Epigallocatechin-3-gallate preferentially induces aggregation of amyloidogenic immunoglobulin light chains

Antibody light chain amyloidosis is a rare disease caused by fibril formation of secreted immunoglobulin light chains (LCs). The huge variety of antibody sequences puts a serious challenge to drug discovery. The green tea polyphenol epigallocatechin-3-gallate (EGCG) is known to interfere with fibril formation in general. Here we present solution- and solid-state NMR studies as well as MD simulations to characterise the interaction of EGCG with LC variable domains. We identified two distinct EGCG binding sites, both of which include a proline as an important recognition element. The binding sites were confirmed by site-directed mutagenesis and solid-state NMR analysis. The EGCG-induced protein complexes are unstructured. We propose a general mechanistic model for EGCG binding to a conserved site in LCs. We find that EGCG reacts selectively with amyloidogenic mutants. This makes this compound a promising lead structure, that can handle the immense sequence variability of antibody LCs.

Epigallocatechin-3-gallate (EGCG) has already been shown to interact with other amyloid proteins like α -synuclein 30 , amyloid-β 31,32 , huntingtin 33 , IAPP 34,35 , transthyretin 36,37 tau 38 and SEVI 39 . For these proteins, detailed mechanistic insights are available, highlighting the role of oxidation of EGCG 40 , redirection to non-toxic species 41 or remodeling of fibrils 42 . EGCG has already been studied in the context of AL amyloidosis. It redirects the LCs into partially SDS-stable aggregates 43 and also causes a considerable delay of aggregation kinetics by interactions with the native state 44 . These in vitro studies support preliminary in vivo reports on the efficacy of EGCG against AL amyloidosis in C. elegans 45 and in human patients [46][47][48] . Two clinical trials in phase II are ongoing (EpiCardiAL, ClinicalTrials.gov Identifier: NCT01511263; TAME-AL, ClinicalTrials.gov Identifier: NCT02015312).
In this study, we report on the mechanism of EGCG interacting with the LC V L domain of MAK33. The murine MAK33 antibody targets muscle-type creatinine kinase and its V L domain is used here as a model system to investigate amyloid aggregation. The κ -type MAK33 V L domain has been studied in detail regarding structure 49 and biophysical properties 19,20 . Several point mutants have been shown to form amyloid fibrils. The strongly destabilising I2E mutant can form fibrils at native conditions 25 . Both the S20N and the D70N substitutions are sufficient to enable fibril formation under destabilising conditions, e.g. sonication or addition of SDS 26 . The latter two mutants are especially interesting, since they display the same thermodynamic stability as the WT protein.
Binding of EGCG was investigated using solution-state nuclear magnetic resonance (NMR) spectroscopy, site-directed mutagenesis and docking studies. Comparison of aggregation kinetics with different mutants indicates increased affinity to LCs with higher amyloidogenicity. Solid-state NMR as well as infrared spectroscopy show that EGCG-V L coprecipitates are mostly unstructured, with only the binding site being structurally well defined.

Results
EGCG binds specifically to proline 59. To derive a general binding principle for EGCG to V L domains, we characterised its interaction with different MAK33 V L protein sequences. We used the S20N mutant for initial experiments to study the interaction with EGCG (Fig. 1a), because its fibril formation can be easily controlled by experimental settings. Addition of EGCG to the S20N variant resulted in moderate, but reproducible chemical shift perturbations (Fig. 1b,d; see SI 1, 2 for further details), with strongest effects on K39, L46, S54, I58 and R61. These residues are located close to each other in the structure. While EGCG is expected to cause strong chemical shift changes due to its aromatic ring systems, the small chemical shift perturbations observed here are in  15 50 . In addition to chemical shift changes, titration of EGCG also caused visible precipitation.
