Barley SWEET candidate genes were identified by combinations of conserved protein domain and BLAST searching methods. The barley reference genome and protein sequences (Mascher et al., 2017) were downloaded from Phytozome (https://phytozome-next.jgi.doe.gov/). A total of 17 AtSWEET proteins and 21 OsSWEET proteins identified previously (Chen et al., 2010) were obtained from corresponding Arabidopsis (Cheng et al., 2017) and rice (Ouyang et al., 2007) files in Phytozome. The Hidden Markov Model (HMM) of the MtN3_slv domain (PF03083), downloaded from Pfam 35.0 (http://pfam-legacy.xfam.org/), was used to identify putative barley SWEET proteins using HMMER (http://hmmer.org/) with the “trusted cutoff and E-value < 0.01” as the threshold. The 17 Arabidopsis SWEET proteins were used as queries to carry out a BLASTP search in the barley protein sequences with the E-value < 1e^-5 as the threshold using DIAMOND (Buchfink et al., 2021) (Version 2.0.11). The barley SWEET proteins, identified by both HMM and BLASTP methods, were manual checked and named in accordance with previously study (Mascher et al., 2017). The R package Peptides (Osório et al., 2015) (Version 2.4.4) was used to calculate the physical and chemical parameters of HvSWEET proteins, including protein length, molecular weight and theoretical pI. The Multiple sequence alignment of HvSWEET proteins was performed using MUSCLE program implemented in MEGA7 (Kumar et al., 2016) with the default parameters, and the phylogenetic tree was constructed by MEGA7 with the bootstrap of 1000 replications using neighbor-joining (NJ) method. The phylogenetic tree of SWEET family proteins of Arabidopsis, rice and barley was constructed using the same method without OsSWEET7d/LOC_Os09g08490, for short protein length (63 aa). The MEME software (Bailey et al., 2009) (Version 5.0.5) was used to investigate the conserved motifs of HvSWEET proteins with the parameters ‘-mod anr –nmotifs 10 –minw 6 –maxw 200’. The amino acid sequences of 10 conserved protein motifs were plotted using R package ggseqlogo (Wagih, 2017). The phylogenetic tree, conserved protein motifs and gene structures of HvSWEET genes were shown with iTOL (https://itol.embl.de/).

The reference primary protein files of Arabidopsis (Cheng et al., 2017), rice (Ouyang et al., 2007), maize (Jiao et al., 2017), sorghum (McCormick et al., 2018), barley (Mascher et al., 2017), and wheat (Zhu et al., 2021) were downloaded from Phytozome, and used to perform genome-wide syntenic analysis within barley or between barley and other plant species. Proteins of plant species were subject to homologous searching by DIAMOND BLASTP with the paramenters ‘–evalue 1e^-10 –max-target-seqs 5’. The MCScanX (Wang et al., 2012) was used to deal with the BLASTP results to identify the gene duplication events and the synteny results were visualized by Circos (Krzywinski et al., 2009) (Version 0.69-8) or python version JCVI (https://github.com/tanghaibao/jcvi) of MCScan. The OrthoFinder (Emms and Kelly, 2019) was used to make gene family clustering. The orthogroups, containing three genes in wheat and only one gene in other plant species, were identified and the genes, with two wheat genes randomly discarded, were derived to assemble super genes to construct maximum-likelihood (ML) species phylogenetic tree using IQ-Tree (Nguyen et al., 2015) (Version 1.6.12) with the parameters ‘-m MFP -bb 1000’.

To investigate the HvSWEET gene expression patterns, the RNA-seq raw sequencing data of 16 different barley developmental tissues (leaf, root, inflorescence, etc) (Mascher et al., 2017) and seed tissues during germination (embryo, aleurone, scutellum and grain) (Betts et al., 2017) were downloaded from NCBI, and filtered using fastp (Chen et al., 2018) (Version 0.12.4) with the default parameters. The high-quality cleaned reads were aligned to the barley reference genome (Mascher et al., 2017) with HISAT2 (Kim et al., 2019). Following alignments, raw counts for each gene were derived using featureCounts implemented in R package Rsubread (Liao et al., 2019), and normalized into the number of transcripts per kilobase of exon sequence in a gene per million mapped reads (TPM) with TMM method (Robinson and Oshlack, 2010). Heatmaps of HvSWEET gene expression profiles were generated using the R package ComplexHeatmap (Gu et al., 2016) (Version 2.10.0) based on the log₂ (TPM +1) transformation or expression proportions of HvSWEET genes in that of the whole gene family calculated by TPM values.

