Main

The ‘seed and soil’ theory of metastasis proposes that the fate of DTCs is dependent on the microenvironment where they lodge1. This supports the organ-specific preference for metastasis in certain cancers. However, the timing of metastasis is largely unpredictable because residual DTCs may be dormant for long periods2,3. Solitary BM DTCs that are mostly negative for proliferation markers have been catalogued as being in a cellular dormancy state functionally defined by quiescence2. Intriguingly, in many malignancies, although 30–40% of patients have BM DTCs, these rarely develop bone metastases (HNSCC, gastric cancer) or develop very late (breast cancer)4,5. A similar scenario for seed and soil mechanisms can be studied in mouse tumour models. Whereas spontaneous metastases vigorously develop in lungs and lymph nodes and, less frequently in liver6, DTCs can be found in the BM and other sites in a non-productive state, at least in the same time frame2,7. Importantly, even in syngeneic models, spontaneous overt bone lesions rarely develop, unless the microenvironment is altered7. This suggests that specific microenvironmental signals could keep DTCs dormant.

We reported a ‘dormancy gene signature’ regulated by p38α/β kinase signalling in quiescent HNSCC cells8. Oestrogen-receptor-positive luminal breast cancer tumours that were enriched for the dormancy signature exhibited longer metastasis-free periods, suggesting that these dormancy genes in DTCs may affect metastasis timing9. Some of these genes (for example, NR2F1; ref. 10, DEC2 (also known as SHARP1); ref. 11, FOXM1; ref. 12) regulate pluripotency, are predictors of delayed recurrence in breast and prostate cancer11, and suppress the malignant behaviour8,11. Another signature gene, TGF β 2, was the most upregulated TGF-β-family cytokine in quiescent HNSCC cells in vivo. TGF-β2 can activate p38 (ref. 13) and induce quiescence, and both these proteins limit self-renewal10,14. Thus, we reasoned that TGF-β2high microenvironments might drive DTC dormancy. Here we reveal a previously unrecognized ‘seed and soil’ mechanism where TGF-β2 signalling through p38α/β regulates DTC dormancy and defines restrictive (BM) and permissive (lung) microenvironments for HNSCC metastasis.

RESULTS

The BM microenvironment induces DTC dormancy

HEp3–GFP cells obtained from xenografts were injected subcutaneously into nude mice and when primary tumours reached ∼500 mm3 (t0) they were surgically removed. Then, HEp3 spontaneous DTCs were followed using Green Fluorescent Protein (GFP) fluorescence and by human Alu sequence-specific quantitative PCR (qPCR; ref. 15) to identify DTCs undetectable using GFP, or Cytokeratin 8/18 (CK8/18) staining. This revealed a DTC prevalence of ∼ 80% in lungs, ∼ 28% in the BM and ∼ 5% in liver and spleen at the time of dissection (Fig. 1a and Supplementary Table S1 and Fig. 1a–d). GFP+ cells imaged in fresh tissues in situ were intact viable cells as determined by confocal imaging of DTCs in situ (Fig. 1a). In the BM the number of DTCs (GFP+) ranged from 10 to 102 per BM sample, was >2 logs lower than in lungs (Fig. 1a,b and Supplementary Fig. 1a,c,d) and remained constant for at least 4 weeks after primary tumour surgery (Fig. 1b). In contrast, lung DTCs that were already present in the lung as single solitary cells at the time of surgery (Supplementary Fig. 1e) initiated vigorous proliferation 2 weeks after surgery (Fig. 1b).

Figure 1: Growth behaviour of BM- and lung-derived DTCs.
figure 1

(a) Phase-contrast and GFP-channel images of a single DTC in a BM flush (upper left panel) and a lung metastasis (upper right panel) 3 wk after primary tumour surgery. Scale bars, 40 μm (BM) and 400 μm (lung). S, stroma. Lower left panel: fluorescence images of lung stroma and a micro-metastasis in a freshly resected lung. Lower right panel: laser scanning confocal image of a DTC cluster in lung. Scale bars, 160 μm (left) and 20 μm (right). n = 16 (BM), 35 (LU) mice. (b) Number of HEp3–GFP DTCs found in lungs and BM after primary tumour surgery (n = 5 mice per time per condition). (c) Tumour volume of PT-HEp3-, Lu-HEp3- and BM-HEp3-derived cell lines from mice, after 1 week on CAM. (n = 5 (Lu6), 5 (Lu7), 4 (PT-2), 7 (PT-4), 5 (BM-T1), 5 (BM-D1) tumours). (d) Phospho-histone3 (P-H3; left column) and p27 (right column) staining in BM-D1 (upper panels) tumour nodules and Lu-7 (lower panels) tumours grown on CAMs. Scale bar, 40 μm. Arrows, positive cells. Lower panel: quantification of P-H3+ and p27+ cells (n = 100 cells assessed per section. 15 sections assessed from 3 different tumours per condition). FOV, field of view. (e) Weight of BM-D1 tumour nodules on CAMs. BM-D1 cells were inoculated into CAM (5×105 cells per animal) and transplanted into a new CAM every week (n = 3 (week 1), 5 (week 2), 5 (week 3), 5 (week 4), 3 (week 5), 4 (week 6), 2 (week 7), 3 (week 8) tumours per time point). (f) Number of tumour cells per CAM nodule produced by primary tumour (PT-HEp3) and BM-DTC cell lines (BM-HEp3) derived from an avian source after 1 week on CAM (n = 4 (BM-D2), 3 (BM-D3), 4 (BM-D4), 10 (PT), 5 (BM-D5), 5 (BM-T2), 4 (BM-T3), 2 (BM-D6), 3 (BM-D7), 3 (BM-D8) tumours). See Supplementary Table S4 for statistic source data. Data in a,b and d, represent mean±s.e.m. In b and d, *P<0.05, **P<0.01 and +++P<0.001 by Mann–Whitney test. In c and f, *P<0.05, **P<0.01 and ***P<0.001 by one-way analysis of variance (ANOVA) with Bonferroni’s multiple comparison test, and in e, **P<0.01 by unpaired t-test. NS, not significant.

Full size image

In 100% of the cases we could expand in vitro HEp3 GFP-tagged DTCs (puro-resistant) isolated from lungs (Lu-HEp3) and this was independent of the initial number of recovered lung DTCs (Supplementary Table S2). In stark contrast, in the ∼ 28% of mice that had BM DTCs, although all of these DTCs (BM-HEp3) survived plating, only 2/15 (13.3%) expanded in culture. The lack of proliferative capacity was persistent as evidenced by 86.6% of all BM DTC isolates resulting in GFP+ puromycin-resistant solitary growth-arrested HEp3 DTCs for >4 weeks in culture (Supplementary Table S2).

