Design and optimization of a kinase-controlled allosteric switch

Qinhao Cao; Jared E Toettcher

doi:10.1038/s41592-026-03163-1

. 2026 Jul 21;23(8):1540–1552. doi: 10.1038/s41592-026-03163-1

Design and optimization of a kinase-controlled allosteric switch

Qinhao Cao ¹, Jared E Toettcher ^1,^2,^✉

PMCID: PMC13442002 NIHMSID: NIHMS2198701 PMID: 42481857

Abstract

Post-translational control enables rapid and precise regulation of cell behavior. Despite these advantages, general strategies to build phosphorylation-based synthetic circuits are limited. Here we reasoned that engineered allostery, a technique that has been applied to design light- and chemically gated protein switches, could also be used to engineer phosphorylation-controlled protein switches (phospho-switches). Using an allosterically controllable Gal4 transcription factor as a scaffold, we show that a classic kinase Förster resonance energy transfer biosensor architecture can be used as a starting point for phospho-switch design. We optimize all features of the phospho-switch to develop an ERK-controlled transcription factor with a 20-fold phosphorylation-dependent change in transcriptional output. The resulting synthetic ERK-responsive transcription factor responds with comparable sensitivity to the c-fos promoter and reveals spatial ERK signaling patterns in mammalian developmental organoids. We further show that our switch architecture can be generalized to other input kinases and allosterically controlled targets. This work provides a general platform for a new generation of kinase-responsive tools for biosensing and synthetic biology applications.

Subject terms: Phosphorylation, Synthetic biology, Sensors and probes

This Article introduces a phospho-switch for allosterically regulating target proteins in response to ERK kinase activity.

Main

Cells rely on protein phosphorylation as a primary carrier of intracellular information. Protein kinases are activated by a variety of stimuli and regulate diverse processes including cytoskeletal assembly, metabolism, protein localization and gene expression. Phosphorylation can also link multiple kinases and substrates into networks that can perform diverse signal processing capabilities, including signal integration¹, amplification² and feedback control³. Yet, while bioengineers have long been able to build synthetic transcriptional circuits^4,5, our ability to engineer post-translational networks has been far more limited.

Recent studies have already begun to demonstrate the power of synthetic post-translational control for sensing and rewiring oncogenic states, constructing neural networks to perform information processing in cells and improving engineered cell-based therapies^6–14. Many of these studies have focused on engineering protease–substrate interactions to perform stimulus-dependent irreversible cleavage of an effector domain. In contrast to protein cleavage, the reversibility of protein phosphorylation presents an opportunity to provide bidirectional control for more fine-tuned responses in space and time. Synthetic phosphorylation-based logic would make it possible to directly interface with the kinome to sense kinase-encoded cell states and trigger desired responses.

Engineered allostery is a powerful concept for achieving stimulus-dependent control over protein function¹⁵. In engineered allostery, a stimulus-regulated protein domain (switch) is inserted at a particular site in a target protein, such that a conformational change in the switch alters the target protein’s function (Fig. 1a). Several domains can act as suitable stimulus-dependent switches, changing conformation in response to illumination or addition of a small molecule^16–20 (Fig. 1b). However, no switches are yet available to convert protein phosphorylation into an allosteric change in protein activity. A kinase-controlled phosphorylation-dependent switch could immediately confer phospho-regulation to the diverse set of target proteins previously engineered to be controlled by light or small molecule addition.

Fig. 1 — a, A schematic of engineered allosteric control of protein function. A stimulus-regulated protein domain is inserted into an allosteric site in a target protein and coupled to its function. b, Mechanism of action for existing light/ligand-sensitive switches, which undergo a stimulus-dependent conformational change. c, A schematic of FRET-based KARs. Fluorescent proteins on the N and C termini are linked by a phospho-binding domain (purple), a hinge region (gray) and a kinase-specific substrate (red) with docking motif (green). d, The Gal4 DNA-binding domain was previously made light-switchable by AsLOV2 insertion between Ser22 and Lys23. Replacing AsLOV2 with the EKAR produces a candidate ERK-controlled phosphoGal4 (EKAR-Gal4). e, An overview of the EKAR-Gal4 testing pipeline. EKAR-Gal4 was expressed in *5xUAS*-dEGFP reporter cells and GFP expression was measured in ERK-ON and ERK-OFF conditions. pp. switch, phospho-switch. f, Serum-dependent GFP expression of reporter cells transfected with WT-Gal4 or EKAR-Gal4. Cells were incubated in serum-free or growth media for 26 h before measurement. Reactivation was measured 6 h after switching back to growth media. Histograms represent >10,000 cells combined from three replicates. g, The GFP expression in response to repeated growth media stimulation in EKAR-Gal4 expressing reporter cells. Error bars represent mean ± s.e.m (n = 3 replicates). h, The GFP expression in reporter cells expressing EKAR-Gal4 (PRTP) or its nonphosphorylatable T-to-A mutant (PRAP) in serum-free or growth media conditions. Curves present GFP expression across a range of transcription factor expression levels. Error bars represent mean ± 95% confidence interval (left plot) or mean ± s.e.m. (right plot) of n = 3 replicates.

Source data

Here, we discover that a commonly used Förster resonance energy transfer (FRET) kinase biosensor provides a starting point for the development of a phosphorylation-sensitive allosteric switch. We show that the EKAR biosensor^21,22 for the extracellular signal-related kinase (ERK) can function as a prototype kinase-controlled switch when inserted into a Gal4 transcription factor at a previously identified allosteric site. We systematically optimize each component of the switch to improve its maximum amplitude and dynamic range, revealing design principles for achieving phosphorylation-dependent allostery that differ from FRET biosensors. Our final switch design results in a >20-fold change in transcriptional output between phosphorylated and unphosphorylated states. We show that our synthetic kinase-controlled transcription factor can serve as an excellent biosensor of ERK kinase activity, comparable in sensitivity to the widely used c-fos promoter and enabling detection of spatial signaling patterns during mammalian organoid development. Our switch generalizes to multiple kinases and target proteins, including a JNK-controlled transcription factor, an ERK-controlled actin nanobody and a synthetic signaling effector that wires ERK activity into cell protrusion. Overall, this strategy expands the synthetic biologist’s toolbox of post-translational regulation to multiple kinases and effector proteins with potential applications in biosensor design, drug screening and cell-based therapy.

Results

Repurposing FRET-based KARs as allosteric switches

Our strategy for achieving kinase-triggered allosteric control requires a phosphorylation-regulated sensory domain (phospho-switch) that can regulate the conformation of a target protein on the basis of the activity of a specific upstream kinase. An ideal phospho-switch would satisfy a few criteria. First, phosphorylation should trigger an open-to-closed conformational change that alters the distance between the switch’s N and C termini, in a similar manner to successful chemogenetic and optogenetic allosteric switches^16,17. Second, it should be rapidly and reversibly regulated by phosphorylation from a defined kinase of interest and dephosphorylation by intracellular phosphatases. Third, the switch design should be modular and generalizable to multiple kinases.

We reasoned that FRET-based kinase activity reporters (KARs) provide an attractive starting point for satisfying each of these criteria. KARs can report on seconds-timescale changes in kinase activity, and owing to their modular design, KARs targeting diverse kinases have already been developed and validated²³. The sensory region in KARs consists of a phosphorylatable substrate and docking motif for a specific kinase, a flexible hinge and a phosphate-binding domain (phospho-binder)²⁴. Binding between the phosphorylated substrate and phospho-binder brings fluorescent proteins on the KAR’s N and C termini into proximity to produce a change in fluorescence, resembling the desired conformational change of an allosteric switch (Fig. 1c).

As an initial test of KAR-based allosteric control, we set out to ‘wire in’ a KAR for the ERK kinase (EKAR-EV) to regulate a Gal4-VP64 transcription factor. The EKAR-EV biosensor contains an ERK-responsive sensory domain that includes an ERK-phosphorylatable peptide (amino acids PRTP) and docking motif (FQFP), an 88 amino acid Gly/Ser/Ala hinge and the Pin1 WW phospho-binding domain^21,22. We previously developed a light-controlled Gal4-VP64 variant by inserting the photoswitchable AsLOV2 domain between positions Ser22 and Lys23 inside the Gal4 DNA-binding domain, resulting in a >100-fold change in activity upon blue-light illumination²⁵. We replaced the AsLOV2 module with the EKAR sensory domain, terming the resulting fusion protein EKAR-Gal4. When ERK is active, the WW domain would be expected to bind the phosphorylated peptide, resulting in a closed switch and a transcriptionally active Gal4 (Fig. 1d).

To assess the kinase-controlled activity of EKAR-Gal4, we used a previously developed clonal HEK293T cell line in which enhanced GFP (EGFP) expression is driven by a Gal4-responsive 5xUAS promoter (5xUAS-dEGFP)²⁵ (Fig. 1e). We transiently transfected these cells with an EKAR-Gal4 plasmid that also included IRES-mCherry marker to assess cells’ relative Gal4 expression levels (Fig. 1e). We then incubated cells in either serum-free media to induce an ‘ERK-OFF’ state or serum-containing growth media to maintain the ‘ERK-ON’ state (Methods and Extended Data Fig. 1a). We observed approximately three-fold higher EGFP expression from cells incubated in growth media compared with serum-free media (Fig. 1f). EGFP expression could be reactivated within 6 h after a switch back from serum-free to growth media (Fig. 1f). By contrast, cells expressing an unmodified Gal4-VP64 transcription factor (‘WT-Gal4’) exhibited high levels of EGFP induction in both serum-free and growth media conditions (Fig. 1f).

Extended Data Fig. 1 — **(A)** PhosphoGal4 were transiently expressed and tested in a clonal *5xUAS*-dEGFP HEK293T cell line (Fig. 1e). Samples were analyzed for GFP induction with flow cytometry. First, cells are gated for proper single cells with Gate A/B using the FSC and SSC channels (size/single cell gating). Next, Gate C selects cells expressing the integration marker (P_PGK-iRFP) on the same construct with *5xUAS*-dEGFP (iRFP linear gate:1×10⁴-2×10⁵). Two analysis pipelines are used for data presentation. (1) Gate C gated cells are then gated for IRES-mCherry expression (mCherry linear gate: 5×10⁴-6×10⁴ or otherwise specified). mCherry-gated cells are used to plot GFP histograms or calculate the mean GFP value for bar graphs. (2) Gate C gated cells are plotted for mCherry vs GFP 2D scatter plots. For visualization, we bin cells on IRES-mCherry expression level and generate the IRES-mCherry vs GFP curve to show the transcriptional activity across all expression levels of phosphoGal4. (B) HEK293T cells expressing live-cell ERK biosensor (ErkKTR) were stimulated with growth media (GM, containing 10% FBS). Two GM changes were done following the same procedure of GM change in Fig. 1g. Nucleus ErkKTR clearance was measured to represent ERK activity (n = 24 cells from 3 biological replicates). ERK activity peaks 7 mins post GM change and quickly decays. By fitting ERK activity curve to an exponential decay model, the adaptation shows a t_1/2 of 14.17 mins. ERK activity can also be reactivated by the 2^nd GM change and then show similar adaptation. Representative images are shown at the time of 1^st/2^nd GM change, ERK activity peaks and 90 mins post GM change. Scale bars, 20 μm.

ERK signaling can be dynamic, with cells only transiently entering an ERK-high state after growth factor stimulation^26–28, which might limit the extent of Gal4 transcriptional activity in our system. To test if this was the case in HEK293T cells, we characterized ERK dynamics using an ERK biosensor (ErkKTR)²⁹. We found that ERK is only transiently activated for ~1 h by serum-containing growth media in HEK293T cells and that after adaptation, cells can be activated again by exchanging the media for fresh growth media (Extended Data Fig. 1b). On the basis of these results, we hypothesized that a series of growth media exchanges might produce a higher Gal4 response. Indeed, the level of EKAR-Gal4-induced EGFP expression was correlated to the number of growth media exchanges delivered within 30 h (Fig. 1g). For all subsequent serum-stimulation experiments, we exchanged media at 0, 6, 18 and 24 h before quantification of EKAR-Gal4 transcriptional output at 30 h.

We next sought to confirm that the change in EKAR-Gal4 transcriptional activity depends on the phospho-peptide’s phosphorylation state. We made an EKAR-Gal4 variant with a threonine-to-alanine mutation in the consensus ERK phosphorylation site (PRTP to PRAP). In ERK-ON conditions, the PRAP mutant exhibited reduced transcriptional activity compared with the PRTP variant, but higher activity than cells in serum-free media (Fig. 1h). These data suggests that EKAR-Gal4 transcriptional activity is controlled by serum factors beyond a single phosphorylation event, a primary target for optimization in subsequent experiments.

In summary, our initial experiments suggested that EKAR insertion is a promising starting point for the development of an ERK-controlled phospho-switch. However, further optimization is essential to improve its low maximum amplitude, limited dynamic range and residual activity in the T-to-A mutant. For version control purposes, we term this original EKAR as ‘phospho-switch v0’ and the corresponding EKAR-Gal4 as ‘ERK-phosphoGal4v0’.

Hinge optimization reveals allosteric switch design principles

As a first target for phospho-switch optimization, we examined the hinge region that links the phosphorylatable peptide to the WW phospho-binding domain. Prior studies of FRET kinase biosensors revealed that hinge length affects biosensor efficacy, with longer hinges producing a lower baseline of FRET activity and higher biosensor gain²², but the generality of these observations beyond FRET biosensors is unclear. We set out to explore a broad design space of hinge sequences, focusing on improving both the maximum activation in ERK-ON conditions and dynamic range between ERK-ON and ERK-OFF conditions.

We constructed a combinatorial Golden Gate cloning system (QCTK) for rapidly generating allosteric switch variants (Fig. 2a and Methods). We designed level 1 ‘a’, ‘b’ and ‘c’ parts consisting of phospho-binding domains, hinge sequences and phosphorylatable substrates, respectively (Supplementary Tables 1 and 2). We first built a hinge library containing 26 hinge designs (1b_001-1b_026) in the context of EKAR’s WW domain phospho-binder (1a_001) and ERK-phosphorylatable substrate (1c_001). We focused on synthetic hinges in three major classes: flexible GS-rich peptides, alpha helical EAAAK-repeat sequences and proline-rich AP-repeat sequences^30,31, as well as some hinge sequences derived from natural proteins (Fig. 2b).

All switch variants were assembled into the Gal4 backbone and transiently transfected into HEK293T 5xUAS-dEGFP cells. Cells were either starved or stimulated with four growth media exchanges (Fig. 1g) and analyzed by flow cytometry to estimate the phosphorylation-dependent change in transcriptional activity. Many hinge variants exhibited stronger maximum activation compared with our initial EKAR sequence (Fig. 2c, v0). Plotting the maximum activation and fold-change between serum-free and serum-stimulated conditions (denoted as serum-free/serum fold change) revealed individual variants with high scores on both metrics (Fig. 2d), and we chose 1b_001 (a seven amino acid EAAAKEA hinge) as our ‘v1’ variant for further development (Fig. 2d, arrow).

