Ultra-rapid auxin metabolite profiling for high-throughput mutant screening in Arabidopsis

A high-throughput screening of auxin metabolites in combination with multivariate data analysis can be used to identify novel regulators of the IAA metabolism in the model plant Arabidopsis thaliana.

Another auxin inactivation mechanism is the formation of conjugates with amino acids or sugars (Tam et al., 2000;Kowalczyk and Sandberg, 2001). Some of these conjugates might be hydrolysed, releasing free active auxin, indicating that they may serve as temporary storage forms of the inactive hormone (reviewed by Ludwig-Müller, 2011). However, in Arabidopsis thaliana, the most abundant amide-linked conjugates, IAA-aspartate (IAAsp) and IAA-glutamate (IAGlu), are not reversibly converted to IAA, and probably serve as degradation intermediates (Kowalczyk and Sandberg, 2001;Woodward and Bartel, 2005). Recent research suggests that conjugation of IAA to amino acids could play important roles in specific developmental processes (Zheng et al., 2016;Di Mambro et al., 2017). Nevertheless, the roles and regulation of the different pathways of auxin biosynthesis and degradation/ conjugation are not well understood.
To improve understanding of auxin metabolism in specific tissues or processes, information on levels of the free hormone, its biosynthetic precursors, and its major metabolites is highly important. However, analysing plant hormones is challenging due to their very low concentrations and the complexity of plant extracts . Furthermore, many of these compounds are unstable and can be easily degraded during extraction and purification. Fortunately, some of the problems associated with phytohormone analysis can be overcome by exploiting recent advances in analytical techniques, such as ultra-fast LC coupled with high-sensitivity tandem MS (LC-MS/MS) Porfírio et al., 2016). For samples containing minute amounts of tissue, the sensitivity of the analytical method can also be improved by miniaturization of the extraction and purification steps, which can minimize analyte losses due to adsorption to surfaces and/or increase analyte recovery in the solid phase extraction (SPE) step (Svačinová et al., 2012;Novák et al., 2017). Micro-purification techniques have previously been used for auxin isolation, and these include solid phase microextraction (SPME) (Liu et al., 2007), in-tip micro-SPE (µSPE) (Liu et al., 2012), and magnetic SPE (Zhang et al., 2010). Nevertheless, further advances are still required.
There is a particular need for a simple, high-throughput analytical approach that provides sufficient robustness, sensitivity, and selectivity for analyses of large numbers of samples, such as from mutant collections. In the study presented here, an ultra-rapid auxin profiling method, involving micro-scale in-tip SPE (in-tip µSPE) and ultra-fast LC-MS/MS analysis, was developed and evaluated. The high-throughput approach was successfully validated against the previously published purification protocol using a polymer-based reversed phase sorbent . The method was then used for screening a collection of Arabidopsis mutant lines (Berná et al., 1999;Pérez-Pérez et al., 2009), and several lines with altered IAA metabolite profiles were identified using multivariate data analysis (MVDA). Compared with previously published methods, the approach presented here is less timeconsuming and allows greater processivity (>100 samples d -1 ). This makes it an ideal tool for the screening of large collections  Mano and Nemoto (2012), Kasahara (2016), Ludwig-Müller (2011), andPorco et al, 2016. Dashed arrows indicate steps in which the enzymes catalysing the reaction are not known. Significantly higher (red) or lower (green) concentrations of individual metabolites in investigated mutant lines showing the greatest difference from wild-type IAA metabolomes are indicated. of lines, and subsequent identification of altered metabolite profiles, which will ultimately help researchers to find novel players in specific IAA metabolic pathways.

