ReviewCell Biology

Cracking the context-specific PI3K signaling code

See allHide authors and affiliations

Science Signaling  07 Jan 2020:
Vol. 13, Issue 613, eaay2940
DOI: 10.1126/scisignal.aay2940

Abstract

Specificity in signal transduction is determined by the ability of cells to “encode” and subsequently “decode” different environmental signals. Akin to computer software, this “signaling code” governs context-dependent execution of cellular programs through modulation of signaling dynamics and can be corrupted by disease-causing mutations. Class IA phosphoinositide 3-kinase (PI3K) signaling is critical for normal growth and development and is dysregulated in human disorders such as benign overgrowth syndromes, cancer, primary immune deficiency, and metabolic syndrome. Despite decades of PI3K research, understanding of context-dependent regulation of the PI3K pathway and of the underlying signaling code remains rudimentary. Here, we review current knowledge on context-specific PI3K signaling and how technological advances now make it possible to move from a qualitative to quantitative understanding of this pathway. Insight into how cellular PI3K signaling is encoded or decoded may open new avenues for rational pharmacological targeting of PI3K-associated diseases. The principles of PI3K context-dependent signal encoding and decoding described here are likely applicable to most, if not all, major cell signaling pathways.

AN OVERVIEW OF CLASS IA PI3K RESEARCH

Class IA phosphoinositide 3-kinase (PI3K) enzymes catalyze the formation of the second messenger phosphatidylinositol 3,4,5-trisphosphate (PIP3). This phospholipid triggers a central signaling pathway in eukaryotic cells and regulates various downstream effectors including protein kinases, such as AKT and mTORC1 (mechanistic target of rapamycin complex 1), and transcription factors belonging to the FOXO family (Fig. 1) (1). The PI3K pathway is best known for its ability to coordinate anabolic metabolism and cell growth downstream of multiple growth factor receptors, including but not limited to those for insulin, insulin-like growth factor (IGF), vascular endothelial growth factor (VEGF), epidermal growth factor (EGF), and platelet-derived growth factor (PDGF). PI3K family members with disease-associated mutations include PI3Kα (encoded by the PIK3CA gene) and PI3Kδ (encoded by the PIK3CD gene) (1), which show ubiquitous or leukocyte-enriched expression, respectively (Fig. 1).

Fig. 1 Simplified schematic of canonical class IA PI3K signaling and cellular outputs.

Class IA PI3Ks exist as heterodimers composed of one of three catalytic subunits (p110α, p110β, or p110δ) bound to one of five regulatory subunits. They are commonly activated downstream of receptor tyrosine kinases when the regulatory subunit binds to phosphorylated tyrosine residues on the cytoplasmic domain of the receptor itself or associated adaptor proteins. The activation of individual PI3K isoforms may be enhanced further by RAS (PI3Kα and PI3Kδ), RAC/CDC42 (PI3Kβ), and/or G protein–coupled receptors (PI3Kβ). Once activated, class IA PI3Ks catalyze the formation of the second messenger phosphatidylinositol 3,4,5-trisphosphate (PIP3), which signals by binding to and recruiting effector proteins containing pleckstrin homology (PH) domains. Among these effectors, the AKT isoforms (AKT1/2/3) are involved in orchestrating key PI3K-dependent cellular phenotypes by acting on various cellular substrates, with some of the best characterized examples illustrated. These substrates also receive input from other pathways, and thus, the final phenotypic output is determined by context-dependent signal integration. Feedback loops are omitted for clarity.

CREDIT: A. KITTERMAN/SCIENCE SIGNALING

PI3K enzymatic activity was discovered 30 years ago (2), and the two decades that followed were focused on fundamental PI3K research. The 1990s saw the discovery of multiple PI3K isoforms and key components of canonical PI3K signaling, linking the activity of this pathway with control of essential cellular processes (3, 4). At the turn of the millennium, the first mouse models with disrupted PI3K activity demonstrated that several components of this pathway are required for organismal homeostasis and normal development (5). In addition, the PI3Kα isoform was found to be among the most commonly mutated oncogenes in solid tumors, whereas the gene encoding the PIP3 phosphatase PTEN emerged as one of the most frequently inactivated tumor suppressors, superseded only by TP53 (6, 7). The third decade of PI3K research has been dominated by the development and testing of PI3K pathway inhibitors as potential therapeutics for cancer and immune dysfunction. Although some PI3K inhibitors have been approved for clinical use, most of these compounds have failed to meet the initial high expectations and their utility in cancer treatment is limited by systemic toxicity and/or tumor drug resistance (1). However, some of these drugs have shown remarkable promise in the treatment of genetic disorders of PI3K dysregulation, including in the activated PI3Kδ syndrome (APDS) (8) and, when used at a lower dose than in cancer therapy, in the PIK3CA-related overgrowth spectrum (PROS) (9).

Although the key PI3K pathway components have now been identified, a fundamental gap in our understanding of PI3K signaling concerns how different cell and environmental contexts determine the functional outcome of pathway activation. It remains unclear how activation of the same set of components can trigger the vast repertoire of PI3K-driven phenotypic responses, which may be glucose uptake and proliferation in one setting, or senescence and even cell death in others. Moreover, the impact of mutational activation of the PI3K pathway on its signaling dynamics is largely undetermined.

With inspiration from progress made in the field of RAS/ERK (extracellular signal–regulated kinase) signaling, this Review summarizes emerging evidence supporting the importance of a context-specific PI3K signaling “code,” governed by distinct dynamics of pathway activation. Additional research in dynamic PI3K signaling may allow better understanding of how it controls normal physiology and becomes corrupted in disease, and how it can be modulated by pharmacological targeting.

