect on the number of micrometastases (less than 300 μm in a single direction) or leptomeningeal metastases. However, the number of large (>300 μM) parenchymal metastases formed from PEDF-231-BR cells (mean=11.9; n=60 mice for the three combined PEDF-expressing clones) was decreased to 66% of controls (mean=18.1; n=48 mice for the three combined control clones; p=0.001).

Staining of tissue sections with anti-FLAG revealed that only 1–2% of tumor cells (352/22,092) retained PEDF expression after 26 days in vivo. Metastases were co-stained for PEDF-FLAG and Ki67 (Figure 2C), and the numbers of PEDF-FLAG, Ki67, and double-positive cells were quantified (Figure 2D). As previously published (41), Ki67 staining was confined to metastatic tumor cells and absent in the surrounding brain cells (Figure 2E). The majority (59%) of cells (7,812/13,317) were proliferating in control metastases (Figure 2D, left column). In the mice inoculated with PEDF-231-BR cells, the proliferation rate among the subset of metastases in which no PEDF-expressing cells were visible was not significantly different from controls (57%; 8,387/14,732 cells). In contrast, the rate of proliferation (42%; 3,133/7,360 cells) was significantly reduced in the subset of metastases with mosaic PEDF expression (Figure 2D, middle column). There was an even greater decrease in proliferation (12%; 42/352 cells) among the individual cells which maintained PEDF expression (PEDF-FLAG and Ki67 double-labeled cells; Figure 2D, right column). The reductions in proliferation, both in PEDF-expressing cells themselves, and in mosaic metastases containing a minority population of PEDF cells, were highly significant (p= 0.002 vs. control).

Vascular density was examined in control and PEDF brain metastases. Blood vessels were stained with anti-CD31, and the blood vessel density was counted both in brain metastases and in an adjacent region of brain devoid of tumor cells (Figure 2F). In agreement with previously published results (42), brain metastases in the 231-BR model had a reduced vascular density relative to the surrounding normal brain (p=0.0005). PEDF expression did not significantly alter the number of blood vessels in brain metastases (p=0.37, n=6 mice per group). This suggests that the observed reduction in proliferation and formation of large brain metastases is not due to an anti-angiogenic effect of PEDF.

In an attempt to increase the fidelity of PEDF expression, a second expression construct was engineered. Transcription of PEDF was placed under control of the human EF1α promoter, and the transgene was flanked with duplicate transcriptional insulator elements derived from the chicken heat shock 4 locus (35). This construct was used to express PEDF in the murine 4T1-BR5 brain metastatic cell line (Figure 3A). PEDF expression appeared to be uniform and stable among clonal populations over many generations in vitro, and had no effect on anchorage-dependent growth of these cells (Supplementary Figure 2C). The expression construct is illustrated in Supplementary Figure 2D.

Control and PEDF-expressing clones of 4T1-BR5 cells were inoculated into the left ventricle of syngeneic, immunocompetent Balb/c mice. The number of brain metastases was quantified after 9 days (Figure 3B). The insulator construct did not improve transgene stability in vivo. Similar to the results with 231-BR cells, PEDF expression was rapidly silenced in the syngeneic mouse model, and no PEDF-positive cells were visible at the endpoint (data not shown). However, transient PEDF expression in 4T1-BR5 cells lead to a significant reduction in micrometastases (median= 73) as compared to controls (median= 432, p=0.018), and a significant reduction in large brain metastases (median=0 vs. 8, p=0.005).

In order to determine whether the tumor microenvironment influenced either the stability of PEDF expression or its effects on breast cancer cell proliferation, we implanted breast cancer cells into the mammary fat pad. The insulator construct was used to generate stable polyclonal (231-BRins) control and PEDF-expressing populations in 231-BR cells (Supplementary Figure 2E). PEDF expression was relatively uniform across the populations in vitro (Supplementary Figure 2F). PEDF expression significantly slowed mammary tumor growth rate (p=0.005, two-sample t-test, Supplementary Figure 3A). Similar to the experimental brain metastasis models, very few cells at the endpoint maintained PEDF expression. However, the rare PEDF-expressing cells were Ki67-negative (Supplementary Figure 3B). Blood vessel density was not significantly different (p=0.55) between the two groups, suggesting that the slower growth of PEDF tumors was not due to anti-angiogenesis (Supplementary Figure 3C). Thus, using two models of breast cancer metastasis, and in two microenvironments, despite achieving only transient PEDF expression, a growth suppressive effect of PEDF was repeatedly observed.