EGCG is known to bind strongly to proline-rich peptides 51 . Since proline lacks the amide hydrogen atom, 1 H, 15 N-HSQC NMR experiments fail to detect this residue. However, mapping of chemical shift perturbations on the structure shows most of the affected residues being close to P59. Therefore, P59 is a likely candidate for interaction with EGCG in the V L domain (Fig. 1b,d). To validate this interaction, we mutated this proline to alanine. P59 is located in a loop area. Thus, the mutation is unlikely to change the structure of the domain. Comparison of NMR chemical shifts revealed only minor differences between the WT, the amyloidogenic S20N and the P59A mutant, confirming this assumption (data not shown). The mutation was, however, sufficient to suppress most chemical shift perturbations induced by EGCG (Fig. 1c,e), confirming the important role of P59 for EGCG binding. At the same time, we find that the P59A mutant is still precipitated by EGCG (SI 3).
EGCG can bind at two distinct sites. To further characterise the binding site, we performed docking studies, both with and without guidance from NMR chemical shift perturbations (Fig. 2). Based on the NMR experiments, we first ran docking experiments with predefined contacts to I58 and P59. This NMR-guided docking suggested a binding site for EGCG involving residues S52, S54, S60, S63 and G64 (Fig. 2a). However, the subsequent MD simulation showed that EGCG is not tightly bound at that position. It fluctuated around the starting structure and eventually dissociated within the 500 ns simulation. This indicates low affinity binding. In a complementary simulation, we docked EGCG to MAK33 V L without experimental restraints. In this case, EGCG bound to another site, with strong interactions with Q38, P44, F87 and F98 (Fig. 2b). During the subsequent 500 ns MD trajectory, EGCG was more stably bound at that site, indicating a higher binding affinity compared to binding around residues I58 and P59. The stable binding in this position originates from π -π stacking interactions between ring A of EGCG and F98, ring B and F87, a hydrogen bond between one of the hydroxyl groups of ring B and Q38, which is stable during the MD simulation, and electrostatic interactions between R45 and one of the hydroxyl groups of ring D of EGCG. In the binding mode obtained from NMR-guided docking, the interactions between EGCG and MAK33 V L are generally less pronounced and especially the π -π stacking interactions are missing. Moreover, the stable binding predicted by unguided docking occurs in a similar position as reported for 6aJL2 R24G 44 , suggesting a common binding site for different immunoglobulin V L domains. In both the NMR-guided and unguided simulations, we observed a decrease in protein flexibility in the region around residues 92-94 as a result of EGCG binding (Fig. 2c,d). In the unguided simulation, there is also an increase of flexibility around residue P59, which is not observed in the guided simulation. This is an interesting observation, as in the unguided simulation EGCG does not bind directly to P59, yet its dynamics is affected by EGCG binding in a neighbouring region on the V L surface. This change in P59 dynamics may give rise to the chemical shift perturbations in that area as observed for MAK33 V L S20N in the presence of EGCG (Fig. 1d).
Interaction with proline 44 leads to precipitation. Using solution-state NMR, we were not able to probe the EGCG binding experimentally, since the residues located in the predicted binding site are only poorly visible or cannot be detected at all. This is caused by dimerisation-related dynamics. Homodimerisation is frequently observed for V L domains 52 . This typically weak interaction results in exchange broadening of the NMR resonances for residues involved in the dimer interface 21 . These exchange dynamics can also be observed for MAK33 V L . The EGCG binding site observed by unguided docking is located in this dimer interface.
