Epithelial-to-Mesenchymal Transition Drives Invasiveness of Breast Cancer Brain Metastases

Simple Summary The prevalence of breast cancer brain metastasis has increased over the last decades, yet those patients are considered untreatable. Surgical removal of brain metastases is impeded by the presence of infiltrative tumor cells; however, the identity and behavior of those cells remain understudied. We demonstrate that cancer cells that invade the brain parenchyma are in a mesenchymal state and undergo EMT. We also demonstrate that removing those infiltrated tumor mesenchymal cells improves the efficacy of surgery and opens new avenues to better treat patients. Abstract (1) Background: an increasing number of breast cancer patients develop lethal brain metastases (BM). The complete removal of these tumors by surgery becomes complicated when cells infiltrate into the brain parenchyma. However, little is known about the nature of these invading cells in breast cancer brain metastasis (BCBM). (2) Methods: we use intravital microscopy through a cranial window to study the behavior of invading cells in a mouse model of BCBM. (3) Results: we demonstrate that BCBM cells that escape from the metastatic mass and infiltrate into brain parenchyma undergo epithelial-to-mesenchymal transition (EMT). Moreover, cells undergoing EMT revert to an epithelial state when growing tumor masses in the brain. Lastly, through multiplex immunohistochemistry, we confirm the presence of these infiltrative cells in EMT in patient samples. (4) Conclusions: together, our data identify the critical role of EMT in the invasive behavior of BCBM, which warrants further consideration to target those cells when treating BCBM.


Introduction
Although early detection of primary tumors and novel treatment strategies have improved outcomes of patients with breast cancer (BC), an increasingly large number of BC patients live long enough to develop lethal BM: 30-50% of patients with Her2+ BC and 25-45% of triple negative BC develop brain metastasis [1][2][3][4][5]. Standard-of-care treatments for BCBM include neurosurgical resection, systemic therapies and stereotactic radiotherapy. However, these treatments largely fail to improve patients' outcomes and BCBM remains a leading cause of BC-related death [5]. In particular, the ability of BCBM cells to infiltrate into the healthy brain tissue renders BCBM difficult to treat. Although tumor resection may seem complete by magnetic resonance imaging (MRI), the limited resolution of this technique does not allow the detection of individual cells that have invaded beyond the resection cavity. Consequently, overall survival rates following surgical removal of BCBM remain dismal, with 60% of patients with confirmed complete resection by MRI locally recurring within 1 year after surgery [6,7]. Nevertheless, the properties of BCBM cells that invade the brain parenchyma and drive patients' relapse remain under-studied. It is therefore critical to better understand the mechanisms that drive the invasive behavior of BCBM cells in order to design efficient treatment strategies for this dreadful disease.
The migratory and infiltrative behaviors of BC cells that escape primary tumors have been extensively studied [8][9][10][11][12][13]. It is generally accepted that in order to become invasive, cells in the primary tumor hijack the developmental program known as EMT [14][15][16][17][18][19][20][21][22][23][24]. For instance, we and others previously used intravital microscopy to show that cells that escape from primary breast tumor and invade the healthy stroma undergo EMT [25][26][27][28]. During EMT, cells downregulate the expression of epithelial markers such as the cell-cell adhesion molecule E-cadherin (E-cad), thereby losing cell polarity and cell-cell adhesion. In parallel, cells undergoing EMT acquire mesenchymal markers that drive stemness and invasiveness such as Zeb1 and Twist [14,18,21,[29][30][31][32][33][34]. To date, the role of EMT in BC has mainly been studied in the context of cancer cells disseminating from the primary tumor and during the outgrowth of extracranial metastases, but the existence and functions of EMT in BCBM remain unknown.
We previously used the mouse mammary tumor virus polyoma middle-T antigen (PyMT) model crossed with fluorescent E-cadherin-mCFP (E-cad-mCFP) reporter mice to report EMT in ductal mammary carcinomas and lung metastases [26,27]. Here, we employed this EMT reporter mouse to generate a BCBM model in order to image, isolate and characterize cells that have undergone EMT in the context of brain metastasis. Using intravital imaging, we identify infiltrative tumor cells in BCBM which preclude the complete removal of the lesion by surgery, leading to increased BCBM recurrence in mice. Moreover, we demonstrate that in contrast to cells in the tumor mass, cells infiltrating into the brain tissue display a mesenchymal expression profile. Additionally, we find that mesenchymal cells can revert to an epithelial state during outgrowth of tumor masses in the brain. Importantly, using multiplex immunohistochemistry, we also detect this infiltrating population of cells in EMT in the brain tissues of BCBM patients, illustrating the clinical relevance of our observations. Collectively, our data warrants further consideration to target EMT as part of the treatment of BCBM.

