A Convergence Model of Bioelectric, Gap Junctional, and Hippo-YAP Signalling in Oral Cancer Stem Cell Maintenance

This mechano-electric bridge to Hippo–YAP is the molecular convergence point of the three-pillar hypothesis advanced in this review, and it is developed further in Section 5 and Section 6 in the context of bioelectric signalling and gap junctional communication respectively.

Membrane potential (Vmem) is not merely a metabolic epiphenomenon but an instructive regulator of cell fate and tissue organisation. Binggeli and Weinstein first proposed in 1986 that membrane depolarisation could constitute a general mechanism of growth regulation and cancer formation [17], a prediction supported by decades of subsequent evidence: resting Vmem is systematically more hyperpolarised in quiescent or differentiating cells and more depolarised in proliferating cells, and tumour cells across multiple cancer types are consistently more depolarised than their normal tissue counterparts [48]. A recent meta-analysis quantified this signature across 41 cancer cell types, finding mean polarisation differentials of approximately 16 mV in human and 32 mV in rodent cancer cells relative to their normal counterparts [49]. Levin and colleagues demonstrated in developmental biology models that Vmem patterns in embryonic tissues encode positional and fate information independently of genetic sequence, and that experimental manipulation of these patterns can redirect tissue-level outcomes including organ identity, regeneration, and tumour suppression [18]. Critically for this review, Saw et al. demonstrated that transepithelial potential difference (TEPD) governs epithelial homeostasis through an electro-osmotic mechanism in polarised epithelia: basal-to-apical TEPD promotes junction formation and uniform morphology, while reversal induces proliferative bud-like structures, establishing that Vmem is a tissue-level organiser, not a cell-autonomous property [46].

As introduced in Section 4.5, this two-scale architecture is fundamental to oral epithelial homeostasis: single-cell Vmem regulates individual cell-fate decisions through Mukherjee’s mechano-electro-osmotic mechanism, while tissue-level TEPD established by tight epithelial barrier functions provides the homeostatic endpoint signal that single-cell Vmem signalling drives toward [16,46]. Both scales are dysregulated in the OSF-to-OSCC continuum: individual oral epithelial cells experience Vmem perturbation from arecoline-induced ROS and altered ion channel function, while OSF tissue architecture compromises tight junction integrity and therefore the TEPD-based homeostatic reference.

The mechanistic bridge between Vmem and Hippo–YAP signalling was established by Mukherjee et al., who demonstrated in a non-oral epithelial system that changes in cellular biomass density alter Vmem through an electro-osmotic mechanism, and that this Vmem shift in turn regulates MST1 kinase activity: hyperpolarisation activates MST1 → LATS1/2 → YAP phosphorylation and cytoplasmic retention, while depolarisation reduces LATS activity and promotes YAP nuclear translocation [16]. Critically, Mukherjee et al. demonstrated that this Vmem–Hippo signalling axis is FAT1-dependent: in MCF10A cells, FAT1 knockdown abolished the YAP localisation response to valinomycin-induced depolarisation, identifying FAT1 not as a parallel co-regulator of Hippo activity but as a required scaffold for MST1 membrane recruitment in response to Vmem changes [16]. The implication for HNSCC is direct: in the 29.8% of HNSCC harbouring FAT1 loss-of-function mutations [36], the bioelectric pillar of the convergence framework is structurally decoupled from the Hippo pillar at the molecular level. Vmem changes, whether arising from arecoline-induced ROS, ECM stiffness perturbation, or ion channel dysregulation, cannot reach LATS1/2 to restrain YAP nuclear retention, because the FAT1-dependent MST1 membrane recruitment step is missing. FAT1 mutation therefore does more than amplify YAP activation; it produces a state in which YAP nuclear retention becomes uncoupled from the homeostatic Vmem feedback that would normally constrain it. This extrapolation from the experimental knockdown phenotype to FAT1-mutant tumours, while biologically plausible, has not been tested in OSCC.

This structural difference implies that the convergence model is not uniform across patients—the upstream bioelectric and gap junctional inputs can modulate Hippo activitiy only where FAT1 is intact, whereas in FAT1-altered tumours, nuclear YAP is sustained more directly through loss of the FAT1–MST1 restraint. The therapeutic consequences of this division are developed in Section 8.4.

