The “adductome”: a limited repertoire of adducted proteins in human cells

Abstract

Proteins form adducts with nucleic acids in a variety of contexts, and these adducts may be cytotoxic if not repaired. Here we apply a proteomic approach to identification of proteins adducted to DNA or RNA in normally proliferating cells. This approach combines RADAR fractionation of proteins covalently bound to nucleic acids with quantitative mass spectrometry (MS). We demonstrate that “RADAR-MS” can quantify induction of TOP1- or TOP2-DNA adducts in cells treated with topotecan or etoposide, respectively, and also identify intermediates in physiological adduct repair. We validate RADAR-MS for discovery of previously unknown adducts by determining the repertoires of adducted proteins in two different normally proliferating human cell lines, CCRF-CEM T cells and GM639 fibroblasts. These repertoires are significantly similar with one another and exhibit robust correlations in their quantitative profiles (Spearman r=0.52). A very similar repertoire is identified by the classical approach of CsCl buoyant density gradient centrifugation. We find that in normally proliferating human cells, the repertoire of adducted proteins — the “adductome” — is comprised of a limited number of proteins belonging to specific functional groups, and that it is greatly enriched for histones, HMG proteins and proteins involved in RNA splicing. Treatment with low concentrations of formaldehyde caused little change in the limited repertoire of proteins in a small number of functional classes. The endogenous adductome may contribute to the burden of adducts requiring repair in order to maintain genomic structure, particularly in cells deficient in adduct repair.

1. Introduction

Protein-nucleic acid adducts can form in the course of enzymatic reactions or as a result of treatment with agents that cause proteins to become crosslinked to DNA or RNA. Adducts can be cytotoxic, and robust pathways carry out adduct repair [1–3]. More than 30 human proteins form transient adducts with DNA as obligatory reaction intermediates (Table S1), among them topoisomerases, methyltransferases, tyrosyl-DNA phosphodiesterases, DNA glycosylases, polymerases and repair proteins with AP lyase activity [1, 4]. Treatment of cells with chemicals or radiation causes a much wider spectrum of proteins to become crosslinked to nucleic acids, and details of processes essential to DNA replication, repair, transcription, RNA processing, and translation have been elucidated by combining chemical and UV crosslinking with precise characterization of interacting sites and motifs (e.g. [5–7]).

The ease with which adducts can be induced experimentally raised the question of whether cells may normally contain some level of adducts formed in response to exposure to endogenous formaldehyde. To address this, we have characterized the repertoire of adducted proteins in normally proliferating human cells, taking advantage of the unbiased detection intrinsic to mass-spectrometry (MS) to identify proteins covalently bound to nucleic acids. Samples were generated for MS by RADAR fractionation [8, 9], a procedure in which adducts are fractionated by cell lysis in chaotropic salts and detergent followed by alcohol precipitation of nucleic acids — both DNA and RNA (Fig. 1A). Adducts of human TOP1, TOP2A, POLβ, and of DNA gyrase and Topoisomerase IV from both E. coli and S. aureus, have previously been recovered by RADAR fractionation and quantified by immunodetection [8–16]. MS enables unbiased identification of proteins independent of the antibodies, which may be limited in availability or specificity, or unable to detect adducts undergoing proteolytic repair that eliminates epitopes critical for antibody recognition.

Fig. 1. RADAR-SILAC analysis identifies adducts formed by TOP1 and TOP2.

(A) Schematic of fractionation: Cells are lysed in chaotropic salts and detergent, then nucleic acids and adducted proteins precipitated with ethanol or isopropanol.

(B) RADAR-SILAC analysis of CCRF-CEM cells treated with the TOP1 poison, topotecan. Proteins enriched in inhibitor-treated cells are shown in red.

(C) RADAR-SILAC analysis of CCRF-CEM CCRF-CEM cells treated the TOP2 poison, etoposide. Proteins enriched in inhibitor-treated cells are shown in red.

(D) Above, diagram of TOP1. Tryptic peptides used to quantify recovery from the N-terminal (residues 205–216, 224–239, 252–271 and 300–310,) and C-terminal (residues 643–750, 693–700, 701–712 and 736–742) are highlighted in yellow and green, respectively. Catalytic tyrosine Y723 indicated in red. Below, ratios of recovery of the indicated peptides from the C- and N-terminal regions of TOP1, based on intensity. Each dot represents relative recovery in one of 28 different experiments reported in the MaxQB database (whole cell), and squares represent relative recovery in one of 11 different MS analyses of RADAR-fractionated cells. The Mann-Whitney test was used to determine p value.

Here we show that RADAR-MS faithfully identifies TOP1 and TOP2 adducts induced by treatment with topotecan or etoposide, respectively; and also identifies TOP1 fragments that are likely intermediates in physiological repair. We show that RADAR-MS identifies repertoires of adducted proteins in two normally proliferating human cell lines derived from two different tissues, CCRF-CEM T cells and GM639 fibroblasts. These repertoires are significantly similar with one another and with the repertoire determined using the classical approach of CsCl buoyant density gradient fractionation to recover adducts. We show that treatment with low doses of formaldehyde increases representation of histones and HMG proteins but has little effect on the overall composition of the repertoire of adducted proteins, consistent with the possibility that reaction with endogenous formaldehyde contributes to adduction. The repertoire of adducted proteins is enriched for histones, HMG proteins and for proteins involved in RNA splicing. These results suggest that human cells contain an adducted proteome, or “adductome”, comprised of a limited repertoire of proteins in a small number of functional classes. The endogenous adductome may contribute to the burden of adducts requiring repair in order to maintain genomic structure, particularly in cells deficient in adduct repair.

