The basic/helix-loop-helix (bHLH) proteins are a superfamily of transcription factors identified in different species (Toledo-Ortiz et al., 2003; Li et al., 2006). To determine all bHLH-related proteins and encoding genes in maize, BLASTp was performed to search against the maize genome dataset (ZmB73_5b_FGS_translations.fasta.gz) with an E <1 ×10⁻⁵, using rice and Arabidopsis bHLH proteins as query sequences. Proteins containing bHLH domain were defined as bHLH superfamily and further verified using SMART (http://smart.embl-heidelberg.de/). The expression data of the bHLH superfamily was mainly based on the RNA-sequence data downloaded from the MaizeGDB (https://www.maizegdb.org/), which included more than 60 distinct tissues representing different development stages of the 11 major organs (Stelpflug et al., 2016). Tbtools (Chen et al., 2020) was used to build expression heat maps.

Co-expression analysis was performed based on the method of Fu and Xue (Fu and Xue, 2010). Twenty-seven highly expressed starch biosynthetic related genes in maize endosperm were considered as guide genes (Nelson and Pan, 1995; James et al., 2003). The Pearson Correlation Coefficients (PCC) between the transcription factors of the bHLH family and guide genes were calculated based on the RNA-sequence data. If the absolute value of a PCC is >0.60, the probe will be considered to be associated with the guide gene. The genes associated with more than six guide genes were classified as candidate factors involved in maize starch biosynthesis.

Maize inbred line Mo17, provided by the Maize Research Institute of Sichuan Agricultural University (SAU), was grown on the Research Farm of SAU in Wenjiang according to the agronomic guidelines. The roots, stems and leaves were obtained after the maize was in the initial jointing stage of growth. Anthers were collected during the dispersal period, and silks were obtained before emerging from the husks. Kernels of 15 days after pollination (DAP) were collected for semi-quantitative RT-PCR analysis. For qPCR analysis, whole kernel, embryo and endosperm were obtained at different development stages after pollination. All tissue samples were immediately frozen in liquid nitrogen and stored in a refrigerator at −70°C for use in further experiments.

Total RNA was isolated using TRIzol reagent (Invitrogen) for specific tissues of the Mo17 according to the manufacturer's instructions. First-strand cDNA was synthesized from 1.5 μg of total RNA using the PrimeScript^TM RT reagent Kit with gDNA Eraser (TaKaRa, Japan). According to RNA-sequence information, the full-length cDNA of ZmbHLH175 (ZmICE1) was cloned from maize seed at 15 DAP. The sense and anti-sense primers were bHLH175F: 5'- TCAGGATCCATGGACGACTCGG-3' and bHLH175R: 5'-GAGCTCCTACATTGCGTTGTGGAG-3'. The PCR was performed under the following conditions: 20 μL of reaction mixture containing 10 μL 2^*KOD Buffer, 3 μL dNTPs, 0.3 μL KOD Neo, 1 μL cDNA template, 4.7 μL H₂O, 0.5 μL bHLH175F and 0.5 μL bHLH175R, was pre-denatured at 94°C for 4 min, followed by 35 cycles of 30 s at 94°C, 30 s at 60°C and 90 s at 72°C. At least three independent PCR replicates were performed. The PCR products were fractionated on 1.5% agarose gel containing GoldView™ and photographed under UV light. Genomic DNA of inbred line Mo17 was used as the template to clone the promoters of starch synthesis genes. All gene fragments were cloned into pMD19-T vectors (Takara, Dalian, China) by using KOD enzymes (Toyobo, Osaka, Japan) for sequencing.

Using ZmICE1 as query sequence, the full-length protein sequences of ICE subfamily in several plant species, including Arabidopsis, rice, maize, sorghum, and foxtail millet, were identified by Blastp. Multiple sequence alignment and phylogenetic tree construction were implemented in MEGA7.0 by the neighbor-joining method (Kumar et al., 2016). Bootstrapping test of tree was performed using 1,000 sampling repetitions. The syntenic relationships of the genes of several plants were analyzed by Tbtool. Exon/intron structures were analyzed using the GENE Structure Display Serve 2.0 (http://gsds.cbi.pku.edu.cn/). The conserved domains of the putative ICE proteins were identified by searching against the NCBI's conserved domain database (https://www.ncbi.nlm.nih.gov/Structure/cdd/wrpsb.cgi). The conserved motifs with default parameters were analyzed using MEME (http://meme-suite.org/tools/meme), and the maximum number of motifs was set as 10.

