Subverted regulation of Nox1 NADPH oxidase-dependent oxidant generation by protein disulfide isomerase A1 in colon carcinoma cells with overactivated KRas

Protein disulfide isomerases including PDIA1 are implicated in cancer progression, but underlying mechanisms are unclear. PDIA1 is known to support vascular Nox1 NADPH oxidase expression/activation. Since deregulated reactive oxygen species (ROS) production underlies tumor growth, we proposed that PDIA1 is an upstream regulator of tumor-associated ROS. We focused on colorectal cancer (CRC) with distinct KRas activation levels. Analysis of RNAseq databanks and direct validation indicated enhanced PDIA1 expression in CRC with constitutive high (HCT116) vs. moderate (HKE3) and basal (Caco2) Ras activity. PDIA1 supported Nox1-dependent superoxide production in CRC; however, we first reported a dual effect correlated with Ras-level activity: in Caco2 and HKE3 cells, loss-of-function experiments indicate that PDIA1 sustains Nox1-dependent superoxide production, while in HCT116 cells PDIA1 restricted superoxide production, a behavior associated with increased Rac1 expression/activity. Transfection of Rac1G12V active mutant into HKE3 cells induced PDIA1 to become restrictive of Nox1-dependent superoxide, while in HCT116 cells treated with Rac1 inhibitor, PDIA1 became supportive of superoxide. PDIA1 silencing promoted diminished cell proliferation and migration in HKE3, not detectable in HCT116 cells. Screening of cell signaling routes affected by PDIA1 silencing highlighted GSK3β and Stat3. Also, E-cadherin expression after PDIA1 silencing was decreased in HCT116, consistent with PDIA1 support of epithelial–mesenchymal transition. Thus, Ras overactivation switches the pattern of PDIA1-dependent Rac1/Nox1 regulation, so that Ras-induced PDIA1 bypass can directly activate Rac1. PDIA1 may be a crucial regulator of redox-dependent adaptive processes related to cancer progression.


Introduction
Protein disulfide isomerase (PDI or PDIA1) is a dithiol/ disulfide oxidoreductase chaperone from the endoplasmic reticulum (ER), where it assists redox protein folding and thiol isomerization. PDIA1 is the prototype of a multifunctional family having > 20 members 1,2 . In addition, PDIA1 is involved in redox cell signaling regulation at distinct levels 1 . PDIA1 can also locate at the cytosol, cell surface, and is secreted by distinct cell types 3 . Cell-surface/secreted PDIA1 regulates virus internalization, thrombosis, platelet activation, and vascular remodeling 1,4 . Overall, PDIA1 is implicated in the pathophysiology of cardiovascular and neurodegenerative disorders, diabetes, and, in particular, cancer 5 . Several PDIs such as PDIA1, PDIA6, PDIA4, and PDIA3 are reportedly upregulated in cancer 6 . PDIA1, in particular, is overexpressed in melanoma, lymphoma, hepatocellular carcinoma, brain, kidney, ovarian, prostate, and lung cancers [6][7][8][9][10] and frequently associates with metastasis, invasiveness, and drug resistance 11,12 . Conversely, lower tumor PDIA1 levels associate with improved survival in breast cancer and glioblastoma 13 . In glial cells, breast and colorectal cancer, PDIA1 overexpression has been proposed as a cancer cell biomarker [13][14][15] . The mechanisms whereby PDIA1 supports tumor progression are yet poorly understood.
An important cancer cell hallmark is the enhanced output of reactive oxygen species (ROS) such as superoxide, hydrogen peroxide, peroxynitrite, etc., which engage into disrupted signaling routes that further support tumorigenesis or metastasis, but in some instances may suppress tumor propagation 16 . Such dual oxidant effects of ROS in tumorigenesis may underlie transition from adaptive to maladaptive responses enabling tumor escape 17 . Therefore, mechanisms of ROS regulation can illuminate the understanding of tumor biology and are potential therapeutic targets. Most of such mechanisms converge to enzymatic ROS sources, such as mitochondrial electron transport and Nox family NADPH oxidases. Noxes, in particular, have been increasingly implicated in cancer pathophysiology 18 . The upstream mechanisms governing Nox-dependent processes in cancer are not fully understood. In vascular cells, our group has shown consistent correlation between PDIA1 and Noxdependent ROS generation. PDIA1 silencing/inhibition abrogates growth factor-dependent Nox1 activation and expression [19][20][21] and, in parallel, significantly disrupts cytoskeletal organization, RhoGTPase activation, and cell migration 4,21 . Acute PDIA1 overexpression supports agonist-independent superoxide production and Nox1 expression in vascular smooth muscle (VSMC) 20,21 . PDIA1 similarly converges with Nox2 in phagocytes 22,23 . We propose that PDIA1 is a relevant upstream regulatory mechanism of ROS generation in tumor cells. Conversely, understanding mechanisms associated with PDIA1/Nox convergence may help to understand the roles of PDIA1 in cancer pathophysiology.
Here, we focused on colorectal cancer cells (CRC), since colorectal tissue basally expresses high protein expression levels of Noxes 24 . In total, Ca.60% of CRC present a mutation on Ras proto-oncogenes or one of their downstream effectors 25 . Cells bearing Ras mutations exhibit enhanced ROS generation 26 . Nox1 expression correlated with KRas-activating mutations in CRC 27 . Ras proteins are GTPases that function as molecular switches regulating cell proliferation and survival through canonical Raf-MEK-Erk and PI3K pathways. Aberrant Ras overactivation often follows single mutations favoring GTP binding, typically at codons 12, 13, or 61 and associates with hyper-proliferative developmental disorders and cancer. Here, we investigated mechanisms associated with PDIA1-mediated Nox1 regulation in CRC exhibiting distinct levels of Ras pathway activation.

