Recombinant proteins (RPs) are proteins produced in heterologous systems, usually used to increase yields relative to the native system from which the protein is derived. Frequently, these proteins are pharmaceuticals, including vaccines and antibodies, which are produced as recombinant proteins in such scales that the biopharmaceutical sector has previously been the largest sector of the pharmaceutical market (Owczarek et al., 2019). Furthermore, protein drugs contributed to approximately 10% of the drug market in 2017 (Usmani et al., 2017). For biopharmaceuticals specifically, plants have already been shown to be a viable expression platform, and have been recently reviewed (Shanmugaraj et al., 2020), especially for the production of antibodies (Malaquias et al., 2021). Traditionally, mammalian cell cultures are the most favourable expression chassis, and between 2014 and 2018 they were used to produce 84% of novel recombinant proteins, while E. coli cell cultures were used for production of 8% and S. cerevisiae for 6.5% (Walsh, 2018). Broadly speaking, proteins for vaccines are not needed in substantial yields, but monoclonal antibody production and proteins produced for structural studies need high yields, requiring cost-efficient production platforms (Legastelois et al., 2016). However, it is unlikely that one optimal expression chassis exists for all RPs, as they each have different capabilities, and different RPs have various levels of complexity. For example, some proteins have eukaryotic-specific post-translational modifications (PTMs), and some are membranous meaning they have lower abundance relative to cytosolic proteins due to the reduced area that plasma membranes make up in most cells compared to the cytoplasm. Where possible, recombinant membrane proteins are produced in engineered E. coli strains (Schlegel et al., 2014). However, for proteins with PTMs, eukaryotic cell cultures including insect and mammalian cell lines are usually used (Hunter et al., 2018).

There is however an emerging potential for plants to be used as RP expression systems. Plant-made antibodies, or plantibodies, have been investigated as early as 1989 (Hiatt et al., 1989), and have had dramatic success in recent years (Malaquias et al., 2021), for example in producing a successful vaccine for the Ebola outbreak (Zhang et al., 2014), demonstrating that the system is suitable for expression of complex proteins with high yields. Moreover, technologies are developing that allow dramatically improved protein yields in plants, such as hyper-translatable viral elements for transgenes (Thuenemann et al., 2013) and commercially viable plant expression tools (Sainsbury et al., 2009). Due to these improvements, plant expression chassis have become a viable alternative to the conventional mammalian and bacterial systems, though further optimisation is still necessary.

The field of plant recombinant protein production has been heavily reviewed in recent years (Schillberg et al., 2019; Schillberg and Finnern, 2021; Schillberg and Spiegel, 2022), with reviews focussing on the advantages of plant RP expression systems, including the low upstream costs (Avesani et al., 2013; Chen and Davis, 2016), high capacity for scalability (Burnett and Burnett, 2019), potential for edible vaccines (Merlin et al., 2014; Kurup and Thomas, 2019), lack of endotoxin production (Fischer and Emans, 2000; Merlin et al., 2014), rapid protein production times that make them particularly suitable for vaccine production (Zhang et al., 2014; Chen and Davis, 2016; Rattanapisit et al., 2020; Makatsa et al., 2021), and the ability to produce complex post-translational modifications (Gomord and Faye, 2004; Millar et al., 2019), summarized in Table 1 . Glycosylation is particularly important for biopharmaceutical proteins, with engineering efforts made to humanize plant glycosylation and ensure safe immunogenicity (Sethuraman & Stadheim, 2006; Castilho et al., 2010; Montero-Morales & Steinkellner, 2018).

