The original Gal2 fragments were cloned from the genomic DNA of the CEN.PK102-3A yeast strain (Entian and Kötter, 1998) and then inserted into EcoR I and Nco I sites of the pYX242-TEF1araA plasmid (Wang et al., 2013), resulting in the pYX242-TEF1araA-Gal2 plasmid (the physical map was listed in the Supplementary Figure S1). The site-directed mutations of Gal2 were constructed by a fusion PCR strategy based on the overlap extension PCR (Urban et al., 1997) and then inserted into pYX242-TEF1araA using Gibson assembly (Gibson, 2011). The promoter of Gal2p and its mutants was the original promoter TPI of plasmid pYX242, and the terminator of them was PGK1. Escherichia coli DH5α was used for plasmid amplification.

The episomal plasmid pYX2422-TEF1araA in a formerly obtained L-arabinose utilizing strain BSW3AP (Wang et al., 2013) was removed to obtain strain BSW4AP (Wang et al., 2017). The Gal2 gene of BSW4AP was then knocked out by transforming the fusion fragments containing two homologous arms of Gal2 and a loxp-KanMX4-loxp segment cloned from pUG6 (Güldener et al., 1996) to obtain the chassis BSW5AP. Plasmids pYX2422-TEF1araA, pYX242-TEF1araA-Gal2, and pYX242-TEF1araA-Gal2m were transformed into BSW5AP or BSW4EYX (Wang et al., 2015) to test recombinant yeast strains for L-arabinose or xylose and glucose transport. The plasmid YIp5-ara (Wang et al., 2013) was further transformed to BSW4EYX to evaluate the co-utilization and co-fermentation of glucose, xylose, and L-arabinose of Gal2p mutation. Yeast transformations were conducted using the conventional lithium acetate method (Gietz et al., 1995).

Saccharomyces cerevisiae strains and plasmids used in this study are listed in Table 1. The primers used in this study are summarized in Table 2.

The yeast synthetic complete (SC) medium used in this study contains 1.7 g L^-1 yeast nitrogen base (YNB, Sangon, China) and 5 g L^-1 ammonium sulfate (Sangon, China). To cultivate the yeast and maintain the required plasmids, the medium was supplemented with the appropriate carbon source and complete supplement mixture (i.e., 0.77 g L^-1 CSM-URA, 0.69 g L^-1 CSM-LEU, or 0.67 g L^-1 CSM-LEU-URA) (MP Biomedicals, Solon, OH, United States). For strains containing the KanMX4 marker, the medium was supplied with 200 μg mL^-1 G418 sulfate (Promega, Madison, WI, United States). Plasmids were amplified in E. coli strain DH5α (TransGenBiotech, China), which was grown on Luria-Bertani (LB) medium with 200 μg mL^-1 ampicillin.

To cultivate BSW3AP-based strains, single colonies were preincubated in SC medium containing 20 g L^-1 glucose for 24 h, and then shifted into medium containing 5 g L^-1 glucose and 15 g L^-1 L-arabinose for 48 h. After that, the cells were collected and used for batch cultivation in the respective SC medium containing 20 g L^-1 L-arabinose at an initial OD₆₀₀ of 1. To cultivate BSW4EYX-based strains, single colonies were pre-incubated in SC medium containing 20 g L^-1 maltose for 24 h, and then the cells were collected and used for batch cultivation in the respective SC medium containing 20 g L^-1 glucose, 20 g L^-1 xylose, or 20 g L^-1 L-arabinose at an initial OD₆₀₀ of 1 or 5. All the yeast strains were batch cultivated in 40 mL cultures in 150 mL aerobic triangular flasks or air-limited flasks at 30°C, 200 r min^-1. All E. coli strains were cultured at 37°C, 200 r min^-1.

The putative homology model of the transporter Gal2p was generated using Discovery Studio software (DS, Accelrys, San Diego, CA, United States). The E. coli homolog of the glucose transporters GLUT1-4 (PDB code 4GBY, 4GBZ, and 4GC0) (Sun et al., 2012) served as template to construct the homology model using the MODELER auto module. The CHARMm-based molecular mechanics was used to refine the loop regions of the protein structure. PROCHECK (Laskowski et al., 1993) was used to validate the refined models. The locations of L-arabinose were determined by core-constrained protein docking and a modified CHARMm-based CDOCKER method. The best position of L-arabinose was chosen by comparing the CDOCKER energy among the determined 10 positions. The critical amino acid residues at the L-arabinose binding position were analyzed using VMD software (Humphrey et al., 1996).

