Aptamers as quality control tool for production, storage and biosimilarity of the anti-CD20 biopharmaceutical rituximab

Detailed analysis of biopharmaceuticals is crucial for safety, efficacy and stability. Aptamers, which are folded, single-stranded oligonucleotides, can be used as surrogate antibodies to detect subtle conformational changes. We aimed to generate and assess DNA aptamers against the therapeutic anti-CD20 antibody rituximab. Six rituximab-specific aptamers with Kd = 354–887 nM were obtained using the magnetic bead-based systematic evolution of ligands by exponential enrichment (SELEX) technology. Aptamer folds were analysed by online prediction tools and circular dichroism spectroscopy suggesting quadruplex structures for two aptamers while others present B-DNA helices. Aptamer binding and robustness with respect to minor differences in buffer composition or aptamer folding were verified in the enzyme-linked apta-sorbent assay. Five aptamers showed exclusive specificity to the Fab-fragment of rituximab while one aptamer revealed a broader recognition pattern to other monoclonal antibodies. Structural differences upon incubation at 40 °C for 72 h or UV exposure of rituximab were uncovered by four aptamers. High similarity between rituximab originator and biosimilar lots was demonstrated. The most sensitive aptamer (RA2) detected signal changes for all lots of a copy product suggesting conformational differences. For the first time, a panel of rituximab-specific aptamers was generated allowing the assessment of conformational coherence during production, storage, and biosimilarity of different products.


Results
In vitro selection of rituximab specific DNA aptamers. A DNA-library consisting of 10 15 different single-stranded oligonucleotides with a random part of 40 nucleotides in length was used for selection of aptamers against the therapeutic IgG1 antibody rituximab. In vitro selection was performed by eight recurring incubations of rituximab-coated protein A magnetic beads using the folded single stranded oligonucleotides (Fig. 1a). Stringency of the SELEX process was increased in the last selection rounds by decreasing the amount of DNA incubated with the beads, increasing the number of washing steps and decreasing the number of PCR cycles. Additionally, a negative selection round was carried out with uncoated protein A magnetic beads before the last cycle. After cloning the DNA fragments of SELEX cycle eight into a cloning vector, plasmids from 50 clones were obtained and sequenced. Analysis of the 40 nucleotide random part revealed fifteen different rituximab aptamer (RA) sequences with occurrences ranging from 1-12 times (Supplementary Table S1).

Six aptamers efficiently bound to rituximab. For the first binding experiments, sequences occurring
≥2 (RA1-9) were obtained as 5′-biotinylated aptamers and assessed using an enzyme-linked apta-sorbent assay (ELASA). For this, biotinylated aptamers were incubated with protein A immobilised rituximab and the interaction was detected using streptavidin-HRP (Fig. 1b). Highest signals were observed with RA2 and RA4, binding capability was medium for RA1 and RA3, while RA5 and RA6 showed lower binding (Fig. 1c). RA7-9 demonstrated very low signals and those sequences were therefore not further pursued in this study. Titrations of RA1-6 ranging from 1.95-1000 nM were performed to identify optimal working concentrations. All aptamers bound to rituximab in a dose-dependent manner, though with different kinetics (Supplementary Fig. S1). Based on these results, aptamer concentrations of 25-50 nM were found to efficiently detect rituximab in the ELASA and were consequently employed in subsequent experiments. Aptamer affinity was measured by surface acoustic wave and K d values ranging from 354-887 nM were experimentally determined (Table 1).