There are several available plant chassis that each have advantages and disadvantages, reviewed in detail by Sabalza et al. (2013) and more recently by (Gerszberg and Hnatuszko-Konka, 2022). Tobacco (Nicotiana spp.) is often used for RP production due to its rapid growth time, scalability, large number of seeds, and well-established transgenesis methods (Ma et al., 2003; Chen and Lai, 2015; Burnett and Burnett, 2019; Moon et al., 2019). Furthermore, the most commonly used plant cell culture is the tobacco Bright-Yellow 2 (BY-2) cell line, however, plant cell cultures rarely present any significant advantages over other eukaryotic cell cultures (Schillberg et al., 2019; Karki et al., 2021; Sethi et al., 2021). Legumes have high protein content (Burnett and Burnett, 2019), rapid growth (Abd-Aziz et al., 2020) and several transformable species such as Mucuna bracteate, but often show unfavourable heterogenous glycosylation (Abd-Aziz et al., 2020; Debler et al., 2021). Fruit and vegetable crops, including Lactuca and Solanacea species have been made to produce RPs, with the main advantage that they can be eaten, negating downstream purification costs (Merlin et al., 2014; Kurup and Thomas, 2019; Moon et al., 2019; Shanmugaraj et al., 2021). For these, several tissue-specific promoters are available (Lim et al., 2012; Ding et al., 2022), and researchers have found that growth conditions can dramatically affect protein content (Matsuda et al., 2009). Aquatic plants can be used, in particular Lemnoideae or Wolffia species, due to their fast growth rates and high yields, but transformation methods are poorly optimised (Khvatkov et al., 2018; Chanroj et al., 2021). Cereals present another promising expression system, recently reviewed (Mirzaee et al., 2022), with the advantage that their seeds have a high protein content and enable RPs to be stably stored (Stöger et al., 2000). Commonly used cereals include maize (Naqvi et al., 2011; Sabalza et al., 2013), barley (Ritala et al., 2008; Magnusdottir et al., 2013; Shanmugaraj et al., 2021), and rice, the latter particularly as cell-cultures (Huang et al., 2002; Kuo et al., 2013; Huang et al., 2015; Liu et al., 2017). Mosses such as Physcomitrella patens have had success for RP production (Baur et al., 2005; Gitzinger et al., 2009; Niederkrüger et al., 2014; Reski et al., 2018), particularly for glycosylated biopharmaceuticals (Huether et al., 2005; Decker & Reski, 2007; Reski et al., 2015), where RPs can be secreted into culture medium which dramatically reduces downstream purification costs (Schaaf et al., 2005). In fact, α-galactosidase A produced in moss has been shown to be more effective in treating enzymatic deficiencies in mice than the more conventional agalsidase alfa (Shen et al., 2015), demonstrating the potential superiority of the system. Finally, some newer expression systems are available, including cell-free protein synthesis systems such as the tobacco BY-2 cell-line lysate-derived system ALiCE, and several wheat germ systems, which have had success in producing cytoplasmic and microsomal proteins (Buntru et al., 2014; Harbers, 2014; Buntru et al., 2015). Similarly, plant cell packs utilise porous plant cell aggregates without cultivation media (Rademacher et al., 2019), compatible with well-plates which enable high-throughput rapid construct screening and can be automated to minimise batch variation (Gengenbach et al., 2020).

Plants can also be transformed using a variety of different methods, depending on a researcher’s needs. Stable transformation requires a single transformation event which is stably inherited in progeny of the primary transformant. This can be nuclear transformation, which typically results in moderate expression levels, but can be subject to transcriptional gene silencing through DNA methylation (Gallego‐Bartolomé, 2020). The site of transgene integration in the genome is generally random (Jaganathan et al., 2018), though some targeted transgenesis methods have been developed (Sanagala et al., 2017). Alternatively, plastids can be stably transformed, enabling very high expression levels due to the high plastid copy number within cells (Oey et al., 2009) and desirable biocontainment through primarily maternal inheritance (Ruf et al., 2007; Gerszberg and Hnatuszko-Konka, 2022). However, plastids are unable to produce complex PTMs. Lastly, transient expression is very rapid, taking days, with well-developed methods and high-expression levels and the ability to produce complex PTMs (Chen and Lai, 2015; Chen and Davis, 2016), with the main drawback that transformation needs to be repeated for each batch of RP. Table 2 provides a range of examples of recombinant proteins that have been expressed in plants.

While there is extensive literature about RP production in plants, little information is available about optimising expression through genetic manipulation and subsequent protein extraction. Consequently, this review focusses on the optimization of RP expression at the level of transcriptional and translational control, and the subsequent extraction methods employed for RP purification, with the aim of providing a comprehensive summary of the necessary considerations for producing RPs in plants.

Recombinant protein expression can be optimised through increasing transcription and translation and decreasing mRNA and protein degradation. This section is dedicated to the transgene design considerations that can help improve in planta RP production (summarised in Figure 1 ).