As developed in Section 4.5, the bioelectric and mechanotransductive routes to YAP nuclear retention are coupled rather than parallel. Mukherjee et al. observe that Vmem is well-positioned to regulate any signalling pathway originating at the plasma membrane, identifying calcium signalling as an “ideal effector” of Vmem-mediated mechanical transduction given the strong voltage-dependence of calcium ion influx [16]. Critically, Piezo channels, the mechanosensors that drive YAP nuclear translocation in OSF [20], are themselves voltage-gated, with channel inactivation kinetics demonstrated experimentally to depend on cellular membrane potential [47]. This intrinsic voltage-sensitivity of Piezo1 means that the calcium influx response to a given mechanical stimulus depends on the cell’s Vmem state. In oral epithelium subjected simultaneously to ECM stiffening and Vmem dysregulation, the prevailing condition in OSF-to-OSCC progression, the two perturbations integrate at the Piezo1 channel itself rather than acting as independent insults. Beyond Piezo1, Mukherjee et al. additionally demonstrated that Vmem modulates mitogen-activated protein kinase (MAPK) signalling, possibly downstream of recently characterised Vmem–K-Ras coupling [16], suggesting that bioelectric dysregulation in OSCC may have regulatory consequences beyond Hippo–YAP that warrant investigation. Independent evidence from a non-epithelial system further supports a direct ion channel to Hippo link; Dupuy et al. demonstrated that transcriptional regulation of the potassium channel KCNA2 (Kv1.2) controls YAP/Hippo signalling and cell proliferation, establishing that specific ion channel activity can directly modulate Hippo pathway output [50].

Together, these findings support a multi-input bioelectric–Hippo axis in which membrane depolarisation, whether arising from altered ion channel expression, disrupted GJIC, or ECM-driven mechanosensory signals, converges on reduced LATS1/2 activity and sustained YAP nuclear retention through FAT1-dependent and -independent routes.

In the attractor framework developed in Section 3, sustained Vmem depolarisation is proposed to function as a landscape-deforming force in OSCC: by persistently reducing LATS1/2-mediated YAP phosphorylation independently of upstream oncogenic signals, it deepens the CSC attractor basin and raises the energy barrier to differentiation. No published study has simultaneously measured Vmem profiles, ion channel transcriptomes, and CSC marker expression within the same OSCC model—this constitutes the primary experimental gap motivating the bioelectric axis of the convergence framework. In the betel quid-associated OSCC context, the OSF tissue environment engages multiple Vmem-perturbing inputs simultaneously: arecoline-induced mitochondrial ROS alter ion channel function and shift Vmem toward depolarisation, providing a chemical carcinogen-driven bioelectric dysregulation route; ECM stiffening drives Piezo1-mediated calcium influx, with the magnitude of this response itself dependent on Vmem state through Piezo voltage-gating; and FAT1 loss-of-function in approximately 30% of HNSCC structurally decouples the bioelectric input from Hippo regulation. The mechanistic implication is that betel quid is uniquely damaging not because it engages multiple pathways additively but because the bioelectric and mechanotransductive systems normally provide redundant fail-safe that fail together; arecoline disrupts the Vmem-mediated route, OSF disrupts the mechanotransductive route, and the two routes are coupled at Piezo1 voltage-gating such that disruption of one amplifies dysregulation of the other. Whether this proposed multiplicative coupling produces measurably greater YAP nuclear retention than would be predicted by additive insults is directly testable and constitutes a tractable experimental question within the convergence framework.

Gap junctions are intercellular channels formed by connexin hemichannels that enable the direct cytoplasmic exchange of ions, second messengers (Ca2+, cAMP, IP3), and metabolites up to ~1 kDa between adjacent cells, constituting the principal mechanism by which individual cell bioelectric states are integrated into tissue-level electrical fields [51]. In oral epithelium, Cx43 (GJA1) and Cx26 (GJB2) are the predominant isoforms, with expression tightly coupled to keratinocyte differentiation state, positioning connexins as markers and mediators of epithelial cell fate [52]. Cx43 function is regulated beyond transcription: phosphorylation by Src and EGFR—both constitutively activated in OSCC—reduces channel conductance and accelerates protein degradation, providing a post-translational route to GJIC disruption that is independent of connexin gene expression [51,53]. In the majority of epithelial cancers, GJIC is substantially reduced relative to normal tissue through a combination of transcriptional downregulation, promoter hypermethylation of GJA1, oncogenic kinase-mediated channel inactivation, and redistribution of connexin from the plasma membrane to intracellular compartments [54,55]. This reduction correlates with tumour grade and invasive capacity across multiple cancer types. The functional consequence most relevant to this review is the loss of collective bioelectric error-correction: when GJIC is intact, a cell entering an aberrant bioelectric state receives corrective signals from its neighbours through shared ion flux; when GJIC is disrupted, local epithelial clusters can sustain aberrant Vmem and attractor states independently of tissue-level cues.

