De Novo Design of Tyrosine and Serine Kinase Driven Protein Switches

Abstract

Kinases play central roles in signaling cascades, relaying information from the outside to the inside of mammalian cells. De novo designed protein switches capable of interfacing with tyrosine kinase signaling pathways would open new avenues for controlling cellular behavior, but to date no such systems have been described. Here we describe the de novo design of two classes of protein switches which link phosphorylation by tyrosine and serine kinases to protein-protein association. In the first class, protein-protein association is required for phosphorylation by the kinase, while in the second class, kinase activity drives protein-protein association. We design systems which couple protein binding to kinase activity on the ITAM motif central to T-cell signaling, and kinase activity to reconstitution of GFP fluorescence from fragments and the inhibition of the protease calpain. The designed switches are reversible and function in vitro and in cells with up to 40-fold activation of switching by phosphorylation.

While there has been progress in the de novo design of stable proteins with a range of folds and functions, the de novo design of proteins which modulate or are modulated by tyrosine phosphorylation has not to our knowledge been described. Native protein motifs such as EF-hands and coiled coils in which aspartate or glutamate residues play critical roles have been made responsive to serine phosphorylation by taking advantage of the approximate mimicry of these residues by phospho-serine. For example, a serine substitution in an EF-hand eliminates calcium binding, and binding activity can be restored by phosphorylation of the same serine¹. Phosphorylation of serine near the positive N-terminal end of an alpha helix is electrostatically stabilizing and has been used to engineer switching². Phosphorylation control has also been achieved by introducing phosphorylation sites at naturally occurring interfaces; phosphorylation at such sites can reduce binding by introducing unfavorable steric and electrostatic interactions³. Serine PKA phosphorylation sites designed into a coiled-coil along the dimer interface in positions normally occupied by glutamate increase stability upon phosphorylation when these sites interact electrostatically with arginine residues placed across the interface^2,4. The glutamate approximate mimicry approach does not work for tyrosine-based phosphorylation, as no natural amino acid has a shape and charge distribution similar to phospho-tyrosine (pTyr). Tyrosine kinases are of particular interest for mammalian cell control, as they are involved in a large number of signaling pathways relaying information from the outside to the inside of cells⁵.

We set out to design proteins that modulate or respond to phosphorylation of tyrosine by tyrosine kinases as outlined in Fig 1A. In the first “binding-dependent phosphorylation” scheme, a de novo designed protein has a kinase target site that is cryptic (cannot be accessed by the kinase) until addition of a designed co-regulator. From the perspective of engineered cell control, the first scheme would enable the incorporation of additional protein control elements regulating the extent of activation of a tyrosine kinase signaling pathway. For example, T-cell activation begins with the phosphorylation of the ITAM (immunoreceptor tyrosine-based activation motif), which then binds to activate the ZAP-70 kinase which continues the pathway⁶. Controlling the accessibility of the ITAM motif via the binding of co-regulator would control the extent of T-cell activation. In the second “phosphorylation-dependent binding” scheme, a binding interface for a target of interest is hidden in the absence of kinase, and becomes exposed upon kinase catalyzed phosphorylation. Such switches could allow for the incorporation of new signaling pathway branches downstream of activated kinases.

Figure 1.

Phosphorylation Switch Design (A) Schematic of our two different mechanisms that use binding to control a phosphorylation event and phosphorylation to control a binding event. “A” represents the protein switch that can be phosphorylated. “B” represents a protein that can bind “A”. The superscript “p” represents a phosphorylation event on “A.” (B) Design model of the caged ζ1 ITAM in a DHD-1:234A_B (PDB ID 6DLC) based LOCKR and model mechanism for phosphorylation controlled by binding. The cage (green), ITAM threading (cyan), and key peptide (blue) are colored on the structures. (C) Key induced phosphorylation of the ζ1 ITAM motif measured by intact-protein mass spectrometry. The N-terminal tyrosine in the ζ1 ITAM motif is mutated to glutamate to solely follow the phosphorylation of the caged C-terminal tyrosine. Data are presented as mean values +/− standard deviation from three independent experiments. (D) Design pipeline for incorporating two-stacked hydrogen-bond networks to contain the Src-phosphorylation site. The combined phosphorylation site is shown for the (i/i+4) arrangement used in the final switch. The percentage of backbones generated containing hydrogen bond networks including the indicated residues for both straight and left-handed helices. (E) Schematic diagram of split-GFP activation by psGFP-2Y by phosphorylation by the Src-kinase. On the left, the design model for psGFP-2Y showing the GFP-11 peptide in magenta. Phosphorylation site tyrosine residues are shown in blue. (F) GFP fluorescence of pGFP-2Y, GFP-1–10 and ATP increases with the addition of Src kinase. Data are presented as mean values +/− standard deviation from three independent experiments.

Design of binding-induced phosphorylation:

We began by exploring the de novo design of protein systems implementing binding-dependent phosphorylation. We chose to focus on the ITAM) central to T-cell signaling. The ITAM contains a dual-tyrosine motif that when phosphorylated initiates T-cell activation⁶. To make phosphorylation of the ITAM tyrosines dependent on a regulatory protein in trans, we aimed to design a system in which competition between inter and intra molecular interactions drives a conformational switch, as in the recently described LOCKR switches ^7,8. We first sought to design a 4-helix bundle in which the C-terminal tyrosine in the ITAM sequence motif forms part of the central core and hence is not accessible in the folded state to the kinase. There is a considerable solvation-free energy cost to burying the polar hydroxyl groups on the tyrosines without making hydrogen bonds to other residues in the core, and hence we focused on designed proteins in which one of the ITAM tyrosines could make hydrogen bonds. We took advantage of a series of de novo designed heterodimers containing extensive buried hydrogen bond networks, and found that the C-terminal ζ1 ITAM tyrosine could be aligned onto a tyrosine, Y123, hydrogen-bond network in the designed heterodimer DHD-1:234A_B, PDB ID 6DLC (Fig 1B)⁹. To construct a single chain based switch, we connected the DHD-1:234A_B heterodimer into a four-helix bundle monomer with the embedded ITAM motif.

To drive exposure of the buried ITAM motif so that it could be phosphorylated by a tyrosine kinase, we explored switch designs analogous to LOCKR in which intermolecular interactions with a helical “key” are stronger than intramolecular interactions with a helical “latch” which can hence be displaced by the key. Caging phosphorylated motifs like the ITAM in the four-helix bundle creates an additional challenge when designing the key peptide. The ideal residue choice for complementing the tyrosine hydrogen bond network in DHD-1:234A_B by the key peptide would be a tyrosine; however if the key peptide tyrosine becomes phosphorylated, the steric and electronic clash of the pTyr would preclude binding. To avoid inactivation of the key by phosphorylation, we replaced the tyrosine with a histidine residue that can also hydrogen bond across the interface.

To test the designed binding induced phosphorylation system, we obtained synthetic genes encoding the caged ITAM motif and the key, expressed the two proteins in E coli, and purified them by Ni-affinity chromatography. We then tested the ability of Lck, the native T-cell kinase, to phosphorylate the C-terminal tyrosine in the ITAM motif in the presence or absence of the key peptide. Without key peptide, no phosphorylation was detected by intact protein mass spectrometry: the ITAM motif as designed is caged by the helical bundle. Upon addition of key peptide, the ITAM motif was readily phosphorylated by Lck. 5μM key activated phosphorylation of 50μM of the design (which we call CD3 ζ1 ITAM LOCKR), indicating multiple turnover activation which could be useful for cellular control applications (Fig 1C). Because phosphorylation is irreversible under our experimental conditions, the key can disassociate from one switch molecule and then bind to and activate another, and hence may be viewed as a multiple turnover catalyst; once phosphorylated the ITAM motif binds with high affinity to ZAP-70 to signal T-cell activation.

