Seed of the th3 stock (CS81) and reference cla1-1S DXS1 stock (CS16003) were obtained from the Arabidopsis Biological Resource Center.¹ Seeds were sown on agar media containing 0, 30 nM or 10 μM thiamin as described in Mimura et al. (2016). Homozygous mutant th3 plants were maintained by soil applications of 1 mM thiamin at intervals through flowering.

Extraction and quantitative analysis of thiamin vitamers was performed as described in Mimura et al. (2016) and Hasnain et al. (2016). Briefly, whole plants (300 mg) were ground in liquid N₂ then extracted in 500 ml of 7.2% perchloric acid by 30 min of bath sonication at 0°C. Extracts were incubated on ice for 15 min with periodic vortex mixing then centrifuged to remove cell debris (14,000 g, 10 min, 22°C). Vitamers in the supernatant fluid were converted to thiochrome derivatives and resolved by HPLC with fluorometric detection (Fraccascia et al., 2007; Hasnain et al., 2016).

A multiplex RNA-seq library was constructed from three biological replicate RNA samples each from Landsberg wild-type and th3 mutant plants grown on media containing 30 nM thiamin as described in Mimura et al. (2016). The library was sequenced in a single direction on a Hiseq500 instrument (Illumina). RNA-seq data were analyzed using TOPHAT (Trapnell et al., 2012) and gene expression quantified using CUFFDIFF (Trapnell et al., 2013). Differentially regulated genes were annotated using the GO-Term Enrichment tool at Arabidopsis.org. Analysis details are summarized in Supplementary Tables S1–S5.

The RNA-seq reads were analyzed for th3 linked single-nucleotide polymorphisms (SNPs) using a kmer-counting approach. In brief, a database of 22-mer oligonucleotide SNP variants was constructed from the set of predicted full-length cDNA sequences for all genes on chromosome 4 using JELLYFISH (Marçais and Kingsford, 2011). In the kmer database, each 22-mer in the wild-type cDNA was represented by four variants including the possible single base substitutions in the first position. Jellyfish was then used to count the frequency of each 22-mer in the RNA-seq data for each replicate. SNP variants were detected by a shift of counts from wild-type kmer to one of the other three variants at that position. SNPs relative to Ler-0 that were present in the mutant replicates but not detected in wild-type RNA-seq reads were selected for further analysis. This method allowed rapid detection of SNPs and quantitative comparison of kmer counts from isogenic wild type and mutant replicates, provided that mutations were sparsely distributed in the coding sequences (i.e., a SNP density well below 1 per 22 bases) – a reasonable expectation for the ems mutagenesis used to generate th3.

Structural models the wild-type and th3 mutant DXS1 proteins were constructed by MODELER (Webb and Sali, 2016) using the Deinococcus radiodurans structure as a template (PDB:6OUW; Chen et al., 2019). Fifty replicate models were constructed for each protein, and the objective function score was used to select the best structure (Webb and Sali, 2016). Molecular graphics images were created using the UCSF Chimera² (Pettersen et al., 2004).

Consistent with Koornneef (1981) characterization of the th3 mutant, when grown in the absence of supplemental thiamin th3 mutant plants formed pale-green leaves and grew slowly (Figure 1). We confirmed that th3 plants could be rescued on media containing thiamin concentrations as low as 30 nM resulting in fertile plants with normal leaves.

Analysis of thiamin vitamers in wild type and th3 plants (Figure 1) showed no discernible differences in thiamin, ThMP and ThDP contents of mutant and wild-type plants grown either with or without supplemental thiamin in the medium. Growth on medium containing 30 nM thiamin resulted in similar increases in ThDP content in mutant and wild type (2.3-fold and 2.6-fold, respectively) compared to plants grown with no exogenous thiamin. In plants grown on 30 nM thiamin, free thiamin increased relative to plants grown on thiamin-free medium, accounting for 14 and 11% of total thiamin in supplemented wild type and mutant plants, respectively. In plants grown on medium containing a high thiamin concentration (10 μM), free thiamin content increased dramatically in both mutant and wild type. While ThDP content was also elevated at least 6-fold in both mutant and wild type, the elevation in ThDP was not proportional to the large increase in free thiamin content. This result is consistent with evidence from THIC and THI4 overexpression studies indicating that thiamin diphosphokinase is a limiting step in conversion of thiamin to ThDP (Dong et al., 2015). Thus, while the possibility of a lesion affecting transport and subcellular localization of specific vitamers was not ruled out by these data, no obvious defects in thiamin biosynthesis and metabolism were detected in the th3 mutant.

