Selenium-Encoded Isotopic Signature Targeted Profiling.

Selenium (Se), as an essential trace element, plays crucial roles in many organisms including humans. The biological functions of selenium are mainly mediated by selenoproteins, a unique class of selenium-containing proteins in which selenium is inserted in the form of selenocysteine. Due to their low abundance and uneven tissue distribution, detection of selenoproteins within proteomes is very challenging, and therefore functional studies of these proteins are limited. In this study, we developed a computational method, named as selenium-encoded isotopic signature targeted profiling (SESTAR), which utilizes the distinct natural isotopic distribution of selenium to assist detection of trace selenium-containing signals from shotgun-proteomic data. SESTAR can detect femtomole quantities of synthetic selenopeptides in a benchmark test and dramatically improved detection of native selenoproteins from tissue proteomes in a targeted profiling mode. By applying SESTAR to screen publicly available datasets from Human Proteome Map, we provide a comprehensive picture of selenoprotein distributions in human primary hematopoietic cells and tissues. We further demonstrated that SESTAR can aid chemical-proteomic strategies to identify additional selenoprotein targets of RSL3, a canonical inducer of cell ferroptosis. We believe SESTAR not only serves as a powerful tool for global profiling of native selenoproteomes, but can also work seamlessly with chemical-proteomic profiling strategies to enhance identification of target proteins, post-translational modifications, or protein-protein interactions.

Introduction

Selenium is a chemical element that was first discovered by the Swedish chemist Jöns Jacob Berzelius in 1817.[1] Since then, researchers over the past 200 years have established that selenium is an essential micronutrient for human health, the imbalance of which causes severe pathophysiological conditions. Excessive ingestion of selenium is the main culprit of both “alkali disease” and “blind staggers”,[2] and selenium deficiency is also detrimental, resulting in a variety of diseases including white muscle disease[3] and mulberry heart disease[4] in livestock, and Keshan’s disease[5] and Kashin-Beck disease[6] in humans.

It is generally accepted that the biological essentiality of selenium is mediated through a unique class of selenium-containing proteins named “selenoproteins”, which are the major organic form of selenium in cells.[1] Selenium is biosynthetically incorporated into selenocysteine (Sec), a natural amino acid structurally identical to cysteine except with an atom of selenium in place of the sulfur. Sec is genetically encoded by a stop codon “UGA” and inserted into certain positions of selenoproteins during translation with guidance of a special cis-acting Sec insertion sequence (SECIS) element on mRNAs.[7] Compared to Cys, the selenol side chain group of Sec has enhanced reactivity, which enables it to accelerate enzymatic reactions and react with oxygen and related reactive oxygen species (ROS) in a readily reversible manner.[8] Therefore, selenoproteins are positioned to perform important redox-active and antioxidant functions in cells.[1]

The first selenoprotein discovered was glutathione peroxidase 1 (GPx1),[9] which was found to play a crucial role in the overall recovery of cells exposed to oxidative stress. Subsequently, a number of additional selenoproteins were also identified by biochemical experiments from a wide range of species.[10,11] A landmark bioinformatics study by Gladyshev and colleagues established that there are 25 selenoproteins in total within the human genome based on strong features of SECIS elements and UGA codons.[7,12] However, given that many of these selenoproteins are expressed in very low abundance, identification directly from complex proteome (cells or tissues) samples remains challenging, and a global atlas of the selenoproteome with tissue distribution is still missing.

Mass-spectrometry-based (MS-based) shotgun proteomics has become a major research tool to identify and quantify proteins from complex proteome samples. It is generally operated in the data dependent acquisition (DDA) mode where only ions with the top N intensities per full MS scan are selected for fragmentation and acquisition of MS/MS spectra.[13] This creates a strong bias against peptides with extremely low abundance, such as those from selenoproteins containing the Sec residue (selenopeptides). Targeted proteomic profiling has emerged as a powerful tool to address this limitation. In this methodology, elements with unique natural isotopic distributions are incorporated into peptides of interest to produce distinctive isotopic envelope signatures (groups of isotopically related peaks), which can be recognized either manually or by computational algorithms. Ions with desired isotopic envelopes are added to an inclusion list for targeted analysis regardless of their actual intensities, which can dramatically increase the sensitivity of detection.[13] Such examples can be traced back to as early as 2000 when Aebersold and co-workers used a dichloride tag to discriminate peptides with and without a cysteine residue from yeast peptide samples.[14] A monobromide-cleavable tag was also used by Hang and co-workers to enrich newly synthesized proteins in bacteria.[15] Recently, Bertozzi and colleagues developed a computational method named “isoStamp” to detect mass envelopes of peptides perturbed with a dibrominated chemical tag in complex proteomes[16,17] and applied it in glycoproteomics for intact glycopeptide discovery and analysis.[18] Woo and co-workers developed an isotopically encoded enrichment tag to enable targeted analysis of binding sites for the nonsteroidal anti-inflammatory drugs in proteomes.[19]

