Genomic Analysis of the DNA Replication Timing Program during Mitotic S Phase in Maize (Zea mays) Root Tips

The time during S phase at which different maize DNA sequences replicate reveals a complex temporal program influenced by genomic features, transcriptional activity, and chromatin structure.

Abstract

All plants and animals must replicate their DNA, using a regulated process to ensure that their genomes are completely and accurately replicated. DNA replication timing programs have been extensively studied in yeast and animal systems, but much less is known about the replication programs of plants. We report a novel adaptation of the “Repli-seq” assay for use in intact root tips of maize (Zea mays) that includes several different cell lineages and present whole-genome replication timing profiles from cells in early, mid, and late S phase of the mitotic cell cycle. Maize root tips have a complex replication timing program, including regions of distinct early, mid, and late S replication that each constitute between 20 and 24% of the genome, as well as other loci corresponding to ∼32% of the genome that exhibit replication activity in two different time windows. Analyses of genomic, transcriptional, and chromatin features of the euchromatic portion of the maize genome provide evidence for a gradient of early replicating, open chromatin that transitions gradually to less open and less transcriptionally active chromatin replicating in mid S phase. Our genomic level analysis also demonstrated that the centromere core replicates in mid S, before heavily compacted classical heterochromatin, including pericentromeres and knobs, which replicate during late S phase.

INTRODUCTION

DNA replication enables a cell to faithfully transmit genetic material to daughter cells while maintaining genome integrity. During each S phase, there must be a complete and accurate duplication of the genome, which must occur in a regulated and reproducible way from one cell cycle to the next (Klein and Gilbert, 2016). To accomplish this task, higher eukaryotes organize the process both temporally and spatially so that replication initiates at multiple loci, or origins, distributed throughout the genome, with different origins becoming active at different times during S phase (Masai et al., 2010). In most higher eukaryotes, there is an association between early replication and euchromatic, transcriptionally active chromatin (Hatton et al., 1988; Hiratani et al., 2008; Lee et al., 2010; Ryba et al., 2010), though there are classes of genes that have a much weaker association with replication time and are developmentally regulated in humans and mice (Hiratani et al., 2008; Rivera-Mulia et al., 2015). Conversely, there is also a long established association between late replication and classical heterochromatin (Lima-de-Faria and Jaworska, 1968; Pryor et al., 1980), although there are a few exceptions (Kim et al., 2003). However, the intervening time between what is considered “early” and “late” has not been characterized thoroughly by whole-genome replication timing analyses, and replication times in this portion of S phase are only inferred from the ratio of signals obtained from early versus late in many studies. Some information is available about mid replicating chromatin from cytological studies. For example, in mammalian cell lines, mid replicating loci are spatially localized to the perinuclear and perinucleolar edges, while early replication occurs in dispersed, punctate foci (Dimitrova and Berezney, 2002; Panning and Gilbert, 2005; Zink, 2006). Such spatiotemporal differences are less pronounced in plants, in which both early and mid replicating loci have been reported to be dispersed throughout the nucleoplasm (Samaniego et al., 2002; Bass et al., 2015).

Replication in mid S phase has also been indirectly discussed in the context of timing transition regions (TTRs), which connect adjoining earlier and later constant timing regions in mammals (Hiratani et al., 2008; Desprat et al., 2009; Ryba et al., 2010). Some have hypothesized that these TTRs consist of unidirectional replication forks spreading from origins active earlier in S phase (Hiratani et al., 2008; Ryba et al., 2010), while others have argued that such regions contain origins activated in a “cascading” or “domino” pattern by earlier firing origins nearby (Guilbaud et al., 2011). There has been much less debate over the patterns of activation of origins firing within the constant timing region domains, as these are thought to activate roughly synchronously in clusters (Jackson and Pombo, 1998; Blow et al., 2011; Boulos et al., 2015; Klein and Gilbert, 2016).

Most studies in mammalian cells agree that much of the replication occurs in large coordinated domains (0.2–2 Mb) that presumably contain multiple origins (Farkash-Amar et al., 2008; Guilbaud et al., 2011; Hyrien, 2016; Klein and Gilbert, 2016). The size of replication domains in mammals is further supported by results from chromatin capture experiments that have defined topologically associated domains. Topologically associated domains align with the early boundaries of TTRs that delineate replication timing domains (Pope et al., 2014). However, in species with smaller genomes, such as Drosophila melanogaster and Arabidopsis thaliana, replication domains are smaller, with a mean domain size in the range of ∼30 to 450 kb (MacAlpine et al., 2004; Schwaiger et al., 2009; Lee et al., 2010).

Although plant and animal lineages diverged 1.6 billion years ago, the mechanisms and core machinery used to duplicate DNA have been largely preserved (Shultz et al., 2007; DePamphilis and Bell, 2010). However, there has been little attention focused on understanding DNA replication programs in plants (Lee et al., 2010; Costas et al., 2011a). Early literature using tritiated thymidine autoradiography indicated that plant-specific environmental and developmental events can impact the number and spacing of replication origins (reviewed in Bryant and Aves, 2011). Recent work revealing differences between transcriptional regulation strategies (Hetzel et al., 2016) also illustrates how fundamental processes, often assumed to be the same in all eukaryotes, may differ in important aspects between plants and animals.

In plants, new organs continue to form throughout an individual’s lifespan, providing a unique and useful model for investigating DNA replication dynamics in both space and time (Costas et al., 2011a). Early research on plant DNA replication focused mostly on the duration of S phase, the rate of fork movement, the size of individual replicons, and the number of replicon families that initiate replication coordinately. These studies were largely based on cell cycle kinetics and DNA fiber autoradiography experiments primarily using root tip meristem cells from various plant species (Van’t Hof, 1976; Van’t Hof et al., 1978a, 1978b; Van’t Hof and Bjerknes, 1981; Kidd et al., 1987). More recently, studies also characterized plant homologs of proteins involved in the prereplicative complex (Kimura et al., 2000; Castellano et al., 2001; Ramos et al., 2001; Witmer et al., 2003; Masuda et al., 2004; Mori et al., 2005; Shultz et al., 2007). However, genomic analyses of DNA replication and replication origins in plants have been sparse and thus far have focused exclusively on Arabidopsis. Replication timing has been profiled for a single chromosome (Lee et al., 2010), and efforts have been made to identify initiation zones or origins of replication (Lee et al., 2010; Costas et al., 2011b).

Because of its small genome size and relative paucity of repeats, it is reasonable to suppose that DNA replication in Arabidopsis may be simpler than for the larger, more complex genomes of crop plants. Maize (Zea mays) has a moderately large (2.3 Gb), fully sequenced genome (Schnable et al., 2009). The complexity of this genome is similar to that of many other crop plants, with a high concentration of repetitive sequences located in heterochromatic blocks and extensively interspersed in the euchromatic portion of the genome. Together with its illustrious history as a model species for genetic and cytogenetic studies, these features make maize an ideal model crop system for investigating the replication timing program.

We developed a system (reviewed in Bass et al., 2014) to examine the spatial, temporal, and genomic patterns of DNA replication in maize, using rapidly cycling cells from root tips pulse-labeled with the thymidine analog 5-ethynyl-2’-deoxyuridine (EdU). EdU offers a substantial improvement over classical replication timing assays, which have typically used 5-bromo-2’-deoxyuridine (BrdU) to label and immunoprecipitate newly replicated DNA (Hiratani et al., 2008; Schwaiger et al., 2009; Chen et al., 2010; Hansen et al., 2010; Lee et al., 2010; Ryba et al., 2011). A key benefit of using EdU is that a heat or acid denaturation step is not required for immunoprecipitation (reviewed in Darzynkiewicz et al., 2011). Hence, it is possible to visualize EdU-labeled DNA while maintaining native subnuclear structure (Kotogany et al., 2010; Bass et al., 2014) and to separate labeled from unlabeled nuclei by flow cytometry prior to DNA isolation (Wear et al., 2016). We previously used the EdU labeling system to investigate the spatio-temporal aspects of DNA replication in proliferating root tip cells (Bass et al., 2015). In that study, we used fluorescence microscopy to show that loci replicating in early and mid S phase are in close proximity to each other but have limited spatial overlap. This observation led us to hypothesize that there are at least two “euchromatic compartments” in replicating nuclei. We envisioned that the early replicating compartment is largely comprised of extended, transcriptionally active, accessible chromatin, while chromatin replicating in mid S phase includes locally compacted repetitive blocks and silent genes (Bass et al., 2015).

In this study, we extend our previous work to include a fine-scale genomic analysis of DNA replication, gene expression, and chromatin accessibility to add detailed molecular evidence to the initial cytological observations. Our study represents the first to characterize a replication timing program in a crop species, and the first whole-genome analysis in any plant species. The root tip system allowed us to work with intact meristems, avoiding potential chromosome aberrations, mutations, aneuploidization, and other genetic and chromatin-related changes that have been documented in both plant and animal cell cultures (Lee, 1988; Phillips et al., 1994; Serrano and Blasco, 2001; Maitra et al., 2005; Tanurdzic et al., 2008; Mayshar et al., 2010; Laurent et al., 2011).

We present whole-genome profiles of replication activity in early, mid, and late S phase and companion data sets for transcription, and a selection of histone marks. We found that maize root tips have a complex replication timing program, including regions of distinct early, mid, and late S replication that each constitute between 20 and 24% of the genome, as well as loci accounting for another ∼32% of the genome that exhibit replication activity in two different time windows, such as early and mid or mid and late. Analyses of genomic, transcriptional, and chromatin features of the euchromatic portion of the maize genome provide evidence for a gradient of early replicating, open chromatin that transitions gradually into less open and less transcriptionally active chromatin replicating in mid S phase. Our results also confirm previous observations in maize nuclei that heavily compacted classical heterochromatin replicates late in S phase (Pryor et al., 1980; Bass et al., 2015).

RESULTS

Whole-Genome Profiling of DNA Replication Timing in Maize Root Tips

A flowchart of the experimental steps used to generate replication timing by sequencing or “Repli-seq” data for maize root tips is detailed in Figure 1A. We pulse labeled 3-d-old B73 seedling roots in vivo with the thymidine analog, EdU, for 20 min to label DNA that was replicated during the pulse period. Over the 20-min labeling period, EdU uniformly penetrates the root and labels cells in the multiple emerging cell lineages present in the meristematic region (Figure 1B). Under the same growth conditions, the average length of S phase of mitotic cells in the terminal 1-mm region of the root was estimated to be between 2.7 and 3.9 h (Mickelson-Young et al., 2016). Hence, a 20-min labeling period represents ∼10% of the length of S phase. The terminal 1-mm root segments (Figure 1B) were excised, fixed, and frozen before isolating nuclei for flow cytometric analysis (Figure 1C; Supplemental Figure 1A). Most of the nuclei from the terminal 1-mm region had DNA contents ranging from 2C to 4C, characteristic of cells undergoing a mitotic cell cycle (Baluska, 1990). Approximately 20% of the nuclei had DNA contents above 4C, indicative of some cells transitioning into a programmed endocycle (Figure 1C; Bass et al., 2014, 2015). In 1-mm root tips, nuclei from endocycling cells can be excluded based on DNA content as shown in Figure 1C and are not considered further in this article.

Figure 1.

Experimental Approach.

(A) Workflow. Roots of 3-d-old maize seedlings were pulse labeled with EdU for 20 min, after which terminal 1-mm segments were harvested and fixed with formaldehyde.