1.5 μg RNA from different tissues was used as template for the synthesis of the first-strand cDNA, which was used for semi-quantitative RT-PCR and real-time PCR. The sense and anti-sense primers, bHLH175QF: 5'-GCTTCAACCCATCAACACC-3' and bHLH175QR: 5'-CTTCCCTCATCCTGACTTCG-3', were used for semi-quantitative RT-PCR and real-time PCR analysis. The β-actin was amplified as the internal control, with primers ACTINF: 5'- CTGGAATGGTCAAGGCTGGT-3' and ACTINR: 5'-CTGGAATGGTCAAGGCTGGT-3'. The PCR was performed using 20 μL of reaction mixture containing 2 × Premix Taq, 8 μL H₂O, 1 μL template, 0.5 μL bHLH175QF and 0.5 μL bHLH175QR. The reaction mixture was pre-denatured at 94°C for 4 min, followed by 30 cycles of 30 s at 94°C, 30 s at 59°C and 30 s at 72°C. All PCRs were repeated for use for at least three independent samples. The PCR products were fractionated on 1.5% agarose gel containing GoldView™ and photographed under UV light.

The quantitative real-time PCR analysis was performed using the iCycler instrument, model 5.0 (Bio-Rad, Hong Kong), in a total reaction volume 10 μL of SYBR Green PCR Master Mix (TaKaRa, Japan). The Ct value was determined using the instrument's software. The relative quantification of gene expression was monitored after normalization to β-actin expression as the internal control. Relative transcription levels were calculated using the -ΔΔCt method based on the internal reference expression of maize β-actin.

The pGBKT7 containing the GAL4 binding domain was used to analyze the transcriptional activity of ZmbHLH175 in yeast system. The ZmbHLH175 was cloned into the pGBKT7 vector with the Sense primer (5'-CATATGGACGACTCGGCGGAG-3') and Anti-sense primer (5'-GGATCCCTACATTGCGTTGTGGAG-3'). The restriction enzyme sites were NdeI and BamHI. The completed carrier pGBKT7-ZmbHLH175 was transformed into the yeast strain AH109 to investigate the transactivation of the transcription factor. The transformants were screened on the SD/-Trp plates, grown for 3 days, and mono-clones were picked for propagation in a 2 ml micro-tube. The colonies were further screened on SD/-Trp-His-Ura plates with X-α-gal and cultured at 28°C for 3 days to analyze transcription activation activity.

The pCAMBIA2300-35S-eGFP was used for subcellular localization analysis. ZmbHLH175 with KpnI and BamHI sites was amplified without the termination codon by PCR and cloned into pCAMBIA2300-35S-eGFP. The sense and anti-sense primers were 5'-GGTACCATGGACGACTCGG-3' and 5'-GGATCCCATTGCGTTGTGG-3', respectively. The constructed pCAMBIA2300-35S-ZmbHLH175-eGFP was bombarded into onion epidermal cells through the helium biolistic gun transformation system (Bio-Rad, USA) according to previous method (Hu et al., 2012), and incubated in darkness for 24 h at 28°C. The sub-cellular localization of eGFP fusion proteins was visualized with a florescence microscope ECLIPSE 80i under blue excitation light with 488 nm wave length.

The modified pBI221 plasmid, pUbi:GUS, was constructed based on previous method (Hu et al., 2012) and was used for transient expression assay in particle bombardment experiments. pUbi:GUS was used as the internal control. Pro:Adh:LUC (Pro represents the different promoters of genes related to starch synthesis) and Ubi:ZmICE1 were constructed as the reporter and effector plasmids, respectively. Pro:Adh:LUC contained the Adh1 intron 1 (Adh) between Pro and Luc to enhance promoter activity (Mascarenhas et al., 1990). The promoter fragments were sub-cloned into the pBI221 with the restriction sites of BamHI and PstI, and all the Pro:Adh:LUC vectors with different promoters were provided by the lab. Maize endosperms were isolated from kernels at 10 DAP and cultured on Murashige and Skoog (MS) medium, and were subsequently transfected by particle bombardment biolistic system. The bombarded tissues were incubated for 24 h and the GUS and LUC activities were measured.

Yeast one-hybrid assays were essentially performed to test whether ZmICE1 can bind to the promoters of genes related to starch synthesis. The full length of the promoters and their truncated derivatives were separately cloned into the pHis2 vector with EcoRI and MluI restriction sites named pHis2-promoter (fragment). The CDS sequence of ZmICE1 was cloned into pGADT7-Rec2 with NdeI and BamHI restriction sites named pGADT7-Rec2-ZmICE1. The pHis-53 and pGADT7-Rec2-ZmICE1 constructs were used as a negative control, and pGADT7-Rec2-ZmICE1 and pHis2-promoter (fragment) constructs as the experimental group. Yeast strain Y187 was used as the host and transformed with the control and experimental constructs. The interaction between promoter (fragment) and ZmICE1 was detected according to the growth status of yeast cells cultured on SD/-His/-Leu/-Trp selective medium with 3-amino-1,2,4-triazole (3-AT) at the final concentration from 50 to 150 mM growth for 3 day.

