1. Cell Biology
Download icon

The quantitative architecture of centromeric chromatin

  1. Dani L Bodor
  2. João F Mata
  3. Mikhail Sergeev
  4. Ana Filipa David
  5. Kevan J Salimian
  6. Tanya Panchenko
  7. Don W Cleveland
  8. Ben E Black
  9. Jagesh V Shah
  10. Lars ET Jansen Is a corresponding author
  1. Instituto Gulbenkian de Ciência, Portugal
  2. Harvard Medical School, United States
  3. Brigham and Women's Hospital, United States
  4. University of Pennsylvania, United States
  5. University of California, San Diego, United States
Research Article
Cited
51
Views
2,972
Comments
0
Cite as: eLife 2014;3:e02137 doi: 10.7554/eLife.02137

Abstract

The centromere, responsible for chromosome segregation during mitosis, is epigenetically defined by CENP-A containing chromatin. The amount of centromeric CENP-A has direct implications for both the architecture and epigenetic inheritance of centromeres. Using complementary strategies, we determined that typical human centromeres contain ∼400 molecules of CENP-A, which is controlled by a mass-action mechanism. This number, despite representing only ∼4% of all centromeric nucleosomes, forms a ∼50-fold enrichment to the overall genome. In addition, although pre-assembled CENP-A is randomly segregated during cell division, this amount of CENP-A is sufficient to prevent stochastic loss of centromere function and identity. Finally, we produced a statistical map of CENP-A occupancy at a human neocentromere and identified nucleosome positions that feature CENP-A in a majority of cells. In summary, we present a quantitative view of the centromere that provides a mechanistic framework for both robust epigenetic inheritance of centromeres and the paucity of neocentromere formation.

https://doi.org/10.7554/eLife.02137.001

eLife digest

The genetic information in a cell is packed into structures called chromosomes. These contain strands of DNA wrapped around proteins called histones, which helps the long DNA chains to fit inside the relatively small nucleus of the cell.

When a cell divides, it is important that both of the new cells contain all of the genetic information found in the parent cell. Therefore, the chromosomes duplicate during cell division, with the two copies held together at a single region of the chromosome called the centromere. The centromere then recruits and coordinates the molecular machinery that separates the two copies into different cells.

Centromeres are inherited in an epigenetic manner. This means that there is no specific DNA sequence that defines the location of this structure on the chromosomes. Rather, a special type of histone, called CENP-A, is involved in defining its location. Bodor et al. use multiple techniques to show that human centromeres normally contain around 400 molecules of CENP-A, and that this number is crucial for ensuring that centromeres form in the right place. Interestingly, only a minority of the CENP-A molecules are located at centromeres; yet this is more than at any other region of the chromosome. This explains why centromeres are only formed at a single position on each chromosome.

When the chromosomes separate, the CENP-A molecules at the centromere are randomly divided between the two copies. In this way memory of the centromere location is maintained. If the number of copies of CENP-A inherited by one of the chromosomes drops below a threshold value, a centromere will not form. However, Bodor et al. found that the number of CENP-A molecules in a centromere is large enough, not only to support the formation of the centromere structure, but also to keep it above the threshold value in nearly all cases. This threshold is also high enough to make it unlikely that a centromere will form in the wrong place because of a random fluctuation in the number of CENP-A molecules. Therefore, the number of CENP-A molecules is crucial for controlling both the formation and the inheritance of the centromere.

https://doi.org/10.7554/eLife.02137.002

Introduction

Centromeres are essential for proper cell division. During mitosis, a transient structure called the kinetochore is assembled onto centromeric chromatin, which mediates the interaction between DNA and the mitotic spindle (Allshire and Karpen, 2008; Cheeseman and Desai, 2008). Intriguingly, although centromeres are directly embedded in chromatin, specific DNA sequences are neither necessary nor sufficient for centromere function. This is best exemplified by the rare occurrence, within the human population, of neocentromeres: functional centromeres that have repositioned to atypical loci on the chromosome (Amor et al., 2004; Marshall et al., 2008; du Sart et al., 1997; Voullaire et al., 1993). Rather than centromeric sequences, the primary candidate for epigenetic specification of centromeres is the histone variant CENP-A, which replaces canonical H3 in centromeric nucleosomes (Palmer et al., 1987, 1991; Stoler et al., 1995; Henikoff et al., 2000; Yoda et al., 2000). CENP-A chromatin is sufficient for recruitment of the downstream centromere and kinetochore complexes (Foltz et al., 2006; Okada et al., 2006; Carroll et al., 2009, 2010; Barnhart et al., 2011; Guse et al., 2011; Mendiburo et al., 2011). In addition, CENP-A is stably transmitted at centromeres during mitotic (Jansen et al., 2007; Bodor et al., 2013) and meiotic (Raychaudhuri et al., 2012) divisions, and its assembly is tightly cell cycle controlled (Jansen et al., 2007; Schuh et al., 2007; Silva et al., 2012). Importantly, targeting of this protein to an ectopic site of the genome is sufficient to initiate an epigenetic feedback loop, recruiting more CENP-A to this site (Mendiburo et al., 2011). However, little is known about the quantity of CENP-A present at centromeres, despite this being an essential parameter for a functional understanding of both centromeric architecture and epigenetic inheritance. Here, we use multiple, independent approaches to determine the absolute copy number of CENP-A at centromeres. In addition, we provide novel insights in the mechanisms of centromere size control.

Results

Modification of endogenous CENP-A alleles in diploid human cells

To determine absolute centromeric CENP-A levels in human cells, we set out to build cell lines in which the entire CENP-A pool is fluorescent. To accomplish this, we removed a significant and essential portion of the CENP-A gene to create a knock-out allele in stably diploid, human retinal pigment epithelium (RPE) cells (Figure 1A, bottom). Subsequently, a fluorescent knock-in allele was created by placing GFP or YFP encoding sequences in frame with the sole remaining CENP-A gene (Figure 1A, middle). Specifically, we have built the following endogenously targeted RPE cell lines: CA+/−, CAG/−, CAY/−, and CA+/F (where + = wild-type; − = knock-out; G = GFP knock-in; Y = YFP knock-in; F = floxed [to control for potential gene-targeting artifacts]; Figure 1—figure supplement 1A). Western blot analysis confirms that CAG/− and CAY/− cells exclusively contain tagged CENP-A (of ∼43 kDa), while CA+/+ (wild-type), CA+/F, and CA+/− cells only express wild-type CENP-A (∼16 kDa) protein (Figure 1B). Importantly, heterozygous expression or tagging of endogenous loci did not interfere with cell viability.

Figure 1 with 1 supplement see all
CENP-A levels are regulated by mass-action.

(A) Schematic of gene-targeting strategy that allowed for the creation of CENP-A knockout and fluorescent knock-in alleles. The region encoding the essential CENP-A targeting domain (CATD, Black et al., 2007) is indicated. (B) Quantitative immunoblots of CENP-A, HJURP, and Mis18BP1 in differentially targeted RPE cell lines. α-tubulin is used as a loading control. (C) Immunofluorescence images of same cell lines as in B. CENP-A intensity is represented in a heat map as indicated on the right. The fold difference ± SEM (n is biological replicates) compared to wild-type RPE cells is indicated below. Scale bar: 10 μm. Note that in contrast to quantification of immunoblots, immunofluoresce detection of untagged and tagged CENP-A is directly comparable. (D) Quantification of centromeric CENP-A levels (from C) by immunofluorescence (IF) and total CENP-A levels (n = 4–9 independent experiments as in B) by western blot (WB). All cell lines expressing untagged CENP-A are normalized to CA+/+ while those expressing tagged CENP-A are normalized to the centromeric CAY/− levels measured in C, as indicated by dashed lines. (E) Correlation of centromeric and total cellular CENP-A levels as measured in D. Dashed line represents a predicted directly proportional relationship with indicated correlation coefficients. Throughout, the average ± SEM is indicated. (F) Quantification of centromeric CENP-A levels in synchronized HeLa cells (based on anti-CENP-A staining) within a single cell cycle after transient transfection of indicated proteins. Asterisk indicates statistically significant increase compared to control or indicated transfections (one-tailed t test; p<0.05; n = 3); NS indicates no significant increase. Average ± SEM of three independent experiments is shown.

https://doi.org/10.7554/eLife.02137.003

Centromeric CENP-A levels are regulated by mass-action

While CENP-A is an essential and constitutive component of centromeres, how the size of the centromeric chromatin domain is controlled is not known. We analyzed the consequences of different CENP-A expression levels in our CENP-A heterozygous knock-out and knock-in lines, as well as in a cell line that ectopically overexpressed CENP-A-YFP (CAY/−+OE; Figure 1B; Figure 1—figure supplement 1A). First, we measured the total protein pool of CENP-A in our cell lines by quantitative immunoblotting. While we found the detection output for CENP-A to be linear over at least a 32-fold range (Figure 2E), due to differences in protein transfer efficiencies this method does not allow for a comparison between proteins of different sizes, for example (GFP- or YFP-) tagged and untagged (wild-type) CENP-A (Figure 2—figure supplement 3). Nevertheless, we could directly compare CAG/−, CAY/−, and CAY/−+OE cell lines (Figure 2—figure supplement 3) and found that cellular CENP-A content spans a sixfold range (Figure 1B,D).

Figure 2 with 3 supplements see all
Human centromeres contain 400 molecules of CENP-A.

(A) Schematic outline of strategy allowing for the quantification of the centromeric fraction of CENP-A compared to the total cellular pool. Scale bars: 5 μm. (B) Quantification of the centromeric fraction of CENP-A in CAY/− cells. Each circle represents one centromere; circles on the same column are individual centromeres from the same cell. Dashed line indicates average of all centromeres. (C) Quantification of the centromeric fraction of CENP-A in indicated cell lines. Each square represents the average centromeric signal from one cell; squares on the same column are individual cells from the same experiment (Exp). Figure 2—figure supplement 2 shows quantification of individual centromeres in CAG/− and CAY/−+OE cells. (D) Representative quantitative immunoblot of purified recombinant CENP-A and endogenous CENP-A from whole cell extracts (WCE). (E) Quantification of D. Solid line represents the best fit linear regression. Dashed line represents the amount of CENP-A from 150,000 cells. (F) Quantification of the total cellular CENP-A copy number. Each diamond represents one replicate experiment; measurement from E is indicated as a gray diamond. (G) Calculation of average CENP-A copy number per centromere (CEN) in wild-type RPE cells. Throughout, the average ± SEM is indicated.

https://doi.org/10.7554/eLife.02137.005

Given its essential role in centromere function, we predicted a tight control of centromeric CENP-A levels. However, instead of maintaining a fixed amount of CENP-A at centromeres, the levels varied extensively (Figure 1C). Both CA+/− and CAG/− cells, which contain a single intact allele, have decreased centromeric CENP-A levels, while the parental CA+/F cells maintain wild-type levels. Surprisingly, despite expressing CENP-A from a single allele, CAY/− cells have increased CENP-A levels, which may be due to adaptations that arose during the creation of this cell line. As expected, CENP-A levels are further elevated in CAY/−+OE cells (Figure 1C). Remarkably, we found a very high correlation (r2 = 84%) for a hypothetical directly proportional relationship between centromeric and total cellular CENP-A-GFP or CENP-A-YFP levels (Figure 1D,E). Similarly, despite an only approximately twofold range of expression, we still observe a high correlation with direct proportionality (r2 = 71%) for cells expressing untagged CENP-A (Figure 1D,E). Thus, our observations indicate that centromeric levels are determined by a mass-action mechanism, where the amount of centromeric CENP-A varies in direct proportion with the cellular content.

An alternative hypothesis is that stable cell lines have undergone long-term adaptation to altered CENP-A expression, which has led to re-equilibrated centromeric levels. For example, proteins involved in depositing CENP-A at the centromere may have adapted to CENP-A expression levels. Indeed, we see a weak correlation between the levels of CENP-A and its histone chaperone HJURP (Dunleavy et al., 2009; Foltz et al., 2009; Barnhart et al., 2011) in our cell lines (Figure 1B, Figure 1—figure supplement 1B). Conversely, no correlation was detected for Mis18BP1 (Figure 1B, Figure 1—figure supplement 1C), another essential protein for CENP-A assembly (Fujita et al., 2007; Maddox et al., 2007), arguing that it is a non-stoichiometric component of the loading pathway. To test whether centromeric CENP-A levels require long-term adaptation, we analyzed the effect of CENP-A and/or HJURP overexpression in a single round of CENP-A assembly. Therefore, we transiently expressed CENP-A and/or HJURP and measured the level of centromeric CENP-A after a single cell cycle in HeLa cells, which can be effectively synchronized in S phase using thymidine. While induction of CENP-A expression leads to a prompt increase in centromeric levels, no (additional) effect was observed by expression of HJURP (Figure 1F). Together, our results strongly suggest that centromeric CENP-A levels are directly regulated by its protein expression levels.

Centromeres contain ∼400 molecules of CENP-A

To understand how CENP-A chromatin is self-propagated and nucleating the kinetochore, it is critical to establish the absolute amount of CENP-A present. In vertebrates, previous estimates range from a few tens of molecules (in chicken DT40 cells, Ribeiro et al., 2010) to a potential maximum of tens of thousands (in HeLa cells, Black et al., 2007). To directly determine the copy number of CENP-A on human centromeres, we developed a 3D imaging strategy (Figure 2A), which we adapted from a method used previously to quantify cytokinesis proteins in fission yeast (Wu and Pollard, 2005; Wu et al., 2008). In brief, we use a non-cell permeable dye (Figure 2A, I) to determine the 3D shape of cells (Figure 2A, II) and measure the total amount of fluorescence within the entire cell volume (Figure 2A, III). Total cellular fluorescence of CAY/− cells (Figure 2A, III) was corrected for autofluorescence measured in wild-type RPE cells (Figure 2A, IV), thus resulting in a measure of total CENP-A-derived fluorescence. Next, centromere-specific fluorescence was measured after correction for local background (Figure 2A, V; Hoffman et al., 2001) and axial oversampling. Importantly, fluorescence lifetime of CENP-A-YFP is similar between highly concentrated centromeric and diffuse cytoplasmic pools (Figure 2—figure supplement 1), arguing that clustering does not lead to changes in fluorescence efficiency. In effect, our 3D-integrated fluorescence strategy measures the centromeric fraction of CENP-A compared to the total cellular pool. We find that while CENP-A is enriched at centromeres, on average only 0.44% of cellular CENP-A is present per centromere in CAY/− cells (Figure 2B). Very similar fractions were observed in CAG/− and CAY/−+OE cells (0.38% in both cases; Figure 2C, Figure 2—figure supplement 2A,B), which provides an additional line of evidence in support of a mass-action mechanism for CENP-A assembly. Furthermore, these findings show that a surprising minority, about one-fifth of the CENP-A protein content (0.44% × 46) is present on the functionally relevant subcellular location, i.e. at the centromeres.

