Differential regulation of host plant adaptive genes in Pieris butterflies exposed to a range of glucosinolate profiles in their host plants

Specialist herbivores have often evolved highly sophisticated mechanisms to counteract defenses mediated by major plant secondary-metabolites. Plant species of the herbivore host range often display high chemical diversity and it is not well understood how specialist herbivores respond to this chemical diversity. Pieris larvae overcome toxic products from glucosinolate hydrolysis, the major chemical defense of their Brassicaceae hosts, by expressing nitrile-specifier proteins (NSP) in their gut. Furthermore, Pieris butterflies possess so-called major allergen (MA) proteins, which are multi-domain variants of a single domain major allergen (SDMA) protein expressed in the guts of Lepidopteran larvae. Here we show that Pieris larvae fine-tune NSP and MA gene expression depending on the glucosinolate profiles of their Brassicaceae hosts. Although the role of MA is not yet fully understood, the expression levels of NSP and MA in larvae that fed on plants whose glucosinolate composition varied was dramatically changed, whereas levels of SDMA expression remained unchanged. In addition, we found a similar regulation pattern among these genes in larvae feeding on Arabidopsis mutants with different glucosinolate profiles. Our results demonstrate that Pieris larvae appear to use different host plant adaptive genes to overcome a wide range of glucosinolate profiles in their host plants.

Although the GLS myrosinase system in Brassicales is known as an efficient defense against many herbivores 15,16 , some insects can use Brassicales plants as hosts 1,4 . Pierid butterflies use Brassicales plants by disarming GLS defense with a protein expressed in the larval gut called nitrile-specifier protein (NSP) 17 . NSP can redirect the GLS-myrosinase reaction to form less toxic nitriles rather than toxic ITCs. Since Pierid butterfly larvae are generally oligophagous and use a distinct range of Brassicales host plants 18 , each species will likely be exposed to a range of GLSs in their host plants. Pieris butterflies encounter a large variety of GLS profiles in their host plants because they rely on various genera in the Brassicaceae family, which seems to have the highest GLS diversity in Brassicales 14,19 .
NSP belongs to the small so-called NSP-like gene family, which also includes genes that encode major allergens (MA) and single domain major allergens (SDMA) 20 . NSP and MA have a similar genetic structure, as they both possess three replicated domains that originated from SDMA 20 . SDMA is generally found across Lepidoptera. Although its specific function is still unknown, SDMA is known to be expressed only in larval guts, suggesting this protein plays a general digestion-related role in Lepidoptera 21 . NSP and MA are specific for Pierid butterfly larvae feeding on Brassicales 20 , suggesting that both proteins may have GLS-disarming functions. The role of MA proteins in modifying the outcome of the GLS-myrosinase defense system, by which MA causes nitriles rather than ITCs to form, was confirmed in a Brassicaceae-feeding Pierid species, Anthocharis cardamines, which also seems to lack an NSP gene 19 . However, in general, the role of MA proteins is still unclear in most Pieris species, especially those which encode both NSP and MA 19 . Given that in at least one Pieris species, not only NSP but also MA have been shown to have GLS-disarming functions, the presence of both genes in some Pieridae could be a molecular adaptive mechanism for overcoming diverse GLSs in the species' host plants.
