Mammalian Cells Lacking HMGB1 Are More Sensitive to Ionizing Radiation
Previous results indicated that a given dose of UV irradiation produced almost twice as many thymidine dimers in mammalian cells lacking HMGB1 and yeast cells lacking Nhp6 proteins compared to wild type cells 
. We then asked whether ionizing radiation also produced more DNA damage in Hmgb1
We irradiated primary wild type and Hmgb1
−/− MEFs with 10 Gy of gamma rays; we measured the formation of single-stranded and double-stranded DNA breaks in individual cells by means of the comet assay 
, whereby in the presence of an electrophoretic field short DNA fragments migrate out of the lysed cell and into the agarose, whereas intact DNA remains confined (, left). The tail moment, which is a measure of DNA fragmentation, indicated that Hmgb1
−/− MEFs contained more DNA breaks before irradiation (, right). The number of DNA breaks induced by irradiation was higher in Hmgb1
−/− cells; this could not be ascribed to defective DNA repair since the cells were subjected to the assay immediately after irradiation, before DNA repair could deal with the breaks. We also quantitated γH2AX levels after irradiation with gamma rays (): substantially more H2AX is phosphorylated in Hmgb1
−/− cells relative to wild type cells after 1 h, but the difference subsides after 6 h. This suggests that Hmgb1
−/− cells can repair effectively double strand breaks.
Hmgb1−/− nuclei are more accessible to DNA damage by ionizing radiation.
Ionizing radiation generates hydroxyl radicals, which in turn react with DNA producing a large number of chemical modifications, including DNA breaks. Our results suggest that the DNA of Hmgb1−/− MEFs is more accessible to hydroxyl radicals.
Cells Lacking HMGB1 Contain a Reduced Amount of Histones
DNA-bound proteins protect DNA from the attack of hydroxyl radicals; this property is exploited in protocols of hydroxyl radical footprinting. Nucleosomes shield DNA from hydroxyl radicals, and chromatin structure is a major factor determining DNA radiosensitivity 
. We then hypothesized that the DNA of Hmgb1
−/− cells is less protected by associated proteins, and in particular by histones. We thus measured histone content in Hmgb1
−/− and wild type cells.
We accumulated by Coulter counting an equal number of Hmgb1−/− and wild type MEFs, blocked in G0/G1 by serum starvation, and measured their DNA content by PicoGreen fluorescence and their histone content by quantitative immunoblotting (). While the amount of DNA was not statistically different between wild type and mutant cells, the amounts of core histones H2A, H2B, H3, and H4, linker histone H1, and the variant histone H2AX were all reduced by about 20% in Hmgb1−/− MEFs. On the contrary, beta-actin content (a common control for protein loading) was about 50% higher in Hmgb1−/− MEFs, whose cytoplasm appears larger than that of wild type MEFs even at an early passage (unpublished data). The abundance of other proteins, like peroxiredoxin-2, did not change in Hmgb1−/− MEFs.
Hmgb1−/− cells contain a reduced amount of histones.
We confirmed these results in HeLa cells stably transfected with a plasmid expressing HMGB1 shRNA (HeLa knockdown, henceforth KD) or a control plasmid. HMGB1 expression was almost abolished by the HMGB1 shRNA (Figure S1A
), and cycling KD cells (Figure S1B
) contained less core and linker histones (about 80% compared to the control HeLa cells) and more beta-actin (about 120%) (Figure S1C
, upper panel).
