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

In addition to the immediate microenvironment, long-range signaling may be an important component of cancer. Molecular-genetic analyses have implicated gap junctions—key mediators of cell-cell communication—in carcinogenesis. We recently showed that the resting voltage potential of distant cell groups is a key determinant of metastatic transformation and tumor induction. Here, we show in the Xenopus laevis model that gap junctional communication (GJC) is a modulator of the long-range bioelectric signaling that regulates tumor formation. Genetic disruption of GJC taking place within tumors, within remote host tissues, or between the host and tumors significantly lowers the incidence of tumors induced by KRAS mutations. The most pronounced suppression of tumor incidence was observed upon GJC disruption taking place farther away from oncogene-expressing cells, revealing a role for GJC in distant cells in the control of tumor growth. In contrast, enhanced GJC communication through the overexpression of wild-type connexin Cx26 increased tumor incidence. Our data confirm a role for GJC in tumorigenesis, and reveal that this effect is non-local. Based on these results and on published data on movement of ions through GJs, we present a quantitative model linking the GJC coupling and bioelectrical state of cells to the ability of oncogenes to initiate tumorigenesis. When integrated with data on endogenous bioelectric signaling during left-right patterning, the model predicts differential tumor incidence outcomes depending on the spatial configurations of gap junction paths relative to tumor location and major anatomical body axes. Testing these predictions, we found that the strongest influence of GJ modulation on tumor suppression by hyperpolarization occurred along the embryonic left-right axis.

Together, these data reveal new, long-range aspects of cancer control by the host's physiological parameters.

Consistent pattern formation during embryogenesis and organ regeneration requires the coordination of cell activities and information across large distances in vivo. Alongside gradients of extracellular secreted molecules, functions an important system of direct cell to cell communication. Gap junctions are intercellular channels that allow small molecule-mediated signals to pass directly from the cytoplasm of one cell to its neighbors (Goodenough et al., 1996; Bruzzone et al., 1996a,b). Connexins, subunits that make up gap junctions (GJs), control the transfer of a wide range of cellular molecules based on their charge, size, and shape (Loewenstein, 1979, 1981; Goldberg et al., 2004), serving as true electrical synapses among non-neural cells (Bennett, 1997; Pereda et al., 2013; Rash et al., 2013). Due to their rich combinatorial properties and gating regulation, gap junction function can implement complex circuits with properties such as memory (Palacios-Prado and Bukauskas, 2009; Pereda et al., 2013). Many biological processes have evolved to take advantage of these versatile elements, using GJs to regulate differentiation, proliferation, and apoptosis (Krysko et al., 2005; Wong et al., 2008), as well as morphogenetic signaling in many contexts, including development of the brain, limb, bone, tooth, heart, and the left-right axis (Levin, 2007; Wong et al., 2008). Not surprisingly, loss of morphogenetic control due to defective gap junctional communication (GJC) has been implicated in numerous disorders, including hearing impairment (Rabionet et al., 2000), neural tube defects (Ewart et al., 1997), parasitic infection (Vega et al., 2013), cardiovascular diseases (Jongsma and Wilders, 2000), and other genetic syndromes (Zoidl and Dermietzel, 2010).

Interestingly, while GJC is most often thought of as mediating local interaction, it can also provide long-range morphogenetic coordination on the scale of the whole organism, for example during the establishment of polarity in regenerating planaria (Oviedo et al., 2010) and left-right pattering of Xenopus and chick (Levin and Mercola, 1998, 1999).

A major area in which GJC has been implicated is tumorigenesis (Yamasaki et al., 1995, 1999; Ruch and Trosko, 2001; Trosko, 2005; King and Bertram, 2005; Mesnil et al., 2005). A role for GJs, as mediators of cell-cell signaling and information exchange, is consistent with a view of cancer as a developmental disorder—a derangement of the interaction of cells with the normally tight field of patterning controls of the body (Tsonis, 1987; Pierce and Speers, 1988; Clark, 1995; Dean, 1998; Rubin, 2006; Bissell and Hines, 2011; Marongiu et al., 2012). Alongside cell-autonomous mutations in so-called cancer stem cells, tumorigenesis and progression are controlled by biophysical properties of the surrounding microenvironment (Chernet and Levin, 2013a) and even by neural inputs (Scharrer, 1953; Pawlowski and Weddell, 1967; Magnon et al., 2013). The molecular nature of the processes by which surrounding cells regulate transformation and metastasis is an important area of research today.