To investigate the function of amino acid residues of Gal2p for L-arabinose, xylose, or glucose transport, we utilized site-directed random mutagenesis. The variety of the residues were introduced by the design of primers, with the coded three bases replaced by NNN (Farwick et al., 2014).

The plasmids containing different mutants of Gal2 were ligated and directly transformed into BSW5AP; then cells were spread onto solid medium with 20 g L^-1 L-arabinose or 20 g L^-1 L-arabinose together with 2 g L^-1 glucose. The colonies that were distinctly larger or smaller than the BSW5AP-AGal2 strain were selected out. The exponential growth rates of the selected colonies were subsequently determined in the secondary screening procedure on 20 g L^-1 L-arabinose. The plasmids in the selected yeasts were then extracted and sequenced to identify specific mutations.

The culture optical density (OD₆₀₀) was measured by a BioPhotometer plus (Eppendorf, Germany) and was used to determine the strain growth. The growth capacities of strains were determined by the exponential growth rates (Young et al., 2014), which were analyzed by the linear regression coefficients of ln OD₆₀₀ versus growth hours from the growth curves (Wang et al., 2015).

The concentrations of glucose, xylose, xylitol, L-arabinose, arabitol, and ethanol were determined using the supernatant of filtered samples collected from batch cultivation. The high-performance liquid chromatography (HPLC) prominence LC-20A (Shimadzu, Japan), which has a refractive index detector RID-10A (Shimadzu, Japan) and an Aminex HPX-87P ion exchange column (Bio-Rad, United States), was used to determine the concentration of the above chemicals at 80°C with a mobile phase of water at a flow rate of 0.6 mL min^-1, as previously reported (Garcia Sanchez et al., 2010; Wang et al., 2013).

To construct L-arabinose transport-deficient chassis and study the L-arabinose transport mechanism, the endogenous and highly efficient L-arabinose transporter Gal2 of BSW4AP was knocked out using the KanMX4 marker to generate the strain BSW5AP. To test the L-arabinose transport capacity of BSW5AP, plasmids pYX2422-TEF1araA and pYX2422-TEF1araA-Gal2 were then transformed to generate the strains BSW5AP-A and BSW5AP-AGal2, respectively. The growth curves of the strains BSW5AP-A, BSW5AP-AGal2, and BSW3AP on L-arabinose were tested, and the exponential growth rates were calculated (Figure 1). It was clear that BSW5AP-A lost the ability to grow on L-arabinose. After overexpressing Gal2 by plasmid in BSW5AP, the BSW5AP-AGal2 strain regained the ability to grow on L-arabinose; however, the exponential growth rate of BSW5AP-AGal2 was lower than BSW3AP, which may have resulted from the different expression strategy. We verified that our strains exhibited Gal2p-dependent L-arabinose transport, and we successfully generated the L-arabinose transport-deficient chassis BSW5AP.

The homologous model of Gal2p for L-arabinose binding was constructed according to the outward-facing and partly occluded conformation of the E. coli xylose permease XylEp (Sun et al., 2012). The homology of Gal2p with XylEp was more than 28.8%, which indicated the feasibility to construct a homologous model. The homologous model of Gal2p was constructed by excluding 63 amino acids from the N-terminus and 39 amino acids from the C-terminus due to the absence of the corresponding sequences in XylEp. A total of 14 amino acid residues within a distance of 5 Å to L-arabinose in the model were predicted (Figures 2A,B and Table 3). The polar or aromatic amino acid residues in different transmembrane sequences (TMSs) of Gal2p were chosen for more in-depth study, including F85, T89, F223, N346, N347, F350, Y351, N376, and Y446.

To investigate the function of the selected amino acid residues of Gal2p in either L-arabinose, xylose, or glucose transport, we used random and site-directed mutagenesis to introduce amino acid substitutions. The exponential growth rates of the expressing BSW5AP strains during aerobic cultivation were used to determine the change in growth capacity (Table 4).

