Implementation of a Rapi-Fluor LC-FLR-MS method for the relative quantification of released N-glycans from monoclonal antibodies

(1)

MSc Thesis Chemistry

Analytical Sciences

Research Project

Implementation of a Rapi-Fluor LC-FLR-MS method for the

relative quantification of released N-glycans from monoclonal

antibodies

By

Carlo Roberto de Bruin

12431737

March 2021 48 ECTS

August 2020 - March 2021

Supervisors: 1st Examiner: 2nd Examiner:

Sipke Sangers Rob Haselberg Andrea Gargano Ewald van den Bremer

(2)

2 MSc Thesis Carlo de Bruin

List of abbreviations

Abbreviation Definition

2-AA 2-aminoanthranilic acid

2-AB 2-aminobenzamide

ADCC Antibody-dependent cell-mediated cytotoxicity

AGC Automatic gain control

Arb Arbitrary units

AX Anion-exchange

BEH Bridged ethylene hybrid

CID Collision induced dissociation

DDA Data dependent acquisition

DMF Dimethylformamide

ETD Electron transfer dissociation

Fab Fragment antigen-binding

Fc Fragment crystallizable

FLR Fluorescence

HCD Higher energy collisional dissociation

HILIC Hydrophilic interaction liquid chromatography ICH International Conference of Harmonisation

IgG Immunoglobulin

LOD Limit of Detection

LOQ Limit of Quantitation

m/z Mass to charge ratio

mAb Monoclonal antibody

MS Mass spectrometry

NHS N-hydroxysuccinimidyl

pmol picomol

PTMs Post translational modifications

PTS Performance Test Standard

RF Rapi-Fluor

RFMS Rapi-Fluor MS

RSD Relative standard deviation

RT Retention time

SPE Solid phase extraction

(3)

Abstract

N-glycans are located at the Fc region of a monoclonal antibody (mAb). These compounds are known for their possible impact on the effector function of a mAb. Therefore, the interest in the biopharmaceutical industry for the characterization and quantitation of N-glycans is increasing. Released N-glycans are challenging to analyze as they do not show UV nor fluorescence absorbance and have low ionization efficacies. In the past years, an effective label was developed called RapiFluor-MS (RFMS), which can accomplish the detection of released N-glycans with sufficient sensitivity. While this technique has great potential, there is a lack of studies that qualified this protocol for multiple species of N-glycans. In this study, a LC-FLR-MS method was optimized and successfully qualified for the characterization and relative quantitation of 19 released N-glycans labeled with RFMS. Within these 19 released RFMS N-glycans, the compounds contained properties such as fucosylation, galactosylation, sialylation and bisection. The method was qualified by following the ICH (International Conference of Harmonization) guidelines (CPMP/ICH/381/95) [30], along the parameters: linearity, range, quantitation limit (LOQ), detection limit (LOD), specificity, accuracy, precision and robustness. In multiple parameters the performance of the MS and FLR detector were compared. Eventually, the FLR data was qualified for the use of relative quantification of the glycan profile and the MS data was qualified for the identification of RFMS N-glycans. This method shows the potential to be expanded for more compounds and other future perspectives for this released N-glycan analysis are indicated and discussed.

(5)

1. Introduction

Monoclonal antibodies (mAbs) are biopharmaceuticals that are used increasingly for the treatment of a wide range of diseases such as cancer and autoimmunity. These are Y-shaped proteins (see Figure 1) which are mainly produced by plasma cells and are part of the immune system to neutralize pathogens. The most common antibody selected for the production of therapeutic mAbs is immunoglobulin (IgG). Therapeutic mAbs can be produced by injecting a mammal with an antigen that induces an immune response, which is a common antibody production technique. The B cells in the blood produce antibodies which are harvested. The isolated B cells are used to produce a hybrid cell line called a hybridoma, immortal cancer cells are fused with the isolated B cells to form these hybridomas. The hybridomas can be grown in culture and produces one type of mAb for each culture [1]. Currently, also in vitro techniques such as phage display and modelling are used in therapeutic mAb production [2]. The effectiveness of therapeutic mAbs is dependent upon multiple mechanisms, for example, their ability to link antigen recognition with an appropriate effector function, to elicit a biological response in vivo that will treat the targeted disease [3]. The mAbs consist of an Fc region and two Fab domains. Within the Fab region there is the binding site where specific antigens can bind and are neutralized. The Fc region is the lower part of the antibody which provides the binding site for Fcγ receptors and hence is responsible for antibody effector mechanisms. In addition, this is the region that contains N-glycans. In Figure 1, the characteristic structure of a mAb and an example of glycosylation is shown.

Figure 1: An antibody comprises of two identical heavy chains in association with two identical light chains arranged in a

characteristic Y-shaped structure. The Fc region is glycosylated with variating N-glycans, the degree of fucosylation, galactosylation, bisection and sialylation has impact on the effector function of the antibody [4].

(6)

Studies over the last decade have determined that the effector function of mAbs is highly dependent upon the structures of the N-linked glycans of the Fc domain of the mAb, because the three-dimensional shape of the mAb can differ for variating glycosylation patterns [5-7]. For example, larger N-glycans, e.g. bi-antennary complex type with terminal galactosylation, open up the Fc part in the CH2 region. Whereas

smaller attached N-glycans like the core structure favor a more “closed” Fc conformation. This open and closed formation can greatly influence the effector functions induced by interactions of the Fc part with receptor molecules [8]. Total removal of the N-glycans is highly detrimental to the effector function of the mAb, but subtle differences in the glycan structure, can improve the bioactivity and function of the mAbs significantly [9, 10].To give some examples, a decrease of fucosylation increases the antibody-dependent cell-mediated cytotoxicity (ADCC), which can enhance the efficacy of the mAb. In addition, the presence of high-mannose glycan structures also increases the ADCC [11]. In summary, the degree and distribution of fucosylation, galactosylation, mannosylation and sialylation of N-glycans is important knowledge when producing therapeutic mAbs [12, 13].

Therefore, it is valuable to analyze and monitor these N-glycan profiles of therapeutic mAbs throughout their developmental process. The host cellular production system including the bioreactor environment can produce mAbs with different glycosylation profiles, hence cell culture conditions should be considered in bioprocess development. Cell culture conditions such as dissolved oxygen, nutrient levels, pH and feed strategies can all have considerable influence on the glycosylation profile of the mAb, which could affect product quality and efficacy. Great improvements have been made in techniques for high resolution and high throughput analysis of N-glycans such as hydrophilic interaction liquid chromatography (HILIC) and mass spectrometry (MS) [14].N-glycan analysis is challenging, because they are largely invisible to optical detection methods which results in laborious sample preparation in order to release them enzymatically from the protein and to label them to allow fluorescence detection. Multiple labels were used to fulfil this purpose, such as 2-aminobenzamide (2-AB) and 2-aminoanthranilic acid (2-AA) [15] . These labels are coupled to the glycan structure via a multi-hour reductive amidation to activate fluorescence detection. While addressing the detectability of the N-glycans, it leaves the N-glycans hard to identify by mass spectrometry due to the low ionization efficiency.

