Proteolytic and N-Glycan Processing of Human α1-Antitrypsin Expressed in Nicotiana benthamiana1[C][W][OPEN]

Eliminating β-hexosaminidase3 activity and terminal sialylation yields a human protease inhibitor that resembles the serum-derived counterpart. Plants are increasingly being used as an expression system for complex recombinant proteins. However, our limited knowledge of the intrinsic factors that act along the secretory pathway, which may compromise product integrity, renders process design difficult in some cases. Here, we pursued the recombinant expression of the human protease inhibitor α1-antitrypsin (A1AT) in Nicotiana benthamiana. This serum protein undergoes intensive posttranslational modifications. Unusually high levels of recombinant A1AT were expressed in leaves (up to 6 mg g−1 of leaf material) in two forms: full-length A1AT located in the endoplasmic reticulum displaying inhibitory activity, and secreted A1AT processed in the reactive center loop, thus rendering it unable to interact with target proteinases. We found that the terminal protein processing is most likely a consequence of the intrinsic function of A1AT (i.e. its interaction with proteases [most likely serine proteases] along the secretory pathway). Secreted A1AT carried vacuolar-type paucimannosidic N-glycans generated by the activity of hexosaminidases located in the apoplast/plasma membrane. Notwithstanding, an intensive glycoengineering approach led to secreted A1AT carrying sialylated N-glycan structures largely resembling its serum-derived counterpart. In summary, we elucidate unique insights in plant glycosylation processes and show important aspects of postendoplasmic reticulum protein processing in plants.

Recombinant protein-based drugs are among the fastest growing areas of development in the pharmaceutical industry. Consequently, there is a demand for exploring new production systems. Plants are increasingly being used for the expression of recombinant proteins, primarily because of their remarkable production speed and yield (for review, see Gleba et al., 2014). The highly conserved secretory pathway between human and plant cells allows similar, if not identical, protein folding, assembly, and posttranslational modifications. Importantly, plants are able to synthesize complex N-glycans, a prerequisite for the in vivo activity of many therapeutically interesting proteins. Despite substantial differences in N-glycan diversity, we and others have shown that plants are highly amendable to glycan engineering and allow proteins with controlled human-type N-glycosylation profiles to be generated . Moreover, it has even been possible to reconstruct entire human glycosylation pathways, which was shown by the introduction of the human sialylation and O-glycosylation processes in Nicotiana benthamiana (Castilho et al., 2010. These accomplishments render plants suitable for the production of human proteins that require a complex glycosylation profile. Notwithstanding, to use plants as a versatile expression host for complex human proteins, it is important to fully understand intracellular processes. Particularly detailed knowledge about constraints along the plant cell secretory pathway, including proteolytic processing, is required, because these constraints may compromise protein integrity and quality. Despite major achievements in controlling protein-bound oligosaccharide formation, some plant glycosylation peculiarities are not entirely understood. For example, plant cells synthesize so-called paucimannosidic N-glycans, a type of truncated glycans usually absent in mammals (Lerouge et al., 1998). The biosynthesis and physiological significance of this N-glycan formation has yet to be completely explained (Strasser et al., 2007;Liebminger et al., 2011). Another process not fully understood in plants is subcellular localization of proteins. Aberrant intracellular deposition and as a consequence, incorrect glycosylation of recombinant proteins are often reported. For example, recombinant proteins designed for secretion are frequently also located in the endoplasmic reticulum (ER) and as a consequence, carry oligomannosidic carbohydrates instead of the desired complex-type glycans (Loos et al., 2011;Schneider et al., 2014a). By contrast, KDEL-tagged proteins designed for ER retention are sometimes partially secreted (Van Droogenbroeck et al., 2007;Niemer et al., 2014). How and at which biosynthetic stage these plant-specific peculiarities arise are largely unpredictable, which makes controlled expression of recombinant proteins with features authentically to their natural counterparts a difficult task.
One human protein that is pharmaceutically interesting, and thus needed in large amounts at high quality, is a1-antitrypsin (A1AT). This highly glycosylated protease inhibitor from the serpin superfamily interacts with a wide variety of proteases (Gettins, 2002). Like other serpins, A1AT is characterized by an exposed and mobile reactive center loop (RCL) with a Met ( 358 M) residue acting as bait for specific target proteinases (Travis and Salvesen, 1983). The main biological role of plasma A1AT is to prevent excessive action of leukocyte-derived Ser proteinases, especially neutrophil elastase, in the circulatory system (Blank and Brantly, 1994). Therapeutic A1AT used in augmentation therapies is currently purified from pooled human serum, and the treatment can cost up to $100,000 per year per patient (Alkins and O'Malley, 2000). Concerns over the supply and safety of the products have urged searches for alternative recombinant sources of A1AT. Recombinant A1AT has been produced in human and nonhuman cell production systems with limited success (Blanchard et al., 2011;Brinkman et al., 2012;Ross et al., 2012;Lee et al., 2013). The production suffers from two major drawbacks: low expression levels and/or incorrect glycosylation (Garver et al., 1987;Chang et al., 2003;McDonald et al., 2005;Hasannia et al., 2006;Karnaukhova et al., 2006;Plesha et al., 2007;Agarwal et al., 2008;Nadai et al., 2009;Huang et al., 2010;Arjmand et al., 2011;Jha et al., 2012). The mature plasma-derived 52-kD protein has three N-linked glycosylation sites that are mainly decorated with disialylated structures (Kolarich et al., 2006). Sialylated N-glycans are a well-known requisite for the plasma half-life of A1AT (Mast et al., 1991;Lindhout et al., 2011;Lusch et al., 2013); the difficulties associated with obtaining them hamper the generation of biologically active A1AT in many expression systems.
Here, we pursued the expression of recombinant human A1AT in glycoengineered N. benthamiana and investigated the system's ability to generate active sialylated variants. Unusually high amounts of A1AT were obtained using a plant viral-based transient expression system. The inhibitor was efficiently secreted to the intercellular space (IF); however, peptide mapping showed that the secreted A1AT was truncated at both the N and C termini. Mass spectrometry (MS) -based N-glycan analysis of IF-derived A1AT showed that vacuolar typical paucimannosidic N-glycans were present. By expressing A1AT in Arabidopsis (Arabidopsis thaliana) knockout plants lacking b-Nacetylhexosaminidase (HEXO) activity (Liebminger et al., 2011), we showed that paucimannosidic structures are generated by the action of HEXO3 located at the plasma membrane.
Coexpression with the mammalian genes necessary for in planta sialylation allowed the synthesis of disialylated A1AT, and sialylation levels could be increased by the synthesis of multiantennary glycans. By contrast, fulllength A1AT purified from total soluble extracts exhibited ER-typical oligomannosidic carbohydrates. Using live-cell imaging, a GFP-tagged A1AT fusion did, indeed, exhibit aberrant ER-associated deposition of full-length A1AT. Elastase inhibition assays showed that ER-retained A1AT exhibits inhibitory activity, whereas the IF-derived truncated form was rendered inactive by cleavage within its RCL.

