Arabidopsis seeds were germinated vertically on 0.5× Linsmaier and Skoog (LS) medium containing 1.5% (w/v) sucrose and solidified with 0.8% (w/v) gellan gum (Gelrite) and transferred at 1–2 weeks after germination to pots containing Fafard Super Fine Germinating Mix (Sun Gro Horticulture Inc., Agawam, MA, USA) supplemented with 0.5× Miracle-Gro All Purpose Plant Food (The Scotts Miracle-Gro Co., Marysville, OH, USA). Plants were grown at 22°C (for seedlings) and 25°C (for soil-grown plants) under a long-day photoperiod (16-h light, 8-h dark cycle). The PDI9 (At2g32920) and PDI10 (At1g04980) genes were previously characterized (Lu and Christopher, 2008; Feldeverd et al., 2020). Seeds of WT Arabidopsis (Columbia, Col-0) and homozygous T-DNA insertion lines pdi9-1 (WiscDsLox445A08, progeny line CS864623), pdi10-1 (SALK_206219C), ire1a (SALK_002316), and ire1b (SAIL_238_F07) were obtained from the Arabidopsis Biological Resource Center (ABRC). The thorough characterization of the T-DNA mapping and verification of the homozygosity for the pdi9-1, pdi9-2, pdi10-1, and pdi9–pdi10 mutants were described in Feldeverd et al. (2020). The IRE1A (AT2G17520) and IRE1B (AT5G24360) genes contain seven and six exons, respectively. The mapping details for ire1a and ire1b were also reported by Lu and Christopher (2008) and Deng et al., (2011), respectively. The ire1a–ire1b double mutant was created by crossing ire1a × ire1b, and double homozygous mutant progeny in the F3 generation were confirmed by PCR. Genotyping of the T-DNA insertion mutants was done by extracting genomic DNA (Klimyuk et al., 1993) and PCR verification (MyFi Mix; Bioline) using T-DNA- and gene-specific primers ( Supplementary Figure S9 , primer table). The T-DNA insertion for ire1a has been mapped to intron 5, and that for ire1b is located in intron 4 ( Supplementary Figure S10 ). Genotyping at the IRE1A locus was performed using primers SALK_002316_ire1a-3-RP and SALK_002316_ire1a-3-LP to detect the WT allele and primers LBa1 and SALK_002316_ire1a-3-RP to detect ire1a. Genotyping at the IRE1B locus was performed using primers IRE1B-Int2-F and IRE1B-Int4-R to detect the WT allele and primers SAIL_LB2 and SAIL_238_F07_RP to detect ire1b. The ire1A–ire1B double-mutant was verified by PCR (35 cycles of 98°C for 15 s, 60°C for 30 s, and 72°C for 1 min), as shown in Supplementary Figure S10 .

The GFP referred to in all of the experiments is eGFP (S65T), which is the main source construct for GFP as described (Yuen et al., 2013). To generate the GFP : IRE1A construct, the CaMV 35S promoter was PCR-amplified from pCAMBIA1302 with primers engineered with KpnI and XhoI restriction sites and was ligated into the corresponding sites of pBluescript KS+. The IRE1A signal peptide, including a portion of the 5′-UTR, was PCR-amplified from Col-0 genomic DNA and ligated into XhoI and ClaI restriction sites. GFP was then PCR-amplified from the previously reported construct, PDI9:GFP-KDEL (Yuen et al., 2013), using primers engineered with ClaI and XmaI restriction sites. The mature polypeptide region of IRE1A (At2g17520) was PCR-amplified from Col-0 genomic DNA using primers engineered with XmaI and BamHI restriction sites and inserted into the intermediate plasmid from above. The nopaline synthase (nos) 3′-UTR fragment was PCR-amplified from the previously cloned GFP construct (Yuen et al., 2013) with flanking BamHI and NotI restriction sites and inserted into the intermediate plasmid to yield the final construct, GFP : IRE1A.

To generate the GFP : IRE1B construct, a portion of the IRE1B 5’UTR (At5g24360), including the signal peptide sequence, was PCR-amplified from Col-0 genomic DNA and ligated into the XhoI and ClaI sites in the 35S promoter-pBluescript KS+ intermediate plasmid (described above). The GFP coding sequence was PCR-amplified from the previously cloned GFP vector (Yuen et al., 2013) using primers engineered with ClaI and XmaI sites and ligated into the respective sites in the intermediate plasmid from above. The IRE1B mature polypeptide region was PCR-amplified from Col-0 gDNA with XmaI and NotI restriction sites, fused to the nopaline synthase (nos) 3′-UTR derived from a previous construct (Yuen et al., 2013) with flanking NotI and SacI restriction sites, and ligated into the intermediate plasmid to yield the final GFP : IRE1B construct.

For generation of the PDI9:mCherry-KDEL construct, the previously described PDI9:GFP-KDEL (Yuen et al., 2013) was digested using two subsequent double digestion reactions with XmaI/BstEII and XmaI/BamHI, followed by ligation with the mCherry insert amplified from the ER-mCherry marker with corresponding XmaI and BamHI restriction sites. For the construction of the mCherry control vector, the mCherry fragment was amplified from PDI9:mCherry-KDEL using primers engineered with XhoI and BamHI restriction sites and ligated into the plasmid template described above containing the 35S promoter. The 3′-UTR nos terminator sequence was amplified from the main GFP construct and inserted between BamHI and NotI restriction sites to yield the final construct.

The FRET positive control dual fluorescent protein fusion construct, pBL(35S:GFP-mCherry) (referred to hereon as GFP:mCherry), was created by replacing the γTIP coding sequence cassette between the SpeI and BamHI sites of plasmid pBL(35S:γTIP-mCherry) (Yuen et al., 2013) with an SpeI/BamHI fragment encoding GFP. The GFP coding sequence was generated by PCR using the primers GFP(Spe)F and GFP(BamHIXmaI)R ( Supplementary Figure S9 ), using the plasmid pBL-GFP (Yuen et al., 2013) as a template.

The construct encoding the hemagglutinin (HA) epitope-tagged PDI9 protein, HA : PDI9, was generated for co-I.P. in transfected protoplast samples using pBluescript KS+. The nopaline synthase (nos) 3′-UTR fragment was PCR-amplified from the PDI9:mCherry-KDEL construct with flanking NotI and SacI restriction sites and ligated into the respective restriction sites in the empty vector. The CaMV 35S promoter sequence was PCR-amplified from the plasmid, PDI9:mCherry-KDEL, using primers engineered with EcoRI and KpnI and ligated into the intermediate plasmid from above. The PDI9 signal peptide sequence (including a portion of the 5’ UTR) and the HA tag were PCR-amplified from Col-0 genomic DNA using primers engineered with KpnI and BamHI restriction sites ( Supplementary Figure S9 ) and inserted into the corresponding restriction sites of the developing vector. The mature polypeptide region of PDI9 was then amplified from Col-0 genomic DNA with flanking BamHI and NotI restriction sites and inserted into the respective sites of the intermediate plasmid from above to yield the final construct, HA : PDI9.

