Higher Order Structure Differences Among Insulin Crystalline Drugs Revealed by 2D heteronuclear NMR

Abstract

During therapeutic protein development, two-dimensional (2D) heteronuclear NMR spectra can be a powerful analytical method for measuring protein higher order structure (HOS) in solution since the spectra exhibit much higher resolution than homonuclear ¹H spectra. However, 2D NMR capabilities for characterizing protein HOS in crystalline states remain to be assessed, given the low ¹³C natural abundance and intrinsically broader lines in solid-state NMR (SSNMR). Herein, high-resolution heteronuclear correlation (HETCOR) SSNMR was utilized to directly measure intact crystal drug products of insulin human, insulin analogs of insulin lispro and insulin aspart. The fingerprint regions in 2D ¹H– ¹³C HETCOR spectra were identified, which distinguished the insulin crystals in their primary structure, HOS heterogeneity and dynamics, as well as the manufacturing processes. The HOS heterogeneity in insulin analogs is consistent with their therapeutic effect of rapid action; while insulin human crystals showed more structural homogeneity, consistent with their slower pharmacokinetics (PK) peak time than insulin analogs. Therefore, heteronuclear NMR could be broadly applicable to study protein drug dosage forms from liquid to solid, yielding improved molecular level structure data for assessing drug HOS in biosimilar drug development.

Introduction

The 51-amino acid protein insulin is composed of two peptide chains, A and B, covalently linked through two disulfide bonds (Figure 1A). Insulin folds into higher order structure (HOS) and forms dimers, hexamers, dodecamers and other multimers, in drug formulation depending on insulin concentration and the presence of zinc ion and phenol (Figure 1B).^[1] Maintaining protein HOS in formulation of either liquid or crystal state is often critical for drug efficacy and safety.^[2] Further, insulin therapeutics can have different analogs with modified amino acid sequences to achieve varied pharmacokinetic profiles through modulating oligomerization.^[3] As an alternative to amino acid mutations, intermediate-release insulin was developed by co-crystalizing fast-acting insulin human or rapid-acting insulin analogs (i.e., insulin lispro or insulin aspart) with protamine, forming Neutral Protamine Hagedorn (NPH) insulin.^[4] Protamine is a biologically sourced mixture of arginine rich positively charged peptides with an average molecule weight of 5 kDa.^[5] NPH insulin has the advantage of conveniently mixing with fast/rapid acting insulin, both of which have similar formulation pH, to allow a single injection with sustained release.^[6] Recently, NPH insulin was suggested as the initial basal insulin therapy instead of long-acting insulins, due to the lower cost.^[7] Development of NPH insulin biosimilars, or crystalline protein therapeutics typically involves structural characterization because local structure element can vary (e.g., helix vs. coil) if the crystallization process undergoes minor changes.^[8] Therefore, an analytical method sensitive to small changes in HOS of crystalline samples is necessary for assessing product quality and process control. SSNMR has been demonstrated to be capable of obtaining ¹H detected 2D spectrum for insulin human crystals and has revealed structure details previously,^[9] however, the extent of its ability to differentiate the HOS variation among different insulin drugs remains unknown.

Figure 1.

(A) Sequences of insulin human, lispro, and aspart. The A chains are the same for all, while mutated residues on the B-chain are indicated in black. Disulfide bonds are represented by dashed lines. (B) Schematic depiction of insulin monomer, dimer, and hexamer with orange and green representing chain A and B, respectively. (from pdb 4F1C)^[10]

In this study, we investigated four commercial combination insulin drug products, namely Humulin 70/30,^[11] Novolin 70/30,^[12] Humalog 75/25,^[13] and Novolog 70/30^[14] (Table 1). They are insulins composed of 70% or 75% of insulin NPH crystal suspension, and 30% or 25% of fast-acting/rapid-acting insulins in aqueous phase. The insulins in the soluble form were characterized using solution NMR, while the evaluation of the HOS of crystalline insulin was performed using SSNMR spectroscopy. The structural information of insulin drug products could facilitate the development of next generation basal insulin drug products, and the SSNMR approach presented here will benefit drug development research for crystalline proteins.

Table 1.

Insulin Drug Products used in the study.

Drug Product	Drug Substance	Solution NMR 1H T1/T2	SSNMR 13C T2 (ms)	SSNMR # of peaks[a]
Humulin	Insulin	49 (± 3)	3.58	50
70/30	human		(± 0.09)
Novolin	Insuzlin	52 (± 2)	3.69	44
70/30	human		(± 0.09)
Humalog	Insulin	44 (± 4)	3.05	51
75/25	lispro		(± 0.09)
Novolog	Insulin	43 (± 7)	3.25	42
70/30	aspart		(± 0.09)

^[a]The detection region is 45–68 ppm of ¹³C and 1–7 ppm of ¹H. The signal detecting threshold is set to 4σ.

