Decomposition profile data analysis of multiple drug effects identifies endoplasmic reticulum stress-inducing ability as an unrecognized factor

Chemicals have multiple effects in biological systems. Because their on-target effects dominate the output, their off-target effects are often overlooked and can sometimes cause dangerous adverse events. Recently, we developed a novel decomposition profile data analysis method, orthogonal linear separation analysis (OLSA), to analyse multiple effects. In this study, we tested whether OLSA identified the ability of drugs to induce endoplasmic reticulum (ER) stress as a previously unrecognized factor. After analysing the transcriptome profiles of MCF7 cells treated with different chemicals, we focused on a vector characterized by well-known ER stress inducers, such as ciclosporin A. We selected five drugs predicted to be unrecognized ER stress inducers, based on their inducing ability scores derived from OLSA. These drugs actually induced X-box binding protein 1 splicing, an indicator of ER stress, in MCF7 cells in a concentration-dependent manner. Two structurally different representatives of the five test compounds exhibited similar results in HepG2 and HuH7 cells, but not in PXB primary hepatocytes derived from human-liver chimeric mice. These results indicate that our decomposition strategy using OLSA uncovered the ER stress-inducing ability of drugs as an unrecognized effect, the manifestation of which depended on the background of the cells.


Results
Characterization of a vector focussed on the ability of chemicals to induce ER stress. From the CMap database, we obtained transcriptome data describing the cellular responses of MCF7 cells treated with 318 compounds and subjected these data to OLSA (Fig. 1a). To select a vector that was related to toxic aspects of drugs, we analysed the component genes and the characteristic compounds for each vector. The genes constituting each vector were sorted by their absolute values and the top 1% genes were subjected to gene ontology (GO) analysis. Hereafter, we use the term PXV to indicate the Principal X th Vector component (X means the rank of the contribution). Based on these results, we focused on vector P14V because the top 1% of vector genes was significantly enriched for GO terms relevant to ER stress induction, such as 0034976, 0035966, and 0006986 ( Fig. 1b, Supplementary Data). The ER stress response is well-known to be involved in DILI, which is often the cause for drugs being withdrawn from the market and is a precise toxic effect of some drugs [10][11][12] . In OLSA, the scores calculated for a vector indicate the strength of the effect represented by the vector. Many of the drugs and compounds with high P14V scores (i.e., a strong P14V-related effect) were previously reported to be associated with the UPR or ER stress (Fig. 1c). Consistency between the component genes and the characteristic compounds of P14V with respect to ER stress motivated us to investigate this vector as a possible detector of latent aspects of drugs.
It was important to emphasize the differences between the scores defined by the focused vector and those identified by other generally used indicators, such as the expression levels of well-known ER stress markers and similarity to well-known ER stress inducers. Therefore, we surveyed the literature to clarify the ability of high-scoring compounds to induce ER stress, as defined by the indicated interquartile range (IQR) of the score distribution of each indicator ( Supplementary Fig. S1). As shown in Fig. 1d, the number of previously confirmed ER stress inducers within the compounds with a high P14V score was relatively high compared with those determined by other well-known markers or inducers, particularly at high IQR, although the reported ratio was comparable for all indicators. Note that the lack of a relevant publication does not mean that a compound does not induce ER stress, just that no information is available. All of these results indicate that the features of P14V response scores are different from those calculated by conventional approaches and are suitable for exploratory use in the detection of the ability of chemicals to induce ER stress.

Selection of test drugs and decomposition analysis.
