Characterization of Critical Quality Attributes of an Anti-PCSK9 Monoclonal Antibody

Abstract

During early development of biopharmaceuticals, suboptimal producing clones and production conditions can result in limited quantities of high-purity products. Here we describe a systematic approach, which requires minimal amounts of protein (~10 mg) to assess critical quality attributes of a monoclonal antibody (mAb). A commercial anti-PCSK9 IgG2 (evolocumab, Repatha^®) and an early-stage biosimilar candidate were compared head-to-head using a range of high-throughput physicochemical and in-vitro binding analytical methods. Overall, both mAbs were shown to be highly pure and primarily monomeric, to share an identical primary structure, and to have similar higher-order structural integrity, apparent solubility, aggregation propensity, and physical stability profiles under temperature and pH stress conditions. Low levels of dimers were detected for the innovator (1.2%) and the biosimilar candidate mAb (0.3%), which also presented fragments (1.2%). Regarding charge heterogeneity, the amount of the main charge isoform was 53.6% for the innovator and 61.6% for the biosimilar candidate mAb. Acidic species were 38% for the innovator and 30% for the biosimilar candidate. Variations in the relative content of a few N-glycan species were found. The in-vitro binding affinity to PCSK9 was monitored, and no differences were detected. The mathematical approach called “error spectral difference” (ESD), proposed herein, enabled a quantitative comparison of the biophysical datasets. The workflow used in the present work to characterize CQAs at early stages is helpful in supporting the development of biosimilar mAb candidates.

1. Introduction

Monoclonal antibodies (mAbs) play a critical role in the biopharmaceutical sector, dominating the overall product approvals (including biosimilars) and market value [1]. While most of these complex glycoproteins are indicated for treatment of cancer and autoimmune disorders, novel mAb products having different targets have been approved [1]. In 2015, two fully human mAbs were approved for reduction of low-density lipoprotein cholesterol (LDL-C): evolocumab (Repatha^®, Amgen) and alirocumab (Praluent^®, Sanofi/Regeneron). Both mAbs act by inhibiting proprotein convertase subtilisin kexin type 9 (PCSK9), an enzyme that regulates LDL metabolism through LDL receptor (LDLR) degradation. PCSK9 inhibition prevents LDLR degradation, lowering LDL-C levels by 50–60% and reducing the incidence of cardiovascular events [2]. Elevated plasma levels of LDL-C are a significant risk factor for cardiovascular diseases, which are the leading cause of death worldwide. Currently, treatment with 3-hydroxy-3-methylglutaryl coenzyme A reductase inhibitors (statins) remains the cornerstone of therapy for patients with hypercholesterolemia. There are, however, limited therapeutic options for patients who are intolerant to statin therapy (more than 20%) [2]. or have severe familial hypercholesterolemia. For these patients, PCSK9 inhibition has been a therapeutic breakthrough in the management of lipid disorders and cardiovascular risk [2,3].

Given the potential of anti-PCSK9 therapies, our group has been investigating the stable expression of an anti-PCSK9 mAb in CHO cells [4], using Repatha® (evolocumab, Amgen) as a reference product. In order to allow extensive mAb characterization despite the early-stage limited availability of high-purity protein, here we employ high-throughput methods for a head-to-head comparison of the anti-PCSK9 mAb and its reference product using minimal protein quantities. This approach is helpful in guiding clone selection and optimization of upstream and downstream processes and can support future scale-up efforts. Biological drugs are complex and sensitive to variations in their multi-step manufacturing process (expression system, cell growth conditions, purification process, formulation, and storage), which can give rise to product heterogeneity. Currently, there is no harmonization of the regulatory approval pathway for biosimilars among different regulatory agencies, such as the European Medicines Agency (EMA), the US Food and Drug Administration (FDA), and others across the world. Despite regional differences, biosimilars typically have a shorter approval pathway due to reduced clinical trial requirements. A strong demonstration of analytical, non-clinical, and clinical similarity to a reference product is critical for approval [5,6].

Comparative quality studies rely on a broad range of analytical techniques for extensive characterization of protein critical quality attributes (CQAs), such as purity, intact molecular mass, primary structure, higher-order (secondary, tertiary, and quaternary) structure, thermal stability, charge and size heterogeneity, post-translational modifications (glycosylation, disulfide bonds), additional modifications (deamidation, oxidation), impurities profile, as well as biological and immunological properties [7]. Several studies in the literature discuss analytical assessments of different biosimilar mAbs or mAb-related products, such as rituximab [8,9], adalimumab [10,11], infliximab [12,13], and etanercept [14]. Detailed analytical characterization is mandatory during different steps of product development, such as early development, stability studies, clinical trials, licensure of the product, and post-marketing changes [5,15].

Here, we use high-throughput methods to perform a head-to-head characterization of an anti-PCSK9 mAb with minimal protein amount requirements (~10 mg). An in-house-produced biosimilar candidate and its reference product (Repatha^®) were extensively characterized in terms of purity, primary structure, size and charge heterogeneity, apparent solubility, overall biophysical stability, and in-vitro binding to PCSK9. We propose the mathematical method “error spectral difference” (ESD) as a tool to enable a quantitative comparison of biophysical datasets. Our results provide important information about CQAs of evolocumab and its biosimilar candidate that could guide process optimization for biosimilar mAbs.