That ∼ 86% of BM DTCs (Supplementary Table S2) are viable but non-proliferative in culture led us to reason that the BM microenvironment may instruct DTCs to activate long-lasting dormancy programs. We could study mechanistically only those BM DTCs that expanded in vitro and focused on BM- versus lung-derived DTCs because of their most divergent in vivo behaviour and clinical relevance16 (Fig. 1a). We defined dormant DTCs as non-proliferating or slow-cycling cells, negative/low for proliferation markers (that is, phospho-H3 (P-H3)) and positive/high for CDK inhibitor expression (that is, p21, p27Cip). Two lung (Lu-HEp3)-, two BM (BM-HEp3)- and two primary tumour (PT-HEp3)-derived cell lines were screened for tumorigenicity in the chorioallantoic membrane (CAM) system17,18,19 or in nude mice8. In all cases, Lu-HEp3 and PT-HEp3 cells were tumorigenic and were P-H3high/p27low (Fig. 1c,d and Supplementary Fig. 1f). In contrast, one BM-HEp3 cell line remained dormant for 4–5 weeks before growing vigorously (henceforth BM-D1; Fig. 1e and Supplementary Fig. 1g); dormant BM-D1 cells were P-H3low/p27high (Fig. 1d and Supplementary Fig. 1f), confirming their quiescent phenotype. Another BM-derived DTC line (BM-T1) that did not proliferate in situ, fully reversed to a proliferative behaviour in culture and in vivo (Fig. 1c). In the nude mouse BM-D1 cells injected subcutaneously also formed tumours later than BM-T1 cells (Supplementary Fig. 1h). These data suggest that although ∼ 85% of DTCs from murine BM will remain dormant and non-proliferative in culture, half of those BM-DTCs that do expand in culture can retain a dormant phenotype when re-injected in vivo (Fig. 1c–e).

We also obtained BM-DTC cell lines from chicken and turkey embryo BM because we could process larger numbers of animals and increase our chances of obtaining BM DTC lines (Supplementary Fig. 1i). DTCs were non-proliferative in avian BM but did proliferate in the liver and lungs where metastases develop20 (Supplementary Fig. 1j). As in mice, the number of DTCs in the BM was 2 logs lower than in the liver and lungs (Supplementary Fig. 1j). Importantly, 7/9 (77%) BM-DTC cell lines remained dormant in vivo (Fig. 1f- and Supplementary Fig. 1k). We conclude that whereas 100% of lung-derived DTC lines were tumorigenic, 13/15 (86.6%) of the BM-isolated DTCs were non-proliferative in vitro (up to 4–8-weeks) and 1/15 (∼ 7%) was dormant in vivo after 5–6 weeks. Together, 14/15 (∼ 93.3%) BM-derived DTCs from mice and 77% of BM-derived DTC lines in the avian system are dormant in vitro and/or in vivo. Thus, dormancy is the predominant phenotype of murine or avian BM-isolated human HNSCC DTCs.

Signalling pathways activated in BM dormant DTCs

We next measured previously identified key regulators of dormancy8,21. Proliferative Lu-HEp3 and BM-T1 cells from the murine system and BM-T2 and BM-T3 cells from the avian system exhibited a high extracellular signal-related kinase (ERK)/p38 activity ratio, consistent with their tumorigenicity. In contrast, the dormant BM-D1 and BM-D2, -D3 and -D4 cells from murine and avian BM, respectively, exhibited a persistently low ERK/p38 activity ratio (Fig. 2a,b); BM-D5 was the only avian BM DTC line in which the ERK/p38 activity ratio (high) was not predictive of dormancy (Fig. 2b). The ERK/p38 ratio correlated also with the metastatic (4T1 and F3II; refs 22, 23; (ERK/p38)high) or dormant (4T07; (ERK/p38)low) behaviour of mouse breast cancer cell lines in lungs23 (Supplementary Fig. 2a). Thus, a high ERK/p38 signalling ratio (see Methods for quantification) is predictive of a tumorigenic and dormant phenotype in 100% (n = 9) and 90% (n = 9) of the DTC lines, respectively (Fig. 2c). BM-D1 cells also upregulated p53 (messenger RNA and protein) and DEC2 mRNA, two genes previously linked to dormancy and quiescence8, when compared with the proliferative variants (Lu-HEp3 and BM-T1) cells. The p53 transcript was almost undetectable in 4T1, 4T07 and F3II cell lines, but DEC2 was vigorously upregulated in 4T07 versus F3II and 4T1 metastatic cells (Fig. 2d and Supplementary Fig. 2b,c).

Figure 2: Signalling pathways and genes regulating dormancy of BM-HEp3 cells.
figure 2

(a,b) Lysates from the indicated DTC-derived cell lines from murine (a) and avian microenvironments (b) were probed by immunoblot analysis for the indicated antigens. Numbers on top of blots, (ERK/p38) activity ratio quantification (Methods). (c) Graph showing the percentage of cell lines with high or low ERK/p38 ratio and their phenotype in vivo. (d) Left panel: heat map showing DEC2 and p53 mRNA expression in the indicated cell lines after 24 h in culture. Scale, log2-fold change and upregulation of mRNAs (red) was significant (P<0.05). Middle and right panels: immunoblots for p53 and GAPDH in the indicated cell lines after 24 h in culture. (e) BM-D1 cells transfected with either scrambled (scr), p53, p38α or DEC2 siRNAs or Lu-HEp3 cells transfected with empty vector (Ctrl) or a constitutively active p38 α construct (p38α-CA–HA) or a DEC2–V5 construct were inoculated in vivo (5×105 BM-D1 cells per CAM and 2.5×105 lu-7 cells per CAM) for 5 days. The graph represents the number of cells per tumour nodule (n = 18 (scr), 12 (p53), 6 (p38α), 7 (DEC2), 20 (Ctrl), 7 (p38αCA–HA), 5 (DEC2–V5) tumours per condition). *P<0.05 and **P<0.01 by one-way ANOVA with Bonferroni’s multiple comparison test. (f) Growth of BM-D1 cells on CAM after RNAi against DEC2 for two weeks in vivo. BM-D1 cells transfected with scrambled (scr) or DEC2 siRNA were inoculate into CAMs (5×105 BM-D1 cells per CAM). One week after the tumour nodules were minced, cells per tumour nodule were quantified, transfected again with either scrambled or DEC2 siRNAs and reinoculated again in vivo. Graph, mean number of cells per tumour nodule±s.e.m. (n = 4 scr siRNA week 1, 5 DEC2 siRNA week 1, 3 scr siRNA week 2, 5 DEC2 siRNA week 2 tumours per condition). See Supplementary Table S4 for statistic source data. *P<0.05 and **P<0.01 by Mann–Whitney test. Right panel: p27 and CDK4 expression in BM-D1 cell lines after DEC2 inhibition with siRNAs. Scr, scrambled. (g) p27, p21 and CDK4 immunoblots in T-HEp3 cells transfected with either a GFP or a DEC2–V5 construct for 24 h. Uncropped images of blots are shown in Supplementary Fig. 6.