Results and Discussion

Solution NMR is a natural method of choice to study protein structure and dynamics in formulation.^[15] However, the limited quantity of soluble insulin present in the aqueous phase of combination insulin drugs is a challenge. As an alternative approach, we conducted solution NMR experiments directly on liquid state insulin drug products, namely Humulin R, Novolin R, Humalog, and Novolog, all of which share the same formulation composition as the combination insulin drug products but with a higher insulin concentration in the aqueous phase (Tables S1 and S2). The 1D ¹H spectra of Humulin R and Novolin R showed similarity in peak positions, which are substantially different from Humalog and Novolog (Figure S1). The distinction could be due to variations in amino acid sequence and/or oligomerization. However, the peak overlap in ¹H 1D spectra prevented us from obtaining more detailed HOS information. Next, ¹H-¹³C HSQC 2D spectra of the side chain methyl region were used to highlight the HOS differences (Figure 2). Notably, the HSQC spectra of Humulin R and Novolin R have remarkably similar chemical shifts, implying a conformational commonality between these two insulin human products despite being manufactured and formulated by different firms. In contrast, the spectra of Humalog and Novolog differ significantly in peak positions from those of insulin human. Even though the mutations are situated at the C-terminus of the B-chain (Figure 1A), extensive chemical shift perturbations were observed for numerous resonance peaks, e.g., A2-Ile, A10-Ile, and many leucine residues, indicating that the mutations exert an allosteric effect on the insulin structure in addition to localized disturbances. Additional T₁/T₂ (longitudinal relaxation time constant/transverse relaxation time constant) ¹H relaxation studies were performed. The results showed T₁/T₂ values of 49 and 52 in Humulin R and Novolin R, respectively, which are higher than T₁/T₂ values of 44 in Humalog and 43 in Novolog (Table 1), suggesting more oligomerization in the insulin human drug products.^[16] The lower amount of oligomerization in the insulin analog drugs is consistent with their therapeutic rapid action.

Figure 2.

The ¹H-¹³C 2D HSQC spectra of liquid formulations of Humulin R (blue), Novolin R (green), Humalog (red), and Novolog (yellow). The assignment is shown in the spectrum of Humulin R.^[15] The HSQC spectra of Humulin R in gray is overlayed with the other spectra for comparison.

Next, SSNMR was performed to characterize the majority (70–75%) of insulin in crystals. The crystal size of the four insulin products are not known, but a size of 30×7.5×5 micron of NPH crystal was reported before.^[17] With the recent advancements in ultra-fast MAS spinning for SSNMR, ¹H-detection NMR experiments can be conducted at spinning rates beyond 100 kHz using 0.7 mm or 0.75 mm rotors, providing remarkable ¹H resolution. ^[18] However, the application of isotope labeling for commercial drug products is impractical, posing a significant sensitivity challenge. In comparison to 0.7 mm rotors, a 1.6 mm rotor can accommodate approximately 12–16 times more sample by volume, with a theoretical sensitivity gain of 6 times.^[18d] The larger rotor capacity allows the collection of 2D spectra for proteins at natural abundance ¹³C (1.1%) within a reasonable timeframe. Therefore, we opted to utilize 1.6 mm rotors at a 40 kHz MAS rate to characterize the NPH insulin crystals in the drug products.

The ¹³C cross-polarization (CP) 1D spectra of insulin NPH crystals confirmed the overall structural likeness of the four insulin drug products but also identified structural heterogeneity (Figure 3A). A consistent spectral pattern with similar spectral intensity was observed across all ¹³C CP 1D spectra, suggesting a uniform overall structure and a consistent crystal sample handling among the four samples. However, the bulk ¹³C T₂ of insulin human (3.6–3.7 ms), average from all ¹³C sites, were longer than insulin analogs (3.1–3.3 ms) (Table 1), which led to narrower ¹³C linewidth and stronger signals in the ¹³CCP 1D spectra of insulin human (see Figure 3B, Figures S2–S5 for overlapped comparison). The average ¹³C linewidths, estimated from bulk T₂, are 88 Hz for insulin human and 106 Hz for insulin analogs. This finding implies that the structure of insulin human NPH crystals is more homogeneous compared to the insulin analogs. However, the ¹³C linewidth of these insulin drugs are wider than the pembrolizumab crystal that had a reported ¹³C linewidth of approximately 29 Hz.^[19] It is also noteworthy that the insulin NPH crystals from Novolin 70/30 and Humulin 70/30 had similar ¹³C T₂s, indicating an overall structural and dynamic similarity between the two insulin human formulations prepared by different manufacturers.

Figure 3.