Waters developed a fast and efficient sample preparation for N-glycan analysis, Rapi-Fluor MS (RFMS) [16].This sample preparation method takes less than one hour in simple steps including deglycosylation, efficient labelling and sample clean-up. For the deglycosylation an enzyme was used named peptide-N4

-(N-acetyl-β-D-glucosaminyl) asparagine amidase F (PNGaseF), this enzyme is the most frequently used for N-glycan release and the efficiency has been well-established [17]. PNGaseF is an amidohydrolase that removes intact asparagine-linked oligosaccharide chains from glycoproteins and glycopeptides [18]. In addition, it is capable to work under high temperatures for fast N-glycan cleavage. The RapiGest surfactant is an anionic surfactant that ensures the accessibility of the N-glycans on the mAb and it prevents the precipitation of the mAb under heat denaturation conditions. The RFMS label consists of an N-hydroxysuccinimidyl (NHS) carbamate, a quinolinyl core and a tertiary amine group (See Figure 2). These components facilitate rapid labelling kinetics, highly sensitive fluorescence and high gas-phase proton affinity for good ionization and sensitive mass spectrometry measurements in positive mode. This label was compared to the other labels by multiple studies, which proved that RFMS labelling is more sensitive in both fluorescence and mass spectrometry [16]. Nevertheless, this labeling technique is rarely used in

(7)

literature compared to the other mentioned labels. Only a few studies were published that use RFMS labeling for different mAbs, without a proper qualification of the used method [19, 20].

Figure 2: The structure of the RapiFluor-MS label, it consists of an N-hydroxysuccinimidyl (NHS) carbamate, a quinolinyl core

and a tertiary amine group.

The challenge regarding qualification of this released RFMS glycan assay is the amount of N-glycans actually taken into account. For example, in the study of Lim et al. [21], the protocol was qualified for only four major abundant RFMS N-glycans (G0F, G1F isomers and G2F), while more N-glycans are present in their results. In addition, it was not mentioned how the recovery (range 90 % - 110 %) was calculated for the qualified compounds. In their study, an IgG standard was used with ranging concentrations to obtain a range in N-glycan concentration and to determine the linearity. This was done because there are no certified RFMS released N-glycan standards for single compounds available on the market, which complicates a validation protocol with a standard addition approach. Despite that this challenge is acknowledged, no relative standard deviations or residual errors were shown within their linearity test. The main challenge of performing a proper qualification is that the initial N-glycan concentration in an IgG cannot be known beforehand. Additionally, the sample preparation procedure is optimized for an amount of 15 µg IgG using a concentration of 2 mg/mL, therefore, it was decided to use a different approach in our study. A separate Quantitative Standard (QS) standard is used to determine the linearity of the method to avoid the approach of ranging IgG concentrations. In addition, a Performance Test Standard (PTS) is used to perform an adequate accuracy and recovery test, which lacks in the qualification of Lim et al. The PTS represents a certified standard with multiple compounds combined, it contains 19 released RFMS-N-glycans. In this way, the protocol should be qualified for more N-glycans that have different properties. To obtain more information on the N-glycan profile in a robust and reproducible manner, which is important information during the production of mAbs.

In this study, a LC-FLR-MS method is implemented for the identification and relative quantitation of mAb N-glycan profiles. In short, the N-glycans are released from mAbs labeled with RFMS, isolated by solid phase extraction and finally analyzed by liquid chromatography coupled to fluorescence and mass spectrometry detection. The capabilities of the method are checked and the method is optimized to increase the performance within all steps (sample preparation, LC separation, MS detection). Afterwards, the method is qualified for 19 released N-glycans (see Table 19) along the following parameters: linearity, range, specificity, accuracy, precision, and robustness. Within these parameters the performance of the FLR and MS detectors are compared to qualify if they are both capable of the identification and relative

(8)

quantification of released N-glycans. Within these 19 released N-glycans multiple species are present such as: fucosylated, galactosylated, sialylated and bisected N-glycans. Differences between the results of the multiple species of N-glycans are indicated and discussed. Future perspectives are also given to further improve the throughput and robustness of this released N-glycan analysis.

2. Material & Methods

2.1 Materials

The RapiFluor-MS kit was acquired from Waters Corporation, which was used for the sample preparation. It consists of an intact mAb check standard (IgG control sample), PNGaseF, RFMS label-reagent, a HILIC µ-elution plate and the necessary SPE reagents. Additional standards were also obtained from Waters such as Performance Test Standard (PTS) and the RFMS Quantitative standard (QS). Two antibody batches were used originating from Genmab BV (IgG-b12) and Sigma (I4506), solvents (ULC-MS quality) were obtained from Biosolve and Merck.

2.2 Sample preparation

The sample preparation consists of three steps: deglycosylation, fluorescence labeling and glycan enrichment by solid phase extraction (SPE). The experimental steps were based on the procedure given by Waters. In Figure 3, a schematic overview of the complete method is displayed. In summary, 6 µL of RapiGest SF solution was added to 7.5 µL of glycoprotein (2 mg/mL). The samples were heated for three minutes at 90 ˚C and cooled afterwards. Then 1.2 µL PNGaseF was added to the samples and mixed, the samples were incubated at 50 ˚C for five minutes, subsequently cooled for three minutes at room temperature. For the fluorescence labeling step, a RapiFluor-MS solution was prepared and 12 µL was added to the samples. The labeling reaction was completed in five minutes, where after the samples were diluted by addition of 358 µL of acetonitrile in order to prepare the samples for the glycan enrichment step with HILIC SPE. A HILIC µ-elution plate and vacuum manifold was used for the SPE step. First, the wells were conditioned and equilibrated by water and 85 % acetonitrile. Then the samples were loaded, and subsequently the wells were washed two times with 1/9/90 (v/v/v) formic acid/water/acetonitrile. Then the N-glycans were eluted with three times 30 µL 200mM ammonium acetate in 5 % acetonitrile. Finally, the samples were diluted with 310 µL of DMF/Acetonitrile solution and were ready for LC-FLR-MS analysis.

2.3 Analysis with UHPLC-FLR-MS

The labeled N-glycans were separated on a Vanquish UHPLC (Thermo Scientific) with a binary pump system equipped with an ACQUITY Glycan BEH Amide column (2.1mm x 150mm, 1.7 µm particle size)(Waters). A gradient was applied with 50mM ammonium formate buffer (pH 4.4) as mobile phase A and 100 % acetonitrile as mobile phase B at a flow rate of 0.5 mL/min. The gradient was linear for 35.0 minutes from 80 % to 63 % of mobile phase B. Then, 100 % mobile phase A was employed for ~5.0 minutes. Afterwards, the column was re-equilibrated with 80 % of mobile phase B for 15 minutes. In Table

(9)

2, the gradient details are shown. The injection volume was 16 µL (~8 picomol of the total amount of N-glycans) and the column temperature was set at 45 ˚C. The labeled N-glycans were detected by fluorescence and mass spectrometry, which was performed by an Orbitrap Eclipse Tribrid MS from Thermo Fischer Scientific. The Vanquish UHPLC (Thermo Scientific) was connected to both detectors by a split tubing (1:1 ratio). The FLR detector was set at 264nm excitation and 425nm emission wavelengths. The MS detection was carried out by data dependent acquisition (DDA) mode. Positive ionization was used at a voltage of 3400V and temperatures of 300˚C and 250˚C were used in the ion transfer tube and as vaporizer temperature. The sheath gas was set at 35 Arb and the auxiliary gas at 7 Arb. The MS1_{scan employed a}

range of 350-2000 m/z at a scan rate of 3 Hz, where a resolution of 120,000 was set. Precursor selection for MS2 scans were restricted by an intensity threshold of 50,000 and dynamic exclusion was set at 6 seconds. The precursor ions were isolated in the quadrupole and fragmented by higher energy collisional dissociation (HCD). The MS2_{resolution was set at 60,000 and the normalized HCD collision energy was 28 % with a}

fixed energy mode.

Figure 3: Schematic overview of the complete method. Step 1: an intact glycoprotein is exposed to the RapiGest surfactant and

the enzyme PNGaseF to deglycosylate the glycoprotein. Step 2: All the released N-glycans were labeled with the RFMS label with a reaction time of 5 min. Step 3: The N-glycans were purified with a sample clean-up step using a HILIC µ-elution plate and vacuum manifold, afterwards the samples were diluted and ready for LC-FLR-MS analysis. The raw data was processed by Genedata Expressionist software.