Further examination of the hinge library responses revealed some design principles (Fig. 2e,f). We found that GS-rich hinges exhibited the highest serum-induced GFP expression but, owing to high background activity²², these hinges had a low serum-free/serum fold change (Fig. 2e). Increasing hinge length reduced both background activity and maximum activity (Fig. 2f). By contrast, EAAAK and AP hinges exhibited low background activity even at short lengths. We fitted and validated a multivariable regression model with experimental hinge data to predict switch performance (Methods, Supplementary Note 1 and Extended Data Figs. 2 and 3). The model recapitulated the observed patterns for hinge rigidity and length, indicating that rigid EAAAK- and AP-repeat sequences exhibited an optimal balance of maximum activation and sensitivity to phosphorylation (Extended Data Fig. 2g–i). These design principles are consistent with our top hit being a short but relatively rigid seven amino acid helical hinge.

Extended Data Fig. 2 — (A) Features were extracted from hinge sequences and input to a multi-variable linear regression model for predicting phospho-switch variants performance. Sparse experimental data were used for model fitting. Fitted model was then used to predict performance of > 6000 *in silico* generated hinge sequences. (B) 5 selected hinge sequence features (length, A%, GS%, EK%, P%) were used as variables and tested for multi-collinearity with variance inflation factor analysis (VIF). All variables show VIF < 5. (C) Linear regression model was trained with experimental data to use hinge features to predict maximum activation in ERK-ON conditions and background activation in ERK-OFF conditions of phosphoGal4 variants. (D) Predicted maximum and background activation values comparing to measured values of variants shown in Fig. 2e. (E) Relative error in different hinge groups. Hinge groups as shown in Fig. 2b: Repeats, synthetic strict repeats (for example EAAAK repeats); Hybrid, synthetic hinges with mixed amino acids (for example EV-AP mix); Natural, natural hinges. Relative error was calculated as (x_predicted – x_measured) / x_measured. (F) Standardized contribution of variables in maximum and background activation models. Inputs for each variable were scaled for coefficient comparison. (G) Computationally generated hinge sequences (GS-rich flexible hinges, EAAAK-repeat alpha helical hinges, and AP-repeat proline-rich hinges) were fed into the model to generate a 2D plot (as in Fig. 2d) and the result recapitulates the pattern on hinge rigidity. (H) Starting from rigid AP-repeat proline-rich hinges, different percentages of the amino acids were computationally mutated to G/S and increasing percentages of mutations gradually shifts the performance of phosphoGal4 towards those with flexible hinges. (I) Model prediction recapitulates the pattern on hinge length. Increasing hinge length correlates with lower maximum activation and higher fold change.

Extended Data Fig. 3 — (A) Validation method 1: Train/test dataset split (details in Supplementary Note 1). In the hinge dataset, 70% of variants were randomly selected as the training dataset to generate a regression model for predicting maximum activation or background activation. The other 30% of variants served as the testing set, which are experimental data that the model has never seen. 3 random split tries were shown for each model to prevent random split bias. The results are shown on predicted vs measured 2D plots. (B) Validation method 2 (details in Supplementary Note 1): Challenge the model with new experimental datapoints. 5 new hinges were designed and tested experimentally. The mean value of 3 replicates was used as measured values. The previous models were used to predict maximum activation or background activation of the new variants, and the results are shown on predicted vs measured 2D plots.

We performed additional simulations and experiments to validate that our phosphorylation-responsive Gal4 system is allosterically controlled. We used AlphaFold3³² to predict the structures of phosphorylated and unphosphorylated ERK-phosphoGal4v1 (Extended Data Fig. 4a–d), which captured the binding event between the WW domain and phosphorylated ERK substrate³³, matching previously solved structures of WW-peptide binding^34–36 and supporting our conceptual model of a phosphorylation-dependent conformational switch (Extended Data Fig. 4e–h and Supplementary Note 2). We also measured ERK-phosphoGal4v1 performance after inserting additional linker residues between the phospho-switch and Gal4 DNA-binding domain, which should weaken allosteric coupling between these domains. Indeed, we observed decreasing ERK-switchable behavior with increasing linker length, consistent with allosteric coupling that is weakened by longer, flexible linkers (Extended Data Fig. 5a,b).

Extended Data Fig. 4 — (A) Crystal structure of wild-type Gal4 homodimer-DNA complex (PDB: 3COQ). Detailed structure of Zn₂C₆ zinc finger of Gal4 shows 6 cysteines (C11, C14, C21, C28, C31, C38) capturing 2 zinc ions (green). Allosteric insertion site between S22 (dark blue) and K23 (red) separates the zinc finger to N-term half (pink) and C-term half (light blue). (B) AlphaFold3-predicted models of ERK phospho-switch v1 (dark red) inserted Gal4 in complex with UAS elements with ERK targeted T95 phosphorylation. (C) Comparison of phosphorylated and unphosphorylated AlphaFold3 models. Arrow indicates the interaction between phosphorylated T95 in ERK substrate (green) and the binding interface in the Pin1 WW domain (purple). (D) Distances between S42 in the WW binding interface and T95 in the ERK substrate are measured in 15 predicted models (30 data points for each model contains 2 zinc fingers as Gal4 forms a dimer when binds to UAS elements). (E) Structure matching between a solved structure (PDB: 1i6c; yellow) and a AlphaFold3-predicted model (purple) of Pin1 WW alone (WW as a single protein). (F) Structure matching between WW (blue) inside a solved structure of Pin1 protein (PDB: 1pin; yellow) and WW (purple) inside the AlphaFold3-predicted phosphoGal4 model (WW as an intra-protein domain). R.M.S.D was calculated only for the WW domains. (G) Solved structure of the Pin1 WW domain in complex with a human phosphorylated Smad3 derived peptide. The phosphorylated 4 amino-acid peptide motif PEpTP (orange) has the phosphate interacting with the Pin1 WW domain (pink). Predicted hydrogen bonds (blue) are shown in zoomed panels. (H) WW-PRpTP interaction in the AlphaFold3-predicted phosphoGal4 model. The phosphorylated motif, PRpTp (orange) interacts with Pin1 WW inside the phospho-switch. Predicted hydrogen bonds (blue) are shown in zoomed panels.

Extended Data Fig. 5 — (A) Generation of ERK-phosphoGal4v1 variants with increasing length of flexible linkers (L1/L2) between ERK phospho-switches and Gal4. Variant 1 is the unmodified ERK-phosphoGal4v1. Variant 2-4 add two-residue GS linkers at N-term side, C-term side or both sides of ERK phospho-switch v1. Increasing length of L1/L2 un-couples the phospho-switch from target protein function. Variants were generated with QCTK Golden Gate assembly. (B) GFP expression of ERK phosphoGal4v1 variants 1-4. For each variant, a non-phosphorylatable mutant was tested as control. Cells transiently expressing variants were stimulated with four growth media (GM) changes in 30 h before flow cytometry measurements. Error bars represent mean± s.e.m (n = 3 replicates). Statistical significance was computed using the two-tailed Welch’s t-test. p-values: var1-PRTP vs var2-PRTP: 0.19; var1-PRTP vs var3-PRTP: 0.031; var1-PRTP vs var4-PRTP: 6.2×10⁻³; var1-PRAP vs var2-PRAP: 0.17; var1-PRAP vs var3-PRAP: 0.079; var1-PRAP vs var4-PRAP: 1.7×10⁻³. (C) Histograms show the performance of ERK-phosphoGal4 (controlled by corresponding versions of ERK phospho-switches) with a WT non-inserted Gal4 (green) and a no-VP64 version of ERK phosphoGal4v3 (gray) as controls. The performance is presented by comparing GFP expressions of the phosphorylatable (PRTP, orange) and non-phosphorylatable (PRAP, pink) phosphoGal4 variants in ERK-ON condition. Histograms represent >10000 cells from 3 replicates. Cells are gated for IRES-mCherry expression range of (5×10⁴-1.25×10⁵) for fold change calculation and histograms (See **Methods**). Top charts summarize the course of optimization. The dashed lines highlight the peak values of the no-VP64 control (left) and the WT non-inserted Gal4 control (right). (D) The threonine at the phospho-site (TP) in ERK-phosphoGal4v4a was mutated to a non-phosphorylatable alanine (AP) or a phospho-mimetic aspartic/glutamic acid (DP/EP). Cells transiently expressing these variants were either stimulated with four growth media (GM) changes or kept in serum-free (SF) media in 30 h before flow cytometry measurements. Error bars represent mean± s.e.m (n = 3 replicates). Statistical significance was computed using the two-tailed Welch’s t-test. P-values: TP vs AP: 0.0010; AP vs DP: 0.29; AP vs EP: 0.060. p-value annotation: ns, p > 0.05; *, 0.01 < p <= 0.05; **, 10⁻³ < p <= 0.01; ***, 10⁻⁴ < p <= 10⁻³; ****, p <= 10⁻⁴.

Improving the substrate site and phospho-binder to generate an optimal phospho-switch

Although our initial phospho-switches exhibited good switching between serum-free and growth media conditions, we found that a nonphosphorylatable mutant switch still retained modest residual EGFP expression in growth media (Fig. 1h). We initially hypothesized that the residual activity might be due to phosphorylation at additional sites in the substrate peptide that leads to WW domain binding and Gal4 activation.

We constructed a series of phospho-switches harboring mutant substrate peptides lacking prolines and phosphorylatable residues that may contribute to WW domain binding (Fig. 3a). We found that the presence of the phosphorylatable PRTP motif ensured strong serum-dependent EGFP expression regardless of these additional sequence modifications (Fig. 3b), and only substrate 5, which lacks all prolines, phosphomimetic residues and phospho-sites, abolished residual EGFP expression in growth media (Fig. 3b). On the basis of these data, we designed a final pair of peptides in which the phosphorylatable PRTP or nonphosphorylatable PRAP motif was re-introduced into substrate 5 (Fig. 3c). A single Pro-directed phosphorylation site was able to rescue high transcriptional output in growth media, and the corresponding nonphosphorylatable PRAP variant exhibited a 42% reduction in background activity compared with the original ERK-phosphoGal4v1 design; we termed this our ‘v2’ variant (Fig. 3c).

Fig. 3 — a, An overview of mutant substrate peptides. ERK-phosphoGal4v1 (substrate 1) was modified to produce variant substrates 2–5 by mutating phosphorylatable, proline and negatively charged residues that might contribute to WW domain binding. b, The GFP expression of phosphoGal4 variants. Cells transiently expressing variants were stimulated with four growth media changes 30 h before flow cytometry measurements. Error bars represent mean ± s.e.m. (n = 3, 6, 3, 3 and 3, respectively). c, Reintroducing the ERK-consensus phosphorylation motif (PRTP or nonphosphorylatable PRAP as control) into substrate 5 increases the switch’s dynamic range. The best-performing variant (a1b1c17) was termed the ‘v2’ phospho-switch. Error bars represent mean ± s.e.m. (n = 3 replicates). d, An AlphaFold3-predicted model of the Pin1 WW domain (pink) interacting with phosphorylated PRTP peptide (PRpTP, orange). Arrows indicate the positions of Arg14 and Phe25. Dashed blue lines indicate predicted hydrogen bonds. e, The characterization of ERK-phosphoGal4v2 variants with WW domain mutations in phosphorylatable (PRTP) and nonphosphorylatable (PRAP) contexts. The resulted optimal variant (a16b1c17) is termed ‘v3’. Error bars represent mean ± s.e.m. (n = 5 replicates). f, The characterization of ERK-phosphoGal4v3 variants with 17 and 27 amino acid helical hinges, termed ‘v4a’ and ‘v4b’. Error bars represent mean ± s.e.m. (n = 5 replicates). g, The overall course of optimization from v0 (EKAR) to v4a/b. GFP expression curves for each variant are presented across a range of transcription factor expression levels. Error bars represent mean ± 95% confidence interval (n = 3, 3, 5, 5, 5 and 5, respectively). Phosphorylatable (PRTP), nonphosphorylatable (PRAP) and a no-VP64 phosphoGal4v3 variant were tested under ERK-ON or ERK-OFF conditions, as in Fig. 2. h, On/off kinetics of ERK-phosphoGal4v4a in stably expressing cells. For off kinetics, cells were seeded in growth media for 24 h and switched to serum-free media for the indicated times. For on kinetics, cells were seeded in growth media for 24 h, switched to serum-free media for 18 h and then switched into growth media for the indicated times. Error bars on the curve represent mean ± s.e.m. (n = 3 replicates). For all panels, statistical significance was computed using the two-tailed Welch’s t-test.

Source data

We next made targeted mutations in the WW domain to further reduce residual affinity for the unphosphorylated substrate. Structural and biochemical studies have determined that specific residues in the WW domain, Arg14 and Phe25, make contacts to the first proline residue in the PRTP site and thus may contribute to phosphorylation-independent binding^37–39 (Fig. 3d). We constructed versions of ERK-phosphoGal4v2 in which these two residues were mutated to Ala and Leu, respectively. We found that the R14A variant still drove strong EGFP expression in growth media and exhibited a further 58% reduction in phosphorylation-independent activity compared with ERK-phosphoGal4v2, resulting in a ‘v3’ variant (Fig. 3e). As a final improvement, we revised our hinge design to further weaken substrate–WW interactions by increasing the length of the helical hinge to 17 or 27 amino acids, producing ‘v4a’ or ‘v4b’ variants with further reduction in background activity (Fig. 3f).

Over the course of optimization from our phospho-switch v0 to v4a/b designs, the TP/AP fold change of the engineered transcription factor increased from 2.7-fold to 24.4-fold and the maximum activation increased 3.5- to 4.6-fold (Fig. 3g and Extended Data Fig. 5c). Our ERK-phosphoGal4v4a reaches a comparable level of gene expression to unmodified Gal4, and the nonphosphorylatable mutant is comparable to a negative control Gal4 construct that lacks the VP64 transactivation domain (Extended Data Fig. 5c). Phosphomimetic variants produced only weak GFP expression (Extended Data Fig. 5d), consistent with a strict phosphate moiety requirement for WW domain high-affinity binding⁴⁰.

We measured the kinetics of EGFP accumulation and loss in response to media changes in 5xUAS-dEGFP HEK293T cells that stably express the ERK-phosphoGal4v4a construct, observing changes in GFP expression with a half-life of ~6 h (Fig. 3h). We also tested the ERK-phosphoGal4v4a system in additional cellular contexts: mouse embryonic stem cells (mESCs) and fibroblasts (Extended Data Fig. 6 and Supplementary Note 3). Overall, the switch was functional in all cases, albeit with different magnitudes of EGFP response that probably reflect nuances of stimulus-dependent ERK activation in each cellular context.