In order to identify the mutation responsible for the th3 phenotype, RNA-seq analysis of triplicate wild type and mutant plant samples was undertaken. Analysis of SNPs in RNA-seq data (Figure 2A) revealed a non-synonymous mutation in the protein coding sequence of AT4G15560, a gene encoding DXS1, also known as Cloroplastos Alterados-1 (CLA1). DXS1 is a highly conserved ThDP requiring enzyme that catalyzes the first committed step in the MEP pathway for isoprene biosynthesis in plants and bacteria (Zhao et al., 2013). The location of DXS1 on chromosome 4 is consistent with the classical genetic map position of th3 locus (Meinke et al., 2009), making the DXS1 SNP a candidate for the th3 mutation. Consistent with this hypothesis, the mutation in th3 causes a non-conservative substitution of a lysine for a highly conserved glutamate at position 323 in the DXS1 protein sequence (Figure 2B). To test the allelism hypothesis, homozygous th3 plants were crossed to plants heterozygous for the cla1-1S reference allele. As shown in Figure 3, the F1 seed segregated 1:1 for the characteristic pale-green th3 phenotype confirming that the th3 mutant is allelic to the cla1-1S reference mutant.

As shown in Figure 2B, the E323K substitution in the th3 mutant occurs in a conserved motif (FEELG) that in bacterial DXS enzymes of known structure forms a short α-helix located at the interface between subunits of the DXS homodimer (Chen et al., 2019). To explore implications of the E323K mutation, molecular models of the wild type and th3 mutant Arabidopsis DXS1 protein were constructed based on homology to the D. radiodurans enzyme (PDB:6OUW, Chen et al., 2019) using MODELER (Webb and Sali, 2016). As shown in Figure 4, the face-to-face homodimer configuration places the E323 residues of each wild-type subunit in close proximity at the subunit interface. While the E to K substitution has no predicted effect on the helix structure, the mutation substantially alters the distribution of surface charge on the protein by introducing positive charge at the normally negatively charged subunit interface (Figure 4, right panel). Interestingly, the FEELG α-helix motif is immediately adjacent to a segment (amino acids 198–243) that is not resolved in crystal structures of E. coli and D. radiodurans DXS proteins due to having a disordered structure. The functional significance of this apparently conserved structural feature is unknown.

The centrally important MEP pathway for isoprene biosynthesis in chloroplasts is highly regulated in plants (Figure 5A). Known mechanisms include diurnal regulation (Pokhilko et al., 2015), retrograde signaling mediated by the pathway intermediate MEcPP (Xiao et al., 2012; Lemos et al., 2016; Bjornson et al., 2017), and feedback control of DXS1 activity via competitive inhibition of ThDP binding by DMAPP (Banerjee et al., 2013). Although th3 plants rescued on media containing 30 nM thiamin have a normal morphological phenotype, analysis of the replicated RNA-seq data revealed gene expression changes that overlap substantially with patterns conditioned by mutant alleles of HDS that accumulate MEcPP (Xiao et al., 2012; Lemos et al., 2016; Bjornson et al., 2017).

Of 344 genes upregulated in th3 relative to wild type, 153 genes (Figure 5B, Supplementary Table S1) are also upregulated by the ceh mutation of HDS (Bjornson et al., 2017) using a threshold of 1.5-fold change for both datasets. The shared suite of upregulated genes is strongly enriched for GO terms associated with biotic stress, pathogen defense responses, and salicylic acid signaling (Supplementary Table S2). Key exemplars of this response are shown in Figure 5C. Interestingly, HDS is the sole MEP pathway gene differentially regulated in th3 plants (Figure 5C). The overlap with hds (ceh) affected gene expression is unexpected because effects of HDS mutations are attributed to retrograde signaling by MEP intermediate MEcPP that accumulates in hds loss of function mutants (Xiao et al., 2012; González-Cabanelas et al., 2015; Benn et al., 2016) and DXS1 over-expressors (Wright et al., 2014). By contrast, attenuation of DXS1 activity causes a reduction in the MEcPP pool size (Wright et al., 2014).