Plant material and growth conditions
Seven-day-old Arabidopsis wild-type seedlings were used as the material for development and validation of the ultra-rapid auxin metabolite profiling method. The 64 Arabidopsis leaf mutants used in this study were isolated in the laboratory of José Luis Micol and have been described by Berná et al. (1999) and Pérez-Pérez et al. (2009). The sur2-1 mutant line (Barlier et al., 2000) was included as a positive control. Arabidopsis thaliana Columbia-0 (Col-0) and Landsberg erecta (Ler) wild-type accessions were obtained from the Nottingham Arabidopsis Stock Centre (NASC). All seeds were surface-sterilized using a bleach solution containing 0.002% Triton X-100 and then sown on Murashige and Skoog square agar plates (4.4 g l -1 Murashige and Skoog, 0.5 g l -1 MES monohydrate, and 8 g l -1 plant agar, pH 5.7). After 3 d of stratification, the plates were placed vertically in long-day conditions (16 h light/8 h dark) at 22 ± 1°C under cool white fluorescent light (maximum irradiance 550 μmol m −2 s −1 ). Whole seedlings were collected in five replicates and weighed, immediately frozen in liquid nitrogen, and stored at -80 °C until extraction.

Extraction and purification of IAA metabolites
Frozen samples were placed in a crushed-ice bath in order to avoid enzymatic degradation of analytes. For quantification of IAA and its metabolites, samples containing 10 mg (FW) of plant material were extracted in 1 ml of ice-cold Na-phosphate buffer (50 mM, pH 7.0, 4 °C) containing 0.1% diethyldithiocarbamic acid sodium salt. The following stable isotopelabelled internal standards were added to each sample: . The samples were homogenized using a MixerMill MM 301 bead mill (Retsch GmbH; http://www.retsch.com) at a frequency of 29 Hz for 6 min after adding 2 mm ceria-stabilized zirconium oxide beads. The plant extracts were incubated at 4 °C with continuous shaking (10 min), centrifuged (15 min, 23 000 g at 4 °C), and purified by in-tip µSPE using self-packed multi-StageTip columns prepared according to Svačinová et al. (2012). The columns contained two types of extraction sorbents (three layers of each type): C 18 and SDB-XC (Empore™, 3M™; http://www.3m.com).
Multi-StageTips with C 18 /C 8 and C 18 /SDB-RPS combinations of sorbent types were also prepared for development of the purification method. Briefly, microcolumns of both kinds were activated sequentially with 50 μl each of acetone, methanol, and water (by centrifugation at 2200 rpm, 10-15 min, 4 °C), and aliquots of the acidified sample extract were applied (3400 rpm, 25 min, 4 °C). The microcolumns were then washed with 50 μl of 0.1% acetic acid (3400 rpm, 15 min, 4 °C), and samples were eluted from the C 18 /C 8 and C 18 /SDB-RPS sorbents with 50 μl of 80% methanol and 50 μl of 0.5 M NH 4 OH in 80% (v/v) methanol (3400 rpm, 15 min, 4 °C), respectively. To validate the final μSPE protocol, extracts were also purified on Oasis HLB columns (30 mg, Waters Corp., Milford, MA, USA), conditioned with 1 ml of methanol, 1 ml of water, and 0.5 ml of Na-phosphate buffer (pH 2.7) as described by Novák et al. (2012). After sample application, the columns were washed with 2 ml of 5% methanol and then eluted with 2 ml of 80% methanol. All eluates were evaporated to dryness and stored as described above.

Quantification of IAA metabolites
The evaporated samples were dissolved in 40 μl of mobile phase prior to LC-MS/MS analysis, using a 1290 Infinity LC system and a 6490 Triple Quadrupole LC/MS system equipped with Jet Stream and Dual Ion Funnel systems (Agilent Technologies, http://www.home. agilent.com). A 20 μl portion of each sample was injected onto a reversed-phase column (Kinetex C18 100A, length 50 mm, diameter 2.1 mm, particle size 1.7 μm; Phenomenex, http://www.phenomenex.com), and the analytes were eluted by a 3 min linear gradient of 5:95 to 35:65 A:B, where A and B are 0.1% acetic acid in methanol and 0.1% acetic acid in water, respectively. The column was then washed with 100% methanol (1.0 min), and re-equilibrated to initial conditions (1.0 min). Throughout the procedure, the flow rate was 0.5 ml min −1 , and the column temperature 40 °C. The effluent was introduced into the MS system with the optimized settings listed in Supplementary Table S2 at JXB online. Analytes were quantified using diagnostic multiple reaction monitoring (MRM) transitions of precursor and appropriate product ions using optimal collision energies and 50 ms dwell time (Table S2). Chromatograms were analysed using MassHunter software (version B.05.02; Agilent Technologies), and the compounds were also quantified by standard isotope dilution analysis (Rittenberg and Foster, 1940).

