Quantitative and Qualitative Differences in the Activation of a Fibroblast Growth Factor Receptor by Different FGF Ligands

Abstract

The FGF system is the most complex of all receptor tyrosine kinase signaling networks with 18 FGF ligands and four FGFRs that deliver morphogenic signals to pattern most embryonic structures. Even when a single FGFR is expressed in the tissue, different FGFs can trigger dramatically different biological responses via this receptor. Here we show both quantitative and qualitative differences in the signaling of one of the FGF receptors, FGFR1c, in response to different FGFs. We provide an overview of the recent discovery that FGFs engage in biased signaling via FGFR1c. We discuss the concept of ligand bias, which represents qualitative differences in signaling as it is a measure of differential ligand preferences for different downstream responses. We show how FGF ligand bias manifests in functional data in cultured chondrocyte cells. We argue that FGF-ligand bias contributes substantially to FGF-driven developmental processes, along with known differences in FGF expression levels, FGF-FGFR binding coefficients and differences in FGF stability in vivo.

Introduction

The development and maintenance of tissue homeostasis depends on extracellular communication signals that control basic cell functions. Fibroblast growth factors (FGFs) represent an important molecular component of this cell-to-cell communication. In mammals, the FGF family consists of 18 extracellular ligands, which have a characteristic beta-trefoil fold and possess mitogenic or metabolic activity (1). FGFs act as morphogens, growth factors, or metabolic hormones to regulate many important processes in development, life and disease (2). FGFs signal over different distances in the organism. This includes relatively short distance auto- and paracrine tissue signaling that is typical for most FGF growth factors, as well as long-range communication between organs in the case of the metabolic hormones FGF19, FGF21 and FGF23 (3).

The morphogenic and growth factor signaling of FGFs is essential for cell proliferation and differentiation during the development of virtually all tissues and organs. Examples of FGF-regulated processes include blastocyst formation (FGF4), gastrulation (FGF8), heart and brain morphogenesis (FGF15, FGF16, and FGF17), and limb patterning and growth (FGF4, FGF8, FGF9, FGF10, FGF18) (4–12). To fulfill their functions, the 18 FGF ligands exhibit a complex spectrum of interactions with their FGF receptors (FGFR1–4) (13, 14).

The FGFRs are single-pass membrane receptors, all of which belong to the receptor tyrosine kinase (RTK) family. FGFRs consist of an extracellular ligand-binding domain containing three immunoglobulin-like (IgI-III) domains, a single transmembrane domain, a juxtamembrane domain, and an intracellular region containing a split tyrosine kinase domain (Figure 1). There are four different genes that code for human FGFRs (FGFR1, FGFR2, FGFR3 and FGFR4). Each FGFR type undergoes alternative mRNA splicing leading to multiple isoforms. The splice variants, especially in the Ig-like domain III, lead to different ligand specificities (15). In FGFR1–3, domain III splicing leads to three versions of the FGFRs (IIIa, IIIb and IIIc). IIIa (“a”) is a secreted extracellular FGF-binding protein with no signaling function, while IIIb (“b”) and IIIc (“c”) are cell surface receptors that bind FGF ligands. The “b” isoform is restricted to epithelial lineages, whereas the “c” isoform is preferentially expressed in mesenchymal lineages. FGFR4 has a single isoform (16, 17). In addition, splicing can give rise to a “short” form of the receptor with two Ig-like domains, which has similar ligand-binding properties to the “long” FGFRs with three Ig-like domains, but possibly a higher affinity for certain FGFs (18).

Figure 1.

FGFR architecture and downstream signaling. FGFR extracellular region is composed of three extracellular Ig-like domains (Ig I - III). The ligand-binding site is located between Ig II and Ig III. An alternative splicing site, localized at the C-terminal part of Ig III, generates “b” or “c” variants of FGFR. FGFR has a single transmembrane domain, a juxtamembrane domain, and a split tyrosine kinase domain. FGF binding stimulates FGFR tyrosine kinase activity, leading to the activation of cytoplasmic proteins such as FRS2, STAT, and PLCγ. This initiates downstream signaling cascades which induce gene transcription and lead to cellular responses such as growth arrest, extracellular matrix (ECM) degradation, proliferation, survival, and motility.