Mice
All experiments were performed under the guidelines of the Animal Welfare Committee of the Royal Netherlands Academy of Arts and the Netherlands Cancer Institute, in accordance with national guidelines. Animals were kept at the Hubrecht animal facility in Utrecht, the Netherlands or at the Netherlands Cancer Institute facility in Amsterdam, the Netherlands. All animals were housed in individually ventilated cages under specificpathogen-free conditions and received chow and water ad libitum. Female ROSA26 mTmG C57BL/6J mice were purchased from Jackson Laboratory and backcrossed to FVB mice at the animal facility of the Hubrecht and of the Netherlands Cancer Institute. Female FVB mice were acquired from Janvier. All mice were injected with tumor cells when they were 8-20 weeks old.

Serial Intracranial Injections to Generate the BCBM Model
End-stage primary breast tumors were isolated from PyMT E-cad-mCFP mice [27] and frozen in freezing medium (Gibco, Cat. No. 12648-010, ThermoFisher Scientific, Waltham, MA, USA). For the first intracranial injection, tumors were thawed and minced on ice with sterile knives and enzymatically digested at 37 • C for 20-30 min while shaking at 900 rpm. The digestion mix was composed of 0.01 g collagenase A (Roche, Cat. No. 10103586001, Basel, Switzerland), 2 mL TrypLE Express (Gibco, Cat. No. 12605010) and 3 mL DMEM/F12 GlutaMAX (Gibco, Cat. No. 10565018) supplemented with 10 mmol/L Hepes (Gibco, Cat. No. 15630106), 100 µg/mL streptomycin and 100 U/mL penicillin (Gibco, Cat. No. 15140122). Undigested tumor pieces were removed by spinning down at 800 rcf at 4 • C and for 5 min. Then, 10 µg/mL DNase I (Roche, Cat. No. 10104159001) in DMEM/F12 GlutaMAX medium were used to dissolve the pellet while manually shaking the tube for 5 min. The final pellet was resuspended in 3 µL per intracranial injection of PBS. After a tumor was formed in the brain of the recipient mouse, the tumor was isolated and cut into smaller tumor pieces prior to freezing (Gibco, Cat. No. 12648-010). This whole process is considered as one enrichment round in the brain. We repeated this process for a total of 6 times using the tumor pieces from the previous serial enrichment round. At the 6th enrichment round, we derived organoids which are described in this study as 'BCBM organoids'.

Cranial Window Implantation and BCBM Injection under the Window
BCBM derived organoids were digested with TrypLE Express (Gibco, Cat. No. 12605010) for~20 min at 900 rpm and 37 • C to produce a single cell suspension. Digestion was stopped by adding DMEM F12 medium. Per injection, 100,000 cells were resuspended in 3 µL of PBS and injected just before the implantation of the cranial window. Cranial window implantation was performed as previously described [36]. In summary, mice were treated with 0.067 mg/mL rimadyl (carprofen, Zoetis) in drinking water 1 day before injection and for 3 days post surgery. Furthermore, 30 min before surgery mice were subcutaneously injected with 0.1 mg/kg temgesic (buprenorphine, Indivior Europe Limited) and 8-12 h after window implantation. Mice were anesthetized via inhalation of 2% (v/v) isoflurane and eyes were covered with duratears (Alcon). Once placed in a stereotactic frame, the mouse head was shaved and the skin was cut in a circular manner. Once the skull was exposed, a local anesthetic (0.01-0.02 mL) made of NaCl supplemented with 1 mg/mL Lidocaine (Fresenius Kabi) and 0.25 mg/mL Bupivacaine (Actavis, Aurobindo Pharma B.V.) was applied. After removal of the periosteum, a 5 mm diameter circle was drilled on the right parietal bone. The bone was removed following removal of dura mater with thin forceps. The tumor cell suspension was injected in the middle of the craniotomy at 0.5 mm depth with a 10 µL glass Hamilton syringe, 30G and point 4-style needle at a speed of 1 µL/minute. Silicone oil covered the exposed brain and a 6 mm coverslip was glued on top with histoacryl (Braun). Dental acrylic cement (Vertex) was applied on the skull surface to glue a 3D printed window which provides fixation when performing intravital microscopy. After surgery, mice were allowed to recover on a heating pad and were closely monitored the following days.