Experimental design and data analysis
Seeds from all Arabidopsis mutant and Ler wild-type lines were sown in five rows per plate, with 50-70 seeds per row. Plates containing the Ler seedlings were randomly placed along the racks, so that any growth variation due to the position of the plate on the growth room shelves would be represented in the reference data. Replicates containing ~10 mg (FW) of 7-day-old seedlings were harvested. For each mutant line, including the sur2-1 positive control, five biological replicates were harvested, while the Ler wild type was represented by 40 replicates.
Multivariate data analysis was conducted using 'Soft Independent Modeling of Class Analogies' (SIMCA) software version 13 (Umetrics AB, Umeå, Sweden). Clustergrams were drawn in R software.

Plant phenotyping
For plant phenotyping, plates containing vertically grown 10-day-old seedlings were scanned. The digital images were used for primary root and hypocotyl length measurements using FIJI software. The number of lateral roots was counted from the images.

Development of a µSPE purification method for IAA metabolites
One of the most crucial steps in the development and optimization of a purification method for metabolite profiling is selection of suitable SPE sorbents. Ideally, they should afford good retention and high recovery of all target compounds (which may have diverse chemical properties), while excluding others. They must also be compatible with any miniaturized equipment to be used. Here we made our own stop-and-go extraction tips (StageTips) from ordinary pipette tips containing very small disks made of beads with reversed-phase, cation-exchange, or anion-exchange surfaces immobilized on a Teflon mesh (Rappsilber et al., 2003). We previously showed that auxin metabolites can be efficiently retained by reversedphase sorbents . Therefore, two sorbents with long alkyl chains (octyl, C 8 ; and octadecyl, C 18 ) and two poly(styrenedivinylbenzene) co-polymer-based sorbents (SDB-RPS and SDB-XC) were selected and used for testing in the study presented here.
Accordingly, IAA and its precursors and degradation products were retained and eluted with varying efficiency ( Fig. 2A). A combination of C 18 /C 8 sorbents retained 19-75% of all of the IAA metabolites, except the most polar compounds (TRA and Trp). StageTips combining C 18 and SDB-based sorbents provided the highest extraction yields. Because SDB-RPS modified with sulphonic acid groups allows reversed-phase and cation-exchange interactions, amine-containing basic analytes, ANT, TRA, and Trp, were enriched using C 18 /SDB-RPS columns with high recoveries: 63 ± 1, 70 ± 2, 82 ± 1, and respectively. However, recoveries of IAA-glc and oxIAAglc were far too low with these columns (1.6 ± 0.1% and 1.0 ± 0.1%, respectively), due to degradation of these conjugates under the alkaline conditions (Supplementary Table S1) required for elution from the SDB-RPS sorbent. In contrast, use of C 18 /SDB-XC columns, which do not exploit the cation-exchange interactions, allowed maximization of yields from the in-tip µSPE step and minimization of losses due to pH lability ( Fig. 2A). Moreover, the C 18 /SDB-XC combination retained the polar compounds (TRA and Trp) more strongly than the reversed-phase sorbents, C 18 /C 8 ; this combination allows lower recovery of Trp during the purification step (10 ± 2%), but this is not limiting due to the very high endogenous levels of Trp in plants . Therefore, we decided to optimize the one-step purification protocol for multi-µSPE columns further by using a combination of C 18 and SDB-XC sorbents.
