Arabidopsis thaliana L. seeds (Col-0, Columbia ecotype; stock number CS6673, Arabidopsis Biological Resource Center, Columbus, OH, USA) were surface-sterilized, and approximately 25 seeds were placed into liquid cultures (100 mL of ½× MS medium containing 2.5 mM KH₂PO₄ and 1% (w/v) sucrose in a 250-mL flask). The medium contained supplemental phosphate so that this nutrient was not limiting for nucleotide metabolism during the course of the experiment. After 2 days at 4°C to synchronize seed germination, flasks were placed on a rotary shaker at 100 rpm under fluorescent lights (approximately 100 µE m⁻² s⁻¹) and a 16L/8D photoperiod at 25°C. The experiment design is shown in Figure 1 . PALA (N-(phosphonoacetyl)-L-aspartate, NSC #224131-F/30; Developmental Therapeutics Program, National Cancer Institute, NIH, Bethesda, MD, USA) was added to 1 mM final concentration to some of the cultures at day 2. Untreated and 1 mM PALA-containing cultures were harvested 6 h into the light cycle on days 5, 7, 9, and 12. Uracil (Ura), uridine (Urd), and thymidine (Thd) (Sigma-Aldrich; St. Louis, MO, USA) were added to 1 mM final concentrations in some PALA-containing cultures on day 5. Seedlings were harvested on days 7, 9, and 12. Data are representative of two separate experiments, except for gene expression data, which were derived from the second experiment only.

Seedlings were weighed, homogenized in 2 mL of 95% (v/v) methanol, then centrifuged at 16,000 × g, 10 min. The pellet was reextracted with an equal volume of solvent. Total chlorophyll (a+b) was determined using the method of Arnon (1949) and expressed on a fresh weight basis. Three biological replicates of 8–10 seedlings each were assayed for each treatment.

Day 9 seedlings were removed from liquid culture, briefly rinsed in pH 5.8 potassium phosphate buffer, then blotted dry and weighed. Two seedlings were placed in a capped 75 × 12-mm plastic tube containing 1 mL of the same buffer containing 1 mM unlabeled Ura + 1 µCi [2-¹⁴C]-Ura (7.4 GBq mmol⁻¹; Sigma-Aldrich, St. Louis, MO, USA). Following a 16-h incubation at 25°C, seedlings were removed from the tube, rinsed in buffer + 1 mM unlabeled Ura, and homogenized in 1 mL of 10% (v/v) perchloric acid (PCA). PCA-soluble and -insoluble fractions were separated by centrifugation at 16,000 × g. PCA-insoluble materials were hydrolyzed overnight in 6 M HCl and the supernatant was collected after 16,000 x g centrifugation. Soluble and insoluble fractions were neutralized with KOH, and ¹⁴C-label was quantified by scintillation counting. In parallel, a CO₂ capture assay was performed to measure Ura catabolism. Seedlings were placed in tubes containing the same reactants as previously described and 6-mm paper discs saturated with 2 M KOH were mounted inside the tube on a plugged 22-gauge syringe needle, inserted through the tube cap. After the incubation period, the assay was terminated by addition of 1 mL of 10% PCA. The tube was re-capped and any additional release of ¹⁴CO₂ was captured for 6 h. After drying, ¹⁴C-label on paper discs was counted directly. Four biological replicates of each assay were used.

Seedling DNA contents were determined using a DAPI assay, as described by Baer et al. (1982). Seedling total RNA was isolated using the Plant RNeasy kit (Qiagen, Valencia, CA), according to the manufacturer’s protocol. RNA samples were treated using a TURBO DNA-Free DNase I kit (Thermo Fisher Scientific, Carlsbad, California) to remove genomic DNA contamination. RNA was quantified by A₂₆₀ using a NanoDrop One spectrophotometer.

Primer3 was used to design primers to Arabidopsis coding sequences. Unique primer pairs generating 100–150-bp amplicons with T_m = 60.0 ± 2.0°C were designed for each of the target genes and tested for self-complementarity characteristics. For target genes producing alternatively spliced transcripts, primers were designed to amplify all transcripts associated with the gene. Primers were also designed for the five spiking controls and four invariant control genes (see below). Primer sequences and product sizes for all genes are listed in Supplemental Table 1 . Primers were synthesized by Operon (Eurofins Genomics, Louisville, KY).