Cell models
To assess the roles of PDIA1-dependent Nox regulation in CRC, we selected colon carcinoma cell lines HCT116 and HKE3, a well-known pair of isogenic cell lines known to differ by KRas constitutive activation ( Fig. 1a and Suppl. Fig.1A). KRas G13D mutant gene is disrupted in HKE3 cells by homologous recombination and insertion of a non-transcribed KRas G12C mutant 28 . Although recent data 29 showed that HKE3 cells do present a reminiscent RNA expression of KRas G13D gene, KRas gene and protein are clearly less expressed in HKE3 than HCT116 29 . We also used a second control Caco2 colon carcinoma cell line, which is a non-mutated wild-type KRas control.
In addition to bioinformatic analysis, we performed immunoblots to validate these results in our cells, showing increased PDIA1 protein expression in HCT116 vs. HKE3 or Caco2 (Fig. 1c), further confirmed through intracellular PDIA1 titration using ELISA (Fig. 1d). ER stress marker expression showed no differences among the cell lines (Suppl. Fig. 1B). As a control, we transfected primary VSMC with overactivated Ras mutant, and observed analogous increase in PDIA1 gene expression (Suppl. Fig. 1C). Thus, PDIA1 protein expression correlates with increased KRas activation.

PDIA1 silencing promotes dual, Ras-dependent, effects on superoxide production
To address the effects of PDIA1 on oxidant generation, we investigated the effects of PDIA1 loss-of-function. Specific superoxide generation was assessed through DHE/HPLC method 30 . There was no difference in baseline superoxide generation between HKE3 and HCT116 cells (Suppl. Fig. 1D). We then examined the effects of PDIA1 loss-of-function through siRNA. Importantly, drastic PDIA1 knockdown is reportedly cytotoxic 31 and unpublished results from our laboratory), and in vivo PDIA1 knockout is lethal for mice embryos 32 . Furthermore, PDIA1 protein is quite stable, with a long half-life 33 , requiring aggressive procedures to promote its complete silencing. Thus, here we performed a conservative loss-of-function using siRNA-mediated silencing, with 54-60% lower PDIA1 expression in our cells (Suppl. Fig. 2A). PDIA1 silencing promoted lower superoxide generation in HKE3 and Caco2 cells (Fig. 1e, f), in line with our previous VSMC data 21 . Contrarily, in HCT116 cells, PDIA1 silencing increased superoxide production (Fig. 1g). These results were confirmed using the lucigenin reductase assay (Suppl. Fig. 3). Thus, overactivated Ras associates with switch in the pattern of PDIA1-mediated superoxide regulation. We interrogated whether ER stress would be differentially induced by PDIA1 silencing in these cells. However, ER stress marker expression was unaltered by PDIA1 silencing in HKE3 and HCT116 cells (Suppl. Fig. 2B). Analogous effects occurred in HT29-D4 cells bearing V600E activating mutation on Ras effector Braf; PDIA1 silencing in these cells promoted increased superoxide production Fig. 1 Role of PDIA1 in oxidant generation by colon carcinoma cells with distinct levels of KRas activation. a KRas expression and activity in Caco2, HKE3, and HCT116 cells. Active KRas was pulled down with GST-RBD beads from lysates of serum-starved cells treated with EGF 25 ng/mL for 10 min. Aliquots of the lysates were blotted for total KRas and GAPDH as loading control. b RNAseq analysis of P4HB (PDIA1) gene in Caco2, HCT116, and HCT15 cells, showing mean expression in FPKM (Fragments Per Kilobase Million), P4HB fold-change expression in HCT15 vs. Caco2 (q < 0.000001) and HCT116 vs. Caco2 (q < 0.000001). c PDIA1 basal protein expression by western analysis, normalized for GAPDH (n = 3) and quantified using Odyssey software. d Intracellular PDIA1 titration, using Human P4HB ELISA Pair Set (SinoBiological). Test t **p < 0.01, (n = 4); e-g ROS production after 72 h of PDIA1 silencing, measured by HPLC analysis of DHE oxidation products. DHE oxidation produces, among many others, two major products: 2-hidroxyethidium (EOH), which is representative of superoxide species and ethidium, representative of other oxidant species. Ctrl negative, si-RNA control; si-PDI, si-RNA against PDIA1 protein. Representative immunoblots of PDI silencing for each cell type Test t *p < 0.05; **p < 0.01. For Caco2 cells, n = 3; HKE3 cells, n = 4; HCT116 cells, n = 5 (Suppl. Fig. 3D). To further address whether overactivated Ras associates with switch in the pattern of PDIA1mediated superoxide regulation, Caco2 cells were transfected with KRas G13D active form, which promoted significant increase of superoxide production (Suppl. Fig. 4A). It is well known that in Caco2 cells, Ras activation induces Nox1 expression and increases ROS production through the MEK-ERK pathway and GATA6 transcription factor 34 . Opposite to wild-type Caco2 cells, in which PDIA1 silencing diminished superoxide production ( Fig. 1e), transfection of the KRas G13D active form blocked such effect (Suppl. Fig. 4B); however, this experiment proved to be technically difficult due to simultaneous transfections of KRas plasmid and PDIA1 siRNA oligonucleotide. Thus, we assessed the effects of thiomuscimol (15 µM, 15 min), a thiol inhibitor acting as a non-specific but cell-permeable PDIA1 inhibitor 35 . Thiomuscimol lead to increase in superoxide production, similar to that observed in HCT116 cells after PDIA1 silencing (Suppl. Fig. 4C, D).

PDIA1 silencing sustains superoxide production in HCT116 through Nox1 NADPH oxidase
To investigate the source of superoxide in our cells, we first assessed the protein expression of Nox complex subunits. Immunoblot analysis showed that Nox1 was more expressed in HKE3 and HCT116 vs. Caco2 control, consistent with the previously described changes in Nox1 due to Ras activation 25 (Fig. 2a). Concerning regulatory Nox subunits (Fig. 2a, c), Nox organizer 1 (NoxO1, a p47phox analog) expression was significantly higher in HCT116 vs. other cell lines, while p47phox was more expressed in HKE3 and Caco2 (Fig. 2a). Caco2 cells expressed significantly more NoxA1 (p67phox analog Nox activator-1) than HCT116 and HKE3; however, p67phox expression was similar among these cells. Expression of RhoGTPases and their regulators was distinct among cell types: HCT116 cells expressed more Rac1 than Caco2 and HKE3, while RhoA expression showed an inverse pattern, with low expression in HCT116 cells. Meanwhile, Caco2 expressed less RhoG-DIα than HKE3 and HCT116 (Fig. 2b). Rac1 activity was higher in HCT116 cells vs. HKE3 (Fig. 2d).
To investigate possible roles of Nox1 in oxidant generation in HCT116 cells, we used the Nox1 peptide inhibitor NoxA1ds 36 . Incubation with this cell-permeable peptide led to decreased superoxide production in HCT116, which was statistically significant after PDIA1 silencing (Fig. 2g). Therefore, Nox1 complex contributes to superoxide production in HCT116 cells. These data are in line with results in HT29-D4 cells, in which superoxide increase after PDIA1 silencing was prevented by concomitant Nox1 silencing (Suppl. Fig.  3D). We next addressed whether PDIA1 silencing affected protein expression of Nox1 or its regulatory subunits. PDIA1 silencing in HKE3 and HCT116 cells did not alter NoxO1 protein expression (Fig. 2f). Rac1 protein expression was slightly decreased by PDIA1 silencing in HKE3, but not HCT116 cells (Fig. 2e). Interestingly, PDIA1 silencing promoted decrease in RhoGDIα expression in HKE3 but not in HCT116 cells (Fig. 2f), in line with the Rac1 results. We propose that