In order to minimize contamination with non-polar extractable substances that could interfere with subsequent MS-based analyses (e.g. pigments and lipids), we used 50 mM Na-phosphate buffer (pH 7.0) as the extraction solvent (Edlund et al., 1995). To maximize recoveries of the analytes when using the µSPE-based approach, we also optimized the loading conditions, by assessing the effects of variations in conditions on analytes classified in terms of their stability and acidity or basicity (in three classes: basic, neutral, and acidic). As shown in Supplementary Fig. S1, we obtained higher recovery under acidic conditions (pH 2.7) for most metabolites (mean recovery, 75 ± 25%, with overall means of 75, 65, and 90% for the basic, neutral, and acidic compounds, respectively) than under neutral pH. As expected, yields of the neutral IAA precursor IAOx were higher in the extraction buffer with neutral pH (96 ± 3%) than under acidic conditions (31 ± 1%), in accordance with the previously reported lability of IAOx in strongly acidic solutions .
IPyA has also previously exhibited instability in solution, therefore derivatization is required for its accurate quantification (Tam and Normanly, 1998;Mashiguchi et al., 2011). For this purpose, we converted IPyA to the thiazolidine product IPyA-TAZ by derivatization with cysteamine following Novák et al. (2012), but miniaturized the derivatization step by using 200 μl of crude plant extract (from ~2 mg FW of plant tissue) with 100 μl of 0.75 M cysteamine (pH 8.2). This afforded 15-fold greater yields of IPyA-TAZ from minute samples than the original derivatization protocol .
We then tested the influence of a complex multicomponent plant matrix on recoveries of diverse IAA metabolites in small amounts of plant tissue. The efficiency of the entire developed method was evaluated by spiking 10-dayold Arabidopsis seedling extracts with a mixture of stable isotope-labelled standards. We first examined the extraction capability of multi-StageTip columns packed with one, three, and five layers of each sorbent type (C 18 and SDB-XC) using a crude extract from 2.5 mg FW of plant tissue. As shown in Fig. 2B, recoveries increased as the number of sorbent multilayers increased (mean total extraction yields were ~30, 40, and 60% with one, three, and five bi-layers, respectively). Thus, IAA metabolites were most effectively enriched by using microcolumns packed with five layers of each sorbent (C 18 and SDB-XC). However, these columns were most prone to clogging due to the high number of layers (10 in total), which hindered the subsequent washing (0.1% acetic acid) and elution (80% methanol) steps. Consequently, multi-StageTips packed with three layers were used in further analyses, and their capacity to isolate IAA metabolites was tested by using them to purify extracts from four different amounts of fresh Arabidopsis tissue (1.0, 2.5, 5.0, and 7.5 mg). As expected, recoveries declined with increasing amounts of fresh plant tissue from overall means of 61 ± 22% for extracts from 1.0 mg to 26 ± 11% for extracts from 7.5 mg, due to overloading of the sorbents (Supplementary Fig. S2A). We concluded that 200 µl of an Na-phosphate buffer extract containing 2 mg FW of plant material was optimal for purifying IAA and its key metabolites using a C 18 /SDB-XC multi-StageTips microcolumn.
More than 100 samples can be extracted and purified per working day using this approach. The final method for high-throughput extraction and purification of Arabidopsis samples is shown in Fig. 2C.