Total RNA (1 µg) was reverse-transcribed (SuperScript III First-Strand Synthesis System; Invitrogen, Carlsbad, CA, USA). Spiking controls (Bacillus subtilis transcripts) were added to each sample before cDNA synthesis (per µg total RNA: lysA, 0.00002 fmol; pheB, 0.0002 fmol; thrC, 0.002 fmol; trpF, 0.02 fmol; dapB, 0.2 fmol). PCR primers were arrayed with sample sets in 384-well optical plates in 10-µL reactions containing 2.5 µL of 0.8 mM primers, 2.5 µL of 1:12.5-diluted sample cDNA, and 5 µL of 2× SYBR Green PCR Master Mix. The master mix and plates were purchased from ABI (Applied Biosystems Inc., Foster City, CA, USA). qPCR analyses were conducted using an ABI Prism 7800HT Sequence Detection System and the SDS v. 2.2.1 software package. PCR amplification steps: 50°C, 2 min; 95°C, 10 min [45 cycles: 95°C, 15 s; 60°C, 1 min]. SYBR Green fluorescence was measured at a threshold value of 0.2, and results were expressed as the C_T value. Dissociation curve analyses confirmed amplification of homogeneous targets. qPCR products were sequenced to confirm target specificity. PCR amplification efficiencies for each template (three biological replicates, two technical replicates of each) were determined using the LinRegPCR v 7.5 analysis package (Ramakers et al., 2003). qBase v. 1.3.5 analysis software (Hellemans et al., 2008) was used to calculate normalized relative quantities (NRQ) from raw expression data (C_T values) with gene-specific efficiency correction and sample-to-sample interrun calibration (IRC) using four reference genes (AT4G26410, AT4G27960, AT5G15710, AT5G46630; Czechowski et al., 2005). Stability of the reference genes under experimental conditions was evaluated using the geNorm algorithm (Vandesompele et al., 2002) within qBase. Expression data for experimental samples were calculated as ratios ± S.E, relative to control samples. Expression data for day 12 seedlings was gathered only for genes analyzed by qRT-PCR.

Total RNA sample quality was analyzed using a Agilent Bioanalyzer 2100 platform and prepared for sequencing using a TruSeq cDNA library prep kit (Illumina). Using an Illumina HiSeq 2000 sequencer, an average 53.9 million reads (50 bp, unpaired) per each biological replicate (n = 2) were obtained. Sequencing was performed by Cornell Weill Medical College genomics core (https://research.weill.cornell.edu/core-facilities/genomics). CLC Genomics Workbench v. 10.0.1 software was used to trim and map reads to exons in the Arabidopsis reference genome (TAIR10.1) using Araport11 annotation (Cheng et al., 2017) and 94.8% of reads mapped to exons and introns. Data set normalization, quality control analyses, and gene differential expression (DE) analyses using the R Bioconductor module DESeq2 were performed according to Veerappa et al. (2019). Differentially expressed genes were defined as those having fold-change values of ≥1.5 (padj ≤ 0.05) in response to treatments, relative to untreated controls. RNA-seq reads data have been deposited in the NCBI Sequence Read Archive (BioProject accession #PRJNA704993). Normalized gene expression data were visualized, hierarchically clustered (1-Pearson, average linkage) and further analyzed using Morpheus (https://software.broadinstitute.org/morpheus/). qRT-PCR expression analyses for 35 genes were used to verify RNA-seq expression data for day 5 and day 7 samples ( Supplemental Figure 1 ).

ATTED-II (Obayashi et al., 2018) was used to identify the 300 most highly coexpressed Arabidopsis genes with input lists of DE genes functioning in de novo synthesis, NTP synthesis, salvaging, catabolism, or nucleobase transport. The PlantRegMap transcription factor (TF) enrichment tool (Tian et al., 2020) was used to identify TF with statistically overrepresented targets in coexpressed genes that were also DE in response to PALA treatment at any time point (day 5, 7, or 9). Lists of significantly enriched TF (q ≤ 0.05) were further limited to those TF that were also DE in response to PALA treatment.

iRegNet (Shim et al., 2021) was used to suggest transcriptional regulatory networks beyond potential direct regulations identified in target genes. This resource was queried with TF that were induced or repressed by PALA treatment in day 5 seedlings and which also had enriched targets within the highly co-expressed nucleotide metabolism gene cohorts described above (Bonferroni-corrected p-values ≤ 1e−10; 1,000-bp upstream and 1,000-bp downstream flanking sequences of genes were analyzed). iRegNet was also used to explore potential upstream regulons for individual TF and small groups of co-regulated genes.