(10)

3. Method optimization

An optimization was carried out in order to check the capabilities of the protocol in comparison to the published results [16]. The optimization of the LC-FLR-MS analysis was performed by variating the following parameters: LC gradient, MS ionization and MS/MS fragmentation. The sample preparation procedure consists of three steps: deglycosylation, labelling and sample clean-up as described in Section 2.2. The procedure was tested through recovery testing by different experimental set-ups, which are described below. Data analysis was performed with Genedata Expressionist software where multiple tools were used to process the data and a final workflow was created for automated data analysis.

3.1 LC optimization

The LC-FLR-MS analysis was set up with the use of a certified standard named RFMS Glycan

Performance Test Standard. This standard contains 19 released N-glycans labeled with RFMS in a given relative abundance ratio. Two linear gradients were evaluated, which were provided by Waters together with the description of the procedure.The first gradient is intended for universal N-glycan profiling and the second gradient for mAb N-glycan profiling (See Table 1 & Table 2 for the details).

Table 1: LC gradient universal N-glycan profiling and column temperature in ˚C. Time (min.) Flow (mL/min.) % A % B T (˚C)

0.00 0.4 25 75 60 35.0 0.4 46 54 36.5 0.2 100 0 39.5 0.2 100 0 43.1 0.2 25 75 47.6 0.4 25 75 55.0 0.4 25 75

Table 2: LC gradient mAb N-glycan profiling and column temperature in ˚C. Time (min.) Flow (mL/min.) % A % B T (˚C)

0.00 0.5 20 80 45 3.00 0.5 27 73 35.0 0.5 37 63 36.5 0.2 100 0 39.5 0.2 100 0 43.1 0.2 20 80 47.6 0.5 20 80 55.0 0.5 20 80

The universal profiling method was not capable to separate all the 19 N-glycans sufficiently. This was due to the overlapping peaks of the N-glycans that contain sialic acids, which complicated annotation for these compounds. However, the mAb glycan profiling gradient was capable of separating all 19 N-glycans, also the present isomers (G1, G1F and G1F+GN) were sufficiently separated. The FLR

(11)

between both. It was decided to continue with the mAb N-glycan profiling method as the elution window for the mAb related N-glycans was slightly larger, allowing for more time between the peaks. In the study of Hilliard et al, the same protocol was used to analyze a NIST mAb, which contains 35 N-glycans [19]. In their chromatography, co-eluting peaks were observed with the use of the universal gradient.

Therefore, they also used the mAb N-glycan profiling gradient, which improved their results as well.

Figure 4: FLR chromatogram of released N-glycans originating from the PTS that was obtained by use of the universal gradient

(see Table 1).

Figure 5: FLR chromatogram of released N-glycans originating from the PTS that was obtained by use of the mAb profiling

gradient (see Table 2).

It has to be noted that these data show slight retention time differences with the data obtained for the qualification experiments. This was due to an instrument maintenance event, where the piston seals of the UHPLC pump were required to be replaced. The difference in retention times between injections of the same batch and different batches of PTS were evaluated. The shift between compounds of the same batch was negligible and the shift between compounds of different batches was also small (<0.1 min).

3.2 MS optimization

For the MS settings, a Thermo glycan template, containing MS and HESI source settings (See Section 2.3), was used and further optimized where necessary. Minor adaptions were made to the original settings, for example ETD fragmentation was tested, which could deliver additional fragment information for larger N-glycans. As ETD fragmentation is known to produce cross ring fragments, which are c- and z- type ions [22].ETD fragmentation is most suitable for ions with high charge states (>2+), while most of the measured N-glycans have a charge of 2+, only larger N-glycans such as A2F can produce a charge state of 3+. It was notable that ETD is not capable of producing sufficient fragmentation for these N-glycans and is therefore not suitable for the identification of released N-glycans (data not shown). Therefore, it was decided to not

(12)

use scan time to acquire ETD fragmentation spectra. CID fragmentation was also tested, but HCD fragmentation produced more N-glycan fragments with higher intensities.

HCD fragmentation is known to produce predominantly Y- and B-ions and depending on the collision energy it may produce some cross ring fragmentation as well [23]. HCD fragmentation produces sufficient fragments for the identification of released N-glycans and complements the specificity of the method. In Figure 6, the fragmentation spectra of G0F is shown with annotated fragments.

Figure 6: The MS2_{HCD (28% collision energy) fragmentation spectra of RFMS labeled G0F (m/z 1774,72), a characteristic}

N-glycan fragment is observed at m/z 204 N-Acetylglucosamine (GlcNAc). The spectrum was dominated by Y- and B-ions and the MS/MS score was 94.8, which confirms the identification of G0F.

The HCD collision energy was set at 28 % and delivered clear spectra, higher collision energies were also tested but this led to loss of specific larger fragments. HCD fragmentation of N-glycans delivers mostly Y- and B-ions, but there were also some low intensity C- and Z-ions present, which are cross ring fragments. Dynamic exclusion was employed with an exclusion duration of 6 seconds, which resulted in multiple spectra across the maximum of each peak.

3.3 Recovery of the sample preparation procedure

The sample preparation procedure was tested to evaluate its suitability. Therefore, the efficiency of the sample clean-up procedure was determined. This was performed by collecting and analyzing all the SPE steps: two wash steps, two load steps and three elution steps (See Table 3). After elution, the samples were diluted in 310µL Dimethylformamide/Acetonitrile. For this experiment, a control IgG-b12 sample was used from Genmab’s inventory. In total nine samples were used, three samples did not pass the SPE clean-up procedure (SPE 0), another three samples passed the SPE-clean-up procedure once (SPE 1), while the last three samples passed this procedure two times (SPE 2). During this optimization step, the experiment was

(13)

performed multiple times due to some random errors. It was observed that the recovery was dependent on a constant low flow rate. Otherwise, the compounds will flush through the column in the load step and the recovery will be very low ~10 % (data not shown). Eventually, a suitable constant low flow rate was found (2.5 Hg vacuum).

Table 3: Scheme clean-up, including the purpose of each step with the corresponding solution and amount in µL.

Step Purpose Solution Amount (µL)

1 Conditioning Water 200

2 Equilibration Water/Acetonitrile (15/85 %) 200

3 Load sample Samples 400

4 Washing Formic acid/Water/Acetonitrile (1/9/90 %) 600

5 Washing Formic acid/Water/Acetonitrile (1/9/90 %) 600

6 Elution 200mM Ammonium acetate in 5 % Acetonitrile 30

As expected, no signal was observed in the following fractions: load fraction (1), wash fractions and 2nd

elution fraction of the SPE 1 samples. The first three samples (SPE 0) that did not passed the SPE clean-up procedure had no signal. This was because of an error in the sample composition (no elution buffer), which caused elution of the compounds within the void time of the LC separation. Therefore, the relative recovery of SPE 2 was calculated by comparing the samples of SPE1 and SPE2. Similar N-glycan profiles were observed in SPE 1 and SPE 2 samples, however, the absolute intensities in SPE 1 were higher than in SPE 2 (See Figure 8A & B).

It was speculated that this was due to the sample composition of the SPE 2 samples, which consisted of 90µL elution buffer and 310µL of diluent. They had this composition because they underwent SPE 1 already. Therefore, some N-glycans already eluted during the load step (2) of SPE 2. It was decided to combine the peak areas of SPE 2 and the load step of SPE 2 to obtain the relative recovery results (See Table 4). The control IgG-b12 has a simple glycosylation profile, the most abundant N-glycans consist of G0F, G1F isomers and G2F.