A total of 10 substitution mutants were identified at the selected F85 site. When F85 was changed to a charged E or R, Gal2p lost the transport capacity of L-arabinose. However, changes to the polar amino acids S, G, C, T, Y, or N of F85 somehow increased the exponential growth rates. Furthermore, the exponential growth rates of the F85S, F85G, F85C, and F85T mutants on L-arabinose significantly increased by 40, 32, 20, and 16%, respectively, compared with the strain with wild-type Gal2p. As such, F85 was identified as a very important amino acid residue for L-arabinose transport (Figures 2C,D). For site T89, except for the slight increase in exponential growth when changed to an I residue, changes to H, R, and K all resulted in a decrease or loss of growth capacity on L-arabinose. Another important amino acid residue for L-arabinose transport of Gal2p is F223. When F223 was changed to Q, C, L, and S, the exponential growth rates increased, and F223Q significantly improved by 16%. We changed the amino acids N346 and N347 together and found that, even when changed to small side chain amino acids A and G, respectively, the L-arabinose transport capacity of Gal2p was not remarkably reduced. Random mutation of amino acids F350 and Y351 resulted in a loss in the L-arabinose transport capacity of Gal2p. Mutations to the amino acid residue N376 in TMS 8 of Gal2p also presented complex effects on L-arabinose transport. Changing N376 to Y, R, and K, resulted in loss of Gal2p transport capacity. The mutations N376T, N376G, N376C, N376S, N376A, N376F, and N376I decreased the exponential growth rates by 16, 16, 16, 24, 28, 64, and 68%, respectively. Mutating the Y446 residue to A, I, C, or S resulted in a loss of transport capacity. This result indicated that Y446 is very conservative and is necessary for L-arabinose transport in Gal2p.

The enhanced L-arabinose transport capacity of positive mutations F85S, F85G, F85C, and F85T of Gal2p was further examined by L-arabinose metabolism. The L-arabinose metabolism capacity was evaluated by oxygen-limited batch cultivation. The L-arabinose metabolizing and ethanol producing were deeply concerned. The growth curves and L-arabinose metabolic trend are presented in Figure 3.

The L-arabinose metabolic efficiency positively correlated with the exponential growth rates on L-arabinose (Table 4). F85S and F85G exhibited rapid growth rates on L-arabinose and utilization of L-arabinose. During the 76 h fermentation, the Gal2p-expressing strain utilized 4.5 g L-arabinose and consumed only 22%. The L-arabinose utilization of F85S and F85G increased to 8.3 and 7.3 g, and the consumption amounts were 42 and 37%, respectively. Compared with the wild-type Gal2p-expressing strain, the consumption of L-arabinose by the F85S and F85G mutants increased by 84 and 62%, respectively. Furthermore, the F85S and F85G mutants produced 1.4 and 1.1 g L^-1 more ethanol than the wild-type Gal2p-expressing strain.

From the results above, we can predict that the change in F85 to polar amino acids will improve the L-arabinose transport capacity of Gal2p. The F85 site of Gal2p is also very conservative in the L-arabinose transporters expressed in S. cerevisiae and is located in the “G-G/F-XXX-G” motif in the first transmembrane span (Figure 4). The universal meaning of this key site was presented.

Endogenous Gal2p of S. cerevisiae can transport many sugars, including glucose, xylose and L-arabinose (Leandro et al., 2009). To test the glucose and xylose transport capacities of the positive L-arabinose transport mutations, F85S, F85G, F85C, F85T, and F223Q of Gal2p, these mutants were transformed into the BSW4EYX strain, which lost the hexose and pentose transport capacities and contained the XR/XDH pathway of xylose (Wieczorke et al., 1999; Wang et al., 2015). The exponential growth rates of the BSW4EYX expressing strains were also used to determine the change of growth capacity on glucose- (Figure 5A) and xylose-containing medium (Figure 5B). Unlike mutation F223Q, the F85G, F85S, F85T, and F85C mutations all decreased the glucose transport efficiency by 67, 45, 38, and 19%, respectively. This result indicated that the benzene ring of F85 in Gal2p might be important for glucose transport. For xylose transport of Gal2p, the F85C, F85S, and F85T mutations improved the xylose transport efficiency, and the exponential growth rates increased by 29, 23, and 15%, respectively. However, the exponential growth rate of the F85G mutant decreased by 19%, and F223Q did not change the growth efficiency on xylose.

To further test the effect of the best positive mutation F85S on the co-utilization and co-fermentation of glucose, xylose, and L-arabinose, we then transformed the plasmid YIp5-ara (Table 1) to the BSW4EYX strains expressing F85S or Gal2p. During a 96 h fermentation for the expressing strains, the mutation F85S decreased the glucose utilization efficiency (Figure 6A), obviously increased the xylose consumption amount (Figure 6B) by 1.1 g L^-1, and also increased the L-arabinose consumption amount (Figure 6C) by 1 g L^-1, compared with the original Gal2p. After about 40 h cultivation, the F85S expressing strain could exhibit more significant co-utilization of xylose and L-arabinose with glucose, and the ethanol production was 10 g L^-1 which was consistent with that of Gal2p.