Intravital Imaging Microscopy
Mice were sedated with 1% (v/v) isoflurane and placed in a custom-designed imaging box while kept under continuous anesthesia with isoflurane. Both the imaging box and microscope were adjusted to 36.5 • C using a climate chamber. During long-term imaging, mice were administered with 100 µL/h of Nutriflex special 70/240 (Braun). Intravital images were acquired with both an inverted Leica SP8 Dive system (Leica, Mannheim, Germany) with a MaiTai eHP DeepSee laser (Spectra-Physics, Milpitas, CA, USA) and an InSight X3 laser (Spectra-Physics) or an inverted Leica SP8 Dive system (Leica, Mannheim, Germany) with only InSight X3 laser (Spectra-Physics). For both systems a Leica Application Suite X (LAS X) software (Mannheim, Germany) was used. For the system with 2 lasers, SHG (crosslinked collagen I), E-cad-mCFP and H2B-Dendra2 were simultaneous excited with 860 nm (Mai Tai) and detected at 420-442 nm (HyD-RLD1), 444-485 nm (HyD-RLD2) and 510-556 nm (HyD-RLD3), respectively. ROSA26 mTmG was excited with 980 nm (Insight X3) and detected at 564-698 nm (HyD-RLD4). For the system with one laser, we performed sequential scanning; sequence 1 used an excitation of 930 nm (Insight X3) to detect SHG at 458-471 nm (HyD-RLD2), H2B-Dendra2 at 491-541 nm (HyD-RLD3) and ROSA26 mTmG at 564-684 nm (HyD-RLD4); sequence 2 used an excitation of 840 nm to image E-cad-mCFP at 442-475 nm (HyD-RLD3). Three-dimensional tile scans were acquired with a 5 µm z-step size. Time-lapse movies were acquired with a 5 µm z-step and imaged every hour for a total of 8 to 12 h. All images were acquired in 12 bit and with a 25×/0.95 NA water immersion objective. Dextran Texas Red 70,000 MW (Invitrogen, Cat. No. D1830) at a concentration of 10 mg/mL was intravenously injected prior imaging. Texas Red was excited at 980 nm and detected at 674-711 nm.

Movies Correction and xy Migration Analysis
Time-lapse movies were corrected for xyz drift movements using Huygens Professional software program (version 20.10, Hilversum, Netherlands). For this correction we used the ROSA26 mTmG channel as a reference. Per corrected position, we made a maximum projection of 3 consecutive z-planes throughout the complete z-stack for the analysis, thus covering the complete imaging depth. Cells were manually tracked using MTrackJ plugin from ImageJ. A total of 120 to 370 cells were tracked per mouse in 3 mice. For each population (cells in tumor lobes or in the brain parenchyma), 350 cells were analyzed. The XY position was determined over time by the MTrackJ plugin and the speed and persistence for each individual cell was calculated with Excel (Microsoft, Redmond, DC, USA). For each population (cells in tumor lobes or in the brain parenchyma), we determined the percentage of migratory cells based on the number of tracked cells which speed was equal or higher than 2.5 µm over the total number of tracked cells per population.

Intracranial Injection
BCBM organoids were prepared as mentioned in 'Cranial window implantation and BCBM injection under the window'. Prior to surgery, mice received 0.067 mg/mL rimadyl (carprofen, Zoetis) in drinking water 1 day before injection and for 3 days post surgery. Additionally, 30 min before surgery and 8-12 h after surgery, mice were subcutaneously injected with 0.1 mg/kg temgesic (buprenorphine, Indivior Europe Limited). Mice were anesthetized via inhalation of 2% (v/v) isoflurane and eyes were covered with duratears (Alcon). After shaving the head, an incision was made in the skin to reveal the skull, followed by the application of a local anesthetic (0.01-0.02 mL). This was prepared with NaCl supplemented with 1 mg/mL Lidocaine (Fresenius Kabi) and 0.25 mg/mL Bupivacaine (Actavis, Aurobindo Pharma B.V.). After removal of the periosteum, a hole was drill 1.5 mm to the right and 1.5 mm posterior to the bregma. At a depth of 1 mm, we injected our tumor cells preparation (100,000 cells in 3 µL) with a 10 µL glass Hamilton syringe, 30G and point 4-style needle at a speed of 1 µL/minute. Skin was sutured after injection and the mouse was allowed to recover on a heating pad, being monitored closely the next days.

Magnetic Resonance Imaging
Magnetic resonance imaging was performed with a 7 Tesla BioSpec 70/20 USR (Bruker; Billerica, MA, USA). Imaging was performed with a T2 RARE sequence with a 39 ms echotime, 2200 ms repetition time and 8 averages. Paravision software (v6.0.1; nBruker) was used for image acquisition and MIPAV NIH software (Bethesda, USA) was used for tumor volume measurements. Mice were anesthetized with 2% (v/v) isoflurane during imaging, eyes were covered with duratears (Alcon) and the heart rate was monitored throughout the procedure.

