FX-Cell: a method for single-cell RNA sequencing on difficult-to-digest and cryopreserved plant samples

Abstract

Single-cell RNA sequencing in plants requires the isolation of high-quality protoplasts—cells devoid of cell walls. However, many plant tissues and organs are resistant to enzymatic digestion, posing a significant barrier to advancing single-cell multi-omics in plant research. Furthermore, for field-grown crops, the lack of immediate laboratory facilities presents another major challenge for timely protoplast preparation. Here, to address these limitations, we developed FX-Cell and its derivatives, FXcryo-Cell and cryoFX-Cell, to enable single-cell RNA sequencing with both difficult-to-digest and cryopreserved plant samples. By optimizing the fixation buffer and minimizing RNA degradation, our approach ensures efficient cell wall digestion at high temperatures while maintaining high-quality single cells, even after long-term storage at −80 °C, and circumvents use of nuclei, which are not representative of the pool of translatable messenger RNAs. We successfully constructed high-quality cell atlases for rice tiller nodes, rhizomes of wild rice and maize crown roots grown under field conditions. Moreover, these methods enable the accurate reconstruction of plant acute wounding responses at single-cell resolution. Collectively, these advancements expand the applicability of plant single-cell genomics across a wider range of species and tissues, paving the way for comprehensive Plant Cell Atlases for plant species.

The cell holds the genetic blueprint of an organism, yet neighboring cells can differ dramatically in morphology and function. Understanding the gene expression patterns that drive these differences can provide critical insights into the roles, developmental trajectories and evolutionary histories of cell types, tissues and entire organisms. Single-cell RNA sequencing (scRNA-seq) has led to important discoveries in both animals^1–4 and plants^5–15; however, in plant biology and agriculture, several challenges have made it difficult to apply scRNA-seq to many important species and tissues.

The first major challenge arises from the plant cell wall, a rigid outer layer surrounding the plasma membrane. Composed primarily of cellulose, along with other polysaccharides, proteins and sometimes lignin, the cell wall provides structural integrity but also makes enzymatic digestion difficult. In many plants, secondary cell wall thickening and specialized cell wall components further impede protoplast isolation—the process of releasing live single cells without cell walls^15–17.

The second limitation is the outdoor cultivation of most plants, especially crops. scRNA-seq requires immediate enzymatic digestion to remove cell walls, yet most farms and field sites lack the necessary molecular biology facilities. As a result, current single-cell transcriptional atlases of crops are derived exclusively from greenhouse-grown materials, which may not adequately capture transcriptome responses under natural conditions.

The third limitation is the time-sensitive nature of plant responses to environmental changes, such as wounding or pathogen infection. Existing protoplast isolation methods require 1–2 h of enzymatic digestion, during which time the transcriptome can change. A fourth challenge is that while single-nucleus RNA sequencing (snRNA-seq) offers an alternative approach^18–21, it often produces lower-quality data than scRNA-seq^16,22–24. Additionally, nuclear RNA may not fully represent the cellular transcriptome, in particular the subset of mRNAs that are likely to be translated into protein. Furthermore, isolated nuclei are prone to aggregation, leading to higher doublet rates. Certain transcripts also exhibit differential enrichment between snRNA-seq and scRNA-seq datasets²⁵.

To tackle these challenges, we refined existing protoplast isolation protocols and developed simple, versatile methods for scRNA-seq that enable its application to difficult-to-digest and cryopreserved plant samples, as well as samples subjected to wounding treatment. These advancements will expand the range of species and tissues amenable to scRNA-seq, accelerating the development of comprehensive Plant Cell Atlases^26,27, similar to the Human Cell Atlas²⁸.

Results

Fixation increases single-cell release from plant tissues

Protoplast isolation is perhaps the greatest technical hurdle in scRNA-seq of plant tissues. We hypothesized that the use of coagulant fixatives, which stabilize cells by coagulating the protein matrix while disrupting lipid membranes, followed by cell wall digestion, could offer two key advantages for cell release: fixation (i) stabilizes the cell cytoplasm so that cells can withstand harsher shear forces without breaking, and (ii) allows enzymatic digestion to occur at higher temperatures (~50 °C) where cellulase enzymes are most active²⁹. Additionally, acid treatment can be used and may further degrade cell walls, as acids are commonly used in industrial hydrolysis³⁰.