Most of the precursor and product ions of IAA metabolites determined under optimized LC-MS/MS conditions corresponded well with previously published data . Moreover, unlabelled and labelled IAA-glc and oxIAAglc were included in the IAA profiling, and detected in negative-ion as well as positive-ion MRM mode. In accordance with previously published MS/MS patterns (Kai et al., 2007a), IAA and oxIAA ions (m/z 174 and 190, respectively) were detected as high-intensity fragments of IAA-glc and oxIAA-glc in negative-ion mode. However, it is well known that negative ion mode (ESI-)-MS is generally less sensitive than positive ion mode (ESI+) (see Supplementary Fig. S3). In efforts to increase sensitivity, we also examined ionization patterns of IAA-glc conjugates in (ESI+)-MRM mode. The neutral losses of a sugar moiety (162 Da) through in-source fragmentation of precursor ions (m/z 338 of IAA-glc and m/z 354 of oxIAAglc) produced high-intensity ions, and subsequent fragmentations in the quadrupole collision cell led to the creation of quinolinium/quinolonium ions (m/z 130 and 146, respectively) (Kowalczyk and Sandberg, 2001). Thus, MRM transitions m/z 176 >130 and m/z 192 >146 were used to detect IAA/IAA-glc and oxIAA/oxIAA-glc, respectively. Moreover, both molecule pairs were fully resolved under our reversedphase LC conditions ( Fig. 2C; Supplementary Table S2).
To determine the limits of detection (LODs) and linear calibration ranges of the ultra-fast LC-MS/MS method, we constructed calibration curves using data obtained from repeated injections of sets of standards in 12 amounts ranging from 1.0 fmol to 500 pmol. We also applied the stable isotope dilution method, comparing response ratios for each pair of unlabelled and labelled compounds. The responses covered a very broad linear range, spanning at least four orders of magnitude with correlation coefficients (R 2 ) exceeding 0.9975 (Supplementary  Table S2), in accordance with previously published linear ranges for LC-MS/MS methods (Matsuda et al., 2005;Pěnčík et al., 2009;Floková et al., 2014). In the optimized MRM mode, the LODs ranged from 2.5 fmol to 50 fmol (Supplementary  Table S2). Overall, the method's sensitivity allowed the straightforward analysis and determination of IAA metabolites in extracts of just 2.0 mg FW of Arabidopsis tissue, and >250 samples could be quantified per day.

Validation of the profiling method
The effectiveness of the method was validated by spiking experiments. The results confirmed the high precision and accuracy of the method (Supplementary Fig. S2B). The mean precision obtained in the spiking experiments with Arabidopsis extracts was 6 ± 3% RSD and the mean accuracy for all compounds was 1 ± 17% BIAS (percentage deviation from the accepted reference value), confirming the robustness of our method.
We also processed an extract from 7-day-old Arabidopsis seedlings using both our new µSPE-based method and the previously published one-step purification protocol, involving purification on reversed-phase Oasis™ HLB columns .
Levels of IAA and its metabolites (precursors, catabolites, and conjugates), analysed by ultra-fast LC-MS/MS and quantified by standard isotope dilution, following the two purification methods, were similar (Table 1). Furthermore, profiles and levels of most known auxin precursors and conjugates/catabolites were very similar to previously published patterns (Kai et al., 2007;Novák et al., 2012). The miniaturized SPE system also provided substantially better method precision (Table 1) than the commercially available polymeric HLB columns, with an overall mean of 3.4% RSD compared with 4.7% RSD. However, it was not possible to calculate IAOx and TRA levels in extracts purified by StageTips microcolumns, probably due to their very low endogenous levels, lower capacity of C 18 /SDB-XC sorbents (Fig. 2), and/or strong effects of the plant matrix.
Taken together, our results in these validation experiments demonstrate the accuracy and robustness of our high-throughput method for routine determination of key auxin precursors and conjugates/catabolites in minute samples of plant material (<10 mg FW).

High-throughput mutant screening of Arabidopsis lines
Next, we tested the ability of our method to identify genotypes with abnormal IAA metabolite profiles within a collection of Arabidopsis mutant lines. We chose 64 lines that were initially isolated based on perturbations in their leaf morphology (Berná et al., 1999;Pérez-Pérez et al., 2009). The Ler accession, which is the genetic background of the mutants, was used as a wild-type control for the analysis, and the auxin-overproducing mutant line sur2-1 (Barlier et al., 2000), which has an altered IAA metabolite profile compared with Ler ( Supplementary Fig. S4), was included among the mutant lines as a positive control.