EXAMPLES OF DYNAMIC INFORMATION TRANSMISSION IN CELL SIGNALING

Dynamic signal encoding and decoding

To sense changes in internal and external conditions, a cell needs mechanisms to encode these changes and, subsequently, to decode them into an appropriate response. For any individual hormone or growth factor signaling response, there is no single protein or gene that preserves signaling specificity; instead, this is achieved through dynamic regulation of multiple signaling effectors (10), hereafter referred to as “dynamic information transmission” (Fig. 2, A to D). Accordingly, a cell’s computational capacity—its ability to receive and process diverse signals—is determined by the intrinsic biochemical properties of its signaling components, including reaction rates, affinity constants, relative expression levels, and the presence of allosteric modulators (11, 12). Cell-specific differences in one or several of these parameters may lead to different and even opposite phenotypes downstream of the same upstream stimulus (13).

Fig. 2 Information transmission in cell signaling.

(A to D) Cells can respond to a signal’s rate (A) and duration (B); they can also respond to a signal’s strength (C) [amplitude modulation (AM)] or a signal’s temporal on/off pattern (D) [frequency modulation (FM)] (147149). Conversely, cells may also use similar changes in the dynamic activity of a shared set of intracellular components to encode the identity of the upstream stimulus (38, 55). (E) Example of low-fidelity signal transmission in cells with oncogenic RAS/ERK pathway activation. Altered dynamics in cells with specific BRAF variants result in misinterpretation of the upstream signal [adapted from (20)]. Similar corruption of information transmission, caused by enhanced BRAF-CRAF dimerization, has also been observed in response to the BRAF inhibitors SB590885 and vemurafenib (20).

CREDIT: A. KITTERMAN/SCIENCE SIGNALING

Although dynamic information transmission ensures that cells are capable of appropriately detecting, responding to, and even memorizing a stimulus, this complexity is difficult to capture experimentally and calls for high-density time course studies of single cells and multiple signaling effectors. In general terms, researchers interested in perturbing signaling dynamics in a controlled fashion first need to identify a system that allows direct manipulation of the input signal under study and high-resolution monitoring of relevant output responses. Beyond pharmacological approaches, synthetic biology tools, including chemically induced dimerization (CID) and optogenetic systems, enable extrinsic control of both temporal and spatial dynamics of signaling pathways [reviewed in (14)]. Key limitations of these technologies include the need for genetic manipulation of cells to express the synthetic protein controllers and additional fluorescent reporters to allow live single-cell imaging. Last, conceptualization of the obtained multidimensional data typically requires the generation of reliable and testable computational models that can predict the dynamic “input-output” response for a given pathway (10, 11).

RAS/ERK signaling dynamics

Studies on the RAS/ERK signaling pathway have been instrumental in demonstrating key aspects of dynamic information transmission in cultured mammalian cell lines and model organisms. Early experiments in the rat PC12 pheochromocytoma cell line by the Cohen and Marshall groups revealed that transient activation of ERK by EGF promotes cell proliferation, whereas sustained ERK activation by neural growth factor (NGF) triggers cell differentiation (15, 16). A systematic study of the EGF-ERK cascade in the human MCF10A mammary breast epithelial cell line demonstrated that the concentration of extracellular EGF becomes encoded in temporal parameters such as the frequency and pulse duration of downstream ERK activation (17). In turn, different patterns of ERK activity are integrated and decoded by its effectors to control the cell’s propensity to enter the cell cycle (17). This study also showed that intracellular ERK activity remains pulsatile, even in the presence of continuously high EGF levels in the medium, establishing frequency modulation (FM) as an important mode of information transmission in this pathway (17). In other words, these cells encode the growth factor dose in the frequency and duration of ERK activity pulses (17).

More direct evidence for the in vivo importance of ERK frequency modulation is now emerging. A light-inducible ERK activation system has demonstrated that different patterns of ERK signaling orchestrate distinct cell fate decisions in the fly embryo in a region-specific manner (18). Similarly, ERK frequency modulation has been linked to cell fate specification in Caenorhabditis elegans (19). These are prime examples of how activation of the same pathway can specify distinct biological outputs by using different signaling patterns.

Corrupted signaling dynamics downstream of RAS/ERK oncogenic mutations

The concept of corrupted signaling dynamics downstream of cancer-associated mutations is not new (11), but direct experimental evidence has only emerged in the last decade. In a seminal study, Bugaj et al. (20) used optogenetic stimulation of RAS to demonstrate that altered dynamic signal transmission properties, and thus not only a high level of baseline activation, contribute to the oncogenic properties of specific BRAF mutations (Fig. 2E). Rather than causing a constitutively “ON” state, some oncogenic BRAF mutations still allow the pathway to perceive upstream signals. However, the decay kinetics of downstream ERK phosphorylation is slower in cells expressing mutant BRAF compared to wild-type counterparts, which leads to loss of fidelity in signal transmission and an aberrant phenotypic response (Fig. 2E) (20). Furthermore, the BRAF inhibitors vemurafenib and SB590885 enhance downstream ERK signaling by corrupting the dynamic signal transmission properties of the system: Instead of rapid ERK signal decay that normally occurs when RAS signaling ceases, these BRAF inhibitors result in slow pathway deactivation and cellular misinterpretation of the original input signal (Fig. 2E) (20).

These findings have potential therapeutic implications by suggesting that it might be necessary for drug treatment to shift away from complete inhibition of the mutated components and instead explore how normal signaling dynamics can be restored (21), possibly by modulating the relevant upstream regulators (22). This notion is consistent with previous computational and experimental testing of the overall concept that signaling dynamics may serve as a pharmacological target (13).

