Candidate bTB proteins were selected by reference to lists derived from published bioinformatics approaches (e.g. (He et al., 2010; Monterrubio-López et al., 2015)) as well as current web sources (Kapopoulou et al., 2011; Dhanda et al., 2016). Throughout the study the genome of Mycobacterium bovis strain AF2122/97 was used as the reference strain. Proteins with a range of putative functions related to vaccine use were included, for example roles in virulence, predicted surface expression, previous use as a candidate vaccine and roles in the immune response and immune evasion.

Sequences representing the candidate ORFs were ordered as custom designed double stranded DNA fragments (Integrated DNA Technologies, Belgium). Sequence predicted to encode transmembrane domains was removed and synthetic genes were codon optimised for Spodoptera frugiperda (Sf9) or E. coli to maximise expression in the relevant host. All DNA fragments were flanked with 18 base pairs of sequence homologous to the vector pTriEx 1.1 (EMD Biosciences), suitable for expression in E. coli under control of the T7 promoter or insect cells under control of the baculovirus very late P10 promoter. All clones were generated by Gibson assembly (Gibson et al., 2009). The sequence designs were such that translation initiated at the bTB ORF initiation codon and fused the C-terminus with the histidine tag present in the vector, except when otherwise indicated. The pTriEx-1.1 cloning vector was linearised by digestion with restriction enzymes XhoI and NcoI (Thermo Fisher Scientific, Paisley, United Kingdom) and 20 ng gene fragment was assembled into 50 ng linearized pTriEx-1.1 cloning vector using the NEBuilder HiFi DNA Assembly Protocol (NEB, United Kingdom). Clones were isolated following transformation of Stellar competent cells (Takara Bio, France) with confirmation by colony PCR followed by DNA sequencing.

The default strain for protein expression in E. coli was SoluBL21 (Kim et al., 2015). Alternate hosts were E. coli T7 Express lysY (NEB) and LOBSTR (Andersen et al., 2013) which were used for low yielding constructs. Initial screening was achieved with cultures of ∼20 ml and scaled according to the outcome. Cells were grown at 37°C to an OD₆₀₀ = 0.6 and induced by the addition of IPTG to 0.1 mM. After induction, growth was continued for 4 h at 37°C. Cultures were harvested by centrifugation at 4,000 × g for 15 min and disrupted for gel analysis or purification as required. Solubility was determined after lysis in 1% Triton-X detergent and gel analysis of the soluble and pellet fractions. Poor solubility was corrected, where possible, with low temperature induction (16°C) or by varying the IPTG concentration used for induction. Intractable expression was abandoned in favour of alternate constructs or the alternate host.

Recombinant baculoviruses were produced by co-transfection of Sf9 cells with transfer vectors and linearised baculovirus DNA as described (Porta et al., 2013). Transfections were done in 6 well tissue culture plates at a monolayer confluency of 50%. Viruses were passaged when significant cytopathic effect was observed and viral stocks were amplified to high titre, typically 3 passages, prior to use for infection and protein detection. All viral stocks were produced in Sf9 cells. For protein expression, 100 ml suspension cultures of AoTni38 cells (Hashimoto et al., 2012) were infected at a density of 2 × 10⁶ per ml with 10 ml of high titre (>10⁷ pfu per ml) virus stock and the culture continued for 3 days. Cells were harvested by centrifugation and processed as described.

Protein samples were prepared using NuPAGE loading buffer (Fisher Scientific UK Ltd., United Kingdom) and heated to 98°C for 10 min before loading the gel. Proteins were separated by SDS-PAGE using 4–12% precast Tris-Glycine SDS polyacrylamide gels (Fisher Scientific UK Ltd., United Kingdom) for 30 min at 200 V. SDS-PAGE loading used the equivalent of 50 μL of bacterial culture or 5 × 10⁴ insect cells per lane of a 10 lane, 10 cm gel. After electrophoresis, gels were subjected to either Coomassie Brilliant Blue R250 staining or transferred to a polyvinylidene difluoride (PVDF) membrane for western blot analysis.

Proteins were transferred electrophoretically to PVDF membranes (Immobilon^®, Merck, Germany) by semi-dry transfer. Membranes were blocked with Protein-Free T20 (TBS) Blocking Buffer (Fisher Scientific UK Ltd., United Kingdom) for 1 h. All subsequent incubations and washes were done in Tris-buffer saline containing 0.1% Tween 80 (TBST). Membranes were incubated with monoclonal Anti-6X His tag^® antibody conjugated to HRP (Abcam, United Kingdom) at 1:2000 dilution followed by three washes of 5 min each. The membrane was finally washed with TBS and antibody reactive bands were revealed using ECL Western Blotting Detection Reagent (Geneflow, United Kingdom) and imaged using a Syngene G-Box (Syngene, United Kingdom).

