Engineering mono- and multi-valent inhibitors on a modular scaffold† †Electronic supplementary information (ESI) available. See DOI: 10.1039/d0sc03175e

We exploit the simple modular architecture of repeat proteins to create a platform for single- and multi-functional display in diverse programmable geometries and demonstrate proof of concept by building potent inhibitors of a key signalling pathway.


Introduction
The relationship between protein structure and function has been a cornerstone of biology for decades. However, in recent years, the unstructured or intrinsically disordered regions of the eukaryotic proteome (40% in humans) have gained increasing interest. This is due in part to the abundance in these regions of short independently functioning binding modules known as MoRFs (molecular recognition features) or SLiMs (short linear motifs). 1 One approach to exploit these motifs for inhibiting protein-protein interactions (PPIs) is to chemically synthesise them in combination with modications such as cross-linking and macrocyclisation designed to improve affinity, half-life and cell penetration. 2 However, in nature high-affinity and high-specicity interactions and more complex regulatory mechanisms are achieved through multivalency and avidity, neither of which are straightforward to realise with conventional peptide technologies. It is also uncertain whether small molecules could ever act as effective inhibitors of such complex networks of multivalent contacts. 3,4 Consequently, in order to interrogate and ultimately to drug such intricate networks of inter-linked motifs, new technologies are needed. There are many multivalent antibody technologies leveraging the natural modularity (i.e. multi-domain nature) of immunoglobulins. 5,6 However, multivalency has been less successful in antibody-like domains, where it has only been achieved by connecting monovalent units in "beads-on-a-string" or by directly assembling them on antibody scaffolds. 7-10 Likewise, functional peptide motifs have been assembled on synthetic chemical scaffolds, 11,12 DNA 13,14 and protein scaffolds, [15][16][17][18][19][20][21] but no modular platform has been developed to date that is capable of combinatorial incorporation of multiple SLiMs in a single stable and reliable scaffold as well as presenting them with varied, precise and programmable geometries. Here we show that tandemrepeat proteins possess all the necessary features with which to build such a platform.
Tandem-repeat proteins comprise tandem arrays of small structural motifs that pack in a linear fashion to produce regular, elongated, quasi-one-dimensional architectures and function in binding to other proteins, small molecules or nucleic acids. The repetitive modular organisation of this architecture makes it straightforward both to dissect and to redesign the biophysical properties. [22][23][24][25][26][27] One of the simplest repeat structures is the tetratricopeptide repeat (TPR), 28,29 a 34residue motif comprising two antiparallel a-helices with adjacent repeats connected by a short turn (Fig. 1A, bottom). 30 Articial proteins comprising multiple copies of a consensusdesigned TPR (CTPR) sequence have been shown to be extraordinarily stable. 31 Moreover, the modular nature of the architecture means that consensus repeats are self-compatible and can be individually designed and put together in any order. We recently showed that the CTPR scaffold can accommodate peptide extensions in the loop between adjacent repeats up to as many as 25 amino acids without compromising the native structure. 32, 33 We then demonstrated that we could gra a SLiM from the protein Nrf2 that recognises the oncogenic protein Keap1 onto the scaffold and that we could not only recapitulate the native binding affinity but also improve it without need of sophisticated computational modelling. Thus, by combining this modular scaffold with SLiM graing we potentially have the capacity for diverse functionalization. 33 As proof of concept, we herein construct a set of monovalent and multivalent CTPR proteins to bind and inhibit the human poly (ADP) ribose polymerase (PARP) proteins tankyrase 1 and tankyrase 2 (referred to subsequently as hTNKS). hTNKS are unique in the PARP family in having a series of ankyrin-repeat cluster (ARC) subdomains (Fig. 1A). 34,35 Four of the ve ARCs (ARCI, ARCII, ARCIV and ARCV) mediate protein-protein interactions including substrate recognition by binding short (8 residue) peptide motifs. [36][37][38][39] hTNKS has been implicated in the regulation of a number of cellular processes, [40][41][42][43] but more recently the role of hTNKS in the regulation of Wnt pathway activity has received particular attention. 44,45 The Wnt pathway is the principle driver of colon cancer, and a number of studies have reported the development of small molecule inhibitors of the hTNKS catalytic activity as a means of modulating Wnt pathway activity. 44,[46][47][48][49] However, by targeting the catalytic PARP domain these small molecules may lack specicity for hTNKS over other PARP family members. [50][51][52] Moreover, they can only inhibit the catalytic function of hTNKS but not the well-documented non-catalytic functions. 53,54 To overcome these limitations, our aim was to target the unique substratebinding ARC subdomains and thereby block both catalytic and non-catalytic functions. Guettler et al. previously determined a consensus sequence that is recognised by the ARC subdomains (referred to subsequently as the tankyrase-binding peptide or TBP) 36 (Fig. 1A), and we have shown that macrocyclised TBPs are able to bind hTNKS and inhibit Wnt signalling. 55 We graed the TBP onto the inter-repeat loop of the CTPR scaffold to generate a series of mono-and multi-valent binding proteins (nTBP-CTPR2n). Our aim was to explore the effects of multivalency on the biophysical properties of the proteins and on their interactions with and cellular inhibition of hTNKS, which is itself multivalent. Further motivation came from the fact that several hTNKS substrates are themselves multivalent as they contain multiple TBPs. 56 In order to further extend the multivalent capabilities and to produce more complex binding geometries beyond what is possible with the linearly arrayed nTBP-CTPR2n format, we engineered a trimeric 'foldon' motif into the constructs (nTBP-CTPR2n-foldon). [57][58][59][60][61] Our results highlight the power of multivalent inhibitors for targeting complex protein-protein interactions, where occupancy-driven inhibition using simple monovalent molecules is unlikely to be effective.

nTBP-CTPR2n proteins: design of mono-and multi-valent tankyrase inhibitors
We functionalised the CTPR scaffold by graing an 8-residue hTNKS-binding consensus peptide (TBP), REAGDGEE, identied from a mutational analysis of hTNKS substrates, 36 onto the loop between two adjacent repeats (Fig. 1A). This minimal hTNKS-binding unit was then tandemly repeated to generate a series of mono-and multi-valent molecules named nTBP-CTPR2n, with n between one and four (Fig. 1C). The CTPR proteins adopt a superhelical conformation with eight repeats required to complete a superhelical turn. 62 Thus, in the nTBP-CTPR2n construct design, the TBP loops will be offset from each other by approximately 90 (Fig. 1D). All nTBP-CTPR2n constructs could be expressed and puried in high yields in E. coli ($20 mg per litre of culture). All proteins were soluble in solution.