2. Materials and methods

2.1. Cells, cell culture, drug treatment and SILAC labeling

The CCRF-CEM T lymphoblastoid cell line (ATCC CCL-119), which was derived from a human acute lymphoblastic leukemia, was cultured in RPMI1640 containing 10% fetal bovine serum and Pen-Strep (Gibco) at 37°C in 5% CO₂. The GM639 cell line, derived from an SV40-transformed human fibroblast, was cultured in DMEM containing 10% fetal bovine serum and Pen-Strep (Gibco) at 37°C in 5% CO₂. Sensitivity to topoisomerase poisons topotecan and etoposide was demonstrated using the CellTiter-Glo® ATPase assay (Promega). Prior to metabolic labeling analyses, cells were shown to be drug sensitive (Supplementary Fig. S1).

For metabolic labeling, light and heavy SILAC RPMI-1640 media were prepared by supplementing RPMI-1640 lacking lysine and arginine with either “light” (normal isotope abundance L-lysine and L-arginine) or “heavy” (stable isotope [¹³C₆, ¹⁵N₂] L-lysine and [¹³C₆, ¹⁵N₄] L-arginine enriched) amino acids [17]. CCRF-CEM cells were metabolically labeled in parallel 40 ml cultures in T-75 flasks at 37°C, 5% CO₂ in light or heavy SILAC RPMI-1640 media supplemented with 10% dialyzed fetal bovine serum for at least five doublings. When cell density reached 5 × 10⁵ cells/ml, cells were treated for 15 min with 10 μM topotecan (Enzo Life Sciences) or 50 μM etoposide (EMD Biosciences), with controls treated with 0.5% DMSO. Cells (2 × 10⁷) were then harvested by 10 min centrifugation at 2,740 RCF. To validate differences between untreated and treated cultures, label-swap replicates were performed by reversing the SILAC labeled state and the drug treatments between replicates, with the expectation that SILAC ratios for bona fide protein adducts would invert accordingly.

2.2. RADAR fractionation

Cells were recovered by centrifugation and lysed in 1 ml of pre-warmed LS1 reagent, consisting of 5 M guanidinium isothiocyanate (GTC), 2% Sarkosyl, 10 mg/ml DTT, 20 mM EDTA, 20 mM Tris-HCl (pH 8.0) and 0.1 M sodium acetate (pH 5.3), adjusted to final pH 6.5 with NaOH. To ensure complete homogenization and reduce viscosity due to high molecular weight DNA, lysates were sonicated on ice using a cuphorn device (QSONICA) or passed 8–12 times through a 22G 1½ inch needle. Each lysate was aliquoted into a pair of Eppendorf tubes, 450 μl/tube (900 μl total), then to each tube was added 150 μl of 8 M LiCl (final concentration 2 M LiCl), followed by an equal volume (600 μl) of isopropanol. Nucleic acids and proteins bound to them were recovered by 10 min centrifugation at 21,000 RCF. Pellets were washed twice by addition of 1 ml 75% ethanol and 5 min centrifugation, then resuspended by addition of 100 μl of freshly prepared 8 mM NaOH followed by shaking on an Eppendorf Thermomixer at room temperature for 15–30 min. Samples were neutralized by addition of 2 μl 1M HEPES per tube and then treated for 30 min at 37ºC with RNase A (Fermentas) at final concentration 50 μg/ml. Then 350 μl LS1 reagent and 150 μl 8M LiCl were added, and samples again precipitated with an equal volume (600 μl) of isopropanol. Nucleic acids were recovered and washed twice as above, resuspended by brief shaking in 50 μl 8 mM NaOH, then neutralized by addition of 1 μl 1M HEPES per tube. DNA and RNA were quantified using DNA/RNA specific fluorescent detection kits (Qubit, Invitrogen). Total protein was measured using a BCA detection kit (Pierce).

2.3. Enrichment of adducts by CsCl buoyant density gradient centrifugation

DNA-protein covalent complexes were isolated by ultracentrifugation on a CsCl buoyant density gradient [18]. Briefly, CCRF-CEM cells (2 × 10⁷) were harvested and lysed in 3 ml TE with 1% Sarkosyl supplemented with protease inhibitor cocktail (Roche). The lysate was passed through a 22G 1½ inch needle to reduce viscosity, then loaded on top of a preformed four-step CsCl gradient (1.37, 1.50 1.72, and 1.82 g/ml) and centrifuged in an SW41 rotor at 30,000 rpm for 20 hr at 20ºC. Peak fractions containing DNA and DNA-protein adducts were pooled and desalted by diafiltration using a Microcon centrifugal filter (EMD Millipore). This fraction contained 82% dsDNA and 18% RNA, as measured by fluorescent Qubit assays specific for dsDNA and RNA.

