A semi-automated pipeline for high-throughput analysis of chromatin reporters

We aimed at high-throughput imaging of plant iPSC ancestors in culture, similar to what had been undertaken previously for analysing cellular morphology (Dawson et al., 2022), but focusing here on chromatin markers. For this, we generated a dual reporter Arabidopsis line expressing both the H1.2–GFP and H2B–RFP markers (selected following the strategy explained above), and used it to establish and benchmark the growth conditions, imaging setup and an image analysis pipeline. Details are provided in the Materials and Methods but, in brief, leaf protoplasts were cultured under sterile conditions in 96-well plates with a coverglass bottom for semi-automated imaging using Cell Voyager (Fig. 2A). The imaging set-up allows the capturing of up to 60 wells (i.e. 60 cultures) per multi-well plate, each being covered by six regions of interest (ROIs). Imaging of a full 60×6 ROIs takes 90 min. After each time point, the plate was returned to the temperature and light-controlled plant growth incubator until the next measurement. Image analysis, illustrated in Fig. 2B, consisted of a supervised, batch processing approach comprising the following steps performed in TissueMAPS (http://tissuemaps.org): (1) maximum intensity projection of the image series; (2) denoising; (3) watershed-based segmentation of H2B-RFP-stained nuclei and image masking using the segmented objects; (4) training, classification and filtering of segmented objects to improve nuclei segmentation; and (5) quantitative measurements using validated nuclei objects.

Open in a separate window

Fig. 2.

A semi-automated pipeline for high-throughput analysis of chromatin reporters in plant iPSC ancestor cultures. (A) Semi-automated microscopy imaging is carried out in cover-glass bottom multi-well plates containing cultures of protoplasts, plant iPSC ancestors, using a Cell Voyager platform. Six Regions of Interest (ROI) are randomly selected per well and imaged for each channel. The procedure is repeated for each time point, the culture being returned to the growth incubator in between two measurements. (B) Supervised image processing followed a workflow as depicted using TissueMAPS and enabled the segmentation of several hundred nuclei with high accuracy. The image analysis package returns several descriptors (features) of the segmented objects describing the morphology, the intensity distribution and texture (see details in the text). These metrics can be plotted or further analysed. (C,D) Comparison of the H1.2-GFP/H2B-H2B ratios obtained following automated (C) or manual (D) image analysis, showing a reproducible reduction of H1.2-GFP relative to H2B-RFP during culturing time (experiment HTI001; see Table S4). (E) Control experiment showing stable signals of a nuclear localised, free YFP marker, relative to H2B–RFP levels (experiment HTI002; see Table S4).

To benchmark the above approach, we cultured protoplasts expressing the dual markers in replicate wells and applied our imaging and image processing procedure between day 0 and day 7 (representative images are shown in Fig. S2A,B). The setup of replicate wells allowed the capture of a total of 500–1000 nuclei per time point, per initial culture. Using fluorescein diacetate (FDA) staining (Saruyama et al., 2013), we confirmed a high density of viable cells at each time points (Fig. S2C). However, we observed a decrease in the number of nuclei (cells) identified following the segmentation, particularly at day 7 (Fig. S2D), indicating loss of viability, consistently with previous reports (Chupeau et al., 2013; Xu et al., 2021). The analysis of GFP to RFP signal intensity ratios per nuclei confirmed a dramatic reduction of H1.2 abundance relative to H2B as soon as day 3 (Fig. 2C). This automated measurement compared very well to a set of manually segmented images used for the same intensity ratio measurement (Fig. 2D). Then, to assess the reproducibility of the culturing–imaging–image analysis workflow, we set up eight replicate measurements in the same multiwell plate with cultures from two independent reporter lines expressing H1.2–GFP and H2B–RFP. The signal intensity ratio distributions were highly consistent between wells at each day (Fig. S2E). Finally, as a control, we analysed cultures co-expressing a nuclear localised (NLS–YFP) and the same H2B–RFP internal control, using the same setup as before. The analysis showed a stable ratio during culturing (Fig. 2E; Fig. S2F).

In conclusion, we established a robust automated high-throughput imaging-image analysis workflow enabling the capture of several hundred nuclei per time point and which is suitable for quantitative measurements of chromatin patterns during culturing of plant iPSC ancestors. In addition, the quantifications confirmed that a decrease in H1.2 abundance is a hallmark of chromatin changes within the first 2 days of culturing.

Nuclei morphology and chromatin patterns change rapidly

Next, we exploited the numerous image features exported by the pipeline to analyse the cytological organisation of chromatin in plant iPSC ancestors, particularly during the first days of culturing corresponding to the dedifferentiation phase, before cells start dividing (Grafi et al., 2011). Three groups of metrics were produced from the analysis – signal intensity features, morphology features and texture features. This corresponds to a total of 193 metrics per channel and 370 for both channels (morphology features were derived from segmentation on the H2B–RFP signal only, Table S1). A principal component analysis (PCA; Metsalu and Vilo, 2015) that included all the features indicated clear changes within the first 2 days of culturing (Fig. 3A, a replicate is shown in Fig. S3A).

