Investigating circulating tumor cells and distant metastases in patient-derived orthotopic xenograft models of triple-negative breast cancer

Despite the tremendous progress made in the diagnosis and treatment of breast cancer, tumors of the breast still remain one of the leading causes of cancer-related deaths in women [1]. The intertumoral and intratumoral molecular heterogeneity of breast cancer challenges its diagnosis and effective treatment [2‐9]. Tailored therapies, such as hormone therapies (e.g., tamoxifen and inhibitors of the enzyme aromatase, involved in estrogen synthesis) for ER-positive disease and trastuzumab (Herceptin®) for HER2-overexpressing breast cancer have led to considerable success in treating some subtypes of breast cancer. However, drug resistance to these regimens can represent a major hurdle to successful treatment [10‐15]. Most importantly, there is still no good targeted therapy for triple-negative breast cancer (TNBC), a very aggressive subtype that remains difficult to treat [16, 17]. Due to the very aggressive nature of TNBC and the lack of well-established molecular therapeutic targets, patients with TNBC tend to have a relatively poorer outcome compared to patients with other subtypes [18, 19]. In breast cancer, and especially in TNBC, dissemination and metastatic growth of tumors at distant sites is the major cause of patient mortality [20]. Despite chemotherapy, fewer than 30% of women diagnosed with metastatic TNBC will survive beyond 3 years, and, unfortunately, almost all women with metastatic TNBC will ultimately succumb to their metastatic disease [21‐23]. Although newer therapies and combinations of therapies for TNBC are under active investigation and hold future promise, including the use of poly (ADP-ribose) polymerase (PARP) inhibitors for TNBC patients with homologous recombination DNA repair-deficient cancers associated with BRCA1 mutations, the use of immune checkpoint inhibitors, approaches that target other signaling pathways, or combination therapies, responses are still only observed in a small fraction of patients with advanced TNBC [24‐30].

Factors that drive tumor metastasis have been a subject of intense scrutiny and research. As circulating tumor cells (CTCs) are considered contributory precursors that seed metastases in many cancers, including breast cancer, studying the biology of CTCs has provided vital clues regarding cancer metastasis [31]. Multiple mouse models may be used to study breast cancer biology, including syngeneic models (immunocompetent models generated from murine breast cancer cell lines, such as 4T1 cells), environmentally induced tumor models, transgenic models (models expressing mouse oncogenes, such as the polyomavirus middle T antigen controlled by the mouse mammary tumor virus long terminal repeat promoter, MMTV-PyMT model), genetically engineered mouse models (GEMMs), cell line-derived xenografts, and patient-derived xenografts [32‐39]. However, the use of in vivo models to study the shedding and biology of human CTCs requires either human breast cancer cell line-derived xenografts [40] or patient-derived xenografts (PDXs). Generation of PDX models involves the transplantation of primary human cancer cells or pieces of tumor tissue into immunocompromised mice. Although most PDX models are generated in mice lacking a functional human immune system, they are still considered to be highly clinically relevant [41, 42], particularly when implanted orthotopically (e.g., human breast tumor tissue from the operating room implanted into the mammary fat pads of mice). Orthotopically implanted PDX models, called patient-derived orthotopic xenograft (PDOX) models, have been shown to recapitulate critical histological, genomic, transcriptomic, and proteomic features of the patients’ tumors from which they were derived [43‐45]; they are also better models of human metastatic disease [46] and of response to anti-cancer therapies [41, 45, 47] and, in the case of TNBC PDOX models, represent more aggressive phenotypes [48]. Currently, there is renewed interest in utilizing PDX models as evident in US National Cancer Institute’s (NCI) plan to replace their NCI-60 cell line resource with PDX samples [49]. However, the background strain of mouse used for PDX studies is of key importance. NOD scid gamma (NSG) mice have been shown to be best at recapitulating the entire metastatic process in breast cancer from implantation in the mammary fat pad to distant metastatic spread, including development of metastases in cell line models that had not been previously associated with metastatic spread [50]. So while there has been an increasing interest in using PDOX models of breast cancer to study human CTCs captured from mouse blood and/or human disseminated tumor cells (DTCs) from mouse bone marrow [51‐57], the studies from these other groups were performed in non-NSG mouse models. We had previously described the isolation of CTCs from two NSG mice from a single TNBC PDOX tumor [58]. In this study, we use NSG mice to examine the distribution of distant metastatic spread in 57 mice from seven PDOX models of TNBC and to analyze CTC shedding in 37 of these mice.

