To achieve the compartment-specific accumulation of SAL1 and expression of SAL1, we generated six chimeric constructs (Supplementary Figure S1). Constructs SAL1_I to SAL1_IV consisted of a truncated SAL1 (SAL1tr) backbone fused to various organellar targeting sequences. This SAL1tr backbone was previously reported to be localized to the cytosol and the nucleus and to complement the sal1 phenotype, in the absence of organellar targets (Kim and von Arnim, 2009). SAL1_I was designed to express the protein exclusively in the nucleus and contained SAL1tr fused to a nuclear localization sequence (NLS) at the C-terminus. Construct SAL1_II was created to express SAL1tr exclusively in the cytosol and was a fusion of the SAL1tr backbone to the nucleus exclusion sequence (NES) from At1g07140. This NES was necessary to avoid SAL1 localization to the nuclei, as described by Kim and von Arnim (2009). To constructs II–V, we added NES at the C-terminus of SAL1 by incorporating it into primers by PCR. Construct SAL1_III was designed to express SAL1 exclusively in plastids and consisted of SAL1tr fused to the chloroplast pre-sequence (cPS) of the Rubisco small subunit (SSU) at the N-terminus and to the NES at the C-terminus. Construct SAL1_IV aimed to express SAL1 exclusively in the mitochondria and contained SAL1tr fused to the mitochondrial pre-sequence (mPS) or transit peptide of heat shock protein 90 (Hsp90) (Krishna and Gloor, 2001) at the N-terminus and NES at the C-terminus. Construct SAL1_V encoded the native SAL1 pre-sequence and full-length SAL1 fused to the NES at the C-terminus and should target SAL1 to plastids and the mitochondria. SAL1tr, as described by Kim and von Arnim (2009), was designated SAL_VI.

To confirm the subcellular localization of SAL1 fusion proteins experimentally, we used Arabidopsis root cell suspension cultures and mesophyll protoplasts (Figure 2 and Supplementary Figure S2). Suspension cells from Arabidopsis roots were transformed with Agrobacterium carrying constructs encoding SAL1_I–VI fused to green fluorescent protein (GFP) at the C-terminus (Berger et al., 2007), and protoplasts were isolated from the mesophyll of leaves and transfected using purified plasmid DNA (Yoo et al., 2007). Fluorescence confocal microscopic analysis, shown in Figure 2 and Supplementary Figure S2, confirmed the expected localization of the transiently expressed SAL1_(I–VI):GFP fusion proteins in the designated compartments. DAPI staining of Arabidopsis root cells expressing SAL1 _I-GFP and SAL1_VI-GFP (Figures 2A,F) confirmed the nuclear localization of both constructs (nuclei of cells containing both DAPI and SAL1-GFP are indicated by white arrows). The cytosolic localization of SAL1_II-GFP and SAL1_VI-GFP was observed in cells showing equal distribution of GFP in the cytosol (Figures 2B,F). Cells transfected with the SAL1-VI construct and showing SAL1-GFP in the cytosol are marked by red arrows (Figure 2F). The chloroplastidic localization of SAL1_III_GFP can be also confirmed (Figure 2C) as chlorophyll autofluorescence coincides with the GFP signal. Similarly, the mitochondrial localization of SAL1_IV (Figure 2D) was revealed by confirming the presence of the MitoTracker signal in the same structures as SAL1_IV-GFP. The GFP signal of SAL1_V was difficult to interpret as the expression level of this construct was weak (adjustments of levels were applied in Photoshop CS3). Nevertheless, the presence of SAL1_V protein in chloroplasts and in some mitochondria-like structures in the cytosol can be assumed (Figure 2E). The co-localization of SAL1_V with the MitoTracker was not successful as the GFP signal was too weak for this assay. Interestingly, the expression of SAL1_V in Arabidopsis suspension cells from roots, which lack chloroplasts (Supplementary Figure S2), showed intensive GFP staining in tiny mitochondria-like structures in the cytosol. In line with this observation Chen et al. (2011) and Estavillo et al. (2011) have previously demonstrated that when the full-length protein of SAL1 is fused to GFP (similar to the SAL1_V construct used in this work), SAL1 will be found in both chloroplasts and mitochondria. The only difference in the SAL1_V construct used in this work from that used by Chen et al. (2011) and Estavillo et al. (2011) is that our construct contained NES. However, as NES did not lead to mislocalization of constructs SAL1_II, SAL1_III, and SAL1_IV, the potential mislocalization of SAL1_V-GFP is less probable.

