NMR detects clustering and ultra-weak excipient interactions governing monoclonal antibody viscosity in formulation-relevant conditions Authors: Liza Boëffard-Dosierre, Camille Dagallier, Géraldine Eudier, Vasco Filipe, Atifa Badar-Majeed, Solène Fraumont, Sophie Zinn-Justin, Francois-Xavier Theillet Published: 2026 ABSTRACT ABSTRACTHigh-concentration formulations of monoclonal antibodies (mAbs) are required for subcutaneous administration but are frequently challenging to develop due to elevated viscosity and colloidal instability. These properties are governed by mAb–mAb interactions that are regulated by additional mAb–excipient interactions. These ultra-weak interactions remain difficult to characterize using conventional techniques. Here, we combined nuclear magnetic resonance (NMR) spectroscopy with rheometry and dynamic light scattering to investigate mAb interactions under formulation-relevant conditions. Rheological measurements showed that, in the case of the chosen mAb, the excipients arginine and lysine strongly reduce the macroscopic viscosity, whereas other excipients display a moderate effect by themselves. Characterization of the mAb oligomeric state by 1H NMR confirmed that arginine and lysine are the most efficient at reducing mAb self-assembly, while proline and glycine promote clustering. Interactions between mAb and excipients were first analyzed by measuring excipient diffusion coefficients, but these are only weakly affected by the addition of concentrated mAb. The mAb–excipient interactions were further detected by measuring excipient 1H and 13C chemical shifts, mAb–excipient saturation transfers, and excipient 1H transverse relaxation rates. These experiments provided complementary information on excipient interactions with the different mAb oligomers. Lysine was identified as the best mAb binder. However, several excipients such as sucrose that do not reduce the macroscopic viscosity also bind to concentrated mAb, highlighting that excipient binding can have various consequences on the transient interactions between mAb species. Altogether, this work proposes a powerful NMR pipeline to dissect ultra-weak molecular interactions that govern viscosity and developability in therapeutic antibody formulations. Introduction Monoclonal antibodies (mAbs) have revolutionized the treatment of various diseases such as cancer and autoimmune disorders.1 The number of clinical-stage therapeutic mAbs has remarkably increased in the past decade, with 178 molecules reported in late-stage clinical studies in 2025, showing their growing success and broad applications.2 However, the development of formulations for these mAbs is complicated by their high molecular weight and susceptibility to various instabilities, as compared to synthetics.3 This applies particularly to preparations for subcutaneous injections (SCs), a desirable delivery mode for patients and healthcare systems.4 SCs require low volumes, which impose high biopharmaceutical concentrations when high doses are necessary, as is commonly the case for mAbs.3,5–8 Thus, key challenges have emerged, including maintaining the physical and chemical stability of mAbs at high concentrations throughout manufacturing and storage without compromising compatibility with patient delivery.5,9,10 At concentrations of 100 to 200 mg/mL aimed for mL-sized SCs, mAb preparations show viscosities that can reach dozens of centipoise (cP), hence becoming unsuitable for simple handling and injection.7,11 To control the stability and viscosity of highly concentrated biopharmaceuticals, excipients are commonly incorporated into formulations: these additives reduce the population of partially unfolded proteins and/or modify protein–protein interactions (PPIs), thereby preventing transient oligomerization and aggregation, and reducing viscosity.12–16 However, the underlying mechanisms are insufficiently understood and the selection of optimal excipients remains largely empirical, based on excipient screenings analyzed by various biophysical methods.12,15,17 As these screens require large quantities of mAb (~10–100 mg), predictive approaches have also been developed, which aim at prognosticating viscosity at high mAb concentration using observables obtained at low mAb concentration. The prevailing model is that reversible mAb–mAb interactions, responsible for the formation of transient clusters, are dominant contributors to viscosity.5,12,16,18–20 Favorable, viscosity-reducing excipients are thought to screen the interactions between the various oligomeric mAbs species, hence diminishing first order and higher order PPIs. A number of methods can give access to proxies of repulsion or attraction between mAbs at low concentrations (1 to 10 mg/mL), including dynamic and static light scattering (DLS, SLS),21,22 PEG-precipitation23,24 and self-association reporting spectroscopies.25,26 Their predictive values are mitigated by the fact that diluted solutions do not always recapitulate the mAb intermolecular interactions observed when the average mAb–mAb distance decreases, as in concentrated solutions. Only effects due to first-order interactions (monomer–monomer and not monomer–oligomer) are captured, which might not be the main contributors to viscosity in formulation-relevant conditions.14,15 The popular interaction parameter kD or second virial coefficient, extracted from DLS measurements, has thus been challenged in its predictive potency, and other coefficients have been proposed.15,16,27–30 Nuclear magnetic resonance (NMR) spectroscopy is a powerful technique to detect and characterize weak interactions in the millimolar range and above.31–34 This can be achieved for both small and large molecules, and even in complex samples.35 Importantly, NMR can provide atomic-level insights, which can be used to identify the excipient moieties that interact with mAbs. NMR methods for evaluating mAb stability and reversible oligomerization include measurements of water diffusion coefficient and transverse relaxation (R2) rates,36,37 as well as mAb signal intensity and diffusion coefficient.38 In addition, 1D and 2D NMR fingerprinting can be used to assess mAb structure and post-translational modifications39–42: peak intensities reveal mAb reversible oligomerization at high concentrations,43–45 while 2D chemical shift perturbations (CSPs) of Fab or Fc fragments provide a mapping of mAb–excipient interactions, even though this mapping can be ambiguous because mAb–mAb transient interactions can generate similar perturbations.46–48 On the excipient side, interactions with mAbs can be detected by measuring peak intensities and relaxation properties.48–50 Classical ligand screening methods like saturation transfer difference (STD)-NMR and WATER-LOGSY31 can be applied to identify binding of highly concentrated excipients to mAbs. However, reported experiments have been performed at low mAb concentrations in order to minimize the mAb signal.49,50 They have shown that the STD-derived nonspecific binding affinities of a range of excipients with a Bristol Myers Squibb mAb correlate with the colloidal stability measured by SLS via the second virial coefficient.50 Here, we sought to further investigate mAb–excipient interactions over a broad range of mAb concentrations, from 0 to 100 mg/mL, using NMR spectroscopy in conjunction with rheometry and DLS. Because of the lack of information on these processes at high mAb concentrations, we decided to map them in different conditions, from diluted samples at 10 mg/mL to mAb preparations at 100 mg/mL like those used in therapeutic formulations. Thus, we aimed at deciphering the consequences of the mAb–excipient interactions on mAb transient oligomerization (also named here clustering) and macroscopic sample viscosity. We used a mAb from Sanofi, which provided us with the large quantities necessary to carry out such a demanding study. We adapted a panel of NMR techniques, including classical 1D and chemical shift perturbation experiments and prolonged R2 relaxation using similar schemes to those published recently,34 STD-NMR, and diffusion measurements, to our highly concentrated samples. Our NMR analyses identified two excipients that significantly limited mAb self-assembly. They also revealed contrasted mAb–excipient interactions for the whole set of nine excipients at low and high mAb concentrations. Based on these data, we characterized the mechanisms mediating the effect of each of the selected excipients (Suppl. Figure S1) on the mAb sample viscosity. Materials and methods Most of the experiments were performed at 10 °C. This temperature was selected to increase the STD-NMR signals and was later applied to all other experiments to ensure consistent experimental conditions. A supplementary subset of rheometry and NMR experiments was recorded at 25 °C (mAb in the presence of arginine, lysine, NaCl, or sucrose). The ranking remained unchanged, for both viscosity and self-association (see Suppl. Figures S2-3, Suppl. Table S1). Sample preparation The investigational, full-length IgG mAb used in this study was provided by Sanofi. Its molecular weight is about 150 kDa. A centrifugal filter unit (Amicon ® Ultra Filters 15 mL 30 kDa, ref: UFC9030) was used to exchange the mAb from its storage buffer to an aqueous solution containing histidine at 10 mM, pH 6.0. The concentration-dilution protocol was as follows: a 2 mL solution of mAb at 157 mg/mL was diluted in 30 mL of the histidine buffer and centrifuged at 5000 g until the volume was reduced to 6 mL; then 6 mL of the 10 mM histidine or citrate buffer at pH 6.0 were added and concentrated again to 6 mL; this dilution and centrifugation process was