We reason that targeted proteomics should be highly suitable for selenoproteome profiling. First, selenoproteins are often expressed with extremely low abundance. Second, most of the selenoproteins contain only a single but functionally critical selenocysteine, which further reduce the possibility of detecting the actual selenopeptide by shotgun proteomics. Lastly, the element of selenium has six stable isotopes with detectable distribution (74Se, 0.89%; 76Se, 9.37%; 77Se, 7.63%; 78Se, 23.77%; 80Se, 49.61%; 82Se, 8.73%) which, when incorporated into peptides, can produce a unique isotopic envelope pattern. In the case of selenopeptide, a distinct mass envelope is readily observable if the sulfur atom in the side chain of a cysteine is replaced by selenium (Cys to Sec) (Figure a). With a robust method to automatically recognize selenium-encoded mass spectra, we can bypass the abundance bias and specifically subject them for fragmentation to generate MS/MS spectra for peptide identification

Figure 1

Conceptual workflow and algorithm of SESTAR. (a) The introduction of selenium induces significant change on the isotopic envelope of a proteogenic peptide. (b) SESTAR enables improved detection of selenium-containing peptides in shotgun proteomics. The envelopes with selenium-encoded isotopic patterns are first detected by SESTAR, which subsequently can be compiled into an inclusion list for targeted fragmentation. (c) Extraction of isotopic envelopes from shotgun-proteomic data by LC-MS/MS analysis. Each full mass scan is used as a separate unit for envelope detection, and peaks separated by one isotopic unit are considered as isotopically related. Envelopes for a selenium-containing peptide, a coeluting proteogenic peptide, or a separated proteogenic peptide were drawn in red, orange, or purple, respectively. (d) For each experimentally observed envelope, the mass is calculated and used to theoretically simulate two predicted envelopes, one for a selenium-containing envelope and the other for a proteogenic envelope. Two scores are calculated to decide whether the observed envelope matches to a selenium-containing envelope: the similarity score (SS) describes the similarity between the observed envelope and the simulated selenium-containing envelope, and the discrimination score (SD) describes the uniqueness of the similarity to only the selenium-containing envelope but not the proteogenic counterpart.

In this work, we report the development of a targeted proteomic approach termed “selenium-encoded isotopic signature targeted profiling” (SESTAR). The core component of SESTAR is a computational algorithm that is able to detect ions with the distinctive isotopic envelope pattern introduced by selenium in full-scan mass spectra. We then directed these ions with selenium-encoded isotopic envelopes for targeted fragmentation to enhance the sensitivity of their detection in complex proteome samples (Figure b). We first validated the performance of SESTAR by spiking synthetic selenopeptides into cell lysates with a series of dilutions and showed that the method can improve the sensitivity of detection by at least 1 order of magnitude as compared to the DDA-based shotgun analysis. When operated in targeted analysis mode with an inclusion list, SESTAR was able to dramatically enhance the detection of selenopeptides in both whole proteomes and chemically enriched proteomes. We further applied SESTAR to screen 66 data sets as collected in the Human Proteome Map[20] and provided a comprehensive atlas of selenoprotein distribution across different cell lines and tissues. We showed that SESTAR could also be applied to detect MS/MS spectra with unique selenium-encoded isotopic patterns, and the prefiltering of these spectra before standard database searches can significantly reduce both search space and time from information-dense MS data. Lastly, we applied SESTAR to aid chemical-proteomic analysis of selenoprotein targets of RSL3, a potent inducer of cell ferroptosis, and found that the compound can covalently modify the active-site Secs of several selenoproteins besides GPX4. We believe that SESTAR will not only be a powerful tool for the profiling of native selenoproteomes, but also be highly compatible with chemoproteomic profiling methods to facilitate detection of novel proteins of interest, post-translational modifications, as well as protein–protein interactions from complex proteome samples.

Results

SESTAR Algorithm

The aim of SESTAR is to detect selenium-encoded isotopic patterns from raw LC-MS/MS data through an algorithm containing two components: envelope extraction and pattern recognition. First, each full-scan spectrum was extracted from raw LC-MS/MS data, and isotopic envelopes belonging to individual peptides were identified (Figure c). We intentionally chose not to operate on three-dimensional chromatographic peaks because many selenopeptides are with extremely low abundance and would more likely be trimmed off by the peak extraction criteria. Every isotopic envelope we identified from full MS spectra is then compared to the two theoretically simulated envelopes based on the element composition calculated using the “averagine” assumption,[21] explicitly including or excluding the selenium atom. Thus, two score matrices are used to make the judgment. The similarity score SS) corresponds to the similarity between the observed envelope and the simulated selenium-containing envelope; the discrimination score (SD) reflects the uniqueness of the similarity to the selenium-containing envelope but not the nonselenium counterpart (Figure d) (see the Supporting Information methods for details).

