METHODS article

Front. Epigenet. Epigenom., 28 October 2026

Sec. Epigenomic Tools

Volume 4 - 2026 | https://doi.org/10.3389/freae.2026.1968465

Unpacking chromatin accessibility with Fiber-seq: general principles with special considerations for plant systems

  • Department of Genome Sciences, University of Washington, Seattle, WA, United States

Abstract

Chromatin accessibility has long been used as a marker for regions of DNA with regulatory potential. Fiber-seq detects chromatin accessibility on individual DNA fibers, enabling analyses beyond the identification of the accessible chromatin regions (ACRs). By providing single molecule level high resolution, Fiber-seq provides unprecedented qualitative descriptions, including potential categorizations of ACRs, identification of internal transcription factor footprints and nucleosome positioning within individual DNA fibers. As with all tools, the power of this technique depends on careful experimental design and data analysis -- incorrect usage will result in incorrect conclusions. Here we offer guidelines and flag potential pitfalls when generating and analyzing Fiber-seq data, such as (1) the optimum levels of adenine methylation per-fiber, (2) the power of per-fiber state inference, (3) the importance of controlling for read depth and methylation rates when comparing across samples, (4) the limitations of long-read sequence mapping, and (5) suggestions for identification of differentially accessible peaks across samples.

1 Introduction

Enzymes like DNase I, Tn5, and Hia5, which preferentially modify non-nucleosome-bound DNA, have proven invaluable to delineate genomic regions accessible to regulatory elements and infer protein-DNA interactions that govern regulation of gene expression. For the first few decades, chromatin accessibility assays were performed on bulk tissue, with accessibility defined as a binary state. However, even with these older accessibility assays, it was apparent that there were differences in both degree and quality of accessible states. The development of methods to purify single cell-types (Deal and Henikoff, 2011) and subsequent advancements in single-cell profiling technologies (Baek and Lee, 2020; Wang et al., 2025) indicated that this variability was not simply due to a mixture of cell types, but an inherent property of chromatin accessibility. The first attempts at measuring single-molecule chromatin accessibility used a GpC methyltransferase to identify accessible regions but were limited to short-read Illumina sequencing (Kelly et al., 2012). The emergence of long-read technologies allowed for the development of methods to observe the entire genome, including repetitive regions, such as SMAC-seq (Shipony et al., 2020) and Fiber-seq (Stergachis et al., 2020). Similar methods have been tailored to specific applications. For example, SAMOSA (Abdulhay et al., 2020) enables users to examine nucleosome states and potential transcription factor interactions in high resolution, while Fiber-TEnCATS (Pavlovic et al., 2026) allows for targeted enrichment of transposon sequences. Reviews comparing these techniques and their utility for addressing various biological questions can be found elsewhere (Hook and Timp, 2023; Krebs et al., 2026).

Fiber-seq (Stergachis et al., 2020) outperforms prior methods by measuring chromatin accessibility on long (>15 kb) single molecules at much higher per-bp resolution than previous methods. Compared with ATAC-seq, Fiber-seq offers three major advantages. First, it enables the analysis of genomic regions that cannot be mapped using short reads, such as long repetitive regions. Second, it reveals the existence of multiple classes of ACRs across the genome, including those with short, irregularly positioned regions of accessibility on underlying single molecules and others with longer, consistently positioned ones. Third, it reveals high-resolution details about individual ACRs, for example, enabling the detection of transcription factors and nucleosome footprints on each underlying molecule. Recent studies suggest that ACRs that appear unchanging at an accessibility level can nonetheless change dramatically in terms of transcription factor occupancy (Hu et al., 2025). However, this higher resolution also increases the potential for new artifacts. For example, within a single Fiber-seq experiment, the degree of adenine methylation varies widely across fibers. In addition, the distribution of per-fiber methylation often differs between Fiber-seq experiments. These technical differences must be accounted for to avoid conflating them with biologically meaningful differences.

