Atomic-Scale View of Protein-PEG Interactions that Redirect the Thermal Unfolding Pathway of PEGylated Human Galectin-3.

A variant of Gal3C containing a single extrinsic cysteine at position 243 and a C-terminal 6 X His tag, Gal3C[T243C], was covalently modified with a single 5.4 kDa monomethoxy-poly(ethylene glycol)-maleimide polymer at position 243 ( Supplementary Fig. 1 ). Native Gal3C contains only one cysteine, C173, buried in the hydrophobic core. Attempts to conjugate native Gal3C with PEG resulted in completely unreacted protein, verified by size exclusion chromatography ( Supplementary Fig. 2 ). We selected C243 for PEGylation because it was predicted to be solvent accessible and located on the opposite surface of the Gal3C sugar-binding pocket where it was anticipated to not alter Gal3C function. To produce reactive PEG, a maleimide group was installed with end-group functionalization of monomethoxy-PEG (mPEG) (see Methods). A maleimide group was chosen as the Michael-acceptor because it is frequently employed for efficient bioconjugation with cysteines [ 40 – 42 ] and used in many PEG-protein conjugates in industrial and clinical applications. Thiol-Michael addition between C243 and the mPEG-maleimide (PEG) produced Gal3C[T243C]-PEG ( Fig. 1A ). Gal3C[T243C]-PEG was produced and purified via two consecutive chromatographic steps to remove excess PEG and unreacted Gal3C[T243C] (see Methods). Baseline separation of PEGylated and unreacted Gal3C[T243C] was achieved in a final size exclusion chromatography step ( Supplementary Fig. 3 ). The resulting monodispersed Gal3C-T243C]-PEG was highly pure, as verified by SDS-PAGE ( Fig. 1A ). MALDI-TOF mass spectroscopy confirmed that PEG was covalently attached, and the expected 1:1 stoichiometric ratio of PEG:protein was observed ( Supplemental Fig. 4 ). The binding affinity of Gal3C[T243C]-PEG was compared with Gal3C[T243C] and Gal3C ( Fig. 1B and Supplemental Figs. 5 – 6 ) via intrinsic tryptophan fluorescence optimized to ensure equilibrium binding conditions [ 43 ] . Binding of lactose or LacNac quenched the fluorescence of W181, located adjacent to the bound sugar in the crystal structure of Gal3C [ 44 ] , permitting direct measurement of ligand K d values. The observed K d for lactose binding to Gal3C[T243C]-PEG was essentially identical to the values measured for Gal3C[T243C] and Gal3C ( Fig 1B and Supplemental Table 1 ). The binding affinity of LacNac for Gal3C[T243C]-PEG was modestly enhanced by 2.4-fold over Gal3C[T243C] ( Fig 1B and Supplemental Table. S1 ). CD spectra of Gal3C[T243C] measured at 30 ºC are consistent with a folded protein containing mostly β-sheet secondary structure and highly similar to CD spectra of Gal3C ( Supplemental Fig. 7 ). Variable temperature CD experiments of Gal3C[T243C] showed a single cooperative unfolding transition with an apparent melting temperature (T m ) of 58.1 ºC ( Fig. 1C , and Supplemental Fig. 8A ), which was indistinguishable from the unfolding behavior of Gal3C ( Supplemental Figs. 9 – 10 and Supplemental Table 2 ) and indicated the T243C amino acid replacement did not alter the thermal stability of the protein. CD spectra of Gal3C[T243C]-PEG measured at 30 ºC were highly similar to unconjugated Gal3C[T243C] ( Supplemental Fig. 7 ). Remarkably, the thermal unfolding of Gal3C[T243C]-PEG was distinct from unconjugated Gal3C[T243C] and showed two separate unfolding transitions with apparent T m values of 56.2 ºC and 84.5 ºC, respectively ( Fig. 1C and Supplemental Table 2 ). The lower T m value of Gal3C[T243C]-PEG was similar to Gal3C[T243C], and the higher T m value was determined from fitting the data using an estimated value for the third plateau. The first T m observed for Gal3C[T243C]-PEG was 56.2 ºC, similar to the measured T m value of Gal3C[T243C]. However, unlike Gal3C[T243C], Gal3C[T243C]-PEG showed a transition to an intermediate state ( Fig. 1C and Supplemental Fig. 8B ) and remained soluble at higher temperatures. The CD spectra for Gal3C[T243C]-PEG measured above the first transition temperature show conjugated Gal3C[T243C]-PEG retained similar β-sheet content at higher temperatures, but the lambda