A1AT Is Efficiently Expressed in N. benthamiana Leaves
A plant viral-based transient expression system (Marillonnet et al., 2005) was used for recombinant expression of human A1AT. Complementary DNA (cDNA) of A1AT lacking the N-terminal signal peptide (Supplemental Fig. S1) and carrying either an N-or C-terminal Strep-II tag was cloned into the magnICON vector pICHa26211 (Schneider et al., 2014a). The recombinant protein was targeted for secretion using the barley (Hordeum vulgare) a-amylase signal peptide present in the vector. The resulting constructs ( Strep A1AT and A1AT Strep ; Fig. 1A) were transferred into Agrobacteria spp. and delivered to plant leaves by agroinfiltration. Expression of strep-tagged A1AT was monitored 5 d postinfiltration by western blotting using antibodies against the protein or the strep tag ( Fig. 2A). Two signals of 40 and 52 kD were detected in total soluble protein (TSP) extracts, and the 52-kD band corresponded to the size of the full-length protein. A band corresponding to the full-length size was also present in Strep A1AT purified from TSP ( Fig. 2A). Analysis of the intercellular fluid, representing the secretome of cells, using anti-A1AT antibodies showed the  presence of only a 40-kD band, and no signal was detected using strep tag-specific antibodies. Similar results were obtained upon expression of A1AT Strep (data not shown). These results indicate the expression of fulllength A1AT but also, an additional A1AT variant with truncations at both termini. The expression level of A1AT in TSP was up to 6 mg g 21 of leaf material, about 60% of which can be attributed to the 40-kD form (Supplemental Fig. S2).
A strong band at position 40 kD but not 52 kD was seen by SDS-PAGE and Coomassie Brilliant Blue staining of IF-derived proteins after infiltration with Strep A1AT (Fig. 3A), indicating that only the truncated version of A1AT is secreted. By contrast, three bands in the range of 50 to 55 kD that reacted with both anti-A1AT and antistrep antibodies were seen in Strep A1AT purified from TSP (Fig. 3B). The different sizes at close proximity may be attributable to differences in glycosylation of Strep A1AT.

A1AT Exhibits Truncated N and C termini
Liquid chromatography (LC)-electrospray ionization (ESI)-MS/MS was used to carry out peptide mapping of the IF-derived 40-kD A1AT and confirmed the presence of the A1AT protein backbone, but also, it revealed truncations at both ends (Fig. 4). The first peptide found corresponded to 11 TDTSHHDQDHPTFNK 25 . One cleavage site in the mature protein sequence (Supplemental Fig. S1) was after 10 K. This may have arisen from tryptic cleavage of the N-terminal sequence, but the expected N-terminal tryptic peptide ( 1 EDPQGDAAQK 10 ) was never detected. Also, several other peptides lacking up to 15 amino acids from the N terminus ( 16 HDQDHPTFNK 25 ) were detected. These fragments differing by 1 amino acid suggest additional trimming by amino-peptidase activity (Fig. 4A). C-terminal analysis of 40-kD A1AT revealed that the four most C-terminal tryptic peptides were missing. The full-length tryptic peptide carrying the RCL ( 344 GTEAAGAMFLEAIPMSIPPEVK 365 ) was also not detected, but several truncated forms of it could be found. The major cleavage occurred after 354 E, thereby destroying the RCL ( Fig. 4B; Supplemental Fig. S1). Other terminal fragments were found to a lesser extent (Fig. 4B).
To investigate which proteinases are involved in the conversion of A1AT from an active (native) to an inactive (truncated) state, we incubated commercially available human plasma a1-antitrypsin (hA1AT) and purified plant-derived A1AT ( Strep A1AT) with N. benthamiana IF isolated in neutral and acidic pH buffers. Acidic pH mimics the conditions of IF in planta (Grignon and Sentenac, 1991;Gao et al., 2004), where several proteases are active (Kaschani et al., 2010). Incubation of hA1AT and purified Strep A1AT with phosphate-buffered saline (PBS) did not alter the protein sizes detected by westernblot analysis (Fig. 5A, lane 1). After incubation with IF at pH 7.0, both proteins were detected as full-length versions, but truncated fragments were also visible (Fig. 5A, lane 2). When acidic IF (pH 5.0) was used, only truncated A1AT was detected (Fig. 5A, lane 3). After incubation of A1AT with IF previously heated to 95°C to inactivate possible proteases present, only full-length A1AT was detected (Fig. 5A, lane 4).

