De-identified brain metastases and their clinical-pathological information (age, ER, PR, and HER2 status at the time of metastases resection, prior therapies, and survival) were obtained from consenting breast cancer patients undergoing neurosurgery. These samples were collected under approved IRB protocols at University of Colorado Anschutz Medical Campus. All animal studies were performed under approved University of Colorado Institutional Animal Care and Use Committee (IACUC) protocols.

Freshly removed brain metastasis samples were placed on sterile ice-cold DMEM and transported to the laboratory for transplantation into mice. Specimens that could not be immediately implanted were maintained at 4°C for no longer than 8 h. Female NSG mice, 6–8 weeks old were purchased from Jackson laboratories or bred at the UC Denver Center for Comparative Medicine breeding facility. Brain metastases were partitioned into 5–10 mm³ pieces, dipped into cultrex, and implanted in the fourth mammary fat pad of anesthesized mice using a 10-gage trochar. In one case, brain metastatic cells were collected from cerebrospinal fluid from a patient diagnosed with meningeal carcinomatosis. Here, cancer cells were collected by centrifugation and divided into two aliquots. One mouse was injected ic with cancer cells suspended in 100 μl PBS, another recipient was injected into the mammary fat pad with cancer cells resuspended in 50 μl cultrex. All mice were implanted with a silastic pellet providing slow release of 17β-estradiol (E2), as prior experience showed that E2 increases tumor uptake of breast cancer PDXs irrespective of tumor subtype. Tumors were palpated weekly to assess tumor take for up to 8 months postimplantation. Once palpable, tumor size was assessed weekly using caliper and volume estimated as length × width²/2. When tumors reached ~1.5 cm in any direction, mice were euthanized and tumors removed. Tumors were divided into several 10 mm³ pieces and reimplanted in the mammary fat pad of NSG mice, cryopreserved in 10% DMSO/90% FBS in liquid nitrogen, stored in trizol RNA extraction, and fixed in 10% formalin for paraffin embedding. BM-PDXs were considered established if they grew over two generations.

A subset of BM-PDXs were labeled with lentiviral particles expressing EGFP-luciferase as we described previously (27). Briefly, >1 cm³ tumors were resected from euthanized mice and digested in Accumax (Stemcell Tech) for 3 h at 30°C. Human cancer cells were separated from mouse stromal cells using a lineage cell depletion kit (MACS) and isolated breast cancer cells were plated in six-well ultralow attachment plates in DMEM-F12 media. Tumor cells were transduced with 30 MOI of lentiviral pHAGE-EF1aL-luciferase-UBC-GFP-W and GFP expression monitored for up to 48 h. Labeled tumor cells were then collected, washed, resuspended in 100 μl Cultrex basement membrane extract and injected in the mammary fat pad of NSG mice. Efficiency of transduction was assessed using luciferase activity imaging (IVIS) or GFP expression when tumors reached >1 cm³. Labeled BM-PDX were cryopreserved, fixed for immunohistological analysis, stored in trizol for RNA extraction, or transplanted into a new recipient.

A cell line (F2-7) was derived from triple-negative BM-PDX. For this, tumor cells were dissociated from BM-PDX F2-7 using Accumax, and dissociated cells were plated in ultralow attachment six-well plates in DMEM-F12 supplemented with 10% of FBS, 1 μg/ml hydrocortisone, 100 ng/ml of cholera toxin, and 1 nM of insulin. After 4 weeks of growth in suspension, human cells free of fibroblasts were plated in collagen-I coated dishes, and purity validated by immunohistochemistry. F2-7 cells were labeled with GFP-luciferase as described for BM-PDXs. Short-Tandem Repeat analysis was performed in the established cell line and deposited at University of Colorado Tissue Culture Core facility for validation and future reference.