(B) A merged confocal image of a 1-mm root tip longitudinal section showing DAPI stained DNA (red) and EdU label in newly replicated DNA (green). There are multiple emerging cell lineages present in the terminal 1 mm of the root. Bar = 100 μm.

(C) Sorting. Nuclei were isolated and EdU incorporated into DNA was conjugated to a fluorescent probe (AF-488) using click chemistry. Nuclei were counterstained with DAPI prior to sorting by flow cytometry using 355-nm (UV) and 488-nm (blue) lasers. A bivariate plot of relative DNA content (DAPI fluorescence) and EdU incorporation (AF-488 fluorescence) is shown, overlaid with the gates (black rectangles) used to sort nuclei representing early (E), mid (M), and late (L) fractions of S phase. Unlabeled nuclei from G1 phase (G1) were also sorted to use as a reference.

(D) Histogram showing relative DNA content (DAPI) for the unsorted nuclei population (black line), overlaid with the position and relative frequency of nuclei that fall in the indicated sorting gates. DNA was extracted from sorted nuclei and EdU/AF-488-labeled DNA immunoprecipitated from the early, mid, and late fractions with an AF-488 antibody, prior to sequencing on the Illumina HiSeq 2000 platform.

(E) to (H) Summary of computational processing of Repli-seq reads.

(E) and (F) The number of reads that mapped uniquely to the maize B73 AGPv3 reference genome was calculated over 1-kb windows (see Methods).

(G) After normalization for sequencing depth, replication activity was expressed as the ratio of EdU/AF-488 reads in early, mid, or late S phase to reads from total DNA from unlabeled G1 nuclei.

(H) The resulting data smoothed with a Haar wavelet function. Representative data tracks from IGV are shown here for early S data, and the corresponding genomic region is shown for early, mid, and late S data in Supplemental Figure 3. Artificial spikes in sequencing coverage (arrowheads) often correspond to tandem repeat regions that have been “collapsed” in the reference assembly, and these regions are subsequently excluded. Scale: E and F, 0 to 1200 read density; G and H, 0 to 5.4 normalized signal ratio.

Maize root tips yield relatively pure nuclei preparations that are amenable to flow sorting (Figure 1C; Supplemental Figure 1A). We developed a protocol for bulk nuclei isolation from root tips excised from ∼500 seedlings (Wear et al., 2016) that is a modification of the original tissue chopping method described by Galbraith et al. (1983). We visualized EdU-labeled nuclei by conjugating an Alexa Fluor-488 (AF-488) fluorophore to EdU using “click chemistry” (Salic and Mitchison, 2008). The EdU-labeled nuclei were then separated from unlabeled nuclei by flow cytometry and sorted into populations representing early, mid, and late stages of the mitotic S phase based on DNA content (rectangle areas in Figure 1C). We intentionally left space between the early, mid, and late sorting gates to reduce the amount of overlap in the sorted populations. Unlabeled G1 nuclei were also sorted to provide an unreplicated DNA reference for subsequent analysis. In Figure 1D, the range of DNA contents for the early, mid, and late S sorting gates are overlaid on the DNA content histogram for total nuclei to demonstrate that nuclei in early and late S cannot be resolved from nuclei in G1 and G2/mitosis on the basis of DNA content alone. Separation of these populations can only be achieved using a two-color sorting strategy (e.g., 4′,6-diamidino-2-phenylindole [DAPI] and EdU/AF-488) to separate labeled and unlabeled nuclei. A small portion of nuclei from each sorting gate was reanalyzed to determine the sort purity (Supplemental Figure 1B). The reanalysis showed that there was very good separation between the histograms of DNA content for each sorted fraction with only ∼5% overlap between adjacent fractions (Supplemental Figure 1C).

EdU/AF-488 labeled DNA from the sorted nuclei in early, mid, and late S phase was immunoprecipitated (IP) using an antibody specific to the AF-488 moiety. In our experience, the AF-488 DNA-IP is highly reproducible, both in efficiency of precipitation (see Methods) and in quality of the sequencing data obtained when using it. EdU/AF-488-labeled DNA from each S phase stage from three biological replicates, as well as unlabeled G1 reference DNA, was sequenced on an Illumina HiSeq 2000 to generate paired-end 100-bp reads. After quality control and trimming, we obtained over 100 million read pairs per S phase stage that uniquely mapped to the maize B73 AGPv3 reference genome, representing 9 to 17x whole-genome coverage for each S phase sample (see Supplemental Table 1 for mapping statistics). Even though the maize genome is highly repetitive and transposon laden, there is sufficient sequence polymorphism to allow mapping of the majority of reads to the reference genome (Schnable et al., 2009; Gent et al., 2013). The use of paired-end sequencing also aided significantly in mapping reads uniquely.

We generated a profile of the replication activity in each S-phase stage across the genome using a custom computational pipeline called Repliscan described in detail by Zynda et al. (2017). The read densities were aggregated into 1-kb windows, and after observing a strong Pearson correlation of 0.8 to 0.98 between the biological replicates (Supplemental Figure 2), the reads in each window of the replicates were summed (Figure 1F). Artificially high or very low read coverage regions were excluded from the analysis by removing statistically outlying coverage. The total read number from each S phase sample and the G1 reference was then scaled to correct for overall sequencing depth differences, allowing comparison of the signal in corresponding genomic intervals between the different S phase stages. Read counts from each 1-kb window in each S-phase stage were then divided by the read counts for the corresponding genomic interval in the unlabeled G1 data set (Figure 1G), which provided a reference in which all genomic sequences have a 2C copy number. Normalizing to G1 corrected for different sequencing efficiencies, yielding an estimate of replication intensity in each 1-kb window. Finally, the data were smoothed by Haar wavelet transform (Figure 1H) to reduce noise without altering peak boundaries (Percival and Walden, 2000). Figures 1E to 1H show a visual summary of these data processing steps for the early S data in an example region of the genome. The corresponding data for mid and late S in the same genomic region are in Supplemental Figure 3. These analyses resulted in whole-genome profiles of the intensity of the DNA replication occurring in 1-kb windows from cells in early, mid, or late S phase (Figure 2B) visualized using the Integrative Genomics Viewer (IGV; Robinson et al., 2011).

Figure 2.

Chromosome 5 Replication Profiles.

(A) Gene and TE coverage were determined using the maize genome AGPv3 annotation and are expressed as the gene or TE percent coverage, respectively, in 10-kb nonoverlapping windows. For IGV visualization, the coverage values were smoothed (see Methods). Gray dashed line represents 50% coverage.

(B) Replication intensity profiles for early, mid, and late S phase cells processed and presented as described in Methods and the Figure 1 legend. Scale for all replication intensity tracks is 0 to 4.5 normalized signal ratio.

(C) Segmentation of replication timing profiles into the predominant replication time classes (RT classes; see Methods) for each 1-kb window. Replication timing was classified as E (early), EM (early and mid), M (mid), ML (mid and late), L (late), EL (early and late), EML (early, mid, and late), and NS (not segmented); see color chart at bottom of figure.

(D) Schematic representation of chromosome 5 with the centromere position approximately marked based on the CENH3 binding region (Zhao et al., 2016). Red rectangles denote the locations of the expanded panels shown below, each 2.5 Mb in size.

(E) A region near the end of the short arm composed predominantly of early replication and E segments.

(F) and (G) Regions near the middle of the short arm composed of various combinations of single and mixed RT class segments. Regions with a relatively high abundance of EM and ML segments often show clear replication activity at both times (F) or occur in regions where replication activity is spreading along the chromosome as S phase proceeds (G).

(H) A pericentromeric region showing predominantly late replication and L segments. Corresponding tracks for gene and TE percentage of coverage and the composite RT class segmentation are displayed in each expanded panel, as described in (A) and (C).

(I) The RT segment classification color legend.

The Temporal Order of Replication on Maize Chromosomes

The maize B73 genome consists of 10 metacentric chromosomes, which range in size from 150 to 301 Mb in the AGPv3 genome assembly. Protein coding genes are most densely represented on the ends of the chromosome arms (Schnable et al., 2009), though genes are present in all major portions of each chromosome, including the centromere (Zhao et al., 2016). Like many other large genomes, maize is notable for an abundance of transposable elements (TEs), which are estimated to account for ∼85% of the genome (Schnable et al., 2009). We observed several global trends for replication activity across the maize chromosomes: (1) the highest intensity of early replication coincides with the gene-dense ends of the chromosome arms (see early S profile in Figure 2B and Supplemental Figure 4); (2) the highest intensity of late replication coincides with the pericentromeric centers of the chromosomes (see late S profile in Figure 2B and Supplemental Figure 4); and (3) mid S replication is more evenly dispersed along the chromosomes and only slightly higher near the middle of the chromosome arms (see mid S profile in Figure 2B and Supplemental Figure 4).

A replication timing profile for the entire chromosome five is illustrated in Figure 2B, which shows tracks for early, mid and late S replication along with corresponding tracks for gene and TE coverage (Figure 2A). When we further expand our view along chromosome five to smaller example regions (2.5 Mb in size), the finer structure of peaks and valleys of early, mid, and late replication interspersed with each other is apparent. The more detailed patterns are consistent with, but more complex than, the global patterns listed above. For instance, peaks of early replication activity often occur in regions with higher gene coverage (Figures 2E and 2G), and peaks of late replication activity are most often associated with regions of lower gene and higher TE coverage (Figures 2G and 2H).

Though all maize chromosomes generally display similar global replication timing patterns, inspection of the patterns for individual chromosomes in detail revealed many interesting differences (Supplemental Figures 5 and 6). For example, chromosome four, which has the lowest gene density (17 genes/Mb), also has several dispersed large blocks (0.7–3 Mb) of late replication that occur in the middle and ends of the chromosome arms (Supplemental Figure 6A). Another example of chromosome variability is chromosome six, which contains both a knob and the nucleolus organizer region on its short arm, as seen by fluorescence in situ hybridization (Albert et al., 2010), and also has an asymmetrically positioned centromere in the reference genome assembly. The replication timing pattern of chromosome six reflects this asymmetry with more intense early replication occurring on the long arm (Supplemental Figure 6B). However, it should be noted that there are known difficulties with portions of the assembly of chromosome six (Bilinski et al., 2015; Gent et al., 2015; Zhao et al., 2016). As the B73 reference genome assembly improves in the future, we will be able to further fine-tune our understanding of the replication timing patterns on this chromosome.

The only similar study profiling DNA replication timing in plants examined Arabidopsis chromosome four (Lee et al., 2010). That study, which used a 1-kb tiling microarray and a 1-h BrdU pulse, concluded that the replication timing program was essentially biphasic, meaning the early and mid S replication intensity profiles were nearly identical for most of the euchromatic portion of chromosome four, while heterochromatin replicated separately in late S. This initial finding prompted us to assess whether the replication program of the much larger maize genome with more repetitive DNA follows a similar pattern or is more complex. We found evidence that maize does have a more complex temporal replication program, in which the global distribution of replication activity in mid S is clearly different from the patterns in early or late S (Figures 2B and 2E to 2H; Supplemental Figures 4 and 5). However, there is some overlap of both early and late S replication timing patterns with mid S. Interestingly, in regions with distinct mid S patterns, peaks of mid replication often alternate with peaks of either early or late replication, consistent with the idea that replication activity may spread from one temporally ordered “zone” or “domain” of coordinate replication into another (Figures 2G and 2H). These replication zones likely contain clusters of origins that initiate replication or “fire” somewhat synchronously, although not every origin in a cluster is expected to fire in every cell or in every round of replication (Blow et al., 2011; Hyrien, 2016).