We then compared the entire proteomes of control and KD cells by stable isotope labeling with amino acids in cell culture (SILAC) 
. Control cells were grown for 8 passages in either light medium (Arg0 Lys0) or medium containing C13,N15-labelled arginine and lysine (Arg10 Lys8); KD cells were grown in light medium only. Light and heavy cells were mixed 1
1 before lysis, subjected to SDS-PAGE and in-gel trypsin digestion; peptides were quantitated by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) (Figure S2A
). HMGB1-derived tryptic peptides were absent in KD cells, as expected (Figure S2B
). In the control experiment (which compared heavy and light HMGB1-containing control cells) the ratios of light to heavy proteins had a narrow log-normal distribution (standard deviation close to 0.13). In contrast, when comparing light KD and heavy control cells these ratios showed a much wider distribution (standard deviation close to 0.37) (Figure S2C
). The abundance of most proteins changed slightly but significantly, albeit few proteins showed a change larger than 2-fold (MS tracings for a few representative peptides are shown, Figure S2B
). We then investigated the relative abundance of all histone-derived peptides that do not bear post-translational modifications; we excluded peptides known to bear modifications because a difference in their abundance could be due to variations in modification, rather than to a difference in the quantity of the histone protein. Notably, peptides from core and linker histones were reduced by about 25% in HMGB1-depleted cells (p
by two-sample Wilcoxon test) (, S2D
). Variant histones H2AX and H2AZ were also significantly reduced (Figure S3
). These experiments were repeated on wild type and Hmgb1
−/− MEFs, with comparable results (unpublished data).
Taken together, SILAC and quantitative immunoblotting indicated that cells lacking HMGB1 contain a coordinately reduced amount of all histones.
The Histone Content of Mammalian Cells Can Be Reduced Transiently and Reversibly
A lower histone content might be due to compensatory mutations selected in response to the lack of HMGB1 in Hmgb1−/− cells. In this case, rare mutant cells might be selected during in vitro culture and expand into viable clones. Alternatively, all cells might be able to modulate histone content in response to their physiological state, including the available level of HMGB1 protein. To distinguish between the possibilities, we transfected HeLa cells with 21-mer double-stranded HMGB1 siRNA, verified the disappearance of HMGB1, and grew the cells for 10 d until the amount of HMGB1 returned to normal (). Notably, the amount of histone H3 decreased concomitantly with the decrease in HMGB1, down to less than 80% of the starting level, and then recovered concomitantly with the recovery in HMGB1 content. HeLa cells transfected with control firefly luciferase siRNA showed no change in either HMGB1 or histone H3 content. Since there was no gross cell death after siRNA transfection, cells with a reduced histone content are not rare clones selected from a large cell population; rather, most cells in the population down-regulate histone content in response to a lack of HMGB1, and this regulation is reversible.
To further establish the physiological interdependence between HMGB1 and histone content we examined MEFs derived from Hmgb1
+/− heterozygous embryos; these have one half the amount of HMGB1 protein and contain about 90% of the normal amount of histones, which is intermediate between the amount in wild type and in Hmgb1
−/− MEFs (unpublished data). Finally, we verified that Hmgb1
−/− embryo livers contain a 20% reduction in histone content (Figure S4
), further excluding that the observed histone reduction in cultured cells can be due to culture conditions.
Cells Lacking HMGB1 Contain Fewer Nucleosomes
Histones are predominantly associated with DNA to form nucleosomes. Thus, a severe reduction in histone content should translate in a corresponding reduction in nucleosomally organized DNA. To verify the hypothesis that the DNA of Hmgb1−/− cells might be wrapped into fewer nucleosomes, cells were partially lysed and chromatin was digested with increasing amounts of micrococcal nuclease (MNase). At higher MNase concentrations, the amount of remaining (nucleosome-protected) DNA was reduced by about 30% in Hmgb1−/− MEFs (, quantification with PicoGreen). Agarose electrophoresis indicated that the total amount of MNase-resistant DNA is reduced in Hmgb1−/− MEFs, at all concentrations of MNase (). However, at low MNase concentration (0.5 U/ml, ), Hmgb1−/− samples contained more of higher molecular weight DNA (). This result was repeated several times and most likely indicates that a minor fraction of the chromatin of Hmgb1−/− cells becomes more resistant to digestion, in contrast to the major fraction which becomes more accessible.
Cells lacking HMGB1 contain fewer nucleosomes and more RNA transcripts.