Open in a separate window

Fig. 3.

Marked changes in nuclear morphology and chromatin organisation during the dedifferentiation phase. (A–D) PCA of chromatin features measured in Arabidopsis leaf protoplasts expressing H1.2–GFP and H2B–GFP, imaged ~4 h after release (day 0), at day 2 and day 5 (experiment HTI004; see Table S4). (A) PCA computed on all features (Table S2), (B) PCA on H1.2–GFP and H2B–GFP intensity features, (C) PCA on morphology features (i.e. size and shape descriptors of the segmented nuclei) and (D) PCA on LBP texture features for H2B–RFP. See also Fig. S3A for replicate PCAs and Fig. S3B for the PCA loading scores per descriptors. The x- and y-axis show principal component 1 and principal component 2 that explain the given percentage of the total variance, respectively. Ellipses: 95% confidence interval. Each point represents a culture replicate (well). (E) Relative changes for each morphology descriptors during culturing. Heatmaps represent the median value for each descriptor, with unit variance scaling applied to rows. (F) Plots of selected morphology descriptors. Data is shown as a violin plot, with a box plot inside, where the box represents the 25–75th percentiles, and the median is indicated as a line. The whiskers show the minimum and maximum values, respectively. (G) Relative changes for LBP texture descriptors for the H2B–RFP signal distribution. Heatmaps represent the median value for each descriptor, with unit variance scaling applied to rows. The list of LBP descriptors (rows) is provided in Table S2. Rows are clustered using correlation distance and average linkage. Number of nuclei, n=1008, 508, 238 at day 0, 3, 5, respectively. ****P<0.0001; ns, not significant (Kruskal–Wallis test followed by post-hoc Dunn's test and Bonferroni correction).

To then estimate the contribution of each feature group, we carried out separate PCA. Clearly, each of the intensity, morphology and texture features contributed to the explanation of the chromatin organisation changes during the dedifferentiation phase (Fig. 3B–D; Fig. S3A).

The observation that intensity features distinguish cells at day 0, day 2 and day 5 indicates that the relationships between chromatin markers change rapidly in the early culturing phase. Consistent with our previous observation that cells barely divide within the first 5 days of culturing, we found that the H2B–RFP intensity distribution did not significantly change during this period of time, but that it had increased by day 7, probably indicating that cells had entered S phase (Fig. S3B). It is likely that the global decrease in linker histone abundance (H1.2–GFP), relative to total chromatin (H2B–RFP), as documented above, has a major contribution to the separation of chromatin features on this PCA. However, the standard deviation, and minimum and maximum intensity values also contribute the principal components (Fig. S3C) suggesting that, beyond the absolute levels of chromatin markers, their spatial distribution, reflecting the occurrence of chromatin regions with varying density and compaction, also change rapidly.

Morphology features are computed on the segmented H2B–RFP signal, and thus provide a proxy for nuclei size and shape (Fig. S3D). A PCA considering all morphology features indicated that nuclei undergo continuous size and shape changes between day 0 and day 5 (Fig. 3C,E). Interestingly, although an increase in nuclear size would be expected from the decreased abundance in H1 variants (Rutowicz et al., 2019; She et al., 2013), we did not detect a positive correlation, apart from a moderate one at day 5 (Pearson correlation r<0.3, Fig. S3E). Instead, the nuclear size distribution shifted towards smaller sizes at day 2 and day 5 (Fig. S3F). The median and the distribution of shape descriptors at day 5 also differed from that at day 0 (Fig. 3E), with, for instance, slightly rounder and less elongated nuclei at day 5 than day 0 (Fig. 3F).

Next, we interrogated the group of texture features, which showed a clear evolution during culturing (Fig. 3D). Texture metrics describe the spatial distribution of signal intensities as a function of scale (Depeursinge et al., 2017; Di Cataldo and Ficarra, 2017) and can be used to analyse patterns in chromatin organisation (Kerr et al., 2010; Lee et al., 2021; Rana et al., 2021). TissueMAPS returns metrics corresponding to four types of texture analysis: Gabor wavelet filter, local binary pattern (LBP), threshold applied statistics (TAS), and Hu invariant moment (Hu) (reviewed in Di Cataldo and Ficarra, 2017; Hamilton et al., 2007; see the list in Table S2). The LBP analysis, based on neighbouring pixel intensity scanning in incrementally growing circles, is particularly interesting as a proxy of chromatin organisation patterns at different length scales (Fig. S3G) (Di Cataldo and Ficarra, 2017; Rana et al., 2021). We detected a rapid, global decrease in LBP values along the different radii for the H2B–RFP signal (Fig. 3G) suggesting a decreasing heterogeneity in chromatin distribution at day 2. The texture metrics of the H1.2–GFP signal also showed rapid changes in iPSC ancestor chromatin (Fig. S3G). These dynamics in H2B–RFP and H1.2–GFP textures, reflecting the distribution pattern of these histone variants, were confirmed with the TAS and Gabor filter methods (Fig. S3H,I). Furthermore, we detected modest, but consistent positive correlations between the LBP metrics and the H1.2–GFP-to-H2B–RFP ratio, indicating that the changes in chromatin distribution patterns are likely to be linked with the relative abundance of linker and nucleosome histones (Fig. S3J).