repeated five times; the sample was finally concentrated to its original volume of 2 mL. The final mAb concentration was deduced from the UV absorbance at 280 nm measured using a UV spectrophotometer (Nanodrop ONE™, ThermoScientific™). Solutions of excipients (arginine, proline, glycine, lysine, NaCl, sucrose, polysorbate 20 (PS20), polysorbate 80 (PS80) and poloxamer 188 (P188)) were prepared at 6x in 10 mM histidine buffer, pH 6.0. All formulations were sterile filtered with a 0.22 μm cellulose acetate filter (Millipore). Bovine serum albumin (BSA) samples were prepared from lyophilized, crystallized BSA (BSA fraction V, Sigma-Aldrich, ref. 810533). A stock solution at 250 mg/mL was obtained by resuspending the powder in a 10 mM histidine buffer at pH 6.0. The suspension was left overnight to ensure complete dissolution, and the final volume was adjusted afterward. This stock solution was used to prepare all BSA samples for NMR analysis through dilution in the same 10 mM histidine buffer at pH 6.0. Size exclusion chromatography To verify that the NMR-detected mAb self-association is fully reversible and does not correspond to irreversible aggregation, size-exclusion chromatography (SEC) was performed. The NaCl 200 mM condition was selected for this experiment because it corresponds to one of the strongest NMR-detected clustering (NMR linearity table). First, 10 mL of mAb solution at 1 mg/mL in NaCl at 200 mM/histidine at 10 mM, pH 6.0, was prepared, and 1 mL was injected 10 times in 10 successive runs in a Superdex 200 Increase 10/300 column, using an ÄKTA Pure (Cytiva). The eluted mAb monomer samples from all 10 runs were pooled and concentrated to approximately 100 mg/mL using an Amicon Ultra centrifugal filter (30 kDa MWCO). This pooling strategy was necessary to obtain sufficient material to reach 100 mg/mL while remaining above the minimum working volume of the centrifugal unit. The concentrated sample was then diluted back to 1 mg/mL and re-injected into the same column under identical conditions. Viscosity measurements by rheometry Solution viscosity measurements were conducted using an Anton Paar MCR 302 rheometer with a cone-plate geometry. The upper part is a C35/1° cone with a 35 mm diameter and a 1 ° angle. A sample volume of 210 µL was typically used. The measurement temperature was precisely controlled at 10 °C or 25 °C with a Peltier plate. Steady shear measurements were performed by ramping the shear rate from 100 to 10,000 s−1 in both ascending and descending directions to check for reversibility and hysteresis. For all the samples, the pH was adjusted at 6.00 ± 0.05 using diluted hydrochloric acid solutions. The samples were finally filtered through 0.22 μm cellulose acetate filters. For surfactant-free samples, the measurements were limited by rheological artifacts at both ends of the shear rate range. Interface effects: at low shear rates (below 1000 s−1), measurements were compromised by interface effects, previously observed in surfactant-free protein solutions. The mAb was adsorbed at the air/water interface, forming a stiff, elastic film that influenced the measured torque, resulting in an artificially high apparent viscosity.51 Consequently, no reliable viscosity data could be obtained at low shear rates.Sample ejection: at high shear rates, measurements were limited by the physical stability of the sample within the geometry that resulted in the possible ejection of a small volume of the sample (even though our tests show the absence of ejection, see Suppl. Figure S4A). Interface effects: at low shear rates (below 1000 s−1), measurements were compromised by interface effects, previously observed in surfactant-free protein solutions. The mAb was adsorbed at the air/water interface, forming a stiff, elastic film that influenced the measured torque, resulting in an artificially high apparent viscosity.51 Consequently, no reliable viscosity data could be obtained at low shear rates. Interface effects: at low shear rates (below 1000 s−1), measurements were compromised by interface effects, previously observed in surfactant-free protein solutions. The mAb was adsorbed at the air/water interface, forming a stiff, elastic film that influenced the measured torque, resulting in an artificially high apparent viscosity.51 Consequently, no reliable viscosity data could be obtained at low shear rates. Sample ejection: at high shear rates, measurements were limited by the physical stability of the sample within the geometry that resulted in the possible ejection of a small volume of the sample (even though our tests show the absence of ejection, see Suppl. Figure S4A). Sample ejection: at high shear rates, measurements were limited by the physical stability of the sample within the geometry that resulted in the possible ejection of a small volume of the sample (even though our tests show the absence of ejection, see Suppl. Figure S4A). To obtain the least biased measurement of the bulk viscosity, the analysis was restricted to the stable shear regime. The viscosity value obtained at 1020 s−1 was consistently defined as the “macroscopic” viscosity of the sample. At this rate, the Newtonian plateau is reached or close to be reached for all excipient conditions (Suppl. Figure S4B) and consequently is a good compromise that allows a fair comparison between excipients without risking sample ejection or mAb aggregation at higher rates. NMR sample preparation NMR samples consisted of 200 µL solutions in 3 mm-diameter NMR tubes. The respective volumes of the mAb, BSA, and various excipients were adjusted to achieve the desired final concentration. A 10 µL aliquot of 99.97% deuterium oxide (D2O, Eurisotop) was added to obtain a final concentration of 5% (vol/vol). Additionally, 0.2 µL of 100 mM 4,4-dimethyl-4-silapentane-1-sulfonic acid (DSS, Sigma-Aldrich) was added as a chemical shift reference, resulting in a final DSS concentration of 0.1 mM in the NMR tube. The pH of all samples was adjusted when needed to 6.00 ± 0.05 by adding small volumes of diluted HCl using a calibrated pH-meter and a microprobe (Orion™ PerpHecT™ ROSS™, ThermoScientific™). NMR experiments All NMR experiments were performed on a 600 MHz Bruker Avance II spectrometer equipped with a cryogenic triple-resonance TXI probe, except for 1D 1H spectra reporting the CSPs, which were acquired on a 950 MHz Bruker Avance NEO spectrometer equipped with a cryogenic triple-resonance TCI probe. Data processing and analysis were conducted using Topspin 4.3.0 (Bruker). All experiments were recorded at 10 °C or 25 °C and pH 6.0. 1D 1H experiments 1D 1H spectra were acquired at different mAb concentrations (10, 25, 50, and 100 mg/mL). The spectrum recorded at 10 mg/mL was used as a reference. Under ideal linear conditions, the expected scaling factors would be 2.5 (10 to 25 mg/mL), 5 (10 to 50 mg/mL), and 10 (10 to 100 mg/mL). To quantify the deviation from the expected linearity, the loss of linearity was deduced for each concentration by subtracting the ratio between intensities from 1 and expressing the result as a percentage. For visualization, the loss of linearity was color-coded: 0–30% in light blue; 30–60% in light purple; 60–80% in light red and 80–100% in dark gray. Diffusion coefficient measurements and microscopic viscosity calculations Translational diffusion coefficients of excipients were measured using the PROJECTED (Periodic Refocusing Of J Evolution by Coherence Transfer Extended to DOSY) pulse sequence with convection compensation.52 This experiment contains intrinsically a T2-filter, which permits to selectively observe faster-diffusing excipient molecules. Most importantly, it is not sensitive to proton exchange with water, in contrast to most of the Pulse Field Gradients experiments. It also suppresses the J-evolution during diffusion delays and convection effects. Square gradients of 1 ms were applied using increments of maximal strength from 5% to 95% of 50 G.cm−1 nominal amplitude, and 15 loops were used including diffusion delays of 15 ms between gradient pulses. Acquisitions were carried out using 16 scans and an interscan delay of 2 s. Diffusion coefficients D were obtained by fitting the resultant signal decay f as a function of the increasing squared-shape gradient strength G using the Stejskal-Tanner Equation (54):(1)fG=e−2.n.D.γ2.G2.δ2Δ−δ3 with D the diffusion coefficient, G the gradient amplitude (G.cm−1), γ the gyromagnetic ratio, ∆ the diffusion time unit, δ the gradient pulse width and n the number of diffusion-encoding echoes. We then used the Stokes–Einstein equation to evaluate the NMR-based “microscopic” viscosity ƞNMR perceived by the excipients:(2)D=kBT6πηNMRRh with kB the Boltzmann constant, T the absolute temperature, ηNMR the viscosity measured by NMR and Rh the hydrodynamic radius of the excipient calculated as following. The macroscopic viscosity of the excipient alone in the histidine buffer was used to calculate the hydrodynamic radii Rh of the excipients at 10 °C via the Stokes-Einstein equation. The resulting calculated Rh values were: arginine 0.17 nm, lysine 0.16 nm, glycine 0.10 nm, sucrose 0.26 nm, PS20 3.6 nm, PS80 3.5 nm and P188 3.4 nm. Chemical shift perturbations To monitor any CSPs, all NMR data were acquired on a 950 MHz Bruker Avance NEO spectrometer equipped with a cryogenic triple-resonance TCI probe. CSPs were measured by assessing changes in the proton resonance positions of the excipient ligands when comparing them with and without the presence of the mAb. 1H 1D spectra were acquired on a Bruker spectrometer using the standard Bruker pulse sequence incorporating excitation sculpting with gradients for water suppression (zgesgp). To probe potential interactions occurring at the excipients’ extremities, arginine, lysine, and glycine samples were prepared with 25% of the total being 13C-15N-labeled amino acids to investigate their respective carbon positions. Sucrose and proline 13C positions were also monitored at natural abundance. To precisely determine shifts resulting from the presence of the mAb, the observed chemical shift values of the free excipients (reference) were subtracted from the values recorded in the presence of the mAb.