To test this hypothesis, we fixed maize anthers in ice-cold Farmer’s solution (3:1 100% ethanol:glacial acetic acid) and quantified cell release from both fresh and fixed maize anthers. Because fixation disrupts all cellular activities, we refer to isolated structures as ‘cells’ rather than ‘protoplasts’. We found that an optimized maize anther protoplasting protocol (1.25% wt/vol cellulase, 0.5% pectolyase, 0.5% macerozyme, 0.5% hemicellulose)³¹ had a mean release of 4,387 protoplasts per anther after 90 min digestion and 11,333 protoplasts per anther when extended to 16 h (Fig. 1a). In comparison, if anthers were fixed before digestion in a reduced enzyme mix (1.25% wt/vol cellulase and 0.4% macerozyme), 15,900 cells were released within 90 min; this increase in cell release is presumably because the cells were stabilized against mechanical lysis by fixation, while unfixed cells are very fragile. When incubation temperature was increased from 30 °C (standard) to 50 °C we observed an average release of 45,033 cells (Fig. 1a), close to the actual number of 50,000 cells in a 2.0-mm maize anther³².

Fig. 1 |. Fixation and digestion at high temperatures facilitate cell release.

a, Quantification of released protoplasts or cells from maize anthers using different protocols. Maize anthers were digested for 90 min or 16 h at 30 °C (gray bars)³¹ with optimized enzyme mix (1.25% wt/vol cellulase-RS, 0.5% pectolyase Y-23, 0.5% macerozyme-R10, 0.5% hemicellulose) or first fixed and then digested for 90 min with reduced enzyme mix (1.25% wt/vol cellulase-RS and 0.4% wt/vol macerozyme-R10) at either 30 °C or 50 °C (blue bars). The number of released protoplasts or cells was quantified using a hemocytometer. Data represent the mean ± s.e.m. from five independent experiments. Different letters indicate statistically significant differences (two-sided Student’s t-test, P < 0.05, with no adjustment for multiple comparisons). Fix., fixation; dig., digestion; temp., temperature. b, Representative image of released cells from fixed maize anthers. Scale bar, 100 μm. c, Quantification of released protoplasts or cells from different tissues and species. Plant tissues were subjected to either fresh protoplasting or fixation and digestion at high temperatures as described in a. Data represent the mean ± s.e.m. from five independent experiments. Different letters indicate statistically significant differences (two-sided Student’s t-test, P < 0.05). Conv., conventional. d, Quantification of released cells from fixed rice tiller nodes, wild rice rhizome nodes and shoot apices of S. martensii. Tissues were enzymatically digested at a high temperature using two enzymes (that is, reduced enzyme mix, C + M, 1.25% wt/vol cellulase-RS and 0.4% wt/vol macerozyme-R10) or four enzymes (C + M + S + P, 1% wt/vol cellulase-RS, 0.5% wt/vol macerozyme-R10, 1% wt/vol snailase and 0.5% wt/vol pectinase). Data represent the mean ± s.d. from three independent experiments. Different letters indicate statistically significant differences (two-sided Student’s t-test, P < 0.05). e, Schematic representation of RNase depletion in the digestion buffer. f, RNA quality assessment of fixed maize anthers under different conditions: (1) without digestion, (2) digested at 50 °C for 90 min in enzyme buffer lacking enzymes, (3) digested at 50 °C for 90 min using commercial enzymes and (4) digested at 50 °C for 90 min using RNase-depleted enzymes. Data represent the mean ± s.e.m. from five independent experiments. Different letters indicate statistically significant differences (two-sided Student’s t-test, P < 0.05, with no adjustment for multiple comparisons).

The optimized maize anther cell release protocol with unfixed tissue yielded fewer than expected epidermal and endothecial cells, both of which tended to remain clumped and undigested, producing a skewed release favoring tapetal cells, middle layer cells and meiocytes. When the anthers were fixed and then digested at 50 °C, we did not observe any cell clumps, debris or undigested material, suggesting that the digestion was complete (Fig. 1b). In addition to increasing the cell release efficiency and cell-type representation, we found that fixation before digestion maintained cells’ natural morphology allowing the potential for cell-type identification after isolation (Fig. 1b and Extended Data Fig. 1a,b).