E.coli cell pellets were lysed in 10% of the original culture volume in a lysis buffer (500 mM NaCl, 20 mM Sodium phosphate buffer pH 7.4, 20 mM imidazole, 1% Triton X-100) supplemented with 1% lysozyme, 1 mM phenylmethanesulfonyl fluoride and 1 mM benzamidine hydrochloride and incubated for 10 min at 4°C. Samples were sonicated for 10 min on ice followed by centrifugation at 15,000 × g for 10 min. Infected insect cells were disrupted similarly but without lysozyme in the lysis buffer. Expressed proteins, with an affinity tag of six consecutive histidine residues, were purified by His-mag sepharose Ni beads (GE Life Sciences) using 200 µL bead slurry per preparation. Beads were incubated with the clarified lysate for a minimum of 2 h on a blood wheel before washing and elution according to the vendor’s instructions. For proteins that demonstrated a tendency to insolubility the lysis buffer also contained 34 mM SDS and the wash and elution buffers included 0.1% sodium sarkosyl (Schlager et al., 2012). For secreted proteins the clarified supernatant was adjusted to 0.5 mM nickel sulphate and used directly for IMAC pull-down. Constructs that were positive for expression but failed purification by one of these routes were abandoned.

The selection of candidates for this study was based on one or more vaccine related properties, viz. their predicted biological function (cell surface protein, virulence or adhesion), predicted immunogenicity, previous use as a candidate vaccine or sequence conservation in multiple mycobacterial genomes. From these considerations the final ORFs selected were grouped into seven loose descriptive categories (Table 1). For ORFs representing complete proteins with a predicted molecular mass of greater than ∼15 kDa, synthetic DNA was produced to the entire coding region with flanking regions that ensured a unique initiator ATG downstream of the Shine-Dalgarno sequence and fusion with a sequence encoding polyhistidine at the carboxyl terminus of the expressed protein, both features of the vector used (Figure 1A). If a signal peptide was present in the original M. bovis ORF, assessed by routine submission to the SigP server (Nielsen et al., 2019), it was changed to that of honeybee melittin, known to be efficiently processed in insect cells (Tessier et al., 1991). Sequences predicted to encode transmembrane domains were deleted. In cases where expression of the recombinant protein was either poor or, more usually, insoluble, the design was changed to incorporate an N-terminal polyhistidine sequence after the initiator codon and a stop codon prior to the vector encoded polyhistidine sequence (Figure 1A).

More than half of the expressed proteins that were problematic in recombinant form with a C-terminal polyhistidine were improved by this strategy, as described below. For proteins that were <15 kDa, or for proteins whose database entry did not contain an initiator methionine, for example some members of the esx family, sequences were concatenated before synthesis so that several ORFs were expressed together, connected by a Gly-Ser linker of 7 residues (Figure 1A). Similarly, for protein pairs such as the proline-proline-glutamate (PPE) and proline-glutamate (PE) complexes, the matching pair was synthesised as a single chain also connected by a flexible linker. The available crystal structures of such complexes show the C-terminus of the PPE partner lying close to the N-terminus of the paired PE (Strong et al., 2006), a distance easily spanned by the added linker. Some M. bovis ORFs were persistently difficult to express in either E. coli or insect cells and in these cases workaround strategies involved fusion with solubility tags including GFP, SUMO or the B1 domain of streptococcal protein G (GB1) (Lindhout et al., 2003; Pengelley et al., 2006; Lee et al., 2008). Some of these strategies improved expression or purification for some of the ORFs but none provided a universal route to improvement.

To maximise the possibility of efficient expression of a diverse range of selected M. bovis ORFs as soluble proteins, two different expression systems were used, E. coli and recombinant baculoviruses. Screening of the clones or recombinant viruses was done in small scale cultures by Western blot with a directly conjugated antibody recognizing the polyhistidine tag (Figure 1B). The default result was a band of the molecular mass predicted for the translated protein but occasionally the band identified by the Western blot had bands additional to the predicted mass (marked in Figure 2A). We assumed smaller sizes were the result of protein degradation, with the residual C-terminal His tagged fragment being detected, whereas larger proteins were post-translationally modified or oligomeric. Notably, the concatenated PPE-PE complexes gave rise to stable products of the predicted molecular mass (marked in Figure 2B) as did the Esx concatenates (marked in Figure 2A), some of which have been reported elsewhere (e.g., (Li et al., 2016; Mearns et al., 2017)). Concatenated constructs Mb1036_Mb1129c, Mb1272_Mb2477c, Mb3070_Mb3641 and Mb3338_Mb3644c also showed predominantly single bands at the predicted molecular mass when probed with an anti-His antibody (marked in Figure 2B).