To convert centromeric fractions to absolute amounts, we determined the total number of CENP-A molecules in RPE cells. To this end, we prepared whole cell extracts of RPE cells and analyzed these alongside highly purified recombinant CENP-A/H4-complexes of known concentration by quantitative immunoblotting (Figure 2D). Importantly, we ensured that recombinant and cellular CENP-A have the same transfer efficiency and can be directly compared to each other (Figure 2—figure supplement 3). Fitting the cellular amount of CENP-A onto a linear regression curve of purified protein (Figure 2E) shows that CA+/+ cells contain an average of ∼9.1 ± 1.1 × 104 (n = 10) molecules of CENP-A per cell (Figure 2F). Because the centromeric fraction of CENP-A is fixed, we can calculate the absolute amount of CENP-A per centromere in our cell lines (Figure 2G, Figure 2—figure supplement 2C) and show that wild-type RPE cells contain ∼400 molecules of CENP-A on an average centromere.

Both the expression and centromeric loading of CENP-A are cell cycle regulated (Figure 3A). In human cells, cellular protein levels of CENP-A peak in late G2 (Shelby et al., 2000), while incorporation into centromeric chromatin occurs in early G1 phase (Jansen et al., 2007). Thus, it is possible that part of the cell-to-cell variation of the centromeric CENP-A ratio observed in Figure 2C is due to differing cell cycle stages. We tested this possibility using a previously developed fluorescent ubiquitin-based cell cycle indicator (FUCCI) that can be used in live cells (Sakaue-Sawano et al., 2008). In particular, we used hCdt1(30/120)-RFP, which is expressed ubiquitously throughout the cell cycle, but is specifically degraded in S, G2, and M phases (Sakaue-Sawano et al., 2008). As a result, protein levels increase as cells enter and progress through G1 phase, peak at the G1/S boundary, and then drop until cells re-enter G1 (Figure 3A). We expressed this protein in CAY/− cells and tracked the RFP fluorescence intensity over time (Figure 3B, Figure 3—figure supplement 1A) to identify cells that entered S phase (see ‘Materials and methods’ for details). We compared the centromeric fraction of CENP-A of S Phase cells to that of randomly cycling cells and found that neither the mean nor the variance differs significantly between these two populations (Figure 3C). Importantly, expression of the FUCCI marker itself has no effect on the measurements performed (Figure 3—figure supplement 1B). While the centromeric fraction of CENP-A is likely low in G2 phase and high just after assembly in early G1, we find that the variation observed in Figure 2C is not a consequence of such cell cycle-induced effects and may instead reflect inherent variation between cells.

Figure 3 with 1 supplement see all
Centromeric CENP-A levels are equivalent between S phase and randomly cycling cells.

(A) Cartoon depicting changes in cell morphology and nuclear levels of hCdt1(30/120)-RFP (in red) throughout the cell cycle (Sakaue-Sawano et al., 2008). Approximate timing of CENP-A expression (Shelby et al., 2000) and centromeric loading (Jansen et al., 2007) are indicated in orange and blue, respectively. The stage at which cells were analyzed to measure the centromeric fraction of CENP-A is indicated in green. (B) An example trace of a cell entering S phase (indicated by a sudden decrease in RFP levels) is shown. The centromeric fraction of CENP-A was measured at this point as outlined in Figure 2A. Peak expression is normalized to 100 and background fluorescence to 0. Micrographs of hCdt-1(30/120)-RFP at indicated timepoints are shown below. (C) As in Figure 2C. Orange squares represent cells that have passed the G1-S transition point, as indicated by decreasing levels of hCdt-1(30/120)-RFP. Gray squares represent randomly cycling cells. No statistically significant differences (NS) were observed between randomly cycling cells and S phase cells.

https://doi.org/10.7554/eLife.02137.009

Although the method we employed to measure centromeric ratios is internally controlled, it relies on measurement of integrated fluorescence of whole cells, including highly dilute cytoplasmic CENP-A. To exclude potential errors in measurements of low protein concentration, we stably expressed H2B-RFP in CAY/− cells (Figure 4A, inset) and determined that 0.73% of nuclear CENP-A is present on each centromere (Figure 4A). In addition, low salt fractionation experiments indicate that ∼74% of cellular CENP-A co-pellets with other chromatin components in CAY/−+H2B-RFP cells (Figure 4B), indicating that this represents the stable nuclear pool. Combined, we find a similar number of CENP-A molecules per centromere when analyzing the nuclear pool (492 molecules; Figure 4C) as when measuring total cellular CENP-A. This argues that the measurements performed above are not significantly influenced by a potential inaccuracy in determining the cytoplasmic pool. Interestingly, it has recently been shown that detectable levels of CENP-A are assembled into non-centromeric chromatin of HeLa cells (Lacoste et al., 2014). We now find that, at least in RPE cells the proporation of chromatin bound CENP-A outside of the centromere is surprisingly high (∼66% in this cell line).

Measurement of nuclear CENP-A confirms centromeric copy number.

(A) As in Figure 2B, except that the centromeric fraction compared to total nuclear pool is indicated. Inset shows a representative image of a CAY/−+H2B-RFP cell (scale bar: 2.5 μm). (B) Quantitative immunoblot showing the soluble fraction and a dilution series from the insoluble fraction of CENP-A-YFP in CAY/−+H2B-RFP cells (left). Tubulin is used as a marker for the soluble fraction and H4K20me2 (exclusively found in chromatin, Karachentsev et al., 2007) for the insoluble fraction. Quantification of insoluble fraction of CENP-A is shown to the right. (C) Calculation of the average CENP-A copy number per centromere (CEN) in wild-type RPE cells, based on results from CAY/−+H2B-RFP cells.

https://doi.org/10.7554/eLife.02137.011

Centromeric CENP-A copy number is confirmed by three independent methods

To further validate that the strategy described above accurately measures centromeric CENP-A copy numbers, we used two additional independent quantification methods. First, we applied a method that employs the statistical properties of fluorescence redistribution (Rosenfeld et al., 2005, 2006). This method relies on the fact that random segregation of fluorescent molecules leads to each daughter receiving an (unequal) fraction, where the distribution of differences relates to the total number of molecules (as outlined in Figure 5A). During mitosis, sister centromeres form individually resolved spots by light microscopy, allowing us to measure the fluorescence intensity of individual sisters (Figure 5B). We find that rather than accurately segregating exactly half of pre-assembled CENP-A onto each daughter chromatid, the difference between sister centromeres follows a random distribution (Figure 5B,C). Previously, Rosenfeld et al. have provided mathematical evidence that measurements of this deviation allow for the determination of the fluorescence intensity of a single heritable, segregating unit (Figure 5A, Rosenfeld et al., 2005, 2006). We measured an average of 75.4 segregating units of CENP-A-GFP per centromere in CAG/− cells (Figure 5D). Because each segregating unit consists of one or more nucleosomes, containing two molecules of CENP-A each (Sekulic et al., 2010; Tachiwana et al., 2011; Bassett et al., 2012; Hasson et al., 2013; Padeganeh et al., 2013), an average CAG/− centromere has a minimum of 150.8 molecules of CENP-A. Correcting the amount of CENP-A measured in CAG/− cells for wild-type levels (Figure 1C) results in ≥377 molecules of CENP-A per centromere (Figure 5D, right y-axis). Importantly, these measurements differ significantly if random centromere pairs are chosen for which no statistical correlation exists (Figure 5—figure supplement 1E). This confirms that fluorescence intensities at sister centromeres co-vary and renders this type of analysis suitable for centromere quantification. Stochastic fluctuation measurements in CAY/− and CAY/−+OE cells indicates that wild-type cells contain ≥188 and ≥149 CENP-A molecules per centromere, respectively (Figure 5—figure supplement 1A–D). Importantly, the number of co-segregating CENP-A nucleosomes is unknown, which can be one or more. Therefore, despite the variation between the cell lines used here, all results obtained from this method provide a minimum estimate of the centromeric CENP-A copy number that is in agreement with the 400 centromeric molecules of CENP-A measured above (Figure 2G).

Figure 5 with 1 supplement see all
Independent quantification methods confirm centromeric CENP-A copy number.

(A) Stochastic fluctuation method: cartoon depicting inheritance and random redistribution of parental CENP-A nucleosomes onto sister chromatids during DNA replication. A hypothetical distribution of the absolute difference between the two sister centromeres, as well as the formula for calculating the fluorescence intensity per segregating unit (α) are indicated on the right. (B) Representative image of mitotic CENP-A-YFP expressing cell. CENP-B staining allows for identification of sister centromeres. Blowup to the right represents a single slice of the boxed region showing that CENP-B is located in between the CENP-A spots of sister centromeres. (C) Frequency distribution of the difference between CENP-A-GFP intensity of sister centromeres in CAG/− cells. (D) Quantification of centromeric CENP-A-GFP based on the stochastic fluctuation method. Each circle represents one centromere; circles on the same column are individual centromeres from the same cell. Left y-axis indicates segregating CENP-A-GFP units in CAG/− cells; right y-axis shows the conversion to minimum number of centromeric CENP-A molecules in CA+/+ (WT) cells. (E) Fluorescent standard method: representative fluorescence images of 4kb-LacO, LacI-GFP S. cerevisiae and human CAG/− cells. (F) Quantification of fluorescence signals of LacI-GFP and CENP-A-GFP spots (n = 2 biological replicates). The left y-axis indicates the fluorescence intensity normalized to LacI-GFP; the right y-axis shows the conversion to maximum number of centromeric CENP-A molecules in wild-type cells. (G) Comparison of independent measurements for the centromeric CENP-A copy number (corrected for CA+/+ levels; Stoch. fluctuations = stochastic fluctuation method [Figure 5A]; Integr. fluorescence = integrated fluorescence method [Figure 2A]). Levels from all strategies are corrected for CA+/+ (WT) levels. Throughout, the average ± SEM and scale bars of 2.5 μm are indicated.

https://doi.org/10.7554/eLife.02137.012

Next, we used a yeast strain that harbors a chromosomally integrated 4 kb LacO-array and expresses GFP-LacI as a calibrated fluorescent standard (Lawrimore et al., 2011). While there is a potential for 204 molecules of GFP-LacI to be bound to this array (Lawrimore et al., 2011), it is unlikely that the entire array is fully occupied at any moment. Because CAG/− cells express the same version of GFP as this yeast strain, direct comparison of fluorescent foci (Figure 5E) provides a maximum estimation of the centromeric CENP-A-GFP copy number. In this way, we determined that CAG/− centromeres contain at most 215 ± 32 CENP-A-GFP molecules, which translates to ≤538 CENP-A molecules in wild-type cells (Figure 5F).

Importantly, the copy number that we measure directly by our 3D integrated fluorescence approach is in close agreement with minimum and maximum estimates of the stochastic fluctuation and fluorescent standard approaches, respectively (Figure 5G). This provides confidence that 400 molecules of CENP-A per centromere in wild-type RPE cells is an accurate measure.

Assessing the critical number of CENP-A nucleosomes

While cells are able to survive with a sixfold range of CENP-A levels (Figure 1D), centromere function may be compromised when levels are not accurately maintained. Indeed, based on a conserved ratio of centromere and kinetochore proteins and kinetochore microtubules between multiple yeast species as well as chicken DT40 cells, it has been hypothesized that centromeres form modular structures by repeating individual structural subunits (Joglekar et al., 2008; Johnston et al., 2010), as originally proposed by Zinkowski et al. (1991). Thus, the amount of CENP-A would directly reflect the number of downstream centromere and kinetochore proteins and microtubule attachment sites. Conversely, experiments in human cells indicate that the centromere is assembled by multiple independent subcomplexes (Foltz et al., 2006; Liu et al., 2006). Here, we analyzed whether altering the levels of CENP-A has an effect on the recruitment of other, downstream centromere or kinetochore proteins. Both CENP-C and CENP-T rely on CENP-A for their centromeric recruitment (Régnier et al., 2005; Liu et al., 2006; Fachinetti et al., 2013) and have recently been shown to be responsible for mitotic recruitment of the KMN network (Gascoigne et al., 2011), including the key microtubule binding protein Hec1/NDC80 (Cheeseman et al., 2006; DeLuca et al., 2006). Interestingly, we found that none of these three proteins were significantly affected by altering the levels of CENP-A between 40% and 240% of wild-type levels (Figure 6A, Figure 6—figure supplement 1). In line with previous findings (Liu et al., 2006; Fachinetti et al., 2013), these results argue against a modular centromere architecture where CENP-A nucleosomes form individual binding sites for downstream components. Rather, a >2½-fold excess of CENP-A appears to be present for recruitment of centromere and kinetochore complexes of fixed pool size.

Figure 6 with 1 supplement see all
Reduction of CENP-A leads to a CENP-C, CENP-T, and Hec1 independent increase in micronuclei.

(A) Quantification of centromeric CENP-A (from Figure 1), CENP-C, CENP-T, and Hec1 levels for indicated cell lines; n = 4 independent experiments in each case. Note that cell lines carrying tagged CENP-A have a slight, yet non-significant impairment in recruiting CENP-C, CENP-T, and Hec1. However, this does not correlate with the CENP-A levels themselves. Below, representative images of indicated antibody staining from CA+/+ cells are shown. Representative images from all cell lines can be found in Figure 6—figure supplement 1. (B) Quantification of the fraction of cells containing micronuclei (MN) for indicated cell lines. Asterisk indicates statistically significant increase compared to wild-type (paired t test; p<0.05; n = 3–4 independent experiments [500–3000 cells per experiment per cell line]); NS indicates no significant difference. Throughout, the average ± SEM is indicated and dashed lines represent wild-type levels. Scale bars: 5 μm.

https://doi.org/10.7554/eLife.02137.014

Intriguingly, despite no quantitative effect on centromeric proteins, we observed that decreasing CENP-A levels leads to an increase in the fraction of cells containing micronuclei (MN; Figure 6B). MN often arise as a consequence of mitotic errors, such as lagging chromosomes during anaphase (Ford et al., 1988), breakage of anaphase bridges (Hoffelder et al., 2004), or multipolar mitoses (Utani et al., 2010). The presence of MN can be scored by DAPI staining (Figure 6B, bottom). A baseline fraction of 0.53% ± 0.07% (n = 4) of wild-type CA+/+ cells contain MN (Figure 6B). Both cell lines that have decreased CENP-A levels show a significantly increased fraction of cells with MN with 2.77% ± 0.48% (n = 3) and 1.95% ± 0.50% (n = 4) in CA+/− and CAG/− cells, respectively. Importantly, these two cell lines were derived independently from the parental CA+/F cell line (Figure 1—figure supplement 1A), which has wild-type levels of CENP-A and no significant increase in MN (Figure 6). In addition, neither cell line with increased CENP-A levels has a larger fraction of MN than CA+/F cells. While the essential role for CENP-A in centromere function is well established (Régnier et al., 2005; Liu et al., 2006; Black et al., 2007), our results indicate that a critical level of CENP-A is passed after reducing the levels to ∼50%. However, the molecular mechanism responsible for MN formation remains unclear, as downstream centromere and kinetochore components of CENP-A remain unaffected.