Effect of TBP insertions on CTPR folding and stability
We rst investigated the effects on protein stability of graing the TBP sequence onto the inter-repeat loop using far-UV circular dichroism (CD). Given the high a-helical content of the nTBP-CTPR2n constructs, all CD spectra have the expected double minima at 208 nm and 222 nm similar to those observed previously for the CTPR proteins (Fig. S1A, † le), 32 conrming that all nTBP-CTPR2n constructs were correctly folded. There was a linear increase in the molar ellipticity at 222 nm with increasing size of the CTPR construct, which further conrms that the proteins were natively folded (Fig. S1A, † right).
Next, thermal unfolding was performed to investigate the stability of the nTBP-CTPR2n constructs. We used the ellipticity at 222 nm as a measure of a-helical content. All proteins were found to be extremely thermostable, with only the smallest protein, 1TBP-CTPR2, undergoing complete denaturation. The larger nTBP-CTPR2n proteins remained partly folded even at the highest temperature, conrming that the stability increases with increasing number of repeats. The midpoint of the unfolding transition (T M ) is therefore provided only for 1TBP-CTPR2 and corresponds to 75 C (Fig. S1B †). Following thermal denaturation, the samples were allowed to return to 20 C, and the CD spectra remeasured to evaluate the reversibility of the reaction. For all of the proteins, there was no signicant difference between the CD spectrum before and aer thermal denaturation, indicating that unfolding is reversible (Fig. S1C †).
Chemical-induced denaturation experiments were also performed, monitoring the uorescence of the tryptophan residues (one per repeat) (Fig. S2 †). A single transition was observed for all of the constructs, and the data were tted to a two-state model to give the midpoints of unfolding (D 50% ), m-values (m D-N , a parameter related to the change in solvent-exposure upon unfolding) and the free energies of unfolding ðDG H2O DÀN Þ (Table S2 †). There was a signicant increase in D 50% and mvalue between 1TBP-CTPR2 and 2TBP-CTPR4 and smaller increases for the larger proteins. Lastly, the nTBP-CTPR2n showed a decrease in the stability compared with their respective counterparts without the TBP loops. As previously shown, 32 loop insertion has a destabilizing effect through both the entropic cost of loop closure and a consequent weakened coupling of adjacent repeats.
We next investigated whether all of the TBP loops in the multivalent constructs are accessible for binding by measuring the affinity and stoichiometry for each nTBP-CTPR2n construct binding to hTNKS2 ARC4 (the fourth ankyrin-repeat cluster of hTNKS2), which contains a single binding site for the TBP ( Fig. 2A). All nTBP-CTPR2n constructs showed similar lowmicromolar affinities, and the stoichiometry was found to increase from 1TBP-CTPR2 to 3TBP-CTPR6 in proportion to the number of TBP loops, conrming that all sites are available for binding (Table 1). For 4TBP-CTPR8, the stoichiometry is lower than 4, likely because steric hindrance precludes binding of four hTNKS2 ARC4 molecules to one 4TBP-CTPR8 molecule. No binding could be detected for a control CTPR6 protein containing no TBP (Fig. S3 †). The changes in the free energy (DG), enthalpy (DH) and entropy (ÀTDS) for the binding of the four nTBP-CTPR2n proteins to hTNKS2 ARC4 are plotted in Fig. 2B and into 1TBP-CTPR2 (50 mM), hTNKS2 ARC4 (500 mM) into 2TBP-CTPR4 (25 mM), hTNKS2 ARC4 (500 mM) into 3TBP-CTPR6 (16.6 mM), hTNKS2 ARC4 (500 mM) into 4TBP-CTPR8 (12.5 mM); experiments were performed at 25 C. The concentrations of nTBP-CTPR2n used were chosen so that the molar ratio between the number of TBP loops and TNKS2 ARC4 protein was constant. (B) Thermodynamic parameters (DG, DH and ÀTDS) obtained by ITC for the binding of the four nTBP-CTPR2n proteins to hTNKS2 ARC4. For each parameter, the average from two independent experiments is plotted. (C) HiBiT-tagged hTNKS2 protein levels measured by luminescence intensity throughout each step of the HiBiT-qIP method. IP was performed in the presence of HA-tagged 3TBP-CTPR6, CTPR6 and an empty vector. Average values and standard deviation from two independent biological replicates are shown.  Table S3. † The results show that in all cases the interaction is enthalpy-driven. There is a slight tendency for both the enthalphic gain and entropic cost of binding to increase with increasing number of hTNKS2-binding loops. The increase in enthalpy with increasing number of binding loops is to be expected, as having multiple binding sites in close proximity should make it more likely for the protein to bind a target again aer dissociation. The greater entropic cost of ordering a greater number of loops upon binding is also as expected. The greater entropic cost is offset by the greater enthalpic gain, and hence the values of DG are similar for all four nTBP-CTPR2n proteins.

nTBP-CTPR2n proteins and hTNKS interact in the cell
Binding between hTNKS and nTBP-CTPR2n in the cell was conrmed using a modied version of the HiBiT-based quantitative immunoprecipitation (HiBiT-qIP) method, that provides a much faster and more sensitive read-out than conventional IP by western blot. 66 The method relies on the split luciferase complementation of the two NanoLuc fragments, HiBiT and LgBiT, providing a sensitive and quantitative method to track any HiBiT-tagged protein. 67 HA-tagged nTBP-CTPR2n was used as bait to pull-down HiBiT-tagged hTNKS2 using an anti-HA resin. The presence of HiBiT-tagged hTNKS2 was followed and quantied through the luminescence signal generated by the HiBiT technology, allowing a real-time measurement of the pulled-down target throughout the process. The results show that HiBiT-tagged hTNKS2 is pulled down by 3TBP-CTPR6 but not by the control CTPR6 construct, conrming that the graed TBP loop mediates the binding to hTNKS2 within the cellular environment also. Moreover, no non-specic binding of HiBiTtagged hTNKS2 to the Anti-HA Agarose resin was detected (Fig. 2C). Nano-Glo HiBiT blotting further conrmed the presence of HiBiT-tagged hTNKS2 in the eluted samples ( Fig. S4 †). Likewise, the bait constructs and tubulin were detected by standard western blot methods (Fig. S4 †).