In silico prediction of aptamer folds. It is well established that aptamer stability depends on nucleotide composition and resulting three-dimensional structure. In specific, an elevated GC-content was noted for RA1, RA5, and RA6 (Table 1). In silico analysis of the sequences was performed using the online tool Mfold and most probable structures with the lowest free energy (dG) were selected and are displayed in Fig. 2a. These predictions suggested hairpin structures of varying length and loop formations with lowest energies of folding noted for RA1 (dG = −4.93 J) and RA4 (dG = −2.62 J). In addition, the capacity to form G-quadruplex structures was studied using the web-based QGRS Mapper program. Interestingly, results indicated that all aptamers except RA4 could adopt G-quadruplex structures and suggested guanine bases involved in G-tetrads are highlighted in Table 1 Based on the calculated G-score values which indicate the probability of a G-rich sequence to form a quadruplex structure, rituximab aptamers reached values of 18-20 out of 105 maximum score as a consequence of four times staples of two guanines. Folded biotinylated aptamers (RA1-RA9, c = 500 nM) were tested for rituximab binding. Bound aptamers were detected by streptavidin-HRP, chemiluminescence ELISA substrate was used for detection and luminescence was measured. Measurements were performed in six replicates; means and standard errors of the mean are given. Circular dichroism spectroscopy indicates B-DNA and quadruplex structures of aptamers. The secondary structure was further experimentally investigated using circular dichroism spectroscopy (Fig. 2b). For this experiment, non-biotinylated aptamers were folded in binding buffer before analysis. CD spectra of aptamers RA1, RA2, RA4 and RA5 were rather similar showing a positive maximum around 280 nm and a negative minimum around 245 nm indicative of a B-DNA helix 30,31 . In contrast, the CD spectrum of RA3 revealed a positive maximum at 295 nm and a negative minimum at 260 nm, characteristic for an anti-parallel quadruplex structure. A typical parallel quadruplex structure with a positive maximum at 260 nm was observed for RA6 32,33 . Aptamers demonstrate robustness to minor changes in buffer composition and aptamer folding.
To investigate the robustness of aptamers within our ELASA setup, different aptamer buffers and folding conditions were analysed. Experiments were exemplarily carried out with RA2 and RA4, which showed highest binding affinities. To determine the influence of buffer composition and potential differences during preparation (e.g. weighing errors), a set of different buffers varying in preparation or ionic strength were assembled (Fig. 3a). Minor to moderate differences were observed when buffers were prepared in different ways or by another researcher (Fig. 3b). Minor variations in ionic strength showed no significant effect on aptamer binding. However, omitting sodium chloride triggered a considerable increase in detection signal, while potassium chloride led only to a moderate enhancement for RA2, a result of unspecific interaction of negatively charged DNA with positively charged protein surface areas. In addition, the influence of aptamer folding conditions was investigated. Aptamers were typically denatured at 94 °C for 8 min, followed by a cooling step on ice for 15 min and a step at ambient temperature for 10 min (designated as original protocol). To test the robustness of aptamer folding, denaturation temperature, time as well as the subsequent cooling steps were amended (Fig. 4a). Generally, the original protocol resulted in highest signals but only minor to moderate signal decreases were found for the other folding protocols (Fig. 4b). Highest impact was detected upon an extended denaturation step for 12 min at 94 °C. Interestingly, completely omitting the folding step did not result in failure to bind rituximab, but showed only decreased binding, similar to those aptamers folded under different conditions. RA1-5 present high specificity for rituximab. To investigate aptamer specificity, rituximab, Fc fragments and other biopharmaceuticals were tested in ELASA. Two different Fc fragments of rituximab were analysed, (i) glycosylated Fc obtained after FabRICATOR (IdeS) digest and (ii) recombinant non-glycosylated Fc/2 obtained from E. coli 34 . In addition, adalimumab (IgG1), bevacizumab (IgG1), and etanercept (Fc part of IgG1) were investigated. All aptamers except RA6 exclusively recognised full-length rituximab pointing at their high specificity. RA6 also efficiently recognised adalimumab; low but positive reactivity was also observed for bavacizumab and the glycosylated Fc fragment (Fig. 5).