Recombinant protein purification is considered the costliest stage in plant RP production, with comparable costs to other expression systems (Kuo et al., 2013; Merlin et al., 2014; Chen and Davis, 2016; 2016; Dong et al., 2019). Extraction of plant proteins has been reviewed by Wilken and Nikolov (2012) with some further developments in physical (Buyel and Fischer, 2015) and chemical extraction (Dong et al., 2019). Plant protein purification often relies on chromatography techniques, although non-chromatographic techniques also exist and have shown success in plants. These include flocculation (Elmer et al., 2009; Buyel and Fischer, 2013; Buyel and Fischer, 2014), ultrafiltration and diafiltration (Buyel and Fischer, 2013; Opdensteinen et al., 2019), heat-precipitation (Menzel et al., 2016; Opdensteinen et al., 2020), and pH-precipitation (Menzel et al., 2016; Opdensteinen et al., 2020), with several methods reviewed by Buyel (2022). These will not be discussed in detail in this section, which will instead focus on the use of purification tags in plants as they are encoded in DNA and are often designed with the expression construct.

The use of affinity tags in RP production has been recently reviewed (Mahmoodi et al., 2019), with a diverse array of tags to choose from. There are many existing examples of affinity tag usage for plant RP purification. Antibody production usually utilises protein A affinity chromatography and does not require an additional fusion tag and is an efficient but expensive process that is only applicable to antibodies (Merlin et al., 2014). As a result, a dual-purification technique was developed by Jugler et al. (2020), which utilised a hydrophobin fusion to protein A (HPA). When the HPA fusion protein was mixed with the monoclonal antibody (mAb), two-phase separation could be performed to isolate the mAb from the bulk of endogenous proteins, which are retained in the aqueous phase. Following this, the addition of a low pH buffer can separate the mAb from the HPA fusion protein, allowing high purity and yield mAb recovery. This method is particularly interesting as it utilises a fusion-protein that does not involve the RP (mAb) of interest. However, despite this methods success, due to the use of protein A affinity purification, this protocol is still only applicable to antibodies.

Consequently, most affinity tags used in plant RP purification use fusions to the RP itself. One study utilised Arabidopsis to produce a recombinant membrane protein, STT3a, and used tandem affinity purification of a FSG tag (comprised of three FLAG–streptavidin binding peptide, a TEV protease cleavage site, and two protein G domains) and retrieved the RP following TEV-mediated cleavage with high enough yield and purity for rough structural studies at resolutions of 30Å (Jeong et al., 2018). A cheaper method is the use of IgG-affinity columns to purify proteins with a Staphylococcus aureus Z-tag. This has been used to produce chloroplast-expressed hIGF, at a cleavage efficiency of only 40%, but using hydroxylamine to cleave the tag (Daniell et al., 2009). Histidine residues have an affinity to nickel beads, so His-tags or proteins containing histidine residues in tandem are also used for purification using immobilised metal affinity chromatography (Wilken and Nikolov, 2012). However, despite their common use, His-tags are not always specific, with some endogenous plant proteins containing natural histidine motifs, so one group developed a hybrid tag, termed ‘Cysta-tag’, comprised of the His-tag with an additional multicystatin domain SlCYS8, which enhances RP levels in tobacco, which enabled a tagged RP to have 25% yield recovery and 90% purity (Sainsbury et al., 2016). Other researchers have created hybrid tags containing several different protein domains. Qi and Katagiri (2009) created a purification tag, called the HPB tag, which utilises a biotin carboxyl carrier protein domain (BCCD) derived from the Arabidopsis 3-methylcrotonal CoA carboxylase, and enables purification of low abundance membrane protein complexes using streptavidin beads, and subsequent cleavage of the tag using PreScission protease, leaving only a single short copy of the hemagglutinin (HA) tag for further purification if desired.

Alternatively, tags have been developed that do not rely on chromatography methods, including oleosin, elastin-like proteins (ELPs) and γ-Zein-derived peptides. Oleosin is similar to hydrophobin proteins, as when fused to a RP oleosin causes it to become hydrophobic and localise to the lipid-rich layer of a cell, allowing separation from the bulk of intracellular proteins through centrifugation (Wilken and Nikolov, 2012; Gerszberg and Hnatuszko-Konka, 2022). Parmenter et al. (1995) managed to purify recombinant aprotinin from maize seeds using an oleosin tag, with yield recovery of 49% but relatively low purity of 79%. ELPs allow similar purification, by causing the RP to become insoluble at raised temperatures, allowing easy separation from endogenous proteins (Rigano et al., 2013). Finally, γ-Zein-derived peptides cause localisation into the ER, and allow purification through differential centrifugation (Torrent et al., 2009; Joseph et al., 2012).