Design of Phosphorylation-induced binding:

We next sought to de novo design a protein system in which protein-protein association is driven by a tyrosine kinase catalyzed tyrosine phosphorylation. We sought to design a helical bundle system containing a “latch” helix with multiple tyrosine phosphorylation sites that regulate the accessibility of a binding peptide motif that is inactive when sequestered within the bundle. In this design scheme, phosphorylation of one or more tyrosine residues causes the latch helix to disassociate from the bundle, allowing the binding peptide to associate with its target. We chose to design systems containing multiple tyrosines to be phosphorylated rather than a single tyrosine to obtain a greater thermodynamic driving force for conformational change following kinase activation. We found that two Src-Family Consensus sites, EXLYXXL, interwoven into an (i/ i+4 straight or i/i+3 left-handed helix) provided a particularly compact arrangement of tyrosine sites¹⁰. Compactness has the advantage of enabling the fitting of more phosphorylation sites in a single designed system, and leaves more flexibility in the bundle to thread any bioactive motif of choice without disrupting the phosphorylation site networks.

Aiming for switches with smaller total size, to facilitate incorporation into cell signaling studies in future work, we focused on designing four-helix bundle (rather than the 6 helix bundle original LOCKR⁷) systems with two tyrosine-containing hydrogen-bond networks involving a terminal latch helix containing a bioactive binding peptide. We found that straight and left-hand twisted helical topologies fit the limited rotameric preferences of tyrosine residues with both residues pointing towards the core with room for forming hydrogen bond networks (Fig 1D)--straight helices are compatible with an i/i+4 arrangement of tyrosine residues, while left-handed helices fit an i/i+3 arrangement. We parametrically generated hundreds of thousands of bundles varying the super-helical radius, super-helical twist, chain axial offset, and helical phase, and used Monte Carlo sampling with the HBNet mover to identify tyrosine containing hydrogen bonding networks¹¹.

We found that only large bundles with a radius of greater than 6.75 angstroms could incorporate both tyrosine residues into hydrogen-bond networks. For the left-hand super twisted topologies, we parametrically generated 100,000 bundles and identified 23,917 hydrogen-bond networks; 519 of the models contained hydrogen-bond networks with both tyrosines. For straight helical topologies, we parametrically generated 1,000,000 bundles resulting in 68,095 hydrogen bond networks with 819 networks involving both tyrosine residues. Structures without both tyrosine residues participating in a hydrogen-bond network were discarded, and combinatorial sequence optimization was used to minimize the energy of the remainder of the structure using the Rosetta energy function which includes terms representing van der Waals packing, hydrogen-bonding, electrostatics and solvation. The resulting designs were filtered based on Rosetta energy, secondary-structure side chain complementary, cavities within the core of the protein, and buried unsatisfied polar residues; polar residues without hydrogen-bonds in the core of the protein are very energetically unfavorable due to high de-solvation energies and are therefore important to avoid (Supplemental Figure 1). Genes encoding 16 designs passing these filters were synthesized, 14 were expressed in the soluble fraction and 10 were largely monomeric by size exclusion chromatography (SEC) (Supplemental Table 1).

GFP Switch:

With the core phosphorylation switch mechanism in place, we sought to incorporate peptides that interact with external protein partners into the latch such that the interaction was only possible in the open state of the switch. We first sought to design a kinase induced fluorescence switch by threading the GFP beta-strand 11 sequence, DHMVLHERVNAAGIT, and some variants onto the latch helix of the monomeric designs. In isolation, the GFP-11 peptide can assemble with the remainder of GFP(GFP1–10) and cause the internal chromophore to mature resulting in green fluorescence; we reasoned this would be possible in the open (phosphorylated) form of the switch but not the closed state due to steric exclusion in the latter¹². We found threadings of the GFP peptide predicted to be compatible with the structure for six of the solubly expressed buried tyrosine network proteins described above and obtained synthetic genes encoding these designed switches. One of the six, which we refer to as pGFP-2Y in the remainder of the paper, was soluble and monomeric by SEC; it is a straight bundle with (i/i+4) stacked-tyrosine networks on the C-terminal helix (Fig. 1E, Supplemental Table 1). To test for phosphorylation induced switching, 2μM pGFP-2Y was mixed with 1μM GFP1–10 and 500μM ATP (Fig 1E). Addition of the Src-kinase catalytic domain resulted in a five-fold increase in GFP fluorescence consistent with the designed phosphorylation switch concept (Fig 1F). However, pGFP-2Y was also observed to equilibrate with a larger oligomeric species which likely limits the extent of activation (Supplemental Figure 2).

We sought to stabilize the bundle by extending each helix by one turn while maintaining the loops. To counteract the decrease in switching expected from the concomitant increase in the free energy of latch disassociation, we searched the newly introduced positions for additional hydrogen bond network placements compatible with the Src-kinase phosphorylation site to provide a larger destabilization upon phosphorylation and increase switching. We were able to incorporate another hydrogen bond network with two additional Src-phosphorylation sites. The resulting pGFP-4Y design is a four helix bundle, 35 amino acids per helix, containing four Src phospho-sites(Fig 2A).

Figure 2.

Phosphorylation Activates GFP Fluorescence (A) The helices of pGFP-2Y were extended (yellow) and a new double-tyrosine hydrogen-bond network introduced. The caged GFP-11 is shown in magenta and the original tyrosine phospho-sites in blue. The graph shows the previous pGFP-2Y in vitro switching plotted next to the new pGFP-4Y data. Data are presented as mean values +/− standard deviation from three independent experiments. (B) GFP fluorescence of pGFP-4Y, GFP-1–10 and ATP increases with the addition of Src kinase. Correlation between phosphorylation and pGFP-4Y activation. Pre-phosphorylated pGFP-4Y was added to GFP1–10. ATP concentration was used as a limiting reagent to control the amount of phosphorylation that occurred. Phosphorylation was measured by denaturing whole protein mass spectrometry. Due to a large number of possible heterogeneously phosphorylated species, we did not attempt to quantify the phosphorylation. Phospho-species for pGFP-4Y detected are shown above each bar. Examples of the de-convoluted mass spectrums for the 0, 200μM, 500μM and 2.5mM ATP samples are shown below the graph. Data are presented as mean values +/− standard deviation from three independent experiments. (C) pGFP-4Y initial phosphorylation event is spread across all tyrosine sites. We performed a limited phosphorylation such that only a single phosphorylation occurred per protein at maximum representing the initial phosphorylation event. The samples were digested with trypsin and analyzed by MS/MS in a quantitative manner to determine which sites were phosphorylated (Supplementary Table 3). The percentages reported are for the total phosphorylation for each phospho-site individually in the sample. Reported error is standard deviation from three experiments.

The pGFP-4Y design was expressed in E. coli at higher levels than pGFP-2Y, and was found to remain monomeric for over a week when stored at 4C (Supplemental Figure 2). pGFP-4Y has a highly helical content as indicated by circular dichroism spectroscopy, consistent with the design model, and was found to be monomeric by SEC-MALs (Supplemental Figure 2). The solution X-ray scattering (SAXS) spectrum of pGFP-4Y matches that expected from the computational design model, suggesting that the overall shape of the protein is close to that designed (Supplemental Figure 2, Supplemental Table 2). We investigated pGFP-4Y switching by comparing fluorescence in the presence and absence of Src kinase. Addition of the Src kinase to pGFP-4Y, ATP, and GFP1–10 resulted in a 10-fold increase in GFP fluorescence (Figure 2A).