Validating the Performance of SESTAR with Synthetic Selenopeptides

We first tested whether the SESTAR algorithm was able to detect selenopeptides in a predefined benchmark system. Two selenopeptides from naturally occurring selenoproteins SELW(VVYCGAUGYK) and SELT(FQICVSUGYR) were chemically synthesized, and both showed the expected unique selenium-encoded isotopic pattern when dissolved in water and analyzed by a high-resolution mass spectrometer (Figure S1a). These peptides were then spiked in a series of dilutions (500, 100, 50, 10, and 5 fmol) into 2.5 μg of HeLa lysate, corresponding to an approximate proteome/selenopeptide ratio (m/m) from 1:4 × 103 to 1:4 × 105. These samples were subjected to standard shotgun LC-MS/MS with a DDA setting of TOP20 (top 20 intensities selected for fragmentation and MS/MS analysis), and the resulting data were analyzed by SESTAR (Figure S1b). In the standard DDA shotgun analysis, selenopeptide ions of SELT and SELW were no longer selected for fragmentation when the dilutions are below 50 and 10 fmol, respectively. However, SESTAR was able to recognize the selenium-encoded isotopic patterns of these peptides at a dilution as low as 10 fmol for SELT and 5 fmol for SELW, which improves the detection sensitivity by 5- and 2-fold, respectively (Figure a). As expected, both SS and SD decreased as the selenopeptides were more diluted, finally approaching the cutoffs of 10 and 1/6, respectively (Figure b). Notably, under these diluted conditions, the ion intensities of the corresponding selenopeptides were ranked only 310th and 197th in the full MS spectra, highlighting the limitation of analyzing low-abundance selenopeptides by traditional DDA shotgun proteomics. (Figure c).

Figure 2

SESTAR recovers known selenopeptides from complex proteome samples. (a) Two synthetic selenopeptides SELW(VVYCGAUGYK) and SELT(FQICVSUGYR) were spiked in a series of dilutions (500, 100, 50, 10, and 5 fmol) into 2.5 μg of HeLa cell lysates and analyzed by DDA-based shotgun proteomics or SESTAR. If the selenopeptides can be identified with either approach, it was marked with a black dot. (b) SESTAR score distribution of these two synthetic selenopeptides at indicated concentrations, showing that SS ≤ 10 and SD ≤ 1/6 is a reasonable cutoff. (c) Representative full-scan mass spectra showing that the corresponding peaks for SELT and SELW were only ranked 310th and 179th, respectively, in terms of ion intensity and cannot be selected for fragmentation by standard DDA-based analysis. Their selenium-encoded isotopic envelopes shown in the zoom-in view can be detected by SESTAR. (d) Recovery of native selenopeptides identified from human proteome map data sets by SESTAR. The standard database search by ProLuCID identified 10 unique native selenopeptides from different cells and tissues for a total of 67 times and SESTAR was able to detect strong selenium-encoded isotopic envelopes in full MS spectra of these peptides for 62 times. The distributions of the corresponding SESTAR scores are shown in the neighboring green box. (e) Heat-map showing the false positive rates of identifying selenium-encoded isotopic envelopes by SESTAR from the selenoprotein-free A. thaliana proteome vary with different combinations of SS and SD as the score cutoff. SESTAR showed an overall low FPR.

Recovery of Native Selenopeptides from the Human Proteome

We next applied SESTAR to detect native selenopeptides from human proteomes. The Human Proteome Map was generated in 2014 with the analysis of 30 histologically normal human samples using DDA-based shotgun-proteomic strategies.[20] In total, proteins encoded by 17 294 genes were identified, accounting for approximately 84% of the total annotated protein-coding genes within the human genome. We downloaded 1715 raw data files corresponding to 17 adult tissues and 6 primary hematopoietic cells (Table S1) for SESTAR analysis in order to generate a global selenoproteome atlas.

It should be noted that native selenopeptides were rarely identified from proteome analysis, likely due to the limitation that many standard search engines do not recognize Sec as an extra amino acid (“U” as the one-letter code) during the database search. Given both Sec and Cys can react with thiol blocking reagents,[22] we bypassed the problem by manually converting all Sec to regular Cys in the database and searching with two different dynamic modifications on Cys (57.021 47 and 104.9659 Da for alkylation on Cys and Sec, respectively). We used ProLuCID[23] followed by DTASelect[15] to search through all of the 1715 raw files and identified 11 unique selenopeptides from different primary cells and tissues for a total number of 67 times (one “time” denotes one unique selenopeptide identification from one specific tissue sample, e.g., three times were counted for SELH, which was identified from three tissues or primary cells—CD4 T cells, ovary, and testis) (Table S2). When we ran SESTAR through these data sets with the cutoffs of SS ≤ 10 and SD ≤ 1/6, we were able to identify corresponding spectra for these selenopeptides for 59 times, accounting for a recovery rate of 88% (Table S2). Five unassigned identifications had low ion intensity and therefore had only one intact envelope that passed the score cutoff (Figure d). The remaining three cases scored slightly worse than the cutoffs (SS ≤ 10.25).