Previously, we characterized the accessible chromatin landscape of the maize reference inbred strain B73 (Bubb et al., 2025) using the Fiber-seq method (Stergachis et al., 2020) and associated tools. We described a variety of types of accessible regions, discovered hundreds of individual LTR retrotransposons with persistent internal promoter and enhancer accessibility, and showed that regions with diffuse chromatin accessibility were more likely to be infiltrated by transposons. We are now in the process of extending our analysis to compare accessibility across trios consisting of two inbred parental lines and their F1 offspring. Our aim is to determine how chromatin accessibility patterns associated with a given genome in a particular tissue change in a different trans-environment. By comparing the chromatin accessibility patterns of the parental inbreds with the F1 hybrid, we also aim to gain insights into the mechanism of hybrid vigor. This task presents new challenges. Unlike with human samples, in which the two genomes (haplotypes) in each diploid sample are methylated in the same experiment and then computationally resolved (Vollger et al., 2025a), we can prepare essentially single genomes in inbred parental lines and compare them to that same entire genome in the F1 to understand the influence of trans-acting factors on chromatin accessibility. A main motivation in preparing this manuscript is to preemptively identify and account for potential technical artifacts that might otherwise be mistaken for biologically meaningful differences.

In this paper, we will (i) highlight how sample preparation quality interacts with important properties of the analytical algorithms and can influence the conclusions drawn, (ii) point out potential sources of false negatives when attempting to identify regions of differential accessibility across samples, and (iii) propose strategies to identify regions with truly differential accessibility across samples. We will not be describing the fibertools and FIRE algorithms that enable initial processing of raw Fiber-seq data, as those details are published elsewhere (Jha et al., 2024; Vollger et al., 2025b). In many ways, this manuscript is similar to our previous analysis of the properties of ATAC-seq (Bubb and Deal, 2020). By outlining common pitfalls and suggesting practical recommendations, the paper aims to promote Fiber-seq as a powerful tool to advance our understanding of the dynamics of chromatin accessibility across diverse biological systems.

2 Initial metrics for evaluating Fiber-seq sample quality differ from those used in short-read accessibility assays

Like ATAC-seq, Fiber-seq is used to identify genomic positions of accessible chromatin regions (ACRs). Unlike ATAC-seq, however, Fiber-seq also generates rich accessibility descriptions of the long individual DNA molecules contained within a single experiment (Vollger et al., 2025b). A user new to Fiber-seq might first be tempted to examine whether it is feasible to simply bulk the raw signal (m6A in this case) and call peaks as is done with previous end-sequencing methods, such as MNase-seq, DNAse-seq, ATAC-seq. We therefore wish to highlight why this is not a good approach and why the fiber-level state inferences are essential for proper peak calling. Furthermore, because of the differences between the two assays, it is not appropriate to use the same quality metrics when evaluating newly prepared samples.

End-sequencing chromatin accessibility methods rely on the generation and sequencing of small fragments (generally ∼200 bps) where each end of the fragment is accessible to the enzyme used in the assay (Figure 1A, top). The enzyme used in Fiber-seq creates single nucleotide modifications which are then detected via long-read sequencing, and thus detection of one modification does not depend on the presence of another one nearby (Figure 1B, top). In end-based assays the first raw data that we can directly observe is the per-base count of ends, most of which are concentrated in regions where two ends are approximately 200 bp apart (Figure 1A, middle). In Fiber-seq, the first raw data that we can directly observe is methylated adenines on individual DNA fibers, some of which are in >200bp accessible regions and some of which are in linker spaces between nucleosomes (Figure 1B, middle). If we simply make a histogram of the raw data from each assay within this 5-kb window, the end-sequencing data produces much cleaner peaks of accessibility than Fiber-seq (Figures 1A,B, bottoms).

FIGURE 1

Because Fiber-seq generates single-nucleotide signal (m6A) in inter-nucleosome linkers, metrics that are typically used to report sample quality, such as a Fraction of Reads in Peaks (FRiP) or the less peak-caller-sensitive Fraction of Reads in TSSs (FRiT), do not accurately reflect the utility of the sample (Figure 1C). For example, the human sample shown (GM12878) was used to develop the FIRE pipeline (Vollger et al., 2025b) and yet, in aggregate, has no more signal in the TSS region than expected due to chance. Although some samples do have FRiP scores substantially exceeding that expected due to chance, notably the previously published maize sample (Bubb et al., 2025), optimizing only for FRiP score favors samples with very low levels of methylation and would result in an inability to detect other important features, such as nucleosomes and other protein footprints.