2.4. NanoLC-MS/MS analysis and quantification

Prior to MS analysis, RADAR fractions in 200 μl volume were treated with 10 units of Cyanase nuclease (RiboSolutions) for 16–18 hr at room temperature, in reactions containing 6 mM MnCl₂. For SILAC experiments, RADAR fractions from untreated and drug-treated SILAC-labeled cells were mixed in a 1:1 ratio based on DNA concentration before processing. For label-free experiments, each fractionated sample was processed separately. To each sample, 8M urea in 50mM Tris pH 8.0 and 75 mM NaCl was added, followed by reduction and alkylation with Tris2(-carboxyethyl)phosphine (TCEP) and chloroacetamide (CAM) at final concentrations of 1 mM and 2mM, respectively. An equal volume of 100 mM triethylammonium bicarbonate (TEAB; Sigma) was then added and samples were then digested with LysC (Wako Chemicals) at 1:100 ratio for 2 hr. Samples were further diluted to adjust urea concentration below 1.5 M, then digested with Trypsin (Thermo Scientific) at 1:100 ratio for 16–18 hr at room temperature. Peptides were acidified at pH 2.0 with 10% trifluoroacetic acid (TFA) and desalted using StageTips [19]. Peptides were eluted in 80% acetonitrile/0.1%TFA and dried to completion using vacuum centrifugation, then resuspended in 5% acetonitrile/0.1% TFA.

For SILAC experiments, peptides were separated on a Thermo-Dionex RSLCNano UHPLC instrument (Sunnyvale, CA) with 10 cm long 100 μm I.D. fused silica capillary columns and packed with 3 μm 120 Å reversed phase C18 beads (Dr Maisch), made in house with a laser puller (Sutter). The LC gradient was 90 min of 10–30% B at 300 nL/min. LC solvent A was 0.1% acetic acid and LC solvent B was 0.1% acetic acid, 99.9% acetonitrile. MS data were collected with a Thermo Orbitrap Elite. Data-dependent analysis was applied using Top15 selection with CID fragmentation using 35% collision energy.

For label-free analysis of sample, peptides were separated on a Thermo EASY-nLC 1200 UHPLC instrument with in-house packed columns as described above. The LC gradient was 90 minutes of 6–38% B at 300 nL/min. LC solvent A was 0.1% acetic acid and LC solvent B was 0.1% acetic acid, 80% acetonitrile. MS data were collected with a Thermo Orbitrap Fusion Lumos Tribrid. Data-dependent analysis was applied using Top10 selection with simultaneous CID fragmentation at 32% CID collision energy detected in the ion-trap and HCD fragmentation using 31% collision energy and detected in Orbitrap. The resolution of the Orbitrap for is 60,000 for MS1 and 30,000 for MS2.

2.5. Data analysis

MaxQuant v.1.5.7.4 and the associated Andromeda search engine was used to search a Uniprot human database (July 2016). The following search parameters were used: Trypsin/P with two missed cleavages, fixed carbamidomethylated cysteines, and variable modifications of oxidized methionines and N-terminal acetylation. Initial FTMS and ITMS MS/MS tolerances were set at 20 ppm and 0.5 Da, respectively. Protein and peptide false discovery rate (FDR) was 1%, minimum peptide length was seven amino acids, and a minimum of two peptide ratios were required to quantify a protein. Data were analyzed with the Perseus and R environments. A published deep whole cell proteome of CCRF-CEM cells [20] containing 6,282 unique gene entries with positive IBAQ values was used as a reference for repertoire and peptide intensities of this cell line. To calculate intensities of individual TOP1 peptides recovered from whole cell proteomes, we used a dataset of 11 cell lines deposited in the MaxQB database (maxqb.biochem.mpg.de) [21, 22].

2.6. Analysis of gene ontologies

A total of 20,199 entries from the manually annotated and reviewed human proteome (Swiss-Prot) were extracted from the Uniprot (www.uniprot.org). Entries that had no matching gene names (some putative proteins and peptides < 20 amino acids) and non-unique gene names from families with high sequence similarity (e.g. HLA-A) were then eliminated, to generate a list containing 19,742 unique entries. We searched the list of unique entries for gene ontologies related to DNA/RNA/nucleic acids binding properties: GO:0003676 (nucleic acid binding, NBP); GO:0003677 (DNA binding, DBP); GO:0003723 (RNA binding, RBP). This yielded 3,919 NBP, 2,438 DBP and1,577 RBP species. Of the 3919 NBP, 265 were able to bind both DNA and RNA, and 169 lacked a particular DNA/RNA binding GO assignment. As a reference, we used a published deep whole cell proteome of CCRF-CEM that contained 7,625 unique gene entries, of which 6282 had positive iBAQ values [20]. Enriched pathways and gene ontologies were identified using STRING database of protein-protein interaction networks (string-db.org).

2.7. Statistical analyses

Hypergeometric distributions were evaluated using the online calculator https://systems.crump.ucla.edu/hypergeometric/index.php. The population sizes, N, and the experimental parameters for each analysis are shown in the corresponding figure legend. For comparisons with whole cell repertoires, the population size was specified at 9000, based on averages of 10,000 proteins per cell and 90% shared repertoire between pairs of human cell lines [21]. Other statistical tests were performed using GraphPad Prism software.

3. Results

3.1. RADAR-MS detects drug-induced topoisomerase-DNA adducts

TOP1, the most abundant topoisomerase, cleaves and rejoins single DNA strands to regulate superhelicity at the promoter and target repair to non-canonical DNA structures [23]. Topotecan is a derivative of camptothecin, a natural product which stabilizes normally transient TOP1-DNA adducts, thereby generating cytotoxic lesions. CCRF-CEM cells are derived from a human acute lymphoblastic leukemia and are topotecan-sensitive (Supplementary Fig. S1). We assayed the ability of RADAR-MS to detect enrichment of TOP1-DNA adducts in CCRF-CEM cells cultured in heavy or light SILAC medium, treated for 15 min with topotecan or DMSO carrier, then RADAR fractionated to recover nucleic acids and adducted proteins. Samples were then treated with Cyanase nuclease to digest nucleic acids and with LysC and trypsin to generate peptides suitable for MS analysis. In two independent experiments, TOP1 was consistently found to be enriched in cells treated with topotecan, as shown by a representative plot (Fig. 1B).