The activity of FGFRs is controlled by their lateral association to dimers. At the cell surface, FGFRs can be found either as monomers, as unliganded dimers or as ligand-bound dimers, depending on the FGFR expression level and FGF concentration (19, 20). FGFRs exert their activity primarily as ligand-bound dimers. Ligand binding stabilizes the FGFR dimers and induces a structural change in the dimers that brings the two receptors into the correct orientation to trigger a conformational switch in the intracellular kinase to the active state (21). Once activated, the kinases phosphorylate multiple tyrosine residues on the neighboring receptor, creating binding sites for signaling mediators such as STATs and PLCγ (22) and activating the FRS2 adaptor protein which is associated with the juxtamembrane region of the receptor (23, 24). These proteins transduce intracellular signals, by initiating downstream signaling pathways such as Ras/Raf-MEK-MAPKs, PI3K-AKT, PLCγ-PKC, and STAT-p21. These various signal transduction cascades regulate embryonic development, tumor growth, angiogenesis, and wound healing (25) (Figure 1).

Both in vitro and in vivo data show that cells differentiate between the different FGF ligands. For example, cultured embryonal carcinoma cells differentiate into cardiac muscle in response to FGF1, but into skeletal muscle in response to FGF2 (26). In rat oligodendrocyte progenitors, FGF2 induces proliferation, whereas FGF8 inhibits terminal differentiation into mature oligodendrocytes (26). Similarly, in human adipose-derived mesenchymal stem cells, FGF2 promotes proliferation while FGF1 has no effect on cell growth but induces differentiation into preadipocytes (27). In mouse embryonic stem cell-derived astrocytes, FGF2 treatment increases proliferation, while other FGFs such as FGF16 and FGF18 mainly affect astrocyte maturation (28). In rat C5.18 calvarial cells, both FGF2 and FGF9 increase proliferation but have opposite effects on chondrocyte differentiation (29). In rat kidney organ cultures, several FGFs modulate branching morphogenesis of ureteric buds (UB), but produce distinctly different effects. While FGF1 induces an elongated UB stalk with well-differentiated, growing ampullary tips, FGF7 causes non-selective proliferation with poor morphological differentiation (30). Different variants of a single FGF can induce strikingly different phenotypes when acting on one cell type, as exemplified by the effects of FGF8 isoforms on brain morphogenesis. Ectopic expression of FGF8a in the developing mouse midbrain leads to midbrain expansion while FGF8b transforms the midbrain into cerebellum (31–33). Differential FGF signaling in different cell types could be due to binding preferences for FGFRs, as cultured cells typically express multiple FGFR variants. However, even when only a single FGFR is expressed, different FGFs can trigger dramatically different biological responses via this receptor, as shown in the regulation of early limb bud growth by FGF4, FGF8 and FGF9 (34).

How is the differential signaling of several FGFs via one FGFR mediated is a critically important question in biology. Here, we provide an overview of both quantitative and qualitative differences in the signaling of one of the FGF receptors, FGFR1c, in response to different FGFs, and describe the novel concept of FGF ligand bias. Bias represents qualitative differences in signaling as it is a measure of differential ligand preferences for different downstream responses.

Hierarchy of FGF effects in Ba/F3 and RCS cells expressing FGFR1c: one vs. multiple measured effects

Ornitz and colleagues were the first to measure the mitogenic activity of different FGF ligands through the same FGFR (13, 14). These studies were performed in the pro‐B cell line Ba/F3 expressing FGFR1c upon treatment with exogenous FGFs. The mitogenic response was assessed by quantification of de novo DNA synthesis. These experiments revealed a hierarchy of effects of the different ligands that activate FGFR1c. The mitogenic response to 5 nM FGF is strongest for FGF4, followed by FGF8 and FGF9 (FGF4 >FGF8>FGF9) (13). The limitation of the Ba/F3 model is that only one type of response (proliferation) to the FGF stimulus can be measured, which does not reflect the full spectrum of cell responses to FGFs.