A set of 552 genes downregulated in th3 (>1.5-fold change, Supplementary Table S3) were enriched for functions associated with abiotic stress responses including water and oxidative stresses (Supplementary Table S2). Of these, 73 genes were also downregulated in ceh (Supplementary Table S3). This shared set of genes showed moderate enrichment for cell wall biosynthetic functions (GO:0071554, p = 4.07E-03).

The hypothesis that the th3 mutation reduces affinity of DXS1 for ThDP is consistent with thiamin responsive mutants studied in other organisms. Notably, there is ample precedence in humans for mutations in thiamin-requiring enzymes having phenotypes that can be rescued by supplemental thiamin (reviewed in Ames et al., 2002). Thiamin responsive mutations that increase the K_m of human branched chain α-ketoacid dehydrogenase for ThDP are not restricted to the ThDP-binding E1 subunit, highlighting the importance of subunit interactions in determining cofactor affinity (Fisher et al., 1991). Interestingly, structural analysis indicates that the th3 mutation replaces a highly conserved glutamate (E323) located in a short α-helix at the interface between subunits of the DXS1 homodimer with a positively charged lysine. While it is unknown whether the helix formed by the conserved FEELG motif has a specific role in ThDP binding and assembly of holoenzyme, it is plausible that the change in surface charge alone would affect interaction of the enzyme with the negatively charged ThDP molecule. Whatever the case, our results for DXS1-th3 illustrate that the physiological effects of single-residue changes far from an enzyme’s active site can be as consequential as more easily rationalized changes in the active-site region (Dalby, 2011).

Because the total ThDP content is normally comparable to the abundance of thiamin-dependent enzymes in leaf cells (Hanson et al., 2016; Joshi et al., 2019), the concentration of free ThDP is most likely below the K_m of DXS1 for ThDP (Banerjee et al., 2013). This suggests that DXS1 is not fully activated under normal conditions rendering the enzyme sensitive to small changes in free ThDP concentration. Indeed, limiting ThDP likely enhances sensitivity of DXS1 to feedback regulation by competitive inhibition of ThDP binding by MEP pathway product, DMAPP (Banerjee et al., 2013; Sharkey et al., 2015).

Activity of the mutant DXS1-th3 enzyme is evidently restored sufficiently by the ~2.5-fold elevation of in vivo ThDP content conditioned by supplementation with 30 nM thiamin. However, the observed fold elevation in total ThDP content most likely understates the impact of the 30 nM thiamin supplement on the concentration of free ThDP in leaf cells. If, for example, 90% of the ThDP pool in un-supplemented control plants is bound to enzymes (leaving 10% free), then most of the excess ThDP contained in supplemented plants will be in the unbound fraction. In that instance, a 2.5-fold increase in total ThDP would correspond to a greater than 15-fold increase in free ThDP [(2.5 − 0.9)/0.1 = 16]. Hence, even without taking compartmentalization into account, order-of-magnitude scale changes in free ThDP concentration in plants supplemented with 30 nM thiamin are plausible.

The induction of a biotic stress response associated with MEcPP signaling suggests that the th3 mutation is a novel genetic perturbation of the MEP pathway. On the one hand, this unexpected biotic stress response in young th3 plants may reflect complex developmental dependencies of MEcPP signaling. Notably, in young hds-3 mutant seedlings expression of WRKY70 and PR1 is suppressed relative to wild type in apparent contrast to their induction by MEcPP in older plants (González-Cabanelas et al., 2015). If MEP pathway signaling actively inhibits the stress response in young plants, then attenuation of MEcPP synthesis could result in de-repression of the biotic stress response during an early phase of vegetative development. Alternatively, there are plausible circumstances in which mis-regulation of DXS1 activity in thiamin-supplemented th3 plants could result in activities that exceed that of wild-type enzyme creating excess flux through the MEP pathway. A key unknown is how th3 impacts relative sensitivities of DXS1 to changes in in vivo concentrations of ThDP and DMAPP. Because DMAPP inhibition of ThDP binding to DXS1 is competitive (Banerjee et al., 2013), implying that cofactor and inhibitor bind to a common site, a DXS1 mutation with a higher K_m for ThDP will likely also have altered affinity for the DMAPP inhibitor, potentially rendering the enzyme less sensitive to feedback inhibition. In that case, elevated ThDP concentrations that compensate for the reduced cofactor-binding affinity could in principle result in DXS activities that exceed wild-type levels due to insensitivity of the mutant enzyme to feedback inhibition.