By regulating the spread of morphogenetic signals, GJs are an ideal candidate for keeping individual cell activities coordinated toward the anatomical needs of the host, or conversely, allowing confounding signals that may induce tumorigenesis (Levin, 2011, 2012b). Disruption in the function of GJs is implicated in a number of cancers (Yamasaki et al., 1995; Duflot-Dancer et al., 1997; Yamasaki et al., 1999; Ruch and Trosko, 2001; Mesnil et al., 2005; Sirnes et al., 2012). For example, Cx26-deficient mice exhibit a 25-fold increased incidence of spontaneous liver tumors (Temme et al., 1997). Moreover, tumor incidence is higher, and clinical prognosis is worse, when cells are gap-junctionally isolated by pharmacological agents or genetic mutation (Loewenstein and Kanno, 1966; Loewenstein, 1979; Rose et al., 1993; Mesnil et al., 2005).

Most clinically-relevant tumor cells are known to exhibit down-regulation in connexin expression, leading to the disruption of effective cell:cell communication (Soroceanu et al., 2001; Gee et al., 2003; Mesnil et al., 2005; Talbot et al., 2013). However, in some some studies, enhanced GJC was suggested as a tumor promoting factor (Saito-Katsuragi et al., 2007; Naoi et al., 2007; Elzarrad et al., 2008; Haass et al., 2010). Breast cancer and melanoma cells take advantage of these connexins to enhance their metastatic potential in Stoletov et al. (2013). Together, the data suggest that it is imperative to understand the signaling mediated by GJs and the information passed among normal cells that could promote (Rose and Wallingford, 1948; Lewalle et al., 1998; Zhang et al., 2003; Donahue et al., 2003), or normalize (Hendrix et al., 2007), cancer in vivo. How might GJC regulate tumorigenesis in the context of tissue and organ patterning?

One carrier of the morphogenetic cues that go awry during tumorigenesis may be current—the movement of charged ions. Gradients of resting potentials established by ion channels and pumps in the cell membrane are now known to be instructive patterning cues that regulate cell behavior during pattern formation (Levin, 2012a; Tseng and Levin, 2013; Levin, 2013, 2014), and gap junctions sculpt the distribution of iso-potential cell fields by allowing specific cells to equalize their trans-membrane voltages (Vmem) in response to various physiological signals. Bioelectric gradients have already been implicated in the control of metastasis (Morokuma et al., 2008; Blackiston et al., 2011) and oncogene-mediated tumorigenesis (Lobikin et al., 2012; Chernet and Levin, 2013a,b, 2014), while specific ion channels are becoming increasingly recognized as oncogenes and important drug targets (Diss et al., 2005; Fraser et al., 2005; Arcangeli et al., 2009; House et al., 2010; Yildirim et al., 2012; Arcangeli et al., 2012; Yang and Brackenbury, 2013; Than et al., 2014; Pardo and Stuhmer, 2014). Interestingly, the resting potentials of distant cells are critical for oncogene-dependent tumorigenesis: modulation of ion channels in locations quite distant to oncogene expressing cells in Xenopus tadpoles significantly reduces the incidence of tumors. This effect is mediated by a butyrate-based mechanism that regulates oncogene-mediated tumorigenesis via histone deacetylase activity (Chernet and Levin, 2013a,b, 2014), but the spatial dynamics of butyrate signaling in this context remain to be elucidated. The known role for ion flows and resting potentials in cancer suggests the possibility that gap junctions participate in bioelectric regulation during carcinogenesis and/or neoplastic progression.

To investigate the complex interplay of physiological and genetic signals in tumorigenesis, we pursued a combination of modeling and experiment to probe the spatial relationships between tumorigenesis, voltage properties, and GJ paths in vivo. We recently showed that ion channel-mediated changes in Vmem can modulate the tumorigenicity of human oncogenes misexpressed in the Xenopus laevis embryo (Chernet and Levin, 2013b, 2014). Here, we use this assay to investigate the interplay between bioelectric controls and gap-junctional connectivity in vivo. Unexpectedly, we found that disruption of cell:cell communication via H7—a chimeric connexin construct that is known to inhibit GJC in Xenopus (Paul et al., 1995; Levin and Mercola, 1998)—is able to suppress tumor formation. Remarkably, the suppression effect was observed regardless of where GJC disruption was occurring (host-wide, within tumors only, or away from tumors). Indeed, the most pronounced suppression was recorded for GJC disruption taking place non-locally to oncogene-expressing cells, revealing a role for distant cell:cell communication in tumorigenesis. Conversely, enhanced GJC within tumors or their microenvironment, achieved via the mis-expression of a constitutively permeable junction-forming connexin Cx26 (Levin and Mercola, 1998), increased tumor incidence. Together, these data suggest that GJC is a mediator of both local (within tumors) and long-range (within the microenvironment and the host) signaling and that specific patterns of physiological isolation may be necessary for tumor suppression. Here we also formulate a mechanistic, quantitative model consistent with these data and with prior work on endogenous left-right asymmetric voltage gradients, and test key predictions of this model.

