To perform a functional screen for suppressors of anoikis, we selected rat intestinal epithelial (RIE) cells owing to their non-malignancy and high sensitivity to anoikis (refs 6, 7; see below). Anoikis can be induced in vitro by transferring epithelial cells from standard, adhesive cell culture dishes (with a hydrophilic surface that supports cell attachment and spreading) to ultra-low cluster (ULC) plates with a covalently bound hydrogel layer that effectively inhibits cellular attachment. Shifting RIE cells from adhesive to ULC dishes resulted in massive cell death (Fig. 1a, left), which was critical in view of the desired minimal number of surviving ‘background’ cells that would represent false positives in the screen. In agreement with previous results7, oncogenic RasV12 protein effectively suppressed anoikis (Fig. 1a, right).

Figure 1: Functional cloning and validation of TrkB as a suppressor of anoikis.
figure 1

a, Parental or RasV12-expressing RIE cells seeded into ULC tissue culture plates and photographed at ×50 magnification after 3 days (top). The bottom panel shows subsequent transfer of the cells to adhesive tissue culture plates, followed by fixation and crystal violet staining 2 days later. b, Outline of the screen. See Methods for details. c, Suppression of anoikis by clone ME-3, relative to a GFP-control, photographed at ×50 (top) and ×1 (bottom) magnification and at indicated time points. d, RIE cells expressing cDNAs as indicated, seeded on adhesive plates and photographed at ×200 magnification. e, RIE cells expressing cDNAs as indicated, seeded into ULC plates and photographed in time at ×50 (day 3) and ×1 (day 14) magnification. f, Single cell suspensions of RIE cells expressing cDNAs as indicated, seeded into soft-agar and photographed at ×50 magnification 1 week later.

Upon retroviral delivery of a high-complexity pEYK retroviral complementary DNA library8, cells were brought into suspension and screened for surviving clones (Fig. 1b). Relative to control green fluorescent protein (GFP)-infected cells that died, a proportion of library-infected pools showed surviving clones (see Methods), including ‘ME-3’; this clone effectively suppressed anoikis and formed large, spheroid aggregates in suspension (Fig. 1c). Isolation and sequencing of its proviral insert revealed a full-length, wild-type cDNA encoding TrkB, a neurotrophic tyrosine kinase receptor that is required for the ontogeny, function and survival of the mammalian nervous system4,5 and other cell types9,10,11,12. Interestingly, neurite development and outgrowth may represent processes that particularly rely on adhesion-independent survival13.

To characterize the biological effects of TrkB, we generated, by retroviral delivery, pools of epithelial cells stably expressing TrkB, its primary ligand (brain-derived neurotrophic factor (BDNF))14,15, or both (Supplementary Fig. S1a). On adhesive plates, TrkB, but not BDNF, disrupted epithelial cell organization, which was aggravated by ligand co-expression, resulting in complete loss of cell–cell contact and, upon reaching confluency, in continued proliferation as spheroids in suspension (Fig. 1d, Supplementary Fig. S1b, c).

To confirm that TrkB suppresses anoikis, we followed the fate of TrkB-expressing epithelial cells in suspension. In ULC dishes, both empty-vector and BDNF-expressing cells underwent cell death, starting at day two and leaving no surviving cells after two weeks (Fig. 1e). By contrast, TrkB effectively suppressed cell death and allowed cells to proliferate as large spheroid aggregates in suspension, explaining the identification of TrkB in the screen. Co-expression of TrkB with BDNF led to a further increase in cell number (Supplementary Fig. S1d). Similar results were obtained for murine and human epithelial cells (Supplementary Fig. S1e). The observation that the entire polyclonal TrkB-expressing cell population was resistant to anoikis argued against clonal events being due to genetic mutations. This was further supported by a soft-agar assay, which reinforced our notion that TrkB expression bypasses the need for anchorage, with virtually every TrkB-expressing cell producing a colony (Fig. 1f).

As anoikis is defined as apoptosis caused by lack of adhesion2, we examined whether it was specifically apoptosis that was prevented by TrkB. First, staining with an apoptosis-dependent fluorescent dye (3,3′-dihexyloxa-carbocyanine iodide, DiOC6(3)) showed that, in agreement with previous results6,7, RIE cells underwent apoptosis upon attachment withdrawal (Fig. 2a, left). By contrast, apoptosis was completely blocked by TrkB (Fig. 2a, right). Second, we determined by immunoblotting the protein levels of cleaved caspase-3, a primary executioner of apoptosis. Whereas no activated caspase-3 was detected in any of the cell lines when grown on adhesive dishes, it was produced by control and BDNF-expressing cells when they were brought into suspension (Fig. 2b). By contrast, cleaved caspase-3 production was completely suppressed by TrkB, regardless of the presence of TrkB's ligand. Together, these results indicate that it is apoptosis, resulting from lack of adhesion, that is effectively suppressed by TrkB.

