TGFβ2 dictates disseminated tumour cell fate in target organs through TGFβ-RIII and p38α/β signalling

The Bone Marrow Microenvironment Induces DTC Dormancy

HEp3-GFP cells obtained from xenografts were injected subcutaneously (s.c.) in nude mice and when primary tumours reached ~500 mm³ (t0) they were surgically removed. Then, HEp3 spontaneous DTCs were followed using GFP fluorescence and by human Alu sequence-specific qPCR¹⁵ to identify DTCs undetectable via GFP, or 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, Supplementary Table S1, Fig S1a–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 10–10²/BM, was >2 logs lower than in lungs (Fig. 1a, b, Supplementary Fig. S1a, 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. S1e), initiated vigorous proliferation 2 weeks after surgery (Fig. 1b).

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, while all 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 hypothesize that the BM microenvironment may instruct DTCs to activate long-lasting dormancy programs. We could only study mechanistically those BM DTCs that expanded in vitro and focused on BM vs. 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 (i.e. phospho-H3 (P-H3)) and positive/high for CDK inhibitor expression (i.e. p21, p27^Cip). 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) system^17–19 or in nude mice⁸. In all cases Lu-HEp3 and PT-HEp3 cells were tumorigenic and were P-H3^high/p27^low (Fig. 1c, 1d, Supplementary Fig. S1f). In contrast, one BM-HEp3 cell line remained dormant for 4–5 weeks before growing vigorously (henceforth BM-D1) (Fig. 1e, Supplementary Fig. S1g); dormant BM-D1 cells were P-H3^low/p27^High (Fig. 1d, Supplementary Fig. S1f), 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 s.c., BM-D1 cells also formed tumours later than BM-T1 cells (Supplementary Fig. S1h). This data suggest that while ~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-DTCs cell lines from the chicken and turkey embryos BM because we could process larger numbers of animals and increase our chances of obtaining BM DTC lines (Supplementary Fig. S1i). DTCs were non-proliferative in avian BM but did proliferate in the liver and lungs where metastases develop²⁰ (Supplementary Fig. S1j). As in mice, the number of DTC in the BM was 2 logs lower than in the liver and lungs (Supplementary Fig. S1j). Importantly, 7/9 (77%) BM-DTC cell lines remained dormant in vivo (Fig. 1f- and Supplementary Fig S1k). We conclude that while 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 Bone Marrow Dormant DTCS

We next measured previously identified key regulators of dormancy^8,21. Proliferative Lu-HEp3 and BM-T1 cells from the murine system and BM-T2 and BM-T3 cells from the avian system displayed a high ERK/p38 ratio, consistent with their tumorigenicity. In contrast, the dormant BM-D1 and BM-D2, D3, and D4 cells from murine and avian BM, respectively, displayed persistently low ERK/p38 activity ratio (Fig. 2a, 2b); 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 F3II²² - [ERK/p38]^high) or dormant (4T07- [ERK/p38]^low) behaviour of mouse breast cancer cell lines in lungs²³ (Fig. S2a). Thus, a high ERK/p38 signalling ratio, (see Methods for quantification) is predictive of tumorigenic and dormant phenotype in 100% (n=9) and 90% (n=9) of the DTC lines, respectively (Fig. 2c). BM-D1 cells also up-regulated p53 (mRNA and protein) and DEC2 mRNA, two genes previously linked to dormancy and quiescence⁸, when compared to 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 vs. F3II and 4T1 metastatic cells (Fig. 2d, Supplementary Fig. S2b, c).

Knockdown of p38α, DEC2 or p53 interrupted the dormancy of BM-D1 cells (Fig. 2e, Supplementary Fig S2d) 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 a constitutively active p38α (p38α-CA) mutant or wt-DEC2 cDNAs in proliferative Lu-HEp3 cells reduced P-H3 levels, did not affect apoptosis and inhibited their in vivo proliferation (Fig. 2f, Supplementary Fig. S2e, S2f); 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 compared to 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. S2g). 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^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 cytokeratin 8/18 (CK) staining⁷, showed that 80% were p27^high, 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 positive DTCs (Supplementary table S3, Fig. 3b). While 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, Fig . 3c, 3d).