Design of trimeric nTBP-CTPR2n-foldon constructs
We introduced the foldon motif at the C-terminus of each nTBP-CTPR2n construct to induce trimerisation. In this way, each homo-trimeric construct will display up to twelve TBP loops and in a different geometry compared to the corresponding monomeric constructs. The foldon is a short (30 residue) but highly stable motif of the protein britin from the T4 bacteriophage. 57 The so-called 'NCCF' construct of britin contains an Nterminal trimeric coiled-coil region followed by the C-terminal foldon. 68 Here, we introduced the foldon and part of its preceding coiled-coil domain into the nTBP-CTPR2n constructs by graing the helical coiled-coil onto a newly introduced Cterminal CTPR helix (Fig. S5, † top). This helix corresponds to an extra helix A (i.e. half of one CTPR). As all a-helices share the same heptad conformation, it is possible to gra residues of a particular heptad position from a 'guest' helix onto a 'host' helix. When graing a new site onto a CTPR helix, the native residues at the interface between the helix and the neighbouring repeat need to be preserved in order for the graed helix to remain folded onto the rest of the CTPR scaffold. These residues were identied as Ala3, Leu7 and Ala10 of helix A, which are involved in the inter-repeat interface with the helix B of an adjacent repeat. Part of the remaining residues of helix A were substituted with the residues of the britin NCCF coiled-coil domain (residues 65 to 77) involved in the trimeric interface. This was followed by an insertion of the remaining C-terminal end of the coiled-coil domain and the foldon (NCCF residues 78 to 109) (Fig. S5, † top). The resulting constructs are referred as nTBP-CTPR2n-foldon, and the protein sequences are given in Table S1. † A model of the trimeric CTPR2-foldon protein is shown in Fig. 1E. This protein is expected to display the TBP loops at an angle of 120 to one another. The nTBP-CTPR2nfoldon proteins were expressed and puried as for their monomeric counterparts. Analysis by native gel electrophoresis is consistent with trimerisation of those proteins that contain the foldon domain (Fig. S5, † bottom). The series lacking the foldon domain (1TBP-CTPR2, 2TBP-CTPR4, 4TBP-CTPR8, calculated molecular weights 11.3 kDa, 20.6 kDa and 39.4 kDa, respectively) run at increasing molecular weights between the 20 kDa and 66 kDa molecular weight markers. In contrast, those with the foldon domain (1TBP-CTPR2-foldon, 2TBP-CTPR4-foldon, 4TBP-CTPR8-foldon, calculated molecular weights as trimers 49.2 kDa, 77.4 kDa and 133.5 kDa, respectively) run at higher molecular weights between the 66 kDa and 146 kDa markers. The oligomerization was further analysed by size-exclusion chromatography coupled with multi-angle light scattering (SEC-MALS). The results show that 3TBP-CTPR6 and 3TBP-CTPR6-foldon (representative of the two types of nTBP-CTPR2n constructs) are, respectively, monomeric and trimeric in solution (Fig. S6 †). The average molecular weight obtained for 3TBP-CTPR6 (30 kDa calculated molecular weight) is 38 kDa, consistent with it being monomeric. The majority of the 3TBP-CTPR6-foldon sample (105 kDa calculated molecular weight as a trimer) elutes at an average molecular weight of 131 kDa, consistent with it forming a trimer with some less-populated oligomeric species at lower and higher molecular weight also present (99, 204 and 321 kDa, respectively).

Multivalent interactions induce the formation of large macromolecular assemblies with variable internal dynamics
When two multivalent proteins interact, they have the potential to engage with multiple partners simultaneously, leading to the formation of large supramolecular assemblies (see for example work by Rosen and colleagues 69 ). This is indeed what we observe with hTNKS2 and the multivalent tankyrase-binding CTPR proteins (schematically represented in Fig. 3A). These species could be detected using a co-precipitation assay, in which the multivalent 3TBP-CTPR6 and 3TBP-CTPR6-foldon constructs, chosen as representatives of the linear and trimeric arrays, respectively, co-precipitate in the presence of the multivalent hTNKS2 construct corresponding to the rst three ARC subdomains (hTNKS2 ARC1-3). It is worth noting that, despite 3TBP-CTPR6 and 3TBP-CTPR6-foldon being incubated at the same TBP molar concentration, the trimeric 3TBP-CTPR6foldon construct precipitated to a higher extent and lead to an increased precipitation rate of hTNKS2 ARC1-3. In contrast, the mono-valent 1TBP-CTPR2 construct and the control CTPR6 construct remained in the supernatant, conrming that multivalency of TBP loops is required for the formation of a large supramolecular assembly with multivalent hTNKS2 (Fig. 3B).
Titration of increasing concentrations of 3TBP-CTPR6 or 3TBP-CTPR6-foldon proteins into hTNKS2 ARC1-3 showed that the extent of co-precipitation was dependent on the molar ratio of the two interacting molecules, until saturation is reached. Analysis of the gel band intensities allowed us to calculate the molar concentration of each species in the pellet, providing an estimation of the stoichiometries of the macromolecular complexes that had precipitated ( Fig. S7 and Table S4 †). The results indicate a 1 : 1 stoichiometry for the 3TBP- CTPR6 : hTNKS2 ARC1-3 complex, and a higher stoichiometry of 1 : 2 for the 3TBP-CTPR6-foldon : hTNKS2 ARC1-3 complex, explaining why the trimeric 3TBP-CTPR6-foldon can induce a greater extent of hTNKS2 ARC1-3 co-precipitation than the linear 3TBP-CTPR6.
Next, these protein assemblies were examined using negative stain Transmission Electron Microscopy (TEM). The results shown in Fig. S8 † conrm that the large, macromolecular clusters are formed only when both interacting partners are multivalent.
To visualise the effect of multivalency inside the cell, HEK293T cells were co-transfected with an hTNKS2 construct encoding the ARC1-5 subdomains fused to eGFP (referred to as hTNKS2 ARC1-5-eGFP) and the multivalent 3TBP-CTPR6 or 3TBP-CTPR6-foldon constructs fused to mCherry. Large, polymeric species were visible in the cytoplasm, where the signals of eGFP and mCherry perfectly colocalised (Fig. 4). In agreement with the ndings of the coprecipitation assays, these large assemblies were formed only when the multivalent 3TBP-CTPR6 and 3TBP-CTPR6-foldon proteins were co-expressed with hTNKS2 ARC1-5, which is also multivalent. In contrast, when the same CTPR proteins were co-expressed with a monovalent hTNKS2 construct comprising only the rst ARC domain (hTNKS2 ARC1-eGFP), no polymeric assemblies were observed (Fig. S9 †). Likewise, when the multivalent hTNKS2 was coexpressed with the monovalent 1TBP-CTRP2, no polymeric species were observed (Fig. 4). To further compare the properties of the macromolecular assemblies generated in the cell by 3TBP-CTPR6 and 3TBP-CTPR6-foldon bound to hTNKS2 ARC1-5, Fluorescence Recovery Aer Photobleaching (FRAP) was used to assess the dynamics by measuring the recovery rate of the eGFP signal (Fig. 5). The assemblies induced by 3TBP-CTPR6 showed uorescence recovery within 30 seconds, suggesting that the proteins within the assembly are dynamic in nature. In contrast, the assemblies generated by the 3TBP-CTPR6-foldon showed no uorescence recovery within the same timeframe, suggesting that the presence of nine hTNKS binding sites on 3TBP- Fig. 4 Fluorescence microscopy of HEK293T cells co-transfected with mCherry-tagged CTPR proteins and eGFP-tagged hTNKS2 ARC1-5. Colocalisation in large macromolecular clusters is observed for eGFP-tagged hTNKS2 ARC1-5 in combination with mCherry-tagged 3TBP-CTPR6 or with mCherry-tagged 3TBP-CTPR6-foldon. These clusters are not observed for the mono-valent mCherry-tagged 1TBP-CTPR2 protein or the control mCherry-tagged CTPR6 protein. Scale bars for all images are 10 mm.
CTPR6-foldon leads to the formation of a more interconnected, rigid assembly.