The 40-kD A1AT Is Generated by Interactions with Plant Proteases
Proteases are abundant in several subcellular compartments along the secretory pathway and in the apoplast. To assess which type of protease might be involved  in A1AT conversion, we preincubated IF samples extracted at pH 5.0 with different protease inhibitors before addition of purified plant Strep A1AT or plasma-derived A1AT (Fig. 5B). A protease inhibitor cocktail designed to eliminate proteolysis in plant extracts completely abrogated A1AT processing. Similarly, the addition of the Ser protease inhibitor phenylmethylsulfonyl fluoride (PMSF), which also displays some reactivity with Cys proteases, completely prevented breakdown of hA1AT (Fig. 5B). Taken together, these results strongly indicate that plant-derived A1AT interacts with endogenous Ser proteases residing in secretory compartments or the apoplast, ultimately resulting in its cleavage and concomitant inactivation.

Plant-Derived A1AT Is Active in Vitro
A1ATs, like other serpins, are molecular mouse traps, consisting of a bait and a swinging arm (Pike et al., 2002). Serpins are only partially stable in their active form, but they can snap into a far more stable form when they interact with a protease. When the protease binds to the bait (i.e. 358 M), it forms a bond and begins to perform its normal cleavage reaction (Gettins, 2002). The flexible loop of A1AT, now cleaved and free to Figure 4. Secreted 40-kD A1AT exhibits N-and C-terminal truncations. A, Identification of the N terminus of secreted tryptic-digested A1AT by LC-ESI-MS/MS. All found N-terminal peptide variants eluted in a window of 3 min, allowing their presentation in the sum spectrum shown. Several peptides lacking up to 15 amino acids at the N terminus ( 16 HDQDHPTFNK 25 ) were detected. Black arrows indicate cleavage sites. B, Identification of the C terminus of secreted trypticdigested A1AT by LC-ESI-MS/MS. Reversed-phase LC separated the differently truncated C-terminal peptides over the whole chromatogram; thus, extracted ion chromatograms for all variants detected are shown. The major peptide representing the C terminus was identified as 344 GTEAAGAMFLE 354 . The peptide ( 344 GTEAAGAMFLEAIPM*SIPPEVK 365 ) carrying the center of the reactive loop (asterisk) was not detected. Black arrows indicate cleavage sites; the major cleavage site is indicated by the white arrow. All peptide variants were confirmed by MS/MS spectra, and their masses are indicated in parentheses. m/z, Mass-to-charge ratio. move, traps the protease. In the case of elastase, A1AT binds almost irreversibly to its active site, and the complex does not dissociate after cleavage. Here, we evaluated the biological activity of the purified plantderived Strep A1AT as an inhibitor of porcine pancreatic elastase (PPE), a natural target protease. Incubation of 10 ng (or 0.38 mM) of elastase with 20 ng (approximately 0.38 mM) of purified Strep A1AT shows 40% of residual elastase activity. The incubation of stoichiometric amounts of the enzymes reveals that 60% of plantderived purified A1AT is biologically active, and higher amounts of Strep A1AT (50 ng and more) do not result in more elastase inactivation (Fig. 6A).
The A1AT inhibition mechanism involves the formation of a 1:1 covalent complex between A1AT and PPE, which is also associated with some cleavage of the inhibitor by the protease. This unique property of the irreversible A1AT-PPE complex permits activity measurements of A1AT by SDS-PAGE band shift assays. Incubation of purified plant-derived A1AT with PPE resulted in the generation of a high M r band (approximately 65) not present in the control sample. In addition, lower M r bands represent truncated A1AT (approximately 40) and PPE (approximately 26) after complex dissociation (Fig. 6B). Peptide mapping revealed that PPE cleavage occurred after 358 M, confirming that plant-derived A1AT interacts with target proteases in the same manner as its natural counterpart (Supplemental Fig. S3).