Based on these data, we suggest that exploiting the bioelectrical signaling that occurs through electrical synapses among somatic cells represents an important target for cancer therapy that is distinct from the targeting of individual channels.

X. laevis eggs were fertilized in vitro, and embryos were cultured according to standard protocols (Sive et al., 2000), in 0.1X Modified Marc's Ringers (MMR; pH 7.8) with 0.1% Gentamicin. Xenopus embryos were housed at 14–18°C and staged according to Nieuwkoop and Faber (1967). All experimental procedures involving the use of animals for experimental purposes were approved by the Institutional Animal Care and Use Committees (IACUC) and Tufts University Department of Lab Animal Medicine (DLAM) under the protocol number M2014-79.

Fertilized Xenopus embryos were transferred into mesh-bottomed dishes with 3% Ficoll and injected with capped, synthetic mRNAs (made using the Ambion Message Machine kit) dissolved in water at the stages indicated. The doses per cell were KRASG12D (Le et al., 2007), 40 pg; H7 (Paul et al., 1995), 70 pg; Cx26 (Levin and Mercola, 1998), 500 pg; and β-gal (lineage tracer), 400 pg. Two hours after injection, embryos were transferred into 0.75X MMR for 45 min before they were washed and cultured in 0.1X MMR until desired stage was reached. KRASG12D-injected embryos were raised to stage ~35, and scored for the presence of tumors using bright field microscopy as described in Chernet and Levin (2013b).

GJC was assessed between tumors and the host using a 1:1 mixture of Rhodamine-Lysinated Dextran (RLD, 10 kDa, Life Technologies) and Lucifer Yellow (LY, 0.522 kDa, Life Technologies) (Figure 1). 16 cell embryos injected with KRASG12D, RLD, and LY were allowed to grow to stage ~35. Cells exhibiting LY signal (which passes through GJs) in the absence of RLD (which does not pass through GJs) signal reveal an open gap-junctional communication (cells with both LY and RLD signal indicate regions that have acquired the two molecules through cell division and/or migration). RLD and LY were detected in live embryos using TRITC and Lucifer Yellow filtersets, respectively, on an Olympus BX61 spinning-disk confocal microscope with Hamamatsu ORCA digital CCD camera.

Data were expressed as the mean unless otherwise noted. The differences between treatment groups were analyzed using Student's t-test and X2 test, and the null hypothesis was rejected at the 0.05 level.

A quantitative model of embryonic growth and response to microinjection treatments was developed in two stages that model different processes (Figures 4, 5). The first-stage “left-right synchronization” model was implemented using JavaScript and the HTML5 “canvas” function. Details of this model are described below (Section A two-stage quantitative model describes the dynamics between Vmem, GJC, and tumor formation); the model is interactive, and can be manipulated and the source code can be examined at http://chrisfieldsresearch.com/convergence-demo.htm (also seen in Supplement 1). The second-stage “left-right communication” model comprises the implications of a set of assumptions that specify a control network as described below (Section A two-stage quantitative model accurately predicts the dynamics between Vmem, GJC, and tumor formation). Numerical predictions were calculated manually from these assumptions.