Unlike the compounds with high P14V scores, some of the compounds with moderate scores had not previously been associated with ER stress (Fig. 2a). We chose five such drugs (protriptyline, perphenazine, phenoxybenzamine, cyproheptadine, and trimipramine) as the candidate drugs for validation based on the following criteria: (1) drugs with a moderate score defined as median plus 0.75 times the IQR of the P14V score distribution, (2) no publications related to ER stress or UPR stress, and (3) drugs approved by the US Food and Drug Administration (FDA) for oral use ( Supplementary  Fig. S2). Three compounds with a low absolute value for P14V score (bottom three compounds, cycloserine, ciclopirox, and delsoline) were also selected as predicted negative compounds. All the candidate drugs clustered in different clusters from that, including the well-known ER stress inducers thapsigargin and ciclosporin A ( Supplementary Fig. S3). Of note, phenoxybenzamine and trimipramine also clustered distantly from the well-known inducers, which indicates that these drugs cannot be listed using representative clustering methods.
Visualization of the response scores of a small compound provided us with valuable information for clarification of the full range of effects of the chemical. To obtain insight for each candidate, we plotted the response scores of test compounds. The P14V scores of thapsigargin and ciclosporin A were the highest and were strikingly high compared with those of the other vectors, consistent with the fact that these are well-known ER stress inducers (Fig. 2b). However, the P14V scores of the candidate drugs were comparable to those for other high-scoring vectors, although, except for cyproheptadine, these P14V scores were not the highest of all vectors. These results suggest that the capacity of the five candidates to induce ER stress is latent, and that decomposition analysis makes it easy to recognize this inductive ability. www.nature.com/scientificreports/

Induction of ER stress in MCF7 cells by test compounds.
We hypothesized that the candidate drugs (five drugs with moderate P14V scores) had a latent ability to induce ER stress at high concentrations while the three predicted negative compounds (P14V scores close to zero) did not. To verify this hypothesis, we investigated the non-canonical splicing of X-box binding protein 1 (XBP1). Under ER stress, XBP1 is non-canonically spliced to its active form by inositol-requiring enzyme 1 α (IRE1α) in the cytosol, translocates into the nucleus, and functions as a transcription factor; therefore, the ratio of spliced to unspliced XBP1 is a good indicator of ER stress 13,14 . We have established a system to evaluate XBP1 splicing using MultiNA, an electrophoresis apparatus that incorporates a microchip. This assay system demonstrated a dose-dependent increase in the XBP1 splicing ratio caused by a well-known ER stress inducer, ciclosporin A ( Supplementary Fig. S4). MCF7 cells were treated with the test compounds at various concentrations for 6 h and XBP1 splicing ratios were measured. Although the cells were adherent and active at all concentrations, the morphology of the cells became rounded (early phenotype of apoptosis) under some conditions ( Supplementary Fig. S5). As shown in Fig. 3a and Supplementary Fig. S6, all the candidate compounds enhanced XBP1 splicing at the highest concentration while the predicted negative compounds did not. Although the magnitude of the increase varied between the drugs, a trend toward a concentration-dependent increase was observed. We also investigated the effects of these drugs on the expression levels of ER stress markers in MCF7 cells. Transcripts of GRP78 and CHOP, well-known  www.nature.com/scientificreports/ ER stress markers 15,16 , were quantified using quantitative polymerase chain reaction (qPCR, Fig. 3b). Three of the five candidates (protryptiline, phenoxybenzamine, and trimipramine) significantly increased the levels of mRNA for both markers, while cyproheptadine enhanced only CHOP and perphenazine had no effect on either marker. Western blotting analysis showed that phenoxybenzamine and trimipramine induced a clear increase in expression of GRP78 and phosphorylated-IRE1α while the other candidate compounds had only small effects (Fig. 3c). These results indicate that the vector signature generated using OLSA detected the latent ability of the candidate drugs to induce ER stress, although the extent of the effects did not correlate with their scores for the signature and varied among the candidate drugs.
