Construction of an inter-organ signalling network model for understanding type 2 diabetes

Abstract

Type 2 diabetes mellitus (T2DM) is believed to be irreversible although no component of the pathophysiology is irreversible. We show here with a network model that the apparent irreversibility is contributed by the structure of the network of inter-organ signalling. A network model comprising all known inter-organ signals in T2DM showed bi-stability with one insulin sensitive and one insulin resistant attractor. The bi-stability was made robust by multiple positive feedback loops suggesting an evolved allostatic system rather than a homeostatic system. Certain evolutionary hypotheses do suggest existence of multiple stable states in a population which are adapted to different environmental conditions and social roles. Similarly, the bi-stability in this case and the preponderance of positive feedbacks in the network suggest co-existence of the diabetic state and the healthy state. The robustness was unlikely to have arisen due to one or a few nodes or links since deleting individual nodes and randomly adding links to the network did not disturb the bi-stability. Sensitivity analysis showed that this result wasn’t due to chance alone or due to any of the assumptions or contradictions. In the absence of the complete network, impaired insulin signalling alone failed to give a stable insulin resistant or hyperglycaemic state. The model made a number of correlational predictions, many of which were validated by empirical data. The current treatment practice targeting obesity, insulin resistance, beta cell function and normalization of plasma glucose failed to reverse T2DM in the model. However certain behavioural and neuro-endocrine interventions like up-regulations of dopamine, ghrelin, oestrogen and osteocalcin ensured a reversal. These results suggest novel prevention and treatment approaches which need to be tested empirically. The model also shows a difference in steady-state and perturbed-state causality and suggests that making steady-state predictions from perturbed-state data might have led to a confused cause-effect relationship in the field. Finally, a design of a network-level clinical study has been suggested with the kind of analysis used to interpret such a dataset.