Characterization of pGFP-4Y switching mechanism:

To characterize the mechanism by which Src kinase activates the pGFP-4Y switch, we investigated the relationship between the extent of phosphorylation and activation of GFP fluorescence. We generated pGFP-4Y with different extents of phosphorylation in the absence of GFP1–10 utilizing ATP as a limiting reagent, and analyzed the protein using intact protein mass spectrometry. As there are four tyrosines which can each be either phosphorylated or not phosphorylated, there are 16 possible species, and we were not able to quantify the populations of each of these species; we focused instead on overall properties of the distribution. The distribution of pGFP-4Y phosphorylation states goes smoothly from completely unphosphorylated at zero ATP to nearly completely phosphorylated at 5mM ATP (Supplemental Figure 3, Fig 2A). There was a strong correlation between the amount of ATP/phosphorylation and the extent of activation of GFP fluorescence upon addition of GFP1–10. The most heavily phosphorylated population of pGFP-4Y activates GFP 40-fold more than the unphosphorylated protein (Fig. 2B).

We next sought to determine whether there was an order in which the tyrosine residues become accessible to the kinase. We performed a limited phosphorylation of pGFP-4Y such that each protein was phosphorylated on no more than one tyrosine (Fig 2C). Using tryptic digestion and MS/MS on the unphosphorylated and singly phosphorylated states, we found that despite the different accessibility of tyrosines in the designed folded state, there was a roughly equal probability for each of the four to be the first to be phosphorylated (Fig. 2C, Supplemental Table 3). The simplest explanation for this lack of preference is that the entire switch must be in the unfolded state for the tyrosine residues to become phosphorylated--in this case all tyrosines would be approximately equally accessible (in native protein systems, randomly ordered phosphorylation has also been observed to an unfolded state of the region being phosphorylated¹³). In vitro, the switch takes approximately 2 hours at room temperature to become fully phosphorylated; this likely reflects the slow intrinsic opening rate of the pGFP-4Y latch (Supplemental Figure 4).

How does the addition of 4 large and negatively charged phosphate groups influence the structure of pGFP-4Y? We found that the fully phosphorylated form of pGFP-4Y is monomeric by native mass spectrometry (nMS) and has a helical circular dichroism spectrum identical to that of the unphosphorylated form (Fig 3A, Supplemental Figure 4). To probe the effect of phosphorylation on the hydrodynamic radius (which reflects both the size and shape of the protein), we analyzed a mixture of 0–4 phosphorylated pGFP-4Y by size exclusion chromatography (SEC) coupled to native mass spectrometry. The eluting monomeric species differ in their m/z due to different amounts of phosphorylation allowing us to distinguish co-eluting species and extract their individual elution profiles (Fig. 3B, Supplemental Figure 4). We found a decrease in elution time with increasing phosphorylation of pGFP-4Y that is not attributed to a change in oligomerization, indicating an increase in hydro-dynamic radii due to a shift in protein conformation.

Figure 3.

Characterization of phospho-states of pGFP-4Y. (A) Helical circular dichroism signals for both phosphorylated and unphosphorylated pGFP-4Y. (B) Increasing phosphorylation results in a larger hydrodynamic radius. A mixed sample of phosphorylated and unphosphorylated pGFP-4Y was separated over a size-exclusion column and detected by native mass spectrometry. Smoothed and overlaid extracted ion chromatograms (XICs) for m/z 2550–2551, 2560–2561, 2570–2571, 2580–2581, 2590–2591 correspond to 8+ pGFP-4Y with 0, 1, 2, 3, and 4 phosphorylations, respectively (C) GFP fluorescence of phosphorylated or dephosphorylated pGFP-4Y by the lambda phosphatase added to GFP1–10. Data are presented as mean values +/− standard deviation from three independent experiments. (D) Activation of pGFP-4Y-Tsk4 in HEK293T cells by constitutively active Src-530F. HEK293T cells expressing GFP1–10 from the chromosome were transiently transfected with pGFP-4Y-Tsk4 (Black) or Src-530F with pGFP-4Y-Tsk4 (Green). EBFP co-expressed with pGFP-4Y-Tsk4 via P2A peptide monitors the concentration pGFP-4Y-Tsk4 via blue fluorescence. Each point represents the fluorescence of a single cell from three separate experiments measured 48 hours after transfection. A linear regression with 95% confidence interval is plotted for each data set.

We next investigated whether the pGFP-4Y phosphorylation switch is reversible. Since the maturation of the GFP chromophore is an irreversible process, we examined whether, in the absence of GFP1–10, the addition of phosphatase could return phosphorylated pGFP-4Y to a caged GFP-11 state. First, we fully phosphorylated pGFP-4Y and measured its phospho-state by MS. We then split the sample in two, and for one half used lambda phosphatase to dephosphorylate the tyrosines (Supplemental Figure 4). When mixed with GFP1–10, the dephosphorylated pGFP-4Y does not induce GFP-fluorescence, indicating that the phosphorylation switch is reversible (Fig 3C).

Since tyrosine phosphorylation-based cellular signaling plays a central role in mammalian cell signaling, we tested if pGFP-4Y could switch in HEK293T cells. The regulation of tyrosine phosphorylation in mammalian cells occurs primarily through co-localization of kinase and substrate with phosphatase activity dominating elsewhere¹⁴. We appended a Src-binding peptide from a native Src substrate, Tsk4, that recruits the Src-kinase to create pGFP-4Y-Tsk4¹⁵. In a GFP1–10 expressing background, we transiently transfected pGFP-4Y-Tsk4 with or without Src-530F, a constitutively active variant of Src, to test phosphorylation based switching. To control for the variable expression from transient transfection, we expressed eBFP with a P2A site behind pGFP-4Y-Tsk4 to monitor the concentration of pGFP-4Y-Tsk4 in the cell, and normalized the switch activation signal (green fluorescence) by the total amount of sensor (blue fluorescence). In the presence of added Src, there was on average 2.5 ± 0.1 fold more switch activation (green fluorescence) than in the absence of added Src (Fig 3D). Thus, the phosphorylation switch functions as designed in mammalian cells. HEK293T cells natively contain the Src-kinase, which can be activated in normal growth media, likely contributing (through increasing the background switching) to the reduced fold-activation relative to the in vitro system.

Calpain Inhibition Switch:

We next explored the versatility of the phosphorylation-induced binding mechanism by changing the functionality of the switch by incorporating in the latch the IA peptide from the calpain inhibitor, calpastatin (Supplemental Fig 5). Calpain is a regulatory cysteine protease involved in cell motility and cell cycle progression. Calpastatin tightly binds calpain with its IA domain, while the IB domain inhibits the active site^16,17. We found a threading of the IA peptide, MDAALDDLIDTLGG, predicted to not destabilize the closed form of the switch. We call this design pDIA-4Y.

We expressed pDIA-4Y in E. coli and found the purified protein to be soluble and monomeric. The level of activation by phosphorylation of the initial design was low, suggesting the binding affinity of the latch for the cage was too strong. We tuned the latch-cage affinity by introducing mutations of Leu and Ile residues on both sides of the interface to Ala or Ser (L17S, L28A, L53A, I56A, I71A, L94A, L108A, L161S). We observed an increase in the degree of phosphorylation of the switch with increasing numbers of mutations, finally yielding pDIA-4Y_M8 (Supplemental Figure 5). We measured the relative binding of phosphorylated pDIA-4Y_M8 and unphosphorylated pDIA-4Y_M8 to their binding partner Domain IV, DIV of calpain by bio-layer interferometry (Supplemental Fig. 5). Phosphorylation of pDIA-4Y_M8 results in an (8.5±1.6) fold increased binding to DIV of calpain immobilized on the tip.