Induction of ER stress in hepatic cancer-derived cell lines by representative test compounds. The results using MCF7 cells supported that OLSA-based detection of latent ER stress-inducing ability worked well in the cell line from which the transcriptome profiles were obtained. We were interested to determine whether this was also true in other cell lines. Because ER stress is one of the major causes of DILI, we selected the hepatic cancer-derived cell lines, HepG2 and HuH7. Because of the scale of the experiments, only two representative candidate drugs, trimipramine and phenoxybenzamine, were subjected to these further analyses. These two drugs were chosen based on their distances from well-known ER inducers and their structural dissimilarity. As shown in Supplementary Fig. S3, these two drugs both clustered distantly from both thapsigar- www.nature.com/scientificreports/ gin and ciclosporin A. In addition, while protriptyline, perphenazine, cyproheptadine, and trimipramine share a similar 3-ring structure, phenoxybenzamine does not ( Supplementary Fig. S2). First, we checked the effects of the representative drugs on the morphology of the two liver-derived cell lines. Interestingly, the morphology of both cell lines was much more strongly affected than that of MCF7 cells, and both HepG2 cells and HuH7 cells were nearly all dead at the highest drug concentration used in MCF7 cells. This suggests that there are differences between cell lines in their morphological responses to, and tolerated dose of, trimipramine and phenoxybenzamine ( Supplementary Fig. S7a). Next, we treated both HepG2 cells and HuH7 cells with the representative drugs for 6 h at a concentration below the highest concentration employed in the experiments with MCF7 cells, and we evaluated the non-canonical splicing of XBP1 and the morphology of the two cell lines ( Supplementary Fig. S7b). Figure 4a,b demonstrates that XBP1 splicing was significantly increased in both cell lines by the representative drug treatments. Cycloserine, the compound with the second-lowest score, had no effect on the splicing ratio in either cell line, while ciclopirox, which had the The arrow, the upper arrowhead, and the lower arrowhead indicate a non-specific band, the unspliced XBP1, and the spliced XBP1, respectively. Each value in the lower graph indicates a ratio of spliced to unspliced XBP1 concentration. Note that TPG was excluded from these experiments because it exhibited quite a high value and was not suitable as a positive control in data visualization. (b) Changes in expression of mRNA for ER stress markers induced by the candidate drugs. MCF7 cells were treated with the candidate drugs for 24 h at X-times the concentration used for transcriptome data acquisition. X: CSP, 1; PTP, 2; PPZ, 1; PBZ, 10; CHD, 3; TPM, 5; DSL, 10; CPX, 10; CSE, 10. cDNA was synthesized from mRNA and subjected to quantitative PCR analysis. (c) Changes in protein expression of ER stress markers induced by the candidate drugs. MCF7 cells were treated with the candidate drugs for 24 h at X-times the concentration used for transcriptome data acquisition. X: CSP, 1; PTP, 2; PPZ, 1; PBZ, 10; CHD, 3; TPM, 5; DSL, 10; CPX, 10; CSE, 10. The whole-cell lysate was analysed by western blotting. A representative result of at least two independent experiments is shown in each figure. All significance tests were conducted with Welch t test, P values were adjusted for FDR (Benjamini-Hochberg method), and only significant differences between DMSO and the tested compounds are shown: *P < 0.05, **P < 0.01, ***P < 0.001. www.nature.com/scientificreports/ third-lowest score, significantly increased the splicing in HuH7 cells at high concentrations, but did not affect it in HepG2 cells (Supplementary Figs S8 and S9). Consistent with these findings, we confirmed a similar trend in the changes in expression of GRP78 and CHOP (Fig. 4c,d). These combined results indicate that, with the exception of ciclopirox in HuH7 cells, the latent ER stress-inducing ability predicted by OLSA is close to that observed in cell lines other than the line utilized in the acquisition of the transcriptome profiles, although the extent of the latent effect varies among cell lines.

Induction of ER stress in PXB cells by representative test compounds.