Here, we focus on members of the NSP-like gene family and their regulation patterns in Brassicaceae-feeding Pieris larvae exposed to a range of GLS profiles, in order to find out how Pieris butterflies respond and adapt to diverse GLSs in their host plants. We conducted feeding experiments using four Japanese Pieris species (Pieris napi, P. melete, P. rapae and P. brassicae) and two Brassicaceae plants (Arabidopsis kamchatica and Cardamine occulta, both of which have diverse GLS profiles) 9,13,22,23 . The four closely related Pieris species are known to have different host ranges. P. napi and P. melete mainly use wild Brassicaceae plants such as Arabis or Cardamine in Japan 24,25 , while P. rapae and P. brassicae are known pests of Brassica crops (Table 1) [26][27][28] . NSP sequences have been identified only in P. rapae and P. brassicae in this genus 19,20,29 . We first identified NSP, MA and SDMA sequences in all of the tested Pieris species, combining high throughput RNA sequencing (RNA-seq) of larval samples from our plant-feeding experiments and de novo transcriptome assemblies. We subsequently measured and compared the expression levels of all of the identified NSP, MA and SDMA genes in larvae that fed on two host plants using the RNA-seq data. To confirm the expression levels of the NSP-like gene family observed in our RNA-seq data with a larger number of replicates, we also conducted real-time quantitative PCR (RT-qPCR) on P. melete, considering it to be a representative species and targeting all members of the NSP-like gene family (Fig. 1).
Since GLS profiles are specific for each Brassicaceae plant species 9,13 , we expected to find that Pieris larvae feeding on the two Brassicaceae plants regulate NSP-like genes in distinct ways. However, since some plant species also differ in their general chemical profiles -excluding GLSs-in several ways, this chemical difference could also affect the regulation of genes in larvae. To obtain feeding and gene expression data using a more controlled background, we also conducted feeding experiments with Arabidopsis thaliana mutants, which differ in their GLS profiles but share the same chemical background. We prepared wild-type lines (Col-0) and three mutant lines of A. thaliana which have different GLS profiles-namely, MAM1 (which lacks chain elongation genes for the C3/C4 chain length variation of Met-derived GLS), MAM3 (which lacks Met chain elongation gene and has no long-chain aliphatic GLS), and quad-GLS (the quadruple mutant myb28myb29cyp79B2cyp79B3 which has no detectable GLS) 30-32 -and used P. napi as a representative species for this feeding assay. Combining these two approaches, we were able to observe how Pieris butterflies regulate NSP-like gene family members in response to a broad range of GLS defenses in their host plants (Fig. 1).   (Fig. 2). The newly acquired NSP sequences of P. napi and P. melete have 86% and 84% amino acid sequence identity, respectively, with NSP from P. rapae. MA proteins also showed high identity (89% each), and SDMA proteins showed slightly higher sequence identity (92% each) to P. rapae.

RNA-seq identified NSP-like gene family sequences of the four
NSP and MA are differentially expressed in response to wild Brassicaceae plants. GLS profiles of Arabidopsis kamchatica and Cardamine occulta, host plants of Pieris butterflies in Japan, were significantly different according to the UPLC-TQMS analysis. In total, we detected 25 types of GLSs in the two tested plant species and sorted these into the three major GLS chemical classes to observe their profile differences (Table S1).
A. kamchatica had high amounts of aliphatic and indolic GLSs, whereas C. occulta contained benzylic GLSs, which were much lower in A. kamchatica (Fig. 3a, Table S1). Although none of the four larvae species feeding on the two plant species showed any sign of stopping feeding or of poisoning, we observed that P. rapae and P. brassicae appeared to grow better on C. occulta than on A. kamchatica. However, this difference was not statistically significant (P = 0.12 and 0.064, Fig. 3b). We acquired expression levels of the NSP-like gene family from all four species of Pieris larvae feeding on the two Brassicaceae plants by analyzing digital expression based on a single RNA-seq data for each treatment (Fig. 3c). Compared to the observed expression levels of NSP in larvae fed with A. kamchatica, the levels increased when larvae fed on C. occulta, while an inverse gene expression pattern was observed for MA. MA was highly expressed in larvae that fed on A. kamchatica, and less expressed in larvae that fed on C. occulta. This inverse trend was observed in all four Pieris species, whereas expression levels of SDMA (which is not involved in disarming GLS) were similar between the larvae fed with the two Brassicaceae plants in all four Pieris species (Fig. 3c). The result of our P. melete larval RT-qPCR experiments also mirrored the same gene regulation pattern of NSP-like gene family members, supporting our RNA-seq results. Expression levels of NSP and MA differed significantly depending on the plant species on which the larvae fed (NSP and MA: P = 0.05, Mann-Whitney U test), whereas levels of SDMA did not (SDMA: P = 0.90, Mann-Whitney U test, Fig. 3d).