Other signaling pathways

In addition to the RAS/ERK module, dynamic information transmission has been demonstrated for other signaling pathways including DNA damage–induced TP53 activation (23), nuclear factor κB (NF-κB) regulation in innate immune signaling (24, 25), Ca2+-regulated NFAT (26, 27), developmental transforming growth factor β (TGFβ)/NODAL (2831), WNT (32), SHH (33, 34) and NOTCH (3537) signaling, and, as will be detailed below, PI3K-dependent insulin signaling (38, 39). Nevertheless, our understanding of the dynamics of the PI3K pathway remains relatively crude. This deficiency is partially due to the underlying complexity that is poorly captured by the “pathway” simplification: In reality, PI3K activation orchestrates a highly branched network of signaling effectors, thus posing a substantial challenge for conventional signaling studies. Moreover, there are limited experimental options for selectively varying individual PI3K signaling features such as the strength and duration of activation of an enzyme and for overcoming complex feedback signaling loops.

EVIDENCE FOR A DYNAMIC PI3K CODE

Given the many parallels between the RAS/ERK and PI3K pathways, including their frequent activation in cancer, dynamic information transmission would be expected to assume similar importance in determining PI3K-based phenotypic output, in both health and disease. PI3K signaling is commonly depicted in static maps of varying complexity as new effectors and modulators are identified (Fig. 1). These conventional PI3K signaling maps may give the false impression of a hard-wired circuit, with identical output irrespective of context. In contrast, PI3K pathway activation can be associated with diverse and even opposite phenotypes, including cell growth, senescence, proliferation, and cell death. At the organismal level, disease-causing mutations in this pathway may promote cancer in one tissue and benign overgrowth in another (40). In short, the phenotypic output of PI3K signaling is remarkably flexible, governed by dynamic activation, cell type–specific gene expression, and changes in the microenvironment, as detailed below.

The first evidence for a dynamic PI3K code

The first direct evidence that cells compute decisions based on the dynamic properties of PI3K signaling was provided 20 years ago. Using HepG2 liver cancer cells stimulated with PDGF, the Kazlauskas group demonstrated the presence of early (0 to 1 hour after stimulation) and late (3 to 7 hours after stimulation) waves of PI3K activity, with the second PI3K wave being essential for induction of DNA synthesis and cell cycle progression (41). In 2002, Tengholm and Meyer suggested the existence of an insulin-specific PI3K signaling code to explain the translocation of cytosolic GLUT4 glucose transporters to the plasma membrane of 3T3-L1 adipocytes upon stimulation with insulin but not with PDGF (Fig. 3) (42, 43). Their data suggested that strong but transient activation of PI3K triggered by PDGF fails to elicit a response because the integrated concentration of PIP3 over time remains below a specific threshold for downstream GLUT4 translocation (Fig. 3) (42). In the same year, Sedaghat et al. (44) published the first computational model of metabolic insulin signaling, demonstrating how mathematical approaches can be used to capture pathway complexity and how they can serve as hypothesis-generating tools for known and unknown signaling mechanisms. Subsequently, several other dynamic models of PI3K signaling, with a particular focus on AKT and mTOR regulation, have emerged, differing with respect to time scale and network complexity [for a review of mTOR models, see (45)]. There is also an increasing appreciation that temporal and spatial regulation must be considered jointly (46), especially when it comes to understanding the exact dynamics and thresholds of pathway activation that are required to control metabolic versus mitogenic outputs (47).

Fig. 3 PIP3 dynamics encode distinct cellular responses.

Using 3T3-L1 adipocytes stimulated with platelet-derived growth factor (PDGF) or insulin, Tengholm and Meyer demonstrated that cells may use different patterns of PIP3 dynamics to encode the identity of the upstream growth factor and subsequently decode these dynamics into different responses. Thus, insulin, but not PDGF, triggers translocation of intracellular GLUT4 storage vesicles to the plasma membrane and subsequent glucose uptake (42).

CREDIT: A. KITTERMAN/SCIENCE SIGNALING

The discovery that growth factor–induced PI3K signaling exhibits a dynamic pattern of activation during the cell cycle suggested that constitutive PI3K activation might lead to cell cycle abnormalities (41), as observed by Klippel et al. (48) upon overexpression of constitutively active PI3Kα. In line with these observations, the Carrera group demonstrated that temporal PI3K down-regulation during the cell cycle is important for increased downstream FOXO1 activity at the time when this transcription factor is necessary for cell cycle completion (49). In a subsequent study, the cell cycle block could be avoided by ensuring near-endogenous levels of a constitutively active form of PI3Kα (50). A separate study also found cell cycle abnormalities in cells transiently overexpressing constitutively active AKT due to dysregulated activation and localization of the AKT substrate cyclin-dependent kinase 2 (CDK2) (51), a mechanism that was also suggested by the first study of this phenomenon by Klippel et al. (48). Subsequent findings have highlighted that these effects are likely to be context-dependent and thus vary as a function of both cell type and culture conditions (5254). Overall, these studies suggest that the temporal pattern of PI3K activation influences the dynamics of the mammalian cell cycle, with biological output determined by both cell-intrinsic and cell-extrinsic factors.

Renewed interest in the dynamic PI3K code

With advances in automated liquid handling systems and “-omics” technologies, more systematic studies of the mechanisms by which cells decode different patterns of PI3K activation have begun to emerge. By combining experiments in the rat pheochromocytoma PC12 cell line with mathematical modeling and concepts from electronic engineering, the Kuroda group offered a detailed characterization of AKT signaling dynamics in response to EGF stimulation (Fig. 4A) (55). A functionally coupled signaling response downstream of EGF receptor (EGFR) was observed for the activation of AKT, meaning that the relative magnitude of the upstream stimulus and the downstream response followed the same pattern (Fig. 4A). This pattern contrasted with the decoupled downstream signal transmission to mTORC1 and S6 kinase (S6K); in other words, S6 phosphorylation occurred most potently in response to weak, sustained EGFR activation, with the pathway effectively filtering out strong but transient signaling events (Fig. 4A) (55). It remains to be determined whether this decoupling has physiological relevance, but the authors speculate that it may ensure that S6-dependent ribosome and protein biosynthesis take place only when the upstream signal is of sufficient duration, thus limiting potential waste of cellular energy (55). Because additional PI3K/AKT-independent inputs may impinge on mTORC1 and S6 regulation in different contexts, further systematic studies are also needed to determine the potential contribution from such cross-talk.