Another metric that can be derived from raw data alone is whether the spacing between methylated adenines on the same fiber indicates the presence of nucleosomes. This metric has the added advantage that it requires very few reads -- typically just a few thousand -- and so can be performed with small sample quantities prior to expensive deep-coverage sequencing. We use an autocorrelation plot of all methylated adenines within a distance (Lag) of 2 kb to visualize nucleosome and linker positions, the local minima and maxima, respectively (Figure 1D, left). The amount of vertical fluctuation in the autocorrelation plot due to nucleosome positioning can be defined as the nucleosome “Signal” (area between the smoothed autocorrelation curve and the y = 0 line), and the vertical fluctuation due to anything else is the “Noise” (area between the observed and smoothed autocorrelation curves, seen in inset, Figure 1D, left). The ratio between these two is called the Chromatin Signal-to-Noise (CSN). We have found that a CSN ratio above 25 indicates that the sample contains both intact chromatin and sufficient adenine methylation. Because identifying nucleosomes is secondary to our primary goal of identifying accessible regions, the aim should not be to maximize this metric but simply be above the aforementioned threshold, while preserving sensitivity and specificity in ACR detection. As indicated in the table (Figure 1D, right), CSN peaks just after the upper end of the recommended 2–8% range, while nucleosome periodicity is unaffected by %m6A.

Finally, a feature of the data that can be seen with Fiber-seq but not with bulk end-sequencing methods is that, within a sample preparation, the per-fiber adenine methylation rate deviates greatly from a null expectation that the Hia5 methyltransferase is methylating adenines on each fiber with equal probability -- even at the same genomic location. Although there is no way of knowing the range of transposition across different genomes/cells in bulk ATAC-seq experiments, this finding is analogous to, but not as extreme as, the variation in ATAC-seq reads obtained in different cells in single-cell experiments. For example, in the window shown in the middle panel of Figure 1B, the most highly methylated fiber has more than 10x as many methylated adenines than the least-methylated fiber. If accessible adenines were methylated at random with respect to fiber, we would expect a much tighter distribution of methyl-adenines per fiber. This discrepancy can be quantified by comparing the number of m6As observed per fiber to the number of m6As expected per fiber, given that methylations occur at random with respect to fiber. Such a comparison between the observed number of m6As per fiber and the expected can be visualized in a Q-Q plot, after normalization of these numbers into z-scores (Figure 1E, black dots). Conveniently, the z-score conversion also normalizes for locus-specific differences in methylation level, allowing us to plot multiple loci on the same Q-Q plot (Figure 1E, purple hexplot). The observed range of z-scores is at least 20x greater than that expected if m6As were distributed randomly across fibers. It remains unclear to us why the level of methylation is so variable across fibers at a single genomic location within a single sample, but we suspect that it is related to the fact that, to preserve the native chromatin state, nuclei must be handled very gently (i.e., cannot be mixed vigorously), and therefore the Hia5 enzyme may not be evenly distributed amongst the nuclei suspension.

Perhaps related to this per-fiber unevenness in methylation levels, we have found it difficult to obtain consistent methylation rates between different preparations of plant samples, even when using the same reaction conditions (e.g., units of methyltransferase, number of protoplasts, incubation time). This might be in part due to the extra care required to obtain nondegraded, high molecular weight DNA from plant tissue. Direct nuclei preps from frozen tissue have resulted in highly degraded DNA and from fresh tissue have limited scale due to the DNA loss during size selection and from additional clean up steps to remove excess polysaccharides and plant contaminants that can interfere with size selection and library preps. To obtain high quantities of quality plant DNA we protoplast, or enzymatically remove the wall from live cells, then chemically denature the plasma membrane and gently spin to release the nuclei (Bubb et al., 2025). Since we standardize the number of protoplasts prior to nuclei extraction, there may be more variation in the amount of DNA exposed to the methyltransferase between different Fiber-seq preps compared to the mammalian studies that directly extract and normalize the number of nuclei. We solve this problem by preparing multiple samples and evaluating sample quality on a pilot set of only tens of thousands of Fiber-seq reads before collecting the millions of reads necessary for a full analysis.

3 The effect of adenine methylation level on inferred fiber-level and genomic chromatin states

The Stergachis and Noble labs developed tools to infer multiple types of chromatin states from the raw adenine methylation data. The fibertools pipeline (Jha et al., 2024) begins by using the positions of enzymatically methylated adenines to infer nucleosome positions (unmethylated regions) and methylation sensitive patches (MSPs) on individual fibers. The FIRE pipeline (Vollger et al., 2025b) can then be used to label certain MSPs as Fiber-seq Inferred Regulatory Elements (FIREs) on individual fibers, with the remainder labeled as linkers between nucleosomes (Figure 2A).

