The Perilla frutescens var. japonica HARA and cultivars Namcheon, Dayu, and Yeopsil were used in this study. Perilla seeds were surface sterilized with 50% Sodium Hypochlorite for 1.5 h, followed by 5 times wash with sterile distilled water and then sown on MS medium (Murashige and Skoog, 1962) supplemented with 0.8% sucrose, 0.6% agar, and pH5.7 plates. Seeds were grown at 22°C for LD conditions (PPFD-100 µmol m^-2s^-1;16h light/8h dark). To measure flowering days and qRT-PCR experiments, Namcheon perilla seeds and genetic resources #22 and #57 were sown in a seedling box and then transplanted in pots when two leaves were developed and grown on a growth chamber.

Arabidopsis thaliana (wild-type Col-0) and ft-1 mutants were used in this study. Arabidopsis seeds were surface sterilized with 70% EtOH, washed 5 times with sterile distilled water, and then sown in MS medium plate. Seeds were grown at 22°C for 2 weeks, and then the seedlings were transplanted into sterilized soil and grown in a growth chamber for LD condition (PPFD-100 µmol m^-2s^-1;16h light/8h dark) and SD condition (PPFD-50 µmol m^-2s^-1;8h light/16h dark).

An mRNA in total RNA was converted into a library of template molecules to make a cluster sequence using the Illumina TruSeq RNA Sample Preparation Kit (Illumina, Korea). Initially, poly-A-containing mRNA molecules were purified by poly-T oligo-attached magnetic beads. Following purification, the mRNA is fragmented into small pieces using divalent cations under elevated temperatures. The cleaved RNA fragments are copied into first-strand cDNA using reverse transcriptase and random hexamers.

This is followed by second-strand cDNA synthesis using DNA Polymerase I and RNase H. These cDNA fragments then go through an end repair process, adding a single ‘A’ base and ligating the adapters. The products are then purified and enriched with PCR to create the final cDNA library. Paired-end sequencing was performed using a NovaSeq 6000 platform (Illumina, Korea). This transcriptional dataset has been submitted to the NCBI (https://www.ncbi.nlm.nih.gov) and will be released with the reference PRJNA858612.

Perilla’ Namcheon’, ‘Dayu,’ and genetic resources #22 and #57 seeds were sown in pots and grown at 25°C under LD (16h light/8h dark) conditions in a growth chamber, and the upper layers, including leaves and stems, were cut and sampled when 3 true leaves were developed. For RNA analysis, total RNA from samples was extracted using NucleoZOL Reagent (MACHEREY-NAGEL, Germany). 5 µg of the total RNA was used for first strand cDNA synthesis using Oligo dT (Invitrogen, USA) Primer, and cDNA was synthesized according to the protocol of the SuperScript IV First-Strand Synthesis System kit (Invitrogen, USA). Quantitative real-time PCR containing specific primers was performed with the TOPrealTM qPCR 2X Premix (SYBR Green with low ROX). The sequence of each gene was obtained from RNA-sequencing data, and primers were prepared for the experiment ( Supplemental Table S2 ).

Initially, we performed complementation experiments on the ft mutant. Perilla cDNA was used to isolate the coding sequence (CDS) of the PfHd3a gene for constructing plants with over-expression of the PfHd3a gene. The obtained DNA fragments (525 bp) were inserted into the pDONR221 vector using BP clonase (Invitrogen, USA) based on the Gateway system and further moved into the pK7FWG2 and pB2GW7 vectors by LR clonase (Invitrogen, USA). All constructs were transformed into Agrobacterium GV3101 using the electroshock method and transformed into Arabidopsis wild-type and ft mutant plants according to the floral dip method.

As Chen et al. (2020) described, a transient expression assay was performed. Overexpressed-PfHd3a transgenic plants were inoculated on MS medium and transplanted into the soil. After two weeks, we collected two primary leaves and thinly cut them with a blade before soaking them in 1M Mannitol for 30 minutes. After incubation, the sample was added to an enzyme solution (1% cellulase R-10, 0.25% mercerozyme R-10, MES, BSA, 500 mM Mannitol, 1mM CaCl₂) and gently shaken before being incubated at 22°C for 12 hours in dark conditions. Subsequently, the enzyme solution was filtered through a 100-mm mesh. The protoplast solution is loaded onto 20 ml of 21% sucrose and centrifuged at 5000 rpm for 10 minutes. The intact protoplast was extracted and observed under a fluorescent microscope after centrifugation. The excitation filter detected the GFP signal in the range 470–550 nm (transmitting only green light).

