N-glycan Remodeling Using Mannosidase Inhibitors to Increase High-mannose Glycans on Acid α-Glucosidase in Transgenic Rice Cell Cultures

Glycoengineering of plant expression systems is a prerequisite for the production of biopharmaceuticals that are compatible with animal-derived glycoproteins. Large amounts of high-mannose glycans such as Man7GlcNAc2, Man8GlcNAc2, and Man9GlcNAc2 (Man7/8/9), which can be favorably modified by chemical conjugation of mannose-6-phosphate, are desirable for lysosomal enzyme targeting. This study proposed a rice cell-based glycoengineering strategy using two different mannosidase inhibitors, kifunensine (KIF) and swainsonine (SWA), to increase Man7/8/9 glycoforms of recombinant human acid α-glucosidase (rhGAA), which is a therapeutic enzyme for Pompe disease. Response surface methodology was used to investigate the effects of the mannosidase inhibitors and to evaluate the synergistic effect of glycoengineering on rhGAA. Both inhibitors suppressed formation of plant-specific complex and paucimannose type N-glycans. SWA increased hybrid type glycans while KIF significantly increased Man7/8/9. Interestingly, the combination of KIF and SWA more effectively enhanced synthesis of Man7/8/9, especially Man9, than KIF alone. These changes show that SWA in combination with KIF more efficiently inhibited ER α-mannosidase II, resulting in a synergistic effect on synthesis of Man7/8/9. In conclusion, combined KIF and SWA treatment in rice cell culture media can be an effective method for the production of rhGAA displaying dominantly Man7/8/9 glycoforms without genetic manipulation of glycosylation.

A deficiency of acid α-glucosidase (GAA) which hydrolyses glycogen to glucose in the lysosome causes Pompe disease with the accumulation of glycogen in lysosomes. GAA is a glycoprotein of 110 kDa with seven glycosylation sites which is targeted to the by the mannose-6-phosphate (M6P) receptor (M6PR) 4 .
Therefore, to develop therapeutic proteins for lysosomal storage disorders, it is essential to generate terminal M6P glycans on therapeutic proteins for M6PR-mediated pathway 5 , except for Gaucher disease which requires terminal mannose for uptake by the mannose receptor. In the M6PR-mediated pathway, in vitro M6P chemical conjugation depends upon the number of terminal mannoses containing α1,2-glycosidic linkages, which are potential phosphorylation sites 6 .
Glycoengineering could modify glycan structures to improve therapeutic function using various strategies such as genetic engineering or use of glycoprocessing inhibitors 7 . To generate high-mannose (HM) glycans, mannosidases or N-acetylglucosaminyltransferase I (gnt1) must be knocked-down and/or knocked-out in the N-glycosylation biosynthetic pathway 8 . One complementary strategy to genetic engineering techniques is the design of culture media, which can be readily implemented into other glycoprotein expression systems 9 . Especially, glycosidase inhibitors acutely suppress specific N-glycan processing enzymes in the N-glycosylation pathway obtaining targeted glycoforms, and thus there is no need to generate knock-out cell lines for the modulation of N-glycan processing 10 . Kifunensine (KIF) was reported to strongly inhibit α-mannosidase I in the ER (ERMI) and/or cis-Golgi apparatus (GMI) 11 , and swainsonine (SWA) inhibits α-mannosidase II in the ER (ERMII) and/or medial-Golgi (GMII) 12 . Changes in glycan profiling by each inhibitor were reported in mammalian cell cultures 13,14 , but these approaches have not been applied to plant cell culture systems for the production of glycoproteins.
In this study, rice-derived recombinant human acid α-glucosidases (rrhGAA), which are therapeutic enzyme of the Pompe disease requiring M6PR-mediated uptake, were produced in transgenic rice cells. We employed a design of experiment (DoE) approach using response surface methodology to investigate the effects of two inhibitors on HM glycan profiles. To quantify the effects of the inhibitors and evaluate the synergistic effect on glycan profiling of rrhGAA, glycovariants derived from each condition were analyzed by hydrophilic interaction ultra-high-performance liquid chromatography (HILIC-UPLC) with fluorescence detection and matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) after 2-aminobenzamide (2AB) labeling. This study noted the effects of two different mannosidase inhibitors, KIF and SWA, on the glycosylation pathway of rice cells and the production of HM glycans without need for cell line development.