2. Materials and Methods

2.1. Anti-PCSK9 Monoclonal Antibody Expression and Purification

A suspension-adapted, stably transfected CHO cell pool previously generated and evaluated at very small scale [4]. was used for production of the anti-PCSK9 biosimilar candidate mAb. Batch cultures of this cell pool were performed in 500-mL Erlenmeyer flasks with 300-mL working volume at 37 °C, 5% CO₂, in an orbital shaker (5-cm stroke) at 180 rpm. Cells were grown in the customized, chemically defined TC-LECC culture medium (Sartorius Xell, Schloß Holte-Stukenbrock, Germany), supplemented with L-glutamine, insulin-like growth factor (IGF), hypoxanthine, and thymidine (HT). After each batch, the cell suspension was clarified by centrifugation (1000× g for 10 min) followed by 0.22 µm membrane filtration. The filtered supernatants collected from different batches were pooled and stored at 4 °C until purification by protein-A affinity chromatography. Purification runs were performed using an ÄKTA Purifier system (GE Healthcare, Uppsala, Sweden) with online measurement of pH, conductivity, and UV absorbance at 280 nm. Briefly, for each purification batch, 2 L of pooled clarified material was directly loaded at 10 mL/min onto a 5-mL Eshmuno A^® column (Merck KGaA, Darmstadt, Alemanha) previously equilibrated with 100 mM sodium phosphate, pH 7.4. Protein was eluted by an isocratic gradient elution method using 100 mM sodium phosphate/phosphoric acid, pH 3.0 [16]. The elution fractions of all purification batches were neutralized, pooled, and concentrated for further use.

2.2. Sample Preparation

The purified anti-PCSK9 mAb was produced at COPPE/UFRJ (Brazil) as described above, whereas the reference product Repatha^® (evolocumab, Amgen) was purchased at the Brazilian market as a 140 mg/mL solution for injection in pre-filled pens. For biophysical analysis of both of them, samples were dialyzed overnight (at 4 °C) using Slide-A-Lyzer^TM Dialysis Cassettes (10,000 Da molecular mass cutoff, Thermofisher, Waltham, MA, USA) against 20 mM citrate-phosphate (CP) buffers (pH 3.5–7.5, 1 unit intervals) with ionic strength adjusted to 0.15 M using NaCl. For Raman spectroscopy, mAb samples were concentrated to 10 mg/mL using Amicon^® Ultra centrifugal filter devices (0.5 mL, 10,000 Da molecular mass cutoff, Merck KGaA, Darmstadt, Alemanha). For biochemical studies, mAbs were buffer-exchanged into phosphate-buffered saline (PBS). Protein concentration was measured at 280 nm using a NanoDrop 2000 spectrophotometer (Thermofisher, Waltham, MA, USA) and using a theoretical extinction coefficient ε_0.1% = 1.53 (g/100 mL)⁻¹ cm⁻¹, previously determined in silico with the evolocumab sequence employing the ExPASy Protparam bioinformatics tool [17].

2.3. Mathematical Methods for Quantitative Data Comparison

2.3.1. Statistical Analysis

To statistically compare the biosimilar candidate with the innovator mAb, a Student’s t-test with a 95% confidence level was conducted using Prism software version 8 (GraphPad, San Diego, CA, USA). A p value of <0.05 was considered significant. In some cases, only duplicate measurements could be used due to the limited availability of samples.

2.3.2. Error Spectral Difference (ESD)

In this study, an adaptation of the previously described spectral difference (SD) method [18]. is proposed and referred to as “Error spectral difference—ESD”. It is based on Equation (1), where n refers to the number of data points and y_A, y_B, σ_A, and σ_B refer to the innovator (reference, A) and biosimilar candidate (sample, B) data points (y_i) and their corresponding standard deviations (σ_i). The ESD values calculated for all biophysical assays at 10 or 25 °C were directly compared. The melting curve differences were also calculated. The addition of the error factor ${({\sigma }_{Ai}+{\sigma }_{Bi})}^2$ makes ESD dimensionless.

$$\mathrm{E}\mathrm{S}\mathrm{D}=\sqrt{\sum _{i=1}^n\left[{\displaystyle \frac{\left(\frac{1}{n}\right){\left(y_{Ai}-y_{Bi}\right)}^2}{{({\sigma }_{Ai}+{\sigma }_{Bi})}^2}}\right]}$$

(1)

2.4. Analytical Assays

See supplementary materials for a detailed description of the analytical methods including: (1) sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE); (2) size-exclusion chromatography (SEC-HPLC); (3) liquid chromatography/mass spectrometry (LC/MS) peptide mapping; (4) N-glycosylation profiling by hydrophilic interaction liquid chromatography (HILIC-HPLC); (5) imaged capillary isoelectric focusing (icIEF); (6) micro-polyethylene glycol (PEG) precipitation assay; (7) in-vitro antibody binding kinetics; (8) far-UV circular dichroism (CD) spectroscopy; (9) Raman spectroscopy; (10) intrinsic and extrinsic fluorescence spectroscopy; (11) dynamic light scattering (DLS), and (12) differential scanning calorimetry (DSC).

