Skip to main content
NCBI home page
Chemical Science logoLink to Chemical Science
. 2026 Jul 22;17(37):17889–17897. doi: 10.1039/d6sc01974a

Reprogramming FGFR isoform specificity in FGF2 by deep mutational scanning

Yuga Okada a, Jumpei Morimoto a, Shinsuke Sando a,b,✉
PMCID: PMC13440468  PMID: 42559254

Abstract

The fibroblast growth factor (FGF) family exhibits distinct yet overlapping specificities toward seven fibroblast growth factor receptor (FGFR) isoforms, forming highly complex signaling networks through broad binding and activation profiles. FGF2 plays important physiological roles and has considerable therapeutic potential; however, its pleiotropic and non-selective activation of FGFRs can lead to adverse effects. Therefore, molecular engineering of FGF ligands with controlled receptor isoform specificity is required. Here, we engineered FGF2 variants with altered FGFR isoform specificity using mRNA display-based deep mutational scanning (DMS), thereby enabling systematic evaluation of residues within the FGFR-binding interface that contribute to FGFR isoform specificity despite limited structural information. We demonstrate for the first time that FGFs with altered FGFR isoform specificity can be generated. Notably, FGFR isoform specificity proved malleable and could be reprogrammed by single–amino acid substitutions. These findings reveal that receptor selectivity in FGF2 is tunable and provide a framework for rationally designing FGF ligands with tailored FGFR isoform specificity, advancing the understanding and control of complex FGF/FGFR signaling networks.

mRNA display-based DMS systematically explores FGF2/FGFR interaction interfaces and evaluates residues contributing to FGFR isoform specificity, enabling engineering of variants that shift broad FGFR binding toward isoform-selective binding.graphic file with name d6sc01974a-ga.webp

Introduction

The fibroblast growth factor (FGF) family comprises a group of critically important growth factors that play essential roles in a wide range of biological functions, including development, metabolism, and the maintenance of tissue homeostasis. In humans, 22 members of the FGF family have been identified, and their signaling is mediated through four fibroblast growth factor receptor (FGFR) tyrosine kinases, FGFR1, FGFR2, FGFR3, and FGFR4.1 FGFRs are characterized by three extracellular immunoglobulin-like (Ig) domains (IgI–III). For FGFR1, FGFR2, and FGFR3, alternative splicing within the IgIII domain gives rise to splice variants encoded by distinct exons (IIIb/IIIc), resulting in a total of seven FGFR isoforms: FGFR1b, FGFR1c, FGFR2b, FGFR2c, FGFR3b, FGFR3c, and FGFR4.2

In vivo, individual FGFs bind to and activate multiple FGFR isoforms, exhibiting distinct yet overlapping receptor specificities.3 Consequently, FGF/FGFR signaling constitutes a highly complex signaling network. While this complexity underlies the physiological diversity of FGF signaling, it also complicates the identification of the specific FGFR isoforms through which FGF-induced phenotypes are mediated.

FGF2 represents a prototypical ligand that exemplifies the challenges posed by receptor isoform diversity. FGF2 is one of the most extensively studied and functionally important members of the FGF family and is known to exhibit broad binding and activation capabilities toward FGFR1b, FGFR1c, FGFR2c, FGFR3c, and FGFR4 (Fig. 1a). Because FGF2 induces cellular proliferation and differentiation, its potential therapeutic applications have been reported across a wide range of medical contexts, including cardiovascular diseases, ulcers, wound healing, and the maintenance of pluripotency in cultures of human embryonic stem cells.4–6 In recent years, its potential application in the treatment of mood disorders has also attracted increasing attention.7 However, this pleiotropic and non-selective activation of FGFRs can simultaneously activate FGFR signaling pathways that are not directly related to the intended therapeutic effect, thereby leading to adverse side effects.8–10 Indeed, aberrant activation of FGF/FGFR signaling has been implicated in the onset and progression of various diseases. For example, ectopic expression of the mesenchyme-associated IIIc isoform of FGFR2 (FGFR2c) in epithelial tissues alters ligand responsiveness and promotes tumor invasion and malignancy.11,12 These observations highlight that FGF ligands, including FGF2, can drive disease-associated signaling through specific FGFR isoforms.

Fig. 1. (a) Conceptual illustration of the FGFR isoform specificity of FGF2. FGF2 is reported to bind to and activate multiple FGFR isoforms, including FGFR1b, 1c, 2c, 3c, and 4. (b) Conceptual illustration of the expected therapeutic potential of FGFs with engineered FGFR isoform specificity. Native FGF ligands can activate multiple FGFR isoforms, whereas FGFs with altered FGFR isoform specificity may enable selective modulation of therapeutically relevant FGFR signaling while reducing activation of FGFR isoforms associated with side effects. HS: Heparan sulfate.

Fig. 1

Open in a new tab

Against this background, precise control of FGFR specificity of FGF ligands, enabling the selective activation or avoidance of particular receptor signaling pathways, is strongly desired for the therapeutic application of FGFs. In particular, the rational design of FGFs with altered FGFR isoform specificity, including enhanced binding selectivity toward therapeutically relevant FGFR isoforms and reduced binding to FGFR isoforms associated with adverse effects, is expected to provide a basis for future strategies aimed at improving therapeutic efficacy while reducing adverse effects (Fig. 1b).