It is widely accepted that the function of metabolic enzymes such as cytochrome P450 (CYP) is impaired in tumour-derived cell lines 17 . Thus, the manifestations of ER stress induced by the candidate drugs may be different in normal cells, which motivated us to test whether the latent effect could be observed in primary cultured cells. Because the liver is the main organ that deals with xenobiotics such as drugs and chemicals, several liver-derived resources have been developed for evaluation of the activities of drug transporters and metabolic enzymes. PXB cells, primary cultured hepatocytes derived from human-liver chimeric mice, are one such state-of-the-art resource 18 . Their normal metabolic activities and reactivity to compounds are relatively well preserved compared with those of hepatic cancer-derived cell lines such as HepG2 and HuH7, although it is well known that metabolic activity rapidly decreases in primary cultured cells after seeding 19 .
We employed PXB cells to address the above question by evaluating the effects in these cells of the representative candidate compounds, trimipramine and phenoxybenzamine. Of note, the morphology of the cells was almost unaffected by treatment with these compounds at the highest concentration used in MCF7 cells, which severely damaged the two hepatic cancer-derived cell lines, HepG2 and HuH7 (Supplementary Fig. S10). PXB cells were treated with trimipramine and phenoxybenzamine and two of the predicted negative compounds, cycloserine and ciclopirox. After 24-h treatment, non-canonical splicing of XBP1 was clearly increased in the cells treated with trimipramine although the increase was relatively small compared with that observed in the other cell lines (Fig. 5a). Of note, neither phenoxybenzamine nor the predicted negative compounds affected the www.nature.com/scientificreports/ splicing ratio. With respect to mRNA expression, trimipramine significantly enhanced levels of mRNA for the ER stress markers GRP78 and CHOP (Fig. 5b). The other compounds had almost no effect on these markers, although ciclopirox induced a small but significant enhancement of CHOP expression, similar to the trend observed in HuH7 cells (Fig. 4d). These results indicate that in some instances, the prediction of the latent ER stress-inducing ability of drugs by analysing transcriptome data derived from tumour cell lines can be extrapolated to primary cultured hepatocytes (true for one of two drugs under the experimental conditions used in this study).

Discussion
Because a small compound interacts not only with its primary target but also with multiple secondary targets, we often observe unexpected effects of chemicals on cellular functions. These unrecognized effects of small compounds are sometimes beneficial, but can also be harmful. The best solution to this problem would be to define the full range of effects of a chemical. However, this is quite difficult because, in most cases, the "full range" remains undefined. To address this issue, we recently developed a new method of profile data analysis, OLSA, to facilitate understanding of the characteristics of a chemical by decomposing its effects represented by a genetic profile 7 . One example of a damaging unrecognized effect is an adverse event related to drug use. In the present study, we investigated whether the ability of a compound to induce ER stress, one of the major causes of adverse events, can be detected by our "decomposition and understanding" strategy, even when this ability is weak and latent.
In this study, we confirmed that all the candidate drugs had the latent ability to induce ER stress as defined by non-canonical splicing of XBP1 13,14 , in MCF7 cells, in which the analysed transcriptome data were obtained (Fig. 3). However, some of the drugs did not enhance the expression of marker genes or proteins. One of the  10. cDNA was synthesized from mRNA and subjected to conventional PCR. The concentrations of spliced and unspliced XBP1 amplified products were quantified using a MultiNA electrophoresis apparatus. The arrow, the upper arrowhead, and the lower arrowhead indicate a non-specific band, the unspliced XBP1, and the spliced XBP1, respectively. Each value in the lower graph indicates a ratio of spliced to unspliced XBP1. (b). Changes in expression of mRNA for ER stress markers. Extracted cDNA as described in (a) was subjected to quantitative PCR analysis. A representative result of at least two independent experiments is shown in each figure. All significance tests were conducted using Welch's t test, P values were adjusted for FDR (Benjamini-Hochberg method), and only significant differences between DMSO and the tested compounds are shown: *P < 0.05, **P < 0.01, ***P < 0.001. www.nature.com/scientificreports/ possible explanations for this is the differences in the time course of these experiments. The expression levels of the markers were measured after 24-h treatment with the candidate compounds to maximize detection sensitivity, while evaluation of XBP1 splicing was performed after a 6-h treatment [20][21][22] . This short treatment duration is suitable for the evaluation and description of the effects of a drug because longer-term treatment can result in feedback reactions and blunt the effects of interest. In fact, the transcriptome profile data in CMap was mainly acquired after 6 h of compound treatment 5 . An important future task would be to investigate the features of these drugs at other time points. Note that this short treatment duration in MCF7 cells, and probably in other cell lines as well, is enough to induce the expression changes of ER stress-related genes, although the changes are too small to be recognized from a macro perspective in the transcriptome. As shown in Supplementary Fig. S3, phenoxybenzamine and trimipramine located into completely different clusters from those containing the two well-known ER stress inducers. This indicated that this unrecognized effect of these drugs could not be determined using the Ward method, a representative method of unsupervised multivariate analysis 23,24 . Because many of the clustering methods utilize whole variables, in practice the results tend to be dominated by the on-target effects and not to reflect the off-target effects. In contrast, as shown in Fig. 2b, our "decomposition and understanding" strategy using OLSA is effective for the evaluation of such off-target effects.