One thing immediately apparent from the genomic replication timing profiles is the presence of regions exhibiting heterogeneity in replication time. There are many examples of regions where some level of replication occurs in more than one, or even in all three, S-phase fractions (e.g., see replication intensity profiles in Figure 2F). This phenomenon may be more obvious because of the way the data are presented. For example, if “replication time” is defined as a ratio of replication activity at two discrete times in S phase (e.g., early versus late), the procedure imposes a single replication time for each locus even though that locus may actually display activity at multiple times. Instead, we display the data from each S-phase fraction separately, allowing us to observe replication at multiple times. There are several potential sources of heterogeneity in replication time, one of which is the presence of several cell types in the root tip, each of which may have different replication timing programs.

Classifying Predominant Replication Time across the Genome

To enable the maize replication timing data to be associated with other genomic and epigenomic data, we sought to identify a predominant replication time for each 1-kb window of the genome while still retaining as much information as possible from the replication intensity profiles. We accomplished this using a robust, custom segmentation algorithm that uses the replication timing data to automatically define segmentation parameters, instead of arbitrarily setting them by trial and error (Zynda et al., 2017). The whole-genome segmentation algorithm first defines a threshold above which signals are considered replicating and then classifies regions containing signals above the threshold with respect to the predominant replication time(s). The classification considers the relative intensity of replication signals in early, mid, and late S phase for each 1-kb window. The largest signal value in each window is always classified as replicating, and the algorithm allows multiple-time classifications (e.g., early and mid) when another signal is within 50% of the highest value.

The segmentation of predominant replication time (RT) classes is illustrated in Figure 2 for all of chromosome five (Figure 2C) and for four 2.5-Mb example regions (Figures 2E to 2H). RT classifications for other chromosomes are shown in Supplemental Figure 6. The classifications include the discrete single-time classes, early (E), mid (M), and late (L), as well as their adjoining multiple-time classes, early-mid (EM) and mid-late (ML). The total genome coverage is similar for E (21.8%), M (23.7%), and L (20.1%), with slightly less classified as EM (17.8%) and ML (14.0%) (Figure 3A, Table 1). The segmentation algorithm also allows for early-late (EL) and early-mid-late (EML, pan S) classes that together only comprise 0.5% of the genome (Figure 3A). The multiple-time classes represent regions where replication is occurring strongly at more than one time. In the end, we classified ∼98% of the assembled maize genome into either a single or multiple-time RT class. Given that 97.4% of the genome fell into the five main classes (E, EM, M, ML, and L), these were the primary focus for further analysis.

Figure 3.

Characterizing RT Class Segments.

A segmentation procedure was performed (see Methods) to identify the predominant replication time for each 1-kb window across the genome, and adjacent windows with the same RT class were merged.

(A) Total coverage of each RT segment class across the entire genome. The percentage of the genome covered by each class is noted inside the bars.

(B) The number of individual segments in each RT class.

(C) Box plot of the distribution of segment sizes in each RT class. Box plot whiskers represent 1.5× interquartile range (IQR).

(D) The percentage of proportion of each RT class on individual chromosomes. The entire genome comprises 65.6% single-time segment classes of E (21.8%), M (23.7%), or L (20.1%). Another 32.3% comprises multiple-time segment classes, EM (17.8%), ML (14.0%), EL (0.2%), and EML (0.3%). In all, ∼98% of the nuclear genome was assigned to a segment class. See Figure 2I for RT class color legend.

Table 1. Summary of Genome Composition and Features of RT Classes.

	Coverage (Mb)								FGS
RT Class	Total	%a	Genic	%b	Intergenic	%b	Top 20 LTR-Retro	%b	Gene Countc	%d	Median FPKMe
E	449.9	21.8	95.7	21	356.8	79	263.7	59	18,832	48	3.09 (0.31,11.45)
EM	365.9	17.8	47.2	13	319.6	87	287.2	79	9,446	24	2.02 (0.14,8.39)
M	487.9	23.7	32.4	7	456.0	93	402.7	83	7,605	19	0.54 (0.04,4.0)
ML	288.5	14.0	11.7	4	276.9	96	252.5	88	2,945	7	0.10 (0.02,1.17)
L	413.3	20.1	10.0	2	403.4	98	331.9	80	2,564	6	0.05 (0.01,0.71)
Analyzed	2014.9	97.8	197.0	10a	1812.8	88a	1538.0	75a	41,392
AGPv3 total	2059.9								39,475

^aPercentage of whole-genome coverage.

^bPercentage of RT class coverage.

^cGenes that overlap a RT class segment boundary are counted in both classes.

^dPercentage of FGS genes.

^eMedian calculated by excluding genes with FPKM = 0; values in parentheses are first and third quartiles.

Characterizing Predominant Replication Time Classes

After segmenting the maize genome into predominant RT classes, we investigated the differences in size and number of individual segments in each class. There are fewer overall segments categorized as predominantly E or L (Figure 3B) but the median size is larger (see Figure 3C for size distribution) than the segments in other classes. The larger segments are consistent with the replication intensity profiles, in which strong early or late replicating regions often occur in large blocks (e.g., see Figures 2E and 2H). M segments, which have the most abundant genome coverage overall (23.7%), are more numerous but generally smaller than E and L segments (Figures 3B and 3C), consistent with the observation that mid S replication activity is fairly evenly dispersed in lower intensity blocks across chromosomes (e.g., see Figures 2F and 2G). Segments categorized as EM and ML are similar to M in segment size (Figure 3C) and are most often found either in the transition regions between spreading peaks (e.g., see Figure 2G) or in regions where there is heterogeneity in replication time (e.g., see Figure 2F). We found a generally consistent distribution of RT classes on all 10 chromosomes, with some variability between chromosomes for more E or L class coverage (Figure 3D), reflecting the unique timing patterns associated with features that are chromosome specific. We also analyzed the percent GC content and found it varied only slightly between RT classes (Supplemental Figure 7).

Association of Genes and Transcriptional Abundance with Replication Time

We next examined the association of the different RT classes with other genomic features in maize. We already noted the high intensity of early replication in the gene-rich ends of chromosome arms, but E segments are also found in many other places, and we wanted to more fully understand the relationship between replication time and all genes, not just the ones at the ends of arms. We used the AGPv3 5b+ filtered gene set (FGS) annotation, which contains 39,475 putative protein coding genes with evidence for functionality. When we considered the total coverage in 1-kb windows of each RT class of segments across the genome and calculated the percentage of each RT class that overlapped the boundaries of a gene, we found that E segments overlap ∼48% of the genes (Table 1). However, only 21% of the 1-kb windows that make up the E segments physically overlap a gene (Figure 4A), indicating that many such regions are not associated with a currently annotated gene. Other RT classes have reduced association with genes, as measured both by the number of genes and percent of the total windows in each class that overlap a gene (Table 1, Figure 4A). We used a permutation analysis approach to assess the statistical significance of the percent overlap values found between each RT segment class and genes (see Methods). We tested the relationships in both directions, representing two different hypotheses. For example, to test the significance of the enrichment of genes in RT segment classes, we randomly shuffled RT segments; conversely, to test the significance of the enrichment of an RT segment class in genes, we randomly shuffled genes. We found that the observed enrichment for genes in both the E and EM classes (asterisks in Figure 4A; Supplemental Table 2) and the enrichment for E and EM segments in genes (Supplemental Table 3) are both highly significant compared with those expected by chance (permutation P value = 0.001).

Figure 4.

Association of Genes and Gene Transcription with Replication Time.

(A) The number of 1-kb windows in each RT class that overlaps an FGS gene, displayed as percentage of all 1-kb windows in that class. Asterisks denote RT classes in which the indicated percent overlap was significantly greater than expected by chance (permutation P value = 0.001; see Methods). For full details of the permutation analysis, see Supplemental Tables 2 and 3.

(B) The distance from a given gene to the nearest neighboring gene was measured from the 5′ and 3′ ends of genes found in each RT class. This analysis did not consider the RT class of the neighboring genes. The distribution of gene distances is shown as a box plot, with whiskers representing 1.5× IQR.

(C) Gene expression values in FPKM were calculated from RNA-seq data for genes in each RT class, and the distribution is shown as a box plot, excluding genes with an FPKM of zero.

(D) and (E) Genes were further categorized into the following expression levels: FPKM = 0; >0 and ≤1; >1 and ≤10; >10 and ≤100; and >100.

(D) The number of genes found in each RT class and expression group is shown.

(E) The gene count from (D) is presented as the percentage of the total number of genes in each expression level group (totals shown at top of graph). See Figure 2I for RT class color legend.

We also asked if replication timing varies with gene density. As a measure of local gene density, we identified genes found in each RT class and measured the distance from their 5′ and 3′ ends to the nearest genes on each side, regardless of the RT class where the nearest gene was found. The median nearest gene distance for genes in E segments is ∼10 kb, with a large increase in the distribution of gene distances for later replication classes (Figure 4B). For example, the median gene distance is 43 kb for M segments and 73 kb for L segments. Thus, E segments are found in the most gene dense regions of the chromosome. In spite of this, 79% of E windows do not actually overlap a gene (Table 1). However, such regions are usually in close proximity to genes but also contain other features, such as TEs.

Additionally, we assayed transcriptional activity by RNA-seq in the same terminal 1-mm root tip region used to generate the Repli-seq data and evaluated the distribution of transcription levels for genes in each RT class (Figure 4C). The normalized gene expression values, expressed as fragments per kilobase of transcript per million mapped reads (FPKM), were calculated for FGS genes. Genes in each RT class were then ranked into five gene expression levels, ranging from no expression (FPKM = 0) to >100 FPKM. The majority of genes (65%) are low to moderately expressed, but all gene expression levels that have an FPKM greater than zero are strongly represented in E segments (Figure 4D). By contrast, genes with an FPKM of zero are more evenly distributed among the RT classes (Figure 4D). Furthermore, even though there are fewer genes expressed at high levels, the percentage of genes replicating early increased with expression level, with the E class including 72% of the genes in the highest expression level (FPKM > 100; Figure 4E). However, there are some exceptions of very highly expressed genes in other RT classes. Taken together, our data show that regions undergoing early replication are enriched for genes, particularly for genes expressed at higher levels.

Association of Transposable Elements with Replication Time

Collectively, TEs of various kinds comprise most of the maize genome. To assess their relationship(s) to replication time, we used the repeat region annotations from AGPv3 to calculate the total TE coverage in 10-kb windows and quantify the TE coverage in segments from each RT class. The 10-kb windows were used to avoid the overwhelming predominance of 0 and 100% coverage values that result when 1-kb windows are used. The median TE coverage in E segments is surprisingly high at 85%, with a steady increase up to 98% in L segments (Figure 5A). However, within E segments there is a much broader distribution of the two central quartiles for TE coverage (52–99%) relative to L segments (89–100%) (Figure 5A). TEs are traditionally thought to be associated with heterochromatic regions of the genome, so the high presence of TEs in E and EM classes is interesting and warranted further investigation. Hence, we decided to examine individual families within the most abundant group of TEs, the class I long terminal repeat retrotransposons (LTR-retros).