To date, several studies have demonstrated successful modulation of FGF signaling through molecular engineering of FGFs. For example, by focusing on affinity for the co-receptor heparan sulfate (HS), a mutant FGF1 with reduced HS-binding ability (FGF1ΔHBS) retained metabolic activity while exhibiting diminished mitogenic activity.13 Similarly, engineering of FGF19 mutants with reduced interaction with βKlotho, another co-receptor, successfully uncoupled metabolic and proliferative activities.14 However, these approaches have been limited to functional dissociation of signaling downstream of a single FGFR, and no studies have yet reported direct modification of FGFR isoform specificity itself with the aim of discriminating signaling among multiple FGFR isoforms.

The primary factor hindering the rational design of isoform-specific FGFs lies in the extremely limited availability of structural information on FGF/FGFR complexes. Although individual studies have discussed the structures and interactions of specific FGF/FGFR pairs,15–18 comprehensive complex structures encompassing all seven FGFR isoforms have not been determined. As a result, comprehensive information on residues within FGFs that contribute to FGFR specificity remains unavailable, severely constraining efforts to rationally engineer FGFs with altered receptor specificity.

In this study, we therefore focused on FGF2 and aimed to systematically identify residues within the FGFR-binding interface that contribute to FGFR isoform specificity. To overcome the limitation of scarce structural information, we employed deep mutational scanning (DMS) based on mRNA display,19,20 a high-throughput approach with minimal dependence on prior structural information, to systematically analyze the interactions between FGF2 and each FGFR isoform.21–23 Through this analysis, we comprehensively elucidated the residues in FGF2 that contribute to FGFR isoform specificity and demonstrated, for the first time, the generation of FGFs with engineered FGFR isoform specificity.

Results and discussion

Workflow for the DMS-based profiling and engineering of FGF2/FGFR isoform specificity

In this study, we focused on FGF2 and systematically evaluated residues within the FGFR-binding interface that contribute to FGFR isoform specificity using mRNA display-based DMS. On the basis of the obtained data, we designed and validated FGF2 variants with altered FGFR isoform specificity. Fig. 2 shows a schematic overview of the experimental workflow of this study. First, FGF2 residues involved in interactions with FGFRs were selected to construct a mutagenesis library, and parallel affinity selection was performed against seven FGFR isoforms to obtain comprehensive affinity data between FGF2 mutants and each FGFR isoform. Next, FGF2 mutants with altered specificity were designed and evaluated based on these data. Finally, for mutants exhibiting pronounced alterations in specificity, structural factors were examined using predicted complex structures.

Fig. 2. Schematic overview of the workflow in this study. A DNA library for DMS analysis was constructed by introducing mutations at FGF2 positions predicted to be involved in interactions with FGFR isoforms. mRNA display was then performed. The DNA library was transcribed into mRNA, and a puromycin linker was ligated to the 3′ end of each transcript. In vitro translation was carried out, resulting in the covalent linkage of the translated FGF2 mutants to their corresponding mRNA via puromycin. After reverse transcription, FGF2–mRNA : cDNA complexes were subjected to parallel affinity selection against individual FGFR isoforms. Recovered DNA was analyzed by NGS, yielding DMS-derived enrichment landscapes of FGF2 mutants toward each FGFR isoform. These data were subsequently used to design, evaluate, and analyze FGF2 variants with altered FGFR isoform specificity.

Fig. 2

Open in a new tab

Library design for DMS analysis

A library of FGF2 variants suitable for DMS analysis was designed. To systematically evaluate residues contributing to FGFR isoform specificity within the FGFR-binding interface, it is necessary to construct a library that includes mutants affecting interactions with each FGFR isoform. However, for FGF2, complex structures have been reported only with FGFR1c (PDB IDs: 1CVS and 1FQ9) and FGFR2c (PDB ID: 1EV2),15,16,24 and interaction sites between FGF2 and other FGFR isoforms have not been experimentally elucidated. Therefore, for complex structures that have not been previously reported, predicted structures generated by AlphaFold3 (ref. 25) were used as substitutes to improve the accuracy of interaction site estimation (Fig. 3a and S1). In the AlphaFold3 predictions, the complex structure of FGF2/FGFR3b exhibited a markedly bent conformation of FGFR3b that differed from those of other FGFRs, and was therefore judged not to represent an appropriate complex structure. Because FGFR3b is known to lack specificity for FGF2, this result suggested that the prediction appropriately reflected this property. In contrast, for complexes with the other FGFR isoforms, reasonable complex structures with clearly formed ligand–receptor interfaces were obtained.

Fig. 3. (a) Structures of FGF2 in complex with FGFR1b and FGFR1c. The FGF2/FGFR1b complex represents a structure predicted by AlphaFold3 (iPTM = 0.88), whereas the FGF2/FGFR1c complex corresponds to a crystal structure (PDB ID: 1CVS). The lower panels show a magnified view of the FGF2/FGFR1c interaction interface and the same view rotated by 180°. Complex structures of FGF2 with all FGFR isoforms, either experimentally determined or predicted, are provided in Fig. S1. (b) Amino acid sequence of the full-length FGF2. At each mutation site, the designed library contains 19 single amino acid substitutions and a single amino acid deletion, with one mutation introduced per position. (a and b) Residues highlighted in red on FGF2 indicate sites selected for mutation in the designed library.