To assess the method’s broader applicability, we tested four additional maize tissues (shoot apical meristem, leaf, root and young ear) and three non-model plant taxa (Amborella trichopoda leaf, Nymphaea colorata leaf, and Capsella bursa-pastoris leaf and stem; Fig. 1c and Extended Data Fig. 1a,c). As shown in Extended Data Fig. 1a, cellular morphology was maintained in each fixed sample, allowing obvious differentiation of the varying cell types. Notably, cell release was 10-fold to 364-fold higher in fixed tissues compared to a fresh sample, except for maize leaves in which there were 3.6 as many cells released from a fresh sample compared to the fixation-based protocol (Fig. 1c). Presumably, cells with large fluid-filled vacuoles, such as maize mesophyll, are very fragile after fixation, because they have too few coagulated proteins.

Specialized cell wall components in some plants necessitated modifications to the digestion buffer. The fixation–digestion protocol at high temperatures using reduced enzyme mix failed to efficiently digest rice tiller nodes, wild rice (Oryza longistaminata) rhizome nodes and lycophyte (Selaginella martensii) shoot apices. To address this, we incorporated 1% cellulase, 1% snailase, 0.5% macerozyme and 0.5% pectinase, which significantly improved digestion by targeting chitin and pectin (Fig. 1d and Extended Data Fig. 1b). Overall, our fixation-based protocol, combined with an expanded enzyme mix, effectively dissociated diverse plant tissues into single cells, broadening the applicability of scRNA-seq to previously challenging species and tissues.

RNase depletion of the digestion enzyme cocktail is necessary for maintaining RNA quality

While fixation itself does not affect RNA quality, it removes the cell membrane and makes the internal RNA contents accessible to RNases in solution. This generates a challenge during enzymatic digestion, because most cell wall digesting enzymes are complex mixtures that contain substantial RNase activity. We tested several RNase inhibitors, including commercial inhibitors, EDTA and vanadyl ribonucleoside complexes, but found none that could effectively inhibit the RNase activity in cell wall digestion enzyme blends (Extended Data Fig. 1d,e). This is partly because many available RNase inhibitors target the RNase A family of enzymes³³, which is only produced in vertebrates. Secreted fungal RNases are primarily of the T1 and T2 families³³.

To surmount this complication, we adapted a column-based method to reduce fungal T1 and T2 RNases by binding them to Sepharose/agarose coupled with guanosine monophosphate (GMP)³⁴ (Fig. 1e). We found that cell wall digesting enzymes readily passed through GMP-Sepharose/agarose columns, while the contaminating RNases remained bound. After column depletion, RNase activity was greatly reduced in the digestion enzyme blend (Extended Data Fig. 1e). The RNase-depleted enzymes remained stable when stored as glycerol stocks for at least one year at −20 °C.

Because GMP-Sepharose/agarose synthesis involves multiple steps and lacks standardized quality control, we implemented an assessment step to verify successful GMP coupling. Specifically, GMP-Sepharose/agarose was incubated in hydrochloric acid in a boiling water bath for 1 h, and its optical density at 248 nm was measured (Extended Data Fig. 1f). Additionally, we introduced a regeneration step using a wash buffer containing 5 M NaCl and 1 mM tri-GMP (a mixture of 1 mM 5′-GMP and 1 mM 2′(3′)-GMP) to remove bound RNases, allowing column reuse (Fig. 1e and Extended Data Fig. 1g; see Methods). Notably, the regenerated column retained high performance after over 8 months of continuous use and repeated enzyme purification cycles. These modifications enable laboratories with limited biochemistry expertise to easily purify digestive enzymes of RNases.

We next tested the effect of the fixed-tissue dissociation procedure on RNA quality. RNA isolated from fixed maize anthers had an average RNA integrity number (RIN) of 9.3 demonstrating fixation did not cause any significant decrease in RNA quality (Fig. 1f). Fixed anthers digested at 50 °C in a commercial enzyme blend had a RIN of 4.1 with very noticeable loss of ribosomal RNA. After fixation then digestion with RNase-depleted enzymes, the RIN was 6.7, demonstrating the fixed-tissue dissociation protocol can produce RNA of reasonable quality, although there is a decrease in RNA integrity relative to undigested tissue. When fixed anthers were incubated in enzyme buffer at 50 °C without enzymes, we observed a similar RIN of 6.1. Therefore, the decrease in RNA integrity during incubation is not an exogenous enzyme-dependent process, rather we suspect this degradation is caused by endogenous anther RNases that survive the fixation process.