The SSNMR ¹³C CPMAS 1D spectra of the insulin NPH crystal drug products for all the four drug products studied (A) and the overlapped comparison of CO region (B, left) and aliphatic region (B, right).

In parallel, the ¹³C–¹H HETCOR 2D spectra were collected for four insulin crystals (Figure 4A). The ¹H linewidth in these 2D spectra is approximately 0.8 ppm, which is comparable to the reported data acquired using smaller rotors at 60 kHz MAS spinning.^[18c] In contrast to the 1D ¹³C spectra, which primarily illustrate the overall structural similarity of the insulin NPH crystals, the 2D spectra revealed more detailed HOS distinctions among the four insulin NPH drug products (Figure 4A). First, the isoleucine side chain region in the ¹³C–¹H HETCOR 2D spectra provided site-specific insights for conformation assessment, as the resonance peaks were well isolated from other signals (Figure 4B). Two isoleucine residues located in the A chain of insulin, Ile2(A) and Ile10(A), were observed in the ¹³C–¹H HETCOR 2D spectra, Shown in Figure 4B, most of the resonance peaks of the isoleucine side chains of Novolin 70/30, i.e., Ile2(A)δ Ile10(A)δ/γ, were split into one major and one minor peak, indicating the coexistence of different conformations; in contrast, only a single set of strong Ile2(A) and Ile10(A) peaks were observed in the 2D spectrum of Humulin 70/30, suggesting the prevalence of a dominant conformation within its crystal. Studies of the X-ray crystallography on NPH crystals have disclosed the existence of R₆^[20] zinc insulin hexamers with bound phenolic ligands.^[17] Notably, a prior investigation revealed that the insulin hexamer undergoes a structure transition from T₃R₃^f[21] conformation in the precursor crystal to a T₆ conformation in the drug microcrystal, differentiated by the helix/coil structure in the N-terminal portion of B chain.^[22] Therefore, the observation of a coexistence of two or more conformations in the NPH drug formulation was not surprising. In the NPH crystals of Novolin 70/30, it is possible that the insulin exists as a mixture of these states, whereas in NPH crystals of Humulin 70/30, insulin predominately adopts one of these states. The different states of the insulin hexamer could be driven by the concentrations of Zn²⁺ and m-cresol/phenol. However, per formulation component table (Table S2), similar quantities of excipients were used, and a confident correlation can’t be made to predict hexametric states of T or R purely based on excipient concentrations.

Figure 4.

(A) The SSNMR ¹³C –¹H HETCOR 2D spectra of insulin NPH crystal products. The regions of the isoleucine side chain and threonine side chain are highlighted by the boxes and shown in (B) and (C), respectively. The dashed-line boxes highlight the extra peaks of Humulin 70/30 and Humalog 75/25. The signal floor is set to 4 times the noise level, and the spectra are presented at comparable contour levels. The peaks in (B) and (C) were assigned based on solution NMR data of the R6 human insulin hexamer.^[1a]

For the insulin analogs, Novolog 70/30 exhibited a singular set of weak peaks with peaks distributed similar to Humulin 70/30, while splitting was observed for the weak Ile peaks in Humalog 75/25 (Figure 4B). Of note, the crystal structure of T₃R₃ was reported for insulin aspart,^[23] the active ingredient of Novolog 70/30. However, the weaker peak intensity (Figure S6), likely attributed to less ordered structure and local dynamics in insulin analogs, introduced ambiguity regarding a conclusive determination of the presence of multiple conformations for the analogs.

Another fingerprint region in the ¹³C–¹H HETCOR 2D spectra is that of the threonine side chains (Figure 4C). Insulin contains three threonine residues: Thr8(A), Thr27(B), and Thr30(B). Two of these threonine residues are located at the C-terminus of the B-chain, which is in proximity to the mutation sites of the insulin analogs. In the ¹³C –¹H HETCOR 2D spectra of Humulin 70/30 and Novolin 70/30, the C_β/H_β resonance peaks of all three threonine residues were identical, indicating the formation of a rigid and well-ordered structure at the C-terminus of the B-chain of insulin human. However, upon replacing Pro28(B) with aspartic acid (Novolog) or altering the position of Pro28(B) and Lys29(B) (Humalog), the C_β/H_β resonance peaks of Thr27(B) and Thr30(B) were attenuated and even disappeared from the spectra. This observation implies a decrease in the conformational stability of the C-terminal residues, which introduces a significant structural disorder and/or dynamics within the C-terminus of the B-chain in the insulin analogs. The J-coupling-based rINEPT 2D experiment was applied to capture the flexible portion in the insulin NPH crystals (Figure S7). However, no signal from the C-terminus of the B-chain was observed in the rINEPT 2D spectra, indicating that the C-termini did not undergo overall fast motions. Therefore, the signal disappearance of residues near the mutation sites are likely due to structural heterogeneity. This finding aligns with the ¹³C T₂ relaxation results among the four drugs (Table 1), which suggested that the mutation induces structural disorder. The C-terminus of the B-chain was recognized for its crucial roles in stabilizing dimers and subsequent hexamer formation.^[24] Moreover, the B-chain must be detached from the central B9-B19 α-helix upon insulin receptor binding to expose buried A1–A3 and A19 residues, which are important for insulin receptor (IR) interaction.^[25] Therefore, one hypothesis is that the induced disorder in the C-terminus of the B-chain, resulting from mutations, favors the disassembly of the hexamer into monomer and promotes IR binding, thus achieving the rapid absorption of insulin in vivo. The SSNMR result indicated that, despite co-precipitation with protamine, the C-terminus of the B-chain in insulin analogs maintained a state of structural disorder in the drug products, hence preserving a structure associated with rapid-acting insulin.