Fig. 3

Open in a new tab

Based on the literature15,26 and on residues within 5 Å of each FGFR in the predicted complex structures, candidate residues were selected. For FGFR1c, because a dimeric structure was available, residues within 5 Å of FGFR1c in the dimerized state were also included as candidates. From these candidates, residues reported to be involved in interactions with HS were excluded, resulting in the final selection of 48 residues (shown in red in Fig. 3a). This design was intended to generate a targeted, interface-focused library that broadly covers candidate residues potentially involved in FGFR isoform recognition, based on multiple lines of evidence from structural information and previous literature. For each selected position, the library contained 20 variants, comprising 19 single-amino acid substitutions and one single-residue deletion, with each variant carrying a single mutation (Fig. 3b).

DMS analysis based on mRNA display

mRNA display was performed using the designed FGF2 library. Parallel affinity selection was carried out using seven FGFR isoforms (Fig. 4a). The recovery rate, defined as the ratio of the amount of DNA recovered after selection to the amount of DNA in the library before selection, was quantified by qPCR. As a result, sufficient recovery rates were confirmed for all FGFR isoforms except FGFR3b (Fig. 4b). For FGFR3b, almost no DNA was recovered; however, a similar result was obtained in mRNA display using a single DNA encoding FGF2 WT instead of the mutant library, indicating that this outcome reflected the binding specificity of FGF2 for FGFR3b.

Fig. 4. (a) Schematic overview of the workflow from affinity selection to NGS analysis. DNA recovered from parallel affinity selection against individual FGFR isoforms was processed into sequencing libraries independently, during which unique molecular identifiers (UMIs) were introduced. Immediately prior to NGS analysis, libraries corresponding to each FGFR isoform were pooled at equal mass. In parallel, the pre-selection library was prepared and pooled in the same manner and subjected to NGS analysis. (b) Recovery rates of the designed FGF2 library following affinity selection against FGFR isoforms. Recovery rates were determined by quantifying DNA amounts before and after selection using qPCR. (c) Results of NGS analysis for FGFR1b and FGFR1c. Values represent DNA copy numbers recovered after selection, normalized to both the pre-selection library and the WT control by dividing by their respective values. Accordingly, higher values indicate increased affinity for the corresponding FGFR relative to WT. The heatmaps for the other FGFR isoforms are provided in Fig. S2. The same color scale, including the same minimum and maximum values, was applied to all heatmaps in (c) and S2, enabling direct comparison across FGF2 mutants and FGFR isoforms. (d) Correlation analysis of NGS results among FGFR isoforms. Correlation coefficients were calculated to assess similarities in the affinity profiles obtained for each FGFR isoform.

Fig. 4

Open in a new tab

The recovered libraries were analyzed by next-generation sequencing (NGS). To reduce PCR amplification bias during library preparation and analysis, unique molecular identifiers (UMIs) were introduced prior to NGS analysis. This enabled quantitative evaluation based on the number of DNA copies at the time of recovery rather than read counts. In addition, DNA recovered for each FGFR isoform was independently processed into sequencing libraries and mixed immediately prior to NGS analysis, thereby equalizing input amounts for NGS while preserving the relative composition of mutants under each FGFR isoform condition. Copy numbers of individual mutants were normalized to the pre-selection library and to WT to correct for differences in initial abundance and experimental conditions (Fig. 4c and S2).

Fig. 4c shows the results for FGFR1b and FGFR1c. The vertical axis indicates the mutated residues, and the horizontal axis indicates the introduced mutations. Values greater than 1.0 are shown in red, whereas values less than 1.0 are shown in blue, with higher values indicating higher affinity of the mutant for the target FGFR relative to WT. Mutations with values of 1.0 are shown in white, indicating no change and affinity equivalent to that of WT. Examination of the results shows that the introduced mutations caused changes in affinity, and that the trends differed depending on the residue. Results for the other isoforms are shown in Fig. S2.

To validate the reliability of the obtained results, we focused on residue E105, which is highly conserved within the FGF family and has been reported to be important for interactions with FGFRs.27 Analysis of DMS data for mutants carrying substitutions at E105 showed that, across all FGFR isoforms, mutations other than substitution with aspartic acid, which is structurally similar to glutamic acid, resulted in reduced affinity compared with WT. Furthermore, more than 90% of the introduced mutations exhibited values of 0.5 or lower, indicating pronounced reduction in interactions relative to WT. These results indicate that E105 plays an important role in FGF2/FGFR interactions irrespective of the FGFR isoform and, consistent with previously reported findings, support the validity of the analytical approach and the results obtained.