Figure 2: Specific suppression of anoikis-associated apoptosis by TrkB.
figure 2

a, Control or TrkB-expressing RIE cells seeded into ULC plates and analysed for apoptosis over time by DiOC6(3) staining for FACS analysis30. Right- and left-hand peaks reflect living and apoptotic cells respectively. b, Immunoblotting for cleaved caspase-3 of lysates from RIE cells expressing indicated cDNAs, on adhesive (left) or ULC (right) plates, for 2 days. CDK4 serves as loading control. c, RIE cells expressing indicated cDNAs, seeded either in the presence (red) or absence (blue) of serum into ULC plates and analysed for apoptosis 3 days later, as in a. Results with 0.1% serum were similar to those with 0% (not shown). d, e, Rat1 fibroblasts co-expressing c-Myc and indicated cDNAs, on adhesive dishes, in the absence of serum for 3 days and analysed by FACS (d) and photographed at ×200 magnification (e). Empty vector-transduced cells in 10% serum (Vector 10%) serve as control. f, Immunoblotting for serine 473-phosphorylated PKB (P-PKB) and total PKB of lysates from RIE cells expressing indicated cDNAs, on adhesive (left) or ULC plates (right) for 2 days. g, TrkB/BDNF co-expressing cells, either in the presence or absence of activated PKB, treated after spheroid formation with LY294002 (20 µM) or solvent and photographed 12 days later at ×1 magnification. h, i, TrkB/BDNF-expressing cells treated after spheroid formation with rapamycin (20 nM), photographed at ×50 magnification 2 days later (h) and subsequently used for immunoblotting for p70S6K and CDK4 as loading control (i). Similar results were obtained for 1-week treatment (not shown).

Integrins are principal mediators of adhesion between cells and extracellular matrix proteins, including fibronectin, and they transduce signals critically required for cell survival16,17. Therefore, we examined whether TrkB-mediated resistance to anoikis depends on fibronectin, or on any other potential survival factor(s) present in cell culture serum. When TrkB-expressing cells were deprived of serum, they were partially protected from apoptosis, relative to control cells (Fig. 2c). However, BDNF-stimulated TrkB rendered cells resistant to anoikis, despite the complete absence of serum. This effect was not due to a general pro-survival function of TrkB, as TrkB, in contrast to the anti-apoptotic Bcl-2 protein, completely failed to protect adherent c-Myc-expressing fibroblasts from apoptosis resulting from serum withdrawal (Fig. 2d, e). Therefore, in untransformed epithelial cells, the BDNF–TrkB axis generates a signal that is sufficiently potent to allow, in the absence of any exogenous factors, survival in the face of a strong and specific pro-apoptotic anoikis signal.

To investigate whether TrkB directly activates survival signalling, as well as to identify its downstream targets, we first examined the potential involvement of protein kinase B (PKB; also known as AKT), which by itself is sufficient to suppress anoikis18,19. Under adherent conditions, TrkB induced production of activated, phosphorylated PKB (Fig. 2f, left). As TrkB/BDNF-expressing cells on adhesive dishes show diminished cell–cell interactions (Fig. 1d), this result demonstrates that, rather than through promoting cell–cell adhesion, TrkB directly activates the PKB survival pathway. This result, however, does not rule out the possibility that aggregate formation may contribute to survival signalling in suspension. Importantly, PKB activation was maintained in cells challenged with anoikis (Fig. 2f, right), raising the possibility that the PKB pathway mediates anoikis suppression by TrkB. Indeed, blocking PKB-dependent signalling by pharmacological inhibition of its major upstream regulator, phosphatidylinositol-3-OH kinase (PI(3)K), with LY294002 significantly interfered with TrkB-mediated survival (Fig. 2g; Supplementary Fig. S2a, b). In turn, cell death was prevented by activated PKB, excluding aspecific cytotoxic effects of LY294002, and further supporting a central role for PKB in anoikis suppression by TrkB. In line with previous data on Madine–Darby canine kidney (MDCK) cells18, PI(3)K was also sufficient to suppress anoikis, albeit to a lesser degree than TrkB (Supplementary Fig. S2c). One of the critical downstream targets of the PI(3)K/PKB pathway is p70S6K, a kinase implicated both in anoikis and lymphomagenesis18,20. However, although rapamycin (a specific inhibitor of mTOR, which is a regulator of p70S6K) blocked p70S6K phosphorylation, it completely failed to affect the survival of TrkB-expressing cells in suspension (Fig. 2h, i). Similarly, dominant interference with signalling by Rac proteins (another target of PKB), or pharmacological inhibition of MEK did not prevent suppression of anoikis by TrkB (Supplementary Fig. S2d, e). We conclude that TrkB activates PI(3)K/PKB signalling, which contributes to anoikis resistance.