Scientific RepoRtS
As shown in Fig. 3, we could confirm the induction of ER stress by the test drugs at concentrations higher than those employed in the original acquisition of transcriptome data. Note also that the actual exposure concentration may vary between this study and the original transcriptome data acquisition, because there may have been undefined differences in the experimental conditions, such as the batch of foetal bovine serum (FBS). However, compared with the test drugs, the predicted negative compounds did not induce ER stress, even at the highest concentration used. Total strength (L2 norm of gene expression changes of each sample) of negative compounds is similar to that of positive compounds (Supplementary Data). Considering that total strength represents the magnitude of drug effects, treatment with the negative compounds at the CMap concentrations, which are the base concentrations, have enough effectivity to be compared with each other 7,25,26 . These results indicated that the compounds with moderate scores on the ER stress vector potentially have ER stress-inducing ability, which, depending on the exposure concentration, can be latent. It should be kept in mind that in clinical practice, the exposure concentration can be unexpectedly high because of many unintended differences between patients such as mutations, drug-drug interactions, and tissue injury [27][28][29] .
The successful validation of our method in MCF7 cells means that our approach worked well in identifying whether a chemical has the potency to cause the effect of interest. In this study, we also investigated the effect in other cell lines. As shown in Fig. 5 , PXB cells exhibited no ER stress responses to phenoxybenzamine, which induced ER stress in all the other cell lines. Note that knowing the potency of a chemical is a totally different issue to prediction of the nature and extent of the effect, particularly in different environments. The factors that affect the manifestation of an effect by a chemical are mainly divided into two types: dynamic factors, such as the affinity of a chemical for the target protein, which determine the magnitude of the effect, and kinetic factors, which regulate the exposure of the chemical to the target. To discuss the former, we should know the target of each chemical, although it is quite difficult to speculate about the direct target molecule only from transcriptome data. For the latter, one of the possible explanations for the specificity of the PXB cell phenotype is a difference in metabolic enzymes. PXB cells are human primary hepatocytes derived from human-liver chimeric mice, and it is well-known that metabolic enzymes such as CYP are downregulated in tumour-derived cell lines compared with primary culture cells 17 . Indeed, differential CYP expression profiles were confirmed by the analyses of transcriptome data comparing MCF7 cells, HepG2 cells, HuH7 cells, and human hepatocytes ( Supplementary Fig. S11). Note that we analysed human hepatocytes instead of PXB cells because we could not obtain transcriptome data for PXB cells. These findings are consistent with the relatively low level of XBP1 splicing induced by trimipramine in PXB cells, because the drug is mainly metabolized by CYP2D6, which is expressed at a higher level in human hepatocytes than in the other cell types 30 (Supplementary Fig. S12). Because of a lack of information about the metabolic pathways for phenoxybenzamine, the ADMET Predictor was employed to investigate the metabolic enzymes, with the results showing that the relevant enzymes such as CYP1A2 are similarly highly expressed in PXB cells 31 . Of course, from these expression data we can analyse only correlations,further studies are necessary to identify the cause of the different phenotype of phenoxybenzamine treatment from the viewpoint of exposure. With regard to the dynamic factors, limited information is so far available for phenoxybenzamine 32,33 . It will be helpful to employ chemoinformatics methods such as docking and simulation in studies to survey dynamic factors, which is an important future task in this field 34 .
