Mechanism-based Proteomic Screening Identifies Targets of Thioredoxin-like Proteins*

Background: Trx (thioredoxin) is a prototypical oxidoreductase, but the functions of numerous Trx-like proteins are unknown. Results: A mechanism-based proteomic screening was developed to identify physiological targets of Trx-like proteins. Conclusion: Targets of Trx1, Rdx12, Txnl1, and Nrx1 were identified. Significance: Redox and non-redox targets could be identified by this method. Thioredoxin (Trx)-fold proteins are protagonists of numerous cellular pathways that are subject to thiol-based redox control. The best characterized regulator of thiols in proteins is Trx1 itself, which together with thioredoxin reductase 1 (TR1) and peroxiredoxins (Prxs) comprises a key redox regulatory system in mammalian cells. However, there are numerous other Trx-like proteins, whose functions and redox interactors are unknown. It is also unclear if the principles of Trx1-based redox control apply to these proteins. Here, we employed a proteomic strategy to four Trx-like proteins containing CXXC motifs, namely Trx1, Rdx12, Trx-like protein 1 (Txnl1) and nucleoredoxin 1 (Nrx1), whose cellular targets were trapped in vivo using mutant Trx-like proteins, under conditions of low endogenous expression of these proteins. Prxs were detected as key redox targets of Trx1, but this approach also supported the detection of TR1, which is the Trx1 reductant, as well as mitochondrial intermembrane proteins AIF and Mia40. In addition, glutathione peroxidase 4 was found to be a Rdx12 redox target. In contrast, no redox targets of Txnl1 and Nrx1 could be detected, suggesting that their CXXC motifs do not engage in mixed disulfides with cellular proteins. For some Trx-like proteins, the method allowed distinguishing redox and non-redox interactions. Parallel, comparative analyses of multiple thiol oxidoreductases revealed differences in the functions of their CXXC motifs, providing important insights into thiol-based redox control of cellular processes.

proteins containing CXXC motifs. Among them, Trx1 (for recent reviews, see Refs. 13 and 14) is an essential and ubiquitous protein, which in combination with thioredoxin reductase 1 (TR1) and peroxiredoxins (Prxs) comprises a key redox system in the cell. Known targets of Trx1 include Prxs, phosphatase and tensin homolog, ribonucleotide reductase (15), methionine-sulfoxide reductase (16), and a wide range of ribosomal and metabolic proteins (10,(17)(18)(19). Rdx12 or C17orf35, herein referred to as Rdx12, is a protein whose catalytic Cys is replaced with selenocysteine in fish (20). It forms a Rdx family with several other selenoproteins, i.e. selenoproteins H, T, V, and W. Recently, Rdx12 has been shown to be anchored to plasma membranes of cancer cells, promoting cell migration (21,22). A pulldown assay identified several Rdx12 targets in mouse liver lysate (20). Among them, glutathione peroxidase 1 (GPx1) was enriched in the CXXS mutant, suggesting that Rdx12 may function as a possible reductase of this enzyme. Thioredoxin-like protein 1 (Txnl1) is an intriguing protein, described to be negatively regulated by glucose deprivation (23) and to act as a redox sensor of endocytosis (24). Its C-terminal domain has been found to bind the regulatory particle of the proteasome (25)(26)(27)(28), whereas the N-terminal domain contains a CXXC motif (25,26). Elongation factor 1-␣1, a protein involved in protein synthesis and also degradation, was identified as a target of Txnl1 in HEK 293 cells (25), but other targets are unknown. Nucleoredoxin 1 (Nrx1) has been described to negatively regulate Wnt-␤-catenin (29) and Toll-like receptor 4 (30) pathways. Its other partners previously described include protein phosphatase 2A (31), transport protein Sec63 (32), and phosphofructokinase 1 (33). The latter two interactions may be redox-dependent, although the exact role of the CXXC motif of Nrx1 in these interactions is unclear.
Considering differences in functions of Trx-like proteins as well as differences in the approaches previously employed to identify their interactors, it is important to examine Trx-like proteins using a common proteomic approach and the same cellular system. The strategy employed here to identify targets of Trx1, Rdx12, Txnl1, and Nrx1 entails the expression of trapping mutants in their physiological locations, to in situ trap cellular targets. In addition, the expression of the corresponding endogenous proteins was knocked down. Following immunoaffinity purification under mild conditions, LC-MS/MS and computational analyses, we identified redox targets of Trx1 and Rdx12, whereas Nrx1 and Tnxl1 did not engage in mixed disulfides with cellular proteins. Thus, these studies provided important insights into the control of cellular processes through thiol-based redox regulation.