(14)

Figure 8: (A) Overlay FLR chromatograms of SPE 1 & SPE 2 samples. (B) Overlay FLR chromatograms of SPE 1 and the load

step fraction of SPE 2.

The recovery is in the 75-95 % range, the recoveries for each N-glycan differ slightly. However, it is observed in Figure 7 that the ratio of N-glycans almost remains intact, independent of its losses in the clean-up and that is the main result of this assay. Therefore, these results were sufficient to qualify the method following the ICH guidelines.

Table 4: Relative recovery of the sample clean-up, obtained by combining the fractions of the load step and elution of the SPE 2

samples. N-Glycan Recovery G0F 95 % G1F-6 80 % G1F-3 75 % G2F 84 %

(15)

3.4 Data analysis

The data analysis was performed with the use of Genedata Expressionist software. This software comprises of numerous activities that can be used to simplify and treat the data. The raw data was processed in a created automated workflow, which was mainly optimized by trial and error of numerous settings within each activity that was selected. The final workflow consists of the following processing steps: chemical noise subtraction, RT alignment, FLR/MS peak detection, charge and adduct grouping, MS/MS consolidation & peak detection and finally a glycan library search was done. These processing steps allow automated analysis of the raw data and provides the N-glycan profile (in relative abundances) of each analyzed sample. In Figure 10, an overview of the complete workflow with all activities is shown. The MS data is visualized as an ion map, the y-axis depicts the retention time and the x-axis depicts the m/z range. Signal Intensities were displayed as a color gradient, as in a heat map or tile plot. Acquired MS/MS spectra are depicted as In Figure 9, an example of such an ion map is given of a typical raw N-glycan data.

Figure 9: Example of ion map of a PTS sample, where the y-axis depicts the retention time and the x-axis depicts the m/z range.

The red dots represent acquired fragmentation spectra.

First, the raw data is preprocessed to reduce the noise and apply chromatographic smoothing. In this activity multiple algorithms were used to preprocess the data and remove low intensity signals that were indicated as noise. A moving average filter with a retention window of five scans was used to smooth the data. In addition, the subtraction of noise was done by a clipping method, which sets all values below the quantile to zero and this was combined with an intensity threshold. This tool also allows restriction of certain retention times and m/z values. The FLR data was preprocessed with a baseline subtraction, where a penalized least squares algorithm was used. After the preprocessing, all the retention times were aligned between samples.

(16)

Peak detection was performed with an ascent-based algorithm for both the MS and FLR data, which was used to calculate the peak boundaries around the signals. Additionally, the algorithm searches for the local maxima of each signal.

After the peak detection, all the peaks were clustered and grouped based on their isotope pattern, charge states and adducts. In this way, all the information belonging to a compound was grouped together. For these activities retention time and m/z boundaries were set at 0.1 minutes and 50ppm. In addition, all the singletons were filtered out (peaks that could not be assigned to a cluster containing at least two isotope peaks).

The MS/MS data was used to confirm the identification of the N-glycans. Therefore, the MS/MS data was processed as well. All MS/MS data that was not located within the peak boundaries of grouped signals was removed with a consolidation tool and the highest TIC was selected from each cluster. Again the ascent-based algorithm was used to provide the MS/MS data that approaches the local maximum of the peak. Afterwards, a deisotoping activity was used to recalculate the MS/MS fragments to singly charged monoisotopic peaks. This simplified the data and improved the probabilistic scoring algorithm that was used in the library activities.

Two library activities were used in this workflow, namely, the released glycan activity and the library MS activity. Both of these activities were used for the annotation of N-glycans, however, the library MS tool is intended for high throughput analysis. This library contains N-glycan names, masses and retention times, the activity annotates peaks based on this features. The released glycan activity was only used when the library MS activity was not able to annotate N-glycans. For example, if retention times are shifting, then the released glycan activity is preferred. It provides manual review options, so it can be observed which peaks to annotate and which not. In addition, the released glycan activity was used for the configuring of MS/MS scores for each glycan. The MS/MS score is calculated from the Andromeda scoring algorithm, which is based on the probability that the observed number of matches between the calculated and observed fragment masses could have occurred by chance. This score indicates the certainty of successful identification based on the detected fragments. After the annotation of all peaks, the data was transferred to Excel where statistical analysis was performed. In Figure 11, the chromatogram of all the 19 released N-glycans was shown with annotated features.

(17)

Figure 11: Chromatogram of all the 19 annotated glycans at their corresponding peak. See Section 7, Table 19, for the details of

each glycan.

For the quantification, different observables were used for the MS and FLR data. The FLR data is one-dimensional data and therefore, the area of the peak was used (Figure 12), where the area under the curve is subdivided into trapezoids and these values are summed. The MS data is two-dimensional data, because peaks were recorded across time and mass, which were plotted in a two-dimensional planar where the axes correspond to retention time and mass/charge ratio (m/z). For Orbitrap instruments the ion population abundance is represented by the amplitude of the signal. In case of isotopically separated signals, the peak width (in the m/z direction) of an isotope depends on the used resolving power used to acquire the transient. Therefore the maximum intensities across the peak (in the RT direction) are integrated to obtain the representative intensity read out for the mass spectrometry data (Figure 13).

Figure 12: Visual explanation of the integration of the peak area from FLR data. (The total value of the integral is obtained by

summing the volumes of all the trapezoids)

Figure 13: Visual explanation of the integrated maximum intensity (corresponds to the peak height integrated along its scans in

(18)

4. Method qualification

After method optimization, the method was qualified for batch profiling, following the ICH (International Conference of Harmonization) guidelines (CPMP/ICH/381/95) [30]. The following parameters were tested: linearity, range, quantitation limit (LOQ), detection limit (LOD), specificity, accuracy, precision and robustness. The performance of each test is explained and the corresponding results are shown afterwards. These experiments were performed using two mAbs: IgG (Sigma, I4506) and the PTS reference (Waters), which consists of 19 RFMS labeled released N-glycans originating from the same IgG (Sigma, I4506). In Table 19, the structures of all 19 N-glycans are shown without the RFMS label. Direct comparisons between the PTS and IgG (Sigma, I4506) could be inaccurate, because of the PTS and the IgG originate from different batches and could therefore show slightly deviating N-glycan profiles.

4.1 Linearity and Range

The linearity and range were assessed by a calibration curve from the Quantitative Standard (QS) from Waters. This standard consist of a RFMS labeled peptide (RFMS -TPTTQSSVSSQTTR-OH), which is derivatized by one mole equivalent of the RFMS label and enables quantitation based on fluorescence. The fluorescence response is linearly proportional to the amount of RFMS label, therefore, fluorescence peak areas can be used to quantify specific RFMS labeled N-glycans. The calibration curve was prepared in a range of 0.04 pmol - 4 pmol (picomol). The acceptance criteria for the linearity and range was set with a correlation coefficient (R2_{) of >0.99 over the specified range and RSD values below 10 % for each}

measured amount of QS. In addition, the residual error was calculated for each amount of QS. The calibration curve was also used to calculate the LOD and LOQ based on the standard deviation of the response and the slope (See Equation 1). The QS was reconstituted in 90µL of 200mM ammonium acetate in 5 % Acetonitrile and 310µL of diluent DMF/Acetonitrile. In this way, the conditions were identical compared to real samples. Different injection volumes (0.1-10 µL) were used to obtain results across the indicated range (0.04-4 pmol) and each amount of QS was analyzed in triplicate (n=3). There was also a blank used with 90µL of 200mM ammonium acetate in 5 % Acetonitrile and 310µL of diluent DMF/Acetonitrile. In Table 8 (see Section 7), an overview is shown of the samples and their features. The linearity and range were evaluated for the FLR and the MS data separately

Equation 1: The LOD and LOQ were calculated based on the standard deviation of the response and the slope of the calibration

curve. In the depicted formulas, σ represents the standard deviation of the response and S represents the slope of the calibration curve.