FIGURE 2

The genomic position of FIRE peaks (ACRs) is inferred from the combination of information provided by fibers underlying that location. Therefore, it is important to understand how methylation levels affect these inferred fiber-level states. For illustration, we display single fibers with varying methylation rates spanning a 3-kb region (Figure 2B, Example fibers from %m6A strata). We then plot the length distributions of inferred nucleosomes (Figure 2B, Nucleosomes), MSPs (Figure 2B, MSPs), and FIREs (Figure 2B, FIREs) on same-sized batches of DNA molecules (fibers) with differing percentages of adenines methylated taken from a single sample preparation. As the percent of adenines methylated increases, we observe (1) the inferred nucleosome sizes decrease, while MSP and FIRE sizes increase, (2) the emergence of a second mode of extra-large (>175-bp) MSPs, and (3) disappearance of the first mode of small FIREs (<125 bp in length), which have been shown in maize to comprise a category of accessible regions largely undetected by ATAC-seq in a tissue-matched sample, yet coincide with ATAC-seq peaks in other tissues (Bubb et al., 2025).

As suggested in Figure 2A, some of the large (>175-bp) MSPs are converted to FIREs. However, especially in highly methylated fibers, many of the large MSPs are not reflective of genomic regions of accessibility but instead are cases where adenine methylation occurs on DNA that is wrapped around a nucleosome, resulting in merging of linkers across nucleosomes (Figure 2C, long MSPs containing asterisks). These erroneously large MSPs become statistically conflated with large MSPs at true sites of accessibility and result in undercalling of true FIREs (Figure 2C, red outlined MSPs). To quantify the fraction of large MSPs in regions of true genomic chromatin accessibility (i.e., those that should be converted to FIREs), we calculated the Fraction of Large-MSP-bps in Peaks (FLiP score), using only well-supported FIRE peaks, having 50% or more underlying fiber-bps called as FIRE (Figure 2B, Fraction of Large MSPs in Peaks). To ensure high FLiP scores and to adequately detect biologically significant small FIREs, we conclude that fibers with 2–8% of adenines methylated provide the most accurate and rich FIRE information.

To highlight some of the genomic features that are detectable with Fiber-seq, we plotted the nucleosome density and degree of nucleosome phasing (Lee et al., 2019) in 1-kb windows along the maize (B73) genome. The vast majority (95%) of 1-kb windows had a nucleosome density greater than 96% and phasing less than 0.1, a quarter of maximum value (Figure 2D, marginal histograms). A hexplot comparing the two metrics reveals that phasing peaks when nucleosomes are at 80% density. Note that the z-axis is on a log scale, so that regions of the plot with few 1-kb instances can be examined. Five example 1-kb windows are shown to illustrate the chromatin feature differences that these metrics reflect. The two high-density windows, outlined in light blue and green and representative of the vast majority of the genome, have few fiber-bps classified as MSPs (purple). The phased window, outlined in pink, contains five columns of stacked MSPs. In the window containing an ACR, outlined in red, most of the MSPs are long and have been converted to FIREs. The window outlined in yellow is part of a larger locus with diffuse accessibility, reported previously as being regions prone to transposition (Bubb et al., 2025).