(3)CSPmAb=δmAb−δmAb=0 STD-NMR To detect and identify interactions between the mAb and excipients using the STD-NMR experiment,53 the stddiffesgp.3 pulse sequence was used. Parameters included an ON-resonance frequency of 0.5 ppm, an OFF-resonance frequency of −80 ppm, 64 scans and an interscan delay of 5 s. The saturation was carried out during 2 s using a series of 2000 270 ° Gaussian-shaped pulses of 1 ms at γB12π = 1820 Hz, spaced by 4 μs delay between pulses. This allowed for depositing high levels of saturation at the expense of pulse selectivity. For this reason, ΔSTD ratios were calculated to correct for residual signals in control spectra, as follows: the STD ratio obtained with the excipient and mAb sample minus the STD ratio obtained with the excipient alone. A T1ρ filter was inserted before acquisition to remove the mAb signals, with a spinlock time of 40 ms.(4)ΔSTD=STDmAb+excipient−STDexcipientonly STD buildup curves were recorded for lysine and arginine at both 10 °C and 25 °C. Buildup curves were acquired at a mAb concentration of 100 mg/mL with excipient concentrations of 200 mM, using saturation times of 0.25, 0.5, 0.75, 1 and 2 s (Saturation pulses: Gaussian-shaped, 50 ms, 40 dB). ΔSTD ratios were calculated according to Equation 4. Transverse relaxation rate measurements The R2 of the excipients in the presence of the mAb were determined using the PROJECT (Periodic Refocusing Of J Evolution by Coherence Transfer) pulse sequence.54 This sequence, a modified Carr-Purcell-Meiboom-Gill experiment, was specifically chosen for its capability to suppress the effects of the potential chemical exchange and the J-modulation: quadrature 90° pulses within each double spin echo create a “perfect echo” and effectively refocus homonuclear J-evolutions. Relaxation delays larger than 20 ms provoke a selective filtration of broad signals originating from the mAb. Because viscosity affects R2 via its effects on correlation time, and because viscosity is not constant among the samples, R2 was decomposed into 2 components:(5)R2=R2exchange+R2tumbling R2exchange being the relaxation due to transient interactions with the mAb (exchange between free and bound forms), and R2tumbling being the basal intrinsic relaxation related to the local rotational correlation time τc of the free molecules. We considered that R2tumbling was proportional to τc, and that τc was proportional to the measured excipient diffusion coefficient (free rotational diffusion is always equal or superior to the theoretical one calculated with the Rh obtained from equation 2 and the experimental translational diffusion). Hence, R2exch=R2−R2freeDfreeD with R2free being the R2 of the excipient alone in the buffer, Dfree the diffusion coefficient of the excipient alone in the buffer, and D the diffusion coefficient in the considered sample mix. Dynamic light scattering A Wyatt DynaPro Plate Reader II instrument (Wyatt Technology) was used to measure light intensities scattered by an 824.9 nm laser through mAb solutions. Based on the Stokes−Einstein (SE) equation, the hydrodynamic diameter DH of the dominant species was calculated from the diffusion coefficient (D), as shown in this equation:(6)D=kBT3πηsDH where kB is the Boltzmann constant, T is absolute temperature, ηs is the solvent viscosity. The solvent viscosity and refractive index of water were used for the analysis. The analysis was performed at 10 °C, and mAb samples at 1, 5, 7, 10, 25, 50, 75 and 100 mg/mL were used. The scattered intensities were recorded at a fixed scattering angle (θ) of 158 °, after temperature equilibration of the samples. Then, 10 acquisitions were performed for one measurement, in a run duration of 5 s per acquisition. All experiments were performed in duplicates. The cuvette was loaded with a sample volume of 50 μL for examination. Dynamics software V7 (Dynapro) was used to calculate the particle size distribution. Results The sample macroscopic viscosity and the microscopic viscosity sensed by the excipients show uncorrelated variations upon addition of mAb First, we evaluated the macroscopic viscosities of solutions of our test mAb using rheometry in the presence of nine common excipients (Suppl. Figure S1). We performed the measurements in a standard formulation buffer (histidine at 10 mM, pH 6.0), a consensus condition for mAb therapeutics, pH 6.0 being within a common range of formulation pHs used for most approved antibody therapeutics since 2015.11 The studied mAb has a pI > 7 and is positively charged at pH 6.0. Without excipient, the viscosity of the mAb sample increased from 1.5 cP (no mAb) to 13.8 cP (mAb at 100 mg/mL) (Figure 1(A); Suppl. Table S2). It increased to 4.5–5.6 cP in the presence of arginine and lysine, to 7.3 cP with NaCl, and to 9.3–12.7 cP with proline, glycine and surfactants. We measured the largest viscosity (14.2 cP) with sucrose. Altogether, we concluded that four excipients, arginine, lysine, NaCl and proline, reduced the sample viscosity by keeping it under 10 cP at [mAb]=100 mg/mL, whereas glycine and surfactants showed moderate to null effects. Conversely, sucrose was the only excipient that increased the mAb sample viscosity. Figure 1.Macroscopic viscosities of the mAb samples and microscopic viscosities sensed by the excipients at 10 °C. (A) The macroscopic viscosities of the samples, measured using a rheometer with cone – plate geometry, are plotted as a function of the mAb concentration. (B) The microscopic viscosities, deduced from the NMR-derived diffusion coefficients of the excipients, are plotted as a function of the mAb concentration. Error bars report discrepancies between measurements from duplicates using independent samples..Two panels figure showing two diagrams. (A) Rheometry-measured viscosity grows somewhat exponentially in samples containing from 0 to 100 mg/mL of mAb, but in very variable regimes depending on the excipients, e.g. from 2 to 14 cP in presence of sucrose at 8% w/v, and from 2 to 4 cP in presence of arginine. (B) NMR-measured viscosity, extracted from diffusion-rates of excipients, grows somewhat linearly in samples containing from 0 to 100 mg/mL of mAb, with low variability, e.g. from 1.4 to 1.5 cP for PS20 and from 1.7 to 2.1 cP for glycine. Macroscopic viscosities of the mAb samples and microscopic viscosities sensed by the excipients at 10 °C. (A) The macroscopic viscosities of the samples, measured using a rheometer with cone – plate geometry, are plotted as a function of the mAb concentration. (B) The microscopic viscosities, deduced from the NMR-derived diffusion coefficients of the excipients, are plotted as a function of the mAb concentration. Error bars report discrepancies between measurements from duplicates using independent samples.. To further evaluate the viscosity at the molecular level, we measured the diffusion coefficients of the excipients in the presence of increasing concentrations of mAbs. We performed pulsed-field gradient (PFG) NMR experiments to quantify them for eight excipients (NaCl cannot be detected by 1H NMR). We observed limited variations of these diffusion coefficients as a function of the mAb concentration (Suppl. Table S3). We used these experimental values to calculate the hydrodynamic radii of the excipients in our conditions. We further deduced the viscosities sensed by the excipients as a function of the mAb concentration. These molecular-level viscosities, so-called microscopic viscosities, were significantly lower than the measured macroscopic viscosities (Figure 1(B); Suppl. Table S4). They rose from 1.4–1.7 cP to 1.6–2.1 cP as the mAb concentration increased from 0 to 100 mg/mL. Meanwhile, the bulk viscosity ranged between 4.4 and 14.2 cP at a mAb concentration of 100 mg/mL. There is no contradiction between macro- and micro-diffusion coefficients: these are known to diverge in the presence of macromolecular crowding, which starts at about 50–100 mg/mL for mAbs.55,56 The diffusion of small molecules is not or only weakly restrained by macromolecular crowders. It reports viscosity values close to the micro-scale viscosity, unless specific, strong binding occurs with the crowder.57–59 We observed that the microscopic viscosity experienced by arginine, lysine, proline, glycine, and sucrose increased progressively by more than 0.4 cP when the mAb was concentrated from 0 to 100 mg/ml. In the case of PS20, PS80, and P188, however, the micro-viscosity changed by less than 0.2 cP. The macroscopic viscosity of mAbs scaled as an exponential function of its concentration (the Ross-Minton model was used for fitting, see Figure 1(A)), while the microscopic viscosity scaled somehow linearly. When plotting the microscopic viscosity as a function of the macroscopic viscosity, we observed that the slope of the curve varied within excipients (Suppl. Figure S5). Lysine and arginine showed the most restricted motions at high mAb concentrations, suggesting that these excipients might interact with the mAb. 1D 1H NMR can discriminate between excipients used for mAb formulation, depending on their impact on mAb self-association We characterized mAb self-association using 1D 1H NMR. The NMR signals from the methyl