Collectively, the analysis of nuclei morphology and of H2B–RFP signal distribution indicate clear changes in nuclear size, shape and in chromatin organisation in the dedifferentiation phase of plant iPSC ancestors (Grafi et al., 2011). These changes occur largely within the first 2 days and are concurrent, but not correlated, with a decrease in the relative abundance of linker histone H1.2. The distinct texture of H2B–RFP distribution at different length scales at day 5 compared to day 0 suggests a reorganisation of chromatin domains at the (sub)micrometre scale.

Plant iPSC ancestor cultures show a marked heterogeneity in their chromatin features, which reduces over time

The above analysis clearly showed a spread distribution of intensity, size and shape descriptors, indicative of a vast heterogeneity of nuclei type, despite the fact that leaf-derived protoplasts consist of ~85–90% mesophyll cells (Xu et al., 2021). To assess this heterogeneity, we first generated density distribution maps of chromatin features summarised by the principal components (PCs) computed previously. The maps confirmed a broad dispersion of the data in the PC landscape, indicating a very heterogenous population in terms of nuclei type and chromatin patterns (Fig. 4A). The distribution however changed over time with an apparent enrichment of nuclei with similar PC values at day 5 (Fig. 4A; Fig. S4).

Open in a separate window

Fig. 4.

Plant iPSC ancestor cultures are characterised by a high entropy of chromatin features, reducing over time. (A) Density distribution of chromatin features summarised by principal components PC1 and PC2 as computed in Fig. 3A. Density contours are coloured according to the frequency (percentage) of nuclei falling in the corresponding PC space. (B) Entropy (H₀) of chromatin features per family (n, number of descriptors per family). (C) Differential entropy (ΔH₀) between day 5 and day 0 for LBP texture features of H2B-RFP and at different length scale (radius). P values, paired Wilcoxon rank test. For box plots, the box represents the 25–75th percentiles, and the median is indicated. The whiskers show the minimum and maximum values. Data is from experiment HTI004; see Table S4.

To quantify this heterogeneity and possible changes during the culturing time, we computed the entropy of the data. Entropy is a useful measure of variability in biological data, capturing both the variance and the shape of the data distribution (Gandrillon et al., 2021). The analysis indeed revealed a marked positive entropy (H₀) for a large fraction of chromatin features (Fig. S4B), changing significantly over time.

We then analysed the family of features separately and found that heterogeneity as assessed by a positive entropy was contributed to by all features (Fig. 4B). Strikingly, the entropy decreased over time, particularly that of texture features and moderately for morphology features, whereas it remained largely positive for intensity features. Interestingly, the heterogeneity of chromatin distribution, measured by LBP texture features on H2B–RFP, was higher at a small length scale (LBP radius 1, Fig. S4C) but entropy reduction over time (ΔH₀) was more significant for the highest length scale (radius 10, Fig. 4C).

These findings were confirmed in a replicate experiment and where an additional imaging time point, at day 7 showed that entropy continue to decrease, but more slowly, after day 5 (Fig. S4D).

In conclusion, entropy analysis indicates a profound heterogeneity of nuclei morphology and of chromatin organisation among iPSC ancestors. Strikingly, heterogeneity reduces progressively, mostly within the first 5 days, and more particularly for chromatin distribution patterns (texture). This suggests a tendency towards homogenisation of chromatin types, although entropy remains high even after 7 days.

Chromatin heterogeneity is influenced by phytohormones

Protoplast cells cultured in the absence of phytohormones do not grow nor do they divide and undergo progressive cell death (Zhao et al., 2001). We thus asked whether the chromatin changes detected within the first days of culturing are part of the cellular responses to phytohormones. For this, we partitioned the initial pool of freshly released cells in two media – either in the regular Gamborg B5's medium (Gamborg et al., 1968), which is rich in macro- and micro-elements, vitamins, and is supplemented with glucose (2%) and phytohormones (auxins and cytokinin), or in the same medium but without phytohormones. First, we asked whether the absence of phytohormones would affect the reduction in the relative abundance of linker histones (H1.2) that was observed above. Quantifications showed that this is not the case and that H1.2 reduction still took place in the absence of phytohormones, although perhaps along a milder gradient (Fig. S5A). This suggests that H1.2 reduction is not a response to phytohormones in the medium but most likely a response to the cellular isolation, away from the source tissue, and culturing. Next, we interrogated the entire family of chromatin features with or without hormones, using PCA. The analysis showed that cells cultured without hormones underwent similar changes in chromatin organisation to that seen in the presence of hormones (Fig. 5A; Fig. S5B). However, and unexpectedly, we detected that chromatin organisation heterogeneity (measured by the entropy on the LBP texture of H2B–RFP distribution) decreased significantly more in the absence of phytohormones (Fig. 5B), whereas the heterogeneity of intensity and morphology features were not affected or were only moderately affected (Fig. S5C). We made the same observation when culturing the cells in a nutrient-poor medium (W5) without phytohormones (Fig. S5D). This suggests that phytohormones contribute to maintaining a certain level of heterogeneity during dedifferentiation in the plant iPSC ancestor cultures. Whether the effect is direct, with chromatin reorganisation responding to phytohormones, or indirect, owing to higher cell viability in the presence of phytohormones (Fig. S1D), remains to be determined. However, the first scenario is supported by the numerous evidence of crosstalk between phytohormones and chromatin modifiers particularly affecting plant cell identity and plasticity (Maury et al., 2019).

Open in a separate window

Fig. 5.

Phytohormones do not influence the trajectory of chromatin changes induced by culturing but dampen entropy reduction. (A) PCA of chromatin features of plant iPSC ancestor cultured in the Gamborg medium, including phytohormones (‘Gamborg’ or ‘+H’ as quick annotation on the graph), or in the same basis but without phytohormones (‘Gamborg/no Hormones’ or ‘-H’ as quick annotation on the graph). The PCA was computed with all features (see Materials and Methods, data is from experiment HTI004; see Table S4). (B) Differential entropy (ΔH₀) between day 5 and day 0 for the chromatin texture features (LBP texture metrics, H2B-RFP) of cells cultured in Gamborg with or without hormones as indicated. P value: Wilcoxon signed-rank test. The box represents the 25–75th percentiles, and the median is indicated. The whiskers show the minimum and maximum values.

Nutrient availability influence chromatin changes during culturing

To answer the question of whether chromatin changes respond to the physiological quality of the culturing medium we partitioned a pool of freshly released leaf protoplasts into W5 (nutrient poor; Yoo et al., 2007) or Gamborg B5 (nutrient rich; Gamborg et al., 1968), both prepared with or without phytohormones (Table S3). We analysed the chromatin features as before focusing on the first 2 days. Clearly, the culturing medium strongly influenced the chromatin changes with distinct trajectories principally influenced by the nutrient basis more than by the phytohormones (Fig. 6A; Fig. S6A).

Open in a separate window

Fig. 6.

Nutrient availability strongly influences the trajectory of chromatin changes in plant iPSC ancestor cultures. (A) PCA of chromatin features in cells cultured either in a nutrient rich (Gamborg) or in a nutrient poor (W5) medium, each with (+H) or without (−H) hormones. The cultures stem from the same, original pool of protoplasts partitioned in the different media and were imaged at day 0 and day 2 in two or three replicate wells (number of datapoint with the same colour). Data is from experiments HTI004 and HTI005; see Table S4). (B) Morphological features of nuclei from cultures either in Gamborg or W5 medium, at day 0, 3,5 and 7 (data is from experiment HTI001; see Table S4), each square unit is a replicate well. Median values are normalised (centred by rows) and the colour scale shows the fold change. (C) Representative nuclei of the same experiment, day 7. Scale bars: 5 μm. (D) Chromatin texture features (LBP method, H2B–RFP signal) of the same culture as in B (labels and key are the same) showing comparatively stronger changes between day 0 and day 2 in Gamborg compared to in W5. The list of LBP descriptors (rows) is provided in Table S2. Rows are clustered using correlation distance and average linkage.

Next, to verify whether the nutrient composition influences chromatin changes further in time we imaged new cultures for 7 days. The analysis confirmed that major changes occur essentially between day 0 and day 3, yet according to different trajectories depending on the medium, and with chromatin features stabilising rapidly after day 3 (Fig. S6B). The typical process of H1.2 reduction also occurred in the nutrient-poor medium (W5) although at a slightly lower rate (Fig. S6C). The medium affected nuclei morphology (Fig. 6B; Fig. S6D) with notably rounder and larger nuclei in the nutrient-poor (W5) medium (Fig. 6C; Fig. S6D,E). The nutrient-poor medium also dampened changes in chromatin texture across all length scales, possibly indicating a slower transition in chromatin reorganisation (Fig. 6D).

Finally, we asked whether the heterogeneity of chromatin features was affected by the culturing medium. The differential entropy (ΔH₀) between day 0 and day 7 per family of features did not reveal an overall significant effect of the medium (Fig. S6F), although, some morphology features showed an increase (e.g. area) or a decrease (e.g. roundness) in entropy in the nutrient-poor medium compared to the Gamborg's medium (Fig. S6G).

In conclusion, nutrients strongly influenced chromatin dynamics during the dedifferentiation phase of plant iPSC ancestor culture, with lower nutrient availability inducing rounder, bigger nuclei with a less differentiated chromatin texture.

Harnessing the informative potential of image features provide a new perspective on chromatin organisation in plant cells

Semi-automated high-throughput imaging, followed by supervised image segmentation, allowed us to record several hundreds of nuclei per day of observation, in 2–4 replicate cultures per experiment. Image analysis computed by TissueMAPS (http://tissuemaps.org) returned three family of features describing chromatin organisation: morphology descriptors of nucleus size and shape (the nucleus is the segmented object based on the H2B-RFP signal); intensity variables for each of the two, jointly expressed, chromatin markers consisting of fluorescently labelled histones H2B and H1.2, respectively; and texture descriptors for the distribution of chromatin markers.

Morphology and intensity features are commonly used to describe characteristics of segmented cells or nuclei because they are intuitive. Beyond the size, shape descriptors inform here on the roundness, elongation, convexity or solidity of the nuclei, which can be apprehended collectively in a multivariate analysis. Intensity features, in addition, inform on the absolute levels of each chromatin marker (intensity sum), their global compactness or density in the nucleus (intensity mean) and their intensity variability in the segmented object (standard deviation, minima and maxima). By contrast, texture features give abstract representations of signal distribution that cannot intuitively be assigned to a particular distribution pattern, that is, a chromatin phenotype in our case. However, those are useful to identify structured and repeated patterns (Depeursinge et al., 2017; Di Cataldo and Ficarra, 2017). Here, we focused on the LBP method (local binary pattern), which computes a local representation of signal distribution, based on pixel neighbourhood analysis, along circles of incrementally bigger radius, and scanning a growing number of positions along each circle (Ojala et al., 2001). The higher the LBP values at a given radius, the more heterogeneous the signal distribution, hence, this method informs on chromatin structures with contrasted density at different length scales. This approach has been reported to be powerful to classify the chromatin types from healthy versus carcinogenic cells presenting different aggregation phenotypes (Rana et al., 2021; Venkatachalapathy et al., 2021). Texture analysis indirectly informs on the homogeneity versus aggregation of chromatin at the mesoscale (determined by the LBP method scanning along window of different pixel size) and conceivably also captures the relative proportions of chromatin masses versus interchromatin compartment (Cremer et al., 2020). This approach thus nicely complements former cytological observations describing the disaggregation of the relatively large-scale heterochromatin domains at day 0 of protoplast cultures and their progressive reassembly (Ondřej et al., 2009; Tessadori et al., 2007; Williams et al., 2003; Zhao et al., 2001). We confirmed here that texture features allowed us to capture changes in chromatin organisation at different length scales (see below).

The rapid, multiscale reorganisation of chromatin along a trajectory is mostly affected by nutrient availability rather than phytohormones

The release of protoplasts from plant tissues induces dramatic chromatin disorganisation, with the disassembly of heterochromatin domains being a landmark confirmed in different species (Ondřej et al., 2009; 2010; Tessadori et al., 2007; Williams et al., 2003; Xu et al., 2021; Zhao et al., 2001). Culturing induces a progressive reassembly of heterochromatin domains and re-entry in the cell cycle (starting at day 4 and beyond) in the presence of phytohormones (Tessadori et al., 2007; Williams et al., 2003; Xu et al., 2021; Zhao et al., 2001). Here, high-throughput imaging allowed us to capture additional characteristics of chromatin dynamics, in particular of heterogenous cultures and the role of culturing media. Multivariate analysis identified a clear shift in chromatin features between day 0 and days 2–3, both when considering all families of features (morphology, intensity and textures) or each separately. This indicates that chromatin reorganisation occurs rapidly, early during the dedifferentiation phase, and is detectable at multiple scales. Specifically, changes were most prominent for nuclear morphology and for the distribution of H1.2 and H2B as revealed by chromatin texture.

Strikingly, nutrient availability, more than the phytohormones, had an influence on the trajectory of chromatin changes. Nutrient mostly affected nuclear morphology and chromatin textures, with H1.2 reduction being largely unaffected. Low nutrient availability prevents protoplasts from re-entering the cell cycle and eventually leads to cell death upon prolonged culturing (Zhao et al., 2001). Here, we show that low nutrient availability has an immediate effect and led to larger, rounder nuclei with less structured chromatin (i.e. displaying low textures), right up until day 5–7. This is reminiscent of the situation in animal iPSC ancestors, where nutrients have a profound impact on chromatin dynamics during cellular reprogramming (Apostolou and Hochedlinger, 2013; Lu et al., 2021). Metabolic fluxes are thought to influence the availability of metabolites used in epigenomic modifications both in plant and animal cells (Apostolou and Hochedlinger, 2013; Lindermayr et al., 2020; Lu et al., 2021, 2023), up to the point where energy metabolism has been shown to influence cell fate decisions (Ly et al., 2020). Here, the nutrient-rich and nutrient-poor media used differed not only in the amount of micro- and macro-nutrients, but also in sugar availability. Glucose levels have been shown to control, via the TOR signalling pathway, the cytoplasmic-to-nuclear ratio of Polycomb-group repressive complex 2 (PRC2) components influencing H3K27me3 deposition in plants (Ye et al., 2022). H3K27me3 reprogramming is essential for the acquisition of pluripotency in tissue explants primed for callus development in vitro (He et al., 2012). Given the ground role of H3K27me3 in cell identity and pluripotency in multicellular organisms, it is tempting to speculate that plant iPSC ancestors might undergo H3K27me3 reprogramming within the first culturing days, corresponding to the dedifferentiation phase (Grafi et al., 2011), under the joint influence of H1 linker histones (Rutowicz et al., 2019; She et al., 2013; Teano et al., 2023) and sugar and/or nutrient availability (Lu et al., 2023; Ye et al., 2022). That other nutrients influence chromatin reorganisation is nevertheless likely given the dramatic impact of nutrient availability on the trajectories of chromatin features.

A role of linker histone in plant cell dedifferentiation?

The pronounced decreased abundance of both linker histone variants H1.1 and H1.2 within the first culture days suggests that there are large-scale changes in chromatin composition, which might explain the enhanced chromatin accessibility previously measured (Xu et al., 2021; Zhao et al., 2001). Yet, interestingly, while reduced H1 levels correlate with chromatin decompaction and increased nuclear size in differentiated tissues (Rutowicz et al., 2019), this was not the case here in leaf-derived protoplasts. This suggests that there are mechanisms controlling nuclear size counteracting the possible effect of chromatin decondensation in protoplasts. In Arabidopsis, H1 abundance largely influences the levels and genomic distribution of DNA methylation, the levels of epigenetic marks such as H3K27me3, H3K4me3 and histone acetylation, heterochromatin domains and genome topology (Bourguet et al., 2021; Choi et al., 2020; He et al., 2024; Rutowicz et al., 2019; 2015; Teano et al., 2023; Wierzbicki and Jerzmanowski, 2005; Zemach et al., 2013). Whether the decrease of H1 levels observed here during dedifferentiation in vitro triggers vast epigenomic changes could therefore be expected, but remains to be explored.

A link between linker histones, chromatin reprogramming and pluripotency has been established in animal cells. On the one hand, gradual and vast changes are observed in the epigenetic landscape, chromatin accessibility and genome topology of animal cells undergoing iPSC reprogramming in vitro (Pelham-Webb et al., 2020). Besides, the abundance and type of linker, histone variants clearly contribute to several epigenomic and topological features of the genome (Fyodorov et al., 2018). On the other hand, depletion in somatic H1 variants is necessary for in vivo pluripotency acquisition in mouse primordial germ cells (Christophorou et al., 2014), and in vitro iPSC reprogramming is enhanced when an oocyte-specific H1 variant (H1foo) is expressed together with the traditional Oct4, Sox2 and Klf4 reprogramming factors (Kunitomi et al., 2016). Thus, it would be interesting to determine in the future whether depletion of canonical linker histone variants plays a role in the reprogramming competence of plant iPSC ancestor cultures and whether non-canonical variants, like H1foo in animals, exist that might facilitate this process. In addition, a comprehensive overview of epigenome reprogramming and topological reorganisation covering the entire process of plant cell dedifferentiation and pluripotency acquisition remains to be established, with single-cell resolution to account for the high cellular heterogeneity.

Chromatin features of plant iPSC ancestors show a high entropy, which reduces over time and is antagonistically modulated by phytohormones and histone deacetylation

Transcriptome analyses of protoplast cultures from different plant species have shown that these have extensive reprogramming compared to their source tissue (Cápal and Ondřej, 2014; Chupeau et al., 2013; Xu et al., 2021). Yet recently, single-cell-based reconstructions have demonstrated a high heterogeneity of transcriptome patterns even in a relatively homogenous protoplast culture composed of 85% mesophyll cells (Xu et al., 2021). Cell-to-cell variability, particularly in animal systems, has been recognised as an important factor controlling the inherent properties of a cellular system prompting investigations to understand its origin and regulation (Eldar and Elowitz, 2010; Eling et al., 2019; Guillemin and Stumpf, 2021; Huang, 2009; Mitchell and Hoffmann, 2018; Mojtahedi et al., 2016; Pelkmans, 2012; Richard and Yvert, 2014; Safdari et al., 2020). One approach to measure the heterogeneity, or information content, of biological systems is based on the calculation of the Shannon entropy, as borrowed from statistical mechanics (Gandrillon et al., 2021; MacArthur and Lemischka, 2013).This has been successfully applied to measure variability versus robustness of gene expression during cell differentiation (Dussiau et al., 2022; Richard et al., 2016; Stumpf et al., 2017; Wiesner et al., 2018) and reprogramming (Guillemin et al., 2019; Ye et al., 2020). In addition, entropy analysis was recently used to describe the heterogeneity of chromatin states at developmentally regulated loci, as an effective information content-predicting local genome topology and the competence for binding transcription factors (D'Oliveira Albanus et al., 2021). Here, consistent with stochastic gene expression (Xu et al., 2021), we identified a high entropy of chromatin features among plant iPSC ancestors. This is reminiscent of mouse and human iPSC cultures displaying a high level of transcriptome and epigenetic heterogeneity, which is thought to correlate with functional heterogeneity (Cahan and Daley, 2013; Yokobayashi et al., 2021).

Interestingly, entropy decreases over culturing time (5–40% depending on features) suggesting a process orienting the trajectory of chromatin changes towards homogeneisation, although entropy remains largely positive after 5–7 days culturing. We found that entropy reduction is attenuated in the absence of phytohormones. Although it cannot be excluded that this is a result of higher cell viability in the presence of hormones, it is conceivable that phytohormone-based signalling directly influences chromatin modifiers and remodellers, with an effect on cell identity maintenance and cellular plasticity (Maury et al., 2019), and is promoted when cells are exposed to an inhibitor of histone deacetylation. In contrast, nutrient availability did not influence the heterogeneity of chromatin features. Interestingly, during erythroid cell differentiation, stochasticity in gene expression is increased in the presence of a drug inhibiting histone acetylation (Guillemin et al., 2019), rather than deacetylation, as in our case. Possibly, the heterogeneity of chromatin patterns and that of transcription profiles are uncoupled to some extent, but more studies on the control of gene expression stochasticity in both animal and plant iPSC ancestor cultures remain necessary.

Is chromatin entropy as functional determinant of pluripotency acquisition?

A link between chromatin dynamics and chromatin pattern heterogeneity with the actual ability of plant iPSC ancestors to reprogramme remains to be established. Indeed, only a small fraction of cells from a plant iPSC ancestor culture will further develop in a pluripotent cell mass with regenerative ability (Pasternak et al., 2020; Sugimoto et al., 2011). Interestingly, in an Arabidopsis, in a leaf-derived protoplast culture consisting of 85% mesophyll cells, and hence that is initially relatively homogenous, only 0.5% cells seem to contribute to an effective regeneration process (Xu et al., 2021). This is reminiscent from the situation in animal iPSC cultures, which show variable levels of cellular reprogramming (0.5–10%) depending on the tissue source and inductive method (Kalra et al., 2021; Romanazzo et al., 2020).

Pluripotency in animal iPSCs has been proposed to be an emerging property of an intrinsically entropic cellular system, rather than from a unique property at the single-cell level (MacArthur and Lemischka, 2013). In this conceptual framework, where pluripotency is a statistical property of a microstate system, uncommitted PSCs have weak regulatory constraints leading to a high entropy and stochasticity in gene expression and chromatin states (MacArthur and Lemischka, 2013). As differentiation progresses, more regulatory constraints apply and the heterogeneity of the cellular system diminishes (MacArthur and Lemischka, 2013). Our findings suggest a similar conceptual framework to explain the properties of plant iPSC ancestors (Fig. 8) – the release of cell-wall-free cells away from the tissue context might abolish regulatory constraints stabilising cellular identity and leading to a high heterogeneity in gene expression and chromatin organisation. Appropriate culturing conditions might progressively restore regulatory signals in the culture, reducing entropy and thereby stabilizing gene expression and chromatin organization patterns (Fig. 8). Interestingly in this process, we found that phytohormones, at least at the specific ratio and concentration used in our conditions, are dampening this process and nutrient availability does not affect entropy reduction. This suggests that intrinsic properties of plant iPSC ancestor cultures re-establish regulatory constraints reducing the entropy of the cellular system.

In addition, the multiscale heterogeneity of chromatin patterns (as captured by textures) is reminiscent of the finding that variations in local chromatin density underscore the differentiation competence of human embryonic stem cells (hESCs) (Golkaram et al., 2017). This heterogeneity, which is influenced by genomic contacts but also by DNA-free space in the nucleus – a variable intrinsically captured by textures in our cytological analysis – is proposed to reflect variable states of molecular crowding, which in turn controls transcriptional bursts and noise underlying cellular reprogramming (Golkaram et al., 2016, 2017).

Protoplast preparation and culture

Protoplasts were isolated from Arabidopsis leaves based on published protocols (Li et al., 2014; Yoo et al., 2007) with some modifications to ensure sterile conditions during isolation and protoplast culturing as described thereafter. The whole procedure was performed under a laminar hood. All solutions were filter sterilised using 0.22-μm filters. Blades, forceps white pieces of paper and tubes were autoclaved. For pipetting, sterile filter tips were used. After leaf tissue digestion, protoplasts were filtered with sterile single-use cell strainers with 70 μm pores. After isolation cells were suspended in the intended medium (Table S3) and distributed into, 96-well plates with coverglass bottoms (Greiner Bio-One, Ref: 655087) with 100 μl culture per well. The outer wells of the plate were excluded due to the limited field of view and travel range at imaging. Each plate was sealed with 3M tape to avoid drying, wrapped in a layer of aluminium foil and placed in the growth chamber for 5 to 7 days depending on the experiment.

For the trichostatin A (TSA) treatment, 100 nM or 200 nM TSA in DMSO (Sigma-Aldrich) was added at day 0 in the culture or the equivalent amount of DMSO (2% or 4%, Mock). To assess cell viability, fluorescein diacetate (FDA; Sigma-Aldrich) was added at either day 0, day 2 or day 5 before imaging.

Microscopy imaging

Microscopy images shown in Fig. 1 were taken with a laser scanning confocal microscope (Leica SP5, Leica microsystems, Germany). For scoring (Fig. 1 graphs) the percentage of cells expressing the chromatin markers were scored manually by a researcher who was not aware of the experimental conditions under an epi-fluorescence microscope (Leica DM6000 Leica microsystems, Germany).

For all other figures, leaf protoplasts cultured in coverglass-bottom 96-well plates were imaged using a confocal microscope Cell Voyager (CV7000, Yokogawa), equipped with a 60× water immersion objective (Nikon, NA1.2) using illumination by 100–200 mW lasers from Coherent (depending on the channel) and filters from Chroma. Images were acquired as 16-bit images using two Neo sCMOS cameras (Andor).

For each well, six regions of interest (ROIs) corresponding to a field of view of 277 µm×234 µm with an image format of 2560×2160 pixels over 16 z-planes with a step of 1 µm were assessed. ROIs were randomly chosen without overlap. The imaging time for one well was 90 s; the imaging time for one plate (60 wells) was 90 min.

Image analysis

Images were loaded into TissueMAPS (http://tissuemaps.org; code available at https://github.com/pelkmanslab/TissueMAPS) where ROIs were grouped as 2×3 grids per well. Illumination correction was based on averaged intensity statistics across all images and maximum intensity projection along the z-dimension was performed in TissueMAPS.

Image processing was undertaken in TissueMAPS v. 0.6.3 for the following steps: (i) gaussian smoothing with a filter-size of 5 pixels; (ii) Otsu thresholding in a user-defined range using the red (H2B–RFP) channel, followed by binary mask filling and filtering objects <200 pixels in area to get segmentation; (iii) feature measurements using the measure_morphology, measure_intensity and measure_texture modules in TissueMAPS for both channels without the smoothing applied (see Table S2); (iv) classification of mis-segmentations by interactive training of Support Vector Machine (SVMs) in TissueMAPS. For the training, two classes of objects were annotated and created: A, correctly segmented and B, non-correctly segmented. Then for each class ~40–50 objects were labelled manually. Object morphology and RFP and GFP intensity features were chosen for training the classifier. In particular, the following parameters: area, circularity, roundness, elongation, convexity, mean intensity in RFP channel and mean intensity in GFP channel were used. In total, 100 objects in 10 images (total=1000) were used for training and resulted in 95% segmentation accuracy (n=417), with 4% of the 5% false positives being truncated nuclei at the edge of the image and 1% corresponding to undersegmented nuclei (Fig. S8). The description of the SVM algorithm which was used here is available at https://scikit-learn.org/stable/modules/svm.html). Features measurements and object classification were then downloaded from TissueMAPS for further analysis. In the R software environment (https://www.r-project.org/), we further filtered away outlier nuclei with an area >5000 pixels [‘giant nuclei’, 1–5% per dataset, possibly from trichomes (Walker et al., 2000)].

Principal component analysis

The PCA was undertaken using Clustvis (Metsalu and Vilo, 2015). The original dataset exported from TissueMAPS (HTI001, HTI002, HTI004 or HTI005; see Table S4) was subset to remove non informative columns such as redundant identifier codes (related to the plate, experiment and objects), position information (such as Morphology_local centroid_X and _Y, well_position, is_border) and features or experiment description not relevant for the analysis. If too big for upload, the subset data was entered using the input type ‘paste data’; the data matrix was transposed (‘data matrix reshape/transpose matrix’) and the option ‘detect column and row annotations’ was unchecked to be adjusted manually. Columns with similar annotations were collapsed by taking the median inside each group. Unit variance scaling was applied to rows. SVD with imputation was used to calculate the principal components. The x- and y-axis show the principal component 1 and principal component 2, respectively, which explain the given percentage of the total variance. Prediction ellipses depict a 95% confidence interval (a new observation from the same group will fall inside the ellipse with probability P=0.95).

Entropy analysis

The initial script for computing Shannon Entropy is described in Dussiau et al. (2022) and is available at https://osf.io/9mcwg/. The adapted script for computing entropy of chromatin features is provided at https://github.com/barouxlab/ChromatinEntropy. When all cells (segmented nuclei) express the same value for a given feature, this entropy of the feature will be null. The more cell-to-cell variability for a given chromatin feature, the higher value of entropy.

We thank Alexis Maizel (COS Heidelberg, Germany) for the Arabidopsis lines expressing the UBQ10::H2B-RFP reporter (von Wangenheim et al., 2016), Alejandro Fonseca for computational help in running entropy analysis, Prof. Ueli Grossniklaus and group members for insightful discussions and Prof. Grossniklaus department lab managers and technicians for daily support.