Antibodies and Recombinant Human Proteins-Anti-Txnl1
(ab26171), anti-Nrx1 (ab88753), anti-HA (ab9110), antichicken IgY-HRP (ab97140), and anti-Rdx12 (ab92499) were purchased from Abcam. Anti-AIF (sc9416) antibody was from Santa Cruz Biotechnology, anti-PDI (P7122) from Sigma, and anti-rabbit IgG-HRP and anti-mouse IgG-HRP from GE Healthcare. Anti-Prx2, anti-Prx5, and anti-GPx4 were from Frontier Antibodies. Anti-Trx1 and anti-TR1 were obtained from Covance. CXXC and CXXS recombinant human Trx1 and CXXU TR1 were prepared as previously described (34). Recombinant human AIF (AIF(⌬1-120)) (35) and human Mia40, with a His tag at their C terminus, were prepared after cloning AIFM1 and CHCHD4 (Mia40) from the HEK 293T cDNA library into pET28a vector (Novagen), both between EcoRI and SalI restriction sites. The expression vector was amplified in DH5␣ cells and the sequence was verified by DNA sequencing. Protein expression was carried out by transforming competent BL21 cells (Novagen), and inducing protein expression in 1.5 liters of culture with 0.4 mM isopropyl 1-thio-␤-D-galactopyranoside for 3 h at 37°C. Recombinant AIF was purified under native conditions (25 mM imidazole and 0.6 M NaCl in PBS) with HiTrap nickel-nitriloacetic acid-agarose resin (GE Healthcare). After cell lysis using an ultrasound probe, the His-tagged proteins were bound to nickel beads, washed with the lysis buffer, and eluted with 500 mM imidazole in the presence of 600 mM NaCl. The eluted yellow fraction was concentrated with a 30-kDa cutoff spin column (Millipore) and desalted (GE Healthcare). The concentration of purified AIF(⌬1-120) was spectrophotometrically quantified (⑀ 450 nm ϭ 12.12 mM Ϫ1 cm Ϫ1 ) (36) and the ratio A 270 /A 450 was determined to be 7.8. The eluted Mia40 fraction was desalted by dialysis. Mia40 concentration was determined using a Bradford reagent (Bio-Rad).
Trx-like Protein Construct-HEK 293 cDNA was synthesized from total RNA with Superscript II and oligo(dT) (Invitrogen) and used to clone Trx1, Txnl1, and Rdx12. Nrx1 was amplified from human NXN plasmid obtained from the Harvard Plasmid Database (clone ID number HsCD00082998). The whole CDSs of these genes were amplified and cloned into pCI-neo vector (Promega), downstream of the HA-TEV nucleotide sequence (37), after double digestion with XbaI and SalI (Fermentas) for 3 h at 37°C and ligation with T4 DNA ligase (Fermentas) for 1 h at 22°C. Competent Nova Blue cells (Novagen) were used to amplify the plasmids. To obtain the CXXS and SXXC mutants, mutagenic primers were designed with the Primer X software. The wild type sequence was used as template, and amplification was carried out with 15 cycles by Pfu Ultra High Fidelity DNA polymerase AD (Agilent). After DpnI (Fermentas) digestion, commercial competent DH5␣ cells (New England Biolabs) were transformed. The sequences were verified by sequencing.
Trx-like Protein Knockdown and Overexpression in Human Cell Lines-Knockdown of various Trx-like proteins was performed by transfecting siRNA duplexes (Dharmacon) targeting the untranslated regions of the genes with Dharmafect 1 reagent, in both HEK 293T and HeLa cells. To this end, 5 ϫ 10 6 cells were plated in 7 P100 dishes. The next day, they were transfected with 50 -100 nM siRNA using 1 l of Dharmafect 1/ml of transfection medium (ϳ7 ml/P100). The medium was kept with the transfection reagent until the next day, when cells were expanded to 16 P150 dishes. On the following day, they were transfected with the expression vectors (ϳ25 g/P150), with calcium phosphate (for HEK 293T) or Lipofectamine 2000 (Invitrogen, for HeLa cells), according to the manufacturer's instructions. Four P150 dishes were typically used for each construct.