These findings add to an emerging picture of a highly interconnected thiamin economy in plant cells comprised of enzymes of central metabolism localized in multiple compartments governed by complex dynamics of ThDP synthesis, transport, holoenzyme assembly, and protein turnover. Remarkably, a modest overexpression of transketolase apo-protein in tobacco conditions a pale-green leaf and plant growth phenotype that can be rescued by supplemental thiamin (Khozaei et al., 2015). The effects of excess transketolase apo-protein can be partly attributed to depletion of the pool of ThDP available to other thiamin-requiring enzymes including DXS1. Consistent with that hypothesis, Khozaei et al. showed that the leaf phenotype of transketolase over-expressors is partially rescued on media containing DXS product, deoxyxylulose 5-P. Less intuitively, transketolase overexpression in tobacco also causes reduction in overall thiamin content. Our results showing that thiamin vitamer levels in DXS1-deficient th3 mutant plants – grown with or without supplemental thiamin (Figure 1) – are normal indicate that reduced flux through the MEP pathway does not detectably alter thiamin metabolism in Arabidopsis. Nevertheless, the complexity of the transketolase overexpression phenotype highlights the apparent sensitivity of the thiamin economy to small perturbations in balance of thiamin-requiring enzymes.

Protein turnover in plastids plays a central role in the thiamin economy. The plastid localized THI4 and THIC proteins that catalyze limiting reactions in the thiazole and pyrimidine branches of the thiamin biosynthetic pathway, respectively, are among the most rapidly degraded proteins in the leaf proteome (Nelson et al., 2014; Li et al., 2017). While DXS1 turnover was not captured in the proteome-wide analyses by Nelson et al. or Li et al., results of Pulido et al. (2016) are consistent with a half-life of less than 5 h in leaves, indicating a turnover rate comparable to those of THI4 and THIC. Thus, three of the most rapidly degraded leaf proteins function in the thiamin economy. The rapid turnover rates of THI4 and THIC thiamin biosynthetic enzymes are likely related to their susceptibility to irreversible inactivation during catalysis. In the most extreme case, THI4 participates in a single-turnover reaction that irreversibly incorporates a sulfur from an active-site cysteine into the thiazole ring of the thiamin precursor (Chatterjee et al., 2011; Joshi et al., 2020). Consequently, one THI4 protein molecule is consumed for every molecule of thiazole synthesized. While THIC is capable of multiple catalytic turnovers, the enzyme is thought to be subject to frequent inactivation during catalysis due to side-reactions involving its radical SAM cofactor (Palmer and Downs, 2013; Hanson et al., 2016). Some thiamin-requiring enzymes may also be susceptible to damage from side-reactions involving reactive intermediates. For example, in addition to its primary activity of catalyzing synthesis of DXS from pyruvate and glyceraldehyde-3-P, bacterial DXS has an oxygenase side-activity that results in decarboxylation of pyruvate forming a reactive peracetate intermediate in the active site (Johnston and Meyers, 2021). Conceivably, if the oxygenase side-reaction occurs in chloroplasts where O₂ concentrations are high, this reaction could be a significant cause of damage to the enzyme, shortening its functional lifetime. In this context, th3 provides a tool for probing the relationship between rates of DXS1 turnover (Pulido et al., 2016) and holoenzyme assembly, a poorly understood process implicated in feedback regulation of the MEP pathway (Banerjee et al., 2013). Thus, th3 is a revealing perturbation of the crucial interface between the thiamin economy and the MEP pathway.