Tumor Resection
This surgery was designed in close collaboration with a neurosurgeon to resemble the clinical setting. During this entire experiment, the researcher performing the surgery and MRI was not aware of the BCBM group from the time of intracranial injection until the 11 weeks after tumor resection. Mice were intracranially injected with 20,000 cells derived from either shScramble or shARPC3 BCBM organoids. The coordinates from the bregma were 1.5 mm to the right, 1.5 mm posterior and 0.5 mm depth. Once mice developed tumors with a volume of 10-18 mm 3 , BCBM resection was performed. Mice received 0.067 mg/mL rimadyl (carprofen, Zoetis) in drinking water 1 day before surgical resection and for 3 days post surgery. Moreover, 30 min before surgery and 8 h after surgery, mice were subcutaneously injected with 0.1 mg/kg temgesic (buprenorphine, Indivior Europe Limited). Mice were anesthetized via inhalation of 2% (v/v) isoflurane and their eyes were covered with duratears (Alcon). The mouse head was shaved and Rymadil (5 mg/kg) and warm NaCl were subcutaneously injected. Once placed on the stereotactic frame, an incision was made in the skin to reveal the skull. A local anesthetic which consisted of NaCl supplemented with 1 mg/mL Lidocaine (Fresenius Kabi) and 0.25 mg/mL Bupivacaine (Actavis, Aurobindo Pharma B.V.) was applied on the skull. Removal of the periosteum was performed with a cotton swab, followed by drilling of a 5 mm diameter circle on the right parietal bone. Bone and meninges were removed with thin forceps. With a vacuum pump, the BCBM was aspirated with a 100 µL sterile tip until no more tumor was visible. To stop bleeding, we used cold NaCl with eyespear pointed tips (Sugi) and once mostly dried we applied spongstan (Ethicon). Once the bleeding stopped and the surgery finished, we used a fluorescent stereomicroscope system (Nightsea fluorescent adapter, Lexington, MA, USA) to visualize how much fluorescent tumor was left behind. Skin was sutured and glued with histoacryl (Braun). After surgery, mice received subcutaneously warm NaCl while recovering on a heating pad. Mice were closely monitored over the following days.

Fluorescence Activated Cell Sorting
BCBM were isolated from mouse brains and minced manually on ice using sterile scalpels. All of these steps were executed inside a flow cabinet. Minced tumors were collected in a 15 mL falcon and enzymatically digested in 3 mL PBS supplemented with 10 µg/mL DNase I (Roche, Cat. No. 10104159001) and 0.96 mg/mL Liberase TH (Roche, Cat. No. 5401151001) at 37 • C for 15 min at 1050 rpm. The remaining pieces were resuspended with a 20-gauge needle. Supernatant was spun down and washed twice with a FACS buffer made of PBS with 2 mmol/L EDTA (Lonza AccuGENE, Cat. No. 51234) and 1% B27 (50× Gibco, Cat. No. 17504044). After the last wash, the pellet was resuspended in FACS buffer and filtered through the lid of a falcon tube with a cell strainer (Falcon, Cat. No. 352235). Sample was transferred into a polypropylene falcon (Falcon, Cat. No. 352063) and just prior to sorting, 20 nmol/L of TO-PRO Iodide-3 642/661 was added (Invitrogen, Cat. No. T3605). Cells were sorted on a FacsAria Fusion 1 (BD Biosciences) (San Jose, USA) and the gating strategy is illustrated in Figure S3. The purity check control was performed using the CFP positive population since we did not have enough CFP negative cells for this purpose ( Figure S3).  Table S1. Thermal cycle conditions were as follows: 2 min at 50 • C, 2 min at 95 • C followed by 45 cycles consisting of denaturation for 15 s at 95 • C, annealing for 1 min at 60 • C, and extension for 1 min at 72 • C. PCR reactions were concluded with incubation for 10 min at 72 • C. RDML raw data was processed using RDML-Ninja (version 0_9_3 , Amsterdam, The Netherlands) and LinReg software (version 2020, Amsterdam, The Netherlands) to determine the baselines and obtain the empirical primer efficiencies. qPCR fold changes between samples were obtained by ∆∆Ct calculations and corrected by primer efficiencies. Average values from the two housekeeping genes were calculated, normalized to E-cad HI tumor cells, log2 transformed and the average plotted. Statistical significance was calculated with GraphPad Prism (San Diego, CA, USA) package using one sample t-test, one-tailed.

RNA Bulk Sequencing and Bioinformatics
Illumina TruSeq mRNA libraries were generated and sequenced with 65 base single reads on a HiSeq 2500 using V4 chemistry (Illumina Inc., San Diego, CA, USA) as previously described [37]. Reads were mapped to the genome with Hisat version 2.1.0. (Amsterdam, Netherlands) Read counts were determined with Icount based on the mouse reference genome GRCm38_87. To correct for sequencing depth variability between samples, a DESeq median-of-ratios approach was implemented [38] followed by a log-transformation of the normalized counts. These corrections were necessary in order to obtain the normalized expression values. The singscore package [39] was used to determine the EMT scores based on the Ecad LO signature of PyMT breast tumors [27]. Statistical significance of both scores to 0 was calculated with one-sample t-test. DESeq2 [38] was used for the differential expression analysis. Pathway analysis was conducted with GSEA preranked [40] by using as metric -log10(FDR) divided by the sign of the log2FoldChange and the Hallmark gene sets as pathways [41].
Samples of the Beerling et al. [27] were downloaded from GEO number GSE77107. The reads were trimmed using Cutadapt [42] to remove any remaining adapter sequences and to filter reads shorter than 20 base pairs. The trimmed reads were aligned to the GRCm38 reference genome using STAR [43]. Quality control (QC) statistics from FastQC [44] and the above-mentioned tools were collected and summarized using MultiQC [45]. Gene expression counts were generated by featureCounts [46] using gene definitions from Ensembl GRCm38 version 100. Normalized expression values were obtained as described above. The entire analysis was implemented by Julian de Ruiter using Snakemake (snakemake version 5.19.2; wrapper version 0.63.0) (Amsterdam, Netherlands) [47] and is freely available on GitHub (https://github.com/jrderuiter/snakemake-rnaseq) (accessed on 4 June 2020).

