Five novel α-level plasmids named α11, α12, α13, α14, and α13R were based on original GB pDGB1α1 plasmid – a derivative of the pGreen vector (Hellens et al., 2000). The lacZ gene was amplified using specific primers listed in Supplementary Table S1 and Phusion DNA polymerase (Thermo Fisher Scientific, United States). The PCR products were excised from 0.8% agarose gel, purified using gel extraction kit (Qiagen, Germany) and subjected to the second round of PCR using extension primers dB1a-extF and dB1a-extR (Supplementary Table S1) that created 35 nt long overhangs homologous to EcoRI digested α1 plasmids on both ends. As a source of vector backbone, we chose to use plasmid with the inserted sequence, since it enabled the use of blue/white screening to identify recombinants. One μg of pDGB1α1 plasmid containing 1 kb long RB7 matrix attachment region (MAR) insert was digested with EcoRI HF (New England Biolabs, United States) and 2.6 kb long plasmid backbone was gel purified (Qiagen, Germany). 100 ng of purified PCR product and 50 ng of linearized backbone plasmid were mixed with 1 μl of SLiCE reagent (Zhang et al., 2012) and incubated for 60 min at 37°C. The SLiCE reaction mixture was then transferred to chemically competent E. coli Top10 (Thermo Fisher Scientific, United States) cells and plated on solid LB media with 50 μg/ml kanamycin and X-gal (Thermo Fisher Scientific, United States). Two blue colonies were selected from each plate, plasmid DNA purified using the GeneJET Plasmid Miniprep Kit (Thermo Fisher Scientific, United States) and verified by Sanger sequencing (GATC Biotech, Germany).

Multigenic DNA constructs were created using existing DNA parts from the GB 2.0 kit obtained from Diego Orzaez (Addgene kit # 1000000076, namely basic UPD, α and Ω plasmids, P19 CDS, NOS terminator), the MoClo Toolkit (obtained from Sylvestre Marillonnet & Nicola Patron, Addgene kit # 1000000044, namely 35S promoter, GUS, GFP and DsRed reporter genes) along with novel parts reported in this work. These novel parts were domesticated either at the Laboratory of Virology IEB, Prague (MAR sequences, short stuffers, novel α 11–14 entry plasmids, pEAQ derived CPMV 5′/3′ UTRs, CCAT-AATG intron) or at the Biopharming Research Unit, University of Cape Town (all BeYDV derived sequences – LIR, SIR, and Rep/RepA). Primers for the domestication of novel standard GB3.0 parts used in this work were designed using GB3.0 automatic primer designer¹. All primers are listed in Supplementary Table S1. For cloning purposes, we have used Phusion DNA polymerase (Thermo Fisher Scientific, United States). PCR products were column purified either from gel or directly from the 50 μl reactions using High Pure PCR Product Purification Kit (Roche, Switzerland).

For restriction/ligation assembly we used the standard protocol described in Sarrion-Perdigones et al. (2013) with either 45 cycles or 20 cycles. Both BsaI and BsmBI restriction enzymes and T4 ligase were obtained from Thermo Fisher. It is important to note that proper warming and resuspension of the 10× T4 ligase buffer (5 min at 40°C followed by 1 min vortex) prior to making aliquots is necessary. The DTT tends to form an insoluble precipitate in the freshly thawed buffer which leads to insufficient BsmBI cleavage due to low DTT levels.

Sequences of novel parts were verified using Sanger sequencing (GATC Biotech, Germany). The sequences of multi-cassette expression vectors are included in the Supplementary File S1. For sequence verification of larger multigene constructs by Sanger sequencing, we also provide sequencing primers directed outside of MAR domains (Supplementary Table S1). Schematic diagrams of the T-DNA regions of the plasmids pGB-R-DsRed-GFP, pGB-R-GFP-DsRed, pGB-G-GFP-DsRed, pGB-R-DsRed, pGB-R-GFP, pGB-E-DsRed-GFP, pGB-E-GFP-DsRed, pGB-E-MAR-GFP, pGB-E-MAR-GFPR, pGB-E-MAR-GFPi, pGB-E-GFP are shown in Figure 1. The resulting plasmids were transformed into E. coli Top10 (Thermo Fisher Scientific, United States), and after isolation and sequence verification they were eventually transformed into Agrobacterium tumefaciens EHA105 by the freeze-thaw method (An, 1987).