Is it possible to speculate about the nature of the pathways regulating latent ER stress inducing ability? First, western blotting analysis revealed an increase in phosphorylated IRE1, which is upstream of XBP1. This result indicated that these drugs activate IRE1-XBP1 signalling, the most preserved of the three well-known ER stress signalling pathways (IRE1-XBP1 signalling, activating transcription factor 6 signalling, and (PKR)-like ER kinase signalling) 35,36 (Fig. 3c). Next, what about upstream of IRE1-XBP1? We would like to suggest that the response scores of the other vectors help us to answer this question. Figure 2b and the Supplementary Data show that although the P14V scores of the candidates were similar, their scores for other vectors varied. Phenoxybenzamine and trimipramine, which clustered differently and have different structures, exhibited relatively large differences in scores for some vectors such as P18V, P19V, P22V, and P23V. Interestingly, combinatorial treatment with these two drugs resulted in synergistic enhancement of XBP1 splicing, which supports the hypothesis that they act at different points ( Supplementary Fig. S13). It is also noteworthy that ciclopirox, which induced biphasic and strange XBP1 splicing profiles in MCF7 cells and HuH7 cells although its P14V score was zero, had a strikingly high P19V score. Because OLSA contracts dimensions based on a linear hypothesis, one of the strong points of the method is the ease with which biological understanding of the meaning of the new indicators could be described as a weighted gene group, although non-linear approaches provide superior resolution 7  www.nature.com/scientificreports/ associated with particular GO terms, which may be clues to investigate their molecular entities (Supplementary Data). We have no clear explanation for the differences between test drugs or the peculiarity of ciclopirox, but consider that the elucidation of these mechanisms was not the main objective of this study. However, to achieve not just an understanding of the effects of a chemical but to be able to predict the way in which they are manifested, it would be important to test whether the molecular-biology-friendly features of OLSA contribute to identifying the upstream pathways. We would like to suggest that unrecognized effects of chemicals may be a novel source of drug discovery. In CMap, one of the most successful in silico platforms for drug repositioning 6,37 , the "entire" features of each compound could be classified by the differentially expressed genes that were significantly affected by the dominant, mainly on-target, effects. In contrast, our decomposition strategy provides an alternative approach that turns the focus to the off-target effects. Although such effects are often weak and can be difficult to be directly repurposed as therapeutics, they do exist as biological responses and have the potential to be drug targets for diseases associated with those responses.
Shedding light on the unrecognized effects of chemicals using OLSA may also be useful in exploring biology, because such chemicals have contributed to the development of biology; chemical biology. For instance, necrostatin-1, an inhibitor of necroptosis, was utilized to characterize an important cell-death mechanism by targeting an at first unknown protein; two years later the target was identified as a key component of necroptosis 38,39 . Thus, the uncharacterized effects of small compounds are useful tools in biology and the unrecognized effects of chemicals can be a source of such interesting unknowns. Indeed, we identified some vectors that were not associated with an existing body of knowledge, such as GO, even though the chemical structures of the highscoring compounds were similar (Supplementary Data). With respect to chemical biology, we believe that our decomposition strategy is appropriate for understanding the properties of natural products, whose effects are often complicated and complex 40 . Although the elucidation of such possibilities is not within the scope of this study, it is an attractive challenge for the future.