Immunofluorescence and Confocal Imaging
Brain samples were fixed overnight in 4% periodate-lysine/4% paraformaldehyde [48] at 4 • C. The next day, tissues were washed with PBS, incubated overnight in 30% sucrose and frozen in OCT. Per tumor, 2 consecutive sections of 5 µm thickness were cut. After cutting, sections were hydrated twice with PBS for 15 min and mounted with anti-fading mounting medium (Vectashield, Vector laboratories (Cat. No. H-1000-10). Slides were imaged using inverted Leica TCS SP8 confocal microscope. H2B-Dendra2 was excited at 488 nm and collected at 495-545 nm. E-cad-mCFP signal was excited at 442 nm and collected at 454-479 nm with a line accumulation of 3. All images were acquired with a 25×/0.95 NA water immersion objective.

Human Samples and Multiplex Immunohistochemistry
All the human BCBM biospecimens were isolated at the Amsterdam UMC (location VUMC). Before the surgery of the brain tumors, patients at the A-UMC/VUMC are informed and have always had the opportunity to object to the use of their biospecimens. For all samples used in our study, patients did not object, and hence, they did therefore agree to the use of their biospecimens. This opt-out procedure is pursuant to the ethical rules and regulations of the Amsterdam UMC, location VUMC. Hence, the procedures comply both with (inter-) national legislative and ethical standards. In the database of the Amsterdam UMC, location VUMC, 8 samples of brain resections with a metastasis of mammary carcinoma were selected. These were all invasive ductal carcinomas, not otherwise specified, with a different receptor status: 2 ER-positive, 2 HER2-positive, 2 triple negative and 1 unknown receptor status. Immunohistochemistry of the Formalin Fixed Paraffin Embedded (FFPE) tumor samples was performed on a Discovery Ultra automated stainer (Ventana Medical Systems). Briefly, paraffin sections were cut with a 3 µm thickness, heated at 75 • C for 28 min and deparaffinized with an EZ prep solution (Ventana Medical Systems). Heat-induced antigen retrieval was carried out using Cell Conditioning 1 (CC1, Ventana Medical Systems) for 64 min at 95 • C. For the triple panel Zeb1 (DAB), Pankeratin (Yellow) and E-cadherin (Purple): Zeb1 was detected in the first sequence using clone E2G6Y (Cell Signaling, 1:200 dilution, 60 min at room temperature). Zeb1 bound antibody was visualized using Anti-Rabbit HQ (Ventana Medical systems) for 12 min at 37 • C followed by Anti-HQ HRP (Ventana Medical systems) for 12 min at 37 • C, followed by the ChromoMap DAB Detection kit (Ventana Medical Systems). In the second sequence of the triple staining procedure, Pankeratin was detected using clone AE1/AE3 (Abcam, 1:400 dilution, 60 min at 37 • C) Pankeratin was visualized using Anti-Mouse NP (Ventana Medical systems, Oro Valley, AZ, USA) for 12 min at 37 • C followed by Anti-NP AP (Ventana Medical systems) for 12 min at 37 • C, followed by Discovery Yellow Detection Kit (Ventana Medical Systems). For the third sequence, E-cadherin was detected using clone NCH-38 (Agilent/DAKO, 1:50 dilution, 60 min at 37 • C). Ecadherin was visualized using Anti-Mouse HQ (Ventana Medical systems) for 12 min at 37 • C followed by Anti-HQ HRP (Ventana Medical systems) for 12 min at 37 • C, followed by the Discovery Purple Detection Kit (Ventana Medical Systems). Slides were counterstained with Hematoxylin and Bluing Reagent (Ventana Medical Systems) and imaged with a 40× magnification with a PANNORAMIC ® 1000 scanner (3DHISTECH, Budapest, Hungary).