The PDI9-promoter:mCherry-KDEL reporter construct was generated by PCR-amplifying the 5′-flanking sequences approximately 2.7 kb upstream of the start codons of PDI9 (containing the promoter regions) from Arabidopsis Col-0 genomic, using primers engineered with KpnI and XhoI restriction sites ( Supplementary Figure S9 ), and inserted into the respective sites in the mCherry control vector described above. Generation of the 35S::bZIP60 intron:GFP construct was previously described by Carrillo and Christopher (2022). To confirm transfection efficiencies between protoplast cells expressing varying fluorescence intensities of the PDI9-promoter:mCherry-KDEL construct, cells were cotransfected with the GFP control.

Protoplast isolation and transfection were performed using the Tape-Arabidopsis Sandwich protocol, as described by Wu et al. (2009) and as modified by Carrillo and Christopher (2022). The transfected protoplasts were incubated in the light at room temperature for 16–18 h before being examined using a Leica TCS SP8 laser scanning confocal microscope at the Biological Electron Microscope Facility (University of Hawaii at Manoa, Honolulu, HI, USA). The excitation/emission filters utilized for fluorescence detection were 488/505–525 nm for GFP and 543/585–615 nm for mCherry. For chemical induction of ER stress and the UPR in protoplasts, samples were inoculated with 2 mM dithiothreitol (DTT) from a 1-M DTT stock (Roche Applied Science, Indianapolis, IN, USA) and incubated for 3 h prior to visualization by scanning confocal microscopy.

FLIM for FRET detection was performed using a Leica TCS SP8 confocal laser scanning microscope at the Biological Electron Microscope Facility (University of Hawaii at Manoa, Honolulu, HI). A diode pulsed laser was used to excite GFP, and the emission from 488 nm was collected in 520 pixel × 520 pixel mode. The GFP lifetime was determined from the fluorescence decay curve per pixel, using a photon count rate of 1 photon per laser pulse. Photon counts were accumulated by line repetition to optimize the setting for live protoplasts. The average lifetime in the selected region(s) of interest (ROI) within each of the protoplast cells examined was analyzed with the FLIM pixels binned to select an optimal total photon count > 1,000. The ROIs containing GFP fluorescence (GFP : IRE1A, GFP : IRE1B, GFP, or GFP:mCherry) in all single- and dual-expressed (with PDI9:mCherry-KDEL or ER:mCherry) protoplast samples were selected for average GFP lifetime values. FRET efficiency (E) was calculated for each ROI using the following formula:

Where τ _i is the mean lifetime for that ROI, and τ₀ is the average lifetime of all ROIs in GFP : IRE1A and GFP : IRE1B single-transfected samples (not expressing detectable mCherry). At least 20 cells were analyzed for each sample, and average GFP lifetime values were obtained over two independent experiments.

Protoplast isolation and transfections were performed using the Tape-Arabidopsis Sandwich protocol (described above) and modified for co-I.P. using a protocol adapted from Shan et al. (2008) and Carrillo et al. (2021). Protoplasts isolated from 4-week-old Col-0 plants were cotransfected with the constructs HAPDI9 and GFP : IRE1A or GFP : IRE1B. Single-construct transfections were prepared as controls. Sample transfection volumes were scaled fivefold to account for a 1-mL final transfection volume per sample (i.e., 150 µg of pDNA per construct mixed with 1 mL of protoplasts in MMg solution and an equal volume of 40% PEG). All subsequent washes were also scaled up fivefold (i.e., 5 mL of washes in W5 solution), and the transfected protoplasts were incubated at RT for 16–18 h prior to protein extraction.

The 1-mL transfected protoplast samples were centrifuged at 100×g for 5 min, and the W5 solution was discarded. The pelleted protoplasts were vortexed vigorously for 30 s with 150 µL of protein extraction buffer (nonreducing 10 mM HEPES at pH 7.5, 100 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% Triton X-100, and 1 mM PMSF) and centrifuged at 13,000 rpm for 10 min at 4°C. The supernatant was collected, and 100 µg of total protein per lysate was incubated with 50 µL of anti-HA magnetic beads for 3 h at 4°C on a rotary shaker (20 rpm). The beads were collected on a magnetic stand using the Pierce HA Tag Magnetic I.P./Co-I.P. kit (Thermo Fisher Scientific, Waltham, MA, USA). Proteins isolated by co-I.P. were separated by reducing SDS-PAGE and analyzed by Western blot with a monoclonal anti-HA antibody (Sigma-Aldrich, St. Louis, MO, USA) or polyclonal anti-GFP antibody (Molecular Probes, San Jose, CA, USA) at 1:2,000 dilution, with a secondary goat antimouse (Advansta Inc., San Jose, CA, USA) or goat-antirabbit (Advansta Inc., San Jose, CA, USA) IgG HRP conjugate antibody, respectfully.

Constructs for heterologous expression of Arabidopsis PDI9 and IRE1A or IRE1B in Escherichia coli were generated in the vector pETDUET-1 (Novagen, 2011), designed for the simultaneous coexpression of the two genes. The signal peptides were omitted for PDI9, IRE1A, and IRE1B cDNAs to enable protein expression in the cytoplasm of E. coli. The mature protein coding sequence for PDI9 cDNA was first amplified by RT-PCR from Arabidopsis Col-0 WT total RNA (isolated using the NucleoSpin RNA Plant Kit, Macherey-Nagel Inc., Duren, Germany), using primers engineered with BamHI and NotI restriction sites and inserted into corresponding sites in the pETDUET-1 vector MCS1, fusing it in-frame to the built-in 6×His tag. The coding sequence for the N-terminal lumenal (ER-sensor) domain (LD) of the IRE1A cDNA was amplified by RT-PCR using primers containing KpnI and PacI restriction sites and an internal Strep-tag. This truncated IRE1A fragment was ligated into the same restriction sites in the pETDUET MCS2, thus fusing it in the frame to the Strep-tag to generate the final HisPDI9-StrepIRE1A_LD construct.

To make the HisPDI9-StrepIRE1B_LD construct, the N-terminal LD of the IRE1B cDNA was amplified by RT-PCR as described above for IRE1A. RT-PCR used modified primers containing a Strep-tag and KpnI and PacI restriction sites. The digested fragment was ligated into KpnI and PacI sites in the MCS2 of the pETDUET-1 containing the PDI9 cDNA. This produced the final HisPDI9-StrepIRE1B_LD construct. Single-insert vectors were generated as controls that contained only the HisPDI9 StrepIRE1A_LD or StrepIRE1B_LD coding regions.

