Predicting and testing a gene network regulating seed germination in Arabidopsis

A strategy for identifying separate groups of co-expressed auxin-responsive genes in the FUN-CSC continuum

The FUN and chalazal seed coat (CSC) form a tissue continuum in seed development. This FUN-CSC continuum is the only route for directly supplying nutrients and other molecules from the maternal tissues to the developing seed. Conceivably, the genes expressed in the FUN-CSC continuum can exert maternal effects on seed development and germination. Transcriptomic datasets for genes in parts of the Arabidopsis seed including FUN, CSC, and distal seed coat (DSC) at different developmental stages are available from Belmonte et al. (2013) and Khan et al. (2015). These datasets were conveniently compiled in a single Excel file by Khan et al. (Dataset S1 in Khan et al. (2015)). Moreover, datasets of auxin-up- and-downregulated genes are available from microarray studies (Goda et al., 2004). It is possible to identify components of auxin signaling networks in the FUN-CSC continuum by searching for genes present in both the FUN and CSC gene expression datasets and the auxin-regulated gene datasets. The expression trends (upward or downward from FUN to CSC) of these genes can also be compared with the expression trend(s) of the AFB genes. If the AFBs exhibit different expression trends in the FUN-CSC continuum, the identified auxin-regulated genes can be further assigned to separate groups based on their expression trends in the FUN-CSC continuum and their modes of response (up or down) to auxin. In particular, Dataset S1 in Khan et al. (2015) were downloaded and reduced to contain only the raw microarray signals of the genes expressed in FUN, CSC (corresponding to CZSC in Khan et al. (2015)), and DSC (corresponding to SC in Khan et al. (2015)) at the mature-green-cotyledon stage. The auxin-regulated genes in this reduced dataset were identified by manually searching for the auxin-regulated loci numbers reported in Goda et al. (2004). These auxin-regulated genes were further manually grouped based on whether they are down- or upregulated by auxin and whether they exhibit an up or a down expression pattern from FUN to CSC, as shown in Table S1. The grouping of AFBs was also conducted based on their up or down expression pattern from FUN to CSC (Table S1). When developing the gene groups in Table S1, genes whose raw microarray signal values in both FUN and CSC are less than 10 were omitted to reduce potential false positives in this investigation.

Linear regression analysis for assigning subgroups of co-expressed auxin-responsive genes in the funiculus and seed coat

The available gene expression data contain microarray signal values for a gene expressed in the FUN, CSC, and DSC regions. The three data points from FUN, CSC, and DSC should enable a linear regression analysis of in vivo mRNA concentrations between gene pairs after proper normalizations, which may provide additional evidence for assigning genes in the above-mentioned groups to subgroups. Assuming that the RNA isolation efficiency was the same for the FUN, CSC, and DSC tissues, the relative in vivo concentration of an mRNA can be defined by its microarray signal/its tissue volume. Because the tissues collected for the RNA isolation were from standard sections that had a fixed thickness, the volume ratios between any two of the FUN, CSC, and DSC regions are approximately equal to their area ratios. Based on the images provided by Belmonte et al. (2013), the areas of the FUN and CSC were approximately the same and the area of the DSC was approximately 26.8 times of that of FUN or CSC (areas were measured using SigmaScan 5.0, Systat Software). Therefore, we normalized the microarray signal values for DSC by dividing them by a factor of 26.8 (Table S1). We then conducted linear correlation analysis using the normalized values along with the original FUN and CSC values in Excel. We used a range of high coefficient values, 1 ≥ R² ≥ 0.90, to determine whether two genes belong to the same subgroup.

Confirmation of T-DNA insertion mutants

A total of 10 mutants at five loci (two at each locus) of the identified genes without a known function in seed germination, all in the Columbia (Col-0) accession, were obtained from the Arabidopsis Biological Resource Center at The Ohio State University. These were T-DNA-insertion mutants that were verified by polymerase chain reactions (PCR) using their genomic DNAs as templates and T-DNA and gene-specific primers and a routine PCR protocol on a thermocycler (Tables S2 and S3; Fig. S1). Transcription of these genes in the corresponding mutants and Col-0 was investigated by reverse transcription (RT)-PCR with a primer in the T-DNA insertion and a gene-specific primer (Tables S3 and S4) and cDNA samples synthesized using Thermo Scientific Verso cDNA synthesis Kit (Thermo Fisher Scientific, Waltham, MA, U.S.A.) and RNA samples extracted from inflorescences using RNeasy Plant Mini Kit (Qiagen, Valencia, CA, U.S.A.).