Intravital Microscopy Identifies Migratory Single Cells at the Invasive Front of BCBM
To characterize BCBM cells that invade the brain parenchyma, we performed a total of 6 serial intracranial transplantations of tumor pieces originally derived from end-stage primary ductal carcinomas that had developed in the mammary glands of PyMT E-cad-mCFP reporter mice [27] (Figure 1A). Fast-growing brain metastases were established after the 6th round of transplantation in the brain. From these tumors, we derived organoids with the aim to maintain 3D heterogenic characteristics of brain metastases. From here on, tumors that develop upon intracranial injection of the single cell suspension of brainenriched organoids are referred to as BCBMs. To study the behavior of BCBM cells in vivo, we intracranially injected these organoids labelled with a nuclear fluorescent marker (H2B-Dendra2) in ROSA26 mTmG FVB mice and monitored BCBM progression through a cranial imaging window using multi-photon microscopy. Two weeks after intracranial injection of BCBM organoids, we observed the formation of a tumor with individual lobes containing a high density of tumor cells ( Figure 1B). Importantly, we observed single tumor cells outside of the tumor lobes ( Figure 1B). These single tumor cells were associated with blood vessels (Figure 1B,C). To assess the behavior of these single cells, we performed time-lapse live imaging through the cranial imaging window for up to 12 h by multi-photon microscopy in ROSA26 mTmG mice (Figure 2A,B). We tracked the position of 350 cells within the tumor lobes and 350 cells at the invasive front over time in a total of 16 positions in 3 mice ( Figure 2C,D). Whilst the vast majority of cells in the tumor lobes were nonmigratory, a large proportion of the cancer cells in the brain parenchyma were motile (defined by cells with a speed equal to or higher than 2.5 µm per hour, Figure 2C,D). Moreover, cell migration in the parenchyma was highly persistent ( Figure 2E). Additionally, we frequently observed cells migrating along structures such as type I collagen fibers as observed by second harmonic generation (SHG) ( Figure 2F). Together, we identify a population of highly migratory cancer cells that escape from the BCBM lobes and invade the surrounding parenchyma.

Impeding Tumor Cell Migration Improves Surgical Resection
Considering that cancer cell invasiveness hurdles complete tumor resection and are associated with worse patient prognosis in BM and glioblastoma [49][50][51], we next tested whether impeding tumor cell migration would improve the efficiency of surgery in BCBM. Cell motility in BC, including PyMT breast tumors, was previously shown to be partly driven by the actin-related protein 2/3 complex (ARP2/3 complex), which mediates actin nucleation [52][53][54]. As such, reducing ARPC3 expression can impair cell migration without altering proliferation [55]. We therefore transduced our BCBM organoids with short-hairpin RNAs against ARPC3 (shARPC3). Western blot analysis confirmed the reduced expression of ARPC3 using two distinct shARPC3 (Figures 3A and S1). For further experiments, we worked with shARPC3 #2, which showed the highest knockdown efficiency. Next, we intracranially injected BCBM organoids engineered with either shScramble or shARPC3 in ROSA26 mTmG FVB mice. We confirmed that reducing ARPC3 expression did not alter tumor cell proliferation, based on Ki67 expression ( Figure 3B,C). Next, we used multiphoton microscopy to monitor BCBMs through a cranial window. In line with previous literature [52,53,55], reducing ARPC3 expression in BCBM cells significantly decreased the number of cells that invaded the surrounding parenchyma ( Figure 3D,E).    Figure S1 for uncropped blot. (B) Representative images and (C) quantification of Ki67 immunohistochemistry staining in BCBM derived from shScramble or shARPC3 organoids intracranially injected into recipient mice. Data are presented as mean +/− standard-error of the mean (SEM); n = 5 mice per group; p-value was determined using a Mann-Whitney t-test, twotailed. (D) Quantification of the number of invasive tumor cells observed by intravital imaging of ROSA26 mTmG mice injected with shScramble or shARPC3 BCBM organoids. Data are presented as mean +/− SEM; n = 4 mice for shScramble and n = 3 mice for shARPC3 #2; p-value was determined using a Mann-Whitney t-test, one-tailed. (E) Representative multi-photon images of BCBM derived from shScramble and shARPC3 BCBM organoids intracranially injected into ROSA26 mTmG mice.
Next, we tested whether impairing migration and invasion by ARPC3 knockdown would improve surgical removal of BCBM. We intracranially injected BCBM organoids engineered with either shScramble or shARPC3 in 16 FVB mice and monitored BCBM growth using MRI ( Figure 4A). Whilst blinded for the experimental group, we performed surgical resection of the BCBM when tumors reached a volume of 10-18 mm3. The average tumor volume was equal between both groups ( Figure 4B). Importantly, in line with the absence of significant changes in proliferation ( Figure 3B,C), the time for tumors to reach the same average volume prior to resection did not differ between the shScramble and shARPC3 tumors, confirming that reducing ARPC3 expression does not influence tumor growth ( Figure 4B,C). From the shARPC3 group, one mouse did not grow a BCBM and another mouse was sacrificed for BCBM-unrelated reasons; therefore, we excluded those mice from the experiment. Since our BCBM cells express H2B-Dendra2, we used fluorescent microscopy to score the amount of tumor cells that were left behind after surgical resection of the BCBMs ( Figure 4D). This revealed that in all mice of the shScramble group, some BCBM cells remained in the surgical cavity; however, this was not the case in the majority of the shARPC3 mice ( Figure 4D,E). Strikingly, even a week after surgical removal, those cells that were left behind were not detected by MRI ( Figure 4F). This suggests that also in the clinic, where MRI is used as a measure of complete BCBM resection, single BCBM cells that invade the parenchyma and are left behind upon resection may not be detected. Additionally, similar to BCBM patients, we observed BCBM recurrence at later time-points with 50% of the shScramble mice recurring versus 33% in the shARPC3 group during the 11 weeks follow-up post-surgery ( Figure 4F,G). Reduced expression of ARPC3 in the recurrent tumors was maintained, as confirmed by RT-qPCR ( Figure S2). Together, these data suggest that blocking cell invasion may improve surgical resection of the BCBM.

