Endogenous molecular network reveals two mechanisms of heterogeneity within gastric cancer.

Intratumor heterogeneity is a common phenomenon and impedes cancer therapy and research. Gastric cancer (GC) cells have generally been classified into two heterogeneous cellular phenotypes, the gastric and intestinal types, yet the mechanisms of maintaining two phenotypes and controlling phenotypic transition are largely unknown. A qualitative systematic framework, the endogenous molecular network hypothesis, has recently been proposed to understand cancer genesis and progression. Here, a minimal network corresponding to such framework was found for GC and was quantified via a stochastic nonlinear dynamical system. We then further extended the framework to address the important question of intratumor heterogeneity quantitatively. The working network characterized main known features of normal gastric epithelial and GC cell phenotypes. Our results demonstrated that four positive feedback loops in the network are critical for GC cell phenotypes. Moreover, two mechanisms that contribute to GC cell heterogeneity were identified: particular positive feedback loops are responsible for the maintenance of intestinal and gastric phenotypes; GC cell progression routes that were revealed by the dynamical behaviors of individual key components are heterogeneous. In this work, we constructed an endogenous molecular network of GC that can be expanded in the future and would broaden the known mechanisms of intratumor heterogeneity.


Cell death module
Cell death plays an essential role during the animal development and tissue homeostasis [25]. Apoptosis, an important form of programmed cell death, mediates physiological gastric epithelial cell loss in the normal gastric mucosa and chronic gastritis [26,27]. Apoptosis is tightly controlled by either extracellular or intracellular signals. The center executor of apoptosis is caspases, of which activation is under dual control by both apoptosis activators and inhibitors such as Bcl-2, Bcl-XL, BAD, BAK-BAX, Bid and XIAP. The relatives of Bcl-2, such as Bad, Bax, Bak, and Bid, strive to pry open channels in the outer membrane of the mitochondria and thereby release cytochrome c into the surrounding cytosol [28,29]. Yet Bcl-2 and Bcl-XL work oppositely to keep channels closed [30]. Once present in the cytosol, cytochrome c molecules associate with the Apaf-1 protein and form apoptosome and then proceeds to activate procaspase 9. The active caspase 9 then proceeds to activate procaspase 3, which executes apoptotic process [31][32][33][34][35][36]. Except the intrinsic apoptotic pathway, apoptosis can also be triggered through activated pro-apoptotic cell surface receptors, such as Fas and TNFR. These receptors act via the FASassociated death domain (FADD) protein to assemble a death-inducing signaling complex (DISC), and the latter proceeds to activate caspases 8 and 10 [28,37]. These then converge on the intrinsic apoptotic cascade by activating the caspases 3 [31][32][33][34][35], the latter cleaves and activates Bid [28], which initiates opening of the outer mitochondrial membrane channel, further activating the apoptotic cascade. Studies had revealed that a variety of stresses could cause a rapid increase in p53 levels. After p53 concentrations increase, the p53 bind to the promoters of a large of target genes and induce their transcription. Among the induced proteins are p21 [18,21], Bax [38,39], Fas [38,39], Bcl-2 [38,39] and IGF1-R [40,41]. The survival signaling pathways through activation of the PI3kinase (PI3K)-Akt/PKB kinase pathway leads to Mdm2 phosphorylation and to the resulting p53 activation [42,43]. The XIAPs bind to and inhibit caspases [44,45]. XIAPs are bound and inhibited by a protein named SMAC/ DIABLO, which is released from mitochondria along with cytochrome c during apoptosis [44]. In response to certain physiologic signals, NF-κB activates a large constituency of target genes involved in apoptosis; included among these are p53 [46], XIAP [46,47], Bcl-2 [48] and Bcl-X L [46,47].

Angiogenesis
Angiogenesis is the physiological process beneficial for normal tissue growth and regeneration. However, it also represents a malignant transformation of tumor progression [70]. Under conditions of hypoxia, the HIF transcription factor accumulate, which then drives the expression of a number of genes whose products encourage angiogenesis. Prominent among these is VEGF [71]. VEGF/VEGFR could also be activated by IL-8 [72], IL-1 [73], HGF [74], Integrin [75] and PI3K/ Akt signaling [76].

Metabolism
During cell growth and division, adjustment of energy metabolism is observed in neoplastic diseases. Cancer cells carry out increased aerobic glycolysis, a phenomenon known as the Warburg effect [77]. Some proteins, such as Myc, HIF, p53, AKT, Ras and β-catenin, are intimately linked to metabolic pathways through transcriptional or post-transcriptional regulation of metabolic enzymes [78].