In the chick embryo system of metastasis²⁵ the liver and lung are permissive sites for metastasis^20,26; liver DTCs (>10⁵ cells/organ) are P-H3^high/p27^low and proliferate actively (Supplementary Fig. S3a). The avian BM is also restrictive as DTCs (10¹–10² cells/BM) were also negative for P-H3 and positive for p27 (Supplementary Fig. S3a). After matching the primary tumours sizes (Supplementary Fig. S3b) we found that knockdown of p38α in T-HEp3 cells (siRNA knockdown can last for 5 days), increased the amount of DTCs in avian BM 5 days after inoculation (Fig 3e, Supplementary Fig. S3c). This increase correlated with a decrease in the percent of p27^high 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. S3d).

Solitary HEp3 lung DTCs identified by detecting human Vimentin (abundantly expressed in these cells - Supplementary Fig. S3e), were mostly negative for P-H3 and 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 raised in lungs, these lesions were P-H3^high and C-casp3^low (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 treatment (Fig. 4c, d). We conclude that p38α/β signalling limits the expansion of DTCs in growth-restrictive sites like 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 dramatic impact on disease burden in this tissue.

TGFβ2 Signalling in the BM Induces Dormancy Hallmarks

Expression profiling of proliferating (T-HEp3 [ERK/p38^high]) and dormant D-HEp3 ([ERK/p38^low]) cells in vivo revealed that TGFβ2 mRNA (not TGFβ1 or TGFβ3) is highly elevated only in dormant cells (Supplementary Fig. S4a). TGFβ2, which is present in the BM²⁷ activated p38, induced DEC2 and TGFβR-I-dependent quiescence in D-HEp3 cells (Supplementary Fig. S4b–d). In addition, dormant BM-D1 cells express >5-fold higher levels of TGFβ2 than parental T-HEp3 cells (T) and Lu-HEp3 cells (Fig. 5a). Importantly, we also found that TGFβ2 mRNA was greatly enriched in naÔve BM over lung tissue (Fig 5b) and that in basal conditions (serum free media) T-HEp3 cells express marginal TGFβ2 mRNA (Fig. 5a, c). Upon treatment of T-HEp3 cells with lung CM, TGFβ2 mRNA is increased. However, treatment with BM CM vigorously upregulated TGFβ2 expression (>10 fold - Fig. 5c), 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. S4e). Lung CM enhanced phospho-ERK1/2 levels (Fig 5f) and none of the organ-CM induced apoptosis (Supplementary Fig S4f). Importantly, TGFβ2-depleted BM CM was incapable of inducing p38 activation and growth inhibition of T-HEp3 cells in vivo (Fig. 5f, 5h) while lung CM supplemented with TGFβ2 inhibited T-HEp3 tumour growth in vivo (Fig. 5h). TGFβ2 depletion from BM CM did not restore ERK1/2 phosphorylation (see discussion) (Fig. 5f) and significantly, depletion of TGFβ1 from the BM CM, did not affect its ability to activate p38 or inhibit tumour growth (Supplementary Fig. S4g). In fact, treatment with TGFβ1, rapidly switched of 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 suggests that TGFβ2 but not TGFβ1 is a candidate mediator of p38 activation, DEC2 expression and dormancy in the BM.