High-throughput RNA sequencing from a cell line derivative (F2-7) and a selected set of BM-PDXs before and after labeling with EGFP-luciferase was performed. RNA was isolated from tumor samples using trizol followed by RNA cleanup using RNEeasy MinElute Cleanup kit (Qiagen), and RNA concentration was measured in a Nanodrop 2000 (Thermo Scientific). The Genomics and Microarray core facility at the University of Colorado AMC performed RNA quality control using an Agilent 2100 Bioanalyzer, and prepared RNA-seq libraries using the Illumina TruSeq Stranded mRNA LT Sample Prep Kit. The resulting libraries were sequenced on an Illumina HiSeq 2500 system (1 × 125 bp). After demultiplexing, the resulting reads were trimmed with cutadapt to remove 3′ adaptor sequences and low quality 3′ bases (Q < 10). The trimmed reads were then aligned to both the human (hg19/GRCh37) and mouse (mm10) genomes using Tophat2 (28). Reads were then assigned to either the human or mouse genome using disambiguate (29) and ambiguous reads were discarded. Unambiguous reads were assigned to features using Rsubread (30) and normalized read counts were produced using the rlog function in DESeq2 (31). The GEO Accession number for this data is GSE104020.

Two tumors from E22-1 BM-PDXs grown in the mammary fat pad were excised at necropsy, cut into 2 mm² pieces and dissociated using Accumax for 3 h. The digestion was stopped using DMEM/F12 10% FBS, and single cells isolated by filtering through 100 and 70 μm mesh filters. Viable cells were counted using trypan blue exclusion and 250,000 cells resuspended in 100 μl PBS were injected in the left cardiac ventricle of recipient female NSG mice (n = 8). Brain metastatic burden was assessed using T1/T2 contrast magnetic resonance imaging (MRI) 8 weeks after ic injection, and mice were euthanized under CO₂ asphyxiation at 8 or 12 weeks after ic injection as mice developed signs of CNS metastatic burden. In all cases, brains were removed at necropsy and brain hemispheres embedded in OCT and stored at −80°C until sectioning. Micrometastases were visualized with H&E and/or pan cytokeratin (PanCK) staining in six serial sections (10 μm thick), one every 300 μm in a sagittal plane through the right hemisphere of the brain.

To non-invasively detect and quantify brain metastatic colonization, brain MR scans were acquired using a Bruker 4.7 T PharmaScan and a bird-cage radio frequency 36 mm coil (Bruker Medical, MA, USA). Animals were injected via tail vain with 0.4 mmol/kg gadolinium contrast Multihance (gadobenate dimeglumine, Bracco Diagnostic) and anesthetized with 2–2.5% isoflurane. High-resolution rapid acquisition with relaxation enhancement (RARE) T2-weighted images with fat suppression were obtained (TR/TE = 4,000/80 ms) followed by a multislice multiecho (MSME) T1-weighted sequence (TR/TE = 700/11 ms). All images were obtained in the axial plane, with the field of view of 3 cm, slice thickness 1 mm, number of slices 16, matrix size 256 × 256. In-plane resolution was 90 μm. T1-weighted MSME images were acquired as well to confirm metastatic lozation. All images were acquired and analyzed (T2-MRI for lesion numbers and diameters) using Bruker ParaVision (v4.3) software.

Tumors were removed from animals and fixed in 10% buffered formalin. Tissue was processed, paraffin embedded, and cut into 5-μm sections. After high-temperature antigen retrieval in citrate buffer, sections were stained with rabbit anti-epidermal growth factor receptor (anti-EGFR, Cell Signaling), rabbit monoclonal anti-C-erB-2 (SP3, Neomarkers), mouse monoclonal antibody anti-cytokeratin 5 (anti-CK5, Vector), rabbit polyclonal anti-PR (DAKO, Carpinteria, CA, USA), and rabbit polyclonal anti-ERα (SP1, Thermofisher). Sections were counterstained with hematoxylin and mounted. Representative photographs were taken under a light microscope at ×20 magnification.