minimum is less than the value observed for Gal3C[T243C] (215 nm versus 225 nm) ( Supplemental Fig. 8B ), indicating formation of an intermediate state at elevated temperatures with subtly different global structure. As a control, we recorded variable temperature CD data with Gal3C[T243C] alone and in the presence of free, non-functionalized PEG. These data were nearly identical to the data measured for unconjugated Gal3C[T243C] ( Supplemental Fig. 11 ), indicating direct interactions between Gal3C[T243C] and covalently attached PEG increased the thermal stability of the conjugated protein and drove formation of the intermediate. Additional control data were recorded with Gal3C[T243C] conjugated to N-methylmaleimide without PEG, Gal3C[T243C]-NMM. Gal3C[T243C]-maleimide showed a thermal unfolding response nearly identical to Gal3C[T243C] with a determined T M of 55.6 ºC ( Supplemental Fig. 12 ), further confirming the covalently attached PEG was response for the altered thermal unfolding profile of Gal3C[T243C]-PEG. To characterize potential non-covalent interactions between Gal3C[T243C] and covalently attached PEG, we recorded 2-dimensional NMR correlation spectra with both conjugated and unconjugated Gal3C[T243C]. Previously reported NMR assignments for Gal3C [ 45 ] and full-length Gal3 [ 46 ] enabled unambiguous assignment of 71% of the backbone amide groups for Gal3C[T243C] and an additional 17% of amide groups with some ambiguities. 2D [ 15 N, 1 H]-heteronuclear single quantum correlation (HSQC) spectra recorded at 30 ºC with both [u- 15 N] Gal3C[T243C] and [u- 15 N] Gal3C[T243C]-PEG were well dispersed, confirming unconjugated and conjugated Gal3C[T243C] were folded ( Fig. 2A ). The 2D HSQC spectra for unconjugated and conjugated Gal3C[T243C] were highly similar globally ( Fig. 2A ), demonstrating PEGylation of Gal3C did not change the overall fold of Gal3C[T243C]. However, comparison of the HSQC spectra for unconjugated and conjugated Gal3C[T243C] revealed several differences for specific amide groups. Significant changes in chemical shifts (>0.05 ppm) and/or changes in line broadening (>25% change in line width) between HSQC spectra of Gal3C[T243C] and Gal3C[T243C]-PEG were observed for about 12 amide groups predominantly from residues numbered 136 to 141 and 193 to 200 ( Fig. 2 , B and D , and Supplemental Table 3 ). These perturbations are local to the site of conjugation but not exclusively adjacent to position 243C, encompassing several loops and the ends of two β-strands near the conjugation site ( Fig. 2C ). As an additional control we recorded 2-dimensional NMR correlation spectra with Gal3C[T243C]-NMM and compared these data to HSQC spectra of Gal3C[T243C]-PEG ( Supplemental Fig. 13 ). Conjugation with NMM induced small chemical shift perturbations and line broadening relative to unconjugated Gal3C[T243C], primarily near position C243. Comparison with the HSQC data of Gal3C[T243C]-PEG showed further chemical shift perturbations and line broadening when both the maleimide linker and PEG were covalently attached in Gal3C[T243C]-PEG, similar to the observations in Fig. 2 . Interestingly, while most amide groups showed single NMR peaks, signals for two residues, E193 and K199, were split in the HSQC spectrum of Gal3C[T243C]-PEG ( Supplemental Fig. 14 and Supplemental Table 3 ), suggesting the existence of multiple local conformations at these positions. This result contrasts with earlier NMR studies of PEGylated plastocyanin [ 47 ] that reported peak doubling for residues adjacent to the site of conjugation attributed to the presence of diastereomers. Because perturbations in our HSQC data could be related to either direct non-covalent interactions between PEG and the Gal3C[T243C] surface or due to an indirect effect of PEGylation on the protein’s local conformation and dynamics, we performed further NMR experiments to more thoroughly investigate the nature of these interactions. To investigate if PEG and Gal3C[T243C] behaved more as two independent domains within the conjugated protein or interacted more closely, we determined the rotational correlation times (τ c ) for Gal3C[T243C] and Gal3C[T243C]-PEG