Development of the FX-Cell method

To evaluate the applicability of the fixation–digestion methods at high temperatures for scRNA-seq, maize anthers were fixed and digested with RNAse-depleted enzymes at 50 °C for 90 min. The resultant cells were sorted and isolated using either BioSorter (Union Biometrica) or Hana (Namocell) machines into 96-well plates, and scRNA-seq libraries were prepared using a modified CEL-Seq2 library preparation protocol^31,35. Of the 384 maize anther single cells examined, 307 exhibited more than 500 unique molecular identifiers (UMIs) and at least 200 detected genes after the removal of cell-cycle genes. An average of 5,885 UMIs and 2,016 transcribed genes per cell were detected (Supplementary Table 1). The dataset was classified into four distinct clusters, two of which were reclustered based on marker gene expression, producing six total clusters (Supplementary Table 1). Notably, clusters corresponding to meiocyte, endothecium, tapetum and epidermis were identified (Extended Data Fig. 2). A more extensive dataset generated using this method enabled the reconstruction of developmental trajectories of all living maize anther cell types, extending from early cell proliferation to subsequent cell fate determination and differentiation over a 20-day time course³⁶.

Encouraged by these findings, efforts were made to determine whether this method could be adapted for commercial high-throughput scRNA-seq platforms such as Chromium (10× Genomics). To facilitate direct comparisons with conventional scRNA-seq and snRNA-seq, rice root tissues with existing scRNA-seq data were selected as the experimental samples³⁷. The data quality obtained using the fixation and digestion high-temperature method was suboptimal. While the median number of transcribed genes detected per cell was 1,230 and 1,369 (Fig. 2b and Supplementary Table 1), the total number of captured cells was only 2,084 and 2,810, respectively (Fig. 2a and Supplementary Table 1). The cumulative cell count across two biological replicates was significantly lower than the 27,469 cells obtained from traditional protoplast-based scRNA-seq³⁷, indicating a considerable number of low-quality cells were generated, leading to their automatic removal by Cell Ranger. Consistently, when data quality (that is, the number of genes per cell) was not considered, the filtering criteria required to retain 8,000 cells resulted in a decrease in the median number of transcribed genes per cell, dropping to 303 and 444 (Fig. 2c and Supplementary Table 1). The data quality for snRNA-seq was similarly unsatisfactory, with single-cell gene counts of 635 and 629, and captured cell numbers of 1,428 and 1,835 (Fig. 2a,b and Supplementary Table 1).

Fig. 2 |. Development and evaluation of FX-Cell for high-throughput scRNA-seq.

a,b, The number of recovered cells (a) and the median number of transcribed genes (b) in the rice root tip cell atlas generated by different scRNA-seq methods. Two biological replicates were performed for each method. c, The median number of transcribed genes in the rice root tip cell atlas under conditions where filtering criteria were adjusted to achieve 8,000 cells in the atlas, without considering data quality (that is, the number of genes per single cell). The same data as in a and b were used. d, Assessment of RNase enzymatic activity in the digestion solution under various conditions. ‘No purif.’, ‘one purif.’ and ‘two purif.’ indicate RNase removal without purification or following one or two rounds of affinity chromatography. ‘tri-GMP’ represents a mixture of 5′-GMP and 2′(3′)-GMP. ‘Blank’ serves as a control using the enzyme-dissolving RNase binding buffer. Data represent the mean ± s.d. from three independent experiments. e, UMAP plots depicting the rice root tip cell atlas generated by FX-Cell. Two biological replicates were performed. The batch effect was not corrected using the Harmony algorithm. Different colors indicate distinct cell clusters. f, Comparison of rice root tip cell atlases generated by FX-Cell and traditional protoplasting-based scRNA-seq³⁷. The datasets were integrated, and batch effects were removed using the Harmony algorithm. Different colors represent distinct cell clusters. To minimize biases due to differences in cell numbers across datasets, both atlases were normalized to approximately 9,000 cells. g, Annotation of the integrated rice root tip cell atlas shown in f based on published cell-type annotation³⁷. Different colors correspond to distinct cell types. h, UMAP plots illustrating the expression patterns of known cell-type marker genes in the rice root tip cell atlas generated by FX-Cell, corresponding to f.