To study the function of SAL1 in different cell compartments in planta, sal1 mutant plants were transformed with the respective SAL1 chimeric constructs. We selected independent transgenic lines that showed varying SAL1 transcript levels and that were either marginally, moderately, or fully complemented in terms of the overall rosette morphology (Figure 3). We isolated 77 independent transgenic lines for each of SAL1_I, SAL1_III, and SAL1_IV, 40 lines for SAL1_V, and 12 lines for each of SAL1_II and SAL1_VI. Although we repeated the transformation process several times, we were not able to significantly increase the number of transgenic lines for constructs SAL1_II and SAL1_VI, indicating that SAL1 localization in the cytosol was probably not indiscriminate for the survival of transgenic plants. We then selected six to nine lines for each construct, which showed different levels of complementation. These lines were first compared to wild type and sal1 in their visual appearance of rosette morphology, SAL1 mRNA levels, shoot fresh weight (biomass) of fully developed 5-week-old adult plants, and PAP levels (Figures 3–5).

Loss of SAL1 function results in low total sulfate levels, a decreased accumulation of GSLs, an increased level of desulfo-precursors, and a decreased level of thiols (Lee et al., 2012). Furthermore, sulfur assimilation in sal1 mutants was not only affected at the metabolic level, but the transcript profile of genes was similar to that of sulfate-starved plants (Lee et al., 2012). Therefore, in addition to the general processes related to growth and stress response, the nutritional status was also impaired in sal1 plants. To address the role of the compartmentalization of SAL1 in sulfur assimilation, we selected the best-complemented transgenic lines, SAL1_I 26, SAL1_II 1, SAL1_III 53, SAL1_IV 8, SAL1_V 13, and SAL1_VI 6 (Figures 4, 5), and analyzed their sulfur metabolite profiles (Figure 6). The accumulation of sulfur metabolites in phenotypically weakly complemented lines is shown in Supplementary Figure S3.

The sulfate content in the selected lines revealed that SAL1 can rescue the low-sulfate phenotype of sal1 in all compartments, even when it is expressed in the cytosol or the nucleus alone (Figure 6A). However, in weakly expressing lines, complementation of low sulfate levels only occurred when SAL1 was expressed in the nucleus, mitochondria, or chloroplasts, but not when SAL1 was present in the cytosol (constructs II and VI) (Supplementary Figure S3).

As sulfate is required for the sulfation of GSLs, we additionally measured the accumulation of indolic (IG) and aliphatic (AG) GSLs and their desulfo-precursors. The desulfo-GSLs (ds-GSLs), which accumulate to high levels in sal1, were fully complemented by SAL1 expression in the nucleus, mitochondria, and by the native SAL1 construct. However, the cytosolic and chloroplastic expression of SAL1 only led to a 50% decrease in the accumulation of ds-GSL, which could not restore the wild-type level of ds-GSL (Figure 6B). In contrast to ds-GSLs, we observed less phenotypic variation among the SAL1-expressing lines for AGs and IGs. The production of AGs was fully complemented in all lines, with the exception of line VI in which the low AG level only slightly increased (Figure 6D). Conversely, the production of IGs was fully complemented only when SAL1 was targeted to the mitochondria (Figure 6C).