Immunoaffinity Purification-Two days after plasmid transfection, the cells were washed with ice-cold PBS and then with ice-cold 20 mM N-ethylmaleimide in PBS. Cells were harvested Targets of Thioredoxin-like Proteins by scraping and incubated with this solution for 10 min at 4°C. The pelleted cells (3,000 ϫ g, 5 min, 4°C) were lysed with 9 -10 ml of lysis buffer (25 mM Tris, pH 7.4, 5 mM MgSO 4 , 150 mM NaCl, 0.3% Triton X-100, 20 mM N-ethylmaleimide and Roche protease inhibitor without EDTA) for 30 min under mild agitation at 4°C (38 -40). The supernatant (16,000 ϫ g, 15 min, 4°C) was collected and its protein concentration was determined by the Bradford assay. Approximately 15-22 mg of total protein were incubated with 100 l of anti-HA-agarose resin slurry (monoclonal antibody, Sigma). After 3 h under agitation at 4°C, the beads were recovered and washed with wash buffer (25 mM Tris, pH 7.4, 5 mM MgSO 4 , 150 mM NaCl, 0.1% Triton X-100), using spin columns (Sigma). The final 2 washes were made with the same wash buffer composition, but without Triton X-100. Finally, the beads were incubated with (4 l/construct) TEV protease (Invitrogen) in 100 l of wash buffer overnight at 4°C, under gentle agitation. The beads were then centrifuged, washed (10,000 ϫ g, 1 min), and the eluates (final volume of 150 l/construct) were collected, aliquoted, and frozen at Ϫ80°C.

SDS-PAGE, Gel Staining and Western
Blotting-Samples (eluates corresponding to ϳ2 mg of the initial protein input or 30 g of total cell lysate proteins to Western blotting) were resolved in 10, 12, or 4 -12% gradient precast Novex gels (Invitrogen), with either MES or MOPS running buffer (Invitrogen), under constant voltage. Reducing or non-reducing conditions were determined by the Novex SDS-loading buffer. Reducing conditions were achieved by the addition of ␤-mercaptoethanol. Gels were stained with silver nitrate using a Pierce Silver Stain kit. To perform Western blotting, proteins were transferred onto PVDF membranes (Invitrogen) for 1 h at 40 V. The membranes were blocked with 5% nonfat milk and incubated with primary and secondary antibodies diluted in 2% nonfat milk. Reactions were developed with West Pico or Femto Pico chemiluminescent kits (Pierce) and exposed to an autoradiogram or a CCD camera. When indicated, membranes were stripped with Restore it (Pierce) and reprobed with a different antibody.
LC-MS/MS-The complex mixture of eluted proteins was freed of detergents by precipitation with methanol/chloroform (41). Proteins were digested with trypsin and subjected to LC-MS/MS with a hybrid linear ion trap Orbitrap Velos Pro mass spectrometer (Thermo Fisher Scientific). The Thermo.RAW files were converted to .mzXML files and searched against a concatenated database (2009-07-06_REVipi.HUMAN.v.3.60.fasta) containing 160,900 protein entries using the Sequest search algorithm (version 28). Fifty ppm precursor and 0.8-Da fragment ion tolerances were used. Methionine oxidation was considered as a variable modification and trypsin was used as the enzyme allowing up to two missed cleavages per peptide. A classifier based on linear discriminant analysis was utilized to distinguish correct and incorrect peptide identifications considering numerous parameters such as Xcorr, DCn, precursor mass error, and charge state as previously described (42). After correlating all peptides with their corresponding proteins, proteins were scored based on their multiplied peptide LDA probabilities. The sorted list was filtered based on reversed protein hits to maximally comprise only 1% false positives (43). An in-house program was used to collapse the peptide data set down to the minimum number of proteins sufficient to explain all the data. Two parameters were used to analyze the dataset obtained for each Trx-like protein: the total number of peptides detected in each construct and the normalized spectral abundance factor (NSAF) (44).
The scores are clearly related as they utilize the same measures; however, by their formal definition, the Z-score is better suited to rank the "prominence" of the trapping mutant (NSAF_CS) over all the other samples, whereas the CS-score is better fit to score and rank abundant targets with different significance (i.e. in different samples). Thus, we considered both Z-scores (to effectively filter out non-enriched NSAF_CS samples) and CS-scores (to better separate significant targets). Overall, to be considered as top targets, candidates should: (i) have Z-score Ͼ2, (ii) CS-score Ͼ10, (iii) show peptide count in the trapping mutant Ն3, and (iv) be present in less than 3 protein datasets (considering the HEK 293T datasets for Trx1, Rdx12, Txnl1, and Nrx1, and the HeLa dataset for Trx1). HEK 293T datasets for Trx1 under basal versus H 2 O 2 conditions were treated separately, using criteria i-iii.
NADPH and NADH Consumption-NADPH and NADH (Sigma) consumption was measured spectrophotometrically at 340 nm. The positive control reaction mixture contained 0.2 mM NADPH, 1 M TR1, 2 M Trx1, and 160 M insulin in 50 mM Tris buffer, pH 7.2, containing 2 mM EDTA, at 30°C. The putative TR1 activity of AIF was tested by replacing TR1 with 1 M AIF(⌬1-120) and using NADH or NADPH.