These results likely stem from RNA degradation, prompting efforts to improve RNA integrity and increase the number of transcribed genes captured per cell. During the digestion of rice root tips, elevated temperatures and prolonged digestion times enhanced RNA degradation and decreased RNA integrity (Extended Data Fig. 1h–k). To address this issue, digestion time was reduced from 90 min to 30 min, and the temperature was lowered from 50 °C to 40 °C. Additionally, to minimize RNase contamination in the digestion buffer, digestion enzymes were purified twice using GMP-Sepharose columns (Fig. 2d). Transfer RNA is commonly added as a competitive inhibitor of RNase to safeguard the integrity of target RNA in RNA in situ hybridization experiments³⁸. tri-GMP, a mixture of 2′-GMP, 3′-GMP and 5′-GMP (where 2′-GMP and 3′-GMP are collectively isolated as 2′(3′)-GMP), can competitively bind to RNase^39,40. Remarkably, the incorporation of tRNA and tri-GMP effectively inhibited RNase activity (Fig. 2d and Extended Data Fig. 1e).

Following these optimizations, the experiment was repeated using rice root tips with two biological replicates. Compared to the original method and the data obtained from snRNA-seq, the optimized protocol resulted in a substantial increase in the number of detected cells (9,474 and 8,874 cells) as well as in the median number of transcribed genes per cell (1,494 and 1,661 genes; Fig. 2a,b and Supplementary Table 1). Notably, when the filtering criteria were adjusted to retain 8,000 cells, the gene counts per individual cell increased to 1,661 and 1,801, which were substantially higher than those obtained using the original method and snRNA-seq (Fig. 2c and Supplementary Table 1). Furthermore, uniform manifold approximation and projection (UMAP) visualization demonstrated high reproducibility between the two independent experiments (Fig. 2e). The dataset generated using the optimized method also displayed high consistency with traditional protoplast-based scRNA-seq data³⁷ (Fig. 2f). Using published cell-type annotation³⁷, most of the cell types present in rice root tips were successfully identified (Fig. 2g and Extended Data Fig. 3a). Importantly, the expression specificity of these marker genes was largely preserved (Fig. 2h and Extended Data Fig. 3b). The lower number of root hair cluster cells in the new atlas may be attributed to developmental differences arising from variations in cultivation conditions (Extended Data Fig. 3c).

To further evaluate the method’s performance, we applied it to Arabidopsis thaliana root tips. The resulting cell atlas demonstrated gene counts of 1,728 and 1,862 genes (n = 9,318 and 6,099 cells), closely mirroring those generated using conventional protoplasting protocols^41,42 (Extended Data Fig. 4a–d).

Comparative analysis of the atlases revealed both common and cell-type-specific protoplasting-associated genes across A. thaliana root tip cell types (Extended Data Fig. 4e,f and Supplementary Table 1). While these genes did not significantly impact overall clustering or the identification of key marker genes, some wound-response genes were incorrectly flagged as cluster-enriched when protoplasting genes were not excluded (Extended Data Fig. 4g,h and Supplementary Table 1). Identifying this gene set lays the groundwork for more accurate future analyses of root cell-specific responses via scRNA-seq.

In summary, a novel high-throughput plant scRNA-seq method, termed FX-Cell, was developed. This method outperforms snRNA-seq and produces data of comparable quality to that obtained using traditional scRNA-seq. Importantly, FX-Cell enhances the efficiency of single-cell release, enabling the application of scRNA-seq to tissues and organs that are otherwise challenging to digest and enhances the likelihood of recovering and analyzing rare cell types.

Development of FXcryo-Cell and cryoFX-Cell for cryopreserved plant samples

Traditional protoplast-based scRNA-seq methods require the use of fresh plant tissues, necessitating immediate processing after sampling, which significantly limits their applicability. The ability to perform scRNA-seq on fixed plant materials following cryopreservation would greatly expand the potential applications of this technology. To test this hypothesis, fixed rice roots were cryopreserved by rapidly freezing them in liquid nitrogen and then stored at −80 °C in an ultralow-temperature freezer. After thawing and subsequent cell wall digestion at 40 °C for 30 min, the morphology of the cells released from cryopreserved rice roots was largely similar to that of non-cryopreserved roots (Fig. 3c). Similar observations were made in other plant tissues, including A. thaliana leaves and S. martensii shoot apices (Extended Data Fig. 5a,b).