Full size image

Knockdown of p38α, DEC2 or p53 interrupted the dormancy of BM-D1 cells (Fig. 2e, Supplementary Fig. 2d) and sustained knockdown of DEC2 for two weeks allowed BM-D1 cells to completely regain tumorigenicity in vivo. This correlated with upregulation of CDK4 and inhibition of p27 (Fig. 2f). Accordingly, overexpression of vectors encoding a constitutively active p38 α (p38a-CA) mutant or wild-type DEC2 in proliferative Lu-HEp3 cells reduced P-H3 levels, did not affect apoptosis and inhibited their in vivo proliferation (Fig. 2e and Supplementary Fig. 2e,f); DEC2 overexpression in T-HEp3 cells also induced p27 and p21 expression while inhibiting CDK4 expression (Fig. 2g). Analysis of human HNSCC primary and metastatic lesions showed that when compared with normal oral epithelium, and stromal cells, in 4/4 patients, DEC2 protein was strongly downregulated in the primary tumours and also in the matched lymph node metastasis (Supplementary Fig. 2g). These results suggest that primary tumour and metastatic growth is associated with downregulation of DEC2 as a possible dormancy escape mechanism.

Systemic p38α/β inhibition fuels occult DTC expansion

Next we used the small-molecule inhibitor SB203580 that faithfully phenocopies the genetic inhibition of at least p38α (refs 8, 19, 21, 24) to systemically inhibit p38α/β activity and test DTC fate after primary tumour surgery. We defined liver, spleen and BM as metastasis restrictive sites in nude mice because DTCs are detectable (Supplementary Table S1) but never grow. Direct analysis of BM DTCs using CK8/18 staining7 showed that 80% were p27high, and only 1% were P-H3+ (Fig. 3a), supporting their dormant phenotype. Importantly, we found that after 4 weeks, SB203580 treatment significantly increased the prevalence and number of DTCs in the BM as detected by GFP imaging, Alu sequence specific qPCR or quantification of CK8/18+ DTCs (Supplementary Table S3 and Fig. 3b). Although the three different detection methods have limitations they all collectively support that p38α/β inhibition after surgery can increase the DTC burden in the BM. Similar results were observed in the liver and spleen where p38α/β inhibition for 4 weeks doubled the prevalence of DTCs, micro- and macro-metastasis in these organs (Supplementary Table S3 and Fig. 3c,d).

Figure 3: Systemic p38α/β inhibition affects HNSCC DTC behaviour in growth-restrictive microenvironments.
figure 3

(a) Representative images of cytokeratin 8/18 (CK), P-H3 (left) and p27 (right) staining in solitary BM DTCs in BM cytospins. Scale bar, 10 μm. Right, graph: percentage of p27 and P-H3 positive DTCs per BM flush. Numbers on top, n for scored DTCs. 6 BM flushes and 3 independent experiments were assessed. (b) Quantification of BM-DTCs in mice treated with dimethylsulphoxide (control) or SB203580 (SB; 10 mg kg−1 every 48 h) for 4 weeks. Left graph: Alu qPCR quantification of BM DTCs shows the human Alu genomic signal normalized to genomic mouse GAPDH (n = 6 (control); 15 (SB) DNA samples were assessed over 3 independent experiments; see methods for details). Right graph: quantification of CK8/18+ cells in mice BM cytospins (n = 6 BM samples per condition were assessed over 2 independent experiments). *P<0.05 by Wilcoxon’s signed rank test. (c,d) Live imaging of HEp3–GFP DTCs, micro and macrometastasis in spleen (c) and liver (d) of control (scale bar, 80 μm) and SB203580-treated mice (spleen, scale bar, 400 μm and liver, scale bar, 80 μm). Numbers, GFP+ cells per organ. (e) Quantification of BM-DTCs in chicken embryos’ BM flushes 5 days after inoculation of T-HEp3–GFP cells transfected with either scrambled (scr) or p38 α siRNAs. (n = 7 (scr siRNA), 6 (p38 α siRNA) BM samples were assessed over 2 independent experiments.) Lower panels: representative fluorescence intravital images of BM DTCs in chicken embryo BM flushes. Scale bar, 160 μm. (f) Graph: percentage of p27+ cells per BM flush (n = 3 BM flushes were assessed over 2 independent experiments). Lower panels: representative images of CK8/18 (right column) and p27 (left column) staining in solitary BM DTCs in chicken embryo BM 5 days after inoculation of T-HEp3–GFP cells transfected with scrambled (scr; upper row) or p38α (lower row) siRNA. Scale bar, 40 μM. Data in a,b and f, represent mean±s.e.m. In b (left graph), e,f, *P<0.05, **P<0.01 by Mann–Whitney test.

Full size image

In the chick embryo system of metastasis25, the liver and lung are permissive sites for metastasis20,26; liver DTCs (>105 cells per organ) are P-H3high/p27low and proliferate actively (Supplementary Fig. 3a). The avian BM is also restrictive as DTCs (101–102 cells per BM) were also negative for P-H3 and positive for p27 (Supplementary Fig. 3a). After matching the primary tumours sizes (Supplementary Fig. 3b), we found that knockdown of p38α in T-HEp3 cells (short interfering RNA (siRNA) knockdown can last for 5 days) increased the amount of DTCs in avian BM 5 days after inoculation (Fig. 3e and Supplementary Fig. 3c). This increase correlated with a decrease in the percentage of p27high DTCs, suggesting an exit from quiescence (Fig. 3f). In the embryo liver (permissive site), p38α knockdown had no effect, possibly because DTCs grow immediately in this site (Supplementary Fig. 3d).