The average spacing between nucleosomes was very similar in Hmgb1
−/− and wild type MEFs (), contrary to what is expected if available nucleosomes were uniformly redistributed over the genome. Similar results were obtained with KD cells (Figure S1C
The conclusion from these experiments is that mammalian cells can survive and proliferate with substantially fewer nucleosomes.
Cells with Fewer Nucleosomes Contain More RNA Transcripts
The availability of cells with fewer nucleosomes allowed us to test the widely held opinion that nucleosomes limit transcription in vivo, as they do in vitro by impeding the progress of RNA polymerases 
. We quantified total nucleic acids in KD and control HeLa cells by FACS after acridine orange staining () 
. Whereas the DNA content was similar in KD and control cells, the RNA content is about 1.3 times higher in KD cells. Both polyA+
mRNA and the 47S rRNA precursor are similarly increased ().
Although global transcript abundance increases in cells lacking HMGB1, so that most transcripts will be more abundant, the expression of individual genes can also change relative to each other. We thus measured the relative representation of each transcript within an identical amount of RNA extracted from cells. Relative representation in a fixed amount of RNA automatically normalizes away the global increase of about 30% in total RNA amount in HeLa KD cells. The comparison in relative amount is instructive to identify which genes deviate the most from the average effect. Our analysis indicates that about 13% of transcripts (1,080 over 8,027 on the Illumina platform; p
<0.01; Figure S5A
) are over-represented (577 genes) or under-represented (503 genes) from the average 30% increase in KD HeLa cells. The Gene Ontology categories significantly affected at the mRNA level (p
<0.05, Wilcoxon test) are indicated in Figure S6A
. These broadly correspond to the Gene Ontology categories significantly affected at the protein level (Figure S6B
HMGB1 Promotes Histone Deposition In Vitro
Since the absence of HMGB1 leads to a decrease in nucleosome number, we investigated whether HMGB1 was directly involved in chromatin assembly, as suggested by early experiments 
We tested the effect of HMGB1 on histone deposition onto naked DNA using a simple, commercially available assay (Chromatin Assembly Kit by Active Motif). Linearized plasmid DNA was mixed with soluble histones, the histone chaperone NAP, and the remodeling factor ACF. After incubation for 15 min at 27°C, the assembled chromatin was digested with micrococcal nuclease, and an aliquot was run on an agarose gel (, upper panel). No DNA remained after nuclease digestion if histones were absent from the assembly reaction (lane 3–4), whereas a clear band of mononucleosomal size was present in the presence of histones (lane 5). We then added to the reaction mix increasing concentrations of HMGB1, and we noted a highly significant dose-dependent increase in the mononucleosome band (lanes 6–9), reaching a maximal yield at 1 µg/ml. At higher HMGB1 concentrations, the efficiency of nucleosome deposition decreased (lanes 10–12). At the optimal HMGB1 concentration, nucleosome formation was 3.5 times faster in the presence than in the absence of HMGB1 (, upper panel). Direct quantification of nuclease-resistant DNA by PicoGreen confirmed the data obtained by gel electrophoresis (, lower panels).
HMGB1 promotes the assembly of chromatin in vitro.
nhp6 Yeast Mutants Recapitulate the Phenotype of Mammalian Cells Lacking HMGB1
Yeast Nhp6 proteins are functionally equivalent to HMGB1 in mammalian cells, and nhp6
yeast mutants are more sensitive to UV irradiation 
. We therefore verified whether yeast nhp6
cells also have reduced histone and nucleosome content. We synchronized yeast cells in G1 by treatment with alpha factor pheromone, collected an equal number of wild type and nhp6
cells, and measured their DNA content with PicoGreen and their histone content by quantitative immunoblotting. nhp6
cells contained about 65% of the amount of histones compared to the wild type, and their chromatin was more accessible to digestion by MNase (). Moreover, 2D gel analysis indicated that the supercoiling of the 7.0 kb yRp17 plasmid was reduced by about three turns in nhp6
cells, equivalent to about three nucleosomes fewer than in the wild type (). Finally, nhp6
cells contain about 1.2 times more RNA than wild type cells (). We conclude that the phenotypes of nhp6
mutants are largely similar.
nhp6 cells contain fewer nucleosomes and more RNA transcripts.