CRISPR/Cas9 reagents were cloned into the pBAtC binary vector having a whole CRISPR/Cas9 cassette to edit a target gene in perilla. Two sgRNAs (sgRNA1: TTACAAATGGCTGTGAATTT and sgRNA2: TACTGGAGCAACCTTTGGAC) were designed to recognize conserved regions in the coding sequence of the PfHd3a gene in perilla. Using Aar1 sites, both sgRNAs were ligated to the pBAtC vector and confirmed by sequencing. Constructs were transformed into Agrobacterium EHA105 using the electroshock method and transformed into perilla using plant tissue culture.

Perilla transformations were performed as previously described (Lee et al., 2005). Perilla Yeopsil seeds were sown in MS medium (1x MS, 0.8% sucrose, and 0.8% agar, pH 5.7) in LD conditions (16h light/8h dark, 22°C). The cotyledons and hypocotyls were cut into 0.8~1cm size from 7 to 10 days old plants and immersed in Agrobacterium suspension for 30 minutes. Then samples were placed on sterilized filter paper to remove moisture and co-culture medium (1% MS medium, 3% Sucrose, 3mg/L 6-benzylaminopurine (BA), 0.1 mg L-1 α-Naphthaleneacetic (NAA), 1mM acetosyringone, 0.4% Gelrite) and cultured at 25°C under dark condition. After 3 days, the explants were washed 3 times using liquid MS medium and put on sterilized filter paper to remove moisture and cultured by adding on shoot induction medium (1% MS medium, 3% Sucrose, 3 mg L-1 BA, 0.1 mg L-1 NAA, 500 mg L-1 carbenicillin, 1.2 mg L-1 phosphinothricin (PPT), 0.4% Gelrite). After 6 to 10 weeks, the new shoots regenerated from the cotyledons, and hypocotyls were cut from the callus sections and put on the shoot elongation medium (1% MS medium, 3% Sucrose, 3 mg L-1 BA, 500 mg L-1 carbenicillin, 1.2 mg L-1 PPT, 0.4% Gelrite), then the roots are induced by putting it on the root induction medium (1% MS medium, 3% Sucrose, 0.4% Gelrite). After that, the transgenic plant was transferred to a pot, grown in a greenhouse, and the seeds were harvested to obtain transformed perilla seeds.

A de novo transcriptome assembly of perilla was performed to generate a reference for high-throughput gene expression analysis and identification of critical flowering activator genes in the second leaf pair stage of young perilla plants grown under SD conditions. Differential expression of genes between perilla plants grown under SD and LD conditions was first analyzed using DEseq2 (Anders and Huber, 2010) with reference genes from the Ensembl Plants database (https://plants.ensembl.org/info/website/ftp/index.html). The raw data were then standardized and log2-transformed to form a scatter plot ( Supplemental Figure S1 ). A total of 5,725 differentially expressed genes [DEGs; false discovery rate (FDR) ≤ 0.05; |fold-change (fc)| ≥ 2] were identified by comparing the two groups of samples (SD and LD). Significantly, 2,739 upregulated and 2,986 down regulated genes were counted by fold change ( Supplemental Figure S2 ). We narrowed down the DEGs to select SD-specific flowering-related genes. We identified 18 novel perilla genes in the SD/LD comparison, of which 9 were upregulated and 9 were down regulated ( Figures 1A, B ). It was found that CONSTANS-like 9, AGAMOUS-like, and PSEUDO-RESPONSE REGULATOR 7 (PRR7) genes were upregulated, and many flowering-related genes such as CONSTITUTIVE PHOTOMORPHOGENIC 1 (COP1), REVEILLE 1 (RVE1),and CIRCADIAN CLOCK ASSOCIATED 1 (CCA1) were downregulated under SD conditions ( Figures 1A, B ). In particular, the Hd3a-like gene was highly expressed under SD conditions ( Figure 1A ), consistent with the levels of the OsHd3a gene, a major flowering activator in rice (Komiya et al., 2008). The perilla Hd3a-like gene was renamed PfHd3a. The expression of PfHd3a was validated using quantitative reverse transcriptase-polymerase chain reaction (qRT-PCR). The transcript level of PfHd3a in SD was > 300-fold higher than in LD ( Figure 1C ). In addition, PfHd3a was highly expressed in mature leaves but not in seedling plants ( Figure 1D ) by the expression of other FT homologs (Wickland and Hanzawa, 2015).