Dual immunofluorescence of brain metastasis was performed 10-μm sections from frozen unfixed-OCT embedded brains. Sections were fixed in acetone and stained with a mouse monoclonal antibody specific for human cytokeratins (Pan-CK, MNF116, Dakocytomation, Glostrup Denmark); in combination with rat anti-GFAP (Invitrogen, CA, USA); rabbit anti-collagen IV (Millipore). Secondary antibodies were anti-mouse Alexafluor-488 or anti-rabbit Alexafluor-565 or anti-Rat-Alexafluor-594 (all from Invitrogen/Thermofisher, CA, USA). Nuclei were stained with 1 μg/ml 4′,6-diamidino-2-phenylindole (DAPI) in methanol for 10 min at room temperature. Photographs were taken under ×20 magnification for the same field using the UV, FITC, and TRITC filters.

We implanted a total of 14 brain metastases specimens from TN (n = 4), ER⁺HER2⁻ (n = 1), ER⁺HER2⁺ (n = 2), and ER⁻HER2⁺ (n = 7) breast cancer subtypes. From these, 8 (57.2%) successfully established as BM-PDXs, defined as those tumors that grew as xenografts in NSG mice at least in one consecutive passage and maintained expression of clinical markers from original patient sample. The frequency of BM-PDXs uptake varied among subtypes with a non-significant trend toward highest take rate in TNBC (3/4 BM-PDXs, 75%) and lower take rate in ER⁻HER2⁺ (3/7, 43%) (Figure 1A). The overall clinical–pathological characteristics of the brain metastases successfully established as BM-PDXs are presented in Table 1. Among HER2⁺BM-PDXs, two specimens had prior history of TNBC but their brain metastasis were diagnosed as HER2⁺; these were classified as HER2⁺BM-PDXs. Time to xenograft tumor formation for each BM-PDX at initial implantation ranked between 28 and 223 days, with an average of 84 days (Figure 1B). Similar to prior reports, in vivo tumorigenic potential of explanted brain metastases (tumors that grew as BM-PDXs) was correlated with decreased survival of their donor patients (P = 0.0011) (Figure 1C). Tumor progression in initial PDX (P0) and subsequent in vivo passaging (P1 to P3) for a TN and HER2⁺BM-PDXs is shown in Figures 1D,E. In one case, an ER-HER2⁺ brain metastases was implanted but an inguinal tumor developed suddenly after 70 days, away from the implantation site. This tumor lacked HER2, EGFR, or CK5 and upon transplantation grew into a large mass within 2 weeks (data not shown). As this suggested either loss of human epithelial markers or most likely, outgrowth of a murine tumor, we did not consider this a successful PDX and excluded it from further analysis.

To determine whether outgrowth of BM-PDXs in the mammary fat pad retained key clinical features of brain metastases donors, we stained sections of BM-PDXs (P0-P1) for ER, PR, and HER2, and—when available—we compared them to clinical specimens at the time of implantation. TNBC brain metastases lack these markers but frequently express EGFR and CK5 (32). Thus, we added these to our validation panel. As show in Figure 2A, TN BM-PDXs expressed EGFR and CK5. For these samples there was no matching donor sample to compare, but lacked ER, PR, and HER2 as expected from TN tumors. ER-HER2⁺BM-PDXs retained HER2, EGFR, and CK5 (Figure 2B), and ER⁺BM-PDXs retained ER, HER2, EGFR, and CK5 expression similar to the donor sample (Figure 2C). Surprisingly the two ER⁺BM-PDXs (G7-1, G13-1), showed increased PR expression as compared to the donor samples (Figure 2C), suggesting that ER is functional in these BM-PDXs and that E2-supplementation in mice upregulates PR in ER⁺BM-PDXs.