TOP2 regulates DNA topology by catalyzing breakage and joining of both DNA strands. Etoposide, a synthetic derivative of a natural toxin, poisons TOP2 by stabilizing normally transient adducts to generate cytotoxic lesions. In two independent experiments, RADAR-MS analysis of cells treated with etoposide documented enrichment of two different human type 2A topoisomerases, TOP2A and TOP2B, as shown by a representative plot (Fig. 1C). As anticipated, TOP1 was not enriched following etoposide treatment. Posttranslational sumoylation can contribute to proteolytic repair, and peptides from SUMO2 were enriched, consistent with previous reports that TOP2A- and TOP2B-DNA adducts are sumoylated [24, 25]. The SILAC enrichment score of TOP2A was four-fold higher than that of TOP2B. This could reflect a variety of factors which are not mutually exclusive: greater abundance of TOP2A, longer persistence of TOP2A-DNA adducts, or a higher selectivity of etoposide for TOP2A. These results demonstrate that RADAR-MS can identify post-translational modifications and quantitatively discriminate among adducted protein species.

3.2. RADAR-MS provides a snapshot of physiological adduct repair

In normally proliferating cells, TOP1 forms transient covalent bonds at some sites, but at AP sites and other endogenous lesions it may form irreversible adducts which must undergo proteolytic repair. This raised the possibility that some of the TOP1 adducts recovered from untreated cells were repair intermediates that had undergone proteolytic processing. The active site tyrosine Y723 in TOP1 that forms stable adducts with DNA is near the C-terminus of the 765 residue TOP1 protein (Fig. 1D). If RADAR fractionation recovers partially proteolyzed TOP1-DNA adducts from untreated cells, then C-terminal tryptic peptides of TOP1 will be enriched in RADAR-MS spectra relative to N-terminal peptides. We confirmed this by comparing the relative recovery of N-terminal (residues 205–310) and C-terminal (residues 643–742) tryptic peptides from whole cells in 28 different experiments reported in the MaxQB database with recovery by RADAR-MS analysis as determined in 11 independent experiments. C-terminal peptides were significantly enriched in the RADAR fraction (Mann-Whitney test, p=0.005; Fig. 1D).

3.3. RADAR-MS identifies a limited repertoire of adducted proteins shared between different cell types

We next sought to characterize species that form endogenous crosslinks with nucleic acids in normally proliferating CCRF-CEM T cells. Two untreated samples (NTa and NTb), each containing 25 × 10⁶ cells, were RADAR-fractionated, yielding 220 and 270 μg DNA, respectively (Supplementary Fig. S2A). They were then subject to label-free quantitative MS. The IBAQ intensity of a protein in a MS sample provides a measure of how observed enrichment relates to theoretical enrichment by calculating relative abundance of each protein (its IBAQ score) from the sum of intensities of all identified peptides divided by the number of theoretically observable peptides in its sequence; and the MS signal of a sample corresponds to the sum of IBAQ intensities for all identified proteins in that sample. The MS signals for samples NTa and Ntb were 1.9 and 2.4 × 10¹⁰, respectively (Supplementary Fig. S2B; Table S2). Combining the separate repertoires of samples NTa and NTb (1449 and 1511 proteins, respectively) yielded a total repertoire of 1664 distinct proteins, with 1296 shared proteins (Fig. 2A). The shared proteins accounted for 78% of the total repertoire and for 99.6% and 98.9% of MS signal in each of the two samples, respectively. Pairwise comparisons of the IBAQ intensities of the 1296 shared proteins showed very high correlation (Spearman r=0.93; p<0.0001; Fig. 2B). Thus, RADAR-MS yielded consistent repertoires of adducted proteins, with relative IBAQ intensities of fractionated proteins well reproduced between samples.

Fig. 2. RADAR-MS identifies a limited set of adducted proteins in normally proliferating human cell lines.

(A) Venn diagram of shared and unshared proteins in two untreated samples of CCRF-CEM T cells, NTa and NTb, containing 1449 and 1511 proteins, respectively; and a combined repertoire of 1664 non-overlapping proteins. Repertoire size and percent repertoire indicated in parentheses.

(B) Pairwise comparison of the IBAQ intensities of the 1296 proteins shared between untreated samples NTa and NTb (Spearman r=0.93; p<0.0001).

(C) Venn diagram comparing repertoires of adducted proteins in untreated CCRF-CEM T cells (unshaded) and GM639 fibroblasts (shaded) as determined by RADAR fractionation. Repertoire size and percent of the GM639 repertoire indicated in parentheses. Hypergeometric p=3.6e-227 was calculated based on the following parameters: population size=9000 ([21]; see Methods); sample size A=1296; sample size B=518; set=384; expected successes=75; observed/expected: 175/38; enrichment=5-fold.

(D) Pairwise comparison of the intensities of the RADAR shared repertoire of CCRF-CEM T cells and GM639 fibroblasts (Spearman r=0.52; p<0.0001).