Direct Interaction between Recombinant Trx1 and AIF or Mia40 -Reaction mixtures consisted of 10 M reduced CXXC or CXXS Trx1 and 2.5 M AIF(⌬1-120) (pre-oxidized or nonoxidized) or 10 M Mia40 (pre-oxidized or non-oxidized), in 50 mM Tris buffer, pH 7.2, at 37°C for 15 min. In addition, Trx1 and non-oxidized AIF(⌬1-120) were incubated in the presence of 0.5 mM H 2 O 2 . Reactions were stopped by the addition of non-reducing loading buffer.
Recombinant AIF and Mia40 Oxidation-AIF(⌬1-120) (107 M) was oxidized by 140 M H 2 O 2 in PBS for 5 min at room temperature. Mia40 (100 M) was oxidized either by 2 mM diamide or 150 M H 2 O 2 in PBS for 15 min at room temperature. The reaction with H 2 O 2 was stopped by the addition of 22 units/ml of catalase, whereas diamide was removed by using an Amicon ultrafiltration unit (Millipore).

RESULTS AND DISCUSSION
Approach to Identify Targets of Trx-like Proteins-Among the 136 predicted thiol oxidoreductases encoded in the human genome (3), we identified 36 Trx-like proteins containing the CXXC motif. For the current study, we selected four of these proteins: Trx1, Rdx12, Txnl1 and Nrx1. They were chosen because, besides having a single CXXC motif, they were predicted to be globular proteins. Three of these proteins are poorly characterized with regard to thiol oxidoreductase functions, whereas Trx1 is a well known oxidoreductase that served as a positive control in the proteomic analyses. Parallel analysis of these four proteins helped in the identification of specific interactions, as proteins detected in multiple datasets (likely to be nonspecific targets) could be filtered out.
We adapted a proteomic strategy to trap redox targets of thiol oxidoreductases in situ, under physiological conditions in cells. The approach was based on the trapping mutant strategy, i.e. the formation of a stable mixed disulfide between the nucleophilic Cys of a Trx-like protein and a Cys of a target protein, which can be achieved when the resolving Cys is mutated in a Trx-like protein (Fig. 1A). Because the second Cys in the CXXC motif is known to function as the resolving Cys in all well studied cases of Trx-like thiol oxidoreductases, the trapping mutants were expected to be the CXXS mutants, whereas the SXXC mutants could not form mixed disulfides and therefore served as controls. However, in case the second Cys in the CXXC motif was the catalytic Cys, the use of both CXXS and SXXC forms of the proteins still allowed us to detect the biological targets (which would be enriched in the CXXS samples, over the SXXC; or vice versa, depending on the nature of the catalytic Cys).
The overall approach (Fig. 1B) consisted of expression of a trapping mutant of each Trx-like protein in a cell line, HEK 293T and/or HeLa, in which endogenous expression of the same protein was decreased by siRNA (Fig. 1B). An additional control included cells lacking an expression construct (Fig. 1C). A diminished expression of endogenous Trx-like proteins was important to limit a possibility that they resolve mixed disulfides between exogenous thiol oxidoreductases and their targets. All expressed Trx-like proteins carried a HA tag, and were immunopurified with an anti-HA antibody coupled to agarose beads. In addition, we introduced a TEV protease cleavage site that allowed the release of expressed proteins together with their targets and interacting proteins complexes under very mild conditions, preserving the interactions. The proteins in eluates were identified by LC-MS/MS, and a computational analysis ranked top targets of each Trx-like protein (Fig. 1B). A somewhat related proteomic approach has recently been described (11); however, our method offered an advantage of having low endogenous expression of Trx-like proteins and using mild elution conditions, thereby allowing trapping and enrichment of less abundant targets, and distinguishing redox and non-redox interactors. Furthermore, in contrast to the methods relying on gel separation prior to MS/MS analysis, we directly analyzed proteins by LC-MS/MS following tandem affinity purification. Finally, based on the comparison of spectral counts of each candidate across 3 constructs (as well as controls lacking expression constructs), we calculated target enrichment, supporting detailed, quantitative, and comparative analyses of the datasets.
We reduced the expression of Trx1, Txnl1, or Nrx1 by 50 -100% by an siRNA knockdown approach, as shown by Western blotting, whereas the endogenous expression of Rdx12 was already undetectable in HEK 293T cells (Fig. 2). We then expressed the trapping mutants of each protein or controls (Fig. 2) and, following immunoaffinity purification and digestion with TEV protease, the proteins eluted were resolved by denaturing gel electrophoresis under reducing conditions (Fig.  3). The results showed that the most abundant proteins eluted were the Trx-like proteins themselves and TEV protease, as expected. Overall, there was no apparent difference in electrophoretic profiles between samples from cells expressing proteins with the CXXC motif and mutant forms of each protein.