It should be noted that connexin biology in cancer is context-dependent: in some settings, connexins facilitate tumour progression through metabolic coupling with cancer-associated fibroblasts or metastatic niche formation, underscoring the need for OSCC-specific rather than extrapolated evidence [54].

Direct OSCC evidence for connexin dysregulation is limited in scope but consistent in direction. Immunohistochemical studies confirm reduced or redistributed Cx43 relative to normal oral mucosa [54]. The most detailed OSCC IHC study found that functional GJIC loss reflects cytoplasmic sequestration of Cx43 rather than transcriptional silencing—cytoplasmic Cx43 was significantly upregulated while membrane Cx43 was not reduced in aggregate [52]. Critically, high membrane Cx43 was the only independent predictor of shortened overall survival (p = 0.0088), suggesting a context-dependent tumour-promoter role for membrane-localised Cx43 in established OSCC that underscores why expression levels alone, without subcellular localisation data, are insufficient to infer GJIC status. This finding is corroborated in a 90-patient HNSCC tissue microarray: Cx43-negative tumours had median disease-specific survival of 5 months versus 15 months in Cx43-positive tumours (HR = 0.509; p = 0.004), though oral cavity primaries comprised only 5% of that cohort [56]. For the betel quid-specific OSF-to-OSCC transition, Mugundan et al. performed IHC for Cx43 across normal mucosa (n = 6), OSF (n = 14), oral epithelial dysplasia (n = 12), and OSCC-with-OSF history (n = 7): Cx43 showed a stepwise reduction across groups (p = 0.033), with OSF retaining moderate suprabasal expression, OED showing progressive loss, and OSCC-with-OSF showing near-complete loss in infiltrating tumour islands (OSF vs. OSCC-with-OSF: p = 0.014) [57]. Separately, Gutierrez-Camacho et al. reported that Cx43 expression serves as a biomarker discriminating OSCC from normal oral mucosa, with additional associations between Cx43 loss and HPV16/18 status [58].

This stepwise loss indicates that GJIC disruption occurs during, rather than prior to, malignant transformation from the OSF precursor field, consistent with a causal role, though causality has not been directly demonstrated. Functional GJIC also governs stem cell fate in progenitor contexts: Ho et al. demonstrated that gap junction-mediated calcium signalling controls blood progenitor cell fate decisions in haematopoiesis, establishing GJIC as an active determinant of stem cell state transitions [59], though the downstream effector in that system was CaMKII–JAK–STAT rather than CaMKII–MST1–Hippo, so the mechanistic analogy remains inferential. The critical gap is that no published study has simultaneously characterised connexin expression, GJIC functional capacity, and CSC marker expression within the same OSCC model.

In the attractor framework described in Section 3, the CSC state is a stable but locally accessible configuration on the gene regulatory network, a basin that cells normally cannot enter easily because the surrounding regulatory landscape maintains differentiated attractor states. The bioelectric model developed in Section 5 proposed that tissue-level Vmem dysregulation deforms this landscape, making the CSC basin more accessible. GJIC fragmentation extends this model by removing the collective error-correction mechanism that normally prevents local cells from entering aberrant attractor states independently of the tissue collective.

The proposed mechanism operates as follows. In normal oral epithelium, GJIC maintains synchronised bioelectric states across the epithelial sheet. A cell that transiently depolarises, perhaps in response to a local mechanical perturbation or a burst of ROS from carcinogen exposure, is rapidly corrected by ion flow from hyperpolarised neighbours through gap junctions, restoring it to the tissue-appropriate Vmem state and maintaining Hippo pathway activity. This collective buffering capacity means that individual cells rarely sustain the prolonged depolarisation needed to maintain YAP in the nucleus long enough to initiate the TAZ–SOX2 positive feedback loop.