We further characterized the best-complemented lines in Figure 6 by quantifying their leaf phenotypes in more detail. We did comparative analysis of the following transgenic lines: 1) plants expressing SAL1 in the compartments in which PAP is proposed to act (nucleus, cytosol, or both) compared to the wild type and 2) plants expressing SAL1 in a single organelle (chloroplast or the mitochondria) compared to both organelles where SAL1 is normally found (wild type). Targeting SAL1 to both the cytosol and the nucleus (SAL1_VI_6) showed an additive effect compared to either the nucleus (SAL1_I 26) or the cytosol (SAL1_II_1) alone in restoring leaf area to wild-type levels (Figures 6, 7). In contrast, SAL1 in either the cytosol or the nucleus alone was similarly effective as SAL1 in both the cytosol and the nucleus in rescuing rosette compactness (Figure 7). In the second set of comparisons, targeting SAL1 to chloroplasts alone (SAL1_III 53) only partially rescued leaf area and rosette compactness. Similarly, targeting SAL1 to the mitochondria (SAL1_IV) did not completely restore leaf area, although it rescued rosette compactness to wild-type levels (Figure 7). Taken together, these results suggest that PAP most likely exerts its effects in both the cytosol and the nucleus, albeit unequally, and that the presence of SAL1 in both chloroplasts and the mitochondria is required for PAP homeostasis.

The unequal complementation obtained with the different constructs (Figures 3–7) was striking and unexpected, given that SAL1 was driven by a strong promoter that, in theory, should drive overexpression in all constructs. Therefore, we first examined for any possible relationship between the different phenotypes and SAL1 mRNA levels in lines of the different constructs. SAL1 overexpression at 5- to 10-fold of the wild-type levels was sufficient to completely restore the PAP levels in SAL1_IV 26, but not in lines of the other constructs such as SAL1_I 53, SAL1_II 2, SAL1_III 53, and SAL1_VI 5 (Figures 4, 5). Similarly, SAL1 expression at levels comparable to those of wild-type Col-0 restored plant biomass and significantly decreased the PAP levels only in construct IV, but not in lines of the other constructs (Figures 4, 5).

To address the possibility that the mRNA levels of transgenic SAL1 are uncoupled from the protein levels of SAL1 in the different constructs, we first compared selected lines with similar SAL1 mRNA expression levels (∼10-fold higher than that of Col-0) irrespective of their targeting. The different constructs showed significantly different degrees of leaf complementation despite their similar levels of SAL1 mRNA (Figure 8). Interestingly, the variance in leaf phenotype complementation appeared linked to substantial variation in the protein levels of SAL1, with the Western blot of lines SAL1_III 16 and 28 in particular showing much weaker SAL1 protein bands compared to SAL1_II 3 and SAL1_VI 6 (Figure 9B). This was further confirmed by comparing the protein and mRNA levels of SAL1 in the best-complemented lines of each construct regardless of their SAL1 mRNA levels (Figure 9A). Although the SAL1_I 26, III 53, and VI 6 lines all expressed SAL1 mRNA at 10- to 15-fold that of the wild type, their SAL1 protein levels varied substantially, with III 53 showing very low SAL1 protein abundance. SAL1_II 1 had similar SAL1 protein abundance to III 53 despite overexpressing SAL1 mRNA to a greater extent (40- vs. 10-fold, respectively). Similarly, SAL1_V 13 accumulated substantially more SAL1 protein than SAL1 IV 8 despite overexpressing SAL1 mRNA to a lesser degree (60 vs. 80-fold). Therefore, instead of SAL1 mRNA being the primary determinant of complementation, targeting SAL1 to different subcellular compartments may lead to different levels of the SAL1 protein in vivo through unknown mechanism(s), thus causing unequal complementation.

Seeds of A. thaliana ecotype Col-0, the T-DNA insertion mutant, and complemented transgenic plants were sown on soil or culture plates containing 1/2 Murashige and Skoog medium. The seeds were stratified for 2–3 days in the dark, and the plants were cultivated under short-day (8-h light and 16-h dark) conditions with an average photon flux density of 100–150 μmol m⁻² s⁻¹. White light was provided by Fluora L58W/77 fluorescent tubes. The temperature was kept at 22°C during the light period and 18°C during the dark period. The relative humidity was ∼40%.