To convert psDIA-4Y_M8 into a phosphorylation inducible inhibitor of calpain, we fused the calpain inhibitory IB peptide downstream of the latch to create pDIA-4Y_M8_IB. In this scheme, phosphorylation of psDIA-4Y_M8 opens the latch, enabling the caged IA peptide to bind calpain, and then the adjacent IB peptide can block the enzyme active site. We assayed the ability of phosphorylated psDIA-4Y_M8_IB to inhibit calpain. Calpain activity was monitored spectroscopically by the fluorescence increase resulting from cleavage of a calpain substrate peptide (Supplemental Figure 5)¹⁸; we compared phosphorylated and unphosphorylated switch to the uncaged IAB calpastatin positive control and the unlocalized IB peptide negative control. Phosphorylation of psDIA-4Y_M8_IB increased calpain inhibition; the fold activation is lower than that observed for the binding interaction in part because the unphosphorylated psDIA-4Y_M8_IB retains considerable inhibitory activity (localization is not essential for inhibition under the conditions we tested).

Design of Phospho-serine PKA Switch:

Given our success of using the phosphorylation of tyrosine residues in core hydrogen-bonding networks to drive switching, we reasoned the same switching mechanism should be extensible to serine-based phosphorylation by designing protein hydrogen-bond networks incorporating a serine based phosphorylation site. We chose to design a switch using the consensus phosphorylation site, RRASL, from the cyclic AMP dependent Protein Kinase A (PKA) due to its centrality in G protein-coupled receptor signaling^19,20. Using a strategy analogous to that for pGFP-4Y, we set out to design a 4 helix-bundle that incorporates 4 ‘RRASL’ motifs with phosphorylated serine involved in core hydrogen-bond networks. Upon phosphorylation of the serine residue, the core of the protein should be disrupted sterically and electronically by the phosphoryl group.

Following the same protocol as the tyrosine phosphorylation switch, we first identified four positions in a four helix bundle for the RRASL site that positions both the serine in the core and the charged arginine residues on the surface; based on the results with pGFP-4Y suggesting phosphorylation through global unfolding, the phosphorylation sites were not limited to the C-terminal helix (Fig 4A). 100,000 nearby helical backbones were generated from helical parameters and the HBNet mover was used to search for hydrogen-bond networks that satisfied the all four serine residues in the RRASL-sites. For the 350 backbones with such networks, the remainder of hydrophobic core and surface residue were designed and filtered by Rosetta energy, secondary-structure side chain complementarity, buried unsatisfied polar residues and cavities. Genes encoding 12 designs passing these filters were synthesized. All designs expressed in the soluble fraction and 4 designs were monomeric by size exclusion chromatography (Supplemental Table 4). We sought to introduce functionality into the three designs with the best solution properties by threading the GFP beta-strand 11 sequence, DHMVLLERVNAAGIT, or a closely related variant onto a helix of the design. From 7 threadings, 6 designs expressed and 1 of those, denoted pGFP-4S, maintained the monomeric form of the protein by SEC (Supplemental Table 4). pGFP-4S showed limited phosphorylation by PKA at room temperature (Supplemental Fig 5).

Figure 4.

Serine Based Phosphorylation Switch (A) Design model for pGFP-4S. The GFP-11 threading (magenta) and PKA phosphorylation sites (cyan) are shown on the structure. (B) Activation of pGFP-4S in K562 cells by the PKA kinase. The serine switch monitored by the RFP fluorescence was used to normalize the GFP fluorescence for each cell. The background (GFP1–10 expression alone) and GFP1–10 activated with an uncaged version of GFP-11 showed no changes upon FSK and IBMX activation of the PKA kinase. Data are presented as mean values +/− standard deviation from three independent experiments. (C) pGFP-4S-M2 phosphorylated by the PKA kinase in vitro preferentially activates GFP1–10. Data are presented as mean values +/− standard deviation from three independent experiments.

We hypothesized that, as in the case of pDIA-4Y, destabilizing pGFP-4S mutations would increase phosphorylation, and generated and screened variants for PKA driven GFP reconstitution by co-expressing them with GFP1–10 in K562 cells. Hypothesizing that high local PKA concentrations would be required to activate the switch in cells, we also tested switches that were C-terminally tagged with a 3X tandem repeat of the PKA consensus sequence (RRASL) as a recruitment motif, denoted by R3 in the name. To activate PKA, we co-treated cells with 50μM Forskolin (FSK) and 100mM 3-isobutyl-1-methylxanthine (IBMX) and screened cells at 8, 30, 55, and 77 hours post treatment for GFP fluorescence. The optimal switch variant, I48A and L49A (called pGFP-4S_M2_R3), demonstrated a small but reproducible fold-change 1.10 (± 0.001) (Fig. 4B), which is on par with most current PKA biosensors albeit on a significantly longer timescale^21–23. The recruitment motif dramatically increased the response to PKA activation. When phosphorylated in vitro pGFP-4S_M2 activates green fluorescence 2.1(± 0.2) times more than unphosphorylated pGFP-4S_M2 (Figure 4C, Supplemental Fig 5). Thus, our phosphorylation switch concept extends to serine phosphorylation and can be activated by endogenous kinases. The reduced activation of pGFP-4S_M2 relative to the tyrosine switch may reflect the greater concentration of phosphorylation sites in the latch helix for pGFP-4Y relative to the dispersed phospho-sites in pGFP-4S.

Discussion:

It is instructive to compare our designed binding-induced phosphorylation and phosphorylation-induced binding switches to naturally occurring kinase switches that are the product of natural selection. Signaling networks utilizing protein phosphorylation have phospho-sites that frequently occur in intrinsically-disordered regions such that the phospho-acceptor is always accessible in the unphosphorylated state²⁴. Often the additional negative charge of the phosphate group promotes a disorder to order transition, for example, the activation of ZAP-70 kinase via binding a doubly-phosphorylated and previously dis-ordered ITAM motif. Our binding induced phosphorylation switch reverses the order of events: Binding of the “key” peptide releases the latch helix containing the ITAM motif which likely becomes disordered in isolation. In this way, the expression of “key” peptide should regulate the functional change brought about by phosphorylation; in the case of ITAM, T-cell activation. Analogously in natural systems, phosphorylation of the Y291 in WASP is controlled by the binding of activated Cdc42. Once Cdc42 is bound, Y291 becomes unstructured and available for phosphorylation resulting in increased actin remodeling through activation of the Arp2/3^25,26. Our designed phosphorylation induced-binding switch employs structured phospho-sites that are disrupted by the electronic and steric clash of the added phosphoryl groups. This is analogous to the mechanism of phosphorylation-based switching for the voltage-dependent potassium channel, K_v3.4, the cell cycle inhibitor p27^Kip1, and the KH domain of the mRNA regulatory protein, KSRP ^27–29. For K_v3.4, phosphorylation of serine residues disrupts the structure of the inactivation domains altering the inactivation kinetics of the channel³⁰. Our current switches activate slowly due to the high cooperativity of unfolding; to increase the rate of activation it should be possible to design switches with low energy intermediates that expose individual tyrosine phosphorylation sites similar to p27^Kip1, although designing multiple intermediates remains a challenge for de novo design²⁸. While slow kinetics increases response time after a signaling event, they can be advantageous for molecular timing and memory, as in the slow re-phosphorylation of the phosducin protein upon complex formation after light stimulation in rod cells³¹.