For transient expression of multigene constructs, we used Agrobacterium tumefaciens strain EHA105. Agrobacteria were grown in 4 ml of LB medium with appropriate antibiotics (100 μg/ml spectinomycin, 50 μg/ml rifampicin) under agitation (250 rpm) at 26°C overnight. The bacteria were pelleted by centrifugation for 10 min at 3700 × g and then resuspended in the infiltration buffer - 10 mM MES pH 5.6, 10 mM MgCl₂ and 0.1 mM acetosyringone. The optical density (OD₆₀₀) was measured directly in clear polystyrene culture tubes (Gama, Czechia) using a densitometer (DEN-1; Biosan, Latvia). We used bacteria at the concentration of OD = 1 if not indicated otherwise. The resulting bacterial suspensions were injected into fully expanded leaves using a syringe without a needle, either alone or as a mixture of several strains, through a small puncture (Huang and Mason, 2004).

For infiltration, we used 5- to 6-week-old Nicotiana benthamiana plants grown in growth rooms. The temperature in the growth rooms was set to 24°C throughout the experiment. All plants were grown in a 16/8 h light/dark regimen. Complete fertilizer (Kristalon, Czechia) was applied weekly. Plants were grown under 1:1 mixture of fluorescent tubes 36W/840 G13 MASTER TL-D (Philips, Netherlands) and plant growth specific LED tubes 8BEN-120cm-15W-GL-230V-T-G13 (Frontier Technologies, Czechia) (Janda et al., 2015). The light intensity at the base ranged from 90 to 140 μmol.s (Janda et al., 2015).

For each experiment (biological replicate) we inoculated 3–5 leaves of 3–5 plants. Three representative leaves were selected on the basis of being mechanically undamaged, with sufficiently large infiltrated areas and showing visible fluorescence when observed under UV light (approximately 400 nm) from a handheld lamp. In total, we prepared three extracts for each construct. Each extract consisted of three disks, originating from the same leaf. First we determined relative fluorescence readings for each construct on each leaf by comparison with the positive control (pGB-R-GFP, respectively, pGB-E-MAR-GFP for GFP, and pGB-R-DsRed for DsRed). Then we calculated the average fluorescence from multiple leaves. Samples were collected 5 days post inoculation (DPI) and were frozen in liquid nitrogen. The tissue was homogenized in tubes with 1 g of 1.3 mm zirconium milling silica beads in 300 μl PBS with 0.05% Tween-20 and 0.03% sodium azide using a FastPrep-24 (MP Biomedicals, United States). Samples were then normalized to total protein concentration of 1 mg/mL using Pierce^TM BCA Protein Assay Kit (Thermo Fisher Scientific, United States). The fluorescence of leaf extracts containing the expressed GFP and DsRed were measured on a Tecan-F200 instrument (Tecan, Austria) using a GFP filter set (485/20 and 535/25 nm excitation/emission, respectively) and a DsRed filter set (535/25 and 586/20 nm). The identity of respective fluorescent proteins was confirmed by western blotting using rabbit polyclonal anti-GFP antibody (EXBIO, Czechia) at 1:1,000 dilution, and rabbit polyclonal anti-RFP antibody (MBL International, United States) at 1:1,000 dilution. The band intensities were quantified with the ImageJ software (Version 1.48). GFP (Abcam, United Kingdom), and DsRed (Fraunhofer IME, Germany) proteins in the range of 0.005–1.0 mg/mL were used to create the standard curves.