Fig. 4 Examples of dynamic information transmission in the class IA PI3K signaling pathway.

(A) In the PC12 rat pheochromocytoma cell line, different patterns of epidermal growth factor receptor (EGFR) stimulation are transmitted differently to S6 kinase (S6K) downstream of AKT activation. Strong but transient EGFR stimulation is not transmitted efficiently from AKT through mTORC1 and S6K, representing a case of decoupled signal transfer in which the magnitude of the downstream response is opposite to that of the upstream signal. Instead, downstream S6 phosphorylation occurs most potently in response to weak but sustained EGFR activation. The EGFR kinase inhibitor lapatinib (dashed line) paradoxically enhances S6 phosphorylation by changing the dynamics of EGF-induced EGFR phosphorylation and activation. Adapted from (55). (B) Insulin levels in the blood oscillate according to specific patterns. These dynamic insulin changes are transmitted through phosphorylation of the insulin receptor and PI3K/AKT activation. Downstream, the different patterns of stimulation are selectively decoded through S6K and glycogen synthase kinase 3 (GSK3) phosphorylation as well as changes in G6P gene expression (38, 58). As a result, the activity of each component is in tune with different aspects of the upstream signal to elicit the most appropriate physiological response to insulin. Adapted from (58). Note that oscillations are not drawn to scale.

CREDIT: A. KITTERMAN/SCIENCE SIGNALING

Physiologically, the pancreas secretes insulin at a low constant level and in a 10- to 15-min pulsatile manner, with additional insulin secretion occurring in response to eating (56, 57). By providing different patterns of insulin stimulation in the rat hepatoma FAO cell line in vitro and in anesthetized rats in vivo, the Kuroda group demonstrated that insulin dynamics are captured differently by AKT and its downstream effectors as a result of differences in feedback regulation and kinetic constants (Fig. 4B) (38, 58). Although phosphorylation of S6K was most responsive to an increase in the rate of insulin exposure and was thus used by cells to detect transient stimulation, it always returned to the same basal level, regardless of stimulus duration or dose, a phenomenon known as “perfect adaptation.” Insulin dose and duration were better captured in the dynamics of glycogen synthase kinase 3 (GSK3β) phosphorylation and transcription of G6P (which encodes the enzyme glucose 6-phosphatase) (38, 58). Similar dynamic information transmission has also been demonstrated in insulin-stimulated mouse 3T3-L1 adipocytes (39). These findings may have important physiological implications if future studies demonstrate that dynamic signal encoding and decoding within individual cells is needed for insulin-responsive tissues to elicit an appropriate metabolic response to different physiological patterns of the hormone.

The dynamics of PI3K activation are also important during B cell selection in early development in which both hyperresponding and potentially self-reactive cell clones, as well as clones with poor response to antigen activation undergo negative selection. Low PI3K signaling occurs in poorly responsive cells, whereas strong PI3K activation characterizes autoreactive immune cells (59). Accordingly, pre-B cell negative selection takes place both when PI3K signaling falls below a certain lower threshold and when it exceeds an upper threshold of hyperactivation (60).

It is important to emphasize that the PI3K code does not exist in isolation and is subject to extensive cross-talk with other pathways. For instance, a quantitative study of NGF signaling in PC12 cells revealed that the cellular decision to differentiate or proliferate is determined by a two-dimensional phospho-ERK/phospho-AKT response map that integrates the activation strength of both pathways (61). In addition, the early and late dynamics of FOXO3 nuclear-cytoplasmic shuttling is differentially regulated by AKT and ERK downstream of different growth factors, potentially serving as a mechanism to encode the identity of upstream ligands (62).

Last, the PI3K code is likely to depend on the spatial distribution of the PIP3 and PI(3,4)P2 lipid products of PI3K activation. For example, only PI(3,4)P2 appears to move from the plasma membrane to early endosomal compartments where it results in preferential activation of AKT2 over other isoforms (63). Future studies are warranted to determine the extent to which spatiotemporal control of PIP3 and PI(3,4)P2 is used to encode distinct cellular phenotypes.

Technological challenges and potential solutions

A quantitative understanding of the dynamic PI3K signaling code requires “forward” experimental testing, using tools that allow precise control of PI3K activation. Several artificial systems have been developed for this purpose (Fig. 5, A and B), but none of these systems allow isoform-specific PI3K activation. Besides technical implementation, a remaining challenge is how best to quantify pathway dynamics, at the level of both PI3K activation and downstream responses. Current single-cell approaches rely on a limited set of PI3K signaling reporters (Fig. 5, C and D) and thus fail to capture the potential existence of a range of effector-specific responses. The use of exogenously expressed biosensors is also not without caveats; the potential for dominant-negative effects on signaling requires optimization of expression levels and use of appropriate controls to rule out either false positives or false negatives (64).

Fig. 5 Synthetic biology tools used in quantitative studies of PI3K signaling dynamics.