Our designed switches provide a mechanism for designed proteins to ascertain or alter the state of the cell through direct interactions with signaling networks. As an illustrative example, the advent of chimeric antigen receptor T-cells (CAR-T) as a cancer therapy has created new synthetic biology challenges such as mitigating T-cell exhaustion from excess tyrosine kinase signaling through the chimeric receptor; exhausted cells have a reduced ability to perform effector T-cell functions^32,33. CAR-T expansion and action occurs over many days making switch kinetics less relevant³⁴. For our binding-dependent phosphorylation switches, the loss of key peptides under the control of repressors associated with T-cell exhaustion could inhibit caged ITAMs and mute CAR signaling during times of excess signaling^35,36. Alternatively a phosphorylation-dependent binding switch that responds to the tyrosine kinase that starts T-cell activation, Lck, could cage an inhibitor of T-cell activation, and dampen excess signaling from the CAR. Going beyond previous designed phospho-switches which control coiled-coil dimerization⁴, our approach enables caging of peptides that interact with other native cellular proteins, including potentially toxic peptides like BIM which drive apoptosis, that are difficult to sequester within a coiled coil. There is still considerable room for improvement in the dynamic range of the switches described in this paper, both by decreasing signal in the absence of kinase, and increasing the extent of activation in the presence of kinase; these improvements will facilitate applications in engineered T-cells and other synthetic biology systems.

Online Methods:

Design of ITAM LOCKR –

The designed heterodimer DHD-1:234A_B was used as base for the ITAM LOCKR, PDB ID 6DLC⁹. Originally crystallized as 3 helices + 1 helix heterodimer we connected the helices with a loop placing the single helix as C-terminal. The loop was composed of the ξ1 ITAM sequence, NQLYNELNLGRREEYDVLD, that was transitioned into the C-terminal helix such that the C-terminal tyrosine of the ITAM aligned with the Y123 in the original single helix in DHD-1:234A_B. We incorporated this loop sequence and helical segment into the DHD-1:234A_B model using the remodel program from the Rosetta suite³⁷. Design was done using Rosetta 2018.19. Using a blue print file we assigned a “loop” secondary structure for residues NQLYNELNLGRR and a “helical” secondary structure for residues EEYDVLD overwriting the native residues of DHD-1:234A_B. We chose the lowest energy structure from 30 runs as the design model. The key peptide for this ITAM LOCKR is the +1 helix from DHD_37. The single tyrosine residue for the key peptide was mutated to histidine to prevent phosphorylation by Lck.

Design of Tyrosine Phosphorylation Switch Scaffold:

Parametric backbones were generated using the Rosetta mover BundleGridSampler. Design was done using Rosetta 2018.19. For the straight helices, we sample a normal distribution with the given values being 2 sigma from the mean of the two values around the following parameters: Omega0 (−.3,.3), r0(6.5,7), helix1 delta omega1 (−15,15), helix1 z0 offset (−1,1), helix2 delta omega1 (−15,15), helix2 z0 offset (1,4), helix3 delta omega1 (−15,15), helix3 z0 offset (−1,1), helix4 delta omega1 (−30,−40), helix4 z0 offset (3,4). Degrees was used for any angle and angstroms for any distance. For the left-handed helices: Omega0 (−3.9,−3.3), r0(6.75,7.1), helix1 delta omega1 (−10,10), helix1 z0 offset (−4,−1), helix2 delta omega1 (−15,15), helix2 z0 offset (1,4), helix3 delta omega1 (−15,15), helix3 z0 offset (−2,1), helix4 delta omega1 (−15,15), helix4 z0 offset (1,4). For both types of bundles. Helix 2 and 4 were inverted for the N to C direction of the helices. Delta omega0 for was 0° for helix 1, 90° for helix 2, 180° for helix 3, 270° for helix 4. Each helix contained 35 amino acids. We mutated in the tyrosine residues, position 107 and 111 for the straight helices and position 31 and 34 for the left-handed helices. Using 100,000 left-handed helices and 1,000,000 straight helices, we seeded the creation of hydrogen bond networks off the tyrosine residues using the HBNet mover¹¹. We then identified the models that included a hydrogen bond network with both tyrosine residues and only moved these models to the next stage of design. We maintained the constraints on the hydrogen bond networks through-out the design process. In this stage, we added an additional hydrogen bond network to the opposite end of the helices from the tyrosine residues. After generating these networks, we did sequence design on the remaining residues. Essentially, we used layer design and used PackRotamers mover to pack hydrophobic residues around hydrogen-bond networks. We then did a 5 cycles of limited use of the FastDesign mover only designing alanine, methionine and phenylalanine residues in a generic monte carlo mover optimizing for the least buried unsaturated hydrogen bonds and secondary structure side chain compatibility^38–40. We also used an amino acid composition constraints to keep the fraction of alanine residues less than 0.07, limited total phenylalanine and methionine residues to 8 each. From these structures, we then filtered on Rosetta score (<−3.0), secondary structure side chain compatibility (>0.65), and cavities by visual inspection using Pymol’s cavity display feature to reduce the number of sequences to 30 of each type (left-handed or straight). We then connected the individual helices with loops using the ConnectChainsMover. We then used FastRelax on the loop segments to get rid of clashes and designed the sequence with the PackRotamers mover. We then picked 12 left-handed helical models structure and 9 straight helical models for experimental validation based on maximizing hydrogen-bond network diversity. We threaded the caged peptides onto the phosphorylation switch scaffolds by visual inspection using Pymol.

Design of PKA Switch Scaffold:

Parametric backbones were generated using the Rosetta mover BundleGridSampler using the following parameters sampling a normal distribution with the given values being 2 sigma from the mean of the two values. Omega0 (−4,−3.3), r0 (6.4,7.0), helix1 delta omega1 (−10,10), helix1 z0 offset (−4,−1), helix2 delta omega1 (−15,15), helix2 z0 offset (1,4), helix3 delta omega1 (−15,15), helix3 z0 offset (−2,1), helix4 delta omega1 (−15,15), helix4 z0 offset (1,4). For both types of bundles. Helix 2 and 4 were inverted for the N to C direction of the helices. Delta omega0 for was 0° for helix 1, 90° for helix 2, 180° for helix 3, 270° for helix 4. Each helix contained 35 amino acids. We mutated in RRASL sites to the poly-alanine backbones with the serine positioned at residues 10, 45, 80 and 101. We performed the hydrogen bond network and the remainder of the design as described in the tyrosine switch scaffold section. After filtering, we selected 12 designs based on achieving hydrogen-bond network diversity. We threaded the caged peptides onto the phosphorylation switch scaffolds by visual inspection.