The homozygous mutant sal1 line (At5g63980, SALK_020882, and SALK_005741) was identified and the insertion position of the T-DNA in the target gene was confirmed by sequencing previously (Ashykhmina et al., 2019). In this manuscript, we present data on the analysis and complementation of SALK_020882 line, named here also as sal1.

To achieve the compartment-specific accumulation of SAL1, we generated six chimeric constructs, as shown in Supplementary Figure S1. The SAL1tr backbone was amplified from Arabidopsis cDNA and fused with different DNA pieces using fusion PCR, as described in Section 2.

In our cloning procedure, all six constructs were inserted into the entry pDONOR207 vector. pDONOR207 vectors containing six different SAL1s were recombined with either the binary vector pGWB5 (for GFP localization studies) or pGWB2 (for the complementation of sal1) under the control of the 35S cauliflower mosaic virus promoter. All constructs of interest were transformed into Arabidopsis mesophyll protoplasts or root suspension cells.

To confirm the subcellular localization of SAL1 fusion proteins experimentally, Arabidopsis mesophyll protoplasts and root cell suspension cultures were transformed with Agrobacterium carrying constructs encoding SAL1_I–VI fused to the GFP at the C-terminus. Transfection of Arabidopsis mesophyll protoplasts was performed as described by Yoo et al. (2007) using 20–40 μg of plasmid DNA. Transformation of Arabidopsis root suspension cells was performed as described previously (Berger et al., 2007).

For staining with 4′,6-diamidino-2-phenylindol (DAPI), Arabidopsis cells were incubated in 2 μg ml⁻¹ (w/v) solution of DAPI for 5 min in the dark, followed by one to two times rinsing with cell culture media (Berger et al., 2007). For MitoTracker staining, the cells were removed from the medium and exposed in 500 nmol concentrated MitoTracker dye for 45–60 min. Cells were rinsed several times with cell culture media prior to imaging.

The GFP expression patterns in dark-grown cultured Arabidopsis or protoplasts from mesophyll tissue were recorded by confocal laser scanning microscopy (Zeiss, Jena, Germany). Images were acquired as z-series with a 1- to 3-μm interval with 25 frames using a Zeiss LSM510 confocal laser scanning microscope. Green fluorescent protein was visualized with a 488-nm laser and a band-pass (BP) 500–530 filter and the MitoTracker stain with a 543-nm laser and a BP 565–615 filter. For the visualization of DAPI, we used ultraviolet light and a BP 385–470 filter.

Co-localization of the GFP signal with chlorophyll autofluorescence indicated chloroplastidic localization. A co-labeling of GFP with the DNA fluorescent stain DAPI was interpreted as a nuclear localization, whereas a co-labeling of GFP with the MitoTracker indicated a mitochondrial localization.

Results were documented with Discus and Zeiss software. Images were processed using Photoshop CS3 (Adobe Systems, San Jose, CA, USA). Adjustment of levels was applied to the SAL1-V construct to make weak GFP signals better visible in protoplasts.

To complement sal1, we utilized constructs SAL1 I to VI without GFP and generated stable Arabidopsis transgenic lines. All six SAL1 constructs were recombined from pDONR207 into the Gateway destination pGWB2 vector (35S cauliflower mosaic virus promoter), and correctness of constructs was verified by sequencing. SAL1 I to VI were delivered to sal1 Arabidopsis plants by Agrobacterium-mediated transformation, and positive transformants were selected using kanamycin. The homozygous SAL1ox lines I to VI were isolated following segregation analysis of populations.

Individual Arabidopsis plants were grown on MS agar media for 4 weeks as described above. To measure the shoot fresh weight of complemented sal1 mutants, the aboveground rosettes were excised at the hypocotyl and individual weight was recorded. At least nine seedlings per construct were analyzed.