KRas overactivation bypasses PDIA1/Nox1 regulation by sustaining high Rac1 activity
To investigate whether Rac1 activation could account for the sustained Nox1 activation in HCT116 cells, HKE3 cells were transfected with overactive Rac1 G12V mutant in a way to mimic HCT116 cells. Contrarily to non-transfected HKE3 cells, HKE3 Rac1 G12V showed increased superoxide production 72 h after PDIA1 silencing (Fig. 3a), similar to results from HCT116 cells (Fig. 1g). Conversely, treatment of HCT116 cells with the Rac1 peptide inhibitor W56 37 decreased basal superoxide levels and prevented superoxide increase after PDIA1 silencing (Fig. 3b). These results suggest that constitutive high Rac1 activation may contribute to the switch of PDI-dependent regulation from supporting to limiting superoxide production, respectively, in cells with low levels vs. overactivated KRas.

PDIA1 interaction with KRas and Rac1
The results so far allow the proposal that PDIA1 may act as a servomechanism (oscillator) basally supporting, while in parallel posing a "capping" limit to superoxide generation, depending on the enhanced activations of Rac1 and KRas. We investigated possible interactions between PDIA1 and KRas. Using cavitation techniques (see Methods) able to preserve sensitive protein complexes and/or microdomains, given that KRas is known to localize to specific labile nanoclusters 38 , we showed that PDIA1 coimmunoprecipitated with KRas ( Fig. 3c) in all cell types, with enhanced detectable interaction in HCT116. PDIA1 immunoprecipitation yielded enhanced protein amounts in HCT116 vs. HKE3 cells, consistent with results from Fig. 1c, d. Moreover, PDIA1 also exhibited detectable interaction with Rac1 ( Fig. 3c). Similar interactions between PDIA1/KRas and PDIA1/Rac1 were also detected in HUVEC (Suppl. Fig. 5), while we have previously detected PDIA1-Rac1 interaction in VSMC 21 .