(A) Reversible, chemically induced dimerization (CID) system used to modulate class IA PI3K signaling. It relies on the expression of a synthetic inter-SH2 (iSH2) construct of p85 interacting with the p110 catalytic subunit in an isoform-agnostic manner (150). Dimerization is induced by rCD1, a synthetic moiety that binds to both the SNAP tag at the plasma membrane and an FKBP fusion protein in the cytoplasm. The interaction can be reversed by addition of FK506 or an inert rapalog, both of which compete for binding to FKBP. (B) One of the first PI3K optogenetic (light-inducible) systems relied on the reversible light-induced interaction between phytochrome-interacting factor (PIF) and phytochrome (PHY) (151). Several other light-inducible PI3K systems have subsequently become available (152, 153). Note that both current CID and optogenetic approaches inevitably perturb the endogenous stoichiometry between p85 and p110, with likely consequences for downstream signaling output (5, 154). (C) Principle behind commonly used genetically encoded PIP3/PI(3,4)P2 biosensors. Different PH domains may bind either one or both lipid species, leading to translocation of the fluorescent reporter from the cytosol to the plasma membrane [for a comprehensive review on these sensors, see (64)]. (D) Fluorescent FOXO-based nucleocytoplasmic translocation reporters are commonly used in dynamic single-cell studies of PI3K signaling (62, 155157). However, FOXO proteins are only responsible for a subset of PI3K-dependent phenotypes (158).

CREDIT: A. KITTERMAN/SCIENCE SIGNALING

Potential solutions are in sight, however. These include (i) the use of CRISPR-mediated tagging of endogenous effector proteins, such as AKT or FOXO, to follow their dynamic translocation live and without stoichiometric changes and (ii) the integration of quantitative, multiplexed immunofluorescence in time course studies that seek to assess a wider repertoire of signaling responses at the single-cell level (6567). Proof-of-concept studies from Sorger and his team illustrate the detailed cell signaling insight that can be obtained with some of these approaches and the translational potential of the acquired knowledge (62, 68). Adoption of these methodologies will likely be instrumental in closing the PI3K signaling knowledge gaps that will be discussed next.

EXAMPLES OF CONTEXT-DEPENDENT DIFFERENCES IN THE PI3K CODE

Differences in the PI3K code according to cell type

The phenotypic outcome of PI3K activation changes according to cell type, reflecting intrinsic differences in the expression of signaling components and downstream effectors. As a result, the same pattern of PI3K pathway activation may lead to distinct responses in two different cell types under otherwise identical conditions. For instance, an adipocyte and a muscle cell differ in their response to insulin-dependent activation of the PI3K pathway. Both induce an anabolic program, but according to different mechanisms, a muscle cell will predominantly engage protein synthesis and glycogen storage, whereas an adipocyte’s response will be biased toward lipid accumulation (69). Although this is an obvious example, differences are also likely to exist in otherwise similar cell types. For example, in their study of ERK- and AKT-dependent FOXO3 regulation, the Sorger laboratory used a panel of breast cancer cell lines and normal controls to demonstrate how differences in network topology result in cell line–specific dynamics of FOXO3 nuclear-cytoplasmic translocation (62).

PI3K signaling studies commonly use transformed cell lines or immortalized, nontumorigenic counterparts, whose signaling principles and phenotypic outputs cannot necessarily be extrapolated to those operating in untransformed cells that are more relevant for understanding normal regulation and mechanisms of early disease progression. Intrinsic biological differences across cell types and species are equally important to consider when evaluating oncogenic mechanisms, with early studies reporting different susceptibilities to transformation and senescence across different human cell lines as well as mouse compared to human fibroblasts (70, 71). Oncogenic activation of PI3K signaling elicits senescence in some cellular contexts but not others (60, 7279), with species, cell lineage, expression of key tumor suppressors (for example, TP53 and retinoblastoma protein), and the strength of PI3K activation emerging as important determinants.

A better understanding of cell type–specific PI3K signaling may also clarify the perplexing phenotypic complexity characterizing diseases of PI3K dysregulation. For example, activating mutations in PI3Kα are frequent in epithelial cancers originating in ectodermal and endodermal tissue derivatives, but when the same mutations are acquired developmentally in mesodermal and neuroectodermal tissues, the common outcome is nonmalignant overgrowth (40).

Differences in the PI3K code according to organismal and cell developmental stage

There are emerging indications that the effect of PI3Kα activation may also differ according to the developmental stage of the cells and the organism. Constitutive hyperactivation of the PI3K pathway has previously been linked to progenitor stem cell loss in mouse hematopoietic (80, 81), skeletal muscle (82), and epidermal lineages (78). Paradoxically, homozygous expression of the oncogenic PIK3CAH1047R variant delays tumor growth in the epidermis of mice expressing the human papillomavirus (HPV) E7 oncogene (78). Mechanistically, homozygosity for PIK3CAH1047R promoted the differentiation of epidermal progenitors downstream of increased phosphorylation of the AKT substrate SH3RF1, resulting in disruption of its scaffolding function and ability to promote c-Jun N-terminal kinase (JNK) signaling, which is critical for the maintenance of skin cell progenitors (78).

In contrast, oncogenic PI3K pathway activation has been linked to long-term stemness in both mouse and human pluripotent stem cells (hPSCs) (8387), which are used as models of the embryonic epiblast before gastrulation. Oncogenic PI3Kα activation downstream of PIK3CAH1047R can also induce multipotency in otherwise lineage-restricted, adult mammary epithelial cells in vivo (88, 89). The mechanistic basis for these observations remains limited, however. Computational network reconstruction and experimental follow-up suggest that homozygous expression of PIK3CAH1047R in hPSCs leads to signaling rewiring and self-sustained TGFβ pathway activation downstream of increased NODAL expression (90). Nevertheless, these findings require further confirmation and may only apply to contexts conducive to embryonic gene expression such as hPSCs and transformed tumor cells (91).

Overall, these observations suggest that developmental context, cell type, and differentiation stage may interact to determine the specific response to PI3K activation. It is notable that this concept was alluded to in 1999 when Rommel et al. (92) demonstrated differentiation stage–specific differences in the ability of AKT to associate with RAF and inhibit downstream ERK activation. The exact dynamics of PI3K activation are likely to modulate such relationships. As mentioned above, chronic PI3K activation in hematopoietic stem cells (HSCs) causes their exhaustion, yet transient pathway activation in response to physiological stress or cytokine stimulation is associated with better HSC regeneration and long-term maintenance (93). A similar phenotype has also been observed in skin adipocyte stem cells undergoing renewal (94).