Solitary HEp3 lung DTCs identified by detecting human vimentin (abundantly expressed in these cells; Supplementary Fig. 3e) were mostly negative for P-H3 and cleaved caspase 3 (C-C3), supporting that initially solitary lung DTCs are viable quiescent or slow-cycling tumour cells (Fig. 4a,b). At 4 weeks, when macro-metastases are detectable and HEp3–GFP cell numbers increased in lungs, these lesions were P-H3high and C-C3 (Fig. 4a–c). Importantly, the p38α/β inhibitor treatment significantly increased HEp3–GFP HNSCC and F3II breast cancer metastatic cell burden after 2 and 4 weeks of treatment (Fig. 4c,d). We conclude that p38α/β signalling limits the expansion of DTCs in growth-restrictive sites such as mouse and chick embryo BM and in mouse liver and spleen. This effect is also observed on the short-term dormancy that solitary DTCs undergo in the lung in mice and the effect of p38α/β inhibition even on this short-term dormancy phase has a marked impact on disease burden in this tissue.

Figure 4: Systemic p38α/β inhibition interrupts DTC dormancy in the lungs.
figure 4

(a,b) Representative images of P-H3 and vimentin (a) or cleaved caspase 3 (C-C3) and vimentin (b) staining in lung DTCs (upper panels, scale bar, 20 μm) and lung micro-metastases (micro-MET; only a) 4 weeks after surgery (scale bar, 40 μm; micro-MET (a), inset (b)). Arrows, marker+ cells. Graphs: percentage of P-H3+ (a) or C-C3+ (b) cells per lung section, mean±s.e.m. of three different lungs, 5 sections per lung (n = 45 (DTCs), n = 416 (Met) DTCs per lungs section (a), n = 36 (DTCs) (b)). *P<0.05 (a) and not significant (NS; b) by Mann–Whitney test. (c) Representative lung images containing HEp3–GFP DTCs or metastasis (scale bars, 80 μm, left column and 400 μm right column) after surgery and control or SB203580 (SB; 10 mg kg−1 every 48 h) treatment for 2 and 4 weeks. Lower graphs: quantification of HEp3–GFP cells in whole lung suspensions. (n = 5 (left graph) and 4 (right graph) lungs per condition). See Supplementary Table S4 for the statistic source data of the right graph. *P<0.05 by Mann–Whitney test. (d) Number of lung F3II breast cancer macro-metastasis per lung in mice treated with dimethylsulphoxide (control) or SB203580 (SB). Top numbers: prevalence (%) of mice with more than 3 metastatic nodules per lung. n = 25 mice, *P = 0.046 by Mann–Whitney test.

Full size image

TGF-β2 signalling in the BM induces dormancy hallmarks

Expression profiling of proliferating (T-HEp3 (ERK/p38high)) and dormant D-HEp3 ((ERK/p38low)) cells in vivo revealed that TGF β 2 mRNA (not TGF β 1 or TGF β 3) is highly elevated only in dormant cells (Supplementary Fig. 4a). TGF-β2, which is present in the BM (ref. 27), activated p38, and induced DEC2- and TGF-β-RI-dependent quiescence in D-HEp3 cells (Supplementary Fig. 4b–d). In addition, dormant BM-D1 cells express >5-fold higher levels of TGF β 2 than parental T-HEp3 cells (T) and Lu-HEp3 cells (Fig. 5a). Importantly, we also found that TGF β 2 mRNA was greatly enriched in naive BM when compared with lung tissue (Fig. 5b) and that in basal conditions (serum free media) T-HEp3 cells express marginal TGF β 2 mRNA (Fig. 5a,c). On treatment of T-HEp3 cells with lung conditioned media (CM), TGF β 2 mRNA is increased. However, treatment with BM conditioned medium (CM) vigorously upregulated TGF β 2 expression (>10-fold; Fig. 5c) and inhibited T-HEp3 tumour growth in vivo, and this was positively correlated with upregulation of p27 and downregulation of P-H3 levels in vivo (Fig. 5d,e). In addition, the BM, but not lung CM, induced p38 phosphorylation and DEC2 and p53 mRNA expression in T-HEp3 and 4T1 cells (Fig. 5f,g and Supplementary Fig. 4e). Lung CM enhanced phospho-ERK1/2 levels (Fig. 5f) and none of the organ CM induced apoptosis (Supplementary Fig. 4f). Importantly, TGF-β2-depleted BM CM was incapable of inducing p38 activation and growth inhibition of T-HEp3 cells in vivo (Fig. 5f,h) whereas lung CM supplemented with TGF-β2 inhibited T-HEp3 tumour growth in vivo (Fig. 5h). TGF-β2 depletion from BM CM did not restore ERK1/2 phosphorylation (see the Discussion; Fig. 5f) and significantly, depletion of TGF-β1 from the BM CM did not affect its ability to activate p38 or inhibit tumour growth (Supplementary Fig. 4g). In fact, treatment with TGF-β1 rapidly switched dormant D-HEp3 and BM-D1 cells from dormancy to rapid tumour growth in vivo while not affecting at all T-HEp3 tumorigenicity (Fig. 5i). These data suggest that TGF-β2 but not TGF-β1 is a candidate mediator of p38 activation, DEC2 expression and dormancy in the BM.