Multiple technologies have been developed to enable CTC isolation from the peripheral blood of human patients with solid tumors. Approaches to capture and/or identify these rare cells rely on positive or negative cell selection, density gradient centrifugation, microfiltration, microfluidic or electrophoretic separation, direct imaging, or functional assays [59‐61]. Here, we isolate human CTCs from mouse blood in PDOX models using the Vortex platform, a technology originally developed to allow fast and label-free isolation of human CTCs over a broad range of blood volumes (200 μL to 16 mL). CTCs are enriched from whole blood based on size and deformability, using inertial microfluidics combined with microscale vortices [62]. This technique had been successfully used to isolate human CTCs from blood samples from patients with metastatic breast, lung, colorectal, and prostate cancer [63‐67]. In the present study, isolation of human CTCs from mouse blood was first optimized using a human breast cancer cell line-derived xenograft model of TNBC, generated from MDA-MB-231 cells, that we had previously shown to shed CTCs and metastasize to the lung [40]. Based on these experiments, a set of seven PDOX models of TNBC, including four that had previously undergone global genomic and transcriptomic analyses [44], were used for replicate isolation and characterization of human CTCs and for identification of sites of distant metastases.

Circulating tumor cells, detected most often in metastatic cancer [71], are a key player in the metastatic cascade [31]. We have recently reported on label-free isolation and analysis of CTCs from two PDOX tumor-bearing mice as part of an evaluation of Vortex’ fully automated VTX-1 Liquid Biopsy System [58]. Here, we investigated CTC shedding and metastases in TNBC PDOX models, examining 57 mice from a cohort of seven TNBC PDOX models for distant metastases in serially sectioned lung, liver, and brain, and using a marker-agnostic CTC isolation platform to isolate and analyze CTCs from 37 of these mice.

As far as we are aware, seven previous studies have used breast cancer PDOX models to capture and analyze CTCs from mouse blood or DTCs from mouse bone marrow. One study [54] used a previous TP53 wildtype PDOX model of non-basal TNBC to generate cell lines that were transduced with lentivirus encoding CBR-luc and mCherry (FUW-CBR-luc-mCherry) and then infected with control retroviruses or retroviruses encoding p53-specific short hairpin RNAs (shRNAs) to knock down p53. They then implanted these wildtype and p53-deficient transduced cells into cleared mammary fat pads of NOD/SCID mice whose stroma was humanized using immortalized and irradiated human mammary stromal fibroblasts derived from a patient undergoing a reduction mammoplasty. Tumor growth; metastases to the lung, bone, liver, brain, and axillary lymph node; and CTC shedding (quantified using flow cytometry) were compared between wildtype and p53-deficient tumors. Although their PDOX models were generated very differently than ours, similar to our findings, metastatic tumor was identified at different sites in differing fractions of mice. Moreover, although CTC shedding increased with time, a finding more pronounced in mice whose tumors were p53-deficient, with numbers of CTCs in the less than 15 range at 9 weeks to the less than 220 range at 18 weeks, CTC shedding appeared more related to total primary and metastatic tumor burden. A different study [55] used PDOX models derived from five tumors of different breast cancer molecular subtypes that were implanted in humanized mammary fat pads (previously cleared and injected with GFP-labeled immortalized human fibroblasts) of NOD/SCID mice to investigate DTCs and distant metastases. Here, mouse bone marrow was analyzed by performing qRT-PCR for human transcripts as well as microarray analysis of gene expression in primary tumors, the bone marrow, and a metastatic lesion in one mouse. Only two PDOX tumor models developed metastases, both of which had human DTCs and which had variable sites of metastases. Global gene expression between diverse sources in the one mouse tested, a TNBC PDOX model, was also variable, but seemed to show patterns of human gene expression “reminiscent of a ‘mesenchymal-like’ phenotype in multiple animals across multiple passages,” with the authors suggesting that DTCs from the bone marrow may have undergone a phenotypic transition that enabled migration to and survival in the bone marrow, similar to many of the CTCs we observed in our current study. Another study [52] measured CTCs from blood and DTCs from the bone marrow, detected using anti-human pan-cytokeratin immunohistochemistry (IHC), in 18 different breast cancer PDOX models, 13 of which were TNBC PDOX models, generated from human breast tumors implanted into the cleared mammary fat pads of SCID/beige mice. Again, differing fractions of individual mice in each model had detectable CTCs (overall, present in 83% of PDOX models) and/or DTCs (overall, present in 63% of PDOX models) and lung metastases (overall, present in 50% of PDOX models). Although no specific mesenchymal-like IHC marker was used, there was still a strong association between CTC and DTC detection. Importantly, all mice with lung metastases had detectable CTCs but not all mice with detectable CTCs had lung metastases, similar to our findings. A strong association between CTC clusters and the presence of lung metastases was also found. In a fourth study [51], single-cell suspensions of seven patient-derived breast tumors were orthotopically injected into SCID mice; two PDOX tumors developed CTCs as detected by an EpCAM-based platform, and no metastases were observed in any of the seven mice. An important study by the Werb group [53] using three PDOX models of TNBC in NOD/SCID mice showed that all models shed CTCs into blood, with variable CTC shedding (7/31 mice = 23% in one TNBC model had detectable cancer cells in their peripheral blood; 1/19 mice = 5% in another TNBC model had detectable cancer cells in their peripheral blood; and 3/22 mice = 14% in a third TNBC model had detectable cancer cells in their peripheral blood). All models developed metastases in different organs, but again, like our study, showed variability among individual mice within a TNBC model. Excitingly, single-cell gene expression signatures of cells from tissues with low and high metastatic burden and CTCs showed that a subpopulation of stem-like CTCs may potentially represent “metastatic seeder cells.” A recent study of CTCs in non-NSG mice [56] used the same PDOX models described in their previous work [45, 52] but with CTCs isolated using a different CTC isolation platform that improved on CTC number and allowed identification of the PIK3CA T1035A mutation in single cells from CTCs, primary tumors, and lung metastases in one of their PDOX models. Finally, an interesting new study that analyzed human RNA transcripts in the blood (positive for CTCs), bone marrow (positive for DTCs), and tumor in severe combined immunodeficient (SCID) xenograft mouse models found that CTCs from an ER-positive PDOX model of lobular carcinoma expressed a possibly uncoupled epithelial-mesenchymal plasticity and lower expression of stem cell-like markers than CTCs from a cell line-derived TNBC xenograft [57].