Metabolic pathways and enzyme and transporter annotation and genes for Arabidopsis nucleotide metabolism were adopted from Zrenner et al. (2006) and Witte and Herde (2020). Pathways described in the present study are also supported by information retrieved from AraCyc (https://biocyc.org/group?id=:ALL-PATHWAYS&org-id=ARA#%20).

The experiment design for investigating developmental and biochemical phenotypes related to changes in pyrimidine availability is shown in Figure 1 . Treatment with 1 mM PALA beginning at 2 days after germination resulted in developmental arrest of seedlings by day 5 and continued through day 9 and day 12 time points ( Figure 2A ). Root growth was especially stunted. Seedlings exhibited severe chlorosis, with chlorophyll contents only 25%–30% of that in untreated control seedlings at days 5, 7, and 9 ( Figure 2B ). At day 9, seedling DNA and RNA contents were 60% and 30% lower than controls, respectively ( Figure 2C ). This pyrimidine starvation phenotype persisted at day 12, indicating that metabolism of PALA was negligible.

Addition of 1 mM Urd to day 5 cultures containing 1 mM PALA reversed seedling developmental arrest beginning around 9, when increased hypocotyl and primary root growth, expansion of cotyledons, and development of leaves were noted ( Figure 2A ). Chlorophyll contents were not significantly different from control seedlings by day 7 ( Figure 2B ). By day 9, DNA contents were the same as controls and RNA contents were 20% higher than in control seedlings ( Figure 2C ). In cultures grown for an additional 3 days (day 12, 7 days of Urd supplementation), seedling development was nearly indistinguishable from control seedlings at the day 9 stage, including significant lateral root development and two sets of rosette leaves ( Figure 2A ). In contrast, development of seedlings supplemented with 1 mM Ura was not different from PALA-treated seedlings in day 9 or day 12 cultures. However, at day 7, chlorophyll contents of these seedlings were 80% of those in control seedlings and were not different from controls at day 9 ( Figure 2B ), suggesting a limited ability to salvage this nucleobase. Seedlings in thymidine-supplemented cultures (not shown) did not look different than PALA-treated seedlings at either day 9 or day 12 ( Figure 2A ). Chlorophyll contents were also not different from those of PALA-treated seedlings, indicating little salvaging of this nucleoside ( Figure 2B ).

The limited salvaging of Ura in PALA-treated day 9 seedlings was supported by assays of [2-¹⁴C]-uracil metabolism ( Figure 3 ). In untreated control seedlings, approximately 40% of the label taken up by the plant was recovered as ¹⁴CO₂, a measure of catabolic pathway activity. This was similar to the catabolic activity previously reported for WT Arabidopsis seedlings (Zrenner et al., 2009). PALA treatment increased catabolism of Ura twofold, with decreased labeling of soluble metabolite and insoluble nucleic acid fractions ( Figure 3 ). Interestingly, PALA-treated seedlings took up only 75% as much labeled Ura as control seedlings, which may reflect a reduced energy status in their tissues.

RNA-seq identified 11,845 unique genes that were differentially expressed (DE) at one or more stages of seedling development (day 5, day 7, day 9) in response to PALA treatment, with or without Urd supplementation. Table 1 summarizes the number of induced and repressed DE genes in response to different experimental treatments at different time points. Remarkably, most genes that remained DE in response to PALA at day 9 were no longer DE after 4 days of Urd supplementation, relative to untreated control seedlings. This condition paralleled the recovery of seedling development.

Clustering of genes by their overall expression profiles ( Figure 4 ) facilitated identification of 12 gene clusters and their enriched GO BioProcess categories that were co-regulated in response to changes in pyrimidine availability over time. Most genes induced by PALA function primarily in abiotic and biotic stress responses. Genes involved in cellular responses to hypoxia or pathogens and in auxin, and Ca²⁺ homeostasis and signaling were regulated beginning at day 5. Genes regulating root development were not induced until day 9, although inhibition of root growth occurred much earlier. In contrast, genes in the cluster 10 expression module were broadly repressed on day 5. Among the genes that belong to cluster 10, many are involved in regulating cell wall loosening during growth, including several expansins and an endoglucanase, and many others that encode proteins which regulate gene expression and cell division activities required for normal seedling growth and development. At day 7, expression of large numbers of genes involved in plastid development and photosynthesis was uniformly decreased. Thus, overall patterns of gene expression supported developmental and biochemical phenotypes but further revealed major abiotic and biotic stress responses resulting from pyrimidine starvation.