We then asked how the repertoire of adducted proteins in CCRF-CEM human T cells compared with the repertoire of adducted proteins in an unrelated cell type, human GM639 fibroblasts. Four untreated samples of GM639 cells were RADAR-fractionated and analyzed by MS, identifying a total of 676 proteins (average 442 per spectrum; range 306–522), with 518 proteins accounting for 99.1% of the MS signal and 77% of the total repertoire (Table S3). Comparison of these 518 proteins with the 1296 protein repertoire in CCRF-CEM cells identified 384 shared proteins (Fig. 2C; Table S3). These shared proteins accounted for 88% and 85% of the MS signals in the two cell types (Supplementary Fig. S2D); and 74% of the GM639 RADAR repertoire, a 5-fold enrichment over 75 common elements expected by chance (hypergeometric p=3.65e-227; Fig. 2C). The correspondence of adducts recovered from the two cell types was further evident upon pairwise comparison of their relative abundance (Spearman r=0.52, p<0.0001; Fig. 2D). Thus, RADAR-MS identified a limited number of adducted proteins in human cells. We will refer to this repertoire of adducted proteins, based on more than 20 independent RADAR-MS analyses, as the “RADAR shared repertoire”, with the caveat that MS analysis is inherently variable and additional experimentation will be required to compile a truly definitive repertoire.

3.4. RADAR enrichment does not correlate with protein abundance

One trivial explanation for repeated identification of a specific protein or class of proteins in independent RADAR-MS analyses is that RADAR fractionation does not effectively eliminate abundant proteins. We therefore asked if abundant proteins were enriched in the RADAR shared repertoire. By intersection of the RADAR shared repertoire (384 proteins) with the CCRF-CEM cell whole cell proteome (6282 proteins, IBAQ>0; http://wzw.tum.de/proteomics/nci60 [20], 342 matching pairs were identified (Table S4), which corresponded to 89% of repertoire. Pairwise comparison established that there was no significant correlation between the IBAQ scores of proteins in the RADAR shared repertoire and the whole cell proteome (Spearman r=0.029, p=0.595, Fig. 3A).

Fig. 3. RADAR fractionation enriches RNA binding proteins.

(A) Pairwise comparison of the IBAQ intensities of the 342 protein pairs common to the CCRF-CEM whole cell proteome and the RADAR shared repertoire (Spearman r=0.029; p=0.595).

(B) Venn diagram illustrating overlap of repertoires of adducted proteins as determined by RADAR fractionation (unshaded) and CsCl buoyant density gradient fractionation (shaded). Repertoire size indicated in parentheses. Hypergeometric p=2.3e-41 was calculated based on the following parameters: population size=6282 sample size A=384; sample size B=96; set=54; expected successes=5.9; observed/expected: 54/5.9; enrichment=9.2-fold.

(C) Pie chart indicating the fraction of proteins classified as DBP, RBP or dual binders in the repertoires of the CCRF-CEM whole cell proteome and the RADAR shared repertoire.

(D) Pie chart indicating the fraction of proteins classified as DBP, RBP or dual binders in the repertoire determined by buoyant density gradient centrifugation.

3.5. Overlap of repertoires of adducted proteins as determined by RADAR and buoyant density gradient fractionation

CsCl buoyant density gradient fractionation has been the standard method for isolation of covalent protein-DNA adducts [26]. To compare enrichment of adducts by RADAR fractionation to this classical approach, samples from untreated CCRF-CEM cells which had been lysed in sarkosyl were fractionated by density gradient centrifugation then analyzed by MS, thereby identifying 96 proteins with positive peptide intensity values (Table S5). The repertoire of adducted proteins determined by density gradient fractionation exhibited significant overlap with the RADAR shared repertoire (hypergeometric p=2.3e-41, Fig. 3B).

3.6. RADAR fractionation enriches proteins known to form adducts

Of the 35 human proteins known to form adducts with nucleic acids (Uniprot; Table S1), 19 were detected in the whole cell CCRF-CEM proteome [20]. Nine of these proteins (DNMT1, GAPDH, PARP1, PTMA, TOP1, TOP2A, TOP2B, XRCC5 and XRCC6) were identified in the RADAR shared repertoire (Supplementary Fig. S3A), where together they accounted for 1.8% of its MS signal (Table S3). This represents a 7.8-fold enrichment over the number of proteins expected in that set by chance (hypergeometric p=5.8e-07). Similarly, the density gradient-fractionated sample (Table S5) included five proteins known to form adducts (DNMT1, GAPDH, PTMA, TOP1 and TOP2A), a significant enrichment over the 0.3 proteins expected in that set by chance (hypergeometric p=7.4e-06, Supplementary Fig. S3B). Thus, proteins previously shown to form adducts comprise a small but significant fraction of the RADAR shared repertoire.

3.7. RADAR fractionation enriches RNA binding proteins

We further characterized the RADAR shared repertoire by comparing the abundance of nucleic acid binding proteins (NBPs) in it and in the publicly available CCRF-CEM whole cell proteome (96282 proteins; [20]). NBP account for 24% of the expressed repertoire of the CCRF-CEM whole cell proteome (Fig. 3C), very similar to fraction of the repertoire (20%) that NBP constitute in the reference human proteome (SwissProt, 19,742 entries). RADAR fractionation enriched NBP, which accounted for 61% of the RADAR shared repertoire. Notably, RADAR fractionation especially enriched RNA binding proteins (RBP, 42%) and dual DNA and RNA binding proteins (14%), and modestly depleted DNA binding proteins (DBP, 5%) relative to the whole cell proteome (Table S6). Similarly, GO analysis of the repertoire of the sample prepared by density gradient fractionation identified 67% of the proteins as NBP, predominately RBP (47%) and dual binders (17%), along with a small fraction of DBD (3%; Fig. 3D).

