The 35S promoter cassette of the vector pAMPAT-GW (GenBank accession no. AY436765, Pesch et al., 2015) was exchanged with the 898 bp 5′ sequence immediately upstream of the start codon of GL3 gene using AscI and XhoI [pAMPAT-GW-GL3(5′-1 kb)]. The 1051 bp 3′ fragment was cloned into the PmeI site of pAMPAT-GW-GL3(5′-1 kb) to create the pAMPAT-GW- GL3(5′-1 kb):LR recombination cassette:(3′-1 kb) vector. All genomic fragments of GL3 were cloned into pDONR201 by BP reactions (Invitrogen). Deletions of single introns within the genomic sequence of GL3 were introduced by PCR based site directed mutagenesis. The entry clone carrying the GL3 gene with the second intron was generated using the following strategy: an entry clone carrying the genomic GL3 was cut with EcoRV and KpnI generating a GL3 fragment that includes the second and third intron of GL3. This fragment was exchanged against the corresponding GL3 fragment without introns in the entry clone carrying the coding sequence of GL3. Thereafter, the third intron was deleted by PCR based site directed mutagenesis. Coding and genomic sequences of GL3 were introduced into pAMPAT-GW-GL3(5′-1 kb):LR recombination cassette:(3′-1 kb) by LR recombination with the respective entry clones to generate the various intron deletion constructs. Plants were transformed using the floral dip method described previously (Clough and Bent, 1998).

Plants were grown on soil at 24°C with 16 h of light per day. All Arabidopsis thaliana used in this study were of the Columbia (Col-0) ecotype. The gl3-3 mutant line has been described previously (Jakoby et al., 2008). egl3-77439 corresponds to the TAIR accession 1008704039.

Total RNA was extracted from 10-day-old true leaves using the RNeasy Mini Kit (Qiagen, Cat No./ID: 74106) and first-strand cDNA was then synthesized from the total RNA (1 μg) using the RevertAid H Minus 1st strand cDNA synthesis (Thermo) as described by manufacturer’s instruction. Real-time polymerase chain reactions (PCR) contained 1 μl of primer mix (10 μM), 1 μl cDNA template (10-fold dilution),10 μl 2 × SYBR Green master PCR mix and 8 μl water to a total of 20 μl. cDNA concentrations in different samples were normalized with reference to AtAct2. Gene-specific primers are listed in Supplementary Table S1.

GUS stainings were essentially done as previously (Sessions et al., 1999; Schroeder et al., 2016). After staining for 16 h at 37°C, tissues were cleared and leaves were inspected by light microscopy and pictures taken using the DISKUS software (Carl H. Hilgers -Technisches Büro, Germany). Trichome numbers were determined on the third and fourth fully expanded leaf of soil-grown seedlings.

It has been previously reported, that a 2.5 kb 5′-promoter fragment of GL3 fused to GUS reveals trichome specific expression in leaves and that a fusion to the GL3 cDNA can rescue the trichome phenotype in gl3 egl3 double mutants (Zhang et al., 2003; Zhao et al., 2012). In addition, it was shown that a genomic GL3 fragment including 1kb of the 5′-promoter was sufficient to rescue the gl3 trichome phenotype (Bernhardt et al., 2005). To test, whether the 1 kb 5′-promoter fragment is sufficient for trichome-specific expression or whether the introns are also important we created a pGL3(1 kb):GUS line. We observed ubiquitous GUS expression in young leaves (Supplementary Figures S1A–D). In older leaves, expression levels were close to background (Supplementary Figures S1E–G).

In parallel, we performed rescue experiments by expressing the GL3 cDNA under the control of the 1 kb 5′-promoter and 1 kb downstream of STOP codon (termed as 3′-1 kb) in the gl3-3 egl3-77439 double mutant. The gl3-3 single mutant shows about 50% reduction in trichome number whereas the egl3-77439 mutant shows a significant reduction of about 10% similar as reported for the egl3 allele in the Ler background (Table 1 and Supplementary Figures S2A,B) (Zhang et al., 2003). The gl3-3 egl3-77439 double mutant is completely glabrous and one would expect that rescued lines should exhibit the egl3 77439 mutant phenotype. Among 80 transformed T1 plants we found no rescue. All plants were completely glabrous (Supplementary Figures S2C,D).