Yeast Cells with a Reduced Nucleosome Number Have a Distinct Transcriptional Profile
A transient model of nucleosome depletion in yeast was examined previously 
. In the UKY403 yeast strain, the sole copy of histone H4 is under the control of the GAL1
promoter. In glucose medium, UKY403 cells lost around 50% of nucleosomes by 6 h, relative to a control strain with a wild type H4 gene, and the expression of 15% of genes increased and the expression of 10% of genes decreased more than 3-fold.
We then looked at the relative expression of genes in wild type and nhp6
cells and compared them to those in the UKY403 strain. By Affymetrix analysis we found that out of 5,447 genes, 219 are up and 251 are down in nhp6
relative to wild type cells (1.5-fold threshold and p
<0.05) (Figure S5B
). The Gene Ontology categories that are significantly affected are shown in Figure S7
. The correlation between gene expression profiles in UKY403 and nhp6
cells () rises from r2
<0, when nucleosomes are not depleted in UKY403, to almost 0.16 after 2 h (p
), and remains almost constant thereafter. Since about half of the genes transcriptionally affected by nucleosome depletion in the UKY403 strain are also affected by slow growth, we asked whether our results were influenced by the slow growth of the nhp6
mutant relative to its wild type counterpart 
. Indeed, the correlation between the two strains is much stronger for the growth-related gene subset; however, the correlation for the genes unresponsive to changes in growth rate is only slightly smaller (r2
) than the one for all genes. Taken together, these data suggest that nucleosomal depletion affects transcription profiles in broadly similar ways in strains where histone H4 is depleted or Nhp6a/b proteins are lacking.
In Yeast Cells a Reduced Nucleosome Number Affects Primarily Nucleosome Occupancy and Not Nucleosome Position
Research in the last few years has highlighted the importance of nucleosome positioning in the control of transcription; we therefore asked how nucleosome depletion affects genomewide nucleosome positioning.
The hypothesis of statistical positioning states that nucleosomes space themselves between barriers 
. In this case, the position of nucleosomes should vary when their number is lower (, hypothesis 1). According to the alternative hypothesis that DNA sequence is the major determinant of nucleosome positioning, nucleosome limitation could lead to the selective loss of a minority of nucleosomes (hypothesis 2). Alternatively, nucleosomes might occupy the same positions, but spend less time on each of them (hypothesis 3; nucleosome “occupancy” of individual sequences is reduced).
nhp6 cells have increased variability in nucleosome occupancy.
When we applied high throughput sequencing to MNase-resistant DNA from nhp6
and wild type cells, we found that the distribution of sequence reads was very similar. Representative snapshots of the nucleosome maps are shown in ; a complete browsable form is available at the website indicated in Materials and Methods
The number of times a specific base pair appears in sequence reads, divided by the total number of sequence reads, is the relative occupancy of that base pair. Relative occupancies of all base pairs can then be compared between strains; a density dot plot allows a visual representation of such a comparison. The comparison between biological replicates gives a density plot where most bases cluster around the diagonal (, right). The comparison between nhp6
and wild type cells (, left) gives a density plot which is more dispersed about the diagonal and has a characteristic skew with more points below the diagonal at low occupancy and more points above the diagonal at high occupancy. This result is inconsistent with a global redistribution of nucleosomes over the genome (hypothesis 1 in ), which would give a smeared density plot with a lot of points close to the axes (base pairs occupied in one strain but not in the other). Our result is also inconsistent with the disappearance of nucleosomes from a minority of sites (hypothesis 2), which would give a density plot with two separate sub-populations, as simulated in Figure S8A
In fact, the similarity of the nhp6/wt density plot to that of biological replicates indicates that most base pairs that are occupied by nucleosomes in wild type cells are also occupied in nhp6 cells. A complete identity between nucleosome positions in the two strains (although with reduced occupancy in one strain) would give the same density plot of biological replicates. Thus all three hypotheses depicted schematically in do not correspond to observation, but hypothesis 3 comes closest.