Clinical observations indicate that activating PI3Kα (PIK3CA) mutations in developmental overgrowth disorders such as PROS are more likely to have been acquired in progenitor stem cells as opposed to pluripotent embryonic stem cells or terminally differentiated cell types (40). It is thus tempting to speculate that PI3K activation–induced negative selection or growth suppression in specific progenitor cells during embryogenesis may underlie the apparent absence of strongly activating PI3Kα mutations in hematopoietic and endodermal lineages in such disorders (40). Moreover, weaker PI3Kα variants are tolerated in a wider tissue distribution (40), perhaps reflecting dose-dependent differences in negative selection downstream of chronic PI3K activation.

Future studies are warranted to address these hypotheses, with consideration of the contribution of non–cell-autonomous effects linked to tissue complexity and niche-specific microenvironments in vivo. For instance, the relative strength of combined AKT and ERK activation in endothelial cells—a commonly affected cell type in PROS patients—balances self-renewal and differentiation of mouse HSCs in vivo (95). Consequently, chronic activation of AKT in endothelial cells promotes long-term maintenance of hematopoietic stem and progenitor cells, whereas concomitant activation of ERK signaling opposes this effect by triggering their differentiation (95).

Differences in the PI3K code according to microenvironmental conditions

Activation of the PI3K pathway is commonly held to lead to enhanced cell survival. However, this outcome depends on environmental context and can be modified by changes in nutrient, growth factor, and oxygen availability. Across a range of cell types, PI3K pathway activation enables survival under adverse conditions such as growth factor or serum removal (96100), ultraviolet-B (UV-B) irradiation (101, 102), and matrix detachment (102, 103). In contrast, when cells are cultured in the presence of growth factors or serum, several studies have reported that oncogenic PI3K pathway activation does not confer additional resistance to cell death (77, 87, 98). This finding suggests the existence of a PI3K activity threshold for survival, beyond which additional activity offers little benefit.

Under other conditions, PI3K activation can even lead to cell death. Using an inducible form of a constitutively active PI3Kα in rat embryonic fibroblasts, Klippel et al. (48) found that prolonged pathway activation (48 hours) in the absence of serum results in apoptosis that can be rescued by rapamycin. Another study noted that strong overexpression of constitutively active viral Akt was not well tolerated by a rat hippocampal cell line, whereas intermediate levels of overexpression offered protection against apoptosis (96). Because of increased energy demand and reactive oxygen species generation (104), cells with chronic PI3K activation are also sensitized to cell death under conditions of glucose deprivation (105, 106), hypoxia (107, 108), and oxidative stress (109111), although this sensitization may once again depend on the pattern and strength of pathway activation (109). The PI3K pathway also promotes the death of necrotic hematopoietic and neuronal cells, giving rise to a seemingly paradoxical rescue of cell viability upon PI3K pathway inhibition (112, 113). A similar response was reported in a mouse epidermal cell line treated with the proapoptotic factor Fas (114).

Microenvironmental conditions can also change as cells multiply and establish physical contacts with one another, coinciding with changes in the extracellular concentration of multiple factors. This, in turn, influences both the dynamics of and the response to PI3K activation, in ways that may not be revealed in conventional population-based cell studies. For example, breast epithelial MCF10A cells exhibit a bimodal distribution of PI3Kα expression and AKT phosphorylation, subject to modulation by both cell density and the expression of oncogenic PI3Kα variants (76). Through Eph receptor activation, cell density also modulates the spatial distribution of EGFR activity, with high densities resulting in selective suppression of downstream AKT activation (115). More generally, nongenetic heterogeneity in cell signaling is known to exist due to intrinsic differences in protein expression within individual cells (76, 116118).

POTENTIAL CORRUPTION OF THE PI3K CODE BY PI3K PATHWAY MUTATIONS

A wealth of information is available on activating PI3Kα and PTEN loss-of-function mutations when it comes to key phenotypes such as cancer growth, survival, and metabolism. However, we know very little about whether these genetic alterations corrupt PI3K signaling dynamics and how this may contribute to the observed phenotypic changes. Given computational evidence that many cancer mutations are likely to result in dynamic and structural rewiring of signaling networks (119, 120), a better understanding of a putative “mutant” PI3K code is warranted.

Distinct PI3Kα mutations differ in their potency to activate the pathway (121, 122), and we have also demonstrated that differences in allele dosage of the same mutation, PIK3CAH1047R, cause near-binary phenotypic differences in hPSCs (87). With evidence that corrupted signaling dynamics comprise a defining feature of oncogenic mutations in the RAS/ERK pathway (Fig. 2E) (20), similar questions await to be addressed in relation to PI3K signaling. Do oncogenic mutations in the PI3K signaling pathway cause an amplitude increase in PIP3—at baseline and/or in response to growth factors? Or do they (also) increase signal duration following external stimulation? How do they affect the natural temporal dynamics of PI3K activation, and would such changes be sufficient to result in corrupted information transmission within the cell? Do oncogenic PIK3CA mutations give rise to mutant p110α proteins with an altered subcellular localization and spatial dynamics of PI3K signaling? Are differences in spatiotemporal signaling dynamics important determinants of the phenotypic variability observed across different mutations in vivo and across different doses of the same mutation? How might disease-associated changes to the dynamic PI3K signaling code be shaped by the cell type and its microenvironment?