nTBP-CTPR2n and nTBP-CTPR2n-foldon proteins inhibit Wnt signalling
We next evaluated the inhibitory capabilities of the monomeric and trimeric nTBP-CTPR2n constructs on Wnt signalling using the TOPFLASH reporter assay. Treatment of HEK293T cells with all the nTBP-CTPR2n constructs, with the exception of 1TBP-CTPR2, led to a reduction in Wnt pathway activity compared with the controls (empty vector, CTPR2 and CTPR6 and 3RL-CTPR6 containing a non-binding peptide named 'RL', random loop) (Fig. 6A). To assess the levels of nTBP-CTPR2n proteins in the cell during the TOPFLASH assay, we used N-terminally HiBiT-tagged nTBP-CTPR2n proteins. The 11-amino-acid HiBiT tag allows a more sensitive and quantitative measurement than western blot via the antibody-free split NanoLuc® luciferase technology (Promega). The results show that the extent of Wnt inhibition is proportional to the nTBP-CTPR2n protein levels as quantied 24 h aer transfection (Fig. S10 †). In particular, it is clear that the CTPR2 protein, whether in the monomeric or the trimeric format, is present at much lower levels than the larger proteins. To test whether the CTPR2 constructs are subject to proteasome-induced degradation, transfected wells were incubated with the proteasome inhibitor MG132 for 5 hours. The results show that there is an approximate 3-fold increase in protein levels in the presence of MG132, the exceptions being the two smallest constructs 1TBP-CTPR2 and CTPR2, whose levels are affected to a greater extent (Fig. S11 †). Thus, the smallest, least thermodynamically stable CTPR protein is the most susceptible to proteasome-mediated degradation, and the foldon domain enhances its intracellular stability. In addition, the formation of the large macromolecular assemblies is likely to protect the multivalent hTNKSbinding CTPR proteins from proteolysis, further explaining why the monovalent 1TBP-CTPR2 and the non-binding control CTPRs do not accumulate in the cell. Lastly, we conrmed that none of the constructs showed cytotoxicity (Fig. S12 †).

Intracellular delivery of 3TBP-CTPR6 by encapsulation with fusogenic liposomes induces high levels of Wnt inhibition
Intracellular delivery of proteins is challenging. 70 We used an encapsulation approach in which puried protein (3TBP-CTPR6) was loaded into fusogenic liposomes. Fusogenic liposomes have been shown to merge with the cell membrane and deliver their cargo directly into the cell cytoplasm. 71 Encapsulation of 3TBP-CTPR6 in liposomes could minimise their premature degradation and adverse immune response, if any. 72 Liposomal formulation of 3TBP-CTPR6 (FL-3TBP-CTPR6) and control liposomes without protein (FL) were prepared as  described in the experimental section. Both FL and FL-3TBP-CTPR6 had highly positive surface with zeta potential (ZP) values of +126 mV and +75.7 mV, respectively, at pH 7.4 (Fig. S13 †). The lower ZP of FL-3TBP-CTPR6 is due to encapsulation/association of the negatively charged 3TBP-CTPR6 (pI $ 4.8) with liposomes. The hydrodynamic sizes of FL and FL-3TBP-CTPR6 were similar, at $106 nm (Fig. S13 †). FL and FL-3TBP-CTPR6 did not show any signicant cytotoxicity even for amounts three times higher than those used in our experiments (Fig. S14 †). To visualize intracellular delivery under the confocal microscope, 3TBP-CTPR6 was uorescently labelled with rhodamine. Liposomes were prepared using the labelled protein and designated as FL-3TBP-CTPR6-RITC. Confocal images of HEK293T cells treated with FL-3TBP-CTPR6-RITC (Fig. S15 †) clearly suggest that protein is inside the cells and is distributed throughout the cytoplasm. Cells treated with FL did not show any signal for 3TBP-CTPR6-RITC (Fig. S16 †).
We next tested the effect of liposome-delivered 3TBP-CTPR6 on Wnt signaling using the TOPFLASH assay. A decrease in the TOPFLASH activity was observed, reaching around 50% aer 6 hours incubation. Importantly, decreasing TOPFLASH activity correlated with increasing concentration of 3TBP-CTPR6 (Fig. 6B, le), indicating a dose-dependent inhibition of hTNKS. Treatment with free 3TBP-CTPR6 did not affect the luciferase expression levels as the protein cannot enter cells on its own. Also, FL did not alter luciferase expression, indicating that membrane fusion of liposomes does not interfere with the intracellular Wnt signalling. Further, control protein, CTPR6, lacking hTNKS-binding units and delivered using liposomes, did not change the intracellular luciferase levels.
For comparison, we also performed the TOPFLASH assay in the presence of three well-characterized small molecule hTNKS inhibitors (XAV939, IWR-1 and G007) that bind the catalytic domain of hTNKS. Critically, whereas the small molecule inhibitors are effective when used prophylactically (i.e. before Wnt pathway activation) (Fig. S17 †), they are ineffective when used interventionally, i.e. aer Wnt pathway activation (as would be the case in the clinic, since patients will have elevated Wnt signaling) (Fig. 6B, right). Strikingly, FL-3TBP-CTPR6 at the highest dose had a signicantly greater inhibitory effect than any of the small molecule hTNKS inhibitors (Fig. 6B). This result shows that targeting the non-catalytic activity of hTNKS could be an effective treatment where small-molecule inhibitors have so far failed.