High-throughput auxin metabolite profiling, combining the one-step in-tip µSPE purification protocol and ultra-fast LC-MS/MS analysis (Fig. 2C), was performed on a total of 365 samples (five biological replicates per mutant line and 40 Values are means ±SD (n=4); ND, not detected.
biological replicates of the Ler wild-type; see the Materials and methods). This approach allowed multiplex quantification of IAA and its precursors (ANT, Trp, IPyA, and IAN) and degradation products (oxIAA, IAAsp, IAGlu, IAA-glc, and oxIAAglc), but not (under our method conditions) of TRA and IAM, as their signal intensities were consistently below their respective LODs, and of IAOx, as this metabolite was detected in only 10 out of 66 analysed genotypes. MVDA was then applied to the data set to identify mutant lines with different IAA metabolite profiles from the Ler wild type. Clustering of the lines' average concentrations of metabolites revealed high diversity in the profiles ( Supplementary Fig. S5), as expected for mutants isolated based on a criterion (perturbations of leaf architecture) that might be associated with diverse metabolic phenotypes. At the same time, such variety in the profiles provides an indication of the high interconnection of the IAA metabolome with many different processes, which might be represented among the mutant lines. Nevertheless, our principal component analysis (PCA) revealed a few lines that are remarkably different from the wild type in terms of their IAA metabolite profiles (Fig. 3). Clustering of the Hotelling distance (T2) and distance to the model (DModXP+) values, which indicate how far a line is from the wild type in different PCA models, ranked the lines by their similarity to Ler, and identified those which were most different (Fig. 3A, top lines). These lines (ero1, sea4-1, ucu2-1, ond1, and rug1-1; Fig. 3B) separated from other samples, mainly due to their different contents of TRP, ANT, IAN, IPyA, IAA, oxIAA, and oxIAAglc (Fig. 3C, D; Supplementary Fig. S6). As expected, the sur2-1 mutant (Fig. 3B) was also found to be very different from the wild type (Fig. 3A) and markedly separated from other lines due to differences particularly in IAA, IPyA, IAN, oxIAA, IAAsp, IAA-glc, and oxIAA-glc levels (Fig. 3C). This is consistent with previously published data from auxin metabolite profiling of the sur2-1 mutant line by Novák et al. (2012), and corroborates the accuracy of our analytical method.
As shown in Fig. 4, metabolite profiles of the most different lines were very different from that of the Ler wild type. The observed differences in IAA metabolite profiles also provide indications of the IAA metabolic pathways affected in these mutant lines (Fig. 1). However, an altered IAA metabolite profile can be a cause (e.g. when the function of an enzyme associated with IAA metabolism, or a regulator of such enzyme, is perturbed) or a consequence (e.g. the result of any other misfunction that indirectly changes the IAA metabolome). It is fair to assume that perturbations in direct regulators of IAA metabolism will create the greatest alterations in the IAA metabolite profile. The sur2-1 mutant, which directly perturbs IAA metabolism, as it impairs the function of the cytochrome P450 monooxygenase CYP83B1 that regulates the levels of the IAA precursor IAOx (Barlier et al., 2000;Bak et al., 2001), was found to be the most different line in terms of its IAA metabolite profile ( Fig. 3A-C). Together with sur2-1, the ero1 and sea4-1 mutants were most separated from the wild type and other lines ( Fig. 3A-C), which raises the possibility that ERO1 and SEA4 genes are also involved in direct regulation of IAA metabolism. However, only the mutated genes in the rug1-1 and ucu2-1 mutants have been identified (Pérez-Pérez et al., 2004;Quesada et al., 2013), not those in the ero1, sea4-1, and ond1 mutants. The RUG1 gene encodes porphobilinogen deaminase, an enzyme of the tetrapyrrole biosynthetic pathway. rug1-1 mutant plants accumulate porphobilinogen, and both their vegetative and reproductive development is perturbed (Quesada et al., 2013). We found that the rug1-1 mutant also exhibited elevated levels of IAA, its precursor TRP, and the downstream products oxIAA and IAA-glc (Figs 1, 4). The particularly high levels of TRP in rug1-1 plants suggests a link between the tetrapyrrole pathway, TRP biosynthesis, and IAA homeostasis.