PEDF expression rapidly inhibits proliferation and activates apoptosis of breast cancer cells in the mouse brain

In order to study both the cancer suppressive and the neuroprotective effects of PEDF simultaneously, metastatic breast cancer cells were directly implanted into mouse brains. A time course was performed to determine the stability of PEDF protein expression in vivo. Five hours post-implantation, PEDF-FLAG expression was visible in the majority of PEDF-231-BRins cells, but was lost in many cells 24 hrs post-implantation and was present in less than 1% of the cells by 3 days post-implantation (Figure 4A).

To determine the effect of PEDF on the viability of the brain metastatic tumor cells, mice were necropsied 24 hrs post-implantation of control or PEDF-expressing tumor cells. A third group of mice received an intracranial injection of saline, in order to determine the effects of the stab wound on the mouse brain. Serial sections were prepared, starting from the needle entry site along the dorsal surface of the brain, and progressing to the bottom of the lesion, at a depth of ~3 mm. Representative H&E stained sections are shown in Figure 4B. Five step sections, evenly spaced over a length of ~1 mm (which spanned most of the depth of the implanted tumor cells), were stained for either Ki67 (Figure 4C and 4D) or Cleaved Caspase-3 (Figure 4E and 4F). The majority (64%, 17,658/27,625 cells) of the implanted 231-BRins cells in control lesions were Ki67-positive (proliferating) (Figure 4C, left). In PEDF-expressing lesions (Figure 4C, right) proliferation was reduced to a minority population (37%, 5278/14,186 cells, p<0.0001). In control 231-BRins tumors (Figure 4E, left), apoptotic cells expressing Cleaved-Caspase-3 were rare (4%, 1,286/28,763 cells). In contrast, more than a third (39%) of the cells in PEDF-231-BRins lesions (5,596/14,461 cells; p<0.0001 vs. control) were undergoing apoptosis (Figure 4E, right). Thus, while PEDF expression was well tolerated by breast cancer cells in vitro, it has a rapid and profound suppressive effect on tumor viability in the brain. Both control (Supplementary Figure 4A–B) and PEDF tumors (Supplementary Figure 4C–D) showed no alterations in their vasulature, suggesting that these effects are independent of the anti-angiogenic effects of PEDF.

PEDF protects neurons in the brain from damage induced by tumor cells

In addition to the direct effect of PEDF on tumor cells, the intracranial model permitted an examination of the effects of secreted PEDF on the brain. Silver stain was used to visualize the morphology of brain neurites, and was evaluated independently by two neuropathologists. Adjacent to control 231-BRins lesions, the bundles of nerve fibers were frequently disorganized, with obvious thickened and thinned neurites indicative of neuronal damage (Figure 5A, middle). A saline-only injection confirmed that neuronal damage outside of the lesion was due primarily to the tumor cells rather than the injection process (Figure 5A, left). Far less damage was apparent adjacent to PEDF-231-BRins tumors, as compared to control tumors (Figure 5A, right).

As the silver stains were only qualitative, a second stain for neuronal damage was employed, Fluorojade-B (Figure 5B). Fluorojade-B has been shown to label damaged brain neurons specifically. This dye appears to label neurons undergoing either necrotic or apoptotic death, and has been used to analyze brain damage resulting from a variety of insults (40, 43–47). Fluorojade-B staining was not visible around breast cancer micrometastases, but became apparent in advanced disease in the hematogenous brain metastasis model (Supplementary Figure 5).

To examine the extent of damage in the intracranial model, ten step sections spanning ~1 mm of the lesion were stained and counted to determine the number of injured neurons surrounding the lesions (Figure 5C). In agreement with the silver stains, a small number of damaged neurons (200 per brain) were evident surrounding the stab wound in saline injection controls (Figure 5B, left). There was a 15-fold increase in damaged neurons (2950 per brain; p<0.0001 vs. saline controls) when control 231-BR cells were implanted (Figure 5B, middle). The majority of this damage was suppressed by PEDF expression in the implanted cells (840 per brain; p=0.008 vs. control tumor) (Figure 5B, right). Moreover, PEDF appeared to decrease the area of the brain impacted by the lesion (Figure 5D). The farthest Fluorojade-positive neurons were only a few cell lengths (mean=86 μM) from the stab wound in saline injected brain. Damage around 231-BRins tumors extended much further (mean=269 μM; p=0.0003 vs. saline controls). The radius of damage around PEDF-231-BRins tumors was significantly smaller than the damage around control tumors (mean=145 μM; p=0.009). These results indicate that PEDF can simultaneously act to promote survival in neurons while activating cell death in cancer cells in the brain.