In this study, we have identified unrecognized effects of five drugs, focusing on ER-stress inducers. We believe that the "decomposition and understanding" strategy using OLSA is useful to illuminate such latent effects of chemicals and will lead us to a deeper understanding of the MOA of chemicals, which could contribute to drug discovery and chemical biology.
Cell culture. MCF7 cells and HuH7 cells were cultured in high-glucose Dulbecco's modified Eagle's Medium (DMEM) (#11995-073, Thermo Fisher Scientific, Waltham, MA) supplemented with 10% FBS. HepG2 cells were cultured in DMEM with 10% FBS and 1% MEM non-essential amino acids (#11140-050, Life Technologies, Waltham, MA). PXB cells (#PPC-P02) and culture medium (#PPC-M100) were purchased from PhoenixBio (Hiroshima, Japan). All cells were maintained at 37 °C under 5% CO 2 . XBP1 splicing test. After a 6-h treatment with compounds, the cells were washed with PBS and cDNA was prepared with SuperPrep II Cell Lysis & RT Kit for qPCR (Toyobo, Osaka, Japan). Note that the wash process removes the dead cells observed under some conditions. Synthesized cDNA was amplified by conventional PCR with KOD One PCR Master Mix (Toyobo); 25 cycles of denaturation: 98 °C, 10 s; annealing: 60 °C, 5 s; extension: 68 °C, 1 s using the following primers (hXBP1 Fw: GAG TTA AGA CAG CGC TTG GG, Rv: ACT GGG TCC AAG TTG TGC AG, mXbp1 Fw: GAG TTA AGA CAG CGC TTG GG, Rv: ACT GGG TCC AAG TTG TGC AG). Each DNA concentration was determined using MultiNA (electrophoresis apparatus incorporating a microchip). Band density was quantified with the pre-installed software of MultiNA. Values were calculated as the ratio of concentrations of spliced to unspliced XBP1. Raw data are summarized in Supplementary Data. Western blotting analysis. Western Blotting analysis was conducted as previously described 7 . Briefly, specimens were separated with sodium dodecyl sulphate polyacrylamide gel electrophoresis on a 10% polyacrylamide gel with a 3.75% stacking gel at 140 V for 90 min. The molecular weight was determined using Precision Plus Protein Standards (#1610373, Bio-Rad, Richmond, CA). Proteins were transferred electrophoretically to a poly(vinylidene difluoride) (PVDF) membrane (Pall, Port Washington, NY) using a blotter (Bio-Rad) at 100 V for 60 min. Non-specific binding sites on the membrane were blocked with PVDF Blocking Reagent for Can Get Signal (Toyobo) at room temperature for 60 min. After blocking, the PVDF membrane was incubated with primary antibodies diluted with Can Get Signal solution 1 (Toyobo) at 4 °C for 24 h. Primary antibodies were used as follows: anti-β-actin (1/2,000), anti-p-IRE1α (1/2,000), and anti-GRP78 (1/1,000).
Quantitative PCR. Total RNA was extracted, and cDNA was reverse transcribed with SuperPrep II Cell Lysis & RT Kit for qPCR (Toyobo). Quantitative PCR analysis was performed using Thunderbird Probe qPCR Mix (Toyobo) and a LightCycler 480 Instrument II (Nippon Genetics Co, Ltd., Tokyo, Japan). Levels of mRNA for each gene of interest were normalized to levels of mRNA for human glyceraldehyde 3-phosphate dehydrogenase (GAPDH). OPC grade PCR primers were purchased from Eurofins Genomics (Tokyo, Japan), with sequences as follows: hGAPDH (Fw: GGG GAG CCA AAA GGG TCA TCA TCT , Rv: GAC GCC TGC TTC ACC ACC TTC TTG )

Statistical analysis.
Welch's t test and permutation test were used to identify significant differences between groups, where appropriate. The data were analysed using the statsmodels and Scikit-learn libraries of Python 3.