FGF ligands can initiate many different signaling pathways when binding FGFRs, and these pathways regulate cell differentiation, migration, metabolism, and apoptosis. Many of these effects can be quantitatively measured in cultured rat chondrosarcoma (RCS) cells, including FGF regulation of the cell cycle, cell proliferation, differentiation, premature senescence, loss of extracellular matrix, interaction of FGF and WNT signaling, cytokine and natriuretic peptide signaling, and others (35–43). By western blot, RCS cells express detectable levels of FGFR1c and FGFR2c. While FGFR3 and FGFR4 cannot be detected by western blot, transcripts for FGFR3 and FGFR4 can be detected by RT-PCR. To isolate the effects of FGFR1c, RCS cells were engineered to express only FGFR1c using CRISPR-Cas9 to inactivate the endogenous FGFR2, FGFR3 and FGFR4 genes (RCS^Fgfr1c cells) (41).

Since RCS cells are known to respond to FGF treatment with growth arrest, RCS^Fgfr1c cells were treated with 1 μg/ml heparin and the FGF ligands FGF4, FGF8 and FGF9 for three days. Cell counting revealed a decrease in cell numbers in response to the FGF ligands (44). The dose-response curves in Figure 2A show that the extent of FGF-induced growth arrest increases with FGF concentration, with the curves flattening and approaching a plateau at high FGF concentration. This type of dose-response curve has the rectangular hyperbolic shape observed in functional studies of other membrane receptors, including RTKs and G-protein coupled receptors (GPCRs) (45, 46). At 5 nM ligand, we observe a very similar degree of growth arrest induced by the three ligands. This result differs markedly from Ornitz’s mitogenic results in Ba/F3 cells (13), where at 5 nM the ligands have a clearly defined hierarchy (FGF4>FGF8>FGF9). The maximal response in Figure 2A, which occurs at the highest ligand concentration in the plateau region, is the efficacy of the ligand. The efficacy of FGF4, FGF8 and FGF9 in inducing growth arrest is similar. Thus, all three ligands can be classified as “full agonists” in terms of their ability to induce growth arrest.

Figure 2.

Functional FGFR1c-mediated responses to different ligands in RCS cells expressing only FGFR1c. (A) Dose-response curves for growth arrest of RCS^Fgfr1c cells after 72 hours in response to FGF4, FGF8 and FGF9. Data are from (44). (B) Dose-response curves describing the loss of collagen type 2. RCS^Fgfr1c cells were treated with FGF4, FGF8 and FGF9 for 48 hours and the levels of collagen type 2 were determined by western blot. Data are from (44). (C) Ligand bias is a phenomenon that can lead to different receptor signaling properties and biological effects. The diagram shows three cases where two ligands cause two distinct biological effects (e.g. growth arrest or collagen 2 loss). The thickness of the arrows represents the efficiency of a response, which is a combination of the potency of the ligand and the magnitude of the response. In case I, only response 1 is triggered more efficiently by ligand A than by ligand B, so ligand A is biased toward response 2 relative to ligand B. In case II, ligand A triggers both responses more efficiently than ligand B, but both responses are equally stronger in response to ligand A and therefore there is no bias. In case III, response 2 is stronger for both ligands, but there is no bias in the responses induced by the two ligands (D). Bias plots showing that FGF8 is biased towards collagen type 2 loss and against growth arrest, as compared to FGF4 and FGF9. Plot is created from data in (44).

In addition to the efficacies, the dose-response curves in Figure 2A provide information on the potencies of the three ligands. Potency is the ligand concentration that elicits 50% of the maximal possible response. A ligand with high potency elicits a response at low concentrations, while a ligand with low potency elicits the same response at higher concentrations. There is a clear hierarchy of potencies, with FGF4 having the highest potency (lowest concentration required for 50% of the maximum response). The hierarchy of potencies in Figure 2A is as follows: FGF4>FGF9>FGF8.