𝐷𝐿 =

3.3𝜎

𝑆

&

𝑄𝐿 =

10𝜎

(19)

Figure 14: Linearity of Quantitative Standard (RFMS -TPTTQSSVSSQTTR-OH peptide) within the range of 0.04 – 4 pmol for

the FLR (A) and MS (B) data, each amount of QS was measured in triplicate (n=3).

In Figure 14 A&B, the calibration curves are shown of the FLR and MS data. The precision of each data point (amounts of QS in pmol) was evaluated to prove efficient detection across the range. This was visualized in Table 6 and Figure 15, where the RSD and residual errors of both data are shown. The LOD is 0.019 pmol and the LOQ is 0.057 pmol for the FLR data (amount of QS), which means that this method can detect 0.019 pmol N-glycan from 4 pmol IgG. Or simplified, from the initial (15 µg) 100 pmol used IgG, 0.475 pmol of N-glycan can still be detected and 1.425 pmol of N-glycan can be quantified. The LOD and LOQ from the FLR data are well within the acceptance criteria of < 1 % (1 pmol) and < 3 % (3 pmol) detected/quantified N-glycan from the total amount of IgG. From the MS data, the LOD is 0.095 pmol and the LOQ is 0.29 pmol, which is 2.375 and 7.25 pmol of N-glycan per 100 pmol of IgG. In Table 5, a summary of the results with the corresponding acceptance criteria was given.

Table 5: A summary of the results of the linearity and range experiment, with their corresponding acceptance criteria.

Parameter Results Acceptance criteria

Linearity & Range

- R2_{of 0.9999 (FLR), R}2_{of 0.9932 (MS)}

- Range: 4 – 0.04 pmol, all points < 10 % RSD

- (FLR) LOD: 0.475 pmol N-glycan/100 pmol IgG (= < 1 %) - (FLR) LOQ: 1.425 pmol N-glycan/100 pmol IgG (= < 3 %) - (MS) LOD: 2.375 pmol N-glycan/100 pmol IgG (= > 1 %) - (MS) LOQ: 7.25 pmol N-glycan/100 pmol IgG (= > 3 %)

- R2_{> 0.99}

- < 10 % RSD - LOD < 1 %

N-glycan from total

amount IgG - LOQ < 3 %

N-glycan from total

amount IgG

Table 6: %RSD values at each measured amount of the QS (n=3), calculated from the triplicate measurements of each amount.

Amount QS (pmol) % RSD FLR data % RSD MS data

0.04 3.3 % 3.2 % 0.1 0.1 % 4.4 % 0.2 3.6 % 4.2 % 0.4 0.5 % 1.3 % 1.0 1.3 % 3.4 % 2.0 0.3 % 8.7 % 4.0 1.3 % 7.7 %

(20)

From the results it can be stated that all FLR data was within the acceptance criteria. However, the MS data had some results that were outside the acceptance criteria. The RSD values were within the acceptance criteria, but it was already observed that lower amounts of QS comprised of higher RSD values. In Figure 15, the residual error for the lowest amount of QS (0.04 pmol) showed a very high value, which clarifies that the LOD and LOQ of the MS data were not within the acceptance criteria. For the purpose of this method, it was not necessary to improve these results for the MS. Because, it was decided to use the FLR data for the relative quantification and MS for identification of N-glycans. Although, it is expected that a reduce of MS1_{scan time may improve the LOD and LOQ of the MS.}

Because, this will result in more data points on each peak, which could be useful for low abundant compounds that comprise of small peaks. This can be done by lowering the orbitrap resolution for the MS1_scan.

(21)

4.2 Specificity

The specificity was evaluated with the calculated mass accuracy, MS/MS score and the fragmentation spectra for each N-glycan. The mass accuracy was calculated by comparing the theoretical mass and the measured mass. This was performed with the reference PTS and treated IgG (Sigma, I4506) samples in triplicate (n=3). Both of these samples contain the same N-glycans and their accuracies (in ppm) were averaged. The acceptance criteria for the mass accuracy was set at < 5ppm. The acceptance criteria of the MS/MS score was set a minimal score of 10 for successful identification. Fragmentation spectra of each N-glycan with the highest score were visually checked if specific N-N-glycan fragments were present, for example the N-Acetylglucosamine (GlcNAc) fragment at m/z 204.

Figure 16: Average mass accuracy in ppm and MS/MS score of each individual N-glycan measured in triplicate (n=3). The red

line indicates the acceptance criteria, for the mass accuracy all N-glycans needed to be <5 ppm, for the MS/MS score all N-glycans needed a score above 10.

The average mass accuracy was < 1.2 ppm which is within the acceptance criteria of < 5 ppm. In addition, all the detected N-glycans had a MS/MS score above the acceptance criteria of > 10 score. In Figure 16, all the ppm values and consolidated MS/MS scores are shown for each N-glycan. In Figure 17, the HCD MS2

spectrum of G0 is shown to provide an overview of specific N-glycan fragments.

The N-glycan G0 is low abundant within the used IgG samples and PTS (See N-glycan profile in Figure 32). The relative abundance is around ~0.8 % and despite of this, the quality of the MS2_{spectra was still}

(22)

Figure 17: HCD MS2_{fragmentation spectra of G0 (m/z 1628.66) . This spectrum shows characteristic fragments of N-glycan}

fragmentation, for example the N-Acetylglucosamine (GlcNAc) fragment at m/z 204.

In the N-glycan profile three isomers were present, which were: G1, G1F and G1F+GN (see Figure 11). The isomers are indicated with -3 and -6, because the difference between these isomers is the terminal galactose residue at the α1.3 or α1.6 branch [24]. In Figure 18, the difference between these branches in the G1F isomers is displayed. While chromatographic separation between isomeric compounds (e.g. G1F isomers) was obtained, identifying the isomers still proposes a challenge. This is due to: the identical mass of the isomers and that the typical MS fragmentation techniques, HCD, CID and ETD, are not able to differentiate these isomers from each other. In Figure 19A & B, the MS2_{spectra of the two G1F isomers}

are shown.

Higel et al. also obtained chromatographic separation of G1F isomers and stated from quantitative data that the higher intense α1.6 isomer elutes first followed by the lower intense α1.3 isomer [25]. In this study, this pattern was the same for all isomers, the first eluting isomer was the most intense peak followed by the other isomer which was less intense (see Figure 11). Because of this, the α1.6 isomer was indicated as first eluting peak and the α1.3 isomer as second eluting peak for all three isomers (G1, G1F, G1F+GN) within this research. However, it is suggested to reveal this structural differences via own research to confidentially annotate these differences. Therefore, some possible solutions were indicated within the future perspectives (see Section 6).

(23)

Figure 18: Schematic representation of the two G1F isomers and their specific α- and β-linkages.

Figure 19: HCD fragmentation spectra of two G1F isomers from an IgG sample, A elutes at 18.45 min. and B elutes at 19.04 min.

Minor differences in intensity of fragments are present, Genedata Expressionist annotates both isomers to each spectrum as the fragmentation spectra are comparable.