Fig. 3 |. Applications of FXcryo-Cell and cryoFX-Cell on rice root tip samples.

a,b, Step-by-step schematic of the FXcryo-Cell (a) and cryoFX-Cell (b) workflows. c, Representative images showing released cells from rice root tips. Cells were obtained from tissues subjected to three different treatments: (1) fixation followed by high-temperature enzymatic digestion (fix.), (2) fixation, freezing, thawing and then high-temperature enzymatic digestion (fix. + cryo.), and (3) freezing, thawing, fixation and high-temperature enzymatic digestion (cryo. + fix.). Scale bars, 50 μm. Three independent experiments were performed. d,e, UMAP plots displaying rice root tip cell atlases generated using FXcryoCell (d) and cryoFX-Cell (e). Two biological replicates were performed for each experiment. Batch effects were not removed by the Harmony algorithm. Different colors indicate distinct cell clusters. f, Joint analysis of rice root tip cell atlases from FX-Cell, FXcryo-Cell and cryoFX-Cell. Batch effects were not removed by the Harmony algorithm. Different colors represent different cell clusters. g, Integrated rice root tip cell atlas, annotated based on published cell-type marker genes (Extended Data Fig. 3a and Supplementary Table 1). Atlases from FX-Cell, FXcryo-Cell and cryoFX-Cell were combined. Different colors represent different cell clusters, with related clusters grouped (dashed outlines). h, UMAP plots showing expression patterns of representative cell-type marker genes in the integrated rice root tip cell atlas. Four representative cell types are displayed. See Extended Data Fig. 5e for additional cell types.

scRNA-seq was then performed on fixed and cryopreserved rice root tips using two biological replicates, yielding median gene counts of 1,637 and 1,711 per individual cell and capturing 6,192 and 8,259 cells, respectively (Fig. 2a,b and Supplementary Table 1). When filtering criteria were adjusted to retain 8,000 cells, the median gene counts per individual cell reached 1,341 and 1,802, surpassing the results obtained from snRNA-seq (Fig. 2c and Supplementary Table 1). UMAP visualization demonstrated distinct cell clustering and high reproducibility between the two independent experiments (Fig. 3d). This method was, therefore, designated as FXcryo-Cell (Fig. 3a).

Under certain circumstances, such as field sampling, immediate fixation followed by freezing can be challenging. To address this, an alternative approach was developed in which plant materials were directly cryopreserved and fixed immediately after thawing, followed by high-temperature enzymatic digestion. The quality of cells and RNAs obtained using this workflow was comparable to that of FXcryo-Cell (Fig. 3c and Extended Data Fig. 5a–c). Notably, scRNA-seq performed on these cells from rice root tips, using two biological replicates, yielded high-quality and reproducible data (Fig. 3e), with median gene counts per individual cell of 1,660 and 1,465 and total captured cell numbers of 7,578 and 5,524 cells, respectively (Fig. 2a,b and Supplementary Table 1). When 8,000 cells were retained in the Seurat analysis, the median gene counts per cell were 1,640 and 1,115 (Fig. 2c and Supplementary Table 1). This method was designated as cryoFX-Cell (Fig. 3b).

To evaluate the consistency among FXcryo-Cell, cryoFX-Cell and FX-Cell, datasets from rice roots generated by these three methods were merged without batch effect correction. The cell types were annotated with published cell-type marker genes (Fig. 3g, Extended Data Fig. 3a and Supplementary Table 1). UMAP visualization demonstrated strong reproducibility across all three datasets (Fig. 3f,g). Transcriptomic profiles of pseudobulk cell clusters generated by different methods showed strong concordance (Extended Data Fig. 5d). Additionally, marker genes specifically expressed in previously published scRNA-seq data of rice roots were also specifically expressed in both FXcryo-Cell and cryoFX-Cell datasets (Fig. 3h and Extended Data Fig. 5e).

In conclusion, FXcryo-Cell and cryoFX-Cell effectively replace traditional scRNA-seq methods, significantly expanding the potential applications of scRNA-seq in plants. In the following sections, three potential application scenarios for these two methods are explored.