In order to confirm this binding site experimentally, we collected the EGCG-protein coprecipitate (Fig. 3a) and conducted solid-state NMR experiments. We performed a 13 C, 13 C correlation experiment, which relies on dipolar couplings. Since anisotropic interactions are averaged to zero in dynamic regions, this experiment displays only the rigid parts of the sample. The spectral quality reflects the homogeneity and rigidity of the sample. We determined a 13 C linewidth on the order of 250 Hz to 300 Hz, which indicates a high degree of structural heterogeneity in the precipitate. By contrast, structured amyloid aggregates display 13 C linewidths of approximately 100 Hz 53 . Still, we obtained a spectrum, in which we could assign the spin systems of nine amino acids (Fig. 3c). The observed amino acids include arginine (only side chain), glutamate, isoleucine, leucine, lysine, proline, serine, threonine and tyrosine (only side chain). Since only nine out of 108 amino acids were visible, we assume that the protein is unfolded in the precipitate, which is in agreement with the broad resonance lines mentioned above. The visible resonances originate from residues, which are presumably rigidified due to interactions with EGCG. Their assignments provide important clues to identify the EGCG binding site in the precipitates. To enable use of native-like conditions for fibril formation, the highly amyloidogenic I2E mutant was employed. Fibrils were grown by incubating a 50 μ M protein solution at a temperature of 37 °C and agitating the sample at 120 rpm (left). The sample with EGCG-induced precipitates (right) was treated identically, but in the presence of a 10-fold excess of EGCG. (b) MAS solid-state NMR analysis of coprecipitates of EGCG with the WT or the P44A mutant. The P44A precipitates yield a drastically reduced signal intensity in 1 H, 13 C-cross polarization experiments. (c) 2D 13 C, 13 C PDSD correlation spectrum of coprecipitates of MAK33 V L WT and EGCG (black). Nine spin systems could be identified, which presumably originate from residues close to the EGCG binding site. In the lower right part of the symmetric spectrum, the proline spin system is marked with green lines. The NMR spectrum of the P44A variant prepared in the same way (red) does not yield a comparable cross peak pattern. (d) Assigned spin systems of EGCG-induced WT-V L aggregates. The identified spin systems were assigned to those residues, which are closest to the EGCG binding site in the unguided docking structure ( Fig. 2b; details in SI 4). Unambiguous assignments are shown in green, ambiguous assignments in yellow.
Scientific RepoRts | 7:41515 | DOI: 10.1038/srep41515 Due to resonance broadening, sequential assignment was not feasible for EGCG-induced MAK33 V L WT aggregates. However, the observed spin systems allow already to restrict the potential binding site by making use of the primary sequence of the protein and the docking experiments. The proline Cβ and Cγ resonances have a chemical shift difference of 5 ppm, which is a clear indication for a trans conformation 54 . Out of the five prolines in MAK33 V L , two prolines are in cis conformation in the native state (P8 and P95) 19,49 . We can assume that binding of EGCG does not alter the proline conformational state, as the trans conformation is fixed by the interaction with EGCG. This leaves three prolines as potential binding sites: P15, P44 and P59. Mapping of the identified amino acids on the sequence and the unguided docking structure suggest P44 as the preferred binding site (Fig. 3d). Except for threonine, all identified amino acids are close to P44 in the structure (SI 4). Isoleucine and threonine have very weak intensities in the spectrum, suggesting that they are further away from the binding site (which is the case for I48), or located in other flexible regions.
As a control, the solid-state NMR experiment was repeated with the coprecipitate of EGCG and the P44A mutant. Even though sample preparation was identical, the signal intensity in 13 C cross polarization 1D experiments was lower compared to the experiment carried out with the WT protein (Fig. 3b). The difference in intensities, corrected for slightly different amounts of sample in the initial setup for precipitation, corresponds to a factor of ca. 15, indicating that the P44A mutant is precipitated with considerably lower efficiency. The 13 C, 13 C correlation experiment that was acquired for these precipitates hardly shows any cross peaks even after three days of measurement time (Fig. 3c). These observations, together with the docking experiments, suggest a crucial role of P44 for binding EGCG.

EGCG interacts stronger with amyloidogenic mutants.
To gain insights into the mode of action of EGCG, we quantified the kinetics of EGCG-induced precipitation using solution-state NMR. First, we monitored precipitation at different concentrations of EGCG. For this experiment, we used the S20N mutant, which has the same thermodynamic stability as the WT, but displays increased fibrillogenicity 26 . In the control experiment, we added water to the sample. In this case, we saw only a fast initial decrease of protein signal intensity induced by mixing of the sample and subsequent unfolding of a fraction of the protein. This initial decrease was similar in all further experiments.
Over the course of more than two days, we saw hardly any decrease in intensity in the control experiment, whereas increasing amounts of EGCG resulted in higher degrees of precipitation (see Fig. 4a). A 5-fold molar excess of EGCG caused reduction of the protein signal by 26% over two days, while 20-fold excess reduced the signal by 65%, indicating a concentration dependent mode of action of EGCG.