The second cysteine residue in the two thioredoxin active sites of PDI9 (positions 63 and 195) and the two cysteine residues in the ER lumenal domain of IRE1A (positions 233 and 257) and IRE1B (positions 107 and 222) were individually mutated to alanine (Kersteen et al., 2005). The previously designed constructs described above were used as template plasmids for generating the respective site-specific mutations for use in both in vivo protoplast fluorescence lifetime imaging microscopy-fluorescence resonance energy transfer (FLIM-FRET) and in vitro coimmunoprecipitation assays. Primers were designed to contain the desired mutation, and mutagenesis was confirmed by DNA sequencing (GeneWiz Inc., Plainfield, NJ, USA). A summary of primers used in this study is shown in Supplementary Figure S9 . For in vitro co-I.P. assays, the following mutants were generated in the pETDUET-1 expression vector: HisPDI9(C63A)-StrepIRE1A_LD, HisPDI9-StrepIRE1A_LD (C233A), HisPDI9-StrepIRE1A_LD (C257A), HisPDI9(C63A)-StrepIRE1B_LD, HisPDI9(C195A)-StrepIRE1B_LD, HisPDI9-StrepIRE1B_LD(C107A), and HisPDI9-StrepIRE1B_LD(C222A), including their associated single-vector controls. For in vivo protoplast expression assays, the following mutants were generated in the pBluescript KS+ expression vector: PDI9(C63A):mCherry-KDEL, PDI9(C195A):mCherry-KDEL, GFP : IRE1A(C233A), GFP : IRE1A(C257A), GFP : IRE1B(C107A), and GFP : IRE1B(C222A).

To promote the formation of disulfide bonds in the cytoplasm of E. coli, the expression host Rosetta-gami 2 (DE3) was used, which carries the glutathione reductase (gor) and thioredoxin reductase (trxB) mutations (Novagen, 2011). The E. coli Rosetta-gami 2 (DE3) was transformed with the following pETDUET-1-cloned plasmids: HisPDI9-StrepIRE1A_LD, HisPDI9-StrepIRE1B_LD, HisPDI9, StrepIRE1A_LD, and StrepIRE1B_LD, including all aforementioned mutant constructs containing the specific cysteine to alanine substitution. Transformed cells were grown to approximately OD600nm = 0.5–0.6 and induced with 0.2 mM IPTG for 3 h at 30°C. After induction, cells were harvested for extraction of soluble proteins under nonreducing conditions using the BugBuster Protein Extraction Reagent (EMD Millipore, Burlington, MA, USA) supplemented with phenylmethylsulfonyl fluoride (PMSF), benzonase nuclease, and rLysozyme (EMD Millipore, Burlington, MA, USA).

Approximately 100 µg of total protein from each of the dual-expressed and single-vector lysates described above were incubated with 100 µL of the nickel-nitrilotriacetic acid (Ni-NTA) His Bind resin slurry (EMD Millipore) under nonreducing conditions and incubated at 4°C on a rotary shaker (20 rpm) for 3 h. The protein–resin mixture was loaded onto a 1-mL Pierce Spin Column (Thermo Fisher Scientific) and centrifuged for 1 min at 1,000×g to collect the flow through. The resin was washed twice with 400 µL 1× Ni-NTA wash buffer and eluted with 30 µL 1× Ni-NTA elution buffer. A total of three elutions were performed with a 1-min incubation with the protein–resin mixture. All fractions (unbound flow through, washes, and elutions) were analyzed by SDS-PAGE under reducing conditions as previously described (Yuen et al., 2017).

A 20-µL aliquot of each fraction collected from the Ni-NTA His binding affinity columns was loaded on 10% polyacrylamide gels. Proteins were separated by SDS-PAGE and electrotransferred to Amersham Protran nitrocellulose membranes (GE Healthcare Life Sciences, Milwaukee, WI, USA). Immunoblot analyses were performed using a monoclonal anti-6× His antibody (Sigma-Aldrich, St. Louis, MO, USA) at 1:2,000 dilution or monoclonal anti-Strep antibody (EMD Millipore, Burlington, MA, USA) at 1:1,000 dilution and a horse-radish peroxidase (HRP)-conjugated antimouse IgG secondary antibody (Advansta Inc., San Jose, CA, USA) at 1:20,000 or 1:10,000 dilutions, respectively. Chemiluminescent detection of HRP was done using the Advansta WesternBright ECL HRP substrate (Advansta Inc., USA).

To analyze transcript levels of UPR marker genes, quantitative PCR (qPCR) was done on RNA from Arabidopsis protoplasts from the WT, pdi9-1, and pdi9–pdi10 genotypes, using a protocol adapted from Carrillo et al. (2022). Total RNA from 4-week-old Arabidopsis protoplasts (n = 3) was extracted using the NucleoSpin RNA Plant, Mini Kit (Macherey-Nagel, Duren, Germany). Protoplasts from each genotype were resuspended in RAP lysis buffer with mercaptoethanol (1% v/v) and immediately vortexed to mix. Respective samples treated with 2 mM DTT were incubated for 3 h prior to the harvesting of cells and subsequent RNA extraction.

The qPCR primers were designed using Primer3 (v. 0.4.0) to identify amplicons around 200 bp with a melting point of around 63°C. The synthesis of first-strand cDNA qPCR was prepared at the University of Hawaii Cancer Center Genomics and Bioinformatics Shared Resource (UHCC GBSR) facility. For qPCR, cDNAs were diluted to 10 ng/µL using water, and 1 µL (10 ng) per qPCR reaction was used. PCR reactions were done under the following cycling conditions: 15 s denaturation at 94°C, 30 s annealing at 60°C, and 30 s extension at 72°C for a total of 40 cycles. Gene expression analysis was done using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific), and each 10 µL reaction consisted of 5 µL of Master Mix, 0.41 µL of each forward and reverse primer (10 µM), 3.18 µL of water, and 1 µL of cDNA. A melting curve was generated to verify sequence-specific amplification of PCR products. The genes of interest were prepared in triplicate, and the expression level was determined using cycle threshold (Ct) values with a standard curve after normalization with the housekeeping gene, actin. Fold change was calculated by this method for each gene, and the data are shown as mean with standard deviations. Significance was calculated using one-way analysis of variance (ANOVA) followed by Tukey’s test to compare differences between multiple means and determine significance at p-values< 0.01 and< 0.05. Primers used for amplification of PDI9 (PDI9-F, PDI9-R), bZIP60t (bZIP60-F, bZIP60-R), bZIP60s (bZIP60-F, bZIP60s-R), BiP2 (BiP2-F, BiP2-R), and actin (Actin2-F, Actin2-R) transcripts are listed in Supplementary Figure S9 . Prior to qPCR, RNA samples were analyzed on denaturing RNA gels and by standard RT-PCR to verify the RNA quality and expression of genes in all of the samples.