4 The effect of adenine methylation levels on peak-calling quantity and quality

Next, we examine the factors affecting genomic-level FIRE peak-calling. To determine whether overall sample methylation rates affect the number of FIRE peaks called, we merged data from two human cell line samples that were treated with either low or high (Figure 3A, top and bottom, respectively) levels of methyltransferase. Note the abundance of reads with very low levels of methylation (<2%), a phenomenon common in plant preparations as well and one that we believe is also related to the necessity for gentle mixing of the methylase with the nuclei. A similar number of FIRE peaks were identified in the low- and high-methylation samples (i.e. 67 thousand vs. 68 thousand, respectively) using the same number of reads from each group (i.e. 4.27 million), suggesting that variation in methylation level does not greatly affect the number of peaks called (Figure 3B, left). Furthermore, the quality of the peaks, defined by the distribution of FIRE accessibility scores per FIRE peak, were remarkably similar (Figure 3B, right). However, only 67% of peaks in either sample overlapped a peak in the other sample by at least 100 bp -- and only 38% of the FIREs in either sample overlapped a FIRE in the other sample -- suggesting that methylation level does affect which loci are called as peaks. Peaks found in the low-methylation sample but not the high-methylation sample appear to be the result of false negative FIREs (large MSPs not converting to FIREs) (Figures 2C, 3C, left), while peaks found in the high-methylation sample but not the low-methylation sample appear to be the result of false positive FIREs (long MSPs erroneously converting to FIREs) (Figures 2C, 3C, right). Local low coverage in one or the other sample did not seem to account for much of these differences, as the distribution of x-coverage for the lower of the two samples at these peaks was similar (Figure 3D). While the Fiber-seq analysis pipeline is robust over a wide range of methylation levels (Vollger et al., 2025b), the high-methylation sample shown here lies outside that range.

FIGURE 3

Read depth, on the other hand, strongly affects both the number and quality of FIRE peaks called. This is expected for any significance-based peak caller, such as the commonly-used MACS2 for ATAC-seq (Gaspar, 2018; Bubb and Deal, 2020), however it is striking when quantified. We used downsampling to examine the relationship between read depth and number and quality (Figure 3E, top and bottom, respectively) of called FIRE peaks, with quality defined as the fraction of total underlying fiber bps labeled as FIREs (FIRE accessibility score). As expected, while the number of FIRE peaks increases with read depth, the fraction of FIRE peaks with low FA score increases, suggesting that above a certain depth, only transiently or partially accessible ACRs are identified -- along with a certain number of false positives. The reason for this, in short, is that FIRE peaks are called as a function of both the quality (p-value assigned by the fibertools algorithm, reflected in degree of “redness,” with darker red having a higher p-value), and number of FIREs underlying that location and are, therefore, sensitive to local read depth. The number of reads collected and analyzed per sample should be guided by the biological features of interest. For example, detecting transcription factor footprints generally requires greater sequencing depth than identifying accessible chromatin regions (ACRs) or nucleosome positioning. The appropriate depth also depends on the acceptable level of false positives. In Section 5, we provide recommendations for mitigating issues associated with peak calling in low-signal regions. Regardless of the application, comparisons across samples should be performed using similar numbers of reads.

5 Comparing Fiber-seq results across divergent genomes

The above analyses establish a baseline of how the FIRE pipeline acts on samples of varying methylation levels and read depths. We next turn to how best to compare across samples with different genomes.

Although Fiber-seq was initially developed in Drosophila (Stergachis et al., 2020), much of the more recent use has been in human tissue. Our lab has further expanded the use of Fiber-seq to maize (Bubb et al., 2025), and we are now expanding to other plant species as well as multiple within-species strains and their hybrids. With this data we can compare accessibility across homologous regions as well as examine the effects of trans-environments on accessibility for identical underlying sequences, specifically how genome accessibility changes in an F1 context. Maize and humans have very different evolutionary histories, but there are certain similarities. Humans have experienced repeated bottleneck and founder effects (Tournebize et al., 2022; Ebert et al., 2021; Liao et al., 2023). Maize, as a domesticated species, has also undergone several population bottlenecks, and yet domesticated maize retains a surprising level of genetic diversity. In a comparison of 26 diverse maize inbred lines, only about 60% of the genes are shared among all lines (Hufford et al., 2021). This diversity is at least partially due to active transposition, leading to transposon-level presence/absence polymorphisms and larger-scale structural variation (Buckler et al., 2006; Munasinghe et al., 2023), as well as persistent introgression by its wild ancestor, teosinte (Yang et al., 2023). The suggestions we propose below were developed for maize but are generally applicable for any cross-genome comparisons.