Next, we investigated the influence of the two prolines P44 and P59 on the interaction with EGCG (Fig. 4b). As expected from our previous experiments, the P44A mutant precipitated slower than the WT in the presence of EGCG, although the effect was not as pronounced as in the solid-state NMR experiments reported above. The P59A mutant precipitated slightly faster than the WT. This might be a consequence of different dynamics due to the mutation. Neither of these mutations had strong effects on protein stability, as determined by temperature-induced unfolding transitions monitored by CD spectroscopy. The T m values for the WT protein, P44A and P59A were determined to be 50.7 °C, 51.3 °C and 50.1 °C, respectively (Fig. 4d). Of note, the thermal transitions are irreversible. Thus, the T m values are apparent values. The slight increase of T m for P44A cannot account for the drastic differences observed regarding the precipitation behaviour in presence of EGCG.
In the following, we compared V L mutants with different propensities to form amyloid fibrils 25,26 with respect to their interactions with EGCG. The S20N and the D70N mutants both have the same melting temperature and thermodynamic stability as WT V L . However, while WT V L forms fibrils only at pH 2, these mutants form amyloid aggregates at native pH, upon ultrasonication 26 . The I2E mutant has a drastically reduced thermodynamic stability and forms fibrils at native-like conditions without addition of destabilising chemicals 25 . We conducted precipitation experiments with all variants, adding a 10-fold excess of EGCG (Fig. 4c). The S20N mutant displayed the same precipitation kinetics as the WT, while the D70N and I2E proteins precipitated considerably faster. Solid-state NMR experiments revealed that these precipitates are mostly unstructured. As I2E is far less stable, it might be expected that this variant forms unstructured aggregates with EGCG faster than the WT. The D70N mutant, in contrast, has the same stability as the WT, but is more likely to undergo a transition to fibrils. This indicates, that EGCG not just precipitates all antibody LCs, but has a higher affinity for the more amyloidogenic species. The S20N mutant's slow precipitation might be related to different pathways for amyloid fibril formation. Since the S20N mutant differs both from the WT and the D70N mutant with respect to the m-value for GdmCl denaturation 26 , it might expose different sites during fibril formation and thus react only moderately with EGCG.
To probe amyloid fibril formation and ThT binding 55 , we employed the most amyloidogenic variant. In the absence of EGCG, the I2E variant started immediatelly to form fibrils and reached the plateau phase after 10 days (Fig. 4e). At a 1:1 molar ratio, EGCG already considerably reduced the amount of fibrils. A 5-fold excess of EGCG was enough to quantitatively prevent fibril formation. We are aware of the fact that EGCG changes the fluorescence spectrum of ThT. Therefore, the results were analysed with caution. In agreement with the ThT results, we found no fibrils in TEM experiments (Fig. 3a). In addition, Fourier-transform infrared (FTIR) spectroscopy confirmed the non-fibrillar structure of the EGCG-protein coprecipitates (SI 5). Upon addition of EGCG, the kinetics of fibril formation remained unperturbed, whereas the amount of formed fibrils is significantly reduced. A comparison between the ThT kinetics and the precipitation kinetics (Fig. 4c) suggests, that EGCG-induced aggregates form much faster than amyloid fibrils.
Finally we quantified stochiometries of the precipitate-EGCG complexes. Pelaez-Aguilar et al. have reported a 1:1 stochiometry for binding of EGCG to 6aJL2 R24G in solution, determined by isothermal titration calorimetry 44 . However, the stochiometry of the coaggregates was not quantified yet. We compared the reduction of the 1 H NMR signal intensities of both MAK33 V L S20N and EGCG over time (Fig. 4f). The stochiometries vary both with time and with respect to the initial stochiometry in solution. We find that 10 to 20 molecules of EGCG co-aggregate per precipitated protein molecule. While the initial binding in solution might be 1:1, more EGCG is found in the coaggregates. As the aggregates were shown to be unstructured, more EGCG seems to be recruited during later stages of precipitation. This might indicate, that precipitation is caused be interactions between EGCG molecules.