The exome SNP data of 360 barley accessions (20 wild accessions, 166 landraces and 174 cultivars) with clearly known breeding history from previously published study (Bustos-Korts et al., 2019) were derived and used to analyze the genetic diversity of HvSWEET genes. The SNPs with missing data >10% or minor allele frequency (MAF) < 5% were filtered using VCFtools (Danecek et al., 2011) (Version 0.1.17), and then the missing data were imputated using the package beagle (Browning et al., 2018) (Version 5.4) with the default parameters. The SNPs in HvSWEET genes were extracted to perform haplotype analysis and annotated according to the barley genome (Mascher et al., 2017) using the package ANNOVAR (Wang et al., 2010) (Version 2019-10-24). The pairwise linkage disequilibrium (R²) of SNPs in HvSWEET genes were calculated using PLINK (Purcell et al., 2007) (Version 1.90) and displayed using R package LDheatmap (Shin et al., 2006) (Version 1.0-6). The network 10 (https://www.fluxus-engineering.com/) was used to construct median-joining network of different haplotypes (Bandelt et al., 1999).

Primers were designed according to open reading frames of HvSWEET1a, HvSWEET4 and AtSWEET1 genes. The corresponding sequences were amplified from cDNA of barley (Golden Promise) or Arabidopsis, and cloned into pDONR201 vector (Invitrogen). The clones were selected by PCR and sequenced to confirmation. For subcellular localization of HvSWEET1a and HvSWEET4 in tobacco leaves, the corrected pDONR201 vectors were recombinant with destination vectors pH7WGF2.0 to obtain GFP-HvSWEET1a and GFP-HvSWEET4 constructs using Gateway system. For complementation of yeast mutant EYB.VW4000 (Wieczorke et al., 1999), lacking 18 hexose transporters, and subcellular localization of SWEET proteins in yeast cells, the seamless cloning was used to introduce SWEET gene and GFP gene amplified from pFA6a-GFP(S65T)-His3MIX6 vector, into yeast expression vector pYEPlac195 vector, which was inserted by ADH1 promoter amplified from pADGT7 vector in advance, to obtain pADH1-SWEET-GFP(S65T) constructs. The primer information is listed in Supplementary Table 1 .

The subcellular localization vectors were transiently expressed in tobacco (Nicotiana benthamiana) leaves by Agrobacterium-mediated infiltration. AtPIP2A-meCherry was used as plasma membrane (PM) marker. The Agrobaterium strain C58C1 harboring p19 was used to prevent the onset of PTGS (post-transcriptional gene silencing) in the infiltrated leaves. Infiltrated tobacco plants were grown for another 3 days for GFP and mCherry imaging using a Zeiss LSM710NLO confocal laser-scanning microscope. Excitation/emission wavelength were 488 nm for GFP, and 561 nm for mCherry.

The yeast complementation vectors or empty vectors were transformed into the hexose-uptake deficient yeast mutant EYB4000 (Wieczorke et al., 1999). Then, transformed yeasts were screened in synthetic dropout (SD)-Ura media, supplemented with 2% maltose. For complementation growth assays, yeasts were grown overnight in liquid SD media to an optical density at 600 nm (OD₆₀₀) of ~0.6, then OD₆₀₀ was adjusted to ~0.3 with water. Five-microliter aliquots of serial dilutions were plated on SD media containing 2% maltose (as control) or 2% other hexoses. Growth photographs were taken after incubation at 30 °C for three days.

For subcellular localization of SWEET proteins in yeast cells, transformed yeasts cultured in SD media supplemented with 2% maltose were collected, washed three times with water, and then applied on microscope slide. Fluorescence signals were detected using a Zeiss LSM710NLO confocal laser-scanning microscope. Excitation/emission wavelength were 488 nm for GFP signal.

A total of 23 SWEET genes were identified in barley, consistent with previously studies (Mascher et al., 2017; Qin et al., 2020), and were named accordingly ( Supplementary Table 2 ). We analyzed the characteristics of HvSWEET genes, including exon number, protein length, the number of MtN3/saliva domain, molecular weight (MW) and isoelectric point (pI), and found that most proteins contained two MtN3/saliva domains and four contained only one domain (HvSWEET7b/c and HvSWEET15b/c), the protein length of HvSWEET ranged from 91 amino acid (aa; HvSWEET7b/c) to 353 aa (HvSWEET13a), the MW ranged from 10.19 kDa (HvSWEET7b/c) to 38.74 kDa (HvSWEET13a), and the pI ranged from 6.51 (HvSWEET12) to 11.3 (HvSWEET15c) ( Supplementary Table 2 ).