A challenge when studying in vivo CTC models has been identifying an appropriate route of blood collection for maximal CTC yield for downstream analyses. We addressed this question first by using a metastatic, human TNBC cell line, MDA-MB-231, and discovered that blood collected via cardiac puncture yielded higher numbers of CTCs and CTC clusters, with some clusters heterotypic and containing both CTCs and WBCs. The importance of blood collection route for CTC analyses in mouse xenograft models was nicely addressed using cell line-derived xenografts and an EpCAM-based CTC isolation platform [51], concluding that cardiac puncture was the best route for obtaining CTCs compared to tail vein, retroorbital, and jugular vein aspiration. Similar to our findings, blood samples from the retroorbital venous plexus were uniformly contaminated by CTC-like normal murine epithelial cells in control mice without tumors. They also found that jugular vein aspiration yielded CTCs in only 15% of samples and also showed normal murine epithelial cell contamination from jugular vein aspiration in control mice. To our knowledge, our results are the first to address blood collection route using a marker-agnostic CTC isolation platform. While the cardiac route of blood collection is superior in its CTC yield, a major limitation of this route is that, as used here, it is a terminal method of blood collection, thereby limiting its application for serial CTC monitoring.

Analyzing CTCs and distant metastases in different breast cancer PDOX models also depends on the genetic background of the mouse model selected. A recent study highlighted challenges and limitations of metastatic TNBC PDOX models using SCID mice [72]. The authors discussed several challenges that included low rate of metastases, long periods for metastases in the lungs to be detected, and development of thymic lymphomas of mouse origin, using their SCID mouse models. SCID mice may also develop spontaneous non-thymic tumors, including mammary adenocarcinomas [73]. Pertinently, while younger NSG models do not develop spontaneous tumors, a recent study of aging female NSG mice showed that tumors may indeed develop in mice of median age 52 weeks, when age was available, including 31% developing spontaneous mammary gland neoplasms [74], a recent finding for which the PDOX community should be made aware.

In another study using NCr nude and SCID mice, only 2/7 PDOX models yielded CTCs and none had detectable distant metastases [51]. Addressing the issue of mouse model background, the Lippman group showed that NSG mice are superior for generating distant metastases in cell line- and patient-derived orthotopic xenograft models [50]. Using NSG mice for our study, most PDOX tumors grew aggressively and produced large tumors within 3 months of implantation, with detectable metastatic lesions in 6/7 of our TNBC PDOX tumor models.

In our study, the percentage of PDOX tumor-bearing mice with metastases ranged from 78% for SUTI319 to 0 for SUTI368. It is interesting to note that although CTCs were detected in 5/5 of SUTI368 tumor-bearing mice, no distant metastases were detected in eight SUTI368 PDOX mice. While serial sectioning of organs was sufficient to detect metastases in the majority of our PDOX models, we may have missed micrometastases in un-sectioned portions of an organ that perhaps a more sensitive PCR-based approach, based on detecting human DNA in mouse whole organs, could have potentially detected, including PDOX model SUTI368 [75]. However, a more likely explanation for the presence of CTCs in 5/5 mice and lack of metastases in 0/8 mice from this model is that metastasis is an inefficient process [76]. In a classic murine melanoma model with intravital video-microscopy, metastatic inefficiency was found to be mainly due to (1) lack of initiation of growth (i.e., dormancy) of solitary cancer cells in a distant organ and (2) lack of growth of initial micrometastases into macroscopic tumors and vulnerability to early destruction [77]. This has also been confirmed in multiple other tumor models, including breast cancer [78], and additional mechanisms have been proposed as to why only a subset of circulating tumor cells may be associated with the development of metastases [79‐81].