One exception was a band (just below 28 kDa) observed only in the CXXS mutant of Trx1. In addition, overexpression of Txnl1 and Nrx1 led to isolation of an increased number of proteins, compared with the respective mock controls (Fig. 3). These eluates were then digested with trypsin and subjected to protein identification by LC-MS/MS. A first analysis of the resulting datasets (supplemental Tables S1-S4) showed that the number of targets was different among different Trx-like proteins ( Table 1). Each target protein detected in each dataset was assigned a Z-score and a CS-score (directly measuring the relative abundance of peptide counts in CXXS samples, as compared with all other samples; see "Experimental Procedures"), which in combination were used to detect and rank the most enriched targets in the CXXS mutants (Tables 2-5). The number of top candidates for Trx-like proteins ranged from 17 to 27 ( Table 1).
Analysis of the Trx1, Rdx12, Txnl1, and Nrx1 datasets (supplemental Tables S1-S4) revealed that the top targets for each protein were non-redundant (Tables 2-5), an evidence of good target specificity provided by our approach. In addition, targets differed with regard to enrichment for CXXS trapping mutants, as demonstrated by the CS-scores (Tables 2-5).
Trx1 Targets-Known Trx1 targets, such as Prxs (Prx1, Prx2, Prx4, and Prx5 were detected), TR1, and several ribosomal proteins, were remarkably enriched in the CXXS mutant dataset in HEK 293T cells (Table 2). Although TR1 is the Trx1 reductase, the approach used was sufficient for its identification. It is possible that Trx1 exists in redox equilibrium with TR1, thereby  . Immunoaffinity purification of Trx-like proteins with their targets. HEK 293T cells were transfected with siRNA targeting the untranslated regions of the Trx-like protein genes, followed by overexpression of the corresponding Trx-like proteins (with CXXC, CXXS, and SXXC constructs or nothing ϭ mock). Cells were lysed and the HA-tagged proteins were immunopurified with agarose beads coupled to anti-HA. The complexes were eluted with TEV protease, and resolved in reducing SDS-PAGE. Bands were detected with silver staining. The overexpressed proteins (arrow) and TEV protease (*) are indicated. The number of top targets refers to the number of proteins enriched in the CXXS trapping mutant and selected by filtering out proteins that (i) appeared in more than 3 datasets, (ii) had a Z-score Ͻ2; (iii) had a CS-score Ͻ10; and (iv) showed a TSC Ͻ3. Tables S1-S4) Top targets (Tables 2-5)   Trx1  379  17  Rdx12  715  27  Txnl1  1655  22  Nrx1 1140 18 Targets of Thioredoxin-like Proteins FEBRUARY 27, 2015 • VOLUME 290 • NUMBER 9

Proteins detected (supplemental
trapping a fraction of the latter protein by the CXXS mutant. Indeed, TR1 has also been detected as a Trx1 target in yeast cells (45). The same experimental approach was applied to HeLa cells (supplemental Table S5). Top Trx1 targets in HEK 293T ( Table 2) and HeLa (Table 6) cells showed significant overlap, e.g. Prxs, TR1, and ribosomal proteins appeared as top candidates in both cell lines. Other targets detected in both HEK 293T and HeLa datasets include elongation factors and apopto-sis inducing factor 1 (AIF), which is a mitochondrial protein. In addition, mitochondrial intermembrane space import and assembly protein 40 (CHCHD4 or Mia40), a protein involved in oxidative import of mitochondrial intermembrane space (IMS) proteins (46,47), emerged as a target in the HEK 293T dataset. Some interactions involving Trx1 were assessed by Western blotting to validate our proteomic approach. The results demonstrated the specific interaction of the CXXS form of Trx1 with Prx2, Prx5 and TR1 (Fig. 4, left panels). Trx1 is known to promptly respond to an oxidative insult, regenerating sulfhydryl groups in target proteins. We hypothesized that oxidative conditions may promote interprotein disulfide bonds involving Trx1. Thus, in an attempt to increase the fraction of Trx1 targets and facilitate their identification, we incubated HEK 293T cells expressing various Trx1 forms in the presence or absence of 0.1 mM hydrogen peroxide. Western blotting experiments showed that hydrogen peroxide treatment led to an increased trapping of Prxs and TR1 by the trapping mutant (Fig. 4, right    Targets of Thioredoxin-like Proteins panels). However, these proteins