groups of the mAbs between 0 and 2 ppm were specifically monitored due to their isolation from the excipients’ peaks and their robust signal intensities in the monomeric state of the mAb. First, we recorded the 1D 1H NMR spectrum of the mAb in 10 mM histidine pH 6.0 at concentrations ranging from 10 mg/mL to 100 mg/mL. A drastic loss of methyl signals was observed above 50 mg/mL (Figure 2(A); Suppl. Figure S6A). Then, we evaluated the beneficial effects of the common viscosity-reducing agent arginine.13,60 The loss of NMR signal was reduced at high mAb concentration (from 50 mg/mL) by adding arginine at 50 mM and was further reduced at all mAb concentrations (25 to 100 mg/mL) with arginine at 200 mM (Figure 2A). Figure 2.1H NMR characterization of the impact of a large set of excipients on mAb self-assembly in formulation-relevant conditions. (A) Effect of two arginine concentrations (50 and 200 mM) on mAb 1H NMR signal intensities. The color scale illustrates the efficiency of arginine in reducing 1H NMR signal linearity loss upon mAb concentration due to both viscosity and mAb self-assembly. This loss is classified as: very low, 0–30%, light blue; low, 30–60%, light purple; medium, 60–80%, light red and strong, 80–100%, dark grey. (B) Effect of a large set of common excipients on mAb 1H NMR signal intensities (Suppl. Table S5). Loss of 1H NMR signal linearity due to both viscosity and mAb self-assembly is color-coded as in (A). (C) Loss of 1H NMR signal linearity due to mAb self-assembly (Suppl. Table 6). Here, the loss of linearity was corrected based on the rheometry-derived macroscopic viscosity values, to remove the contribution of viscosity and extract the contribution of mAb self-assembly (same color code as above).Multi-panel figure. (A) Overlaid 1D NMR spectra at mAb concentrations of 25, 50 and 100 mg/mL of mAb in presence of arginine at 50 or 200 g/L, and color-coded tables indicating the progressive loss of NMR signal at growing concentrations of mAb (10, 25, 50, 100 mg/mL), mitigated by growing arginine concentrations (0, 50, 200 mM). (B) Color-coded tables indicating the progressive loss of NMR signal at growing concentrations of mAb (10, 25, 50, 100 mg/mL) in absence and in presence of the excipients studied here, showing the beneficial effects of arginine lysine, and the absence of effects from the other excipients (proline, glycine, sucrose, PS20, PS80, P188). (C) same as (B) but after correction of the NMR signal by the rheometry-measured viscosity, showing the same results than in (B). 1H NMR characterization of the impact of a large set of excipients on mAb self-assembly in formulation-relevant conditions. (A) Effect of two arginine concentrations (50 and 200 mM) on mAb 1H NMR signal intensities. The color scale illustrates the efficiency of arginine in reducing 1H NMR signal linearity loss upon mAb concentration due to both viscosity and mAb self-assembly. This loss is classified as: very low, 0–30%, light blue; low, 30–60%, light purple; medium, 60–80%, light red and strong, 80–100%, dark grey. (B) Effect of a large set of common excipients on mAb 1H NMR signal intensities (Suppl. Table S5). Loss of 1H NMR signal linearity due to both viscosity and mAb self-assembly is color-coded as in (A). (C) Loss of 1H NMR signal linearity due to mAb self-assembly (Suppl. Table 6). Here, the loss of linearity was corrected based on the rheometry-derived macroscopic viscosity values, to remove the contribution of viscosity and extract the contribution of mAb self-assembly (same color code as above). The intensity of the NMR signal depends on both the viscosity experienced by the mAb and its degree of reversible oligomerization. Viscosity slows molecular tumbling linearly, which results in linear intensity losses.43 The relevant viscosity here is the one related to rotational diffusion, which is more linked to the microscopic viscosity sensed by the mAb than to the macroscopic viscosity. Translational microscopic viscosity appears to follow the macroscopic viscosity in the case of concentrated mAbs,56 but it is often a high threshold for rotational viscosity in crowded environments,61–63 even though protein-specific effects can emerge from attractive or repulsive interactions.64 We analyzed the 1D 1H NMR spectra based on the hypothesis that the mAb diffuses in a solution with a viscosity equal to that measured by rheometry, to estimate the maximum possible intensity loss due to increased viscosity. To extract the information related to mAb self-assembly from the 1D 1H NMR spectra, we calculated the signal intensities of the mAb’s methyl groups upon removal of the viscosity effect, as explained above. In the absence of mAb self-assembly, these “viscosity-corrected” intensities should be proportional to the concentration of the mAb. Deviation from this linearity marks the presence of mAb clusters or aggregates that cause fast relaxation and signal losses. Hence, we used viscosity-corrected 1D NMR methyl intensities as a proxy to detect the formation of mAb clusters. No opalescence or cloudiness was observed in the samples, so that deviations from linearity were interpreted as being only due to reversible oligomerization. Moreover, we recovered exactly the original NMR peak intensities upon diluting concentrated samples, which supported the reversibility of the oligomerization processes. Finally, we performed SEC experiments before and after concentration of the mAb at 100 mg/mL in NaCl at 200 mM, a condition in which strong oligomerization appears. The two chromatograms (before and after the concentration/dilution cycle) were virtually identical in peak retention time, width, and area, with no detectable increase in high-molecular-weight species (Suppl. Figure S7). This confirmed that concentrating the mAb does not produce irreversible aggregates in this excipient condition, and that the self-association detected by 1D 1 H NMR at high concentration is reversible. We applied this approach to the analysis of our 1D 1H NMR spectra recorded on increasingly concentrated mAb samples by taking as a reference the NMR intensity at a mAb concentration of 10 mg/mL. We calculated the NMR signal linearity loss for each mAb concentration from 25 mg/mL to 100 mg/mL and illustrated the amplitude of this linearity loss with a color scale that we used throughout the study. Thus, we observed a loss of NMR signal linearity as a function of the mAb concentration above 50 mg/mL, as indicated using our color code in Figure 2A. In the presence of arginine at 50 mM and 200 mM, this loss of linearity was reduced. We also observed that the addition of arginine at 200 mM partially restored the 1H NMR signal linearity of mAb samples in the presence of other excipients such as lysine, sucrose, proline, and P188, so that combining excipients improved the NMR-detected mAb behavior (Suppl. Figure S6B). We finally extended our NMR analysis to mAb samples concentrated at 10, 25, 50, and 100 mg/mL in presence of the nine selected excipients (the samples contained only one of these excipients). We found that the losses of NMR signal linearity (Figure 2(B); Suppl. Table S5) and the extracted propensions of mAb to form clusters (Figure 2(C); Suppl. Table S6) strongly varied within excipients. Only arginine, lysine, and NaCl significantly reduced the NMR signal loss at high mAb concentrations (Figure 2(B)). Once the effect of viscosity was removed, arginine was confirmed as the most efficient excipient for limiting mAb reversible oligomerization (Figure 2(C), Supp. Figure S4C). Lysine also largely prevented mAb reversible oligomerization up to 100 mg/mL, whereas NaCl, which was less efficient at reducing the macroscopic viscosity, did not modify mAb self-assembly. Sucrose reduced mAb oligomerization only at low mAb concentration (up to 25 mg/ml), and P188, PS20 and PS80 showed no detectable impact on the distribution of the mAb species. Proline and glycine even slightly increased mAb self-assembly. The results do not depend strongly on the mode of viscosity correction: taking into account macroscopic viscosity (Figure 2(C)) or NMR-derived microscopic viscosity (Suppl. Figure S6C) for correcting 1D NMR intensities does not change the ranking and the conclusions. Overall, we found that, under formulation-relevant conditions, (1) our mAb forms reversible intermolecular interactions that are associated with increased sample viscosity, (2) mAb self-assembly can be decreased with high efficiency by only two excipients, which are arginine and lysine, and (3) mAb self-assembly is not the only determinant of viscosity, as proline and glycine favored the formation of clusters but did not lead to the highest sample viscosities. DLS detects only very weak populations of stable mAb clusters and delivers questionable diffusion coefficients To further characterize mAb clustering, DLS measurements were conducted on the mAb alone and in the presence of the nine excipients (Figure 3(A-C)). The autocorrelation curves did not reveal any presence of stable mAb clusters, in the presence of proline at 200 mM, which provokes a poorly reproducible population of large, stable oligomers (Figure 3(C)). Fitting the other autocorrelation curves using standard models of stable monodisperse or polydisperse particles,65 namely cumulant and regularization fits, led to the identification of one major species at 1 mg/mL of mAb, with apparent diffusion coefficients between 2.8 and 3.8 × 10−7 cm2/s depending on the excipient (Suppl. Table S7). At higher mAb concentrations, analyzing the distributions of particle sizes did not reveal the presence of larger stable mAb oligomers. Such large species (radii of ~100–1000 nm) were