Functional effects of PDIA1 silencing on cell proliferation and migration
Since ROS closely correlate with cell proliferation, migration, and survival 39,40 , we sought to investigate functional readouts of PDIA1 effects in tumor dynamics. The spheroid assay, also termed 3D culture, can provide information of cell proliferation and evasion, mimicking cell interaction, and escape from the core tumor. The growth of tumor cell spheroids was investigated by measuring total spheroid area at T0 and T48 h, calculated as the ratio T48h spheroid area/T0 initial spheroid area. HCT116 growth was expectedly higher vs. HKE3 cells. PDIA1 silencing induced decreased spheroid growth for HKE3, but not for HCT116 cells (Fig. 4a, b). To assess cell evasion after spheroid formation, spheroids were placed on fibronectin 2D matrix and their areas assessed at T0 and T48 h. Cell evasion was calculated as (T48h total evasion area-T0 initial spheroid area)/T0 initial spheroid area. PDIA1 silencing promoted decrease in cell evasion in HKE3 and no modification in HCT116 (Fig. 4c, d). These data are consistent with results of a 2D random migration assay in HT29-D4 cells (Suppl. Table 1) and are in line with previous data in VSMC 21 .
Among the investigated pathways, Stat3 and GSK3β showed the most consistent results and were differentially affected by PDIA1 silencing between HKE3 and HCT116 (Fig. 5b, c). GSK3β (Glycogen synthase kinase-3 beta) is a constitutively active protein kinase inactivated by Ser9 phosphorylation. GSK3β supports Stat3 activation 41 . Stat3 is a transcription factor for many cytokines and growth factor receptors, activated by Tyr705 phosphorylation, which induces its dimerization, nuclear translocation, and DNA binding. In HKE3 cells, PDIA1 silencing induced Ser9 GSK3β phosphorylation (suggestive of inactivation) and reduced Tyr705 Stat3 phosphorylation (suggesting impaired activation) upon PDIA1 silencing. Conversely, in HCT116 cells, Ser9 phosphorylation of GSK3β was significantly elevated already at baseline and stayed elevated upon PDIA1 silencing. Meanwhile, Stat3 Tyr705 phosphorylation was also enhanced at baseline and remained high after PDIA1 silencing. Results with GSK3β were validated by immunoblot (Suppl. Fig. 8). These results indicate that GSK3β/Stat3 regulation is disrupted in HCT116 cells. Since Stat3 is a well-known target of Rac1 42 , we propose this may be a possible mechanism to sustain Stat3 activation in HCT116 cells.

Enrichment pathway analysis of protein interaction networks
To contextualize our results, we constructed a protein-protein interaction network based upon enrichment pathway analysis, fashioned using string functional association network, in which nodes signify each protein and edges represent protein interactions. Interactions among Nox1, PDIA1, RhoGDIα, Rac1, B-Raf, KRas, GSK3β, STAT3, and E-cadherin were investigated and confronted to the database (Fig. 6a). The network connecting these nine proteins of interest associated with more edges (16) than those predicted by the software (4), suggesting that these proteins interact among themselves at a degree higher than that expected from an analogous stochastic group. Such an enrichment indicates that these proteins display some degree of biological connection as a group. Importantly, this analysis showed Rac1 as a likely hub of interaction with most proteins. Top 10 ranked enriched proteins by KEGG pathway analysis (Fig. 6b) showed five cancerrelated pathways (black bars), including those for CRC. Gene Ontology (GO) analysis (Fig. 6c) identified proteins localized in the plasma membrane, focal adhesion, cell periphery, and cell junction.