This suggests that FT functions as the main flowering activator in plants, and FT homologs are well conserved across many flowering plant species (Wickland and Hanzawa, 2015). To examine how widespread the PfHd3a protein is, a phylogenetic tree of PfHd3a was constructed using other plants. PfHd3a is highly homologous to other FT homologs, especially Malus x demestica FT2 (MdFT2), Beta vulgaris FT2 (BvFT2), and MdFT1 with 89%, 84%, and 84% identity, respectively ( Supplemental Figure S3 ). In addition, OsHd3a and AtFT showed 82% and 80% amino acid sequence similarity to PfHd3a, respectively ( Supplemental Figure S3 ). Notably, the amino acid alignment of PfHd3a with other FT homologs revealed that an external loop of protein sequences known as segment B, which is essential for FT function, was well conserved among FT homologs ( Supplemental Figure S4 , S5 ). It has also been reported that Tyr-85 and Gln-140 residues in segment B are critical for forming the ligand-binding pocket wall and are core residues for FT function (Laurie et al., 2011). Analysis of PfHd3a segment B showed that both residues were well conserved in the perilla homolog, similar to other FT homologs ( Supplemental Figure S5 ), suggesting that PfHd3a may function as the effective flowering activator in perilla.

Rice Hd3a and Arabidopsis FT transcript levels oscillate in distinct rhythmic patterns (Tsuji et al., 2015). We investigated the diurnal rhythm of expression of PfHd3a under SD and LD conditions by qRT-PCR. Interestingly, PfHd3a mRNA levels were diurnally regulated under SD and LD conditions. PfHd3a transcripts accumulated from the first appearance of light reached a peak at 8 h lights and then decreased until the end of the dark period ( Figures 2A, B ). To ascertain whether the circadian clock-controlled PfHd3a transcript levels, perilla grown under LD and SD conditions were transferred to continuous light or dark conditions, respectively, and PfHd3a transcript levels were verified. The rhythmic expression of PfHd3a quickly disappeared under continuous light and dark conditions ( Figures 2C, D ), indicating that PfHd3a is expressed rhythmically and is regulated by the photoperiod but not by the circadian clock.

To assess the potential role of PfHd3a in flowering time regulation, the gene was overexpressed in Arabidopsis using the cauliflower mosaic virus 35S promoter. The 35S:PfHd3a construct fused to green fluorescence protein (GFP) was transformed into Arabidopsis (Col) with Agrobacterium tumefaciens, and its flowering times were determined. At least 12 independents homozygous T3 lines were generated, and representative overexpressed-PfHd3a-GFP plants (III-1 and VI-1) showed early flowering under LD and SD conditions compared to Col plants ( Figures 3A–D ). PfHd3a-GFP protein expression was observed in protoplasts extracted from the overexpressing lines ( Figure 3E ). PfHd3a-GFP proteins were mainly expressed in the cytoplasm, as confirmed by western blotting using an anti-GFP antibody ( Figures 3E, F ). These results suggest that PfHd3a is involved in the regulation of flowering. To verify whether PfHd3a can complement the late-flowering of the Arabidopsis ft-1 mutant, the same 35S:PfHd3a-GFP construct was introduced into this mutant. As predicted, overexpressed-PfHd3a was able to fully complement Arabidopsis FT with rapid flowering compared to Col plants ( Figures 3G–I ). This adds to the findings that PfHd3a may act as a floral activator, similar to other FT orthologs.