Testing seed germination in mutants of genes predicted to function in seed germination

The seeds for testing germination of a mutant line and the wild-type control Col-0 were harvested from plants that were grown at approximately the same time (planted on the same date and harvested zero to a few days apart) in the same growth chamber as previously described (Wang et al., 2022). The mutant and Col-0 seeds were air-dried at the same location and room temperature for the same duration (5–6 weeks) and were stored at the same location and room temperature in 1.5 mL cryogenic tubes with the screw cap tightened. At the time of the germination experiment, the seeds were less than a year old from the date of harvesting. Germination tests were conducted similarly to those previously described (Wang et al., 2022). Specifically, each piece of filter paper (9 cm in diameter) was placed in the lid of a petri dish (10 cm × 1.5 cm) and wetted with 2 mL double-deionized water. More than 100 seeds of each genotype were sprinkled onto the filter paper and covered with the bottom of the petri dish. The petri dishes were incubated upside down in a growth chamber at approximately 22 °C with approximately 100 µmol photons m⁻²s⁻¹ in light intensity and a 16 h-light/8 h-darkness regime. Seeds were scored for germination under a dissecting microscope at the time of 50-h imbibition. A seed was considered germinated if a region of the seed coat was broken. If the germination status could not be assessed due the downfacing of the radicle, the seed was flipped over with a syringe needle to determine its germination status. More than 100 seeds in each petri dish were scored to determine the germination frequency. The germination frequency of each genotype was determined eight or 12 times in separate petri dishes. For each mutant petri dish, one Col-0 petri dish was included as control; they were placed as one cluster of petri dishes in the growth chamber. For minimizing the difference in incubation time between the petri dishes due to the time spent in the seed counting process, four or fewer mutant-wild-type pairs of petri dishes were investigated in a single experiment. These measures should minimize the confounding effect of variations in seed germination between experiments at different times, and enable paired t-test between the mutant and corresponding wild-type samples.

Constructing a gene network for controlling seed germination

Three groups of highly coexpressed genes identified earlier (Table S1) included components in the auxin, ABA, and brassinosteroid signaling pathways that are known to act in the seed germination process. Therefore, it is possible that all genes in the three groups act in a gene network that regulates seed germination by integrating the hormonal signaling processes. To explore such a possibility, datasets of genes whose expression is regulated by the three hormones were downloaded from the supplementary data linked to the publications (Goda et al., 2004; Okamoto et al., 2010; Tian et al., 2020; Liu et al., 2020). The potentially seed-germination-related genes in the three groups identified in this investigation were manually searched in these datasets. These searches revealed in-group and between-group connections for the genes in the three groups, enabling the construction of a gene network. This network was first constructed in Cytoscape_3.9.1 (Fig. S2 and Data S1–S3) and then, for simplicity, in Microsoft PowerPoint. This network contains 30 identified genes in the three group, the six AFBs, BES1 and BZR1 involved in brassinosteroid signaling, and ABI3, 4, and 5 involved in ABA signaling.

Identification of auxin-down- or upregulated coexpressed genes in FUN and CSC

Following the strategy described in Materials and Methods, in silico searches uncovered that TIR1, AFB1, and AFB4, (the AFB1 group) were expressed in a downward trend (Fig. 1A) whereas AFB2, AFB3, and AFB5 (the AFB5 group) in an upward trend (Fig. 1B) from FUN to CSC in the dataset generated by Belmonte et al. (2013) and Khan et al. (2015). For all AFBs, the expression values of the distal seed coat (DSC), after normalization (see Materials and Methods for the rational for the normalization), were small compared to those of FUN and CSC (Fig. 1). Moreover, 118 genes, either down- or upregulated by auxin according to Goda et al. (2004), were also identified in the same dataset, and were assigned into four groups based on whether they had a downward or an upward expression trend from FUN to CSC and whether they were down- or upregulated by auxin (Table S1).

Identification of genes that are highly correlated in expression with genes known to be involved in seed germination

To further investigate how the AFBs in the AFB1 group and the AFB5 group are correlated in expression, respectively, linear correlation analysis was conducted as described in Materials and Methods. The three AFBs in the AFB1 group were found to be highly correlated (R² = 0.99, TIR1 vs. AFB1; R² = 1, AFB4 vs. AFB1; Fig. 2A), and so were the AFBs in the AFB5 group (R² = 0.99, AFB3 vs. AFB5; R² = 0.92, AFB2 vs. AFB5; Fig. 2B). However, the R² value between AFB1 and AFB5 was only 0.57. These relationships among the members within the AFB1 group or the AFB5 group, and between AFB1 and AFB5 are consistent with the AFB1 group and the AFB5 group exhibiting opposite trends in expression in the FUN-CSC continuum.