One early response to pyrimidine starvation was the strongly increased expression of genes encoding enzymes that synthesize the indole glucosinolate camalexin, the major phytoalexin in Arabidopsis which plays an important role in defense against pathogens and herbivores ( Supplemental Figure 2 ; Lemarié et al., 2015). Several genes in this pathway are regulated by MYB51/HIGH INDOLIC GLUCOSINOLATE 1 (Frerigmann et al., 2015), which was shown to be strongly induced at day 5 ( Supplemental Figure 2 ). In contrast, expression of myrosinase 1 and 2, which activate glucosinolates, producing toxic isothiocynante and thiocyanate defense compounds (Shroff et al., 2008), was repressed ( Supplemental Figure 2 ). This suggests that this defense response would be ineffective in PALA-treated seedlings, whereas metabolic diversion of L-tryptophan from auxin synthesis (Zhao, 2012) to camalexin synthesis would be expected to impact root development (Overvoorde et al., 2010) and other auxin-regulated processes.

The severe chlorosis associated with pyrimidine limitation in the light-grown seedlings was previously noted. Genes in chlorophyll ( Supplemental Figure 3 ) and carotenoid ( Supplemental Figure 4 ) synthesis pathways were coordinately downregulated by PALA treatment by day 7, although genes involved in plastid organization were repressed earlier, beginning on day 5. Protochlorophyllide reductase A (PORA), which functions in phototransformation of protochlorophyllide to chlorophyllide (Liu et al., 2017) was repressed nearly 40-fold in the tetrapyrrole synthesis pathway. Genes encoding MEP pathway enzymes and prenyltransferase activities synthesizing the isoprenoid geranylgeranyl diphosphate (GGPP) precursor for phytyl diphosphate, the phytol tail donor to chlorophyllide a in the final step of chlorophyll a synthesis (Kim et al., 2013; Kopcsayová and Vranová, 2019), were also repressed. PALA induction of chlorophyll catabolism enzymes chlorophyllase-2 and the magnesium dechelatase STAY-GREEN1 (Hörtensteiner and Kräutler, 2011) on day 5 preceded repression of chlorophyll synthesis genes beginning on day 7 ( Supplemental Figure 3 ) and likely contributed to decreased chlorophyll contents of these seedlings. Expression of GATA1, GATA2, and GLK2, transcription factors which activate genes promoting chlorophyll biosynthesis and photosynthetic membrane organization (Liu et al., 2017; Zubo et al., 2018) was repressed ( Supplemental Figure 3 ). Additionally, numerous genes encoding chlorophyll a/b binding proteins of photosystems I and II were repressed (data not shown). For example, LHCB2.4 transcript levels continued to decrease in response to PALA treatment to only 5% of control levels by day 12, but Urd supplementation on day 5 restored levels of this transcript to 60% of those in control seedlings by day 12 ( Supplemental Figure 1 ).

Carotenoids are also essential photosynthetic pigments which are synthesized from GGPP and repression of isoprenoid pathway genes would limit their production, in addition to the synthesis of isoprenoid-derived hormones abscisic acid and gibberellins (Kopcsayová and Vranová, 2019), which regulate a wide range of physiological and developmental processes. Phytoene synthase, which regulates carotenoid synthesis (Stanley and Yuan, 2019), and other pathway genes were further repressed in response to pyrimidine limitation ( Supplemental Figure 4 ).

Supplemental Table 2 lists 152 genes encoding enzymes or transporters involved in nucleotide metabolism and the co-regulated Arg synthesis pathway. qRT-PCR expression analyses ( Supplemental Figure 1 ) verified the expression of 35 individual genes at day 5 and day 7 time points but also included day 12 expression data, which are not represented in transcriptome datasets. Day 12 expression data identified genes that responded more slowly to changes in pyrimidine availability. In most cases, RNA-seq and qRT-PCR expression data were highly similar.

Among the 2,163 genes that were downregulated by PALA on day 5 but not at later stages, genes in the GO BioProcess category “de novo” pyrimidine nucleobase biosynthetic process (GO: 0006207) were the most enriched, although de novo purine synthesis genes were also repressed ( Supplemental Table 2 ). Approximately half of the genes involved in nucleotide metabolism and Arg synthesis were DE in response to PALA treatment and expression profiles for most of these genes were reversed by Urd supplementation. Genes that function in salvaging of extracellular nucleotides were strongly induced by pyrimidine starvation, but, unexpectedly, pyrimidine and purine catabolic pathway genes were also induced under limited pyrimidine availability.

