4S-fluorination of ProB29 in insulin lispro slows fibril formation

Recombinant insulin is a life-saving therapeutic for millions of patients affected by diabetes mellitus. Standard mutagenesis has led to insulin variants with improved control of blood glucose; for instance, the fast-acting insulin lispro contains two point mutations that suppress dimer formation and expedite absorption. However, insulins undergo irreversible denaturation, a process accelerated for the insulin monomer. Here we replace ProB29 of insulin lispro with 4R-fluoroproline, 4S-fluoroproline, and 4,4-difluoroproline. All three fluorinated lispro variants reduce blood glucose in diabetic mice, exhibit similar secondary structure as measured by CD, and rapidly dissociate from the zinc- and resorcinol-bound hexamer upon dilution. Notably, however, we find that 4S-fluorination of ProB29 delays the formation of undesired insulin fibrils that can accumulate at the injection site in vivo and can complicate insulin production and storage. These results demonstrate how subtle molecular changes achieved through non-canonical amino acid mutagenesis can improve the stability of protein therapeutics.


Enzymes:
Gibson Assembly enzymes were purchased as the Repliqa HiFi assembly mix from Quantabio.Trypsin was purchased from MilliporeSigma.Carboxypeptidase B was purchased from Worthington Biochemical.Glu-C peptidase was purchased from Promega.

Strains and plasmids:
The proline-auxotrophic E. coli strain CAG18515 was obtained from the Coli Genetic Stock Center (CGSC) at Yale University.Strain DH10B was used for all cloning applications.
The plasmid pQE80_H27R-PI-KP_proS contains an IPTG-inducible proinsulin-lispro gene and an E. coli prolyl-tRNA synthetase gene controlled by its endogenous promoter.Proinsulin is translationally fused to an N-terminal leader peptide (H27R) that increases expression yields, (42) and a 10x-his tag to facilitate proinsulin enrichment after refolding.The ProB28-LysB29 inversion present in insulin lispro was installed in plasmid pQE80_H27R-PI_proS.(25) We used a two-part Gibson Assembly approach with the two sets of primers described below.The two overlap regions were at the site of mutation, and within the ampicillin resistance gene selection marker.Correct installation of the desired mutation was verified by Sanger sequencing.

Primers:
DNA oligos were purchased from Integrated DNA Technologies (IDT).Nucleotides responsible for the proline-lysine inversion are underlined.All lispro variants were expressed as the corresponding proinsulins using strain CAG18515/pQE80_H27R-PI-KP_proS.This is a proline auxotrophic strain of E. coli which carries a plasmid for prolyl-tRNA synthetase over-expression and inducible expression of proinsulin-lispro.
When growth reached mid-exponential phase (OD600 ~0.8), a medium shift was performed: cells were pelleted by centrifugation (5 kg, 5 min, 4°C) and washed twice with ice-cold 0.9% NaCl.Cells were resuspended in 1 L of 1.25x AMM -Pro, a 1.25x concentrated form of AMM that omits proline.To deplete residual proline, cells were incubated for 30 min at 37°C.A 250 mL volume of a solution containing 2.5 mM ncPro and 1.5 M NaCl was then added (0.5 mM ncPro and 0.3 M NaCl working concentrations).For the incorporation of 44-diF, the concentration of NaCl was 2.5 M (0.5 M working concentration).After 30 min of incubation at 37°C for ncPro uptake, proinsulin expression was induced by the addition of isopropylthio--galactosidase (IPTG, 1 mM).Cultures were incubated overnight at 37°C; cells were then harvested by centrifugation and stored at -80°C until further processing.
Proline-containing proinsulins were expressed in rich medium: proinsulin-lispro was expressed using strain CAG18515 harboring plasmid pQE80-H27R-PI-KP_proS in 6 L (as 6 x 1.0 L cultures) of Terrific Broth (TB).IPTG (1 mM) was added at mid-log phase (OD600 ~0.8) to induce proinsulin expression.Cultures were incubated at 37°C for 3 h, after which cells were harvested by centrifugation and stored at -80°C until further processing.
Proinsulin refolding: Cell pellets were warmed from -80°C to room temperature, then resuspended in 5 mL IB buffer (50 mM tris, 100 mM NaCl, 1 mM EDTA, pH 8.0) per gram cell pellet.Lysozyme (1 mg L -1 ) and phenylmethylsulfonyl fluoride (PMSF, 1 mM) were added, and the slurry was placed on ice for 30 min.Cells were lysed by sonication, the lysate was centrifuged (14,000 g, 30 min, 4°C), and the soluble fraction was discarded.The pellet was washed twice with IB buffer + 1% Triton X-100, once with IB buffer, and once with water; this final step required extended centrifugation (14,000 g, 45 min).A minimal amount of water was used to resuspend the washed inclusion body pellet, and the mass of proinsulin in the inclusion body pellet was estimated by SDS-PAGE.
To prepare for proinsulin refolding, the proinsulin concentration was adjusted to 1 mg proinsulin per L total slurry by resuspending the inclusion body in 3 M urea and 10 mM cysteine in water.
The pH was adjusted to 12 and sample stirred for 1 h at room temperature to dissolve proinsulin.At this stage, ncPro incorporation was assessed by MALDI-TOF, as described in the section entitled "MALDI-TOF MS" below.The solubilized proinsulin solution was diluted ten-fold into refolding buffer (10 mM N-cyclohexyl-3-aminopropanesulfonic acid, CAPS; pH 10.6) that had been pre-cooled to 4°C.The pH of the refolding solution was adjusted to 10.7 and the sample stored at 4°C; throughout the refolding process, the solution pH was periodically adjusted so that it remained between 10.6 and 10.8.Proinsulin refolding progress was monitored by reversephase HPLC, and usually reached completion within 50 h.
Proinsulin was enriched from the refolding solution after adjusting the pH to 8.0 and incubating the sample overnight with Ni-NTA resin and 10 mM imidazole.The resin was washed with wash buffer (25 mM imidazole in PBS, pH 8.0), and proinsulin was eluted with elution buffer (250 mM imidazole in PBS, pH 8.0).Fractions that contained proinsulin were combined and dialyzed extensively against 10 mM sodium phosphate, pH 8.0.

