Abstract

Background

Redox signaling caused by knockdown (KD) of Glutathione Peroxidase 2 (GPx2) in the PyMT mammary tumour model promotes metastasis via phenotypic and metabolic reprogramming. However, the tumour cell subpopulations and transcriptional regulators governing these processes remained unknown.

Methods

We used single-cell transcriptomics to decipher the tumour cell subpopulations stimulated by GPx2 KD in the PyMT mammary tumour and paired pulmonary metastases. We analyzed the EMT spectrum across the various tumour cell clusters using pseudotime trajectory analysis and elucidated the transcriptional and metabolic regulation of the hybrid EMT state.

Results

Integration of single-cell transcriptomics between the PyMT/GPx2 KD primary tumour and paired lung metastases unraveled a basal/mesenchymal-like cluster and several luminal-like clusters spanning an EMT spectrum. Interestingly, the luminal clusters at the primary tumour gained mesenchymal gene expression, resulting in epithelial/mesenchymal subpopulations fueled by oxidative phosphorylation (OXPHOS) and glycolysis. By contrast, at distant metastasis, the basal/mesenchymal-like cluster gained luminal and mesenchymal gene expression, resulting in a hybrid subpopulation using OXPHOS, supporting adaptive plasticity. Furthermore, p63 was dramatically upregulated in all hybrid clusters, implying a role in regulating partial EMT and MET at primary and distant sites, respectively. Importantly, these effects were reversed by HIF1α loss or GPx2 gain of function, resulting in metastasis suppression.

Conclusions

Collectively, these results underscored a dramatic effect of redox signaling on p63 activation by HIF1α, underlying phenotypic and metabolic plasticity leading to mammary tumour metastasis.

Discussion

Our findings highlight the far-reaching effects of GPx2 dysregulation on ROS/HIF1α signalling underlying EMT and metastasis. Our study provides novel insights into the diverse mammary tumour subpopulations and transcriptional regulators driving phenotypic and metabolic plasticity.

Single cell transcriptomics unravelled dramatic effects of GPx2 KD on EMT, resulting in tumour subpopulations drifting through an EMT continuum. Notably, GPx2 KD caused de novo induction of mesenchymal genes in luminal clusters as well as enhanced the mesenchymal signature of cluster 3. By arbitrarily placing cluster 3 at the origin of the lineage trajectory as the mesenchymal transcriptional node, we uncovered closest proximity between cluster 5 and 3 in pseudotime distance, implying cluster 5 might be an early hybrid cluster. This is consistent with the hybrid metabolic status of cluster 5, underscoring a link between phenotypic and metabolic adaptation [9]. In fact, studies in skin, breast and pancreatic cancers unravelled early and late hybrid EMT clusters, with the early clusters in breast exhibiting robust tumour initiating and metastatic potential [15, 16].

These studies argue against mesenchymal tumour cells as the pivotal drivers of metastasis, which precludes mesenchymal cluster 3 from initiating this process [39]. In fact, others have shown that E/M cells were tumour-initiating and metastatic relative to E or M cells [13, 14]. We speculate that the M subpopulation (cluster 3) lies at the apex of the hierarchy, giving rise to bipotent progenitors that differentiate into an E or M state. Alternatively, E carcinoma cells may convert into E/M and then M trough a stepwise transition process. Interestingly, we showed that luminal clusters in primary tumour gained mesenchymal genes, whereas the mesenchymal cluster 2 in the lungs acquired de-novo luminal gene expression. Hence, we speculate that E/M or M/E cells may be bipotent progenitors that differentiate into E or M clusters in response to TME cues [40].

HIF1α appears to play a pivotal role in EMT dynamics by promoting mesenchymal gene expression in luminal as well as in mesenchymal clusters at the primary tumour, thereby generating hybrid E/M and extra-M subpopulations. HIF1α is known to activate Snail1/2, which in turn represses Cdh1 among other epithelial genes, causing E to M switch. However, HIF1α may independently activate mesenchymal genes such as Vim, Twist2, and Cdh2 in luminal clusters [25, 41], thus generating an E/M phenotype. Of note, Twist, and not Vim, was upregulated in cluster 2 in lung mets relative to cluster 3 in primary tumour, thus supporting other findings that N-CAD (downstream of TWIST) was more critical than VIM in driving the metastatic process [42].