Whole plants were imaged using an Olympus E-PL3 digital camera with a Zeiss Jena Pancolar (50/1.8) lens. GFP was visualized using self-assembled UV LED fixtures (approximately 400 nm) and LEE 124 Dark Green filter. Similarly DsRed fluorescence was visualized using green LEDs (approximately 530 nm) and LEE 019 Fire filter (both filters Thomann, Germany). For fluorescent microscopy we used a Leica CTR 5000 instrument with GFP and N2.1 filter cubes (Leica Microsystems, Germany).

The copy number of geminiviral replicons was estimated by qPCR on a LightCycler^® 480 instrument (Roche, Switzerland). We compared the accumulation of vector DNA in infiltrated plant tissue. Two similar constructs were tested: these were pGB-R-GFP-DsRed (which includes functional Rep/RepA) and pGB-G-GFP-DsRed (Rep has been replaced with gene coding for GUS). The scheme of the constructs is shown in Figure 4. For the quantitation, we used primers designed for DsRed (Supplementary Table S1), with normalization performed using Nb actin primers. The N. benthamiana leaves were infiltrated with replicating (pGB-R-GFP-DsRed) and non-replicating (pGB-G-GFP-DsRed) constructs. DNA was extracted from infiltrated leaves 5 DPI using salt and SDS extraction (Dellaporta et al., 1983) and used as a template for qPCR. Two samples for each constructs were measured. Each sample contained 3 × 2 leaf disks from upper, middle and lower leaves. Calibration curves based on spiked known concentrations of purified plasmid DNA were constructed.

We replaced the existing α1 plasmid of the Golden Braid 3.0 system with four new plasmids named α 11 to 14 (Figure 1). The 4 nt overhang generated during the BsmBI reaction (Table 1) on the 5′ end of α1 is identical with the 5′ overhang of the α11 fragment. Similarly, the 3′ end overhang of α1 is identical with that of α14. The engineered ends of plasmids 11 through 14 are shown in Table 1. With the exception of these unique 4 bp overhangs, the vectors are otherwise identical to the parental pDGB1α1 plasmid. Plasmids α2 and all Ω plasmids were retained for compatibility and can be used to shuffle existing modules between assembly systems.

The novel α plasmids were generated using PCR and homology based in vitro recombination (SLiCE) as described in the Methods section. Plasmid DNA from blue colonies (two colonies per construct) were isolated and verified by sequencing. One of the unique features of the GB plasmid set is that inserts can be easily released by restriction with one common and inexpensive restriction enzyme (RE). This RE is unique for each plasmid, so that in case of uncertainty, both the backbone and the insert can be easily confirmed by restriction digest. In all plasmids of this new set we decided to use the same flanking enzyme recognition site: this was EcoRI. The benefit of simpler management of restriction mapping verification of multiple constructs processed in parallel outweighs the impossibility of identifying plasmid backbones by RE digestions.

To increase the design flexibility of the extended set of plasmids we have also prepared a new α13 plasmid with an inverted insertion cassette named α13R. In future, other plasmids in the set (11, 12, and 14) can also be created in inverted orientation, however, at this point we believe that the flexibility offered by the current option of having 2 (α13R and α2R) out of 5 TU in reverse orientation is sufficient.

Since the number of parts in the assembly is fixed to five, there is also an increased need to use spacers or stuffer fragments in projects where more than two but fewer than five TUs are used. For this purpose, we have generated several spacers. As a first option, we have generated two short random DNA fragments of 35 and 55 bp in length that were cloned into pUPD2 as GGAG-CGCT fragments. Since most of our projects are aimed at overexpression of proteins, we have also domesticated four different matrix attachment regions (MARs) that can be used both as spacers and expression enhancers at the same time.

We set up in total 144 5-fragment reaction mixes, of which 115 resulted in plasmids with the expected restriction digestion patterns. Normally we use just two colonies to isolate plasmid DNA; in rare cases when neither corresponds to the expected map, two additional colonies are used for plasmid isolation. If neither of four colonies is correct the whole plate is discarded, and a new reaction is set up. Such failed attempts were recorded 29 times in total. In most of these cases, the reasons that lead to failed reaction could be identified afterward and can be seen in Table 2. Overall, the success rate of the five fragment assembly in our experience is similar to the original binary assembly.

