Long-range gap junctional signaling controls oncogene-mediated tumorigenesis in Xenopus laevis embryos

In particular, we assume that the expression of a dominant negative H7 disrupts response to the contralateral signal (Figures 5C,D) on that side in two ways, by blocking the % for % tumor-suppressing capability of increased contralateral signal, and by causing cells in that same region to increase production of their own (ipsilateral) signal by 50%. We assume the presence of tumor cells on either side disables the increased signal production on that side. The model also requires that tumor cells, but not normal cells, on the contralateral side respond to this increased signal by further suppressing cell division; otherwise the division of normal cells would be over-suppressed. This restriction corresponds biologically to a response saturation on the part of normal cells that is disabled in tumor cells. We model Cx26 as having the opposite effect of increasing response to contralateral signal and decreasing production of ipsilateral signal. We also assume that local disruption of GJC by H7 decreases KRASG12D tumor incidence by 20%, while local enhancement of GJC by Cx26 increases tumor incidence by 20%; these local effects are insensitive to contralateral signal.

With these assumptions, it is straightforward to calculate the results expected from the experimental manipulations shown in Figures 2, 3. A single injection of KRASG12D on one side of the embryo produces 35% tumors (Figures 2, 3, treatment mode A); this baseline number provides the starting point for calculations. An ipsilateral H7 background decreases cellular response to KRASG12D by 20%; hence the expected tumor incidence following Figure 2, treatment mode B is 80% of 35%, i.e., 28%, a prediction error of only 0.3% compared to the observed value (Figure 6A, treatment mode B). The more interesting case is a contralateral H7 background (treatment mode C). Here contralateral H7 causes the contralateral cells to increase signal production by 50%; intuitively, the contralateral cells cannot perceive the signal produced by their neighbors due to disruption of GJC by H7, so they turn up their own signal production to compensate. The KRASG12D-ipsilateral side, therefore, receives 50% more contralateral signal. Increased contralateral signal turns KRASG12D response down on a % for % basis, so the expected tumor incidence following Figure 2, treatment mode C is 50% of 35%, i.e., 17.5%, a prediction error of 2.2% compared to the observed value. In treatment mode D, both sides of the embryo have an H7 background. Signal production is increased on the side contralateral to KRASG12D, but the tumor-suppressing capability of this increased contralateral signal is disrupted by the ipsilateral H7. The situation is, therefore, the same as in treatment mode A; KRASG12D activity is only decreased by 20% due to the ipsilateral H7 background. This prediction of 28% tumors has a prediction error of only 1.2% (Figure 6A, treatment mode D).

Treatment modes E and F both involve KRASG12D injections on both sides of the embryo; here the 44.9% tumor incidence observed with treatment mode G is used as the baseline, i.e., a baseline of 22.45% tumors is assumed on each of the two sides. With KRASG12D present on both sides, neither side can increase signal production in response to H7. In treatment mode E, H7 is present on one side, so the predicted tumor incidence is 20% less on that side only; hence the total predicted incidence is (80 × 22.45%) + 22.45% = 40.4%, a prediction error of 1.7%. In treatment mode F, H7 provides 20% tumor suppression on both sides, so the total predicted incidence is (80 × 22.45%) × 2 = 35.9%, a prediction error of 2.2% compared to observations (Figure 6A, treatment mode F). The predictions for Cx26 backgrounds are performed similarly (Figure 6B).

The model assumptions used here effectively specify a control structure (Figure 5E). Each arrow in this structure represents a cellular process regulating response to KRASG12D, H7, or Cx26. To objectively test the apparent ability of the model to correctly predict the observed data (Figures 2, 3), we performed a chi-squared test between the model's predicted outcomes (“expected”) and the actual data (“observed”) for the various experimental setups. For the H7 injections, X2 = 0.547 (df = 4), p = 0.97, while for the Cx26 injections, X2 = 5.264 (df = 4), p = 0.26, thus showing no significant difference between the model's predictions and the experimental data. We conclude that the emergent dynamics of this quantitative model accurately reproduce the complex experimental dataset linking GJC, Vmem control, and resulting tumor incidence in various spatial configurations.

We next tested a novel, surprising prediction of this model that would not have been made without it (by inspection of the experimental data alone): that effectiveness of ion channel injections on tumor suppression will be different depending on which embryonic axis is used to separate the sites of injection.