3. Results

3.1. Purity and Size Distribution

The innovator and biosimilar candidate mAbs were analyzed by SDS-PAGE under reducing and non-reducing conditions to assess protein purity (Figure 1A). Both mAbs were of high purity and had similar migration behavior. Under non-reducing conditions, two major bands (around 150 kDa) were observed for both mAbs, reflecting disulfide isoforms. Evolocumab is an IgG2, and disulfide bond isoforms with comparable biological activities were also reported for the innovator mAb [19]. IgG2 molecules usually have distinct isoforms caused by multiple disulfide bond pairings [20]. The theoretical molecular mass (MM) for intact IgG is expected to be around 150 kDa, which is in accordance with the observed MM for both mAbs. In reduced samples, two protein bands, corresponding to the heavy and light chains, were observed for both mAbs since the intermolecular disulfide bonds present in the hinge region were disrupted.

The formation of aggregates or fragments can reduce bioactivity and increase the immunogenic and toxicological risks of biopharmaceuticals [7]. Thus, SEC-HPLC was used to characterize size heterogeneity by analyzing the distribution of high molecular mass species (HMMS), monomers, and low molecular mass species (LMMS). Representative SEC chromatograms of the innovator and the biosimilar candidate mAb are shown in Figure 1B,C, respectively. The results indicated no significant differences in monomer levels for both mAbs, which were primarily monomeric (>98%) and eluted at ~13 min. Low levels of HMMS, probably arising from soluble mAb dimers, were detected for the innovator mAb (1.2%) and for the biosimilar candidate mAb (0.3% ± 0.1). Additionally, the SEC data also indicated the presence of low levels of LMMS for the biosimilar candidate mAb (1.2%), which are probably related to mAb fragments or residual host-cell proteins. According to the EMA assessment report of Repatha^® [19], the size heterogeneity assessment of the innovator mAb shows that it exists predominantly as monomers, with low levels of dimeric HMMS, and LMMS comprised of free LC-LC and LC structures.

3.2. Peptide Mapping Analysis

According to regulatory agencies, the primary structure of a proposed biosimilar candidate must be identical to the reference product. The amino acid sequence of both mAbs was compared by reverse-phase nano-ultra-performance liquid chromatography/tandem mass spectrometry (RP-nanoUPLC-MS/MS) peptide mapping. As shown in Figure 2, the chromatograms of the trypsin/chymotrypsin digested mAb samples were overall similar in peak intensities and retention times, with minor differences probably due to variations in ionization and different deamidation and N-glycosylation patterns. MS/MS analysis of individual trypsin, trypsin-chymotrysin, trypsin/LysC, and GluC digested peptides confirmed approximately 86% of the amino acid composition of the innovator mAb (see Supplementary Figure S1A). For the biosimilar candidate mAb, the trypsin-chymotrysin digestion analysis achieved ~64% coverage (see Supplementary Figure S1B). The covered sequence confirmed identical amino acid sequences for both mAbs.

3.3. N-Glycosylation Profile

Glycosylation is one of the most complex post-translational modifications in biopharmaceuticals due to its heterogeneity and sensitivity to variations in the production process (cell line, cell culture media, cultivation conditions) [21]. In this study, N-glycan structures of mAb samples were preliminarily identified and quantified by high-performance hydrophilic interaction liquid chromatography (HILIC-HPLC). Representative HILIC chromatograms of the innovator and the biosimilar candidate mAb are shown in Figure 3A,B, respectively. Among all the peaks detected for both mAbs, two of them were unique for each mAb, but with low abundance (<3%). Supplementary Table S1 shows the possible N-glycan structures found according to the GlycoStore [22]. database. For most peaks, there is more than one possible structure, and only a more in-depth study would allow the definitive assignment of the glycoform. However, it is known that the most prevalent glycoforms in human IgGs are the ones usually known as G0, G0F, G1F, G2F, and Man5. Also, it is known that bisected glycoforms are very rare in mAbs expressed by CHO cells. Therefore, it is suggested that peaks 3, 5, 8 + 9, 11, and 6 refer to glycoforms G0, G0F, G1F, G2F, and Man5, respectively. Although there are differences in the relative abundance of possible N-glycan species, in both mAbs, about 95% of them refer to the aforementioned most prevalent glycans in human IgGs.

In the study by Goetze et al. [23], for instance, the N-glycosylation profile of three different IgG2 mAbs, expressed in CHO cells and developed by Amgen (the same supplier of Repatha^®), was analyzed in detail. As shown in Figure 3C, the glycoform distribution presented by our biosimilar candidate showed higher similarity to these three IgG2 mAbs reported in literature than to the reference innovator Repatha^®.

Additionally, the most abundant glycoform for both the innovator and the biosimilar candidate was a fucosylated biantennary complex species (G0F), which is the most abundant glycoform in IgG2 molecules (~60%) [23]. For the biosimilar candidate mAb, N-glycan species that can be potentially immunogenic in humans, such as those containing N-glycolylneuraminic acid (NGNA) or terminal α1,3-galactose (α-gal) [24]., were not observed. On the other hand, for the innovator mAb, a low abundance (1.2% ± 0.2%) of glycans that could possibly contain α-gal was detected, although according to the EMA’s assessment report [19], no potentially immunogenic species were detected during Repatha^® development. Further HILIC- and mass spectrometry-based assays using sequential exoglycosidase digestion can be performed to provide a complete picture of the N-glycosylation profile of both mAbs.