The main objective of this study was to assess the feasibility of using anoikis suppression as a basis to identify metastasis-inducing (onco)genes. When non-malignant cells expressing TrkB, either alone or in conjunction with its ligand, were introduced into nude mice, they rapidly formed tumours with high efficiencies, whereas vector and BDNF-expressing cells failed to do so (Supplementary Table S1). Therefore, expression of TrkB is sufficient to convert non-malignant epithelial cells into highly tumorigenic cells.

To address whether TrkB endows epithelial cells with metastatic potential as well, we first re-engineered our cell panel to express luciferase, which allows non-invasive in vivo imaging in live mice21. Each polyclonal cell line was administered intravenously to nude mice, and outgrowth and the location of tumours was monitored. Luciferase imaging showed that, although at 90 min after injection control cells were apparently viable and had accumulated in the lungs to at least the same extent as TrkB/BDNF co-expressing cells, no luciferase signal was detectable at 6 days (Fig. 3a). Indeed, control and BDNF-expressing cells failed to form tumours within 100 days after inoculation (Fig. 3b), consistent with the hypothesis that non-malignant epithelial cells fail to survive for a prolonged period in the absence of appropriate attachment2. By contrast, TrkB-expressing epithelial cells colonized the lungs and heart, where they formed rapidly growing tumours (Fig. 3c). This phenotype was even more pronounced for TrkB/BDNF-expressing cells, which caused metastases throughout the body (Fig. 3a, c; Supplementary Fig. S3). The cooperation between TrkB and BDNF in inducing metastasis therefore recapitulated their collaboration in suppressing anoikis under low-serum conditions (which may better approximate the situation in vivo than does 10% serum; compare Fig. 2c to Fig. 3c).

Figure 3: Tumorigenic potential of TrkB.
figure 3

ac, Mice injected intravenously with RIE cells co-expressing a luciferase gene and indicated cDNAs, and subjected to in vivo imaging at indicated time points, with relative light units per pixel indicated in colour (a, c), or represented in a Kaplan–Meier survival plot (b). Note the relatively short latency of TrkB/BDNF-induced tumours and different imaging sensitivities used (c).

Microscopic pathological analysis showed the capacity of TrkB to induce solid, non-encapsulated tumours with undifferentiated properties (Fig. 4 and Supplementary Fig. S4). The tumours showed malignant cellular characteristics, including a high mitotic index, a high nucleus:cytoplasm ratio, and notable cellular and nuclear pleomorphism. TrkB-specific immunohistochemistry showed that these tumours were able to efficiently invade both blood and lymphatic vessels in various tissues, including liver, kidney, lung and heart (Fig. 4a–f, i, j and Supplementary Fig. S4). This infiltrative character of TrkB-driven tumours was also markedly illustrated by their capacity to destruct and invade bones (Supplementary Fig. S4g–i). Furthermore, these tumours, when present as large tumour masses, were viable and proliferating, as judged by their very low apoptotic and high proliferative indices respectively (Fig. 4 g, h, k). Importantly, this viable and non-apoptotic character was retained when these tumours extended into vessels (Fig. 4i, j), conceivably contributing to their metastatic potential and consistent with our finding in vitro that TrkB suppresses anoikis. The tumours were so invasive, destructive and fast-growing that only 18 days after administration, all mice had been euthanized (Fig. 3b, Supplementary Fig. S3).