Several important trends in the receptor-binding properties of FGF2 were revealed. Despite comprehensive mutagenesis of the FGFR-binding interface, fewer than 4% of mutations exhibited values of 2.0 or higher, indicating enhanced interactions relative to WT for any FGFR. This suggests that FGF2 is already highly optimized for interactions with FGFRs, making further enhancement of affinity difficult. In addition, correlation analysis of relative affinity changes of FGF2 mutants across FGFR isoforms showed high correlations (Fig. 4d). Even for the combination of FGFR1b and FGFR2c, which showed the lowest correlation, the correlation coefficient was 0.66. Particularly strong positive correlations, with correlation coefficients greater than 0.9, were observed among FGFR1b, FGFR2b, and FGFR3c, as well as among FGFR1c, FGFR2c, and FGFR4. These results suggest that FGFR specificity of FGF2 depends more strongly on differences in the splicing variant regions of the IgIII domain than on differences among FGFR1–4 as receptor types. Taken together, these findings indicate that the data obtained in this study appropriately reflect changes in interactions between FGF2 mutants and each FGFR isoform, and provide important insights into residues contributing to FGFR isoform specificity.

Design and evaluation of FGF2 variants with altered FGFR specificity

Based on comprehensive interaction profiles between FGF2 mutants and individual FGFR isoforms obtained by DMS analysis, mutants with enhanced specificity toward each FGFR isoform were identified. Candidate mutations were narrowed down on the basis of the following three criteria: (1) the value for the target FGFR is higher than those for the other FGFRs; (2) the value for the target FGFR is not markedly reduced; and (3) structural stability is maintained. For criterion (1), candidate mutations were identified not only based on the DMS data for the target FGFR, but also through comparison with the DMS data for the other FGFRs, such that the selected mutations showed a relatively higher value for the target FGFR than for the other FGFRs. This criterion was intended to enhance selectivity toward the target FGFR. Criterion (2) was included to avoid a substantial decrease in affinity for the target FGFR. However, from the viewpoint of selectivity, mutations were not excluded solely because their values for the target FGFR were reduced relative to WT, provided that they showed relatively higher values for the target FGFR than for the other FGFRs and that the decrease for the target FGFR was not substantial. Criterion (3) was intended to exclude effects arising from reduced activity due to structural destabilization. For evaluation of criterion (3), an in silico approach based on Cartesian_ddG,28 a Rosetta-based modeling method, was employed (Fig. S3). Using the crystal structure of FGF2 (PDB ID: 2FGF) as the initial structure, ΔΔG values for each mutation were estimated, and mutations with ΔΔG values of 1.5 kcal mol−1 or lower were selected.

For each FGFR isoform, mutations satisfying all three criteria were selected as candidates and identified as mutants with enhanced specificity toward the target FGFR. When multiple candidates were obtained, the mutation exhibiting the largest difference between the target FGFR and the other FGFRs under criterion (1) was preferentially selected. As a result, six mutants (K55A, L107P, Y112W, N110D, R69W, and L149Q) were identified as the final candidates for FGFR1b, 1c, 2b, 2c, 3c, and 4, respectively.

To verify whether the identified mutants exhibited enhanced FGFR specificity as intended, individual DNA sequences were prepared for each mutant, and affinity selection against each FGFR isoform was performed by mRNA display. Comparison of recovery rates for each mutant showed that, despite the introduction of only a single mutation, clear differences in recovery trends among FGFR isoforms were observed for most mutants (Fig. S4a). Furthermore, changes in binding selectivity toward each FGFR isoform relative to WT were quantitatively evaluated according to eqn (1). In this equation, R represents the recovery rate, and N denotes the number of FGFR isoforms analyzed. Specifically, for a given FGFR isoform x (FGFRx), specificity was defined as the recovery rate for FGFRx divided by the geometric mean of the recovery rates for the remaining FGFR isoforms, thereby representing the extent to which recovery for FGFRx is favored relative to other FGFRs. This value was further normalized to the corresponding value for WT, allowing relative changes in receptor selectivity to be quantified.

graphic file with name d6sc01974a-t1.webp 1

x = 1b, 1c, 2b, 2c, 3c, and 4.

As a result, with the exception of K55A targeting FGFR1b, enhanced specificity toward the target FGFR was confirmed for all mutants, although to varying degrees (Fig. 5a). In particular, the FGFR1c-targeting mutant L107P and the FGFR4-targeting mutant L149Q exhibited approximately 50- and 100-fold increases in relative selectivity toward FGFR1c and FGFR4, respectively, compared with WT. In contrast, binding to the non-target FGFR isoforms showed little increase relative to WT, indicating a pronounced enhancement of specificity toward the target FGFR. This improvement in specificity appears to be driven primarily by a reduction in affinity for non-target FGFRs rather than by an increase in affinity for the target FGFR. This interpretation is consistent with the fact that FGF2 WT already exhibits high affinity for the target FGFR. N110D exhibited a distinctive behavior, showing a tendency for selective binding to FGFR1c, FGFR2c, and FGFR4, consistent with the similarity among these receptors suggested by the correlation analysis. By contrast, for Y112W and R69W, which are mutants targeting FGFR2b and FGFR3c, respectively, although an increase in relative selectivity toward the target FGFR was observed, the magnitude of the increase was relatively small and the differences from other FGFRs were limited. This may be attributable to the intrinsically low affinity of FGF2 for FGFR1b, FGFR2b, and FGFR3c, which limits the extent to which selectivity differences can be generated.