A considerable number of proteins in the RADAR shared repertoire (149) were classified as non-NBP by GO analysis (Fig. 3C). The universe of proteins that bind RNA has undergone recent rapid expansion [27], and the possibility that this might not yet be reflected in GO classifications prompted us to test the possibility that some proteins designated non-NBP in GO classifications might be appropriately reclassified as RBP. We therefore compared the non-NBP GO species of the RADAR shared repertoire to an experimental proteomics database of RNA-binding proteins recovered from three human cell lines following UV-induced RNA-protein crosslinking (ihRBP, 1753 proteins [7]). Most of the species (79/149; 60%) identified as non-NBP by GO analysis of the RADAR shared repertoire intersected with the ihRBP set (hypergeometric p=1.2e-43; Supplementary Fig. S3C). Adjusting the composition of the RADAR shared repertoire by reclassification of those 79 proteins increased the fraction of RBP to 63% and reduced the fraction of non-NBP to 18%. This reaffirms the enrichment of RBP among adducted proteins.

3.8. Treatment with exogenous formaldehyde increases abundance of adducts present in the repertoires of untreated cells

The results described above show that specific proteins are covalently bound to nucleic acids in normally proliferating human cells, and that most of these proteins are not known to form adducts as part of their mechanism of action. What might cause protein-nucleic adducts to form? Formaldehyde is a highly reactive aldehyde that is generated naturally as a product of demethylation reactions in living cells (reviewed by [3, 28]. The formaldehyde concentration in human plasma exceeds 0.1 mM, and intracellular concentrations may be four-fold higher (European Food Safety Authority, https://efsa.onlinelibrary.wiley.com/doi/pdf/10.2903/j.efsa.2014.3550). Formaldehyde exhibits clear selectivity. It reacts with only some amino acid residues, including exposed N-terminal amino groups and side chains of cysteine, histidine, arginine, lysine and tryptophan [29]. It does not crosslink proteins that are not bound to nucleic acids, even abundant proteins like serum albumin; and it only crosslinks some DNA bound proteins: while it efficiently crosslinks histones, it fails to crosslink the formidable DNA binding protein, lactose repressor [30].

If endogenous formaldehyde promotes adduct formation by some proteins or classes of proteins, then treatment with exogenous formaldehyde is predicted to increase the abundance of those same proteins or classes of proteins. To test this, cells were treated for 1 hr with 0, 0.5, 1.0 or 2.0 mM formaldehyde, a narrow range of concentrations slightly above the endogenous level and in the range of concentrations previously used to study formation and repair of crosslinked proteins in mammalian cells [14, 31, 32], but below levels used to ensure the extensive crosslinking necessary for quantitative immunoprecipitation (5 min, 130 mM formaldehyde [33]). Rather than perform two exact replications, parallel analyses were carried out on CCRF-CEM cells that were cultured in medium lacking or containing 10% fetal bovine serum (Groups A and B, respectively). RADAR fractionation yielded on the order of 265 μg DNA per sample, independent of formaldehyde dose, similar to levels observed in untreated CCRF-CEM samples (Supplementary Fig. S4A). The number of proteins detected ranged from 855 to 1081 in the eight samples; and the total MS signal (a sum of intensities of all identified peptides in the sample) was consistent with the number of detected proteins (Supplementary Fig. S4B; Table S7). Neither the MS signal nor repertoire size showed an appreciable increase with the formaldehyde dose, indicating that massive chromatin crosslinking did not occur at these formaldehyde concentrations.

To determine which proteins were most reactive with formaldehyde, the proteins that exhibited positive IBAQ values in all treated and untreated samples in both groups were identified, and then the response slopes (the change in relative abundance at each formaldehyde concentration from 0–2 mM) were calculated for each of these 625 proteins using the SLOPE function in Excel (Table S7). Response slopes ranged from +1.91 to −2.06. Proteins were ranked based on response slope and the top 100 proteins in each group (16% of total repertoire) compared to identify common proteins (Fig. 4A). The two groups included 46 common proteins, far more than the 16 shared elements predicted by chance in the absence of any concentration-dependent increase in adduction in response to treatment with formaldehyde (hypergeometric p=1.8e-15; Fig. 4A). Summing up MS signals of these 46 species in each of the two groups established that signals increased in response to increasing formaldehyde, with an average 4-fold increase in both groups at the highest formaldehyde concentration (Fig. 4B).

Fig. 4. Formaldehyde treatment has little effect on the composition of the repertoire of adducted proteins.

(A) Venn diagram of shared and private (unshared) proteins among the 100 proteins with greatest response slopes in Groups A and B. Hypergeometric p=1.8e-15 was calculated based on the following parameters: population size=625 (all proteins with positive intensities); sample size A=100; sample size B=100; set=46; expected successes=16; observed/expected: 46/16; enrichment=2.9-fold.

(B) Fractions of MS signal contributed by the 46 shared proteins ranked in the top 100 based upon response slopes.

(C) Quantitation of contribution to signal of four proteins/classes of proteins that contributed to 90% of the increased signal at the highest dose of formaldehyde.