Discussion
We have shown how the polyphenol EGCG interacts with the murine κ -type MAK33 V L antibody domain. EGCG was already shown to bind to 6aJL2 R24G, which is a highly amyloidogenic λ -type V L sequence 44 . Andrich et al. reported anti-amyloidogenic activity of EGCG for another LC sequence, 1AL1 43 . This suggests, that EGCG interacts with both κ -and λ -type AL sequences.
Many residues important for the interaction with EGCG are conserved between MAK33, 6aJL2 and 1AL1 (Fig. 5a). The highest conservation is obtained for P44 and the aromatic residues Y36, F87 (corresponding to tyrosine in 6aJL2 or 1AL1) and F98 (residue numbering as in MAK33). This is valid even if the comparison is expanded to a very large set of human V L domains (Fig. 5b). In addition to the highly conserved residues Y36, P44, F87 and F98, a basic residue is found at position 45 (arginine in case of MAK33). Positions 42 and 43 are more variable. Considering that also 6aJL2 and 1AL1 differ from MAK33 in these positions, they seem to be less crucial for EGCG binding. Despite minor differences in the primary sequences, the EGCG binding site reported for 6aJL2 matches well with the P44 site described here 44 . Even though we used EGCG in excess, the match between binding sites for MAK33 and 6aJL2 indicates specific binding. For 6aJL2, an affinity of K A = 1.35 × 10 4 M −1 has been reported 44 . Considering the conservation of the binding site, the vast majority of V L domains should be able to bind EGCG in a way that induces unstructured precipitation.
It has been found recently that EGCG did not inhibit fibril formation according to ThT and TEM using the patient-derived λ -V L sequence Mcg 29 . We would like to mention that similar controversies were also reported for Aβ and EGCG 56 . We speculate that this is related to a potentially lower amyloidogenicity of Mcg. Our results showed stronger effects of EGCG on more amyloidogenic mutants. For 6aJL2 R24G, the effect of EGCG on the ThT kinetics was reported already at lower stochiometries than observed here. This matches our expectations, as 6aJL2 R24G is known to form amyloid fibrils quite fast and thus might be more sensitive to EGCG.
Mutants with a higher amyloidogenicity induce faster precipitation. We were therefore wondering, whether it is possible to derive general principles to explain this difference. Hence, we suggest the following mechanism (Fig. 6). We assume that EGCG binds initially to a native V L domain. EGCG can interact either with P44 or P59. The mutagenesis experiments suggest that interaction with the P59 site is not important for precipitation. By contrast, a more stably bound EGCG at P44 could recruit further EGCG molecules, considering its intrinsic tendency to form oligomers 57 . EGCG is mostly monomeric at the concentrations used here. However, also under these conditions, 1-2% of EGCG is in a dimeric state 58 . Similar models have been suggested earlier for binding of EGCG to β -casein 59 . Subsequently, two or more V L domains, which are already bound to EGCG, could form a complex, with the self-associating EGCG molecules acting as a molecular glue. This complex should be still soluble, as both the native V L domain and EGCG are rather hydrophilic. To explain the observed precipitation of the complex, V L proteins must expose hydrophobic patches upon binding to EGCG. We suggest, this could be triggered by partial unfolding of already recruited V L domains. Our results indicate that EGCG reacts stronger with more amyloidogenic mutants, even if they show similar thermodynamic stability as the wild-type. This Frequency plot of residues in the binding site for human V L sequences. Most of the residues of the P44 binding site are conserved in human V L sequences. The frequency plot was created using amino acid frequencies from more than 11,000 human LC sequences (taken from abYsis 2.3.3; http://www.bioinf.org.uk/abysis/).

Figure 6. Mechanism of EGCG-induced V L precipitation.
Native monomers can be converted via partially unfolded intermediates into amyloid fibrils. EGCG can bind either to P59 (low affinity, no precipitation) or to P44 (higher affinity). More EGCG molecules are recruited to the P44 binding site, which in turn results in association of further proteins. Locally increased protein concentration and partial unfolding, as it presumably occurs prior to amyloid formation, are favourable conditions for precipitation.