Sequences of bHLH proteins in Arabidopsis and rice (Toledo-Ortiz et al., 2003; Li et al., 2006) were used to search the maize genome annotation database by BLASTp. In all, 202 genes encoding the bHLH proteins were identified and named from ZmbHLH1 to ZmbHLH202, including all the 175 annotated ZmbHLHs (from ZmbHLH1 to ZmbHLH175) in the MaizeGDB database (Supplementary Table S1). To investigate the expression pattern of the ZmbHLH superfamily, we analyzed the expression of all ZmbHLH genes based on the RNA-sequence data. The results showed that the genes ZmbHLH167 (GRMZM2G147685; ZmO11) and ZmbHLH175 (GRMZM2G173534; ZmICE1) were significantly expressed in maize seed, especially in seed endosperm (Figure 1). It was suggested that ZmO11 and ZmICE1 might exert certain functions in endosperm. Co-expression analyses showed that both ZmO11 and ZmICE1 showed a positive co-expression relationship with 14 starch biosynthetic genes, including ZmGBSSI, ZmSSIIa, ZmBT1 and ZmBt2 (Supplementary Figures S1A,B, Supplementary Table S2). The result suggests that ZmO11 and ZmICE1 may be candidate genes involved in the regulation of starch biosynthesis in maize endosperm. The role of ZmO11 in developmental seeds as a hub regulator had been recently reported (Feng et al., 2018). However, to date, ZmICE1 and its orthologous gene in Arabidopsis (AtICE1) were reported to be involved in abiotic stress processes (Chinnusamy et al., 2003). The role of ZmICE1 gene in seed development and starch biosynthesis is uncertain yet.

Based on the protein sequence of ZmICE1, the ICE subfamily members were identified in 5 plant species, including Arabidopsis thaliana, Zea mays, Sorghum bicolor, Setaria italica and Oryza sativa (Supplementary Table S3). There are 3 members in maize, and 2 members in each of the other plant species. To reveal the origin of each member, syntenic and phylogenetic relationships were investigated. The member of the ICE1 gene is located in the syntenic chromosome regions of both monocots and dicots (Figure 2A), instead of AtICE2, because AtICE2 originated from a recent duplication within Brassicaceae (Kurbidaeva et al., 2014). For cereals, the origin of duplication of ICE1 and ICE2 was earlier than the species differentiation of Poaceae, so the two members exist in all cereals and formed two main branches in the phylogenetic tree (Figure 2B). In maize, the synteny analysis revealed the possible origin of ZmICE2 and ZmICE3 by maize species-specific whole-genome duplication (mWGD) (Wang et al., 2015). We obtained 20 homologous gene pairs that are present in both genomic fragments encompassing ZmICE2 on chr2 and ZmICE3 on chr4 (Supplementary Figure S2, Supplementary Table S4). Moreover, the order of the genes and the orientation of the chromosome fragment of the 20 homologous gene pairs are conserved. The gene names and Ks of each gene pair are listed in Supplementary Table S4. The Ks of most gene pairs were similar to that of nucleotide substitution of mWGD. These data indicated that mWGD led to the origin of ZmICE2 and ZmICE3.

The analysis of gene structure revealed that the ICE subfamily genes had three to four exons, and all intron phases were “0.” As expected, most members of the ICE subfamily had similar intron-exon compositions, including the number and length of exons. All ICEs contain two conserved domains, the basic helix-loop-helix (bHLH) domain shown in green, and ACT domains shown in yellow (Figure 2C). The basic region may bind DNA to a consensus hexanucleotide sequence. ACT domains are commonly and specifically involved in the binding of amino acids or other small ligands. MEME analysis of the ICE subfamily proteins identified 10 putative conserved motifs. The number and distribution of motifs in each protein are similar, except that two AtICEs lack motif8 (Figure 2D).

Based on the reference genome sequence (Jiao et al., 2017), the full-length CDS of ZmICE1 (1350bp) was amplified by RT-PCR from the maize Mo17 inbred line. The nucleotide sequence of ZmICE1 was predicted to encode 449 amino acids, which is slightly longer than that of the genome annotation. Semi-quantitative RT-PCR was performed to analyze the expression of ZmICE1 in different tissues (Figure 3A). The results showed that ZmICE1 was expressed in all the tested tissues (root, stem, leaf, anther, silk and kernel), but showed a higher expression in seed. Then, the gene expression pattern of ZmICE1 in maize kernel, endosperm and embryo at different developmental stages was further investigated by quantitative PCR analysis. During the kernel filling period, ZmICE1 exhibited a high expression level between 15 and 30 days after pollination (DAP). The transcript accumulation was observed mainly in endosperm (not in embryo), with a peak at 25 DAP (Figure 3B). Thus, it is suggested that ZmICE1 may play an important role in seed endosperm development and also in starch biosynthesis.