Effects of the combination SWA and KIF on N-glycosylation of rrhGAA derived from wild type.
A combination of KIF and SWA more effectively suppressed early stage N-glycan processing in the ER and Golgi, resulting in elevation of Man9 glycan compared to KIF treatment alone (Figs 2d, 3d and Table S2). While the combination of KIF and SWA led to a moderate increase in the total amount of HM glycans compared to treatment with KIF alone [from 76.8% at 5 µM KIF (#7) to maximum abundance of 81.8% with 5 µM KIF and 10 µM  Table S1). The hybrid type glycans, abundant in SWA-treated rrhGAA, were reduced in the presence of both KIF and SWA.
The changes in Man8 were dependent on the concentration of KIF in the combination groups (#2, #3, #5, #6, #9 and #10). An increasing SWA concentration (from 5 µM to 10 µM) led to a slight decrease in the total amount of Man8 in the presence of 2.5 µM KIF (#2, #5, #10 vs #3) while increasing Man8 in combination with 5 µM KIF (#6 vs #9). Quantification of Man8 isomers revealed that 2.5 µM KIF (#1) dominantly produced Man8 isomer C (8.4%), a similar or slightly less amount of isomer A (8.2%), and a low amount of isomer B (2.4%) ( Fig. 2c and Table S2). This trend in Man8 isomer distribution appeared to be more evident with a higher concentration of KIF at 5 µM (#7), producing 9.3% of Man8 isomer C followed by 7.9% of Man8 isomer A and only 1.8% of Man8 isomer B. Interestingly, additional SWA in KIF-treated cells altered this trend by reducing the relative amount of Man8 isomer C. As a result, Man8 isomer A was the prominent isomer, followed by isomers B and C (Fig. 2d).
Depending on the concentrations of SWA and KIF, the total amount of Man8 isomers varied, ranging from 14.6 to 22.3%. However, the trend in Man8 isomers remained unchanged in the presence of SWA and KIF, as Man8 isomer A was the most abundant (8.2~12.3%), followed by isomers B (3.5~6.0%) and C (3.0~4.4%) (Table S2).
DoE-based approach for analyzing effects of KIF and SWA on the production of rrhGAA with HM glycans in transgenic rice cell cultures. We hypothesized that a combination of KIF and SWA might increase the proportion of HM glycovariants, especially Man7/8/9 which are desired glycan structures containing potential phosphorylation sites for M6P chemical conjugation in rice cell cultures 10 . As shown in Fig. 4, N-glycans derived from SWA alone (#4 and #8) presented relatively low amounts of Man7/8/9, indicating that SWA did not significantly increase Man7/8/9. In contrast, KIF alone at 2.5 µM and 5 µM (#1 and #7) resulted in 63.8% and 65.4% Man7/8/9, respectively, demonstrating that KIF significantly enhanced the proportion of Man7/8/9 in a dose-dependent manner within this range. More interestingly, the other discrete coordinates in the design space, where both KIF and SWA were added, showed further elevation of Man7/8/9 (up to 80.3%), implying an interaction effect of KIF and SWA on Man7/8/9 formation, although SWA itself showed a weak effect on the formation of Man7/8/9 (Table S1). In fact, these contrasting results are derived from the different N-glycan processing enzymes inhibited by KIF and SWA; inhibition of ERMI and GMI by KIF in the ER and/or cis-Golgi and inhibition of ERMII and GMII by SWA in the ER and/or medial-Golgi. The combined treatment could provide a synergistic effect on enhancing formation of Man7/8/9, especially Man9. Based on the results of the experimental design, specifically Man7/8/9 contents, a response surface model was constructed into the contour plots with input variables (KIF and SWA) and output response (Man7/8/9 contents). These results including contour plot (Fig. 5a) and scaled estimates of main effects on Man7/8/9 (Table S3) revealed that KIF was more influential on Man7/8/9 than SWA. Interestingly, the interaction term (KIF*SWA, p = 0.023) appeared to be statistically significant, indicating a considerable synergistic effect. From the regression model, optimal concentrations of KIF and SWA (3.84 µM KIF and 8.20 µM SWA) achieving the maximum level of Man7/8/9 (88.1 ± 2.0%) were calculated within the experimental ranges ( Fig. 5c and Table S4). To validate these optimal values of input and output, additional cultures were conducted with both optimal point (combination of 3.84 µM KIF and 8.20 µM SWA) and control condition (without mannosidase inhibitors) ( Fig. 6 and Table S5). While the measured content of Man7/8/9 at optimal point (91.3%) was slightly higher than expected content (88.1 ± 2.0%), the overall glycan profiles were consistent with the expected ones from the regression model, thus providing the statistical confidence of response surface methodology for media optimization. More importantly, purified rrhGAAs in both conditions showed similar levels of in vitro enzymatic activity compared with Myozyme (Fig. S1), demonstrating that the treatment of mannosidase inhibitors in rice cell cultures to modulate N-glycan structure of rrhGAA may not significantly affect functional activity of rrhGAA.