Some evidence that constitutive activation of PI3K alters the cellular decoding of growth factor stimulation was provided by Klippel et al. (48) in their study of rat embryo fibroblasts constitutively expressing membrane-targeted forms of PI3Kα or AKT. Subsequently, work from the Sorger group on AKT- and ERK-dependent control of FOXO3 was the first—and remains the only—study to touch upon this complexity in a systematic manner. Although the study does not extend to cellular decision making, it demonstrates that oncogenic PIK3CA mutations reduce the dynamic range over which FOXO3 can respond to growth factors in human breast cancer cell lines (62). Thus, similar to the discovery of corrupted information transmission in cancer cells with oncogenic BRAF mutations (Fig. 2E) (20), cells with PIK3CA mutations may exhibit low-fidelity transmission of upstream signals.

The benefits of efforts to capture this complexity extend beyond the realms of oncology and are equally relevant for our understanding of diseases such as APDS and PROS. As alluded to by Kubota et al. (58), insight into pathological changes to PI3K signaling dynamics may also contribute to a better understanding of the phenomenon of selective insulin resistance in which only insulin-dependent glucose regulation but not lipid or protein synthesis is compromised.

THE CONTEXT-DEPENDENT PI3K CODE: A CHALLENGE AND AN OPPORTUNITY FOR THERAPEUTIC TARGETING

Multilevel pathway dynamics and limited therapeutic success of PI3K targeting in cancer

BRAF inhibitors can corrupt the signaling dynamics of the RAS/ERK pathway in cultured cells, resulting in paradoxical pathway activation and loss of signaling fidelity akin to that observed with specific oncogenic BRAF mutations (Fig. 2E) (20). Similarly, the EGFR inhibitor lapatinib can lead to a paradoxical increase in S6 phosphorylation in rat pheochromocytoma cells (Fig. 4A) (55). Such findings of unexpected pathway rewiring illustrate an important limitation in conventional thinking about pharmacological targeting of disease-associated signaling pathways. In cancer, the most common approach relies on a priori predictions about the right dosing regimen (123). Although systems biology studies suggest otherwise (124126), such predictions are mainly founded on the belief that effective disease management can be achieved through direct pharmacological manipulation of one or several molecular targets identified through genomic sequencing efforts. This approach fails in the face of continued phenotypic cancer evolution driven by nongenetic heterogeneity and drug-induced signaling rewiring (127).

Therapeutic targeting of cancers with mutational PI3K pathway hyperactivation is commonly based on continuous high-dose inhibitor administration (often the so-called maximum tolerated dose defined in phase 1 clinical trials). This strategy has so far had limited success in cancers associated with PI3K mutations and is further compromised by adverse effects due to on-target PI3K inhibition in normal tissues (128). In particular, hyperglycemia is a major problem because it feeds back to the pancreas to trigger rapid insulin secretion, which, in turn, activates the PI3K pathway and counteracts drug-induced PI3K inhibition (129).

Beyond toxicity and systemic feedback, pharmacological inhibition of PI3K signaling is also dampened by cell-intrinsic adaptive and acquired resistance (1), which are rooted in the context-specific properties of the PI3K code. For example, negative feedback regulation within the PI3K pathway allows for extensive adaptation to external perturbation through both rapid phosphorylation of key proteins and delayed transcriptional responses (1). Transcriptional changes may also be accompanied by changes to the signaling code through epigenetic modifications (130), thereby enabling adaptive resistance to spread across an entire cell population during subsequent division. Similarly, existing cells may acquire genetic alterations that make them resistant to PI3K inhibition (1). The selective expansion of a few resistant cells may eventually result in tumors exhibiting full-blown drug resistance. Predicting drug-induced rewiring remains a challenge, however, with network analyses suggesting extensive plasticity and heterogeneity in the signaling response of cancer cells that have become resistant to PI3K pathway inhibition (131). These findings underscore the importance of systematic analyses of the PI3K code in a cell type– and context-dependent manner (131).

Integration of drug therapy approaches with systems biology

The pattern of limited therapeutic success is not unique to cancers with PI3K pathway activation. Dynamic mechanisms of adaptation operate within most, if not all, signaling pathways and, as demonstrated for the PI3K pathway, often extend beyond individual cells to encompass tissue cross-talk. It is therefore unsurprising that the results of traditional reductionist approaches have insufficient predictive power when it comes to therapeutic success—such methods are simply unable to capture the complexity of the system under study.

The incorporation of knowledge about the PI3K code into rational therapeutic design may benefit from input from the rapidly maturing field of systems biology, which is aimed at dealing with higher-order complexity. Briefly, systems biology approaches rely on dynamic, high-content datasets and mathematical abstractions in the form of computational models. The best models are able to capture causal signaling relationships and can simulate their dynamics in response to various perturbations, be it pharmacological targeting or a mutation in a key component. The quality of such models itself depends on information from conventional studies, including the biochemical properties of individual signaling components, their temporal behavior, and their spatial organization (11, 132).

A system can take many forms—an individual cell, homogeneous cell populations, heterogeneous tissues in vivo, or entire organisms. More complex systems can be addressed with so-called multiscale modeling approaches (11, 133). Multiscale models of different tumors are emerging, taking into account nutrient diffusion rates, blood vessel density, and individual probabilities for cell division, migration, and death (133). Similar models could be developed to integrate knowledge about the context-dependent PI3K signaling code in cell culture systems with the higher-order complexity of physiological or pathophysiological systems in vivo.

Optimizing drug dose, drug combinations, and temporal delivery

The rationale for using mathematical models of signaling dynamics for improved therapeutic targeting, particularly in cancer, has been covered extensively (11, 13, 123, 134). Here, we will use examples from diseases of PI3K activation to illustrate more specifically how a quantitative understanding of the context-specific PI3K code may benefit clinical drug development in this area.