The ucu2-1 mutation, which causes increases in levels of IAA, IAAsp, oxIAA, and oxIAA-glc (Figs 1, 4), perturbs AtFKBP42, a peptidyl-prolyl cis-trans isomerase of the FK506-binding protein family that participates in auxin and brassinosteroid signalling (Pérez-Pérez et al., 2004). AtFKBP42 also mediates polar auxin transport through its requirement for proper localization of the ABCB/PGPefflux carriers to the plasma membrane (Wu et al., 2010;Henrichs et al., 2012). As found in this study, inefficient ABCB/PGP-mediated auxin efflux results in elevated levels of IAA and IAA degradation products in the ucu2-1 plants (Fig. 4), probably caused by feedback promotion of IAA biosynthesis in the sink tissues, devoid of auxin as a result of the IAA gradient disruption in the mutant (Wu et al., 2010). A detailed study of the lines identified by the present approach might reveal new players in the IAA metabolic pathways, as well as new interconnections between IAA metabolism and other pathways operating in plants.

Phenotypical analysis of the mutant lines
To explore the relationship between IAA and the levels of other IAA metabolites among the lines, we focused on the endogenous IAA contents of the 64 mutant lines, and selected eight lines with lower and seven lines with higher IAA contents than the wild type ( Supplementary Fig. S7A). PCA on the selected lines with low and high IAA levels revealed that, together with IAA, a major component explaining the separation of the lines is IAAsp, IAN, and IPyA on one side, and IAA-glc and oxIAAglc on the other side (Fig. 5A). This indicates, regardless of the phenotype of the mutants, a behaviour of the IAA metablome in which IAAsp, IAN, and IPyA levels are positively correlated, Fig. 4. IAA metabolite concentrations in the Ler wild type and the ero1, rug1-1, sea4-1, ucu2-1, and ond1 mutants. Seedlings of wild-type Arabidopsis Ler and mutant lines were collected in five replicates of 10 mg, and the IAA and IAA metabolites were analysed by LC-MS/MS. Error bars represent the SD. Asterisks indicate a statistically significant difference in a two-tailed Student t-test at a significance level of 0.01. while IAA-glc and oxIAA-glc levels are negatively correlated with IAA levels in these particular lines (Fig. 5A).
We then explored how the IAA metabolite profiles correlated with IAA-related phenotypes. We determined the length of the primary root and the hypocotyl, as well as the density of lateral roots, in the selected lines ( Supplementary Fig. S7B-D), as these phenotypes are highly dependent on IAA levels (Porco et al., 2016;Zheng et al., 2016). Orthogonal Projections to Latent Structures (OPLS) models were constructed using the phenotypic measurements and the metabolite concentrations in these lines. While no model could be generated for the hypocotyl length, due to the lack of correlation of this phenotype with the IAA metabolite profiles in these specific lines, OPLS models for primary root length and lateral root density were generated (Fig. 5B). These two models show very similar patterns, in which IAA, together with IPyA, IAN, oxIAA, and IAAsp, levels are negatively correlated with the primary root length and the lateral root density in the lines examined (Fig. 5B). The fact that these metabolites are negatively correlated with these phenotypes should be understood, however, as a reflection of the positive correlation between IAA and IPyA, IAN, oxIAA, and IAAsp (Fig. 5A), which is the bioactive metabolite. Finally, it is important to note here that OPLS models were constructed based on metabolite profiles from whole Arabidopsis seedlings, whilst morphological phenotypes may depend on local increases or decreases in IAA levels. Tissue-specific IAA metabolite profiling should be the choice to create stronger models from which to draw solid conclusions about the relationship between the IAA metabolome and specific developmental phenotypes.