Estimating False Positive Rate of SESTAR

It is obvious that the two score cutoffs can impact the balance between detection sensitivity and false positive rate by SESTAR, and they therefore need to be carefully chosen. In order to evaluate the false positive rate (FPR) of SESTAR, we prepared a proteome sample from Arabidopsis thaliana, a plant predicted to contain no selenoprotein genes and, therefore, a proteome free of selenoproteins. We analyzed the digested proteome of A. thaliana by DDA-based shotgun proteomics and ran SESTAR to count the number of selenium-encoded isotopic patterns identified with various combinations of the score cutoffs, allowing us to estimate the FPR of SESTAR. It should be noted that the FPR obtained here is an overestimation because we cannot formally exclude the possibility of other forms of selenium-containing proteins (e.g., a post-translational modification with selenium or potential bacterial contaminations) from the sample preparation (see the Supporting Information methods). As shown in Figure e, even at the least stringent cutoffs of SS ≤ 15 and SD ≤ 1/4, only 4095 envelopes were identified by SESTAR that calculates to a FPR of 2.7%. At the more stringent end with SS ≤ 8 and SD ≤ 1/9, there were only 41 envelopes that can be identified by SESTAR (FPR ≪ 1‰), and under this condition, 50% of native selenopeptides identified from the Human Proteome Map could be recovered by SESTAR (Figure S2). These results collectively demonstrate that SESTAR was able to recognize selenium-encoded isotopic patterns with great sensitivity and robustness.

Discovery of Additional Spectra for Native Selenopeptides from the Human Proteome Map

SESTAR was able to recover spectra for more than 90% of the native selenopeptides from the Human Proteome Map database that have been identified by traditional database search workflow. During the same analysis, SESTAR also identified a large number of selenium-encoded isotopic envelopes, many of which do not have ensuing MS/MS fragmentations and therefore are not amenable for database searching. We reason that these mass spectra might come from additional native selenopeptides as the human genome has 25 predicted selenoprotein products with 36 Sec sites accounting for 32 and 2 unique fully tryptic selenopeptides with mono and dual Sec sites, respectively. We therefore tried to match these full MS spectra with selenium-encoded isotopic envelopes to the predicted precursor masses of known selenopeptide sequences. Depending on the levels of confidence, the assignment of these spectra can be divided into three groups. (1) There was an associated MS/MS spectrum, but it needs manual validation; for example, SESTAR detected a matching envelope for the selenopeptide “GFVCIVTNVASQUGK” from GPX4 in adult monocyte (Figure a). Though it had an associated MS/MS, the peptide–spectrum match did not pass the standard cutoffs set by the search engine. We manually assigned the corresponding b and y fragment ions, and confirmed its identity as the selenopeptide from GPX4. (2) There was no associated MS/MS spectrum, but its retention time is similar to that of a selenopeptide identified from another data file; for example, the selenopeptide “VVYCGAUGYK” from SELW was identified by SESTAR in adult CD8 T cells with a retention time at 37.16 min, and was identified by ProLuCID in adult colon tissues with a similar retention time (36.37 min). (3) There was no associated MS/MS spectrum, but the detected envelopes have very good SESTAR scores. This group of peptides could, in principle, be verified by targeted analysis with an inclusion list. Overall, SESTAR was able to identify matching full MS spectra for 17 unique selenopeptides from 16 selenoproteins across these primary cells and tissues for a total of 133 times. This represents a dramatic increase in coverage as compared to detection by the traditional database search (11 unique selenopeptides from 10 selenoproteins, identified for 67 times) (Figure b and Table S3). Comparison to transcriptome profiling data by RNA-seq[24] (https://www.proteinatlas.org/) showed that the selenoproteins detected by ProLuCID and SESTAR tend to be with higher transcription levels as expected (Figure c,d and Figure S3). Furthermore, additional selenoproteins whose spectra are identified only by SESTAR have either low transcription levels (such as SELV and TXNRD2) or are predicted to have unique subcellular localizations (such as SELT, SELN, and SELK, all membrane proteins and SELP, a secreted protein), further highlighting the detection power of our isotope-encoded pattern recognition strategy (Figure c).