3.4. Charge Heterogeneity

The charge heterogeneity profile of both mAbs was evaluated by imaged capillary isoelectric focusing (icIEF), and representative electropherograms are shown in Figure 4A for the innovator mAb and in Figure 4B for the biosimilar candidate. Both mAbs displayed similar profiles, with four isoforms each. The pI observed for the main charge variant peak in each mAb sample was 7.8–7.9, which agrees with the theoretical pI (7.89) of evolocumab based on its amino acid sequence. However, the biosimilar candidate mAb contained a larger proportion of the main isoform (61.59 ± 2.30%) than the reference product (53.59 ± 1.07%). Two acidic species were detected in both mAb samples, with a pI range of 7.5–7.7 for the innovator and 7.5–7.6 for the biosimilar candidate mAb, representing ~38% of all species for the innovator and ~30% for the biosimilar candidate mAb. A basic variant with pI 8.1 was also observed in small amounts (~8%) for both mAbs. Charge heterogeneities can originate from deamidation, oxidation, aggregation, charged glycans (sialylation), fragmentation, and glycation, among other reasons, and could potentially affect binding affinity and pharmacokinetic and pharmacodynamic (PK/PD) profiles [25]. The same isoforms (acidic, main, and basic) were also reported in the EMA assessment report for the innovator Repatha, where further analysis confirmed the presence of multiple asparagine deamidation sites in the acidic fractions, while in the basic fractions an enrichment of HMMS, methionine oxidation, heavy chain C-terminal lysine variants, and light chain N-terminal truncated variants were detected [19]. In summary, the present results demonstrate that both mAbs present similar pI profiles with some differences in the relative content of isoforms.

3.5. Relative Solubility Analysis

Solubility is an important aspect that can affect the stability and activity of biopharmaceuticals and their formulation development [26], especially for therapeutic mAbs formulated at high concentrations such as evolocumab (140 mg/mL). As shown in Figure 5, PEG precipitation curves were plotted to estimate two parameters: (1) the concentration (% m/v) of PEG required to reduce the protein concentration by 50% of its initial value (% PEG midpoint); (2) the apparent solubility (or apparent thermodynamic activity), estimated from the y intercept (when PEG concentration is zero) found by extrapolation when the PEG solubility curve is plotted on a logarithmic scale. The PEG precipitation curves for both mAbs were highly similar, showing that the amount of both mAbs in solution decreased similarly when the concentration of the molecular crowding agent PEG-10,000 increased. Both the innovator and the biosimilar candidate mAb showed similar %PEGmidpt values (10.76% and 10.57%) and apparent solubilities (log 2.96 and 2.72 mg/mL). No statistically significant differences were observed.

3.6. Higher-Order Structure (HOS)

Far-UV circular dichroism (CD), fluorescence, and Raman spectroscopies were used to provide insights into the overall secondary and tertiary structure of both mAbs. As shown in Supplementary Figure S2A, the innovator and biosimilar candidate mAbs share similar far-UV CD spectra (200–260 nm) at 10 °C. A characteristic minimum peak around 217 nm was observed at all pH values, which is a signature of the predominant β-sheet structure of IgGs. The error spectral difference (ESD) approach was used to compare the spectral differences between both mAbs. The CD spectral differences followed the order pH 7.5 (ESD = 6.9) > pH 4.5 (ESD = 3.7) > pH 3.5 (ESD = 2.6) > pH 5.5 (ESD = 2.2) > pH 6.5 (ESD = 1.4), for all pH conditions at 10 °C. To investigate the effect of temperature on the secondary structure of both mAbs, thermal melting curves were generated by monitoring the mean residual molar ellipticity (MRME) at 217 nm as a function of temperature. Overall, the unfolding curves overlapped for both mAbs (Supplementary Figure S2B). The MRME variations were pH-dependent, with one major thermal unfolding event. The calculated T_m values for the innovator and the biosimilar candidate were similar, as shown in

Table 1. Summary of the thermal unfolding temperatures (T_m) of mAbs assessed by biophysical assays *.