To demonstrate the usefulness and greater flexibility of our extended set of vectors we decided to adapt existing a geminiviral expression vector into the GB standard. Only few relatively short sequences are necessary to initiate circularization and rolling circle replication of geminiviral replicons in host cells. These essential parts are long and short intergenic regions (LIR and SIR, respectively), and the Rep/RepA ORF. Typically the T-DNA with the replicon contains tandem copies of LIR that provide the limits of the excised circularized replicon DNA. The SIR sequence is placed in between the LIR copies. Insertion of a Rep/RepA module in the replicon is optional; it can be provided in cis on the same plasmid or in trans on separate plasmids. If it is present on the same backbone it is typically placed in the antisense orientation between the SIR element and the second copy of the LIR (Hefferon et al., 2004; Regnard et al., 2010). Our domestication strategy (Figure 2) was to create two different versions of the LIR sequence. LIR1, that is used as the 5′ limit of the replicon, has the syntax of the entire TU (GGAG-CGCT, A1-C1 according to GB convention). LIR2 which is used for the 3′ limit of the replicon has the syntax of a standard promoter (GGAG-AATG, A1-B2). The Rep/RepA CDS was cloned as a standard protein coding region (AATG-GCTT, B3-B5). Finally, the SIR region was cloned as the standard terminator (GCTT-CGCT, B6-C1). Only one BsaI site (344 bp from the Rep ATG) had to be mutated in the Rep/RepA gene: fortunately, it is located in a sequence encoding just one reading frame and thus it was possible to introduce a silent mutation.

After domestication and sequence verification the LIR1 was cloned into vectors α 11 to 13. For the c-sense part of the genome (LIR2-Rep/RepA-SIR) we took advantage of pDGB1α2R vector that allows the assembly of TU in antisense orientation. The resulting clone contained a 1 nt mutation before the ATG start to accommodate the GB syntax, a 1 nt silent mutation in the BsaI site, and a 5 nt insertion after the RepA stop codon to accommodate the GB syntax. As a non-replicating control, we made a similar cassette containing the GUS gene (LIR2-GUS-SIR). This flexible design allows the construction of replicating geminivirus-derived vectors carrying up to three standard transcription units (Figure 2). In this work, we have used vectors carrying either one or two transcription units. The assembled vectors were verified using restriction mapping and were then used for infiltration experiments.

To verify the ability of the novel GB derived vectors with BeYDV elements to replicate the DNA within the plant cell nucleus we attempted to quantify the vector DNA using qPCR. The plant tissue was infiltrated with two similar constructs: these were pGB-R-GFP-DsRed (which includes Rep) and pGB-G-GFP-DsRed (contains GUS gene instead of Rep, see Figure 3). Based on the qPCR results we estimate that replication-capable constructs accumulated 150–170-fold more copies per nuclear genome than a similar construct without the functional Rep (Figure 4).

Simultaneous expression of two or more genes is important for the production of multimeric proteins such as antibodies. The pEAQ vectors based on hypertranslatable 5′/3′ UTRs from CPMV have been successfully used to express antibodies in plants (Sainsbury and Lomonossoff, 2008; Sainsbury et al., 2009). Replicating geminiviral vectors were also successfully used for the same purpose (Hefferon and Fan, 2004; Huang and Mason, 2004; Regnard et al., 2010). Here we hypothesized that combination of both high transgene copy number provided by the replicating vector with highly efficient 5′/3′ UTRs derived from CPMV could lead to even higher expression levels.