Discussion
Our results suggest the possibility that PDIA1, known to correlate with sustained tumor growth and metastatization 12,45 , may play significant regulatory roles in redox-related transition from less to more aggressive stages of tumorigenicity. Increased aggressiveness correlates with disabled PDIA1-mediated oxidant generation, resulting in PDIA1-supported restriction of oxidant generation. Whether this process is a possible component of a tumoral escape program deserves further discussion. ROS production can support either pro-or anti-tumoral effects, depending on the type, levels, and sources of ROS, as well as associated activated pathways. ROS production is a wellknown hallmark of cancer initiation via DNA damage and genomic instability 46 . Redox signaling sustains crucial oncogenic processes, such as proliferation and migration 39,40 . On the other hand, high level of ROS can have anti-tumoral effects by inducing senescence and apoptosis 16 . Similarly, ROS generation interplays in different ways with responsiveness to anti-cancer drugs. Antioxidant supplementation shows no benefit and can even worsen cancer treatment in animal models 47,48 and patients 49 . Efficacy of radiotherapy or some chemotherapeutics depends on their ability to induce ROS production. In HT29-D4 and Caco2 cells, Nox1 silencing significantly decreases oxaliplatin efficiency 50 , while adjuvant ROS-generating molecules are promising 51 . In contrast, sustained ROS production may associate with chemoresistance. CRC resistant to oxaliplatin present increased basal Nox1 activity and supportive of cell survival 52 . Ras overactivation also presents a dual pattern of associated ROS-related effects. In Caco2 cells, Ras activation induces Nox1 expression through the MEK-ERK pathway and GATA6 transcription factor 34 . Ras overactivation-increased ROS production 26 may associate with senescence and apoptosis 53,54 , but also with cancer initiation and sustained tumorigenicity 55 . In NRK cells (rat kidney fibroblast), cancer initiation by KRas G12V requires Nox1, while KRas G12V -induced Nox1 upregulation requires Ras-Raf-MEK-ERK axis 56 . Therefore, Nox1 is important to maintain Ras-induced malignant transformation. Meanwhile, mitochondrial ROS also support Ras-dependant cancer cell transformation 57 . Mutant overactive KRas is also able to induce NRF2 upregulation, which buffers cellular oxidant levels but confers resistance to platine compounds 58,59 . The identification that the supportive vs. inhibitory patterns of PDIA1-dependent Nox-dependent regulation accompany Ras activation levels indicates that PDIA1 acts as an upstream determinant of redox signaling programs associated with distinct stages of tumorigenicity. Whether PDIA1-mediated servomechanism-like behavior at these distinct stages can modify tumor evolution remains to be determined.
Mechanisms involved in PDIA1-dependent superoxide modulation converge to RhoGTPases. We show that basal Rac1 levels closely correlate with the dual effects of PDIA1 on superoxide, since forced Rac1 overactivity in HKE3 induces PDIA1 to inhibit, while Rac1 inhibition in HCT116 induces PDIA1 to support Nox1-dependent superoxide. Since constitutive Rac1 G12V mutation likely bypasses its regulation by PDIA1, our experiments (Fig. 3) suggest that loss of Rac1 regulation by PDIA1 can be a key mechanism underlying transition from PDIA1-supported to PDIA1-inhibited ROS generation. In parallel, since Ras activity correlates with such PDIA1 effects, we speculate that Ras-mediated Rac1 activation underlies PDIA1 bypass and transition in oxidant regulation. PDIA1-Rac1 interaction accounts for PDIA1 effects on VSMC Nox1mediated processes, as PDIA1 silencing disables the associated Rac1 activation and PDIA1 interacts with Rac1 21 . Here, we confirmed a similar PDIA1-Rac1 interaction both in CRC and endothelial cells. Moreover, PDIA1 has been associated with significant effects on cytoskeletal regulation, which is the hallmark of RhoGT-Pase effects 60,61 . The importance of PDI-RhoGTPase convergence is further evident from our recent study showing an extremely conserved evolutionary pattern of gene clustering involving PDI and RhoGDI families. Since RhoGDIs are essential regulators of RhoGTPase activity and PDIA1 closely interacts with RhoGDIα 62 , RhoGDIs may be a mechanism whereby PDIA1 regulate Rac1, as indeed suggested by our expression data (Fig. 2c-e). These considerations further support Rac1 as a crucial node at the center of a signaling hub involving PDIA1, RhoGDIs, Ras, and other proteins, as indeed shown in our protein-protein interaction model (Fig. 6).
In HCT116 and HT29-D4 cells, in which ROS production stays elevated after PDIA1 silencing, there is associated insensitivity of cell migration to PDIA1 silencing (Fig. 5). An important mechanism supporting cell migration is EMT, known to associate with ROS production in breast cancer cells 63 . In melanoma, Nox1-derived ROS sustains and Nox1 inhibition reverts EMT 64 . PDIA1 silencing promoted switch to epithelial phenotype in HKE3 cells, but a switch to mesenchymal phenotype in HCT116 cells. It is possible that PDIA1 supports EMT through its effects on Nox1 and ROS. Our results are in line with previous reports in hepatocellular carcinoma showing that PDIA1 supports tumorigenesis by enhancing EMT through Grp78 downregulation 65 . Furthermore, E-cadherin protein expression negatively correlates with Stat3 activation in CRC 66 . Indeed, in our HKE3 cells, PDIA1 silencing induces Stat3 inhibition and increased E-cadherin expression. Rac1 sustains Stat3 activation; 42 we believe that Rasinduced sustained Rac1 underlies the lack of effect of PDIA1 silencing on Stat3 (Fig. 4). Basal saturation of these pathways may explain why HCT116 cells, despite exhibiting more mesenchymal phenotype after PDIA1 silencing, does not further enhance cell invasion. Of note, Stat3 activation has been correlated with Nox1 expression [67][68][69] . Overall, the GSK3β/Stat3 axis deserves further investigation regarding the pathways whereby PDIA1 interplays with Nox1 activation during Ras overactivation and EMT.
Altogether, our results indicate (Fig. 7) a novel dual effect of PDIA1 on Nox1-dependent superoxide regulation in CRC, as a servomechanism correlating with the extent of Ras activation. Such Ras-induced bypass of the otherwise normal support of Nox1 by PDIA1 seemingly involves an independent activation of Rac1. Several studies indicate that primarily adaptive responses, e.g., senescence, can be hijacked to promote tumor escape responses, such as stemness 70 . PDIA1 could thus act as a regulatory mechanism underlying adaptive redox switches from a tumor-suppressive to a vicious adaptive program promoting tumor escape. These results reinforce the emerging potential therapeutic implications of PDIA1 inhibition against cancer progression.