To assess whether GFP and luciferase labeling would allow in vivo imaging of these BM-PDXs, we dissociated cells from >1 cm³ xenografts BM-PDXs F2-7, CSF-1, and G5-3 and transduced them with high titer viral particles of a GFP-luciferase vector as described (27). Labeled cells were regrown in the mammary fat pad of NSG mice and tumor labeling was assessed by measuring luciferase activity (IVIS) and GFP expression (Figure 3A). Spontaneous metastases to surrounding areas were detected in some mice during tumor excision (Figure 3B), but without spontaneous metastases to brain or other organs. Since tumor cell dissociation and cell transduction with lentiviral vector might results in selection of subclones of the original tumor, we assessed whether labeled PDXs recapitulated the heterogeneity observed in parental PDXs. IHC staining showed that labeled BM-PDXs retained expression of EGFR, CK5, and HER2 (Figure 3C), suggesting that dissociated/labeled cells are capable of reconstituting tumor heterogeneity of BM-PDXs.

As tumor-dissociated cells survived short-term in vitro culture during labeling, we sought to determine whether dissociated brain-metastatic cells could be cultured as cell lines. We cultured dissociated cells from F2-7, E22, and G5-3 BM-PDXs in plates coated with collagen-I or ultralow attachment plates. In either condition, only cells from F2-7 BM-PDXs survived in vitro culture and remained proliferative after multiple passages of repeated freezing and thawing (Figure 4A). This F2-7 cell line-derivative retained expression of EGFR as its BM-PDXs counterpart (Figure 4B), and retained tumor initiating capability in vitro (measured as ability to form colonies in the absence of extracellular matrix in mammosphere assays, not shown). To further assess whether BM-PDXs labeling or cell line derivation maintained features of parental BM-PDXs, we performed RNA sequencing followed by hierarchical gene clustering analysis of BM-PDX before and after GFP-luciferase labeling, and F2-7 BM-PDX and its cell line-derivative (Figure 5). Key genes (EGFR, KRT5, NTRK2) were expressed at similar levels in parental PDXs (i.e., E22-1 PDX-P0) compared to its labeled counterpart (E22-1 PDX-P0-I1), and in the F2-7 cell line (F2-7 CL) compared to its BM-PDXs parental (F2-7 P5). Taken together, these data suggests maintenance of clinical markers through passaging and manipulation of BM-PDXs.

Breast cancer PDXs grown in the mammary fat pad rarely metastasize to distant organs, but dissociated cells can colonize lung, bones and brain after ic injection (27). To assess whether BM-PDXs retain their ability to colonize the brain, we induced experimental brain metastasis using dissociated cells from TN E22-1 BM-PDX. For this, 250,000 dissociated cells were injected in the left ventricle of 8–week-old female NSG mice (n = 8) supplemented with estradiol, and metastases were allowed to grow until mice showed >15% weight loss or neurological impairment. Nine weeks after ic injection mice were imaged using T1/T2 MRI and mice without symptomatic metastases were left alive for 3 additional weeks. MRI-detectable brain metastases were found in 4/8 (50%) of injected mice, with metastatic lesions ranking from 0.29 to 0.62 mm in size (Figure 6A, top). One additional mouse was injected with E22⁻BM-PDX dissociated cells but MRI was performed 14 weeks after ic injection. This mouse showed multiple large MRI-detectable metastases (Figure 6A, bottom). Histological analysis showed 8/8 mice (100%) harboring micrometastases (defined as >50 μm cancer cell foci counted in six sagittal brain sections 300 μm apart) with a 10.25 median number of micrometastasis per mouse (Figure 6B). To determine whether brain metastases formed by BM-PDXs showed pathophysiological features similar to those encountered in humans, we performed double immunofluorescence staining of brain metastatic cells (pan-cytokeratin⁺, green) and reactive astrocytes (GFAP⁺, red) or blood vessels (Col-IV, red) in brain sections from mice injected with E22-1 BM-PDXs. Brain metastatic clusters were located to the brain parenchyma (Figure 6C), were associated with blood vessels (Figure 6D) and were surrounded by GFAP⁺ reactive astrocytes (Figure 6E); all of these characteristics of breast cancer brain metastases. Taken together, these studies demonstrate that BM-PDXs retain their ability to form large brain metastases and micro metastases at high frequencies, making them suitable models for studies of brain metastatic colonization and preclinical testing of drugs in preventive and therapeutic settings.