were detected also in the case of CXXC and SXXC forms of Trx1 (Fig. 4, right panels). We attribute this observation to an enhanced affinity of Trx1 for its targets as they become more oxidized. To test this hypothesis, we analyzed the immunopurified fractions by SDS-PAGE. The electrophoretic profiles obtained confirmed an increased number and amounts of target proteins, compared with non-oxidative conditions (Fig. 5). A MS analysis was carried out with such samples (supplemental Table S6). In agreement with the Trx1 results in the absence of stress (Tables 2 and 6), computational analyses confirmed the enrichment of Prx1, Prx2, Prx4, Prx5, TR1, and AIF (supplemental Table S7). However, these targets were not more enriched in the CXXS mutant under oxidative stress conditions (supplemental Table S7), presumably because they were already present in considerable amounts in the CXXC and/or SXXC constructs, as demonstrated by Western blotting (Fig. 4) and by the analysis of the immunopurified proteins (Fig. 5, supplemental Table S6). Overall, these results indicate that our approach successfully detects known Trx1 targets and provides a dynamic view of trapped targets of Trx1. However, the use of hydrogen peroxide did not improve the approach regarding target enrichment by the trapping mutant. Therefore, we focused on characterizing the targets of Trx-like proteins under basal conditions. Novel Targets of Trx1-AIF was present in all Trx1 proteomic datasets (in both cell lines), and Mia40 was detected in all HEK 293T datasets under basal conditions. Thus, these proteins emerge as major new targets of Trx1. An interaction between AIF and Trx1 was also detected in a yeast two-hybrid study (48). Western blotting of immunopurified eluates confirmed that AIF specifically binds to the CXXS mutant of Trx1 (Fig. 6A). Following hydrogen peroxide treatment, this interaction was enhanced, and similarly to other targets tested, AIF could also interact with both CXXC and SXXC forms of Trx1. These latter interactions were much weaker than with the CXXS mutant and appeared to represent increased affinity of the oxidized AIF toward Trx1 (Fig. 6A), as previously discussed. AIF is a mitochondrial protein, whose N terminus faces the mitochondrial matrix, whereas its larger portion is in the IMS. The domain organization of AIF resembles that of TR1, with an NADHbinding domain located between two FAD-binding domains (49). AIF has 3 cysteine (Cys) residues, although the roles of these residues in its structure or function are unknown. Therefore, a reasonable possibility for the interactions observed in our experiments could be that AIF is a Trx1 reductase in the IMS. Although Trx1 is primarily found in the cytosol, it as well as other primarily cytosolic oxidoreductases, such as glutaredoxin 1 (50) and TR1 (51), and also mitochondrial matrix Trx2 (52), have been found in the IMS. The IMS is known to be a very redox-active compartment and has a system for disulfide bond formation.
To test for possible Trx1 reductase activity of AIF, a recombinant human AIF, without the mitochondrial localization sequence and transmembrane domains (AIF (⌬1-120)), but still FIGURE 4. Validation of the proteomic approach by immunoblotting of the immunoaffinity purified targets of Trx1 under basal and oxidative conditions. HEK 293T cells were transfected with siRNA targeting the untranslated regions of Trx1 gene, followed by transfection with Trx1 constructs (with CXXC, CXXS, and SXXC constructs or nothing ϭ mock). Two days after transfection, cells were treated or not with 0.1 mM H 2 O 2 in PBS for 3 min before lysis and immunoaffinity purification with agarose beads coupled to anti-HA. The complexes were eluted with TEV protease, and resolved in reducing SDS-PAGE. Resolved proteins were immobilized in PVDF membranes and probed with the indicated antibodies. A whole cell lysate was used as a positive control of the immunoblotting. IP, immunoprecipitation. FIGURE 5. Immunoaffinity purification of Trx1 with their targets under basal and oxidative conditions. HEK 293T cells were transfected with siRNA targeting the untranslated regions of the Trx1 gene, followed by transfection with Trx1 constructs (with CXXC, CXXS, and SXXC constructs or nothing ϭ mock). Two days after the transfection, cells were treated or not with 0.1 mM H 2 O 2 in PBS for 3 min before lysis and immunoaffinity purification with agarose beads coupled to anti-HA. The complexes were eluted with TEV protease, and resolved in reducing SDS-PAGE. Bands were detected with silver staining. The overexpressed Trx1 (arrow) and TEV protease (*) are indicated. FEBRUARY 27, 2015 • VOLUME 290 • NUMBER 9