To measure the transcript levels in the wild type and the different complemented mutant plants, total RNA was isolated from leaves, cDNA was reversely transcribed, and reverse transcription quantitative PCR (RT-qPCR) was performed as described previously (Gigolashvili et al., 2009). The relative quantification of the expression levels was performed using the comparative delta C _t method, and the calculated relative expression values were normalized to Actin2 and compared with the expression level in wild-type plants (Col-0 = 1).

GSLs and their desulfo-precursors were extracted from the lyophilized plant material as reported previously (Ashykhmina et al., 2019). Sulfate was detected as previously reported in Mugford et al. (2009).

Plant material was collected and frozen in liquid nitrogen. The extraction and chromatographic detection of PAP were performed as reported previously (Ashykhmina et al., 2019).

Individual Arabidopsis plants were grown on separate soil-filled pots for 25 days under long-day conditions (18-h light and 6-h darkness at ∼100 µmol photons m⁻² s⁻¹, 22°C/20°C day/night cycle). For imaging, the aboveground rosettes were excised at the hypocotyl with sharp scissors and immediately photographed using an in-house imaging platform comprising a Canon DSLR camera and controlled fluorescent lighting. The images were analyzed in ImageJ to quantify the leaf area (area of green pixels) and the rosette compactness (ratio of leaf area relative to the convex hull) (Vanhaeren et al., 2015).

Approximately 100 mg leaf tissue harvested from 25-day-old soil-grown Arabidopsis plants was snap-frozen in liquid nitrogen, ground to fine powder using a ball mill at 20 Hz for 1 min (Retsch, Haan, Germany), and then resuspended in 300 µl RIPA extraction buffer (1% NP-40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate, and 10% glycerol) supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF). Cellular debris was removed by centrifugation (14,000 × g for 10 min at 4°C) and the supernatant moved to a fresh tube. Total protein was quantified using the Bradford assay. For SDS-PAGE and Western blotting, 20 µg total protein was loaded per sample onto 12% Mini-PROTEAN® TGX™ Precast Gels (Bio-Rad, Hercules, CA, United States) and resolved at 200 V for 30 min. The proteins were then transferred unto PVDF membranes using the Trans-Blot Turbo Transfer System (Bio-Rad). The membranes were incubated in blocking solution (5% milk, 1 × PBS, and 0.05% Tween) for 1 h at room temperature. This was followed by an anti-SAL1 antibody (Agrisera AS07 256; 1:1,000 dilution in 1% milk, 1 × PBS, and 0.05% Tween) overnight at 4°C and a secondary antibody (ECL^TM Donkey anti-rabbit IgG, horseradish peroxidase linked; GE Healthcare, Chicago, IL, United States) for 1 h at room temperature, with three 15-min washes (1 × PBS and 0.05% Tween) after every incubation step. The blots were developed using ECL imaging solution (Clarity Western Peroxide Reagent and Clarity Western Luminol/Enhancer Reagent; Bio-Rad). Chemiluminescence was visualized using the ChemiDoc system (Bio-Rad). To visualize all proteins present on the blot, the membrane was incubated for 5 min in Ponceau S solution (Sigma-Aldrich, St. Louis, MO, USA) and washed with distilled water to remove background staining.

Comparison of means was performed to determine statistical significance using a two-sample Student’s t-test or an ANOVA (Tukey’s test). Constant variance and normal distribution of data were verified prior to statistical analysis.

In the first phase of the project, comparison to wild type with repeated t-tests seemed to be the most accessible method of choice, as we were most interested in whether an individual line was not significantly different from the wild type (indicating complete complementation) rather than possible differences between certain lines (partial complementation). Prior to the t-test, a two-sample F-test for variances was implemented to check for scedasticity, shown in Figures 4–6. If the variances differed, a heteroscedastic two-tailed t-test was performed, while equal variances led to the choice of a homoscedastic t-test. For Figures 7, 8, homogeneity of the variances was verified using the Levene’s test for equality of variances in SPSS Statistics version 27 (IBM, Armonk, NY, USA). All datasets passed the test for equal variances, returning significance values between 0.3 and 0.6 (significance values higher than 0.05 indicate equal variance). The datasets were subsequently analysed by ANOVA.