PDIA1 quantification by ELISA
In total, 2 x 10 6 cells was seed in 100 mm cell culture dish, 24 h after seeding cell were lysed in 300 µL of lysis buffer A (HEPES 20 mM, NaCl 150 mM, Glycerol 10%, triton 1%, EGTA 1 mM, MgCl 2 1.5 mM). Following cell lysis, soluble PDIA1 antigen was measured using the Human P4HB Pair Set enzyme-linked immunosorbent assay (ELISA) (SinoBiological Inc) according to the protocol described by the supplier. Briefly, 96-well microplate was coated with 100 µl per well of the diluted capture antibody and incubated overnight at 4°C. Thereafter, each sample was added for 2 h at room temperature. Following 100 µl of detection antibody conjugated to horseradishperoxidase (HRP) was incubated for 2 h at room temperature. Plates were washed three times after each incubation. Finally, 200 µl of tetramethylbenzidine (TMB) solution were added for 30 min, and optical density of each well were determined immediately using a microplate reader (SpectraMax 340, Molecular Devices) set to 450 nm. The values were determined according to a standard curve, and normalized according to the total protein dosage.

Coimmunoprecipitation Experiments
Cells 10 × 10 6 were grown to confluence in 150 mm plates, washed three times in PBS buffer and 2 ml of lysis buffer (20 mM Tris-HCl pH 7.8, 250 mM sucrose, 1 mM MgCl 2 , and 1 mM CaCl 2 ) supplemented with protease (1 mM PMSF, 1 µg/ml leupeptin and aprotinin) and phosphatase (50 mM sodium fluoride, 2 mM sodium orthovanadate, 10 mM sodium pyrophosphate) inhibitors. Cells kept for 20 min over ice were scraped and collected in a final volume of 5 ml. The cell suspension was transferred to a 35 -ml nitrogen cavitation bomb for 30 min in 400-psi nitrogen pressure on ice. Intact cells, large debris and nucleus were removed by centrifugation at 1 000 g for 10 min at 4°C. Lysates were incubated overnight at 4°C under agitation with 8 µg of PDIA1 antibody, followed by incubation with 70 μl of protein G-coated magnetic beads (ref. 28-9513-79 from GE Health Care) for 4 h at 4°C. Beads were successively washed in sucrose buffer to remove contaminating material, ressuspended in modified FLAG lysis buffer (50 mM Tris-HCl pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100 and 1% CHAPS) supplemented with a protease and phosphatase inhibitors. After 1 h of incubation at room temperature, Laemmli sample buffer was added and incubated at room temperature for additional 1 h. KRas was detected by immunoblot.