detected only in the graphs of particle distributions by DLS intensity, but not in the distributions by mass, indicating that they represented a negligible fraction of the whole species (Suppl. Figure S8). Moreover, in classical models of oligomerization thermodynamics, the size and population of stable clusters would grow with mAb concentrations, of which there was no sign in the distributions. Thus, DLS did not report any clear signs of sizable populations of stable mAb clusters. Figure 3.DLS-derived mAb apparent diffusion coefficients as a function of the mAb concentrations and in the presence of various excipients. DLS experiments were performed (A) in the absence of excipient and in the presence of lysine at 200 mM, glycine at 200 mM, sucrose 8%, P188 0.5%, PS20 0.1% and PS80 0.1%, (B) in the presence of arginine at 200 mM and NaCl at 200 mM, and (C) in the presence of proline at 200 mM. Left panels: autocorrelation curves recorded in selected conditions, including without excipients (from light blue to dark red following mAb concentration increase). Right panels: apparent diffusion coefficients (as obtained from fitting autocorrelation curves using a model of stable particles) of the mAb without excipient (light blue) and in the presence of each of the excipients (dark blue). Diffusion coefficient profiles were calculated using the polydisperse fit of the DYNAMICS V7 software by Wyatt technology. The displayed profiles correspond to the population with the lowest radius. Error bars report discrepancies between measurements from duplicates using independent samples..Multi-panel figure. (A) on the left: 3DLS autocorrelation curves of the mAb are overlaid at growing concentrations of mAbs in absence of excipients (top) in presence of scurose 8% (middle) and of PS80 0.1% (bottom), and on the right, 6 graphs show the extracted diffusion rates of the mAb in function of the mAb concentration in presence of 6 excipients (lysine, glycine, sucrose, P188, PS20, PS80) that provoke an initial increase of the diffusion rate from 0 to 20 mg/mL of mAb (from 3.10−7 to 6.10−7 cm2/s), and then a decrease of the diffusion rate from 20 to 100 mg/mL of mAb (from 6.10−7 to 4.10−7 cm2/s). (B) Same as (A) but in presence of arginine or NaCl, showing a continuous decrease of the diffusion rate on the right from 0 to 100 mg/mL of mAb (from 3.10−7 to 1.5.10−7 cm2/s). (C) Same as (A) but in presence of proline, showing the detection of large species in the autocorrelation curves on the left, and on the right, an evolution of the diffusion rate similar to those in (A) for the major mAb species. DLS-derived mAb apparent diffusion coefficients as a function of the mAb concentrations and in the presence of various excipients. DLS experiments were performed (A) in the absence of excipient and in the presence of lysine at 200 mM, glycine at 200 mM, sucrose 8%, P188 0.5%, PS20 0.1% and PS80 0.1%, (B) in the presence of arginine at 200 mM and NaCl at 200 mM, and (C) in the presence of proline at 200 mM. Left panels: autocorrelation curves recorded in selected conditions, including without excipients (from light blue to dark red following mAb concentration increase). Right panels: apparent diffusion coefficients (as obtained from fitting autocorrelation curves using a model of stable particles) of the mAb without excipient (light blue) and in the presence of each of the excipients (dark blue). Diffusion coefficient profiles were calculated using the polydisperse fit of the DYNAMICS V7 software by Wyatt technology. The displayed profiles correspond to the population with the lowest radius. Error bars report discrepancies between measurements from duplicates using independent samples.. Then, we compared the DLS-reported apparent diffusion coefficients of the major species, i.e., most likely the mAb monomer. In most conditions, (including without excipients), the fitted diffusion coefficients increased for mAb at 1 to 25 mg/mL, and then decreased until 100 mg/mL, except in the presence of arginine and NaCl, which provoked a steady progressive decrease. (Figure 3A,B). The resulting graphs showing coefficient diffusion as a function of mAb concentrations reveal initial slopes, which are commonly analyzed by calculating the protein interaction parameter kD.kD=DD0−1mAb0−10mg/mL This interaction parameter is interpreted as reflecting the attractive or repulsive forces exerted between mAbs molecules at low concentrations in a given buffer.21,22 Typically, a positive kD value (measured here without excipient; Figure 3(A)) would indicate a dominant repulsion between mAbs, which, in turn, has been proposed as a predictor of low propensity to generate clusters and thus low viscosity at high mAb concentration. A negative kD value (like the one measured with arginine and NaCl; Figure 3(B)) would predict the opposite, i.e., favored clusters and high viscosity. In line with this interpretation, our DLS measurements report that, for our mAb without excipient, the interactions between mAbs at low concentrations are repulsive, as deduced from the positive kD value, consistently with the limited viscosity increase measured by rheometry upon mAb concentration. However, samples formulated with two of the most viscosity-reducing excipients, arginine and NaCl, exhibited negative kD values, while samples with two other viscosity-reducing species, lysine and proline, led to positive kD values (Figure 3(A,B)). Therefore, we thought to verify the basic hypothesis of classical DLS fitting, i.e., the stability of the characterized species. The possibility of measuring chemical exchange using DLS was evoked as early as the 1970s,66 but was not further developed, to the best of our knowledge. Here, we knew from NMR that reversible oligomerization took place in many conditions when increasing mAb concentration. Because NMR detects exchange at the µs-ms timescale, we tried to fit the initial autocorrelation curves up to ~30 µs. We noticed that the resulting diffusion coefficients were almost constant at all mAb concentrations (Suppl. Figure S9), indicating the presence of mAbs as monomers in all samples. We obtained similar results by fitting the late autocorrelation curves, from 100 µs to 1 s (Suppl. Figure S9). Hence, it is possible that exchange between monomeric and reversible oligomeric forms occurring at high mAb concentrations generates autocorrelation curves that cannot be accurately interpreted using classical DLS fitting procedures, which rely on the hypothesis of the stability of the observed particles. Altogether, DLS measurements provided surprising variations in diffusion coefficients of mAb species: in contrast with most previous reports, kD values were anti-correlated with self-assembly and viscosity at high mAb concentrations. This might reflect the presence of different types of transient interactions upon mAb concentration. Moreover, DLS detected only minor populations of high molecular weight species, which was not in agreement with the 1D 1H NMR observations reporting reversible self-assembly for mAb concentrations above 25 mg/mL. We observed that the classical stable-particles model for fitting DLS autocorrelation curves is probably not adapted to the present situation, where the mAb might exchange between monomeric and reversible oligomeric forms at high concentrations. A solution would be the use of improved fitting equations of the DLS autocorrelations curves incorporating monomer-oligomer exchange in the microsecond time scale, which is beyond the scope of this study. 1H and 13C chemical shift perturbations reveal weak excipient interactions with monomeric and clustered mAb species To detect interactions between the excipients and the mAb and to identify their binding moieties, we monitored the excipient 1H and 13C CSPs at 950 MHz in the presence of increasing concentrations of the mAb. CSPs report for variations in average chemical environments of species experiencing fast chemical exchange (>106 Hz).31 We were careful to keep pH (<0.05 pH unit accuracy) and temperature constant. To accelerate the detection of the 13C signals, the excipients arginine, lysine, and glycine were a mix of 75% unlabeled amino acids and 25% 15N,13C-labeled amino acids. We observed reproducible CSPs at 950 MHz for all the excipients in the presence of mAb up to 100 mg/mL. The averaged CSP among all 1H and 13C of a given excipient ranged from 0.006 to 0.014 ppm (6 to 13 Hz) and from 0.006 to 0.033 ppm (1.5 to 8 Hz), respectively (Figure 4(A,B); Suppl. Tables S8 and S9). Accordingly, these CSPs could be discriminated only at a high magnetic field, which does not necessarily mean that the detected interactions are limited and rare: the weak CSPs values may result from the averaging of the chemical shifts generated by the numerous binding sites explored by the excipients on the mAb’s surface. This would also be consistent with the similarities observed between CSPs for the different atomic positions of every given excipient. Figure 4.NMR chemical shift perturbations (CSPs) measured at 950 MHz for every proton and carbon of the excipients, reflecting excipient–mAb interactions. The graphs show the (A) 1H and (B) 13C CSP for the different proton and carbon positions of the excipients, respectively (Suppl. Tables S8 and S9). The color scale is overlaid, providing a visual reference to the self-association behavior of the mAb (Figure 2C, Suppl. Table S6). Error bars report discrepancies between measurements from duplicates using independent samples.Multi-panel figure. (A) 8 graphs show the chemical shift perturbations (CSPs) for all protons of the 8 excipients in function of the mAb concentration from 0 to 100 mg/mL, showing