PI3K signaling through a biochemical systems lens

Abstract

Following 3 decades of extensive research into PI3K signaling, it is now evidently clear that the underlying network does not equate to a simple ON/OFF switch. This is best illustrated by the multifaceted nature of the many diseases associated with aberrant PI3K signaling, including common cancers, metabolic disease, and rare developmental disorders. However, we are still far from a complete understanding of the fundamental control principles that govern the numerous phenotypic outputs that are elicited by activation of this well-characterized biochemical signaling network, downstream of an equally diverse set of extrinsic inputs. At its core, this is a question on the role of PI3K signaling in cellular information processing and decision making. Here, we review the determinants of accurate encoding and decoding of growth factor signals and discuss outstanding questions in the PI3K signal relay network. We emphasize the importance of quantitative biochemistry, in close integration with advances in single-cell time-resolved signaling measurements and mathematical modeling.

Keywords: AKT; PI3K; growth factors; signaling; systems biology.

Figure 1

Context-dependent PI3K signaling. The same core PI3K signaling scaffold can be deployed differently in different cell type and lineages, exemplified here with commonly used adherent (HELA) and suspension (JURKAT T) cell line models and stimuli. The first set of subspaces show how the core PI3K signaling scaffold becomes modified because of cell line–specific expression of individual components. For example, HELA cells do not express BTK and ITK and have lower expression of PI3Kδ compared with JURKAT T cells. Within each cell line, multiple yet poorly understood instantiations of the modified PI3K signaling scaffold are possible depending on culture conditions and the application of specific stimuli. For example, the amplitude and/or cellular localization of a phosphorylated component may change as a function of the dose of and the time after a stimulus and may differ yet again for different stimuli. Cells use such changes in spatiotemporal PI3K signaling to achieve specificity in biochemical information transfer and downstream phenotypic control. For simplicity, this figure does not capture the full scale of the signaling network nor the extensive crosstalk with other classical pathways or the impact of non–cell-autonomous inputs. Created with BioRender and AffinityDesigner.

Figure 2

Feedback control and dynamic signaling plasticity. Feedback regulation is represented by two main motifs: negative (A) and positive (B) feedback loops. These endow signaling systems with different properties, examples of which are given in the respective boxes. For each regulatory motif, an example from the PI3K signaling network is given below. The example of negative feedback regulation reflects a synthesis of work presented in Refs. (93, 113, 114, 115, 116, 201, 202, 203, 204, 205, 206, 207, 208, 209, 210, 211). The example of positive feedback regulation was adapted from Ref. (165). Feedback motifs can operate at different time scales, as determined by the underlying molecular mechanism (e.g., post-translational feedback mechanisms are faster than transcriptional feedback mechanisms). This, in turn, enables the precise tuning of the signaling response, which is used to specify a phenotypic outcome. While the examples in the figure indicate separate usage of positive and negative feedback motifs, the two can be combined to generate complex signaling functions. For example, mixtures of positive and negative feedback featuring integration of PI3K and RAS activity as part of an excitable network can be used to generate self-propagating cytoskeletal waves or cell polarization (11, 12, 212). Created with BioRender.

Figure 3

Biochemical determinants of dynamic signal encoding. Available knowledge of signaling topology and biochemical parameters (rate constants, affinity constants, and concentrations) can be used to construct predictive mathematical models of signaling phenomena, provided that such models are calibrated with the relevant experimental data. Shown here is how one of the first models of short-term epidermal growth factor (EGF) signaling in primary hepatocytes could be used to simulate the nonlinear kinetics of GRB2 recruitment to the activated EGF receptor (EGFR) as a function of concentration changes in GRB2 itself or SHC. GRB2 and SHC represent key adaptor proteins for activation of RAS–ERK and PI3K signaling downstream of EGF. Created with BioRender and adapted from Ref. (34). ERK, extracellular signal–regulated kinase.

Figure 4

Single-cell heterogeneity and limitations of snapshot signaling measurements. Across an otherwise isogenic cell population treated with an increasing concentration of IGF1, a snapshot measurement of AKT activity will reveal a wide response distribution at the single-cell level (86). This heterogeneity is obscured by bulk immunoblotting for AKT substrate phosphorylation. Population snapshot measurement cannot reveal whether the observed response heterogeneity is caused by high intracellular response variability, high intercellular variability with otherwise stable individual responses, or a mixture of both. To resolve this, one needs dynamic signaling measurements at high temporal resolution, particularly in the case of an ergodic system (one where the time average of a rapidly fluctuating signaling response in a single cell equals the population average in a snapshot measurement). Live-cell measurements of AKT reporter activity in single cells suggest that individual cell responses are stable, with population heterogeneity arising from a mixture of stable high and low responders (86). Such low intracellular variability in PI3K–AKT signaling may enable individual cells to encode extracellular signals reliably. Created with BioRender based on Refs. (86, 97). IGF1, insulin-like growth factor 1.

Figure 5

Probabilistic signaling response maps.A, when rat pheochromocytoma (PC12) cells are stimulated with neural growth factor (NGF) versus epidermal growth factor (EGF), they exhibit differences in their propensity for differentiation versus proliferation. Quantitative single-cell analyses have revealed the existence of a sharp nonlinear decision boundary governed by a two-dimensional pAKT–pERK plane. PC12 cells with a higher probability of proliferation have higher pAKT and lower pERK, and vice versa for cells that are more likely to differentiate. For each perturbation, a response vector is drawn to indicate the direction of the population shift. The position of the cell population along the decision boundary is determined by a feedback mechanism that involves PI3K signaling, RAS–ERK signal modulation, and the Ras GTPase-activating protein (GAP) Rasa2 (B). Combined with stochastic variation in protein expression levels of individual signaling components, this constellation ensures that the same stimulus can give rise to coexisting cell populations with distinct phenotypes, according to specific probabilities. Both (A and B) are adapted from Ref. (106) and represent approximations of the original plots. ERK, extracellular signal–regulated kinase.

Figure 6

PI3K signaling through a biochemical systems lens. A systems biology of PI3K signaling requires a cyclic integration of theoretical knowledge, computational models, and experimental studies. Accurate models of PI3K signaling responses must feature high-quality biochemical measurements, attention to model- and experiment-specific variables, including dose and time. Careful spatiotemporal measurements are needed to resolve the nonlinear complexity of PI3K signaling responses, and how this “code” contributes to reliable information transmission within individual cells and, ultimately, across entire cell populations. Addressing this gap in quantitative PI3K signaling in a time- and cost-efficient manner requires systematic interdisciplinary effort to enable standardized experimentation and downstream data integration.