BCBM Invading Tumor Cells Are in a Mesenchymal State
Next, we investigated the role of EMT in these invasive tumor cells using the E-cad-mCFP reporter, as we previously did in the primary tumor [26,27]. In the tumor lobes, cancer cells express E-cad-mCFP ( Figure 5A). Importantly, we observed that all cells that invaded the parenchyma lost E-cad-mCFP expression (arrowheads, Figure 5A,B). From here on, we refer to cells with high E-cad-mCFP expression as E-cad HI cells and cells that lost E-cad-mCFP expression as E-cad LO cells. In line with our previous observations in primary PyMT breast tumors [26,27], E-cad HI cells were on average non-motile, whilst a large fraction of the E-cad LO cells were migratory ( Figure 5C,D).

BCBM Invading Tumor Cells Are in a Mesenchymal State
Next, we investigated the role of EMT in these invasive tumor cells using the mCFP reporter, as we previously did in the primary tumor [26,27]. In the tumo cancer cells express E-cad-mCFP ( Figure 5A). Importantly, we observed that all ce invaded the parenchyma lost E-cad-mCFP expression (arrowheads, Figure 5A,B here on, we refer to cells with high E-cad-mCFP expression as E-cad HI cells and ce lost E-cad-mCFP expression as E-cad LO cells. In line with our previous observat primary PyMT breast tumors [26,27], E-cad HI cells were on average non-motile, w large fraction of the E-cad LO cells were migratory ( Figure 5C,D).
To characterize the Epithelial/Mesenchymal (E/M) states of E-cad HI cells and cells, we isolated both populations using fluorescence activated cell sorting (FAC ures 5E and S3). Imaging analysis of E-cad-mCFP confirmed the successful isolatio cad HI and Ecad LO cells ( Figure S3). Next, we performed bulk mRNA sequencing o cell populations ( Figure 5F,G) and analyzed the differentially expressed genes t refer to as the E-cad LO gene set. The E-cad LO gene set contains typical EMT genes, E-cadherin (Cdh1), N-cadherin (Cdh2), Vimentin (Vim), Fibronectin (Fn1) and tra tion factors such as Snail (Snai1), Zeb1 and Zeb2 ( Figure 5G). We confirmed th  p-values were determined with a one-sample t-test to calculate the significance to 0.
To characterize the Epithelial/Mesenchymal (E/M) states of E-cad HI cells and Ecad LO cells, we isolated both populations using fluorescence activated cell sorting (FACS) (Figures 5E and S3). Imaging analysis of E-cad-mCFP confirmed the successful isolation of E-cad HI and Ecad LO cells ( Figure S3). Next, we performed bulk mRNA sequencing on both cell populations ( Figure 5F,G) and analyzed the differentially expressed genes that we refer to as the E-cad LO gene set. The E-cad LO gene set contains typical EMT genes, such as E-cadherin (Cdh1), N-cadherin (Cdh2), Vimentin (Vim), Fibronectin (Fn1) and transcription factors such as Snail (Snai1), Zeb1 and Zeb2 ( Figure 5G). We confirmed the bulk mRNA sequencing results by RT-qPCR ( Figure 5G). In line with this data, the E-cad LO gene set was enriched for genes involved in the "Hallmark Epithelial_Mesenchymal_Transition" [42] ( Figure 5H). Lastly, we compared the expression profile of E-cad LO cells isolated from BCBM to the expression profile of their counterparts isolated from primary PyMT tumors. Importantly, in our prior study, we showed that E-cad LO cells have low expression levels of the epithelial marker E-cadherin and high expression of the mesenchymal marker Vimentin [27]. In line with the enrichment of genes involved in EMT, we found that the E-cad LO gene set but not E-cad HI gene set of BCBM tumors is enriched in genes that were found to be significantly upregulated in the E-cad LO gene set of the primary breast tumor [27]. Collectively, these data show that the differential expression of E-cad-mCFP in Ecad HI and E-cad LO cells reports the epithelial and mesenchymal states of cells, respectively, in our PyMT BCBM model.

Mesenchymal Cells Can Form New Tumors and Revert to an Epithelial State
Since the presence of invasive BCBM cancer cells is involved in relapse following surgical resection (Figure 4 and [49,51]), we next investigated whether the E-cad LO cells have the capacity to grow BCBMs. We isolated~2000 E-cad LO cells from BCBM grown in ROSA26 mTmG FVB mice by FACS and injected this population intracranially in tumor free FVB mice ( Figure 6A). An assessment of the purity suggested a potential contamination of the sorted fraction of at most 2.5% with other cells. Therefore, we injected 2.5% of E-cad HI cells in tumor free mice to serve as a purity control group ( Figure S3). In 75% of mice injected with E-cad LO cells, BCBMs had developed during approximately 7 weeks post-intracranial injection, while no tumors developed during the same follow-up period in the purity control group ( Figure 6B). Next, we isolated the BCBMs generated by the E-cad LO cells and studied the expression of the E-cad-mCFP reporter ( Figure 6C). In all tumors, we detected the E-cad-mCFP reporter in tumor cells, which was further validated by staining tumors with an E-cadherin antibody ( Figure 6C). This demonstrated that in our BCBM model, mesenchymal cells have the ability to form new tumors and to concomitantly revert to an epithelial phenotype, a process often referred to as mesenchymal-to-epithelial transition (MET).