at 30 °C ( Fig. 3A ). τ c values were determined from measurements of 15 N longitudinal (T 1 ) and transverse (T 2 ) relaxation times of backbone amides from the well folded cores of conjugated and unconjugated Gal3C[T243C] ( Fig. 3 , A and B , and Supplemental Fig. 15A ). This approach has been utilized routinely for determination of rotational correlation times of globular proteins up to ~25 kDa [ 48 – 49 ] . Gal3C[T243C] PEGylation increased the rotational correlation time from 10.7 (± 0.7) to 12.0 (± 0.8) ns, consistent with a decreased transverse relaxation time (T 2 ) ( Fig. 3 ) and increased line broadening in 15 N-edited 1D 1 H NMR spectra of Gal3C[T243C]-PEG ( Supplemental Fig. 15A ). The increased rotational correlation time for Gal3C[T243C]-PEG is in line with expectations for a single polypeptide chain of increased molecular weight based on literature data of globular proteins of different molar masses [ 49 ] ( Supplemental Fig. 15B ), indicating the protein and attached polymer do not diffuse independently. To probe whether the noncovalent interactions between Gal3C[T243C] and PEG persist over longer time scales, we recorded hydrogen-to-deuterium exchange (HDX) NMR data with Gal3C[T243C]-PEG at 30 °C ( Fig. 4A ). A [ 15 N, 1 H]-HSQC spectrum was recorded with [u- 15 N] Gal3C[T243C]-PEG, then the same sample was concentrated and diluted with buffer containing 99.9% 2 H 2 O and a second [ 15 N, 1 H]-HSQC spectrum was recorded at the same temperature. The final concentration of 2 H 2 O in the NMR sample was calculated to be 73%. Reduction in amide signals intensities proportional to the change in 2 H 2 O concentration indicated an amide was fully solvent accessible, while a reduction in the NMR signal by less indicated some degree of solvent protection ( Supplemental Fig. 16 ). Regions of Gal3C[T243C]-PEG protected from solvent exchange correlated with Gal3C secondary structure ( Supplemental Fig. 16 ) but not with the region of amino acids perturbed by PEGylation ( Fig. 4A ), showing that non-covalent interactions between Gal3C[T243C] and PEG do not persist long enough to prevent bulk solvent molecules from accessing the protein surface. In line with these observations, 1 H- 1 H NOESY experiments recorded with Gal3C[T243C]-PEG at 30°C failed to show persistent contacts between the protein and polymer (data not shown). Taken together, these results suggest PEG interactions with the protein surface are relatively transient. To probe the impact of PEGylation on Gal3C[T243C] at higher temperatures, we recorded variable temperature [ 15 N- 1 H]-HSQC spectra of Gal3C[T243C] and Gal3C[T243C]-PEG at 40 ºC, 50 ºC and 55 °C, the latter temperature being the highest temperature at which the NMR cryoprobe could be safely operated. For each temperature, we compared chemical shift perturbations and line broadening in NMR data sets of unconjugated and conjugated Gal3C[T243C] to visualize perturbations from PEGylation ( Supplemental Fig. 17 A and B ). For all temperatures, we observed chemical shift perturbations and line broadenings at the same residues as data measured at 30 °C ( Fig. 2D ), indicating effects of PEGylation persisted over the entire temperature range. We also empirically observed that PEGylation offered prolonged stability at 55 ºC. Changes in 15 N- 1 H HSQC signals that occurred between data recorded at 30 °C and 55 °C, for both Gal3C[T243C] and Gal3C[T243C]-PEG were also analyzed. The observed chemical shift perturbations and/or line broadening are globally similar for the protein and the conjugate, but we observed some local differences ( Supplemental Fig. 17C ). Interestingly, a number of these differences are clustered on the site of Gal3C[T243C] where PEGylation induced chemical shift perturbations or line broadening at 30 °C. To probe the effect of temperature on Gal3C[T243C]-PEG above 55 ºC, we heated an NMR sample of Gal3C[T243C]-PEG to 62 °C for 3 minutes before quickly cooling the sample to 30 ºC and acquiring a [ 15 N, 1 H]-HSQC spectrum. This temperature was above the observed T m of Gal3C[T243C] and selected to partially populate the intermediate state while avoiding irreversible