Perfect and Imperfect Nucleosome Positioning in Yeast

Abstract

Numerous studies of nucleosome positioning have shown that nucleosomes almost invariably adopt one of several alternative overlapping positions on a short DNA fragment in vitro. We define such a set of overlapping positions as a "position cluster", and the 5S RNA gene positioning sequence is presented as an example. The notable exception is the synthetic 601-sequence, which can position a nucleosome perfectly in vitro, though not in vivo. Many years ago, we demonstrated that nucleosome position clusters are present on the CUP1 and HIS3 genes in native yeast chromatin. Recently, using genome-wide paired-end sequencing of nucleosomes, we have shown that position clusters are the general rule in yeast chromatin, not the exception. We argue that, within a cell population, one of several alternative nucleosomal arrays is formed on each gene. We show how position clusters and alternative arrays can give rise to typical nucleosome occupancy profiles, and that position clusters are disrupted by transcriptional activation. The centromeric nucleosome is a rare example of perfect positioning in vivo. It is, however, a special case, since it contains the centromeric histone H3 variant instead of normal H3. Perfect positioning might be due to centromeric sequence-specific DNA binding proteins. Finally, we point out that the existence of position clusters implies that the putative nucleosome code is degenerate. We suggest that degeneracy might be a crucial point in the debate concerning the code.

1. Definition of nucleosome position

The nucleosome core contains ~147 bp of DNA wrapped around a central core histone octamer, composed of two molecules each of the four core histones - H2A, H2B, H3 and H4 [1]. In vivo, nucleosomes are formed into regularly spaced arrays. Nucleosome core particles can be isolated by careful digestion with micrococcal nuclease (MNase), by virtue of the protection of 145–150 bp by the histone octamer. We have discussed this in detail previously [2].

It is worth defining what we mean by nucleosome position, since there has been some debate [3, 4]. We define the position of a nucleosome by the ~147 bp sequence occupied by the histone octamer [5]. It follows that all nucleosomes have a position with respect to the DNA sequence. A useful concept is to imagine each chromosome as a series of overlapping 147 bp windows (i.e. positions), 1 bp apart, each of which might be bound by the histone octamer. Thus, there are (n - 146) potential positions on a linear DNA molecule, where 'n' is the number of base pairs. The important question is, which of these potential positions are actually occupied by the octamer, and to what extent?

This definition of position is clear to those groups who do experiments on nucleosomes in vitro, because the traditional methods available for mapping nucleosomes in vitro have sufficient resolution to resolve overlapping positions. In general, several overlapping positions with a range of occupancies are observed. These overlapping positions are often rotationally related (i.e., they differ from one another by multiples of ~10 bp, such that the rotational orientation of the DNA bent around the histone octamer is preserved). Nucleosomes at different positions on the same DNA fragment can be resolved in native gels, because their mobility depends on how close the nucleosome is to the centre of the DNA molecule. Their precise positions can be measured by making core particles using MNase and analysing the DNA by restriction mapping or sequencing. There are almost no examples of nucleosomes which adopt exactly the same position on every DNA molecule in vitro. The notable exception is the synthetic 601-sequence [6].

In contrast, the traditional method for mapping nucleosomes in vivo, indirect end-labelling, is a low resolution method that cannot resolve overlapping positions. The method was generally employed because the high resolution methods used in vitro could not be applied in vivo for technical reasons. Consequently, positioning in vivo has been defined more loosely than in vitro, leading to some disagreement about what position means. The chromatin structure deduced by indirect end-labelling is a simplified version of the structure, reflecting occupancy rather than precise positions. See [2] for a more detailed discussion.

2. Perfect and imperfect nucleosome positioning in vitro

Many years ago it was discovered that a nucleosome preferentially occupies a specific position with respect to the 5S RNA gene in vitro [7]. Although a dominant position is dictated by the 5S gene sequence, it was appreciated early on that several alternative positions are possible, which overlap the major position (Fig. 1A) [8–10]. The extent to which the major position is dominant depends on the length of the DNA fragment and on the nature of the flanking sequences, both of which influence the degree of competition with the strong 5S position for the histone octamer during reconstitution. In contrast, the synthetic 601 sequence, discovered by Jon Widom and colleagues using a selection technique [6], is so strong that it out-competes all alternative positions for the histone octamer, resulting in a unique nucleosome, formed at the same position on every 601-DNA fragment. That is, the 601 nucleosome, unlike the 5S-nucleosome, is perfectly positioned (Fig. 1B). Consequently, most laboratories have switched from the 5S sequence to 601 for studies in vitro. However, there are very few natural sequences with the positioning strength of 601 [11]. When reconstituted in vivo, they almost always give rise to multiple overlapping positions.