Tumor Cells Undergoing EMT Are Present at the Invasion Front in Human BCBM
Finally, we investigated whether mesenchymal infiltrative tumor cells are also present in BCBM patients (Figure 7). We obtained tumor resections from 8 patients with BCBM. Using multiplex immunohistochemistry, we simultaneously stained for Pankeratin and E-cad to mark cells in an epithelial state, and Zeb1 to mark cells in a mesenchymal state ( Figure 7A). In the tumor mass, we indeed observed that cancer cells stain positive for Pankeratin and E-cad and negative for Zeb1, as expected for epithelial cancer cells. Moreover, and in line with previous observations [56], the brain parenchyma that surrounds the BCBM contained cells that were positive for Zeb1 and negative for E-cad. Yet, most of these cells were also negative for Pankeratin, suggesting that those cells belong to a population of glial and immune cells that also express Zeb1 [57]. However, in seven out of eight patients, we also observed a population of Zeb1 positive; E-cad negative cells with moderate to high expression of the epithelial marker Pankeratin ( Figure 7B). In line with the data of the infiltrative E-cad LO cells that are present in the brain parenchyma in our mouse model, these Pankeratin-positive, Zeb1-positive, E-cad-negative cells have a spindle shape nucleus, indicative of a migratory state (encircled in Figure 7A). Together, this suggests that BCBM patients also have infiltrative tumor cells in EMT in the brain parenchyma that surrounds the BCBM.

Conclusions
The role of EMT in the formation of metastases has been heavily debated by the field. However, the role of EMT at later stages of the metastatic cascade in growing metastatic masses remains under-studied. Here, we used intravital microscopy combined with mRNA sequencing to demonstrate that cells undergoing EMT are present in established BCBM, invade the brain parenchyma and have the capacity to revert to an epithelial state when forming secondary lesions. We also reveal that the presence of these invasive cells impairs the successful removal of BCBM lesions. Importantly, even if surgery is followed by chemotherapy or radiotherapy, these invasive cells may not be efficiently depleted, since EMT endow cells with properties that can render them resistant to those therapies [21,[58][59][60]. Therefore, targeting EMT can improve surgical resection and may sensitize cells that are left behind to subsequent therapies.
In our study, we combined an E-cad fluorescent reporter with mRNA sequencing to study EMT in BCBM. We showed that the E-cad LO cells upregulate mesenchymal markers and are migratory. However, recent work has provided evidence that EMT is not a binary switch but that cells can adopt various hybrid E/M states with distinct metastatic and chemoresistance properties [17,19,25,61,62]. Here, in our mouse experiments, we have neither investigated the presence nor the role of cells with different E/M states. However, in our analysis of human BCBM, we observed cancer cells that escaped from the tumor mass and invaded the brain parenchyma contained a typical mesenchymal profile of expressing Zeb1 and losing E-cadherin, but interestingly still expressing the epithelial marker Pankeratin. Expression of the latter marker suggests that these infiltrating cells may not have lost all epithelial traits which can indicate a hybrid state. Future studies should shed light on the role of specific hybrid E/M states in BCBM and their distinct contributions to patient relapse following surgery. Additionally, developing therapeutic strategies to block EMT and/or mesenchymal cells is required to target this process and to ultimately improve BCBM patients' outcomes. Nevertheless, our data warrant future studies to target EMT processes for better treatment of BCBM.

Supplementary Materials:
The following supporting information can be downloaded at: https: //www.mdpi.com/article/10.3390/cancers14133115/s1, Figure S1: Western blot analysis of ARPC3 knockdown in BCBM organoids. (A) Uncropped Western blot showing ARPC3 expression in BCBM organoids engineered with all ARPC3 shRNA's. Tubulin was used as a loading control; Figure S2: ARPC3 knockdown validation by RT-qPCR in recurrent tumors. Relative Arpc3 expression in recurrent tumors in mice injected with BCBM organoids engineered with either the shARPC3 or shScramble constructs. Data are presented as mean +/− SD. n = 1 mouse with 3 technical replicates; Figure S3: Gating strategy of BCBM tumors FACS (A) Gating strategy applied to BCBM tumors to sort Ecad HI and Ecad LO tumor cells. (B) Representative images of Ecad HI and Ecad LO sorted tumor cells. (C) Gating strategy to analyze the purity of our FACS strategy. Due to the limited amount of Ecad LO cells that we could obtain for this experiment, we performed the purity control using the Ecad HI tumor cells; Table S1: Primer sequences.