Aptamers were able to detect changes induced by thermal stress and UV exposure. In order to assess the use of aptamers as potential tools for quality control during protein production and storage, differently stressed rituximab samples were investigated by ELASA. For this purpose, aliquots of rituximab were stored at different temperatures, exposed to UV light or stressed mechanically. All aptamers except RA1 and RA6 were able to detect differences between the untreated and treated rituximab samples (Fig. 6). Storage of proteins at 40 °C    for 72 hours led to reduced binding of RA2-RA4, while signals remained similar when kept at RT or 4 °C or 24 °C for the same time. Interestingly, UV exposure of rituximab resulted in increased but also decreased signals while mechanical stress (vortexing) and repeated freeze-thaw cycles did not considerably influence aptamer binding.

Investigation of different rituximab biopharmaceuticals.
Rituximab aptamers were further employed to investigate the similarity between different batches of the originator (MabThera, Roche), a proposed biosimilar (GP2013, Sandoz) as well as a copy product (Reditux, Dr. Reddys) (Fig. 7). The assay was performed in triplicates on independent days and given values represent the mean of six replicates of each measurement (Fig. 7). For statistical analysis, all rituximab samples were compared with lot 1 from MabThera using one-way ANOVA. Generally, a high degree of similarity was found for the different rituximab products. However, while RA1, RA3-6 showed no statistically significant difference, RA2 was able to reveal a difference in the signals towards the copy product Reditux. Slightly increased signal intensities were also observed for different lots of MabThera and the proposed biosimilar GP2013, which were however beyond statistical significance.

Discussion
Biologics or biopharmaceuticals represent an important branch of the pharmaceutical industry. Patents for some well-established biopharmaceuticals have expired or are approaching expiration. Currently, there are 38 approved biosimilars in the EU and 7 in the US [35][36][37] . Several technologies are accepted for characterisation of the physicochemical and biological properties of biosimilars 17 . However, methods to investigate and compare the tertiary structure of proteins are rather limited and laborious. This encouraged us to employ aptamers for monitoring conformational similarities between different biologics and as a quality control tool for the detection of differently treated samples. In our study, we identified the first panel of aptamers reactive against the monoclonal anti-CD20 antibody rituximab using the in vitro selection process FluMag SELEX 38 . Most studies Figure 6. Analysis of differently stressed rituximab using ELASA. Prior to protein A binding, aliquots were exposed to different temperatures for 72 hours, UV light or mechanical stress. Mean of triplicates and standard error of mean are depicted. use oligonucleotide libraries with a length of 20-80 nucleotides; for our study we chose a 40 nucleotide random part to obtain stable structures with high affinity to the protein target 15,39,40 . For selection of specific aptamers, rituximab was immobilised on the surface of magnetic beads. Several immobilisation methods like capillary electrophoresis, flow cytometry, or electrophoretic mobility shift assays are available. The use of magnetic beads is however easy to handle, adaptable for many kinds of targets and requires only small amounts of target 15,[41][42][43] . To avoid potential loss of aptamers during initial counter selection, we chose to perform this step later based on an optimized protocol for FluMag SELEX by Stoltenburg et al. 38 .