Figure 4: Metastatic potential of TrkB.
figure 4

aj, Pathological characteristics of tumour processes after tail vein injection into nude mice. a, Large tumour nodule within liver. b, Clusters of tumour cells located in liver sinusoids and a tumour cell located in vena hepatica branch. c, Tumour nodule within liver, invading a vena portae branch. d, Large tumour within kidney. e, Detail of a renal tumour growing within lumen of large vein. f, Another plane of the section shown in e, demonstrating renal tumour extending into vein, activating the blood coagulation system and causing fibrin deposition (F). g, Detail of tumour in d, showing tumour cells located between renal tubules, and high proliferative index, as evident from nuclear MCM7 immunoreactivity. h, Representative photomicrograph of section in g, showing lack of apoptosis, as evident from absence of cleaved caspase-3 immunoreactivity. i, Tumour nodule (with edge of large tumour mass at bottom side of microphotograph), spreading into branch of coronary vein of the myocardium and showing high proliferative index, as evident from MCM7 immunopositivity. j, Representative photomicrograph of adjacent section of i, showing lack of apoptosis of tumour cells infiltrating coronary vein, as evidenced by a lack of expression of cleaved caspase-3. k, Splenic lymphoma from an unrelated mouse, showing numerous apoptotic cells, serving as positive control for cleaved caspase-3-specific immunohistochemistry. lp, Pathological characteristics of tumour processes upon subcutaneous injection. Representative microphotographs showing the metastatic process in one mouse. l, Subcutaneously injected tumour cells infiltrating muscle tissue (arrows). m, Clusters of tumour cells infiltrating lymphatic vessels, at a distance from the subcutaneous tumour. n, Tumour infiltrating axillary lymph node (L) draining the tumour site. o, Large metastases at pleural side of the lungs (apical lobe of the right lung was removed to allow a better view of the tumour nodules). p, One of the metastases shown in o. Immunohistochemistry was performed for TrkB (ae, m, n), MCM7 (g, i), or cleaved caspase-3 (h, j, k). f, l, p, Haematoxylin and eosin stainings. T and V mark tumours and vessels, respectively. Scale bars: 500 µm (a, p); 200 µm (c, e); 100 µm (f, k, l, n); 50 µm (b, gj); 20 µm (m); 1 mm (d); 5 mm (o). Rows indicate four individual sets of photographs.

As the intravenous metastasis system precludes an assessment of the early stages of metastasis (tissue invasion and endothelial or lymphatic intravasation by disseminated tumour cells), we also introduced TrkB/BDNF co-expressing cells subcutaneously. To increase the time window in which metastasis could occur, tumours that arose within a week were surgically removed, and upon relapse mice were killed. In agreement with, and extending the results described above, pathological examination showed that the subcutaneous tumour cells had efficiently infiltrated muscles, lymphatic vessels and regional lymph nodes, eventually causing several metastases at the pleural side of the lungs (Fig. 4l–p). Together, these results demonstrate that TrkB acts as a potent oncogene that converts non-malignant epithelial cells into metastasizing tumour cells.

TrkB and BDNF are frequently overexpressed in human cancer, including pancreatic and prostate carcinomas, Wilms' tumour and neuroblastomas, particularly those with aggressive behaviour and poor prognosis22,23,24,25. Moreover, recent sequence analysis of the tyrosine kinome in colorectal cancers has shown mutations within the kinase domain of TrkB26. Although these observations have associated TrkB with cancer, it has been unclear whether deregulated TrkB can act oncogenically. We find that TrkB generates a specific and potent pro-survival signal that renders epithelial cells resistant to anoikis. This is accompanied by the acquirement of potent tumorigenic, invasive and metastatic capacities, arguing that TrkB may directly contribute to human malignancies. Although our data are consistent with the hypothesis that suppression of anoikis contributes to the acquisition of a metastatic phenotype, other properties of TrkB, including induction of invasiveness, may contribute to its aggressive tumorigenic effects.

The propensity to metastasize may be determined by the gene expression signature already present in the primary tumour27. The results presented here raise the possibility that TrkB activation may represent such an early event during multistep tumorigenesis, endowing primary cells with a wide spectrum of capacities contributing to cancer. Mainly on the basis of the overexpression of TrkB in poor-prognosis human tumours, pan-Trk-inactivating drugs are already being developed28. Our findings predict that drug-mediated inactivation of specifically TrkB may have a notable impact on the tumorigenic and metastatic capacities of human tumours with overexpressed or mutated TrkB. Finally, our results not only illustrate the feasibility of using anoikis suppression as a functional property to identify metastasis-associated oncogenes, but also show its potential use in high-throughput screens for TrkB-inhibitory cancer therapeutics.