	Tm 1 (°C)			Tm 2 (°C)			Tm 3 (°C)			Tonset (°C)
Analytical Method/pH	Innovator	Biosimilar Candidate	p-Value	Innovator	Biosimilar Candidate	p-Value	Innovator	Biosimilar Candidate	p-Value	Innovator	Biosimilar Candidate	p-Value
CD
pH 3.5	52.63 ± 0.89	53.82 ± 1.16	0.3688	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 4.5	67.01 ± 1.10	66.51 ± 0.11	0.5879	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 5.5	66.23 ± 0.35	66.80 ± 0.81	0.4574	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 6.5	66.86 ± 0.37	67.12 ± 0.35	0.5453	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 7.5	69.13 ± 0.00	69.00 ± 0.72	0.8283	NA	NA	NA	NA	NA	NA	NA	NA	NA
Intrinsic FL
pH 3.5	51.58 ± 0.30	52.42 ± 0.11	0.0653	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 4.5	62.60 ± 0.39	64.12 ± 0.20	0.0391 *	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 5.5	66.47 ± 0.19	68.56 ± 0.20	0.0086 *	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 6.5	67.87 ± 0.21	67.79 ± 0.11	0.6803	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 7.5	67.38 ± 0.09	67.79 ± 0.11	0.0552	NA	NA	NA	NA	NA	NA	NA	NA	NA
Extrinsic FL
pH 3.5	34.87 ± 1.29	36.81 ± 0.00	0.1673	51.44 ± 1.60	51.44 ± 1.52	1.0000	NA	NA	NA	NA	NA	NA
pH 4.5	59.48 ± 4.64	56.84 ± 1.52	0.5844	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 5.5	66.17 ± 0.07	66.17 ± 0.07	1.0000	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 6.5	67.41 ± 2.61	68.03 ± 2.52	0.8315	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 7.5	67.49 ± 2.56	66.72 ± 1.32	0.7417	NA	NA	NA	NA	NA	NA	NA	NA	NA
DSC
pH 3.5	36.63 ± 0.47	37.45 ± 0.08	0.1355	54.17 ± 0.59	54.27 ± 0.04	0.8333	59.13 ± 0.21	59.33 ± 0.08	0.3352	24.85 ± 0.58	26.69 ± 0.06	0.0467 *
pH 4.5	59.16 ± 0.08	59.04 ± 0.01	0.1699	66.68 ± 0.02	66.62 ± 0.03	0.1429	74.26 ± 0.05	74.08 ± 0.01	0.0379 *	51.44 ± 2.23	51.90 ± 0.28	0.7995
pH 5.5	68.57 ± 0.02	68.67 ± 0.01	0.0241 *	77.74 ± 0.04	77.57 ± 0.02	0.0329*	NA	NA	NA	59.99 ± 0.39	61.14 ± 0.29	0.0789
pH 6.5	69.45 ± 0.12	69.56 ± 0.01	0.3255	77.84 ± 0.01	77.81 ± 0.01	0.0955	NA	NA	NA	61.58 ± 0.25	61.44 ± 0.07	0.5254
pH 7.5	69.26 ± 0.01	69.50 ± 0.01	0.0017	77.58 ± 0.03	77.56 ± 0.06	0.7143	NA	NA	NA	61.80 ± 0.30	61.75 ± 0.14	0.8507
DLS
pH 3.5	78.83 ± 0.88	79.62 ± 0.09	0.3339	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 4.5	73.95 ± 0.86	73.10 ± 0.33	0.3218	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 5.5	73.83 ± 1.05	72.16 ± 1.93	0.3949	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 6.5	70.66 ± 0.39	70.16 ± 2.73	0.8216	NA	NA	NA	NA	NA	NA	NA	NA	NA
pH 7.5	74.57 ± 2.41	74.47 ± 0.00	0.9585	NA	NA	NA	NA	NA	NA	NA	NA	NA

* Asterisks indicate statistically significant differences (p < 0.05). NA: not applicable, FL: fluorescence.

. The stability of the overall secondary structure was slightly higher at pH 7.5 (~69 °C), comparable to pH 6.5, 5.5, and 4.5 (~66 °C), and lower at pH 3.5 (~53 °C).

For further confirmation of the secondary structure, Raman spectroscopy was used as a complementary method to CD, providing information on the vibrational states of the mAb molecules. The Amide I (1600–1700 cm⁻¹), tryptophan (1520–1580 cm⁻¹), and tyrosine (780–920 cm⁻¹) bands were analyzed. A comparison of the normalized Raman spectra of the innovator and the biosimilar candidate mAb at different pH values is shown in Supplementary Figure S3. Overall, both mAbs displayed very similar spectra for all three regions, with minimal differences as calculated by ESD. Minor differences related to the Amide I peak maximum and tyrosine mean band were detected at pH 7.5.

Intrinsic and extrinsic fluorescence spectroscopies were performed to evaluate the overall tertiary structure of both mAbs and to monitor their conformational stability as a function of temperature under different pH conditions. Intrinsic Trp fluorescence spectroscopy is a routine method to study tertiary structural changes in proteins. The emission spectrum of the fluorescent indole ring of Trp residue is a sensitive probe of its microenvironment. Evolocumab contains 18 Trp residues (2.6% of its amino acid composition). As shown in Supplementary Figure S4A, the innovator and the biosimilar candidate mAb presented highly similar emission spectra at 10 °C. A wavelength of maximum intensity at ~335 nm was observed at all pH values, indicating a relatively hydrophobic environment for the average Trp residue. The ESD values were relatively small (<5.0), indicating minimal differences within the method error between samples. Thermal curves were plotted by monitoring the mean spectral center of mass (MSM) peak position versus temperature. This method was chosen due to its higher sensitivity for the detection of subtle conformational changes [27]. Similar melting curves were observed for both mAbs at all pH values (Supplementary Figure S4B). A major thermal transition was observed under all conditions for both mAbs, with an increase in the Trp fluorescence peak position observed with temperature increase. The T_m values for the innovator and the biosimilar candidate were overall similar, with higher T_m values detected for the biosimilar candidate mAb at pH 4.5 and 5.5, as summarized in Table 1. The stability of the tertiary structure was observed to be the highest at pH 7.5, 6.5, and 5.5 (~66–68 °C), slightly lower at pH 4.5 (~62–64 °C), and much lower at pH 3.5 (~51–52 °C), as expected.