Targets of Thioredoxin-like Proteins
with functional ability to induce apoptosis (35), was prepared. An assay with NADPH or NADH, recombinant human Trx1, bovine insulin, and recombinant human TR1 was employed, and the reaction was followed by cofactor consumption. Upon the addition of TR1 to the reaction mixture, clear consumption following a lag phase was observed, with the slope of approximately Ϫ0.04 ⌬A 340 nm /min. Subsequently, the absorbance increased due to insulin precipitation. However, when TR1 was replaced with AIF(⌬1-120), no NADPH or NADH consumption was detected. These results suggest that AIF(⌬1-120) is a poor Trx1 reductase compared with TR1 under these experimental conditions. Therefore, a second hypothesis was tested, i.e. that AIF is reduced by Trx1. To characterize an interaction between Trx1 and AIF involving mixed disulfide, in vitro reactions with recombinant proteins were carried out. Human recombinant CXXC and CXXS Trx1 forms were first reduced with DTT. Then, the reduced CXXC and CXXS proteins were incubated with AIF(⌬1-120). The latter protein was either in a reduced state or was oxidized with 0.5 mM H 2 O 2 in the reaction mixture or was oxidized prior with 0.15 mM H 2 O 2 at pH 7.0 for 30 min. Formation of the AIF(⌬1-120)-Trx1-mixed disulfide was analyzed by Western blotting following non-reducing SDS-PAGE. Immunoblotting with anti-Trx1 resulted in the detection of 12and 28-kDa bands, which represent Trx1 monomer and dimer, respectively. Interestingly, two bands with the molecular masses between 62 and 98 kDa were detected in all CXXS Trx1 reactions, and in the reaction of CXXC Trx1 with AIF(⌬1-120) in the presence of H 2 O 2 (Fig. 6B). The mixed disulfide Trx1-AIF(⌬1-120) has the expected mass of ϳ70 kDa, which is consistent with the more intense lower band (labeled with an arrow). This blot was further probed with anti-AIF antibody (Fig. 6B, left panel). Although the non-reacted AIF appeared as the most evident band (molecular mass between 49 and 62 kDa), a band slightly above 62 kDa was present in all CXXS Trx1 reactions and in the reaction of CXXC Trx1 with AIF(⌬1-120) in the presence of H 2 O 2 , similar to the anti-Trx1 immunoblotting. The fact that incubations of CXXS Trx1 with non-oxidized AIF also presented the ϳ70 kDa band suggests that AIF(⌬1-120) has at least one reducible Cys residue. These data indicate that the CXXS mutant of Trx1 can trap oxidized AIF. Thus, Trx1 is a reductant of oxidized AIF(⌬1-120), which places Trx1/AIF in a redox regulatory pathway, in agreement with the findings showing that low levels of AIF are associated with increased oxidative stress (53,54). Finally, we also validated the interaction between Mia40 and Trx1, using recombinant proteins. Reduced CXXC or CXXS Trx1 was incubated with Mia40. The incubation of CXXS Trx1 with oxidized Mia40, particularly with diamide, led to the detection of a band with the molecular mass between 28 and 38 kDa, which is consistent with the predicted molecular mass (33 kDa) of the Mia40-Trx1 mixed disulfide (Fig. 6C). This band was absent when Mia40 was not oxidized as well as in all CXXC Trx1 incubations. This result confirms that Mia40 is a Trx1 redox target in vitro. In accordance, it has been recently demonstrated that Trx1 is partially responsible to keep cytosolic Mia40 in a reduced state to allow its import into the mitochondrial IMS (55). Rdx12, Txnl1, and Nrx1 Targets-The three other Trx-like proteins were analyzed based on their top candidates (Tables 3-5) in HEK 293T cells. In the case of Rdx12, several mitochondrial and ribosomal proteins appeared as top targets (Table 3). Among them, GPx4 was especially interesting, because another glutathione peroxidase, GPx1, has already been found as a Rdx12 target (20). Like GPx1, GPx4 can be reduced by GSH, but it has propensity to oxidize Cys residues in other proteins. The Rdx12-GPx4 redox interaction was confirmed by immunoblotting (Fig. 7A), potentially opening new avenues on the possible participation of Rdx12 in the control of phospholipid oxidation.