To study the role of GJC in oncogene-mediated tumorigenesis, we made use of a simple assay that utilizes expression of an exogenous human oncogene in X. laevis embryos (Figure 1A). This model system is ideal for probing connexin-based GJC (Swenson et al., 1989; Barrio et al., 1997; Cao et al., 1998; Lee et al., 2009) and has been used to study GJC-mediated morphogenesis and large-scale patterning (Warner, 1992; Levin and Mercola, 1998; Levin, 2002). The oncogene KRASG12D (Le et al., 2007) was used to induce Xenopus tumors (Figures 1B–D, green arrowhead). These exhibit the same major hallmarks of tumors as do their mammalian counterparts: increased mitotic activity, induced vasculogenesis, disorganization of normal cellular architecture, increased hypoxia, acidic microenvironment, and ability to illicit innate immune response (Chernet and Levin, 2013b, 2014). We began by testing GJC between induced tumors and the host, which revealed considerable spreading of the small GJC-permeant tracer Lucifer Yellow (Figures 1B,D; yellow signals, white arrowheads) away from descendants of one of the 16 progeny cells that were injected with oncogene mRNA and a large, GJC-impermeant Rhodamine Dextran (5 out of 5 tested; Figures 1C,D; red signal). Having observed active GJC between the tumor and surrounding host tissue, we began functional studies by investigating how the growth of these oncogene-induced tumors is affected by spatially targeted disruption of GJC. All of the reagents used below were titered to levels low enough that the normal background processes of embryogenesis and growth proceeded normally.

To determine the effect of GJC disruption on tumorigenesis in vivo, mRNAs of dominant-negative connexin H7 (Paul et al., 1995; Levin and Mercola, 1998) and KRASG12D were injected in X. laevis embryos. The oncogene and GJC disrupting constructs were introduced in different spatial arrangements to investigate what happens when GJC is disrupted host-wide, within tumors, and away from tumors. Taking advantage of the embryonic fate-map (Dale and Slack, 1987; Moody and Kline, 1990), we were able to control which cells received oncogene mRNA and which cells were inhibited with respect to GJC. To achieve suppression of GJC within the tumor (“local” mode), H7 mRNA was injected into 1 cell of a 2-cell embryo and the KRASG12D mRNA was introduced into one of the progeny of that cell (at the 16-cell stage). For remote (“long range” mode) GJC suppression in tissue outside the tumor, H7 mRNA was injected into 1 cell of a 2-cell embryo, which was raised to the 16-cell stage and then injected with KRASG12D mRNA into one of the cells on the opposite side of the embryo from the cells bearing the H7. For host-wide GJC disruption, both of the 2 cells of the 2-cell embryo were injected with H7, and KRASG12D mRNA was then injected in 1 of the 16 progeny cells. Figure 2 shows the various combinations of which sides of the embryo received the GJ blocker (H7) and which received the oncogene (KRASG12D).

Compared to tumor incidences observed in the baseline KRASG12D-only injected embryos (Figure 2; treatment mode A), local (treatment mode B) and host-wide (treatment mode D) disruption of GJC showed significant suppression of tumor formation by 6.6 and 5.8%, respectively (t-test; p < 0.05). Interestingly, long-range disruption of GJC (treatment mode C) had the strongest suppressive effect, showing a 15.8% decrease in the number of embryos with tumor (t-test; p < 0.001). Injection of H7 host-wide with KRASG12D injected in 2 cells on opposite sides of the 16-cell embryo (Figure 2; treatment mode F) did not affect tumor incidence when compared to KRASG12D-only injected embryos (Figure 2; treatment mode A). However, the host-wide H7 introduction (Figure 2; treatment mode F) reduced tumor incidence resulting from oncogene expression on both sides of the embryo (compare to Figure 2; treatment mode G) by more than 11% (t-test; p < 0.05). Similarly, tumor incidence from doubling the oncogene level can be reduced by the introduction of H7 to only one side of the embryo (treatment mode E). Together, the results imply that GJC disruption suppresses KRASG12D-induced tumors: the most pronounced effect on tumor incidence is observed when H7 is introduced at a distance from the oncogene-bearing cells, unless H7 is also expressed within the oncogene-expressing cells, in which case suppression is not as effective.

To further test the hypothesis that GJC plays a role in oncogene-mediated tumorigenesis, we scored tumor incidence in embryos injected with Cx26, a short connexin protein which lacks most of the intracellular regulatory region, and thus facilitates GJC by forming constitutively permeable gap junctions (Mesnil et al., 1997; Levin and Mercola, 1998). Similar to the H7 (GJ blockade) experiment, mRNA injections of Cx26 and KRASG12D aimed at enhancing GJC host-wide, within tumors, and away from tumors were performed (Figure 3). Compared to KRASG12D-only injected embryos (Figure 3, treatment mode A), long-range and host-wide Cx26 treatments (treatment modes C, D, F) both showed an increase in number of embryos with tumor by 6.4 to 11.4%. This is in contrast to the H7 data where modulation of long-range GJC had the most impact on tumor incidence. Interestingly, enhanced GJC within tumors (treatment modes B and E) slightly (but not significantly) decreased tumor incidences when compared to treatment modes A and F, respectively. Together, these data indicate that facilitating GJC host-wide and in the tumor microenvironment enhances the tumorigenesis process. In contrast, enhanced GJC within oncogene-expressing cells had minimal effect on tumor formation.