First, when collecting Fiber-seq reads from hybrid individuals, we can only compare accessibility of the haplotypes in regions where the Fiber-seq reads are uniquely mappable. To ensure that fibers are mapped to their correct genome of origin, we align our reads to a compound genome consisting of both parental genomes concatenated and maintain only uniquely mappable reads (mapq = 60 using pbmm2). For consistency, we also map fibers from parental lines to this compound genome, even though we know that all fibers originate from this genome. We determined the fraction of the genome that is uniquely mappable by our long-reads by simulating reads from an in silico F1 genome, using read lengths and per-base error distributions empirically derived from previous Fiber-seq reads, and mapping them to either parental-only or compound genomes. We then report the percent of the genome uniquely mappable for each parent genome with either mapping strategy (Figure 4A). For all maize strain pairs examined, only 70–80% of either parental genome is uniquely mappable by F1 reads to an F1 (compound) genome. Note that this fraction can vary for the same parent when it is paired with a different parent, depending upon how similar the genomes of the two co-parents are. For example, LH244 has better alignment with its parental genome when concatenated to Ki3 vs. Oh43. In contrast, more than 95% of the genome is mappable if simulated reads are mapped to parental-only genomes rather than compound genomes.

FIGURE 4

We then determined whether different types of base pairs were more likely to be in uniquely mappable regions, looking specifically at LH287 in the LH244 x LH287 parental pair. While 76% of all genomic bps were uniquely mappable, less than 40% of LTR retrotransposon bps were in uniquely mappable regions. Transcribed or accessible (FIRE peak) bps were more likely to be in uniquely mappable regions as compared to all genomic bps (Figure 4B).

Second, presence/absence polymorphisms, such as transposon insertions, structural variation, or copy number variation, should not be mistaken for accessibility differences. We soon noticed that although the genomes in our sequenced inbred lines and F1s were nominally the same (e.g., P1 = B73; F1 = B73xMo17; P2 = Mo17) there were genomic presence/absence differences between them. For example, an LTR retrotransposon is present in the mapped-to reference B73 (v5) and the sequenced parental B73 but not in the F1 B73 genome (Figure 4C, yellow highlighted region). In some cases, regions missing in the parental version, but not the F1 version, of the Mo17 genome were quite large--on the scale of megabases (Figure 4D, yellow highlighted region). Note that regions lacking coverage in both the parental and F1 Mo17 genomes are likely not uniquely mappable in the B73xMo17 compound genome (Figure 4D, blue highlighted region). In general, if the specific strain that we sequence is missing genetic material present in the mapped-to reference sequence, there will be no Fiber-seq coverage of those regions. Thus, there are two broad reasons for missing coverage: (i) lack of unique mappability, and (ii) the sequenced genome is missing a part of the mapped-to genome (Figure 4E, blue and yellow shading, respectively). Such regions should not be included in lists of differential accessibility.

Third, reads should be aligned to genomes that include organellar genomes, to prevent ectopic alignment. Substantial stretches of the nuclear genome have sequence homology to organellar genomes, and fibers originating from organellar genomes should be allowed to align to their true genome of origin. Organellar genomes typically have completely accessible chromatin and are present at much higher copy number than the nuclear genome. Ectopic mappings can introduce both (i) false positives, because the accessibility and coverage of fibers matching such regions are much higher than nuclear-derived fibers, and (ii) false negatives, because this region may indeed contain accessible regions that should not be ignored or discarded. For example, when we align Fiber-seq reads to only the nuclear genome, regions with homology to chloroplast genome have more than 10x higher coverage than the rest of the genome and are correctly flagged as regions of “unreliable coverage” by the FIRE pipeline (Figure 4F). To prevent this erroneous mapping, the chloroplast and mitochondrial genomes should be appended to the compound nuclear genomes before alignment, allowing fibers originating from the organellar genomes to uniquely map to their correct location.

Fourth, partial alignments should be removed prior to running the FIRE pipeline, as they can introduce another source of artifact to both FIRE peak calling and FIRE accessibility score quantification. For example, sections of fibers deriving from the parental LH287 genome align to regions immediately flanking an intact LTR retrotransposon in the parental LH244 genome (Figure 4G). Such alignments can and should be removed early in the analysis pipeline.

6 Detecting differentially accessible FIRE peaks

Finally, after appropriately normalizing and controlling for artifacts, we can begin looking for truly differentially accessible FIRE peaks across samples.

FIRE peak calling is a useful way to identify the small fraction of the genome that is above some threshold of accessibility. However, among regions identified as FIRE peaks, there are different degrees and types of accessibility. For example, some called FIRE peaks cover more contiguous bps than others. Some FIRE peaks consist of smaller per-fiber FIREs that do not occupy the exact same sequence, which can result in large FIRE peaks. Some FIRE peaks have a smaller fraction of underlying fibers that contain FIREs. We are only beginning to understand the significance of these differences. Likewise, when identifying differential FIRE peaks, one should use metrics that are sensitive to these state differences and not simply rely on peak presence/absence.