Figure 5.

Replication Times for TEs.

(A) The percentage of total TE coverage was calculated for RT class segments (see Methods). The distribution of percentage of coverage values in each RT class is shown as a box plot (see Methods).

(B) The total coverage in megabases of the top 20 most abundant LTR-retro families in each RT class. The percentage of each RT class that is composed of these top 20 families is shown inside the bars.

(C) The distance from individual LTR-retro family members to the nearest neighboring gene was measured for the top six most abundant families, and the distribution is shown as a box plot.

(D) The coverage in megabases of individual families from the top six most abundant LTR-retro families in each RT class (x axis shared with [E]). The families are grouped based on their RT class abundance (“earlier,” “middle,” and “later”) and then ordered by total abundance. Asterisks denote RT classes in which the observed percentage of overlap with each family, as indicated inside the bars, was significantly greater than expected by chance (permutation P value = 0.001; see Methods). For full details of the permutation analysis, see Supplemental Tables 2 and 3. The RT class coverage in megabases of the top 20 most abundant LTR-retro families is shown in Supplemental Figures 8B to 8D.

(E) The distribution of distance to the nearest gene for family members within each RT class from (D). The number of family members found in each family and RT class is indicated above the boxes. Box plot whiskers represent 1.5× IQR. See Figure 2I for RT class color legend.

In the maize genome, the LTR-retros constitute ∼75% of the genome sequence (Baucom et al., 2009). Although there are over 400 LTR-retro families, the top 20 most abundant families make up ∼70% of the genome and are comprised of four families from the RLC/copia superfamily, 12 from the RLG/gypsy superfamily, and four from the RLX/unknown superfamily (Baucom et al., 2009). Annotations in AGPv3 were used to calculate the coverage for each of the top 20 LTR-retro families in each RT class. Taken together, the distribution of the top 20 LTR-retros do not differ dramatically from the overall genome distribution of the RT segment classes (compare Figures 3A to 5B), except the total coverage in E segments is lower than expected for a completely even distribution (Figure 5B). The percentage of coverage for the top 20 LTR-retros is 59% for E segments and ranges from 79 to 88% for the other RT classes (Figure 5B, Table 1). However, individual families differ dramatically with respect to their distribution across chromosomes (Supplemental Figure 8A; Mroczek and Dawe, 2003; Lamb et al., 2007). We grouped the families according to replication time trend (“earlier,” “middle,” and “later”) and ordered them by abundance. The six most abundant of the top 20 families, which display replication timing trends representative of the rest in their group, are in Figure 5D, while results for all 20 families are shown in Supplemental Figures 8B to 8D. Again, we used permutation analysis to assess if the enrichment of elements from each of the top six families found in each RT class was statistically significant (permutation P value = 0.001) compared with that expected by chance. We found that the two highest abundance RLC/copia families, Ji and Opie, are significantly enriched in the E, EM, and M classes (percentage of overlaps and asterisks indicated in Figure 5D). The four most abundant RLG/gypsy families displayed different trends. The RLG Huck family is significantly enriched in the E and M classes, but not the rest of the RT classes. This enrichment in noncontiguous RT classes may relate to the presence of several Huck subfamilies with different characteristics (SanMiguel and Vitte, 2009). The RLG Xilon-diguus family is significantly enriched in the M, ML, and L classes, and finally the RLG Cinful-zeon and Flip families are significantly enriched in the ML and L classes (asterisks in Figure 5D). All of these relationships are also significant in the reverse permutation tests (shuffling LTR-retro family members; see Methods). However, in these reverse tests, the M class is also significantly enriched in RLG Cinful-zeon elements, even though RLG Cinful-zeon elements are not significantly enriched in M segments (compare Supplemental Tables 2 and 3).

We further investigated these top six most abundant LTR-retro families to determine whether they differ in their proximity to genes. We found that the RLC Ji and Opie families are typically closer to genes than the RLG families (Figure 5C), which was previously reported as a general feature of the RLC/copia superfamily (Schnable et al., 2009). For each family, we also asked if distance to the nearest gene varied with RT class. Interestingly, the distribution of distance to the nearest gene for elements within a particular RT class does not vary widely between families, and elements in earlier replicating classes are consistently closer to genes (Figure 5E). For example, for elements found in E segments, the median distance to the nearest gene ranged from 8 to 13 kb in different families, while in L segments the median ranges from 49 to 68 kb (Figure 5E). Although the distance distributions are similar for different families, it is important to note that the number of elements in each RT class varies widely between the families (see the numbers above the boxes in Figure 5E). This result further emphasizes the apparent insertion and/or retention biases of these families.

Maize Functional Centromeres Replicate Predominantly in Mid S

Maize centromeres also contain highly repetitive DNA sequences, including several centromeric retrotransposons (CRM1 and CRM2) and the CentC tandem repeat (Zhong et al., 2002; Wolfgruber et al., 2009). However, functional centromeres can only be clearly differentiated from pericentromeres by binding of centromeric histone H3 (CENH3), a histone H3 variant (Gent et al., 2012). Using the reported locations of CENH3 binding domains (Gent et al., 2015; Zhao et al., 2016), we found that in most cases the centromere core replicates predominantly in the M class and is flanked by ML segments that transition into L segments at the edges of the CENH3 binding regions (Figure 6A). Interestingly, close inspection uncovered low levels of early replication activity in the functional centromere (Figure 6A, blue early S replication intensity track), resulting in some small EM segments in some of the centromeres. The predominantly M and ML replication pattern is clearest in the fully assembled centromeres of chromosomes two and five (Figure 6A; Supplemental Figure 9), but holds true with some variation across all centromeres except for chromosome one, which is particularly poorly assembled with an extremely small mapped CENH3 binding domain of ∼10 kb (Gent et al., 2015).

Figure 6.

Replication Times for Centromeres and Tandem Repeat Sequences.

(A) The functional centromere of chromosome 5, as defined by CENH3 binding (black rectangle; from Zhao et al., 2016), replicates predominantly in M and transitions to ML and L near the ends of the CENH3 binding region. See Figure 2I for RT class color legend.

(B) and (C) Tandem repeat consensus sequences were blasted against the trimmed Repli-seq reads, independent of mapping to the reference genome, to estimate the abundance of these tandem repeat sequences in the Repli-seq reads (see Methods).

(B) The percentage of reads corresponding to each tandem repeat sequence in each replication time sample.

(C) The fold enrichment of each tandem repeat relative to the amount in G1. The mean and sd (error bars) for three biological replicates of early and mid S and two biological replicates of late S are displayed.

Analysis of Tandem Repeat Sequences in Repli-Seq Reads

Like many other large genomes, the maize genome contains a large number of tandemly repeated arrays (Plohl et al., 2008). Some of the high-copy tandem repeats in maize include the knob-associated knob180 and TR-1 repeats (Peacock et al., 1981; Ananiev et al., 1998a), the 5S and 45S rDNA repeats (Rivin et al., 1986), and the centromere-associated CentC repeat (Ananiev et al., 1998b). Despite the large amount of DNA estimated to physically be in these tandem repeat arrays (30-Mb knob180, 2-Mb TR-1, 35-Mb 45S rDNA, 3-Mb CentC [Schnable et al., 2009], and ∼0.8-Mb 5S rDNA [Rivin et al., 1986]), they are not well represented in the genome assembly (Schnable et al., 2009). Regions that match these repeat sequences are often visible in our raw mapped Repli-seq reads as large spikes of “collapsed” signal (e.g., see Figure 1F, arrowheads). Given that reads mapping to such regions are derived from many identical copies that are sometimes from different locations, we excluded them from our replication intensity profiles and segment classification. To investigate the replication times of these biologically important tandem repeat arrays, we used a different approach that took into account all of the Repli-seq sequencing reads, even those that did not uniquely map to the genome assembly. We used consensus sequences for the knob180, TR-1, 5S and 45S rDNA, and CentC tandem repeats (courtesy of J. Gent, K. Dawe, T. Wolfgruber, and G. Presting) to individually query trimmed Repli-seq reads using BLAST software and a nonstringent E value to allow for variants of each repeat (Gent et al., 2014). The resulting read counts for each tandem repeat type in each S phase or the G1 genomic DNA sample were then normalized to the total number of reads in that sample to obtain the percentage of reads that align to each tandem repeat.

As expected, we found that reads corresponding to 45S rDNA and knob180 repeats are much more abundant than those corresponding to 5S rDNA, TR-1, and CentC (Figure 6B). To address this difference, the percentage of early, mid, and late reads matching each tandem repeat was normalized to the percentage in G1 genomic DNA to obtain a measure of the fold enrichment relative to G1 (Figure 6C). The 5S rDNA repeat has over 3-fold enrichment in early S reads and is depleted in mid and late S reads. The opposite pattern was clear for the 45S rDNA repeat sequence, which showed an almost 2-fold enrichment over G1 in the late S reads. A much smaller fraction of the 45S rDNA replicates in early and mid S and likely corresponds to a small fraction of 45S copies that are transcriptionally active (Buescher et al., 1984; Tucker et al., 2010). Both of the knob repeat sequences, knob180 and TR-1, replicate almost entirely in the late S fraction, where they are enriched by 1.7- and 3-fold, respectively, relative to G1. This result is consistent with our previous cytological analysis of late S nuclei using a fluorescence in situ hybridization probe for knob180 repeat clusters (Bass et al., 2015), and with an older report that knobs are some of the last sequences to replicate in maize (Pryor et al., 1980). Finally, the CentC repeat sequence is much more distributed across the reads in all three S phase fractions. Although most of this sequence replicates with the late S fraction, significant portions replicate in mid S and even early S fractions. Although the distribution of the CentC repeat in our Repli-seq reads was initially surprising, it may reflect the presence of some smaller clusters of CentC repeats outside of centromeric regions (Bilinski et al., 2015) and our observation that the functional centromere replicates primarily in the M and ML RT classes with lower amounts of activity in E and L.

Association of Other Chromatin Features with Replication Time

To explore potential associations of chromatin structure with replication time, we used chromatin immunoprecipitation sequencing (ChIP-seq) to profile the genomic locations of three histone modifications in G1 nuclei. The histone marks were acetylation of lysine 56 (H3K56ac), trimethylation of lysine 4 (H3K4me3), and trimethylation of lysine 27 (H3K27me3). H3K4me3 is a euchromatic mark associated almost exclusively with genes and is highly conserved among eukaryotes (Fuchs and Schubert, 2012). H3K4me3 also shows a consistent distribution in cytologically visible euchromatic regions in cells from various zones of the developing maize root (Yan et al., 2014). H3K56ac has roles in several different biological processes, including transcription, DNA replication and repair, and nucleosome dynamics in yeast and other eukaryotes (Masumoto et al., 2005; Xu et al., 2005; Han et al., 2007; Kaplan et al., 2008; Williams et al., 2008). H3K27me3 is a mark for facultative heterochromatin and is associated with repressed transcription, developmental gene regulation, and imprinting in maize (Wang et al., 2009; Makarevitch et al., 2013; Zhang et al., 2014). We expected H3K4me3 and H3K56ac to mark open chromatin in the euchromatic portion of the genome and H3K27me3 to mark chromatin that is more condensed, though not as tightly packaged as constitutive heterochromatin. We obtained over 41 million mapped read pairs for each histone mark (Supplemental Table 1). Globally, the distribution of called peaks for H3K4me3 and H3K56ac closely follows the gene distribution on maize chromosomes (Supplemental Figure 10A). The distribution of H3K27me3 also parallels the gene distribution to some degree, in that it is most abundant near the ends of chromosomes, but it is also abundant all across the chromosome arms except for near the centromere (Supplemental Figure 10A). Supplemental Table 4 is a summary of the number and average size of called peaks in G1 cells from root tips.