The tumor-incidence results were challenging to explain for the following reason. While perturbing GJC in the same side as KRASG12D expression (Figure 2, treatment mode B) reduced tumor incidence, the effect became more pronounced when the GJC perturbation is on the opposite side of KRASG12D injection (treatment mode C); it is puzzling why having H7 on the opposite side from KRASG12D injection could make the KRASG12D expressing side much less likely to form a tumor. More strikingly, why would perturbing GJC on both sides of the embryo, while only introducing KRASG12D on one side (Figure 2, treatment mode D), have the same effect as perturbing GJC on the KRASG12D injected side only?

In order to mechanistically integrate the main components of our experiments (Vmem manipulated by ion channel injections, GJC manipulated by Cx injections, and long-range effects on tumors induced by oncogene injections), a two-stage model was developed to explain the tumor-incidence data shown in Figures 2, 3 and to make additional testable predictions. The first stage of this model (Figure 4) establishes regions of different Vmem within the embryo and provides the input required by the second stage (Figure 5), which shows the dynamics that regulate tumor outcome. The dynamics postulated by the second-stage model require that the left and right sides of the embryo exchange a signal; it is assumed that this alternating signal is implemented, or at least enabled, by coupled oscillations of Vmem on the two sides as described in more detail below.

Our two-stage model is based on the observation that large-scale bioelectric cues can provide patterning information during development (Nuccitelli, 2003a,b; McCaig et al., 2005, 2009; Levin, 2012a; Zhao et al., 2012). Let us call the left and right sides of the Xenopus embryo A and B, in no particular order. Assume that the two sides exchange a long-range signal that controls proper patterning and subsequent maintenance of order, and that both A and B also respond to cell-autonomous and local signals. We hypothesized that it is an interaction between long-range and short-range signals that produces the non-local effects that we see.

If the left and right sides of the embryo are to produce a coherent long-range bioelectric signal, their long-range signaling activity must be coordinated or synchronized. Cells are known, for example, to alternate between hyperpolarization and depolarization with cell cycle (Bregestovski et al., 1992; Arcangeli et al., 1995; Blackiston et al., 2009). If this alternation between hyperpolarization and depolarization was synchronized so that the left side was fully polarized when the right side was fully depolarized and vice-versa, it could produce a coherent long-range signal. The first, “left-right synchronization” stage of our model generates synchronized left and right cell populations from a random initial state. The model assumes a total starting population of 100 cells, corresponding roughly to embryonic cell division seven (27 = 128). Below this number of cells, nearest-neighbor interactions are insufficient to produce global order because too many neighborhoods overlap. The model also assumes an initial left-right bias in the polarization state of the cells [as has been observed experimentally to derive from consistently-asymmetric localization of ion channels on the left and right sides of the 2-cell embryo as it divides (Levin et al., 2002; Qiu et al., 2005; Adams et al., 2006; Morokuma et al., 2008)]. The left-right bias was treated as a free parameter and varied to optimize model behavior. The cells are assumed to occupy a planar array representing the embryonic animal cap epithelium. Each cell is assumed to communicate only with its four nearest neighbors. Each square of the 10 × 10 array (Figure 4) represents one cell at the 100-cell stage; as embryonic cells continue to divide, each square represents the daughter cells in the epithelial “patch” occupied by their progenitor at the 100-cell stage.

The left and right boundaries of the planar array are mathematically identified (joined) so that neighborhoods “wrap” around this boundary; the 10 × 10 planar array thus has the topology of a cylinder. Making the upper and lower boundaries join also, to form a spherical topology, was found to have no significant effect on the model's behavior.

The initial random state of the 10 × 10 array was generated by assigning each square of the array two independent integer values: a random “polarization” value between 0 and 255 and a random “depolarization” value between 0 and 255. These random values were then biased by adding a single bias value both to the random “polarization” values of the squares on the left side of the grid and to the random “depolarization” values of the squares on the right side of the grid. At the 100-cell stage, these distinct values represent the potentially-distinct polarization values of distinct cell-membrane patches within a single cell (Wallace, 2007; Adams and Levin, 2013); their sum represents the average Vmem of that cell. Following the 100-cell stage, these distinct values also represent the potentially-distinct polarization values of distinct daughter cells within an epithelial patch; their sum represents the average Vmem of that patch. The JavaScript Math.random function was used to generate random values. Cell behavior was modeled by a single local rule uniformly implemented by all squares of the array. On each cycle, each square computed the ratio between the total polarization and depolarization values of its four neighbors. If the ratio was greater than or equal to 1, the square multiplied its own polarization value by 1.2 and its own depolarization value by 0.8; if the ratio was less than 1, it did the opposite. These multipliers represent the local response of cells to the average bioelectric state of neighboring cells; cells surrounded by polarized cells become more polarized, while cells surrounded by depolarized cells become less polarized.