Lispro maturation and purification:
Refolded and dialyzed proinsulin was warmed to 37°C and digested with trypsin (20 U mL -1 ) and carboxypeptidase-B (10 U mL -1 ) at 37°C for 2.5 h to remove the N-terminal tag and C-chain.The pH was adjusted to ~3 with 6 N HCl to halt digestion.
Lispro variants were immediately purified after proteolysis by reverse-phase HPLC on a C4 column (Penomenex Jupiter 5 µm particle size, 300 Å pore size, 250x10 mm) using 0.1% trifluoroacetic acid (TFA) in water (solvent A) and 0.1% TFA in acetonitrile (solvent B) as mobile phases.A gradient of 25-32% solvent B was applied over 65 min, and fractions that contained lispro were collected.Aliquots were removed at this stage for purity analysis; the remaining portion of each lispro-containing fraction was lyophilized.Each fraction was analyzed by analytical reverse-phase HPLC, MALDI-TOF MS (Figure 1h-k), and SDS-PAGE (Figure S1) to verify sample quality and ensure ≥95% purity for all downstream analyses.Lyophilized powders were stored at -20°C until further use.

Circular dichroism spectroscopy:
Equilibrium measurements: The circular dichroism spectra of insulin or lispro samples (60 µM in 100 mM sodium phosphate, pH 8.0) were measured at 25°C in 1 mm quartz cuvettes using a step size of 0.5 nm and averaging time of 1 s on an Aviv Model 430 Circular Dichroism Spectrophotometer.A reference buffer spectrum was subtracted from each sample spectrum.
Kinetic measurements: Insulin or lispro samples in 100 mM sodium phosphate buffer pH 8.0 were dialyzed overnight against 28.6 mM tris buffer, pH 8.0 (Slide-A-Lyzer dialysis cassettes, 3.5 kDa MWCO, ThermoFisher).Samples were formulated as follows: 600 µM lispro, 250 µM ZnCl2, 25 mM resorcinol, 25 mM tris buffer, pH 8. A 20 µL volume of the insulin formulation was injected into a stirred buffer solution containing 2.98 mL of 25 mM tris, pH 8.0 in a 10 mm quartz cuvette (150-fold dilution).Ellipticity was monitored at 222 nm over 120 s (1 s kinetic interval, 0.5 s time constant) at 25°C.A typical run led to a rapid drop in CD signal as mixing occurred (~5 s), then a gradual rise to an equilibrium ellipticity representative of an insulin monomer.Data preceding the timepoint with the greatest negative ellipticity represented the mixing time; these data were omitted from further analysis.Runs were discarded if the maximum change in mean residue ellipticity from equilibrium did not exceed 750 deg cm 2 dmol -1 , which indicated poor mixing.The remaining data were fit to a mono-exponential function using Scipy (Python); data presented here are from at least two separate HPLC fractions, measured on two different days.
An equilibrium spectrum for each protein was obtained after dilution; all spectra were indicative of an insulin monomer (Figure S3).The CD spectrum of lispro under pre-dilution formulation conditions was obtained using a 0.1 mm quartz cuvette.A blank spectrum containing all buffers and ligands was subtracted from the sample spectrum.

