Background
Metastasis is the primary cause of human cancer mortality, accounting for >90% of deaths due to cancer [
1]. There is now abundant evidence that, independent of the process of cellular transformation, the metastasis phenotype is genetically controlled [
2]. Metastasis is a multistep process that involves local tumor invasion followed by dissemination to, and re-establishment at, distant sites. Families of genes have been described which have no effect on cell proliferation but which can suppress or promote metastasis [
3,
4]. Thus, targeting metastasis may prove to be effective in reducing cancer mortality if specific targets can be identified that suppress this phenotype. Here, we present a robust
in vivo system for rapidly and accurately evaluating the effectiveness of candidate suppressor molecules.
Much of the analysis of metastasis pathways is conducted in tightly controlled
in vitro cell systems, usually involving overexpression or ablation of a particular gene. Assays such as wound healing, transwell motility, invasion assays and hanging drop assays have been developed which provide readouts of cellular phenotypes related to metastasis [
5‐
7]. These assays, however, do not address the issue of intravasation of tumor cells into blood vessels and extravasation into distant organs, a process requiring an
in vivo assay system. Typically, such assays are performed in mice using experimental or spontaneous metastasis models [
8,
9]. While it is ultimately necessary to demonstrate that a pathway identified
in vitro also affects invasion and metastasis
in vivo, mouse models have significant drawbacks: 1) it is difficult to study early stages of the process where it is necessary to rapidly evaluate whether a particular drug or genetic manipulation has affected the metastasis phenotype, 2) evaluating the complete process in mice can require up to 6 months (depending on the cell system), 3) these experiments are expensive, immunosuppressed mice are required to study human cells and per diem charges in barrier facilities are costly, 4)
in vivo imaging of small metastatic lesions is not possible in the deep tissues of the mouse, thus typically requiring termination and autopsy, thus extrapolation across experimental populations to realize the result, 5) popular immunosuppressed mice such as, nude (nu/nu), the severe combined immunodeficiency (SCID), or mice null for the recombination activating gene (Rag), have residual immune competence, which can actually prevent metastasis and, 6) the cohort size in these experiments is often pragmatically limited by high costs, thus statistical verification of metastasis modulation cannot be adequately assessed when the effect is mild.
Zebrafish provide an experimentally and genetically tractable animal model of a wide variety of human diseases [
10]. Recent studies have demonstrated that zebrafish form spontaneous tumors with similar histopathological and gene expression profiles as human tumors [
11‐
13]. The zebrafish-cancer model overcomes the drawbacks of murine xenograft models and offers alternative options for studying human tumor angiogenesis and metastasis [
14‐
21]. Following early reports of the application of zebrafish to evaluate metastasis [
22], we now tested whether metastasis in fish faithfully reports the metastatic potential of a broad range of cancer cells. To do so, we correlated
in vitro invasion efficacy to
in vivo metastasis metrics following manipulation of the metastatic phenotype. Without exception, we show that gene manipulations that affect
in vitro invasion, alter metastasis in fish in a corresponding manner, demonstrating that the zebrafish is a tractable model to assay metastatic potential of human cancer cells. We also show that primary human cancer cells can metastasize in fish and that this ability can be used to predict metastatic potential in a clinical setting.
Discussion
Dissection of the functional aspects of genes that impact the metastasis phenotype requires a robust assay for tumor spread. While it is accepted that, in the final analysis, rodent models should be used to evaluate metastasis, this approach is costly and inefficient as an up-front assay to determine whether a particular genetic manipulation affects the metastasis phenotype. The spontaneous metastasis assay [
8,
9] has shortcomings, since it cannot be used to evaluate intravasation into the blood vessels. The zebrafish model, on the other hand provides a solution to the time-consuming and costly mouse experiments, since in many cases the assay can be performed within 24–36 hours of xenotransplantation, in large cohorts of fish, providing statistical power to the results. Although we have only studied cancer cells from breast, pancreas, colon and sarcomas thus far, in all cases
in vitro invasion ability correlated with the metastatic potential of tumor cells to spread
in vivo. Importantly, we have shown that genetic manipulations of human cancer cells which affect invasion, also affect metastasis in fish. Although the technical dexterity needed to inject 48 hpf zebrafish can be demanding, the absence of an adaptive immune response for the first 14 days post fertilization (dpf) [
33] avoids side effects and dosing issues related to using immunosuppressants [
34,
35]. Early studies targeting the yolk sac as a site of injection truly challenged cancer cells to enter the blood stream. The blood supply to the yolk sac is extensive since this sustains the fish for the first 5 dpf and maximizes the opportunity for intravasation. It has been shown using the
cloche mutant fish, which do not develop a vasculature or circulation, that metastatic human cells injected into the yolk sac cannot metastasize in these fish, demonstrating the requirement for a functional circulatory system in this process [
15]. Injection into the yolk sac, however, has complications apparently associated with poor resealing of the yolk sac membrane which leads to spillage of the cancer cells or yolk sac contents. The perivitelline space between the body of the fish and the yolk sac provides an alternative, which does not suffer from these associated problems. The technical challenge is successfully targeting the perivitelline space, and avoiding injection directly into the circulatory system. For this reason we examined fish after 12 hours for the presence of cells in the vasculature and excluded these fish from the analysis. Since the injection involves large numbers of fish, excluding those that have been compromised during the injection process does not have any impact on the final analysis.