linear evolutions for arginine, lysine, glycine and PS20, and irregular evolutions for proline, sucrose, PS80 and P188, all CSPs being below 0.015 ppm. (B) Same as (A) but for carbons’ CSPs of the arginine, lysine, glycine, proline and sucrose, the results are very similar to those in (A). NMR chemical shift perturbations (CSPs) measured at 950 MHz for every proton and carbon of the excipients, reflecting excipient–mAb interactions. The graphs show the (A) 1H and (B) 13C CSP for the different proton and carbon positions of the excipients, respectively (Suppl. Tables S8 and S9). The color scale is overlaid, providing a visual reference to the self-association behavior of the mAb (Figure 2C, Suppl. Table S6). Error bars report discrepancies between measurements from duplicates using independent samples. The CSPs did not vary linearly with the concentration of the mAb. This suggests that the interfaces between the excipients and the mAb change depending on the mAb transient oligomeric states. The measured CSPs result from the addition of CSPs due to the mAb monomers and CSPs due to the mAb clusters, and reflect the diverse populations of mAb species, which depend on the mAb concentration and the nature of the excipient. In conditions of low mAb concentrations and low self-assembly (light blue and purple in Figure 4(A,B)), the excipients showing the largest CSP values are arginine, lysine, and sucrose. In conditions of strong mAb self-assembly at [mAb]=100 mg/mL, as revealed by vanishing methyl 1D NMR signals (red and black backgrounds), we measured the largest CSPs in the presence of sucrose and the surfactants polysorbates and poloxamer. Overall, CSPs detected excipient–mAb interactions, but comparing the strength of these interactions is difficult due to the diversity of the explored binding sites and the evolution of the accessible surface of the mAb as a function of its transient oligomeric states. STD-NMR identifies excipients with slower exchange regimes at the surface of the mAb We investigated further the interactions between the mAb and the excipients using the STD NMR method, which can also provide insight into the binding epitopes of the excipients.53 Due to the very high concentrations of these excipients, we observed imperfections in the spectra: the STD-NMR signals were never completely null, even in the absence of mAb.53 To eliminate these imperfections, we calculated the difference in STD values in the presence and absence of the mAb to obtain ΔSTD values (Figure 5; Suppl. Table S10). Moreover, residual dispersion-looking signals emerged from the subtraction of the on- and off-resonance spectra, probably because of imperfect electronics and faint differences in pulses and acquisition time delays, and weak direct saturation were observed for signals at 1 ppm from the saturation offset (Suppl. Figure S10). Despite these artifacts, reproducible ΔSTD-NMR values were generally observed in the presence of the mAb (Figure 5). We used saturation times of 2 s, which provided the highest expectable ΔSTD-NMR levels according to buildup curves recorded with arginine and lysine (Suppl. Figure S11). They increased with the concentration of the mAb, although non-linearly in most cases. Here again, it seems that monomeric mAbs do not interact with excipients in the same way as mAb clusters. While larger cluster species should normally transfer their saturation more efficiently to binding species,31 we observed an inversion of the STD slopes when NMR-detected mAb self-assembly becomes strong (i.e., [mAb]>50 mg/mL) in the presence of sucrose. This indicates probably that this excipient interacts with different kinetics with mAb monomers compared to large clusters. Figure 5.ΔSTD–NMR values plotted against mAb concentration for every proton position of the excipients. The color corresponding to the mAb self-assembly scale is in background, providing a visual reference for the self-association behavior of the mAb (Figure 2C, Suppl. Table 6). The proline plot has a different y-axis scale, due to large ΔSTD–NMR values. Error bars report discrepancies between measurements from duplicates using independent samples.Eight graphs show the ΔSTD–NMR values for all protons of the 8 excipients in function of mAb concentration from 0 to 100 mg/mL, showing somewhat linear evolutions for arginine (ΔSTD–NMR < 2%), lysine (ΔSTD–NMR < 8%), glycine (ΔSTD–NMR < 2%) and PS20 (ΔSTD–NMR < 2%), no STD for PS80 and P188, growing STD for sucrose from 0 to 50 mg/mL of mAb (ΔSTD–NMR < 2%) and a decrease at 100 mg/mL, and and very irregular and hectic evolutions for proline’s protons (ΔSTD–NMR < 25%). ΔSTD–NMR values plotted against mAb concentration for every proton position of the excipients. The color corresponding to the mAb self-assembly scale is in background, providing a visual reference for the self-association behavior of the mAb (Figure 2C, Suppl. Table 6). The proline plot has a different y-axis scale, due to large ΔSTD–NMR values. Error bars report discrepancies between measurements from duplicates using independent samples. Conversely, larger ΔSTD-NMR values emerged with the appearance of mAb clusters for lysine and proline. In the case of proline, using DLS, we already observed that the mAb sample is a mix of monomers and of minor populations of large and poorly reproducible clusters. Here again, we found that it is intrinsically difficult to produce two samples with the same large clusters, close to aggregation. The ΔSTD-NMR values measured upon addition of proline to increasing mAb concentrations were enhanced due to more efficient spin diffusion from the specifically large mAb clusters. Because viscosity slows down rotational correlation times and thus speeds up spin diffusion, it can also contribute to the increase of ΔSTD-NMR values at higher mAb concentrations. To evaluate these effects, we carried out STD-NMR experiments with BSA, which allows high protein concentrations without generating any cluster (Suppl. Figure 12A-C). We observed a rapid increase in the ΔSTD-NMR values for arginine, proline and sucrose at high concentrations of BSA, suggesting that protein–excipient interactions can be measured in our conditions even in the absence of clusters or in low viscosity samples. Altogether, it was difficult to give a univocal interpretation of ΔSTD-NMR values because of the mixed nature of the mAb samples, where monomers and clusters contribute differently to saturation transfers, a situation further complicated by viscosity effects on spin diffusion at high mAb concentrations. However, we can conclude that arginine, glycine, polysorbates, and poloxamer interact with the mAb surface at extremely fast regimes that do not allow spin diffusion and saturation transfer to develop, as compared to probably slower regimes for lysine and proline. In the latter case, large clusters are also responsible for the high saturation values. Finally, sucrose interacts differently with monomers and clusters, as shown by the significant but non-linear STD effects measured in the presence of increasing mAb self-assembly. 1H R2 measurements discriminate the different mAb binding capacities of the excipients Finally, we measured the relaxation properties of the excipients in the presence of increasing concentrations of mAb. When using prolonged relaxation delays, 1H R2 relaxation allows for the detection of ultra-weak interactions, as demonstrated recently by Gossert and colleagues.34 However, the 1H R2 rate depends primarily on the correlation time τc of the free species, with a linear relationship between R2free and τc, to which the binding exchange contribution adds itself. To extract the exchange component, we subtracted to the measured 1H R2 values the 1H R2 of the free excipients multiplied by the local viscosities measured earlier using PFG-NMR. We named the resulting parameter R2exchange, as it reports on the exchange between the free and mAb-bound excipient. Strictly speaking, the excipient R2exchange parameter is also influenced by the mAb R2, which might be negligible because of the supposed extremely low residence time of the excipient at the surface of the mAb. It is worth mentioning that we had to extrapolate 1H R2 values from the measured viscosities because measuring them in viscosity-matched samples was not reliable. Indeed, at high concentrations, the usual viscosity-matching molecules, like glycerol or sucrose, cannot be considered as non-binding species. This is exemplified in Suppl. Figure S13, where we report the 1H R2 values of arginine in a buffer supplemented with glycerol at 7.5% v/v, i.e. at a concentration generating the viscosity measured with mAb at 100 mg/mL in arginine at 200 mM. Arginine’s protons appear to experience varying chemical shift environments when loosely interacting with glycerol or water molecules, which provokes 1H R2 values higher than those expected if viscosity was the only influencing parameter. Most R2exchange values above 1 Hz were found for mAb concentrations of 50 and 100 mg/mL, showing again that the interactions between the mAb and the excipients are difficult to detect below formulation-relevant mAb concentrations (Figure 6; Suppl. Table S11). However, R2exchange measurements are more straightforward to interpret than their ΔSTD-NMR counterparts. The R2exchange graphs obtained with either the mAb or BSA share more similarities than those resulting from ΔSTD-NMR experiments (Suppl. Figure S12D). R2exchange measurements are not blurred by the complications due to different spin diffusion efficiencies between species met in ΔSTD-NMR measurements. They can still be sensitive to the mAb R2 in case of very large clusters: consistently, the R2 