Extended Data Fig. 6 — (A) Schematic of single-construct ERK-phosphoGal4v4a/UAS system. Mouse embryonic stem cells (mESCs) were engineered to stably express the system. Engineered cells were seeded in 2i+LIF media and switched to the conditioned media for 24 h. UCOE, ubiquitous chromatin-opening element. PGK, phosphoglycerate kinase 1 gene promoter. (B) Representative images of stem cell colony morphology and GFP expression in LIF-only media (No inhibitors) or in 2i+LIF media. Scale bar, 250 μm. Images are representative of n = 3 replicates. (C) GFP expression of engineered mESCs in response to drug treatments. Chiron (CHIR-99021), 3 μM. MEKi, PD0325901. Error bars represent mean ± s.e.m (n = 3 replicates). Histograms represent >10000 cells. (D) Schematic of NIH 3T3 cells engineered with the single-construct system in A. Engineered cells were seeded in growth media (contains 10% FBS) and switched to the conditioned media for 24 h. (E) GFP expression of engineered NIH 3T3 cells in response to drug treatments. Serum, 10% FBS. MEKi, PD0325901. Error bars represent mean ± s.e.m (n = 3 replicates). Histograms represent >10000 cells. (F) Fold change of ERK-phosphoGal4v4a transcriptional activity in HEK293T and NIH3T3 cell lines. HEK293T and NIH3T3 cells stably expressing the single-construct ERK-phosphoGal4v4a/UAS system were either stimulated with one growth media (GM) changes or kept in serum-free (SF) media in 24 h before flow cytometry measurements. n = 3 replicates. Fold changes are calculated as direct ratios between two conditions. (G) Differential ERK-phosphoGal4v4a/UAS responses to various stimuli in HEK293T and NIH3T3 cell lines. 24 h post seeding, two cell lines were maintained in previously in-well growth media for another 18 h and then stimulated with various stimuli for 4 h before flow cytometry measurements. MEKi (PD0325901), 10 μM. EGF, 100 ng/mL. PDGF, 50 ng/mL. n = 3 replicates. (H) Differential ERK-phosphoGal4v4a/UAS responses to various seeding density in HEK293T and NIH3T3 cell lines. 24 h post seeding at various densities, the two cell lines were kept in serum-free (SF) media with or without EGFR inhibition for 24 h before flow cytometry measurements. EGFRi (Gefitinib), 10 μM. Error bars represent mean± s.e.m (n = 3 replicates).

Direct and specific activation of ERK-phosphoGal4 using optogenetics

Media switch experiments are a simple method to alter ERK activity for characterizing candidate ERK-phosphoGal4 designs. However, growth media contains complex mixtures of growth factors and other molecules that can alter the activity of many intracellular pathways. We thus set out to characterize our ERK-phosphoGal4 system in response to stimuli that specifically and uniquely activate the ERK cascade. Such an approach might also aid in better understanding the residual EGFP expression observed from our v0–v2 nonphosphorylatable switch variants in growth media compared with serum-free conditions (Figs. 1h and 3g).

We turned to our previously developed OptoSOS optogenetic system as a method to directly activate the Ras/ERK pathway^41,42. We engineered 5xUAS-dEGFP HEK293T cells that stably expressed a blue-light-responsive variant of this OptoSOS system^42,43, the ERK-phosphoGal4v2 transcription factor and an infrared fluorescent protein-fused ErkKTR biosensor (ErkKTR-iRFP) (Fig. 4a,b). Continuous blue-light stimulation drove sustained ErkKTR-iRFP nuclear export, indicative of sustained ERK activity, and also produced strong EGFP expression in serum-free media (Fig. 4c and Extended Data Fig. 7a–c). The light-induced response could be eliminated by treatment with 10 μM of the MEK inhibitor PD0325901 (Fig. 4d and Extended Data Fig. 7d). Light-induced EGFP expression could be tuned by varying the intensity or duty cycle of illumination, with increasing EGFP output as the duration of light pulses increased from 2.5 min every 30 min to continuous illumination at 2 mW cm⁻² (Fig. 4e,f). These data indicate that ERK-phosphoGal4 activity can be tuned to intermediate levels by minute-timescale changes in ERK activity, suggesting that the transcription factor is able to rapidly switch between active and inactive states upon phosphorylation in response to these fast changes in illumination conditions.

Fig. 4 — a, A schematic of optogenetic ERK stimulation and biosensor measurement. ERK is activated by blue-light stimulation of the OptoSOS system and measured using the ErkKTR biosensor and phosphoGal4v2-induced GFP expression. b, A HEK293T stable cell line was constructed that harbors optoSOS, ErkKTR-iRFP, ERK-phosphoGal4v2 and a *5xUAS*-dEGFP reporter. c, Representative images of light-induced ErkKTR translocation and GFP expression. Top: ErkKTR localization before and after 1 h blue-light illumination (scale bar, 25 µm). Bottom: GFP expression before and after 10 h blue-light illumination (scale bar, 100 µm). Images are representative of n = 3 replicates. d, Light-induced GFP expression downstream of ERK-phosphoGal4v2 is MEK-dependent. Cells were incubated in dark or continuous light illumination at 2 mW cm⁻² for 22 h with or without the presence of the MEK inhibitor PD0325901 and analyzed by flow cytometry. Error bars represent mean ± s.e.m. (n = 3 replicates). e, GFP expression varies with the light intensity of optogenetic stimulation. Cells were continuously illuminated at various intensities for 22 h and analyzed by flow cytometry. Error bars represent mean ± 95% confidence interval (n = 3 replicates). Data were fitted to a Hill curve for EC₅₀ estimation. f, GFP expression varies with the duty cycle of optogenetic illumination. Cells were exposed to 2 mW cm⁻² blue light at schedules ranging from 2.5 min every 30 min to continuous illumination for 22 h and analyzed by flow cytometry. Error bars represent mean ± s.e.m. (n = 3 replicates). g, The crystal structure of Gal4 Zn₂C₆ zinc finger (PDB: 3COQ). Six cysteine residues (yellow) coordinate with two zinc ions (green). The phospho-switch is inserted between Ser22 (blue) and Lys23 (red), potentially interfering with zinc coordination. h, GFP expression for cells incubated in serum-free media supplemented with various levels of ZnCl₂ in dark or under continuous illumination at 2 mW cm⁻² for 22 h and analyzed by flow cytometry. Error bars represent mean ± s.e.m. (n = 5, 5, 5, 5, 6 and 4, respectively). Histograms show >25,000 cells pooled from all replicates.

Source data

Extended Data Fig. 7 — (A) Schematic of live-cell imaging experiments for long-time optogenetic ERK stimulation. Engineered optoSOS/ERK-phosphoGal4v2/UAS HEK293T cells were seeded in growth media (GM) and then switched to serum-free (SF) media for 24 h. Cells were kept in dark before imaging. After starvation, cells were imaged and stimulated with blue light for >15 h. (See **Methods**). (B) Representative ErkKTR images at 0, 5, 10 and 15 h in dark or under blue light illumination. Scale bar, 100 μm. (C) Quantification of ErkKTR response to long-term optogenetic ERK stimulation in B. Mean trajectories (solid lines) are shown with error bars representing the 95% confidence interval. Average ~250 cells in 3 replicates are quantified at each time points. (D) HEK293T cells stabling expressing ERK-phosphoGal4v2 or ERK-phosphoGal4v4a were seeded and transiently transfected with the optoSOS-expressing plasmid 6 h post-seeding. Cells were kept in dark for 18 h and then switched to serum-free media and incubated in dark or continuous light illumination at 2 mW/cm² for 22 h with or without the presence of 10 μM of the MEK inhibitor PD0325901 (MEKi) before flow cytometry measurements. Error bars represent mean ± s.e.m (n = 3 replicates). Fold changes are calculated as direct ratios between two conditions.

The S22K23 insertion site is located inside the Zn₂C₆ zinc finger domain, near cysteine residues that are essential for Zn²⁺ coordination⁴⁴ (Fig. 4g). We hypothesized that the allosteric switch might toggle the Gal4 DNA-binding domain between zinc-bound and zinc-unbound states. In this model, high zinc ion concentrations could shift the equilibrium to favor the zinc-bound, active Gal4 conformation even in the absence of ERK phosphorylation. Such a model could help explain the difference in activity observed for a nonphosphorylatable variant in serum-free and growth media conditions, as the 1–40 μM levels of Zn²⁺ in tissue culture media is primarily provided by serum^45,46. To test this model, we used the OptoSOS system to activate ERK in serum-free media supplemented with varying concentrations of ZnCl₂. Indeed, we observed that high zinc concentrations increased EGFP induction in both illuminated (ERK-ON) and unilluminated (ERK-OFF) conditions (Fig. 4h). These data are consistent with a model where high zinc concentrations act in concert with residual WW-peptide binding to shift the equilibrium of ERK-phosphoGal4v2 to the active conformation. Importantly, our improved v4 phospho-switches already eliminate serum-induced activity from the nonphosphorylatable PRAP variant, probably by reducing the WW-peptide binding affinity to shift the equilibrium to the switch-open, zinc-unbound state even under high-zinc conditions provided by growth media.

Applying the ERK-phosphoGal4 switch as a biosensor for developmental cell signaling

Transcriptional biosensors are widely used for assessing signaling pathway activity in high-throughput screens and complex tissues^47–49. While they lack the ability to sense the rapid, complex dynamics that are captured by FRET and translocation-based biosensors²³, they are still useful for transducing a cell’s integrated kinase activity over time to a simple change in fluorescence, or for constructing complex gene circuits to record information about cells’ signaling history^47,50,51. Most transcriptional biosensors are constructed from signaling-responsive endogenous enhancers and promoters^52–54. However, endogenous enhancers typically lack specificity to a single kinase, as they may contain binding sites for many endogenous transcription factors that are themselves subject to complex regulation. A second challenge is availability: well-defined transcriptional biosensors are only available for a few kinases. We reasoned that engineered kinase-responsive transcription factors could be used immediately as biosensors that convert total kinase activity over time into a simple fluorescent signal.

We first set out to compare our optimized ERK-responsive transcription factor (ERK-phosphoGal4v4a) to a 2.5-kb c-fos upstream regulatory sequence (P_FOS) that has been widely used for monitoring ERK, calcium and PKA signaling^{50,53,55–59} (Fig. 5a). We constructed stable HEK293T cell lines containing either 5xUAS-dEGFP or P_FOS-dEGFP and a constitutively-expressed iRFP marker using PiggyBac-mediated genome integration⁶⁰. Both lines were sorted to similar iRFP levels to ensure similar reporter strengths. We also stably introduced the ERK-phosphoGal4v4a kinase-sensitive transcription factor into the 5xUAS-dEGFP cell line (Fig. 5a). We stimulated both P_FOS and ERK-phosphoGal4v4a HEK293T cell lines with varying doses of epidermal growth factor (EGF) in the presence or absence of inhibitors of EGF receptor or ERK’s kinase MEK (Fig. 5b,c). We found that both P_FOS and ERK-phosphoGal4v4a were similarly sensitive to EGF, with EC₅₀ values of 40 and 20 ng ml⁻¹, respectively. Interestingly, P_FOS produced a strongly switch-like response to EGF (Hill coefficient n_H = 20.0) compared with a more graded response of the ERK-phosphoGal4 system (Hill coefficient n_H = 1.4), consistent with prior observations of strongly switch-like c-fos expression in response to graded ERK stimuli⁵⁷.

Fig. 5 — a, A schematic comparing ERK-dependent transcription for a traditional P_FOS reporter and the ERK-phosphoGal4v4a/UAS system. b, EGF-stimulated responses of the P_FOS promoter and the ERK-phosphoGal4v4a/UAS system. Cells were tested under various combinations of EGF (100 ng ml⁻¹), MEKi (PD0325901, 10 μM) and EGFRi (Gefitinib, 10 μM) for 4 h before flow cytometry measurements. Data was normalized to GFP expression in the unstimulated condition. Error bars represent mean ± s.e.m. (n = 4, 4, 3 and 3, respectively). c, EGF dose–response curve for the P_FOS promoter and the ERK-phosphoGal4v4a/UAS system. Cells were treated with 0.03–600 ng ml⁻¹ EGF for 4 h. Each curve was normalized to its minimum and maximum GFP expression. Error bars represent mean ± 95% confidence interval (n = 3 replicates). The data were fitted to a Hill curve for EC₅₀ and Hill coefficient estimation. d, A schematic of the gastruloid generation protocol. mESCs stably expressing the ERK-phosphoGal4v4a/UAS system are seeded into low-adhesion wells and treated with the Wnt agonist CHIR-99021 from days 2 to 3 to trigger symmetry breaking and elongation. e, ERK-phosphoGal4v4a/UAS responses in individual elongating gastruloids over time. Snapshots are shown at 90, 96 and 108 haa for the GFP and bright-field channels. A ↔ P, anterior-to-posterior. Dashed line shows gastruloid boundaries. Scale bar, 100 μm. Images are representative of n = 6 gastruloids. f, ERK-phosphoGal4v4a/UAS response in early gastruloid development from 48 to 96 haa. Images of representative fixed gastruloids are shown at 48, 72, 84 and 96 haa for the GFP channel. Pixel intensity in the 48 haa image was scaled down 4× to prevent pixel saturation. Scale bar, 100 μm. g,h, Quantification of gastruloid ERK-phosphoGal4v4/UAS activity by measuring dEGFP intensity (g) and dEGFP polarization as described in the Methods (h). For f–h, n = 6, 7, 6 and 15 gastruloids were analyzed across the indicated conditions. i, FGFR inhibition of gastruloid ERK-phosphoGal4v4a/UAS responses. Gastruloids were cultured in the presence or absence of 100 nM PD173074, an FGFRi, from 72 to 96 haa. Left: representative gastruloid images at 96 haa. Scale bar, 100 μm. Right: GFP expression along the gastruloid major axis at 96 haa. Curves show mean ± 95% confidence interval for n = 14 and 14 gastruloids, respectively. j, A schematic of brachyury integration of Wnt and FGF/ERK signaling. k, Representative images of ERK-responsive GFP (top) and brachyury (bottom) in gastruloids at 72, 84 and 96 haa. Scale bar, 100 μm. l, The quantification of brachyury expression polarization, computed as in h. m, Pearson’s correlation of pixel intensities between GFP and brachyury expression for individual gastruloids. For k–m, n = 7, 5 and 7 gastruloids for 72, 84 and 96 haa timepoints, respectively. For g, h, l and m, all gastruloids are plotted as scattered points and box plots indicate the median (center line), upper and lower quartiles (box limits) and 1.5× the interquartile range (whiskers). Statistical significance was computed using the two-tailed Welch’s t-test.

Source data

We next set out to apply the ERK-phosphoGal4 system to characterize ERK signaling in a novel context: 3D gastruloids, which are organoids composed of ESCs that grow over 5 days, break symmetry and elongate along an anterior–posterior axis to form a structure resembling the mammalian tailbud^61,62 (Fig. 5d). Gastruloid development is triggered by addition of a Wnt pathway activator CHIR-99021 from 48–72 h after aggregation (haa), leading to establishment of an anterior–posterior body axis and pronounced elongation from 96–120 haa. During symmetry breaking and elongation, gastruloids are made up of 10⁴–10⁵ highly motile cells in a three-dimensional aggregate, making quantification of biosensor responses challenging. Intriguingly, single-cell RNA sequencing revealed differential expression of FGF ligands during this symmetry-breaking period⁴⁷. Nevertheless, the FGF/ERK pathway’s potential involvement in gastruloid symmetry breaking remains uncharacterized, owing in part to a lack of validated ERK transcriptional biosensors in ESCs.