The values of 1.2 and 0.8 were chosen to produce convergence to a stable state within less than 50 cycles on most model runs; small variations in the values of these multipliers had no significant effect on the qualitative behavior of the model.

This first stage of the model represents the dynamics of bioelectric state change in the early embryo. Its prediction that nearest-neighbor interactions are sufficient to drive convergence to a stable bioelectric state is borne out by timelapse videos using voltage-sensitive fluorescent dyes (Adams and Levin, 2012) in Xenopus embryos, which reveal precisely this kind of fluctuation that then settles on a consistent difference in resting potential. Interestingly, the model shows that despite considerable stochastic variability in individual cells' states, such a system consistently converges to a stable state in which the embryo exhibits large regions of identical voltage. In this way, this model for the first time provides an explanation for the observed robustness of bioelectric patterns in development despite the known physiological noise (variability) in individual cells' states.

The model was run until all squares had converged to either fully polarized, defined as a “polarization” value of at least 230 and a “depolarization” value of at most 15, or fully depolarized, defined with the reverse criteria. Replicate model runs that differed only in their initial random states were evaluated on the basis of two criteria: whether the fraction of either polarized or depolarized cells was greater than 65%, and whether more than one region, accounting for “wraps” around the mathematically-identified left and right boundaries, in each state existed, i.e., whether there were “ectopic islands” of polarized or depolarized cells. Due to the model's cylindrical topology, regions of one color, representing one polarization, sometimes “wrap” around the mathematically-identified left and right boundaries, i.e., blue squares appear on the left border and/or red squares appear on the right border. A region was considered “ectopic” only if it comprised one or more squares completely surrounded by squares of the opposite color, or surrounded by squares of the opposite color on three sides if the remaining side formed part of the top or bottom border. The requirement that the number of solutions failing to satisfy either of these criteria is negligible (< 3%) in 100 consecutive runs yields a value of 27% for the initial, 1st-division bias in Vmem among the left and right side blastomeres. This prediction of the model is in accordance with data, since direct measurements of voltage states in early embryos, as well as protein and mRNA analyses show consistent and significant biases in the bioelectric states of 2- or 4-cell frog embryos.

Decreasing the percent adjustment in polarization and depolarization values on each cycle downward from 20% increases the convergence time for this model without qualitatively changing the results, as does running the model on a 20 × 20 grid (i.e., 400 squares total).

The existence of bioelectrically-synchronized cell populations on the left and right sides of the embryo that is predicted by the stage-1 model is used by the stage-2, “left-right communication” model to explain the observed results of combining KRASG12D with either H7 or Cx26 (Figure 2). Experimental data show that 35% of KRASG12D-injected embryos produced tumors; compare this to 44% tumor incidence in embryos injected with KRASG12D on both sides (Figure 2). The non-linearity of tumor incidence by itself is an indication of growth control as a result of cross-talk (i.e., a contralateral signal) between sides A and B. Our stage-2 model is based on two primary assumptions. First, we propose that the Vmem values of the cells on the left and right sides of the embryo oscillate, in synchrony, with cells on the left polarizing when cells on the right depolarize and vice-versa as discussed above. Normal, non-tumor cells respond to the long-range, synchronized, oscillatory bioelectric signal produced by the contralateral side of the embryo (i.e., to the contralateral signal), but not to the long-range signal that they themselves produce (i.e., the ipsilateral signal). We conceptualize this contralateral signal as a cell-division suppressor; the A and B sides of the embryo are, on this model, alternately signaling to the other side of the embryo to turn the rate of cell division down (Figure 5A), thus maintaining approximate equality of cell-division rates on the two sides. Increased (decreased) contralateral signal is assumed to decrease (increase) tumor formation probability on a % for % basis. Second, we propose that disrupting wild-type GJ with either H7 or Cx26 alters both production of and response to the long-range signal.