values of proline 1H are significantly larger in the presence of the mAb as compared to BSA (Suppl. Figure S12D). Figure 6.R2exchange values as a function of mAb concentration for the different excipients. Overlaid is the color scale, providing a visual reference for the self-association behavior of the mAb (Figure 2C, Suppl. Table 6). Error bars report discrepancies between measurements from duplicates using independent samples.Eight graphs show the R2exchange values for all protons of the 8 excipients in function of mAb concentration from 0 to 100 mg/mL, showing somewhat linear evolutions for arginine (R2exchange< 2 Hz), glycine (R2exchange< 1 Hz), proline (R2exchange< 2 Hz), sucrose (R2exchange< 6 Hz), almost no R2exchange for PS20, PS80 and P188 (R2exchange< 1 Hz), and a disrupted evolution for lysine with a much steeper growth of R2exchange from 50 to 100 mg/mL of mAb (R2exchange< 1 Hz at 50 mg/mL, and about 4 Hz at 100 mg/mL). R2exchange values as a function of mAb concentration for the different excipients. Overlaid is the color scale, providing a visual reference for the self-association behavior of the mAb (Figure 2C, Suppl. Table 6). Error bars report discrepancies between measurements from duplicates using independent samples. Our R2exchange analysis confirmed the nearly undetectable binding of glycine, polysorbates, and poloxamer to the mAb (Figure 6). We could detect the weak interactions of arginine, and the stronger interactions of lysine and sucrose with the mAb surface. We could also identify interactions of proline with mAb clusters. Overall, these results are consistent with the CSP and STD data. Furthermore, the R2exchange analysis appears to be less susceptible to the nature of mAb species. Discussion Viscosity is an emergent property arising from molecular architecture and noncovalent intermolecular interactions resulting in the formation of a dynamic particle network. To control the viscosity of highly concentrated therapeutic mAbs, excipients are commonly incorporated into formulations.12–16 Several mechanisms of action have been proposed for these excipients, which are non-exclusive: they can weaken or disrupt the interaction network formed by individual mAbs molecules or clusters; they can influence solvent organization at the surface of the different mAbs species, forcing either a better solvation or solvent layer depletion; they can modify the solvent polarizability and hydrogen-bond capacities67; they can even stabilize self-repulsive nanoclusters ensuring good colloidal properties.68 Understanding the relationship between excipient interactions and mAb self-assembly is key to guiding the selection of excipients that will best reduce the viscosity of a mAb sample. Here, we took advantage of the capacity of NMR to characterize both highly concentrated mAbs and excipients in formulation-relevant samples. We combined NMR spectroscopy, rheometry and DLS to describe the species present in our samples. Then, we analyzed the correlation between NMR-detected mAb–mAb as well as mAb–excipient interactions and viscosity at high mAb concentrations. Importantly, while NMR has great advantages for detecting weak interactions, NMR observables are sensitive to a whole set of binding parameters, so that it is often difficult to extract quantitative information on the interactions. In the present case, excipients can interact with an undetermined but certainly large number of binding sites on the surface of the mAb, which makes a quantitative interpretation of the NMR difficult. Indeed, all the corresponding binding event come with varying thermodynamics and kinetics, and also with varying chemical shifts. These parameters influence CSP, STD and R2 values. Hence, the ranking that we provide between excipients must be considered as a qualitative one. First, focusing on the relationship between viscosity and mAb–excipient interactions, we did not perceive any singular rule from our characterization. The studied mAb showed a high viscosity behavior in a simple buffer containing histidine at 10 mM, pH 6.0: at 100 mg/mL, the viscosity of the sample was 13.8 cP (at 10 °C). Rheometry measurements showed that almost all the excipients except sucrose reduced the viscosity (down to 4.4 cP for the best tested viscosity reducer: arginine); surfactants had nevertheless a limited impact on the sample viscosity (Figure 7). NMR-detected diffusion gave access to the microscopic viscosities sensed by the excipients: these were much lower and did not correlate with the macroscopic ones, reaching 2.0–2.1 cP for arginine, lysine, proline, and glycine, 1.8 cP for sucrose and 1.6 for polysorbates and poloxamer. It was tempting to interpret these differences as sensors of mAb–excipient interactions, and to conclude that arginine and lysine interact most with the mAb, but this was not in full agreement with the other NMR observables. Indeed, the NMR CSPs, STD-NMR and R2exchange measurements reported weak interactions between a large set of excipients and the mAb. All these interactions were difficult to measure below 50 mg/mL, which underlines the importance of measuring them under formulation-relevant conditions. Moreover, NMR observables reported different interactions between excipients and either the mAb monomers or the NMR-detected mAb clusters. This generated non-linear variations between NMR observables and mAb concentrations, posing difficulties to deconvolute the respective effects of interactions with monomers and with clusters. Figure 7.Summary of the effects of excipients on sample viscosity, mAb self-assembly, and mAb–excipient interactions as revealed by rheometry and NMR. For macroscopic viscosity (rheometry), 1D 1H NMR of mAb methyl signals, and microscopic viscosity derived from NMR diffusion measurements, the color scale classifies each excipient according to its overall impact: blue indicates the most favorable behavior (strong reduction of macroscopic viscosity, efficient limitation of mAb self-assembly, and the largest change in excipient diffusion behavior upon mAb addition), yellow denotes intermediate behavior, and red corresponds to the least favorable behavior (poor viscosity reduction, strong mAb self-assembly, and negligible effect of mAb on excipient diffusion). For chemical shift perturbation (1H/13 C CSP) experiments, blue identifies excipients exhibiting linear CSPs variations as a function of the mAb concentration (similar interactions with all mAb species), whereas red indicates significant slope variations (different interfaces with monomers versus clusters). For STD-NMR and R2exchange measurements, the color scale reports the strength of detectable mAb–excipient interactions, with blue corresponding to the highest NMR response (strongest apparent interactions) and red to the lowest response (weakest or undetectable interactions). Proline is shown in black in the STD-NMR column because, despite exhibiting in average a very high STD-NMR signal, the STD-NMR results displayed low reproducibility.Color coded table show the influences of the 8 excipients on the 6 observables (rheometry-measured macroscopic viscosity, 1D NMR signal intensities, NMR-measured microscopic viscosity,1H and13C CSPs, STD-NMR and R2exchange). Arginine and lysine have overall beneficial effects on viscosity and cluster formation, but arginine interacts less than lysine with the mAb according to CSPs, STD-NMR and R2exchange, glycine and proline have limited effects and poorly interact with the mAb according to CSPs, STD-NMR and R2exchange(even though large clusters provoke large STD in the case of proline, marked in black), sucrose, PS20, PS80 and P188 have no beneficial effects and poorly interact with the mAb according to CSPs, STD-NMR and R2exchange. Summary of the effects of excipients on sample viscosity, mAb self-assembly, and mAb–excipient interactions as revealed by rheometry and NMR. For macroscopic viscosity (rheometry), 1D 1H NMR of mAb methyl signals, and microscopic viscosity derived from NMR diffusion measurements, the color scale classifies each excipient according to its overall impact: blue indicates the most favorable behavior (strong reduction of macroscopic viscosity, efficient limitation of mAb self-assembly, and the largest change in excipient diffusion behavior upon mAb addition), yellow denotes intermediate behavior, and red corresponds to the least favorable behavior (poor viscosity reduction, strong mAb self-assembly, and negligible effect of mAb on excipient diffusion). For chemical shift perturbation (1H/13 C CSP) experiments, blue identifies excipients exhibiting linear CSPs variations as a function of the mAb concentration (similar interactions with all mAb species), whereas red indicates significant slope variations (different interfaces with monomers versus clusters). For STD-NMR and R2exchange measurements, the color scale reports the strength of detectable mAb–excipient interactions, with blue corresponding to the highest NMR response (strongest apparent interactions) and red to the lowest response (weakest or undetectable interactions). Proline is shown in black in the STD-NMR column because, despite exhibiting in average a very high STD-NMR signal, the STD-NMR results displayed low reproducibility. Despite these difficulties, since each NMR measurement is sensitive to a different interaction regime, our data provided a description of a large set of mAb–excipient interactions, characterized by different strengths and kinetics. From this analysis, we found that several mechanisms can regulate the sample viscosity. Arginine provides a clear illustration of the pipeline used to correlate mAb interaction and viscosity. In its presence, increasing