We then identified the replication time of regions of the genome containing each of the three histone marks or any combination in close proximity (within 1 kb), giving each 1-kb window a histone mark “signature.” We found that H3K56ac and H3K4me3 by themselves or in combination have a strong association with earlier replicating regions, and H3K4me3 is rarely found without H3K56ac in close proximity (Figure 7A). Permutation analysis also showed significant enrichment (permutation P value = 0.001) for these marks in E and EM segments compared with that expected by chance (asterisks in Figure 7A; Supplemental Table 2). Altogether, 1-kb windows marked with H3K56ac and H3K4me3 peaks alone or together comprise 13% of the regions classified as E segments, with the majority (71%) overlapping within 1 kb of a gene. Regions marked by H3K27me3 are less abundant (∼4–7% of each RT class) and follow two different replication time trends. When H3K27me3 is colocated with one or both of the active marks it is significantly enriched in the E and EM segments. Conversely, when H3K27me3 is alone, there is a small but significant enrichment in the ML class (asterisks in Figure 7A). However, the reverse tests, assessing the enrichment of RT classes in H3K27me3 peak regions, sometimes yielded different significance outcomes (compared in Supplemental Tables 2 and 3), further indicating the complexity of the relationships with this mark.

Figure 7.

Replication Times for Chromatin-Related Features.

(A) The number of 1-kb windows in each RT class that overlaps called peak regions for H3K56ac, H3K4me3, H3K27me3, or any combination of these three marks, presented as a percentage of the total number of 1-kb windows in each RT class. The inset shows the percentage of a lower abundance histone mark signature, H3K4me3 without H3K56ac or H3K27me3, on an expanded y axis. The histone mark signature labeled H3K56ac/H3K4me3/H3K27me3 represents any combination of either H3K56ac or H3K4me3 with H3K27me3.

(B) MNase hypersensitivity (HS) region data from whole shoots and roots from Rodgers-Melnick et al. (2016) were overlaid with the segmented RT classes and the number of HS regions counted in each RT class. The count of HS regions per megabase covered by each RT class is displayed. Asterisks denote RT classes in which the observed percentage of overlap of the indicated feature was significantly greater than expected by chance (permutation P value = 0.001; see Methods). For full details of the permutation analysis, see Supplemental Tables 2 and 3. See Figure 2I for RT class color legend.

We calculated the median FPKM values for FGS gene-containing windows with each histone mark signature to test whether the expected expression patterns are present. As expected, genes with H3K56ac or H3K4me3 alone or in combination have relatively high median expression levels (Supplemental Figure 10B). However, within the group of windows containing these active marks, gene expression is also associated by the replication time of the genes, such that genes in E segments have higher median expression than those in M segments. Conversely, genes with H3K27me3 alone or in combination with either of the two active marks have zero or extremely low median expression levels regardless of RT (Supplemental Figure 10B).

To further investigate the relationship between RT and chromatin accessibility, we also compared the RT classes with recently published data for micrococcal nuclease hypersensitivity (MNase HS) regions in whole shoots and roots of 9-d-old maize B73 seedlings (Rodgers-Melnick et al., 2016). In maize, MNase HS regions are indicators of open chromatin that are enriched in the areas surrounding genes, are associated with meiotic recombination hot spots, and explain a large amount of heritable phenotypic variation (Vera et al., 2014; Rodgers-Melnick et al., 2016). We counted the number of MNase HS regions in genomic regions corresponding to our RT class segments (Figure 7B) and also calculated the percentage of overlap of each class with HS regions. We found that MNase HS regions reported for both the whole shoot and root samples are significantly enriched in E and EM segments (permutation P value = 0.001; Supplemental Table 2) compared with values expected by chance. There was a sharp decrease in HS regions in M and later segments (Figure 7B). Additionally, we found that 54% of HS regions are located in E segments, and another 23% are located in EM segments (Supplemental Table 3), further supporting their significant association with early replication.

DISCUSSION

Characterizing DNA Replication in Maize

We used a novel adaptation of the Repli-seq assay by labeling newly replicated DNA with a short EdU pulse and sorting nuclei based on EdU incorporation (as AF-488 fluorescence) in addition to the traditional DNA content. The mild conditions and rapid detection of EdU by click chemistry (Salic and Mitchison, 2008; Darzynkiewicz et al., 2011) are an improvement to the classical BrdU labeling protocol we used previously (Lee et al., 2010). The use of EdU also enables another level of purification of S phase nuclei by two-color sorting before immunoprecipitation of labeled DNA. This additional purification reduces unlabeled DNA contamination of the immunoprecipitates, especially in the case of early and late S samples, which otherwise would contain a large excess of unlabeled nuclei.

In the course of our work, we also developed a novel and publically available computational pipeline called Repliscan for analyzing Repli-seq data. This analysis pipeline automatically sets parameters based on the data themselves, is specifically tailored to identify regions showing heterogeneous replication times, and can be applied to data from different species (Zynda et al., 2017). With these tools, we characterized the whole-genome replication timing program for maize, an important crop species with a sequenced genome two-thirds the size of the human genome (Lander et al., 2001; Schnable et al., 2009). Our use of root tips allowed us to analyze the replication program in a naturally occurring, intact organ with none of the manipulations involved in creating a cell culture system. Although we cannot separate the different cell types in the terminal 1 mm of the root at this time, our analysis pipeline identified the predominant replication time for any given locus and thus provided a consensus view of replication programs in actively dividing cells of the root tip. Both the raw sequencing data and processed replication timing data files are available, and the processed files can be visualized using the link to download a preformatted IGV session file (see Methods and Supplemental Tables 5 and 6).

Temporal Order of DNA Replication

Sorting three fractions of the mitotic S phase and displaying the data from each fraction separately uncovered a complex replication timing program. This program includes regions of distinct early, mid, and late S phase replication timing, as well as many regions that exhibit significant replication activity in more than one portion of S phase. Viewed globally, the highest intensity of early replication is at the gene-dense ends of chromosome arms, the highest intensity of late replication is in the pericentromeric regions, and the highest intensity of mid replication is located mid-way between the other two times. However, at the local level, we observed many fine scale patterns of interspersed early, mid, and late replication. In many cases, the patterns of interspersion indicated that sequences are replicating over consecutive fractions of S phase in a continuous, spreading pattern, as was also reported for some mammalian studies that sorted more than two fractions of S phase (Chen et al., 2010; Hansen et al., 2010). In other regions, such spreading is less apparent and, instead, there is significant replication at more than one time in S phase. This timing heterogeneity would not have been captured if the data had been expressed as a ratio of early to late replication like many earlier metazoan studies (Hiratani et al., 2008, 2010; Schwaiger et al., 2009; Ryba et al., 2010).

There are several possible sources for this heterogeneity, each requiring its own specific experimental approach to detect. First, it is well documented from single DNA fiber studies that individual origin use and firing efficiency can vary from cell to cell and from one cell cycle to the next in animal systems (reviewed in Hyrien, 2016). From population averaging studies of replication time, it is thought that this flexibility in origin use is constrained to a broadly defined time in S phase (e.g., early versus late) by a higher level of regulation in large regions termed replication domains (reviewed in Klein and Gilbert, 2016). However, when more than two fractions of S phase have been analyzed in human or mouse cell line populations, it has been estimated that up to 10% of the genome replicates at more than one time (Farkash-Amar et al., 2008). A second possible source of heterogeneity arises from the presence of multiple cell types in the root meristem region, as individual cell types might have different timing programs. Between any two cell types of human or mouse differentiated embryonic stem cells, up to 20% of the genome changes replication time (Hiratani et al., 2008; Gilbert et al., 2010; Hansen et al., 2010; Ryba et al., 2010). It has also been reported that 12% of the genome of human primary erythroblasts exhibits allele-specific differences in replication time, highlighting a third potential source of timing heterogeneity. Such regions can be up to ∼4 Mb in size and are enriched in imprinted genes (Mukhopadhyay et al., 2014). In our root tip system, we classified ∼32% of the genome as replicating at more than one time during S phase. This level of heterogeneity suggests that cellular and/or allelic heterogeneity are important in the maize root tip system. We cannot yet distinguish between these two possibilities. In the future, analysis of separated cell types, as well as comparisons with other meristems in the maize plant, may better resolve the complexity of the replication timing program.

Regions of Coordinate Replication

Large regions of mammalian genomes containing multiple replicons have been reported to exhibit coordinate replication. These regions (termed replication domains) average 400 to 800 kb but can be up to several megabases in size (Hiratani et al., 2008; Ryba et al., 2010; Klein and Gilbert, 2016). In our replication timing profiles, we can find a few examples of coordinate regions on this scale (e.g., the large central peak in Figure 2H), but regions of this size are not typical for maize. Instead, regions of coordinate replication in maize are usually much smaller (∼50–300 kb; Figures 2G and 8A) and similar in size to regions reported for Drosophila and Arabidopsis (MacAlpine et al., 2004; Schwaiger et al., 2009; Lee et al., 2010). The smaller (∼30–450 kb) coordinate regions in these species were attributed originally to their small genome sizes (Lee et al., 2010; Rhind and Gilbert, 2013). However, the maize genome is comparable in size to mammalian genomes and almost 20 times larger than those of Drosophila and Arabidopsis, suggesting that differences in coordinate region size cannot be solely a function of genome size.

Figure 8.

Models of DNA Replication Timing Progression in Maize.

(A) Replication timing intensity profiles for early (blue), mid (green), and late (red) S-phase cells, as described in Figures 1 and 2, are overlaid to highlight the spreading pattern over consecutive fractions of S phase. Two representative regions from chromosome 5 are shown, one in the middle of the chromosome arm (left panel) and a second in the pericentromere (right panel). Tracks containing annotated regions for total TEs, the top six most abundant LTR-retro families from Figure 5 (LTR-retro), and genes, as well as a segmentation track showing the predominant replication time (RT class) are also included for reference.

(B) and (C) Two nonmutually exclusive models for how replication proceeds through S phase in maize. In both models, replication begins at origins or initiation zones (circles) and proceeds bidirectionally (arrows). In the “cascade” model (B), replication initiates in early S and cascades to adjacent origins initiating in mid and then late S phase. In the “elongation” model (C), replication initiates at origins in early S and proceeds through mid S regions by passive elongation of replication forks. In this model, there are no origins initiating specifically in mid S phase. In the pericentromere, which predominantly replicates in late S, the elongation model envisions that small regions with early initiation could passively elongate through mid S, followed by a second round of initiation events in late S phase.