To investigate the evolutionary relationship between HvSWEET and SWEET proteins from other species, including Arabidopsis and rice, we constructed a neighbor-joining (NJ) phylogenetic tree and found that HvSWEET proteins could be clustered into four clades, with clade I containing five members (HvSWEET1a/b, HvSWEET2a/b and HvSWEET3), clade II containing seven members (HvSWEET4, HvSWEET5, HvSWEET6a/b and HvSWEET7a/b/c), clade III containing 10 members (HvSWEET11a/b, HvSWEET12, HvSWEET13a/b, HvSWEET14a/b, HvSWEET15a/b/c) and clade IV containing one member (HvSWEET16), respectively ( Figure 1 ). A small extension of HvSWEET11, HvSWEET13, HvSWEET14 and HvSWEET15 were observed, with each containing two or three members in barley compared with only one orthologue in Arabidopsis and rice, consistent with previous study (Mascher et al., 2017).

Given that segmental and tandem duplications play important roles in gene family expansion during evolution, we detected these duplication events, involving HvSWEET genes, in barley. Two segmental duplications (HvSWEET1a/b, HvSWEET11a/b) and two tandem duplications (HvSWEET6a/b, HvSWEET15b/c) were observed ( Figure 2A ). Additionally, HvSWEET genes were observed uneven distributed over barley chromosomes, with chromosome 6 containing five members, chromosome 1 and 7 each containing four members, chromosome 3 and 4 each containing three members, chromosome 2 containing two members, and chromosome 5 and chromosome unscaffold each containing one member, respectively ( Figure 2A ).

To explore the evolution relationships of SWEET family genes between barley and other plant species, we performed synteny analysis between barley and Arabidopsis or Poaceae plant species, including rice, maize, sorghum and wheat. While only one pair of orthologous SWEET genes was observed between barley and Arabidopsis, which might be caused by evolutionally far genetic relationship between dicots and monocots ( Figure 2B ), there were six, eight, nine and 51 pairs of orthologous genes identified between barley and rice, maize, sorghum and wheat, respectively ( Figures 2C, D ). The closer evolution relationship and high ploidy levels of wheat might lead to more syntenic orthologous SWEET gene pairs between barley and wheat ( Figure 2B ).

To gain more insights into the characteristics of HvSWEET genes, the gene structural diversity was examined, and 11 members were found to have five exons, six members six exons, five members four exons and one member three exons ( Figure 3C ). The exon lengths were similar, whereas the intron lengths varied, with HvSWEET4, HvSWEET15c and HvSWEET16 containing very long introns ( Figure 3C ). Furthermore, we found clade I members contain six exons except HvSWEET2b containing five exons, clade II members contain five exons except HvSWEET7b and HvSWEET7c containing four exons, clade IV member contains six exons, and clade III members contain three to six exons ( Figures 3A–C ), suggesting that the gene structures of clade I and clade II are relatively conserved.

To further investigate the gene structural diversity, the conserved motifs of all HvSWEET proteins were examined. In total, ten conserved protein motifs, with the range from eight to 41 aa ( Supplementary Figure 1 ), were identified, among which the group containing motif 1, 8/10 and 2 and the group containing motif 1, 3 and 7 were annotated as MtN3_slv domain ( Figure 3B ). In addition, the result demonstrated that the motif 1, 2, 3, 6 and 7 exist in most HvSWEET members (96%, 83%, 91%, 61% and 74%, respectively), and motif 4 was identified only in the clade III. The HvSWEET protein in the same clade share relatively similar conserved motifs.

To comprehensively study the physiological functions of HvSWEET genes, the raw RNA-seq sequencing data of 16 different developmental tissues and seed tissues during germinating were derived from previously study (Betts et al., 2017; Mascher et al., 2017), and used to map against barley reference genome (Mascher et al., 2017). Different expression patterns were observed for HvSWEET genes in the detected tissues ( Figure 4A ). Five HvSWEET genes, including HvSWEET1a, HvSWEET2a/b, HvSWEET4 and HvSWEET15a, were expressed in almost all detected tissues, whereas four HvSWEET genes, including HvSWEET5, HvSWEET12 and HvSWEET15b/c, were nearly not detected ( Figure 4A ; Supplementary Tables 3, 4 ). Moreover, several HvSWEET genes were expressed during seed germination stages, such as HvSWEET7b/c, and several genes showed high expression levels in specific tissues, such as HvSWEET11b in developing caryopsis and inflorescence rachis, and HvSWEET13a/b in senescing leaf and epidermal strips ( Figure 4A ; Supplementary Tables 3, 4 ). Additionally, we examined the expression patterns of duplicated HvSWEET gene pairs, and found that some duplicated genes exhibited divergent expression patterns, such as SWEET1a/b and HvSWEET11a/b, suggesting they underwent neofunctionalization after duplications ( Figure 4A ; Supplementary Tables 3, 4 ).