The proteins that contributed to 90% of the increased signal at the highest dose of formaldehyde belonged to four classes: histones, HMG proteins, SUMO2 and a diverse group of nuclear RBP (Fig. 4C). The abundance of adducts formed by these proteins increased in response to increasing doses of formaldehyde independent of inclusion of serum in the culture medium (Fig. S5). The increased signal of SUMO2 may reflect posttranslational modification that targets proteins for proteolytic repair, as was also seen in RADAR-SILAC analysis of etoposide-treated cells (Fig. 1C). Thus, a limited repertoire of proteins appears to be highly reactive to exogeneous formaldehyde at doses only slightly surpassing its physiological level.

3.9. The RADAR shared repertoire is enriched for proteins with specific functions

The distribution of proteins of the RADAR shared set by relative abundance showed that a very limited number of species accounted for most of the MS signal in each of the two cell types (Fig. 5A): more than 50% of the signal could be assigned to a total of 13 proteins, and more than 75% of the signal to 50 proteins. Based on GO analysis, RBP accounted for most of the MS signal in the RADAR shared repertoire (65%; Fig. 5B). This represents at least a 2.2-fold enrichment over the whole cell proteome, a minimum estimate because not all RBPs are scored by GO classification (e.g. Supplementary Fig. S3C). Among the 50 most abundant proteins identified by RADAR-MS, ranked by IBAQ score (Fig. 5C), were nuclear RNA binding proteins, histones and HMG proteins; these same proteins were also found to be reactive with formaldehyde (Fig. 4C).

Fig. 5. Proteins enriched in the repertoire of adducted proteins.

(A) Frequency distribution of IBAQ scores in the RADAR shared repertoire of CCRF-CEM T cells and GM639 fibroblasts.

(B) Pie chart indicating the fraction of MS signal contributed by proteins classified as DBP, RBP or dual binders in the CCRF-CEM whole cell proteome and the RADAR shared repertoire. (A pie chart based on repertoire rather than MS signal is presented in Fig. 3C.)

(C) The fifty most abundant proteins in the RADAR repertoire of unperturbed CCRF-CEM T cells which account for over 70% of the total MS signal. Orange diamonds mark species shared with 30 topmost proteins of RNA-crosslinked proteome of MCF7 cells recovered after XRNAX fractionation (see discussion).

Analysis of protein-protein interaction networks identified specific GO terms for biological processes associated with the adducted proteome with very high significance (FDR<e-30): mRNA metabolism, RNA splicing, RNA processing, mRNA processing and RNA splicing via transesterification reactions (Fig. 6A). Molecular function GO terms (FDR<e-36) and KEGG pathways (FDR<e-08) provided further support for associations of these proteins with RNA processing and biogenesis. Strikingly, proteins characterized by a single GO term, RNA splicing, comprised 61% of the MS signal of adducted proteins, a 10-fold increase over the whole cell proteome (Fig. 6B, left). In contrast, MS signals of proteins characterized by the GO term translation were not enriched among adducted proteins (Fig. 6B, right). Thus, the strong associations with splicing appear to reflect participation in a specific nuclear pathway rather than general, RNA-related activities of proteins that are abundant in the RADAR fraction.

Fig. 6. Pathway analysis of adducted proteins.

(A) Highly significant functional enrichments in the RADAR shared repertoire based on analysis of protein-protein interaction networks (STRING database).

(B) Pie chart indicating the fraction of MS signal contributed by proteins associated with the GO terms RNA splicing and translation in the CCRF-CEM whole cell proteome and the RADAR shared repertoire.

Discussion

The results described here demonstrate that proteins adducted to nucleic acids in human cells can be identified and discovered by combining RADAR fractionation with MS. RADAR fractionation is rapid, high throughput and cost-effective, and requires only standard laboratory equipment. RADAR-MS analysis will be useful not only for identifying adducted proteins, but also for determining how adducts form and how they are repaired.

We showed that RADAR-MS can provide a snapshot of ongoing repair by using it to identify likely intermediates of proteolytic repair of Top1 bound to DNA. Following RADAR lysis in chaotropic salts, which interrupts proteolysis, processing intermediates were identified as enriched peptides surrounding the active site tyrosine of TOP1 (Fig. 1D). Post-translational sumoylation frequently marks a protein for proteolytic degradation, and the observed enrichment of SUMO2 in etoposide-treated CCRF-CEM cells (Fig. 1C) and in extracts of formaldehyde-treated cells (Fig. 4C) provides further evidence of the ability of RADAR-MS to detect such intermediates. Immunodetection cannot reliably provide equivalent information. Intermediates of proteolytic repair may lack epitopes critical for antibody recognition and evade immunodetection. Application of RADAR-MS to cells deficient in specific repair factors is thus a promising approach for defining the molecular mechanisms of adduct repair.

Two steps are key to RADAR fractionation: cell lysis in chaotropic salts and detergent, and alcohol precipitation of nucleic acids and proteins bound to them. We present several kinds of evidence that validates the ability of RADAR fractionation to enrich for proteins covalently bound to nucleic acids. Cell lysis in chaotropic salts and detergent effectively disrupts non-specific interactions among components of a cell extract, as evidenced by the absence of significant concordance between proteins enriched by RADAR fractionation and highly abundant proteins (Fig. 3A). The extract is sonicated or sheared and treated with RNase A to limit trapping of protein by long DNA or RNA molecules; relatively short regions of nucleic acids that are tightly or covalently bound to protein will be survive sonication or shearing and be protected from nuclease digestion. The extract is then precipitated with alcohol, which effectively recovers covalently bound protein, as evidenced by the correspondence between the repertoires determined by RADAR fractionation and by buoyant density fractionation (hypergeometric p=2.3e-41; Fig. 3B), two techniques that depend on very different properties to enrich for proteins covalently bound to nucleic acids.