Scientific RepoRts | 7:41515 | DOI: 10.1038/srep41515 could be explained by lowly populated partially folded intermediates on pathway to fibril formation, which are more often sampled by more amyloidogenic species [60][61][62] . A recent study on β 2-microglobulin has investigated in detail such point mutants, which have reduced kinetic barriers for fibril formation, while at the same time the thermodynamic stability of the native state was not affected 63 . In AL amyloidosis, the C-C' loop around P44 is presumably involved in initial steps of amyloid formation 22,24 . In such an early intermediate, the P44 binding site might be recognised by EGCG. Similarly, a higher affinity of EGCG for unfolded states has been reported earlier for BSA 41 . If several V L proteins are already recruited to a cluster of EGCG molecules, the likelihood that multiple monomers enter such an intermediate state at the same time increases. Exposure of hydrophobic patches then could initiate aggregation and precipitation. The final result is a particle with EGCG in its center and a number of aggregated proteins covering it.
Drug discovery in AL amyloidosis is hampered by specificity. Potential drug candidates should prevent amyloid formation, but are not supposed to interfere with the necessary activity of antibodies. EGCG, despite being infamous as a promiscuous ligand, fulfills these requirements: It precipitates preferentially the amyloidogenic V L domains. A similar selectivity of EGCG for amyloidogenic mutants was recently reported for TGFBIp 64 . At the same time, the EGCG binding site in functional antibodies is hidden, as the binding site is protected by the heavy chain. In LC dimers, which are known to be incompatible with amyloid formation 65,66 , the EGCG binding site is also not accessible.
We consider EGCG a valuable lead structure for further drug development, also considering the plethora of literature reports for its activity against different amyloidogenic proteins. Much progress has been made recently with issues concerning the in vivo stability of EGCG 67-69 : Different crystal forms or nanoparticles as transporters help to increase bioavailability. Even without special formulations, a one time oral application of 500 mg EGCG yields a plasma concentration of ca. 2 μ M of free EGCG 70 . The median value for serum concentration of free λ -LCs in a cohort of AL-patients was ca. 160 mg/l 71 . Assuming an average V L MW of 12 kDa, this corresponds to 13 μ M, hence EGCG could already be bio-available at comparable concentrations. The question of in vivo activity is currently adressed by the clinical trials EpiCardiAL and TAME-AL (EpiCardiAL, ClinicalTrials.gov Identifier: NCT01511263; TAME-AL, ClinicalTrials.gov Identifier: NCT02015312) and it remains to be seen, how effective EGCG is as a therapeutic agent.

Material and Methods
If not specified otherwise, all chemicals were purchased from Sigma Aldrich (Taufkirchen, Germany). EGCG (Sunphenon EGCG) was a kind gift by Stefan Schönland, supplied by Taiyo (Yokkaichi, Japan). All EGCG solutions were prepared freshly from EGCG powder. Aqueous stock solutions of EGCG were set to 20 mM and sonified to ensure solubilisation.
Protein expression and purification. MAK33 V L WT was expressed and purified as described previously 25 . Briefly, E. coli BL21 cells were grown in M9 minimal medium to OD 600 = 0.6-0.8. Expression was then induced with 1 mM IPTG, using a pET28b vector. After expression over night at 37 °C, cells were harvested and inclusion bodies were separated from the soluble fraction. The inclusion bodies were dissolved in 25 mM Tris, 5 mM EDTA, 8 M urea and 2 mM β -mercaptoethanol and purified as the flow-through fraction by ion exchange chromatography with a 20 ml Q Sepharose column (GE Healthcare, Munich, Germany). Refolding was achieved by dialysis over night into 250 mM Tris, 100 mM L-arginine, 1 mM oxidized glutathione and 0.5 mM reduced glutathione (pH 8.0, 4 °C). The protein was purified by gel filtration using a 120 ml Superdex 75 column (GE Healthcare, Munich, Germany). As elution buffer, a solution of 20 mM sodium phosphate, 50 mM NaCl at pH 6.5 has been employed. For MAK33 V L WT, the final yield after purification from 1 L M9 was on the order of 15-20 mg.