Extrinsic fluorescence spectroscopy was used as a complementary method to monitor changes in tertiary structure. SYPRO Orange is a fluorescent dye used to probe changes in the accessibility of hydrophobic regions of proteins since it becomes strongly fluorescent upon interacting with less polar environments. To investigate the effect of temperature on the average surface hydrophobicity of both mAbs, thermal melting curves were plotted by monitoring the SYPRO Orange fluorescence emission peak intensity as a function of temperature (Supplementary Figure S4C). The plots had consistent trends for both mAbs. Initially, the SYPRO Orange intensity kept nearly constant until the temperature reached the onset temperature (T_onset), where the peak intensity sharply increased, indicating improved accessibility of the dye to the apolar surfaces of both mAbs due to structure changes. A decrease in the intensity for all pH values was observed at temperatures higher than the offset temperature (T_offset). This could be explained by protein aggregation causing the release of the dye from previously exposed hydrophobic regions and/or by an increase of fluorescence quenching at higher temperatures [28]. The T_m values determined by this technique were also similar for the innovator and the biosimilar candidate mAb (Table 1). At pH 3.5, two earlier thermal transitions were detected around 34–36 °C and 51 °C, suggesting that the mAbs were thermally more labile at this pH value. At higher pH values, a single transition was observed at ~56–59 °C (pH 4.5), ~66 °C (pH 5.5), ~67–68 °C (pH 6.5), ~66–67 °C (pH 7.5), showing pH-dependent stability. In summary, these results indicate that the higher-order structure of the innovator and the biosimilar candidate mAb is quite similar even across a broad range of temperature and pH conditions.

3.7. Aggregation Propensity and Overall Conformational Stability

The aggregation behavior of both mAbs was studied by DLS. The hydrodynamic diameter of each mAb sample was probed as a function of temperature (10–80 °C) to assess their colloidal stability, as shown in Supplementary Figure S5. Overall, temperature-induced aggregation behavior of both mAbs had similar trends. At pH 3.5, an increase in mAb size from 9 to 18 nm was observed, suggesting microaggregation of the samples under very acidic conditions, as previously reported for another IgG2 molecule [28]. At all other pH values, the hydrodynamic diameter was similar (10–13 nm) until the temperature reached 70–74 °C, which then resulted in a significant size increase due to aggregation. T_onset values were similar for both mAbs (Table 1). A second DLS analysis was performed at 25 °C using five replicates of each mAb at 0.7 mg/mL in PBS. The hydrodynamic diameters observed for the innovator and the biosimilar candidate mAb were 9.58 ± 0.33 nm and 10.46 ± 0.94 nm, respectively, thus not statistically different (p = 0.20), indicating similar sizes. Monomeric mAb molecules usually have a size of 9–12 nm [29], which is consistent with our DLS data.

Temperature-induced changes in the overall conformation of both mAbs were also analyzed by DSC from 10–110 °C. This method measures changes in the heat capacity of a sample resulting from disruption of forces that stabilize the native protein structure. The thermal unfolding midpoint (T_m) values are indicative of the degree of conformational stability. DSC thermograms for the innovator and the biosimilar candidate mAb were superimposed (Supplementary Figure S6), with two or three major endothermic transitions as previously reported for IgG2 molecules [30]. The thermograms at pH 3.5 and 4.5 were further fitted to three transitions. T_onset for the first thermal transition was also calculated for all pH conditions. The resulting T_m and T_onset values for both mAbs were similar, with some very minor differences, as summarized in Table 1. At pH 3.5, three thermal transitions were detected, around 36–37 °C, 54 °C, and 59 °C. At pH 4.5, transitions appeared at approximately 59 °C, 66 °C, and 74 °C, suggesting that the mAbs were thermally more labile in more acidic environments. At higher pH values (5.5, 6.5, and 7.5), just two apparent transitions were observed at ~68–69 °C and ~77 °C. Overall, both mAbs presented similar aggregation propensity and conformational stability, with only some very minor differences.

3.8. Physical Stability Profiles (pH versus Temperature)

The physical stability profiles of both anti-PCSK9 mAbs were evaluated in citrate-phosphate (CP) buffer across a wide range of pH (3.5–7.5) and temperature (10–80 °C). T_m was determined using the biophysical methods described above, and the results are listed in Table 1. A strong pH dependence was observed, and, as expected, the mAb samples were less stable under acidic conditions. Far-UV CD and intrinsic fluorescence spectroscopy probed secondary and tertiary structure changes that had similar T_m values (>50 °C). Extrinsic SYPRO Orange fluorescence and DSC methods detected lower temperature transitions (~35 °C) at acidic pH values, indicating the presence of partially unfolded states with exposed hydrophobic regions and earlier structural changes in mAb domains. DLS demonstrated that significant aggregation occurred at temperatures >70 °C, several degrees higher than other structural changes.