Figure 5: Role of TGF-β2 in dormant HEp3 DTCs.
figure 5

(a) qPCR analysis of TGF β 2 mRNA levels in the indicated cells cultured in serum-free media for 24 h. (b) qPCR analysis of TGF β 2 mRNA levels in BM and lungs mice organs. (For a and b, n = 6 RNA samples per condition were assessed over 3 independent experiments). (c) qPCR for TGF β 2 mRNA in T-HEp3 cells treated with serum-free media (SF) or conditioned media from lung (Lu CM) or BM (BM CM) (24 h). (n = 6 RNA samples were assessed over 3 independent experiments.) (d) Growth after 5 days of T-HEp3 cells on CAMs (2×105 cells per CAM) pre-treated in culture for 24 h and then on the CAM every day with SF, BM CM or Lu CM (n = 11 (SF), 3 (Lu CM), 8 (BM CM) tumour nodules per condition). (e) Immunohistochemistry for P-H3 (upper row) and p27 (lower row) in T-HEp3 tumour nodule sections after treatment with SF (left column), Lu CM (middle column) or BM CM (right column) for 5 days in vivo; scale bar, 40 μm. Arrows: marker+ cells. Lower graphs: quantification of P-H3+ (left) and p27+ (right) cells (n = 100 cells assessed per section. Fifteen sections assessed from three different tumours per condition). (f) Detection of the indicated antigens in lysates of T-HEp3 cells treated with, SF, Lu CM or BM CM immunoprecipitated (IP) with IgG (IP-IgG) or with TGF-β2 antibody (IP-TGF-β2) for 24 h. Top numbers: (ERK/p38) ratio quantification. (g) qPCR for p53 and DEC2 mRNA in T-HEp3 cells treated as in c (n = 6 RNA samples were assessed over 3 independent experiments). (h) Top panel: immunoblot against TGF-β2 in the BM CM after IgG or anti-TGF-β2 immunodepletion. Lower panel: tumour growth of T-HEp3 cells treated with full or TGF-β2-immunodepleted BM CM or with Lu CM supplemented with PBS (CM) or with 2 ng ml−1TGF-β2 (n = 22 (SF), 15 (IP-IgG), 8 (IP-TGF-β2), 7 (Lu CM), 5 (Lu-CM TGF-β2) tumours per condition). (i) Effect of TGF-β1 (2 ng ml−1) on T-HEp3 and dormant D-HEp3 and BM-D1 tumour growth on CAMs for 5 days in vivo (n = 6 (DHEp3-TGF-β1), 9 (DHEp3+TGF-β1), 10 (BM-D1-TGF-β1), 10 (BM-D1+TGF-β1), 13 (THEp3-TGF-β1), 11 tumours per condition). Data in a–c,e and g, represent mean±s.e.m. In a–c,e and g, *P<0.05, **P<0.01 by Mann–Whitney test. In d,h and i *P<0.05, **P<0.01 by One-way ANOVA with Bonferroni’s multiple comparison test. NS, not significant. Uncropped images of blots are shown in Supplementary Fig. 6.

Full size image

We also found that the high basal p38 phosphorylation in BM-D1 cells was not stimulated by TGF-β2 treatment but was inhibited by LY-364947, suggesting maximally activated TGF-β-receptor signalling in these cells (Fig. 6a). TGF-β2 upregulated P-p38 and p21 levels in PT-HEp3, BM-T1 and T-HEp3 cells (Fig. 6a,b). Further, TGF-β2 induced DEC2 and/or p53 (not in the mouse lines) mRNAs in human HNSCC and mouse breast cancer metastatic cells whereas LY-364947 treatment reduced their expression in BM-D1 and D-HEp3 cells (Fig. 6c,d and Supplementary Figs S4c and 5a). Further, knockdown of TGF-β2 or the TGF-β-RI inhibitor interrupted the dormancy of BM-D1 cells (Fig. 6e,f and Supplementary Fig. 5b). TGF-β2 also inhibited 4T1 tumour sphere formation (Supplementary Fig. 5c), a measure of self-renewal that was linked to their metastatic capacity28. Similarly, treatment of the dormant 4T07 cells28 with LY-364947 or SB203580 enhanced tumour sphere formation or proliferation per tumour sphere (Supplementary Fig. 5d,e). We conclude that dormancy of D-HEp3 and BM-D1 cells is dependent on TGF-β2, p38α/β, DEC2 and p53, which in turn induce p21, p27 and growth arrest. TGF-β2, TGF-β-RI and p38 signalling also seem to regulate self-renewal properties of breast cancer 4T07 cells that spontaneously enter solitary dormancy in lungs28.

Figure 6: Effect of TGF-β signalling on dormancy and self-renewal markers.
figure 6

(a) Immunblots for the indicated antigens on PT-HEp3 (PT), BM-D1 and BM-T1 cell lysates after TGF-β2 (2 ng ml−1) or TGF-β-RI inhibitor (LY-364947, 5 μM) treatment for 24 h in SF media. (b) Immunoblots for the indicated antigens on T-HEp3 cell lysates after treatment with TGF-β2 (2 ng ml−1) for 24 h in SF media. (c,d) qPCR-measured DEC2 (c) and p53 (d) mRNA levels in the indicated cells after treatment with 2 ng ml−1TGF-β2 or LY-364947, 5 μM for 24 h in SF media (n = 6 RNA samples were assessed over 3 independent experiments). (e) BM-D1 growth on CAMs for 4 days after TGF β 2 knockdown. BM-D1 cells were transfected with either scrambled (scr) or TGF β 2 siRNA and inoculated into CAMs (5×105 cells per CAM) for 4 days (n = 20 (scr), 19 (TGF β 2) tumours per condition). (f) BM-D1 growth on CAMs for 4 days after TGF-β-RI inhibition with LY-364947 (5 μM). BM-D1 cells were pre-treated with TGF-β-RI inhibitor (LY-364947, 5 μM) for 24 h in SF media and then inoculated into CAMs (5×105 cells per CAM) for 5 days. Tumour nodules were treated with LY-364947 (5 μM) every day in the CAMs (n = 9 tumours per condition). Data in c,d represent mean±s.e.m. In c–f, *P<0.05, ***P<0.001 by Mann–Whitney test. Uncropped images of blots are shown in Supplementary Fig. 6.

Full size image

TGF-β-RIII regulates TGF-β2 signalling for dormancy

In analysing the mechanism by which TGF-β1 and TGF-β2 signalling differs, we found that at short time points (10–30 min) both TGF-β1 and TGF-β2 equally activate SMAD1/2/3/5 phosphorylation with no effect on SMAD4/6 total protein levels (Fig. 7a); at 2 h, TGF-β1 and TGF-β2 also showed a similar response, but now total SMAD3/4/6 were upregulated. However, we found that only TGF-β2 treatment induced a sustained phosphorylation of SMAD1/5, SMAD2 and to some extent SMAD3 at longer times (24 h; Fig. 7b). In addition, TGF-β2 but not TGF-β1 activated p38α, induced p27 and DEC2 expression and downregulated CDK4 at 24 h (Fig. 7b–d). TGF-β2-induced SMAD2 phosphorylation and p27 upregulation was also inhibited by SB203580 (Fig. 7e). TGF-β2 requires the type III TGF-β receptor (TGF-β-RIII) for signalling, whereas this receptor is dispensable for TGF-β1 signalling29,30. T-HEp3 cells express all three TGF-β-receptors in vivo (Supplementary Fig. 5f) and RNAi against TGF β R3 had no significant effect on basal growth of these cells, but it completely eliminated the capacity of TGF-β2 to inhibit growth in vivo (Fig. 7f and Supplementary Fig. 5g). This correlated with the inability of TGF-β2 to induce long-term SMAD1/5 phosphorylation and p27 induction in TGF-β-RIIIlow cells (Fig. 7g). TGF-β-RIII was not required for SMAD2 or p38 activation by TGF-β2 (Supplementary Fig. 5h). TGF-β1 did not stimulate T-HEp3 in vivo growth and TGF-β-RIII downregulation did not affect this lack of effect by TGF-β1 on these cells (Figs 5i, 7h). We propose that TGF-β-RIII is required for the growth inhibitory function that TGF-β2 exerts on malignant HEp3 cells.