The levels of fluorescent protein expression from the non-replicating constructs were higher than from their replicating counterparts (Figure 5), despite the fact that exactly the same cassettes were used for the expression of the fluorescent marker in both variants. While the replicating vectors lead to early development of leaf necrosis, the non-replicating variants did not show a tendency to cause necrosis even after longer periods (Supplementary Figure S2). To test the longevity of transient protein expression we replaced the GFP gene in pGB-E-MAR-GFP with a mutant version of DsRed called Timer (Takara Bio Inc., Japan). This mutant fluorescent protein produces a green fluorescent signal for several hours after protein synthesis, which then slowly matures to red fluorescence. Even 2 weeks after infiltration we observed cells showing green fluorescence, suggesting that protein expression from hypertranslatable mRNA with CPMV 5′/3′ UTRs did not stop even after such long periods (see Supplementary Figure S1).

Another phenomenon we observed was a stronger expression of the first transcription unit (about 2–3-fold) in both replicating and non-replicating constructs. This is an important finding for future strategy design to express multi-subunit proteins where their proper function depends on correct ratios of individual subunits, such as antibodies.

Based on the results summarized in the graph (Figure 5) we hypothesized that in order to achieve the highest co-expression of two proteins, the best strategy would be to use two independent non-replicating plasmids in a mixture of two Agrobacterium strains. However, it was impossible to estimate the co-expression at the cellular level using a plate reader, so the infiltrated area was examined under a fluorescent microscope. This analysis indicated cells showing higher GFP or DsRed signal in tissue infiltrated by two independent plasmids (Figure 6).

It has been shown previously that the necrosis elicited by BeYDV replication in N. benthamiana leaves can be reduced by the use of different strain of Agrobacterium (Diamos et al., 2016), and/or insertion of a MAR element downstream from the expression cassette. However, we did not observe a difference between GV3103 and EHA105 strains and saw only a relatively mild effect when RB-7 MAR was used. Other MAR elements (TM2, TM6 derived from tobacco and P1 derived from soybean) showed no effect. Next, we tested the hypothesis that lower density of A. tumefaciens during infiltration leads to faster reduction of expression from non-replicating constructs compared to replicating constructs. We tested 2-fold Agrobacterium suspension dilution series starting from OD600 = 0.25 to OD600 = 0.015. We also hoped that the dilution of A. tumefaciens might reduce or delay the necrosis. Leaves infiltrated with the non-replicating vector showed no signs of necrosis even at the highest bacterial concentration (OD600 = 0.25), and gradual decrease of fluorescence intensity corresponding with decreasing bacterial density. On the other hand, all dilutions of replicating constructs except the most dilute showed necrosis by the fourth DPI (Supplementary Figure S2). Thus, the non-replicating constructs are easier to use, since they are not that sensitive to Agrobacterium concentration, and can tolerate later harvest after infiltration.

One of the features of our novel extended set of GB vectors is that successful assembly requires exactly five DNA fragments to be combined. For assembly of just 2 TU modules, preferably the conventional binary GB assembly could be used. For assemblies of 3 or 4 TU parts, however, it is preferable to use vectors from our novel set along with appropriate spacers. For this purpose we have domesticated four matrix attachment regions: these are RB-7 (Allen et al., 1996; Halweg et al., 2005; De Bolle et al., 2007), TM2 (Xue et al., 2005), TM6 (Ji et al., 2013) from tobacco and P1 from soybean (Breyne et al., 1992; Petersen et al., 2002). We have also cloned a 700 bp intron from the pKANNIBAL vector (Wesley et al., 2001) and two shorter random sequences of 35 and 55 bp (for the complete sequences see Supplementary File S1).

It has been shown previously that MAR sequences can improve transgene expression and reduce position bias (Halweg et al., 2005), and can also be helpful in transient expression (Diamos et al., 2016). Here we wanted to test the effect of MAR sequences on the expression of non-replicating constructs with a single TU – GFP. As can be seen from Figure 7 the MAR sequences improved the expression of the reporter gene by about 25%. Additional improvement was obtained with pGB-E-MAR-GFPi where the CPMV 5′ UTR contained a short intron (A. thaliana ribose 5S EF intron). The relative orientation of the reporter cassette with respect to vector backbone does not seem to play a role.