Clones positive for expression by western blot were assessed for purification by pull-down from detergent lysates using IMAC magnetic beads and the eluates assessed directly by SDS-PAGE and gel staining (Figure 3). The default outcome was a predominant band of the predicted molecular mass that also agreed with the mass identified by western blot with an anti-His antibody. The concatenated forms of the Esx proteins and many of the similarly linked PPE-PE fusions purified from the soluble fraction as single band products indicating little if any breakdown of the fusion protein at the linker junction (Figure 3B). For example, purified EsxRST and EsxU (Figure 3, panel B, lanes 7 and 8) match those blotted (Figure 2, panel A, lanes 6 and 8) and PPE-PE fusions PPE41_PE25, PPE14_PE7, PPE15_PE8 and PPE20_PE15 (Figure 3, panel B, lanes 10, 14, 16 and 17) matched the proteins identified by blot in crude lysates (Figure 2, panel B, lanes 4, 7, 1 and 2). Sequence concatenation as a viable strategy to reduce clone numbers and to produce single polypeptide versions of multimeric bTB protein complexes, therefore appears feasible, as has been shown for other proteins (reviewed in (Matsushima et al., 2008; van Rosmalen et al., 2017)). Overall, the number of recombinant proteins purified directly without any form of optimisation was ∼50% of the number of clones tested (Table 1), low expression levels or insolubility accounting for the majority of the failures.

To improve the overall recombinant bTB protein recovery rate a limited number of ORF sequences were re-cloned as N-terminal His tagged variants and purification repeated. Cell wall proteins (Mb3338 and Mb3070), conserved hypothetical (Mb0854), ribosomal protein (Mb1656), three metabolic proteins (Mb0130, Mb3652 and Mb3871), and six proteins of unknown function (Mb2898, Mb2900, Mb2656, Mb1319, Mb1228 and Mb2244) were all rescued to purified soluble proteins by this approach (Figures 4A,B). Others, such as outer membrane proteins (Mb3644, Mb1445 and Mb2653) remained unworkable. These proteins may require folding partners to be co-expressed in order to be solubilised, as was the case for the original expression trials of PPE-PE family members (Strong et al., 2006).

For some problematic proteins the case for their inclusion on grounds of likely candidacy as a vaccine component was such that several alternate forms of expression were assessed. EspA (Mb3646c) has been suggested to be part, with EspC, of the type VII secretion apparatus (discussed in (Ates and Brosch, 2017)) and an obvious candidate for inclusion (Chiliza et al., 2017). Recombinant EspC has been described (Lou et al., 2017; Salemi et al., 2020) and used diagnostically (Yan et al., 2018; Srinivasan et al., 2019; Villar-Hernández et al., 2020), but EspA has proven more difficult (Garces et al., 2010). EspA expression as either a C- or N-terminally His tagged protein was not successful and mutation of the single cysteine reported to drive dimer formation (Garces et al., 2010) had no effect. Fusion of EspA with EspC at either the N- or C- termini via a Gly-Ser linker did not improve expression but, of a number of enhancement tags investigated (GFP, SUMO and GB1), fusion of GFP to the N-terminus rescued detectable expression at the predicted molecular mass in both E. coli and insect cells and enabled modest levels to be purified (Figure 5A). Similarly, adding GB1 as an N- terminal fusion tag rescued expression of outer membrane protein Mb1762 and two unknown proteins, Mb1916 and Mb2410, allowing purification to reasonable levels, albeit with contaminating host derived proteins (Figure 5B). To assess if problematic expression was the result of any common feature in the proteins concerned, we inspected the AlphaFold predicted structures (Jumper et al., 2021) of the full length problematic proteins (Jumper et al., 2021) (Table 2). Of the 20 candidates analysed, only 4, Mb2244, Mb2656, Mb3652 and Mb3871 had available full length structures but in each case the proteins concerned were multimeric, some to a very high degree (Mb3871 is a 24mer), which may have limited expression here. Five proteins, Mb1656, Mb1916c, Mb2410, Mb2898 and Mb2900 had partial structures obtained with protein fragments suggesting that full length protein expression was not possible, and a further 5, Mb0130, Mb1228, Mb1319, Mb3070 and Mb3338, had fragment structures predicted on the basis of homology with other proteins whose structures have been solved. Of these 10 proteins, AlphaFold predictions suggested unstructured termini, predominantly at the N-terminus, as a likely limitation to stable protein expression. Unstructured termini, also with a predominance of unstructured N-termini, were associated with the 6 proteins (Mb0854, Mb1445, Mb1762, Mb2653, Mb3644 and Mb3646c) with no available structure, only 3 of which were recoverable here. Submission of the 20 candidate to the FuzDrop server (Hatos et al., 2022) showed a high score for likely aggregation and liquid-liquid phase separation associated with about half of the proteins for which there was either no structure or only a partial structure available (Table 2, indicated). Extensive protein engineering to reduce these scores might be necessary if these proteins were to be produced at a scale and purity required for a vaccine candidate.