The contribution of cell type and local centromere features to centromeric CENP-A levels

Interestingly, we find that not all centromeres of the same cell have equal amounts of CENP-A (e.g., Figure 5D). This could either be due to in cis features driving differential regulation of CENP-A on individual centromeres, or by stochastic, yet unbiased, effects at centromeres. To distinguish between these possibilities, we measured the centromeric levels of endogenous CENP-A on specific chromosomes. First, we analyzed a monoclonal HCT-116 cell line that has an integrated Lac-array in a unique position in the genome (Thompson and Compton, 2011). While the site of integration is unknown, expressing LacI-GFP allows for the identification of the same chromosome in a population of cells (Figure 7A). Both the average and variance of CENP-A at this centromere does not differ statistically from the bulk (Figure 7B, Figure 7—figure supplement 1A), arguing against centromere specific features driving CENP-A levels on the Lac-marked chromosome. Conversely, we found that the Y-centromere, uniquely identified by the lack of CENP-B (Figure 7C; Earnshaw et al., 1987), of two independent male cell lines had a slight yet significant reduction of CENP-A (19% in wild-type HCT-116 and 13% in DLD-1; Figure 7D, Figure 7—figure supplement 1B,C), consistent with an earlier report (Irvine et al., 2004). Finally, we used a human patient-derived fibroblast cell line (PDNC-4) where one centromere of chromosome 4 has repositioned to an atypical location (Amor et al., 2004), which we designate as NeoCEN-4 (Figure 7E). As has been observed in other cell lines derived from this patient (Amor et al., 2004), we found that the NeoCEN-4 has a ∼25% decrease in centromeric CENP-A (Figure 7F, Figure 7—figure supplement 1D). Taken together, these results show that while CENP-A expression drives centromeric levels, local sequence or chromatin features can also contribute to the average amount of CENP-A at specific centromeres. Nevertheless, even on these centromeres, the variance in CENP-A levels is maintained, indicating that other stochastic processes contribute to CENP-A levels.

Figure 7 with 1 supplement see all
Centromere and cell specific distribution of CENP-A.

(A, C, E) Representative micrograph of mitotic spreads for LacI-GFP::LacO expressing HCT-116 cells (A); wild-type HCT-116 cells (C); and PDNC-4 cells (E). Blowups show the chromosome containing the integrated Lac-array (A); the Y-chromosome (outline indicated; CENP-B negative) as well as an autosome (CENP-B positive) (C); and the neocentric chromosome 4, containing 2 pairs of ACA spots (staining both CENP-A and CENP-B), but only 1 pair of CENP-A spots (E). (B, D, F) Quantification of CENP-A levels on the centromere of the chromosome containing the Lac-array (CEN-Lac; n = 29; B); the Y-chromosome (CEN-Y; n = 18; D); and neocentric chromosome 4 (NeoCEN-4; n = 39; F) of indicated cell lines compared to all other centromeres within the same cell (Other CENs; n = 1008, 620, and 1592, respectively). Median (line), interquartile distance (box), 3 × interquartile distance or extremes (whiskers), and outliers (spots) are indicated. Figure 7—figure supplement 1 shows results of individual centromeres. Asterisk indicates statistically significant difference (t test; p<0.05); NS indicates no significant difference. (G) Representative images of CENP-A antibody staining in indicated cell types. Images of RPE cells are shown as independent reference. Primary fibrobl. indicates primary human foreskin fibroblasts. (H) Quantification of G. Mean ± SEM for n = 3–4 independent experiments is shown. Left y-axis represents centromeric CENP-A levels normalized to RPE cells; right y-axis shows number of CENP-A molecules per centromere (CEN). (I) Combined results from AH allow for the determination of CENP-A copy numbers on individual chromosomes as indicated. (J) Statistical map of the distribution of 216 CENP-A nucleosomes on the NeoCEN-4 at three different scales. The top 216 peaks are indicated in blue. Y-axis indicates the probability of CENP-A occupancy for each nucleosome. (K) Histogram of the CENP-A nucleosome occupancy. Inset shows the distribution of 216 neocentric CENP-A nucleosomes on the 10% highest occupancy peaks (green) and 90% lowest occupancy peaks (red).

https://doi.org/10.7554/eLife.02137.016

Next, to determine whether the CENP-A copy number of our model cell line is representative for functionally different cells, we performed comparative immunofluorescence against CENP-A (Figure 7G). We analyzed four different cancer cell lines (HeLa, U2OS, HCT-116, and DLD-1), as well as the PDNC-4 neocentromere cell line discussed above, and primary human foreskin fibroblasts that were cultured for a limited number of passages (<15) since their isolation from a patient (Figure 7G). Using these cell lines, we found a sixfold range of centromeric CENP-A levels (Figure 7H), indicating that there is substantial variance between different cell lines. However, we find that the primary cells have a similar amount of CENP-A as RPEs (Figure 7H), arguing that our measure of absolute CENP-A copy numbers made in RPE cells is relevant for healthy, human tissues as well.

We combined these results with our measurements of individual centromeres and determined that, while an average centromere in PDNC-4 cells contains ∼579 molecules of CENP-A, the NeoCEN-4 only contains ∼432. Average Y-centromeres contain ∼143 or ∼87 molecules in HCT-116 and DLD-1 cells, respectively (Figure 7I). In conclusion, we find evidence that cis-elements can have an effect on CENP-A levels, at least on human Y- and neo-centromeres.

A statistical map of CENP-A occupancy at individual nucleosome positions

The number of CENP-A nucleosomes we find at individual centromeres is much smaller (∼25-fold, see Figure 8A) than the total number of nucleosome positions on human centromeric DNA. This indicates that either CENP-A is randomly distributed at a low level throughout the centromere domain or that it occupies specific ‘hotspots’. Due to their repetitive nature, it is not possible to map individual CENP-A nucleosomes on canonical centromeres. However, a recent high-resolution ChIP-seq analysis of the (non-repetitive) NeoCEN-4 identified 1113 unique CENP-A nucleosome positions spanning a ∼300 kb locus (Hasson et al., 2013). By combining the relative height of individual peaks with the total number of CENP-A nucleosomes at this neocentromere, we were able to determine the fraction of cells containing CENP-A at each nucleosome position (Figure 7J). This statistical map of CENP-A occupancy shows that, while the median is ∼6% (Figure 7K), individual positions feature CENP-A with a surprisingly high occupancy (up to 80% of all cells; Figure 7J, arrow). Remarkably, more than one third of all CENP-A nucleosomes are located on the top 10% potential positions (Figure 7K, inset). This strongly suggests that, at least on the NeoCEN-4, a number of nucleosome positioning sequences exist that strongly favor CENP-A over other H3 variants.

A quantitative view of human centromeric chromatin.

(A) Distribution of CENP-A. Estimated ratio of CENP-A (red) to H3 (gray) at the centromere and on non-centromeric loci (genome) in interphase cells. Estimations are calculated assuming 2 CENP-A molecules per nucleosome (Sekulic et al., 2010; Tachiwana et al., 2011; Bassett et al., 2012; Hasson et al., 2013; Padeganeh et al., 2013), an average nucleosome positioning distance of 200 base pairs, an average centromere size of 2.5 × 106 base pairs (Sullivan et al., 1996; Lee et al., 1997) of which approximately 40% (1 Mbp) contains CENP-A (Sullivan et al., 2011), a diploid genome size of 6 × 109 base pairs, 200 CENP-A nucleosomes per centromere, and 2.5 × 104 CENP-A nucleosomes outside of centromeres (9.1 × 104 CENP-A molecules per cell [Figure 2F], of which 74% is in chromatin [Figure 4B] and 0.44% in each centromere [Figure 2B]). The fraction of CENP-A on centromeres, non-centromeric chromatin, and unincorporated CENP-A are indicated in green, blue, and black, respectively. CENP-A nucleosomes are represented as though evenly spread throughout the centromeric domain. Alternatively, they could be distributed into one or more clusters within this domain. (B) Mitotic organization of centromeric chromatin. 200 nucleosomes are redistributed to 100 nucleosomes per centromere on replicated sister chromatids (Jansen et al., 2007; Bodor et al., 2013). The exact CENP-A copy number at the centromere depends on the available total pool (mass-action). Excess CENP-A could either lead to an increased CENP-A domain or lead to a higher density of CENP-A within a domain of fixed size. This pool forms an excess to recruit downstream centromere and kinetochore complexes and ultimately provides microtubule binding sites for ∼17 kinetochore microtubules (McEwen et al., 2001). To avoid mitotic errors, a critical amount of CENP-A is required (dashed lines). (C) Graph representing the chance of at least one chromosome in a cell (with 46 chromosomes) reaching critically low levels of CENP-A by random segregation of pre-existing CENP-A nucleosomes. Calculations were performed for varying levels of critical nucleosome numbers at a fixed steady state of 200 (left), or by varying the steady state number at a fixed critical level of 22 (right). Red bars represent identical calculations.

https://doi.org/10.7554/eLife.02137.018

Discussion

It has been proposed that centromeres in budding yeast feature a single nucleosome of CENP-ACse4 (Meluh et al., 1998; Furuyama and Biggins, 2007). For this reason, the yeast centromere cluster has been extensively used to calibrate fluorescence intensities of CENP-A and other proteins from a number of species (Joglekar et al., 2006, 2008; Johnston et al., 2010; Schittenhelm et al., 2010). However, the single nucleosome hypothesis has recently been challenged (Coffman et al., 2011; Lawrimore et al., 2011; Haase et al., 2013). To avoid dependency on any single reference, we used three independent methods to measure the human centromeric CENP-A copy number. One strategy uses intrinsically controlled fluorescence ratios of cellular and centromeric CENP-A-YFP signals (Figure 2A). The second method does not rely directly on fluorescence intensities, but rather on the stochastic redistribution of CENP-A (Figure 5A). Finally, we compared CENP-A signals directly to a calibrated fluorescent standard (Figure 5E). Importantly, despite the independent nature of these strategies, they all come to a very similar conclusion. Therefore, we demonstrate that typical centromeres in human RPE cells contain ∼400 molecules of CENP-A. While there is a continuing debate on the composition of CENP-A nucleosomes (Black and Cleveland, 2011; Henikoff and Furuyama, 2012), current evidence, at least in human cells, strongly favors an octameric arrangement harboring two copies of CENP-A (Sekulic et al., 2010; Tachiwana et al., 2011; Bassett et al., 2012; Hasson et al., 2013; Padeganeh et al., 2013). Hence, our numbers, correspond to 200 CENP-A nucleosomes in interphase, which are split into 100 nucleosomes on mitotic chromosomes (Figure 8B).

Epigenetic centromere inheritance is achieved by quantitative inheritance of CENP-A across cell division cycles (Jansen et al., 2007; Bodor et al., 2013). We find that rather than accurately ensuring that each daughter receives exactly half, redistribution of CENP-A occurs in a random fashion (Figure 5B,C). Because this type of regulation has the potential for individual centromeres to stochastically inherit critically low levels of CENP-A, the steady state must be sufficiently high to avoid chromosome loss. Although the critical amount of CENP-A is not known, we have previously shown that HeLa cell viability is lost if CENP-A levels are reduced to ∼33% (Black et al., 2007), i.e. 44 nucleosomes (see Figure 7H). Conversely, we show here that CAG/− cells are viable at 40% of RPE levels (80 nucleosomes). Consequently, we estimate that the critical number of nucleosomes that must be inherited, which is half of the steady state level and is replenished during G1 phase, lies between 22 and 40. We used these values to calculate the chance that any one centromere per cell inherits critically low levels of CENP-A for different steady state and critical CENP-A nucleosome levels (Figure 8C). We demonstrate that at a steady state of 200 CENP-A nucleosomes per centromere, less than one in 1016 cell divisions will give rise to a centromere containing 40 CENP-A nucleosomes or less (Figure 8C, left). Thus, the chance of inheriting a critical amount of CENP-A at wild-type steady state levels is negligible. Conversely, with 100 CENP-A nucleosomes at steady state, the chance of a chromosome inheriting even the most stringent critical level of 22 nucleosomes is close to 10−6 (Figure 8C, right), which may pose a significant problem, for example during the development of a human organism. Conversely, although critical levels would be reached even less frequently if centromeres contained a steady state of, for example 300 CENP-A nucleosomes, this degree of accuracy may be superfluous and not outweigh the cost of maintaining a large centromere size (Figure 8C, right). Therefore, we argue that the number of CENP-A molecules found on human centromeres is optimized for robust epigenetic inheritance and centromeric function.

Previously, it has been shown that CENP-A is interspersed with both H3.1 and H3.3 at the centromere (Blower et al., 2002; Sullivan and Karpen, 2004; Ribeiro et al., 2010; Dunleavy et al., 2011; Sullivan et al., 2011). Indeed, based on the average size of the centromeric chromatin domain, we estimate that 200 CENP-A nucleosomes represent only ∼4% of all centromeric nucleosomes (see Figure 8A for calculation). Surprisingly, we find that the majority of chromatin bound CENP-A is located outside the centromere. Indeed, a recent study found that a proportion of CENP-A containing nucleosomes also exist in non-centromeric chromatin of HeLa cells, and is assembled by DAXX, a major chaperone of histone H3.3 (Lacoste et al., 2014). In addition, detectable levels of non-centromeric CENP-A have been observed in budding yeast (Camahort et al., 2009) and chicken DT40 cells (Shang et al., 2013). Here, we quantify this pool in human RPE cells and while there is more than twice as many non-centromeric CENP-A nucleosomes than there are centromeric ones, this only represents <0.1% of all nucleosomes in the genome and thus CENP-A is ∼50-fold enriched (per unit length of DNA) at centromeres (Figure 8A). This result may explain how, despite being outnumbered 25:1 by other H3 variants at the centromere, CENP-A can still accurately specify the centromeric locus. This hypothesis may potentially be tested by creating artificial CENP-A binding sites (e.g., using the LacO/LacI system) of different known sizes and determining the threshold at which centromeres can be formed.

Interestingly, the study by Lacoste et al. showed that the extra-centromeric CENP-A is not randomly distributed, but enriched at sites of high histone turnover (Lacoste et al., 2014). Our finding that CENP-T, CENP-C, and Hec1 do not quantitatively correlate with CENP-A levels (Figure 6A) argues that not each (non-centromeric) CENP-A nucleosome is able to recruit downstream centromere components. It would be interesting to determine to what extent other centromere and kinetochore proteins are present throughout the genome and whether they are also enriched at extra-centromeric CENP-A ‘hotspots’. This question is particularly relevant since it has been observed that downstream centromere components may affect centromeric CENP-A levels (Okada et al., 2006; Carroll et al., 2009, 2010; Hori et al., 2013). A critical combination of components at such hotspots may trigger neocentreomere formation, the mechanisms of which are still unresolved.

Previously, it has been observed that at very high levels of overexpression, CENP-A ceases to be centromere restricted (Van Hooser et al., 2001; Heun et al., 2006; Gascoigne et al., 2011). Nevertheless, here we show that within a sixfold range of expression levels, the CENP-A loading machinery has a constant efficiency, which maintains a strict ratio between the centromeric and total pools of CENP-A. Thus, within a physiological range, centromeric CENP-A levels are regulated by a mass-action mechanism, where the loading efficiency is independent of the expression levels. This mechanism ensures that with fluctuating expression levels, CENP-A remains mainly centromere restricted, and may prevent potential neocentromere seeding.

Remarkably, varying the amount of CENP-A at centromeres during perpetual growth in culture does not affect the levels of several other centromeric proteins. One possible explanation for this is that there is a fixed subset of ‘active’ CENP-A nucleosomes that is responsible for recruiting downstream components, even if the total amount of CENP-A is variable. Alternatively, an excess of CENP-A could form a chromatin domain that provides a ‘platform’ for recruitment of a centromere complex of fixed size. Surprisingly, however, we find that a critical amount of CENP-A for prevention of micronuclei is reached even before downstream centromere and kinetochore protein levels are affected (Figures 6 and 8B).