We next moved from coverage at individual base pairs to examination of nucleosome positions. We used template filtering 
to call nucleosome positions and confirmed that they are highly conserved (, “nuc calls”). Almost half of the nucleosomes are centered around the same position in both strains, many are offset by about 10 base pairs and some by 20 pairs (Figure S8B
); we note that 10-bp shifts correspond to those expected from the rotational periodicity of DNA wrapped around the nucleosome. Only about 30% of nucleosomes had shifted by more than 20 base pairs. This confirms that most nucleosomes occupy approximately the same location in the two strains. However, in nhp6
cells fewer nucleosomes were unambiguously called (45,441 versus 53,643), and the read peaks that identify nucleosome edges were broadened in the nhp6
sequencing data (). This suggests that some nucleosomes may shift from a single favored position into a superposition of multiple overlapping positions (“fuzzy nucleosomes”; 
); beyond a certain degree of fuzziness, nucleosomes would not be called by the algorithm. The length of DNA predicted to be covered by nucleosomes was reduced on average and had increased variability in nhp6
cells (Figure S8C
), consistent with increased fuzziness.
As observed at the single base pair level (), many nucleosomal sites are either less or more relatively occupied in nhp6
cells. This is clearly visible in the snapshots in , showing three different loci with decreased, unchanged, and increased relative occupancy, respectively. Absolute occupancy is proportional to the nucleosome number, and thus is reduced by about 30% in nhp6
cells. As a result, on some sites absolute occupancy in nhp6
cells may be comparable to that in the wild type (but not higher), whereas on the vast majority of sites it will be reduced or very reduced. High-resolution primer extension analysis confirmed a similar position of nucleosomes in the ars1
locus, but with higher accessibility of nucleosome-covered sequences (and thus lower absolute occupancy) in nhp6
cells (Figure S9A
). Nucleosome ChIP (using an antibody against histone H3) also was in agreement with reduced nucleosome occupancy of the ars1
locus in nhp6
cells (Figure S9B
Overall, these results are in accordance with increased chromatin accessibility in the nhp6 mutant and suggest that nucleosomes have increased mobility on the sites they occupy (either intrinsic or catalyzed by nucleosome remodelling complexes).
Nucleosome Position and Occupancy Over Gene Control Regions in nhp6 Cells
Nucleosomal organization of the control regions of genes is considered most important for gene expression. In yeast, nucleosomes are regularly arranged on protein-coding genes, starting from the transcriptional start site (TSS). A nucleosome-depleted region (NDR, also called nucleosome-free region, NFR) of about 140 bp is generally found just upstream of the TSS and is surrounded by two well-positioned nucleosomes, called −1 and +1 nucleosome, respectively. We aligned genes by their TSS and ranked them by the severity of nucleosome loss in nhp6 cells relative to wild type (, heatmap in the center). All genes had reduced occupancy of the −1 nucleosomes (green streak in the heatmap), and genes with more severe nucleosome loss at the 5′ end also had reduced nucleosome occupancy over the gene body. Once again, we observed that the genes with more severe nucleosome loss in nhp6 cells (, center) were the ones already low in nucleosome occupancy in the wt (, right, red line). A few genes appeared to have relatively increased occupancy in nhp6 cells (red genes in the bottom of ); these genes belong primarily to the Gene Ontology categories “metabolism” and “cell wall”. None of these genes, however, appeared to have increased absolute nucleosome occupancy (i.e., after considering that nucleosome number is reduced by about 30% in nhp6 cells).
Nucleosomal occupancy on yeast coding genes.