Fig. 1.

Perfect and imperfect nucleosome positioning in vitro: the 601 and 5S rRNA gene positioning sequences. Position maps (top panels), dyad position plots (middle panels) and occupancy plots (bottom panels). The dyad position plot indicates the position of each nucleosomal dyad axis and the relative amounts of each nucleosome. The occupancy plot is the sum of the occupancies of all individual nucleosome positions, each indicated by a "square" peak of nucleosome-length with a height proportional to the amount of that nucleosome. (A) The Xenopus borealis somatic 5S gene. The major position is indicated by the grey oval; all positions overlapping the major position are shown, with their proportions (%) (adapted from [9]). The combined occupancy plot has been smoothed. (B) The synthetic 601 sequence. A single position is observed.

Perfect positioning by the 601-nucleosome is characterised by a single, isolated peak in the dyad position plot, with no other peaks within 150 bp on either side of that peak (i.e., no overlapping positions), and by a "square" occupancy peak (Fig. 1B). Nucleosome occupancy may be defined as the probability of a particular base pair being present in a nucleosome. In the case of the 601 nucleosome, the occupancy is 1 for all base pairs within the 601 sequence and zero for all base pairs outside it, resulting in a squared-off occupancy peak covering the 601 sequence. In contrast, the occupancy peak for the 5S nucleosome (Fig. 1A) is the sum of the proportional contributions of the square peaks representing each of the five overlapping positions shown, resulting in a rounded peak with an irregular shape. We will refer to a set of overlapping nucleosome positions such as this as a "position cluster". The histone octamer must select one of the positions within the cluster. The population of reconstituted nucleosomes will reflect the relative affinities of each position for the octamer; the position with the highest affinity will have the highest occupancy. The distribution of nucleosomes amongst these positions may depend on the salt concentration, temperature, and on whether equilibrium was attained during reconstitution. The occupancy profile of a position cluster reflects the weighted average of all overlapping positions. Thus, the occupancy peak indicates an average position and obscures the underlying position complexity.

3. DNA sequence plays a major role in determining nucleosome positions in yeast CUP1 chromatin in vivo

A number of years ago, we obtained a high resolution nucleosome map for native yeast CUP1 chromatin using the monomer extension technique [12]. This technique requires purified chromatin, otherwise the background is too high. We devised a method to purify plasmid chromatin containing CUP1 from cells in which the gene is inactive, or induced by copper. The DNA isolated from nucleosome core particles was mapped, revealing several position clusters on the inactive CUP1 gene, separated by short nucleosome-depleted regions. Activation coincides with the appearance of additional position clusters over the nucleosome-depleted regions (Fig. 2B). That is, positions between clusters are occupied when the gene is activated.

Fig. 2.

DNA sequence plays a major role in determining nucleosome positions in yeast CUP1 chromatin in vivo. (A) Nucleosome positions adopted on transcriptionally active CUP1 in vivo. Shaded ovals indicate positions also observed in vitro. (B) All of the nucleosomes in (A) are superimposed to visualise position clusters on copper-induced CUP1, and on inactive CUP1 isolated from cells lacking the transcriptional activator. Only a subset of the nucleosomes present on the activated gene are present on the inactive gene. Adapted from [12,13].

A comparison of nucleosome positions in native CUP1 chromatin with those adopted in reconstituted CUP1 chromatin shows that most of the positions are the same [13] (Fig. 2A). However, the occupancies of these positions (i.e., the relative amounts of each nucleosome) are quite different, with some strong positions in vitro being only weakly occupied in native chromatin, and vice versa. Thus, DNA sequence plays a major role in determining nucleosome positions in CUP1 chromatin in vivo, but the relative affinity of each position for the octamer is not the only determinant of occupancy.

The nucleosome is intrinsically stable in vitro, because the energy barrier against spontaneous movement under physiological conditions is very high for most sequences. However, in vivo, nucleosome re-positioning is catalysed by ATP-dependent remodelers. Position occupancies probably depend both on intrinsic affinity and on the activities of ATP-dependent remodeling complexes [13]. Thus, a nucleosome can be moved by a remodeling complex to an energetically less favourable position and become marooned there.

4. The 601-nucleosome in vivo

The 601-nucleosome is a case in point, since it is perfectly positioned when reconstituted in vitro, but not in vivo. Overlapping positions are detected on the 603-sequence (which is closely related to 601) when it is inserted into the 5'-untranslated region of CUP1, in both non-induced and induced yeast cells [14]. Similar observations were reported for 601 in yeast [15]. In mouse liver cells carrying episomal 601, the 601-nucleosome is very well positioned early in development, with a dominant position and some closely related overlapping positions [16]. Later in development, the 601-nucleosome is present but its occupancy is much reduced [16]. There is also some evidence that 601 strongly positions a nucleosome in human cells transfected with a 601-containing plasmid [17], although the mapping method used measures occupancy and cannot distinguish between 601 and overlapping positions.