Given the functional redundancy of HvSWEET genes, the relative expression level determines which member plays the main role in sugar transporting in the specific tissue within the gene family. We calculated proportion of each HvSWEET gene expression in that of the whole family, and observed that the sugar transporting activities of the gene family in leaf, senescing leaf, epidermal strips and root were mainly dependent on HvSWEET13a, the inflorescence, rachis and lodicule mainly dependent on HvSWEET11b and HvSWEET15a, the lemma and palea mainly dependent on HvSWEET4, the developing caryopsis mainly dependent on HvSWEET11b, and the germinating seed aleurone, embryo and scutellum mainly dependent on HvSWEET1a, HvSWEET15a and HvSWEET4, respectively ( Figure 4B ). These results demonstrate HvSWEET genes function redundantly and divergently to mediate sugar transport in barley developmental growth.

Barley was selected by early humans in the Fertile Crescent ground 10,000-12,000 years ago and is primarily used for animal feed, and malting and brewing, with a small percentage devoted to human food (Stein and Muehlbauer, 2018). The uniform and fast germination is a typical domestication syndrome trait, which ensure even maturity and enable crop management (Fuller and Allaby, 2009). Moreover, high malting quality is of importance for barley and also requires uniform and fast seed germination, during which the malting grains develop the enzymes required for modifying starches into various types of sugar to fuel the embryonic axis growth (Ma et al., 2017). Given overexpression of the sugar carrier AtSWEET16 resulted in improved seed germination rates (Klemens et al., 2013), we thought HvSWEET1a and HvSWEET4, relatively high expressed in aleurone and scutellum during seed germination, respectively, might be related with seed germination and could undergo artificial selection during barley domestication (wild accessions versus landraces) and improvement (landraces versus cultivars) ( Figure 4B ). To verify our hypothesis, we detected the genetic variations of HvSWEET1a using the exome SNP information of 360 accessions, consisting of 20 wild accessions, 166 landraces and 174 cultivars, from previously study (Bustos-Korts et al., 2019). In total, 20 SNPs, consisting of two SNPs in 5’ UTR, seven SNPs in introns and 11 SNPs in 3’ UTR, were detected in HvSWEET1a genetic region and classified the population into 27 haplotypes, each represented by one to more than one hundred accessions ( Figure 5A ). Median-joining network analysis sorted the 27 haplotypes into three main groups, namely H_I group (including H_1-1 to H_I-8), H_II group (including H_II-1 to H_II-16) and H_III group (including H_III-1 to H_III-3) ( Figure 5B ). Allele-frequency analysis demonstrated that the proportion of H_I significantly increased in landraces compared with that of wild accessions, and further increased in cultivars, indicating strong artificial selection of HvSWEET1a during barley domestication and improvement ( Figure 5C ). In addition, two groups of SNPs in the 3’ UTR region of HvSWEET1a gene showed strong linkage disequilibrium (LD), which might be the artificial selected regions ( Figure 5A ).

Similarly, the genetic variations of HvSWEET4 were also examined and nine SNPs were detected, which sorted the population into 12 haplotypes ( Supplementary Figure 2 ). Median-joining network sorted the 12 haplotypes into three main groups, namely H_I group (including H_I-1), group H_II (including H_II-1 to H_II-8) and group H_III (including H_III-1 to H_III-3) ( Supplementary Figure 2 ). However, no significant allele-frequency variations of three main groups were observed among wild accessions, landraces and cultivars ( Supplementary Figure 2 ).

To validate the sugar transporter function of HvSWEET1a and HvSWEET4 in yeast, the N-terminal HvSWEET1a or HvSWEET4 in fusion with GFP were transformed into hexose-uptake deficient yeast EYB.VW4000 (Wieczorke et al., 1999), which lacks 18 hexose transporters and could not grow on media with hexose as the only carbon source, with AtSWEET1 as the positive control. The results revealed that the plasma membrane (PM) localized HvSWEET1a and HvSWEET4 could transport glucose, galactose and fructose in yeast cells, and HvSWEET1a showed higher transport activity than HvSWEET4 ( Figures 6A, B ). To study the subcellular localization of HvSWEET1a and HvSWEET4 in plant cells, GFP fused with HvSWEET1a or HvSWEET4 was co-expressed with Arabidopsis PM intrinsic protein 2A fused to mCherry (AtPIP2A-mcherry) (Prak et al., 2008) in tobacco leaves. GFP fluorescence signals of HvSWEET1a or HvSWEET4 coincided to mCherry signals of PM maker in tobacco leaves ( Supplementary Figure 3 ). These results demonstrate that HvSWEET1a and HvSWEET4 function as PM hexose transporters.