In an initial step, binding of nine aptamers with ≥2 occurrence was investigated by ELASA 44 . Even though all nine tested aptamers showed similar sequence occurrences of 2-12 times, RA7-9 were negative in binding properties suggesting that their high representation after cloning might be due to an amplification bias during PCR. Indeed, it was shown that some motifs (e.g. GC-rich) are preferred by DNA polymerases, especially when several templates are amplified simultaneously 45,46 . The aim of our study was the generation of relatively small aptamers consisting of 40 nucleotides. Therefore and owing to the fact that recent studies suggested limited influence on aptamer functionality upon primer binding site truncation, our initial ELASA-based studies were carried out using only the random sequence part 39,47,48 . In the literature, little influence of 5′ or 3′-labelling was noted and the impact of biotinylation was found to be lower compared to other labellings 48,49 . Using protein A for immobilisation of the Fc antibody part considerably enhanced the replicate reproducibility in the assay compared to direct coating of rituximab 49 . In summary, we were able to select six aptamers reactive to rituximab showing K d levels in the upper nanomolar range. GC-rich sequences were proposed to stabilise DNA structures and all aptamers except RA2 showed elevated contents 50 . However, sequences were especially enriched in repetitive blocks of two guanine bases, and such motifs are able to form G-quadruplex structures [50][51][52] . Indeed, in silico prediction and more importantly experimental CD data concurrently suggest an anti-parallel and a parallel quadruplex structure for RA3 and RA6, respectively 32,33 . The predicted quadruplex structures for the other aptamers (except RA4) could not be experimentally confirmed by CD which rather suggested a B-DNA form 30,31 . In general, formation of quadruplex structures strongly depends on buffer composition, in particular the concentration of cations (Na + , K + , Mg 2+ ) and pH. Generally, increasing the Na + and K + concentration leads to a stronger CD signal of anti-parallel quadruplex structures, and Mg 2+ cations have a positive effect on the stability of the aptamer structure 53 . For consistency, we analysed the aptamers in the same buffer as used for aptamer folding in the ELASA experiments. Besides, secondary structures were also predicted using Mfold, which suggested various stem-loop structures. Of note, negative folding energy showed very low values for RA3 and RA6, which is in line with our experimental data revealing both as G-quadruplex structures. Interestingly, also a low dG was observed for RA2 even though it shows highest binding affinity. Tools for nucleotide fold prediction typically rely on classical base pairing while alternative structures like G-quadruplexes seem currently challenging to adequately predict. To date, the most reliable method is (co)-crystallisation of aptamers but there is still limited experimental data available due to the resource intense procedure 54 . Once tools have enough reference data for accurate prediction of 3-dimensional structures analogous to protein modelling, there is enormous potential for the aptamer technology in terms of a priori selection and design 55 .
Biopharmaceuticals need to fulfil stringent quality standards for approval by the authorities. Methods used to assess the product quality thus need to demonstrate sufficient robustness regarding different buffer batches, days, laboratories or analysts 56 . Based on the ELASA results, we conclude that the herein presented setup with aptamers results in comparable reading outputs with respect to minor differences in buffer composition and folding conditions. It has to be noted that -analogous to ELISA experiments -most reliable results are obtained when comparing different analytes using identical buffer conditions applied within the very same assay. Especially RA2 showed high robustness if minor weighing mistakes of chemicals were considered. Completely omitting NaCl, and to a lesser extent KCl, led to a tremendous increase in signal which is however due to unspecific binding where negatively loaded phosphate groups of the DNA can bind to positively charged protein surfaces irrespective of the aptamer sequence 57,58 .
Five aptamers exclusively recognised rituximab suggesting an interaction within the antigen binding sites. Solely RA6 additionally bound to adalimumab and bevacizumab as well as to the glycosylated Fc fragment of rituximab. Constant regions of IgG antibody subclasses are highly conserved 59,60 , and indeed rituximab and investigated antibodies' sequence identities of constant domains range from 98-100%, while variable region identities are only between 44-62%. As adalimumab was well recognised by RA6, either a conserved stretch within the CH1/CL domain or the hinge region might be a potential interaction site. Although the amino acid sequences are highly similar, the hinge region shows high structural flexibility and could therefore also lead to different epitopes 60 .