It is difficult to know to what extent estimates of the size of coordinate regions are related to differences in the scale or resolution of the analysis, as opposed to true biological differences in genome arrangement or regulation. Our data were handled in 1-kb static windows across the genome and only lightly smoothed, providing an in-depth, highly granular view of the complexity of replication timing. A similar view of mammalian replication might reveal more complex patterns within replication domains. However, there are substantial differences in the structure of the genes and genomes in maize versus humans or mice that may contribute to the observed differences in replication patterns. For example, maize differs from mammals in total gene density on chromosomes, gene length, and intron length (Lander et al., 2001; Rafalski and Morgante, 2004; Schnable et al., 2009). In addition, maize genes are concentrated near the ends of chromosome arms (Schnable et al., 2009; Wang et al., 2016; Figure 2A), while human genes are clustered in regions of increased gene density and expression (“ridges”) dispersed across chromosomes (Caron et al., 2001; Versteeg et al., 2003). These differences in chromosome organization should be considered together with other larger scale differences in the spatial organization of replication within the nucleus. In maize, mid S replicating loci do not cluster around the nuclear and nucleolar periphery (Bass et al., 2015) as they do in the mammalian nucleus (Dimitrova and Berezney, 2002; Panning and Gilbert, 2005; Zink, 2006). Instead, numerous foci are distributed throughout the nucleoplasm during both early and mid S phase (Bass et al., 2015).

Because mammalian replication domains sometimes exhibit coordinated changes in replication time during development (Gilbert et al., 2010), it may eventually be possible to identify similarly coordinated domains in maize by comparing replication timing programs in other maize meristems, tissues, or cell types. Such an analysis would provide insight into the possible existence and importance of large replication domains in maize.

Replication Timing in Relation to Chromatin Packaging

In eukaryotes, there is no single chromatin feature that correlates exactly with replication time, but instead complex interactions between the ensemble of histone modifications and chromatin binding proteins are thought to relate to replication timing (Rhind and Gilbert, 2013). However, in all higher eukaryotes studied, early replication is usually associated with active chromatin modifications (Karnani et al., 2007; Schwaiger et al., 2009; Lee et al., 2010; Ryba et al., 2010; Eaton et al., 2011; Mechali et al., 2013). We found a similar association of early replication and active modifications in maize. Many active modifications (e.g., H3K4me2/3, H3K36me3, and H3ac) are typically found in the same regions of the genome (Roudier et al., 2011; Julienne et al., 2013) and are usually more indicative of gene expression than replication time (Ryba et al., 2010). By selecting H3K4me3 and H3K56ac, we have likely captured a reasonable consensus of the open regions containing active histone marks within the cell types in the maize root tip. However, it is important to note that only a small fraction (16%) of E class regions are marked by H3K4me3 or H3K56ac, suggesting that the presence of these active marks is not necessary for early replication to occur.

Chromatin accessibility, indicated by susceptibility to endonuclease cleavage, is also associated with early replication in metazoans (Gilbert et al., 2004; Audit et al., 2009; Hansen et al., 2010; Takebayashi et al., 2012; Rhind and Gilbert, 2013). In maize, we saw a strong association between early replication and published data for MNase HS sites profiled from either root or shoot tissues of 9-d-old maize B73 seedlings (Rodgers-Melnick et al., 2016). This result seems highly significant in light of the fact that in human cells, a simple model of DNA replication produced extremely accurate predictions of replication timing profiles when DNase I HS sites were used to construct a “probability landscape” for initiation (Gindin et al., 2014). Additionally, other studies have noted that DNase I or MNase HS sites are enriched in or near some classes of origins in humans and yeast (Audit et al., 2009; Rodriguez and Tsukiyama, 2013; Mukhopadhyay et al., 2014; Cayrou et al., 2015).

There is less consensus on the association of late replication with repressive chromatin modifications (e.g., H3K27me3, H3K9me2/3, and H3K20me3) (e.g., Lee et al., 2010; Julienne et al., 2013; Rhind and Gilbert, 2013) and observed associations vary among cell types (Hiratani et al., 2008; Ryba et al., 2010; Rhind and Gilbert, 2013). For example, the repressive mark H3K27me3 has been reported to associate with early to mid replication and with early and mid origins (Chandra et al., 2012; Julienne et al., 2013; Picard et al., 2014), as well as with late replication (Thurman et al., 2007). In our data, peaks of H3K27me3 appear to follow two different patterns. When H3K27me3 is in close proximity to the two active marks, it is enriched in the E and EM class. Alternately, when H3K27me3 is not colocated with active marks, it is more evenly distributed across all RT classes. Interestingly, genes containing both of these histone mark signatures (H3K27me3 with or without active marks) have repressed gene expression (Supplemental Figure 10). These observations are consistent with the replication of facultative heterochromatin during any portion of the S phase, not just in late S when constitutive heterochromatin has long been known to replicate (Pryor et al., 1980).

Diversity of Replication Timing in TE Families

TEs, which typically contain high levels of DNA methylation and the repressive histone modification H3K9me2 in maize (Eichten et al., 2012; Regulski et al., 2013; West et al., 2014), are traditionally thought of as silenced chromatin. Specifically, all of the six most abundant LTR-retro families that we investigated exhibit high overall levels of internal DNA methylation and H3K9me2 in B73 aerial tissues, as well as spreading of these heterochromatic modifications into adjacent regions of ∼1 to 2 kb (Eichten et al., 2012). Nonetheless, members of many of the highly abundant LTR-retro families are closely interspersed with genes in the euchromatic arms of maize chromosomes (SanMiguel et al., 1996; Liu et al., 2007; Baucom et al., 2009). Interestingly, in our data we found that two of these highly abundant LTR-retro families, RLC Ji, and Opie, were significantly enriched in E, EM, and M regions, and a third family, RLG Huck, was significantly enriched in E and M regions. This observation appears contrary to the idea that early replicating regions associate with genes, open chromatin, and active histone marks. However, the median distance to the nearest gene for the early replicating members within each LTR-retro family is only ∼10 kb, which is less than the size estimates of a single replicon in monocots (34–60 kb; Van’t Hof, 1996). This result suggests that the earlier replicating elements in each family represent a subset of gene-proximal elements that replicate in association with their neighboring genic regions. Whether or not these earlier replicating subsets still maintain the same levels of heterochromatic modifications as the rest of the family is an interesting question that will require further investigation. However, in Arabidopsis and rice (Oryza sativa), DNA methylation levels are known to vary among individual LTR-retro elements as a function of family, genomic location, age, and length, and these patterns can be masked when averaging methylation across many different insertions (Hollister and Gaut, 2009; Vonholdt et al., 2012). These studies suggest that similar variation in the level of heterochromatic modifications may also exist in maize LTR-retro families, which could contribute to some of the differences in replication time. In future investigations, replication timing data may be a useful tool to help sort out different functional classes or modes of TE silencing regulation, even within a given family.

Models for Maize DNA Replication Timing

When comparing consecutive fractions of S phase, many regions of the maize genome show a clear pattern of early replication spreading bidirectionally into neighboring parts of the genome during mid S phase. In these regions, the progress of replication in mid S can be envisioned as elongation from origins that initiate in early S and/or as initiation and elongation specific to mid S phase. Figure 8 shows a model in which these two possibilities are diagramed in two example regions, the middle of a chromosome arm, and a pericentromere. While not all initiation regions are associated with peaks in a population average replication timing profile, sharp peaks reflect a population preference for initiation in that region (Yang et al., 2010; Hyrien, 2016). The presence of some relatively sharp peaks in mid S suggests these may be initiation zones (Figure 8B). However, mid peaks are generally lower intensity with gentler slopes than the peaks found in early S, consistent with the idea that high efficiency origins fire in early S followed by a “cascade” of lower efficiency origins in mid S (Guilbaud et al., 2011). The generally lower intensity mid S pattern is also consistent with the possibility that many mid S loci are passively replicated by elongation from earlier initiation events (Figure 8C). In this second model, there is also a potential for larger regions to be replicated by unidirectional forks, as has also been hypothesized for TTRs between early and late replicating domains of mammals (Farkash-Amar et al., 2008; Hiratani et al., 2008; Desprat et al., 2009). Our data do not distinguish between the two models, and it is possible that both scenarios occur in different regions of the maize genome. If the elongation model were to apply across large portions of the genome, we would expect to find some places where single replicons are much larger than those so far reported in the plant literature (Van’t Hof, 1996). Detecting such large replicons is technically challenging (Berezney et al., 2000). However, as single-cell sequencing and related technologies improve, we may be able to address this question more directly.

In a previous cytological analysis, we observed that maize euchromatin exists as an intermingled mixture of components replicating in early and mid S phase, with the mid S components exhibiting a higher condensation state. We hypothesized that this pattern might reflect an alternation along the chromosome of gene-rich regions with an extended chromatin structure and mostly replicating during early S, followed by intergenic repetitive regions replicating during middle S phase (Bass et al., 2015). Our molecular data are generally consistent with this model. Early replicating regions contain the highest concentration of genes and highly expressed genes (with some gene-proximal TE family members), while mid replicating regions are relatively more gene poor and TE dense. Genes in mid replicating regions also show generally lower expression. Early replicating regions are also enriched for histone marks associated with active chromatin (H3K4me3 and H3K56ac) and MNase hypersensitive sites, indicative of a more open chromatin structure. Finally, peaks of early and mid replication are arranged in an alternating pattern across large portions of the chromosome arms (Figure 8A, left panel). However, our genomic data also suggest the presence of regions with intermediate replication time and chromatin structure, which may reflect gradients of chromatin accessibility between the cytologically detectable compartments.

In summary, Repli-seq data are a new class of genomic data available to the maize research community. Replication timing is associated with multiple different genomic and epigenetic features, including gene expression, chromatin accessibility, and the spatial organization of chromatin in the nucleus. In metazoans, replication timing is considered a functional readout of the many factors affecting large-scale chromatin structure (Rivera-Mulia et al., 2015; Rivera-Mulia and Gilbert, 2016). Recalling that the spatial arrangement of replication activities in maize nuclei is quite different from human cells (Bass et al., 2015; Savadel and Bass, 2017), it will be of particular interest to examine the extent to which maize replication timing is regulated by large-scale features such as chromatin domains, as opposed to local features such as hypersensitive sites. Such questions will best be approached through a comparative analysis of maize cell types, developmental states, and genetic variants. With the rapidly growing resources available for other maize lines (Lu et al., 2015; Andorf et al., 2016; Hirsch et al., 2016; Jiao et al., 2017), the ability to integrate the multiple kinds of information represented in replication timing profiles will greatly facilitate comparative analyses of the maize pan-genome.

METHODS

Plant Material

Maize (Zea mays) inbred line B73 seeds were imbibed overnight, surface sterilized, and germinated in Magenta boxes at 28°C under constant, dim light (∼500 lux, F15T8 plant and aquarium bulb) for 3 d (Wear et al., 2016). Between 450 and 550 seedlings were pooled for each of three biological replicates for the Repli-seq experiments. Biological replicate material was grown independently and harvested on different days. After 3 d of growth, the seedling roots were immersed in sterile water containing 25 μM EdU (Life Technologies) for 20 min at room temperature with gentle agitation. After rinsing well with sterile water, the terminal 1-mm segments were excised from primary and seminal lateral roots. The root segments were fixed for 15 min in 1% formaldehyde in 1× PBS, the formaldehyde reaction quenched by adding 0.125 M glycine, and the roots washed three times in PBS and snap-frozen (Wear et al., 2016).