HEXO3 Forms Paucimannosidic N-Glycans on Secreted A1AT
Two HEXOs (HEXO1 and HEXO3) responsible for the formation of paucimannosidic N-glycans have been identified in Arabidopsis (Liebminger et al., 2011). Here, we used Arabidopsis single-knockout (hexo1 and hexo3) and double-knockout plants (hexo1/hexo3) to investigate how these enzymes contribute to the generation of paucimannosidic structures on secreted A1AT. A binary vector encoding full-length A1AT, including the native signal peptide (1-418 amino acids; 35S:A1AT Full ; Fig. 1B), was used to stably transform wild-type (Columbia-0 [Col-0]), hexo1, hexo3, and hexo1/hexo3 plants. Western-blot analysis of TSP extracted from transformed plants using anti-A1AT-specific antibodies showed two signals: one at position approximately 52 kD and the other at approximately 40 kD (similar to N. benthamiana; Supplemental Fig.  S4). The 40-kD form was recovered from IF and subjected to N-glycan analysis using LC-ESI-MS. The glycosylation profile of A1AT derived from the Arabidopsis wild type and hexo1 lacking the vacuolar HEXO revealed the presence of mainly paucimannosidic structures (i.e. MMXF), as observed for A1AT expressed in N. benthamiana ( Fig. 8A; Supplemental Fig. S4). By contrast, the glycosylation profile of A1AT expressed in hexo3 and hexo1/hexo3 mutants lacking the b-hexosaminidase associated with the plasma membrane exhibited primarily complex-type N-glycans (i.e. GnGnXF; Fig. 8, B and C). Only minor amounts of MMXF structures were detected on A1AT purified from IF of hexo3. Because both HEXO1 and HEXO3 are capable of converting GnGnXF into MMXF (Strasser et al., 2007), these results firmly establish that the paucimannosidic N-glycans decorating A1AT are generated within the secretory pathway and not caused by transient passage of the protein through vacuolar compartments.

N-Glycans of Secreted A1AT Can Be Efficiently Decorated with Terminal Sialic Acid
The extent of sialylation is known to play a significant role in the serum half-life of A1AT (Mast et al., 1991). We set out to generate sialylated plant-derived A1AT that mimics the glycosylation profile of plasma-derived A1AT (Supplemental Fig. S5) by expressing Strep A1AT in N. benthamiana ΔXT/FT together with mammalian genes (in a total of six genes) necessary for in planta sialylation of N-glycans (Castilho et al., 2010). N-glycosylation profiling of IF-derived 40-kD A1AT ( Strep A1AT Sia ) using LC-ESI-MS showed that almost exclusively complex-type sialylated glycans were present (up to 80%; Fig. 9A; Supplemental Table S1). This glycosylation profile highly mimics the plasma-derived counterpart. Because several studies have reported that raising the sialylation extent increases A1AT half-life (Lindhout et al., 2011;Lusch et al., 2013), we aimed to generate A1AT that carries additional sialic acid residues. One way of boosting sialic acid content is to increase the number of glycan antennae providing the appropriate acceptor  substrates for terminal sialylation. Thus, we expressed the human enzymes needed for the formation of tri-and tetraantennary glycans (i.e. N-acetylglucosaminyltransferease IV [GnTIV] and GnTV; Castilho et al., 2011bCastilho et al., , 2013. GnTIV and GnTV expressed individually with the genes for sialylation were able to efficiently branch the a1,3 and a1,6 arms, respectively (about 50%; Strep A1AT TriSia ; Fig. 9B; Supplemental Fig. S5; Supplemental Table S1). Compared with Strep A1AT, Strep A1AT TriSia protein revealed a clear shift in size when isolated from the IF (Fig. 3A). In comparison, synthesis of tetraantennary glycans by simultaneous expression of GnTIV and GnTV was less effective (about 4%; Supplemental Table S1). Interestingly, it seems that protein sialylation is an effective way to protect the trimming of b-1,2-GlcNAc residues, because paucimannosidic (MM) structures were present only at very low levels (3%-5%; Fig. 9, A and B; Supplemental Table S1).
Glycan analysis of A1AT purified from TSP (the wild type or DXT/FT) with and without coexpression of sialylation genes revealed the presence of peaks corresponding to oligomannosidic structures (Man 8 GlcNAc 2 and Man 9 GlcNAc 2 ; Fig. 9C; data not shown). The presence of such carbohydrates, typical for ER-resident proteins, and the absence of full-length A1AT in the IF support the idea that the truncation of A1AT largely occurs along the secretory pathway or in the apoplast.

A1AT-GFP Is Located in the ER
A binary vector containing a C-terminal GFP fusion was generated (35S:A1AT-GFP; Fig. 1B) to monitor subcellular localization of recombinant A1AT. The integrity of the expressed A1AT-GFP fusion protein was analyzed using western blots with anti-A1AT and anti-GFP antibodies (Fig. 2B). The results showed that the intact A1AT-GFP fusion protein (approximately 80 kD) was detected when either anti-A1AT or anti-GFP antibodies were used. In addition, a 40-kD protein band was detected with anti-A1AT antibodies but not anti-GFP antibodies. This indicates that the 40-kD protein is truncated at least on the C terminus. In addition, anti-GFP antibodies detected an approximately 30-kD band in the A1AT-GFP sample, which was slightly larger than the GFP used as the control (35S:GFP; Fig. 1B). This band probably represents GFP that still carries the A1AT fragment C terminus to the cleavage site in the RCL (approximately 4 kD; Fig. 2B).
As a control, the expression of GFP (35S:GFP) in N. benthamiana leaves was first monitored by live-cell confocal laser-scanning microscopy (CLSM). As expected, the small GFP protein was able to diffuse to the nucleus and localized mainly in the cytoplasm (Fig. 10A). By comparison, the CLSM image of cells expressing 35S:A1AT-GFP showed signals in primarily structures resembling the ER network and also, punctuate structures associated with the plasma membrane (Fig. 10B). No fluorescent staining was detected in the nucleus or the cytoplasm. To confirm ER retention of the intact A1AT fusion protein, we performed colocalization studies using the ER marker monomeric red fluorescent protein (mRFP) GnTI-C AAA TS-mRFP (a fusion protein that is predominantly retained in the ER; Schoberer et al., 2009). The two fluorescent signals showed a significant overlap, suggesting at least partial ER retention of A1AT-GFP (Fig. 10C).
The combined results of subcellular localization studies and glycan analysis of full-length and truncated A1AT strongly suggest that a fraction of the active Figure 9. A1AT N-glycan engineering allows the generation of highly sialylated structures. LC-ESI-MS profiles of trypsin-digested Strep A1AT (glycopeptide2) coexpressed with the genes necessary for in planta sialylation in DXT/FT. The 40-kD peptide collected from the IF was analyzed. A, Strep A1AT was coexpressed with the genes necessary for the synthesis of disialylated N-glycans ( Strep A1AT Sia ). B, Strep A1AT was coexpressed with the genes necessary for the synthesis of multiantennary sialylated N-glycans ( Strep A1AT TriSia ). C, Strep A1AT purified from TSP extracts and coexpressed with sialylation genes. Oligomannosidic peaks corresponding to the detected structures are indicated. Peaks were labeled using the ProGlycAn system (www.proglycan.com). Adjacent illustrations display the respective N-glycans using standard symbols. m/z, Mass-to-charge ratio. [See online article for color version of this figure. ] protein is retained within the ER and thereby, escapes attack by nontarget proteinases. These data show that the A1AT-cleaving enzymes reside in a post-ER compartment.