Although zebrafish have been used as a model for metastasis previously, protocols between different groups were not consistent in terms of the number of cells injected, the site of injection, the age of the fish used and the method of quantitation of metastasis [
18,
22]. To evaluate this metastasis model more robustly, we have used a standardized protocol with a variety of different cancer cells and cell systems. In this report we clearly demonstrate that metastasis in the zebrafish correlates with
in vitro invasion assays and in one case (DU145 cells, Figure
3) with metastasis potential in murine models of metastasis. We observed that metastatic spread in the fish was achieved as early as 24 hpi, and possibly sooner, and that the maximum tumor spread was achieved within 48 hours in most cases, without a significant increase over subsequent days. A reduction in the numbers of cancer cells, however, occurred when the analysis was extended to 5 dpi. The metastasis assay, however, if initiated at 2 dpi, can be completed before it is necessary to feed the fish (4 dpi) and, at least in the cells we tested, do not need to be followed for more that 2–3 days to evaluate metastasis. In prior studies, the number of cells injected into the fish varied between 50–2000 [
14‐
17,
19‐
22]. We observed that injecting too many cells can lead to mortality in our system and, following preliminary evaluation of the optimal number to demonstrate metastasis, we consistently injected cells ~300 cells per fish.
In many of our studies, the difference in the number of disseminated cells between the metastatic and non-metastatic tumors was usually striking. The same was observed in experimental cell systems where inactivation of a particular gene led to almost complete loss of invasion or metastasis, e.g. the JAK1 deficient 2C4 cells or the DU145 WASF3 knockdown cells. In these experiments, however, cells were seen outside the yolk sac region, which raised the issue of how metastasis is defined? In cell lines such as T47D, LNCAP and HT29, which are generally considered non-metastatic, our analysis showed that, even if there were disseminated cells, in the majority there were usually only between 1–5 cells in the body of the fish, where 5 cells was the exception. There are several reasons why small numbers of cells may appear in the fish body. Firstly, on rare occasions, the injection procedure could have inadvertently penetrated the vasculature and cells were introduced directly, although we screened all fish 12 hours after injection and excluded any that already showed cells outside the yolk sac region to overcome this being a major factor in the analysis. In practice this was only a very small number (<2%) for each cohort. It is also possible that cells defined as non-metastatic, are in fact weakly metastatic, and so occasional cells will disseminate into the fish. This is particularly true in experimental systems when, for example, shRNA knockdown of a particular gene is not complete, leaving some cells with gene expression levels above the threshold that will allow metastasis. It is important to note, however, that in many of our experiments involving apparently non metastatic cells, the presence of >5 disseminated cells was only seen in a minority of fish in the cohort. The main criterion for metastasis, therefore, is the presence of >5 cells in the majority of fish. In practice, however, the numbers of metastatic cells throughout the various cohorts for metastatic cells was far greater than 5 as shown in Figure
4 (35–55 cells after 48 hours). The number of disseminated cells becomes particularly important when defining relative metastatic potential, as seen in the MCF10A continuum. In this case the overall number of cells found outside the yolk sac area correlated with the invasiveness of the cells in vitro. We expect, however, most metastasis assays will want to determine whether ablation or overexpression of a gene leads to changes in metastatic potential, and following the protocol described here will facilitate this determination.
Although we were limited by the number of clinical samples available to us, the demonstration that primary human cancer cells can survive, grow, and metastasize in zebrafish provides a very encouraging proof-of-principle and opens opportunities to evaluate the metastatic potential in primary cells from biopsies or following surgery, which can have important advantages for clinical management of the patient. Even though it may take 2–3 weeks to establish the primary cell cultures, the presence of metastatic cells in a tumor for which there is a well-differentiated histopathology, could affect the future screening and management protocol. In addition, the well- established systems [
36‐
38] for drug screening in zebrafish, opens up the possibility of identifying therapeutics that can target metastasis on a tumor-by-tumor basis, so providing a personalized approach to individual tumors.
Competing interests
The authors declare no competing financial interests.
Authors’ contributions
YT and JC designed the experiments; YT performed the molecular and cellular experiments; YT, XX and DW performed the fish husbandry and imaging, YT and SW performed the image analysis; JM provided input into zebrafish experiments; YT, JC and JM wrote the manuscript. All authors read and approved the final manuscript.