To quantify total fluorescence levels in Arabidopsis protoplasts from each genotype transiently expressing either the PDI9-promoter:mCherry or 35S::bZIP60 intron:GFP, protoplasts were transfected as described above, and DTT was added to a final concentration of 2 mM where indicated. Protoplasts were imaged on a single plane at ×40 magnification (512 by 512 frame resolution) using a Leica TCS SP8 laser scanning confocal microscope (Leica Inc., Teaneck, NJ, USA). All images were imported as Tif files in ImageJ software (NIH) for quantification of total cell fluorescence. Individual cells were selected and outlined using the “freehand selection” tool and measured for area, integrated density, and mean gray value. Background intensity values were measured by selecting five random areas in the image that do not contain cell fluorescence. Corrected total cell fluorescence (CTCF) values were calculated for each cell using the following equation:

an CTCF values for mCherry and GFP lifetimes were plotted for each reporter construct expressed in the protoplasts of the Col-0, pdi9-1, pdi9-2, and pdi9–pdi10 genotypes with 30–50 cells analyzed for each construct in two independent experiments. Significance was calculated using the two-way ANOVA followed by the Tukey’s test to compare differences between multiple means and determine significance at p-values< 0.01.

Of the Arabidopsis PDI family, the PDI9 protein sequence is most similar to the human ortholog, PDIA6, sharing 42% identity and 61% similarity ( Supplementary Figure S1 ). Since PDIA6 interacts with IRE1A in the ER of mammals (Eletto et al., 2014), we tested if PDI9 also interacted with IRE1A and IRE1B in the ER of Arabidopsis. First, we examined whether PDI9 colocalizes with IRE1A and IRE1B in the ER, as PDI9 is well-documented to be a resident of the ER lumen (Yuen et al., 2013; Feldeverd et al., 2020). Colocalization was tested by transiently expressing the corresponding PDI9 and IRE1 proteins fused to either mCherry or GFP reporters, respectively, in mesophyll protoplasts ( Figure 1 ). When GFP : IRE1A or GFP : IRE1B were coexpressed with PDI9:mCherry-KDEL, and examined by confocal laser scanning microscopy, significant colocalization of the fluorescent fusions was observed in the ER ( Figure 1 ). Localization patterns were similar when compared with the ER marker, ER:mCherry.

Since our transient expression system correctly detects PDI9 with IRE1A and IRE1B in the ER, the next step was to determine if PDI9 interacted with either IRE1A or IRE1B. The following two methods were chosen for this determination: (a) in vivo fluorescence resonance energy transfer (FRET) analysis using fluorescence lifetime imaging microscopy (FLIM) of reporter fusions; and (b) in vitro coimmunoprecipitation. For the FLIM-FRET, the PDI9:mCherry-KDEL and GFP : IRE1 fusions were dually expressed in Arabidopsis protoplasts, and GFP lifetime decay values were measured in regions of colocalization relative to that of various control samples, which included expressing GFP : IRE1 alone and GFP : IRE1 with mCherry alone. In addition, the dimer, GFP:mCherry direct fusion, was created for this study as a positive control to indicate maximum GFP quenching for comparison with the test samples.

The mean lifetime of GFP fluorescence (GFP : IRE1A and GFP : IRE1B) in cells coexpressed with PDI9:mCherry-KDEL was significantly diminished relative to GFP : IRE1 control cells ( Figures 2A, B ). This reduction in GFP lifetime indicated that quenching occurred for the fluorescence of the GFP donor in the presence of PDI9-mCherry. Quenching of the donor lifetime was not observed when GFP : IRE1 fusions were coexpressed with the negative control, the mCherry construct (lacking PDI9) ( Figure 2B ). Furthermore, no statistical difference in GFP lifetime values was observed between cells coexpressing GFP : IRE1 fusions and PDI9:mCherry-KDEL and cells expressing the positive control fusion, GFP:mCherry. Therefore, in cells coexpressing either GFP : IRE1A or GFP : IRE1B with PDI9:mCherry-KDEL, the decreases in GFP lifetime values were due to the interaction of IRE1 (A and B) with PDI9.

FRET efficiency (E) values were calculated ( Figure 2B ) for each of the dual-expressed samples based on their mean GFP lifetime values relative to the GFP : IRE1 baseline values (single-transfected control samples). The FRET efficiency between the PDI9:mCherry-KDEL and GFP : IRE1A donor:acceptor pair yielded 6.82%, while that for the PDI9:mCherry-KDEL and GFP : IRE1B pair yielded 5.97% ( Figure 2B ). These values were 1% to 1.8% less than, but comparable to, the FRET efficiency of the positive control, GFP:mCherry direct fusion, at 7.84%. A negative value was obtained for GFP samples coexpressed with mCherry alone, as the mean GFP lifetime values were slightly higher than the single-transfected controls ( Figure 2B ).

To further test if the interaction occurred between full-length IRE1 and PDI9 in vivo, anti-HA co-I.P. assays were conducted on extracts from Arabidopsis protoplasts coexpressing HA-tagged PDI9 (HA : PDI9) with either GFP : IRE1A or GFP : IRE1B ( Figure 2C ). Both HA : PDI9 and GFP : IRE1 signals were detected together in the eluates of protein extracts from cotransfected protoplasts containing HA : PDI9 coexpressed with either the GFP : IRE1A or GFP : IRE1B fusions ( Figure 2C ), relative to single vector controls ( Supplementary Figure S2 ). Bands of approximately 119 kDa and 125 kDa in size were detected correlating to GFP : IRE1A and GFP : IRE1B fusion proteins, respectively, using an anti-GFP antiserum on immunoblots of the eluates. The 48-kDa HA : PDI9 was also detected in the same fraction, suggesting corelease of GFP : IRE1A and GFP : IRE1B from the column and thus binding between the two putative interactors, which are the HA : PDI9 with the GFP : IRE1 fusions ( Figure 2C ). These results, together with the FLIM-FRET analysis ( Figures 2A, B ), indicated that PDI9 interacted with both IRE1A and IRE1B in vivo in Arabidopsis.

To further examine the mechanism of the PDI9-IRE1 protein–protein interactions, a series of co-I.P. experiments were conducted in E. coli. The recombinantly expressed WT HisPDI9 and the ER lumenal/sensor domains of IRE1A and IRE1B (fused to the Strep-tag) were used ( Figures 3A, B , 4A, B ). To obtain the ER lumenal/sensor domains, the mature polypeptide sequences of the IRE1A and IRE1B Strep fusions were truncated by removing the cytoplasmic TMD and kinase domains (referred to as StrepIRE1A_LD and StrepIRE1B_LD). This truncation also prevented protein aggregation when expressed in E. coli cells because the presence of the insoluble TMD shunted the foreign IRE protein into inclusion bodies. Expressing the specific ER lumenal/sensor domains overcame this problem.

Both the 36-kDa StrepIRE1A_LD and the 40-kDa StrepIRE1B_LD proteins were enriched in the HisPDI9 fractions that were eluted from the nickel column, as revealed using the anti-Strep antibody in the immunoblot analysis ( Figures 3B , 4B ). The wash-resistant presence of StrepIRE1A_LD and StrepIRE1B_LD only in the eluted HisPDI9 samples indicated the binding of IRE1A_LD and IRE1B_LD to PDI9. Based on the eluted band intensity, the interaction of IRE1A_LD with PDI9 ( Figure 3B ) was stronger than the interaction with IRE1B_LD ( Figure 4B ). A portion of the Strep protein alone present in the extract did not bind to HisPDI9 and was also detected in the flow-through ( Supplementary Figure S3 ). Strep was not detected in the eluted fractions when using extracts from single vector controls of the Strep peptide on the column, suggesting binding was specific to HisPDI9 ( Supplementary Figure S3 ). These results confirm that PDI9 interacted with the ER lumenal domains of IRE1A and IRE1B.