Txnl1 top targets, in contrast, showed no ribosomal proteins, instead including several proteins related to microtubules (Table 4). These data are in line with a recent demonstration that Txnl1 indirectly associates with microtubules in Toxoplasma gondii (56). The most striking feature of this dataset was, however, the low enrichment of targets in the CXXS mutant. Indeed, in contrast to Trx1 and Rdx12, Txnl1 showed only a very limited enrichment of targets in the CXXS mutants, as can be estimated by the distribution of CS-scores (Table 4;  compare with Tables 2 and 3 for Trx1 and Rdx12, respectively). In addition, because these top candidates (Table 4) were also found in the CXXC form of Txnl1, and were absent in the mock sample (supplemental Table S3, Fig. 3), they may represent non-redox interactions. In agreement with this view, it has been reported that whereas the N-terminal domain of a recombinant Txnl1 has a reductase activity, the whole protein does not (57). It is possible that Txnl1 reduces other forms of oxidized proteins or compounds. For example, the reduction of methionine sulfoxide or hydrogen peroxide by designated thiol oxidoreductases does not lead to inter-protein mixed disulfides, and as such similar redox activities (if any, for Txnl1) would go undetected in our protein-protein interaction based approach. Nev-ertheless, our Txnl1 dataset (supplemental Table S3) may provide useful information regarding non-redox interactions of Txnl1.
As for the fourth protein in our analysis, Nrx1, the top targets detected were mostly proteins that associate with RNA (Table  5). However, some of the proteins that were previously found to interact with Nrx1 (29,32,33) were not identified in our analysis. Although these interactions have been reported to depend on the CXXC motif, our results would suggest that they might not result from a reductive mechanism. Indeed, phosphofructokinase 1, a recently described Nrx1 partner (33), appears in our dataset as enriched in the CXXC, but not in the CXXS forms of Nrx1 (supplemental Table S4). Thus, similar to Txnl1, Nrx1 does not seem to be a reductant of protein disulfides or oxidized Cys residues, and the targets found to interact with Nrx1 could represent cases of non-redox-dependent protein-protein interactions. Protein-disulfide isomerase (PDI or P4HB), a target whose NSAF were higher in the overexpressed Nrx1 lysates (supplemental Table S4), was confirmed to be enriched in the Nrx1 immunocomplexes, independent of its CXXS motif (Fig.  7B). PDI is a well characterized redox protein of the endoplasmic reticulum. But recent studies have shown that its functions are not restricted to assisting protein folding, with the protein being redox-active also in the cytosol (58). The role of the nonredox interaction of PDI with Nrx1, however, remains unknown. Overall, this finding indicates that our method also identifies non-redox targets for the studied proteins.
Concluding Comments-The approach developed to detect targets of Trx-like proteins proved to be especially useful when a thiol oxidoreductase has a CXXC motif that interacts with oxidized protein thiols, such as interactions involving Trx1. Indeed, in addition to confirming known redox targets of Trx1, we were able to identify and validate AIF and Mia40 as novel redox targets of Trx1. We found that AIF had no reductase activity with oxidized Trx1, whereas Trx1 could reduce oxidized AIF in in vitro assays. The identities of the Cys residues of AIF involved in the interaction with Trx1 remain to be determined. A possible implication of this interaction, which should occur in the IMS, is that AIF functions as a redox regulatory molecule and is regenerated by Trx1. Mia40 was also validated as a Trx1 target. This association is likely involved in keeping cytosolic Mia40 in a reduced state to allow its import into the IMS. In regard to other Trx-like proteins examined, we identified GPx4 as a novel, previously unreported Rdx12 redox interactor. Experiments with Txnl1 and Nrx1, in contrast, revealed no redox targets, a finding made possible due to the comparative nature of our approach (utilizing different redox partners and non-redox controls, analyzed in parallel). Our Txnl1 and Nrx1 datasets present a list of novel non-redox targets for both Txnl1 and Nrx1, of which PDI-Nrx1 was further confirmed. Although we believe that Nrx1 and Txnl1 are not canonical cellular reductases, a more detailed investigation of redox targets of Trx-like proteins could be achieved by a combination of our method with a stable isotope labeling by amino acids in cell culture-based approach. For instance, a comparison of the targets determined in the case of CXXS mutants of Trx-like proteins, in the absence or presence of hydrogen peroxide, could identify redox targets that are affected by peroxide treatment.

Targets of Thioredoxin-like Proteins
To be noted, because the targets were trapped in situ, they could represent natural, physiological targets for all Trx-like proteins investigated in this study.
On a different, more methodological note, a salient feature of our proteomic screen is the ability to detect both reductants and oxidants of thiol oxidoreductases. It was previously thought that only the oxidants could be trapped. Our findings also suggest that multiple appropriate controls are needed before concluding on the identity of the targets as well as on the redox interactions involving specific Cys residues in Trx-like and target proteins. Clearly, many previous proteomic screens have not considered these issues and may have both false positive and false negative targets. Finally, our study may be viewed as a pilot screen, which proved to be useful in characterizing redox and non-redox targets and can be applied to characterize all thiol oxidoreductases containing CXXC motifs.