DISCUSSION
In this study, we pursued the recombinant expression of human A1AT in N. benthamiana. This complex serum protein served as a model to monitor constraints along the plant cell secretory pathway compromising protein integrity. We obtained unusually high expression levels of Strep A1AT (estimated at 6 mg of recombinant protein per 1 g of leaf material). Such extraordinary amounts are usually only reported for expression of reporter proteins like GFP (Marillonnet et al., 2004(Marillonnet et al., , 2005, and would certainly meet the requirements for commercial production. Nevertheless, we were confronted with the fact that the 52-kD full-length Strep A1AT accounts for only about 40% of the recombinantly expressed protein. The majority of Strep A1AT exists as a truncated 40-kD polypeptide. The protein lacks up to 15 amino acids at the N terminus and up to 41 amino acids at the C terminus. Importantly, truncation at the C terminus destroys the RCL and thereby, abolishes the inhibitory activity of the protein. Similar hydrolysis at the C terminus ( 354 E) has been observed when human A1AT is proteolytically inactivated by bacterial Cys proteinases (Nelson et al., 1999). Although target proteinases (e.g. elastase, trypsin, and a-chymotrypsin) cleave A1AT at 358 M, other nontarget proteases have different cleavage sites on the RCL (Nelson et al., 1998). Cleavage of the RCL at 354 E has been found to be typical of the lysosomal Cys proteinase cathepsin L (Johnson et al., 1986) and nontarget Ser proteases (Nelson et al., 1998). Considerable amounts of truncated protein were also obtained in previous attempts to express recombinant A1AT in rice (Oryza sativa) cells (Terashima et al., 1999;Huang et al., 2001;Trexler et al., 2002;McDonald et al., 2005), N. benthamiana (Sudarshana et al., 2006), and Nicotiana tabacum chloroplasts (Nadai et al., 2009). Nevertheless, A1AT produced in transgenic tomato (Solanum lycopersicum) plants does not show degradation (Jha et al., 2012).
The accumulation of the 40-kD polypeptide in the IF and the confinement of the residual full-length protein to the ER indicate that IF-resident proteases contribute to the conversion process. This is consistent with the notion that the plant apoplast is a proteinase-rich environment (Doran, 2006;van der Hoorn, 2008;Goulet et al., 2012). In support of this notion, in vitro assays showed that, when exposed to N. benthamiana IF, both plasmaderived and purified plant-derived A1AT are converted into lower M r versions. This conversion was enhanced in vitro at acidic pH, mimicking the in planta conditions (Grignon and Sentenac, 1991;Gao et al., 2004), and blocked when proteinases were inactivated by high temperature. In addition, our results indicated that Ser proteases are the most likely contributors to the conversion process. Similar results have been reported for hA1AT when spiked in N. tabacum cell culture supernatant (Huang et al., 2009) and the IF-mediated fragmentation of proteolysis-prone monoclonal antibodies (Niemer et al., 2014).
The mode of action of A1AT is to trap target proteases that bind to its bait on the RCL and cleave it. The protease becomes destabilized and partially unfolded, changing the shape of its active site so that it is no longer active (Elliott et al., 1996). However, various proteinases can cleave A1AT within its RCL without being trapped. Previous studies have shown that the inhibitor can be inactivated by interaction with bacterial proteases, thereby causing the conversion of native A1AT into an inactive form of decreased M r (Potempa et al., 1986;Nelson et al., 1998Nelson et al., , 1999. This seems to have also been the case in our studies. Indeed, it seems likely that Strep A1AT accumulates first as a full-length active protein (native state), which acts as bait for host Ser proteases. However, in contrast to elastase (a target protease), the complex formed between A1AT and the nontarget plant proteases is reversible, with the 40-kD A1AT reflecting the converted state of the protein after complex dissociation. A1AT analyzed after incubation with target proteases (chymotrypsin, trypsin, and elastase) or nontarget proteases (papain and bacterial proteases; Johnson and Travis, 1978;Potempa et al., 1986) exhibited C-terminal truncations highly similar to those observed in our studies. Certainly, activity assays of N. benthamiana leaf extracts suggest an abundance of Ser-type proteases (i.e. trypsin-and chymotrypsin-like proteases) in the apoplast (Goulet et al., 2012). Although C-terminal truncations of recombinant A1AT have been previously reported, N-terminal cleavage has so far not been investigated. However, analysis of the migration pattern of serum A1AT by isoelectric focusing suggested N-terminal truncations to be present (Kolarich et al., 2006). Interestingly, engineered A1AT with designed truncations at the N terminus showed higher inhibitory activity than the native protein (Pirooznia et al., 2013).
Another result of our studies, which we did not expect, was the presence of paucimannosidic N-glycans on IFderived A1AT (MMXF or MM depending on whether the expression host was the wild type or ΔXT/FT). Although these truncated N-glycans are frequently found in plants, secreted recombinant glycoproteins usually carry complex N-glycans, like GnGnXF Castilho et al., 2011aCastilho et al., , 2011bSchneider et al., 2014a). Nevertheless, two studies have reported the presence of paucimannosidic oligosaccharides on recombinant proteins, maize (Zea mays)and N. tabacum-derived human lactoferrin (Samyn-Petit et al., 2003), and bovine follicle-stimulating hormone expressed in N. benthamiana (Dirnberger et al., 2001). It is well known that the formation of this glycoform relies on the activity of HEXOs, which remove terminal GlcNAc residues from complex N-glycans (Strasser et al., 2007;Liebminger et al., 2011). Despite the presence of HEXO3 in the plasma membrane, it is commonly suggested that this glycan structure is generated in the vacuoles (Gomord et al., 2010). In this study, vacuolar HEXO1 does not seem to play a major role in the N-glycan processing of A1AT. Instead, our results indicate that complex N-glycans of A1AT are converted to paucimannosidic structures, most probably in the extracellular space by the action of HEXO3. This is a highly interesting finding, because most endogenous and recombinant proteins targeted to the apoplast are not processed by HEXO3. Why some proteins, like A1AT, are processed, whereas others, like human transferrin, are not processed by b-hexosaminidases when expressed and isolated in a similar way (Castilho et al., 2011b) is not known. Although it cannot be completely excluded that spatial separation is responsible for the observed differences, it seems more likely that the intrinsic properties of these proteins, such as their tertiary structures, are the major determinants. Recent advances in computer modeling of complex proteins and glycan structures designed to supplement experimental data (Nagae and Yamaguchi, 2012;Loos et al., 2014) should make it easier to explain this largely unexplored field.