While the SUTI368 model is capable of generating CTCs, the tumor cells in circulation may not be capable of establishing metastatic growth because they may lack the ability to initiate the critical tumor-host interactions or may not possess intrinsic biomechanical or other molecular properties necessary for establishing metastases [82‐85]. Moreover, it is interesting to note that even when all the individual animals were implanted with tumor tissue fragments from the same PDOX tumor, there were differences in tumor growth rates between the individual tumor-bearing animals as well as differences in the development of and site of metastases within the same PDOX tumor model. Heterogeneity in the amount of tumor tissue and associated stromal tissue in each fragment implanted, differences in the number of viable tumor cells, and the molecular heterogeneity of tumor cells in individual tumor fragments could trigger such different growth rates in vivo. To overcome such discrepancies, we are also using single-cell suspensions of tumor tissue for generating and passaging more uniform PDOX tumors (data not shown).

CTCs can exist as single cells or clusters, and CTC clusters are predicted to have a greater potential to establish metastases than individual CTCs [86‐90]. In mouse models of breast cancer, CTC clusters appear to be derived from oligoclonal groupings of primary tumor cells and have been shown to be a highly metastasis-competent subset of CTCs, compared to single circulating breast cancer cells [86, 87]. CTC clusters also may provide additional prognostic value in human breast cancer [91‐93], and in patients with breast, prostate, and small cell lung cancer, the presence of single or increasing numbers of CTC clusters in sampled blood is found to correlate significantly with reduced progression-free survival rates [86, 91‐96]. As a result, there has been an interest in isolating and understanding CTC clusters in different cancers and, in addition to filter and other technologies that identify clusters, some CTC isolation technologies have been specifically designed for isolating CTC clusters [97‐99] or exploring their behavior [100]. In our studies, using the Vortex CTC isolation platform, we demonstrate rapid and efficient isolation of CTCs and CTC clusters from both cell line-derived xenograft and PDOX models of breast cancer. As this platform is marker-agnostic, we were able to capture heterogeneous clusters of CTCs that included individual cells that varied in the expression of CTC markers tested, demonstrating its utility for examining the biology and heterogeneity of CTC clusters in different cancer models.

Another notable finding in our current study is the difference in the heterogeneous distribution of cytokeratin (epithelial), vimentin (mesenchymal), and mixed cytokeratin and vimentin markers in CTCs and CTC clusters within and between the TNBC PDOX tumors. For example, while some CTCs from these PDOX tumor models showed high expression of the epithelial marker cytokeratin or the mesenchymal marker vimentin, many CTCs stained strongly for both cytokeratin and vimentin. The distribution of epithelial-like, mesenchymal-like, and mixed epithelial and mesenchymal CTC phenotypes represents an important aspect of CTC heterogeneity. CTC heterogeneity, including differences in gene expression of vimentin and five other EMT markers, was previously highlighted by our group using high-dimensional single-cell transcriptional profiling of human CTCs from breast cancer patients, where cell-to-cell heterogeneity of CTCs was noted even within the same blood draw [101]. This was also demonstrated through the work of other groups that has shown mixtures of epithelial- and mesenchymal-like CTCs in human blood samples of breast and other cancers [102‐104]. It has been postulated that hybrid EMT phenotypes may promote the development of cancer metastases [105‐107], although how the role of EMT plasticity and the different EMT states precisely contribute to the metastatic process still remains under active investigation [108‐114]. Heterogeneity among CTCs and resulting metastases may also be attributed to (i) differences in methylation patterns of specific genes [115], (ii) selective and differential expression of particular genes in a subset of CTCs [101, 116‐119], and/or (iii) expression of certain genes at different stages of the disease [120‐123].

Our data describes the heterogeneous distribution of lung, liver, and brain metastases in a group of seven TNBC PDOX models and confirms the shedding of CTCs captured by a marker-agnostic CTC isolation platform. It supports further use of PDOX models, generated in NSG mice, to study PDOX-derived CTCs, distant metastases, and for testing the impact and outcome of different anti-cancer agents on CTC shedding and metastasis in breast cancer.