Our analysis indicates that the distribution of CENP-A among centromeres within one cell is generally uniform. However, in agreement with prior publications, we find that both the Y-centromere and a human neocentromere have lower CENP-A levels (Amor et al., 2004; Irvine et al., 2004). Interestingly, both these centromere types are atypical in that they are formed on relatively small genomic loci: ∼600 kb for the Y-centromere (Abruzzo et al., 1996) and ∼300 kb for the NeoCEN-4 (Hasson et al., 2013), whereas autosomes and the X-chromosome have alpha-sattellite arrays of several magabases in size (Wevrick and Willard, 1989; Mahtani and Willard, 1990; Lo et al., 1999). In addition, in contrast to canonical centromeres, neither the Y-centromere nor neocentromeres recruit the sequence-specific DNA binding protein CENP-B (Earnshaw et al., 1987; Amor et al., 2004), which has been hypothesized to alter the 3D structure of centromeric chromatin (Pluta et al., 1992). The presence of CENP-B binding sites has recently been shown to have a role in phasing CENP-A nucleosomes (Hasson et al., 2013), and to cooperate with CENP-A in kinetochore function (Fachinetti et al., 2013), and may therefore be involved in regulation of centromeric CENP-A levels as well. Furthermore, high resolution analysis of a human neocentromere reveals a non-random distribution of CENP-A (Hasson et al., 2013), where individual nucleosome positions are occupied in anywhere between 0.5% and 80% of cells (Figure 7J,K). Thus, despite specific DNA sequences being neither sufficient nor required for centromere identity (Earnshaw and Migeon, 1985; Voullaire et al., 1993; Amor et al., 2004; Marshall et al., 2008), the amount of CENP-A at centromeres likely results from a combination of a systematic cellular mechanism with a contribution of local sequence or chromatin features.

In conclusion, several key mechanistic insights follow from our findings. First, while CENP-A nucleosomes are highly enriched at the centromere, most CENP-A is distributed at low levels throughout chromatin. This indicates that there is no exclusive pathway that restricts CENP-A assembly to centromeres. Nevertheless, we propose that the ample number of CENP-A nucleosomes facilitates a robust epigenetic signal that can absorb fluctuations in CENP-A inheritance and assembly in order to faithfully maintain centromere identity. Secondly, the requirement for a sizable number of CENP-A nucleosomes to perpetuate an active centromere reduces the likelihood for inadvertent detrimental neocentromere seeding without the need for a tightly restricted assembly mechanism. In addition, the fixed ratio between total and centromeric CENP-A levels may prevent excess CENP-A from accumulating at high density at non-centromeric loci, thus further reducing the probability of neocentromere formation. Finally, the number of centromeric CENP-A nucleosomes represents an ample pool of which only a subset is required to nucleate otherwise self-organized centromere and kinetochore complexes. In summary, from our analysis an integrated view of centromeric architecture, size, and regulation emerges (Figure 8) that provides a basis to explain the self-propagating nature of the epigenetic centromere.

Materials and methods

Cell culture and construction

All human cell lines used were grown at 37°C, 5% CO2. Cells were grown in DMEM/F-12 (RPE), DMEM (HeLa, U2OS, PDNC-4), MEM (primary fibroblasts; Coriell GM06170), McCoy's 5A (HCT-116), or RPMI-1640 (DLD-1) cell culture media. Media were supplemented with 10% fetal bovine serum (FBS), 2 mM glutamine, 1 mM sodium pyruvate (SP), 100 U/ml penicillin, and 100 μg/ml streptomycin, with the following exceptions: for RPE cells SP was substituted for 14.5 mM sodium bicarbonate; for HeLa newborn calf serum was used instead of FBS; for fibroblasts 15% FBS was used; for DLD-1 cells SP was omitted; and both SP and glutamine were omitted for HCT-116 cells. During live cell imaging, culture medium was replaced with Leibowitz's L-15 medium containing 10% FBS and 2 mM glutamine. LacI-GFP::LacO HCT-116 cells (gift from Duane Compton, Thompson and Compton, 2011) were selected alternatingly with 2 μg/ml blasticidin and 300 μg/ml hygromycin; PDNC-4 cells were selected with 100 μg/ml neomycin. All media and supplements were purchased from Gibco (Paisley, UK).

All targeted cell lines are derived from wild-type hTERT RPE cells (CA+/+). Gene targeting was achieved by Adeno-associated virus (AAV) mediated delivery of targeting constructs essentially as described (Berdougo et al., 2009), except in the case if CAG/−cells (see below). The CA+/F cell line was created by inserting loxP sites surrounding CENP-A exons 2 and 4 as described previously (Fachinetti et al., 2013). The CA+/− cell line was created by targeting the floxed CENP-A allele of CA+/F cells with a construct lacking 1373 bp of the CENP-A gene (from 43 bp upstream of exon 2 to 134 bp downstream of exon 4) encompassing the essential CENP-A targeting domain (Black et al., 2007). CAY/− cells were created by sequential targeting of a first CENP-A allele with the targeting construct inserting loxP sites flanking exon 3 and 4 as described above and the second allele by targeting EYFP (carrying citrine and monomerization mutations: Q69M, A206K) in frame with the CENP-A gene, immediately prior to the stop codon in exon 4. The floxed allele was subsequently removed by retroviral delivery of HR-MMPCreGFP, a ‘Hit and Run’ Cre vector, as described (Silver and Livingston, 2001). CAG/− cells were created from an independent CA+/− clone where the remaining intact CENP-A allele was targeted with EGFP using a FACS-based strategy that we developed previously (Mata et al., 2012). Targeting resulted in insertion of the EGFP ORF directly downstream the last coding sequence in exon 4, just upstream of the endogenous stop codon, without insertion of any selectable marker gene. CAY/−+OE cells were created by stable transfection of and selection (5 μg/ml blasticidin) for a CENP-A-YFP expression vector (pBOS-Blast) bearing a CENP-A-YFP fusion protein identical to that of the endogenous knockin locus in CAY/− cells. CAY/−+H2B-RFP and CA+/++H2B-RFP cell lines were created by stable transfection of and selection (5 μg/ml puromycin) for a H2B-RFP expression vector (Black et al., 2007) in CAY/− and CA+/+ cells, respectively. All cell lines were monoclonally sorted by FACS.

For the transient transfection experiment (Figure 1F), wild-type HeLa cells were first synchronized in S phase by addition of 2 mM thymidine. After 17 hr, cells were released using 24 μM deoxycytidine and simultaneously transfected with untagged, wild-type CENP-A and/or HJURP expression vectors (or an empty vector) in combination with an EYFP-CENP-C expression vector (Shah et al., 2004) (2:2:1 proportion). 9 hr later, thymidine was re-added for an additional 15 hr, at which point cells were again released with deoxycytidine for 9 hr. A final thymidine arrest was performed and after 15 hr cells were fixed. Only cells expressing the positive transfection marker EYFP-CENP-C were analyzed. All stable and transient transfections were performed using Lipofectamine LTX (Invitrogen; Carlsbad, CA) according to the manufacturer's instructions.

Immunoblotting and cell fractionation

All samples were prepared in 1X Laemmli sample buffer, separated by SDS-PAGE, and transferred onto nitrocellulose membranes. Whole cell extracts were prepared by lysing cells directly in sample buffer, to ensure that the entire cellular protein pool remained present in the sample. Recombinant CENP-A/H4-complexes were purified as described previously (Black et al., 2004), concentration was determined by A280 measurement and mixed with protein extracts from chicken DT40 cells to bring the overall protein concentration of the purified CENP-A protein preps to a level comparable to the RPE cell extracts. Absence of cross-recognition of human CENP-A antibody to chicken protein was confirmed by omission of recombinant human CENP-A protein in DT40 extracts (Figure 2D, second lane). Alternatively, recombinant CENP-A/H4 was spiked into RPE cell extracts. Results obtained from the two methods are comparable (95.3 ± 14.0 ng [n = 8] and 75.4 ± 5.4 ng [n = 2], respectively; p>0.5). Cellular CENP-A quantity was determined by comparison of fluorescence derived from cellular and purified CENP-A. The following antibodies and dilutions were used: CENP-A (#2186; Cell Signaling Technology, Danvers, MA or Ando et al., 2002) at 1:1000 or tissue culture supernatant at 1:400, respectively; α-tubulin (DM1A; Sigma-Aldrich, St. Louis, MO) at 1:5000; HJURP (gift from Dan Foltz, Foltz et al., 2009) at 1:10,000; Mis18BP1 (A302-825A; Bethyl Laboratories, Inc., Montgomery, TX) at 1:2000; H4K20me2 (ab9052; Abcam, Cambridge, UK) at 1:1000. IRDye800CW-coupled anti-mouse or anti-rabbit (Licor Biosciences) and DyLight680-coupled anti-mouse or anti-rabbit (Rockland Immunochemicals, Gilbertsville, PA) secondary antibodies were used prior to detection on an Odyssey near-infrared scanner (Licor Biosciences, Lincoln, NE). Immunoblot signals were quantified using the Odyssey software, and a linear response was confirmed over a 32-fold range (Figure 2E). Target protein signals were normalized to the α-tubulin loading control signal to correct for slight deviations in cell concentration between extracts of different RPE cell lines.

Cell fractionation was performed for CAY/−+H2B-RFP cells after cell lysis in ice cold buffer consisting of 50 mM Tris-HCl (pH 7.5), 150 mM NaCl, 0.5 mM EDTA, 1% Triton-X 100, 1 mM DTT, and a mix of protease inhibitors (1 mM PMSF, 1 μg/ml leupeptin, 1 μg/ml pepstatin, and aprotinin [A6279; Sigma, 1:1000 dilution]). Soluble proteins were separated from the insoluble fraction by centrifugation at 21,000×g at 4°C and resuspended in an equal volume of lysis buffer. Both supernatant and pellet fractions were incubated with 1.25 U/μl of benzonase nuclease (Novagen, San Diego, CA) on ice for 30 min prior to denaturation in Laemmli sample buffer.

Microscopy

Imaging was performed on an Andor Revolution XD system, controlling an inverted microscope (Eclipse-Ti; Nikon, Tokyo, Japan), an iXonEM+ EMCCD camera (DU-897; Andor, Belfast, UK), a CSU-X1 spinning disk unit (Yokogawa, Tokyo, Japan), a laser combiner/multi-port switch system (Andor) and a motorized stage (Prior Scientific, Cambridge, UK), controlled by MicroManager software (Edelstein et al., 2010). Images were collected using a Nikon 100X, 1.4 NA, Plan Apo oil immersion objective (fixed cell imaging) or a Nikon 60X, 1.2 NA, Plan Apo VC water immersion objective (live cell imaging) at 1× binning. For live cell imaging, the temperature of the chamber was maintained at 37°C.

Fluorescence lifetime measurements

Cells grown on glass coverslips were fixed and mounted as described (Bodor et al., 2012) and imaged using a Zeiss LSM710 coupled to a motorized stage of an upright Zeiss Axio Examiner microscope equipped with a Zeiss 63X, 1.4 NA, Plan Apo oil immersion objective lens. A Coherent Chameleon Vision II multi-photon Ti-Sapphire laser was used to excite EYFP samples. All images were 512 × 512 pixels in size, with a pixel size of 0.09 μm. For all samples, an optimal setting of the laser power and PMT voltage was chosen to avoid pixel saturation and minimize photobleaching. The CLSM settings were kept constant so that valid comparisons could be made between measurements from different samples. Fluorescence lifetime imaging microscopy (FLIM) was performed by measuring the decay rate of EYFP using a Becker & Hickl time-correlated single photon counting hybrid detector coupled to the confocal LSM710 setup. The SPCImage (Becker & Hickl, Berlin, Germany) software was utilized to perform single exponential fitting for each pixel location.

Immunofluorescence and mitotic spreads

Cell fixation, immunofluorescence, and DAPI staining was performed as described previously (Bodor et al., 2012). The following antibodies and dilutions were used: CENP-A (gift from Tatsuo Fukagawa, Ando et al., 2002) tissue culture supernatant at 1:100, rabbit polyclonal CENP-B (sc22788; Santa Cruz Biotechnology, Dallas, TX) at 1:100, tissue culture supernatant from mouse hybridomas expressing monoclonal CENP-B (Earnshaw et al., 1987) at 1:4, CENP-C (Foltz et al., 2009) at 1:10,000, CENP-T (gift from Dan Foltz, Barnhart et al., 2011) at 1:1000, Hec1 (9G3.23; MA1-23308; Pierce, Rockford, IL) at 1:100, ACA (anti-centromere antibodies; 83JD, gift from Kevin Sullivan) at 1:100. Fluorescent secondary antibodies were obtained from Jackson ImmunoResearch (West Grove, PA) or Rockland ImmunoChemicals and used at a dilution of 1:200. Immunofluorescence signals of Figures 1C, 5E, 6B, 7G were automatically quantified using the CRaQ method as described previously (Bodor et al., 2012) using CENP-T or CENP-C as a centromere reference. Hec1 levels were measured exclusively in prometaphase or metaphase (based on DAPI staining) of unperturbed cells. Micronuclei were scored based on DAPI staining.

Mitotic spreads were performed after mitotic shake-off of cells arrested overnight (∼16 hr) in 250 ng/ml nocodazole. 25,000 cells/ml were swollen in 75 mM KCl and 5000 cells were cytospun onto coverslips using a Cytopro 7620 cytocentrifuge (Wescor Inc., Logan, UT) for 4 min, at 1200 rpm, high acceleration. Cells were then fixed and processed for immunofluorescence as described above. Average centromere signals of both sisters were measured after background correction, by subtracting the minimum pixel value from the maximum of a box of 5 × 5 pixels around each sister centromere. Specific chromosomal markers were used as described in the text to detect centromeres of interest and signals were normalized to the average of all centromeres of the same cell spread.

Quantification of the centromeric CENP-A copy number

CA+/+ cells were mixed with CAY/−, CAG/−, or CAY/−+OE cells at a ∼1:4 ratio on 35 mm glass-bottom petri dishes (MatTek Corporation, Ashland, MA). Non-cell permeable dextran-AlexaFluor647 (10,000 MW; Molecular Probes, Eugene, OR) was added at 2–4 μg/ml to stain the medium outside of cells (Figure 2A,I). To minimize oversampling, individual live cells were imaged at 500 nm axial resolution (close to the resolution limit of the objective) spanning the entire cell volume. Images were flatfield corrected for unequal illumination using the signal of a uniform fluorescent slide and the ‘Shading Corrector’ plugin for FIJI. For each axial section, the cell outline was determined based on absence of dextran-AlexaFluor647 staining, and the integrated fluorescence intensities inside the cell outline as well as those of 1–3 independent background regions per section were determined. Background corrected signals from all sections were summed to determine the total cellular fluorescence. Fluorescence measurements of CAY/−, CAG/−, or CAY/−+OE cells were corrected for autofluorescence by subtraction of average per pixel fluorescence intensity of non-fluorescent CA+/+ cells from the same dish. Alternatively, CA+/++H2B-RFP and CAY/−+H2B-RFP cells were mixed and no dextran was added to the medium. In this case, the H2B-RFP signal was used to determine the nuclear volume, and the total nuclear fluorescence was determined as described above for the total cellular volume. Automated centromere detection was performed by an analogous algorithm to a previous study (Bodor et al., 2012, 2013), where diffraction limited spots are detected based on their size, circularity, and feret's diameter. Centromere signals were measured by integrating the intensity of a 5 pixel diameter surrounding each centromere in the appropriate axial section. Local background fluorescence was derived by measuring the difference in intensity between concentric circles of 5 and 7 pixel diameter, and subtracted from centromeric signals (Hoffman et al., 2001). In addition, centromeric signals were corrected for axial oversampling. For this, diffraction limited spots of yellow/green PS-Speck fluorescent beads (Molecular Probes) were measured in multiple plains. The sum intensity of individual beads from all these plains was compared to the signal in the plain with the maximum signal (i.e., the focal plane). The percentage of centromeric fluorescence was determined in relationship to the total fluorescence of each individual cell.