So why doesn't the 601-sequence position nucleosomes perfectly in vivo? Firstly, since ATP-dependent remodelers are capable of moving the 601-nucleosome to lower affinity positions in vitro [14], it is likely that they can do so in vivo, again arguing that remodelers can use ATP to catalyse the formation of thermodynamically less stable chromatin structures. Secondly, 601 does not position well under certain conditions in vitro [18] which may be relevant if physiological conditions are more akin to these reconstitution conditions than those typically employed. Better positioning on 601 might be expected in chromatin where DNA sequence is the deciding factor in positioning, perhaps where there are no active remodelers, as discussed above.

5. Nucleosome energetics and ATP-dependent remodelling machines

The analysis above implies that nucleosome positions in vivo cannot be predicted simply by calculating the energetics of nucleosome formation. Since different ATP-dependent remodeling activities are targeted to specific chromatin regions at different times (e.g. when a gene is activated or repressed), their activities will be transient and localised. Nucleosomes may be moved to less favourable positions in some chromatin regions, but not in others. It may be concluded that the distribution of nucleosomes on DNA in vivo is not at equilibrium. Thus, non-equilibrium thermodynamics might be more appropriate for understanding nucleosome positioning, at least where ATP-dependent remodelers are active.

Another important difference between native and reconstituted chromatin is that nucleosomes are regularly spaced in vivo, presumably by ATP-dependent remodelling complexes (e.g. ISW1 and ISW2 [19]). In yeast, the average spacing is ~165 bp, corresponding to a nucleosome core of 145–150 bp and a linker DNA of 15–20 bp. This places a lot of constraint on the positioning information in the DNA, since a nucleosomal array dictated by affinity for DNA sequence could be formed only if high affinity positions are encoded with the required spacing.

6. Nucleosome position clusters are present genome-wide

We have also observed position clusters on the HIS3 gene [20]. How general are position clusters? To address this question, we used paired-end sequencing to analyse nucleosome positions genome-wide [21]. The advantage of paired-end sequencing is that it provides the length of each DNA fragment, which allows us to sort the properly trimmed nucleosome cores from those that are poorly trimmed or over-digested by MNase. These data provide accurate nucleosome positions and make it possible to distinguish closely related overlapping positions. We found position clusters on all genes examined, with very few examples of perfect positioning. These position clusters usually contain a dominant position (i.e., a position with high occupancy), flanked by weaker positions (lower occupancies), similar to that for the 5S gene in vitro (Fig. 1A). A position cluster organisation was also observed [21] at the well-studied PHO5 promoter [22] and at ARS1 [23].

The effect of induction on nucleosome positioning can be dramatic, as in the case of the ARG1 gene (Fig. 3). In the inactive state, the occupancy profile for ARG1 shows a fairly regular set of rounded peaks across the entire gene (Fig. 3A), corresponding to a series of position clusters, most of which have a dominant position flanked by weaker positions (Fig. 3B). This is also true of the gene downstream (YOL057W). The gene upstream, GPD2, has a much less regular occupancy profile, corresponding to position clusters containing a less dominant major position. Activation of ARG1 results in a dramatic loss of nucleosome occupancy over the entire coding region, which extends into YOL057W downstream (Fig. 3A). This corresponds to a major disruption of the position clusters on ARG1 and the 5' half of YOL057W, involving the loss of the dominant positions (Fig. 3C). Thus, activation of ARG1 results in a loss of canonical nucleosomes over the entire gene, extending into flanking sequences [21]. A genome-wide analysis indicates that ~50 genes show a similar loss of nucleosome occupancy over the coding region in response to activation [21]. A similar effect has been observed when the Drosophila Hsp70 loci are activated [24] (see [21] for a more detailed discussion of the effects of activation on nucleosome positioning).

Fig. 3.

Nucleosome position clusters formed on ARG1 are disrupted when the gene is activated. (A) Occupancy plot for ARG1 and flanking genes. Binding sites for the transcriptional activator (Gcn4) are indicated (note the nucleosome-depleted regions at the ARG1 and YOL057W promoters). Control (red line); activated (green line); the oval indicates the size of a nucleosome. (B) Dyad position plot for control cells. Note the position cluster organisation of the ARG1 gene; some clusters have dominant positions which account for about half of the occupancy peak (e.g., the first two nucleosomes downstream of the nucleosome-depleted promoter), but the dominant positions in most of the other clusters account for well under half of their occupancy peaks. Thus, the alternative positions account for a significant fraction of the occupancy in most cases. (C) Dyad position plot for activated cells. Adapted from [21].