Gene Construction:

Synthetic genes were constructed for bacterial expression by either Genscript into a pET21b+ vector or with IDT into the pET29b+ vector using the NdeI and XhoI restriction sites. For the pET21b+ vector, a stop codon was included to prevent the expression of the C-terminal his-tag. For purification, a N-terminal his-tag was synthesized with the gene, GSSHHHHHHSSGS. The original designs for phosphorylation dependent binding (no caged peptide), psGFP_2Y, psGFP_4Y were expressed using a pET21b+ vector. The remaining designed used the pET29b+ vector. For the pET29b+ vector, a stop codon was not included allowing the expression of C-terminal his-tag after the XhoI site. Plasmid were transformed into chemically competent Lemo21(DE3) cells (NEB). Constructs that required his-tag cleavage were synthesized with a N-terminal His-tag followed by TEV site in the pet29b+ vector with a stop codon to prevent the expression of the C-terminal His-tag. Maltose binding protein, MBP, fusions (key peptides and uncaged control peptides) were synthesized into the pET28b+ vector from IDT that includes a N-terminal His-tag followed by a thrombin site before MBP. The fused gene was placed downstream of MBP after a TEV cleavage site. DIV of calpain, uniprot P07384, was expressed with a N-terminal Avi-tag, LNDIFEAQKIEWHE, for in vivo biotinylation by BirA. For mammalian cell expression of pGFP_4Y and derivatives, genes were synthesized by Genscript in the pCMV vector using the KpnI/XhoI restriction sites. An EBFP was added to the C-terminal end of pGFP_4Y_Tsk4 behind a P2A site to monitor concentration in the cell. The Src-kinase domain (Addgene #79700), YopH phosphatase (Addgene #79749)⁴¹, Src-530F (Addgene #124659)⁴², pEGFP-GFP-11 Clathirin Light Chain (Addgene # 70217)⁴³ were ordered from Addgene. The Lck kinase domain in the pET28b+ was a gift from the Andreotti Lab⁴⁴. The GFP strand 1–10 were taken from sfGFP. All numbering reflects the design model without the histidine tags. For the caged peptides, we used residues 216–230 from sfGFP. We used residues 153–166 from DIA of calpastatin, Uniprot P20810, threaded at residue 135 to make psDIA_dY. We fused residues 170–222 from calpastatin IB to make psDIA_M8_IB from psDIA_M8. Residues 466–516 of Tsk4 were appended to pGFP_4Y to make pGFP_4Y_Tsk4¹⁵. We removed any tyrosine residues that could be phosphorylated but were not part of the designed phosphorylation sites. In this way, we could assign any phosphorylation events identified by LC-MS to the designed phosphorylation sites. For this, we made a Y225F (native sequence numbering) mutation in the DIB peptide from calpastatin. Additionally, the N-terminal ITAM tyrosine (in the loop) was mutated to glutamate to reduce the construct to one tyrosine and simplify analysis by mass spectrometry. ZAP-70 binding and activation strictly requires a doubly-phosphorylated ITAM so controlling the phosphorylation of a single tyrosine in the ITAM is sufficient to control ZAP-70 activation⁴⁵.

Plasmids for testing pGFP_4S and variants were constructed via using the Mammalian Cloning Toolkit (MTK) ⁴⁶. Human codon-optimized gene blocks for all de novo designed phosphoswitch variants were ordered from IDT. Phosphoswitch variants were assembled into multi-cistronic transcriptional units containing an Ef1a promoter driving GFP1–10^12,43 with or without a 32 amino acid linker upstream of helix 2 of the phosphoswitch, a P2A element, either the phosphoswitch variant or IRFP713⁴⁷, an optional recruitment motif, a T2A element, mCherry as a transduction marker, and the WPRE transcriptional terminator, all in a lentiviral packaging backbone. All plasmids using the lentiviral backbone were grown in the Stbl3 E. coli strain, and all other constructs were grown in Mach1 E. coli (QB3-Berkeley Macrolab). Protein sequences are included as a supplementary file.

Bacterial Protein Expression and Purification:

Starter cultures were grown overnight in Luria-Bertani (LB) media with 50μg/mL carbenicillin (pET21b+), 50μg/mL kanamycin (pET28b+, pET29b+), or 10μg/mL spectinomycin (YopH expression, Addgene #79749). Starter cultures inoculated 500mL of Studier TBM-5052 auto-induction media without anti-biotic at 1:100 ratio. The cells were grown at 37°C for 3.5h and switched to 18 °C for expression for 18h. The cells were collected by centrifugation and lysed by sonication in lysis buffer (20mM Tris, 300mM NaCl, 30mM Imidazole pH 8.0) on ice. De novo proteins were lysed by sonication. Cells expressing native proteins were lysed at 80% amplitude for 5s on/ 25s off (cool) while cells expressing designed proteins were lysed for 10s on/ 10s off (warm). Lysates were cleared by centrifugation at 22,000 rcf for 25 min. Supernatants were applied to (1mL) Ni-NTA columns equilibrated in lysis buffer. The columns were washed twice with 10 column volumes of wash buffer (20mM Tris, 300mM NaCl, 50mM Imidazole pH 8.0). The columns were eluted with 5mL of elution buffer (20mM Tris, 300mM NaCl, 250mM imidazole pH 8.0). If the his-tags were removed by TEV cleavage, the sample was buffer exchanged into 50mM Tris pH 8.0, 0.5mM EDTA, 1mM DTT over a PD-10 column. Then 0.5mg of TEV protease was added to the sample and let sit overnight at room temperature (4°C for the Src-kinase domain). The TEV protease was removed by batch binding with 1mL of Ni-NTA resin equilibrated in (10mM HEPES pH 8.0 150mM NaCl) for 10min. The flow through of the column was collected and the protein was further purified over a Superdex 75 Increase 10/300 GL (GE) size exclusion column pooling fractions with the desired monomeric protein. Split-GFP was purified using Ni-affinity as above but with 10% glycerol added to the buffer. The Src kinase domain was co-expressed with the YopH phosphatase to prevent toxicity⁴¹. DIV of calpain was co-expressed with soluble BirA ligase and with 50mM biotin added to the cell culture after the culture grew to ~0.6 OD600⁴⁸. Lck was expressed in ArcticExpress(DE3) cells at 16°C overnight. The cells were lysed with lysozyme and freeze-thaw in 3 cycles and purified at 4°C by Ni-affinity resin and the his-tag removed via the TEV protease⁴⁴.

In vitro Phosphorylation Assay:

[Lck Phosphorylation] 500nM of the Lck kinase domains was mixed with 50μM of the ITAM LOCKR in 50mM HEPES pH 7.0, 10mM MgCl₂, 1mM DTT, 1mg/mL BSA, 1mM Pefablock and 500μM ATP for at least 2h. The amount of key peptide (fused to MBP) was included as indicated.

[Src Phosphorylation] 500nM of the Src kinase domain was mixed with 50μM of the phosphate switch in 10mM HEPES pH 8.0, 10mM MgCl₂, 2mM ATP and let to react for at least 5h. The non-phosphorylated samples were generated by incubating in the same solution without the Src-kinase domain. To generate the singly phosphorylated sample of pGFP-4Y the reaction was quenched with 20mM EDTA after 10 min.

[PKA Phosphorylation] 200nM of PKA Kinase (NEB) was mixed with 25μM of the switch in 50mM Tris pH 7.5, 10mM MgCl₂, 2mM ATP and let to react for at least 5h at room temperature. The non-phosphorylated samples were generated by incubating in the same solution without the kinase.

LC-MS:

Phosphorylation was monitored by whole-protein denaturing mass spectrometry on a via electron spray ionization using a Waters Synapt-QTOF at the University of Washington. The protein was desalted in line to the mass spectrometer using an AdvanceBio RP-Desalting (2.1mm, 12.5mm) column using an acetonitrile/water gradient in 0.1% formic acid with a flow rate of 0.5 ml/min on a Waters Acquity Ultraperformance LC. The sample was bound to the column using 10% acetonitrile isocratic flow for 1 min and then eluted by shifting to 95% acetonitrile using a linear gradient during the next minute. The concentration was held at 95% acetonitrile for 2 min. The column was shifted back to 10% acetonitrile for the next minute and then held at 10% acetonitrile for another minute. The spectrum was deconvoluted using the Mass Lynx 4.1 software using the m/z range from 600–1500. A single phosphorylation event on a protein can be accurately quantified using peak ratios⁴⁹. Therefore, we quantified the percent phosphorylation for the ITAM LOCKR by using the ion ratios between the phosphorylated and non-phosphorylated peak.