Figure 3

Comprehensive selenoproteome atlas enabled by SESTAR. (a) The corresponding MS/MS spectrum of the selenopeptide (GFVCIVTNVASQUGK) from GPX4 in adult monocytes shows a unique selenium-encoded isotopic envelope that was detectable by SESTAR but not by database search. The y10+ fragmentation ion, which contains the selenocysteine, also displays a selenium-encoded isotopic envelope. (b) Venn diagram showing the number of selenopeptides detected by ProLuCID and SESTAR, respectively, from Human Proteome Map data sets covering 17 adult tissues and 6 primary hematopoietic cells. SESTAR was able to identify native selenopeptides for 71 more times from these samples. (c) Comparison of selenoprotein detection by SESTAR with transcriptome analysis by RNA-seq SESTAR was able to detect most of the selenoproteins with reasonable mRNA levels. (d) Averaged transcription level of selenoproteins (data are represented as median with interquartile range; mean value is marked by a red star), which were categorized into three groups: (1) validated by the regular database search; (2) additional detections by SESTAR only; and (3) not detected by either method. Selenoproteins detected by the database search or by SESTAR tend to have higher transcription level than those not detected. Secreted proteins were excluded in the calculation because they were secreted after maturation so that no correlation would be expected between the transcription and protein level. (e) Number of detected selenoproteins categorized by tissue or cell type; the number of selenoproteins detected by ProLuCID only, SESTAR only, or both are colored in green, orange, or brown, respectively. For those identified by SESTAR only, the numbers were further broken into three subcategories: with manual validation of MS/MS, with similar retention time, and with confident SESTAR scores.

The improvement in selenopeptide detection is observed across all tissue and cell types (Figure e). In more than 2/3 of the tissues and primary cells analyzed, SESTAR detected full mass spectra for at least 5 selenopeptides per sample, some of which are supported with manually validated MS/MS spectra or similar retention times. In particular, no selenopeptides were identified from pancreas by the traditional database search; however, SESTAR was able to detect envelopes with strong isotope patterns in this tissue that could be assigned to 10 unique selenopeptides based on either similar retention times or envelope scores. Consistent with the fact that GPX1 and GPX4 have broader expression profiles,[7] spectra of selenopeptides from these highly abundant selenoproteins were detected by SESTAR from 21 and 18 out of the 23 tissues, respectively. Another less well-studied selenoprotein, SELW, was detected in 15 tissues, suggesting it may also have a ubiquitous role in redox regulation. In contrast, another functionally cryptic selenoprotein, SELH, was only detected in 3 tissues (CD4+ T cells, adult ovary, and adult testis), and the narrow expression profile may indicate a specific role for this protein in reproductive organs. Considering there are eight membrane selenoproteins that are challenging to analyze by the experiment procedure applied in the project, SESTAR was able to detect full MS spectra matching a large percentage of selenopeptides as predicted from the human genome. This comprehensive atlas of selenoproteomes can provide researchers with important information to further explore the biological functions of selenoproteins, especially for those which remain functionally unannotated.

Application of SESTAR in Targeted Proteomics

We next demonstrate that SESTAR can work seamlessly with targeted analysis to improve the detection of native selenopeptides from proteomes (Figure a). We digested mouse liver lysates with trypsin and analyzed one aliquot first by regular LC-MS/MS. Data were then processed by SESTAR for pattern search. We generated an inclusion list of all selenium-encoded isotopic envelops detected by SESTAR and used it to direct the targeted LC-MS/MS analysis on a replicated sample. Regular LC-MS/MS analysis could only identify three selenopeptides from GPX1, GPX4, and SELW. With the targeted analysis instructed by SESTAR, selenopeptides from four additional selenoproteins (MSRB1, TRXR1, TRXR2, and SEPP1) could be validated (Figure b). For example, the selenopeptide (FUIFSSSLK) from MSRB1 has a precursor ion (m/z = 568.75, RT = 47.99 min, fraction 1) which displays a strong selenium-encoded isotopic pattern. Despite interference by a coeluting peptide with similar m/z values, SESTAR confidently selected this specific envelope for targeted fragmentation, and the MS/MS spectrum confirmed its sequence as expected (Figure b).

Figure 4

Application of SESTAR in targeted proteomics. (a) Schematic workflow of the application of SESTAR in two targeted proteomic or chemoproteomic experiments: (1) detection of native selenopeptides in mouse whole proteomes; and (2) detection of native selenopeptides from chemically enriched mouse liver proteomes. (b) Regular shotgun proteomics identified three native selenopeptides (from GPX4, SELW, and GPX1) in mouse liver proteomes while SESTAR-directed targeted profiling detected four more native selenopeptides (from SEPP1, TRXR1, TRXR2, and MSRB1); right is a representative isotopic envelope of native selenopeptide (FUIFSSSLK) from MSRB1 in the full-scan mass spectrum and its associated MS/MS spectrum generated by targeted fragmentation. (c) SESTAR-directed targeted profiling detected two more native selenopeptides (SPS2 and MSRB1) from chemically enriched mouse liver proteomes.