To illustrate these points, we are comparing accessible regions between the B73 genome in the parental inbred context and in an F1 (B73 x Mo17) context. However, a similar approach could be used to compare accessibility between homologous regions in B73 and Mo17 parentals or across any pair of genomes. The difference between genomes at each union FIRE peak can be described with an effect size and a significance score. Using both the effect size (absolute difference in either per-base or per-fiber FIRE proportions between compared samples) and significance (p-value from a two-proportion Z-test, comparing either the proportion of bps called as FIREs or the proportion of fibers containing FIREs in one sample vs. the other), we can identify a spectrum of differential accessibility across samples at each peak (Figures 5A,B). Effect size and significance, while correlated, do not provide identical information. Three examples of FIRE peaks, each shown within their 3-kb window contexts, are displayed to highlight the features that can affect these four metrics.

FIGURE 5

First, union peaks with high fiber-bp coverage (many fibers, long union peak) and differences primarily in the size of called FIREs will have moderate per-bp and per-fiber effect sizes and high significance in the per-bp proportions test but not in the per-fiber proportions test (Figures 5A–C). Additionally, such regions are likely to be identified as FIRE peaks in both samples. Intermediate accessibility levels may reveal a snapshot of the transition from accessible to inaccessible, or vice versa. For example, we previously used Fiber-seq to identify FIRE peaks within long terminal repeat retrotransposons (LTRRTs). LTRRTs with two peaks in each LTR, putative enhancers and promoters, had higher per-bp accessibility and low rates of CpG methylation, whereas LTRRTs with only a single peak per LTR had both lower per-bp accessibility and higher rates of CpG methylation (Bubb et al., 2025).

Second, small union peaks with a high proportion of fibers containing FIRE in one sample and not the other will attain a high significance score in the per-fiber proportions test. The per-fiber test statistic is not affected by the size of the peak, so excels at finding differences between narrow peaks (Figures 5A,B,D). Third, union peaks with low fiber coverage will have low significance by the per-fiber proportions test but may have higher significance by the per-bp proportions test if the peak is large (Figures 5A,B,E).

Finally, and perhaps most importantly, peak presence/absence is generally a poor method of identifying regions with true accessibility difference between samples, even for samples with equal read depth. For example, two such regions are present in our windows (Figures 5C,D, open circle and open triangle). They have low values for all four metrics (Figures 5A,B). Therefore, it is important to critically assess differential peak types to fully harness the power of Fiber-seq in identifying differential biological contexts rather than oversimplifying to binary peak calling.

In summary (Figure 5F), we recommend comparing samples with similar per-fiber rates of adenine methylation, ideally with the majority of fibers having a percent of adenines methylated within the range of 2–8%. In our hands, it has been useful to prepare multiple samples and sequence them at low coverage in order to determine if the adenine methylation rates and nucleosome size distributions are within the optimal range before obtaining the full coverage. Next, if analyzing hybrids, align to a compound genome that consists of both parental genomes, ideally assembled de novo from the Fiber-seq data itself, rather than a published genome, as well as any plastid genomes (i.e., chloroplast and mitochondrial genomes. Remove partial and organellar alignments, then randomly subsample reads such that each sample has the same genomic coverage, with the hybrid sample having roughly twice as many reads as either parent to obtain similar coverage over each parental genome. Next run fibertools (Jha et al., 2024) to infer nucleosomes and MSPs, and then the FIRE pipeline (Vollger et al., 2025b) to call FIREs on individual fibers and FIRE peaks at genomic locations. Generate a set of union FIRE peaks by merging peaks called on each sample to compare (e.g., using bedops (Neph et al., 2012) or bedtools (Quinlan, 2014)). Finally, define thresholds for effect size (e.g., absolute difference in FIRE accessibility) and p-value (e.g., using a two-proportion Z-test) that identify the type of differential accessibility that is of interest to you.