Quantitative models of the relationship between PI3K signaling thresholds and context-specific phenotypes could be used to determine the level of pathway inhibition that is needed to achieve suppression of a specific disease phenotype. In particular, simulations may predict that lower and potentially less toxic PI3K inhibitor doses are clinically effective, thereby offering a broader therapeutic window. For example, continuous inhibition of PI3Kα with low-dose BYL719 (trade name, alpelisib) has proven to be therapeutically beneficial in some individuals with PROS, with no or minimal adverse effects (9). This treatment strategy contrasts with high-dose PI3K inhibition to treat cancer, which, as mentioned above, is associated with glucose-mediated metabolic feedback and hyperinsulinemia (129).

The remarkable effect of low-dose BYL719 in PROS begs the question whether the same therapeutic strategy should be tested in cancer (135). Such low-dose PI3K pathway inhibition would not necessarily reduce excess cancer growth or proliferation, but could potentially allow for “normalization” of PI3K signaling and thereby dampen ongoing tumor evolution (77, 136). It is tempting to speculate that one may even consider a low-dose cocktail of targeted drugs to simultaneously dampen multiple oncogenic pathways (136).

It is plausible that dynamic computational models of the PI3K code will not reveal a beneficial effect of low-dose PI3Kα inhibition in some or all tumor contexts. In such cases, in silico experiments can be performed to identify alternative strategies, including intermittent high-dose PI3K inhibition. Computational simulations may also identify critical protein-protein interactions responsible for specificity in dynamic signal encoding. Rather than inhibiting the oncogenic PI3K enzyme directly, modulation of such interactions will serve to dampen some aspects of pathway activation but not others. The potential promise of this strategy has also been discussed in the context of therapeutic targeting of the RAS/ERK pathway, where blockade with an allosteric SHP2-targeting drug would limit the signaling flux to downstream oncogenic proteins (22).

Last, quantitative tumor models that capture the PI3K code in various healthy and PI3K mutant cells could provide insight into the interaction between tumor cells and their stroma and how this interaction may be modulated by therapeutic targeting. For example, PI3Kδ-targeting inhibitors, which are clinically approved for specific B cell malignancies, not only act on the cancer cells themselves but also disrupt the tumor cell–stroma interactions, a major aspect of their therapeutic effect (137). Conversely, systemic high-dose PI3Kδ inhibition also leads to adverse effects, inducing elements of immune activation and immunosuppression (138140), once again highlighting the importance of getting PI3K signaling dynamics “just right.”

Computational pan-cancer modeling has already demonstrated that oncogenic PIK3CA mutations are associated with context-specific regulatory programs and signaling networks in different cancers (141), highlighting ways in which such knowledge can be used in the development of improved therapies. Thus, although the context-specific code of PI3K signaling presents a challenge for optimal therapeutic targeting, its quantitative understanding and incorporation into mathematical models may allow rational improvements of current and future clinical strategies.

Once computational models of the context-dependent PI3K code become widely available, subsequent in silico testing of dynamic drug dosing regimens comes at a relatively low cost and has the power to test multiple conditions within a short amount of time. This contrasts with current trials of dynamic dosing of PI3K inhibitors in cancer (142146), which are limited to a handful of regimens and may lack sufficient preclinical evidence to determine the optimal dosing pattern for in vivo application.

SUMMARY AND FUTURE DIRECTIONS

The first study providing experimental evidence for cellular encoding and decoding based on distinct PI3K signaling waves was published in 1999 (41). Two decades later, our understanding of the underlying PI3K signaling code and how it changes in different contexts remains limited. Thus, although we have a detailed understanding of the pathway’s hardware, we know little about the controlling software and how it is programmed.

With this Review, we aimed to highlight the need for a better understanding of PI3K signaling, particularly how stimulus dynamics integrate with cell type, developmental stage, microenvironment, and mutational status to elicit distinct biological outputs (Fig. 6). These parameters are poorly captured by the conventional studies of the pathway performed to date. For example, proliferation and cell survival are widely linked to PI3K pathway activation, yet they are far from being its universal outcomes and are critically context dependent.

Fig. 6 The context-specific PI3K signaling “tune”.

Similar to the melody from an accordion, the output of PI3K signaling is shaped by the integration of multiple input parameters.

CREDIT: A. KITTERMAN/SCIENCE SIGNALING

Beyond its fundamental value, understanding dynamic PI3K signaling could also provide a framework to rationalize drug targeting approaches in cancer, such as intermittent high-dose regimens of PI3K inhibitors or continuous exposure to low-drug doses. Fundamentally, the key questions outlined in this Review are generalizable and equally important to address in the context of most, if not all, other cell signaling pathways.

Tackling context-dependent PI3K signaling dynamics will be challenging, but continued technological advances and cross-disciplinary collaborations between biologists and computational scientists should allow studies to connect the well-known PI3K signaling hardware with its underlying software, a task that is likely to shape the fourth decade of research into this fascinating and druggable biological pathway.

REFERENCES AND NOTES

Acknowledgments: We are grateful to R. Semple, B. Bilanges, S. Conduit, L. Gerosa, and J. Longden for their helpful feedback during the preparation of this manuscript. Funding: Work in the laboratory of B.V. is supported by PTEN Research, Cancer Research UK (C23338/A25722), the UK Biotechnology and Biological Sciences Research Council (BB/I007806/1, BB/M013278/1, and BB/R017972/1), and the UK NIHR University College London Hospitals Biomedical Research Centre. Author contributions: R.R.M. and B.V. conceptualized the review. R.R.M. wrote the original draft and prepared the figures. Both R.R.M. and B.V. reviewed and edited the final manuscript version. Competing interests: B.V. is a consultant for Karus Therapeutics (Oxford, UK), iOnctura (Geneva, Switzerland), and Venthera (Palo Alto, USA) and has received speaker fees from Gilead (Foster City, USA).
View Abstract

Stay Connected to Science Signaling

Navigate This Article