MS analysis is inherently variable. Even though the repertoire of adducted proteins as determined thus far was based on more than 20 independent RADAR-MS analyses, it is not intended to be definitive. Further analysis is necessary, especially to determine whether the low abundance proteins do indeed form adducts with nucleic acids or are recovered with adducts because of some sort of inherent stickiness. Nonetheless, application of RADAR-MS to normally proliferating human cells identified a shared repertoire of adducted proteins with a quantitative profile that was significantly concordant in two unrelated cell types, CCRF-CEM T cells and GM639 fibroblasts (Spearman r=0.52; p<0.0001; Fig. 2D). The composition of this shared repertoire provides experimental insights into the classes of proteins that form adducts in normally proliferating cells.

The repertoire of adducted proteins — or the “adductome” — contained a relatively limited number of proteins. A minor but significant fraction of the adducted repertoire (hypergeometric p=5.8e-07, 1.8% MS signal) was comprised of proteins for which formation of adducts with DNA is key to mechanism of action, among them TOP1, TOP2, DNMT1 and Ku (XRCC5/6). Histones and HMG proteins, which are in close contact with DNA, were considerably enriched among adducted proteins and further enriched upon treatment with formaldehyde (Fig. 4C). This enrichment could reflect a propensity to undergo crosslinking, as evident in the response to formaldehyde treatment.

Enrichment of proteins specifically involved in mRNA splicing was unexpected. This may reflect some shared feature of protein composition which enables them to bind tightly but not covalently to RNA in a complex that is resistant to RADAR extraction. For example, only some amino acid residues can form adducts, and splicing proteins may share some feature of primary sequence that promotes spontaneously adduct formation. In addition, it is an intriguing possibility that RNA splicing proteins form crosslinks with their substrates to facilitate the long-range interactions necessary for RNA processing.

More generally, enrichment of nuclear RNA binding proteins was unexpected, in part because the majority of adduct-forming proteins described thus far form crosslinks with DNA. This may reflect more intensive study of protein-DNA adducts; or it may be a consequence of methods used to fractionate adducted proteins. Buoyant density fractionation takes advantage of density differences to separate free protein from DNA and RNA. The species recovered will depend on the exact density of the solutions used to form the gradients used for this separation step. The buoyant density of RNA is slightly greater than that of DNA, and a density gradient protocol designed to enrich protein-DNA adducts may not efficiently recover protein-RNA adducts (e.g. [35]). In contrast, the alcohol precipitation step in RADAR fractionation recovers proteins covalently bound to either RNA or DNA in a single pellet, which is then subject to further analysis. It is also possible that some proteins currently classified as RBP can bind to and form adducts with DNA, as suggested by the considerable enrichment in the adductome of proteins classified as “dual binders” (5.3-fold, hypergeometric p=6.7e-26; Table S6. Repeated recovery of these proteins makes it unlikely that they are contaminants, as does the absence of enrichment of abundant RNA binding proteins with other functions, including proteins associated with translation.

Recent proteome-based analyses have discovered novel RNA binding proteins that could not have been predicted based on prior knowledge of RNA-binding motifs [27]. This prompted us to compare the RADAR adductome with species found to be covalently bound to RNA following UV crosslinking of MCF7 cells, a human cell line derived from an invasive ductal carcinoma of the breast (“XRNAX fractionation”; [7]). MS analysis of the XRNAX showed that the 30 most abundant species in the fraction accounted for approximately 50% of the MS signal, and most of those species were also greatly enriched by UV crosslinking. Intriguingly, 29 of those 30 species were also present in the RADAR fraction of untreated CCRF-CEM cells, where they contributed to over 50% of its MS signal (Table S2); and 12 of those 30 species were among the most abundant proteins identified by RADAR-MS (Fig. 5C, orange diamonds).

Formaldehyde is highly reactive with some nucleic acid binding proteins, but spares others, making it an especially plausible candidate for the agent that promotes formation of the limited repertoire of protein-nucleic acid adducts identified in normally proliferating cells. Treatment of cells with low concentrations of formaldehyde increased the abundance in the adductome of histones, HMG proteins and nuclear RBD. These same classes of proteins are enriched in the adductome of normally proliferating cells, consistent with the possibility that endogenous formaldehyde generated by ongoing metabolism promotes adduct formation in normally proliferating cells.

Formaldehyde is generated naturally as a product of demethylation reactions in living cells, reaching concentrations in blood in the range of 0.1 mM even in the absence of environmental exposure [28]. Alcohol dehydrogenase 5 (ADH5) normally eliminates formaldehyde in mammalian cells, but impairment of this pathway can result in accumulation of endogenous formaldehyde [34]. Highly reactive proteins such as those we have identified may be especially prone to form adducts in response to exogenous exposures or genetic deficiencies that impair adduct repair. This raises the possibility that proteins enriched in the adductome as defined by RADAR-MS could serve as biomarkers for formaldehyde exposure or genetic deficiencies in adduct repair.

Highlights.

• The “adductome” in human cells includes a limited set of proteins

• RADAR and buoyant density fractionation define highly concordant adductomes

• RADAR fractionation combined with MS can discover novel protein-nucleic acid adducts

• Proteins involved in mRNA processing are overrepresented in the adductome

• Proteins reactive with exogenous formaldehyde are enriched in the adductome