NSP and MA are differentially regulated in P. napi feeding on A. thaliana mutants with different GLS profiles. We identified 9 types of GLSs from the four tested A. thaliana lines (Col-0, MAM1, MAM3 and quad-GLS) by LC-UV, confirming previously described GLS profiles for all the mutant lines ( Fig. 4a) 30,31,33,34 . Col-0 had higher levels of short-chain aliphatic GLSs, and we confirmed that the quad-GLS mutant had no detectable GLSs. In the MAM1 mutant, we detected a higher amount of 3-(Methylsulfinyl)propyl GLS (3MSOP) and 8-(Methylsulfinyl)octyl (8MSOO) but less 4-(Methylsulfinyl)butyl GLS (4MSOB) compared to in the wild type (Col-0) 30 . MAM3 lacked long-chain aliphatic GLSs (8MSOO or 7-(Methylsulfinyl)heptyl (7MSOH) GLS), as described previously 31 . The feeding assay performed with P. napi and the four A. thaliana mutant lines showed that P. napi larvae that fed on MAM1 grew more slowly than did those feeding on MAM3 (P = 0.018, FDR-adjusted pairwise t test) (Fig. 4b). We focused on four Pieris butterflies, and conducted feeding experiments with two Brassicaceae plants in order to see gene expression patterns of NSP-like gene family members to different host plants with distinct GLS profiles by RNA-seq. We confirmed the result of RNAseq based expression analysis by RT-qPCR using P. melete as a representative with more replicates. We also conducted feeding experiments with Arabidopsis thaliana mutants which differ in their GLS profiles to get gene expression data in more controlled chemical background. For this mutant experiment, we used Pieris napi as a representative.
www.nature.com/scientificreports www.nature.com/scientificreports/ Regarding the expression patterns of NSP-like gene family members in P. napi, gene regulation differed significantly for NSP and MA but not SDMA in larvae as a response to the mutant lines they fed on (Fig. 4c). In the quad-GLS mutant, which does not contain GLS, NSP and MA were down-regulated in larvae; these were not down-regulated in larvae that fed on wild-type (Col-0) lines. In MAM1, which had higher levels of 3MSOP but lower levels of 4MSOB, we observed NSP in larvae to be significantly down-regulated compared to larvae that fed on Col-0 (wild type); in contrast, MA did not show this trend and had higher expression levels than in the larvae that fed on wild-type lines. When larvae fed on MAM3 lacking long-chain aliphatic GLS, NSP was expressed at levels similar to those found in larvae that fed on Col-0 but MA expression was highest and comparable to MAM1 mutant plant-feeding larvae.

Discussion
In this study, we conducted feeding experiments combined with gene expression analysis using four Pieris butterflies raised on two Brassicaceae plants and four A. thaliana mutant lines. We aimed to reveal if and how NSP-like gene family members of Pieris butterflies respond to a broad range of GLS profiles. We found that the gene expression levels of NSP and MA in Pieris larvae responded to the presence of GLSs and to GLS profile differences in the plants they fed on, but levels of SDMA remained the same. In addition, gene regulation differed between NSP and MA in response to GLS profiles in host plants; surprisingly, the two members of the NSP-like gene family actually showed inverse expression patterns to the plant species. The results suggest that both NSP and MA are involved in disarming plants' defense systems by targeting different GLSs, and both of these proteins are the result of an adaptive mechanism Pieris employs to overcome a wide range of GLS profiles in their host plants.