As previously mentioned, flowering time is much earlier in seed varieties than in vegetable varieties. To evaluate the exact comparison of flowering time between seed perilla and vegetable perilla varieties, 90 perilla cultivars or perilla genetic resources were cultivated in the field, and their characteristics were analyzed ( Table 1 ). The flowering of the perilla varieties ranged from 57–175 d after sowing ( Figure 4A ). For seed perilla planted on May 20th, the flowering period was usually from the beginning of August to the middle of September. In contrast, vegetable perilla varieties flowered from the beginning of October to the center of November ( Figure 4B ).

The difference in flowering time between seed and vegetable perilla varieties prompted us to examine the induction time of PfHd3a transcripts under different day-length conditions. One seed variety, Dayu (#1), a vegetable variety, Namcheon (#40), and two perilla genetic resources, #22 and #57, which showed very late flowering, were chosen for this assay. First, when we examined the flowering time of the selected perilla under natural light, and SD (8L/16D) conditions; Dayu flowered rapidly under both conditions, followed by Namcheon, while #22 and #57 flowered late under both conditions ( Figures 4C–F ). To validate whether PfHd3a is involved in the different flowering times of these perilla varieties, the transcript levels of PfHd3a in each perilla were compared by qRT-PCR under different day-length conditions. PfHd3a transcript levels were deficient in all perilla under 14.5 L/9.5 D conditions, except for Dayu, which started increasing under 13.5 L/10.5 D conditions, and increased further with darker conditions ( Figure 4G ). In the case of Namcheon, PfHd3a transcript levels were detected from 12.5 L/11.5 D conditions, whereas PfHd3a levels appeared under 11.5 L/12.5 D conditions in #22 and #57 ( Figure 4G ). These results suggest that PfHd3a transcription is induced earlier in seed varieties compared to vegetable varieties and that the flowering time of seed perilla varieties is earlier than that of vegetable perilla varieties.

To further confirm the function of PfHd3a in perilla, a Korean perilla cultivar, Yeopsil (Lee et al., 2005), was transformed with the 35S:PfHd3a construct. As a control, we transformed additional perilla plants with an empty vector and regenerated the perilla from cotyledon explants that had not been infected with Agrobacterium tumefaciens. Eleven independently generated 35S:PfHd3a perilla was obtained, and the flowering time of T3 transgenic perilla (II-1 and III-1) was observed. Both transgenic perillas flowered significantly earlier than the regenerated control and parental perilla under SD conditions (10 L:14 D) ( Figures 5A, B ). When comparing the transcript levels of PfHd3a among transgenic and control perilla, approximately 11 and 13-fold higher levels were observed in the II-1 and III-1 lines, respectively ( Figure 5C ), reiterating that PfHd3a can induce flowering when overexpressed in perilla. To provide more direct evidence for the role of PfHd3a in flowering, we identified perilla plants carrying mutations in this gene. Gene-editing (GE) technology was employed using a reverse genetic approach. Two CRISPR-Cas9 target sites (guide (G) 1 and G2) were selected within the first and second exon of the PfHd3a gene ( Figure 6A ), and after preparing the oligo dimer, it was constructed into the CRISPR-Cas9 vector. Sequences were validated using U6 promoter primers. The correct construct was transformed in Yeopsil perilla using an Agrobacterium-mediated perilla transgenic method (Lee et al., 2005), and 10 positive plants were selected using phosphinotricin. To identify the mutation in the first and second exons of PfHd3a in transgenic lines, PCR amplification was performed, and the adjacent sequence of PfHd3a was sequenced in T0 generation plants. Five plants had mutations only in G1 of the first exon region. We measured the frequency of all mutant genotypes at the G1 target sites. We found that the main mutation types were single-base and amino acid frameshift mutations ( Figure 6B ). In the T2 generation, plants without exogenous DNA were selected. The flowering phenotype was observed under SD conditions (10 L/14 D). The gene-edited perillas showed significantly late flowering, resulting in approximately 50% leaf production enhancement compared to the control ( Figures 6C–E ), reiterating that PfHd3a is the main flowering activator in perilla.