unfolding. To prepare this sample, CD spectroscopy was used to monitor the global fold of the protein as a continuous thermal ramp was applied ( Supplemental Fig. 18 ). After reaching 62 ºC, a slight decrease in CD intensity was observed, the sample was cooled down to 30 °C, and we immediately recorded a [ 15 N- 1 H]-HSQC spectrum. This spectrum was overall highly similar to the HSQC spectrum of freshly prepared samples measured at 30 ºC ( Supplemental Fig. 19A ), indicating the global fold was preserved during the heating and cooling procedure. However, we observed local changes in chemical shifts or line widths for the signals of several backbone amides ( Supplemental Fig. 19B and Supplemental Table 3 ). These amino acids were almost exclusively located in loops and the ends of β-strands, and most of the residues perturbed by heating were the same residues displaying chemical shift and/or line width changes upon PEGylation ( Fig. 4 , B and C and Supplemental Table 3 ). This suggests non-covalent PEG interactions with the Gal3C[T243C] surface induced local changes in protein structure and dynamics related to the formation of the unfolding intermediate and increase in thermal stability. We next investigated potential determinants driving localization of PEG to the region of the surface of Gal3C[T243C] near the site of conjugation ( Fig. 2C ), and the potential relationship between PEG localization and redirection of the thermal unfolding pathway of the PEGylated protein. In earlier computational studies of PEGylated proteins [ 27 , 50 – 51 ] , results from molecular dynamics simulations proposed lysines could act as non-covalent ‘anchors’ with PEG on the protein surface through hydrogen bond interactions between the positively charged amine groups on the protein surface and the ether oxygen atoms in PEG. Small PEGs were also observed to form crown ether-like formations around lysines in crystal structures of some proteins [ 52 ] . Three lysines (K139, K196 and K199) formed the perimeter of the region of Gal3C[T243C] for which chemical shift perturbations were observed upon PEGylation ( Fig. 5A ). This observation and the earlier computational studies motivated us to sequentially replace these specific lysines with isoleucines, which would not be able to participate as hydrogen bond donors and could potentially disrupt Gal3C[T243C]-PEG interactions. We generated three Gal3C[T243C] variants: Gal3C[T243C,K139I], Gal3C[T243C,K139I,K196I] and Gal3C[T243C,K139I,K196I,K199I]. Among these, the variants containing one and two lysine replacements could be expressed and purified for NMR and CD experiments, but Gal3C[T243C,K139I,K196I,K199I] was not stable enough to produce a PEGylated sample, likely because K199 appears to participate in a salt bridge with D215 located in an adjacent loop [ 44 ] . We recorded HSQC spectra with [u- 15 N] Gal3C[T243C,K139I], [u- 15 N] Gal3C[T243C,K139I,K196I] and the corresponding stable isotope-labeled PEGylated variant proteins at 30 °C. [ 15 N, 1 H]-HSQC spectra of the variant proteins were well dispersed, showing the overall fold of the variants was highly similar to Gal3C[T243C] ( Supplemental Fig. 20 ). The PEGylated variants were also well folded and largely retained the same global fold as Gal3C[T243C], evidenced by similar [ 15 N, 1 H]-HSQC correlation spectra ( Supplemental Fig. 20 ). Comparing the HSQC spectra of Gal3C[T243C] with spectra of both variant proteins, we observed subtle differences in the changes of chemical shifts and/or line widths due to PEGylation ( Supplemental Fig. 21 ). These differences tended to shift the perturbed region away from K139 and K196 ( Fig. 5A – C ) and were more widely dispersed across the surface of the proteins, indicating that K139 and K196 participate in localizing PEG. However, replacing these residues with isoleucine was not sufficient to completely abolish PEG interactions with Gal3C[T243C]. Variable temperature CD spectra for Gal3C[T243C,K139I], Gal3C[T243C,K139I,K196I] displayed cooperative one step unfolding similar to Gal3C[T243C] ( Fig. 5 D , F , G , I ). Gal3C[T243C,K139I] showed a lower T m value of 53.0 °C, and Gal3C[T243C,K139I,K196I] exhibited