We also applied the same strategy on samples enriched by chemical proteomics. Recently, an activity-based protein profiling method was developed to chemically label native selenocysteines by an iodoacetamide probe under low pH conditions in order to enrich selenopeptides for proteomic analysis (Bak et. al, Cell Chem. Biol.2018, in press, doi: 10.1016/j.chembiol.2018.05.017). A total of 5 selenopeptides were identified from soluble proteomes of mouse liver using the traditional DDA method. We reanalyzed the same sample by SESTAR and detected two additional selenopeptides with excellent selenium-encoded isotopic patterns (Figure c), which were confirmed by targeted fragmentation (Figure S4). Furthermore, we synthesized an N-hydroxysuccinimide (NHS) ester probe containing a selenium atom (Figure S5a) and used it to chemically label lysines in a mouse liver proteome. The targeted analysis enabled by SESTAR was able to improve the detection of chemically tagged selenium-containing peptides by 2-fold (Figure S5b).

Detection of Selenium-Encoded Tandem Mass Spectra by SESTAR

The development of modern mass spectrometers enables collection of spectra at faster speed and higher resolution, which inevitably increase the time of database searching by orders of magnitude. In order to solve this problem, faster search engines such as MSFragger have been developed,[25] and an alternative approach would be to filter out irrelevant MS/MS spectra before the database search to reduce search time. For example, machine learning was used to select tandem mass spectra of potential phosphorylated peptides[26] for database searching, which reduces both the search space and time by half without losing detection power.

When we manually verified the tandem mass spectra of native or chemically labeled selenopeptides, we observed that certain daughter ions exhibit similar selenium-encoded isotopic patterns as those spectra from the full MS scan (Figure a). These results confirm the existence of selenium in the target peptides and further inspired us to apply SESTAR at the tandem MS/MS level to improve the detection of selenopeptides. Four data sets from the Human Proteome Map were chosen, and the standard database search identified 20 MS/MS spectra that were mapped to 13 native selenopeptides from a total of 3.55 × 105 MS/MS spectra on average per data set. We filtered these spectra by SESTAR and kept only the ones with selenium-encoded isotopic patterns for the following database search (Figure S6a). The preliminary filtering reduced the number of MS/MS spectra and search time, on average per data set, by 13- and 8.5-fold, respectively (Figure S6b). The power of detection remained unchanged as all the 13 native selenopeptides could still be identified, and only three redundant MS/MS spectra with low quality were missed. We also tested this strategy in a sample of chemically labeled selenium-containing peptides by the Se-NHS ester probe. The initial filtering by SESTAR reduced the number of MS/MS spectra to be searched and the search time by 3- and 3.5-fold, respectively (Figure S6c). Compared to the standard database search, more probe-labeled peptides were identified from the enriched MS/MS spectra by SESTAR, which was probably due to the shifted score distribution favoring the detection of selenium-containing peptides after spectra for the nonlabeled peptides were trimmed out. Collectively, all these data suggest that SESTAR can operate at both full MS and tandem MS/MS levels to enable targeted profiling that can greatly improve the detections of either native selenopeptides or chemically labeled selenium-containing peptides.

New Selenoprotein Targets of RSL3 Identified by SESTAR-Aided Chemical Proteomics

Lastly, we applied SESTAR to direct targeted analysis of selenoproteomes modified by RSL3, a potent inducer of cell ferroptosis. Ferroptosis is a recently discovered nonapoptotic cell death that is hallmarked by its dependence on intracellular iron and elevated lipid peroxidation.[27] Stockwell and colleagues discovered that RSL3 inhibits the activity of GPX4, a selenoprotein whose function is to reduce hydroperoxide species resulting from lipid peroxidation, providing the first link between selenoprotein and ferroptosis.[28,29] More recently, Ingold et al. elegantly showed that the utilization of selenocysteine in the active site of GPX4 is indispensable for preventing hydroperoxide-induced ferroptosis.[30] Due to the unique chemical reactivity of selenocysteines, many selenoproteins have been postulated with similar redox-regulating and ROS-eliminating activities in cells.[7] We therefore hypothesized that, in addition to GPX4, RSL3 might target some other selenoproteins at their active-site selenocysteines to induce ferroptosis, and SESTAR-enabled targeted chemical-proteomic analysis would be an ideal approach to explore such a possibility (Figure a).