Cell adhesion
Cell-to-cell interactions and their underlying extracellular matrix (ECM) present critical signals that regulate and maintain gastric epithelial differentiation, Those interaction mediator including E-cadherin and Integrins [79,80]. In addition, cancer cells that participated in invasion and metastasis developed alterations in their attachment to other cells and to the extracellular matrix (ECM). The alteration best characterized by losing of E-cadherin, a key cell-to-cell adhesion molecule [81]. Those regulators include TGF-β [79] and a set of transcriptional factors like Zeb1/2 [82].

Gastric differentiation
The gastric epithelial associated factors that participate in gastric epithelium development and differentiation have been indentified as growth factors and transcription factors, such as Sox2, sonic hedgehog (Shh), Indian hedgehog (Ihh), gastrin, FGF10, Gata4, Gata6, BMP4, Runx3, Bapx1, Foxa1/2 and HoxA5. Recent data indicate that the most critical reprogramming factor, Cdx1/2, which inhibit expression of differentiated genes in gastric epithelium, is sufficient to direct the reprogramming of gastric epithelium cells into intestinal cells. On the other hand, the direct reprogramming factor of intestinal cells into gastric-liked epithelial cells has been demonstrated in several models, such as Sox2 and Shh. Therefore, the gastric cells state and intestinal cell state should be maintained by a set of master regulatory factors respectively.
Cdx2 is believed to be critical for maintaining intestinal epithelial cell phenotype [83,84] and there have been several reports of it expression in intestinal metaplasia and intestinal type gastric carcinomas [85,86]. Gastric expression of Cdx2 alone was sufficient to induce intestinal metaplasia in mice [87][88][89]. Beside, increased β-catenin expressions were significantly more frequent in intestinal-type gastric cancer compared with diffuse-type gastric carcinomas [90]. On the other hand, since, in adults, Sox2 expression is found in stomach and absent from the intestine [91]. Moreover, Sox2 participates in the development of foregut-derived organs, such as esophagus and stomach [92,93], and are involved in regulation of stomach specific genes, pepsinogen and Muc5ac [94,95]. Ectopic Sox2 expression is sufficient to redirect development fate of intestinal epithelium towards a premature gastric phenotype [96]. In addition, the loss of Sox2 expression and aberrant Cdx2 expression occur in intestinal metaplasia.

Growth factors
Various growth factors and hormones are expressed by mesenchymal cells but act on the epithelial compartment: for example, hepatocyte growth factor (HGF) is known to influence proliferation and differentiation of intestinal epithelial cells. The presence of EGF, HGF, IGF and KGF receptors in all glandular compartments clearly suggests the potential role of these growth factors either in the development or in the maintenance of specific gastric epithelial functions [3]. When the gastric mucosa is injured by inflammation or other insults, expression of growth factors and their receptors, such as EGF, IGF, EGFR family, and VEGFR subtypes, are upregulated in epithelial or mesenchymal cells, and subsequently mucosal repair is promoted [97]. Ras can be activated by a number of signaling cascades radiating from growth factor receptors, such as EGF [98], HGF [99], VEGF [100] and Integrins [42,43,101,102]. Ras activates Erk1 and Erk2 (MAPKs) via the Raf kinase [103]. The latter can then phosphorylate kinases in the cytoplasm that regulate transcription as well as transcription factors (Ets, Elk-1, SAP-1), which can then proceed to stimulate the expression of growth-regulating genes, such as Cyclin D [10], Myc [104], p21 [105] and HGF [106]. Ras protein is also capable to phosphorylate and activate phosphatidylinositol 3-kinase (PI3K) and AKT/PKB pathway. Once activated, Akt/PKB proceeds to phosphorylate and activate/inhibite a series of protein substrates, such as Bad [107], caspase-9 [108,109], p21 [110] and GSK3β [111]. The PI3K-AKT pathway mainly aid cell survival and stimulate cell proliferation. The activation of PI3K and Akt/PKB is negatively control by PTEN [112].

DIFFERENTIAL EQUATIONS TO QUANTIFY THE WORKING ENDOGENOUS NETWORK
We translated the endogenous molecular network for gastric epithelium Figure 1 into mathematical format by employing sigmoidal functions. The quantitative descriptions are a set of coupled ordinary differential functions as follow. The meanings of parameters are in Methods. The computational codes are available upon request.
x(39) = Fas Fas ligand or CD95L, and its receptors FasR and DcR3.    a  x  n  a  x  n  a  x  n  a  x  n   a  x  n  a  x  n  a  x  n  a  x  GSK3β * [12] * The molecular mechanism of inhibition and activation interactions in the network. * Components Phosphorylate or dephoephorylate targets, which result in their activation or inactivation; § Components cleave tagets, which result in their activation or inactivation; ‡ Components transcript target genes, and induce expression of targets; † Components bind to targets, which inhibit or activate the activity of targets, or sequester them in the cytoplasm or nuclear; ‖ Components activate or inhibit targets indirectly, intermediates were abbreviated; ※ Components activate or inhibit targets through mechanisms still unknown.  Figure S1.