Discussion
In this study, we created potent hTNKS inhibitors by combining target specicity with multivalency, two features that have not been explored in previous drug development efforts against this target. We demonstrate that modular CTPR proteins can be engineered to display both single as well as multiple copies of short hTNKS binding motifs by graing the TBP onto the loops between adjacent repeats. Helical graing was also applied to engineer trimeric CTPR constructs with enhanced multivalent capabilities. We show that CTPR proteins with one or more graed hTNKS-binding motifs are stable, correctly folded and exceptionally active in inhibiting Wnt signalling. The effect of multivalency in both partners of the interaction -hTNKS and the designed nTBP-CTPR2n proteinsmanifests in the formation of large, intracellular macromolecular assemblies with different dynamics dependent on the conguration (i.e. monomeric versus trimeric). Thus, hTNKS inhibition by these multivalent molecules is enhanced by clustering of the protein within these structures. Such a mechanism of action warrants further investigation and should be explored for future therapeutic efforts against hTNKS and other multivalent targets. Our experiments using fusogenic liposome delivery of 3TBP-CTPR6, one of our most potent molecules, further demonstrated rapid inhibition of Wnt pathway activity within 6 hours of treatment of HEK293T cells pre-stimulated with Wnt3A ligand. This is a noteworthy result that demonstrates interventional inhibition of the pathway, the scenario required for targeting tumours dependent on deregulated Wnt pathway activity in vivo. Moreover, the luciferase protein reporter of Wnt pathway activity has a relatively long halflife of approximately 12 hours, and therefore inhibition aer 6 hours treatment likely reects a much higher attenuation of Wnt pathway activity beyond the measured 50% inhibition.
We compared our molecules with three well-characterised small molecule hTNKS inhibitors and, in agreement with previous studies, we observed around 90% inhibition of Wnt pathway activity for two of them when used prophylactically: i.e. when they are applied either before Wnt pathway activation by Wnt3A ligand or in combination with it. Critically, however, to be of clinical use, such drugs will need to work interventionally; interventional treatment with these small molecule hTNKS inhibitors has been shown by our group and others to be much less effective due to a cell-intrinsic feed-forward mechanism preserving Wnt pathway activity. 73 Thus, the effectiveness of liposome-delivered 3TBP-CTPR6 for interventional inhibition of Wnt pathway activity demonstrated here further highlights the power of our approach.
The work presented here points to the tremendous potential of the repeat-protein scaffold for the rational design of multivalentand thereby potentially multi-specicmolecules for inhibiting PPIs: (1) First, CTPR proteins are small and ultrastable without need of disulphide bonds. As a result, they can accommodate small or even large sequence insertions yet remain folded and stable. 32,74 Moreover, multiple such insertions are also possible without compromising the fold; it is doubtful whether other small single-domain scaffolds based on globular structures could do the same. (2) The second key consequence of the repeat-protein stability is that they can be produced recombinantly in very high yields. (3) Lastly, the repetitive, modular nature of the scaffold is a particularly unique selling point; we have shown here that we can exploit these characteristics to display one or multiple binding motifs in a precise and programmable manner. It has been estimated that there are around 100 000 such SLiMs in the human proteome, 75 each one of which provides a potential starting point for drug discovery. The simple cut-and-paste approach used here, requiring little or no in silico design procedures, could be applied to harness some of these SLiMs, and the platform thus has potential to be used as a synthetic tool in diverse applications ranging from target validation to PPI inhibition, pathway modulation and ultimately molecular therapeutics.
Our condence in the broad applicability of the scaffold to the display of diverse SLiMs arises from the nding that we could effectively gra two SLiMs with very different binding conformations onto the CTPR loops. The Keap1-binding peptide from Nrf2 adopts a tight turn-like conformation, 33 whereas the TBP in this work binds to ARC domains in a highly extended conformation. 36,55 It might appear that a limiting factor in this multivalent platform would be the potential for steric clashes between target and scaffold as well as between targets in a multivalent display. The rst challenge should be approached on a case-bycase basis, as it is common to all peptide display methodology. The second problem can be mitigated by exploiting the modular nature of this technology. For example, the ITC data show that both 3TBP-CTPR6 and 4TBP-CTPR8 bind to approximately three hTNKS2 ARC4 molecules, suggesting that a maximum has been reached. We limited our graing to one peptide for every two repeats, but peptides could be set further apart by increasing the number of intervening repeats. The rigidity of the CTPR scaffold ultimately works as a molecular ruler of predened length and pitch. Furthermore, we have shown that trimerisation by means of the small foldon domain enhances the multivalent capability of the CTPR scaffold, presumably by providing more degrees of freedom for the formation of macromolecular complexes. Future work will focus on building bi-and multi-functional CTPR arrays using multiple SLiMs to bring two or more different targets together in a predened geometry, for example to redirect enzyme activity, alter subcellular localisation or reprogramme signaling pathways.