A typical transcription factor contains specific functional domains, such as nuclear localization signal, activation domain, DNA-binding domain and oligomerization sites. The subcellular location of ZmICE1 protein was investigated in onion epidermal cells (Figure 4A). The 35S:eGFP construct was used as a control. As shown in Figure 4A, the ZmICE1:eGFP fusion protein was localized in the nucleus, and the 35S:eGFP control was located both in the cytoplasm and in the nucleus, indicating that ZmICE1 was a typical transcription factor and functioned in the nucleus.

The GAL4 system was used to analyze the transactivation activity of ZmICE1. The pGBKT7-Lam and pGBKT7-53 vectors were used as negative controls, and pGBKT7-GAL4 as the positive control. The constructs were transformed into yeast strain AH109 and cultured on the SD medium for detection. As shown in Figures 4B,C, the yeast cells without vector could not grow on the SD/–Trp and SD/–Trp–His–Ura; Negative control transformants could grow on SD/–Trp medium, but were not able to grow on the SD/–Trp–His–Ura and turn blue. The transformants of the positive control and ZmICE1 constructs were able to grow to turn blue on the SD/–Trp–His–Ura. These results suggest that ZmICE1 has transcription activation activity. So it is inferred that ZmICE1 has the potential to promote or inhibit the expression of different target genes, just as ZmbZIP91, which acts as a transcriptional activator of starch biosynthesis (Chen et al., 2016).

As mentioned above, ZmICE1 gene shared a positive co-expression relationship with 14 genes related to starch biosynthesis. To explore the potential regulatory role of ZmICE1 in starch biosynthesis, we analyzed the cis-elements in the promoter regions (upstream 2000bp of translation start site) of the 14 starch biosynthesis genes by a Python script. The result revealed that most of the promoters of the starch synthesis-related genes contained the consensus sequence “CANNTG” (Supplementary Table S5), which was the potential recognizable binding sites of ICE subfamily (Meshi and Iwabuchi, 1995).

Yeast one hybrid assay was then used to analyze the binding of ZmICE1 to promoters of starch synthesis genes. Promoters (upstream 1500–2000bp of the translation start site) of 12 key genes of starch biosynthesis (for example, ZmBt2 and ZmSh2) were cloned and used to evaluate their interaction with ZmICE1 in this study; except for ZmAGPL3 and ZmSSV, whose promoters were not constructed successfully. The constructs, pGADT7-ZmICE1-Rec2 and pHIS2 harboring the 12 promoters, were co-transformed into yeast strain Y187 and screened on the selective media. The results showed that ZmICE1 could directly bind to the promoters of ZmBt2, ZmGBSSI, ZmPul, ZmSSIIa, ZmISA1, and ZmSBEI (Figure 5).

The promoters of 14 co-expressed starch synthesis-related genes contained the core binding elements recognized by the ICE subfamily, with some of the promoters having the core elements at many sites (Supplementary Table S5). This indicated that ZmICE1 may recognize and bind to them. Yeast one-hybrid assay showed that ZmICE1 binds to the promoters of 6 starch biosynthetic genes. To further study its regulatory effect on the expressions of these starch synthesis related genes, transient gene expression assays were performed, and the relative activities of β-glucuronidase (GUS) and luciferase (LUC) were measured. The reporter, internal reference and effector plasmids were co-transformed into the maize endosperm as shown in Figure 6, except for ZmPul and ZmSBEI promoters that exhibited incredibly low promoter activity. The results showed that ZmICE1 could significantly enhance the promoter activity of ZmSSIIa. Conversely, ZmGBSSI promoter activity was significantly inhibited by ZmICE1. No obvious influence was observed on the activities of ZmBt2 and ZmISA1 promoters. These data suggest that ZmICE1 can regulate the expressions of ZmSSIIa and ZmGBSSI, and therefore participate in starch biosynthesis.

To further explore the binding of ZmICE1 to the ZmSSIIa and ZmGBSSI promoters, we amplified 3 fragments of ZmSSIIa promoter (pZmSSIIa-1, a-2 and a-3), and 2 fragments of ZmGBSSI promoter (pZmGBSSI-1 and I-2), which contained 1–4 putative core elements of “CANNTG” in varying numbers (Figure 7A, Supplementary Table S6). The pHIS2 constructs harboring these promoter fragments were constructed for the yeast one hybrid assay. The results showed that ZmICE1 could directly bind to the fragments of pZmSSIIa-2, pZmSSIIa-3 and pZmGBSSI-2, but could not bind to the fragments of pZmSSIIa-1 and pZmGBSSI-1 (Figure 7B). Based on the positions of the binding promoter fragments, it could be suggested that the fragments that are nearer to the translation start site, are easier to be bound by ZmICE1.