The individual datasets for each of the techniques used are shown sequentially in Supplementary Figures S2B, S4B,C, S5 and S6.

In order to quantitatively assess the biophysical dataset differences, we established a mathematical method called “Error spectral difference—ESD”. Figure 6 shows a comparison of three different mathematical approaches to compare the second derivative FTIR Amide I spectra of the innovator and the biosimilar candidate mAb: the spectral difference [18], the weighted spectral difference (WSD) [31], and the ESD proposed herein. The WSD is considered an improvement of the SD method, with an additional factor $\left({\displaystyle \frac{\left|y_{Ai}\right|}{{{|y}_A|}_{aver.}}}\right)$ magnifying differences in spectral regions with higher signal intensity. Although the ESD is a simpler adaptation of the SD method, the error factor ${({\sigma }_{Ai}+{\sigma }_{Bi})}^2$ makes the metric dimensionless and permits the direct comparison of different techniques. As shown in Figure 6B, the WSD discriminated well the differences in the higher signal intensity region (1650–1630 cm⁻¹), as expected. On the other hand, both the SD and ESD metrics detected differences in a broader range of the spectra. The error factor in the ESD method added better resolution to discriminate between different regions of the spectra. The averaged ESD values for each pH as a function of temperature and the overall ESD for each biophysical technique are reported in

Table 2. Biophysical methods ranking according to their discriminative power based on error spectral difference (ESD) mathematical approach.

		Error Spectral Difference (ESD)
Rank Position	Biophysical Assay	pH 3.5	pH 4.5	pH 5.5	pH 6.5	pH 7.5	Average
1	Dynamic Light Scattering	4.8	251.2	48.1	7.5	387.2	139.8
2	MSM Intrinsic Trp Fluorescence	22.8	8.0	7.5	4.7	7.4	10.1
3	Circular Dichroism	5.2	1.2	1.3	1.5	1.3	2.1
4	Differential Scanning Calorimetry	0.0	0.0	1.4	0.1	0.0	0.3
5	Extrinsic SYPRO Orange Fluorescence	0.1	0.2	0.2	0.1	0.3	0.2

. Using this approach, it is possible to rank methods according to their discriminative power. In this work, the discriminative power of each technique was ranked as follows: DLS > intrinsic tryptophan fluorescence > Far-UV CD > DSC > extrinsic SYPRO Orange fluorescence. The application of such mathematical and visualization methods enables a less subjective comparison of biophysical datasets and, thus, a more efficient assessment of biosimilarity.

3.9. In-Vitro Binding Analysis

The in-vitro affinity of both mAbs for the target protein PCSK9 in solution was studied using bio-layer interferometry (BLI). Representative sensorgrams for association and dissociation curves are shown in Figure 7A for the innovator mAb and in Figure 7B for the biosimilar candidate mAb. Both mAbs showed similar high affinity for PCSK9, with K_D values of 2.58 ± 0.41 nM and 2.30 ± 0.19 nM, respectively. No significant differences were observed for any of the kinetic parameters, as summarized in tabular form in Figure 7. These values (in the nanomolar range) are in accordance with other studies comparing the in-vitro affinity of several anti-PCSK9 mAbs using different platforms, including Octet [32]. However, Amgen reported picomolar binding affinity to human PCSK9 (16 pM) for Repatha^® measured by surface resonance plasma [19]. This difference between reported data and our measurements for Repatha^® is probably related to detection limit issues (K_D < nM) associated with the BLI technology. It is interesting to note that the differences in N-glycosylation observed for both mAbs did not affect their binding properties since no significant differences in their kinetic binding parameters were detected (summarized in Figure 7). These results are not surprising, since complete deglycosylation had no impact on the biological potency of Repatha^® [19], indicating that N-glycosylation is not critical for the biological activity of evolocumab. The mode of action (MOA) of evolocumab does not involve Fc effector functions since PCSK9 is a soluble target. This agrees with the minimal immune-effector functions of human IgG2 molecules [33,34].

4. Discussion

In this study, we described a head-to-head characterization of the anti-PCSK9 innovator mAb Repatha^® and a biosimilar candidate developed in-house at COPPE/UFRJ (Brazil). Considering the limited amounts of product available at early stages of development, a rational strategy was designed to assess as much information as possible about CQAs using only 10 mg of protein. By using high-throughput analytical methods, the biosimilar candidate and its reference product (Repatha^®) were extensively characterized. Overall, both mAbs shared similar high purity, primary structure, higher-order structure, size and charge heterogeneity, apparent solubility, in-vitro binding to PCSK9, and biophysical stability under temperature and pH stress conditions. Considering the in-vitro binding results for the anti-PCSK9 mAb samples, the N-glycosylation variations observed by HILIC-HPLC are unlikely to impact biological activity, as expected for mAbs of the IgG2 subclass and mAbs targeting soluble molecules. However, as any comparative differences identified are considered in terms of their effect on biosimilarity, more investigations via further in-vitro and in-vivo studies must be performed to assure the safety, efficacy, and PK/PD properties of this biosimilar candidate mAb. Early identification of any differences is important, and here we show that despite using a limited amount of material, they can be rapidly identified and provide a basis for further investigations, including comparative studies at therapeutically relevant concentrations (e.g., 140 mg/mL).