Whole-Root Confocal Microscopy

Maize seedlings were grown, EdU labeled, and fixed as described above. Roots were harvested and embedded in 5% agarose in 1× PBS and 100-μm-thick longitudinal sections made using a Vibratome. Sections were washed and permeabilized, and the incorporated EdU was conjugated to AF-488 using a Click-iT EdU Alexa Fluor 488 imaging kit (Life Technologies) according to the manufacturer’s instructions. Sections were counterstained with 0.1 μg/mL DAPI in 1× PBS and imaged on a Zeiss LSM 710 confocal laser scanning microscope with 405-nm (DAPI) and 488-nm (AF-488) lasers at the Cellular and Molecular Imaging Facility at North Carolina State University.

Nuclei Isolation

The fixed, frozen roots described above were ground in cell lysis buffer (CLB from Wear et al., 2016) supplemented with a “Complete Mini” protease inhibitor cocktail tablet (Roche) in a small commercial food processor (Cuisinart Mini-Prep Processor, model DLC-1SS) at 4°C. The resulting homogenate was filtered and centrifuged as previously described (Bass et al., 2015; Wear et al., 2016). Isolated nuclei were washed in modified CLB buffer (CLB without EDTA or β-mercaptoethanol), and the incorporated EdU was conjugated to AF-488 using a Click-iT EdU Alexa Fluor 488 imaging kit (Wear et al., 2016). Finally, the nuclei were resuspended in CLB containing 2 μg/mL DAPI and 40 μg/mL Ribonuclease A and filtered through a CellTrics 20-μm nylon mesh filter (Partec) just before flow cytometry and sorting.

Flow Cytometry and Sorting

Isolated, fixed nuclei used for Repli-seq experiments were sorted and recovered with an InFlux flow cytometer (BD Biosciences) equipped with UV (355 nm) and blue (488 nm) lasers. Events were triggered on forward-angle light scatter and data collected using 90° side scatter and 460/50-nm and 530/40-nm band-pass filters (Bass et al., 2014, 2015; Wear et al., 2016). Nuclei prepared from the terminal 1-mm root segments were sorted into 1× NaCl-Tris-EDTA (STE) buffer, pH 7.5, using substage gates to collect populations of EdU/AF-488-labeled nuclei with DNA contents in three defined gates between 2C and 4C, corresponding to early, mid, and late S phase (Figure 1C). For each biological replicate, 0.4 to 1 × 10⁶ nuclei were sorted for each fraction of S phase, and 1 × 10⁶ unlabeled G1 (2C DNA content) nuclei were sorted to use as a reference. Additionally, a small sample of nuclei (∼50,000) were also sorted from each gate into CLB buffer containing DAPI and reanalyzed to determine the sort purity (Supplemental Figures 1B and 1C). Flow cytometry data were analyzed using FlowJo software v10.0.6 (Tree Star). Plots of side scatter versus DNA content (460/50 nm) were used to set analysis gates that excluded cellular debris in the flow cytometry plots (Supplemental Figure 1A).

Genomic DNA Extraction from Sorted Nuclei

Formaldehyde cross-links were reversed and DNA was solubilized by incubating the sorted nuclei in 50 mM EDTA, 1% sodium lauroyl sarcosine, and 230 μg/mL proteinase K for 1 h at 42°C and then at 65°C overnight in the dark. To inactivate the proteinase K, samples were treated with 8 mM PMSF for 40 min at room temperature prior to extraction of genomic DNA using phenol/chloroform/isoamyl alcohol and a phase lock gel (5 Prime). The upper aqueous phase was mixed with 150 μg/mL GlycoBlue (Ambion), and the DNA was precipitated in 0.3 M sodium acetate and 0.6 volumes of cold isopropyl alcohol. The DNA was pelleted by centrifugation at 21,130g and 20°C for 30 min, washed with 70% ethanol, and centrifuged at 21,130g and 20°C again for 15 min, dried for 5 min using a SpeedVac concentrator (Savant), and resuspended in 130 μL PCR grade water. DNA was sheared using a S2 focused-ultrasonicator (Covaris; Sonolab Simple) with 10% duty cycle, intensity of 5, 200 cycles per burst, and a 4-min cycle length to achieve an average fragment size of ∼250 bp.

EdU/AF-488-Labeled DNA Immunoprecipitation

One to 2.3 μg of sheared input DNA for each IP reaction was brought to a volume of 500 μL with ChIP dilution buffer (Gendrel et al., 2005) and precleared by slowly mixing (8 rpm) for 1 h at 4°C with 20 μL magnetic protein G beads (Dynabeads Life Technologies) preequilibrated in ChIP dilution buffer. The beads were magnetically captured and the supernatant was transferred to a clean tube. The samples were then incubated overnight at 4°C with a 1:200 dilution of anti-Alexa Fluor 488 antibody (Molecular Probes; A-11094, lot 895897) followed by incubation for 2 h at 4°C with 25 μL of preequilibrated protein G beads. The beads were recovered with a magnet and washed as previously described by Gendrel et al. (2005), except an additional 5-min wash was added for each wash step. EdU-labeled, newly synthesized DNA was eluted from the beads by incubating with 250 μL of elution buffer (1% SDS and 100 mM sodium bicarbonate) at 65°C for 15 min. Beads were magnetically captured and the supernatant was transferred to a new tube. The elution was repeated once more and both supernatants combined for a final volume of 500 μL. Immunoprecipitated DNA, yielding from 0.3 to 0.7 ng DNA per 10,000 nuclei, was purified with a QIAquick PCR purification kit (Qiagen) following the manufacturer’s instructions and was eluted from the QIAquick columns in 32 μL elution buffer. The Alexa-488 DNA IP efficiencies ranged from 1 to 2.2%.

Library Construction and Sequencing for Repli-Seq Experiments

Repli-seq paired-end libraries were constructed from 5 to 10 ng of DNA using the NEXTflex Illumina ChIP-Seq Library Prep Kit (Bioo Scientific) and the ultra-low input protocol. After adapter ligation, the libraries were amplified using 18 cycles of PCR. Individual samples from three biological replicates were bar-coded, pooled, and sequenced using three lanes of the Illumina HiSeq 2000 platform.

Read Trimming and Alignment

To improve alignment rates and reduce errors, Trim Galore! v0.3.7 (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) was used to remove 3′ universal adapters from the reads, trim the 5′ ends with FastQ quality scores below 20, and discard reads trimmed shorter than 40 bp. The quality controlled reads were then aligned to the B73 RefGen_v3 (AGPv3) genome downloaded from Ensemble Plants (Kersey et al., 2016) using BWA-MEM v0.7.12 (Li, 2013) with default parameters. Due to the repetitive nature of the maize genome and the sensitivity of replication signals, only unique alignments whose orientations made them proper pairs for our downstream analysis were used. See Supplemental Table 1 for mapping statistics and total sequencing coverage.

Replication Timing Data Analysis

Repli-seq data were analyzed using Repliscan as described in detail by Zynda et al. (2017). Read densities were calculated in 1-kb windows across the genome and the correlation between biological replicates was assessed (Supplemental Figure 2). After observing a strong Pearson correlation of 0.8 to 0.98 between the biological replicates of each sample, the replicates were then summed (Zynda et al., 2017). Genomic windows with artificially high or extremely low log-transformed coverage in the upper and lower 2.5% tails of a calculated gamma distribution were removed (Zynda et al., 2017), and then data were normalized using sequence depth scaling (Diaz et al., 2012). In each 1-kb window, the data from each of the S phase samples were divided by the nonreplicating G1 reference data to further normalize for sequencing biases. To reduce noise, without spreading peak boundaries, Haar wavelet smoothing was performed using the software package wavelets from Percival and Walden (2000). Haar wavelet level three was chosen because it removed low-amplitude noise, while also preserving replication peak boundaries.

Classifying Predominant Replication Time

The strategy and details of classifying a predominant time of replication for each 1-kb window across the genome is described by Zynda et al. (2017). To classify a predominant time of replication, a threshold of replication was first calculated. Our experimental protocol labels DNA that is replicating, so it is hard to discount any signal. Therefore, an automatic analysis was used that maximized chromosomal inclusion while also excluding low signals from previously included windows. Starting from the point of the largest absolute change in coverage (slope) for each chromosome, the replication threshold was lowered (increasing chromosome coverage) until the absolute change in coverage went below 0.1, meaning very few new chromosomal windows were included if the threshold was lowered further. The algorithm automatically set segmentation thresholds for each chromosome; in this case, normalized signal thresholds were between 0.84 and 0.86. Only values above the threshold were considered when segmenting the genome into predominant replication time classes. The predominant time in which a 1-kb window replicates was determined by considering the proportion of total replication signal above the threshold occurring in early, mid, and late S. All signals in each 1-kb window were divided by the maximum value (infinity-norm), scaling the largest value to 1 and all others between 0 and 1. A window was then classified as predominantly replicating in any S phase time with signal greater than 0.5. The infinity-norm ensured that the largest value was always classified as replicating, and this classification method allowed for a window to be called predominantly replicating at more than one time in S phase (e.g., both early and mid) when other signals were within 50% of the maximum value. The final classifications of predominant replication time include early (E), early and mid (EM), mid (M), mid and late (ML), late (L), early and late (EL), pan S (EML), and not segmented (NS; unable to be called replicating at any time).

Replication Intensity and Relative Distance from the Centromere

The normalized and smoothed replication intensity profiles for early, mid, and late S phase in Figure 2 and Supplemental Figure 6 were used to calculate the percent of total replication in consecutive windows, each representing 10% of a given chromosome arm, and plotted as a function of relative distance from the centromere. Centromere positions in the B73 AGPv3 genome were defined as CENH3 binding domains previously reported (Gent et al., 2015; Zhao et al., 2016). Centromere locations for chromosomes 1, 6, and 7 are more uncertain due to low CENH3 mappability and/or poor reference genome assembly (Gent et al., 2015; Zhao et al., 2016).

Genomic Features

For comparison with Repli-seq data, the GC content and annotations for TEs and genes were taken from the B73 genome assembly AGPv3 and averaged across 10-kb static windows. For visual representation in IGV, these data were further smoothed using the R function ksmooth with a Gaussian (normal) kernel and a bandwidth of 5.

Gene Expression Analysis

Maize seedlings were grown as described above, with pooled root tissue from 45 seedlings used for each of three independently grown and harvested biological replicates. The roots were rinsed quickly with sterile water and the terminal 1-mm root segments were excised, snap-frozen in liquid nitrogen, and stored at –70°C. To isolate total RNA, 10 mg of frozen root segments were ground in a mortar and pestle with liquid nitrogen, and the powder was added to a tube of 0.5 mL cold PureLink Plant RNA reagent (Ambion). The manufacturer’s instructions for a small-scale RNA isolation were followed, except that 90 μg/mL of GlycoBlue carrier (Ambion) was added before the chloroform extraction step. The RNA was resuspended in 250 μL RNase-free water and stored at –70°C. Contaminating DNA was removed from 8 μg of total RNA using a Turbo DNA-free kit (Ambion) following the manufacturer’s instructions, except that 4 μL of Turbo DNase was used and the sample was incubated with the DNase for 45 min. DNA-free RNA was collected by isopropyl alcohol precipitation, washed with 75% ethanol, and resuspended in RNase-free water. DNA-free RNA was quantified using a Qubit (Molecular Probes) and the yield was ∼7 μg of RNA. Four micrograms of DNA-free RNA was processed further to deplete rRNA using a Ribo-Zero Magnetic Kit for plant seed/root (EpiCentre) according to the manufacturer’s instructions. The effectiveness of the rRNA depletion was assessed by RT-qPCR using a qScript One-Step SYBR Green qRT-PCR kit, Low ROX (Quanta Biosciences), and primers for maize, 16S, 18S, 23S, and 26S rRNA. For each biological replicate, 77 to 133 ng of rRNA-depleted RNA was used for cDNA conversion and Illumina library construction using a ScriptSeq v2 RNA-Seq Library Preparation Kit (Epicentre), and samples were bar-coded with ScriptSeq Index PCR Primers (Epicentre). After sequencing, Illumina adapter sequences were trimmed with Trim Galore! (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) and mapped to the B73 genome AGPv3 using Bowtie v2.1.0 (Langmead et al., 2009) and TopHat v3.2.3 (Trapnell et al., 2009) with default parameters (for mapping statistics and genome coverage, see Supplemental Table 1). Mapped reads were sorted using SAMtools v0.1.19 (Li et al., 2009), and normalized gene expression values were calculated using Cufflinks v0.9.3 (Trapnell et al., 2010) for gene annotations from the AGPv3 5b+ FGS. The total number of genes overlapping each RT class was divided into five groups based on FPKM values derived from Cufflinks: FPKM = 0, 0 > FPKM ≥ 1, 1 > FPKM ≥ 10, 10 > FPKM ≥ 100, and FPKM > 100 in the same manner as Regulski et al. (2013).