If two neighboring populations of cells are alternating between polarization and depolarization with opposite phase, so that A-side cells are polarized when B-side cells are depolarized and vice-versa, then the cells forming the border between the two populations must maintain a polarization value very near the average for the two populations. These cells, which form the borders between red and blue in Figure 4 can, therefore, be expected to have qualitatively different behavior from cells away from these borders. Therefore, the model also predicts that the ability of GJC reagents to alter the incidence of tumorigenesis should be different depending on whether the opposite sides of the oncogene/H7 axis are oriented with respect to the left-right or dorso-ventral embryonic axes, since the left-right (LR) axis is the one along which gradients of resting potential are known to exist (Levin et al., 2002; Aw et al., 2008). We tested this prediction (Figure 7), and found that indeed the GJC state is most relevant when the opposite-sided distribution of H7-oncogene mRNA coincides with the early embryo's left-right axis. This stage-1 model thus makes unexpected predictions that are supported by experimental test.

We examined the role of GJC in oncogene-mediated tumorigenesis in Xenopus embryos by targeted molecular-genetic modulation. Injection of mRNA constructs that modify gap junctional states in Xenopus significantly affected the incidence of tumorigenesis, as detected by increases or decreases in tumor formation after oncogene expression (although the absolute magnitude of the effect of GJ-targeting reagents was limited by the need to use low doses of mRNA to avoid disrupting normal development). Reasonably, the reagent that disrupts GJC and the one that enhances GJC affect tumor incidence in opposite directions. However, we were surprised to see that abrogation of GJC actually suppressed the effects of oncogenes, in contrast to prior suggestions that loss of GJC was a hallmark of incipient cancer (Vine and Bertram, 2002; Sirnes et al., 2012). Interestingly, in addition to confirming in vivo the cell-autonomous role inferred for GJCs by cell culture and clinical test data, we found that the greatest impact on tumor incidence occurs when we alter the GJC of cells far away from the tumor. While the microenvironment is increasingly seen to play an important role in cancer (Kenny and Bissell, 2003; Maffini et al., 2005; Hendrix et al., 2007; Kenny et al., 2007; Kasemeier-Kulesa et al., 2008; Tarin, 2012), long-range developmental signals are not yet widely addressed. It should be noted however that two very early workers in the bioelectricity of cancer had previously suggested that such a long-range effect should exist (Burr et al., 1938, 1940; Burr, 1941; Nordenström, 1983).

Targeted expression of wild type connexin Cx26 and dominant negative connexin H7 allowed us to probe how cell:cell communication affects KRASG12D activity. Induction of GJC host-wide or non-local to KRASG12D-expressing cells led to increases in tumor incidence. Enhanced GJC within KRASG12D-expressing cells did not affect tumor incidence. While the increases in tumor incidence seen as a result of GJC induction by Cx26 are significant, the impact of GJC on tumorigenesis was less than expected, partly due to the non-linearity in dose dependence of KRASG12D tumor formation. We demonstrated this non-linearity in dose dependence by showing that doubling KRASG12D only increases tumor incidence by 9%. Conversely, inhibition of GJC by the dominant negative H7 led to tumor suppression regardless of where H7 was introduced. Both local and host-wide GJC significantly reduced tumor incidence, and long-range disruption of GJC had the most impact on tumor incidence, resulting in 15.8% less embryos with tumor. These results suggest that oncogene-expressing cells utilize the network of GJs between tumors and healthy tissue (Figure 1) to drive tumorigenesis. Most importantly, these data suggest treatment strategies in which oncogene-expressing cells can be junctionally isolated in order to counteract neoplastic transformation.

In contrast to these results, there are significant clinical data implicating GJC as a carcinogenesis suppressor because of its ability to mediate growth control, and that disturbance of GJC between cells is a characteristic of several cancers (Yamasaki, 1990; Rose et al., 1993; Grossman et al., 1994; Hirschi et al., 1996; Yamasaki et al., 1999). However, some tumor cells have been shown to increase expression of functional connexin upon leaving their primary growth sites (Kamibayashi et al., 1995; Zhang et al., 1999). For example, melanoma cells have low levels of Cx26 expression when residing in the epidermal basal layer. However, upon acquiring sustained growth capability, they detach from the epidermis and up-regulate Cx26 expression, which is thought to help them couple with endothelial cells and infiltrate secondary sites (Ito et al., 2000). Overall, the studies presented here highlight scenarios in which enhanced GJC is favorable to tumors and detrimental to the host.