Given that PDI9 is a redox-active protein folding enzyme, we next sought to investigate the role of the active site cysteine residues of PDI9 in their interaction with IRE1 in Arabidopsis. To accomplish this goal, site-specific mutagenesis was used to modify these key cysteine residues in PDI9 and IRE1 to the redox-inactive residue, alanine. For example, the second cysteine (Cys-63) in the first thioredoxin active motif of PDI9(CGHC) was mutated to alanine (CGHA, designated C63A). In the second thioredoxin active site motif of PDI9, the second cysteine (Cys-195) was also mutated to alanine (CGHC to CGHA, designated C195A). The single mutant versions of PDI9 were re-tested for potential interaction using in vitro co-I.P. firstly ( Figures 3 , 4 ) and then in vivo FLIM-FRET analysis secondly ( Figure 5 ). Interestingly, the C63A mutation in the first redox-active site of PDI nearly abolished interaction with IRE1A in vitro, as observed by immunoblot analysis from in vitro co-I.P. of HisPDI9(C63A) with StrepIRE1A_LD ( Figure 3B ). However, for the second C195A mutation of PDI9, the binding between HisPDI9(C195A) and StrepIRE1A_LD remained robust. The StrepIRE1A_LD was detected in the coeluted fractions with HisPDI9(C195A) at an almost similar level with respect to that observed when incubated with WT HisPDI9 ( Figure 3B ).

The converse experiment was next conducted, in which the key cysteines of IRE1A were mutated (C233A and C257A), and these mutant proteins tested for co-I.P. with the WT PDI9 of the HisPDI9 fusion. The StrepIRE1A_LD(C233A) mutant exhibited diminished binding with HisPDI9 in vitro. Although the StrepIRE1ALD(C257A) mutant also exhibited diminished binding relative to WT StrepIRE1A_LD, we observed slightly more binding of the IRE1A(C257A) to HisPDI9 compared to the IRE1A(C233A) mutant ( Figure 3B ).

Although a similar alteration in the binding pattern was also observed between the PDI9(C63A) mutant and IRE1B both in vitro and in vivo, the effect of the mutation was weaker relative to that with IRE1A ( Figures 4A, B ). In addition, whereas the C107A mutation in IRE1B reduced its interaction with HisPDI9, the IRE1B(C222A) mutation had only a minor effect on the interaction with HisPDI9 ( Figures 4A, B ).

To complement the co-I.P. approach to defining protein–protein interactions, the same protein mutations were subjected to FLIM-FRET analysis via transient expression in mesophyll protoplasts. The WT PDI9 sequence and two PDI9 mutations, C195A and C63A, were fused to mCherry-KDEL and tested to interact with the WT IRE1A sequence fused to GFP and the single C233A and C257A mutations of IRE1A, also fused to GFP ( Figure 5 ). Likewise, the PDI9:mCherry-KDEL was tested to interact with the WT IRE1B fused to GFP and single mutant versions of IRE1B, C107A, and C222A fused to GFP ( Figure 5 ). The two negative controls consisted of GFP : IRE1A and GFP : IRE1B, each coexpressed with mCherry alone. The positive control consisted of the direct fusion, GFP:mCherry.

The GFP lifetimes ( Figure 5 ) for GFP : IRE1A and GFP : IRE1B coexpressed with mCherry alone were 2.49 ns and were comparable to GFP alone ( Figure 2 ), indicating no GFP quenching in these negative controls. Significant quenching of donor GFP fluorescence was observed as 7.6% decreased lifetimes, 2.29 ns and 2.31 ns ( Figure 5 ) when PDI9:Cherry-KDEL was coexpressed with either GFP : IRE1A or GFP : IRE1B, respectively. These quenching values were similar to the lower GFP fluorescence lifetime from the positive control GFP:mCherry ( Figure 5 , lane 5). The degree of quenching of the GFP donor fluorescence lifetime (2.42 ns) was decreased by half in protoplast cells coexpressing the mutant PDI9(C63A):mCherry-KDEL with GFP : IRE1A and PDI9(C63A):mCherry-KDEL with GFP : IRE1B ( Figure 5 , lanes 6 and 10). However, the lifetime values did not return to the same levels as the GFP : IRE1A control or when this control was dually expressed with the mCherry (no PDI9) control ( Figure 5 , lanes 1 and 2). Therefore, the C63A mutation in PDI9 decreased its affinities for IRE1A and IRE1B to intermediate levels. The C195A mutation in PDI9 ( Figure 5 , lanes 7 and 11) did not statistically alleviate the GFP lifetime quenching in the coexpressed pairs: PDI9(C195A):mCherry-KDEL with GFP : IRE1A; and PDI9(C195A):mCherry-KDEL with GFP : IRE1B.

We next examined the effects of single cysteine mutations in IRE1A ( Figure 5 , lanes 8 and 9) and IRE1B ( Figure 5 , lanes 12 and 13) on their ability to affect quenching of GFP lifetime fluorescence when paired with WT PDI9:mCherry-KDEL. The mutation C233A in IRE1A in the GFP : IRE1A(C233A) fusion when paired with PDI9:mCherry-KDEL pair (lane 8) significantly reduced the GFP fluorescence quenching. However, the C257A mutation in the fusion GFP : IRE1A(C257A) when paired with PDI9:mCherry-KDEL, did not statistically affect GFP fluorescence quenching. Similarly, the mutation C107A in IRE1B in the GFP : IRE1B(C107A) fusion, when paired with PDI9:mCherry-KDEL pair (lane 12), significantly reduced the GFP fluorescence quenching. Yet, the C222A mutation in the fusion GFP : IRE1B(C222A), when paired with PDI9:mCherry-KDEL, did not statistically affect GFP fluorescence quenching.

In summary, the FLIM-FRET results ( Figure 5 ) indicate the C63A mutation in PDI9 was more effective than the C195A mutation in disrupting the quenching of GFP fluorescence. Likewise, the C233A mutation in IRE1A was more effective than the C257A mutation in affecting GFP lifetime fluorescence in the fusion pairing. However, for IRE1B, the C107A mutation was more effective than the C222A mutation.