a slightly higher T m value of 60.4 °C ( Supplemental Table 2 ). Gal3C[T243C,K139I,K196I,K199I] displayed a significantly altered unfolding profile and a precise T m value could not be determined. For Gal3C[T243C,K139I]-PEG and Gal3C[T243C,K139I,K196I]-PEG ( Fig. 5 E , F , H , I ), lysine-to-isoleucine substitutions did not abolish the formation of the unfolding intermediate nor thermal stabilization. However, the unfolding intermediate of Gal3C[T243C,K139I]-PEG was extended to higher temperatures ( Fig. 5 , E and F ). The thermal unfolding profile of Gal3C[T243C,K139I,K196I]-PEG was more similar to Gal3C[T243C] but showed the presence of multiple plateaus ( Fig. 5 , H and I ). These changes appeared to correlate with changes in the localization of PEG on the protein surface, suggesting modulating the PEG location changes the thermal unfolding profiles.

Mono-PEGylation increased the thermal stability of Gal3C[T243C] and formed a thermal unfolding intermediate with similar, but distinct, secondary structure. Integrating the thermal unfolding data with quantitative NMR data and functional assays enabled us to develop a more precise view of Gal3C interactions with covalently-attached PEG related to increased thermal stabilization. Localization of chemical shift perturbations and line broadening observed by NMR suggested protein–PEG interactions occurred in a specific region of the protein surface. An analogous observation was made in an earlier NMR and x-ray diffraction study with PEGylated plastocyanin [ 47 ] , which found chemical shift perturbations local to the site of conjugation [ 47 ] . The NMR-observed perturbations presented here implied PEG does not exhibit a random coil structure, but rather has collapsed upon the protein surface in a defined region. This conclusion is further supported by the slower rate of rotational diffusion for Gal3C[T243C]-PEG and indicates PEG does not behave as an independent domain ( Fig. 3A ), in contrast with the earlier NMR study of PEGylated plastocyanin, which concluded there were minimal to no interactions between the protein and PEG [ 47 ] . This conclusion was based on comparison of NMR linewidths between the PEGylated and unconjugated proteins and also from the observation that PEG was not observed in crystals of the protein but lacked an explicit comparison of diffusion properties of the two molecules [ 47 ] . Altogether, our results are more in line with literature data of PEGylated human Interferon-α 2a , which reported that PEG formed a “shield” for the protein but did not permanently cover the protein surface [ 24 ] . This is also consistent with our presented HDX data, which showed that on longer (hours) timescales, PEGylation did not prevent solvent accessibility. Earlier computational studies of PEGylated proteins proposed lysines could play an important role in orienting PEG on a protein through hydrogen bonding [ 27 , 50 – 51 ] . In the present study, we replaced two lysines that bordered a region of the protein surface affected by PEGylation with isoleucines. For lysine-to-isoleucine variants, we observed PEG-dependent chemical shift perturbations, indicating PEG interacted with the surface of these Gal3C variants, though at subtly different positions than Gal3C[T243C] ( Figure 5 ). Intriguingly, though changes in chemical shifts were subtle between the PEGylated lysine-to-isoleucine variants, thermal unfolding data of the PEGylated variants were distinct from each other and from Gal3C[T243C]-PEG ( Figure 5 ). These data suggest protein–PEG interactions are likely influenced by multiple amino acid types rather than only one specific type. Future experiments probing different conjugation sites and the impact of additional amino acids are likely to be highly informative. While mutations resulted in measurable changes to protein thermal unfolding properties, PEGylation increased the thermal stability for all variants compared with the unconjugated proteins. This observation suggests the benefits of PEGylation may be broadly observed across different variants of the same protein, potentially enabling a wider range of protein engineering applications.