To allow for cell cycle staging of CAY/− cells, transduction with hCdt1(30/120)-RFP was performed using the BacMam 2.0 baculovirus system (Invitrogen). Expression levels of transduced protein were allowed to stabilize for 3 days prior to analysis. Individual cells were followed by live cell microscopy using DIC and RFP signals. Nuclear RFP signals were tracked every ∼2 hr to monitor their cell cycle progression. Imaging of CENP-A-YFP and cellular volume were performed as described above. Analysis of the centromeric CENP-A ratio was performed as described above, but restricted to cells in which RFP levels were decreasing at the specific timepoint of analysis (to exclude cells in G1 phase) and which did not enter mitosis or showed an increase in RFP levels for at least the following 3–4 hr (to exclude cells in G2 phase). Centromeric ratio was compared to non-transduced, randomly cycling cells (Figure 3C) or randomly cycling cells that were transduced, but not followed over time (Figure 3—figure supplement 1). For these experiments, wild-type cells used to measure cellular autofluorescence were seeded on a separate dish.

Stochastic fluctuation measurements

CAY/−, CAG/− or CAY/−+OE cells were treated with nocodazole (250–500 ng/μl) for ∼9 hr, after which cells were fixed and processed for immunofluorescence as described above. Sister centromere pairs were identified by CENP-B staining and GFP or YFP fluorescence intensity of each sister was measured and background corrected by subtracting the minimum pixel value of a 5 pixel diameter circle from the maximum value. The difference (δ) in fluorescence intensity and the sum (Σ) intensity of the two sisters were determined. The fluorescence intensity per segregating unit (α) was determined from the average δ2/Σ of all centromere pairs of the same experiment and cell line. The number of segregating units on each centromere was calculated as Σ/α, as described previously (Rosenfeld et al., 2005, 2006) and in Figure 5A. In addition to sister centromeres, three independent rounds of random centromere pairing between all centromeres measured in a single experiment on CAG/− cells were performed and centromeric CENP-A-GFP units based on these pairings were quantified in Figure 5—figure supplement 1E.

Yeast growth and imaging

4 kb-LacO, LacI-GFP Saccharomyces cerevisiae (gift from Kerry Bloom, Lawrimore et al., 2011) were grown in minimal synthetic media (Yeast nitrogen base [Sigma] + complete synthetic defined single drop-out medium lacking uracil and histidine [MP Biomedicals, Solon, OH]), supplemented with 2% D(+)Glucose (Merck, Darmstadt, Germany). Prior to imaging, log-phase cells (OD600 of ∼0.7) were transferred onto a 2% low melting agarose pad and sealed under a coverslip with VALAP (1:1:1 vaseline:lanolin:paraffin). CAG/− cells were grown on 35-mm glass-bottom petri dishes and yeast and human cells were imaged using identical settings during the same microscopy session. Fluorescence intensity of centromeres and Lac-arrays were quantified after background correction (maximum minus minimum of a 5 × 5 pixel box).

Integrating ChIP-seq and quantitative data of CENP-A at a human neocentromere

CENP-A ChIP-Seq data from the PDNC-4 neocentromere cell line (Accession #GSE44724) was processed as previously described (Hasson et al., 2013). Briefly, paired-end ChIP-Seq reads were aligned to the human genome build hg19 with Bowtie2 version 2.0.0 using paired-end mode. Reads were aligned by using a seed length of 50 bp, and only the single best alignment per read with up to two mismatches was reported in the SAM file. The aligned mate pairs were joined in MATLAB by requiring ≥95% overlap identity. The joined reads were aligned to the PDNC-4 neocentromere and only reads which mapped with 100% identity were used in the subsequent analysis. Nucleosome positions at the neocentromere were determined using the ‘findpeaks’ function in MATLAB. The probability of CENP-A occupancy at a given position was determined according to the following formula: (total reads overlying that position) × (216 CENP-A nucleosomes [Figure 7I])/(total reads mapping to the entire neocentromere).

Calculation of the chance of reaching critical CENP-A levels after random segregation

All calculations represented in Figure 8C were performed in R. For these calculations we assume that CENP-A is inherited following a binominal distribution, consistent with our findings (Figure 5, Figure 5—figure supplement 1A,C). To determine the chance (X) of any chromosome reaching critical levels of CENP-A, the ‘pbinom’ function was used to calculate the fraction of a binomial distribution (where p=0.5 and n [steady state number of nucleosomes] = 200 or was varied as indicated) that is either below a critical value (c = 22, or varied as indicated) or above a critical value (n−c). To determine the chance that any chromosome in a cell (containing 46 chromosomes) reaches critical levels, we calculated the chance that 46 independent centromeres do not reach critical levels and subtracted this chance from 1; [1 − (1 − X)46].

References

  1. 1
  2. 2
  3. 3
    Human centromere repositioning “in progress”
    1. DJ Amor
    2. K Bentley
    3. J Ryan
    4. J Perry
    5. L Wong
    6. H Slater
    7. KHA Choo
    (2004)
    Proceedings of the National Academy of Sciences of the United States of America 101:6542–6547.
    https://doi.org/10.1073/pnas.0308637101
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
    Analysis of protein turnover by quantitative SNAP-based Pulse-Chase imaging
    1. DL Bodor
    2. MG Rodríguez
    3. N Moreno
    4. LET Jansen
    (2012)
    Current Protocols in Cell Biology, Chapter 8, Unit8.8, 10.1002/0471143030.cb0808s55.
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
    Computer control of microscopes using µManager
    1. A Edelstein
    2. N Amodaj
    3. K Hoover
    4. R Vale
    5. N Stuurman
    (2010)
    Current Protocols in Molecular Biology, Chapter 14, Unit14.20, 10.1002/0471142727.mb1420s92.
  26. 26
  27. 27
  28. 28
  29. 29
    Chromosome elimination in micronuclei: a common cause of hypoploidy
    1. JH Ford
    2. CJ Schultz
    3. AT Correll
    (1988)
    American Journal of Human Genetics 43:733–740.
  30. 30
  31. 31
    Centromere identity is specified by a single centromeric nucleosome in budding yeast
    1. S Furuyama
    2. S Biggins
    (2007)
    Proceedings of the National Academy of Sciences of the United States of America 104:14706–14711.
    https://doi.org/10.1073/pnas.0706985104
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
    Heterochromatic deposition of centromeric histone H3-like proteins
    1. S Henikoff
    2. K Ahmad
    3. JS Platero
    4. B van Steensel
    (2000)
    Proceedings of the National Academy of Sciences of the United States of America 97:716–721.
    https://doi.org/10.1073/pnas.97.2.716
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49
  50. 50
  51. 51
  52. 52
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57
  58. 58
  59. 59
  60. 60
  61. 61
  62. 62
    Purification of the centromere-specific protein CENP-A and demonstration that it is a distinctive histone
    1. DK Palmer
    2. K O’Day
    3. HL Trong
    4. H Charbonneau
    5. RL Margolis
    (1991)
    Proceedings of the National Academy of Sciences of the United States of America 88:3734–3738.
  63. 63
  64. 64
  65. 65
  66. 66
  67. 67
    A super-resolution map of the vertebrate kinetochore
    1. SA Ribeiro
    2. P Vagnarelli
    3. Y Dong
    4. T Hori
    5. BF McEwen
    6. T Fukagawa
    7. C Flors
    8. WC Earnshaw
    (2010)
    Proceedings of the National Academy of Sciences of the United States of America 107:10484–10489.
    https://doi.org/10.1073/pnas.1002325107
  68. 68
  69. 69
  70. 70
  71. 71
  72. 72
  73. 73
  74. 74
  75. 75
  76. 76
  77. 77
  78. 78
  79. 79
  80. 80
  81. 81
  82. 82
  83. 83
  84. 84
  85. 85
    Chromosome missegregation in human cells arises through specific types of kinetochore-microtubule attachment errors
    1. SL Thompson
    2. DA Compton
    (2011)
    Proceedings of the National Academy of Sciences of the United States of America 108:17974–17978.
  86. 86
  87. 87
    Specification of kinetochore-forming chromatin by the histone H3 variant CENP-A
    1. AA Van Hooser
    2. II Ouspenski
    3. HC Gregson
    4. DA Starr
    5. TJ Yen
    6. ML Goldberg
    7. K Yokomori
    8. WC Earnshaw
    9. KF Sullivan
    10. BR Brinkley
    (2001)
    Journal of Cell Science 114:3529–3542.
  88. 88
    A functional marker centromere with no detectable alpha-satellite, satellite III, or CENP-B protein: activation of a latent centromere?
    1. LE Voullaire
    2. HR Slater
    3. V Petrovic
    4. KH Choo
    (1993)
    American Journal of Human Genetics 52:1153–1163.
  89. 89
    Long-range organization of tandem arrays of alpha satellite DNA at the centromeres of human chromosomes: high-frequency array-length polymorphism and meiotic stability
    1. R Wevrick
    2. HF Willard
    (1989)
    Proceedings of the National Academy of Sciences of the United States of America 86:9394–9398.
  90. 90
  91. 91
  92. 92
  93. 93

Decision letter

  1. Jon Pines
    Reviewing Editor; The Gurdon Institute, United Kingdom

eLife posts the editorial decision letter and author response on a selection of the published articles (subject to the approval of the authors). An edited version of the letter sent to the authors after peer review is shown, indicating the substantive concerns or comments; minor concerns are not usually shown. Reviewers have the opportunity to discuss the decision before the letter is sent (see review process). Similarly, the author response typically shows only responses to the major concerns raised by the reviewers.

[Editors’ note: although it is not typical of the review process at eLife, in this case the editors decided to include the reviews in their entirety for the authors’ consideration as they prepared their revised submission.]

Thank you for sending your work entitled “The quantitative architecture of centromeric chromatin” for consideration at eLife. Your article has been favorably evaluated by a Senior editor, a Reviewing editor, and 3 reviewers.

The Reviewing editor and the other reviewers discussed their comments before we reached this decision, and the Reviewing editor has assembled the following comments to help you prepare a revised submission.

Two primary concerns were raised.

1) Whether the quantitative immunoblotting is accurate. Reviewer 2 recommends quantitative blots of the discarded insoluble material to demonstrate that all CENP-A has been extracted. Then to demonstrate that CENP-A is not passing through the filters blotting should be done with a stack of two or (better) three filters, and demonstrating that CENP-A cannot be detected on the second and third filter.

2) Reviewer 3 suggests that averaging the quantitation over an asynchronous population leads to an inaccurate estimate of CENP-A loading. This may explain the variation between individual cells mentioned by reviewer 1. To control for this the reviewers suggest selecting cells at G1/S to represent cells that completed centromeric CENP-A assembly (100% centromere occupancy and minimal soluble pool) and late G2 or mitotic cells to represent cells with maximal nascent soluble pool, then perform the quantitation on these isolates. These populations should be obtained without using drug synchronisation that could perturb the process, for example by mitotic shake off. In this respect, Reviewer 3 emphasises (point 5) that your argument that low levels of CENP-A could account for mis-segregation would not stand if sisters with less CENP-A are 'replenished' during the subsequent G1.

These points will require more experimental data; the other points below should be addressed by re-writing the manuscript notably those concerning potential over-interpretation of the data. None of the referees was persuaded by your conclusions that the generation of micronuclei is a surrogate marker for centromere dysfunction.

Reviewer 1 Comments:

1) It would have been interesting to complement 1E by showing a similar graph for a protein that does not follow mass-action (e.g. a kinetochore component?).

2) Why do GFP and YFP constructs yield considerably different results with the stochastic fluctuation method?

3) The authors assume that a large portion of CENP-A is assembled into non-centromeric chromatin. Is there any evidence that these CENP-A molecules are indeed incorporated into nucleosomes/chromatin (rather than being chromatin-associated in a less defined manner)?

4) 2B/3A [currently Figures 2B and 4A]: In my opinion the average should have been calculated by averaging the mean of single cells rather than averaging over all centromeres (which gives some cells more weight than others).

5) The y error bars in 1E do not match the error bars in 1D, although the data should be identical.

6) I would have appreciated yet more detail on the experimental methods. For example: (a) although described in the previous paper, briefly mention by which criteria centromeres are included in or excluded from the quantification; (b) how is the value for axial oversampling determined?; (c) I do not understand the rational for subtracting the minimum pixel value, since this is already background-subtracted (Immunofluorescence and mitotic spreads section); (d) comment on differences/similarities between mixing recombinant CENP-A/H4 into chicken or human cell extract.

7) The text covering Figure 1B/C could be improved. Make clear that immunostaining can be compared between cell lines. In contrast, I would not state that GFP and YFP blotting can be compared.

8) 5B [currently Figure 6A] suggests that the tagged versions are not fully functional, which could be mentioned.

9) 7B [currently Figure 8B]erroneously implies that abundance correlates with spatial expansion - should be removed or revised. 7C [currently Figure 8C]: labeling could be improved to make the figure easier to understand.

Reviewer 2 Comments:

1) Measuring the cellular level of a protein by immunoblotting is tricky. Here a bit more work is needed by the authors. They need to provide some sort of convincing control that they are measuring all of the cellular CENP-A.

Firstly, we need quantitative blots of the discarded insoluble material to demonstrate that all CNP-A has been extracted. Secondly, they need to demonstrate that CENP-A is not passing through the filters they are using for quantitative immunoblotting. This is done by doing the blots with a stack of two or (better) three filters, and demonstrating that CENP-A cannot be detected on the second and third filter. When an analogous experiment was done with CENP-B, the protein was detected even unto the third filter. Despite this concern, I am reasonably content with the conclusion that the average centromere contains ∼400 molecules.

2) In my opinion, the authors state, but do not fully explore their most important conclusion - e.g. that “about one-fifth of the CENP-A protein content (0.44% x 46) is present on the functionally relevant subcellular location”. The authors should say “centromeres” instead of “the functionally relevant subcellular location”.

3) The authors state that neocentromeres are “fully functional centromeres that have repositioned to atypical loci on the chromosome”. Can the authors cite a study that has actually demonstrated that neocentromeres are fully functional (e.g. a study that has measured their efficiency at directing chromosome segregation in a quantitative manner)?

4) I am less convinced by the attempts of the authors at assessing the critical number of CENP-A nucleosomes required for centromere function. Clearly budding yeast can do this with ∼1 nucleosome, and Bruce Nicklas showed years ago that a single microtubule could move a huge newt chromosome in the cytoplasm. Furthermore, the authors' failure to see alterations in levels of CNP-C, -T or Ndc80 when the minimum “critical” threshold level of CENP-A is passed suggests that the explanations for changes in the micronucleus index that they see when CENP-A levels drop below 50%.

5) I was interested by the statement that “Interestingly, we find that not all centromeres of the same cell have equal amounts of CENP-A”. The authors should carry out a similar quantitation on CENP-C and CENP-T.