Another well-established response of RCS cells to FGF ligands is the degradation of their abundant extracellular matrix (ECM), which consists of collagen type 2 and aggrecan (47). In (44), the expression of collagen 2 after 48 hours of treatment with FGFs at different concentrations was determined by western blot to generate the response curve shown in Figure 2B. At 5 nM, differences in the effect of the three ligands can be seen. The plateau values indicating the efficacies are also different and show a clear hierarchy: FGF4>FGF9>FGF8. Thus, FGF4 is a full agonist in the induction of extracellular matrix degradation, while FGF8 and FGF9 are partial agonists. The potency of FGF4 is clearly different from the potencies of FGF8 and FGF9, but the potencies of FGF8 and FGF9 are very similar (indistinguishable by statistical tests). Thus, the hierarchy of potency of the ligands for collagen 2 loss is FGF4>FGF8=FGF9.

Thus, we see that FGF4, FGF8 and FGF9 have very different effects on the two functional responses (growth arrest and ECM loss) in RCS cells. First, the hierarchies in ligand efficacy are different when different responses are examined. Second, the hierarchies in ligand potency are different when different responses are studied. These differences are referred to in the pharmacological literature as “quantitative differences” (48, 49).

It has long been assumed that the dose-response curves depend on the ligand-receptor dissociation constants (13, 14), which have concentration units (molar, just like EC₅₀). This is indeed the case, but the ligand-receptor dissociation constant is only one of many factors that contribute to a dose-response curve (45, 50). Methods exist to directly measure dissociation constants; some of these are listed in Table 1. The most biologically relevant measurements quantify ligand binding to full-length receptors in cells using radiolabeled or fluorescently labeled ligands (51, 52). However, the experiments in the literature generally quantify ligand binding to the extracellular (EC) domain of a receptor, either as a soluble EC domain monomer or as a dimer (53).

Table 1.

Experimental methods that can be used to measure ligand binding constants, as opposed to methods that only provide potencies and efficacies for the response being tested.

Methods that yield ligand binding constants	Methods that CANNOT yield ligand binding constants
Radioligand binding (51)	Proliferation assays (76)
Fluorescent ligand binding (52)	Motility assays (77)
Surface plasmon resonance (18, 78)	Growth arrest assays (44)
BioLayer interferometry (68)	Extracellular matrix degradation (44)
Isothermal titration calorimetry (79)	Receptor/effector proteins phosphorylation (53)
LigandTracer (80)	Mitogenic activity assays (13)

In Table 1 we list assays that cannot be used to measure ligand-receptor dissociation constants because they can only give the potency (EC₅₀) for a response, which can be orders of magnitude smaller than the ligand dissociation constant due to signal amplification. This amplification was first demonstrated in 1983 by Black and Leff in their so-called “operational model” (50), on which classical pharmacology is based. Furthermore, the EC₅₀ can be different when different responses to the same ligand are studied. Indeed, in the example given in Figure 2, the EC₅₀ of FGF8 is 1.6 nM when growth arrest is studied and 0.085 nM when type 2 collagen degradation is studied. This is a difference by a factor of 19, for the same ligand-receptor dissociation constant.

Qualitative Differences in Signaling or Biased Signaling

We can gain additional insight into differential activation of FGFR1c by FGF4, FGF8 and FGF9 by comparing the data in Figure 2A with Figure 2B. The responses to FGF8 and FGF9 are similar when loss of collagen 2 is assayed, but the two responses are very different when growth arrest is assayed. FGF8 thus acts similarly to FGF9 when it comes to triggering collagen 2 loss, but is a much weaker ligand than FGF9 when it comes to triggering growth arrest. Such a difference is referred to in the pharmacological literature as “qualitative differences” or “bias” (48, 54). By definition, “ligand bias” is the ability of different ligands to differentially activate different signaling pathways downstream of the same receptor (Figure 2C) (55). To identify the bias, we construct the bias plot in Figure 2D as a graphical representation showing one response as a function of a second response for each ligand (45). The bias plot is constructed from the dose-response data by plotting the magnitude of one response against the magnitude of the second response for each ligand concentration (54). The bias trajectories for different ligands are directly comparable in a bias plot (45).