The disadvantage of affinity tags is that they sometimes must be cleaved to ensure normal RP function, a step that is usually inefficient and reduces RP yields further. Cleavage mechanisms are also highly varied, with many options available. Often, protease-mediated cleavage is used, where the protein coding sequence incorporates a protease recognition site between the RP and the tag. TEV protease is a widely used and well characterised protease, but several others are available including endoproteases such as factor Xa, enteropeptidase, thrombin and 3C protease, and exoproteases such as aminopeptidases and carboxypeptidases, though exoproteases are rarely used for RP production (Waugh, 2011). Frey & Görlich (2014a) discuss a series of alternative proteases with reportedly improved functions, including SUMO-specific SENP1 and NEDD8-specific proteases from Brachypodium distachyon, NEDP1 protease from Salmo salar, Atg4p from Saccharomyces cerevisiae and Usp2 from Xenopus laevis. Of these, bdSENP1 showed 10,000-fold higher activity than TEV protease at 0-4°C. It also showed increased salt tolerance which is useful as extraction buffers often contain high salt-concentrations. In an accompanying paper, the researchers tested these proteases on several recombinant proteins (Frey & Görlich, 2014b). Similarly, inteins can be used to remove tags, and have been reviewed by Lahiry et al. (2017). These are protein sequences that, under the correct environmental conditions, can ligate flanking peptides called exteins, excising themselves in a process termed protein splicing. Intein activity can be regulated by pH, temperature, the presence of reducing agents, salts, small molecules, and light (Lahiry et al., 2017). However, limitations of these include premature cleavage, product losses, the potential need for unfavourable reducing agents, and varying suitability depending on the amino acid sequence surrounding the exteins. A successful example comes from Islam et al. (2018) who developed a plant-based affinity tag with a family-3 cellulose-binding domain (CBD), CBM3, which binds microcrystalline cellulose (MCC). This was used in conjunction with bdSENP1 which enabled efficient tag cleavage following purification. This method is cheap due to the low cost and high availability of cellulose for purification. One potential disadvantage that the researchers acknowledged was the large size of the CBD (18kDa), which may affect the function and folding of the fused RP. The researchers demonstrated that the CBD tightly bound MCC, allowing RPs to be purified at over 95% purity in a single step. The researchers also noted that traditionally used proteases, including TEV, are expensive, and intein-based cleavage was inefficient and time consuming, requiring high concentrations of potentially damaging reducing agents to activate. Islam et al. (2018) utilised the SUMO-based cleavage as there is no additional overhanging sequence left following cleavage, and the process is highly efficient.

To summarise, protein tags substantially aid purification efficiency, but at the cost of yield due to multi-step purification processes. Thus, purification without the use of tags is often desirable. mAbs can be produced in plants rapidly, cost-effectively, and with high yields (Diamos et al., 2020). For non-antibody proteins, tags are often directly fused to the RP, and must be cleaved. Fortunately, cheaper binding partners are now available and more efficient tags have recently been developed allowing some flexibility when deciding on purification methods. Ultimately, tag usage should also be chosen on a case-by-case basis, paying careful attention to the properties of the RP in different conditions. Table 3 summarises the purification tags discussed. A non-plant-specific table is available in a review by Kimple et al. (2013).

Overall, plants constitute a promising recombinant protein production platform, with several unique advantages that may make them specialists at producing recombinant proteins. When producing RPs, plants give researchers large amounts of flexibility, in that several different modes of transgenesis can be used, whether this is stable or transient, nuclear or chloroplastic, each with their own unique advantages that can cover the breadth of requirements for any recombinant protein. This flexibility results in a series of co-dependent decisions that a researcher can make when expressing RPs in plants, summarised in Figure 2 . It is possible that plant-based protein production platforms could become cheaper than prokaryotic systems, as downstream processing methods begin to rely on cheaper reagents which have been developed in recent years. While plants are able to produce nearly any protein, it is most likely they will only be the dominant production system for proteins that have complex post-translational modifications and are required rapidly. To some degree, this is already realised as they are emerging to be the dominant antibody production platform, with pharmaceuticals not far behind, due to their unique advantage that RP-containing tissue can be consumed. Fortunately, recent and seemingly continuous developments in plant RP production and purification are rapidly advancing the field, with researchers now able to dramatically improve transcriptional and translational efficacy, maximise the number of transformed plant cells, and extract RPs with higher yields and purities than before. Consequently, due to these advancements, extremely cheap upstream costs, unparalleled scalability, and relative lack of specialist equipment, it would be unsurprising if plants eventually became the dominant protein production platform worldwide.

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication. RC generated the figures, tables and led the writing.