Discussion
Herein, we investigated production media using two different mannosidase inhibitors, KIF and SWA, for N-glycosylation control in rrhGAA in rice cell cultures. To quantify the effects of the inhibitors, a DoE approach was adopted in this study. Response surface models for Man7/8/9 contents in rrhGAA was established using KIF and SWA as input variables, thus enabling prediction of optimal concentrations of KIF and SWA to achieve the target level of Man7/8/9 with statistical confidence and further understanding of the interactions among the inhibitors behind the relevant mechanism.
Most N-glycan in rrhGAA produced in the wild-type control (#11) was fully processed in the Golgi apparatus and vacuole, resulting in abundant plant specific complex type and paucimannose type glycans. KIF treatment conditions (#1 and #7) exhibited large amounts of HM glycans, including primarily Man9 followed by Man8, indicating strong inhibition of ERMI and/or GMI in the N-glycosylation pathway. KIF treatment yielded smaller amounts of Man8 isomer B, a product of ERMI, than Man8 isomers A and C, which can be rapidly degraded by GMI in wild-type rice cells 15 . The increased concentration of KIF from 2.5 µM to 5 µM elevated Man9 levels and reduced Man8 isomer B level. There was a minor amount of hybrid type glycans (1.9~2.0%), confirming that KIF can act as a weak inhibitor of GMII 16 . On the other hand, SWA conditions (#4 and #8) resulted in mainly hybrid type glycans and HM glycans composed of Man6 and Man7, possibly due to potent GMII inhibition and minor ERMII inhibition. This study demonstrated for the first time that the combination of KIF and SWA (#2, #3, #5, #6, #9 and #10) strongly inhibited the mannose trimming pathway related to ERMI, GMI, and ERMII in the N-glycosylation pathway of rice cells and led to dramatic increases in Man7/8/9 of rrhGAA derived from rice cells (up to 80.3%). The amount of Man9 was most prominently affected among HM glycans in response to the combination of SWA and KIF compared to SWA or KIF treatment alone. As high contents of Man7/8/9 glycans are critical for efficient in vitro M6P chemical conjugation to enhance M6PR-mediated uptake pathway 17 , this result is valuable for improving the quality of the rrhGAA producing-rice cell culture process. Interestingly, addition of SWA to rice cell culture in the presence of KIF altered the relative distribution of Man8 isomers. The amount of Man8 isomer C, which is the major structure among Man8 isomers produced by KIF alone (Fig. 2c), decreased while the amount of Man8 isomer A increased (Fig. 2d). Man8 isomer A, the product of endomannosidase, is able to circumvent the glycosylation process inhibited by SWA and KIF 15 . This result provides evidence that SWA could act as a potent ERMII inhibitor only when treated along with KIF, as SWA treatment alone was insufficient to block ERMII. In addition, hybrid type glycans, which were the most abundant glycan species in SWA alone, accounted for only 1.8~3.8% of rrhGAA obtained from the combination of SWA and KIF. It is highly possible that the N-glycosylation process was more efficiently arrested at Man8 and Man9 in the ER by the combination of SWA and KIF, suggesting that SWA is not effective in producing hybrid type glycans in the Golgi apparatus.