The bias plot in Figure 2D is a visual aid to determine whether there is bias in FGFR1c signaling in response to FGF4, FGF8, and FGF9. In Figure 2D, FGF4 data points deviate toward the direction of the “growth arrest” axis, while FGF8 data points deviate toward the direction of the “collagen 2 loss” axis. We interpret this as “FGF4 is biased towards growth arrest and against collagen 2 loss” and “FGF8 is biased towards collagen 2 loss and against growth arrest”. Thus, comparison of the data in Figure 2 shows a difference in the preference of FGF ligands to induce either collagen 2 loss or growth arrest. This preference or bias can only be detected when at least two ligands and two responses are compared and dose-response curves are generated (45). Based on this comparison, we see that FGF8 has a preference for collagen 2 loss and against growth arrest compared to FGF4, despite the fact that FGF8 has the lowest potency and the lowest efficacy in inducing collagen 2 loss.

Importantly, the strength of ligand binding (ligand-receptor dissociation constant) is not related to the ligand bias, as it effectively cancels out when the bias plots are constructed. This is because both the x and y values in the bias plot depend on ligand binding. It has been reported that FGF4 binds more strongly to FGFR1c than FGF9 (56), and this fact probably contributes to the lowest EC₅₀ (the highest potency) of FGF4. However, when one response is plotted against the other, effects that contribute equally to both responses cancel out. Thus, a bias plot uncouples the effect of ligand binding strength from ligand preference for one response over another. A formal proof that the magnitude of bias does not depend on the binding coefficients can be found in (46).

The question arises as to how the bias identified in Figure 3A can arise. For GPCRs, it is well established that different GPCR ligands stabilize different receptor conformations, with each of these conformations showing a preference for binding a subset of downstream signaling molecules (either G proteins or arrestins) (57, 58). For FGFR1c, it has been hypothesized that the bias arises in the first step of signal propagation, due to differential phosphorylation of different tyrosines in FGFR1c and adaptor proteins (21). The data shown in Figure 3 below support this view.

Figure 3.

FGF-mediated phosphorylation of FGFR1 and FRS2. Measured are the phosphorylation of Y766 on FGFR1 and the phosphorylation of the adaptor FRS2, which is constitutively associated with FGFR1, in HEK 293T cells. (A) Dose-response curves for Y766 phosphorylation from western blot experiments. Data are from (44). (B) Dose-response curves for FRS2 phosphorylation from western blot experiments. Data are from (44). (C) Bias plots showing that FGF8 is biased towards pFRS2 phosphorylation and against Y766 phosphorylation. Statistical analysis shows that the preferences of FGF8 are significantly different from the preferences of FGF4 and FGF9 (44). Plot is created from data in (44).

The FGFR1 phosphorylation dose-response curves in Figure 3A were acquired in human embryonal kidney (HEK) 293T cells by western blot, while the concentrations of FGF4, FGF8 or FGF9 varied over a wide range. Several phosphorylation sites were monitored, including phosphorylation of Y766 in the kinase domain of FGFR1, which serves as a binding site for PLCγ (59), and phosphorylation of the adaptor FRS2, which is constitutively associated with FGFR1 independent of phosphorylation status (24, 60). The bias plots are shown in Figure 3B. They show that FGF8 is biased towards the phosphorylation of FRS2 and against the phosphorylation of Y766. FGF9, on the other hand, is biased in favor of phosphorylation of Y766. Thus, there is ligand bias in the first step of FGFR1c signal transduction that likely underlies the ligand bias in the functional responses in Figure 2C.

