Driver versus navigator causation in biology: the case of insulin and fasting glucose

Abstract

Background In biomedicine, inferring causal relation from experimental intervention or perturbation is believed to be a more reliable approach than inferring causation from cross-sectional correlation. However, we point out here that even in interventional inference there are logical traps. In homeostatic systems, causality in a steady state can be qualitatively different from that in a perturbed state. On a broader scale there is a need to differentiate driver causality from navigator causality. A driver is essential for reaching a destination but may not have any role in deciding the destination. A navigator on the other hand has a role in deciding the destination and the path but may not be able to drive the system to the destination. The failure to differentiate between types of causalities is likely to have resulted into many misinterpretations in physiology and biomedicine. Methods We illustrate this by critically re-examining a specific case of the causal role of insulin in glucose homeostasis using five different approaches (1) Systematic review of tissue specific insulin receptor knock-outs, (2) Systematic review of insulin suppression and insulin enhancement experiments, (3) Differentiating steady state and post-meal state glucose levels in streptozotocin treated rats in primary experiments, (4) Mathematical and theoretical considerations and (5) Glucose-insulin relationship in human epidemiological data. Results All the approaches converge on the inference that although insulin action hastens the return to a steady state after a glucose load, there is no evidence that insulin action determines the steady state level of glucose. Insulin, unlike the popular belief in medicine, appears to be a driver but not a navigator for steady state glucose level. It is quite likely therefore that the current line of clinical action in the field of type 2 diabetes has limited success largely because it is based on a misinterpretation of glucose-insulin relationship. The insulin-glucose example suggests that we may have to carefully re-examine causal inferences from perturbation experiments and set up revised norms for experimental design for causal inference.

Reduced blood to brain glucose transport as the cause for hyperglycemia in a model resolves multiple anomalies in type 2 diabetes

“…The question whether tight glycemic control leads to subtle neuronal changes in the long run as expected by our model needs careful investigation. The difference between fasting and post meal regression correlation parameters between glucose and insulin is an important epidemiological line of evidence we have used.Chawla et al (2017) and Diwekar-Joshi andWatve (2020) showed this pattern across four different data sets. How generalized the pattern is can be easily tested in multiple populations.…”

“…By classical models the regression correlation parameters of glucose-insulin relationship are not different in fasting state versus post glucose load although the range of variables is different as shown previously by Diwekar-Joshi andWatve (2020). Also, if we assume HOMA-IR to faithfully reflect insulin resistance and HOMA β to faithfully represent β cell response, then there is no reason why the two indices should be correlated.The assumption behind our model that there are different mechanisms at work under fasting versus post glucose load condition is necessary to explain the large difference in the regression correlation parameters in fasting versus post meal levels.…”

“…A unique assumption of our model based on the analysis by Diwekar-Joshi and Watve (2020) is that the mechanism of regulation of glucose and insulin is different in the steady state and post meal state. In the steady state the central mechanisms are more important whereas in post meal state mainly the peripheral mechanisms are at work.…”

“…The rise in FG stimulates insulin production and the rise in insulin levels normalizes glucose again. However, this mechanism fails to achieve a steady state hyperinsulinemic normoglycemic condition (Diwekar-Joshi & Watve, 2020).…”

“…A model should explain this observation. Diwekar-Joshi and Watve (2020) further differentiated between the consequential steady state (CSS) and targeted steady state (TSS) models and demonstrated that the results of the insulin receptor knock out and insulin suppression experiments can be explained by a TSS but not by a CSS model. In effect, to be able to account for these results a model needs to be a TSS model.…”