Minor charge and N-glycosylation differences to reference products have been reported for other biosimilar products, like SB5 (a biosimilar of adalimumab) [10]. The differences in acidic and basic variants or N-glycan species content did not affect the biological activity, efficacy, and safety of this mAb. There are several other studies regarding analytical assessment of mAb biosimilars in the literature, e.g., for infliximab. For instance, Fang et al. [35] performed an analytical assessment study using an infliximab biosimilar (Inflectra^®) and three different batches of Remicade^® (infliximab innovator). The N-glycosylation profiling analysis of the different lots of Remicade^® showed to be quite consistent, while some differences were observed in the biosimilar Inflectra^®. Two unique N-glycans were detected for Remicade^®, and differences regarding the contents of low-abundance species, such as glycans terminated with N-glycolylneuraminic acid and possessing α-linked galactose pairs, were observed. Their results also demonstrated slight differences to the higher-order structures and host-cell protein content. Lee et al. [36] also reported differences in CQAs, such as N-glycosylation profile and biological activity, for two biosimilar products (Flixabi^®/Renflexis^® and Remsima^®/Inflectra^®) of Remicade^®. They observed differences in the relative amount of galactosylated, afucosylated, high-mannosylated, and charged N-glycan species and also detected differences in the Fc-gamma receptor binding and ADCC response. Another study reported significant differences between Remicade^® and the biosimilars Remsima^® and Inflectra^® regarding the relative abundance of eight N-glycan species, detected by both CE-ESI-MS and LC-MS/MS [37]. Furthermore, Hong et al. [12] showed a reasonable N-glycosylation variability (2–20%) in 80 Remicade^® batches produced in the US and the EU. None of the differences observed in the previous studies were clinically meaningful or prevented approval of the biosimilar products.

The antigen-binding affinity and specificity of a biosimilar and its reference product must be comparable. The four IgG subclasses (IgG1, IgG2, IgG3, and IgG4) differ in their N-glycosylation profiles and hinges in the constant region, leading to different immunological effector functions and modes of action [33,38]. For biosimilar mAbs whose mode of action requires effector function, Fc-gamma receptor binding and ADCC activity must be carefully characterized. On the other hand, if effector functions are not considered to contribute to the mode of action (e.g., when the target is a soluble molecule, such as PCSK9), differences in N-glycosylation or Fc receptor binding are not likely to have a significant impact on the potency of the mAb. However, the N-glycosylation profile is quite important to other aspects such as pharmacokinetics and pharmacodynamics (PK/PD), solubility, and stability of mAbs [27,33].

The HOS and overall conformational stability of biosimilar products are important CQAs that can be investigated by a wide range of biophysical methods, among which CD, FTIR, and DSC are the most commonly employed to provide fast and useful data at relatively low costs. Despite the importance of HOS for protein function and stability, most biosimilarity assessments report such data qualitatively, using a simple visual comparison to the reference product in a given condition. Here, we propose a more robust biophysical comparative characterization that can not only characterize CQAs of both anti-PCSK9 mAbs but that can also guide quality control, process optimization, and formulation strategies for biosimilar development. Exploring the potential of biophysical methods across different stress conditions, such as temperature and pH, enables a more comprehensive comparative structural and functional characterization.

Characterization of HOS is a very important aspect for biopharmaceutical development/approval, so there is a need for the use of robust and comprehensive HOS characterization strategies [39,40]. In this context, the ESD numerical approach proposed in the present work is a useful tool to enable a more quantitative comparison of biophysical datasets (spectra and melting curves) as compared to the conventional visual and qualitative evaluation of results obtained from biophysical data plotted in comparative graphs. The application of spectral comparison algorithms like the weighted spectral difference (WSD) had been successfully implemented to calculate spectral differences [31], but we added an error term ${({\sigma }_{Ai}+{\sigma }_{Bi})}^2$ to the original SD [18] approach in order to have an estimate of accuracy and so to enable a statistical comparison. Since the additional error term also makes ESD dimensionless, it was possible to compare and rank multiple techniques regarding their informative power. The ESD is thus a chi-squared (χ²)-like metric that calculates the differences in datasets for a simplified interpretation: values greater than one indicate that the differences detected for the technique are greater than the error of the method, while values less than one mean the differences are less than the error. The ESD can be used to numerically probe dissimilarities between the innovator and the biosimilar candidate across different temperatures and pH values by using different experimental techniques, so in this work it was possible to rank methods according to their overall discriminative power, which represents a potential approach for setting quality control strategies and benchmarks as well as for instrumentation comparability.

5. Conclusions

Overall, our results proved that both mAbs (biosimilar candidate and innovator) are similar in terms of purity, homogeneity, primary structure, higher-order structural integrity, apparent solubility, aggregation propensity, in-vitro binding to PCSK9, and physical stability profile under different temperature and pH stress conditions.

Moreover, we have established a straightforward workflow to enable characterization of CQAs of mAbs and other biotherapeutics at very early development stages, when the availability of the purified molecule is still very limited. This workflow provides a useful tool to guide further biosimilar development, including screening of cell clones and optimization of upstream and downstream processes.