Despite the unusual glycosylation profile of secreted A1AT, coexpression with genes of the sialylation pathway resulted in the generation of fully disialylated N-glycans similar to those in plasma A1AT. The formation of paucimannosidic structures could be largely inhibited.
The results indicate that the presence of sialic acid residues protects terminal GlcNAc residues from being removed by HEXO3. In insect cells, the synthesis of paucimannosidic structures could be prevented by capping terminal GlcNAc residues with b1,4-Gal (Ailor et al., 2000). In contrast, our attempts to b1,4-galactosylate A1AT by the overexpression of the highly active human b1,4-galactosyltransferase ) proved unsuccessful (data not shown), probably because of the presence of potent b-galactosidases in the secretory pathway of plant cells. Hence, only capping of the newly synthesized b1,4-Gal with sialic acid residues prevented their removal as well as GlcNAc residues from complex N-glycan structures.
Purified full-length Strep A1AT was found to carry only ER-typical oligomannosidic structures. This finding complies with that of the subcellular localization studies, in which A1AT-GFP was found to exhibit an ER-associated staining pattern. Partial aberrant subcellular localization of mammalian recombinant proteins expressed in plants has been reported (Loos et al., 2011;Schneider et al., 2014b). It seems that the repertoire of chaperones and folding factors, which is mainly responsible for correct protein folding and ER protein export, differs to some extent between plant and mammalian cells (Gupta and Tuteja, 2011;Denecke et al., 2012). The molecular interactions that allow the ER to retain proteins with no defined ER retrieval or retention signal are not entirely understood, but it is possible that interactions with ER quality components retain incompletely folded or aberrant proteins in the ER. Notwithstanding, we observed that the purified A1AT displayed elastase inhibitory activity, which indicates correct folding.
Processed versions of A1AT were obtained in three different ways: from in planta IF deposition, purified Strep A1AT incubated with PPE, and plasma A1AT after exposure to IF. The three derivatives showed different migration patterns on SDS-PAGE, despite having virtually identical protein backbones, which was confirmed by peptide mapping. Differences in migration can be explained by distinct glycosylation profiles (i.e. paucimannosidic [IF-derived Strep A1AT], oligomannosidic [purified Strep A1AT], and sialylated glycans [plasmaderived A1AT]).
Here, we efficiently decorated plant-derived A1AT with sialylated structures. By coexpressing a2,6sialyltransferase (ST6), secreted plant-derived A1AT carries exclusively a2,6-linked terminal sialic acid (Castilho et al., 2010) corresponding to the native therapeutic counterpart (Prolastin) and in sharp contrast to A1AT expressed in Chinese hamster ovary cells, where mainly a2,3-linked sialylation is present (Lee et al., 2013). Pharmacokinetic (PK) studies in rats have shown that the mean residence time of A1AT can be modified by varying the sialic acid linkage (Brinkman et al., 2012). Using the glycoengineering approach described here, it has been possible to generate recombinant proteins decorated with a2,6or a2,3-sialic acid or a mixture of them, depending on the sialyltransferase used (ST6, ST3, or both; A. Castilho and H. Steinkellner, unpublished data). This procedure allows A1AT to be generated with tailored PK values. This means, for example, that the PK of A1AT can be artificially increased by a2,3-sialylation, which would reduce the dose needed per patient. However, using the a2,6-sialylated counterpart may be advantageous because of the more natural linkage. Apart from the sialic acid linkage, the overall sialylation content of A1AT influences PK values (Lindhout et al., 2011;Lusch et al., 2013). Here, we were able to increase the levels of A1AT sialylation in planta through the generation of triantennary structures, which indeed, showed a significant increase in the sialic acid content over its plasma-derived counterpart. This could contribute to improved PK profiles, which has also been shown for other therapeutic proteins (Elliott et al., 2003).
In summary, we provide unique insights into how complex human glycoproteins are generated and processed in plants. Although the plant-based expression of some therapeutically interesting proteins (e.g. antibodies and certain enzymes) seems to be a success story (Grabowski et al., 2014;Hiatt et al., 2014), our results highlight that it is hard to predict how complex human proteins actually perform when expressed in plants. Upon expression of human A1AT, we observed some unexpected phenomena, such as aberrant subcellular localization, post-ER processing at the N and C termini, and unusual glycosylation of the protein. Although we could only partially explain the first two phenomena, we could gain valuable unique insights into the proteolytic network of the secretory pathway, which might enable the design of strategies to overcome the unwanted processing of plant-produced A1AT. Importantly, we were able to uncover the process that underlies the synthesis of paucimannosidic structures on secretory proteins and provided two independent approaches (knockout of HEXO3 activities and sialylation) to direct their glycosylation pattern toward human-like structures. Overall, our results contribute to a better understanding of the secretory pathway in plant cells.