RNA-seq based gene expression analysis showed that NSP was more highly expressed when larvae fed on C. occulta, which had more benzylic GLSs compared to A. kamchatica; in contrast, MA was highly expressed in larvae that fed on A. kamchatica, which had more aliphatic and indolic GLSs (Fig. 3a,c). Unlike NSP and MA, SDMA was expressed at a similar level, regardless of which plant the larvae fed on (Fig. 3c). These expression patterns of NSP-like gene family members were observed in all four Pieris species (Fig. 3c), and this pattern was also P. melete NSP  www.nature.com/scientificreports www.nature.com/scientificreports/ confirmed with more replicates by RT-qPCR analysis in P. melete (Fig. 3d). Both P. napi and P. melete are known to be able to use both of these Brassicaceae species as hosts in the field 24,25 ; therefore, these expression patterns cannot be dismissed as a response to host or non-host plant differences. Furthermore, the larval growth rate of the four species showed no correlation with the expression patterns of NSP and MA, suggesting that the observed pattern was not the result of larval performance differences in the two plants (Fig. 3b). The involvement of NSP in disarming GLSs has been shown earlier; however, our results suggest that both NSP and MA are involved in functions related to disarming GLS, since these two genes responded to two host plants with dissimilar GLS profiles. Unlike Pierid-specific NSP or MA, SDMA, which can be found across Lepidoptera, did not seem to respond to GLS profile differences, suggesting SDMA is not involved in GLS-specific detoxification or in processes redirecting the breakdown products (Fig. 3c,d). Interestingly, not only did NSP and MA respond to different host plants, but they also showed inverse expression patterns in larvae feeding on the two plants (Fig. 3c,d). NSP seems to respond to benzylic GLSs, whereas MA appears to respond to aliphatic and indolic GLS. However, since we used different Brassicaceae species in this feeding experiment, it can be assumed that non-GLS based chemical differences between the two plant species exist. Thus, these differences might also affect the regulation of larval NSP and MA gene expression.

P. brassicae NSP
We used A. thaliana GLS mutants (wildtype Col-0; MAM1 and MAM3 with different aliphatic GLS compositions compared to Col-0, and quad-GLS with no detectable GLS) in a feeding experiment with P. napi larvae. We found that NSP and MA but not SDMA was down-regulated in larvae that fed on quad-GLS mutants; as these mutants lack GLSs, the expression of NSP and MA must be triggered by the presence of GLSs, supporting our hypothesis that in Pieris larvae, both genes have GLS-disarming roles. Our results are supported by a recent study that also reported a down-regulation of NSP in P. brassicae when larvae fed on an A. thaliana mutant lacking GLSs 6 . These results suggest that there could be costs associated with high level expression of NSP and MA proteins in the larval gut. Therefore, the observed context-specific down-regulation of either NSP or MA, respectively, could be observed in larvae that feed on plants with overall lower GLS amounts, such as certain  www.nature.com/scientificreports www.nature.com/scientificreports/ Brassicaceae crop plants. Furthermore, mRNA levels of NSP and MA also responded differently to GLS profiles. We observed that NSP was down-regulated in larvae that fed on MAM1 plants compared to those that fed on Col-0 plants (Fig. 4c). The MAM1 mutant has a different aliphatic GLS profile (especially 4MSOB) than Col-0; therefore, the level of NSP might vary in response to a type of aliphatic GLS. MA was only down-regulated in larvae that fed on quad-GLS mutants with no GLSs (Fig. 4c). Since MA was not down-regulated in larvae fed on MAM1 and MAM3 plants (which have different aliphatic GLS profiles), MA might not respond to aliphatic GLSs. Therefore, the observed down-regulation of MA in larvae fed on quad-GLS plants might be caused by a lack of indolic GLS.
When comparing the NSP and MA gene expression profile between the two feeding experiments (with Brassicaceae plant species and with A. thaliana mutants), we found discordant expression patterns only in NSP. While the level of NSP did not respond to the presence of aliphatic GLS in the two plant species assays (Fig. 3a,c), it did respond to differences in the aliphatic GLS profiles in A. thaliana mutants (MAM1 and MAM3). Although the expression response of NSP to aliphatic GLSs seems contradictory, we lack the entire benzylic GLS, which is abundant in C. occulta, in our A. thaliana mutant assay, and so cannot state definitively what such a contradiction means. However, it may be that benzylic GLSs are a major target of NSP and so able to trigger its expression, resulting in the different patterns of NSP expression observed in the two feeding experiments. Therefore, further analyses are necessary, especially those that focus on larval responses to benzylic GLSs.