Molecular biology
The gene encoding 1TBP-CTPR2 was synthesised by Integrated DNA Technologies. It was designed with appropriate restriction sites at each terminus to allow ligation and the generation of multivalent constructs by concatemerisation. 76 We used a variant of the original CTPR sequence developed by Regan and coworkers modied at two positions (D18Q and E19K) 63,64 to further enhance the stability of the CTPR scaffold. Our CTPR proteins do not have the extra C-terminal helix that has been used by some groups to improve solubility (Fig. 1B). 25,31,77,78 Likewise, the gene encoding the foldon motif was synthesized and ligated at the 3 0 end of each nTBP-CTPR2n construct using the appropriate restriction enzymes (Thermo Fisher Scientic). The genes encoding hTNKS2 fragments were also amplied and ligated using the same method. Different combinations of tags (His, HA, HiBiT, eGFP, mCherry) were fused at the N-or Cterminal end of each construct. Genes were ligated into a modied pRSRET B vector (Thermo Fisher Scientic) for bacterial expression, or into the pcDNA3.1(À) vector (Thermo Fisher Scientic) for mammalian expression. Correct clones were conrmed by DNA sequencing (Eurons Genomics). Protein sequences of all of the generated CTPR constructs are listed in Table S1. † An E. coli expression plasmid encoding TNKS2 ARC4 was previously generated. 55

Protein purication
The pRSET B vectors encoding the His-tagged nTBP-CTPR2n constructs were transformed into chemically competent Escherichia coli C41 cells. Colonies were grown at 37 C in 2xYT media (Formedium) containing ampicillin (50 mg mL À1 ), shaking at 220 rpm until the optical density at 600 nm was $0.6. Protein expression was induced with 0.5 mM isopropyl b-D-1thiogalactopyranoside (IPTG) for 16-20 hours at 20 C. Cells were harvested and resuspended in buffer A (50 mM Tris-HCl, 150 mM NaCl, pH 8.0) including EDTA-free SIGMAFAST protease inhibitors (Sigma-Aldrich), Lysozyme and DNase I. Cells were lysed by high-pressure homogenization using an Emulsiex C5 homogenizer (Avestin) at 15 000 psi, and cell debris was removed by centrifugation steps at 40 000g for 40 minutes. nTBP-CTPR2n constructs were puried by immobilised metal ion affinity chromatography (IMAC) on a 5 mL HisTrap Excel column according to the manufacturer's instructions (Cytiva). The column was washed with 20 CV of buffer A containing 20 mM imidazole to prevent nonspecic interaction of lysate proteins to the beads. Proteins were eluted with buffer B (50 mM Tris-HCl, 150 mM NaCl, 500 mM imidazole, pH 8.0). All proteins were subsequently puried by sizeexclusion using a HiLoad 16/600 Superdex 75 pg column (Cytiva) pre-equilibrated in buffer A. Puried protein was ashfrozen and stored at À80 C until further use.
Proteins to be analysed by size-exclusion chromatography coupled with multi-angle light scattering (SEC-MALS) were further puried by SEC using a Superdex 200 10/300 GL (Cytiva) preequilibrated in buffer A. Puried protein samples (100 mL) were subjected to SEC-MALS, performed on a SEC Superdex 200 increase 10/300 GL column (Cytiva) preequilibrated in buffer A, in line with a multi-angle light scattering module (DAWN-8+; Wyatt Technologies) and a differential refractometer (Optilab T-rEX; Wyatt Technologies). The light scattering and protein concentration at each point across the peaks in the chromatograph were used to determine the absolute molecular mass from the intercept of the Debye plot using Zimm's model as implemented in the ASTRA v7.3.0.11 soware (Wyatt Technologies). To determine interdetector delay volumes, band-broadening constants and detector intensity normalization constants for the instrument, BSA (Thermo Fisher Scientic) was used as a standard prior-to sample measurement. Data were plotted with the program PRISM 8 (GraphPad Soware Inc.).
His-tagged Tankyrase-2 Ankyrin Repeat Cluster 4 (residues 488-649, referred to as hTNKS2 ARC4) and Tankyrase-2 Ankyrin Repeat Clusters 1-3 (residues 2-485, referred to as hTNKS2 ARC1-3) were expressed and puried as reported above and previously, 55 with all buffers containing 2 mM DTT. To remove the His tag, proteins were incubated with 125 U of thrombin (Sigma-Aldrich) overnight at RT on a rotating mixer. Cleaved protein was further run over IMAC, and the ow-through was collected. The nal purity and identity of all the proteins were determined by SDS-PAGE and MALDI mass spectrometry (PNAC, Department of Biochemistry, Cambridge, UK), respectively. Concentrations were determined by absorbance at 280 nm using the calculated extinction coefficient (ExPaSy ProtParam) 79 for each protein.

Circular dichroism spectroscopy (CD) and thermal denaturation experiments
All protein samples were prepared in 10 mM sodium phosphate buffer, 150 mM NaCl, pH 7.4, at a nal concentration of 20 mM. Circular dichroism (CD) measurements were performed on a Chirascan CD spectrophotometer (Applied Photophysics) at 20 C. CD spectra were acquired at wavelengths between 200 nm and 280 nm using a 1 nm bandwidth at a scan speed of 120 nm per minute. Readings were repeated in triplicate and the data averaged. Thermal denaturation experiments were carried out by increasing the temperature of the protein samples from 20 C to 92 C in 1 C steps, and the ellipticity at 222 nm was monitored. Readings were repeated ve times, averaged and tted to a sigmoidal Boltzmann equation including a sloping baseline term using PRISM (GraphPad Soware Inc.). Subsequently, the sample was allowed to cool back to 20 C and the CD spectrum was re-acquired.