We generated a polyclonal mESC line expressing the ERK-phosphoGal4v4a transcription factor and 5xUAS-dEGFP target gene cassette using the same PiggyBAC transduction strategy described above (Extended Data Fig. 6a–c). Gastruloids were cultured and captured for imaging at various time points. We first examined ERK-phosphoGal4v4a activity in live elongating gastruloids, where prior fixed-cell measurements had confirmed that ERK activity forms a gradient that peaks at the posterior pole⁶³. We observed a single pole of EGFP expression that coincided with the elongating posterior domain (Fig. 5e and Supplementary Video 1). We next measured EGFP activity from 48 to 96 haa, a time period when Wnt signaling breaks symmetry from a uniform to a polarized state. We observed pronounced changes in ERK-driven EGFP expression, from a uniform high state at 48 haa to a single posterior pole by 96 haa (Fig. 5f and Extended Data Fig. 8a), dynamics which correspond closely in space and time to the changes in Wnt pathway activity observed previously⁴⁷. Quantification across multiple gastruloids confirmed the marked reduction in overall EGFP levels and a progression to a polarized state (Fig. 5g,h). Within these overall spatial patterns, EGFP was expressed in a salt-and-pepper fashion, with substantial variability between nearby cells, possibly reflecting stochastic silencing of our engineered construct or true variability in ERK signaling. EGFP expression was abrogated by treatment with the FGFR inhibitor (FGFRi) PD173074, indicating that the ERK-phosphoGal4v4a system faithfully reflected FGF/ERK activity (Fig. 5i and Extended Data Fig. 8b–d). Wnt and ERK signaling are known to converge on a transcription factor, brachyury, which is expressed in the mammalian posterior tailbud (Fig. 5j). Consistent with this logic, we found that brachyury was expressed in the same regions of the gastruloid where EGFP⁺ cells were found (Fig. 5k,l and Extended Data Fig. 8e). Overall, these experiments reveal that ERK signaling is active and dynamic during gastruloid symmetry breaking, suggesting that it may cooperate with Wnt signaling to orchestrate this developmental process.

Extended Data Fig. 8 — (A) Additional 3 gastruloid examples were shown here at 48, 72, 84 and 96 hours after aggregation (haa) as supplement to Fig. 5f. GFP signal is scaled down 4x at 48 haa. Dashed line shows gastruloid boundaries. Scale bar, 100 μm. (B) Schematics of the gastruloid protocol for FGFR inhibition. Gastruloids were cultured in the presence or absence of 100 nM PD173074, a FGFR inhibitor from 72 to 96 hours after aggregation (haa). At 96 haa, gastruloids were fixed and imaged for analysis. (C) Additional 3 gastruloid examples were shown at 96 haa with or without FGFR inhibition as supplement to Fig. 5i. Dashed line shows gastruloid boundaries. Scale bar, 100 μm. (D) Quantifications of gastruloid ERK-phosphoGal4v4a/UAS response with or without inhibition: overall UAS-dEGFP intensity (left panel) and UAS-dEGFP signal polarization (right panel). Number of gastruloids per condition: DMSO, n = 14; FGFRi, n = 14. For D, all gastruloids are plotted as scattered points and boxplots indicate: the median (center line); upper and lower quartiles (box limits); 1.5x the interquartile range (whiskers). Statistical significance was computed using the two-tailed Welch’s t-test. p-values: overall intensity: 1.62×10⁻⁶; polarization: 1.58×10⁻³. p-value annotation: ns, p > 0.05; *, 0.01 < p <= 0.05; **, 10⁻³ < p <= 0.01; ***, 10⁻⁴ < p <= 10⁻³; ****, p <= 10⁻⁴. (E) Additional 3 representative gastruloids of Brachyury immunostaining were shown here at 72, 84 and 96 hours after aggregation (haa) as supplement to Fig. 5k. Dashed line shows gastruloid boundaries. Scale bar, 100 μm.

Phospho-switch control is generalizable to other input kinases and effector proteins

Beyond their use as biosensors, kinase-controlled allosteric switches have the potential to be powerful tools for synthetic biology because of their flexibility at both the input and output levels. At the input level, FRET biosensors have been developed for many kinases, and these phospho-sites could also be incorporated into our optimal switch backbone to rewire input specificity. At the output level, a phospho-switch can be inserted into a growing number of allosterically controlled target proteins^{16,25,64–66}. We thus set out to test whether our design could be extended to other input kinases and effector proteins.

We first tested whether the phosphoGal4 design could be rewired to another input kinase (Fig. 6a). We turned to the JNK kinase, which also has a well-characterized FRET biosensor (JNKAR), but for which transcriptional biosensor options are limited. We assembled a JNK-phosphoGal4 system using the ‘v2’ (wild-type WW phospho-binder) and ‘v3’ (WW^R14A) architecture, but where the ERK-specific substrate was exchanged for the JNK-specific JNKAR substrate sequence. Each phospho-switch design was inserted into Gal4 at the same S22K23 allosteric site (Fig. 6b) and transfected into 5xUAS-dEGFP cells. Indeed, we found that cells expressing a phosphorylatable JNK-phosphoGal4 construct induced EGFP expression in response to the JNK activator anisomycin, with the ‘v3’ design outperforming ‘v2’ (Fig. 6c). EGFP expression varied with anisomycin concentrations previously seen to activate JNK²⁹ (Fig. 6d). To validate the selectivity of the JNK-phosphoGal4 and ERK-phosphoGal4 for their cognate kinases, we stimulated cells expressing either biosensor with anisomycin in the presence of small-molecule inhibitors of either kinase. While ERK-phosphoGal4v4a was also mildly activated by anisomycin, this activity was reversed by MEK inhibition but not JNK inhibition, and the converse was observed for JNK-phosphoGal4v3 (Extended Data Fig. 9a,b). These data confirm that the phospho-switch can be rewired to accept phosphorylation from distinct input kinases.

Extended Data Fig. 9 — (A) Dose response curve of JNK inhibitor VIII. HEK293T cells stable expressing JNK-phosphoGal4v3 were stimulated with 25 ng/mL anisomycin for 24 h in the presence of various concentration of JNK inhibitor VIII. Error bars represent mean ± s.e.m (n = 3 replicates). (B) Kinase selectivity of ERK-phosphoGal4v4a and JNK-phosphoGal4v3. HEK293T cells stable expressing ERK-phosphoGal4v4a or JNK-phosphoGal4v3 were tested under various conditions with anisomycin (25 ng/mL), JNKi (JNK inhibitor VIII) and MEKi (PD0325901). Heatmap shows the mean value of 3 replicates. (C) Serum-induced ERK-phosphoNanobody binding. Top panels: images of a representative cell before and after 7 min treatment with serum. Dashed lines indicate the position of a line scan for quantifying ERK-phosphoNanobody intensity. Scale bar, 10 μm. Bottom panels: quantification of ERK-phosphoNanobody fluorescence along the line scans before (gray) or after treatment (red) reveals translocation to the cell membrane upon serum stimulation as expected for cortical actin binding. (D) Representative ERK-phosphoNanobody and ErkKTR images before and after serum treatment in the presence of 10 μM U0126, a MEK inhibitor as supplement for Fig. 6h. Scale bar, 20 μm. (E) Representative images of ERK-phosphoNanobody v2 or v3 7 mins post serum stimulation. For each version, a non-phosphorylatable (PRAP) variant was tested as control. A 50-pixel line scan (11.8 μm) was drawn to visualize the cell periphery localization of tagRFP-tagged ERK-phosphoNanobodies. Scale bar, 10 μm. (F) Comparison of ERK-phosphoNanobody v2 and v3 in response to serum stimulation. Error bars represent mean ± s.d (v2 PRTP, n = 21 cells; v2 PRAP, n = 21 cells; v3 PRTP, n = 30 cells; v3 PRAP, n = 31 cells. Cells are quantified from 3 independent experiments). Statistical significance is verified with Mann-Whitney-Wilcoxon tests (two-sided with Bonferroni correction). p-values: v2-PRTP vs v2-PRAP, 2.28×10⁻⁴; v2-PRTP vs v3-PRTP, 0.505; v2-PRAP vs v3-PRAP, 1.95×10⁻³. (G) Representative ErkKTR images corresponding to Fig. 6n. Scale bar, 10 μm. (H) Quantification of PRTP/PRAP phosphoVav2-mCherry expression levels to confirm similar phosphoVav2 concentration in tested cells. Error bars represent mean ± s.d (PRTP, n = 22 cells; PRAP, n = 22 cells). Statistical significance is verified with Mann-Whitney-Wilcoxon tests (two-sided with Bonferroni correction). p-value: 0.545. p-value annotation: ns, p > 0.05; *, 0.01 < p <= 0.05; **, 10⁻³ < p <= 0.01; ***, 10⁻⁴ < p <= 10⁻³; ****, p <= 10⁻⁴.

We next tested whether the phospho-switch might also extend to other target proteins (Fig. 6e). We first turned to OptoNanobodies, nanobodies into which we previously inserted AsLOV2 to confer light-controlled binding to various target proteins: mCherry, EGFP and endogenous F-actin⁶⁶. We replaced AsLOV2 in the F-actin OptoNanobody with our v3 phospho-switch and introduced it into HEK293T cells that also expressed the ErkKTR biosensor to monitor ERK activity²⁹ (Fig. 6f,g). Serum stimulation of ERK-phosphoNanobody cells resulted in rapid TagRFP relocalization from the cytosol to the cell periphery (Fig.6h,i and Supplementary Video 2), consistent with nanobody binding to cortical F-actin⁶⁶. This effect was not observed in cells expressing a nonphosphorylatable nanobody variant and was also abrogated by treatment with 10 μM of the MEK inhibitor U0126, which blocked ERK activity as measured by redistribution of the ErkKTR biosensor (Fig. 6h,i, Extended Data Fig. 9c,d and Supplementary Videos 3–5). Quantification revealed that the phosphoNanobody was redistributed from cytosol to membrane within 6 min after serum stimulation, before partially adapting back to baseline within 10–15 min (Fig. 6i, left), closely matching the dynamics of ErkKTR translocation (Fig. 6j, left). These data suggest that, as with the ErkKTR biosensor, the ERK-phosphoNanobody rapidly tracks changes in ERK kinase activity. We observed more residual activity in the nonphosphorylatable ‘v2’ phosphoNanobody compared with the ‘v3’ variant (Extended Data Fig. 9e,f), suggesting again that optimization was crucial for phospho-switch performance across protein contexts.

Can the phospho-switch be used to rewire kinase activity to trigger a novel cellular response? To address this question, we targeted the Rac1 activator Vav2, a member of the guanine nucleotide exchange factor (GEF) family (Fig. 6k,l), based on a prior light-controlled Vav2 variant generated through allosteric AsLOV2 insertion¹⁶. We inserted the v3 ERK phospho-switch between amino acid positions 154 and 155 of the Vav2’s GTPase-binding DH domain (Fig. 6m), using a nonphosphorylatable variant (PRAP) as a negative control. ERK-phosphoVav2 cells exhibited dramatic ERK-dependent changes in cell morphology (Fig. 6n, Extended Data Fig. 9g,h and Supplementary Video 6). Within 10 min of EGF stimulation, cells began to exhibit a spreading phenotype that persisted for tens of minutes. The spreading phenotype was reversed within ~10 min after addition of MEK inhibitor, leading cells to adopt a constricted morphology with dynamic retractions. In nonphosphorylatable (PRAP) control cells, EGF stimulation produced a similar increase in ERK activity without an effect on cell shape (Fig. 6o,p, Extended Data Fig. 9g,h and Supplementary Video 7). These data confirm that the ERK kinase activity can be ‘rewired’ to directly control a novel cell phenotype: Rac1-dependent cell spreading.

Discussion

Building customizable input/output interfaces with the cell’s endogenous kinase networks is a grand challenge in synthetic biology, with diverse applications in cell engineering, biosensing and the detection and reversion of disease-associated cellular states. However, our ability to engineer novel phosphorylation-regulated processes in the cell is still limited. Given a kinase and target protein of interest, how might the synthetic biologist draw a new link to trigger an output protein’s activity in response to phosphorylation by an input kinase?

Here, we show that engineered phosphorylation-controlled allostery provides an attractive route toward addressing this question. Many kinase-specific phospho-switches are already available to convert a phosphorylation event to a conformational change in the biosensor²³, and engineered allostery has been successfully used to design stimulus-responsive variants of enzymes, kinases, transcription factors and protein binders^{16,17,25,66–69}. We link these two approaches and show that an optimized ERK phospho-switch can confer up to a 20-fold change in gene expression when inserted at a previously identified allosteric control point in the Gal4 DNA-binding domain. Our switch design also generalizes to multiple input kinases (ERK and JNK) and output proteins (Gal4, an F-actin nanobody and the Vav2 guanine exchange factor).

Our study reveals that the design constraints of allosteric phospho-switches differ from the FRET biosensors from which they are derived. Long, flexible linkers that perform well for FRET make poor allosteric switches, whereas short, rigid hinges achieve an optimal combination of high maximum activation and low leakiness. We further find that mutations that weaken binding affinity between the phospho-binding domain and unphosphorylated substrate can improve the phospho-switch’s dynamic range. Thus, while an ERK-responsive FRET biosensor serves as a useful starting point for switch engineering, optimal phospho-switch performance required substantial refinement.

One immediate application of our phosphoGal4 system is as a simple and specific biosensor of integrated kinase activity over time^70,71. Such a system could in principle be used for high-throughput, nondestructive isolation of ERK-active cell populations (for example, rare drug-resistant tumor cells with hyperactive kinase signaling)⁷² or as a component of more complex gene circuits to trace the future positions and fates of ERK-active cells during development^47,51. The ability to confer kinase-based regulation over target proteins could be broadly useful beyond biosensing. Kinase-directed signaling proteins could be harnessed as synthetic biology tools to trigger novel responses to a cell’s signaling state. We demonstrate this capability by rewiring cells’ endogenous ERK signaling to reversibly activate Rac1-dependent cell spreading. This principle could be further extended to misdirect tumor-associated ERK hyperactivity to cytostatic or apoptotic outcomes¹¹, or as a method to add feedback loops to intracellular signaling pathways to alter their signal processing⁷³.

This work is only a first step toward a fully generalizable approach in which any kinase could be ‘wired in’ to control any target protein. Our optimized phospho-switch is based on the WW domain, which is ideally suited for detecting phosphorylation by proline-directed kinases of the mitogen-activated protein kinase and cyclin-dependent kinase families. Additional phospho-binding domains (for example, FHA1; SH2) may be required to broaden the range of kinases and substrates that can be used^8,74,75. It may also be possible to extend WW-based switches to phosphorylated substrates beyond proline-directed targets, as the intramolecular configuration of the phospho-switch makes affinity less crucial than reduced ‘leaky’ binding to nonphosphorylated peptides. It will also be essential to broaden the set of effector proteins with sites for potent allosteric control^16,25,66,68. These tools and others may help to usher in a new era of kinase-controlled transcriptional responses and synthetic post-translational intracellular logic circuits.