6) The fact that the Y centromere has less CENP-A seems to impress the authors, but they seem to ignore an obvious possible explanation for this: the Y chromosome also has by far the smallest centromeric alpha-satellite DNA array. In my experience, neocentromeres also have lower levels of core centromere proteins (these authors also look at this and confirm this conclusion). Thus, I believe that in their speculations on the explanation for this phenomenon, the authors should include the overall size of the alphoid array in addition to specific sequence preferences.

7) The authors state that “cis-elements can have an effect on CENP-A levels, at least on atypical human Y- and neocentromeres”. Is the Y centromere atypical? Every male human has one.

8) To me the authors have chosen to make light of the most amazing result of this paper - that 80% of the CENP-A on chromosomes is not at centromeres! Others have found that there are significant amounts of CENP-A outside centromeres, but have not tried to quantify it as convincingly as these authors. For example in Camahort et al. (Mol Cell v. 35), the fact that significant amounts of Cse4 were found outside the centromere was mentioned. Recently in the neocentromere paper by Hori et al. (Dev. Cell), their ChIP-Seq data revealed significant amounts of CENP-A outside the chicken centromeres Z and 5, but again overall amounts were not given. The authors should expand their discussion to explore this. For example, how much of a non-centromeric pool is there likely to be for other CCAN components? The answer could be similar to that for CENP-A. Thus, the key question is what causes the critical local concentration that nucleates the centromeric chromatin?

9) It would not detract from the import of the present work if the authors here mentioned the fact that their numbers are not terribly dissimilar from those estimated by Riberio et al. (PNAS), particularly given that the latter were looking at a different organism.

10) The statement “Previously, it was shown that CENP-A is interspersed with both H3.1 and H3.3 at the centromere (Blower et al. 2002; Dunleavy et al. 2011).” should also reference the Ribeiro et al. study.

11) The authors go on in the same paragraph to say “Nevertheless, this non-centromeric pool only represents <0.1% of all nucleosomes in the genome and thus CENP-A is ∼50-fold enriched (per unit length of DNA) at centromeres (Figure 7a [currently Figure 8A]). This explains how, despite being outnumbered 25:1 by other H3 variants at the centromere, CENP-A can still accurately specify the centromeric locus.” Are they claiming to offer a convincing explanation of how the CENP-A accurately specifies the centromere? If they really believe this, perhaps they could expand and come up with a more directly testable model.

12) Where the authors say “Indeed, it has recently been shown that this sequence specific DNA binding protein has a role in phasing CENP-A nucleosomes (Hasson et al. 2013)”, they might like to read, and possibly cite, the rather extensive discussion of this in Pluta et al. (JCB 116, 1081, pp. 1091, 1092). The discussion of the enigmatic CENP-B aside, I feel (as already stated above) that the reason for the low levels of CENP-A on the Y chromosome most likely arises from the small size of the alphoid array, which the authors should mention here as well.

13) The authors state “the fixed ratio between total and centromeric CENP-A levels prevents excess CENP-A from accumulating at high density at non-centromeric loci”. I believe that this is an overstatement of what the data actually show. This is still, in my mind, a hypothesis - not a rigorously proven fact, and should be stated as such.

14) Lastly, I do not understand Figure 7b [current Figure 8B]. What is the “critical size” and why is this bigger than the kinetochore?

Reviewer 3 Comments:

The authors should discuss possible limitations of their findings, and be careful when making concluding statements. For example, the authors tried to correlate further reduction of CENP-A from the parental RPE line with increases in micronuclei. I am not convinced that the slightly increased frequency of micronuclei is due to a moderate decrease in CENP-A. They would need to show that restoring CENP-A suppresses the phenotype. It is especially a problem when the authors attempt to say that this amount of CENP-A represents the number in “normal” tissue. The results are meaningful in the background of the RPE cell line, but even it is a relevant number, it still has limited value since they have shown that CENP-A levels vary greatly among cell lines and there is no strict correlation with the level of genome instability. This suggests that other 'background' mutations probably contribute greatly to the level of genome instability, and that they only made a weak correlation between CENP-A reduction and increased micronuclei. Overall, these issues raise concerns about using micronucleus frequencies as a direct measure of the impact of altering CENP-A levels, and whether CENP-A reduction accounts for the observed phenotype. The authors determined the critical number based on the observations that HeLa cells with 33% CENP-A are not viable but RPE cells with 40% CENP-A are viable, so they reason that the critical number must lie somewhere in between these two numbers. However, these are completely different lines, with HeLa cells carrying hundreds or thousands of other mutations not in the RPE background. Viability is a combined readout of more than one related pathway. Viability of HeLa cells and RPE cells may have different sensitivity to the levels of CENP-A due to potential mutations in related pathways.

There are several places the authors cite improper literatures, please cite original studies. For example, in introduction, to make the point that CENP-A is so far the primary candidate for centromere identity, the authors need to cite papers beyond just studies on human CENP-A; in the next sentence, to show the point that CENP-A is stably transmitted through cell divisions, there are much more authoritative papers on humans and other organisms than the authors' own papers; the paper cited for primary DNA sequence being neither necessary nor sufficient for centromere determination is Amor et al. in 2004, which is not the best for making this point.

In several places the authors stated results as if they are original observations. However, some of the principles have been revealed or suggested previously, and are here reinforced by the measurements in this manuscript. For example, the correlation between reduction of CENP-A and increased genome instability (if indeed due to CENP-A reduction in the RPE background), and different amounts of CENP-A are at different centromeres. The authors should be careful and need to cite the literature properly.

1) Local background correction for Figure 2 is problematic. The basis of doing local background correction is the assumption that signals immediately surrounding centromeres are noise. What is the factual foundation of this methodology considering that recent papers and the authors' own data indicated the existence of non-centromeric CENP-A in chromatin? If the local background is indeed extremely low, treating it as background noise may not distort the actual number significantly.

2) Figure 1, to overcome the difficulty of different transfer efficiency for untagged or tagged CENP-A, the authors could also use known quantities of purified tagged and untagged CENP-A on the same gel to calibrate tagged CENP-A cell lines.

3) What is the cause of the observed different levels of CENP-A for different constructs in Figure 1? Did the authors compare different clones? This may suggest that other mutations occurred during generation of these lines along with changes in CENP-A levels.

4) Figure 1f, this experiment needs to be performed in the RPE lines not in HeLa. In addition, since transient CENP-A overexpression leads to incorporation into non-centromeric chromatin through a DAXX dependent pathway, this must represent a re-equilibrium of two different loading systems that utilize different assembly factors (HJURP and DAXX), and between centromeres and non-centromeric loci. Does DAXX mediated incorporation happen in the various RPE cells? The authors only considered canonical CENP-A loading at the centromere, but I don't think that it makes much sense to talk about mass action without considering the DAXX dependent assembly that may compete with HJURP for centromeric assembly. This will clearly influence the size of the available CENP-A pool for centromeric loading.

5) Figure 4b and 4c [current Figure 5B,C], the authors measured fluorescence associated with mitotic centromeres. They found that different centromeres behave differently in the same cell. Using data from multiple cells, they concluded that segregation of CENP-A nucleosomes is random. To conclude this, the authors should have compared the behavior of an individual pair between different cells. It is possible that segregation is non-random for a given centromere of the same chromosome, but varies between centromeres of different chromosomes. Combining data from all centromeres without separating individual chromosomes may give rise to the frequency graph in Figure 4c [current Figure 5C]. In addition, the authors need to consider that any differences in the levels of CENP-A at different centromeres/sisters observed in mitosis could be counteracted during new CENP-A assembly in the subsequent G1 phase.

6) The variation of numbers of CENP-A per centromere using different cell lines is over 2.5 fold, which is fairly large. What could be contributing to this variation?

7) Figure 5 [current Figure 6], the micronuclei phenotype has very limited value in assessing centromere defects. It is actually problematic as the end phenotype for centromere dysfunction, since there could be defects in sister separation (cohesion), as well as genome instability/breakage. It is more direct and informative to look at chromosome segregation. Without additional experiments, e.g. live studies, the results and their implications are hard to interpret. The authors also need to evaluate more than one clone for each line.

8) For the phenotype in Figure 5 [current Figure 6], how is it striking to see a correlation between reduced CENP-A and increased genome instability? More important, it is not known whether this is due to the problem of the tags or other potential mutations arising during generation of the cell lines. To conclude that the micronuclei phenotype is caused by reduction of CENP-A, the authors would have to do a rescue experiment to restore CENP-A to normal.

9) Figure 6 [current Figure 7], the authors made a statement about the variance of CENP-A between untransformed cells and cancer cells based on few lines from each category. This statement should be removed. It is safe to say that variance in cancer cell lines is big (6-fold range), but I don't think the evidence is sufficient to say anything about untransformed cells.

10) In the Discussion, please rephrase “the fixed ratio between total and centromeric CENP-A levels prevents excess CENP-A from accumulating at high density at non-centromeric loci, thus further reducing the probability of neocentromere formation.” Fixed ratio is an observation it should not be taken as a mechanism to explain how neo-centromeres are prevented.

11) Also the model figure shows widely spaced single CENPA nucleosomes and could easily be clustered more.

https://doi.org/10.7554/eLife.02137.019

Author response

1) Whether the quantitative immunoblotting is accurate. Reviewer 2 recommends quantitative blots of the discarded insoluble material to demonstrate that all CENP-A has been extracted. Then to demonstrate that CENP-A is not passing through the filters blotting should be done with a stack of two or (better) three filters, and demonstrating that CENP-A cannot be detected on the second and third filter.

In response to the first point raised, regarding discarded insoluble material, we would like to emphasize that to ensure no protein was lost during preparation of the lysates, whole cell extracts were prepared without clearance of the insoluble protein pool in all of our experiments (except for the fractionation experiments in Figure 4). We now clarified this in the main text, in Figure 2d, and in the Methods section.

In addition, we performed the experiment suggested by the reviewer, where we blot CENP-A onto a stack of three membranes (Figure 2–figure supplement 3). The reviewer was correct that CENP-A can indeed transfer through the membranes. However, importantly, we show that the amount of CENP-A retained on the first membrane (which is quantified throughout) is the same between recombinant and cellular (untagged) CENP-A. Therefore, we remain confident that a quantitative comparison between CENP-A derived from these two sources is accurate. In addition, we found that tagged CENP-A does not pass through the first filter, which reinforces our statement that we cannot quantitatively compare proteins of different sizes by immunoblotting.

2) Reviewer 3 suggests that averaging the quantitation over an asynchronous population leads to an inaccurate estimate of CENP-A loading. This may explain the variation between individual cells mentioned by reviewer 1. To control for this the reviewers suggest selecting cells at G1/S to represent cells that completed centromeric CENP-A assembly (100% centromere occupancy and minimal soluble pool) and late G2 or mitotic cells to represent cells with maximal nascent soluble pool, then perform the quantitation on these isolates. These populations should be obtained without using drug synchronisation that could perturb the process, for example by mitotic shake off. In this respect, Reviewer 3 emphasises (point 5) that your argument that low levels of CENP-A could account for mis-segregation would not stand if sisters with less CENP-A are 'replenished' during the subsequent G1.

The reviewer is correct that one possible source of variation between individual cells may be cell cycle related. However, performing the suggested experiment is challenging at best. We found that cells rapidly lose synchrony after mitotic shake off (even before a majority of cells reaches S phase). Instead, we address this issue in a manner that avoids using drug-based synchronization using part of the FUCCI system developed by Sakaue-Sawano et al. (Cell, 2008). With this system, we were able to selectively measure the centromeric CENP-A ratio in S phase cells. Because in human cells CENP-A expression starts in (late) G2 phase (Shelby et al., JCB 2000) and incorporation occurs in (early) G1 phase (Jansen et al., 2007), by selecting cells in S phase, we likely discard any variation of centromeric to whole cell fluorescence that would be induced by cell cycle related effects.

We expressed hCdt1(30/120)-RFP in CAY/- cells using a commercially available baculovirus system. This fluorescent marker is expressed in all stages of the cell cycle, but degraded during S and G2 phase and mitosis. Thus, RFP fluorescence of cells entering into and progressing through S phase decreases, while signal increases again in G1 phase. To select for S phase cells, we followed cells live by microscopy and screened the RFP signal. We selectively analyzed CENP-A fluorescence in cells in which the RFP signal decreased with respect to previous time points (thus excluding cells in G1 phase), and which did not enter mitosis and/or started increasing RFP fluorescence for the following 4 hours after analysis (thus excluding G2 and mitotic cells).

In summary, we found that in S phase, neither the mean centromeric ratio nor the variance differs from randomly cycling cells. The description and results of this experiment have been added to our manuscript in the main text and Methods and in the current Figure 3 and Figure 3–figure supplement 1.

Reviewer 3 states that our argument that low levels of CENP-A could account for mis-segregation would not stand if sisters with less CENP-A are 'replenished' during the subsequent G1. We are aware of such proposals where centromeres with critically low levels would selectively recruit more CENP-A (e.g. Brown and Xu, BioEssays 2009). However, to our knowledge there is no experimental evidence for such a mechanism. In addition, when CENP-A drops below critical levels e.g. during S phase, this may lead to mitotic failure in the same cell cycle, even before replenishment which happens later in G1 phase. Therefore, our argument that centromeres with critically low CENP-A levels are likely resulting in centromere failure is not unreasonable and with the present state of knowledge the most parsimonious.

These points will require more experimental data; the other points below should be addressed by re-writing the manuscript notably those concerning potential over-interpretation of the data. None of the referees was persuaded by your conclusions that the generation of micronuclei is a surrogate marker for centromere dysfunction.

We have now extensively rewritten the section of our manuscript that concerns the determination of critical CENP-A levels to more accurately reflect what the results show. In brief, we have attenuated our conclusions regarding the increased number of micronuclei present in cells with lowered CENP-A levels. We removed the suggestion that this is indicative of mitotic errors and do not make any conclusive statement regarding the molecular nature of micronucleus formation in these cells. In addition, we have now put more emphasis on the fact that downstream centromere/kinetochore protein recruitment does not quantitatively rely on the amount of CENP-A present. Importantly, this last result argues against an existing model of a modular centromere architecture that is assembled from repeating structural substructures, initially proposed by Zinkowski et al. in 1991.

Reviewer 1 Comments:

1) It would have been interesting to complement 1E by showing a similar graph for a protein that does not follow mass-action (e.g. a kinetochore component?).

We agree that this is an interesting issue. However, we feel that it lies outside of the scope of our study.

2) Why do GFP and YFP constructs yield considerably different results with the stochastic fluctuation method?

The stochastic fluctuation method measures the number of what we designated as ‘segregating units,’ i.e. the number of fluorophores that co-segregate. There is extensive (albeit indirect) evidence that existing octameric nucleosomes containing two CENP-A molecules do not split in two during DNA replication. Therefore, a minimum of two molecules of CENP-A co-segregate, thus forming a segregating unit. However, the extent of co-segregation may also depend on the density of CENP-A nucleosomes in the centromere, which would lead to the difference we observed between our cell lines (i.e. CENP-A-YFP being expressed at higher levels). Thus, without making assumptions on the exact nature of the co-segregating unit, we cite measurements from the stochastic fluctuation method as a ‘minimal estimate’ of the CENP-A copy number throughout the manuscript. Importantly, independent of the cell line used, we find that results obtained with the stochastic fluctuation method agree within an approximately 2-fold range with the results from the integrated fluorescence approach (Figure 2). We have modified the section on this quantification method to help clarify the issue raised above.