A1AT Expression Vectors
Human A1AT cDNA (BC0156429) was purchased from the German Resource Centre for Genome Research (IRALp962E2033Q2). Two vector systems were used to express A1AT in plants: the plant viral-based magnICON vector pICH26211a carrying the barley a-amylase signal peptide for protein secretion (Fig. 1A;Schneider et al., 2014a) and standard binary vectors (Fig. 1B). A1AT cDNA (lacking the endogenous signal peptide) was appended with a strep tag on the N ( Strep A1AT) and C (A1AT Strep ) termini by means of PCR (Supplemental Table S2). Flanking BsaI restriction sites were included in the primer sequences to allow cloning of the PCR fragment into the pICH26211a vector. A1AT was also expressed in plants using standard binary vectors. This was done by amplifying the full-length A1AT sequence (amino acids 1-418, including the native signal peptide) as an XbaI-BamHI fragment and cloning it into the binary vector pPT2 (35S:A1AT Full ) or p20 to generate a C-terminal GFP fusion (35S:A1AT-GFP' Castilho et al., 2008).
All binary vectors were transferred into Agrobacterium tumefaciens strain UIA143, and magnICON constructs were transformed into strain GV3101 pMP90. Primers used in this investigation are listed in Supplemental Table S2.

Binary Vector for Modulation of A1AT N-Glycosylation
The binary vectors used to modulate the A1AT glycosylation pattern are described elsewhere. Briefly, biosynthesis of nucleotide-activated N-acetylneuraminic acid was achieved by the concomitant expression of UDP-GlcNAc 2-epimerase/N-acetylmannosamine kinase, N-acetylneuraminic acid phosphate synthase, and CMP-N-acetylneuraminic acid synthase (Castilho et al., 2008). b1,4-Galactosyltransferase was expressed as a fusion of the catalytic domain to the cytoplasmic tail, transmembrane domain, and stem region of ST6 ). The transport of sialic acid into the Golgi and its transfer to terminal Gal was achieved by the expression of the CMP-sialic acid transporter and ST6 as described previously (Castilho et al., 2010). For the synthesis of multiantennary glycans, we used binary vectors containing chimeric forms of GnTIV and GnTV targeted to medial Golgi compartments (Castilho et al., 2011b).
Agrobacterium spp. carrying the appropriate magnICON and binary vectors was used in agroinfiltration experiments to transiently express recombinant proteins (A1AT and proteins for glycan engineering) in N. benthamiana. magnICON and binary vectors were infiltrated at optical density at 600 nm (OD 600 ) = 0.1 and OD 600 = 0.05, respectively. For subcellular localization experiments, binary vectors were diluted to OD 600 = 0.03, and samples were harvested at 3 d postinfiltration.
Floral dipping was used for stable expression of A1AT in Arabidopsis wild-type Col-0 and HEXO knockout lines (Liebminger et al., 2011) using 35S:A1AT Full .