Fig. 5. (a and b) Relative FGFR specificity of the identified FGF2 mutants compared to WT. Single mutants are shown in (a), whereas the double mutant E67K/L149Q is shown in (b). In (a), magnified views of K55A, Y112W, N110D, and R69W are shown in the lower left panel. Ligand specificity for each FGFR was calculated by dividing the recovery rate obtained from mRNA display by the geometric mean of the recovery rates for the remaining FGFR isoforms. Changes in FGFR specificity for each mutant were then determined by normalizing these values to the corresponding WT value (see eqn (1)). For mutants with recovery rates below the detection limit, detection-limit value was used for calculations. (c and d) Evaluation of binding affinities between WT or mutants and FGFR isoforms by ELISA. Dot plots represent measured absorbance values, and line plots indicate curves fitted using a one-site binding (Langmuir) model assuming 1 : 1 ligand–receptor interactions. Experiments were performed in triplicate, and error bars represent SD of technical replicates. (c) Binding assays using WT, L107P, and E67K/L149Q as ligands and FGFR1c, FGFR2b, FGFR2c, and FGFR4 as receptors. WT/FGFR1c and L107P/FGFR1c were analyzed at ligand concentrations of 0.1–100 nM, whereas all other ligand–receptor combinations were analyzed at 0.1–1000 nM. (d) Binding assays using FGF2 WT, L149Q, and E67K/L149Q as ligands and FGFR1c and FGFR4 as receptors. FGF2 WT was analyzed at concentrations of 0.1–100 nM, whereas mutant ligands were additionally analyzed at 300 and 1000 nM. (e) Apparent dissociation constants (Kd) determined by ELISA binding assays. Kd values were obtained by fitting binding curves using the Langmuir model and are reported as Kd ± SE, where SE represents the standard error of the fitted Kd. N.D., not determined due to lack of saturation.

Fig. 5

Open in a new tab

To demonstrate that the DMS-derived data obtained in this study can be utilized for the rational modification of FGFR specificity, FGF2 variants with further enhanced specificity toward the target FGFR were designed. Attention was focused on the FGFR4-targeting mutant L149Q, which exhibited relatively high specificity while retaining residual binding to FGFR1c. Based on the selection criteria described above, candidate mutations were re-evaluated, and a double mutant (E67K/L149Q) was designed by introducing the E67K mutation, which was identified as the next most promising candidate after L149Q. Evaluation of this mutant by mRNA display revealed that, compared with the L149Q single mutant, selectivity toward FGFR1c was reduced, while a further enhancement of selectivity toward the target FGFR4 was suggested (Fig. 5b and S4b). Taken together, these results suggest that the DMS-derived interaction profiles obtained in this study can serve as a guideline for the rational design of variants with modified FGFR specificity of FGF2.

Interaction evaluation by mRNA display is performed with translated proteins bound to RNA, potentially leading to behavior distinct from actual protein–protein interactions. To confirm that FGFR specificity changes were not an artifact of mRNA display, we validated the results using an independent assay. Specifically, L107P, L149Q and the double mutant E67K/L149Q, which exhibited especially high selectivity, were individually expressed and purified as proteins (Fig. S5), and their affinities toward four representative FGFR isoforms with distinct interaction profiles (FGFR1c, 2b, 2c, and 4) were evaluated by ELISA (Fig. 5c–e and S6).

FGF2 WT was confirmed to bind to all FGFRs with affinities in the nanomolar range. L107P exhibited a dissociation constant (Kd) toward the target FGFR1c comparable to that of WT, whereas its affinities toward the non-target FGFR2b and FGFR2c were markedly reduced. This indicates that binding to FGFR1c was maintained while interactions with non-target FGFRs were suppressed.

For the double mutant E67K/L149Q, affinities toward the non-target FGFR2b and FGFR2c were reduced. Although affinity toward FGFR1c was also decreased compared with WT, the extent of this decrease was less pronounced than that observed by mRNA display (Fig. S4b). In addition, affinity toward FGFR4 was reduced relative to WT. Comparison with the single mutant L149Q showed that the Kd values toward FGFR4 were nearly identical between the two mutants, whereas the Kd toward FGFR1c was approximately three- to four-fold higher for E67K/L149Q. Thus, introduction of the E67K mutation resulted in selective attenuation of FGFR1c binding while maintaining FGFR4 binding.

Taken together, these results demonstrate that trends in FGFR specificity modification were reproduced using an evaluation system distinct from mRNA display, supporting that the DMS data obtained in this study reflect changes in actual receptor–ligand interactions. Although quantitative differences in the magnitude of specificity changes were observed between the evaluation systems, the two datasets provide complementary information that captures interaction properties under different experimental conditions. They serve as practical indicators for the direction of receptor selectivity changes and for guiding variant design.

Analysis of structural factors underlying FGFR specificity modulation

The double mutant E67K/L149Q, in which a further modification of FGFR4 specificity was observed through a combination of single mutations, was selected for structural analysis. The structural basis underlying this behavior was examined using predicted complex structures generated by AlphaFold3 (Fig. 6 and S7). Comparison of the environment surrounding L149 revealed that FGFR2b and FGFR2c, for which affinity was markedly reduced, contain hydrophobic residues such as valine, proline, and alanine, forming a hydrophobic environment. Loss of hydrophobic interactions resulting from the substitution of leucine with glutamine, therefore, likely contributes to the decreased affinity. In FGFR4, the environment around L149 is also hydrophobic, which is consistent with the more than ten-fold reduction in affinity observed in Kd measurements. By contrast, the relatively attenuated decrease in affinity toward FGFR4 compared with FGFR2b and FGFR2c may be attributable to the presence of leucine, rather than valine, at the corresponding position in FGFR4. Unlike valine, leucine has greater side-chain flexibility, which may allow partial compensation for the loss of hydrophobic interactions through adjustment of side-chain orientation.