Mutagenesis protocols were adapted from the QuikChange site-directed mutagenesis kit (Agilent, Santa Clara, USA) using Pfu Ultra DNA polymerase (Agilent, Santa Clara, USA). After PCR, plasmids were digested with DpnI and then transformed into E. coli XL1.
Solution-state NMR spectroscopy. Backbone assignments for MAK33 V L WT have been reported previously 25 . Heteronuclear single quantum coherence (HSQC) experiments were recorded using a protein concentration of 50 μ M at a temperature of 25 °C. All NMR experiments were performed using a Bruker Avance III spectrometer (Bruker BioSpin, Karlsruhe) operating at a 1 H Larmor frequency of 500 MHz. The spectrometer was equipped with a triple resonance cryogenic probe. Experiments were recorded and processed using Topspin 3.2 (Bruker BioSpin, Karlsruhe) and analysed with CcpNMR Analysis 2 72  Ti swinging bucket rotor and a rotor filling device (Giotto Biotech, Florence, Italy) were used to pack the protein aggregate into an MAS rotor. The rotation frequency of the centrifuge was set to 28,000 rpm. Samples were packed into 3.2 mm thin wall ZrO 2 rotors with vespel caps (CortecNet, Voisins Le Bretonneux, France) employing house-made teflon spacers. Cross-polarization (CP) and proton driven spin diffusion (PDSD) experiments were recorded using a 500 MHz Bruker Avance II spectrometer (Bruker BioSpin, Karlsruhe) or a 750 MHz Bruker Avance III spectrometer (Bruker BioSpin, Karlsruhe). Both spectrometers are equipped with triple-resonance MAS probes. The magic angle spinning frequency was set to 11 kHz and 10 kHz, respectively. The set temperature of the probe was adjusted to 0 °C. For 13 C, 13 C mixing, a PDSD mixing time of 50 ms was employed.
Melting curves. Melting temperatures were determined using circular dichroism (CD) spectroscopy.
Measurements were performed using a J715 spectropolarimeter (Jasco, Gross-Umstadt, Germany). Melting curves were recorded at a wavelength of 212 nm at a scan rate of 20 °C/h at 10 μ M protein concentration. Curves were fitted with Matlab version R2014b (TheMathWorks, Natick, USA) using the equation To test the stability of the obtained complexes and for allowing MAK33 V L to adjust to the EGCG bound to it, MD simulations were performed using Gromacs 4.6.4 76 . The systems were modeled with the Amber99sb-ildn force field 77 and the Tip4p-Ew water model 78 . EGCG was parametrised using acpype 79 , with bonded and Lennard-Jones parameters taken from the general Amber force field (GAFF) 80 and RESP charges calculated with antechamber 81 from a HF/6-31 G* single point calculation after the geometry was optimised at the B3LYP/6-31 G* level. Before starting the MD simulations, we performed steepest descent energy minimisations. The systems were then equilibrated with a 1 ns NVT simulation followed by a 1 ns NPT simulation. Afterwards we performed 500 ns production runs. In all MD simulations a cutoff of 1.0 nm was used for short-range van der Waals and electrostatic interactions, while long-range electrostatic interactions were calculated using the particle mesh Ewald method 82 . Periodic boundary conditions and a 2 fs timestep were used. The temperature was kept constant at 310 K with the Nosé-Hoover thermostat 83 and the pressure was kept constant at 1.0 bar using the Parrinello-Rahman barostat 84 . For the analysis, we discarded the first 20 ns of each simulation, considering it as equilibration. The trajectories were analysed using Gromacs tools 76 and MDAnalysis 0.8.1 85 . Contacts between EGCG and the protein were considered when the distance between any atom of the two molecules was less than 0.3 nm.

Sequence analysis.
Protein sequence alignments were calculated in Jalview 2.8.2 86 using Clustal W 87 . The likelihood for a particular amino acid to occur at a specific positions was taken from the database abYsis 2.3.3. (http://www.bioinf.org.uk/abysis/). Amino acid frequencies were calculated for human LC sequences, where the number of available sequences varied between 11,000 and 15,000, depending on residue. Amino acid frequency plots were created with WebLogo 2.8.2, using the frequency plot option 88 .