The same linear correlation analysis was also carried out with the four groups of genes in Table S1. Based on the R² values, Group 1 was further divided into two subgroups, Group 1A and 1B, Group 2 into three subgroups, Group 2A–C, Group 3 into three subgroups, Group 3A–C, and Group 4 into four subgroups, Group 4A–D (Table S1). As an example for demonstrating that the genes within a subgroup were highly correlated, the sums of the expression levels of all genes in FUN, CSC, and DSC in Group 1A, excluding those of a member of the group, At2g28470 (arbitrarily chosen), were plotted against those of At2g28470, which exhibited a linear correlation with R² = 0.98 (Fig. 3A). A similar level of linear correlation was found between the sum of Group 1B excluding At4g20460 (arbitrarily chosen) and At4g20460 with R² = 0.99 (Fig. 3B). These two group values, on the other hand, had R² = 0.59 when plotted against each other (Fig. 3C). These results indicated that Group 1A and 1B were separate groups, suggesting that they were not regulated by the same set of transcriptional factors. As conducted with Group 1A and 1B, the separation of Group 4A from Group 4B is shown in Fig. 4, with the intragroup R² values of 0.98 and 0.95, respectively, and the intergroup R² = 0.86 (Fig. 4). Among the 12 subgroups identified, only Group 1A, 1B, and 4A were found to contain genes that are known or highly likely to function in the seed germination process (Tables 1 and 2). In particular, Group 1A and 1B (Table 1) and Group 4A (Table 2) contained five, seven, and nine genes apparently functioning in the seed germination process, respectively. Nine genes in these three subgroups were unknown for a role in seed germination, but their close associations with the other members in the subgroups raised the possibility that they also participate in the seed germination process. Auxin should downregulate the genes in Group 1A and 1B and upregulate the genes in Group 4A. Furthermore, the expression trends of these genes and the AFBs in the FUN-CSC continuum suggest that the AFB1 group downregulates Group 1A and 1B and the AFB5 group upregulates Group 4A. It is noted here that the statistical p values of the above correlations are not provided because each of the variables only had three data points that were too few for producing meaningful p values.

Experimental confirmation of the functions of five newly identified genes in seed germination

To experimentally test whether the genes unknown for their involvement in seed germination in Tables 1 and 2 indeed function in seed germination, we attempted to obtain two T-DNA insertion mutant alleles for each of them. We were able to confirm two homozygous mutant alleles for each of the five loci that included At1g78090, At4g35060 (HMP39), At1g51170, At3g13380 (BRL3), and At2g23060. One or two mRNA transcripts that contained a part of the T-DNA insertion were detected in each of the mutant alleles by RT-PCR, indicating that the insertions likely disrupted the open reading frames of these genes (Fig. 5). Seed germination tests were then conducted for these mutants and Col-0 (wild-type control). The results indicated that both mutant alleles of each locus showed the same trend in germination frequency change; their frequencies were either higher or lower than those of Col-0 (Fig. 6). In particular, mutants of At1g51170 and BRL3 had increased germination frequencies and those of At2g23060, HMP39, and At4g35060 reduced germination frequencies when compared with their respective controls (paired two-tail t-test, p < 0.05, n = 8 or 12). These results indicated that the AT1G51170 and BRL3 proteins normally negatively affect seed germination and the AT2G23060, HMP39, and AT4G35060 proteins positively affect seed germination. The fact that all tested mutants showed an altered seed germination frequency supports the validity of the approach used in this investigation for predicting genes functioning in seed germination. It is noted here that it was necessary to conduct the seed germination test with paired wild-type and mutant samples, and accordingly, paired t-test, because even in the same growth chamber with temperature control, the germination frequencies of Col-0 and the mutants significantly fluctuated between experiments conducted at different times for the same batches of seeds. Such variations, therefore, were likely caused by slight temperature variations in the growth chamber as a result of room temperature variations.

An integrated gene network for controlling seed germination

Because PYL6 and SNRK2-7 in Table 1 were expected to be involved in ABA signaling and BRL3 in Table 2 was expected to be involved in brassinosteroid signaling, the identified genes seemed to be part of a gene network for seed germination that responds to auxin, ABA, and brassinosteroids. Based on the publicly available information on the genes regulated by these hormones (Goda et al., 2004; Okamoto et al., 2010; Tian et al., 2020; Liu et al., 2020), we constructed a gene network that included the six AFBs, the 30 genes in Tables 1 and 2, BES1 and BZR1 involved in brassinosteroid signaling, and ABI3, 4, and 5 involved in ABA signaling (Fig. 7; Fig. S2). The network has interesting features: (1) the genes that are positive for seed germination, including RBOHB, TIP2.2, and HMP39 (this investigation), are negatively regulated by both the AFB1 group and the AFB5 group. (2) The genes that are negative for seed germination, including PRX69 and At4g03140, are positively regulated by both the AFB1 group and the AFB5 group. (3) The AFB1 group and the AFB5 group seem to negatively and positively regulate At2g51170 (inhibiting seed germination, this investigation), At2g34080 (promoting seed germination), and At1g78090 (promoting seed germination, this investigation), respectively. (4) Based on how the AFB1 group and the AFB5 group are expected to affect the expression of ABI3, 4, and 5, the AFB1 group and the AFB5 group seem to negatively and positively regulate ABA signaling, respectively. (5) Outside the aforementioned pathways, the AFB1 group and the AFB5 group seem to negatively and positively regulate genes that are positive for seed germination, respectively. These features suggest that this network can produce either a non-germination or a germination outcome, depending on the environmental and/or physiological conditions.