Figure 7: TGF-β2/TGF-β-RIII signalling regulates DTC dormancy.
figure 7

(a,b) Immunoblots for the indicated antigens on T-HEp3 cell lysates after treatment with either TGF-β1 or TGF-β2 (5 ng ml−1) for 10, 30 min and 2 h (a) and 24 h (b) in SF media. (c,d) qPCR for DEC2 (c) and p27 (d) mRNA levels in T-HEp3 cells treated with either TGF-β1 or TGF-β2 (5 ng ml−1) for 6 (d) and 24 h in SF media (n = 6 RNA samples were assessed over 3 independent experiments) *P<0.05 by Mann–Whitney test. Error bars denote s.e.m. (e) Immunoblots for the indicated antigens on cell lysates of T-HEp3 treated with 5 ng ml−1TGF-β2 for 24 h in SF media with or without SB203580 (SB; 5 μM). (f) Tumour growth on CAMs for 4 days of T-HEp3 cells with TGF-β-RIII knockdown (siRNA TGF β R3) treated with 5 ng ml−1TGF-β2 (scr, scrambled) (n = 16 (scr, SF), 11 (scr, TGF-β2), 15 (TGF β R3, SF), 17 (TGF β R3, TGF-β2)), *P<0.05 by one-way ANOVA with Bonferroni’s multiple comparison test. (g) Immunoblots for the indicated antigens on cell lysates of T-HEp3 treated with TGF-β2 (5 ng ml−1) for 24 h in SF media and after TGF-β-RIII knockdown (siRNA TGFR β 3). (h)Tumour growth on CAMs for 4 days after treatment with TGF-β1 (2 ng ml−1) in T-HEp3 cells after TGF-β-RIII knockdown (TGF β R3 siRNA) (n = 6 (scr, SF), n = 7 (scr, TGF-β1), n = 6 (TGF β R3 siRNA, SF), n = 6 (TGF β R3 siRNA, TGF-β1)), P = 0.3246 by one-way ANOVA with Bonferroni’s multiple comparison test. NS, not significant. Uncropped images of blots are shown in Supplementary Fig. 6.

Full size image

TGF-β-RI inhibition favours DTC escape from dormancy

Whereas we found that TGF-β2 may prevent DTC expansion by inducing dormancy, TGF-β1 signalling can promote metastasis31. To determine how the apparently competing functions of TGF-β1 and TGF-β2 signalling affect DTC behaviour, we systemically inhibited TGF-β-RI using LY-364947 and monitored DTC fate as in Fig. 3. Treatment with LY-364947 after primary tumour surgery approximately doubled the prevalence of spleen and liver DTCs (Fig. 8a) and increased BM DTCs by ∼ 5.6-fold (Fig. 8b). We also found that when compared with control animals, treatment with LY-364947 increased the prevalence and the total metastatic burden in lungs by ∼ 3-fold (Fig. 8c). These data support that in restrictive and permissive sites dormancy is a default state for DTCs and that signalling through TGF-β-RI (and possibly TGF-β2–TGF-β-RIII) might be at least in part responsible for the lack of DTC growth that is eliminated following LY-364947 treatment.

Figure 8: Effect of TGF-β-RI inhibition on DTC burden.
figure 8

(a) Prevalence of HEp3 DTCs (%) per mouse in control (n = 20) and LY-364947-treated (n = 32) animals (2 independent experiments). (b) Alu qPCR quantification of HEp3 DTCs in BM of mice treated either with dimethylsulphoxide or with LY-364947 10 mg kg−1 every 48 h for 2 weeks. Graph: mean±s.e.m. of Alu amplification signal normalized to mice GAPDH (n = 6 DNA samples from 6 different BM samples were assessed over 2 independent experiments), *P<0.05 by Mann–Whitney test. (c) Fluorescence micrographs of HEp3–GFP DTCs in mouse lungs, 2 weeks after control (scale bar, 80 μm) and LY-364947 treatment (scale bar, 120 μm). Lower graph: quantification of HEp3–GFP DTCs in lungs of control and LY-364947-treated animals. n = 4 mice per condition, *P<0.05 by Mann–Whitney Test. (d) Scheme of the proposed mechanism for TGF-β2-induced dormancy. Binding of TGF-β2 to TGF-β-RIII recruits TGF-β-RII and TGF-β-RI into the complex and activates TGF-β2 signalling, which in turn activates Smad1/5 and induces p27. In addition, TGF-β2 also activates p38α in a TGF-β-RIII-independent manner. In response to TGF-β2, p38α activates SMAD2 and DEC2, which induces p27 and inhibits CDK4. All of these signals integrate and contribute to TGF-β2 induction of quiescence.

Full size image

DISCUSSION

We explored the mechanistic basis of microenvironment-driven DTC dormancy, taking advantage of the HEp3 system of overt spontaneous metastasis, which in mice mimics that of HNSCC patients32. Despite their limitations, this model also mimics the presence of non-proliferative DTCs in the BM, as observed in HNSCC and breast cancer patients3. Although our monitoring of DTC dormancy in the BM spanned 4 weeks, which is equivalent to ∼ 3 years in humans, dormancy periods may span decades in patients. The fact that mice never develop BM metastasis argues that monitoring DTCs in this site for longer periods may be an approximate model of the human situation. Our data suggest that BM DTCs undergoing prolonged dormancy exhibit a set of markers (TGF-β2high,(ERK/p38)low,DEC2high,p53high,p27high and P-H3low) indicative of quiescence. 4T07 cells that enter dormancy in lungs when spontaneously disseminating23 also exhibited a dormancy profile ((ERK/p38)low and DEC2high) suggesting a metastasis-restrictive role for these pathways also in breast cancer cells. Our work now provides a set of microenvironment (for example, TGF-β2) and DTCs markers ((ERK/p38)low and DEC2high, and so on) that may be potentially used to determine the proliferative versus dormant state of DTCs in BM or other sites. Undoubtedly, testing these markers in DTCs from patients with advanced or no evidence of disease, is key to fully validating our findings33.