(24)

4.3 Accuracy

To assess the accuracy of the method, two experimental set-ups were performed. The first experimental set-up (Test A) evaluates the impact of the sample preparation on the already released N-glycans. While the second experimental set-up (Test B) evaluates the impact of the SPE clean-up step on individual N-glycans. In the first experimental set-up (Test A), spiked samples with and without sample preparation steps were compared. The PTS was reconstituted in 25µL (200 pmol of total N-glycan) water and aliquoted into two sets of triplicate (2 x n=3) samples. One set of samples (n=3) was directly injected (1µL= 8 pmol) into the LC-MS-FLR instrument for analysis. While the second set of samples was first processed using the sample preparation procedure (deglycosylation, RFMS-labeling and clean-up). These samples were eventually diluted in 90µL 200mM ammonium acetate in 5 % acetonitrile and 310 µL of DMF/Acetonitrile solution, and 16µL (8 pmol) was injected for LC-MS-FLR analysis. Afterwards the two sets were compared with each other with respect of the obtained glycan profiles. The FLR data was used for this tests and the criteria for acceptable accuracy boundaries were set at 80-120 %. The results of the first test are shown below in Figure 20. The N-glycan profiles are shown in Section 7.

Figure 20: Accuracy of each N-glycan from the total N-glycan profile, the red dashed lines represent the boundaries of the

acceptance criteria (80-120 %).

In the results from the first experimental set-up (Test A), the accuracy of N-glycans (12 compounds) without a sialic acid residue were within the acceptance criteria. Overall, it can be seen that the N-glycan profiles of the two sample sets are comparable (see Section 7, Figure 32). However, the N-glycans which contain sialic acid residues (Figure 20, A1F-Gal, A1, A1F, A1F+GN, A2, A2F and A2F+GN) are outside the acceptance criteria. The results of N-glycans that contain sialic acids are corresponding to the results of the precision and therefore will be discussed in Section 4.4.

In the second experimental set-up (Test B), two sets of triplicate (2 x n=3) IgG (Sigma, I4506) samples were compared. The first set was processed by the sample preparation procedure while the second set was excluded from the SPE clean-up step. Both sample sets were compared to each other based on the N-glycan ratio, which resulted in the accuracy (see Figure 22). In addition, this experiment provides

(25)

information on the recovery of the SPE clean-up step. In Figure 21, an overlay of the chromatograms from the samples “with SPE clean-up” and “without SPE clean-up” were shown.

Figure 21: Overlay of the chromatograms from the samples “including the SPE clean-up step” in blue and the samples “excluded

from the SPE clean-up step” in red, which represents a visualization of the 50 % recovery.

Figure 22: Accuracy of each N-glycan for the comparison between the samples “including SPE clean-up” and “excluded from

SPE clean-up”.

In the results from the second experimental set-up (Test B), the N-glycan profile of both set of samples were very comparable. All expected N-glycans show accuracy within the 80-120 % acceptance criteria. Although, the recovery of the clean-up step was low (45-55 %, range of all N-glycans), this does not have an impact on the accuracy of the glycosylation pattern. This was due to the comparable recovery values of all N-glycans. This result was also observed in the method optimization, see Section 3.3, and by Lauber et al. [16]. Possible improvements in terms of recovery were indicated at the future perspectives (see Section 6).

(26)

4.4 Precision

The precision was assessed with nine homogeneous IgG (Sigma, I4506) samples, which were analyzed divided over three days (3 x n=3). The repeatability was expressed in Relative Standard Deviation (RSD) for each day (variance within one day, 3 x n=3) and the intermediate precision was also expressed in RSD, which was derived from all samples together (variance between days, n=9). The RSD values were calculated based on the percentage of each N-glycan. The precision was calculated for the FLR and the MS data separately. The acceptance criteria for the repeatability and intermediate precision were at <10 % and <20 % RSD. In Figure 23, the repeatability of each N-glycan for the FLR and MS data is shown. Additionally, in Figure 24, the intermediate precision of the FLR and MS data is compared. The tables including the RSD values for all individual N-glycans are shown in Section 7.

Figure 23: The repeatability (3 x n=3) of each N-glycan for the FLR and MS data, the red dashed line represents the acceptance

criteria.

Figure 24: The comparison of the intermediate precision (n=9) between the FLR and MS data for each N-glycan, the red dashed

line represents the acceptance criteria.

The repeatability of both FLR and MS data were within the acceptance criteria, as the overall RSD of all N-glycans was below 10 % at each day (see Figure 23). It can be observed that N-glycans containing sialic acid residues have higher RSD values than the N-glycans without a sialic acid residue. The

(27)

N-27 MSc Thesis Carlo de Bruin

glycans was below 20% (see Figure 24). In Figure 25, an overlay is shown of the nine precision samples. The chromatograms are similar to each other, only some small variation in absolute intensity was

observed.

Figure 25: Overlay of the all the chromatograms from the nine IgG samples which were used for the precision tests distributed

over three days.

However, the intermediate precision of the MS data was higher than 20 % RSD and therefore not within the acceptance criteria (see Figure 24). In the MS data, some low abundant compounds (G1-3 & G1F-3+GN) and especially sialic acid containing compounds caused more variation between days. This result is corresponding with the residual errors in the MS data from low amounts of QS (See Section 4.1). Additionally, this could also be due to variating ionization efficiency of these compounds. This result strengthened the decision of using FLR data for the relative quantification and MS the for identification of N-glycans.

In the results of the accuracy, precision and robustness (upcoming results, see Section 4.5), it was found that the seven N-glycans containing sialic acids (N-Acetylneuraminic acid) did not always meet the acceptance criteria. In Section 7, the individual results (% RSD and % accuracy) of sialic acid containing N-glycans for each test are given. These results could be caused by challenges in the chromatographic separation, which results in shoulder/tailing peaks or broader peaks compared to the other N-glycans (see Figure 11). An example of shoulder/tailing peaks from A2F and A2F+GN is shown in Figure 26. This can cause more variation in the peak areas for these compounds, as the peak integration is more complicated and less robust.

Figure 26: An example of the peaks of the sialic acid containing N- glycans. The shoulder/tailing peaks of A2F and A2F+GN

(28)

The peak tailing of these compounds was caused by secondary interactions with the stationary phase, this is probably due to the negative charge of sialic acids groups [26]. However, solutions and suggestions for this challenge are made within the future perspectives, where possible follow up studies are indicated (see Section 6).

4.5 Robustness

For the robustness, three different variations in the sample preparation procedure were evaluated, different IgG concentrations, deglycosylation temperatures and three different sample formulations.

1. The first experiment to test the robustness was the influence of the amount of IgG. Next to the standard amount of 15µg, two other amounts were tested: 7.5µg and 30µg in triplicates (n=3). 2. In the second experiment, the influence of different deglycosylation temperatures on the IgG

samples was evaluated. The standard deglycosylation temperature is 90˚C and the two tested temperatures were 80˚C and 100˚C, all samples were tested in triplicates (n=3).

3. In the third experiment the influence of different drug formulations was tested. The tested drug formulations were: Phosphate Buffered Saline (PBS) pH 7.4, 20 mM histidine, 250 mM sorbitol, 0.04 % polysorbate 80, pH 6.0 (buffer A) and 30 mM sodium acetate, 150 mM D-sorbitol, 0.04 % polysorbate 80, pH 5.5 (buffer B). All samples were tested in triplicates (n=3).

For all these experiments the results were compared with original IgG (Sigma, I4506) samples, which followed the standard sample preparation procedure. The FLR data was used for this tests and the acceptance criteria for the robustness was set at <10 % RSD and 80-120 % accuracy for each N-glycan. The tables including the RSD values for all individual N-glycans and the N-glycan profiles of all variations within each robustness test are shown in Section 7.