TE Families

Genome locations in the B73 AGPv3 annotation of LTR retrotransposon families from the maize TE consortium database were used (Baucom et al., 2009; Schnable et al., 2009). Overlapping and redundant sequence intervals were collapsed using a custom R script (R Development Core Team, 2016) (script and collapsed files courtesy of M. Stitzer). Of these LTR retrotransposon families, the top 20 most abundant families across the superfamilies of RLG/gypsy, RLC/copia, and RLX/unknown as defined by Baucom et al. (2009) were used for our analysis.

Tandem Repeat Sequences

Reference sequences were acquired for maize tandem repeat classes, including CentC, TR-1, knob180 (courtesy of J. Gent and K. Dawe), and 5S and 45S rDNA (courtesy of T. Wolfgruber and G. Presting). Given that the majority of tandem repeat sequences are not included in the B73 AGPv3 reference genome assembly, the abundance of these repeat sequences in our Repli-seq data independent of the reference genome was measured, as described by Gent et al. (2014), allowing us to query all reads and not just uniquely mapping ones. To do so, filtered, trimmed, and adapter-free DNA fragment reads from individual biological replicates of G1, early, mid, and late S samples were aligned to consensus sequences for each tandem repeat family using BLAST software (parameter “-e 1e-8”). For each sample and biological replicate, the number of reads that aligned to each repeat family was normalized to the total number of reads in the sample. Finally, the relative abundance of each family in early, mid, or late reads was normalized to the relative abundance of the same family in the G1 reference.

ChIP-Seq Analysis of Histone Modifications

Maize seedlings were grown as described above. For each of the three ChIP-seq experiments listed below, root tissue was pooled from between 430 and 720 seedlings for three independently grown and harvested biological replicates. Seedling roots were pulse-labeled with 25 μM EdU for 1 h, and the terminal 3-mm (H3K4me3 and H3K27me3) or 5-mm (H3K56ac) root segments were excised and fixed as described above. After nuclei isolation, the incorporated EdU was conjugated to AF-488, total DNA was stained with DAPI, and then nuclei were flow sorted as described above. The unlabeled G1 (2C) nuclei were collected in 2× extraction buffer 2 (EB2) (Gendrel et al., 2005) diluted to 1× with the sorted drops of 1× STE sheath fluid. The antibodies used for ChIP were anti-H3K56ac rabbit polyclonal 1:200 dilution (Millipore; 07-677, lot DAM1462569), anti-H3K4me3 rabbit monoclonal 1:300 dilution (Millipore; 07-473, lot DAM1779237), and anti-H3K27me3 rabbit polyclonal 1:300 dilution (Millipore; 07-449, lot 2,275,589). ChIP procedures were adapted from Gendrel et al. (2005). Briefly, fixed, sorted nuclei in EB2 buffer were centrifuged at 12,000g at 4°C for 10 min. The supernatant was discarded and the pellet was resuspended in 110 μL of nuclei lysis buffer (Gendrel et al., 2005). Chromatin was sheared using a S2 focused-ultrasonicator (Covaris; Sonolab Simple) with 10% duty cycle, intensity of 5, and 200 cycles per burst for 10 min to achieve an average fragment size of ∼200 bp. After shearing, the ChIP protocol of Gendrel et al. (2005) was followed from their step 17, except for the following changes: The chromatin sample was initially brought up to 1 mL in ChIP dilution buffer, Dynabeads protein G magnetic beads (Life Technologies) were used, an additional 5-min wash was added for each wash step, and a treatment of 55 μg/mL RNase A at 37°C for 1 h was added after reversing the cross-links. The final DNA purification after the ChIP was done using a QIAquick PCR purification kit (Qiagen). ChIP-seq libraries were constructed and sequenced in the same way as the Repli-seq libraries described above, except that 0.6 to 15 ng of DNA was used for library construction. After sequencing, Illumina adapter sequences were trimmed and mapped to AGPv3 as described for Repli-seq data. Redundant reads were removed from each of the BAM alignment files using PICARD (http://broadinstitute.github.io/picard/) and SAMtools (Li et al., 2009) (for mapping statistics and genome coverage, see Supplemental Table 1). Enriched ChIP binding regions, also known as “peaks,” were called using MACS v2.1.0 (Zhang et al., 2008) with parameters “–nomodel–nolambda–broad” and a q-value threshold of 0.01. The called peaks for all three histone modifications were intersected with the RT classes using intersectBed in the BEDTools suite (Quinlan and Hall, 2010) to assign each 1-kb window a histone mark “signature” containing any combination of each mark. The median gene expression level and first and third quartile values were computed for 1-kb windows containing a gene and each histone mark combination in the E, M, and L RT classes using the FPKM data described above.

MNase Hypersensitivity Site Analysis

Published data sets of the genomic locations of MNase HS regions from whole root and shoot tissues from 9-d-old B73 seedlings, as described by Rodgers-Melnick et al. (2016), were provided by E. Buckler. The number of HS regions found in the genomic regions corresponding to our RT classes were counted and normalized by the total number of megabases in each class. The percentage of coverage of both HS regions that overlap with each RT class and RT classes that overlap with HS regions was also calculated.

Association of Genetic and Chromatin Features with Replication Time Classes

Various genetic and chromatin features were associated with the RT classes to determine the overlap of a particular feature with each RT class. To calculate the coverage of genes, LTR-retro families, or histone mark signatures, the genomic locations and calculated values were represented in bedGraph format with the window size of 1 kb. The values for individual genomic or chromatin features that overlapped with different RT classes were stored using intersectBed in the BEDTools suite (Quinlan and Hall, 2010). For computing the count of genes in each RT class, the GFF3 file format was used to identify the gene coordinates from the maize B73 AGPv3 annotation and was first computed into bedGraph format and then the same methods were used as above. FPKM values were also appended onto this bedGraph file for associating expression levels with genes overlapping with different RT classes. The median gene distance for genes found in each RT class was calculated by measuring the genomic distance from the 5′ and 3′ ends of each gene to the next nearest gene, regardless of the RT class of the nearest gene. The distance from elements within the top six most abundant LTR-retro families to the nearest gene was calculated using closestBed (parameter –d) using BED formatted coordinates for both features.

Permutation Analysis

A permutation or feature randomization test, similar to that described previously (De and Michor, 2011; Bartholdy et al., 2015), was used to assess the statistical significance of the observed overlap values between RT segment classes and other features. To test the significance of the enrichment of various features (including genes, LTR-retro families, histone marks, and MNase HS regions) in each of the RT segment classes, the RT segments were randomly shuffled (Supplemental Table 2). To test the reverse relationship, namely, the significance of the enrichment of RT segment classes in each feature, the feature was randomly shuffled (Supplemental Table 3). To do this, shuffleBed in the BEDTools suite was used (Quinlan and Hall, 2010) with default parameters to generate 1000 random genomic location lists as a null distribution for each feature or RT class, preserving the number and size of the original intervals. Several shuffle parameters were tested in shuffleBed to determine if the significance outcomes were consistent, irrespective of randomization strategy. These parameters included allowing or disallowing overlaps and excluding “bad spots” in the genome assembly (see replication timing data analysis). None of these parameters on their own produced different significance outcomes; however, the combinations of several of these parameters did produce different outcomes because they greatly limited the shuffle step for larger features or segments. Thus, for the final tests, the default parameters (overlaps allowed, no other exclusions) were used, which impose minimal assumptions on the null distributions (De et al., 2014). The percentage overlap between the two original data sets was calculated as the test statistic (observed value) using intersectBed. An empirical P value was estimated by calculating the proportion of N data sets (n = 1000 shuffled + 1 observed value) with a percentage of overlap value greater than or equal to the observed percent overlap (Ernst, 2004). Permutation P values of 0.001, indicating that none of the randomly shuffled data sets had a percentage of overlap value greater than or equal to the observed value, were accepted as evidence for enrichments significantly greater than expected by random chance.

Accession Numbers

Processed data files formatted for the IGV, as well as a pregenerated IGV session containing the data files, are available for download from the CyVerse Data Store (previously iPlant Collaborative; Merchant et al., 2016) via the links listed in Supplemental Table 5. Sequence data from this article can be found in the NCBI Sequence Read Archive (SRA) under the umbrella accession number PRJNA335625. The SRA numbers for each experiment are listed in Supplemental Table 6.

Supplemental Data

• Supplemental Figure 1. Flow cytometric sorting of nuclei and assessment of purity.

• Supplemental Figure 2. Correlation of Repli-seq biological replicates.

• Supplemental Figure 3. Bioinformatic analysis of Repli-seq data using Repliscan.

• Supplemental Figure 4. Distribution of replication activity on chromosome arms.

• Supplemental Figure 5. Replication activity on individual chromosome arms.

• Supplemental Figure 6. Replication intensity profiles for all ten maize chromosomes.

• Supplemental Figure 7. The percent GC content distribution varies little between RT classes.

• Supplemental Figure 8. Replication time and genomic distribution of individual LTR-retrotransposon families.

• Supplemental Figure 9. Replication time of chromosome two centromere.

• Supplemental Figure 10. Histone mark genomic distribution and transcriptional activity of windows with histone mark signatures.

• Supplemental Table 1. Sequence mapping statistics for Repli-seq and companion data sets.

• Supplemental Table 2. Percent of RT segment classes overlapping with features and corresponding permutation P values.

• Supplemental Table 3. Percent of features overlapping with RT segment classes and corresponding permutation P values.

• Supplemental Table 4. Histone mark called peak region summaries.

• Supplemental Table 5. Processed data availability information.

• Supplemental Table 6. SRA accession numbers.

Glossary

TTRs

timing transition regions

EdU

5-ethynyl-2’-deoxyuridine

BrdU

5-bromo-2’-deoxyuridine

AF-488

Alexa Fluor-488

DAPI

4′,6-diamidino-2-phenylindole

IGV

Integrative Genomics Viewer

TE

transposable element

RT

replication time

FGS

filtered gene set

FPKM

fragments per kilobase of transcript per million mapped reads

LTR-retros

long terminal repeat retrotransposons

ChIP-seq

chromatin immunoprecipitation sequencing