Fig. 6. Predicted models of FGF2 WT and the E67K/L149Q mutant in complex with FGFR isoforms. Structures were generated using AlphaFold3. The left inset shows a magnified view of the L149Q region, and the right inset shows a magnified view of the E67K region. Putative hydrogen bonds are indicated by light blue dashed lines, and mutation sites are highlighted in red. FGF2 WT is shown in sky blue, E67K/L149Q in deep purple, FGFR1c in dark blue, FGFR2c in dark green, and FGFR4 in purple. iPTM scores are provided for the predicted complexes (WT–FGFR4: 0.85; E67K/L149Q–FGFR1c: 0.84; FGFR2c: 0.86; FGFR4: 0.85).

Fig. 6

Open in a new tab

In addition, for the E67K/L149Q mutant, affinity toward FGFR1c was reduced compared with that of the L149Q single mutant upon introduction of the E67K mutation. This reduction may be due to disruption of a hydrogen bond formed in the wild type between E67 and the backbone of V316 in FGFR1c, which is lost upon substitution with lysine. Taken together, these results suggest that FGFR specificity can be tuned in a stepwise manner by combining the contributions of individual mutations.

Conclusions

In this study, by targeting FGF2 and applying mRNA display–based DMS, we obtained a broad, interface-focused map of residues involved in FGFR isoform binding affinity and successfully generated FGF2 variants with altered FGFR isoform specificity for the first time. This study demonstrates that FGFR specificity can be substantially altered by single amino acid substitutions even among FGFR isoforms with high sequence similarity, highlighting that the specificity of FGF/FGFR interactions may be finely regulated by a limited number of residues. Furthermore, because FGFR specificity was enhanced in a stepwise manner by the addition of mutations, it is suggested that FGF variants with rationally modified FGFR isoform specificity can be designed by combining multiple mutations.

In addition, for variants in which specificity toward individual FGFR isoforms was altered, the structural factors underlying these changes could be rationally explained using complex structures predicted by AlphaFold3. These results indicate that the combined use of comprehensive mutational data obtained by DMS and structural prediction information generated by AlphaFold3 provides a powerful framework for understanding the molecular basis of FGFR specificity modification even under conditions where experimental complex structure information is limited, and further suggest the potential applicability of this approach to future rational specificity-modification design.

Several limitations should be noted. Given the high sequence and structural similarity among FGFR isoforms and the limited availability of experimental complex structures, AlphaFold3-derived complex structures should be regarded not as definitive evidence of stable complex formation, but as hypothesis-generating models for structural interpretation. Although some of the mutants designed based on the obtained DMS data exhibited high specificity, others showed only limited improvements in specificity. In particular, it was suggested that the acquisition of mutants that clearly discriminate among FGFR1b, FGFR2b, and FGFR3c proved challenging, indicating that the difficulty in specificity modification may differ depending on the target FGFR isoform. Moreover, because the present study primarily focused on binding affinity in the design, further investigation will be required to evaluate receptor specificity control at the level of functional activity.

Nevertheless, this study demonstrates, for the first time, that FGF2 variants with altered FGFR isoform binding specificity can be practically designed. Importantly, the results establish that the FGFR isoform-binding selectivity of FGF2 is tunable and can be reprogrammed even through single amino acid substitutions. Because this approach is applicable not only to FGF2 but also to other members of the FGF family, it provides a basis for understanding and controlling complex FGF/FGFR signaling networks and for designing FGF ligands with tailored FGFR isoform specificity as a basis for future therapeutic development. Furthermore, the approaches described here may extend to the design of molecules that modulate interactions not only with receptors but also with co-receptors,29 as well as to the rational design of synthetic FGFR agonists, such as peptides and aptamers, whose specificity can be controlled independently of natural ligands.30–34

Author contributions

Y. O. and S. S. contributed to conceptualization; Y. O., J. M., and S. S. contributed to methodology; Y. O. performed investigation, formal analysis, validation, and visualization with support from J. M. and S. S.; S. S. provided resources, supervision, and funding acquisition; Y. O. and S. S. wrote the original draft; all authors contributed to writing – review and editing and approved the final manuscript.

Conflicts of interest

There are no conflicts to declare.

Supplementary Material

SC-017-D6SC01974A-s001
SC-017-D6SC01974A-s002
SC-017-D6SC01974A-s003

Acknowledgments

The authors would like to thank Dr K. Tsuboyama for assistance with NGS data analysis. The computations were performed using Research Center for Computational Science, Okazaki, Japan (Project: 21-IMS-C078). Molecular graphics and analyses were performed with UCSF ChimeraX,35–37 developed by the Resource for Biocomputing, Visualization, and Informatics at the University of California, San Francisco. This work was supported in part by JSPS KAKENHI (Grant Number 25KJ0937) to Y.O., by the grants from the Asahi Glass Foundation and in part by JST-CREST (Grant number JPMJCR21N5) for S.S.

Data availability

The data supporting this article have been included as part of the supplementary information (SI).