The intriguing evidence for a strong interaction of PDI9 with IRE1A and IRE1B suggests a potential role for PDI9 in the UPR signaling pathway. To test this hypothesis further, we examined the impact of the loss of the PDI9 gene family on UPR signaling by using a novel biosensor-reporter system that monitors UPR in living plant cells (Carrillo and Christopher, 2022). The intron from the bZIP60 locus is spliced out in a UPR-dependent fashion. This intron was incorporated into the 5′-end of the GFP gene, creating the 35S::bZIP60 intron:GFP construct. The bZIP60 intron disrupts GFP mRNA from being translated under non-UPR conditions. Upon activation of UPR, the IRE1 kinase/ribonuclease splices the intron from the GFP RNA, permitting translation, resulting in GFP fluorescence. In Figure 6A , the effects of the PDI9 gene family on the UPR were studied via monitoring bZIP60 intron splicing from the modified GFP in protoplasts of WT and the previously well-characterized single pdi9-1, pdi9-2, and double pdi9–pdi10 mutant backgrounds (Feldeverd et al., 2020). The ire1a–ire1b double mutant was included as a control to ensure the measurement of GFP fluorescence and UPR-dependent bZIP60 intron splicing via an IRE-based signal transduction mechanism ( Figure 6A ).

CTCF levels of GFP were quantified in representative protoplasts from WT and mutant genotypes transfected with the 35S::bZIP60 intron:GFP construct and treated with and without DTT ( Figure 6A ). Relative to WT protoplasts in the absence of DTT, significantly greater CTCF values were observed in protoplasts from the pdi9-1, pdi9-2, and pdi9–pdi10 genotypes. Cells were chemically treated with the ER stress inducer, DTT, under short-term (3 h) conditions to avoid secondary effects of DTT (Carrillo and Christopher, 2022). In cells treated with DTT, all genotypes exhibited significantly greater CTCF values relative to nonstressed conditions. Most notably, relative to WT, the pdi9–pdi10 double mutant exhibited nearly fivefold greater GFP fluorescence without DTT treatment and nearly threefold greater GFP fluorescence with DTT treatment ( Figure 6A ). CTCF levels were reduced to< 10% in the ire1a–ire1b double mutant (genotype confirmed in Supplementary Figure S10 ), indicating that GFP fluorescence and bZIP60 intron splicing were via the IRE-based signaling pathway ( Figure 6A ). There was no significant difference in transfection efficiency between the protoplasts expressing the 35S::bZIP60 intron:GFP reporter used in this assay in either genotypes or treatments, as indicated by the cotransfected mCherry and GFP alone controls ( Figure 6A ). Representative protoplast cells expressing the 35S::bZIP60 intron:GFP construct in the aforementioned genotypes and treatments are shown in Supplementary Figures S5 , S6 . These results indicate that the loss of PDI9 increases the activity of the UPR pathway and suggest that PDI9 attenuates the splicing of bZIP60 mRNA through IRE1 RNase activity under ER stress.

A second test for the role of PDI9 in UPR signaling involved the use of the PDI9-promoter:mCherry reporter. The PDI9 gene was previously shown to be transcriptionally induced by UPR via the bZIP60 transcription factor (Lu and Christopher, 2008). Therefore, the PDI9 promoter is a useful tool for measuring UPR signaling in WT, pdi9-1, pdi9-2, and pdi9–pdi10 mutant backgrounds and would complement the aforementioned data ( Figure 6A ). This PDI9 promoter was incorporated upstream of the mCherry reporter gene, creating the PDI9-promoter:mCherry reporter construct. To further investigate the modulatory role of PDI9 within the UPR, this construct was transiently expressed in protoplasts from WT, pdi9-1, pdi9-2, pdi9–pdi10, and ire1a–ire1b genotypes with and without DTT treatments ( Figure 6B ). The mCherry expression was an indication of PDI9 promoter activity, and CTCF intensities were quantified per protoplast.

Relative to WT, the pdi9-1, pdi9-2, and pdi9–pdi10 mutants exhibited significantly greater CTCF intensities in the absence of DTT ( Figure 6B ). The CTCF values were nearly threefold greater in the pdi9–pdi10 genotype without DTT. DTT treatment markedly induced mCherry fluorescence in all genotypes relative to the untreated protoplasts. However, no significant difference was observed in CTCF values between WT and either pdi9-1, pdi9-2, or pdi9–pdi10 mutants within the DTT treatments ( Figure 6B ). Therefore, the activation of the PDI9 promoter was upregulated under DTT-induced ER stress in all samples and also in the pdi9-1, pdi9-2, and pdi9–pdi10 mutants under normal, noninduced conditions. The activation of the PDI9 promoter was severely reduced in both treatments in the ire1a–ire1b mutant. We observed no difference in fluorescence of the GFP transfection control when cotransfected with the PDI9-promoter:mCherry reporter in all genotypes under normal and ER-stress-induced conditions ( Figure 6B ). Representative protoplast cells expressing the PDI9-promoter:mCherry construct in the aforementioned genotypes and treatments are shown in Supplementary Figures S7 , S8 . Thus, transfection efficiencies remained relatively constant between genotypes and treatments. These results further validate that the PDI9 promoter is induced under ER stress via IRE1-mediated signaling. Furthermore, these results are relatively consistent with the results in Figure 6A , in which the loss of the function of the PDI9 gene family leads to an increase in UPR-regulated gene expression mediated by IRE1.

As additional internal controls for the in vivo reporter assays done in protoplasts, an analysis of the effects of the loss of PDI9 on UPR-regulated gene expression was also measured in protoplasts by qPCR ( Supplementary Figure S4 ). Relative levels of RNAs from the UPR marker genes BiP2, PDI11, and bZIP60 (bZIP60t and bZIP60s transcript variants) were measured in the protoplasts (treated with and without DTT) from WT (Col-0), pdi9, and pdi9–pdi10 mutant backgrounds that were treated with DTT prior to qPCR analyses. All RNAs were higher in WT protoplasts exposed to the ER-stress inducer DTT, relative to untreated cells. However, relative to WT, the BiP2, PDI11, and bZIP60 mRNAs were significantly elevated in the pdi9 and pdi9–pdi10 mutants under both normal (untreated) and ER stress-induced (DTT-treated) conditions ( Supplementary Figure S4 ). The induction of RNA levels was greatest in the pdi9–pdi10 double mutant. The removal of PDI9 and PDI10 leads to a marked increase in the levels of several key UPR gene transcripts. These qPCR results are directly consistent with the results from the two reporter assays, which are the 35S::bZIP60 intron:GFP ( Figure 6A ) and PDI9-promoter:mCherry ( Figure 6B ) reporters. Taken together, these results indicate that PDI9 and PDI10 are negatively impacting the UPR pathway.