In both feeding assays, the expression profile of MA was similar. The expression of MA was elevated in larvae that fed on A. kamchatica, which had more indolic GLS than did C. occulta (Fig. 3c,d), as well as in larvae that fed on mutants which have indolic GLSs. If, as mentioned above, the observed down-regulation of MA in quad-GLS A. thaliana mutants was caused by a lack of indolic GLS, MA regulation could be a result of the same GLS trigger. Although the responses of NSP and MA should be tested in specific and controlled experiments, our results suggest that NSP and MA might share GLS-disarming functions that are related but not identical in the many species of Pieris butterflies. www.nature.com/scientificreports www.nature.com/scientificreports/ The regulation of detoxification-related gene expression in herbivores fed on different host plants have been compared in many studies [5][6][7][8] . In most of these, the authors focused mainly on differential gene expression in generalist herbivores as a result of dietary metabolites or on host plant family differences to understand the molecular mechanisms which enable a wider host plant range than in specialist herbivores. However, larval responses to gene regulation in different types of host plants in specialist herbivores have not been well tested 6 . Here we show that even specialist herbivores may fine-tune adaptive gene expression in response to variations in a class of host plants' major chemical defenses, the GLSs.
Our study demonstrates that NSP and MA, which are members of an important gene family involved in host plant adaptation in Pieridae, are induced by the presence of GLSs and differentially expressed in larvae that were fed on plants that have different GLS profiles. In addition, we also found that the regulation patterns of NSP and MA were mostly conserved in the four Pieris butterfly species used in this study. Although the function of MA is still unclear, our results strongly support the idea that not only NSP but also MA is involved in the molecular adaptation mechanisms relied on by Pieris butterflies to overcome the GLS defense system in their host plants. This hypothesis is also supported by the fact that A. cardamines, which seems to lack NSP and has only MA, can use Brassicaceae plants as hosts 19,35 . Furthermore, our results also indicate that NSP and MA have different functions. Although the functional difference between the two genes still needs to be confirmed biochemically, our results suggest that a dynamic gene family has enabled Pieris butterflies to overcome the diversity of GLSs and radiate widely, becoming one of the most successful herbivore groups that feed on Brassicales plants. Further understanding the relationships between gene evolution and the function of the NSP-like gene family and the host plant spectrum of the Pieridae can therefore help to shed light on the molecular mechanisms that mediate the coevolutionary arms race between plants and herbivores.

Materials and Methods
Feeding experiments using two wild Brassicaceae plant species with GLS analysis. We conducted feeding experiments using four closely related Pieris butterfly species (P. napi, P. melete, P. rapae, and P. brassicae) and two Brassicaceae plants from different genera (Arabidopsis kamchatica and Cardamine occulta). We collected female egg-laying butterflies from three out of the four Pieris species from the wild population in Hokkaido (P. napi, P. rapae) and Chiba (P. melete), Japan. For P. brassicae, we collected final instar larvae in Hokkaido (Japan), reared them to adults and mated them by hand-pairing to get fertilized females. We placed the fertilized butterflies in chambers with their host plants (Cardamine leucantha for P. napi and melete, Brassica oleracea var. capitata for P. rapae and brassicae) under high-intensity light conditions. Acquired eggs were incubated at 25 °C and neonates were used for feeding experiments immediately after hatching. We collected seeds from the two species of Brassicaceae plants from the wild population. These seeds were watered, and germinated seeds were transplanted to vermiculite soil. We watered plants once a week with optimally diluted Hyponex solution (N:P:K = 6:10:5; Hyponex, Osaka, Japan). We reared the plants for 2 months under these conditions: 25 °C, with 60% relative humidity and L16:D8. Next, for the feeding experiments, 3 neonates were applied to one plant using a soft-haired brush. We replicated this set twice for each Pieris species and harvested 6 individuals in total from each plant species after 120 hours of feeding. After the harvested larvae were individually weighed (within 0.1 mg), they were flash-frozen in liquid nitrogen immediately and stored at −80 °C until RNA extraction. We conducted FDR-adjusted pairwise t tests to identify statistically significant differences in larval growth among the treatments for each species.