Equilibrium denaturation monitored by uorescence spectroscopy
High-throughput equilibrium denaturation experiments were performed as previously described. 74 All protein samples (at concentrations between 6-10 mM) were prepared in 50 mM sodium phosphate buffer, 150 mM NaCl, pH 6.8. Protein samples were dispensed into 96-well, half-area, black polystyrene plates (Corning) using a Microlab ML510B dispenser (Hamilton) containing different concentrations of guanidinium hydrochloride (GdmHCl) between 0 M and 6 M in increments of 0.1 M per well. Plates were covered with 96-well Microplate Aluminium Sealing Tape (Corning) to avoid evaporation and incubated at 25 C for 1 hour. Plate measurements were carried out on a CLARIOstar Plate Reader (BMG LABTECH) with a tryptophan-detection lter set consisting of an excitation lter (275-285 nm), a dichroic PL325 nm, and an emission lter (350-370 nm) at 25 C. Each measurement was performed in triplicate. The plots of uorescence intensity versus denaturant concentration were tted to a two-state model with sloping baselines, in which only fully folded and fully unfolded states are populated, using the following equation: 80

Isothermal titration calorimetry (ITC)
ITC was performed at 25 C using a MicroCal VP-ITC (Malvern Panalytical). Proteins were dialysed into 10 mM sodium phosphate buffer, 150 mM NaCl, 0.5 mM TCEP, pH 7.4. TNKS2 ARC4 was titrated into the sample cell containing one of the nTBP-CTPR2n constructs. Injections of TNKS2 ARC4 into the cell were initiated with one injection of 5 mL over 6 seconds, followed by 29 injections of 10 mL over 12 seconds. Raw data were rst subjected to baseline determination using NITPIC soware 81 and were tted using the OneSite model within Origin 7.0 soware to a non-linear regression.

Immunoprecipitation (IP)
HEK293T cells were cultured in 15 mL cell growth media (Dulbecco's Modied Eagles Medium (DMEM), high glucose, pyruvate, (Thermo Fisher Scientic) supplemented with 10% Fetal Bovine Serum (Sigma-Aldrich) and 1Â Penicillin/ Streptomycin (Gibco). HEK293T cells in 10 cm dishes were transfected with 3.5 mg of pcDNA3.1(À) vector encoding HiBiTtagged hTNKS2 and 3.5 mg of pcDNA3.1(À) vector encoding a HA-tagged 3TBP-CTPR6, CTPR6 or empty vector. Transfection was performed using Lipofectamine 2000 transfection reagent according to the manufacturer's protocol (Thermo Fisher Scientic). Following 48 hours from transfection, cells were washed twice in PBS and lysed for 30 minutes in 1 mL cold lysis buffer (50 mM Tris-HCl, 225 mM KCl, 1% NP-40, pH 7.5) including protease inhibitor cocktail (Sigma-Aldrich), 10 mM NaF, 1 mM Na 3 VO 4 , 1 mM PMSF. Cell debris was removed by centrifugation at 16 000g for 20 minutes. IP of HiBiT-tagged hTNKS2 bound to HA-tagged 3TBP-CTPR6 was performed by incubating the supernatant for 4 hours with 20 mL monoclonal anti-HA Agarose (Sigma-Aldrich), pre-equilibrated in lysis buffer. Anti-HA agarose resin was washed 4 times with 500 mL lysis buffer. Elution was performed by adding 20 mL 2Â loading dye containing denaturant SDS to the settled resin. All steps following cell washing were performed at 4 C. The presence of HiBiT-tagged hTNKS2 throughout the IP process was detected by mixing 5 mL of each sample with the Nano-Glo HiBiT Lytic Detection System according to the manufacturer's instructions (Promega). The amount of HiBiT-tagged hTNKS2 bound to the beads was quantied by mixing 5 mL of 50% anti-HA Agarose resin slurry with the Nano-Glo HiBiT Lytic Detection System before elution. Data were obtained from two biological replicates.
Co-precipitation assay 20 mL of 10 mM hTNKS2 ARC1-3 was mixed with an equal volume of 1TBP-CTPR2, 3TBP-CTPR6, 3TBP-CTPR6-foldon or CTPR6 at the indicated concentration in co-precipitation buffer (50 mM Tris-HCl, 150 mM NaCl, 0.5 mM TCEP, pH 7.3). Samples were incubated at room temperature for 1 hour and centrifuged at 20 000g for 30 minutes at room temperature. The supernatant was collected and the pellet resuspended in an equal volume of buffer. Samples were run on a polyacrylamide gel, and the gels were imaged using the Li-COR Odyssey Fc Imaging System and protein band densities were analysed using the Image Studio Lite 5.2 soware.
Fluorescence microscopy and FRAP 1 Â 10 5 HEK293T cells were seeded overnight in 700 mL of cell growth media in a m-Slide (Ibidi). The following day, cells were transfected with 400 ng of m-Cherry tagged nTBP-CTPR2n, nTBP-CTPR2n-foldon or the control CTPR construct in combination with 600 ng of eGFP-tagged TNKS2 ARC1-5 or 100 ng of TNKS2 ARC1, using 1.75 mL lipofectamine 2000 transfection reagent according to the manufacturer's protocol (Thermo Fisher Scientic). Following 48 hours from transfection, cells were imaged on a Leica TCS SP5 confocal microscope with a 100Â oil-immersion objective lens (1.4 numerical aperture). Excitation and lters were as follows: eGFP, excitation at 488 nm, emission from 500-540 nm; mCherry, excitation at 543 nm, emission from 600-630 nm. For FRAP, individual circular regions of interest (ROI) were bleached using the 488 laser at 100% power for 5 seconds. Fluorescence intensity changes were recorded comparing 5 pre-bleaching frames with 60 post-bleaching frames (1.3 seconds per frame). Data were analysed using Leica LAS AF suite soware and data were normalised.
Transmission electron microscopy (TEM) hTNKS2 ARC1-3 was mixed with an equal volume and concentration (5 mM) of 1TBP-CTPR2, 3TBP-CTPR6, 3TBP-CTPR6-foldon, CTPR6 or co-precipitation buffer. Samples were incubated at room temperature for 1 hour. They were then dispensed on carbon support lm, 400 mesh, 3 mm nickel grids (EM Resolutions Ltd, Saffron Walden, UK) and stained with 2% uranyl acetate (w/v). The samples were imaged on a FEI Tecnai G2 transmission electron microscope in the Cambridge Advanced Imaging Centre (CAIC, University of Cambridge, Cambridge, UK). Images were analysed using the SIS Megaview II Image Capture system. TOPFLASH dual-luciferase reporter assay of activity of transfected CTPR constructs Wnt pathway activity was induced by treating cells with conditioned media obtained from L-cells expressing Wnt3A (ATCC CRL-2647) for 8 days, according to ATCC guidelines. For the TOPFLASH assay, HEK293T cells in 24-well plates were transfected with 100 ng of TCF7L2-Firey plasmid, 10 ng of CMV-Renilla plasmid and 100 ng of pcDNA3.1(À) vector encoding HiBiT-tagged nTBP-CTPR2n, nTBP-CTPR2n-foldon or the control CTPR constructs using 0.5 mL lipofectamine 2000 transfection reagent according to the manufacturer's protocol (Thermo Fisher Scientic). Transfected cells were allowed to recover in cell growth media for 8 hours, followed by treatment with Wnt-conditioned media (derived from L-cells expressing Wnt3A; ATCC CRL-2647) at a 1 : 1 ratio for a further 16 hours. The TOPFLASH assay was performed using the Dual-Luciferase Reporter Assay System (Promega), as previously described. 82 Relative luciferase values were obtained from three independent experiments by dividing the Firey luciferase values by the Renilla luciferase values. The luminescence intensity of each HiBiT-tagged construct was measured by mixing 20 mL of the same cell lysate with the Nano-Glo HiBiT Lytic Detection System according to the manufacturer's instructions (Promega).