Figure 5

Application of SESTAR to aid targeted profiling of selenocysteine sites covalently modified by RSL3. (a) Left: scheme showing that RSL3 targets the active-site selenocysteine site of GPX4 to induce ferroptosis. Other selenoprotein targets of RSL3 remain to be identified. Right: basic workflow for targeted profiling of selenocysteine sites covalently modified by RSL3 by SESTAR-aided TOP–ABPP. (b) Chemical structures of (1S, 3R)-RSL3, a ferroptosis inducer and its bio-orthogonal (1S, 3R)-RSL3-alkyne probe. The synthesized (1S, 3R)-RSL3-alkyne probe was able to induce ferroptosis with similar potency as the original compound does. (c) List of selenoprotein targets covalently modified by the RSL3-alkyne probe identified by SESTAR-aided TOP–ABPP at both protein and selenopeptide levels. There are 10 selenoproteins that can be identified as the targets of the RSL3-alkyne probe. Seven selenopeptides were identified to be covalently adducted by the RSL3 probe at the selenocysteine site by SESTAR-aided TOP–ABPP analysis, including those by standard ProLuCID search (green dots), those by SESTAR-directed parallel reaction monitoring (PRM) analysis (orange dots), and those by SESTAR-directed manual analysis (yellow dots). New selenoprotein targets of RSL3 that were identified in this study were shaded by cyan color in the table. (d) Manual analysis of the MS/MS spectrum of the RSL3-alkyne probe-adducted selenopeptide from SELT found an ion (indicated by red arrow) with strong selenium pattern whose mass was 280.19 Da less than that of the precursor ion. Based on the probe structure, it was predicted to be generated by unexpected fragmentation at the ester bond (red dotted line in the left structure) in the probe-adducted selenopeptide. Daughter ions from this broken adduct (labeled by boxed b and y) were also observed in the same MS/MS spectrum, which greatly increases its complexity for standard database search. Manual assignment of these ions properly allow identification of more selenopeptides from the PRM data, including SELS, SELM, and EPT1.

Previously Yang et al.[28] showed that only the (1S, 3R)-RSL3 but not the (1R, 3R)-RSL3 was able to induce ferroptosis, and when they used a fluorescein derivatized analogue of (1S, 3R)-RSL3 to perform affinity purification on cell lysates, only two selenoproteins, GPX4 and SELT, were identified as creditable targets; the approach did not identify the actual adducts to the active-site selenocysteines of these two proteins either. To circumvent this limitation, we synthesized a bio-orthogonal (1S, 3R)-RSL3-alkyne probe with minimum modifications to the core structure of RLS3 (Figure b), and the probe, when implemented in a tandem orthogonal proteolysis–activity-based protein profiling (TOP–ABPP) strategy[31] aided by SESTAR (Figure S7a), should enable target profiling of RSL3 with a site-specific precision directly in living cells.

We verified that the synthesized (1S, 3R)-RSL3-alkyne probe was able to induce ferroptosis with similar potency as the original compound does (Figure b). The probe was then used to treat HT1080 cells for different lengths of times (15, 30, 60, 180, or 360 min), and its proteome reactivity was confirmed by in-gel fluorescence (Figure S7b). Following the standard TOP–ABPP protocol, proteomes from the probe-treated HT1080 cells (1 μM for 360 min) were conjugated with an acid-cleavable azide-biotin tag,[18] and the enriched proteomes were subjected to tandem proteolysis by trypsin and formic acid treatment, respectively. While analysis of the trypsin-digested sample can reveal target of RSL3 at the protein level, the acid-cleaved sample contains actual probe-adducted peptides which allows mapping of detailed sites of modification by RSL3. We first analyzed the samples from trypsin digestion by LC-MS/MS and identified 10 selenoproteins including GPX4 and SELT, the two known targets identified by Yang et al. previously[28] (Figure c). When the acid-cleaved sample was analyzed by regular DDA-based shotgun proteomics, we identified based on the MS/MS spectra the direct adducts of the RSL3 probe with the active-site selenocysteine of three selenoproteins, GPX4, TRXR1, and SELK (Figure c). To date, modification of RSL3 on the active-site selenocysteine of GPX4 was only confirmed with mutagenesis study,[29] and our data provided the first piece of analytical evidence for this interaction by mass spectrometry.

We further applied SESTAR to screen the raw LC-MS/MS data of the acid-cleaved sample and directed any additional MS spectra with selenium-containing isotopic patterns for a second round of targeted analysis by paralleled reaction monitoring (PRM) (Figure S7a). Researching with ProLuCID on the PRM data identified one more selenopeptide adducts from SELT. Furthermore, a deeper analysis of the PRM data revealed that ions with masses which are about 280.19 Da less than the expected adduct were repeatedly identified in the MS/MS spectra of the identified selenopeptides (e.g., those from SELT, GPX4, TRXR1, and SELK) (Figure d, Figure S8). Based on the probe structure, it was predicted that they were generated by unexpected fragmentation at the ester bond in the probe-adducted selenopeptides. Moreover, daughter ions from this “unexpected” broken adduct were also observed within the same MS/MS spectra, which greatly increased their complexity for analysis by standard database search methods. (Figure d, Figure S8). For example, a low-quality peptide–spectrum match was found for the probe adduct of EPT1’s selenopeptide (KNPSDULGMEEK) when only considering the daughter ions from full adduct. However, if the daughter ions from the unexpected cleavage adduct were taken into consideration, the confidence for the match was greatly improved (Figure S8). With help of such manual analysis, the probe-adducted selenopeptides were identified for 3 more selenoproteins, including SELS, SELM, and EPT1 (Figure c). These examples highlight the unique advantage of SESTAR in identifying target selenopeptides of interest when the modification was unexpectedly changed due to structural instability. Collectively, our SESTAR-enabled targeted chemical-proteomic analysis revealed that RSL3 could covalently modify at least 5 more selenoproteins at their active-site selenocysteines as well as a list of reactive cysteines in proteomes (Table S4), and the roles of these additional modifications in mediating RSL3-induced ferroptosis will be subjected to functional investigation in the future.