Additional datasets generated during the current study are available in the UTokyo Repository at https://doi.org/10.15083/0002014301.

Supplementary information: the processed NGS data, including sequence information and read counts for each mutant, are provided as a CSV file. See DOI: https://doi.org/10.1039/d6sc01974a.

Notes and references

  1. Ornitz D. M. Itoh N. Wiley Interdiscip. Rev.: Dev. Biol. 2015;4:215–266. doi: 10.1002/wdev.176. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Gong S. G. J. Cell. Physiol. 2014;229:1887–1895. doi: 10.1002/jcp.24649. [DOI] [PubMed] [Google Scholar]
  3. Ornitz D. M. Xu J. Colvin J. S. McEwen D. G. MacArthur C. A. Coulier F. Gao G. Goldfarb M. J. Biol. Chem. 1996;271:15292–15297. doi: 10.1074/jbc.271.25.15292. [DOI] [PubMed] [Google Scholar]
  4. Seo J. H. Yu J. H. Suh H. Kim M. S. Cho S. R. PLoS One. 2013;8:e74405. doi: 10.1371/journal.pone.0074405. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Sun B. K. Siprashvili Z. Khavari P. A. Science. 2014;346:941–945. doi: 10.1126/science.1253836. [DOI] [PubMed] [Google Scholar]
  6. Levenstein M. E. Ludwig T. E. Xu R. H. Llanas R. A. VanDenHeuvel-Kramer K. Manning D. Thomson J. A. Stem Cells. 2006;24:568–574. doi: 10.1634/stemcells.2005-0247. [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Perez J. A. Clinton S. M. Turner C. A. Watson S. J. Akil H. J. Neurosci. 2009;29:6379–6387. doi: 10.1523/JNEUROSCI.4829-08.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Akl M. R. Nagpal P. Ayoub N. M. Tai B. Prabhu S. A. Capac C. M. Gliksman M. Goy A. Suh K. S. Oncotarget. 2016;7:44735–44762. doi: 10.18632/oncotarget.8203. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Tan Y. Qiao Y. Chen Z. Liu J. Guo Y. Tran T. Tan K. S. Wang D. Y. Yan Y. Front. Cell Dev. Biol. 2020;8:223. doi: 10.3389/fcell.2020.00223. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Zittermann S. I. Issekutz A. C. Am. J. Pathol. 2006;168:835–846. doi: 10.2353/ajpath.2006.050479. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Ranieri D. Rosato B. Nanni M. Magenta A. Belleudi F. Torrisi M. R. Oncotarget. 2016;7:5440–5460. doi: 10.18632/oncotarget.6706. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Ranieri D. Rosato B. Nanni M. Belleudi F. Torrisi M. R. Mol. Carcinog. 2018;57:272–283. doi: 10.1002/mc.22754. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Huang Z. Tan Y. Gu J. Liu Y. Song L. Niu J. Zhao L. Srinivasan L. Lin Q. Deng J. Li Y. Conklin D. J. Neubert T. A. Cai L. Li X. Mohammadi M. Cell Rep. 2017;20:1717–1728. doi: 10.1016/j.celrep.2017.06.063. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Niu J. Zhao J. Wu J. Qiao G. Gu J. Zhou C. Li Q. Ying L. Wang D. Lin H. Li X. Mohammadi M. Huang Z. Proc. Natl. Acad. Sci. U. S. A. 2020;117:29025–29034. doi: 10.1073/pnas.2010984117. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Plotnikov A. N. Schlessinger J. Hubbard S. R. Mohammadi M. Cell. 1999;98:641–650. doi: 10.1016/S0092-8674(00)80051-3. [DOI] [PubMed] [Google Scholar]
  16. Plotnikov A. N. Hubbard S. R. Schlessinger J. Mohammadi M. Cell. 2000;101:413–424. doi: 10.1016/S0092-8674(00)80851-X. [DOI] [PubMed] [Google Scholar]
  17. Pellegrini L. Burke D. F. von Delft F. Mulloy B. Blundell T. L. Nature. 2000;407:1029–1034. doi: 10.1038/35039551. [DOI] [PubMed] [Google Scholar]
  18. Liu Y. Ma J. Beenken A. Srinivasan L. Eliseenkova A. V. Mohammadi M. Structure. 2017;25:1325–1336.e3. doi: 10.1016/j.str.2017.06.016. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Nemoto N. Miyamoto-Sato E. Husimi Y. Yanagawa H. FEBS Lett. 1997;414:405–408. doi: 10.1016/S0014-5793(97)01026-0. [DOI] [PubMed] [Google Scholar]
  20. Roberts R. W. Szostak J. W. Proc. Natl. Acad. Sci. U. S. A. 1997;94:12297–12302. doi: 10.1073/pnas.94.23.12297. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Forsyth C. M. Juan V. Akamatsu Y. DuBridge R. B. Doan M. Ivanov A. V. Ma Z. Polakoff D. Razo J. Wilson K. Powers D. B. mAbs. 2013;5:523–532. doi: 10.4161/mabs.24979. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Rogers J. M. Passioura T. Suga H. Proc. Natl. Acad. Sci. U. S. A. 2018;115:10959–10964. doi: 10.1073/pnas.1809901115. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Benz C. Maassen L. Simonetti L. Mihalic F. Lindqvist R. Tsitsa I. Konstantinou A. Jemth P. Överby A. K. Davey N. E. Ivarsson Y. Protein Sci. 2025;34:e70174. doi: 10.1002/pro.70174. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Schlessinger J. Plotnikov A. N. Ibrahimi O. A. Eliseenkova A. V. Yeh B. K. Yayon A. Linhardt R. J. Mohammadi M. Mol. Cell. 2000;6:743–750. doi: 10.1016/S1097-2765(00)00073-3. [DOI] [PubMed] [Google Scholar]