Analytical ultracentrifugation:
Lispro variants were formulated at 140-206 µM in 100 mM phosphate buffer, pH 8.0.Velocity sedimentation experiments were performed at the Canadian Center for Hydrodynamics at the University of Lethbridge using absorbance optics.High concentrations samples were measured at 237 nm, intermediate concentration samples at 232 nm, and low concentration samples at 225 nm.All samples were measured at 60,000 RPM and 20°C in standard Beckman Coulter cell housings fitted with 0.3 cm epon-charcoal centerpieces for the high concentration samples, and 1.2 cm epon-charcoal centerpieces for the intermediate and low concentration samples.All cells were fitted with sapphire windows.Data were analyzed with UltraScan III version 4.0 release 6606.(38) Velocity data were initially fitted with the two-dimensional spectrum analysis(39) to determine meniscus position and time-and radially-invariant noise, and to generate molecular weight distributions.2DSA results were refined using the genetic algorithm (GA) approach.(40) Sedimentation and diffusion coefficients derived from the GA analysis were transformed to molar mass distributions (see Figure S2), assuming a partial specific volume of 0.7248 ml/g for all lispro variants.The enhanced van Holde-Weischet analysis (41) was used to determine diffusioncorrected sedimentation coefficient distributions.For the Kd analysis, AUC data were fitted with a discrete model genetic algorithm (DMGA) approach,(29) assuming the known molar mass of the monomer, and floating the Kd, koff, total concentration, partial specific volumes, and frictional ratios as reported in Table S3. Fibrillation: Lispro samples (60 µM in 100 mM sodium phosphate, pH 8.0) were centrifuged at 22,000 g for 1 h at 4°C, before 1 µM thioflavin T (ThT) was added.Each lispro (200 µL) was added to a 96-well, black, clear bottom plate (Greiner Bio-One) and sealed.Samples were shaken continuously at 960 rpm on a Varioskan multimode plate reader at 37°C, and fluorescence readings were recorded every 15 min (444 nm excitation, 485 nm emission).Fibrillation runs were performed on at least two separate HPLC fractions, each in triplicate or quadruplicate, and on two different days.The growth phase of each fibrillation replicate was fit to a linear function; fibrillation lag times were reported as the x-intercept of this fit.Fibril samples were stored at 4°C until analysis by transmission electron microscopy (TEM).

ANS fluorescence:
Lispro variants (1 µM) were mixed with 5 µM ANS in 100 mM phosphate buffer, pH 8.0.Fluorescence emission spectra were measured in 1 cm quartz cuvettes at ambient temperature using a PTI QuantaMaster fluorescence spectrofluorometer.A 2 nm s -1 scan rate and 350 nm excitation wavelength were used.Measurements for each variant were performed in triplicate from three separate HPLC fractions.

Figure S4
. Changes in CD signal after dilution are not due to protein denaturation.At 60 µM, human insulin is expected to exist as a dimer at pH 8, as a monomer in 20% ethanol, and in denatured form in 8 M guanidinium chloride.These spectra are overlaid with equilibrium spectra collected before and after lispro dilution for kinetic CD measurements.Spectra below 210-215 nm were omitted for some samples due to high levels of buffer absorbance at these wavelengths.

Figure S2 .
Figure S2.Effects of varying doses of lispro on the blood glucose levels in diabetic mice.Commercial lispro (Humalog, Eli Lilly) was injected subcutaneously in diabetic mice at 5, 10, and 100% of the standard dose (35 mg kg -1 ), and blood glucose was measured over time.The data presented here represent the mean ± standard deviation of 4 biological replicates.

Figure S3 .
Figure S3.Diffusion-corrected van Holde -Weischet integral sedimentation coefficient distributions.Lispro variants were formulated in 100 mM phosphate, pH 8.0 at the indicated concentrations, and analyzed by sedimentation velocity analytical ultracentrifugation.

Table S3 . Summary of lispro variant hydrodynamic and thermodynamic characterization by analytical ultracentrifugation.
Values in brackets represent the 95% confidence interval.s, sedimentation coefficient; D, diffusion coefficient; PSV, partial specific volume.