mAb concentrations generate 1D 1H NMR signal intensities scaling almost linearly (after correcting for viscosity effects), indicating a reduced mAb self-assembly associated with a limited sample viscosity. CSPs revealed that arginine binds through different interfaces to the mAb monomer and clusters. The associated STD-NMR and R2exchange contributions remain low and can be interpreted as caused by short-lived surface contacts. These transient interactions are sufficient to strongly reduce self-assembly and viscosity. Lysine is also an efficient viscosity-reducing excipient, but its mechanism of action appears to be different. Indeed, we measured the largest interaction signals for lysine with STD-NMR and R2exchange experiments. Thus, lysine–mAb interactions are probably characterized by more extended binding times than arginine. These contacts are efficient in decreasing mAb self-assembly and the sample viscosity.31 Proline interactions led to moderate CSPs and R2exchange and high STD-NMR values; the mAb species distribution was variable from one sample to the other because proline triggers the formation of large clusters or aggregates according to DLS measurements. Consistently, proline has only weak viscosity-reducing properties on this mAb. Sucrose interactions produced strong CSPs as well as moderate ΔSTD-NMR values peaking at 50 mg/mL and large R2exchange values; sucrose binds to all mAb species, but the sucrose free and bound forms probably exchange more slowly in the presence of mAb monomer and small clusters, and quicker in the presence of large clusters. Interaction with the mAb monomer significantly decreases mAb self-assembly up to 25 mg/mL, but interactions with larger mAb species is no more efficient at preventing oligomerization and limiting the viscosity increase. Conversely, polysorbates and poloxamer interact apparently more with the mAb clusters, but in a very fast exchange regime according to the hardly detectable STD-NMR and R2exchange values, and these interactions do not help reduce the viscosity of the sample. Overall, we concluded that the excipient impact on viscosity depends not only on the excipient–mAb interactions, but also on how these interactions affected the different types of mAb clusters. When we further characterized mAb–mAb interactions and their correlation with viscosity, we obtained two apparently inconsistent descriptions of the mAb clusters using either the NMR or DLS techniques. The simplest NMR-observable, 1D 1H-NMR intensities from mAb’s methyls, was very informative: it revealed the appearance of reversible mAb–mAb interactions, resulting in reduced NMR intensities. Two excipients decreased mAb NMR-detected self-assembly, i.e., arginine and lysine, whereas NaCl, proline, glycine, sucrose and surfactants had little impact, and proline and glycine slightly increased mAb self-assembly (Figure 7). These mAb–mAb interactions arose at varying mAb concentrations depending on the excipient. The early occurrence of NMR-detected self-assembly partially correlated with high viscosity at high mAb concentrations: from our 1D NMR analysis at 10 to 25 mg/mL, arginine could be predicted as the best excipient to reduce the viscosity, and NaCl, proline, glycine, and surfactants to be non-potent candidates for viscosity reduction. Sucrose shall be treated separately. Its addition significantly reduced self-assembly at low mAb concentration, but not at high mAb concentration, and was not efficient at decreasing the sample viscosity. DLS is commonly used to predict mAb self-assembly and sample viscosity at high concentration (100 mg/mL) from data acquired at low concentrations (below 10 mg/mL). We fitted the DLS autocorrelation curves using a model of stable diffusing particles to calculate kD parameters. Positive kD can report for repulsive interactions at low mAb concentrations and has been used to predict low viscosity at high mAb concentrations.21,22 Here, positive kD values were obtained for the majority of excipients (except arginine and NaCl) regardless of their rather weak capacities in reducing viscosity at high mAb concentration. A positive kD was even obtained with sucrose, which has viscosity enhancing properties on this mAb. Moreover, negative kD values were estimated in the presence of arginine and NaCl, which both reduce the viscosity of this mAb. These results contradicted the approach using kD at low mAb concentration to predict viscosity at high mAb concentration.21,22 Several hypotheses can be proposed to explain this situation. First, such a relationship between kD and viscosity at high mAb concentration is not necessarily impossible: kD reports a priori on the first-order interactions dominating at low concentrations, and not on the higher-order ones involving clusters, the latter of which can be driven by different exposed surfaces and different local and global charges. We found using NMR that, for example, in the case of sucrose, interactions with either mAb monomers or clusters differ, and consistently sucrose reduces mAb self-assembly only at low mAb concentrations, which is not sufficient to decrease viscosity under formulation-relevant conditions. Moreover, promiscuity at high concentrations forces a new balance between attractive and repulsive forces, which can have different distance dependencies. Very weak salt concentrations can favor high repulsion at low mAb concentrations, but be incapable of mitigating cluster formation and high viscosity at high mAb concentrations. This has previously been described in the literature.27,69 Arginine and NaCl may enhance self-repulsive interactions between the nanoclusters detected by NMR and DLS, thereby improving their colloidal behavior and leading to a lower viscosity compared with the reference sample containing the mAb alone. A second explanation regarding the kD behavior could be that the classical model underlying fitting equations of DLS autocorrelation curves, i.e., stable particles without any transient binding, is not adapted to our mAb samples, especially in highly self-assembled mAb conditions. 1D 1H NMR analysis suggested that transient exchange between monomers and clusters exists in the µs-ms time scale. The kD values obtained from the classical stable-particle model clearly did not fully correlate with NMR-derived mAb self-association in formulation conditions. This should prompt caution when considering DLS to assess reversible oligomerization. The self-assembly of mAbs can also be detected using small-angle X-ray scattering, SLS, or composition-gradient multi-angle static light scattering, for example,67,70 but NMR is unique in providing information on both the mAb and excipients in formulation relevant conditions. Finally, because the two most efficient excipients are the charged arginine and lysine, the contribution of the electrostatic screening might be questioned. We recorded rheometry and NMR experiments in conditions where ionic strength was matched between arginine, lysine and NaCl samples (205 mM) and between Gly, Pro, sucrose, PS20/PS80, and P188 samples (5.2 mM), while osmolality was also matched between arginine, lysine and NaCl samples (~400 mOsm/kgH2O) or set to weaker values for other excipients (Arg/Lys/NaCl > sucrose >glycine >surfactants, see Suppl. Table S12). Altogether, we observe a correlation between high ionic strength and viscosity reduction, whereas no clear correlation appears to exist between osmolality and viscosity. However, the beneficial effects of arginine and lysine are larger than those of NaCl, and thus are not entirely due to generic electrostatic screening.30,71,72 In conclusion, our study demonstrated that combining NMR techniques with rheometry and DLS provides complementary perspectives on the weak and transient interactions between mAbs and excipients in highly concentrated biopharmaceutical solutions. Our analyses highlighted the importance of the NMR-detected parameters in describing the complex interaction landscape of the highly concentrated mAb, which regulates the viscosity of the mAb formulation. In the case of our mAb, provided by Sanofi, most of the excipients decreased the viscosity of the formulation, even faintly, arginine and lysine being the best viscosity-reducing molecules. The NMR analysis, even if complicated by the high concentrations of both mAb and excipients, revealed that arginine interacted through transient interactions with the mAb, exploring different interfaces when binding to either mAb monomers or mAb clusters. The resulting complex interaction network was the most efficient at reducing the sample viscosity under formulation-relevant conditions. Lysine was detected as the best mAb binder, associated with a strong reduction in mAb self-assembly and sample viscosity, even if weaker than arginine. Importantly, all these measurements are specifically linked to the nature of the studied mAb, of its subclass and more generally of its amino acid composition. Our experimental strategy is nevertheless applicable to any other mAb and should help in identifying the best excipients depending on its constitution. To further characterize the mechanisms of these interactions and distinguish between diffusion at the surface of the mAb and site-specific binding, further relaxation analyses should be performed on a large range of magnetic fields. Such relaxometry studies would permit disentanglement of the broad range of time frame at which the various mAb–mAb and mAb–excipient interactions take place. This would improve our understanding of the complex link between mAb–excipient interactions and mAb developability. Supplementary Material Supplementary_information_2026_05_233.docx