  25. Abramson J. Adler J. Dunger J. Evans R. Green T. Pritzel A. Ronneberger O. Willmore L. Ballard A. J. Bambrick J. Bodenstein S. W. Evans D. A. Hung C. C. O'Neill M. Reiman D. Tunyasuvunakool K. Wu Z. Žemgulytė A. Arvaniti E. Beattie C. Bertolli O. Bridgland A. Cherepanov A. Congreve M. Cowen-Rivers A. I. Cowie A. Figurnov M. Fuchs F. B. Gladman H. Jain R. Khan Y. A. Low C. M. R. Perlin K. Potapenko A. Savy P. Singh S. Stecula A. Thillaisundaram A. Tong C. Yakneen S. Zhong E. D. Zielinski M. Žídek A. Bapst V. Kohli P. Jaderberg M. Hassabis D. Jumper J. M. Nature. 2024;630:493–500. doi: 10.1038/s41586-024-07487-w. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Springer B. A. Pantoliano M. W. Barbera F. A. Gunyuzlu P. L. Thompson L. D. Herblin W. F. Rosenfeld S. A. Book G. W. J. Biol. Chem. 1994;269:26879–26884. doi: 10.1016/S0021-9258(18)47101-3. [DOI] [PubMed] [Google Scholar]
  27. Zhu H. Ramnarayan K. Anchin J. Miao W. Y. Sereno A. Millman L. Zheng J. Balaji V. N. Wolff M. E. J. Biol. Chem. 1995;270:21869–21874. doi: 10.1074/jbc.270.37.21869. [DOI] [PubMed] [Google Scholar]
  28. Frenz B. Lewis S. M. King I. DiMaio F. Park H. Song Y. Front. Bioeng. Biotechnol. 2020;8:558247. doi: 10.3389/fbioe.2020.558247. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Okada Y. Eguchi A. Kuroda D. Tsumoto K. Ueki R. Sando S. ChemBioChem. 2025;26:e202500353. doi: 10.1002/cbic.202500353. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Park J. S. Choi J. Cao L. Mohanty J. Suzuki Y. Park A. Baker D. Schlessinger J. Lee S. Cell Rep. 2022;41:111545. doi: 10.1016/j.celrep.2022.111545. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Ueki R. Atsuta S. Ueki A. Hoshiyama J. Li J. Hayashi Y. Sando S. Chem. Commun. 2019;55:2672–2675. doi: 10.1039/C8CC08080A. [DOI] [PubMed] [Google Scholar]
  32. Hoshiyama J. Okada Y. Hayata Y. Eguchi A. Ueki R. Sando S. Chem. Lett. 2021;50:1949–1952. doi: 10.1246/cl.210505. [DOI] [Google Scholar]
  33. Hoshiyama J. Hayata Y. Eguchi A. Morimoto J. Ueki R. Sando S. Anal. Sci. 2024;40:2251–2258. doi: 10.1007/s44211-024-00660-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Eguchi A. Ueki A. Hoshiyama J. Kuwata K. Chikaoka Y. Kawamura T. Nagatoishi S. Tsumoto K. Ueki R. Sando S. JACS Au. 2021;1:578–585. doi: 10.1021/jacsau.0c00121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Goddard T. D. Huang C. C. Meng E. C. Pettersen E. F. Couch G. S. Morris J. H. Ferrin T. E. Protein Sci. 2018;27:14–25. doi: 10.1002/pro.3235. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Pettersen E. F. Goddard T. D. Huang C. C. Meng E. C. Couch G. S. Croll T. I. Morris J. H. Ferrin T. E. Protein Sci. 2021;30:70–82. doi: 10.1002/pro.3943. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Meng E. C. Goddard T. D. Pettersen E. F. Couch G. S. Pearson Z. J. Morris J. H. Ferrin T. E. Protein Sci. 2023;32:e4792. doi: 10.1002/pro.4792. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

SC-017-D6SC01974A-s001
SC-017-D6SC01974A-s001.xlsx (97.4KB, xlsx)
SC-017-D6SC01974A-s002
SC-017-D6SC01974A-s002.pdf (3MB, pdf)
SC-017-D6SC01974A-s003
SC-017-D6SC01974A-s003.csv (987.6KB, csv)

Data Availability Statement

The data supporting this article have been included as part of the supplementary information (SI).

Additional datasets generated during the current study are available in the UTokyo Repository at https://doi.org/10.15083/0002014301.

Supplementary information: the processed NGS data, including sequence information and read counts for each mutant, are provided as a CSV file. See DOI: https://doi.org/10.1039/d6sc01974a.

Articles from Chemical Science are provided here courtesy of Royal Society of Chemistry

RESOURCES