In summary, the method presented here is a powerful tool for rapid IAA metabolome screening, which provides a valuable resource of informative data and facilitates the identification of novel regulators of IAA metabolism. It is important to note, however, that the IAA metabolome can be significantly perturbed by many different processes, as suggested by the rug1-1 and ucu2-1 mutants and by the huge diversity of IAA profiles among the studied genotypes ( Supplementary Fig. S5). A case-by-case study of selected mutants after PCA is decisive to discern between direct and indirect effects.

Concluding remarks
Overall, the results show a high potential of our method for powerful, rapid metabolome-based screening to measure IAA metabolites, characterize the pathways and the interactions involved, and to help identify novel regulators of the homeostasis of IAA and its metabolites. We have developed a robust ultra-rapid MS-based approach for extracting, purifying, and quantifying most known IAA metabolites, including IAA and its key precursors and conjugates/catabolites, from minute plant samples for high-throughput mutant screening. Sample extraction and micro-scale purification conditions were optimized using a selection of appropriate sorbent types. The method was tested using a collection of previously isolated mutant lines.  Supplementary Fig. S7 and 10 IAA metabolites. The scatter plot of the scores (left) shows the separation of the lines with higher (green) and lower (blue) IAA levels than in the wild type, based on the metabolites shown in the scatter plot of the loadings (right). The percentage of the variance explained by the components t1 and t2 is 41.9% and 16.5%, respectively. (B) OPLS predictive loadings for PCA models of the primary root length (left) and lateral root density (right) using the selected lines in Supplementary Fig. S7  Affording analysis of >100 samples per day, the procedure is less time-consuming and much more effective than previously published methods. It will allow researchers to quantify auxin metabolites, in large numbers of samples containing a few milligrams of fresh plant material, highly accurately and reproducibly. In addition, the reduction in the amount of plant material greatly facilitates a tissue-specific metabolome-based screening of IAA and IAA metabolites. In combination with multivariate data analysis, the method provides a powerful tool for mutant screening and potential identification of novel regulators of the IAA metabolism.
Furthermore, the same approach could be applied to other classes of plant hormones and metabolites, and used for highthroughput metabolic phenotyping of plants with different genetic backgrounds (e.g. knockout and overexpressing mutant and transgenic lines). The application of the present method to wild-type accessions or crop cultivars will be of great use for gene discovery (e.g. in genome-wide association studies based on hormone metabolomes) and for breeding programmes. Finally, we highlight the importance of interdisciplinary collaboration between plant biologists, mathematicians, and chemists for advancing our understanding of IAA and its functions as a crucial plant growth regulator.

Supplementary data
Supplementary data are available at JXB online. Fig. S1. Effects of tested loading conditions on the recovery (%) of IAA metabolites after purification by in-tip μSPE. Fig. S2. Method optimization and validation. Fig. S3. Comparison of signal sensitivities of IAA-glc and oxIAA-glc in analyses by LC-MS/MS using negative-ion (ESI-) and positive-ion (ESI+) multireaction monitoring (MRM) modes. Fig. S4. IAA metabolite profiles in 7-day-old Arabidopsis seedlings of the wild-type accessions Landsberg erecta (Ler) and Columbia (Col-0), and the IAA-overproducing mutant line sur2-1. Fig. S5. High-throughput IAA metabolite profiling of the lines. Fig. S6. Separation of the mutant lines according to their IAA metabolite levels. Fig. S7. Auxin-related phenotypes in the Arabidopsis mutant lines selected by an ultra-rapid auxin metabolite profiling method. Table S1. Stability of IAA-glc and oxIAA-glc in indicated solutions with pH 3-12, 0.1% acetic acid and 80% methanol. Table S2. Diagnostic MRM transitions, optimized collision energies, retention time stability, limits of detection (LOD), dynamic linear range, and linearity (correlation coefficients, R 2 ) of the LC-MRM-MS method.