Methods

Our research complies with all relevant ethical regulations (Princeton University Institutional Biosafety Committee).

Plasmid construction

Important plasmids used in this study are listed in Supplementary Table 3. pQC060 was constructed by cloning the wild-type Gal4-VP64 expression cassette from pHR_SFFV_Gal4_VP64_IRES_mCherry (Addgene, no. 203909) into a PiggyBac backbone. EKAR (88 amino acid hinge version) was amplified from pPBJ-EKAR-EV-NES (a gift from University of California-Davis, Albeck laboratory) with the 116 amino acid hinge shortened to 88 amino acis. pQC061 was constructed by inserting the EKAR amplicon into pQC060 at the Ser22/Lys23 allosteric site. pQC171/235 was constructed by replacing the original AsLOV2 insertion with ERK phospho-switch v2/v3 in pHR_SFFV_Actin_optoNB_tagRFP (Addgene, no. 159595). Phospho-site T-to-A mutations were introduced via site-directed mutagenesis. The no-VP64 control plasmid, pQC238, was constructed by backbone PCR to remove the VP64 coding sequence in 0_001_v2_a16b1c17 (this study). Plasmids discussed above were assembled using In-Fusion assembly (TaKaRa, no. 638911). All other plasmids were generated with the QCTK and will be separately discussed in following sections.

QCTK development

The QCTK for fast phospho-switch generation includes level 0, 1a, 1b and 1c parts (Fig. 2a). To generate level 0 parts, target protein expression vectors were first cloned with desired backbones, promoter sequences and expression markers. For example, a wild-type Gal4-VP64 (as the target protein) was first cloned into a PiggyBac vector with P_EF1a as the promoter and IRES-mCherry as the expression marker. The cloned expression vectors were opened at the allosteric insertion site with BsaI overhangs by PCR amplification. A bacterial EGFP expression cassette (BBa_J72163_P_GlpT_sfGFP) was amplified from pYTK001 (Addgene, no. 65108) with corresponding BsaI overhangs. BsaI_HFv2 (New England Biolabs, no. R3733S) digestion and ligation with T4 DNA Ligase (New England Biolabs, no. M0202S) was then conducted to insert the bacterial EGFP marker into the target protein expression vector and leave BsaI sites on both ends of the EGFP marker. All level 0 parts are carbenicillin-resistant, and transformants express EGFP on LB agar plates with carbenicillin (Gold Biotechnology, no. C-103-5).

To generate level 1 parts, pYTK001 was used as the backbone. Desired insert sequences were generated by PCR amplification, oligonucleotide annealing or DNA synthesis with corresponding 1a, 1b or 1c BsmBI overhangs. BsmBI_v2 (NEB, no. R0739S) digestion and T4 ligation was conducted to replace the bacterial EGFP marker in pYTK001 with target inserts. All level 1 parts are chloramphenicol-resistant and transformants were picked by EGFP counter-selection on chloramphenicol (Gold Biotechnology, no. C-105-5) plates. Standardized overhang sequences for level 0, 1a, 1b and 1c parts are shown in Fig. 2a. See Supplementary Table 1 (QCTK1b parts) and Supplementary Table 2 (QCTK1a/c parts) for lists of all QCTK parts used in this study.

Variant assembly with QCTK

Golden Gate assembly was performed in a one-pot reaction with level 0, 1a, 1b and 1c parts. Reaction mixes were prepared with all four DNA parts (50–100 ng each), 0.5 μl BsaI_HFv2, 0.5 μl T4 DNA ligase and 1 μl T4 DNA Ligase Reaction Buffer (New England Biolabs, no. M0202S) to a total volume of 6.5 μl. The reaction mix was incubated for 50× digestion/ligation cycles (2 min at 37 °C followed by 4 min at 16 °C), followed by 10 min of 37 °C digestion and 10 min of 80 °C heat inactivation. Reaction mixes were then transformed into Stellar Competent Cells (TaKaRa, 636766) and spread on carbenicillin plates. Transformants were picked by EGFP counter-selection. See Supplementary Table 3 for all important plasmids generated by QCTK Golden Gate assembly.

Cell culture

Lenti-X HEK293T cells and NIH 3T3 cells were maintained in DMEM (Gibco, no. 11995073) with 10% FBS (R&D Systems, no. S11150), 1% penicillin–streptomycin (Gibco, no. 15140-122), 2 mM l-Glutamine (Gibco, no. 25030-081). Stem cell basal growth medium was made with GMEM (Millipore Sigma, G6148), 10% ESC-qualified FBS (R&D Systems, S10250), 1× GlutaMAX (Gibco, 35050-061), 1× MEM nonessential amino acids (Gibco, 11140-470050), 1 mM sodium pyruvate (Gibco, 11360-070), 100 μM 2-mercaptoethanol (Gibco, 47121985-023) and 1% penicillin–streptomycin. The mESC line, E14tg2a (ATCC, CRL-1821), was maintained in 0.1% gelatin coated 25-cm² tissue culture flasks and cultured in 2i+LIF media comprising basal growth medium further supplemented with 1,000 units ml⁻¹ LIF (Millipore Sigma, ESG1107), 2 μM PD0325901 (Tocris, 4192) and 3 μM CHIR-99021 (Tocris, 4423). Cells were verified by STR profiling (ATCC). Cells were not cultured in proximity to commonly misidentified cell lines.

Cell transient transfection and ERK stimulation with conditioned media

HEK293T reporter cells were seeded in 48-well plates. Then, 24 h post-seeding, cells were transfected at 60–80% confluency. Transfection mixtures for each well were prepared in 15 µl Gibco OptiMEM (Gibco, 31985070) containing 2.5 µl FuGENE HD (Promega, no. E2311) and 150 ng target plasmid. Then, 24 h post-transfection, culture media was switched to serum-free media (DMEM with 1% penicillin–streptomycin and 0.00476 mg ml⁻¹ HEPES (Sigma-Aldrich, H4034)) or fresh growth media (as mentioned in the ‘Cell culture’ section). For reactivation, media was replaced with growth media at a defined time after the first media switch with serum-free media (Figs. 1f and 3h). For delivering multiple pulses of ERK activation with media changes, 75% of the existing medium in wells were replaced with fresh growth media at 6, 18 and 24 h after the first media change (0 h). All ERK-ON experiments (as seen in Figs. 1h, 2 and 3 and Extended Data Fig. 5) were done with four growth media changes in 30 h. After treatment, cells were then detached and resuspended for flow cytometry.

Cell line generation

Engineered HEK293T stable cell lines were generated using the piggyBac random integration system. A chassis cell line was first cultured to 60–80% confluency in 12-well plates. Transfection mixtures were prepared in 45 µl OptiMEM containing 4 µl FuGENE HD, 800 ng vector plasmid and 250 ng PBase ‘helper’ plasmid (System Biosciences). Transfection mixtures were equilibrated at room temperature for 15 min and then added dropwise to the culture. Cultures were propagated for at least 4 days post-transfection. Cells were then detached and resuspended for fluorescence-activated cell sorting (FACS). Sorted fluorescence-positive cells were collected for experiments.

Flow cytometry analysis and FACS

All flow cytometry/FACS experiments were conducted with a Sony SH800S equipped with Sony 100-μm Sorting Chip. EGFP was measured with 487.5-nm lasers and 525/50-nm filters. Similarly, mCherry with 561-nm lasers and 600/60-nm filters, and iRFP with 639-nm lasers and 720/60-nm filters. Gain and fluorescence compensation was performed following manufacturer’s instructions. Detailed strategies of flow cytometry gating, analysis and visualization are shown in Extended Data Fig. 1a. For flow cytometry quantifications, UAS-dEGFP values were measured as the mean of the EGFP readings for all gated cells in each replicate.

Protein structure prediction with AlphaFold3 and structural analysis

Phospho-switch inserted protein structures were predicted by AlphaFold3 with a default setting (seed: auto). The phosphorylated/unphosphorylated models were separately predicted. The phospho-state was defined in AlphaFold3 by adding phosphorylation on the threonine inside the PRTP motif. Three predicted phospho-protein structures were shown in this study (phosphoGal4, phosphoNanobody and phosphoVav2). To predict PhosphoGal4 models, phosphoGal4 (without VP64, two copies), dsDNA of UAS elements and zinc ions (four copies, two for each phosphoGal4 molecule) were fed into AlphaFold3. PhosphoNanobody and PhosphoVav2 structures were predicted with only the polypeptide sequences.

In Fig. 3d and Extended Data Fig. 4g,h, hydrogen bonds were predicted using UCSF ChimeraX with default setting (a distance tolerance of 0.400 Å and an angle tolerance of 20.000°). The search was limited, in that at least one end of the hydrogen bond needed to be in the phosphate. In Extended Data Fig. 4e,f, structural alignment and r.m.s. deviation calculations were performed with UCSF ChimeraX MatchMaker tool. In Extended Data Fig. 4d, distances between the Cα of indicated amino acids were measured with UCSF ChimeraX Distances tool.

Fold change calculation

Cells were first gated for IRES-mCherry expression level between 5 × 10⁴ and 1.25 × 10⁵, and the fold change of the phosphoGal4s were calculated as

FC = \frac{TP - ctrl}{AP - ctrl},

where FC is the fold change reported, TP is the response of phosphoGal4s with phosphorylatable substrate, AP is the response of phosphoGal4s with nonphosphorylatable substrate and ctrl is the response of the control phosphoGal4 lacking VP64.

Multivariable linear regression model to predict hinge performance

A multivariable linear regression model was built to use five selected hinge sequence features (length, A%, GS%, EK% and P%) as input variables and output the hinge performance (including the maximum activation in the ERK-ON condition and background activation in the ERK-OFF condition). The model was trained using experimentally tested hinges. Features extracted from experimentally tested hinges were tested negative for multicollinearity with all variance inflation factors <5. The trained model was then used to predict the performance of a library of >6,000 in silico generated hinge sequences. The library includes GS-rich flexible hinges, EAAAK-repeat alpha helical hinges, AP-repeat proline-rich hinges and computationally random mutated hinges at various lengths. Details of the model are explained in Supplementary Note 1 and shown in Extended Data Figs. 2 and 3.

Optogenetic activation of ERK-phosphoGal4 with optoWELL and flow cytometry

Polyclonal HEK293T cells stably expressing the optoSOS system and the ERK-phosphoGal4v2/UAS system (Fig. 4a,b) were seeded in glass-bottom, black 24-well plates (Cellvis, P24-1.5H-N) and kept in the dark. Then, 24 h after seeding, the culture was switched to serum-free media and then placed on a light-stimulation device (opto biolabs, optoWELL). Each well can be illuminated with 450-nm light at a defined intensity and duty cycle profile. Light intensity (in milliwatts per square centimeter) was measured with a power meter (Thorlabs, PM100D). After 22 h of light stimulation, cells were detached and analyzed by flow cytometry. For MEK inhibition, PD 0325901 (Tocris, 4192) was diluted in serum-free media at a final concentration of 10 μM and cells were kept in PD 0325901-contaning media during the 22 h stimulation. To supplement the system with zinc, additional ZnCl₂ (Sigma-Aldrich, 208086) was added to the serum-free media.

Signaling pathway stimulation

For short-term 4-h stimulation experiments (Fig. 5b,c and Extended Data Fig. 6g), stable cells (for example, polyclonal HEK293T cell lines harboring P_FOS-dEGFP or ERK-phosphoGal4v4a/UAS-dEGFP reporting system) were seeded in 48-well plates. Then, 24 h after seeding, cells were maintained in previously in-well media overnight (~18 h; cells adapt to a low level of basal ERK activity after sustained culture in growth media as we show in Extended Data Fig. 1b). Drugs used for stimulation were pre-diluted in serum-free media in a 2× concentration. Half of previously in-well media was replaced with 2× stimulation media. Then, 4 h after stimulation, cells were detached and analyzed by flow cytometry. The drugs used were human PDGF (Sigma-Aldrich, P3201), human EGF (R&D Systems, 236-EG-200), PD0325901 (MEKi; Tocris, 4192) and Gefitinib (EGFRi; Cell Signaling Technology, 4765).

For long-term 24 h stimulation experiments (Extended Data Figs. 6c,e,f,h and 9a,b), stable cells were seeded in 48-well plates. Then, 24 h after seeding, cells were switched to conditioned media containing various stimuli and inhibitors. Subsequently, 24 h after stimulation, cells were analyzed as in the short-term experiments. In Fig. 6c,d, cells were seeded and transfected as previously described. Then 24 h after transfection, cells were switched to conditioned media. In JNK-related experiments, the following drugs were used: anisomycin (Sigma-Aldrich, A9789) and JNK inhibitor VIII (Sigma-Aldrich, 420135).

Dose–response curve fitting

The dose–response curves in Figs. 4e and 5c were fit with a four-parameter logistic model. The model was defined as

y = D + \frac{A - D}{1 + {(\frac{x}{EC 50})}^{S}}

where D is the maximum response, A is the minimum response, EC₅₀ is the half-maximal effective concentration and S is Hill’s slope of the curve. The concentration or intensity of the input and the mean response were used to fit the model.

Gastruloid protocol

Gastruloids were grown in N2B27 media, a 1:1 mixture of DMEM/F-12 (Gibco, 11320033) and neurobasal medium (Gibco, 21103049) supplemented with 100 μM 2-mercaptoethanol, 1:100 N-2 (Gibco, 17502048), 1:50 B-27 (Gibco, 17504044) and 1:100 penicillin–streptomycin.

An mESC single-cell suspension was washed twice with DPBS to completely remove 2i+LIF media. The washed pellet was then resuspended in pre-warmed N2B27 media. Cells were then sorted into a 96-well, round-bottom, ultra-low attachment microplate (Corning, 7007). To minimize evaporation at the plate edges, the perimeter wells were filled with 100 μl water. The inner 60 wells were filled with 40 μl N2B27. In total, 200 mESCs were sorted into each inner well. At 48 h, 150 μl pre-warmed N2B27 with 3 μM CHIR-99021 was added to gastruloid-containing wells. At 72 h and 96 h, 150 μl of medium was removed and replaced with 150 μl fresh N2B27. Subsequently, 100 nM FGFRi PD173074 (Selleck Chemicals, S1264) treatments were carried out over 72–96 h.

Gastruloid fixation and immunofluorescence staining

To fix gastruloids, 150 μl of media was removed from each well and a 200-μl pipette with the tip cut off was used to transfer gastruloids to 4% paraformaldehyde (PFA) in PBS (Fisher Scientific, AAJ19943K2) under a dissection microscope. Gastruloids were fixed at 4 °C for 2 h. Then, 4% PFA was carefully aspirated and replaced with DPBS. Then, a 15 min wash was repeated three times. Fixed gastruloids were then imaged directly or prepared for immunostaining.