Isolation of Intercellular Fluid and Purification of A1AT
IF was collected using fully expanded leaves as described previously (Castilho et al., 2011b). The pH of the IF isolation buffer (100 mM Tris-HCl, pH 7.5, 10 mM MgCl 2 , and 2 mM EDTA) was adjusted to 5.0 or 7.0 according to the experiment. One N. benthamiana leaf yields 500 mL of IF, whereas 200 mL of IF is recovered from 20 Arabidopsis leaves. For TSP extraction, infiltrated leaf material was ground in a swing mill (MM2000; Retsch) for 2 min at amplitude 70; then, 1 volume (v/w) of 13 PBS was added. After incubation on ice for 10 min, the extracts were centrifuged (10,000g for 5 min at 4°C). A1AT was purified from TSP extracts by affinity chromatography using Strep-Tactin MacroPrep (IBA2-1505) according to the manufacturer's instructions.
Quantification of A1AT expression was carried out by immunoblot analysis using A1AT-specific antibodies. TSPs were extracted from leaves expressing Strep A1AT and diluted (1:100; 1:200; 1:500, and 1:1,000); 5 mL was loaded on SDS-PAGE and analyzed by western blotting using antibodies against A1AT. Commercially available plasma-derived A1AT (A6150; Sigma) was used as a reference. Western-blot images captured with Gel Doc EQ were analyzed using Quantity One (Bio-Rad Laboratories) software to measure relative band intensities.

A1AT Incubation Studies
Purified plant-derived A1AT and plasma-derived A1AT (approximately 60 ng) were incubated for 1 h at 37°C with 20 mL of concentrated IF (pH 5.0 and 7.0) isolated from wild-type N. benthamiana; 10 mL of Ser and Cys proteases inhibitors (14 mM E-64, 10 mM leupeptin, and 20 mM PMSF; Sigma), cathepsin B inhibitor (10 mM CA-074; Sigma), and proteinase inhibitor cocktail (1003; Sigma) was added to 20 mL of concentrated IF (pH 5.0) and incubated overnight at 37°C. Afterward, 60 ng of hA1AT or purified plant-derived A1AT was added, and the reaction mixtures were incubated at 37°C for 1 h.
All samples were mixed with 33 Laemmli sample buffer, and proteins were denatured by incubation at 95°C for 5 min. Finally, samples were loaded on 12% (w/v) SDS-PAGE and analyzed by western blot. This digestion allows site-specific analysis of three N-glycosylation sites: Gp1, R/ 40 QLAHQSNSTNIFFSPVSIATAFAMLSLGTK 69 ; Gp2, K/ 70 ADTHDEI-LEGLNFNLTEIPEAQIHEGFQELLR 101 ; and Gp3, K/ 243 YLGNATAIFFLPDEGK 259 . Arabidopsis-derived A1AT was analyzed using a different nano-LC-ESI-MS setup to account for the small amounts of material available. Peptides were separated with a PepMap 100 C-18 column (150 3 0.075 mm, 3-mm particle size; LC packings) using a Dionex Ultimate 3000 nano-HPLC System with a flow rate of 0.4 mL per minute. The gradient applied was identical to the standard setup gradient, but the aqueous solvent was changed to 0.1% (v/v) formic acid. Peptides were ionized with a captive spray source and analyzed on a Bruker maXis 4G in the datadependent acquisition mode using collision-induced (activated) dissociation. Glycopeptides were identified by their mass and the appearance of glycan fragment ions in MS2 spectra.

Glycosylation Analysis and Peptide Mapping
Identification of N and C termini was carried out by tryptic peptide mapping using LC-ESI-MS/MS (Pabst et al., 2012).

CLSM
Expression of A1AT C-terminally tagged with GFP (35S:A1AT-GFP) was monitored using an upright Leica TCS SP5 CLSM. The vector 35S:GFP was used to compare the subcellular localization of the A1AT fusion protein with that of GFP. GnTI-C AAA TS-mRFP (Schoberer et al., 2009) was used as an ER marker in colocalization studies. Dual-color imaging of GFP-and mRFPexpressing cells was performed simultaneously using a 488-nm argon laser line and a 561-nm helium/neon laser line. The images obtained were processed in Adobe Photoshop CS4.

Estimation of Functional Activity of A1AT
The elastase inhibitory activity of the plant-derived strep tag-purified A1AT was assayed using the SensoLyte Rh110 Elastase Assay Kit (Fluorimetric; 72179; Anaspec). The inhibitory capacity of A1AT was calculated as the residual activity of elastase measured using a fluorogenic substrate, and the procedure followed the manufacturer's instructions. In each reaction, 10 ng of elastase was incubated with plant-derived purified Strep A1AT at different concentrations ranging from 0 to 80 ng. Elution buffer used for Strep A1AT purification was used as a negative control to test sample autofluorescence.
A band shift assay was used to analyze A1AT and PPE complex formation. Briefly, 500 ng of plant-derived A1AT was incubated with 1,300 ng of PPE (E7885; Sigma) in 20 mM Tris-HCl buffer (pH 8.0) at 37°C for 20 min. All samples were analyzed by western blot using anti-A1AT antibodies or stained with Coomassie Brilliant Blue.

Supplemental Data
The following materials are available in the online version of this article.
Supplemental Figure S1. Amino acid sequence of the mature hA1AT protein lacking the signal peptide (amino acids 1-394).
Supplemental Figure S2. Determination of Strep A1AT expression level in N. benthamiana.
Supplemental Figure S3. Identification of the amino-and carboxylterminal amino acid sequences of the Strep A1AT 40-kD band after interaction with elastase.
Supplemental Figure S4. Expression profiling and glycan analysis of A1AT produced in Arabidopsis Col-0.
Supplemental Table S1. Relative abundance of major glycoforms detected on secreted Strep A1AT (given in percentage).
Supplemental Table S2. List of primers used in this investigation to express hA1AT in plants.