We found that at least TGF-β2 may define metastasis-restrictive versus -permissive microenvironments. However, other cues may also be important for BM DTC dormancy because TGF-β2 depletion from BM CM reduced p38 phosphorylation, but it did not activate ERK1/2. This could be due to the presence of BMP7, which induced an (ERK/38)low ratio and prostate tumour cell dormancy in BM (ref. 34), or to other signals such as the GAS6/Axl pathway35 that also regulate dormancy and would have to be fully eliminated to restore a (ERK/p38)high ratio and metastatic growth. In addition, higher TGF-β1 levels36 or downregulation of TGF-β-RIII37 may be needed to reprogram DTCs out of dormancy.

Our work also shows that TGF-β2, which owing to its low affinity for TGF-β-RII, requires TGF-β-RIII (refs 38, 39), creates a unique signal that results in prolonged SMAD1/2/5 activation and p27 upregulation to induce quiescence. How TGF-β2 and TGF-β1 activate different signals is unclear but it may depend on the composition and/or activity of the TGF-β-RI/TGF-β-RII complex40 (Fig. 8d). Our results showing that TGF-β2 did not require TGF-β-RIII for p38 activation further argues that different downstream complexes funnel the signals through the canonical (that is, SMADs) and non-canonical (that is, p38α) pathways. Together, our data support that at least in HNSCC cells, TGF-β2 signalling through TGF-β-RI and TGF-β-RIII provides a qualitatively different signal from TGF-β1 to induce growth arrest. As TGF-β-RIII can activate p38 to induce growth arrest41 and it can suppress tumorigenesis37,42,43, it is possible that TGF-β-RIIIhigh DTCs may be more prone to enter dormancy in TGF-β2-rich microenvironments such as the BM.

Our results using kinase inhibitors of p38α/β and TGF-β-RI support that both of these kinases hold multi-organ residual disease from expanding into metastasis. The fact that lung DTCs proliferated after TGF-β-RI inhibition suggests that to some extent the TGF-β2–TGF-β-RIII-dependent dormancy is transiently operational and then reversed by opposing signals, perhaps TGF-β1. In this regard, whereas TGF-β1 signalling has been linked to EMT and metastasis31, a large body of work supports the growth-suppressive function of TGF-β2 in normal and cancer cells44,45. Interestingly, the association of TGF-β-RIII loss with prostate bone metastasis development46,47 suggests that TGF-β-RIII makes DTCs impervious to the TGF-β2 growth-suppressive signal in the BM. We propose that dormancy is an initial default program for DTCs and that TGF-β2 growth-suppressive function through TGF-βR-I and -III might mediate these events. Other models and ours provide insight into the regulation of DTC biology. Validation in humans of our findings and those in other models34,35,48 will further our understanding of DTC dormancy and aid the development of rational therapies to target dormant disease.

METHODS

Cell lines and tumour growth studies.

Tumorigenic (T-HEp3) HEp3 cells were derived from a lymph node metastasis from a HNSCC patient as described previously49. Tumour growth on chick embryo CAMs or Balb/c nude mice was performed as described previously21. All animal studies were approved by Institutional Animal Care and Use Committees (IACUC) at Mount Sinai School of Medicine Protocol ID: 08-0366. Start: 10 Nov 2011. The mammary carcinoma cell line F3II is a highly invasive and metastatic variant, established from a clone of a spontaneous Balb/c mouse mammary tumour22. The 4T1 and 4T07 lines are derived from a single mammary tumour that arose spontaneously in a wild-type BALB/c mouse23. The lung-HEp3 (Lu-HEp3) and bone marrow-HEp3 (BM-HEp3) derived cell lines were established from lung and BM preparations respectively (see below). To have enough samples to analyse and account for the number of animals that are lost in each experiment (mice negative for DTCs or chicken embryos that die during the process) we have used at least 6–8 CAMs or 10 mice per group in all of the experiments. For the HEp3 xenograft studies, 5×105 cells were injected subcutaneously into BALB/c nu/nu female mice (4–6 weeks) in the interscapular region. Mice were inspected every 48 h and were kept (∼ 14 days) until primary tumours developed and grew up to ∼500–800 mm3. At this point the tumours were resected and mice were randomized and divided into control or treated groups without any knowledge of the primary tumour volume or latency. For the SB203580 (Calbiochem, CAS 152121-47-6) and LY-364947 (Calbiochem, CAS 396129-53-6) treatments, mice were injected with 10 mg kg−1 of drug intraperitoneally (i.p.) every 48 h for 2 or 4 weeks following day 2 after surgery. In the case of the F3II model, 4–6-week-old Balb/c female mice were injected with 10 mg kg−1 of drug intraperitoneally 24 h after surgery and then daily for 10 days. For TGF-β2, (R&D systems, 302-B2), TGF-β1 (R&D systems, 240-B) and LY-364947 experiments on CAM, HEp3 cells were pre-treated in culture for 24 h and then inoculated on CAMs for 5 days. CAMs were treated every day with TGF-β2 or TGF-β1 (2 ng ml−1) or LY-364947 (5 μM). Unless noted otherwise, all of the experiments were repeated at least three times.

Isolation and quantification of bone marrow and lung DTCs.

The bone marrow was flushed with a 26G needle in 1 ml of PBS with 2 mM CaCl2 and MgCl2 (PBS++) into a 15 ml tube. After Percoll density gradient centrifugation, the suspension was inspected under the fluorescence microscope for GFP+ cell detection. For cytokeratin 8/18 detection of BM DTCs the BM suspensions were fixed with 4% paraformaldehyde (PFA) in PBS for 20 min at 4° and cytospun onto slides (see the Immunofluorescence section). Bone marrow-HEp3 (BM-HEp3)-derived cell lines were prepared by counting the bone marrow suspension and seeded at 106 total cells in 10 cm dishes. At 48 h after plating, cells were washed with PBS. Selection for GFP+ cell populations was performed using 2.5 μg ml−1 puromycin.

To isolate lung GFP+ DTCs, the lungs were excised, dissected and placed in a Petri dish filled with PBS++. Following collection, the lungs were minced and enzymatically dissociated into single-cell suspensions by incubation with type 1A collagenase (Sigma-C9891) for 30 min at 37 °C. Lung-HEP3–GFP (Lu-HEp3) cells were counted with a haemocytometer under the fluorescence microscope. Quantification and outcome assessment of the in vivo studies were performed in a blinded fashion. For the Lu-HEp3 cells line preparation, the cell suspension was adjusted to 5×106 cells per flask and plated in 75 mm flasks; 48h after plating, cells were washed with PBS, selected with puromycin 2.5 μg ml−1 and expanded as indicated above. GFP+ cell integrity and viability was verified using phase contrast microscopy and trypan blue exclusion tests. All GFP+ cells were viable and not aggregates or cell fragments.

BM and lung condition media preparation.