3) The authors assume that a large portion of CENP-A is assembled into non-centromeric chromatin. Is there any evidence that these CENP-A molecules are indeed incorporated into nucleosomes/chromatin (rather than being chromatin-associated in a less defined manner)?

Although nucleosome incorporation is likely the case (in part taking into account a study in Molecular Cell by the Almouzni lab that appeared in print during submission of our manuscript to eLife), it is true that we have no direct evidence for the status of chromatin bound CENP-A. We now refer to this pool CENP-A as “chromatin bound”.

4) 2B/3A [currently Figures 2B and 4A]: In my opinion the average should have been calculated by averaging the mean of single cells rather than averaging over all centromeres (which gives some cells more weight than others).

We realize that there are valid arguments for either way of averaging centromeric fractions. However, in our particular datasets the different methods for averaging give near-identical results. Specifically, averaging as suggested by the reviewer would result in the following values (number between brackets shows averages over all centromere, as performed in the original submission): CAY/-: 0.43% (0.44%), CAG/-: 0.37% (0.38%), CAY/-+OE: 0.38% (0.38%), and CAY/-+H2B-RFP: 0.77% (0.73%).

5) The y error bars in 1E do not match the error bars in 1D, although the data should be identical.

We thank the reviewer for pointing out this error in the figure. We had inadvertently used the standard deviation in Figure 1E rather than the standard error of the mean, which is the measure used in Figure 1D and throughout the manuscript. We have now corrected this in Figure 1E and Figure 1–Figure supplement 1.

6) I would have appreciated yet more detail on the experimental methods. For example: (a) although described in the previous paper, briefly mention by which criteria centromeres are included in or excluded from the quantification; (b) how is the value for axial oversampling determined?; (c) on page 25, I do not understand the rational for subtracting the minimum pixel value, since this is already background-subtracted (Immunofluorescence and mitotic spread section); (d) comment on differences/similarities between mixing recombinant CENP-A/H4 into chicken or human cell extract.

We have added experimental details as suggested for points (a), (b), and (d). Regarding point (c), we have now clarified in the text that subtraction of the minimum pixel value is in fact the background correction method applied.

7) The text covering Figure 1B/C could be improved. Make clear that immunostaining can be compared between cell lines. In contrast, I would not state that GFP and YFP blotting can be compared.

The first point is now emphasized in the legend of Figure 1C.

Regarding the second point: despite the few (8) amino acid differences (and small apparent size difference) between CENP-A-YFP and CENP-A-GFP, we feel that these fusion proteins are still directly comparable by Western Blot and have no indications that this would be problematic. This is in sharp contrast to the comparison between proteins of vastly different sizes such as untagged CENP-A (140 amino acids; ∼16 kDa) and tagged CENP-A-GFP or -YFP (385 amino acids, ∼43 kDa). See also the new Figure 2–figure supplement 3 regarding this issue.

8) 5B suggests that the tagged versions are not fully functional, which could be mentioned.

This is now mentioned in the figure legend (current Figure 6A)

9) 7B erroneously implies that abundance correlates with spatial expansion - should be removed or revised. 7C: labeling could be improved to make the figure easier to understand.

Current Figure 8B: The figure legend was adapted to include the alternative possibility of having an increased density of CENP-A within a domain of fixed size.

Current figure 8C: The y-axis was relabeled for clarity and further explanation was added to the legend.

Reviewer 2 Comments:

1) Measuring the cellular level of a protein by immunoblotting is tricky. Here a bit more work is needed by the authors. They need to provide some sort of convincing control that they are measuring all of the cellular CENP-A.

Firstly, we need quantitative blots of the discarded insoluble material to demonstrate that all CNP-A has been extracted. Secondly, they need to demonstrate that CENP-A is not passing through the filters they are using for quantitative immunoblotting. This is done by doing the blots with a stack of two or (better) three filters, and demonstrating that CENP-A cannot be detected on the second and third filter. When an analogous experiment was done with CENP-B, the protein was detected even unto the third filter. Despite this concern, I am reasonably content with the conclusion that the average centromere contains ∼400 molecules.

As this issue has been raised as one of the primary concerns regarding our manuscript, we addressed it above.

2) In my opinion, the authors state, but do not fully explore their most important conclusion - e.g. that “about one-fifth of the CENP-A protein content (0.44% x 46) is present on the functionally relevant subcellular location”. The authors should say “centromeres” instead of “the functionally relevant subcellular location”.

We agree with the reviewer that our finding that only a minority of CENP-A is centromere localized is an important one. We have now emphasized this point in the text. We have also emphasized that by the functionally relevant location we do in fact mean at the centromere.

3) The authors state that neocentromeres are “fully functional centromeres that have repositioned to atypical loci on the chromosome”. Can the authors cite a study that has actually demonstrated that neocentromeres are fully functional (e.g. a study that has measured their efficiency at directing chromosome segregation in a quantitative manner)?

We have removed the word ‘fully’, as it has in fact been shown that at least one particular neocentromere is less efficient at error correction than endogenous centromeres (Bassett et al., JCB 2010)

4) I am less convinced by the attempts of the authors at assessing the critical number of CENP-A nucleosomes required for centromere function. Clearly budding yeast can do this with ∼1 nucleosome, and Bruce Nicklas showed years ago that a single microtubule could move a huge newt chromosome in the cytoplasm. Furthermore, the authors' failure to see alterations in levels of CNP-C, -T or Ndc80 when the minimum “critical” threshold level of CENP-A is passed suggests that the explanations for changes in the micronucleus index that they see when CENP-A levels drop below 50%.

Unfortunately, the last part of this comment was lost and it is not completely clear to us what the reviewer meant. Nevertheless, we accept the criticism that the increase in micronuclei at low CENP-A levels does not necessarily result from centromere failure. As outlined above, we have rewritten this section of the manuscript to more accurately represent the observations made.

5) I was interested by the statement that “Interestingly, we find that not all centromeres of the same cell have equal amounts of CENP-A”. The authors should carry out a similar quantitation on CENP-C and CENP-T.

We show in Figure 6A that CENP-C and -T levels do not correspond directly the levels of CENP-A. Furthermore, we provide evidence in Figure 7 that a large proportion of the variation seen in the levels of CENP-A is stochastic in nature. Thus, while the CENP-C and -T levels are likely also variable between centromeres, it is unlikely that this variation would be caused by variable CENP-A levels. Based on these arguments we feel that the rather extensive analysis proposed by the reviewer lies beyond the scope of our study.

6) The fact that the Y centromere has less CENP-A seems to impress the authors, but they seem to ignore an obvious possible explanation for this: the Y chromosome also has by far the smallest centromeric alpha-satellite DNA array. In my experience, neocentromeres also have lower levels of core centromere proteins (these authors also look at this and confirm this conclusion). Thus, I believe that in their speculations on the explanation for this phenomenon, the authors should include the overall size of the alphoid array in addition to specific sequence preferences.

The reviewer is correct that the size of Y centromere and neocentromere is indeed another difference between these centromeres and other centromeres and could be causative for the different levels of CENP-A. We have added this hypothesis to the Discussion, regarding why these centromeres carry less CENP-A.

7) The authors state that “cis-elements can have an effect on CENP-A levels, at least on atypical human Y- and neocentromeres”. Is the Y centromere atypical? Every male human has one.

We have removed the term atypical from this section of the paper. We explain in what sense we consider these two types of centromeres atypical, namely that they lack CENP-B and are formed on relatively small genomic loci.

8) To me the authors have chosen to make light of the most amazing result of this paper - that 80% of the CENP-A on chromosomes is not at centromeres! Others have found that there are significant amounts of CENP-A outside centromeres, but have not tried to quantify it as convincingly as these authors. For example in Camahort et al. (Mol Cell v. 35), the fact that significant amounts of Cse4 were found outside the centromere was mentioned. Recently in the neocentromere paper by Hori et al. (Dev. Cell), their ChIP-Seq data revealed significant amounts of CENP-A outside the chicken centromeres Z and 5, but again overall amounts were not given. The authors should expand their discussion to explore this. For example, how much of a non-centromeric pool is there likely to be for other CCAN components? The answer could be similar to that for CENP-A. Thus, the key question is what causes the critical local concentration that nucleates the centromeric chromatin?

We thank the reviewer for pointing this out, as well as for his enthusiasm regarding our finding. We have now added a reference to the non-centromeric pool of CENP-A described in a recent study by Lacoste et al. (Mol Cell 2014), as well references to Camahort et al. and Shang, Hori, et al. as suggested. In addition, we have included a paragraph in the Discussion regarding the questions raised by the reviewer. In particular we discuss the hypothesis that other centromeric proteins exist at low levels throughout the genome, possibly recruited by non-centromeric CENP-A, and ask how this may influence neocentromere formation. Finally, as suggested by the reviewer, we highlighted that the most novel finding regarding this issue is the surprisingly large proportion of CENP-A that is not centromere bound.

9) It would not detract from the import of the present work if the authors here mentioned the fact that their numbers are not terribly dissimilar from those estimated by Riberio et al. (PNAS), particularly given that the latter were looking at a different organism.

The estimate from Ribeiro et al. is stated as follows (direct quote): “A rough estimate of the amount of CENP-A in the fiber in Fig. 4A can be obtained from the number of localizations (a total of 123). Assuming that, on average, each Dronpa molecule switches approximately three to five times (31), the number of labeled CENP-A molecules is approximately 25 to 40. This value should be taken with caution, as Dronpa can switch as many as 170 times (28). For this estimation, we also assume that most CENP-A in the fibers is labeled with Dronpa and that the amount of endogenous CENP-A is not significant.

Given the fact that the estimate given in their study is more than an order of magnitude lower than the 400 CENP-A molecules per centromere that we find, that the measurements were made in a cell line from a different organism, and the admitted experimental limitations of their method, we do not feel that it is appropriate to cite this estimate as similar to ours in this section of the manuscript. Nevertheless, we appreciate the effort made by Ribeiro et al., and it is not our intention to disregard their work. Indeed, we do (and did already in the original submission) appropriately cite their CENP-A estimate when introducing the motivation of our own quantifications.

10) The statement “Previously, it was shown that CENP-A is interspersed with both H3.1 and H3.3 at the centromere (Blower et al. 2002; Dunleavy et al. 2011).” should also reference the Ribeiro et al. study.

Indeed, the Ribeiro et al. study shows that H3 is interspersed with CENP-A at centromeres. However, this finding was originally described in 2002 by Blower et al. Since then, there have been many studies that have confirmed and expanded on this finding, including the Ribeiro study, as well as e.g. Sullivan & Karpen, NSMB 2004 and Sullivan et al., Chrom Res 2011. For the sake of conciseness, we had previously chosen to only reference the original study demonstrating this, as well as Dunleavy et al., which was the first to show that both H3 variants mentioned exist at the centromere. For the sake of completeness, we have now added a reference to the Ribeiro et al. study, as well as the other studies mentioned above.

11) The authors go on in the same paragraph to say “Nevertheless, this non-centromeric pool only represents <0.1% of all nucleosomes in the genome and thus CENP-A is ∼50-fold enriched (per unit length of DNA) at centromeres (Figure 7a [currently Figure 8A]). This explains how, despite being outnumbered 25:1 by other H3 variants at the centromere, CENP-A can still accurately specify the centromeric locus.” Are they claiming to offer a convincing explanation of how the CENP-A accurately specifies the centromere? If they really believe this, perhaps they could expand and come up with a more directly testable model.

We have toned down our conclusion by saying that “it may explain” this. In addition, we have added a suggestion for a testable hypothesis (artificially tethering differential amounts of CENP-A).

12) Where the authors say “Indeed, it has recently been shown that this sequence specific DNA binding protein has a role in phasing CENP-A nucleosomes (Hasson et al. 2013)”, they might like to read, and possibly cite, the rather extensive discussion of this in Pluta et al. (JCB 116, 1081, pp. 1091, 1092). The discussion of the enigmatic CENP-B aside, I feel (as already stated above) that the reason for the low levels of CENP-A on the Y chromosome most likely arises from the small size of the alphoid array, which the authors should mention here as well.

We thank the reviewer for pointing out the rather interesting Pluta et al. paper, of which we were indeed not aware. We have included their hypothesis that CENP-B disrupts the 3D structure of centromeric chromatin in our Discussion. In addition, as outline above (point 7 of this reviewer), we have modified the text to reflect the concern raised regarding the size of the alphoid array.

13) The authors state “the fixed ratio between total and centromeric CENP-A levels prevents excess CENP-A from accumulating at high density at non-centromeric loci”. I believe that this is an overstatement of what the data actually show. This is still, in my mind, a hypothesis - not a rigorously proven fact, and should be stated as such.

We have modified the text to clarify that this is indeed a hypothesis

14) Lastly, I do not understand Figure 7b. What is the “critical size” and why is this bigger than the kinetochore?

In the model figure (currently Figure 8) we represented CENP-A chromatin as a region of which the size is independent of the downstream centromere complex and kinetochore (as based on data in Figure 6). We show that reduction in centromere CENP-A levels by half results in modest but measurable defects. This suggest that centromeric CENP-A is in excess and the critical amount of CENP-A is less than this, possible hovering around half the full amount or less. We aimed to reflect this finding in the model. We realized however, that “critical size” may not be the right term as changes in CENP-A quantity do not necessarily reflect centromere size (as measured as the part of the genome covered by CENP-A). We have now changed this to “critical quantity”

Reviewer 3 Comments:

The authors should discuss possible limitations of their findings, and be careful when making concluding statements. For example, the authors tried to correlate further reduction of CENP-A from the parental RPE line with increases in micronuclei. I am not convinced that the slightly increased frequency of micronuclei is due to a moderate decrease in CENP-A. They would need to show that restoring CENP-A suppresses the phenotype. It is especially a problem when the authors attempt to say that this amount of CENP-A represents the number in “normal” tissue. The results are meaningful in the background of the RPE cell line, but even it is a relevant number, it still has limited value since they have shown that CENP-A levels vary greatly among cell lines and there is no strict correlation with the level of genome instability. This suggests that other 'background' mutations probably contribute greatly to the level of genome instability, and that they only made a weak correlation between CENP-A reduction and increased micronuclei. Overall, these issues raise concerns about using micronucleus frequencies as a direct measure of the impact of altering CENP-A levels, and whether CENP-A reduction accounts for the observed phenotype. The authors determined the critical number based on the observations that HeLa cells with 33% CENP-A are not viable but RPE cells with 40% CENP-A are viable, so they reason that the critical number must lie somewhere in between these two numbers. However, these are completely different lines, with HeLa cells carrying hundreds or thousands of other mutations not in the RPE background. Viability is a combined readout of more than one related pathway. Viability of HeLa cells and RPE cells may have different sensitivity to the levels of CENP-A due to potential mutations in related pathways.

As outlined in the beginning of this response, we have toned down the conclusions that we make based on the micronucleus results. Furthermore, we agree with the reviewer that the percentages at which CENP-A become critical will vary between cell lines.

However, the objective of this section is to provide a discussion on the theoretical probabilities that centromere function can be lost due to stochastic loss of CENP-A. The percentages of CENP-A occupancy in HeLa and RPE cells are used to guide this discussion. Our main findings that 1) the loss rate of most CENP-A on any centromere is exceedingly low when starting with 200 nucleosomes and 2) that this chance increases dramatically with a smaller CENP-A domain size, still stand. To reflect the concern of the reviewer we have now modified the text to highlight that these numbers are estimates.