Despite the stronger effect of KIF on HM glycan formation, KIF treatment in rice cell cultures had a negative effect on culture performance parameters such as cell mass, cell viability, and rrhGAA titer. In contrast, SWA did not significantly impair these cell culture profiles or HM glycan formation. Although these combination conditions had a negative effect on cell culture performance, these impacts were not worse compared to KIF alone. We found that the combination of KIF and SWA dramatically increased HM glycan contents on rrhGAA compared to each of the inhibitors alone, which has not been previously reported. Interestingly, this approach successfully eliminated plant-specific glycan variants in rrhGAA, thus overcoming the major drawback of plant cell culture systems for mammalian protein production. Moreover, detailed analysis of glycovariants enabled to understand the N-glycosylation pathway in callus-derived suspension rice cells (O. sativa), thereby providing more information on recombinant protein development in the rice cell culture platform (Fig. 7). Compared to the genetic manipulation of the rice host cell lines for desired glycan profile of the products requiring considerable time and efforts, this strategy is easily and immediately applicable to the rice cell culture process and thus more feasible in the biopharmaceutical industry. While knock-out approach such as gnt1 mutant used in this study may be efficient to modulate phenotypes in on/off mode, media optimization approach with specific enzyme inhibitors can offer more rapid and flexible operating space identification depending on the target quality attributes, thus Experimental designs and statistical analysis. Determination of the concentration ranges of mannosidase inhibitors was carried out based on previous literature 11,20 , and experiments were designed based on the combinatorial structure of the face-centered response model using JMP 11 software (SAS Institute). Design of the experiment along with values of HM glycan contents and titer is given in Table S1. Experiments and sample analysis were carried out in triplicates, and data from cell culture profiles was expressed as the mean of triplicate  (Table S4). Quantification of rrhGAA. Indirect enzyme-linked immunosorbent assay was used for quantification of rrhGAA in media. As a standard substance, Myozyme ® (Genzyme, Cambridge, MA, USA) was diluted to prepare standard solutions ranging from 10 to 100 ng/mL in bicarbonate buffer solution (15 mM Na 2 CO 3 and 35 mM NaHCO 3 ; pH 9.6). Culture media containing rrhGAA were diluted from 1:1000 to 1:2000 in bicarbonate buffer. Standard and samples were loaded in a 96-well microplate (Sigma-Aldrich) and coated overnight at 4 °C. Custom rabbit anti-GAA polyclonal antibody (Abfrontier, Seoul, Republic of Korea) was used as a primary antibody and loaded with 100 µL/well of the 1:2000 dilution. Peroxidase-labeled goat anti-rabbit IgG (Thermo Fisher Scientific, Waltham, MA, USA) was used to detect primary antibody and loaded with 100 µL/well of the 1:5000 dilution. Development step was performed using 100 µL/well of ultra tetramethylbenzidine (Thermo Fisher Scientific) and then stopped by 2 M sulfuric acid. The absorbance was measured by microplate reader (Thermo Fisher Scientific) at 405 nm.
Purification of rrhGAA. Suspension cultured media were filtered using a 0.22 µm Millipore Express ® PLUS membrane filter (Millipore, Volketswil, Switzerland). His-tagged rrhGAAs in the filtrate were purified using Ni-NTA Superflow Cartridges (Qiagen, Valencia, CA, USA) according to the manufacturer's protocols. Confirmation of purified rrhGAA was performed by SDS-PAGE analysis (Fig. S2).

MALDI-TOF MS analysis of N-glycans.
MALDI-TOF MS was performed with the Ultraflex III system (Bruker Daltonics, Bremen, Germany) controlled by Flex Control 3.0 (Bruker Daltonics). The mass spectra were measured with reflectron positive ionization mode within 500-3,000 Da. For all analyses, 2,5-dihydroxybenzoic acid (Sigma-Aldrich) was used as the matrix at a concentration of 10 mg/mL in acetonitrile/water (50:50, v/v). Dried 2AB-labeled glycans were dissolved in water, and each sample solution (1 μL) was mixed with appropriate matrix solution (1 μL). These samples were then loaded onto the stainless steel target and dried at room temperature.