Detection of ROS production
Intracellular cell ROS production was assessed by HPLC analysis of dihydroethidium (DHE)-derived oxidation products, as described 30 . DHE oxidation produces, among many others, two major products: 2-hydroxyethidium (EOH), which is representative of superoxide species, and ethidium, representative of other oxidant species. Cells were starved for 4 h and incubated or not 2 h with 10 µM of the Nox1 peptide inhibitor NoxA1ds or 50 µM of W56 Rac1 peptide inhibitor. Cells were washed with HBSS without phenol red, Ca 2+ , and Mg 2+ , and incubated for 30 min with 100 μM DHE, plus the inhibitors, on HBSS without Ca 2+ and Mg 2+ . Cells were washed with cold PBS, harvested in 500 µl acetonitrile and centrifuged (12,000 × g for 10 min at 4°C). The homogenate was dried under vacuum and analyzed by HPLC with fluorescence detectors (Waters 2475 HPLC, Colum Synergi 4µ Polar-RP 80A from Allcrom ref. 00F-4336-E0). Quantification of DHE, EOH, and ethidium concentrations was performed by comparison of integrated peak areas between the obtained and standard curves of each product under identical chromatographic conditions. EOH and ethidium were monitored by fluorescence detection with excitation 480 nm and emission 580 nm, whereas DHE was monitored by ultraviolet absorption at 245 nm. Results were expressed as calculated EOH or ethidium concentrations (micromolar), normalized for consumed DHE (i.e., initial minus remaining DHE concentration in the sample).

PathScan ® intracellular signaling array
Signaling pathways associated with the effects of PDIA1 loss-of-function were investigated by the PathScan ® Intracellular Signaling Array Kit (ref. 7323), which is a slide-based antibody array optimizing the performance of sandwich immunoassays. The kit allows the simultaneous detection of 18 signaling molecules (described in Fig. 5), which are either phosphorylated or cleaved. Targetspecific capture antibodies were spotted in duplicate onto nitrocellulose-coated glass slides. Quantification of all spot intensities was performed using ImageJ software.

Cell migration assays spheroids
Three-dimensional cell invasion assay was adapted from previously published works 71,72 . Twenty-four hours after cell transfection with PDIA1, si-RNA cells are trypsinized, counted, and re-suspended in complete medium containing 2.4 mg/ml methylcellulose. In order to start the experiment with equivalent-size T0 spheroids, 1,500 cells per well were seeded for HKE3 vs. 1000 for HCT116. The suspension (100 µl) was added into each well of a Ubottom 96-well plate, allowing the formation of one spheroid per well. Twenty-four hours after plating (T0), spheroids were transferred to a flat-bottom 24-well plate coated with 10 µg/mL fibronectin. Pictures were taken at T0 and T48 h in a Olympus microscope, objective 2X. Invasion was quantified by measuring the area occupied by cells spheroid expansion at T0 and T48 h using ImageJ software. Spheroid expansion was calculated as: (T48 h total evasion area-T0 initial spheroid area)/T0 initial spheroid area. For the proliferation analysis, 24 h after plating (T0), spheroids stayed on a U-bottom 96-well plate with methylcellulose media, and spheroid surface areas were measured at T0 and T48 h using ImageJ software. Spheroid growth was calculated as: T48 h spheroid area/T0 spheroid area. T0 spheroids out of mean size range were excluded from the analysis.
Experiments of the lineages listed were retrieved from NCBI's SRA through searches during the months of November and December 2017. Samples containing metadata information indicating any type of treatment were discarded. Samples that were included had indications that they were experimental controls or did not show any metadata indicating otherwise. Expression correlation analysis of transcripts between samples was performed using the Poisson Distance and pheatmap functions of the PoiClaClu (v.1.0.2) and pheatmap (v.1.0.8) R packages. Samples that had very different behavior in relation to the majority of the same cell line or similar with different strains other than their own were discarded. That is, according to the position of the sample in the hierarchical clustering procedure and the distance. Quality control was done with FastQC (v0.11.5) and MultiQC (v1.0) with default settings. For sequence mapping, the HiSat2 (v2.0.5) aligner and the preformatted index of the reference GRCh38 release 84 of the H. sapiens genome from the Ensembl project was used, including dbSNP (b144) variants, splice site, and exon position information. For transcript assembly, StringTie (v1.3.1c) with strict GRCh38 annotation was used. The transcript data were tested for differential expression with the BallGown (v2.6.0) package in the R (v3.4.0) environment. A differential expression relevance cut was used for false discovery rates of less than 0.05, expression change rates greater than two, and FPKMs greater than one in at least half of the lineage samples.

Statistical analyses
Data are presented as mean ± SD. Comparisons were performed by paired Student t test, one-way ANOVA with Tukey's multiple comparisons test post hoc test using GraphPad Prism 7.0 (GraphPad Software Inc., CA, USA). Significance level was p ≤ 0.05.