Expression data for DE genes were mapped to the de novo pyrimidine and purine synthesis pathways ( Figure 5 ) and pathways for pyrimidine ( Figure 6 ) and purine ( Figure 7 ) salvaging (intracellular and extracellular) and catabolism to better understand possible metabolic responses to changes in pyrimidine availability, at the level of transcriptional regulation. Abbreviations for enzymes and transporters are provided in Supplemental Table 2 and abbreviations for pathway intermediates are listed in Supplemental Table 3 .

Gene expression profiling for transcription factors (TF) which were DE in response to PALA treatment is shown in Figure 8 . As in Figure 4 , clustering revealed early (day 5) and later (day 7, 9) responses and associated enriched GO BioProcess categories. Most induced TF genes in clusters 1–4 are involved in regulating abiotic and biotic stress responses, consistent with the functional characterization of early regulated genes ( Figure 4 ) and included numerous WRKY and ethylene-responsive TF. Early, repressed TF mainly regulate developmental processes.

We identified TF with enriched targets (Tian et al., 2020) within individual cohorts of highly coexpressed genes representing the major metabolic pathways of nucleotide metabolism and Arg synthesis. Enriched TF and their target genes are arrayed in Supplemental Table 4 , and annotation and expression data for these TF are given in Supplemental Table 5 . Many of the co-regulated genes function in different pathways and have common TF targets that may serve to coordinate nucleotide metabolism at the level of transcription. Induction of a large number of ERF TF is correlated with repression of genes involved in nucleotide synthesis and Arg synthesis, and induction of nucleobase transporter genes. Targets for induced WRKY TF are strongly overrepresented in CPSA, whose repression would decrease synthesis of carbamoyl-P, further limiting pyrimidine and Arg synthesis.

To better understand transcriptional control of nucleotide metabolism, we modeled transcriptional regulatory networks for TF whose induction or repression by PALA on day 5 was reversed by Urd. These TF represented the earliest responses to changes in pyrimidine availability. iRegNet (Shim et al., 2021) was used to generate regulatory network graphs for day 5 responses ( Figure 9 ). This resource identifies significantly overrepresented upstream regulators of query genes, integrating gene-to-gene co-expression data to generate realistic regulatory networks. As is shown in Figure 9A , nearly all TF induced by PALA on day 5 are potentially regulated by one or more upstream transcription factors ERF11, ERF105, ZAT10, WRKY33, and WRKY40. Transcription factors repressed by PALA treatment are downstream of ERF039, SPL15, DOF2.4, DOF5.8, and CDF3 ( Figure 9B ).

Some of the predicted upstream regulators of DE TF are themselves regulated by PALA ± Urd treatments. Several have binding sites that are enriched within co-regulated genes of nucleotide metabolism and, thus, have potential for direct regulations of those genes. Many of the potentially regulated downstream TF in the network graphs also have enriched targets in nucleotide metabolism genes. This information can be used to construct hypothetical regulons for individual genes or groups of genes that may help coordinate metabolic responses to altered pyrimidine levels in seedlings. For example, ASIL2 has binding sites in genes that regulate de novo nucleotide synthesis, intracellular and extracellular salvaging, and Arg synthesis activities ( Supplemental Table 2 ) and is repressed by PALA at day 5. A hypothetical transcriptional regulatory network for ASIL2 is shown in Figure 10A . In this model, induction of ethylene-responsive TF, including ABR1, with targets in the same nucleotide metabolism genes, could further coordinate their expression. Additional direct and indirect regulations involving ABR1 are shown in Figure 10B . These network graphs also validate TF enrichment analyses ( Supplemental Table 4 ) as a general approach to identifying nucleotide metabolism genes that are likely to be co-regulated by individual TF.

Attempts to model similar regulatory networks for DE catabolic pathway genes and TF with enriched targets in those genes were unsuccessful. Using all DE catabolic genes ( Supplemental Table S2 ) as queries in iRegNet, the hypothetical regulon shown in Figure 11 was produced. In this model, highly induced WRKY70 and NAC047, which function in abiotic and biotic stress responses, would activate genes of pyrimidine and purine catabolism.