Nucleosome by nucleosome, median relative occupancy over the promoter and the TSS of all genes (from the −1 to the +1 nucleosome) was lower in nhp6
cells, and median relative occupancy for the +2, +3, and +4 nucleosomes was slightly higher (thick lines in ). Relative occupancy of all nucleosomal sites is more variable in nhp6
cells (the thin lines in indicate the lower and upper quartiles of occupancy). Lower nucleosome occupancy in the control regions correlates with increased gene expression (, left, blue line, and Figure S8D
A Simple Model Predicts Nucleosome Occupancy in nhp6 Cells Starting from Nucleosome Occupancy in Wild Type Cells
In all our analyses, from the correlation of base pair occupancy to the distribution of nucleosomes over genes, a theme stands out: the reduction in nucleosome number is associated with an increase in the variability of relative occupancy. From the point of absolute occupancy, we have already pointed out that some sites might be similarly occupied in wild type and nhp6 strains, while sites that are intermediately occupied in the wild type are less occupied in nhp6 cells, and weakly occupied sites in the wild type are much less occupied in nhp6 cells. The skew in the density plot of visually represents this pattern of more pronounced loss of occupancy in nhp6 cells from sites that are already less occupied in the wild type.
Unequal occupancy of nucleosomal sites in vivo is expected, since (1) in vitro the probability of nucleosome occupancy on different sites can vary by a factor of up to 5,000 
and (2) histone octamers are insufficient to package all the genome 
. Thus, in physiological conditions some sites will be occupied close to 100% of the time (“saturated”) and some much less. Based on these considerations, we designed a model to account for the characteristic pattern of nucleosomal occupancy in nhp6
cells. We assume that all sites compete for a finite pool of histones that is insufficient for all of them, and that each site has a certain probability of being occupied, that depends from histone availability. We also posit that the probability of occupation versus availability of histones is a hyperbolic function and is different for each site (). This model recalls formally the formation of a complex between two macromolecules, and we can thus assign a dissociation constant ki
to each nucleosome. The occupancy O of site i is then Oi
), where ki
is the dissociation constant and x
is the concentration of available histones. A decrease in the availability of histones will result in a skewed desaturation, with heavy nucleosome loss at sites of high dissociation constant and mild loss at sites of low dissociation constant (). This will increase the variability of relative occupancy.
Mathematical model describing nucleosomal occupancy.
Based on the relative occupancy of wild type sites, the model should be able to predict the genomewide occupancy for a certain decrease in available histones (details in Figure S8E
). We then used our model to simulate the relative occupancies in a population of cells which have a 30% reduction in histone content (). The density dot plot comparing simulated
occupancy in nhp6
cells (, right) is almost symmetrical about the diagonal and corrects the observed systematic skew in the density dot plot of nph6
/wt relative occupancies (, left), although the dispersion of values is not decreased substantially. We also plotted the distribution of the number of nucleosomes at each occupancy value (); our model correctly predicts the approximate shape of the distribution for nhp6
cells and the position of the mode. The fitting between the observed and predicted nhp6
occupancies is optimal at the nucleosome content actually observed for nhp6
cells. An alternative model based on statistical positioning does not justify our observations, since it predicts that both position and spacing of nucleosomes would be changed when histone content is reduced (Figure S8F
, red line), contrary to what we observed.
Overall, our model justifies the disproportionate loss of nucleosomes from weakly occupied sites, and the increase in relative occupancy at the more occupied sites (, far tail of the distribution). We then asked whether the sites with reduced occupancy in nhp6
cells are the ones with lower intrinsic ability to form nucleosomes. To this aim, we compared our dataset with the dataset obtained by reconstituting yeast chromatin in vitro 
. The comparison of changes in nucleosomal occupancy between our nhp6
/wt dataset and Kaplan's in vitro/in vivo dataset is shown in . The Pearson correlation coefficient between datasets is r2
), indicating that the sites that most lose occupancy in nhp6
cells correspond to the ones with lower occupancy in reconstituted chromatin; conversely, the sites that lose less occupancy in nhp6
cells correspond to the ones that most easily reform chromatin in vitro ().