We have previously determined that the nonclassical PDI-M subfamily member, PDI9, exhibited strong protein folding activity by disulfide bond-mediated folding of alkaline phosphatase in vivo, and it complemented the dsbA protein folding mutation in E. coli (Yuen et al., 2013). The PDI9-based protein folding activity is associated with pollen exine formation and protecting pollen development from heat stress (Feldeverd et al., 2020). In this study, we provided multiple lines of evidence demonstrating a new role for PDI9 in which it interacts in the ER with the UPR stress sensor, IRE1, in Arabidopsis and E. coli expression systems, which were simultaneously coexpressing PDI9 with either IRE1A or IRE1B homologs. These data are further supported by previous evidence of IRE1B localization to the ER in Arabidopsis protoplasts (Bao et al., 2018). The primary evidence for the physical interaction of PDI9 with full-length IRE1 was demonstrated in vivo in leaf mesophyll protoplasts via the stringent assay FLIM-FRET. FRET only occurs when the donor (GFP : IRE1) and acceptor (PDI9:mCherry-KDEL) proteins are in close physical proximity (< 10 nm) and in the proper polar orientation to permit energy transfer (Bajar et al., 2016; Kaufmann et al., 2020). FRET from the donor GFP : IRE1 to the acceptor PDI9:mCherry-KDEL was observed as a PDI9-mCherry-dependent quenching of GFP fluorescence. This decrease in the fluorescence lifetime of the GFP donor ( Figures 2 , 5 ) is indicated by an increase in FRET efficiency. The FRET efficiencies of GFP : IRE1 with PDI9:mCherry-KDEL ( Figures 2 , 5 ) ranged from 5.97% to 6.82% and were slightly less than the FRET efficiencies of 7.4% to 7.8% obtained with the positive control, GFP:mCherry direct translational fusion.

Further evidence supporting the interaction between PDI9 and IRE1 was obtained via co-I.P. of the proteins coexpressed in mesophyll protoplasts and in E. coli. The full-length versions of IRE1A and IRE1B were used in the plant protoplast expression system, whereas the lumenal domains of IRE1A and IRE1B were successfully expressed for co-I.P. in E. coli. The soluble ER lumenal domain of IRE1 is the region expected to be available to interact with PDI9, which also resides in the ER lumen (Feldeverd et al., 2020). In each expression system, the IRE1A and IRE1B proteins were detected in the co-I.P. fractions with PDI9 after stringent washes.

It is worth noting that we observed minor differences in the interaction patterns between IRE1A and/or IRE1B with PDI9. IRE1B displayed a consistently weaker interaction with PDI9 relative to IRE1A ( Figures 2 – 5 ). It is hypothesized that functional and structural differences in IRE1B contribute to this difference in binding affinity with PDI9 relative to IRE1A. Arabidopsis IRE1A and IRE1B have functional redundancy, yet differences such as in expression patterns (Koizumi et al., 2001) and their roles in autophagy (Bao et al., 2018), pathogen response (Verchot and Pajerowska-Mukhtar, 2021), and gametogenesis (Pu et al., 2019) suggest specialization. The lumenal domains of the IRE1 homologs are less conserved among plant species relative to the cytosolic kinase/RNase domain. In Arabidopsis, IRE1A and IRE1B share 57% identity in the cytosolic domain and only 19% in the ER lumenal domain (Li and Howell, 2021). Therefore, PDI9 may preferentially bind to IRE1A in Arabidopsis as a result of structural or biochemical properties unique to the isoform. Questions remain as to what functional differences exist between IRE1A and IRE1B in the ER stress response, as well as what role PDI9 has with respect to these specializations.

Site-specific mutagenesis was used to individually disrupt single cysteines in each of the two thioredoxin catalytic domains of PDI9 and two critical cysteines each in IRE1A and IRE1B. The resulting co-I.P. and FLIM-FRET results with the mutated versions of PDI9 indicated that its interaction between IRE1A and IRE1B is largely dependent on the first thioredoxin domain’s active site (PDI9 C63), such that the C63A mutation abolished the interaction ( Figures 3 – 5 ). Mutation of IRE1A (C233) and IRE1B (C107) sharply decreased interaction with WT PDI9. Therefore, the C63 may form a disulfide bond with IRE1 through a paired cysteine residue (IRE1A C233 and IRE1B C107) in the lumenal/sensor domain. In contrast, the C257A and C222A mutations in IRE1A and IRE1B, respectively, only slightly diminished interaction with PDI9 as opposed to the C233A and C107A mutations. Although PDI can engage in promiscuous disulfide introduction with various target cysteines (Yuen et al., 2013; Feldeverd et al., 2020), its isomerization and proofreading functions can rearrange disulfide bonds to the final most stable cysteine partners for oxidative binding, folding, and/or transfer of electrons in redox-based regulation of enzymes (Hatahet and Ruddock, 2009; Sato et al., 2013). We suggest that the cysteine positions in the lumenal domain of IRE1 may play a cooperative but nonequivalent role in binding to PDI9 for the proper structural conformation to enable a stable interaction.

Similarly, point mutations in the active sites of the mammalian ortholog, PDIA6, reveal that the first N-proximal CXXC active site motif exhibits greater catalytic activity with respect to the C-terminal active site (Lyles and Gilbert, 1994). The mammalian ortholog PDIA6 was shown to interact with IRE1 through the critical cysteine residue Cys148 (Liu et al., 2003; Eletto et al., 2014) and form a disulfide bond with PDIA6, and this interaction was dependent on the phosphorylation-based dimerization and activation of IRE1 (Eletto et al., 2014). The influence of PDI9 on the phosphorylation-activation status of IRE1 in plants has not been determined. However, the data with the PDI9 gene knockout mutants described below supports the hypothesis that PDI9 attenuates IRE1 activity.

Three distinct experimental approaches were used to indicate that the loss of PDI9 resulted in the upregulation of UPR gene expression. Each approach targets a different step in the UPR pathway. First, the absence of PDI9 increased the UPR-dependent splicing of the bZIP60 intron from the GFP mRNA when the 35S::bZIP60 intron:GFP reporter was expressed in protoplasts. This led to significantly higher levels of GFP fluorescence detected in the two single pdi9 mutants and the double pdi9–pdi10 mutant relative to WT under normal conditions ( Figure 6 ). Moreover, with respect to WT, a near threefold elevation of bZIP60 mRNA intron splicing was observed in pdi9–pdi10 cells under DTT-induced ER stress ( Figure 6 ). This splicing is one of the first steps in IRE1 kinase/RNase output at the start of the UPR signal transduction pathway. Secondly, the activation of the PDI9 promoter, which is a downstream target of the IRE1-induced bZIP60 transcription factor in UPR signaling (Lu and Christopher, 2008), was also highly upregulated in the single and double mutants in the absence of ER stress, and in all genotypes under ER stress ( Figure 7 ). Thirdly, in leaf mesophyll protoplasts, mRNA levels significantly increased for UPR-induced chaperones BiP2 and PDI11 plus bZIP60 total and spliced RNAs in the single pdi9 and double pdi9–pdi10 mutants relative to WT with and without DTT treatment ( Supplementary Figure S4 ). Thus, taken together, we conclude that PDI9 plays an important role in the UPR, and its loss leads to a significant stimulation of the UPR pathway and downstream UPR-regulated gene expression.