Importantly, our data highlight the notion that transcriptional regulation of partial EMT or partial MET occurs via p63, which corroborates findings showing tight transcriptional control of the hybrid state by p63 [37]. Indeed, we showed that hybrid cells in GPx2 KD primary tumour and lung mets were enriched in p63, thus supporting a critical role in the onset of hybrid populations at both sites. Remarkably, p63 expression was enriched in micro-, but not, in macro- metastases, suggesting p63 regulates partial MET that progress into full MET upon proliferation/differentiation of metastatic clusters. Others have shown that p63 was co-localised with a partial EMT programme at the leading edge of head and neck tumours [43], and was also enriched in collectively migrating cells or CTCs containing E/M cells [44, 45]. Of note, p63 promotes mammary gland maturation via transactivation of NRG1 in the luminal layer by p63 in the basal layer, resulting in ERBB4/STAT5 signalling driving luminal differentiation [36], a process which remains to be validated in vivo in the tumour context.

Interestingly, GPx2 KD appears to regulate p63 via HIF1α, leading to the onset of partial EMT. Of note, p63 KD in SUM159 basal-like BC cells suppressed lung colonisation, suggesting a critical role for p63 in metastasis [37, 46]. Indeed, we showed that echinomycin inhibits p63 and E/M marker expression while eradicating metastasis, thus suggesting p63 promotes tumour spread via the EMT hybrid state. Whether this regulation is direct or indirect remains to be validated. In favour of an indirect mechanism, we posit that HIF1α perturbation, which inhibits mesenchymal gene expression, destabilises p63 and hence the E/M state. Moreover, GPx2 KD may regulate this process via oxidation/inactivation of tyrosine phosphatases by ROS [47], which results in EGFR/STAT3 signalling, leading to p63 gene transcription [48, 49]. Otherwise, HIF1α may activate Wnt signalling in hypoxic niches, thereby upregulating p63, a well-documented Wnt target gene [47, 49, 50].

Other than promoting EMT hybrid states, HIF1α regulates tumour metabolism, and these processes might be tightly interwoven [30, 51]. In fact, echinomycin suppressed VIM, SNAIL, pAMPK and GLUT1 implying far-reaching effects of HIF1α on EMT and metabolism. Indeed, echinomycin suppressed OXPHOS (OCR) and glycolysis (ECAR) as well as pAMPK and GLUT1 which are markers of these metabolic processes. These data were consistent with activation of AMPK by HIF1α, which in turn stimulates fatty acid oxidation (FAO) and hence OXPHOS [52,53,54]. By contrast to the clusters in the primary tumour which mostly utilised glycolysis, lung mets clusters, especially cluster 2, used OXPOS likely as an adaptive response to the oxygenated environment in the lungs [5, 6, 51, 55]. The relationship between EMT and metabolism remains unclear despite reports of mutual cross regulation [56]. A study of 180 cancer cell datasets showed that the mesenchymal phenotype positively correlates with glycolysis and inhibition of OXPHOS, whereas the epithelial phenotype favours OXPHOS/FAO [57, 58]. Hence, it is not surprising that cells transiting through an hybrid EMT state may use OXPHOS and glycolysis to allow for metabolic flexibility in an evolving TME. Thus, cross regulation between EMT and metabolism warrants further investigation.

The plasticity of the hybrid EMT state remains debatable. Studies using gene regulatory networks from lung cancer single cell datasets, were able to map teams that were “strong” yielding stable phenotypes such as the epithelial or mesenchymal state, or “weak” yielding unstable or plastic phenotype such as the E/M hybrid state [13]. We annotated published team factors [57] in our single cell datasets but were unable of detecting any of the candidate team factors across our clusters. This is however not surprising in light of the high sparsity/drop out rate that is associated with the single cell RNA technology. Team players set aside, empirical evidence suggests that the hybrid EMT state may reside in a transiently stable state [40, 59]. In fact, E/M cells which were isolated from basal-like BC cell lines based on CD104 and CD44 expression, were stable in culture over several passages [13]. Moreover, E/M and M cells were shown to be dependent on Snail and Zeb-1 expression respectively. Hence, trapping carcinoma cells in the E/M state, was achieved by Snail overexpression combined with Zeb-1 knockdown [13]. Since Snail is negatively regulated by Mir 34, and Zeb-1 by Mir 200, we surmise that regulatory networks causing differential expression of these microRNAs contributes to the establishment of the hybrid EMT state [57, 58]

Altogether, our findings illustrate the complex dynamics regulating tumour progression and metastasis with emphasis on GPx2KD/HIF1α/p63 signalling driving the onset of tumour cell subpopulations undergoing phenotypic and metabolic plasticity. Such novel insights have the potential to inform therapeutic strategies for metastatic breast cancer.