In vitro GFP Switching Assay:

Tyrosine Switch [Pre-phosphorylated Assay]: 50μM pGFP_4Y was mixed with 500nM Src kinase in 10mM HEPES pH 8.0, 10mM MgCl₂, 150mM NaCl and 2mM ATP for a total of 200μL. The reaction was let sit for 24h in a covered 96-well plate. After reactions with or without kinase were mixed at 2μM pGFP_4Y with 1μM GFP1–10 for a total of 200μL and let to equilibrate for 1d. [Continuous phosphorylation assay]: 2μM of the pGFP_4Y was mixed with 1μM GFP1–10 and 500nM Src kinase in 10mM HEPES pH 8.0, 10mM MgCl₂, 150mM NaCl, and 500μM ATP in 200μL total volume. The reaction sat in the dark at room temperature for 3d in a covered 96-well plate. Serine Switch [Pre-phosphorylated Assay]: 25μM pGFP_4S_M2 was mixed 200nM PKA kinase in 50mM Tris pH 8.0, 10mM MgCl₂ and 2mM ATP for at least 5 hours. After reactions with or without kinase were mixed at 1μM pGFP_4S_M2 with 1μM GFP1–10 for a total of 200μL and let to equilibrate for 1d. For all preparations, GFP Fluorescence was measured on a BioTek Synergy Neo2 Multi-mode plate reader with excitation at 488nm and emission measured at 509nm using a 10nm bandwidth.

Phosphatase Reaction -

We purchased the Lambda Protein Phosphatase from New England BioLabs (NEB). We mixed 50μL of 50μM of phosphorylated psGFP-4Y in the supplied reaction buffer with 400units of the lambda phosphatase and let the reaction go overnight at 30°C.

Circular Dichroism Measurements –

Proteins were buffered exchanged into 10mM phosphate 100mM NaCl at pH 7.4 via size exclusion chromatography. The CD spectrum from 260nm to 190nm was recorded with 0.2mg/mL protein in 1mM cuvette on a Jasco J-1500 CD Spectrometer.

SAXS –

Samples were separated by sec exclusion chromatography. psGFP_4Y at 75μM in 10mM HEPES pH 8.0 100mM NaCl for SAXS analysis. Data was collected at the SIBYLS 12.3.1 beamline at the Advanced Light Source. The X-ray wavelength was 1.27 Angstroms at 11keV. The sample to detector distance of the Pilatus3 2M detector was 2m corresponding to a scattering vector q (q = 4π sin θ/λ, where 2θ is the scattering angle) range of 0.008 to 0.36 Å⁻¹ Data sets were collected at 10°C with 10 second exposure divided into 33 frames every 0.3 seconds. Frames were merged using the FrameSlice utility from SIBYLS to exclude radiation damage at low q values but reduce noise over the rest of the curve. Data was analyzed using the ScÅtter (version 2.5) software package to calculate SAXS statistics^50,51. The FoXS web server was used to compare design models to experimental scattering profiles and calculate quality of fit, χ-values^52,53.

SEC-MALS –

We ran 3mg/mL of the indicated sample in 50mM Tris-HCL, 150mM NaCl pH 8.0 at 1mL/min over a Superdex 6 10/300 column in line with a Heleos multi-angle static light scattering and an Optilab T-rEX detector (Wyatt Technology Corporation). The data was then analyzed using ASTRA (Wyatt Technologies) to calculate the weighted average molar mass (Mw) of the selected species and the number average molar mass (Mn) to determine monodispersity by polydispersity index (PDI) = Mw/Mn.

Native Mass Spectrometry (MS)

Sample purity and integrity was analyzed by on-line buffer exchange MS using a Vanquish UHPLC system (Thermo Fisher Scientific) coupled to a Q Exactive UHMR Hybrid Quadrupole-Orbitrap instrument (Thermo Fisher Scientific). 100 pmol protein were injected and on-line buffer exchanged to 200 mM ammonium acetate, pH 6.8 (AmAc) by a self-packed buffer exchange column (P6 polyacrylamide gel, BioRad) at a flow-rate of 100 μL per min⁵⁴. Mass spectra were recorded for 800 – 16000 m/z at 17500 resolution as defined at 400 m/z. The injection time was set to 200ms. Voltages applied to the transfer optics were optimized to allow for ion transmission while minimizing unintentional ion activation. Mass spectra were deconvoluted with UniDec version 4.0.0 beta⁵⁵. For size exclusion chromatography coupled to native MS, unphosphorylated and phosphorylated samples were buffer exchanged into 200 mM AmAc and combined to ensure the presence of 0–4 phosphorylated pGFP-4Y at sufficient concentration for subsequent MS detection. 40 pmole protein were injected onto a MabPac SEC-1 column (5 μM 300Å, 2.1 × 300 mm; Thermo Fisher Scientific) and eluted with 200 mM AmAc at a flow rate of 50 μL/min. RNAseA, Cytochrome C, Myoglobin, Carbonic anhydrase and Alcohol dehydrogenase were used as standards to calibrate the SEC column.

Tryptic Digest MS/MS –

pGFP-4Y samples (25 μl of 1 mg/ml or 50 μM) in 10 mM HEPES pH 7.5 containing 5 mM MgCl2 and 100 mM NaCl were mixed 1:1 with 0.2% PPS Silent Surfactant (Expedeon, UK) in 50 mM Tris pH 8. Tryptic digestion was initiated by the addition of one μg trypsin overnight at 37 C. Samples were diluted 100-fold to a final concentration of 0.25 pmol/μl in LC loading buffer (2% acetonitrile with 0.1% trifluoroacetic acid), and three μl were injected onto the LCMS system.

Synthetic peptides to be used as standards were ordered from Genscript. Dilutions of the synthetic peptides were solubilized in LC loading buffer where the non-phosphorylated peptide was kept at a constant 400 fmol/μl and the phosphopeptide was diluted 1-, 4-, 20-, and 40-fold in order to assess the quantitative change in mass spectrometric signal upon addition of a phosphate group. Three μl of the peptide mixtures were injected onto the LCMS system.

Using a Waters Nanoacquity HPLC system, samples were injected via the autosampler onto a 150-μm Kasil fritted trap packed with Reprosil-Pur C18-AQ (3-μm bead diameter, Dr. Maisch) to a bed length of 2 cm at a flow rate of 2 μl/min. After loading and desalting using a total volume of 8 μl of loading buffer, the trap was brought on-line with a pulled fused-silica capillary tip (75-μm i.d.) packed to a length of 30 cm with the same Dr. Maisch beads. The column and trap were mounted to a homemade ion source that was operated at room temperature. Peptides were eluted off the column using a gradient of 2–32% acetonitrile in 0.1% formic acid over 1h, followed by 32–60% acetonitrile over 5 minutes at a flow rate of 250 nl/min.

The mass spectrometer was operated using electrospray ionization (2 kV) with the heated transfer tube at 275 C using data dependent acquisition (DDA), whereby one ms1 scan (m/z 400–1600) was acquired with up to 10 ms2 spectra. All data was acquired at unit resolution. The automatic gain control targets for MS was 3e4, and 8e3 for MS/MS. The maximum fill times were 10 and 50 ms, respectively. The MS/MS spectra were acquired with an isolation width of 2.0 m/z and CID collision energy of 35%.