Discussion

In this study, we developed a method named “SESTAR” to specifically detect mass spectra of native selenopeptides in proteomes by recognizing their special isotopic patterns. As compared to the traditional database search method, SESTAR could dramatically improve the detection of native selenopeptides from shotgun-proteomic data sets collected by the Human Proteome Map. This allowed us to generate a comprehensive atlas of native selenoproteins from different cell types and tissues. We further demonstrated that SESTAR, in combination with the targeted proteomic analysis driven by inclusion lists, could improve the detection of selenium-containing peptides from chemically enriched or labeled proteomic samples. We also showed that SESTAR could be applied at the tandem MS/MS level to keep only spectra with selenium-encoded isotopic patterns before database search, which greatly reduced search space and time. Lastly, we applied SESTAR to direct targeted chemical-proteomic analysis of the selenoproteome modified by RSL3 and found additional active-site selenocysteines are covalently adducted by the ferroptosis-inducing compound in addition to that of GPX4.

More and more studies have shown that selenoproteins play important roles in biology, which are mainly mediated by their active-site selenocysteines. Due to their low abundance and multiple oxidative forms, detection of native selenopeptides from proteomic samples remains challenging. Low-pH iodoacetamide chemical labeling has recently been demonstrated as an effective method to enrich native selenopeptides for proteomic (Bak et. al, Cell Chem. Biol.2018, in press, doi: 10.1016/j.chembiol.2018.05.017). Here, we took a novel computational approach to apply the isotope-encoded pattern search to enable global profiling of native selenoproteomes with unprecedented sensitivity and coverage. In addition, our method has also proven highly complementary to the aforementioned chemical-proteomic strategy. These tools not only allow accurate quantification of the expression profile of selenoproteins from different cells and tissues, but also enable focused analysis of specific modifications or conversions on selenocysteines upon environmental perturbations, such as drug treatment or nutritional deprivation.[32] In this regard, our SESTAR-aided targeted chemical-proteomic analysis of RSL3-reactive selenoproteomes revealed new potential links between selenoproteins and cell ferroptosis, which remains to be functionally characterized in the future.

It has been well-established that the major organic form of selenium in a biological system is selenocysteine,[33] and bioinformatics and biochemistry analysis have identified 25 selenoproteins in humans.[7,12] However, the possibility of other selenium-containing bio-macromolecules, functioning in biological systems through undiscovered mechanisms, cannot be formally excluded. Our preliminary analysis of data sets from Human Proteome Maps have discovered more than 1000 MS spectra with strong selenium-encoded isotopic patterns that cannot be assigned to any known selenoproteins. They may originate from existing selenopeptides with unknown modifications, peptides with unknown selenium-containing modifications, or simply contaminations from environmental microbes. Interestingly, our manual analysis of existing MS/MS spectra indeed identified an unexpected form of selenopeptide of SELH, which seems to have an intramolecular S–Se bond between cysteine and selenocysteine and a simultaneous deamination from Asn to Asp in the sequence (Figure S9). Novel selenopeptides could be another possibility as Gladyshev and co-workers have conducted a blastx query that aligned several cysteine sites to the UGA codon with downstream SECIS element.[34] Since most of these additional selenium-containing envelopes either have associated MS/MS spectra with poor quality or do not have them at all, it will be extremely challenging to confidently confirm their identity using the existing proteomic data. SESTAR-directed targeted proteomics analysis (e.g., by PRM), especially on cells/tissues lacking general selenoprotein expressions,[30] should help reveal the nature of these “cryptic” selenium-encoded spectra in the future.

We showed that SESTAR is applicable to improve detection of chemically labeled peptides with a selenium-containing tag. As the development of SESTAR was partially inspired by the dibromide tag as reported by Bertozzi and colleagues,[16−18] we believe that selenium will be an alternative element of choice for encoding tags used in chemical-directed proteomics. Particularly, selenium has unique chemical properties that can be utilized for special purposes; for example, the oxidative cleavage of a selenium–carbon bond was applied to separate cross-linked protein complexes.[35] In all cases, SESTAR should work seamlessly with these chemical-proteomic approaches to improve detection of tagged peptides from complex samples. Lastly, we envision that SESTAR should be readily applicable to targeted metabolomics analysis so that selenium-containing metabolites can be profiled with enhanced sensitivity. The numerous applications enabled by SESTAR will greatly broaden the scope of future study of selenium biology, including but not limited to functions of selenoproteins and selenometabolites, selenium-directed chemical proteomics, as well as selenium-based drugs.