Table 1.
Accession numbers of all human genes from the model.
Figure 1.
Definition of the TERT transcriptional neighbourhood in A2780 cells by transfection screening.
(A), overexpression of TERT activators. A2780 cells were transfected with the luciferase reporters shown on the vertical axis. Each reporter was co-transfected alongside vector control or transcription factor expression plasmid shown in the right hand boxes. Each bar type represents a different expression vector relative to control. 48 h post-transfection, promoter activities were analysed by luciferase assay. (B), overexpression of TERT repressors, transfected as in (A). (C), overexpression of E2F1 against the promoter panel, transfected as above. Because of the very strong self-regulatory effect on its own promoter, E2F1 is shown on a different scale and separately from the other TERT repressors. Mean ± SEM of 3 experiments (ns: not significant; *: p<0.5; **: p<0.01).
Table 2.
Common, concordant interactions in literature and data-derived models.
Table 3.
Common, non-concordant interactions in literature and data-derived models.
Table 4.
Candidate novel interactions included from the screen.
Figure 2.
Topology, steady states, and statespace structure of the TERT transcriptional neighbourhood model.
(A), topology and steady states of the basal TERT model. Transfection screening data were used to assign activating or repressive network interactions according to the direction of regulation of each promoter and using the cut-offs of minimum fold-change 1.5 up- or down-regulation of promoter activity and p-value (ANOVA)<0.01. Topology of the final model was visualised in Cytoscape [105]. Arrows indicate activation, T-shape indicates repression. Left and right panels show steady states 1 and 2, respectively. Red colour indicates the node is on, green colour indicates the node is off in each steady state. (B), core statespace structure of the model. Statespace was calculated by brute force and visualised in Pajek [103]. Basins of attraction were extracted as weak components of the statespace. To visualise the core structure, all nodes with in-degree ≥1 were extracted as new networks from each weak component and visualised with transient states in blue and attractor states in yellow. Left panel corresponds to state 1, right panel corresponds to state 2.
Figure 3.
Modelling inhibitor effects on the TERT transcriptional neighbourhood.
A2780 cells were transfected with each luciferase reporter shown and 32(A), 5 µM SU6656, (B), 10 µM FR180204. Left panels show mean ± SEM of 3 experiments (ns: not significant; *: p<0.5; **: p<0.01). Central panels: luciferase assay results meeting model cut-off of FC>1.5, p<0.01 were modelled as rule table modifications. Heat-map representation of new model steady states obtained by setting rule tables for constitutive activation or suppression at those nodes significantly affected in the luciferase assay. Red colour indicates the node is on, green colour indicates the node is off. Right panels: analysis of TERT expression after repeat inhibitor treatments. Control and treated samples from treatment time points shown were analysed by RT-QPCR for TERT expression normalised to RPS15. Mean ± SEM of TERT expression in treated cells relative to control from three experiments (ns: not significant; **: p<0.01).
Table 5.
IUPAC names and CAS numbers of the compounds used in the study.
Figure 4.
Simulated effects of GSK3 inhibition and network noise on TERT transcription.
(A), Effect of BIO on the transcription factor promoter panel and simulation in the model. Top panel: A2780 cells were transfected with each luciferase reporter. 32 h later transfectants were treated for 16 h prior to luciferase assay with DMSO or 5 µM BIO. Luciferase assay results meeting model cut-off of FC>1.5, p<0.01 (Fos and STAT3) were modelled as rule table modifications. Lower panel: heat-map representation of new steady states obtained by setting Fos to be constitutively suppressed and STAT3 to be constitutively active. Red colour indicates the node is on, green colour indicates the node is off. (B), noise simulation by the bit-flip method under basal or BIO modified rules in the model. State coherence [73] of each attractor under the model basal rule-set (top) or BIO simulation (bottom) was evaluated as described in the text. Heat-maps of both attractors under either rule-set are shown, with colouration as above. Horizontal arrows between attractor states indicate that a transient state change of the adjacent node caused a shift to the alternate steady state. Vertical and diagonal arrows indicate the state changes resulting from the rule-set change (basal → BIO, or the reverse).
Figure 5.
Topological analysis of the TERT model, and prediction of robust MYC dependent TERT repression.