The results presented here revealed that blocking GJC suppressed tumor formation, while promoting GJC enhanced tumor formation. Even more remarkable was the spatial range over which tumorigenesis could be affected: modulation of GJC at the maximal possible distance from oncogene expressing cells had the most impact on tumor incidence, suggesting a long-range morphogenetic signal that controls tumor growth. In order to formulate a model that quantitatively predicts and explains this puzzling dataset, we undertook a mechanistic/mathematical approach featuring a reasonable set of assumptions.

We propose that H7 disrupts the coordination of polarization-depolarization cycles on the affected side and makes the voltage swings on the affected side bigger; a bigger detected swing suppresses tumor formation on the contralateral side, possibly by affecting the activity of Vmem dependent channels. Enhancing GJC with Cx26, in contrast, may disrupt the stability of the boundary separating polarized and depolarized domains. In other words, altering local cell-cell synchrony with GJC-altering reagents may prevent cells from responding in a coordinated way to changes in the polarization states of both their near and distant neighbors, and hence disrupt the oscillating “breathing” pattern required for normal control of cell proliferation. This can be attributed to the disruption of multi-cellular physiological networks established by electrical coupling of gap-junctionally connected cells. One of the interesting aspects of this kind of dynamics is that it is eminently suitable for integration of complex decision-making as would be needed for developmental patterning; for example, the ability of plasmodium organisms to optimize food gathering strategies is currently thought to be implemented by information processing that arises from the integration of oscillatory patterns (Tsuda et al., 2009).

While we do not at this time know the molecular nature of the two signals that implement these dynamics, butyrate and serotonin are good candidates, given previous data on the role of these small signaling molecules in long-range bioelectrical events that regulate cancer (Blackiston et al., 2011; Lobikin et al., 2012; Chernet and Levin, 2013b, 2014). We propose that gap junctions function in this process by regulating ion flow that synchronizes distant cells, enabling global oscillation—the periodic change in the overall electric field within the embryo that provides long-range information exchange. Our model is able to quantitatively account for this surprising dataset and to make new predictions that were verified by experiment. Subsequent work will experimentally test additional predictions of the model. In addition to its implications for control of tumorigenesis, the first phase of the model (establishment of consistent LR gradient from stochastic and highly variable initial Vmem cell states) has significant implications for the problem of left-right patterning. If correct, it suggests that the focus of new experiments and theory should be not on stable voltage differences arising directly and immediately from early ion channel localization at 4-cell stage, but on understanding the output of a biased dynamical system operating at 128-cell to blastula stages. It is also significant that our model makes experimentally-verified predictions which functionally link left-right asymmetry and cancer. While connections between these two disparate-seeming fields have been suggested previously (McManus, 1992; Sandson et al., 1992; Wan et al., 2011; Wilting and Hagedorn, 2011; Sauer and Klar, 2012; Veltmaat et al., 2013), this is the first quantitative model of the molecular mechanisms by which these different aspects of pattern regulation are unified.

Taken together, our data suggest a complex cross-talk between distant regions of the body that impinge on the stochastic nature of tumorigenesis and physiological dynamics. Additional parameters related to gap junction density include properties such as tissue stiffness, which is known to influence cancer cell behavior and alter cell responses to genetic and epigenetic signals (Bizzarri and Cucina, 2014; Pisanu et al., 2014). Future work will integrate these factors into ever more comprehensive models of tumor formation. While the frog embryo is not considered large by the standards of human medicine (long-range signaling occurs on a scale of from 1.3 to at least 4 mm), it represents about 300 cell diameters (a unit of cell diameter = 13.3 μm)—distinctly non-cell-autonomous (Figure 8). Subsequent work will mechanistically test other predictions of our model, by tracking real-time bioelectrical communication through GJs during tumor induction (a technically very challenging task at the edge of current capability). Our data suggest specific tests of this type of signaling in a mammalian model, and imply that biomedical cancer strategies must consider not only the events of the tumor itself and its microenvironment, but perhaps also signals moving to and from quite remote tissues in the body.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.