When GJIC is fragmented, whether by connexin downregulation, Cx43 phosphorylation by oncogenic kinases, or fibrotic ECM remodeling, this buffering capacity is lost. Local clusters of cells become electrically isolated from the surrounding epithelium, allowing transiently depolarised cells to sustain their altered Vmem state without correction.

Computational models of coupled bioelectric-transcriptional networks formalise this principle. Cervera et al. demonstrated through simulations of multicellular ensembles that intercellular junction conductance is the parameter determining whether individual-cell Vmem perturbations are buffered through a community effect of neighbouring cells or whether they persist as stable local regionalisation [60]. In their model, strong intercellular coupling produces a synchronised polarisation state across the multicellular system, with individual aberrant cells rapidly corrected toward the tissue-level Vmem; weak coupling permits sustained local heterogeneity, with depolarised regions persisting independently of neighbouring cells and feeding back into distinct local transcriptional programmes. This computational framework provides theoretical grounding for the proposed mechanism: GJIC fragmentation in OSCC is expected to convert single-cell bioelectric noise, which would otherwise be transient and rapidly corrected, into stable multicellular regionalisations of aberrant Vmem state, each capable of sustaining local YAP nuclear retention long enough to engage to the TAZ–SOX2 self-reinforcing circuit. Although the Cervera et al. [59] model was developed for embryogenesis and regeneration contexts rather than carcinogenesis, the underlying principle—that GJIC determines whether cells experience their bioelectric environment collectively or autonomously—applies directly to the OSF-to-OSCC continuum, where progressive Cx43 downregulation and fibrotic disruption of gap junction plaques would predict precisely the kind of coupling collapse that converts buffered bioelectric perturbation into sustained local CSC-permissive states.

Sustained depolarisation, acting through the Mukherjee mechano-electro-osmotic mechanism, maintains YAP nuclear retention in these isolated clusters. If this state persists long enough to initiate the TAZ–SOX2 feedback circuit, the cluster transitions into the CSC attractor basin, a self-stabilising configuration that no longer requires the initiating bioelectric stimulus.

This mechanism predicts that GJIC disruption is not merely correlated with CSC emergence but is causally upstream of it, a prediction that is directly testable by restoring GJIC pharmacologically (e.g., with connexin-enhancing agents such as danegaptide or rotigaptide) and assessing whether CSC frequency is reduced even in the presence of ongoing carcinogenic exposure.

The relationship between GJIC and Hippo–YAP signalling has not been studied directly in OSCC, but indirect evidence from adjacent systems supports a mechanistic connection. Cx43 interacts physically with ZO-1 [61], a tight junction scaffolding protein that also regulates LATS1/2 activity through its association with the Hippo pathway component PTPN14 [62]. Disruption of the Cx43–ZO-1 interaction, which occurs downstream of Src-mediated Cx43 phosphorylation, itself activated by oncogenic signalling, may therefore reduce LATS1/2 activity and favour YAP nuclear retention through a connexin-proximal scaffolding mechanism, independent of Vmem effects.

Furthermore, intracellular calcium, the primary second messenger propagated through gap junctions during collective mechanosensing, is a known regulator of MST1/2 kinase activity. Calcium/calmodulin-dependent protein kinase II (CaMKII) has been shown to phosphorylate and activate MST1 in the context of Piezo1-mediated mechanosensing in macrophages, linking calcium influx to Hippo pathway kinase activation, as reviewed in [63]. By analogy, in an epithelium with intact GJIC, it is plausible that a mechanical stimulus induces a calcium wave that activates CaMKII → MST1 → LATS1/2 → YAP phosphorylation across the tissue collective—a homeostatic response that suppresses proliferation in response to crowding. When GJIC is disrupted, calcium wave propagation would be blocked, CaMKII–MST1 activation would not spread, and YAP nuclear retention could persist in isolated cell clusters even under conditions of normal tissue density. This mechanistic inference, though not yet demonstrated in an epithelial or OSCC context, provides a plausible direct biochemical link between GJIC disruption and Hippo pathway inactivation that is distinct from, and potentially additive to, the Vmem-mediated mechanism described in Section 4.