(A), structure of FFL types I–IV. Structures visualised in Pajek [103]. Bold lines indicate activation, dashed lines indicate repression. X, Y represent generalised transcription factors, Z represents a regulated gene. (B), activation and repression modules in the TERT transcriptional neighbourhood model. Subnetworks were extracted from the main model and visualised in Pajek [103]. Extraction was achieved as described in materials and methods. As an indicator of topological importance, node betweenness centralities were calculated and are given in table 6. Additionally, we calculated flow betweenness which is not dependent only on geodesics [77]. (C), Effect of single- and double-node targeting on TERT on-states. Rule-sets for each node were modified in turn individually (black bars) to simulate constitutive repression or activation. For each rule-set change, statespace was derived and the proportion of system states evolving to attractor states with TERT stably on was quantified. The analysis was repeated for each node in the context of double knockouts with MYC also suppressed in each case (grey bars). (D), MYC dependent TERT repression and reversal by AR. A2780 were transfected with 200 nM non-specific control siRNA (Con), 100 nM MYC with 100 nM non-specific (MYC), or 100 nM MYC and 100 nM each specific siRNA. Cells were harvested after 48 h and RNA extracted for analysis of TERT expression normalised to RPS15 by RT-QPCR. Mean ± SEM of TERT expression in treated cells relative to control from three experiments (ns: not significant; *: p<0.05; **: p<0.01). (E), Knockdown of TERT regulatory transcription factors by RNAi. A2780 were transfected with 100 nM each specific siRNA (RNAi) or non-specific control (NS) and harvested after 48 h. 20 µg protein samples were analysed by western blotting against the respective targets. ERK counter-blots were also performed. Each experiment was performed twice. Representative blots are shown.
Table 6.
Node betweenness centrality values for the activation and repression modules.
Figure 6.
Simulation of Ets family transcription factor gain of function at the TERT promoter.
(A), overexpression of ETS2 against the promoter panel. A2780 cells were transfected with the luciferase reporters shown. Each reporter was co-transfected alongside vector control or pCMV-ETS2. 48 h post-transfection, promoter activities were analysed by luciferase assay. Mean ± SEM of three experiments (ns: not significant; *: p<0.05; **: p<0.01). (B), regulation of the ETS2 promoter by the transcription factor panel. A2780 cells were co-transfected with ETS2-luciferase reporter alongside vector control or expression vectors shown. 48 h post-transfection, promoter activities were analysed by luciferase assay. Mean ± SEM of three experiments (ns: not significant; *: p<0.05; **: p<0.01). (C), effect of ETS2 expression on the TERT promoter under MYC inhibition. A2780 cells were co-transfected with the TERT-luciferase reporter and with pCMV control or pCMV-ETS2 with non-targeting or MYC-specific siRNA. 48 h post-transfection, promoter activities were analysed by luciferase assay. Mean ± SEM of three experiments (D), interactions in the ETS2 subnetwork added into the model with cutoffs FC>1.5, p<0.01 from the transfection data. The subnetwork was visualised in Cytoscape [105]. Arrows indicate activation, T-shape indicates repression. (E), Effect of single- and double-node targeting on TERT on-states in the ETS2 modified model. Rule-sets for each node were modified in turn individually (black bars) to simulate constitutive repression or activation. For each rule-set change, statespace was derived and the proportion of system states evolving to attractor states with TERT stably on was quantified. The analysis was repeated in the background of the MYC suppressed rule-set (grey bars).
Figure 7.
Topological control of TERT on-state multiplicity in the model.
(A), influence of activation module dominance on TERT on-state multiplicity. Topology of the model was altered by a series of 600 random attacks deleting activation and repression module interactions with increasing probability. Direct interactions with TERT were left unaltered in all attacks. The remaining sub-networks were extracted from each model variant as described in materials and methods and the number of edges in each were counted to determine the edge ratio AM/RM. The statespace of each model was calculated and the number of stable on states present for the TERT node was quantified and plotted against the calculated AM/RM edge ratio for each variant network. Significance of edge ratio population differences was tested in Matlab by Wilcoxon rank-sum test (**: p<0.01). (B), influence of AM dominance in random networks. A series of 300 (15 node) networks was generated with semi-random edge seeding and increasing edge density. All networks were constrained to have one regulated node which was connected downstream of all others. The number of activators and repressors of the node was allowed to vary randomly. Statespace and AM/RM edge ratios were calculated for each network and compared as in (A), calculating number of stable on-states for the fully connected node. Significance of edge ratio population differences was tested in Matlab by Wilcoxon rank-sum test (**: p<0.01).