We also found that the high basal p38 phosphorylation in BM-D1 cells was not stimulated by TGFβ2 treatment but was inhibited by LY-364947 suggesting maximally activated TGFβ-receptor signalling in these cells (Fig. 6a). TGFβ2 upregulated P-p38 and p21 levels in PT-HEp3, BM-T1 and T-HEp3 cells (Fig. 6a, b). Further, TGFβ2 induced DEC2 and/or p53 (not in the mouse lines) mRNAs in human HNSCC and mouse breast cancer metastatic cells while LY364947 treatment reduced their expression in BM-D1 and D-HEp3 cells (Fig. 6c, 6d, Supplementary Fig. S4c, S5a). Further, knockdown of TGFβ2 or the TGFβRI inhibitor interrupted the dormancy of BM-D1 cells (Fig. 6e, 6f, Supplementary Fig S5b). TGFβ2 also inhibited 4T1 tumour sphere formation (Supplementary Fig S5c), a measure of self-renewal that was linked to their metastatic capacity²⁸. Similarly, treatment of the dormant 4T07 cells²⁸ with LY364947 or SB203580 enhanced tumour sphere formation or proliferation per tumour sphere (Supplementary Fig S5d, e). We conclude that dormancy of D-HEp3 and BM-D1 cells is dependent on TGFβ2, p38α/β, DEC2 and p53, which in turn induce p21, p27 and growth arrest. TGFβ2, TGFβRI and p38 signalling also appear to regulate self-renewal properties of breast cancer 4T07 cells that spontaneously enter solitary dormancy in lungs²⁸.

TGFβ-receptor-III regulates TGFβ2 signalling for dormancy

In analyzing the mechanism by which TGFβ1 and TGFβ2 signalling differs we found that at short time points (10 – 30 minutes) both TGFβ1 and TGFβ2 equally activate SMAD1/2/3/5 phosphorylation with no effect on SMAD4/6 total protein levels (Fig. 7a); at 2 hrs, TGFβ1 and TGFβ2 also showed a similar response, but now total SMAD3/4/6 were upregulated. However, we found that only TGFβ2 treatment induced a sustained phosphorylation of SMAD1/5, SMAD2 and to some extent SMAD3 at longer times (24h - Figure 7b). In addition, TGFβ2 but not TGFβ1 activated p38α, induced p27 and DEC2 expression and downregulated CDK4 at 24 hrs (Figure 7b–d). TGFβ2-induced SMAD2 phosphorylation and p27 upregulation was also inhibited by SB203580 (Figure 7e).

TGFβ2 requires the type III TGFβ receptor for signalling, while this receptor is dispensable for TGFβ1 signalling^29,30. T-HEp3 cells express all three TGFβ-receptors in vivo (Supplementary Fig S5b) and RNAi to TGFβRIII had no significant effect on basal growth of these cells, but it completely eliminated the capacity of TGFβ2 to inhibit growth in vivo (Figure 7f, Supplementary Fig. S5c). This correlated with the inability of TGFβ2 to induce long-term SMAD1/5 phosphorylation and p27 induction in TGFβRIII^low cells (Figure 7g). TGFβRIII was not required for SMAD2 or p38 activation by TGFβ2 (Supplementary Fig. S5d). TGFβ1 did not stimulate T-HEp3 in vivo growth and TGFβRIII downregulation did not affect this lack of effect by TGFβ1 on these cells (F5i, 7h). We propose that TGFβRIII is required for the growth inhibitory function that TGFβ2 exerts on malignant HEp3 cells.

TGFβRI inhibition favours DTC escape from dormancy

While we found that TGFβ2 may prevents DTC expansion by inducing dormancy, TGFβ1 signalling can promote metastasis³¹. To determine how the apparently competing functions of TGFβ1 and TGFβ2 signalling affect DTC behaviour, we systemically inhibited TGFβ Receptor-I using LY-364947 and monitored DTC fate as in Fig3. Treatment with LY-364947 after primary tumour surgery approximately doubled the prevalence of spleen and liver DTCs (Figure 8a) and increased BM DTCs by ~5.6 fold (Fig. 8b). We also found that compared to 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 via TGFβRI (and possibly TGFβ2-TGFβRIII) might be at least in part responsible for the lack of DTC growth that is eliminated upon LY-364947 treatment.