Implications

The concept of “ligand bias” is a relatively novel concept in RTK research (21). It describes the ability of ligands to differentially activate signaling pathways (34, 35, 42) and can only be determined when two ligands and two responses are directly compared (45, 49, 61). Ligand bias has been studied mainly for GPCRs (55), and these studies have led to the establishment of consensus protocols to identify and quantify the bias. The protocols, including the construction of bias plots, have been shown to be directly applicable to RTKs (46). The bias plots in Figures 2 and 3 show that FGF8 preferentially activates some downstream FGFR1 responses (FRS2 phosphorylation in HEK 293T cells and ECM loss in RCS cells), whereas FGF4 and FGF9 preferentially activate different responses (Y766 FGFR1 phosphorylation in HEK 293T cells and growth arrest in RCS cells).

Understanding the mechanisms of the different FGF-FGFR interactions is key to understanding the complex functions of the FGF system in development, where a single FGFR variant transduces extracellular signals delivered by multiple FGF ligands. In limb development, for example, FGFR1c responds to multiple FGF ligands, and the nature of the response determines the critical processes involved in limb patterning. The limb bud in the developing embryo is composed of mesenchymal cells that express FGFR1c and are enveloped by ectoderm. The proliferating mesenchymal cells form a bulge under the ectoderm. Subsequently, the ectodermal cells form the apical ectodermal ridge (AER), which defines the proximal-distal axis of the limb and controls limb growth (62, 63). FGF4, FGF8 and FGF9 are secreted by the AER, diffuse into the mesenchymal region adjacent to the AER and activate FGFR1c, promoting mesenchymal cell survival, proliferation and differentiation.

FGF4, FGF8 and FGF9 show similar spatio-temporal expression pattenrns in AER. Genetic ablation of different FGF combinations results in markedly different limb malformations, demonstrating that FGF4, FGF8 and FGF9 convey different information to FGFR1c-expressing mesenchymal cells (34, 62). Despite this, it has been proposed that “the simplest explanation” for the genetic ablation results is that the different FGFs are “functionally equivalent”, differing only in potency and concentrations (34, 62). The experiments in Figures 2 and 3 show that these three ligands are not functionally equivalent and challenge the previous paradigm. Indeed, Figures 2 and 3 show both quantitative differences in the effects induced by different FGF ligands when signaling via FGFR1c, as well as qualitative differences that represent biased signaling. In pharmacology, the qualitative differences are considered the most important as they lead to fundamentally different biological outcomes (45, 49). It is therefore very likely that the qualitative differences discussed here contribute to the differential signaling of FGF4, FGF8, and FGF9 during limb bud development, along with anticipated differences in ligand expression levels, ligand-receptor binding coefficients, and differences in ligand stability in vivo.

Eighteen FGFs deliver morphogenic signals for patterning virtually all embryonic structures (64, 65). The discovery of biased FGF signaling uncovers a layer of complexity in FGF regulation of developmental processes that has not been appreciated thus far. Further, we believe that the potential of FGFs and FGF analogs as drugs for cancer and tissue regeneration has likely been underutilized to date because the concept of bias has not been fully considered. In prior work, FGF1 and FGF2 have been used as anticancer drug carriers (66–69). Highly stable variants of FGF1 (70), FGF2 (71), and FGF21 (72) have been shown to hold promise for stem cell research and various medical and pharmaceutical applications. Constitutively dimerized FGF2 exhibits increased stability, mitogenic potential, and anti-apoptotic activity (73, 74). Cyclized or oligomerized FGF2 variants demonstrate enhanced stability, enhanced potential for wound healing, and reduced dependence on heparin (75). Additionally, a cyclized and dimerized FGF2 variant, when conjugated with highly cytotoxic anticancer agents, shows significantly increased toxicity against cancer cells overexpressing FGF receptors (75). Currently, it is not known if these novel compounds induce biased signaling as compared to natural ligands. By quantifying bias for these compounds, researchers can determine which signaling pathways are preferentially activated by them in comparison with natural ligands, allowing for more targeted therapeutic strategies. Such advances can open new avenues for the development of precise anti-cancer therapies and regenerative medicine approaches.

We are confident that future studies with other FGFRs and FGF ligands will uncover many more examples of biased FGF signaling. We look forward to exciting new discoveries in the area of FGF-mediated regulation of developmental processes in mammals, and to novel therapies in the clinic.