BM and lung single-cell suspension from 6- to 10-week-old Balbc nu/nu mice were prepared as described above. The cell suspensions were counted and adjusted to 106 cells ml−1. Cells (5×106) were plated in 75 mm flasks in full serum media. At 24 h after plating, cells were washed and the medium was changed to serum-free medium and at 48 and 72 h the supernatant was collected, treated with protease and phosphatase inhibitors, filtered, concentrated using a Vivaspin concentrator (Viva products, VS20913) and frozen.

Metastasis assays in chicken embryos.

The experiments were performed as described previously26. Briefly, suspensions of single HEp3 cells obtained from 7-day-old CAM tumours that were dissociated with collagenase were not transfected or transfected for 24 h with either scrambled or p38 α siRNA and then inoculated into CAMs of embryonic day 10 (E10) chicken embryos (2.5×105 cells per CAM). On day 5 (E15) after inoculation, the liver and lungs of the embryos were excised, minced and dissociated with collagenase. The single-cell suspension was then processed as described above for the isolation of DTCs from mice lungs. The chicken BM was flushed with a 26G needle in 1 ml of PBS++ into a 15 ml tube and processed as described above for the isolation of mice BM DTCs.

Immunoblotting, RT–PCR and quantitative PCR (qPCR).

Immunoblotting and PCR with reverse transcription (RT–PCR) were performed as described previously8,50. All western blots were repeated at least three times with lysates from 3 independent experiments. A representative image is presented in the main and supplementary figures. To calculate the ERK to p38 activity ratio, the gel bands were quantified using ImageJ. The optical density (OD) values for P-ERK1/2 were divided by the OD values for total ERK1/2 and the same was done for P-p38 and total p38α. Then the P-ERK/ERK OD ratio value was divided over the P-p38/p38 OD ratio value.

For the RT–qPCR, 2 μg of total RNA isolated from HEp3 cells (Trizol reagent, Invitrogen) was reverse-transcribed using MMuLV RT (NEB) and then amplified by standard PCR using Taq DNA polymerase (NEB) following the manufacturer’s instructions. Primers were purchased from IDT. See Supplementary Table S5 for primer sequences.

Detection of DTCs through Alu−qPCR.

Human cells within mice BM were detected by real-time Alu qPCR, essentially as described previously15,25,51,52. Briefly, the BM flushes were snap frozen in liquid nitrogen and the genomic DNA was extracted using the Extract-N-Amp kit from Sigma (XNAT2-1KT). Thirty nanograms of genomic DNA was used per PCR reaction. Amplification of mice GAPDH was used as an internal control for the total amount of tissue. The ratio between hAlu and mGAPDH is used for comparative purposes between experimental groups. Each assay included 2 negative controls (water and genomic DNA extracted from BM from a mouse that had not been injected with human cells), a positive control (human genomic DNA), and the experimental samples in quadruplicate. The actual number of tumour cells present in each tissue sample was determined using a standard curve generated by serial dilution of HEp3 cells as previously described15,25. Briefly, BM cells were collected from nude mice that did not bear tumours. HEp3 cells were serially diluted into mice BM single-cell suspensions and the genomic DNA was extracted and amplified. By interpolating the Alu signal from experimental samples with the standard curve, the actual number of tumour cells per BM could be determined over a range of 10–1,000 cells per BM.

TGF-β2 and TGF-β1 depletion from BM conditioned media.

For depletion of TGF-β from the BM CM, agarose beads were pre-incubated with 10 μg of a monoclonal anti-TGF-β2 antibody (R&D systems, MAB612) or anti-TGF-β1 antibody (Santa Cruz Biotechnologies, sc-52893) or matched isotype IgG for 1 h at 4°. Then the beads were washed 3 times with serum-free media and incubated with 1 ml of the BM condition media overnight at 4 °C under rotation. Antibody antigen complexes were precipitated with protein G-agarose beads by a gentle spin and washed three times. The supernatant, CM depleted from either TGF-β2, TGF-β1 or undepleted (IgG-control), was then used to treat the cells for 24 h. IgG or anti-TGF-β1, and -TGF-β2 monoclonal-antibody-treated CM were also used to detect TGF-β1 or TGF-β2 expression using western blotting.

Tumour sphere assays.

Mammosphere assays were performed as previously described53. Single-cell suspensions of 4T1 and 4TO7 cells containing 250 cells were cultured on ultralow-attachment 24-well plates (Corning Costar) in serum-free mammosphere medium (DMEM F12 supplemented with 1:50 B27 (Invitrogen) and 20 ng ml−1 EGF (BD Biosciences)). Cells were treated with either 2 ng ml−1 TGF-β2, 5 μM SB203580 or 5 μM LY-364947, cultured for 7 days and then the tumour spheres were counted. For counting the cells per sphere, tumour spheres were collected, dissociated with trypsin for 15 min to obtain single-cell suspensions andcounted.

RNA interference.

Transfections of siRNAs targeting the desired sequences or, scrambled siRNA, were performed using siPORT NeoFX (Ambion). Detached cells were resuspended and overlaid onto the transfection complexes. After 24 h under normal cell culture conditions, cells were either used for in vivo experiments or lysed for immunoblotting and/or qPCR. DEC2 siRNA was from Santa Cruz Biotechnologies (sc37769), p38 α siRNA was from Cell Signaling (6564s), and p53 siRNA (AM51331) and TGF β 2 (s14061) were from Ambion.

Immunohistochemistry.

Paraffin-embedded sections from human primary tumours, lymph node metastasis, and T-HEp3, Lu-HEp3 and BM-D1 tumours were stained for DEC2, p27, phospho-histone 3 and cleaved caspase-3 expression after quenching endogenous peroxidase activity. Binding of the primary antibody (see Supplementary Table S6) was carried out overnight at 4°, followed by secondary antibody incubation (1 h at room temperature) and revealed using DAB Chromagen.

Immunofluorescence.

BM suspensions were fixed with 4% paraformaldehyde (PFA) in PBS and cytospun onto slides. Cytospins were prepared according to standard protocols. The BM cytospins and paraffin-embedded sections from lungs containing DTCs and micro-metastasis were permeabilized with 0.1% Triton X-100 and 0.01% SDS after performing antigen retrieval with citrate buffer. Cells were then blocked with 3% PBS/BSA containing 3% normal goat serum (NGS) for 30 min and stained overnight with the primary antibodies (see Supplementary Table S6) followed by secondary antibody incubation (1 h at room temperature).

Replication of experiments.

Unless noted, all of the experiments presented in the manuscript were repeated at least 3 times.