In the first robustness test, different IgG amounts (7.5 & 30µg) were tested next to the standard amount (15µg). Although, the variance of the tested amounts (7.5 & 30µg) was higher compared to the variance of the standard amount (15µg), most of the compounds were within the acceptance criteria (<10 % RSD). Also the accuracy was within the acceptance criteria (80-120 %). In Figure 27, an overlay of the chromatograms from the different amounts is shown. In Figure 28, the RSD% and accuracy values of each N-glycan are shown.

(29)

Figure 28: The RSD% (n=3) and accuracy values for each individual N-glycan of robustness test one, the red dashed lines

represent the boundaries of the acceptance criteria.

In Figure 27 & Figure 28, it is observed that the obtained N-glycan profiles are comparable to each other for each of the tested IgG amounts. In addition, the intensity profile of the three amounts indicated a possible linear relationship. Therefore, the total concentration of the detected N-glycans of each IgG amount was compared with each other. This was done by summing the area and integrated maximum intensity of all N-glycans for the FLR and MS data. This was displayed in Figure 29A&B.

Figure 29: A: The summed area of all the detected N-glycans from the FLR data for each tested amount of IgG. B: The summed

integrated maximum intensity of all the detected N-glycans from the MS data for each tested amount of IgG.

In this data, it is observed that the amount of N-glycans in 15µg IgG (100 %) is approximately twice as much as the amount of N-glycans in 7.5µg IgG (52 & 46 %). This was not the case for the 30µg of IgG (152 & 143 %) relative to the 15µg of IgG (100 %). This was probably because, the sample preparation procedure was optimized for an amount of 15 µg IgG, and therefore, some steps of the sample preparation procedure may not be fully effective for an amount of 30 µg IgG. Nevertheless, the accuracy of the N-glycan profile of 30µg IgG was still within the acceptance criteria. This comparison between 7.5 & 15 µg of IgG, strengthens the assumption that all N-glycans were successfully deglycosylated from the IgG, as the absolute intensity (FLR & MS) increased by a factor two.

(30)

To evaluate the labeling efficiency, unlabeled N-glycan standards (6.4 pmol per standard) were directly injected into the LC-FLR-MS setup. The integrated maximum intensity of these peaks were compared with the observed signals of unlabeled N-glycans in the original samples. Only four minor signals of unlabeled N-glycans (G0F, G1F isomers and G2F) were detected in the original samples, which were the most abundant N-glycans in the total N-glycan profile. The signal of unlabeled G0F (Integrated max. intensity= 1.0x104_{) in the original sample was compared with the signal of unlabeled G0F (Integrated}

max. intensity= 1.8x106_{) in the standard based on their integrated maximum intensity. From this ratio it}

was calculated that, the observed signal of unlabeled G0F in the original sample was 0.036 pmol, which corresponds to 2% of the total injected amount of labeled G0F (1.8 pmol) of the original sample. Thus the labeling efficiency of the method is 98% for G0F and it is assumed that this labeling efficiency is

comparable for the other N-glycans. These results confirm the statements of Lauber et al., in their study the deglycosylation and labeling efficiency was optimized and stated to be accurate for complete deglycosylation and labeling of N-glycans [16].

In the second experiment the influence of different deglycosylation temperatures on the IgG samples were evaluated. The standard deglycosylation temperature is 90˚C and the two tested temperatures were 80˚C and 100˚C. In Figure 30, the RSD% and accuracy values of each N-glycan are shown. The tables including the RSD values for all individual N-glycans and the N-glycan profiles of all variations within each robustness test are shown in Section 7.

Figure 30: The RSD% (n=3) and accuracy values for each individual N-glycan of robustness test two, the red dashed lines

Also in this test, the variance within the results of the tested deglycosylation temperatures was compared to the variance of the standard deglycosylation temperature. All the variances within the tested

temperatures were within the acceptance criteria (<10 % RSD), except for some N-glycans containing sialic acids. Also the accuracy was within the acceptance criteria (80-120 %), which means the method is robust for minor deviations in deglycosylation temperatures.

(31)

In the third experiment the influence of different drug formulations was tested. The tested drug formulations were: Phosphate Buffered Saline (PBS) pH 7.4, buffer A (20 mM histidine, 250 mM sorbitol, 0.04 % polysorbate 80, pH 6.0) and buffer B (30 mM sodium acetate, 150 mM D-sorbitol, 0.04 % polysorbate 80, pH 5.5). In Figure 31, the RSD% and accuracy values of each N-glycan are shown. The tables including the RSD values for all individual N-glycans and the N-glycan profiles of all variations within each robustness test are shown in Section 7.

Figure 31: The RSD% (n=3) and accuracy values for each individual N-glycan of robustness test three, the red dashed lines

From Figure 31 it is clearly that these drug formulations do not influence the N-glycan profile, as the variance is comparable or even lower compared to the standard drug formulation (water). The variance and accuracy of the tested drug formulations were within the acceptance criteria (<10 % RSD, 80-120 % accuracy). Although, there were some exceptions for N-glycans that contained sialic acids within the PBS buffer. In the second and third test of the robustness, it was observed that the results are robust against deviations in deglycosylation temperatures and drug formulations. This an important feature of this method, especially for the drug formulations. Because the method is intended to profile Genmab batches, which are mostly formulated in the tested drug formulations. It is assumed that the clean-up step of the sample preparation makes this possible and is therefore a critical step despite of its losses in recovery.

(32)

5. Conclusions

In this study, a LC-FLR-MS method for released RFMS N-glycans was implemented through a method optimization and qualification. During the optimization it was found that HCD fragmentation provides sufficient information for successful identification of released N-glycans and that ETD fragmentation was not useful for this purpose. It was also found that the recovery of the sample clean-up was sensitive for random errors, but has no significant impact on the accuracy of the N-glycan profile. The method was successfully qualified for the relative quantification of the RFMS labeled N-glycan profile from mAbs. Following the ICH guidelines, the following parameters were tested: linearity & range, specificity, accuracy, precision and robustness. During the qualification, the performance of the FLR and MS detectors were compared in multiple parameters. From these results, it was decided to use the FLR data for the relative quantification and the MS data for the identification of N-glycans. A summary of the results is given in Table 7. Compared to the studies in literature, this method was qualified for multiple species of N-glycans. For example, the study of Lim et al. [19], to the best of our knowledge, the only study which described validation of the RFMS method in literature, validated the RFMS method for four glycans.

From the total 19 N-glycans, 12 N-glycans were all within the acceptance criteria and therefore they are suitable for the identification and relative quantification. Within these 12 N-glycans, compounds contained properties such as fucosylation, galactosylation and bisection. On the other hand, the remaining 7 N-glycans were not within all acceptance criteria. These were N-glycans that all were sialylated and are suitable for identification purposes, but not for relative quantification. The N-glycans that contain sialic acid residues comprised of difficulties in separation because of secondary interactions with the stationary phase (see Section 4.4). Nevertheless, this method shows potential to be expanded for more N-glycans. Therefore, other mAbs with different glycosylation profiles should be tested.

Table 7: Summary of the qualification results.