There are several places the authors cite improper literatures, please cite original studies. For example, in introduction, to make the point that CENP-A is so far the primary candidate for centromere identity, the authors need to cite papers beyond just studies on human CENP-A; in the next sentence, to show the point that CENP-A is stably transmitted through cell divisions, there are much more authoritative papers on humans and other organisms than the authors' own papers; the paper cited for primary DNA sequence being neither necessary nor sufficient for centromere determination is Amor et al. in 2004, which is not the best for making this point.

Original papers have now been included in the above mentioned sections of the paper. Although we have addressed the issue raised by citing more literature on different species, we feel that the relevance of these is limited since our study focusses exclusively on human CENP-A and our findings may not be relevant in all systems. Indeed, in some species (most prominently budding yeast), specific DNA sequences rather than CENP-A nucleosomes are the primary candidate for centromere identity. On the point of CENP-A stability, to our knowledge there are no studies in human cells, other than our own (Jansen et al, JCB 2007 and Bodor et al, MBoC 2013), that convincingly demonstrate stable transmission of CENP-A. Indeed, these two studies represent the original finding and most compelling evidence for this, respectively.

In several places the authors stated results as if they are original observations. However, some of the principles have been revealed or suggested previously, and are here reinforced by the measurements in this manuscript. For example, the correlation between reduction of CENP-A and increased genome instability (if indeed due to CENP-A reduction in the RPE background), and different amounts of CENP-A are at different centromeres. The authors should be careful and need to cite the literature properly.

We have now clarified in the text that the finding that CENP-A is important for centromere function is well established. However, our finding at which level CENP-A becomes critical is novel. Further, we are aware that others have observed chromosome specific levels for CENP-A which we highlighted and cited in the text in the original submission. For both the Y-centromere and the neocentromere on chromosome 4 we have in fact already cited all appropriate papers in our original submission. We hope this is now more clearly highlighted.

1) Local background correction for Figure 2 is problematic. The basis of doing local background correction is the assumption that signals immediately surrounding centromeres are noise. What is the factual foundation of this methodology considering that recent papers and the authors' own data indicated the existence of non-centromeric CENP-A in chromatin? If the local background is indeed extremely low, treating it as background noise may not distort the actual number significantly.

We do not understand the reviewer’s argument that we assume that the surrounding background is exclusively noise-derived. The aim in our experiments is to determine the centromeric levels of CENP-A only. Indeed, the reason that we decided to perform local background subtraction on a per centromere basis, rather than overall background correction for the entire cell, is to get rid of all potential sources of background fluorescence. This includes noise, as well as cellular autofluorescence, randomly distributed non-centromeric CENP-A nucleosomes, and soluble CENP-A.

2) Figure 1, to overcome the difficulty of different transfer efficiency for untagged or tagged CENP-A, the authors could also use known quantities of purified tagged and untagged CENP-A on the same gel to calibrate tagged CENP-A cell lines.

This solution would be helpful, however, unfortunately we do not have fluorescently tagged CENP-A at high purity available for this. In addition, we feel that this is not essential given that we are primarily concerned with wildtype CENP-A levels and use the fluorescently-tagged CENP-A cell lines as mere fluorescent reference lines. In addition, we verified these measurements with two additional independent methods that arrive at similar numbers. As such we feel that expanding the quantitative Westerns would be excessive.

3) What is the cause of the observed different levels of CENP-A for different constructs in Figure 1? Did the authors compare different clones? This may suggest that other mutations occurred during generation of these lines along with changes in CENP-A levels.

The reduction of CENP-A levels in CA+/- and CAG/- cells is likely due to the fact that these cell lines express CENP-A from a single allele, as was already mentioned in the original submission. However, it is indeed surprising that CENP-A levels are increased in the CAY/- cell line. Although we do not have a good explanation for this, we hypothesize that this is indeed due to random mutations or some other form of adaptation. We have added this hypothesis to the text. Importantly, since these cell lines are used as fluorescent reference cell lines, we can normalize for these differences and they do not impact on our conclusions.

4) Figure 1f, this experiment needs to be performed in the RPE lines not in HeLa. In addition, since transient CENP-A overexpression leads to incorporation into non-centromeric chromatin through a DAXX dependent pathway, this must represent a re-equilibrium of two different loading systems that utilize different assembly factors (HJURP and DAXX), and between centromeres and non-centromeric loci. Does DAXX mediated incorporation happen in the various RPE cells? The authors only considered canonical CENP-A loading at the centromere, but I don't think that it makes much sense to talk about mass action without considering the DAXX dependent assembly that may compete with HJURP for centromeric assembly. This will clearly influence the size of the available CENP-A pool for centromeric loading.

The experiment described in Figure 1f required synchronization of cells in S phase to ensure a single round of CENP-A incorporation. However, drugs that lead to an S phase arrest, such as thymidine can induce p53 dependent apoptosis in RPE cells. Furthermore, upon thymidine release a large proportion of surviving RPE cells are unable to exit the induced S phase arrest and thus synchrony is lost almost immediately. However, p53 deficient HeLa cells can be synchronized using thymidine and we thus opted to use these cells for this particular experiment. An explanation of this has been added to the text.

Very recently, it has been shown that stable overexpression of CENP-A in HeLa cells leads to DAXX mediated incorporation of CENP-A in non-centromeric chromatin (Lacoste et al., Mol Cell 2014). Indeed, we also find that a significant pool of CENP-A is incorporated into non-centromeric chromatin in RPE cells. While this is likely DAXX mediated in RPE cells as well, we do not have any evidence for this. Indeed, the non-centromeric assembly pathway of CENP-A lies well beyond the scope of our study. Nevertheless, the centromeric assembly is most likely still mediated by HJURP, as was observed in CENP-A overexpressing HeLa cells as well (Lacoste et al., 2014). The Lacoste et al study was published during the process of review of our manuscript, and thus no mention was made of it originally. We now cite this study in the text and throughout the Discussion.

Given that our data show a strong correlation of CENP-A protein expression levels and centromere incorporation (Figure 1d, e) and that the centromeric fraction is independent of CENP-A expression levels (Figure 2c), we feel that we provide sufficient evidence for a mass-action mechanism, irrespective of the specific pathway used to reach this equilibrium.

5) Figure 4b and 4c, the authors measured fluorescence associated with mitotic centromeres. They found that different centromeres behave differently in the same cell. Using data from multiple cells, they concluded that segregation of CENP-A nucleosomes is random. To conclude this, the authors should have compared the behavior of an individual pair between different cells. It is possible that segregation is non-random for a given centromere of the same chromosome, but varies between centromeres of different chromosomes. Combining data from all centromeres without separating individual chromosomes may give rise to the frequency graph in Figure 4c. In addition, the authors need to consider that any differences in the levels of CENP-A at different centromeres/sisters observed in mitosis could be counteracted during new CENP-A assembly in the subsequent G1 phase.

We appreciate the reviewer’s concern on this point regarding the current Figure 5. However, we would argue that it is unlikely that this is a major issue, as all of the following would need to be true simultaneously: 1) specific centromeres display a non-random CENP-A distribution; 2) the non-randomness is different for different centromeres; and 3) the different degrees of non-randomness between centromeres of different chromosomes add up to be Gaussian in their average distribution. The consistency of our data with random segregation is further emphasized by our finding that different specific centromeres display a similar variance across cells (see Figure 7). Finally, it would be technically challenging, if not impossible, to track sister centromeres of a specific chromosome for the purpose of ruling out this theoretical possibility.

6) The variation of numbers of CENP-A per centromere using different cell lines is over 2.5 fold, which is fairly large. What could be contributing to this variation?

The reviewer raises an interesting point here. Although we do not currently have any positive evidence regarding the cause of this variability, we have now excluded the possibility that it is cell cycle dependent (as outlined above and presented in current Figure 3). Although this would be challenging, if at all possible to test, we propose that this reflects inherent variation present between cells.

7) Figure 5 [current Figure 6], the micronuclei phenotype has very limited value in assessing centromere defects. It is actually problematic as the end phenotype for centromere dysfunction, since there could be defects in sister separation (cohesion), as well as genome instability/breakage. It is more direct and informative to look at chromosome segregation. Without additional experiments, e.g. live studies, the results and their implications are hard to interpret. The authors also need to evaluate more than one clone for each line.

This point is shared by the different reviewers and we agree that interpretation of these results is by necessity fairly limited. We have now extensively modified and subdued our original discussion on this point, as outlined above.

8) For the phenotype in Figure 5 [current Figure 6], how is it striking to see a correlation between reduced CENP-A and increased genome instability? More important, it is not known whether this is due to the problem of the tags or other potential mutations arising during generation of the cell lines. To conclude that the micronuclei phenotype is caused by reduction of CENP-A, the authors would have to do a rescue experiment to restore CENP-A to normal.

While loss of CENP-A is predictably leading to centromere failure, mitotic defects and ultimately cell death, our finding that a reduction to half the normal levels leads to defects is of interest because it represents the level at which the centromeric CENP-A pool becomes critical. In other words, we do not report here that CENP-A is important for mitosis (which is indeed well established), but we report at what stage it becomes critical. We have now focused our discussion on this point, which we believe is still a valid one, and have attenuated the interpretation of the mechanisms underlying the micronuclei formation.

9) Figure 6, the authors made a statement about the variance of CENP-A between untransformed cells and cancer cells based on few lines from each category. This statement should be removed. It is safe to say that variance in cancer cell lines is big (6-fold range), but I don't think the evidence is sufficient to say anything about untransformed cells.

We modified this statement regarding the current Figure 7, and simply made the point that the numbers we find in RPE cells are similar to those in primary cells.

10) In the Discussion, please rephrase “the fixed ratio between total and centromeric CENP-A levels prevents excess CENP-A from accumulating at high density at non-centromeric loci, thus further reducing the probability of neocentromere formation.” Fixed ratio is an observation, it should not be taken as a mechanism to explain how neo-centromeres are prevented.

As mentioned above, this sentence has been revised to state “may prevent” rather than “prevents” to indicate that this is a hypothesis.

11) Also the model figure shows widely spaced single CENPA nucleosomes and could easily be clustered more.

The reviewer is correct in this assessment, and this alternative possibility was added to the figure legend.

https://doi.org/10.7554/eLife.02137.020

Article and author information

Author details

  1. Dani L Bodor

    Instituto Gulbenkian de Ciência, Oeiras, Portugal
    Contribution
    DLB, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article
    Competing interests
    The authors declare that no competing interests exist.
  2. João F Mata

    Instituto Gulbenkian de Ciência, Oeiras, Portugal
    Contribution
    JFM, Contributed to construction and characterization of knockout/knockin cell lines
    Competing interests
    The authors declare that no competing interests exist.
  3. Mikhail Sergeev

    1. Department of Systems Biology, Harvard Medical School, Boston, United States
    2. Renal Division, Brigham and Women's Hospital, Boston, United States
    Contribution
    MS, Acquisition of fluorescence life time data
    Competing interests
    The authors declare that no competing interests exist.
  4. Ana Filipa David

    Instituto Gulbenkian de Ciência, Oeiras, Portugal
    Contribution
    AFD, Performed acute CENP-A/HJURP induction experiment
    Competing interests
    The authors declare that no competing interests exist.
  5. Kevan J Salimian

    Department of Biochemistry and Biophysics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, United States
    Contribution
    KJS, Analysis of sequencing data
    Competing interests
    The authors declare that no competing interests exist.
  6. Tanya Panchenko

    Department of Biochemistry and Biophysics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, United States
    Contribution
    TP, Analysis of sequencing data
    Competing interests
    The authors declare that no competing interests exist.
  7. Don W Cleveland

    1. Ludwig Institute for Cancer Research, University of California, San Diego, La Jolla, United States
    2. Department of Cellular and Molecular Medicine, University of California, San Diego, La Jolla, United States
    Contribution
    DWC, Contributed to the establishment of CENP-A floxed and knockout alleles
    Competing interests
    The authors declare that no competing interests exist.
  8. Ben E Black

    Department of Biochemistry and Biophysics, Perelman School of Medicine, University of Pennsylvania, Philadelphia, United States
    Contribution
    BEB, Analysis of sequencing data, Contributions to concept and design
    Competing interests
    The authors declare that no competing interests exist.
  9. Jagesh V Shah

    1. Department of Systems Biology, Harvard Medical School, Boston, United States
    2. Renal Division, Brigham and Women's Hospital, Boston, United States
    Contribution
    JVS, Analysis of fluorescence life time data, Contributions to concept and design
    Competing interests
    The authors declare that no competing interests exist.
  10. Lars ET Jansen

    Instituto Gulbenkian de Ciência, Oeiras, Portugal
    Contribution
    LETJ, Conception and design, Drafting or revising the article
    For correspondence
    ljansen@igc.gulbenkian.pt
    Competing interests
    The authors declare that no competing interests exist.

Funding

European Molecular Biology Organization (EMBO Installation grant)

  • Lars ET Jansen

European Commission (FP7 Marie Curie Reintegration grant)

  • Lars ET Jansen

European Research Council (ERC-2013-CoG-615638)

  • Lars ET Jansen

National Institutes of Health (GM082989)

  • Ben E Black

National Institutes of Health (GM077238)

  • Jagesh V Shah

Burroughs Wellcome Fund

  • Ben E Black

Rita Allen Foundation

  • Ben E Black

Beckman Laser Institute and Foundation

  • Jagesh V Shah

Fundação para a Ciência e a Tecnologia (Foundation for Science and Technology) (SFRH/BD/74284/2010)

  • Dani L Bodor

Fundação para a Ciência e a Tecnologia (Foundation for Science and Technology) (BIA-BCM/100557/2008, BIA-PRO/100537/2008)

  • Lars ET Jansen

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Tatsuo Fukagawa (National Institute of Genetics, Shizuoka, Japan), Dan Foltz (University of Virginia, Charlottesville, VA), Kevin Sullivan (National University of Ireland, Galway, Ireland), David Livingston (Dana-Farber Cancer Institute, Boston, MA), Bernardo Orr, and Duane Compton (Dartmouth Medical School, Hanover, NH), and Kerry Bloom (University of North Carolina, Chapel Hill, NC) for reagents, Nitzan Rosenfeld (Cancer Research UK, Cambridge, UK) for advice, and Jorge Carneiro (Instituto Gulbenkian de Ciência, Oeiras, Portugal) for help using R. We thank the Confocal and Light Microscopy core facility at Dana Farber Cancer Institute (Harvard Medical School) for providing access to the FLIM setup. We are grateful to Alekos Athanasiadis and Monica Bettencourt-Dias (both at Instituto Gulbenkian de Ciência, Oeiras, Portugal) for helpful comments on the manuscript.

Reviewing Editor

  1. Jon Pines, Reviewing Editor, The Gurdon Institute, United Kingdom

Publication history

  1. Received: January 10, 2014
  2. Accepted: June 17, 2014
  3. Version of Record published: July 15, 2014 (version 1)

Copyright

© 2014, Bodor et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 2,972
    Page views
  • 252
    Downloads
  • 51
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Comments

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Biochemistry
    2. Biophysics and Structural Biology
    Lindsay D Clark et al.
    Research Article Updated