Together, the Vmem-mediated and proposed calcium wave-mediated mechanisms suggest that GJIC fragmentation acts on the Hippo–YAP convergence node through at least two parallel routes, making GJIC disruption a particularly potent contributor to CSC attractor stabilisation, and a correspondingly important therapeutic target.

The evidence presented in this section is explicit about its limitations. The connexin–OSCC CSC connection is built from general principles of connexin biology and cancer, limited OSCC-specific expression data, a single proof-of-concept study in haematopoietic progenitors, and mechanistic inference from connexin–Hippo pathway molecular interactions. No primary study directly tests the causal relationship between GJIC and CSC state in OSCC.

The gap is, however, precisely what makes it scientifically valuable. Unlike the bioelectricity gap, where the mechanistic framework exists but OSCC-specific application is absent, the GJIC–CSC gap in OSCC represents an almost entirely uncharted territory with clear biological plausibility, established adjacent evidence, and tractable experimental approaches. Characterising connexin expression across OSCC CSC subpopulations, testing whether GJIC restoration reduces CSC frequency in OSCC cell lines and patient-derived organoids, and examining whether connexin–Hippo pathway interaction is altered in OSCC, these are experiments that are feasible with existing tools and that would directly test a novel hypothesis.

The canonical pathways governing CSC maintenance and plasticity in OSCC—Wnt/β-catenin, Notch, Hedgehog, TGF-β, and PI3K/AKT/mTOR—have been reviewed in detail elsewhere, including our own comprehensive survey of the experimental CSC literature in OSCC [4]. The framework developed here is not intended to displace these pathways but to identify a biophysical convergence point that operates alongside and frequently intersects them; a complete account of the CSC state must situate the Hippo–YAP node within this broader network.

Several of these pathways are mechanically entangled with the convergence node rather than parallel to it. YAP/TAZ participate in the β-catenin destruction complex, coupling Hippo activity to Wnt output and allowing the two effectors to co-regulate overlapping stemness targets [64,65]; Notch and Hedgehog/GLI signalling likewise crosstalk with YAP through shared transcriptional programmes [66]. Two pathways are especially pertinent to the betel quid/OSF context: TGF-β, a principal driver of the fibrotic ECM modelling and EMT that characterise OSF, requires YAP/TAZ as co-effectors of SMAD-driven EMT, placing it upstream of and convergent with the Hippo node during malignant transformation [67,68]; and PI3K/AKT/mTOR, frequently activated in HNSCC functions [69,70] in parts as an effector arm of the convergence node, since YAP drives NRG1/AXL–mTORC1 signalling in oral tumour-initiating cells (Section 4.3) [40]. m-TOR-directed therapy, therefore, targets a downstream output of nuclear YAP rather than an independent input.

Positioned against this landscape, the present framework does not claim that Hippo–YAP is the dominant CSC pathway in OSCC; it proposes that biophysical inputs—membrane voltage and gap junctional coupling—feed a convergence node that is itself embedded in, and reinforced by, this wider stemness network. Because several canonical pathways channel through that node (Wnt–YAP, TGF–β–YAP, mTOR downstream of YAP), their intersection is consistent with the convergence logic rather than contradictory to it and clarifies why convergence-directed combination strategies (Section 8.3) may need to account for these parallel inputs.

The preceding sections have established three mechanistically interconnected lines of evidence. Section 4 demonstrated that Hippo–YAP dysregulation is genomically frequent and functionally essential in OSCC, and that the TAZ–TEAD4–SOX2 positive feedback loop stabilises the CSC transcriptional state as a self-reinforcing attractor. Section 5 established that Vmem dysregulation, through the mechano-electro-osmotic mechanism identified by [16], is an upstream determinant of YAP nuclear localisation, linking the bioelectric state of the oral epithelium to Hippo pathway activity. Section 6 proposed that GJIC fragmentation amplifies this through two parallel routes: loss of collective Vmem buffering and interruption of the proposed Ca2+ wave-mediated CaMKII–MST1–LATS1/2 activation. These three systems converge on a single regulatory output: nuclear YAP. When all three are simultaneously dysregulated—Hippo pathway inactivated, membrane depolarised, GJIC fragmented—nuclear YAP persists without correction, activates the TAZ–TEAD4–SOX2 GRN, and drives the oral epithelial progenitor into a stable CSC attractor state. The attractor is then self-sustaining: nuclear YAP activates SOX2, SOX2 represses NF2 and WWC1 (relieving upstream Hippo activation), reduced LATS1/2 activity sustains YAP nuclear retention, and the circuit runs independently of further external stimulus.