Genome-wide retroviral cDNA screen

RIE cells were infected, in the presence of 5 µg ml-1 polybrene, with a high-complexity retroviral pEYK cDNA library8, prepared from a whole mouse embryo and propagated in Phoenix packaging cells as eight pools representing two size fractions (1–3 and >3 kilobases (kb)). Filtered (0.45 µm) viral supernatant pools representing the two size fractions were used to infect 0.75 × 106 cells each, and 2 days later, cells were collected by trypsinization and transferred to four ULC six-well cell culture dishes (Costar) in total. After 6 days, cells were transferred back to regular, adhesive tissue culture dishes, to allow anoikis-resistant cells to expand. After 10 days, cells were re-challenged with an anoikis insult, by transferring them into ULC plates. One week later, surviving colonies were isolated and transferred to 24-well adhesive plates, and allowed to expand. Then, genomic DNA was isolated and digested with either NotI or AscI, re-ligated and transformed into Escherichia coli. Plasmid DNA was isolated and transfected onto Phoenix cells. Virus supernatant was subsequently used to infect fresh RIE cells and a second-round screen was performed to validate the biological activity of the respective proviral cDNAs. Finally, proviral cDNA inserts were isolated from positive clones, shuttled and analysed by DNA sequencing. One of the anoikis-resistant clones (ME-3) contained a wild-type, full-length murine cDNA encoding TrkB, which was sub-cloned into the pBabe-puro retroviral vector. Of the 24 wells screened, 12 contained anoikis-resistant clones, sometimes several per well, giving 20 anoikis-resistant clones in total. Polymerase chain reaction (PCR) analysis for retroviral TrkB inserts showed that, in addition to ME-3, TrkB was present in seven clones (derived from four different wells). The remaining 13 clones remain to be analysed further.

A murine full-length cDNA encoding BDNF was isolated from a mouse cDNA library (Clontech) by PCR (primers available upon request), and sub-cloned into the pBabe-hygro retroviral vector.

Cell culture and in vitro assays

RIE cells were maintained in high-glucose Dulbecco's modified Eagle's Medium (DMEM) (Life Technologies) supplemented with 9% fetal bovine serum (PAA Laboratories) and antibiotics. c-Myc-expressing Rat1 fibroblasts were provided by K. Berns. Phoenix packaging cells were used to generate ecotropic retroviruses, as described29. In general, a single-infection round of 6 h was sufficient to infect at least 90% of the population. RIE cells were infected with pBabe retroviral vectors carrying selectable markers and cDNA inserts, and at 1 to 2 days post infection selected with puromycin (1.5 µg ml-1) or hygromycin (75 µg ml-1) for at least 5 days. After confirming that all mock-infected cells were dead, the cells were used in various assays.

Proliferation curves were performed with two independent cultures for each cell line, and data points were collected in triplicate, as described29.

To determine the cloning efficiency of RIE cell lines in semi-solid medium, soft agar assays were performed, as described29.

To measure apoptosis, cells were collected from either adhesive or suspension plates (both populations were subjected to trypsinization), treated with DiOC6(3) (a fluorescing dye that accumulates in mitochondria as a function of membrane potential30; Molecular Probes) in PBS for 15 min at 37 °C, and subsequently analysed by fluorescence-activated cell sorting (FACS).

Pharmacological inhibition was performed with LY294002 (20 µM), PD98059 (20 µM), K-252A (300 nM) or rapamycin (20 nM), all from Calbiochem. Retroviral vectors encoding myristylated PKB and activated p110-CAAX PI(3)K were gifts from F. Scheeren and J. Collard respectively.


Paraffin-embedded tissue was cut into 4-µm sections, deparaffinized and rehydrated. Upon antigen retrieval performed with citrate buffer (pH, 6) and incubation with primary and secondary antibodies, immunocomplexes were detected with either DAB/H2O2 or AEC (Dakocytomation).


For western blotting and immunohistochemistry, the following antibodies were used: MS-862 for MCM-7 (NeoMarkers); AC-74 for actin (Sigma); 9661 for cleaved caspase-3 and 9202 for p70S6K (Cell Signalling); SC-8316 for TrkB, SC-260 for CDK4, SC-7985 for Ser 473-phosphorylated PKB and SC-8312 for PKB (all Santa Cruz).

In vivo assays

To determine tumorigenicity and metastatic potential, the RIE cell panel (1 × 106 cells) was injected subcutaneously into athymic nude mice in each flank. Mice were inspected daily and were killed at the time tumours reached a diameter of 10 mm.

To measure metastatic potential and allow in-vivo bioluminescence luciferase imaging, mice were injected intravenously with the RIE cell panel (1 × 106 for Fig. 3b, c, and 3 × 106 for Fig. 3a; administration of 107 cells gave identical results; not shown), co-expressing a luciferase gene and cDNAs of interest. In vivo bioluminescence imaging of luciferase expression was performed as described21. Generally, for each cell line, 2–4 mice were imaged in time, with similar results.