For immunostaining, PBS was completely removed from fixed gastruloids and replaced with PBSFT (10% FBS and 0.2% Triton X-100 (Millipore Sigma, 648466) in DPBS). Gastruloids were permeabilized in PBSFT overnight at 4 °C. The following day, the PBSFT was removed and replaced with the primary antibody diluted in PBSFT. Gastruloids were incubated in the primary antibody overnight at 4 °C. The following day the primary antibody was washed three times (∼15 min intervals) with PBSFT. After the final wash, the PBSFT was completely removed and replaced with the appropriate secondary antibody diluted in PBSFT. After overnight incubation, the same wash procedure was carried out. The antibodies used in this study were the goat anti-brachyury antibody (R&D Systems, AF2085; 10 μg ml⁻¹ working concentration) and the chicken anti-goat Alexa Fluor 594 conjugate (Invitrogen A21468; 1:500).

Gastruloid live imaging

Gastruloids were embedded in 50% Matrigel (Corning, 356231)/50% N2B27 by volume to restrict lateral movement during live imaging. Matrigel was thawed at 4 °C overnight. The next day, 50% Matrigel/50% N2B27 solution was prepared at 4 °C. This mixture was pipetted into the center of a 35-mm glass-bottom imaging dish (Ibidi, 81218). The desired amount of gastruloids was deposited in the Matrigel mixture and distributed evenly. The dish was incubated at 37 °C for 10 min for Matrigel solidification before 2 ml of N2B27 was then added to the dish. The media in the dish was covered with mineral oil (Sigma-Aldrich, 330779) to prevent evaporation. Images were acquired using a 20× air objective. z-Stacks were taken at each frame to cover the depth of gastruloids. Detailed microscopy information can be found the following live-cell imaging section.

Gastruloid image analysis and quantification

To quantify gastruloid imaging, a maximum-intensity z-projection was first obtained. A binary mask was hand-drawn to mark the boundary in the bright-field gastruloid images. Fluorescence intensities were measured as the mean inside the mask. To quantify fluorescence signal polarization, the center of mass of both the binary mask (as a unpolarized uniform shape) and the masked fluorescence channel was calculated. The distance shift between the two centers of mass was used to represent the degree of polarization. To quantify the fluorescence profile along the gastruloid major axis, a bounding box centered along the major axis was calculated for the mask. The bounding box was then sectioned to 10-pixel-wide small regions, and the mean pixel intensity of each region (gated by masks) was measured, thus providing an intensity profile at discrete points along the major axis. These profiles were averaged for multiple gastruloids to generate average plots with 95% confidential intervals. To calculate the spatial correlation between two fluorescence channels, pixel intensities of both channels were first normalized, and the Pearson correlation coefficient was calculated between the two channels.

Live-cell imaging

Cells were kept at 37 ° C with 5% CO₂ for the duration of all imaging experiments with an environmental control unit (Okolab). Mineral oil was pipetted onto the wells to prevent media evaporation. Imaging was done using Nikon Eclipse Ti microscope with a Prior linear motorized stage; a Yokogawa CSU-X1 spinning disk; an Agilent laser line module containing 405, 488, 561 and 650-nm lasers; an iXon DU897 EMCCD camera; and objective lenses as specified for individual experiment method sections. Images were acquired using NIS Elements v4.4 software (Nikon).

Optogenetic activation of ERK-phosphoGal4 with live-cell imaging

Polyclonal engineered HEK293T cells stably expressing both the optoSOS system and the ERK-phosphoGal4v2/UAS system (Fig. 4a,b) were seeded in glass-bottom, black 96-well plates (Cellvis, P96-1.5H-N) and kept in the dark in growth media. Then, 24 h after seeding, culture media was changed to serum-free media and the cells were starved in the dark for ~24 h to ensure a low baseline EGFP before imaging and stimulation. Cells were then imaged with the previously specified microscope configuration with a 20× air objective lens. A 450-nm LED light source was used for photoexcitation, which was delivered through a Polygon400 digital micro-mirror device (DMD, Mightex Systems). The LED power was adjusted to 75% to deliver ~15 mW cm⁻² blue light, as measured with a PM100D power meter (Thorlabs). Light was delivered as 1-s pulses of illumination every 1.5 min throughout 17 h of imaging. ErkKTR-iRFP was captured every 3 min, and EGFP induction and IRES-mCherry expression were captured every 1 h.

For ErkKTR quantification (Extended Data Fig. 7c), cells were segmented using Cellpose3⁷⁶. Fluorescence signals were measured inside individual cell masks. Inside the mask, the tagBFP signal was thresholded to generate a cytosolic mask owing to the natural cytosolic localization of the optoSOS system. Whole cell masks can then be further divided into cytosolic and nucleus masks. ErkKTR-iRFP intensity was measured separately in both regions for ERK activity calculation.

Live-cell imaging for the F-actin ERK-phosphoNanobody

HEK293T cells expressing ErkKTR-iRFP were seeded at low cell density in fibronectin-coated (Sigma-Aldrich, FC010) 96-well glass-bottom plates (fibronectin coating concentration of 5 μg cm⁻²). Then, 24 h after seeding, cells were transfected with F-actin ERK-phosphoNanobody expression plasmids using FuGENE HD (with the same procedure as previously described). Subsequently, 24 h after transfection, cells were starved overnight (>12 h) and then live imaged with a 60× oil immersion objective lens. Then, 2.5 min after the start of imaging, 2.5 μl FBS was added to activate ERK signaling. For MEK inhibition conditions, cells were pre-treated with U0126 (Cell Signaling Technology, 9903) at a 10 μM final concentration for >30 min prior serum stimulation. Images of ErkKTR-iRFP and nanobody-tagRFP were taken every 30 s for >20 min.

Live-cell imaging and analysis for ERK-phosphoVav2

HEK293T cells expressing ErkKTR-iRFP were seeded in six-well plates and transfected with ERK-phosphoVav2 expression plasmids using FuGENE HD (with the same procedure as previously described). Then, 24 h after transfection, cells were detached, resuspended and re-plated at very low density (~1,000 cells per well) in fibronectin-coated 96-well glass-bottom plates. Then, 6 h after re-plating, cells were visually checked to ensure most cells attached as a single cell with no neighbors touching. Cells were then starved for >2 h before imaging. Note, it is important to experiment on individual cells, as neighbor cells could alter the target cell morphology by cell–cell junctions and interfere with the cell area analysis.

Live-cell imaging was performed with a 60× oil immersion objective lens. The EGF or MEK inhibitor, U0126 was pre-diluted in 50 μl serum-free media as a 2× concentration. Then, 50 μl of existing 100 μl medium in each well was carefully removed and the 2× pre-diluted drug solution was then added. The final concentration was 100 ng ml⁻¹ EGF or 10 μM U0126. The time-lapses were first segmented on the phosphoVav2-mCherry channel with Cellpose3⁷⁶. Segmented mask stacks were then tracked using an intersection-over-union method to register the same cell across frames. Cell area profiles were then obtained from mask area profiles of each tracked cell.

Statistics and reproducibility

No data were excluded from the final analyses, the experiments were not randomized, the investigators were not blinded to allocation during experiments and outcome assessment, and no statistical methods were used to predetermine sample size. The reproducibility of experimental findings was verified across multiple independent sessions with consistent results. Statistical significance was determined using a two-tailed unpaired t-test or Mann–Whitney test, with P < 0.05 considered statistically significant. All data are presented as mean ± s.e.m. unless otherwise indicated.

Biological material availability

There are no restrictions on obtaining biological materials (cell lines and plasmids). Plasmids are available from Addgene (plasmid nos. 257515, 257516, 257517, 257518 and 257520), and cell lines will be available upon request from Jared Toettcher (toettcher@princeton.edu).

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41592-026-03163-1.

Supplementary information

Supplementary Information^{(461.4KB, pdf)}

Supplementary Notes 1–3, Video legends and Tables 1–4.

Reporting Summary^{(1.6MB, pdf)}

Supplementary Video 1^{(10.7MB, mp4)}

Live imaging of ERK-phosphoGal4 activity in elongating gastruloids. ERK-phosphoGal4v4a/UAS responses in the elongating gastruloid over time. Gastruloids were imaged for 17.5 h starting at 90 haa. 7.5 h after the start of imaging, the field of view was adjusted for the same gastruloid to prevent the elongating posterior tip to exceed the field. Left: UAS-dEGFP. Right panel: bright-field. Scale bar, 100 μm.

Supplementary Video 2^{(1.9MB, mov)}

F-actin ERK-phosphoNanobodyv3 (PRTP) and ErkKTR response to serum stimulation with no MEK inhibition. Cells were stimulated with serum 2.5 min post-imaging in the absence of MEK inhibition. Left: ERK-phosphoNanobodyv3-tagRFP (PRTP). Right panel: ErkKTR-iRFP. Scale bar, 20 μm.

Supplementary Video 3^{(1.7MB, mov)}

F-actin ERK-phosphoNanobodyv3 (PRTP) and ErkKTR response to serum stimulation with MEK inhibition. Cells were stimulated with serum 2.5 min post imaging in the presence of 10 μM U0126, a MEK inhibitor. Left panel: ERK-phosphoNanobodyv3-tagRFP (PRTP). Right: ErkKTR-iRFP. Scale bar, 20 μm.

Supplementary Video 4^{(1.9MB, mov)}

F-actin ERK-phosphoNanobodyv3 (PRAP) and ErkKTR response to serum stimulation with no MEK inhibition. Cells were stimulated with serum 2.5 min post imaging in the absence of MEK inhibition. Left panel: ERK-phosphoNanobodyv3-tagRFP (PRAP). Right: ErkKTR-iRFP. Scale bar, 20 μm.

Supplementary Video 5^{(1.9MB, mov)}

F-actin ERK-phosphoNanobodyv3 (PRAP) and ErkKTR response to serum stimulation with MEK inhibition. Cells were stimulated with serum 2.5 min post-imaging in the presence of 10 μM U0126, a MEK inhibitor. Left: ERK-phosphoNanobodyv3-tagRFP (PRAP). Right panel: ErkKTR-iRFP. Scale bar, 20 μm.

Supplementary Video 6^{(8.6MB, mp4)}

ERK-phosphoVav2 (PRTP) and ErkKTR response to EGF stimulation and MEK inhibition. Cells were stimulated with 100 ng ml⁻¹ EGF 5 min post-imaging. Then, 20 min after EGF stimulation, cells were treated with 10 μM U0126, a MEK inhibitor. Left: ERK-phosphoVav2-mCherry (PRTP). Right: ErkKTR-iRFP. Scale bar, 10 μm.

Supplementary Video 7^{(5MB, mp4)}

ERK-phosphoVav2 (PRAP) and ErkKTR response to EGF stimulation and MEK inhibition. Cells were stimulated with 100 ng ml⁻¹ EGF 5 min post-imaging. Then, 20 min after EGF stimulation, cells were treated with 10 μM U0126, a MEK inhibitor. Left: ERK-phosphoVav2-mCherry (PRAP). Right: ErkKTR-iRFP. Scale bar, 10 μm.

Source data

Source Data Figs. 1–6^{(464KB, xlsx)}

Source data for all data elements in the figures.

Source Data Extended Data Figs. 1–9^{(677.2KB, xlsx)}

Source data for all data elements in the figures.

Acknowledgements

We thank all members of the Toettcher laboratory for helpful discussions. We also thank B. Ramm, E. Kolenbrander Ho and L. Nguyen for comments on the manuscript. This work was supported by the National Institutes of Health (grants no. R01GM144362 and R35GM164185) and NSF RECODE (grant no. 2134935 to J.E.T.).

Extended data

Author contributions

Conceptualization by Q.C. and J.E.T. Methodology by Q.C. and J.E.T. Investigation by Q.C. Funding by J.E.T. Writing and editing by Q.C. and J.E.T. Supervision by J.E.T.

Peer review

Peer review information

Nature Methods thanks Nikolay Dokholyan and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Primary Handling Editor: Rita Strack and Madhura Mukhopadhyay, in collaboration with the Nature Methods team.

Data availability

There are no restrictions on data availability and additional requests will be fulfilled by the corresponding author (toettcher@princeton.edu). Source data are provided with this paper.

Code availability

The regression model used to predict hinge sequences and the flow cytometry data are available via GitHub at https://github.com/toettchlab/Cao2025. No other code was developed as part of this manuscript.

Competing interests

J.E.T. is a scientific advisor for Prolific Machines and Nereid Therapeutics. The authors have also submitted a provisional patent application related to kinase-controlled protein switches. The other authors declare no competing interests.

Footnotes

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

is available for this paper at 10.1038/s41592-026-03163-1.

Supplementary information

The online version contains supplementary material available at 10.1038/s41592-026-03163-1.