Measurements of cellular levels of HiBiT-tagged CTPR proteins
HEK293T cells in 96-well plates were transfected with 30 ng of pcDNA3.1(À) vector encoding HiBiT-tagged nTBP-CTPR2n, nTBP-CTPR2n-foldon or the control CTPR constructs using Lipofectamine 2000 transfection reagent according to the manufacturer's protocol (Thermo Fisher Scientic). Following 19 hours incubation, cells were treated with 10 mM MG132 (Calbiochem) for 5 hours. The number of viable cells was measured using the CellTiter-Fluor Cell Viability Assay (Promega) according to the manufacturer's instructions. HiBiT-tagged constructs were quantied using the Nano-Glo HiBiT Lytic Detection System (Promega) as above. HiBiT values obtained from three independent replicates were normalised using the corresponding cell viability measurements.

Intracellular delivery of liposome-encapsulated 3TBP-CTPR6 protein
To directly visualise intracellular protein delivery using confocal microscopy, 3TBP-CTPR6 was labelled with rhodamine B isothiocyanate (RITC) (Sigma). In a typical procedure, 50 mL of RITC (1 mg mL À1 in DMSO) was slowly added to a 1 mL solution of 3TBP-CTPR6 (2 mg mL À1 ) in 0.1 M sodium carbonate buffer pH 9.0, with stirring. The solution was stirred at 4 C for 8 hours and ammonium chloride (Sigma) was added to a nal concentration of 50 mM and stirring was continued for a further 2 hours. RITC labelled 3TBP-CTPR6 (3TBP-CTPR6-RITC) was isolated from unreacted RITC using a PD10 desalting column (Cytiva). Liposomal formulation of 3TBP-CTPR6-RITC (FL-3TBP-CTPR6-RITC) was performed as described above. 1.4 Â 10 5 HEK293T cells were seeded overnight in 700 mL of cell growth media in a m-Slide (Ibidi). The following day, medium was replaced with 700 mL cell growth media without FBS and containing 28 mL of FL-3TBP-CTPR6-RITC. Samples were incubated for 15 minutes at 37 C and then washed twice with 1Â PBS. Confocal images were acquired using a Leica TCS SP5 confocal microscope.
Cell viability assay 4 Â 10 4 HEK293T cells were seeded for 24 hours in 100 mL of cell growth media in a 96-well plate. Cells were incubated with 100 mL liposomes in cell growth media without FBS and containing different volumes (1-12 mL) of FL and FL-3TBP-CTPR6 for 15 minutes at 37 C. Aer washing twice with 1Â PBS, 100 mL of CellTiter-Glo Reagent (Promega) was added, and luminescence was measured using a Clariostar microplate reader (BMG LAB-TECH) according to the manufacture's protocol. Untreated cells were used as control. Data were obtained from triplicate samples, and the standard deviation was calculated from two independent experiments.
TOPFLASH assay of cellular activity of liposome-encapsulated 3TBP-CTPR6 protein and small molecule hTNKS inhibitors HEK293T cells were seeded in 24-well plates as described above. Cells were transfected with 100 ng of TCF7L2-Firey plasmid, 10 ng of CMV-Renilla plasmid per well using Lipofectamine 2000 transfection reagent according to the manufacturer's protocol (Thermo Fisher Scientic). Transfected cells were allowed to recover in cell growth media for 8 hours, and treated with Wnt3A-conditioned media, as above. 16 hours post Wnt pathway activation, proteins were delivered into the cells by liposomal treatment. Cells were incubated with cell growth media minus FBS containing 20 mL of liposomes, for 15 minutes at 37 C. Following one wash in 1Â PBS, Wnt3A conditioned media was added and cells were incubated for 6 hours. The TOPFLASH assay was performed, as described above, in triplicate (from two independent experiments).
For the dose-dependence analysis, the volume of liposome added was kept constant (20 mL) and the protein concentration was varied. The FL-3TBP-CTPR6 samples were prepared in the following way: lipid cakes were hydrated with 10 mM HEPES pH 7.4 and 3.125 mM, 6.25 mM, 12.5 mM, and 25 mM 3TBP-CTPR6. 20 mL of these liposomes in 500 mL cell growth media resulted in 3TBP-CTPR6 concentrations of 0.125 mM, 0.25 mM, 0.5 mM and 1 mM respectively. Samples of the liposome-encapsulated unfunctionalised control protein, referred to as FL-CTPR6, were prepared by hydrating lipid cake with 25 mM CTPR6 in 10 mM HEPES pH 7.4.
For testing the small molecule hTNKS inhibitors XAV939, IWR-1 and G007 in the TOPFLASH assay, cells were treated prophylactically with inhibitors mixed with Wnt-conditioned media and incubated for 16 hours, or interventionally with inhibitors added aer overnight treatment with Wntconditioned media and incubated for another 6 hours. All small molecule hTNKS inhibitors were used at a nal concentration of 1 mM in 0.5% DMSO. The TOPFLASH assay was then completed as described above.

Conflicts of interest
Authors APR, LSI and PJER are inventors on patent application PCT/EP2018/068580. Authors APR, LSI and PJER are founders of PolyProx Therapeutics Limited (company number 11664980).