Development and production of biologics are highly complex processes, where changes in the manufacturing process can impact the structure and/or function of a protein. A panel of powerful tools is used to characterise such products, but minor conformational changes are often difficult to detect by bioanalytical methods 10 . In a pioneer study by Zichel et al., aptamers were able to detect differences between native and heat-treated thrombin, whereas antibodies failed to reveal them 17 . In our study, four rituximab aptamers showed differences predominately in binding to samples stored at 40 °C or exposed to UV light. As those stress conditions led to different recognition pattern, the involvement of distinct rituximab epitopes can be anticipated. From a quality control point of view, RA2, RA3 and RA4 seem adequate as an orthogonal method to detect changes during production and storage. In comparison to antibodies, large numbers of aptamers can be generated against any given target as immunogenicity is no prerequisite for development. With respect to rituximab, only a limited number of anti-idiotypic antibodies are commercially available (www.abnova.com; www.biorad.com). Advantages of aptamers are the in vitro selection process, chemical synthesis avoiding batch-to-batch variability of antibodies, introduction of modifications and functional groups as well as their cost-efficiency 61 . In contrast to antibodies, aptamers can easily re-adopt their fold upon denaturation without loss of activity.
Aptamers could thus also be implemented as additional analytical method for development and approval of biosimilars 17 . For approval of biosimilars, high similarity to the originator molecule in terms of safety and efficacy needs to be demonstrated 62,63 . Besides an array of analytical methods 3 , there is still a gap in rapid and routine monitoring of minor conformational changes in biologics that might not be detectable by circular dichroism or Fourier transform infrared spectroscopy. Using six aptamers in our ELASA setup, no significant differences between the originator MabThera and the proposed biosimilar candidate GP2013 were detectable supporting the high similarity of the products. Interestingly, RA2 coherently revealed increased signals for all three batches of Reditux, the "copy product" or "similar biologic" from the Indian market in three independent experiments. Mass spectrometry based analyses of these products revealed only little differences between Reditux and the originator molecule MabThera apart from lysine variants at the C-terminus (Supplementary Fig. S2). However, within the Fab fragment, only minor discrepancies beyond statistical significance were identified for Reditux mostly involving pyro-glutamate formation, a common modification observed in mAbs involving charge variants 64 . As aptamers specifically recognize the Fab fragment of rituximab, the different signals of RA2 towards Reditux indicate recognition of structural variations that could not be resolved by mass spectrometry. In this respect, it should be noted that the regulatory process for biosimilar approval in India differs from the process in the EU or US 65 . Reditux was already approved 2007 in India, comparison of efficacy and safety regarding MabThera and Reditux and phamacokinetic studies have been performed thereafter 66,67 . We now showed for the first time that aptamers are also suitable for comparison of originator and biosimilars in their native conformation. In specific, Scientific RepoRts | (2019) 9:1111 | https://doi.org/10.1038/s41598-018-37624-1 RA2 can be regarded as a highly specific and robust aptamer revealing differences upon prolonged storage or UV exposure and is able to detect conformational variations in native biopharmaceutical products. In summary, the first aptamer panel against the therapeutic antibody rituximab was generated. Based on our results, we suggest including the aptamer technology as orthogonal analytical approach in the portfolio of analytical techniques for characterisation of biologics.

Material and Methods
Immobilisation of rituximab to protein A magnetic beads. 700 µl (1.89 × 10 9 beads) of protein A magnetic beads (Dynabeads ® Protein A, Life Technologies AS, Oslo, Norway) were washed with 500 µl 137 mM NaCl, 2.7 mM KCl, 10 mM Na 2 HPO4, and 1.8 mM KH 2 PO4, 0.05% v/v Tween 20 (PBST). 240 µg of rituximab (MabThera, Roche, Basel, Switzerland) in 700 µl PBST were added and incubated for 4 h at room temperature with gentle rotation. The supernatant was removed and beads were washed 3 x with PBST. Binding efficiency of rituximab to the magnetic beads was verified by reducing sodium dodecyl sulfate polyacrylamide gel electrophoresis.   68 . Cycles were repeated six times, whereby stringency was enhanced by decreasing the amount of DNA for binding to rituximab to half, doubling the number of washing steps and decreasing PCR cycle numbers to 20. In cycle seven, a counter selection against protein A magnetic beads was performed to avoid unspecific binding. After this, a final cycle with the rituximab coated beads was performed.