A critical structural question for the convergence model is whether the three pillars are independent parallel inputs to nuclear YAP or a mutually reinforcing triangle in which dysregulation of any one amplifies the others. The available evidence supports the latter. First, nuclear YAP drives c-Src kinase activation downstream, and activated Src phosphorylates Cx43 at Tyr247/Tyr265, preventing membrane localisation and closing gap junctions [53]; conversely, Cx43 restoration in glioma stem cells inhibits c-Src, suppresses Id1 and SOX2, and reverses the CSC phenotype [71], making the Cx43–Src–YAP interaction a self-reinforcing loop. The three nodes of this chain (Cx43 reduced in OSCC, c-Src constitutively activated, SOX2-driven self-renewal defining the CSC state) are each independently documented in OSCC, though their causal connectivity in oral squamous epithelium has not been directly tested. Second, Vmem dysregulation impairs GJIC directly through connexin hemichannel gating, while GJIC loss prevents tissue-level electrical buffering that normally corrects aberrant single-cell Vmem. This mutual coupling means that dysregulation of any one pillar tends to amplify dysregulation of the others, explaining why single-pathway targeting is predicted to produce incomplete and transient CSC attractor displacement, and why targeting the convergence node, nuclear YAP, is the most rational intervention point (Figure 2).

Conventional CSC-targeting strategies aim at elimination of marker-positive stem-like cells. The attractor framework reveals the fundamental limitation of this approach: if CSCs represent dynamic attractor states rather than a fixed subpopulation, elimination will be followed by regeneration as surviving non-CSC cells stochastically re-enter the attractor basin—precisely the phenotypic equilibrium reconstitution demonstrated by Li et al. [30]. Selective cytotoxic pressure may additionally deepen the attractor basin by eliminating non-CSC cells, enriching for therapy resistance. The attractor framework argues instead for deforming the landscape itself: raising the energy barrier around the CSC basin and simultaneously lowering barriers toward differentiated attractors. This reprogramming logic—targeting the regulatory landscape rather than any particular cell population—has proof-of-principle support in multiple cancer systems [72,73] and is the conceptual foundation for the therapeutic strategies discussed below.

The most direct therapeutic approach is inhibition of the YAP–TEAD transcriptional complex. Verteporfin and CA3 suppress YAP-driven EMT, migration, and xenograft growth in OSCC preclinical models [38]; in an arecoline-induced BALB/c mouse model of OSF, verteporfin treatment specifically suppressed YAP-driven endothelial–mesenchymal transition and reduced submucosal collagen accumulation, providing direct in vivo proof-of-concept for YAP–TAD inhibition as an OSF-stage chemoprevention strategy [23]. TEAD inhibitors IK-930 (NCT05228021) and VT3989 (NCT04665206) are in Phase I/II trials across Hippo-altered solid tumours; VT104, a related compound, significantly reduced clonogenic growth of YAP-activated oral epithelial progenitors in vitro [40], and Sato et al. identified FAT1-altered HNSCC as a potentially enriched-response population [36]. Downstream of YAP, Faraji et al. [40] established that YAP transcriptionally activates NRG1 and AXL, driving mTORC1 signalling, and that rapamycin markedly reduced carcinoma formation in vivo in YAP-activated oral progenitors, identifying TEAD inhibitor + mTOR inhibitor combination as mechanistically motivated (caveat: HPV-driven mouse model; betel quid-specific validation needed). Faraji et al. [40] additionally identified YAP-activated TI cells as the source of CXCL1/CXCL2 that recruit granulocytic MDSCs supplying MMP8/MMP9 collagenases for basement membrane invasion; both G-MDSC depletion (anti-LY6G) and CXCR1/2 inhibition (ladarixin) significantly reduced carcinoma incidence in vivo, motivating a second combination: TEAD inhibitor + CXCR1/2 blockade targeting cell-autonomous CSC programme and non-cell-autonomous invasion in parallel.