Parameter Results Acceptance criteria

Linearity & Range

- R2_{of 0.9999 (FLR), R}2_{of 0.9932 (MS)}

- Range: 4 – 0.04 pmol, all points < 10 % RSD

- (FLR) LOD: 0.475 pmol N-glycan/100 pmol IgG (= < 1 %) - (FLR) LOQ: 1.425 pmol N-glycan/100 pmol IgG (= < 3 %) - (MS) LOD: 2.375 pmol N-glycan/100 pmol IgG (= > 1 %) - (MS) LOQ: 7.25 pmol N-glycan/100 pmol IgG (= > 3 %)

- R2_{> 0.99}

- < 10 % RSD - LOD < 1 %

N-glycan from total amount IgG

- LOQ < 3 %

N-glycan from total amount IgG

Specificity - Average mass accuracy < 1.2 ppm - MS/MS score > 25

- < 5 ppm - > 10 score

Accuracy - 12 N-glycans without sialic acids: ~108 % average accuracy

- 7 Sialic Acid containing N-glycans: ~70 % average accuracy

- 80-120 % Accuracy

Precision - Repeatability: (FLR) ~3.0 % average RSD,

(MS) ~3.6 % average RSD

- Intermediate precision: (FLR) 4.6 % average RSD, (MS) 22.6 % average RSD

- < 10 % RSD - < 20 % RSD

Robustness - Test 1: variating IgG amount

- Test 2: variating deglycosylation temperatures - Test 3: variating drug formulations

- All below < 10 % RSD and ~99.9 % average accuracy

- < 10 % RSD

(33)

Finally, for mAb characterization studies the RapiFluor (Waters) released N-glycan method was successfully implemented and optimized to be used within the Genmab environment and sample types. Additionally, standardized sample preparation procedure and Genedata Expressionist data analysis workflows were established. Within this method, future improvements and perspectives were indicated (see Section 6). Therefore, this study is a step forward in released N-glycan analysis.

6. Future perspectives

For future research perspectives, possible solutions were proposed for the N-glycans that contain sialic acids. For the used IgG (19 N-glycans), the gradient could be made less steep to optimize the separation of the sialic acid containing compounds. In literature, other separation methods have been found for N-glycans that contain sialic acids. For example the study of Torii et al., in this study major sialylated N-glycans were analyzed where a fractionation was performed with anion exchange and subsequently the fractions were separated on a reversed phase column [26, 27]. Alternatively, another column could be used which is specifically intended for sialic acid containing N-glycans. For example, the Acquity Premier Glycan BEH C18 AX Column with the IonHance Glycan C18 AX Ammonium Formate Concentrate as mobile phase [32]. This is a mixed-mode (anion-exchange/reversed-phase) column that provides a charge based separation and extra resolution for acidic glycans. This might also simplify to distinguish between the different types of sialic acids (N-Acetylneuraminic acid and N-glycolylneuraminic acid). It would be interesting and valuable additional research to investigate these options. Such a complementary addition could help significantly in expanding the range of N-glycans that could be measured.

Some other additional improvements for this method have been identified during method optimization. For instance, while the sample preparation is fast and efficient, its recovery was found just to be 50 %. It was observed that this low recovery was the result of the SPE-clean up step which requires careful execution, to ensure a low flowrate through the SPE-wells. Otherwise, the compounds will flush through the column in the load step and the recovery will be very low ~10 % (data not shown). Nowadays, there are multiple options available to improve the recovery and handling of the clean-up step. The first option is the use of a different SPE elution mechanism, which applies positive pressure with nitrogen instead of negative pressure with a vacuum manifold. This will improve the repeatability and avoids human errors in vacuum control, but an additional nitrogen connection would be needed [33]. Another option is the implication of the Andrew Alliance robot, which is capable to execute the complete clean-up step [28]. This automation will also enhance the precision and discards human errors completely. These improvements could make the method more automated and robust, this could justify the additional implementation costs.

Another interesting research topic within glycan analysis is the distinguishment between isomeric compounds. The chromatographic separation of isomers was achieved, but it was not possible to distinguish them from the MS2_{spectra. For this challenge, some possibilities were indicated and shortly described. In}

the study of Zhao et al., they made use of MSn_{experiments in negative ion mode to distinguish isomeric}

N-glycans based on signature D-ions, which are unique for specific isomers [29]. These ions are generated by cleaving the core N-Acetylglucosamine (GlcNAc, blue boxes) residues and the 3-antenna. Such an in-depth MSn_{study might be able to reveal the isomeric identity of these N-glycans.}

(34)

Additionally, an enzymatic approach is suggested. Enzymes like β-N-Acetylhexosaminidase and α-Mannosidase could aid in the identification of these G1F isomeric variants. The β-N-Acetylhexosaminidase would cleave the terminal N-Acetylglucosamine (GlcNAc) monosaccharides followed by an α-Mannosidase digestion which will cleave a single α1-3 linked mannose from the core ß-mannose, leaving the single α1-6 linked mannose from the core ß-mannose intact. Subsequent N-glycan analysis would reveal the isomeric difference due to change in retention time of the peaks.

7. Supplementary Figures & Tables

In this section, additional Figures and Tables are displayed. The caption of the Figures and Tables explains the data and where it belongs referring to the results. In the tables, the results that are outside the acceptance criteria of >10 % RSD and 80-120 % accuracy were underlined.

Table 8: Samples (Lin_0 – Lin_7) with corresponding amounts in pmol and their peak areas.

Sample Amount of QS

(pmol)

Average Peak Area (FLR) Integrated Max. Intensity (MS) Injection volume (µL) Lin_0 0 0 0 1 Lin_0 0 0 0 1 Lin_0 0 0 0 1

Lin_1 0.04 2.30E+04 3.55E+05 0.1

Lin_1 0.04 2.45E+04 3.79E+05 0.1

Lin_1 0.04 2.43E+04 3.67E+05 0.1

Lin_2 0.1 8.17E+04 1.29E+06 0.25

Lin_2 0.1 8.17E+04 1.39E+06 0.25

Lin_2 0.1 8.16E+04 1.30E+06 0.25

Lin_3 0.2 1.75E+05 3.20E+06 0.5

Lin_3 0.2 1.65E+05 2.96E+06 0.5

Lin_3 0.2 1.76E+05 3.16E+06 0.5

Lin_4 0.4 3.64E+05 7.00E+06 1

Lin_4 0.4 3.62E+05 7.18E+06 1

Lin_4 0.4 3.65E+05 7.03E+06 1

Lin_5 1.0 9.51E+05 2.13E+07 2.5

Lin_5 1.0 9.46E+05 1.99E+07 2.5

Lin_5 1.0 9.69E+05 2.10E+07 2.5

Lin_6 2.0 1.96E+06 4.85E+07 5

Lin_6 2.0 1.95E+06 4.12E+07 5

Lin_6 2.0 1.94E+06 4.29E+07 5

Lin_7 4.0 3.87E+06 9.24E+07 10

Lin_7 4.0 3.90E+06 8.63E+07 10

(35)

Table 9: Precision data of day 1 from the qualification, the relative standard deviations of all N-glycans for both the FLR and MS

data.

Precision Day 1

N-glycan FLR RSD Average RSD N-glycan MS RSD Average RSD

G0 0.8 % 2.4 % G0 0.4 % 3.2 % G0F 0.9 % G0F 1.0 % G0F+GN 1.2 % G0F+GN 1.3 % G1-6 0.9 % G1-6 2.4 % G1-3 0.0 % G1-3 2.1 % G1F-6 1.1 % G1F-6 1.5 % G1F-3 1.0 % G1F-3 2.7 % G1F-6+GN 2.3 % G1F-6+GN 3.6 % G1F-3+GN 2.0 % G1F-3+GN 6.2 % G2 0.7 % G2 1.8 % G2F 1.6 % G2F 2.8 % G2F+GN 0.9 % G2F+GN 3.6 % A1F-Gal 0.5 % A1F-Gal 0.9 % A1 2.1 % A1 6.4 % A1F 0.7 % A1F 1.5 % A1F+GN 4.3 % A1F+GN 6.3 % A2 6.8 % A2 0.5 % A2F 6.9 % A2F 5.2 % A2F+GN 11.1 % A2F+GN 9.6 %

Table 10: Precision data of day 2 from the qualification, the relative standard deviations of all N-glycans for both the FLR and MS