In vitro
Cloning and sequencing. Selected oligonucleotides from cycle eight were amplified with the forward and unmodified reverse primer (5′-ATACCAGCTTATTCAATT-3′) and GoTaq ® DNA Polymerase (Promega, Madison, USA). The PCR product was cloned into the pGEM ® -T easy vector (Promega) and transformed into the E. coli NovaBlue. Fifty colonies were selected, plasmid DNA was purified using EZ-10 spin column plasmid DNA Miniprep Kit (Bio Basic Inc., NY, USA) and the aptamer sequence of each clone determined (Eurofins Genomics, Ebersberg, Germany). DNA sequences were analysed with Clustal Omega 69 . To keep aptamers short, nine selected aptamer sequences without primer binding regions were synthesised as non-labelled and 5′-biotin-lablled oligonucleotides (Eurofins Genomics).
Robustness test of the eLAsA setup. Ten varying buffer and ten folding conditions were used to investigate the robustness of the ELASA setup (for details see Figures 3a and 4a). RA2 and RA4 were used at a concentration of 25 nM and the ELASA was performed as described above.
Binding analysis of rituximab, Fc fragments and other biologics. Assay conditions of the ELASA were carried out as described above. The following proteins were included in the assay and captured by the coated protein A: rituximab (MabThera lot N7075B10, Roche), glycosylated Fc fragment of rituximab (MabThera lot N7075B10, Roche) were prepared using the FragIT ™ Kit (Genovis, Lund, Sweden) according to the provided protocol, and Fc/2 fragment was obtained from E. coli as described elsewhere 34 . In addition, the IgG1 antibodies adalimumab (Humira, AbbeVie Ltd., North Chicago, US), bevacizumab (Avastin, Roche), and etanercept (Enbrel, Amgen, Thousand Oaks, US) an antibody containing the Fc part of an IgG1 antibody were tested. Aptamers RA1, RA3, RA5 and RA6 were used at a concentration of 50 nM and aptamers RA2 and RA4 at 25 nM, respectively.
Analysis of stressed rituximab samples using an eLAsA. To simulate protein stress conditions, aliquots of rituximab (MabThera, lot N7075B10) were treated as follows: ten freeze/thaw cycles, incubation at 4 °C, 25 °C and 40 °C for 72 h, exposure to UV light for 30 seconds or two minutes on an UV light box (setting 100%) and vortexing 10 times for 1 min. The respective samples were captured by protein A and analysed using the different aptamers at a concentration of 50 nM using the protocol described above.
Comparison of MabThera, GP2013 and Reditux. Samples of the rituximab originator (MabThera, lot 1 N7075B10, lot 2 N7025B04, Roche, Basel Switzerland), the biosimilar GP2013_DP (lot 1 EH7223_ID_12_7 and lot 2 EU8142_ID_11_7, Sandoz, Basel, Switzerland) and the copy product Reditux (lot 1 RIAV2416, lot 2 RIAV2616, lot 3 RIAV0517, Dr. Reddy's Laboratories, Hyderabad, India) were captured by the coated protein A and analysed using the different aptamers. All biologics were formulated in 25 mM sodium citrate, 0.5 mM polysorbate 80, 154 mM NaCl, pH 6.5 at a concentration of 10 mg/ml. The ELASA setup was as described above; RA1, RA3, RA5 and RA6 were used at a concentration of 50 nM and aptamers RA2 and RA4 at 25 nM, respectively. Three independent assays were performed and results were normalised for MabThera lot 1 as 100% value.
statistical analyses. Statistical analyses were performed in GraphPad Prism 7.03. A one-way ANOVA with corrections for multiple comparisons (Dunnett) to the control column (MabThera originator lot 1) was performed. Results with p ≤ 0.05 were considered statistically significant.

Data Availability
All data generated or analysed during this study are included in this published article (and its Supplementary Information files).