The peak counting analysis shows that Humulin 70/30 and Humalog 75/25 exhibit a greater number of resonance peaks at the C_α/H_α region compared to Novolin 70/30 and Novolog 70/30 (Table 1). The extra number of peaks was attributed to the presence of resonances in the 55–60 ppm range of the ¹³C dimension and 1–3 ppm range of the ¹H dimension for Humulin 70/30 and Humalog 75/25, as highlighted within the dashed boxes in Figure 4A. These resonance peaks fell outside the typical ¹³C/¹H range of proteins, also not observed in solution state 2D ¹H–¹³C HSQC spectra. Given the absence of such peaks in the 2D spectra of Novolin 70/30 and Novolog 70/30, these resonances may be linked to process related components introduced in manufacturing processes of Humulin 70/30 and Humalog 75/25. Therefore, the ¹³C–¹H HETCOR 2D experiment, which is capable of identifying the conformational variation and the presence of non-insulin components, can be valuable as a process control approach in crystalline protein therapeutics development. It is worth mentioning that alternations in crystallization patterns and formulation components have been reported to potentially influence the action duration of insulin.^[26] However, there is currently no evidence indicating significant differences in the drug activity or duration of action between Novolin 70/30 and Humulin 70/30.

An unresolved question regarding the insulin-protamine complex is the mode of protamine binding in insulin NPH crystals. Structural studies from X-ray crystallography show that the electron density for protamine was diffusive or absent, even in co-crystals where the insulin hexamers otherwise display high resolution.^[17] This suggests that protamine is likely to be structurally disordered or flexible within the NPH crystals. In the rINEPT 2D spectra (Figure S7), no signal from any protein was detected, indicating that protamine does not undergo submicrosecond random motion upon forming crystals with insulin. The lack of observable protein signals is consistent with protamine being tightly bound and lacking mobility but also structurally disordered when forming NPH crystals with insulin. Electrostatic interactions have been demonstrated to play an important role in protamine-mediated precipitation.^[27] Therefore, non-specific interactions may contribute to the observed disorder in the protamine structure within NPH crystals. It is also noteworthy that while signals from glycerin were present in all of the rINEPT spectra, signals from phenol and m-cresol were only observed in the Novolog 70/30 and Humalog 75/25 spectra, suggesting the presence of a higher abundance of free phenolic ligand in the analogs, consistent with their higher formulation content (Table S2).

Conclusions

In summary, high-resolution 1D and 2D NMR data were collected on four combination insulin drug products, revealing notable HOS distinctions between different manufacturing processes. The brand-to-brand differences in HOS identified here is assumed to be more than batch to batch difference, which was demonstrated to be less than inter-brand differences.^[15,28] The solution NMR results on the soluble insulin indicated similar structures of insulin human prepared by different manufacturers. The mutations in insulin analogs allosterically affected their conformations, thus leading to less oligomerization and associated changes in the 2D spectra. For the insulin NPH crystals, the ¹³C 1D SSNMR spectra and relaxation data showed that the insulin human crystal was more structurally ordered compared to insulin analog crystals. Further, the ¹³C–¹H HETCOR 2D spectra highlighted HOS and chemical species differences between four insulin drug products. The differences in peak count and the side chain resonances underscored the impact of the manufacturing process and sequence mutation on crystal HOS differences. While the observed HOS difference may not affect clinical efficacy significantly, the availability of analytical tools and data could ensure protein drug quality once biosimilar, or manufacture change effects are assessed. In conclusion, after solution heteronuclear NMR has been applied to liquid protein drugs providing structural information at atomic resolution,^[15,29] the application of SSNMR in protein crystal drug^[19,28,30] provided additional details of HOS in the solid state due to the sensitivity to protein crystal structure and dynamics. We envision that the SSNMR approach presented in this study can be extended to other therapeutic peptide and protein drugs in the solid state, serving as a potential tool for both process monitoring and quality assessment.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.