References

1.Venta, R. et al. A processive phosphorylation circuit with multiple kinase inputs and mutually diversional routes controls G1/S decision. Nat. Commun.11, 1836 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
2.Huang, C. Y. & Ferrell, J. E. Ultrasensitivity in the mitogen-activated protein kinase cascade. Proc. Natl Acad. Sci. USA93, 10078–10083 (1996). [DOI] [PMC free article] [PubMed] [Google Scholar]
3.Barkai, N. & Leibler, S. Robustness in simple biochemical networks. Nature387, 913–917 (1997). [DOI] [PubMed] [Google Scholar]
4.Elowitz, M. B. & Leibler, S. A synthetic oscillatory network of transcriptional regulators. Nature403, 335–338 (2000). [DOI] [PubMed] [Google Scholar]
5.Gardner, T. S., Cantor, C. R. & Collins, J. J. Construction of a genetic toggle switch in Escherichia coli. Nature403, 339–342 (2000). [DOI] [PubMed] [Google Scholar]
6.Chen, Z. & Elowitz, M. B. Programmable protein circuit design. Cell184, 2284–2301 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Gao, X. J., Chong, L. S., Kim, M. S. & Elowitz, M. B. Programmable protein circuits in living cells. Science361, 1252–1258 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
8.Yang, X. et al. Engineering synthetic phosphorylation signaling networks in human cells. Science387, 74–81 (2025). [DOI] [PubMed] [Google Scholar]
9.Pincus, D. et al. Engineering allosteric regulation in protein kinases. Sci. Signal.11, eaar3250 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Gordley, R. M. et al. Engineering dynamical control of cell fate switching using synthetic phospho-regulons. Proc. Natl Acad. Sci. USA113, 13528–13533 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
11.Chung, H. K. et al. A compact synthetic pathway rewires cancer signaling to therapeutic effector release. Science364, eaat6982 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
12.Chen, Z. et al. A synthetic protein-level neural network in mammalian cells. Science386, 1243–1250 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Samaniego, C. C., Moorman, A., Giordano, G. & Franco, E. Signaling-based neural networks for cellular computation. In 2021 American Control Conference 1883–1890 (IEEE, 2021).
14.Daniels, K. G. et al. Decoding CAR T cell phenotype using combinatorial signaling motif libraries and machine learning. Science378, 1194–1200 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
15.Panigaj, M. et al. Autonomous nucleic acid and protein nanocomputing agents engineered to operate in living cells. ACS Nano19, 1865–1883 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
16.Dagliyan, O. et al. Engineering extrinsic disorder to control protein activity in living cells. Science354, 1441–1444 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
17.Dagliyan, O. et al. Rational design of a ligand-controlled protein conformational switch. Proc. Natl Acad. Sci. USA110, 6800–6804 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
18.Vishweshwaraiah, Y. L., Chen, J., Chirasani, V. R., Tabdanov, E. D. & Dokholyan, N. V. Two-input protein logic gate for computation in living cells. Nat. Commun.12, 6615 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Chen, J., Vishweshwaraiah, Y. L., Mailman, R. B., Tabdanov, E. D. & Dokholyan, N. V. A noncommutative combinatorial protein logic circuit controls cell orientation in nanoenvironments. Sci. Adv.9, eadg1062 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
20.Stein, V. & Alexandrov, K. Synthetic protein switches: design principles and applications. Trends Biotechnol.33, 101–110 (2015). [DOI] [PubMed] [Google Scholar]
21.Harvey, C. D. et al. A genetically encoded fluorescent sensor of ERK activity. Proc. Natl Acad. Sci. USA105, 19264–19269 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Komatsu, N. et al. Development of an optimized backbone of FRET biosensors for kinases and GTPases. Mol. Biol. Cell22, 4647–4656 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Greenwald, E. C., Mehta, S. & Zhang, J. Genetically encoded fluorescent biosensors illuminate the spatiotemporal regulation of signaling networks. Chem. Rev.118, 11707–11794 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
24.Gest, A. M. M. et al. Molecular spies in action: genetically encoded fluorescent biosensors light up cellular signals. Chem. Rev.124, 12573–12660 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
25.Zhu, L., McNamara, H. M. & Toettcher, J. E. Light-switchable transcription factors obtained by direct screening in mammalian cells. Nat. Commun.14, 3185 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
26.Goglia, A. G. et al. A live-cell screen for altered Erk dynamics reveals principles of proliferative control. Cell Syst.10, 240–253 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
27.Albeck, J. G., Mills, G. B. & Brugge, J. S. Frequency-modulated pulses of ERK activity transmit quantitative proliferation signals. Mol. Cell49, 249–261 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
28.Aoki, K. et al. Stochastic ERK activation induced by noise and cell-to-cell propagation regulates cell density-dependent proliferation. Mol. Cell52, 529–540 (2013). [DOI] [PubMed] [Google Scholar]
29.Regot, S., Hughey, J. J., Bajar, B. T., Carrasco, S. & Covert, M. W. High-sensitivity measurements of multiple kinase activities in live single cells. Cell157, 1724–1734 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
30.Sato, M., Ueda, Y., Takagi, T. & Umezawa, Y. Production of PtdInsP3 at endomembranes is triggered by receptor endocytosis. Nat. Cell Biol.5, 1016–1022 (2003). [DOI] [PubMed] [Google Scholar]
31.Merutka, G., Shalongo, W. & Stellwagen, E. A model peptide with enhanced helicity. Biochemistry10.1021/bi00231a020 (1991). [DOI] [PubMed]
32.Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature630, 493–500 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
33.Macias, M. J. et al. Structure of the WW domain of a kinase-associated protein complexed with a proline-rich peptide. Nature382, 646–649 (1996). [DOI] [PubMed] [Google Scholar]
34.Aragón, E. et al. A Smad action turnover switch operated by WW domain readers of a phosphoserine code. Genes Dev.25, 1275–1288 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
35.Ranganathan, R., Lu, K. P., Hunter, T. & Noel, J. P. Structural and functional analysis of the mitotic rotamase Pin1 suggests substrate recognition is phosphorylation dependent. Cell89, 875–886 (1997). [DOI] [PubMed] [Google Scholar]
36.Wintjens, R. et al. 1H NMR study on the binding of Pin1 Trp-Trp domain with phosphothreonine peptides. J. Biol. Chem.276, 25150–25156 (2001). [DOI] [PubMed] [Google Scholar]
37.Kato, Y., Ito, M., Kawai, K., Nagata, K. & Tanokura, M. Determinants of ligand specificity in groups I and IV WW domains as studied by surface plasmon resonance and model building. J. Biol. Chem.277, 10173–10177 (2002). [DOI] [PubMed] [Google Scholar]
38.Kato, Y. et al. Common mechanism of ligand recognition by group II/III WW domains: redefining their functional classification. J. Biol. Chem.279, 31833–31841 (2004). [DOI] [PubMed] [Google Scholar]
39.Verdecia, M. A., Bowman, M. E., Lu, K. P., Hunter, T. & Noel, J. P. Structural basis for phosphoserine-proline recognition by group IV WW domains. Nat. Struct. Mol. Biol.7, 639–643 (2000). [DOI] [PubMed] [Google Scholar]
40.Kliche, J. et al. Large-scale phosphomimetic screening identifies phospho-modulated motif-based protein interactions. Mol. Syst. Biol.19, MSB202211164 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
41.Toettcher, J. E., Weiner, O. D. & Lim, W. A. Using optogenetics to interrogate the dynamic control of signal transmission by the Ras/Erk module. Cell155, 1422–1434 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]
42.Johnson, H. E. et al. The spatiotemporal limits of developmental Erk signaling. Dev. Cell40, 185–192 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
43.Guntas, G. et al. Engineering an improved light-induced dimer (iLID) for controlling the localization and activity of signaling proteins. Proc. Natl Acad. Sci. USA112, 112–117 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
44.Hong, M. et al. Structural basis for dimerization in DNA recognition by Gal4. Structure16, 1019–1026 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
45.Glassman, A. B., Rydzewski, R. S. & Bennett, C. E. Trace metal levels in commercially prepared tissue culture media. Tissue Cell12, 613–617 (1980). [DOI] [PubMed] [Google Scholar]
46.Lo, M. N., Damon, L. J., Wei Tay, J., Jia, S. & Palmer, A. E. Single cell analysis reveals multiple requirements for zinc in the mammalian cell cycle. Elife9, e51107 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
47.McNamara, H. M., Solley, S. C., Adamson, B., Chan, M. M. & Toettcher, J. E. Recording morphogen signals reveals mechanisms underlying gastruloid symmetry breaking. Nat. Cell Biol.10.1038/s41556-024-01521-9 (2024). [DOI] [PMC free article] [PubMed]
48.Inglés-Prieto, Á et al. Light-assisted small-molecule screening against protein kinases. Nat. Chem. Biol.11, 952–954 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
49.Durick, K. & Negulescu, P. Cellular biosensors for drug discovery. Biosens. Bioelectron.16, 587–592 (2001). [DOI] [PubMed] [Google Scholar]
50.Ravindran, P. T., McFann, S., Thornton, R. H. & Toettcher, J. E. A synthetic gene circuit for imaging-free detection of signaling pulses. Cell Syst.13, 131–142 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
51.Chen, W. et al. Symbolic recording of signalling and cis-regulatory element activity to DNA. Nature632, 1073–1081 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
52.Serup, P. et al. Partial promoter substitutions generating transcriptional sentinels of diverse signaling pathways in embryonic stem cells and mice. Dis. Model. Mech.5, 956–966 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]
53.Joo, J.-Y., Schaukowitch, K., Farbiak, L., Kilaru, G. & Kim, T.-K. Stimulus-specific combinatorial functionality of neuronal c-fos enhancers. Nat. Neurosci.19, 75–83 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]
54.Morgani, S. M. et al. A Sprouty4 reporter to monitor FGF/ERK signaling activity in ESCs and mice. Dev. Biol.441, 104–126 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
55.Ghosh, A., Ginty, D. D., Bading, H. & Greenberg, M. E. Calcium regulation of gene expression in neuronal cells. J. Neurobiol.25, 294–303 (1994). [DOI] [PubMed] [Google Scholar]
56.Chung, L. A brief introduction to the transduction of neural activity into Fos signal. Dev. Reprod.19, 61–67 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
57.MacKeigan, J. P., Murphy, L. O., Dimitri, C. A. & Blenis, J. Graded mitogen-activated protein kinase activity precedes switch-like c-fos induction in mammalian cells. Mol. Cell. Biol.25, 4676–4682 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]
58.Herschman, H. R. Primary response genes induced by growth factors and tumor promoters. Annu. Rev. Biochem.60, 281–319 (1991). [DOI] [PubMed] [Google Scholar]
59.Janknecht, R., Cahill, M. A. & Nordheim, A. Signal integration at the c-fos promoter. Carcinogenesis16, 443–450 (1995). [DOI] [PubMed] [Google Scholar]
60.Yusa, K., Zhou, L., Li, M. A., Bradley, A. & Craig, N. L. A hyperactive piggyBac transposase for mammalian applications. Proc. Natl Acad. Sci. USA108, 1531–1536 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
61.van den Brink, S. C. et al. Symmetry breaking, germ layer specification and axial organisation in aggregates of mouse embryonic stem cells. Development141, 4231–4242 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]
62.Beccari, L. et al. Multi-axial self-organization properties of mouse embryonic stem cells into gastruloids. Nature562, 272–276 (2018). [DOI] [PubMed] [Google Scholar]
63.Underhill, E. J. & Toettcher, J. E. Control of gastruloid patterning and morphogenesis by the Erk and Akt signaling pathways. Development150, dev201663 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
64.Bubeck, F. et al. Engineered anti-CRISPR proteins for optogenetic control of CRISPR-Cas9. Nat. Methods15, 924–927 (2018). [DOI] [PubMed] [Google Scholar]
65.Wolf, B. et al. Rational engineering of allosteric protein switches by in silico prediction of domain insertion sites. Nat. Methods22, 1698–1706 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
66.Gil, A. A. et al. Optogenetic control of protein binding using light-switchable nanobodies. Nat. Commun.11, 4044 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
67.Karginov, A. V., Ding, F., Kota, P., Dokholyan, N. V. & Hahn, K. M. Engineered allosteric activation of kinases in living cells. Nat. Biotechnol.28, 743–747 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]
68.Reynolds, K. A., McLaughlin, R. N. & Ranganathan, R. Hot spots for allosteric regulation on protein surfaces. Cell147, 1564–1575 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]
69.Carrasco-Lopez, C. et al. Development of light-responsive protein binding in the monobody non-immunoglobulin scaffold. Nat. Commun.11, 4045 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
70.Lin, W. et al. Light-gated integrator for highlighting kinase activity in living cells. Nat. Commun.15, 7804 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]
71.Sun, D. et al. Molecular recording of cellular protein kinase activity with chemical labeling. Nat. Chem. Biol.21, 1818–1827 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]
72.Gerosa, L. et al. Receptor-driven ERK pulses reconfigure MAPK signaling and enable persistence of drug-adapted BRAF-mutant melanoma cells. Cell Syst.11, 478–494 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
73.Mitchell, A., Wei, P. & Lim, W. A. Oscillatory stress stimulation uncovers an Achilles’ heel of the yeast MAPK signaling network. Science350, 1379–1383 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
74.Farahani, P. E. et al. pYtags enable spatiotemporal measurements of receptor tyrosine kinase signaling in living cells. Elife12, e82863 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
75.Gao, X. & Zhang, J. Spatiotemporal analysis of differential Akt regulation in plasma membrane microdomains. Mol. Biol. Cell19, 4366–4373 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]
76.Stringer, C. & Pachitariu, M. Cellpose3: one-click image restoration for improved cellular segmentation. Nat. Methods22, 592–599 (2025). [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

Supplementary Information^{(461.4KB, pdf)}

Supplementary Notes 1–3, Video legends and Tables 1–4.

Reporting Summary^{(1.6MB, pdf)}

Supplementary Video 1^{(10.7MB, mp4)}

Supplementary Video 2^{(1.9MB, mov)}

Supplementary Video 3^{(1.7MB, mov)}

Supplementary Video 4^{(1.9MB, mov)}

Supplementary Video 5^{(1.9MB, mov)}

Supplementary Video 6^{(8.6MB, mp4)}

Supplementary Video 7^{(5MB, mp4)}

Source Data Figs. 1–6^{(464KB, xlsx)}

Source data for all data elements in the figures.

Source Data Extended Data Figs. 1–9^{(677.2KB, xlsx)}

Source data for all data elements in the figures.

Data Availability Statement

There are no restrictions on data availability and additional requests will be fulfilled by the corresponding author (toettcher@princeton.edu). Source data are provided with this paper.

[CR1] 1.Venta, R. et al. A processive phosphorylation circuit with multiple kinase inputs and mutually diversional routes controls G1/S decision. Nat. Commun.11, 1836 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR2] 2.Huang, C. Y. & Ferrell, J. E. Ultrasensitivity in the mitogen-activated protein kinase cascade. Proc. Natl Acad. Sci. USA93, 10078–10083 (1996). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR3] 3.Barkai, N. & Leibler, S. Robustness in simple biochemical networks. Nature387, 913–917 (1997). [DOI] [PubMed] [Google Scholar]

[CR4] 4.Elowitz, M. B. & Leibler, S. A synthetic oscillatory network of transcriptional regulators. Nature403, 335–338 (2000). [DOI] [PubMed] [Google Scholar]

[CR5] 5.Gardner, T. S., Cantor, C. R. & Collins, J. J. Construction of a genetic toggle switch in Escherichia coli. Nature403, 339–342 (2000). [DOI] [PubMed] [Google Scholar]

[CR6] 6.Chen, Z. & Elowitz, M. B. Programmable protein circuit design. Cell184, 2284–2301 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR7] 7.Gao, X. J., Chong, L. S., Kim, M. S. & Elowitz, M. B. Programmable protein circuits in living cells. Science361, 1252–1258 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR8] 8.Yang, X. et al. Engineering synthetic phosphorylation signaling networks in human cells. Science387, 74–81 (2025). [DOI] [PubMed] [Google Scholar]

[CR9] 9.Pincus, D. et al. Engineering allosteric regulation in protein kinases. Sci. Signal.11, eaar3250 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR10] 10.Gordley, R. M. et al. Engineering dynamical control of cell fate switching using synthetic phospho-regulons. Proc. Natl Acad. Sci. USA113, 13528–13533 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR11] 11.Chung, H. K. et al. A compact synthetic pathway rewires cancer signaling to therapeutic effector release. Science364, eaat6982 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR12] 12.Chen, Z. et al. A synthetic protein-level neural network in mammalian cells. Science386, 1243–1250 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR13] 13.Samaniego, C. C., Moorman, A., Giordano, G. & Franco, E. Signaling-based neural networks for cellular computation. In 2021 American Control Conference 1883–1890 (IEEE, 2021).