Biologics are highly effective, targeted treatments for a range of diseases, constituting approximately 30% of drugs approved by the FDA over the past 5 years [ 1 – 2 ] . Despite promise of expanding on this success, biologics are inherently challenging to produce due to their complexity and frailty compared with small molecules. A promising approach to overcoming these challenges is covalent attachment of a polymer such as poly(ethylene glycol) (PEG) to reactive groups in protein side chains. PEGylation is one of the most widely used approaches in a vast array of industrial and pharmaceutical applications [ 3 – 7 ] . Since approval of the first PEGylated biologic Adagen in 1990, nearly 30 FDA-approved PEGylated biologics have entered the clinic to treat a wide array of illnesses including numerous cancers [ 3 ] . Ideally PEGylation should preserve or enhance the activity of the biologic while conferring improved robustness. Benefits of PEGylation to the conjugated protein include increased thermal stability, resistance to chemical denaturation, reduced renal clearance and improved therapeutic half-lives [ 3 , 8 – 10 ] , and improved functional activity [ 6 , 11 – 14 ] . However, these benefits are not necessarily obtained with PEGylation at arbitrary positions in the protein sequence [ 15 – 18 ] . PEGylation has also been documented to affect protein properties negatively, alter protein functions in undesired ways, or have no observable impact [ 15 – 18 ] . Currently there exist no clear criteria for predicting the effects of PEGylation on protein properties, and a grand challenge is designing protein–polymer conjugates a priori with predictable chemical properties. Current molecular models mostly based on lower resolution methods proposed two different solution conformations of protein–PEG conjugates that differ in the degree of non-covalent interactions between PEG and the protein. In one model, the protein and covalently attached PEG are thought to exist as a ‘dumbbell’-shaped conjugate with minimal to no protein–PEG interactions [ 19 – 22 ] . The second model describes a ‘shroud’ conformation where PEG forms more extensive interactions with the protein surface [ 23 – 24 ] . Some experimental data and molecular dynamics simulations support a more nuanced view where the conformation of the protein–PEG conjugate depends on the protein size and the length and chemical structure of the conjugated polymer [ 25 – 27 ] . High resolution magic angle spinning solid-state NMR data showed the conformation of PEGylated protein was largely preserved in sedimented solid samples [ 28 – 29 ] . Motivated by this need, we recorded NMR spectroscopic data with aqueous solutions containing PEG conjugated to the carbohydrate recognition domain of human galectin-3 (Gal3), a β-galactoside-binding lectin required for cell adhesion and formation of extracellular matrices [ 30 ] . Galectin-3 has well-documented roles in multiple diseases, including cancer [ 31 – 34 ] , joint failure [ 35 ] and endometriosis [ 36 ] . Gal3C itself is a potential cancer biologic, inhibiting tumor growth [ 37 ] and enhancing anti-cancer compounds [ 38 ] . High resolution NMR data provided an atomic level view into the effects of PEGylation on the structure of PEGylated Gal3C. Using circular dichroism (CD) spectroscopy, we show PEGylation redirects the Gal3C unfolding pathway, forming a stable intermediate proposed for other protein–polymer conjugates but not previously observed [ 16 , 39 ] . Local Gal3C-polymer interactions observed by NMR correlated with redirection of the thermal unfolding pathway, which could be influenced by replacing specific charged residues. Surprisingly, PEGylation of these Gal3C variants still provided improved thermal stability despite altered interactions. Together these data provide a path toward the rational design of protein–polymer conjugates with predicted properties.