7. Perfect positioning of yeast centromeric nucleosomes in vivo

Our genome-wide data also revealed an example of perfect positioning in vivo: the centromeric nucleosome [25]. A single, perfectly positioned, nucleosome is located directly over each centromere (CEN), demarcated by an obvious square peak (Fig. 4A). This nucleosome is somewhat smaller than the canonical nucleosome. The centromeric nucleosome contains the centromeric H3 variant (CenH3 or Cse4) instead of normal H3, but its composition is otherwise controversial, with three different models proposed (Fig. 4B) (see [25] for further discussion). Although the CEN nucleosome appears to be an exception to the position cluster rule, it is actually a special case because of its altered composition. We suggest that its perfect positioning might be due to the DNA sequence-specific centromeric factors, Cbf1 and CBF3.

Fig. 4.

Perfect positioning of CEN nucleosomes in vivo. (A) Occupancy plot for CEN15 and flanking genes. Control (red line); activated (green line). Note that activation is not expected to have any effect on CEN chromatin structure. (B) The CEN nucleosome is shown positioned according to our data: it covers the entire centromere (white box) which is divided into three regions: CDEI, II and III. The sequence-specific factors Cbf1 and CBF3 bind to CDEI and CDEIII, respectively. They might be integral components of the CEN nucleosome or they might bind on the outside of the nucleosome (adapted from [25]).

8. Position clusters and a degenerate nucleosome code

It has been suggested, with good evidence, that yeast genomic DNA contains a nucleosome positioning code [26,27]. That is, genomic DNA contains information specifying where nucleosomes should be formed. There is some debate over whether or not it is technically a code [3], but we shall use the term here to mean only that DNA contains nucleosome positioning information, which is clearly the case. Our early data support this hypothesis [12,13]. There is more controversy over the relative importance of DNA sequence in determining positioning [2–4,28,29]. Other important contributors include sequence-specific factors, ATP-dependent remodelers, and the transcription and replication machineries [30,31]. As discussed above, nucleosomes in vivo may not be positioned as predicted by simple equilibrium thermodynamics, because ATP-dependent remodelling activities can move nucleosomes to less favourable positions. Thus, occupancies will not necessarily be the same in vitro and in vivo.

Clearly, cells growing under different conditions have different chromatin structures at differentially expressed genes. In addition, cells within a population will differ transiently in their chromatin structures as they transcribe the same genes asynchronously. The genome sequence is unaffected by any of these processes and therefore neither is the positioning information constituting the proposed code. A simple version of the code hypothesis is therefore untenable, since it predicts that nucleosome positioning will be invariant.

However, we believe that the crucial feature of the nucleosome positioning code is that it is degenerate: the code does not specify unique positions, but multiple overlapping positions. We propose that these alternative positions are made use of by the ATP-dependent remodelers when they move nucleosomes, either during gene activation or repression, or to facilitate the organization of nucleosomes into appropriately spaced arrays. Position clusters probably reflect the presence of populations of cells having different nucleosome arrays. We speculate that cells might switch from one array to another as events occur in the chromatin (Fig. 5) [2,20,21].

Fig. 5.

Hypothetical model showing how overlapping nucleosomal arrays might account for position clusters. (A) Five alternative arrays of 5 nucleosomes on a gene, all with 165 bp spacing to account for the repeat length of yeast chromatin. In this example, 45% of cells have the dominant array (black ovals), 25% of cells have an array shifted 20 bp downstream relative to the dominant array, 10% of cells have an array shifted 40 bp downstream, 15% of cells have an array shifted upstream by 20 bp, and the remaining 5% have an array shifted 40 bp upstream. (B) Occupancy plots for the arrays in (A). Each array gives rise to five square peaks with a height proportional to the fraction of cells having this array. The sum of the five occupancy plots (thick grey line) is smoothed (thick black line), resulting in a series of rounded peaks similar to those observed in vivo (e.g. Fig. 3A). (C) Dyad position plot for all of the nucleosomes in (A), showing position clusters. Alternative arrays might be dynamically related, with cells continuously switching nucleosomes from one array to another. Adapted from [2,21].

Highlights.

• >A canonical nucleosome chooses one of several alternative overlapping positions.
• >Such "position clusters" are observed both in vitro and in vivo, genome-wide.
• >Nucleosome position clusters are disrupted on transcriptionally active genes.
• >Perfect positioning is rare. The centromeric nucleosome is an example.
• >The nucleosome code is degenerate, specifying alternative nucleosomal arrays.