[CR14] 14.Daniels, K. G. et al. Decoding CAR T cell phenotype using combinatorial signaling motif libraries and machine learning. Science378, 1194–1200 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR15] 15.Panigaj, M. et al. Autonomous nucleic acid and protein nanocomputing agents engineered to operate in living cells. ACS Nano19, 1865–1883 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR16] 16.Dagliyan, O. et al. Engineering extrinsic disorder to control protein activity in living cells. Science354, 1441–1444 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR17] 17.Dagliyan, O. et al. Rational design of a ligand-controlled protein conformational switch. Proc. Natl Acad. Sci. USA110, 6800–6804 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR18] 18.Vishweshwaraiah, Y. L., Chen, J., Chirasani, V. R., Tabdanov, E. D. & Dokholyan, N. V. Two-input protein logic gate for computation in living cells. Nat. Commun.12, 6615 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR19] 19.Chen, J., Vishweshwaraiah, Y. L., Mailman, R. B., Tabdanov, E. D. & Dokholyan, N. V. A noncommutative combinatorial protein logic circuit controls cell orientation in nanoenvironments. Sci. Adv.9, eadg1062 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR20] 20.Stein, V. & Alexandrov, K. Synthetic protein switches: design principles and applications. Trends Biotechnol.33, 101–110 (2015). [DOI] [PubMed] [Google Scholar]

[CR21] 21.Harvey, C. D. et al. A genetically encoded fluorescent sensor of ERK activity. Proc. Natl Acad. Sci. USA105, 19264–19269 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR22] 22.Komatsu, N. et al. Development of an optimized backbone of FRET biosensors for kinases and GTPases. Mol. Biol. Cell22, 4647–4656 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR23] 23.Greenwald, E. C., Mehta, S. & Zhang, J. Genetically encoded fluorescent biosensors illuminate the spatiotemporal regulation of signaling networks. Chem. Rev.118, 11707–11794 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR24] 24.Gest, A. M. M. et al. Molecular spies in action: genetically encoded fluorescent biosensors light up cellular signals. Chem. Rev.124, 12573–12660 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR25] 25.Zhu, L., McNamara, H. M. & Toettcher, J. E. Light-switchable transcription factors obtained by direct screening in mammalian cells. Nat. Commun.14, 3185 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR26] 26.Goglia, A. G. et al. A live-cell screen for altered Erk dynamics reveals principles of proliferative control. Cell Syst.10, 240–253 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR27] 27.Albeck, J. G., Mills, G. B. & Brugge, J. S. Frequency-modulated pulses of ERK activity transmit quantitative proliferation signals. Mol. Cell49, 249–261 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR28] 28.Aoki, K. et al. Stochastic ERK activation induced by noise and cell-to-cell propagation regulates cell density-dependent proliferation. Mol. Cell52, 529–540 (2013). [DOI] [PubMed] [Google Scholar]

[CR29] 29.Regot, S., Hughey, J. J., Bajar, B. T., Carrasco, S. & Covert, M. W. High-sensitivity measurements of multiple kinase activities in live single cells. Cell157, 1724–1734 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR30] 30.Sato, M., Ueda, Y., Takagi, T. & Umezawa, Y. Production of PtdInsP3 at endomembranes is triggered by receptor endocytosis. Nat. Cell Biol.5, 1016–1022 (2003). [DOI] [PubMed] [Google Scholar]

[CR31] 31.Merutka, G., Shalongo, W. & Stellwagen, E. A model peptide with enhanced helicity. Biochemistry10.1021/bi00231a020 (1991). [DOI] [PubMed]

[CR32] 32.Abramson, J. et al. Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature630, 493–500 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR33] 33.Macias, M. J. et al. Structure of the WW domain of a kinase-associated protein complexed with a proline-rich peptide. Nature382, 646–649 (1996). [DOI] [PubMed] [Google Scholar]

[CR34] 34.Aragón, E. et al. A Smad action turnover switch operated by WW domain readers of a phosphoserine code. Genes Dev.25, 1275–1288 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR35] 35.Ranganathan, R., Lu, K. P., Hunter, T. & Noel, J. P. Structural and functional analysis of the mitotic rotamase Pin1 suggests substrate recognition is phosphorylation dependent. Cell89, 875–886 (1997). [DOI] [PubMed] [Google Scholar]

[CR36] 36.Wintjens, R. et al. 1H NMR study on the binding of Pin1 Trp-Trp domain with phosphothreonine peptides. J. Biol. Chem.276, 25150–25156 (2001). [DOI] [PubMed] [Google Scholar]

[CR37] 37.Kato, Y., Ito, M., Kawai, K., Nagata, K. & Tanokura, M. Determinants of ligand specificity in groups I and IV WW domains as studied by surface plasmon resonance and model building. J. Biol. Chem.277, 10173–10177 (2002). [DOI] [PubMed] [Google Scholar]

[CR38] 38.Kato, Y. et al. Common mechanism of ligand recognition by group II/III WW domains: redefining their functional classification. J. Biol. Chem.279, 31833–31841 (2004). [DOI] [PubMed] [Google Scholar]

[CR39] 39.Verdecia, M. A., Bowman, M. E., Lu, K. P., Hunter, T. & Noel, J. P. Structural basis for phosphoserine-proline recognition by group IV WW domains. Nat. Struct. Mol. Biol.7, 639–643 (2000). [DOI] [PubMed] [Google Scholar]

[CR40] 40.Kliche, J. et al. Large-scale phosphomimetic screening identifies phospho-modulated motif-based protein interactions. Mol. Syst. Biol.19, MSB202211164 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR41] 41.Toettcher, J. E., Weiner, O. D. & Lim, W. A. Using optogenetics to interrogate the dynamic control of signal transmission by the Ras/Erk module. Cell155, 1422–1434 (2013). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR42] 42.Johnson, H. E. et al. The spatiotemporal limits of developmental Erk signaling. Dev. Cell40, 185–192 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR43] 43.Guntas, G. et al. Engineering an improved light-induced dimer (iLID) for controlling the localization and activity of signaling proteins. Proc. Natl Acad. Sci. USA112, 112–117 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR44] 44.Hong, M. et al. Structural basis for dimerization in DNA recognition by Gal4. Structure16, 1019–1026 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR45] 45.Glassman, A. B., Rydzewski, R. S. & Bennett, C. E. Trace metal levels in commercially prepared tissue culture media. Tissue Cell12, 613–617 (1980). [DOI] [PubMed] [Google Scholar]

[CR46] 46.Lo, M. N., Damon, L. J., Wei Tay, J., Jia, S. & Palmer, A. E. Single cell analysis reveals multiple requirements for zinc in the mammalian cell cycle. Elife9, e51107 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR47] 47.McNamara, H. M., Solley, S. C., Adamson, B., Chan, M. M. & Toettcher, J. E. Recording morphogen signals reveals mechanisms underlying gastruloid symmetry breaking. Nat. Cell Biol.10.1038/s41556-024-01521-9 (2024). [DOI] [PMC free article] [PubMed]

[CR48] 48.Inglés-Prieto, Á et al. Light-assisted small-molecule screening against protein kinases. Nat. Chem. Biol.11, 952–954 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR49] 49.Durick, K. & Negulescu, P. Cellular biosensors for drug discovery. Biosens. Bioelectron.16, 587–592 (2001). [DOI] [PubMed] [Google Scholar]

[CR50] 50.Ravindran, P. T., McFann, S., Thornton, R. H. & Toettcher, J. E. A synthetic gene circuit for imaging-free detection of signaling pulses. Cell Syst.13, 131–142 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR51] 51.Chen, W. et al. Symbolic recording of signalling and cis-regulatory element activity to DNA. Nature632, 1073–1081 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR52] 52.Serup, P. et al. Partial promoter substitutions generating transcriptional sentinels of diverse signaling pathways in embryonic stem cells and mice. Dis. Model. Mech.5, 956–966 (2012). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR53] 53.Joo, J.-Y., Schaukowitch, K., Farbiak, L., Kilaru, G. & Kim, T.-K. Stimulus-specific combinatorial functionality of neuronal c-fos enhancers. Nat. Neurosci.19, 75–83 (2016). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR54] 54.Morgani, S. M. et al. A Sprouty4 reporter to monitor FGF/ERK signaling activity in ESCs and mice. Dev. Biol.441, 104–126 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR55] 55.Ghosh, A., Ginty, D. D., Bading, H. & Greenberg, M. E. Calcium regulation of gene expression in neuronal cells. J. Neurobiol.25, 294–303 (1994). [DOI] [PubMed] [Google Scholar]

[CR56] 56.Chung, L. A brief introduction to the transduction of neural activity into Fos signal. Dev. Reprod.19, 61–67 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR57] 57.MacKeigan, J. P., Murphy, L. O., Dimitri, C. A. & Blenis, J. Graded mitogen-activated protein kinase activity precedes switch-like c-fos induction in mammalian cells. Mol. Cell. Biol.25, 4676–4682 (2005). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR58] 58.Herschman, H. R. Primary response genes induced by growth factors and tumor promoters. Annu. Rev. Biochem.60, 281–319 (1991). [DOI] [PubMed] [Google Scholar]

[CR59] 59.Janknecht, R., Cahill, M. A. & Nordheim, A. Signal integration at the c-fos promoter. Carcinogenesis16, 443–450 (1995). [DOI] [PubMed] [Google Scholar]

[CR60] 60.Yusa, K., Zhou, L., Li, M. A., Bradley, A. & Craig, N. L. A hyperactive piggyBac transposase for mammalian applications. Proc. Natl Acad. Sci. USA108, 1531–1536 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR61] 61.van den Brink, S. C. et al. Symmetry breaking, germ layer specification and axial organisation in aggregates of mouse embryonic stem cells. Development141, 4231–4242 (2014). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR62] 62.Beccari, L. et al. Multi-axial self-organization properties of mouse embryonic stem cells into gastruloids. Nature562, 272–276 (2018). [DOI] [PubMed] [Google Scholar]

[CR63] 63.Underhill, E. J. & Toettcher, J. E. Control of gastruloid patterning and morphogenesis by the Erk and Akt signaling pathways. Development150, dev201663 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR64] 64.Bubeck, F. et al. Engineered anti-CRISPR proteins for optogenetic control of CRISPR-Cas9. Nat. Methods15, 924–927 (2018). [DOI] [PubMed] [Google Scholar]

[CR65] 65.Wolf, B. et al. Rational engineering of allosteric protein switches by in silico prediction of domain insertion sites. Nat. Methods22, 1698–1706 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR66] 66.Gil, A. A. et al. Optogenetic control of protein binding using light-switchable nanobodies. Nat. Commun.11, 4044 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR67] 67.Karginov, A. V., Ding, F., Kota, P., Dokholyan, N. V. & Hahn, K. M. Engineered allosteric activation of kinases in living cells. Nat. Biotechnol.28, 743–747 (2010). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR68] 68.Reynolds, K. A., McLaughlin, R. N. & Ranganathan, R. Hot spots for allosteric regulation on protein surfaces. Cell147, 1564–1575 (2011). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR69] 69.Carrasco-Lopez, C. et al. Development of light-responsive protein binding in the monobody non-immunoglobulin scaffold. Nat. Commun.11, 4045 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR70] 70.Lin, W. et al. Light-gated integrator for highlighting kinase activity in living cells. Nat. Commun.15, 7804 (2024). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR71] 71.Sun, D. et al. Molecular recording of cellular protein kinase activity with chemical labeling. Nat. Chem. Biol.21, 1818–1827 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR72] 72.Gerosa, L. et al. Receptor-driven ERK pulses reconfigure MAPK signaling and enable persistence of drug-adapted BRAF-mutant melanoma cells. Cell Syst.11, 478–494 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR73] 73.Mitchell, A., Wei, P. & Lim, W. A. Oscillatory stress stimulation uncovers an Achilles’ heel of the yeast MAPK signaling network. Science350, 1379–1383 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR74] 74.Farahani, P. E. et al. pYtags enable spatiotemporal measurements of receptor tyrosine kinase signaling in living cells. Elife12, e82863 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR75] 75.Gao, X. & Zhang, J. Spatiotemporal analysis of differential Akt regulation in plasma membrane microdomains. Mol. Biol. Cell19, 4366–4373 (2008). [DOI] [PMC free article] [PubMed] [Google Scholar]

[CR76] 76.Stringer, C. & Pachitariu, M. Cellpose3: one-click image restoration for improved cellular segmentation. Nat. Methods22, 592–599 (2025). [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

Design and optimization of a kinase-controlled allosteric switch

Qinhao Cao

Jared E Toettcher

Abstract

Main

Fig. 1. Repurposing FRET kinase activity biosensors as allosteric switches.

Results

Repurposing FRET-based KARs as allosteric switches

Extended Data Fig. 1. Flow cytometry gating strategies and ERK activity adaptation/reactivation in HEK293T cells.

Hinge optimization reveals allosteric switch design principles

Fig. 2. ERK phospho-switch hinge optimization.

Extended Data Fig. 2. Multi-variable linear regression model to recapitulate the hinge design rules.

Extended Data Fig. 3. Validation of the multi-variable linear regression model.

Extended Data Fig. 4. AlphaFold3-predicted structures of ERK-phosphoGal4v1, WW domain and WW-PRpTP interactions.

Extended Data Fig. 5. Allosteric un-coupling, course of optimization and phospho-mimetic mutations of the ERK phospho-switch.

Improving the substrate site and phospho-binder to generate an optimal phospho-switch

Fig. 3. Substrate and phospho-binder optimization for a final ERK phospho-switch.

Extended Data Fig. 6. ERK-phosphoGal4 performance across cell lines.

Direct and specific activation of ERK-phosphoGal4 using optogenetics

Fig. 4. Direct and specific activation of ERK-phosphoGal4 with optogenetics.

Extended Data Fig. 7. Optogenetic activation of ERK and ERK-phosphoGal4 variants.

Applying the ERK-phosphoGal4 switch as a biosensor for developmental cell signaling

Fig. 5. Applying ERK-phosphoGal4 as a biosensor for developmental cell signaling.

Extended Data Fig. 8. Additional gastruloid examples and quantifications of ERK-phosphoGal4 activity.

Phospho-switch control is generalizable to other input kinases and effector proteins

Fig. 6. Phospho-switch control is generalizable to other input kinases and effector proteins.

Extended Data Fig. 9. Additional characterization of JNK-phosphoGal4, ERK-phosphoNanobodies and ERK-phosphoVav2.

Discussion

Methods

Plasmid construction

QCTK development

Variant assembly with QCTK

Cell culture

Cell transient transfection and ERK stimulation with conditioned media

Cell line generation

Flow cytometry analysis and FACS

Protein structure prediction with AlphaFold3 and structural analysis

Fold change calculation

Multivariable linear regression model to predict hinge performance

Optogenetic activation of ERK-phosphoGal4 with optoWELL and flow cytometry

Signaling pathway stimulation

Dose–response curve fitting

Gastruloid protocol

Gastruloid fixation and immunofluorescence staining

Gastruloid live imaging

Gastruloid image analysis and quantification

Live-cell imaging

Optogenetic activation of ERK-phosphoGal4 with live-cell imaging

Live-cell imaging for the F-actin ERK-phosphoNanobody

Live-cell imaging and analysis for ERK-phosphoVav2

Statistics and reproducibility

Biological material availability

Reporting summary

Online content

Supplementary information

Source data

Acknowledgements

Extended data

Author contributions

Peer review

Peer review information

Data availability

Code availability

Competing interests

Footnotes

Extended data

Supplementary information

References

Associated Data

Supplementary Materials

Data Availability Statement

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases