ABSTRACT
Newly synthesized histone H4 that is incorporated into chromatin during DNA replication is acetylated on lysines 5 and 12. Histone deacetylase 1 (HDAC1) and HDAC2 are responsible for reducing H4 acetylation as chromatin matures. Using CRISPR-Cas9-generated hdac1- or hdac2-null fibroblasts, we determined that HDAC1 and HDAC2 do not fully compensate for each other in removing de novo acetyls on H4 in vivo. Proteomics of nascent chromatin and proximity ligation assays with newly replicated DNA revealed the binding of ATAD2, a bromodomain-containing posttranslational modification (PTM) reader that recognizes acetylated H4. ATAD2 is a transcription facilitator overexpressed in several cancers and in the simian virus 40 (SV40)-transformed human fibroblast model cell line used in this study. The recruitment of ATAD2 to nascent chromatin was increased in hdac2 cells over the wild type, and ATAD2 depletion reduced the levels of nascent chromatin-associated, acetylated H4 in wild-type and hdac2 cells. We propose that overexpressed ATAD2 shifts the balance of H4 acetylation by protecting this mark from removal and that HDAC2 but not HDAC1 can effectively compete with ATAD2 for the target acetyls. ATAD2 depletion also reduced global RNA synthesis and nascent DNA-associated RNA. A moderate dependence on ATAD2 for replication fork progression was noted only for hdac2 cells overexpressing the protein.
INTRODUCTION
Acetylation of histone tails is an important component of the epigenetic code of histone posttranslational modifications (PTMs) involved in regulating the functioning of eukaryotic genomes (1–3). Acetylated chromatin assumes a more “open” conformation, facilitating access to DNA. Acetylated histone tails also become substrates for the binding of specific protein regulators via the latter’s acetyl recognition modules, such as YEATS, DPF, and bromodomain (4). Bromodomain-containing proteins are by far the most numerous, with 42 proteins comprising several families identified to date (5, 6). Many of these proteins are involved in transcriptional regulation, and several are also implicated in DNA repair and replication (5–7).
During replication, parental and de novo-synthesized histones H3 and H4 are assembled into separate nucleosomes; however, both “old” and “new” nucleosomes are deposited onto each daughter DNA strand (8). De novo H4 carries acetyl marks on lysines 5 and 12 (reviewed in reference 9). Histone H3 also has de novo marks, in particular on lysine 9 (10–12). De novo marks on H3 and H4 are not, however, used exclusively to distinguish these new histones and are also found in transcriptionally active chromatin. These marks also appear dispensable for new histone deposition by histone chaperones, at least in Saccharomyces cerevisiae (13). More research is required in this area, but one possibility is that de novo marks play a role in DNA replication and/or postreplicative maturation of chromatin.
Whatever their role, de novo marks have to be actively removed if gene- or chromatin domain-specific marks are to be reconstituted and the overall genome acetylation level is to be maintained (14, 15). Indeed, removal of de novo acetyls occurs at the replication fork and postreplicatively and is carried out by at least two of the class 1 histone deacetylases (HDACs), the highly related HDAC1 and HDAC2 (14, 16). The extent to which HDAC1 and HDAC2 are redundant is not known. The two enzymes can homo- and heterodimerize (17–19) and are found associated with many of the same proteins, with a few exceptions (20). In transcriptional regulation, the two HDACs have demonstrated partially nonoverlapping roles (reviewed in references 16 and 21; see also references 22 and 23). HDAC2 was enriched both on the bodies of actively transcribed genes and on promoters, while HDAC1 was found on promoters only (24). In replication, both enzymes are found in the vicinity of replication forks but likely do not travel with the fork (14, 25). Chemical inhibition of both deacetylases results in an increase in total and replication fork-associated levels of H4K5ac and H4K12ac (26). Importantly, this is accompanied by a reduction in replication fork speed (25, 26), which, while counterintuitive to the notion that open chromatin always facilitates DNA metabolism, suggests that the act of histone deacetylation may be coupled to fork progression.
In this study, we knocked out HDAC1 or HDAC2 in human fibroblasts and examined the effects of the knockouts (KOs) on newly replicated chromatin, particularly with regard to HDAC histone PTM targets and the PTM readers that recognize them. The data uncover functional differences between HDAC1 and HDAC2 in the way in which the deacetylases execute the removal of the PTM in vivo. Our study suggests that the level of H4 acetyl PTMs in nascent chromatin is a result of a balancing act between HDAC1 and HDAC2 on the one hand and the acetyl PTM reader proteins on the other, one example of which we examined in the bromodomain protein ATAD2, a potential anticancer target. We suggest that ATAD2 can affect the rate of chromatin maturation by protecting de novo H4K12ac marks from removal. In addition, we show that a high level of ATAD2 drives global transcription and replication-transcription cooccurrences that may include RNA-DNA hybrids. At the same time, DNA replication at the level of fork progression is not sensitive to ATAD2 loss unless HDAC2 is knocked out. The data provide a framework for understanding the biology of cancer cells overexpressing ATAD2.
RESULTS
Ablation of HDAC1 or HDAC2 leads to compensatory increases of the remaining HDAC in nascent chromatin.We used lentiviral Cas9- and double guide RNA (dgRNA)-expressing constructs to disrupt open reading frames (ORFs) of the closely related HDAC1 and HDAC2 genes in the simian virus 40 (SV40)-transformed human fibroblast line GM639. The selection of candidate clones was based on the loss of HDAC1 or HDAC2 protein expression. The relevant genomic regions of the clones used in this study were sequenced (Table 1). We observed small deletions in both alleles of the HDAC1 or HDAC2 gene at the site of the 5′ guide RNA (gRNA) in all clones sequenced. Deletions at the 3′ gRNA or between the 5′ and 3′ gRNA sites were less common. We were not able to generate stable cell lines with null mutations in both HDAC1 and HDAC2, in agreement with previous findings (27, 28). In all clones tested, knockout of HDAC2 and, to a lesser degree, HDAC1 resulted in a slowing of cell growth (Fig. 1A).
Sequence analysis of individual hdac-null clonesa
The deletion of HDAC1 is associated with a compensatory increase in the abundance of HDAC2, and vice versa. (A) Clones of GM639 fibroblasts derived using lentiviral CRISPR-Cas9 and dgRNA vectors were seeded into a 96-well plate at the three indicated densities (from 1,000 to 4,000 cells per well) in triplicates, grown for 3 days, and subjected to a CellTiter Glo assay. Absorbance values from the assay were averaged and normalized to the value for the universal control (UC), i.e., GM639 cells with the integrated Cas9-expressing cassette and empty vector cassette for dgRNA expression (GM639-Cas9-EV). The P value shown is an example of several significant differences. (B) Representative Western blots demonstrating the absence of expression of a disrupted HDAC and a compensatory increase in the level of the remaining HDAC in the same sample. (C) HDAC2 and HDAC1 levels quantified in Western blots from multiple independent experiments were normalized to the corresponding levels in HDAC+ controls. Diamond symbols are means, and asterisks denote statistically significant differences. (D) Extracts of cells harvested in 3 independent experiments were fractionated into nucleoplasm and chromatin as described in Materials and Methods and subjected to Western blotting. Levels of HDAC1 and HDAC2 in chromatin were calculated as fractions of total levels, i.e., of the sum of the levels in chromatin and the nucleoplasm. The derived values for HDAC2 (left) and HDAC1 (right) are expressed as fold changes over values measured for HDAC+ controls. Diamond symbols are means. All P values for panels A through D were calculated by 2-sided, paired t tests. The changes observed in panel D did not reach statistical significance.
A reduction in the protein level of one of the two HDACs (HDAC1 or HDAC2) can result in the upregulation of the other (25, 27). This compensatory phenotype was reproduced in our engineered cell lines of both backgrounds, GM639 (Fig. 1B and C) and RD (25) (not shown). In multiple independent measurements on three or more each of hdac1- and hdac2-null clones in the GM639 background, the HDAC2 level was on average 1.5- to 2-fold higher in hdac1-null clones than in the wild type, and the HDAC1 level was 3-fold higher in hdac2-null clones than in the wild type (see Materials and Methods for more details on the generation of wild-type controls). This may suggest that HDAC2 is more efficient at compensating for HDAC1 than vice versa. Of note, different clones of hdac1-null cells averaged somewhat different levels of compensation.
Compensatory increases in total levels of HDAC1 were paralleled by changes in the recruitment of these deacetylases to chromatin. There was an average 1.3-fold increase of HDAC1 on chromatin in an hdac2-null clone compared to the control, while enrichment of HDAC2 on chromatin in hdac1-null cells did not change (Fig. 1D). We and others have shown by iPOND (immunoprecipitation of nascent DNA) that both HDAC1 and HDAC2 are present on nascent DNA labeled by 5-ethynyldeoxyuridine (EdU) (14, 25). We then established a proximity ligation assay (PLA) of HDACs with EdU as a fast, quantitative, and nonaveraging method that requires substantially fewer cells than iPOND. EdU was clicked to biotin and recognized by antibiotin antibody (Fig. 2A), as we described previously (29). The specificity of the PLA was confirmed by performing HDAC1-EdU and HDAC2-EdU PLAs in hdac1- and hdac2-null clones, respectively (Fig. 2C and I). Since non-S-phase cells that do not incorporate EdU are not expected to display foci and the percentage of non-S-phase cells may vary between cell lines, cells with zero or one focus were excluded from the quantitations in order to more appropriately measure the actual distributions of the signal among replicating cells. This cutoff value was adequate for most of the experiments in this series due to their low background. In general, the cutoff value may vary depending on the antibody and the batch of PLA reagents and was adjusted where indicated (see below). We quantified both the mean nuclear intensity of the PLA signal and the number of PLA foci, since in some cases, one or the other metric may be inadequate. For example, overabundant foci may merge together and result in an underestimation of their number. On the other hand, mean fluorescence levels may be affected by differences in backgrounds across samples.
A proximity ligation assay (PLA) demonstrates enrichment of the remaining HDAC on the nascent chromatin of cells with a single, hdac1- or hdac2-null mutation. (A) Diagram of the PLA approach used in the study. Cells are labeled with 10 to 20 μM EdU for 30 min and fixed. (B through M) EdU is clicked to either biotin-azide only (B through L) or a mix of biotin-azide and Alexa Fluor 488-azide at a molar ratio of 40:1 (M). (B) Representative image of HDAC2-EdU PLA foci in wild-type cells. Red fluorescence was reserved for the PLA throughout this study. (C) Numbers of HDAC2-EdU foci per nucleus were quantified in a representative experiment and plotted. Nuclei with zero foci were included in this measurement to accurately represent numbers of foci in the negative control (hdac2-36 null cells). Here and elsewhere, the UC (universal control) is GM639-Cas9-EV (see Fig. 1A legend). (D) Mean intensities of red fluorescence per nucleus were plotted separately for cells with zero or one focus (background) and in cells with two or more foci of HDAC2-EdU PLA signals. Here and elsewhere, MFI indicates the mean fluorescence intensity, and a.u. indicates arbitrary units. (E) Distributions of numbers of HDAC2-EdU PLA foci per nucleus in control and hdac1-50 cells. Only cells with two or more foci are included. (F and G) Same as panels D and E above, respectively, showing the values for HDAC2-EdU PLA signals in a different HDAC1-null clone, hdac1-47. Note that the experiments represented by panels B through E were performed with rabbit anti-HDAC2 and mouse antibiotin antibodies, and data represented in panels F and G were generated with mouse anti-HDAC2 and rabbit antibiotin antibodies. (H) Representative image of HDAC1-EdU PLA foci in the wild-type control. (I) Numbers of HDAC1-EdU foci per nucleus were quantified in a representative experiment and plotted. Nuclei with zero foci were included in this measurement to accurately represent the numbers of foci in the negative control (hdac1-50 null cells). WT1 and WT2 are GM639-Cas9 cells that have received HDAC1 or HDAC2 dgRNA-expressing vectors, respectively, but upon clonal isolation were shown to express sequence-verified wild-type HDACs. (J) Mean intensities of red fluorescence per nucleus were plotted separately for cells with zero or one focus (background) and in cells with two or more foci of HDAC1-EdU PLA signals. (K) Distributions of numbers of HDAC1-EdU PLA foci per nucleus in the controls and hdac2-21 cells. Only cells with two or more foci are included. (L) Same as for panel K, performed on a separate HDAC2-null clone, hdac2-36. (M) EdU was clicked to a mix of biotin-azide and Alexa Fluor 488-azide to distinguish EdU-positive (S-phase) (red) and EdU-negative (non-S-phase) (gray) cells. The numbers of HDAC1-EdU PLA foci were plotted separately for these two subsets of cells. Blue lines are medians. P values for mean fluorescence intensities were calculated by K-S tests, and those for numbers of foci were calculated by Wilcoxon tests. The number of cells analyzed per sample for each PLA was at least 200 and typically 400 to 500, and the HDAC1-EdU and HDAC2-EdU PLA data shown represent results from 4 independent experiments each. (N and O) Native iPOND confirms elevated recruitment of HDAC1 to nascent DNA in hdac2-null cells, as shown on a Western blot (N) and by quantitation of data from 2 independent experiments (O), where the HDAC1 level in pulldowns was first normalized to PCNA levels in pulldowns and then expressed as a fold change in hdac2-null cells over the wild-type control. Red circles are means. DAPI, 4′,6-diamidino-2-phenylindole.
The results showed that in hdac1-50 null nuclei, the abundance of HDAC2 on nascent DNA increased compared to the wild type based on the mean fluorescence intensity (MFI) per nucleus although not based on the number of foci (Fig. 2D and E). In hdac1-47 null nuclei, the increase of HDAC2 on nascent DNA was more pronounced, affecting both the PLA MFI and the number of foci (Fig. 2F and G). Note that in these experiments, two different pairs of antibodies were used (rabbit anti-HDAC2–mouse antibiotin in Fig. 2D and E and mouse anti-HDAC2–rabbit antibiotin in Fig. 2F and G). Overall, three independent experiments using the latter combination of antibodies confirmed the increased recruitment of HDAC2 to nascent chromatin in hdac1-null cells.
Similarly, more HDAC1 was detected on nascent chromatin in two clones of hdac2 cells than in the wild type (Fig. 2H through J). In particular, the HDAC1/EdU PLA MFI (Fig. 2J) and/or the number of PLA foci (Fig. 2K and L) was dramatically higher in hdac2-null clones than in controls. To distinguish between S-phase and non-S-phase cells, we also introduced a two-azide approach whereby in some Click-It reactions, biotin-azide was spiked with Alexa Fluor 488 (A488)-azide at molar ratios of 1:40 to 1:50 (A488 to biotin), as we documented previously (29). PLA foci were scored separately in A488-positive (EdU-positive [EdU+], replicating) versus A488-negative (EdU-negative [EdU−], nonreplicating) cells to confirm that the HDAC1-EdU PLA signal was enriched in replicating cells (Fig. 2M). Finally, native iPOND (25) assays also showed a tendency toward increased HDAC1 recruitment to nascent DNA upon HDAC2 KO, although not enough data points were collected for statistical significance (Fig. 2N and O). Of note, the increased abundance of HDAC1 or HDAC2 on nascent DNA was not due to the increased EdU incorporation in null cells (see below for more details) or the increased DNA damage, since γH2AX levels in the knockouts were not elevated (data not shown). Overall, the data suggest that compensation between the two HDACs clearly involved an increased presence of the remaining HDAC on nascent chromatin. Since HDAC1 and HDAC2 can homo- and heterodimerize (17, 18), and structural studies suggest a stoichiometry of two HDAC molecules per active complex (30, 31), one interpretation of the data is that in the null cell lines, homodimer complexes of the remaining HDAC are recruited to nascent chromatin instead of HDAC1/HDAC2 heterodimers. Alternatively, only one type of homodimer complex is recruited instead of a mixture of homodimer complexes of either HDAC.
HDAC1 and HDAC2 are only partially redundant for deacetylation of histones in nascent chromatin.Given the increased abundance of the compensating HDACs in nascent chromatin, we next measured the levels of their major target on replicating DNA: newly incorporated hyperacetylated histone H4. We used a PLA to detect the presence of H4K12ac on EdU-labeled, newly replicated DNA as we described previously (29). The specificity of the H4K12ac PLA signal for replicating cells was confirmed previously (29) by clicking EdU to a mixture of biotin- and Alexa Fluor 488-azides at molar ratios of 1:10 to 1:40 (A488 to biotin) and scoring the number of PLA foci and the PLA signal MFI per nucleus in A488-positive (replicating) versus A488-negative (nonreplicating) cells.
We found that hdac2-null lines displayed markedly higher levels of H4K12ac-EdU PLA signals than the controls (Fig. 3A through E). hdac1-null cells also showed elevations of H4K12ac on nascent DNA but to a lesser degree (Fig. 3B through D). Of note, this H4K12ac-EdU signal was nevertheless decreased in hdac1- or hdac2-null cells harvested 4 and 8 h after EdU labeling, indicating that the deacetylation of this PTM was not entirely abolished (data not shown). We also observed that another prominent acetyl mark on histone H4 associated with nascent chromatin, H4K5ac, was elevated in hdac1- and more so in hdac2-null cells (Fig. 3F and G). Additionally, histone H3 lysine 9 acetylation on nascent DNA, which may be another mark characteristic of a newly incorporated histone (11), was also elevated in hdac2-null cells albeit not in an hdac1-null clone (Fig. 3H and I).
Single HDAC deficiency changes the levels of histone acetyl PTMs in nascent chromatin. (A) Example of histone H4K12ac-EdU PLA staining in EdU-positive (S-phase) (green) and EdU-negative (non-S-phase) GM639-Cas9-EV cells. The cell-containing areas of the original image were spliced to increase the cell density for demonstration purposes. (B and C) Mean fluorescence intensities per nucleus (B) and numbers of foci per nucleus (C) of the H4K12ac-EdU PLA signal in EdU-positive and EdU-negative subsets of cells of the indicated cell lines. (D and E) Numbers of foci of H4K12ac-EdU PLA signals in EdU-positive and EdU-negative subsets of cells of additional hdac1- or hdac2-null clones. (F) Example of histone H4K5ac-EdU PLA staining in EdU-positive (S-phase) (green) and EdU-negative (non-S-phase) hdac2-36 cells. (G) Numbers of H4K5ac-EdU PLA foci in EdU-positive and EdU-negative subsets of cells of the indicated cell lines. (H) Numbers of H3K9ac-EdU PLA foci in the indicated cell lines. EdU was clicked only to biotin-azide, and only the cells showing two or more foci are included in the plot. (I) Mean fluorescence intensities of H3K9ac-EdU PLA signals per nucleus in cells with zero to one and two or more foci. In all experiments shown, cells were labeled with 10 to 20 μM EdU for 30 min and fixed. Horizontal dashes in strip chart plots are medians of distributions. P values for mean fluorescence intensities were calculated by K-S tests, and those for numbers of foci were calculated by Wilcoxon tests. ns, not significant.
We also used iPOND to measure H4K12ac in nascent DNA in four independent experiments and found that it was upregulated on average 2.3-fold over the control in hdac2-null cells, although the data did not reach statistical significance (Fig. 4A and B). Finally, to confirm that the remaining HDAC is responsible for the observed level of histone acetylation, we also depleted over 80% of HDAC1 in hdac2-null cells using small interfering RNA (siRNA) (Fig. 4C) and measured the levels of H4K5ac and H4K12ac marks on nascent chromatin by a PLA. Both marks showed dramatic increases upon the depletion of HDAC1 (Fig. 4D and E), suggesting that in hdac2 cells, the histone H4 acetylation level is indeed governed by HDAC1.
Elevated levels of histone H4K12ac in nascent chromatin of hdac2 cells are controlled by HDAC1. (A) iPOND measurement of H4K12ac levels in nascent DNA. The indicated HDAC+ and hdac2-null clones were labeled with 10 μM EdU for 30 min and harvested for nascent DNA precipitation as described by us previously (see Materials and Methods). (B) Data from four independent iPOND experiments performed as described above for panel A with two hdac2-null clones (hdac2-21 and hdac2-36) were quantified. That is, H4K12ac abundances were first quantified by normalizing values to EdU levels in inputs, and for each experiment, the derived values were then compared between wild-type and hdac2-null cells and expressed as fold changes in an hdac2-null clone over the wild type. An open diamond is a mean value. (C) Western blot showing depletion of HDAC1 in hdac2 cells using siRNA. The percentage of remaining protein is indicated below the images. NCL, nucleolin. (D and E) Mean fluorescence intensity per nucleus of H4K12ac-EdU (D) and H4K5ac-EdU (E) PLA signals in EdU-positive and EdU-negative subsets of cells of the indicated cell lines. The graphs represent data from two independent experiments each. P values were derived by K-S tests.
Taken together, the data indicate that HDAC1 and HDAC2 are not equivalent in their abilities to carry out histone deacetylation in nascent chromatin in vivo. In particular, the observed levels of acetyl PTMs on H4 and H3 in the mutant and wild-type lines suggest that despite the compensatory recruitment of extra HDAC1 to nascent chromatin in hdac2 cells, HDAC1 alone will not deacetylate histones to the same level as HDAC2 alone or both proteins together.
Compensation and histone acetylation phenotypes of hdac2 cells are complemented by ectopic expression of HDAC2.We next complemented hdac2 null cells with a lentivirally delivered HDAC2 gene. After transduction and selection, EF-1a promoter-driven ectopic HDAC2 was expressed variably and at an overall lower level than the endogenous HDAC2 in wild-type control cells (Fig. 5B, lanes 1 to 3). Nevertheless, the reexpression of HDAC2 at least in part reversed the hdac2 phenotypes described above. First, it downregulated total cellular levels of HDAC1 compared to the empty vector control (Fig. 5B, compare lanes 4 and 5). Moreover, in individual cells, levels of ectopic HDAC2 and endogenous HDAC1 were inversely related (Fig. 5C). Interestingly, this inverse relationship was not as pronounced in the wild-type control, where the levels of endogenous HDAC1 and HDAC2 were more balanced (Fig. 5D). This is consistent with the idea that endogenous HDAC1 and HDAC2 may autorepress as well as repress each other’s expression at the promoter level (23).
Ectopic expression of HDAC2 reverses elevated HDAC1 and histone acetylation in nascent chromatin. (A) Western blot showing reexpression of HDAC2 in hdac2 cells from an integrated lentiviral vector. e.v., empty vector. (B) Quantitation of immunofluorescence (IF) in situ results of staining of the indicated cells for HDAC1 and HDAC2. The data in the graph are representative of results from two independent experiments. (C and D) Scatterplots of IF in situ results similar to those in panel B showing levels of HDAC1 and HDAC2 in each nucleus of hdac2-27 cells expressing the HDAC2 transgene (C) and the wild-type control, GM639-Cas9-EV (D). (E) Mean fluorescence intensities of the HDAC1-EdU PLA signals in the indicated cell populations, subsetted by the number of PLA foci. (F) Mean fluorescence intensities of histone H4K5ac-EdU (left), H3K9ac-EdU (center), and H4K12ac-EdU (right) PLA signals in the indicated cell populations subsetted by the numbers of PLA foci. Graphs represent data from 2 independent experiments each. P values were determined by K-S tests.
Second, the ectopic expression of HDAC2 in hdac2-null cells also reduced HDAC1 recruitment to nascent DNA compared to the isogenic empty vector-expressing population (Fig. 5E). We also observed reductions in H4K5ac, H3K9ac, and, to a lesser degree, H4K12ac PLA signal intensities on nascent DNA in the HDAC2-reexpressing population (Fig. 5F). In summary, the data support the conclusion that the HDAC1 and histone H3 and H4 acetylation changes that we observed in hdac2-null cells are directly attributable to HDAC2 loss rather than to clonal evolution or heterogeneity.
Bromodomain protein ATAD2 is found on nascent DNA, where its level can be elevated in the absence of HDAC2.Elevated H4K5ac and -K12ac marks can affect histone acetyl PTM readers at the replication fork. H4K5ac and/or H4K12ac marks may be recognized by several bromodomain proteins that have been detected on nascent DNA by iPOND followed by mass spectrometry (MS): BRD2, BRD3, BRD4, and ATAD2 (32–36) and, potentially, BAZ1A (37) and BAZ1B (34, 38). We performed label-free mass spectrometry on iPOND pulldowns of GM639 cells using two different platforms and data-independent acquisition (DIA) (Fig. 6A and B) as well as data-dependent acquisition (DDA) (Fig. 6B through D) modes. ATAD2, BAZ1A, and BAZ1B (Fig. 6B through D) were robustly detected by both approaches at forks in an EdU-biotin-dependent manner in wild-type, hdac1, and hdac2 cells. In contrast, our analyses detected only 1 peptide each of BRD2 and BRD4 and 2 peptides of BRD3 (Fig. 6B). No other bromodomain-containing proteins known to recognize acetyl-histones H3 and H4 (6) were detected in these experiments.
Proteomics of nascent chromatin identifies H4 acetyl PTM reader proteins. (A) Data-independent acquisition (DIA) workflow for iPOND samples. (B) Comparison of two independent label-free LC-MS/MS analyses performed on iPOND samples of HDAC+, hdac1, and hdac2 cells of the GM639 background. Shown are maximum numbers of unique peptides detected for each of the indicated proteins for all three genotypes (experiment 1, Orbitrap Fusion in DDA mode; experiment 2, Q-Exactive HF in DDA and DIA modes). (C) Workflow diagram for label-free iPOND-MS and MaxQuant data for detection of ATAD2 on EdU-labeled DNA in a biotin-specific manner (i.e., pulled down exclusively via the EdU-biotin/streptavidin bead interaction) in wild-type or hdac1- or hdac2-null GM639 cells. (D) Heat map of a cluster of proteins detected in all three genotypes in a biotin-specific manner. ATAD2 is marked by an asterisk. Red, protein is detected; green, protein is absent. Shades of red reflect ion intensities. Note that the experiment was not designed to reliably detect quantitative differences between levels of proteins present on nascent DNA. (E) Sample preparation workflow for a SILAC iPOND experiment. (F) Normalized heavy (H)/light (L) ratios for the wild type (WT) (GM639-Cas9-EV) over hdac2-21 (left) or over hdac2-36 (right) cells were rank ordered and plotted. Data points corresponding to ATAD2 are highlighted in red.
In this study, we followed up on the finding of ATAD2 on nascent DNA since it is a potential new therapeutic target overexpressed in many cancers, including breast, cervical, and endometrial cancers, where its elevated level correlates with decreased survival (39–44). ATAD2 is highly expressed in our GM639 cell line, making it a model useful for understanding the role of ATAD2 in cancer cells. The peptides of ATAD2 detected by MS spanned most of its open reading frame, indicating that the full-length protein was bound to nascent chromatin (data not shown). ATAD2 has a paralog, ATAD2B, that is 94.9% similar (45); however, we detected multiple peptides uniquely attributable to ATAD2 and no peptides unique to ATAD2B.
While ATAD2 was clearly detectable at forks in all three genotypes (HDAC, hdac1, and hdac2), we speculated that its level may be affected by changes in the abundances of its target PTMs, i.e., H4K12ac and H4K5ac, in hdac-null cells. We performed stable-isotope labeling by amino acids in cell culture (SILAC) iPOND-MS with the goal of quantifying levels of proteins on nascent chromatin in wild-type versus hdac2-null cells (Fig. 6E). While this experiment yielded only 40 to 80 proteins for which the heavy/light isotope ratio could be identified, in two hdac2 clones, ATAD2 was among the top proteins enriched over the wild type (Fig. 6F). Also, of these two clones, hdac2-21 cells appeared to have a higher enrichment of ATAD2 than hdac2-36 cells.
We next used a PLA to confirm the association of ATAD2 with nascent chromatin (Fig. 7A) and quantify it across the genotypes. The specificity of the PLA signal was established by demonstrating that it was reduced upon siRNA-mediated ATAD2 depletion (Fig. 7B). ATAD2-EdU PLA foci were more numerous in hdac2-21, hdac2-27, and hdac2-51 cells than in HDAC+ controls (Fig. 7C and D). hdac2-36 cells did not show ATAD2 enrichment in the PLA, suggesting perhaps a lower resolution of the PLA than of SILAC MS. Of note, hdac2-36 cells also have the least altered level of the H4K12ac mark (compared to the HDAC+ control). On the other hand, only hdac1-47 but not hdac1-50, hdac1-5, and hdac1-12 cells had elevated numbers of ATAD2-EdU PLA foci (Fig. 7D and E). The hdac1-47 clone stands out among other hdac-null clones in that it has an in-frame deletion at the 5′ gRNA site (Table 1). While we did not detect HDAC1 protein expression in this clone using two different antibodies, it is possible that the mutant protein is expressed at a very low level, in which case it may have an altered function rather than a null phenotype.
The ATAD2 level in nascent chromatin is affected by HDAC status. (A) Example of an EdU-dependent ATAD2-EdU PLA in GM639-Cas9-EV cells. EdU-labeled cells were clicked to a mixture of biotin- and Alexa Fluor 488-azides. Red and green channels of the same field are shown separately. (B) Western blot demonstrating siRNA-mediated depletion of ATAD2 in hdac1-50 cells and quantitation of ATAD2-EdU PLA fluorescence in the same experiment. Cells with zero foci are included in the plot. (C and D) Numbers of ATAD2-EdU PLA foci in cells of the indicated cell lines. Only cells showing two or more foci are included. (E) Numbers of ATAD2-EdU PLA foci in EdU-negative and EdU-positive subsets of cells of the indicated cell lines. Results shown in panels C through E represent data from 5 independent experiments in total. P values were calculated by Wilcoxon tests. Black lines denote medians of distributions.
Overall, the results establish ATAD2 recruitment to nascent chromatin and its enrichment in the absence of HDAC2 and reinforce the notion that HDAC1 and HDAC2 deficiencies may have different impacts on nascent chromatin.
ATAD2 regulates the level of histone H4 acetylation on nascent DNA.ATAD2 is involved in the transcriptional regulation of many genes (46), including the genes key to the G1/S transition (47), but its role in nascent DNA is not yet clear. Based on the in vitro results, a recent report suggested that by binding to K5 and K12 acetyls of histone H4, ATAD2 may interfere with deacetylation by HDACs (36). To investigate this, we used siRNA-mediated depletion of ATAD2 in our wild-type and hdac-null cells using three siRNAs to the ATAD2 ORF, which were equally effective at depletion (Fig. 7B, Fig. 8A, and data not shown). Of note, siATAD2-3 is the only siRNA of the three that targets both the full-length form of the protein and its putative alternatively spliced shorter isoform 2.
The ATAD2 level affects de novo histone H4 acetyls in nascent chromatin. (A) Western blot demonstrating ATAD2 depletion. NCL (nucleolin) is the internal control. (B) Numbers of H4K12ac-EdU PLA foci in cells of the indicated genotypes transfected with nontargeting siRNA or ATAD2 siRNA. Only cells with two or more foci are included. (C) Pairwise comparisons of numbers of H4K12ac-EdU PLA foci in EdU-negative and EdU-positive subsets of cells of the indicated genotypes transfected with nontargeting siRNA or ATAD2 siRNA. Data in panels B and C represent results from 8 independent experiments in total, 2 of which also included 2 biological replicates. (D and E) IF in situ measurements of C-terminally V5-tagged ATAD2 transgene expression in GM639-Cas9-EV (universal control) cells transduced with the ATAD2 construct (ATAD2) or an empty vector without the tag (e.v.). Endogenous ATAD2 was depleted in the indicated cells by 5′-UTR-targeting siRNA. Expression was assessed with ATAD2 antibody (D) and V5 antibody (E). Graphs represent data from 3 independent experiments. (F) Numbers of H4K12ac-EdU PLA foci in the same cells as in panels D and E. Only cells with more than 5 foci are included. (G) H4K5ac-EdU PLA mean fluorescence intensities (left) and numbers of foci (right) in the same cells as in panels D and E. Only cells with more than 10 foci are included. Data in panels F and G represent results from 3 independent experiments for H4K12ac and from 2 independent experiments for H4K5ac PTM. P values throughout were determined by Wilcoxon tests for numbers of foci and by K-S tests for MFIs.
We measured the level of H4K12ac on nascent chromatin in ATAD2-depleted cells by a PLA using both single-azide (biotin-azide) (Fig. 8B) and two-azide (biotin-azide plus Alexa Fluor 488-azide) (Fig. 8C) approaches. Remarkably, we observed that in ATAD2-depleted cells, the level of H4K12ac on nascent chromatin was typically reduced. This reduction was most pronounced in hdac2 and wild-type cells (Fig. 8B and C), while in comparison, in hdac1-null cells, the level of H4K12ac was less affected by ATAD2 depletion (Fig. 8C). The clones showing the least effect (hdac1-50 [Fig. 8C] and hdac1-12 [data not shown]) were the ones with higher compensatory increases in HDAC2 levels (Fig. 1C). Of note, total cellular H4K12ac was not reduced upon ATAD2 depletion (data not shown). Thus, the data suggest that ATAD2 may interfere with H4K12ac deacetylation, either by virtue of binding to it or through a less direct mechanism. Furthermore, HDAC1 alone is less able to counteract this PTM-protective function of ATAD2 than HDAC2 alone or HDAC1 and HDAC2 together.
To confirm that this protection phenotype is directly linked to ATAD2, we engineered wild-type cells to express ectopic ATAD2 with a C-terminal V5 tag (ATAD2-V5) from an integrated lentiviral cassette and then depleted the endogenous ATAD2 with a Western blot-validated (data not shown) 5′-untranslated-region (UTR)-targeting siRNA, siATAD2-6. Figure 8D and E show immunostaining in situ with antibodies to ATAD2 (Fig. 8D) and to V5 (Fig. 8E), confirming the effective depletion of ATAD2 and expression of ATAD2-V5, which was at a lower level than the endogenous protein. The expression of this ATAD2-V5 as the sole source of ATAD2 in the cell was associated with higher levels of H4K12ac (Fig. 8F) and, to a lesser degree, H4K5ac (Fig. 8G) in nascent chromatin than in cells with depleted ATAD2 and no transgene expression. Thus, we showed that the depletion and reconstitution of ATAD2 in the cell lead to a reduction and an increase, respectively, of histone H4 de novo acetyls, consistent with a role of ATAD2 in preserving this PTM in nascent chromatin.
ATAD2 exerts mild effects on the replication fork progression rate.By immunofluorescence (IF) in situ, ATAD2 protein levels in the nuclei were on average 2-fold higher in the S phase of GM639 cells than during the rest of the cell cycle (data not shown), consistent with previous work (47). ATAD2 depletion was shown to reduce EdU incorporation 2-fold in S-phase cells of the MCF7 breast cancer cell line (36). These data prompted us to examine the contribution of ATAD2 to DNA replication in our model.
Surprisingly, we did not detect a substantial dependence of replication on ATAD2. We saw on average a 5% increase in the S-phase fraction upon ATAD2 depletion in wild-type but not hdac1- or hdac2-null GM639 cells (Fig. 9A). We quantified EdU incorporation during S phase and measured differences in the distributions of EdU incorporation values between EdU-positive cells of the control and ATAD2-depleted lines grouped by HDAC genotype (Fig. 9B and C; see Materials and Methods for quantitation details). For HDAC+ cells, several independent experiments demonstrated no ATAD2-specific effect (Fig. 9C). hdac1 and, more so, hdac2 cells tended to reduce EdU incorporation in ATAD2-depleted compared to mock-depleted cells, although the effect was quite mild (Fig. 9B and C). Similarly, no substantial effect was seen for RD cells that express about 20% of the ATAD2 level seen in GM639 fibroblasts, normal human fibroblasts that express about 10% of the GM639 cell level, and the MCF7 and SKBR3 cell lines (data not shown). The latter two cells lines express some of the highest levels of ATAD2 among breast cancer cell lines (Cancer Cell Line Encyclopedia).
ATAD2 depletion moderately suppresses fork progression in hdac2 cells. (A) Effects of ATAD2 depletion on percent S-phase cells in cell lines grouped by genotype. (GM639-Cas9-EV, WT1, and WT2 for HDAC+; hdac1-50 and hdac1-47 for hdac1; and hdac2-21 and hdac2-36 for hdac2). Percentages of S-phase cells were determined by calculating the fraction of EdU-positive cells after EdU pulse-labeling for 20 min followed by immunofluorescence in situ. The y axis shows differences between percent S-phase values measured for cells transfected with ATAD2 siRNA and those for cells transfected with control siRNA. Data were derived from over 10 independent experiments, each measuring at least 300 cells. The P value was determined by a Wilcoxon test. (B) Representative experiment showing profiles of EdU incorporation in cells with the indicated genotypes and siRNAs. Raw MFI values of at least 300 cells per cell line/siRNA were collected from digital images and normalized as described in Materials and Methods. The left panel includes all cells, EdU negative and EdU positive, and the right panel shows only EdU-positive cells. Vertical marks are distribution means. (C) Summary of data from multiple independent experiments similar to the one shown in panel B. For each experiment, EdU incorporation value data sets were processed as follows. The K-S test D statistic was calculated as a metric of the difference between EdU incorporation of siControl and siATAD2 cells for each genotype. The results were grouped by genotype as described above for panel A and plotted. For visualization purposes, in cases where siATAD2 cells showed reduced EdU incorporation compared to siControl, the value of the D statistic was assigned a negative sign. Black circles denote highly significant differences (P < 0.01), and white circles denote differences with P values of >0.01. Lines indicate means of the D statistic distributions. (D) Experimental design to measure fork progression rates by maRTA and data from a representative maRTA experiment. Cells were labeled sequentially with CldU and IdU for 30 min, harvested, and processed as described in Materials and Methods. Lengths of CldU and IdU segments in 100 to 300 two-segment tracks corresponding to ongoing forks were measured and plotted. P values were calculated by using K-S tests. (E) Summary of data from several independent maRTA experiments. For each experiment, differences between lengths of the 1st-label (CldU) segments in ongoing forks of siControl and siATAD2 cells were expressed as a K-S test D statistic, and the D statistic values were plotted as described above for panel C for each genotype. Black circles are the values that correspond to statistically significant differences (P < 0.01) (GM639-Cas9-EV for the WT; hdac1-50, hdac1-47, hdac1-12, and hdac1-5 for hdac1; and hdac2-21 and hdac2-36 for hdac2).
To assess the effect of ATAD2 depletion on replication at a higher level of resolution, we analyzed the lengths of replication tracks corresponding to ongoing replication forks, as described by us previously (48). Overall, fork progression phenotypes were moderate. Interestingly, in wild-type cells, ATAD2 depletion tended to increase the fork progression rate. In contrast, in two clones of hdac2-null cells, ATAD2 depletion had a moderate but statistically significant negative effect (Fig. 9D and E). hdac1-null clones showed interclonal variation (Fig. 9E). Replication fork progression was no more affected in ATAD2-depleted normal fibroblasts (data not shown). We conclude that ATAD2 contributes to replication fork progression in the absence of HDAC2, suggesting that HDAC1 alone fails to provide the optimal level of the relevant function. Nevertheless, the overall conclusion from our data is that the acute loss of ATAD2 does not suppress DNA replication, regardless of whether the protein is overexpressed or maintained at a normal level.
ATAD2 affects the global RNA synthesis level and replication/transcription cooccurrences.Given the moderate replication phenotypes of ATAD2 depletion and the fact that the protein is considered a transcription facilitator (46), we wanted to verify that ATAD2 affects transcription in the GM639 cell line background. We measured the level of bromouridine (BrU) incorporation into RNA in the nuclei of control and ATAD2-depleted cells by IF in situ. Wild-type and hdac2-null cells were examined since they had demonstrated different replication phenotypes upon ATAD2 depletion. Cells were colabeled with BrU and EdU (the latter in order to distinguish S-phase and non-S-phase cells). The specificity of BrU staining was confirmed by IF analyses of cells labeled with only BrU or EdU (Fig. 10A) and of cells treated with cordycepin, a transcription inhibitor (Fig. 10B). We found that ATAD2 depletion in the GM639 background led to a dramatic reduction of BrU incorporation in both non-S-phase (not shown) and S-phase (Fig. 10C) cells. This was consistently observed in HDAC+ cells in multiple experiments and was a predominant tendency in hdac2 cells (in 6 out of 8 data sets derived from four independent experiments) (Fig. 10D). Overall, the data demonstrate that ATAD2 supports high levels of RNA synthesis independent of the HDAC genotype. Of note, the depletion of ATAD2 from normal human fibroblasts had moderate to no effects on transcription compared to GM639 cells (Fig. 10E), suggesting that elevated RNA synthesis may be associated with the overexpression of ATAD2.
ATAD2 depletion reduces RNA synthesis levels. (A) IF in situ tests for specificity of BrU and EdU detection. GM639-Cas9-EV cells were labeled for 30 min with 5 mM BrU only (left), 10 μM EdU only (middle), and both labels (right). All samples were clicked to biotin-azide and then incubated with both an antibody to BrdU (without denaturation) and an antibody to biotin. (B) GM639-Cas9-EV cells were labeled with EdU and BrU, as described above for panel A, in the absence or presence of 50 μM cordycepin added concurrently with the labels, and subjected to IF in situ as described above. BrU MFI data were plotted separately for EdU− and EdU+ cells. (C) Data from a representative experiment showing distributions of BrU incorporation in cells with the indicated genotypes and siRNAs. Cells were labeled simultaneously with 5 μM EdU and 5 mM BrU for 1 h, harvested, stained for IF in situ, and imaged. The acquired data were subsetted by EdU incorporation status (EdU positive [S phase] and EdU negative [non-S phase]). Mean BrU incorporation levels in 200 to 300 S-phase nuclei per genotype/siRNA are shown. Vertical marks are distribution means. (D) Statistics summary for independent experiments similar to the ones for panel C. For each experiment, differences between BrU incorporation in siControl and siATAD2 cells were expressed as a K-S test D statistic and plotted (GM639-Cas9-EV for the WT [3 independent experiments with biological replicates] and hdac2-21, hdac2-27, and hdac2-51 for hdac2 [4 independent experiments with biological replicates]). Black circles are the values that correspond to statistically significant differences (P < 0.01). (E) Levels of BrU incorporation in EdU-positive (S-phase) primary fibroblasts upon ATAD2 depletion. Cells were labeled simultaneously with 5 mM BrU and 10 μM EdU, fixed, and analyzed by IF in situ, similarly to panel C.
The level of transcription can have an impact on the abundance of RNA-DNA hybrids in a cell (49, 50) and deregulate the spatiotemporal separation of replication and transcription (51), increasing the likelihood of replication-transcription cooccurrences (52). We evaluated total levels of RNA-DNA hybrids in ATAD2-depleted cells by dot blotting of the purified, RNase A-treated genomic DNA with the S9.6 antibody in five independent experiments. The S9.6 antibody recognizes RNA-DNA hybrids and may cross-react with double-stranded RNA (dsRNA) and single-stranded RNA (ssRNA) (53–55). Thus, the specificity of the signal was confirmed by treating DNA samples with RNase H1. The assay showed a <20%, on average, increase in total RNA-DNA hybrids upon ATAD2 depletion, albeit the result was not statistically significant (data not shown).
To probe specifically for the presence of RNA next to replication forks, we developed an S9.6-EdU PLA (Fig. 11). In IF in situ, S9.6 antibody typically produces nuclear staining with significant enrichment in nucleoli as well as some cytoplasmic staining (56). In contrast, the S9.6-EdU PLA signal was overwhelmingly nuclear (Fig. 11A). As expected, this PLA signal was dependent on transcription (Fig. 11B and C) and EdU incorporation (Fig. 11D and E). Also, the expression of ectopic RNase H1 in cells (Fig. 11F) reduced (albeit did not eliminate) this S9.6-EdU PLA signal (Fig. 11G), similar to the way that it reduced the nuclear S9.6 signal measured by IF in situ (Fig. 11H) (57). Enrichment of the ectopic RNase H1 in nucleoli (57) also allowed us to observe, upon costaining for this enzyme, that the nuclear S9.6-EdU foci were not restricted to nucleoli (data not shown). In summary, the assay clearly recognizes RNA-DNA hybrids in nascent DNA, but a contribution of other S9.6-reactive RNA species to the overall signal cannot be ruled out. Remarkably, the assay demonstrated that ATAD2 depletion reduced the S9.6-EdU PLA signal in both wild-type and hdac2-null cells (Fig. 11I and J).
ATAD2 depletion reduces replication-transcription cooccurrence. (A) Example of an S9.6-EdU PLA in GM639-Cas9-EV (UC) cells. (B and C) S9.6-EdU PLA signal quantification as MFI (B) or number of foci (C) in WT24 cells. Cells were incubated with 100 μM cordycepin (or without the drug) for 1 h and then labeled with 20 μM EdU for 20 min with or without cordycepin, respectively. (D and E) S9.6-EdU PLA signal quantification as MFI (D) or number of foci (E) in HDAC+ GM639 cells (WT24). Cells were labeled with 20 μM EdU for 20 min, and EdU was clicked to a mixture of biotin- and Alexa Fluor 488-azides at a 50:1 molar ratio. Data were subsetted by EdU incorporation status. (F) IF in situ demonstrating overexpression of RNase H1 in GM639-Cas9-EV (UC) cells stably transfected with the RNase H1 expression vector. (G) Quantitation of S9.6-EdU PLA results (representative of data from 3 independent experiments) demonstrating a reduction of the nuclear mean fluorescence intensity of the PLA signal in EdU-positive UC cells overexpressing RNase H1 compared to the untransfected parental cells. Cells were labeled with EdU and processed as described above for panels A and B. (H) Quantitation of IF results (representative of data from 2 independent experiments) demonstrating a reduction of nuclear S9.6 fluorescence in UC cells overexpressing RNase H1 compared to the untransfected parental cells. (I and J) Representative data from 2 independent experiments showing a reduction of S9.6-EdU PLA signals as mean PLA fluorescence intensities (I) and numbers of foci (J) in the indicated cells upon depletion of ATAD2 with a mixture of siATAD2-3 and siATAD2-5. Cells were incubated with 20 μM EdU for 30 min prior to harvest, fixed, and clicked to biotin-azide. Cells containing 10 or more foci were considered positive based on the data shown in panels D, E, and G. P values for mean fluorescence intensities were calculated by K-S tests, and those for numbers of foci were calculated by Wilcoxon tests.
In summary, our results suggest that a high expression level of ATAD2 in GM639 cells is associated with elevated levels of general transcription. The data also suggest the possibility that the presence of ATAD2 is associated with more RNA transcription specifically in the vicinity of newly replicated DNA, which can include the formation of RNA-DNA hybrids. At the same time, the total, steady-state level of RNA-DNA hybrids in a cell is unlikely to be affected by the acute depletion of ATAD2.
DISCUSSION
In this study, we took advantage of the functionalities offered by a novel approach to the detection and quantitation of histone and nonhistone proteins in nascent chromatin, the proximity ligation assay, to monitor the fate of de novo histone PTMs in nascent chromatin and their dependence on HDAC1, HDAC2, and the bromodomain acetyl reader ATAD2. We found that HDAC1 and HDAC2 affect each other and are not completely redundant with respect to histone deacetylation in nascent chromatin. Knockout of one of the HDACs resulted in compensatory increases of not only the whole-cell level of the other HDAC, as was shown previously, but also its nascent chromatin-bound level, as we demonstrate here for the first time.
Importantly, despite this compensatory feedback, the levels of H4K12ac and H4K5ac PTMs in nascent chromatin were nevertheless elevated in hdac single mutants. For hdac2-null cells, we also confirmed that these elevated levels of H4K12ac and H4K5ac depend on HDAC1 and can be complemented by the reintroduction of HDAC2. These data suggest that active complexes homodimeric for HDAC1 or HDAC2 are not as efficient at accessing and/or removing these PTMs as complexes containing both HDAC1 and HDAC2.
In addition, our data consistently revealed that knocking out HDAC2 produced stronger phenotypes than the HDAC1 knockout. In particular, increases of H4K12ac, H4K5ac, and H3K9ac on nascent chromatin were either more pronounced in or limited to hdac2 cells. Also, the ablation of HDAC1 versus HDAC2 had different outcomes for the binding of the H4 acetyl reader ATAD2 to the H4K12ac mark and, importantly, for the way in which the level of H4K12ac responded to the manipulation of the ATAD2 level, as we discuss in more detail below.
Putative PTM-protective role of ATAD2 in relation to HDAC1 and HDAC2.One way in which the elevated acetyl marks can affect DNA metabolism is by upregulating the binding of the cognate reader proteins that can perform diverse activities once bound. We used quantitative proteomic analyses and proximity ligation assays to demonstrate an increased binding of ATAD2 to the nascent chromatin of hdac2-null cells compared to the wild type. As a rule, this was not the case for hdac1 cells. In other words, despite the inevitable interclonal variation, clonal evolution in the absence of HDAC2 readily yielded the outcome where ATAD2 binding was upregulated, compared to that in the absence of HDAC1. While the degree to which ATAD2 binding was upregulated appeared moderate, it should not be discounted. Competition with other bromodomain proteins and the possibility that only a subset of nascent DNA sites attract ATAD2 can mask the true magnitude of the change.
The observation that hyperaccumulation of ATAD2 is an epigenetic trait that is more often observed in hdac2 clones, i.e., when only HDAC1 is present, may suggest that HDAC1 is less able to compete with ATAD2 for access to H4K12ac than HDAC2. This is consistent with an observation made with purified ATAD2 and HDAC1 in vitro where the binding of ATAD2 to histone tail peptides precluded the binding of HDAC1 (36). If so, then the depletion of ATAD2 should have a more pronounced impact on H4K12ac levels in hdac2 cells than in hdac1 cells. This is, in fact, what we observed. ATAD2 depletion triggered a marked reduction of H4K12ac PTM from nascent chromatin in hdac2 and wild-type cells, and the reconstitution of ATAD2 expression in wild-type cells reversed this effect. The reduction in H4K12ac upon ATAD2 loss was smaller or nonexistent in hdac1 cells, and the two clones that had a higher compensatory elevation of HDAC2 levels (hdac1-50 and hdac1-12), showed the smallest effect of ATAD2 depletion on H4K12ac levels. Overall, these data not only are consistent with the notion that ATAD2 can protect the H4K12ac PTM from removal in vivo but also suggest that HDAC1 may be unable to effectively remove the PTM when ATAD2 is bound to it.
Compared to HDAC1, HDAC2 may compete with ATAD2 for binding to H4ac effectively. However, since H4K12ac is still elevated in hdac1-null clones, it is possible that HDAC2 is less efficient than HDAC1 at removing the PTM after having gained access to it. In summary, our results are consistent with the notion that the dynamics of the removal of de novo histone H4 acetyls is determined not only by the rate at which HDAC1 or HDAC2 can deacetylate H4 but also by the efficiency with which these HDACs can gain access to the target PTM. This model needs further testing to determine whether the H4 acetyl-recognizing module of ATAD2, the bromodomain, is responsible for the protective effect; at present, however, we were unable to express a bromodomain mutant of ATAD2 at a level comparable to that of the wild-type protein in order to address this question.
Nevertheless, our data suggest that the high level of ATAD2 expressed by GM639 SV40-transformed fibroblasts keeps the abundance of de novo H4 PTMs in nascent chromatin abnormally high, and the PTM level is adjusted once ATAD2 is downregulated, an effect not seen in primary fibroblasts with low steady-state levels of ATAD2. These findings illustrate the notion that PTM readers may control cognate PTM levels, particularly if they are overexpressed and/or if PTM erasers are mutated or silenced, as may occur in cancer cells.
Roles of ATAD2 in replication and transcription.ATAD2 was originally characterized as a transcriptional coactivator in cancer cells, particularly for proliferation-related genes (41). Embryonic stem (ES) cells are another instance where ATAD2 is overexpressed, and in mouse ES cells, it is enriched on transcribed genes rather than on promoter regions (46). ATAD2 knockdown was shown to decrease histone turnover in ES cells (46), although previous work by the same group described the opposite effect in a cancer cell line, H1299 (40). ATAD2 was suggested to be a generalist facilitator of transactions on chromatinized DNA similar to the FACT complex. As for FACT (58), ATAD2 was found on nascent chromatin by our work and in a previous study (36). FACT is involved in replication, albeit its exact role may be unclear, and its contribution can vary in different cells (reviewed in reference 59; see also reference 60). Prompted by our findings and the literature (36, 44), we thoroughly searched for a similar role for ATAD2 in DNA replication. However, we observed only mild replication phenotypes of an acute loss of ATAD2 in the SV40-transformed human fibroblast background that overexpresses ATAD2 and in other backgrounds. Only HDAC2 deficiency led to a discernible negative effect of ATAD2 depletion on the replication fork progression rate. On the other hand, we noted a marked negative effect of ATAD2 loss on global transcription, which is not always found in FACT deficiency (59).
It is still possible that the absence of ATAD2 affects only a subset of replication forks. It has been suggested that heterochromatin replication may be more dependent on ATAD2 (36). We have not ruled this out; however, in our results thus far, the rate of bromodeoxyuridine (BrdU) incorporation in late-S-phase cells closely matched that in early-S-phase cells in ATAD2-depleted fibroblasts (P. Lazarchuk and J. M. Sidorova, unpublished data).
A high level of ATAD2 may affect replication positively by increasing histone mobility and keeping chromatin open and negatively by also enabling a high level of transcription during S phase, which may reduce the spatiotemporal separation of replication and transcription (51) and increase the likelihood of replication-transcription conflicts (50). We addressed this by quantifying the proximity of S9.6 antibody-reactive RNA species to nascent DNA by a PLA. ATAD2 depletion markedly reduced this PLA signal in wild-type and hdac2-null cells, which is consistent with a reduction of RNA synthesis in ATAD2-depleted cells. It can be speculated that the S9.6-EdU PLA signal is more likely to mark transcription on newly duplicated DNA behind the fork rather than the more toxic head-on replication-transcription collision with an RNA polymerase ahead of the fork (61). Our result does not rule out that ATAD2 overexpression increases toxic R loops at forks. However, the more conservative interpretation is that a high ATAD2 level reduces the healthy separation between replication and transcription, whereby RNA synthesis at a given site is halted for approximately 30 min after fork passage (51).
Our data can be summarized in a working model that envisions a buffered balance between de novo PTM erasers and readers/protectors, which together contribute to the observable rate of nascent chromatin maturation and its influence on DNA replication and transcription near nascent DNA (Fig. 12). Highly abundant ATAD2 is recruited to de novo histone H4 acetyls, where it “locks” an open chromatin state by keeping histone mobility high and delaying deacetylation. This may positively affect both replication and transcription in its vicinity; however, replication-transcription cooccurrences may partially offset this positive effect. The HDAC1/HDAC2 heterodimer, most likely in the context of the nucleosome remodeling and deacetylation (NuRD) complex that is found at forks (15, 58), limits the open window of nucleosome mobility at the fork by being able to reposition nucleosomes while deacetylating them (16). Moreover, via HDAC2, this complex may compete with ATAD2 and thus limit its activity in nascent chromatin. This also buffers the effect of the acute loss of ATAD2 on replication in HDAC+ cells. In hdac2 cells, however, the HDAC1 homodimer may be less able to compete with ATAD2, making the contribution of ATAD2 more prominent. Thus, the depletion of ATAD2 in hdac2 cells may have a larger effect. In addition, upon the loss of ATAD2, the complex homodimeric for HDAC1 may deacetylate nascent chromatin too fast compared to the HDAC1/HDAC2 or homodimeric HDAC2 complexes, potentially impeding replication.
Working model of the relationship between HDAC1, HDAC2, ATAD2, maturation of the nascent chromatin, and DNA metabolism. The open (hyperacetylated)-to-closed (hypoacetylated) dynamic in chromatin, which takes place during replication, may affect replication itself as well as local transcription. Activities and levels of HDAC1, HDAC2, and ATAD2 may delay (e.g., when the ATAD2 level is high) or speed up (e.g., when there is no ATAD2 and no HDAC2) chromatin maturation. See Discussion for more details. RNApol, RNA polymerase.
In conclusion, the coexistence of the facilitator and PTM protector functions of ATAD2 in the context of ATAD2 overexpression may be a factor particularly relevant to cancer cells. By comparing normal, transformed, and cancer-derived cell lines of a fibroblastic lineage, we observed that overexpression of ATAD2 means that the cells are more dependent on it for their transcription and their balance of acetylated H4. We also uncovered elevated levels of nascent DNA-associated RNA in ATAD2-overexpressing fibroblasts, which may offset the advantages brought by elevated ATAD2 levels. These findings provide a framework for understanding how cancer cells may become addicted to high levels of ATAD2 and the therapeutic opportunities that this may offer.
MATERIALS AND METHODS
Cells and culture.The SV40-transformed human GM639 fibroblast cell line was described previously (62–64). The human embryonic rhabdomyosarcoma cell line RD and the primary human fibroblast line HFF4 were described by us previously (25, 65).
GM639, RD, and HFF4 cells were grown in Dulbecco modified minimal essential medium (DMEM) supplemented with l-glutamine, sodium pyruvate, 10% fetal bovine serum (FBS) (HyClone, Ogden, UT), and antibiotics. All cell lines were kept in a humidified 5% CO2, 37°C incubator. Mycoplasma testing was performed regularly using the UW/FHCRC Cancer Consortium Shared Resource specimen-processing service (https://sharedresources.fredhutch.org/services/mycoplasma-testing). If required, cells were treated with Plasmocin (InvivoGen) according to the manufacturer’s protocol.
Drugs and other reagents.A stock solution of 5-iododeoxyuridine (IdU) was made at 2.5 mM in phosphate-buffered saline (PBS), a stock solution of 5-chlorodeoxyuridine (CldU) was made at 10 mM in PBS, and a stock solution 5-ethynyldeoxyuridine (EdU) was made at 10 mM in dimethyl sulfoxide (DMSO). IdU and CldU were purchased from Sigma-Aldrich, and EdU was purchased from Sigma-Aldrich and Click Chemistry Tools. IdU and CldU were used at concentrations of 50 μM, and EdU was used at 10 or 20 μM. A stock solution of cordycepin (Thermo Fisher) was made at 50 mM in DMSO and used at 50 μM, and a stock solution of BrU (TCI America or Sigma-Aldrich) was made at 125 mM in PBS and used at 5 mM. All reagent stocks were stored at −20°C.
RNA interference (RNAi)-mediated depletion.siRNAs against ATAD2 ORFs (Hs_ATAD2_2, Hs_ATAD2_3, and Hs_ATAD2_5), the 5′ UTR (Hs_ATAD2_4 and HS_ATAD2_6), and HDAC1 (Hs_HDAC1_6) and a negative-control nontargeting siRNA (siControl) were purchased from Qiagen and transfected using Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer’s protocol. Experiments were performed at 36 to 48 h posttransfection. Depletion was verified for each transfection by Western blotting.
CRISPR-Cas9-mediated gene deletion.For HDAC1 or HDAC2 deletion, the two gRNAs per gene and the procedures were described previously (25). Briefly, GM639 cells were transduced with a lentiviral vector with Cas9 and a red fluorescent protein (RFP) expression cassette and flow sorted for positive cells. These cells, GM639-Cas9, were further transduced with dgRNA-expressing or empty lentiviral vectors carrying puromycin- and green fluorescent protein (GFP)-selectable markers. Transduced mass cultures were selected with puromycin or via cell sorting for GFP-positive cells. The selected mass culture carrying an empty vector, GM639-Cas9-EV, was used as a universal control (UC) in the subsequent assays. For dgRNA-expressing cells, reduction in the expression of HDAC1 or HDAC2 was verified by Western blotting. Individual clones were subsequently derived from these mass cultures and Western blot verified, and the regions surrounding gRNA sites were PCR amplified and sequenced. Some of the selected clones that had wild-type levels of HDAC1 and HDAC2 expression and the wild-type sequence in both gRNA sites were used as additional wild-type controls in functional assays (WT1, WT2, WT24, and WT41).
Constructs.To complement the HDAC2 gene, a FLAG-tagged HDAC2 ORF in the pcDNA3.1 vector was purchased from GenScript. This ORF, which lacked exon 14 of the gene, was PCR amplified with primers that added BamHI and NheI sites and cloned into the BamHI- and NheI-digested pLenti-EFS-T2A-BSD lentiviral expression vector. The full ORF was restored using a Q5 site-directed mutagenesis kit (NEB) to delete the C-terminal FLAG tag and insert exon 14 DNA. The resulting construct, which contains 20 extra amino acids at its C terminus before the T2A linker and the blasticidin resistance gene, as well as the parental empty vector were transduced into an hdac2-null cell line, and the recipient cells were selected with blasticidin.
Overexpression of C-terminally tagged RNase H1 was carried out with ppyCAG_RNaseH1_WT, a gift from Xiang-Dong Fu (Addgene plasmid 111906 [http://n2t.net/addgene:111906]; RRID, Addgene_111906), after transfection and hygromycin selection. Overexpression was confirmed by IF in situ with antibodies to RNase H1 and to the tag (V5), which showed concordant results.
The C-terminally tagged ATAD2 ORF (ATAD2-V5) was derived from GFP-ATAD2 (a gift from Kyle Miller) (Addgene plasmid 65370 [http://n2t.net/addgene:65370]; RRID, Addgene_65370) and transferred into the pLenti-EFS-T2A-BSD lentiviral expression vector by using a Gibson assembly kit (NEB).
Antibodies.Antibodies were as follows: mouse anti-HDAC1 (catalog no. 5356; Cell Signaling), rabbit anti-HDAC1 (catalog no. NB100-56340SS; Novus Biologicals), mouse anti-HDAC2 (catalog no. 5113; Cell Signaling), rabbit anti-HDAC2 (catalog no. 57156; Cell Signaling), anti-PCNA (catalog no. sc-56; Santa Cruz), antinucleolin (catalog no. 396400; Life Technologies), rabbit anti-ATAD2 (catalog no. 61369; Active Motif), rabbit anti-ATAD2 (catalog no. 23894-AP; Proteintech), antibiotin (mouse) (catalog no. MB-9100; Vector Laboratories), antibiotin (rabbit) (catalog no. A150-109A; Bethyl), anti-H4K12ac (catalog no. 39165; Active Motif), anti-H4K5ac (catalog no. 39699; Active Motif), anti-H3K9ac (catalog no. 9649; Cell Signaling); rabbit anti-BrdU (catalog no. GTX128091; GeneTex), rat anti-BrdU/CldU (catalog no. NB500-169; Novus Biologicals), mouse anti-BrdU/IdU (catalog no. 347580; BD Biosciences), mouse anti-RNA-DNA hybrid (clone S9.6) (catalog no. MABE1095; Millipore-Sigma), rabbit anti-RNA-DNA hybrid (clone S9.6) (catalog no. Ab01137-23.0; Absolute Antibody), rabbit anti-RNase H1 (catalog no. 15606-1-AP; Proteintech), and goat anti-V5–fluorescein isothiocyanate (FITC) (catalog no. A190-119F; Bethyl).
Proteins were visualized on Western blots by ECL (Thermo Scientific) and quantified using a FluorChem imager (Alpha Innotech). For presentation, images were saved in TIFF format, adjusted for brightness/contrast, cropped using Adobe Photoshop or CorelPhotoPaint, and then assembled into figures in CorelDraw. Image brightness/contrast adjustments were made across all lanes of each protein measured. In some cases, the lane order was changed, and extra lanes were deleted.
Microfluidics-assisted replication track analysis.Microfluidics-assisted replication track analysis (maRTA) was done as described previously (48, 63, 65). Microscopy of stretched DNAs was performed on a Zeiss Axiovert microscope with a 40× objective, and images were captured with a Zeiss AxioCam HRm camera. Lengths of tracks were measured in raw merged images using Zeiss AxioVision software. Fluorochromes were Alexa Fluor 594 for CldU and Alexa Fluor 488 for IdU. Details of statistical analysis are described in the figure legends.
Proximity ligation assay.A proximity ligation assay (PLA) was performed using the DuoLink red detection kit and DuoLink anti-mouse and anti-rabbit antibodies (catalog no. DUO92008, DUO92001, and DUO92002, respectively; Millipore-Sigma) as described by us previously (29).
Immunofluorescence in situ for EdU and BrU incorporation.Cells grown on coverslips were fixed in 2% formaldehyde in PBS, permeabilized in 0.5% Triton X-100 in PBS, and incubated in a Click-It reaction mix as described previously (65). Alexa Fluor 488-azide (Click It Chemistry Tools) was used in Click-It reactions for EdU incorporation analyses, and biotin-azide (Click It Chemistry Tools) was used for PLAs and for EdU-BrU costaining. In some PLAs, a mixture of Alexa Fluor 488 and biotin-azide was used at a 40:1 molar ratio, respectively. For BrU-EdU costaining after Click-It reactions with biotin-azide, cells were blocked in PBS–3% bovine serum albumin (BSA)–3% normal goat serum and then incubated with anti-BrdU (GeneTex) and antibiotin (Vector Laboratories) antibodies, followed by Alexa fluorophore-conjugated secondary antibodies (Invitrogen). Note that both RFP and GFP signals, if still expressed in the cell lines analyzed, are too low to affect the immunofluorescence (IF) signal; in addition, GFP fluorescence is inhibited by the click reaction.
Quantitative image-based cytometry (QIBC) of IF in situ and PLAs.Images of cells were collected under a Zeiss Axiovert 200M microscope with a 40× objective using AxioVision or Micro Manager software. Digital images were analyzed with the Fiji ImageJ software package with custom macros. Mean fluorescence intensity (MFI) values per nucleus in up to three channels were collected for several hundred nuclei per sample. To count PLA foci, the macro was modified, as described in detail previously (29), to include the identification of maxima of PLA fluorescence within the nuclear region of interest (ROI). Raw EdU MFI values showed a bimodal distribution reflecting the presence of EdU− (non-S-phase) and EdU+ (S-phase) cells. For quantification of PLA or BrU signals in S-phase versus non-S-phase cells, the data were subsetted based on the EdU value, and the PLA MFI and the number of foci or BrU MFI distributions were plotted separately for the subsets. For EdU signal quantitation in EdU+ cells, the following algorithm was employed. To account for potential differences in the background across samples, nuclear MFI values collected in the EdU fluorescence channel for each sample were divided by the mean of the data set’s first quartile (which corresponds to EdU-negative cells, by a very conservative margin). The resulting normalized distributions were subsetted to focus exclusively on the values corresponding to EdU+ cells. The EdU+ subsets were subjected to pairwise Kolmogorov-Smirnov (K-S) tests to determine maximum differences between them (the D statistic) and the corresponding P values. For all other IF experiments, MFI values were not normalized. Analyses and graphing of the data were done in R Studio.
Chromatin fractionation.Fractionation into nucleoplasm and chromatin was done as described previously (25, 66). Briefly, isolated nuclei were extracted with B1 buffer (25 mM NaCl, 2 mM EDTA, 50 mM Tris-HCl [pH 8.0], 1% NP-40, protease inhibitors) by incubation for 15 min on ice, followed by one round of sonication (10 s on and 30 s off) in a Bioruptor Pico instrument (Diagenode), and pelleted at low speed. Supernatants were collected for the nucleoplasm, and chromatin pellets were sonicated in a cup horn sonicator (Misonix) in B1 buffer at 20 W for 6 min in 30-s-on, 30-s-off mode. Chromatin lysates were centrifuged at 16,100 × g for 10 min, and supernatants were collected.
iPOND and mass spectrometry.Immunoprecipitation of nascent DNA (iPOND) was performed as described previously (25), except that lysates were generated by sonication in a Bioruptor Pico instrument (Diagenode). The following modifications were introduced for iPOND samples designated for mass spectrometry: streptavidin pulldowns were washed once for 5 min each, in the following order: radioimmunoprecipitation assay (RIPA) buffer, 1 M NaCl, SDS buffer (50 mM Tris-HCl [pH 8.0], 1% SDS), low-salt buffer (1% Triton X-100, 20 mM Tris [pH 8.0], 2 mM EDTA, 150 mM NaCl), and lithium chloride buffer (100 mM Tris [pH 8.0], 500 mM LiCl, 1% Igepal). Beads were then rinsed once with PBS and treated as follows.
For experiment 1, a label-free liquid chromatography-tandem mass spectrometry (LC-MS/MS) run (in DDA mode), beads were boiled in lithium dodecyl sulfate (LDS) sample buffer (Invitrogen) and loaded onto a Bis-Tris polyacrylamide gel. Samples were run into the gel for 15 min at 150 V, and the entire lanes between the bottom of the well and the dye front were excised, washed in 5% acetic acid–30% methanol for 20 min and then with 1 to 1.5 ml deionized water for 20 min, and stored at 4°C. After reduction and alkylation, proteins were in-gel digested with trypsin, and tryptic peptides were recovered and desalted using a C18 micro ZipTip (Millipore). Liquid chromatography coupled to mass spectrometry was carried out on a NanoEasy-1000 system (Thermo Scientific) coupled to an Orbitrap Fusion mass spectrometer (Thermo Scientific). Mass spectrometry data were analyzed using MaxQuant v1.6. Data analysis and visualization were done in Perseus.
For experiment 2, label-free LC-MS/MS (DDA and then DIA modes), streptavidin bead pulldowns were washed twice with 50 μM ammonium bicarbonate and stored at 4°C. For further processing, pulldowns were heat denatured, reduced, alkylated, and then digested with trypsin. The digested samples were desalted, resuspended in 0.1% formic acid in water, and stored at −20°C until use. Data acquisition was done as described previously (67), using a NanoAcquity ultraperformance liquid chromatography (UPLC) system (Waters) and a Q-Exactive HF quadrupole Orbitrap mass spectrometer (Thermo Scientific).
For SILAC iPOND-MS, heavy amino acids were l-arginine-[13C6, 15N4]HCl and l-lysine-[13C6, 15N2]HCl (Cambridge Isotope). DMEM for SILAC (catalog no. PIA33822) and dialyzed FBS (catalog no. 26-400-044) (Thermo Fisher) were used for cell growth. Cells were fully substituted with heavy amino acids by growing for at least five passages.
In SILAC and label-free iPOND-MS experiments (DDA mode) for Fig. 6C through F, iPOND pulldowns resolved by PAGE were excised and washed for 15 min each with water, a 50:50 mixture of acetonitrile-water, acetonitrile, and 100 mM ammonium bicarbonate, followed by a 50:50 mixture of acetonitrile–100 mM ammonium bicarbonate. After the removal of the final solution, the gel slices were dried by vacuum centrifugation. The dried gel slices were reduced in 10 mM dithiothreitol in 100 mM ammonium bicarbonate at 56°C for 45 min. The solution was discarded, and the gel slices were alkylated in 55 mM iodoacetamide in 100 mM ammonium bicarbonate in the dark at ambient temperature for 30 min. The gel slices were next washed with 100 mM ammonium bicarbonate for 10 min on a shaker, an equal amount of acetonitrile was added, and washing was continued for 10 min with shaking. The solution was discarded, and the gel slices were dried by vacuum centrifugation. The gel slices were cooled on ice, and an ice-cold solution of 12.5 ng/μl trypsin (Promega, Madison, WI) in 100 mM ammonium bicarbonate was added to the gel slices. After 45 min, the trypsin solution was discarded, and a volume of 50 mM ammonium bicarbonate was added to the gel slices for overnight incubation at 37°C on a shaker. Samples were centrifuged, and the supernatant was saved. Peptides were further extracted from the gel slices with 0.1% trifluoroacetic acid (TFA) and vortexing for 30 min. An equal volume of acetonitrile was added, and the samples were vortexed for an additional 30 min. The samples were centrifuged, and the digestion supernatant and the extraction supernatant were pooled and lyophilized by vacuum centrifugation. The resulting peptides were desalted using C18 micro ZipTips (Millipore) and eluted with 70% acetonitrile–0.1% TFA. The desalted material was dried by vacuum centrifugation.
The samples were reconstituted in 20 μl of 2% acetonitrile in 0.1% formic acid, and 18 μl was analyzed by LC-electrospray ionization (ESI)-MS/MS with an Easy-nLC1000 instrument (Thermo Scientific, Waltham, MA) coupled to an Orbitrap Fusion mass spectrometer (Thermo Scientific, Waltham, MA) using a trap column configuration. In-line desalting was accomplished using a 100-μm by 20-mm reversed-phase trap column packed with Magic C18 AQ (5-μm, 200-Å resin; Michrom Bioresources, Auburn, CA), followed by peptide separations on a 75-μm by 250-mm reversed-phase column packed with Magic C18 AQ (5-μm, 100-Å resin; Michrom Bioresources, Auburn, CA) directly mounted on the electrospray ion source. A 90-min gradient from 7% to 28% acetonitrile in 0.1% formic acid at a flow rate of 300 nl/min was used for chromatographic separations. A spray voltage of 2,100 V was applied to the electrospray tip, and the Orbitrap Fusion instrument was operated in the DDA mode, switching automatically between MS survey scans in the Orbitrap instrument (automatic gain control [AGC] target value of 400,000, resolution of 120,000, and injection time of 50 ms) and collision-induced dissociation MS/MS spectrum acquisition in the linear ion trap (AGC target value of 2,000 and injection time of 300 ms) with a “top-speed” setting of 2 s. Monoisotopic precursor selection (MIPS) was set to “peptide,” and charge states from 2 to 6 were selected for MS/MS. Dynamic exclusion was enabled with the duration set to 30 s.
The collected data were analyzed with MaxQuant v1.6 using the built-in label-free quantification (LFQ) and SILAC quantification settings. Filtering, statistical analyses, and visualization were done in Perseus v1.5.8.5.
ACKNOWLEDGMENTS
We are grateful to Ray Monnat, Jr., Eleanor Chen, Michael Phelps, and the Chen lab for support and fruitful discussions. We thank Amina Mohammed for technical assistance with RNA labeling experiments. This study was made possible by the Proteomics and Cellular Imaging FHCRC/UW Cancer Consortium Cores, and we are thankful to the Core leaders Phil Gafken and Julio Vazquez and staff members Lisa Nader Jones and Dave MacDonald for all their help.
This work was supported by the National Institutes of Health (GM115482 to J.M.S. and GM103533, CA192983, and AG013280 to M.M.) and by the American Cancer Society (PF-17-036-01 to A.F.). This work was also supported by the Proteomics and Cellular Imaging Shared Resources of the Fred Hutch/University of Washington Cancer Consortium (National Institutes of Health grant CA015704).
We declare no conflicts of interest.
FOOTNOTES
- Received 4 September 2019.
- Returned for modification 25 September 2019.
- Accepted 21 January 2020.
- Accepted manuscript posted online 3 February 2020.
- Copyright © 2020 American Society for Microbiology.