Our data indicated that the 5′ promoter region and the 3′-1 kb are not sufficient for rescuing the trichome phenotype of gl3-3 egl3-77439 double mutant. We therefore tested the possibility that introns are relevant for the proper regulation of GL3 during trichome formation. Toward this end, we created a gateway construct containing 1 kb of the 5′ promoter region and 1 kb of the 3′-1 kb [called pGL3:GL3(genomic):3′-1 kb] such that the coding region can be replaced by recombination (Figure 1). This construct was used to study the rescue ability in gl3 egl3 mutants in the T1 generation. As expected, we found a range of rescue phenotypes in the T1 generation. The average rescue efficiency was used as a reference for subsequent analysis (Table 1). Next, we created a series of constructs each lacking one of the six introns (Figure 1). We found a clear rescue with constructs missing the third, fourth, or fifth intron. The deletion of the first or the sixth intron resulted in a weaker rescue of trichome formation. No rescue was observed in plants carrying the pGL3:GL3 (genomicΔintron 2):3′-1 kb construct (Table 1) indicating that the second intron is essential. We therefore focused in the following on the function of the second intron.

To test the relevance of the 3′-1 kb region we studied the rescue in lines harboring the pGL3:genomic GL3 construct (Figure 1). These lines showed a rescue of the trichome phenotype. Thus, the 3′-1 kb of GL3 is not necessary for trichome rescue.

In order to demonstrate that the second intron together with the 1 kb 5′-promoter fragment is sufficient for the transcriptional regulation of GL3 in the leaf, we expressed the GL3 cDNA containing the second intron at its original site under the 1 kb 5′-promoter [pGL3:GL3(Intron 2)] in gl3 egl3 mutants (Figure 2). The majority of T1 lines showed rescue of the trichome phenotype (Table 2). In an attempt to map potential relevant regions in the second intron, we compared two deletion constructs missing either the 5′ 125 nt (pGL3:GL3(Intron 2 delta 3-125) or 454 nt at the 3′end [pGL3:GL3(Intron 2 delta 126-579)] of the second intron. Only the construct containing the 125 nt at the 5′end rescued the gl3 egl3 mutant trichome phenotype (Table 2) suggesting that this fragment contains all regulatory sequences. In a next step, we assessed whether the position of the intron 2 is important. Toward this end we placed intron 2 in front of the 5′ promoter in both directions (Figure 2). Neither construct was able to rescue the gl3 egl3 trichome mutant phenotype (Table 2) indicating that intron 2 does not act as a transcriptional enhancer element. This suggested to us that its position in the transcribed region is important for its function. One well-characterized regulatory mechanism that requires the intron within the transcribed sequence in its correct orientation is intron mediated enhancement (IME) (Rose, 2002; Gallegos and Rose, 2017). To address this possibility, we created two constructs in which the second intron of GL3 was replaced by introns for which their ability to mediate IME is well characterized (Figure 2). The first intron of UBQ10 resulted in a 2- to 10-fold higher GL3 expression as compared to wild type Col (Figure 3). Insertion of the first intron of the Cor15a gene lead to wild-type levels or up to about two-fold increased GL3 expression (Figure 3). By comparison, constructs containing intron 2 or the first 125 nt of intron 2 could enhance GL3 expression up to three-fold (Figure 3). However, neither the UBQ10 nor the Cor15a constructs rescued the trichome phenotype (Table 2).

These results suggest that it is not sufficient to merely increase the GL3 expression by replacing intron 2. It is therefore conceivable that intron 2 is important for the proper regulation of the temporal and spatial expression of GL3.

Our analysis revealed that a 1 kb promoter fragment combined with intron 2 in the transcribed region of GL3 is sufficient for complete rescue. To study the expression pattern mediated by this construct we fused the GUS marker gene directly after the second intron. As the signal levels were very low when using X-Gluc as a substrate, we used the more sensitive magenta-Glc-A as a substrate (Schroeder et al., 2016). We detected GL3 expression in young leaves in all stages of trichome development Figures 4A,B). In addition we noted weak expression in epidermal pavement cells in young leaves (Figure 4A). In older leaves with young trichome stages at the leaf bases and mature trichomes at the tip of the leaf trichomes exhibited much stronger expression then the mature trichomes (Figure 4B). Low levels of GL3 were maintained during further leaf growth but disappeared in fully mature leaves (Figures 4C,D).