We used these plants not only for the feeding experiments but also for GLS profile analyses. We harvested leaves from three undamaged individual plants and froze them with liquid nitrogen. Leaves were freeze-dried and ground with metal beads. An aliquot of each powdered sample was pooled for each species and analyzed three times by tandem quadrupole mass spectrometry (TQMS) coupled with ultra-performance liquid chromatography (UPLC) 36,37 . We extracted peaks that showed >30 signal/noise ratios as detected peaks and identified GLSs following Sawada et al. (2009) and Sawada et al. (2017) 36,37 . The relative concentrations of each GLS among samples were calculated by comparing the peak area with the internal standard (10-camphorsulfonic acid). Detected GLSs were sorted into major GLS chemical classes: short-chain aliphatic (-C5), long-chain aliphatic (C6-8), benzylic and indolic GLS.
RNA extraction, RNA-seq, de novo assembly, NSP-like gene family sequence identification and gene expression level analysis. For each butterfly species, we selected one representative larva from the two treatments (two plant species). We chose 8 larvae for RNA sequencing (larvae of 4 Pieris species fed on 2 different plant species each) in total. We extracted RNA with RNeasy Mini Kit (QIAGEN). Extracted RNA samples were quality checked with an Agilent 2100 Bioanalyzer, and all samples were confirmed to have high-quality total RNA. The library for RNA-seq was prepared by Sure Select Strand-Specific RNA Library Preparation Kit for Illumina Multiplexed Sequencing. We sequenced the samples individually on a HiSeq. 1500 (100 bp paired-end read technology). Acquired reads were trimmed by trimmomatic software with the following options (LEADING:10 TRAILING:10 SLIDINGWINDOW:4:20 MINLEN:40-normalize_reads) 38 . For de novo assembly, we pooled all of the trimmed reads from the same species. We conducted de novo assembly with Trinity ver. 2.0.6 for each species 39 . For identifying NSP-like gene family sequences, we used tblastn with setting the assembled contigs (backbone) as databases and NSP-like gene family protein sequences from P. rapae as queries (GenBank accession number AAR84202, ABY88945, ABY88946), and the e-value threshold was set as 1.0e-4 40 . We extracted hit contigs for each query from each species, aligned and trimmed with MEGA6 to reference sequences of NSP-like gene family members from Pieris species stored in GenBank 41 . We made a ML molecular phylogeny of acquired sequences with reference sequences to confirm our annotation (in amino acid level). To measure the relative expression level of each extracted gene, we excluded redundant isoforms of NSP-like gene family members observed in assembled contig backbones and replaced them with trimmed representative www.nature.com/scientificreports www.nature.com/scientificreports/ sequences. Relative expression levels of each gene were estimated by mapping trimmed reads on an assembled backbone by RSEM 42 . Fragments per kilobase of exon per million reads mapped (FPKMs) were used as a relative expression level for each gene.

Gene expression analysis by qPCR.