This investigation raises a major question about the mechanism by which the loss of PDI9 leads to the hyperactivity of the UPR. We propose the following dual roles for PDI9: (a) the compensation mechanism; and (b) the attenuation mechanism. The first, PDI9, has been well-documented to catalyze oxidative disulfide bond-mediated protein folding in the ER (Yuen et al., 2013; Feldeverd et al., 2020). Therefore, the loss of PDI9 in the pdi9 knockout mutants results in a deficiency of PDI9-mediated protein folding capacity, leading to more unfolded proteins in the ER. This increased demand for protein folding could be satisfied by a general stimulation of UPR and the accumulation of downstream protein folding chaperones. The resulting stimulation of UPR partially compensates for the PDI9 deficiency. In the second mechanism, we postulate that PDI9 interacts with the lumenal domain of IRE1 in the ER to attenuate the IRE1-mediated UPR pathway. The loss of PDI9, therefore, derepresses IRE1, leading to stimulation of UPR and its downstream components.

The model in Figure 7 depicts in further detail the proposed dual mechanisms by which PDI9 fits into the current thinking of UPR. Under unstressed conditions, the Hsp70 cognate, BiP, associates with the lumenal face of monomeric IRE1, preventing its activation and phosphorylation-based dimerization (Kopp et al., 2018). A low level of basal splicing of bZIP60 mRNA occurs through the IRE1 dual kinase/RNase activity (Eletto et al., 2014). In response to ER stress, however, unfolded and misfolded proteins accumulate and bind to BiP (Bravo et al., 2013; Vitale et al., 2019). BiP then dissociates, liberating IRE1 to dimerize and activate the UPR (Adams et al., 2019). It is thought that the IRE1 lumenal domain can also bind to the exposed hydrophobic residues on unfolded/misfolded polypeptides, triggering dimerization and subsequence activation of the cytosolic kinase and RNase domains (Li and Howell, 2021). Activated IRE1 catalyzes the unconventional splicing of bZIP60 mRNA in the cytosol, recognizing and cleaving two characteristic RNA stem-loop structures in a 23-nt intron. This bZIP60 RNA intron maintains an encoded reading frame for a C-terminal transmembrane domain that normally localizes the protein to the ER membrane. However, splicing out the intron-encoding sequence produces a new translational reading frame for the bZIP60 transcription factor that translocates to the nucleus to activate UPR target genes. Induction of such genes is mediated through the ER stress response element (ERSE) and UPR element (UPRE), cis-regulatory sequences located in their promoters (Iwata and Koizumi, 2012).

During this process, we propose two competing molecular functions of PDI9 that govern the regulatory system. One is involved in oxidative protein folding, and the other binds to dimeric IRE1 to attenuate IRE1 signaling. The stoichiometric balance of these PDI9 activities maintains proteostasis in the ER between stressed and unstressed conditions. When the PDI9 population is consumed by a high demand for protein folding, less PDI9 is available to interact with the IRE1 complex, thus activating UPR. The need for UPR and protein folding would be high in this scenario of ER stress. In turn, as ER stress and the demand for protein folding subside, less PDI9 is involved in protein folding. A free population of PDI9 is thus available that can associate with the IRE1 complex to attenuate it. The need for UPR would be diminished in this phase.

This layered and multiplex attenuation prevents excessive and continuous UPR signaling, which can be deleterious (Lin et al., 2007; Rubio et al., 2011; Tufo et al., 2014). For example, attenuation of yeast UPR is essential for survival and is mediated by inhibition of IRE1 (Chawla et al., 2011). Research suggests that excessive UPR signaling can promote programmed cell death (PCD) in plants when ER stress remains constitutive (Manghwar and Li, 2022), highlighting the importance of high sensitivity and responsiveness with respect to both the activation and attenuation of this essential signaling pathway (Manghwar and Li, 2022). The model (Eletto et al., 2014) for the human homolog of Arabidopsis PDI9, PDIA6, denotes that it interacts with IRE1 to limit the duration of UPR activity and inhibit the response. However, unlike for plant PDI9 (Yuen et al., 2013), no protein folding activity has been detected for human PDIA6. Due to the importance of the UPR (Zhang et al., 2016) and the conserved primary structure of the PDIs between plants and other eukaryotes (Yuen et al., 2013), we suggest the basic regulatory mechanism of a PDI that represses IRE1 could be conserved among eukaryotes. This mechanism modulates the output of UPR and highlights the critical need for multiple forms of regulation of the ER stress response, including a safeguarding mechanism for attenuation to ensure cellular homeostasis is maintained.

Although the interaction of PDI9 with IRE1 is especially strong, it does not preclude that other PDIs in Arabidopsis could interact with IRE1A/B and influence IRE1A/B and UPR. It is conceivable that simultaneously knocking out other PDI genes could decrease protein folding activity and thereby increase the UPR. Further experimentation would be required to support the broader role of the entire PDI family, which would include interaction and gene knockout studies. However, in mammals, only the PDI9 homolog PDIA6 interacts with IRE1 (Eletto et al., 2014). In addition, we conducted quantitative yeast-two-hybrid screening of Arabidopsis PDI1 and PDI2 (Cho et al., 2011) and PDI5 (Ondzighi et al., 2008) and found no interaction with IRE1. The studies reported here provide a solid foundation for further experimentation to test the proposed model.

With the exception of BiP (Humbert et al., 2012; Li and Howell, 2021), much is unknown regarding the factors that regulate the UPR once activated in plants, particularly within the ER lumen. In mammals, proteins have been identified that function to attenuate the UPR from the cytosolic face of activated IRE1, including the Bcl-2 family protein, BI-1, and several phosphatases (Eletto et al., 2014). Zhu et al. (2019) recently showed that the anti-apoptotic-like proteins GAAP1 and GAAP3 interact with the cytoplasmic domain of IRE1A and IRE1B in Arabidopsis to negatively modulate the UPR and reduce the extent of ER stress-induced cell death. However, the UPR originates from inside the ER and responds to the dynamic secretory protein load in the lumen. This investigation sheds light on the PDI9-based mechanisms by which the unfolded protein signals in the ER are being communicated to ensure an appropriate UPR. In summary, this work highlights the importance of understanding how plants regulate the intricacies of the ER stress response to maintain and protect cellular homeostasis. We propose a model of the UPR pathway in plants in which PDI9 binds to IRE1 in the ER lumen to modulate the UPR pathway. This interaction takes place largely through the formation of a critical disulfide bond between the two protein partners and ultimately affects downstream splicing of the UPR-targeted transcription factor, bZIP60, as well as the expression of UPR-targeted genes such as BiP and PDI9 itself. PDI9 also functions as a critical foldase (Yuen et al., 2013) to assist with the oxidative folding of nascent or misfolded polypeptides within the ER involved in pollen exine formation (Feldeverd et al., 2020). The subgroup of PDI9 engaged in protein folding influences the amount of PDI9 available for the PDI9–IRE1 interaction identified here. The balance of such PDI9 activities, therefore, serves as an important signaling mechanism for the cell that regulates the protein folding load to maintain homeostasis and prevent excessive UPR signaling in an adaptive manner. The multiplex ability of the cell to modulate and control such a critical stress signaling pathway would promote the levels of UPR to be precisely proportional to the demand. These results highlight the need for better understanding the intricate mechanisms by which plants mitigate ER stress-inducing events such as heat stress and promote future agricultural advancements that may improve plant tolerance to stress.