Bio-layer Interferometry –

The affinity for psGFP_DIA to domain IV of calpain was measured on an Octet Red 96 by ForteBio. We used 10mM HEPES pH 7.0, 150mM NaCl, 0.05%(v/v) Tween, 1mm CaCl₂, 0.25% (w/v) BSA as binding buffer. Domain IV was bound to streptavidin functionalized tips (18–5020 ForteBio) using the biotinlyated Avi-tag. Samples were diluted from 50μM phosphorylation buffer into the binding buffer to a final concentration of 500nM. The tips were equilibrated in binding buffer before initiating the binding reaction by dipping the tips into well containing pDIA-4Y at 500nm. The signal response reported after 500s of association to the tip.

Calpain Inhibition Assay –

Calpain protease kinetics were monitored by incubation of calpain-1 (C6108) with calpain-1 substrate II (208772 Sigma). Calpain-1 substrate is an internally quenched peptide functionalized with an EDANS group on the N-terminus and DABCYL group on the C-terminus. After cleavage by calpain, the peptide becomes fluorescent. 30nM Calpain with mixed with 20μM calpain-1 substrate II in 50mM HEPES pH 7.7, 0.1% (v/v) 2-mercaptopethanol, and 1mM CaCl₂ for 2 minutes. After 2 min, 500nM of the indicated pDIA_4Y or uncaged control peptides were mixed and then the fluorescence was excited at 335nm and the emission was measured at 505nm by a BioTek Synergy Neo2 Multi-mode plate reader for 2h.

Mammalian Cell Activation of pGFP_4Y :

Human embryonic kidney cells (HEK-293T) were obtained from ATCC (CRL-3216) and were not authenticated after acquisition. To generate a stable cell-line expressing GFP1–10, HEK-293T cells were transduced with lentiviral particles delivering the GFP1–10-P2A-Puro gene driven by the MND promoter. Cells were then selected with puromycin over a period of two weeks. GFP1–10 expression and complementation with GFP11 was verified through transfection of plasmid encoding GFP11, followed by fluorescence detection. Cells were cultured in DMEM (Gibco) that was supplemented with 1 mM L-glutamine (Gibco), 4.5 g/L D-Glucose (Gibco), 10% fetal bovine serum (FBS) and (1x) non-essential amino acids (Gibco). Cells were cultured at 37°C and 5% CO2 and passaged three times per week. For the transfection, HEK293T GFP1–10 cells were plated at 40% confluence. 24h later, the cells were transiently transfected with SRC-530F, pGFP_4Y_Tsk4, and pGFP_4Y_Tsk4/SRC-530F using linear 40-kDa polyethyleneimine, PEI-MAX (Polysciences).

Cell images were taken using the IN Cell Analyzer 2500 (GE) on a Greiner Screenstar microplate 655866 using a Nikon Plan Apo 60X/0.95mm objective. The EBFP was excited at 390nm and recorded with 0.008 second exposure. Activated split-GFP was excited at 473nm and recorded with 0.1s exposure time. Pictures were analyzed using INCarta software (GE, Version: 1.11.3667461). Cell areas were identified using blue fluorescence with a sensitivity at 50 (out of 100). These identified cells were used to record the fluorescence in both the green and blue channel. A region without cells was used as a background subtraction. The linear regression and confidence interval was produced by SciPy stats library using the linregress method. The mean green fluorescence was normalized to the mean blue fluorescence for each of 3 separate experiments, before determining the fold-activation from Src-530F expression.

Mammalian cell culture and cell line generation for pGFP_4S constructs:

K652 cell were acquired from the UCSF Cell and Genome Engineering Core (UCSF CGEC Catalog Number CCLZR466) and were not authenticated after acquisition. K562 cells were maintained in RPMI 1640 (Life Technologies) medium also supplemented with 10% FBS and 1% Anti-Anti. Cells were grown at 37 °C and 5% CO₂. Lenti-X 293T cells (Takara Bio) were transfected at 50% confluency in a 24-well plate (Corning) with lentiviral plasmids mixed with packaging plasmids (pMD2.G and pCMV-dR8.91), Lipofectamine 2000 (Thermo Fisher) and Opti-MEM (Life Technologies) according to manufacturer’s instruction. After 72h of virus production, the supernatant containing accumulated virus was filtered through a 0.45 μm filter and dispensed onto K562 cells seeded at 50% confluency. The media was replenished after 24h, and transduced cells were expanded to 6-well plate (Corning). Using a BD FACSAria™II Cell Sorter, the top 5% of mCherry-expressing cells were sorted and propagated in a 24-well plate. Sufficiently expanded cell lines were seeded into round-bottom Corning 96-well plates round-bottom for K562) at 2,000 cells/well for induction and flow cytometry experiments.

Drug preparation and cell line induction: Forskolin (Tocris) and IBMX (Tocris) were dissolved in DMSO (Sigma-Aldrich) in stock concentrations of 50 mM and 500 mM, respectively. Drugs were aliquoted in volumes suitable for single time-point inductions and stored at −20 °C, and thawed immediately before usage. At timepoints of 77, 54, 30, and 8h prior to analysis, drug stocks were vortexed and dissolved in appropriate media for each cell type and added atop triplicate wells for final concentrations of 50μM and 100μM for Forskolin and IBMX respectively. Corresponding volumes of DMSO were added to triplicate control wells.

Flow cytometry: All cells were washed with DPBS and resuspended in DPBS + 10% FBS. Flow cytometry experiments were conducted on a LSRFortessa (BD) with a four laser configuration (488 nm, 635 nm, 355 nm, 405 nm) using the software FACSDiva 8.0.1. GFP1–10 (excitation at 488 nm, 530/30 emission filter) and mCherry (excitation at 561 nm, 610/20 emission filter) fluorescence was collected for at least 10,000 single-cell events were recorded at a flow rate at 3 μL/s.

Data processing and analysis: Exported FCS files were analyzed with a custom Python script. Cells were density gated on SSC-A vs SSC-H distributions to capture singlets. For every gated cell, the phosphoswitch response level was calculated by dividing signal of interest (GFP1–10) by cassette expression level (mCherry), and the mean of the response was determined for each individual triplicate at each timepoint. The mean response of each DMSO treated well was determined for all cells at every timepoint, representing a baseline signal for each construct and timepoint pair. To determine the fluorescent fold-change upon Forskolin and IBMX induction of PKA activity, the mean normalized GFP1–10 responses were divided by the mean baseline signal for each triplicate well. Analysis depended on the Python libraries numpy (version 1.18.1) ⁵⁶, pandas, matplotlib, seaborn, scipy( version 1.4.1)⁵⁷, and fcsparser. Source code is available at available at https://github.com/weinberz/phosphoswitch, data is available at https://zenodo.org/record/5095560.

Data Availability:

Source data for Figures 1,2,3,4 and Supp. Fig 5 has been uploaded to the journal. The PKA switch in vivo data is available https://zenodo.org/record/5095560 DOI: 10.5281/zenodo.5095560. Tryptic digest MS/MS available via ProteomeXchange with identifier PXD027295. All protein sequences used in experiments are available as a supplementary file Protein_Sequences.xlsx. The design models experimentally tested are available as a supplementary protein databank file. The source backbone for the caged ITAM is PDB ID: 6DLC. The remainder of the experimental data and computational data are available at https://files.ipd.uw.edu/pub/PhosphateSwitch/DenovoPSwitch.zip

Code Availability:

Custom Rosetta Scripts protocol for design with Rosetta 2018.19 included at https://github.com/NickWoodall/PhosphoSwitch_Design. Source code for PKA in vivo analysis is available at available at https://github.com/weinberz/phosphoswitch.