We also conducted RT-qPCR to confirm the gene expression levels of NSP-like gene family members. We chose P. melete as a representative species, and 3 larvae from each treatment were used for RT-qPCR. We designed primers for RT-qPCR analysis with the following Primer3Plus settings: product size = 70-180 bp, Tm = 59-61 °C, GC% = 40-60%, Max Poly-base = 3 for members of NSP-like gene family 43 . We also designed primers for EF1α, which is frequently used as a housekeeping reference gene in insects for qPCR 6 . Designed primers are listed in Table S2. We extracted RNA as described above, and after confirming the quality of RNA by Agilent 2100 Bioanalyzer, we digested gDNA from each extracted RNA sample using TURBO DNA-free Kit (QIAGEN). We synthesized cDNA with Prime Script RT reagent Kit with gDNA Eraser (Perfect Real Time) (TAKARA) after RNA purification by RNA Clean & Concentrator kit (ZYMO research). We ran RT-qPCR reactions with a CFX Connect Real-Time PCR Detection System (BIO-RAD) using SYBR Premix Ex Taq (Tli RNase H Plus) with two technical replicates for each sample. We verified specific amplification by performing a melting curve analysis from 65 °C to 95 °C. We calculated relative gene expression levels by the ΔCT method normalized by EF1α 44 . We conducted one tailed Mann-Whitney U test to see expression level differences between the treatments follow the trend we found in RNA-seq based expression analysis with software Rstudio ver. 1.0.136 45 . Raw qPCR data are available in Table S3.
Feeding experiments using Arabidopsis thaliana mutants with different GLS backgrounds. We prepared one wild-type (Col-0) and three mutant lines of Arabidopsis thaliana which have different GLS profiles (MAM1, MAM3, quad-GLS). We grew these four lines under short day conditions (25 °C, 8L16D, 60% humidity), and used them for feeding experiments 5 weeks after germination. In this experiment, we used P. napi as a representative species. We collected P. napi larvae in Fukushima, Japan, and reared them to adults. Adults were paired by hand, and acquired neonates were used for the feeding assay. We followed the same protocol as we used for the feeding experiments with the two wild Brassicaceae plants described above. We applied 5 larvae to each mutant individual and replicated this set 4 times (n = 20). We harvested larvae after 120 h feeding and weighed them. We conducted FDR-adjusted pairwise t tests to identify statistically significant differences in larval growth among the treatments. 5 larvae from each treatment were randomly chosen and dissected for further expression level analysis. Mid-gut samples were flash-frozen and stored at −80 °C until RNA extraction. RNA was extracted with innuPREP RNA Mini Kit (Analytik Jena, Germany). We conducted RT-qPCR as described above to measure the expression levels of the NSP-like gene family. We conducted FDR-adjusted pairwise t tests to identify statistically significant differences in gene expression levels among the treatments. Raw qPCR data are available in Table S4.

GLS analysis of A. thaliana mutant lines.
We harvested entire rosettes of 5 individuals from each A. thaliana mutant line and froze them with liquid nitrogen. The samples were freeze-dried and ground by metal beads in a shaker. 10 mg of grounded leaf powder was used for chemical analysis. We added 80% of methanol with 50 µM of 4-hydroxybenzyl GLS (Sinalbin), which is absent in A. thaliana, to each mix as an internal standard. After 5 minutes of incubation with 230 rpm of shaking, we spun down the samples with 130,000 rpm for 10 minutes. We added the supernatant to filters conditioned with DEAE sephadex A-25. We washed the filter columns once with 500 µl of 80% MeOH and twice with 1 ml of water. After a final washing step with 1 ml of MES buffer pH5.2, we added 30 µl sulfate to convert GLS into desulfo GLS and incubated each sample overnight at room temperature. We eluted each column with 0.5 ml water and analyzed each using HPLC-UV with a reverse-phase C-18 column (Nucleodur Sphinx RP, 250 mm × 4.6 mm, 5 μm, Machrey-Nagel, Düren, Germany). Desulfo GLSs were identified based on the retention time and UV spectra with known standard libraries 46 .

Data Availability
The RNA-seq short read data have been deposited in the EBI short read archive (SRA) with the following sample accession numbers: ERX2829492-ERX2829499. The complete study can also be accessed directly using the following URL: http://www.ebi.ac.uk/ena/data/view/PRJEB29048.