7 Methods

7.1 Plant tissue processing and Fiber-seq reactions

Protoplasts were isolated from plant tissue as described previously (Tonnies et al., 2023) using up to 1.5 g of tissue per 20 mL of enzyme solution. Aliquots of 3 M protoplasts were pelleted at 100 × g for 3 min at room temperature and the pellets were resuspended in 300 µL working buffer (15 mM Tris-Cl, pH 8, 15 mM NaCl, 60 mM KCl, 1 mM EDTA pH 8, 0.5 mM EGTA pH 8, 0.5 mM spermidine). Protoplasts were lysed by adding a final concentration of 0.2% IGEPAL in working buffer and incubating on ice for 10 min. Nuclei were pelleted by spinning down at 350 × g for 3 min at 4 °C. The nuclei pellets were resuspended in 100 µL working buffer supplemented with 0.6 mM S-adenosyl methionine and 200 u of Hia5 and incubated for 10 min at 25 °C. The reaction was stopped by adding SDS at a final concentration of 1%, and the DNA was purified using the Promega Wizard High Molecular Weight DNA Extraction Kit (Promega A2920). High Molecular Weight DNA was submitted to the University of Washington PacBio Sequencing Services for library preparation using PacBio HiFi SMRTBell Prep kit v3 with PippinHT size selection and sequencing on a Revio SMRT Cell using Revio SPRQ chemistry retaining full kinetics data.

7.2 Fiber-seq data processing

Denovo assemblies of genomes were generated using hifiasm (Cheng et al., 2021; 2022; 2024). Sequencing data was aligned to the genome using pbmm2 (https://github.com/PacificBiosciences/pbmm2). Filtering of non-uniquely mapped reads (mapq<60), partial mappings and reads mapping to plastids was performed with samtools (Li et al., 2009). Methylated adenines are called using Jasmine (https://github.com/PacificBiosciences/jasmine). Extensive documentation for running fibertools and the FIRE pipeline can be found on the respective github pages (https://github.com/fiberseq/fibertools-rs; https://github.com/fiberseq/FIRE).

Statements

Data availability statement

The original contributions presented in the study are publicly available. ATAC‐seq and Fiber‐seq sequencing data for the maize strain B73 are available from the NCBI Short‐Read Archive (SRA) under Bioproject PRJNA1119563 (https://www.ncbi.nlm.nih.gov/bioproject/?term=PRJNA1119563). ATAC‐seq and Fiber‐seq sequencing data for the human cell line GM12878 are available from the NCBI Short‐Read Archive (SRA) available under the Bioprojects PRJNA1233341 (https://www.ncbi.nlm.nih.gov/sra/SRX19461963) and PRJNA1233341 (https://www.ncbi.nlm.nih.gov/sra/?term=PRJNA1233341), respectively.

Author contributions

KB: Writing – original draft, Methodology, Investigation, Visualization, Software, Conceptualization, Validation, Writing – review and editing, Formal Analysis. MP: Data curation, Writing – review and editing. JC: Visualization, Resources, Writing – review and editing, Funding acquisition. CQ: Writing – review and editing, Funding acquisition, Resources.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This work was supported by the National Science Foundation (IOS Division of Integrative Organismal Systems grant no. 2519784 to CQ and PlantSynBio grant no. 2240888 to CQ), and the National Institutes of Health (NHGRI grant no. R01-HG013025 to CQ, and NIGMS MIRA grant no. 1R35GM139532 to CQ).

Acknowledgments

We thank Bing Yang (University of Missouri) for rice samples, Cinta Romay (Cornell University) for maize seeds, Mitchell Vollger (University of Utah), Shane Neph and the Stergachis Lab (University of Washington), Bryan Ramirez-Corona, Morgan Hamm and the Queitsch/Cuperus Lab (University of Washington) for helpful discussions, and Andrew Stergachis (University of Washington) for the human cell line data.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Summary

Keywords

analysis artifacts, chromatin, chromatin accessibility, Fiber-seq, nucleosomes, plants

Citation

Bubb KL, Perchlik M, Cuperus JT and Queitsch C (2026) Unpacking chromatin accessibility with Fiber-seq: general principles with special considerations for plant systems. Front. Epigenet. Epigenom. 4:1968465. doi: 10.3389/freae.2026.1968465

Received

14 August 2026

Revised

02 October 2026

Accepted

06 October 2026

Published

28 October 2026

Volume

4 - 2026

Edited by

Roger Deal, Emory University, United States

Reviewed by

Michael-Christopher Keogh, EpiCypher Inc., United States

Felix Mueller-Planitz, Technical University Dresden, Germany

Updates

Copyright

*Correspondence: Kerry L. Bubb,

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Article metrics