Application of FXcryo-Cell on difficult-to-digest plant samples

The shoot architecture in cereals is largely influenced by tillers, which are lateral shoots arising from axillary buds at basal nodes⁴³ (insets in Fig. 4a). Dissecting the development of basal tiller nodes at single-cell resolution is, therefore, crucial for understanding crop architecture and yield. As previously mentioned, however, traditional enzymatic digestion methods struggle to obtain protoplasts from these structures. FXcryo-Cell was found to effectively remove cell walls from basal tiller nodes of 5-day-old rice plants, releasing a greater number of individual cells (Fig. 1d and Extended Data Fig. 5f). Consequently, an atlas of 5,822 high-quality cells was obtained, with a median transcribed gene count of 1,504 (Supplementary Table 1). Established marker genes were used to annotate cell clusters corresponding to various cell types (Fig. 4a, Extended Data Fig. 6a and Supplementary Table 1). A central meristem-like cell cluster (cluster 0) was surrounded by differentiating cell types, including phloem (clusters 8 and 9), xylem (cluster 10), cambium (cluster 13), endodermis (clusters 1 and 3) and epidermis (cluster 5). Additionally, consistent with morphological observations indicating that both root and tiller bud formation occurs at the basal tiller node (insets in Fig. 4a), a root cap cell cluster (cluster 11) was identified. The annotation was finally validated through sequential fluorescence in situ hybridization (seq-FISH) experiments (Extended Data Fig. 7a).

Fig. 4 |. Cell atlases of difficult-to-digest and cryopreserved plant samples.

a, UMAP plot of the cultivated rice basal tiller node cell atlas generated using FXcryo-Cell. Different colors indicate distinct cell clusters, with related clusters grouped (dashed lines). Insets: rice seedlings (upper) and a longitudinal section of a basal tiller node (lower). The harvested tissue used for FXcryo-Cell is highlighted by a white dashed box. The basal node can give rise to both tiller buds and roots (red arrows). The atlas was annotated based on published cell-type marker genes (Extended Data Fig. 6a and Supplementary Table 1). Scale bars, 0.5 cm (upper) and 200 μm (lower). One experiment was performed. b, UMAP plot of the wild rice rhizome node cell atlas generated using FXcryo-Cell. Insets: growing rhizomes (upper image, red arrow) and the harvested rhizome node (red dashed box) used for FXcryo-Cell. The atlas was annotated based on published cell-type marker genes (Extended Data Fig. 6b and Supplementary Table 1). Scale bar, 2 mm. One experiment was performed. c, UMAP plots showing the expression patterns of shoot and root meristem identity genes in the wild rice rhizome node cell atlas (b). QC-like/root meristem and shoot meristem-like cell clusters are outlined (dashed lines). d, UMAP plot of the cell atlas of field-grown maize emerging crown roots generated using cryoFX-Cell. Insets: field-grown maize plants (upper) and the harvested crown roots (red dashed box) used for cryoFX-Cell. The atlas was annotated based on published cell-type marker genes (Extended Data Fig. 6c and Supplementary Table 1). Scale bar, 0.5 cm. One experiment was performed. QC, quiescent center.

To further validate the applicability of FXcryo-Cell for difficult-to-digest plant samples, rhizomes—underground, horizontal stems that facilitate the perennial growth habit in wild rice (O. longistaminata) (insets in Fig. 4b)—were analyzed^44–46. Similarly to tiller nodes, rhizome nodes could not be effectively digested into protoplasts using traditional digestion methods. FXcryo-Cell successfully overcame this limitation (Fig. 1d and Extended Data Fig. 5f). As shown in Fig. 4b, the application of FXcryo-Cell on rhizome nodes at the budding stage generated a high-quality single-cell resolution map, capturing 9,233 cells with a median gene count of 891 per cell (Supplementary Table 1). The majority of cell types were characterized using published rice and A. thaliana cell-type marker genes^37,47,48 (Fig. 4b, Extended Data Fig. 6b and Supplementary Table 1) and validated by seq-FISH experiments (Extended Data Fig. 7b).

New buds on rhizomes either grow underground as new rhizomes or emerge above ground as aerial shoots⁴⁵. Accordingly, shoot meristem-like clusters (clusters 0 and 10) were identified in the rhizome node atlas (Fig. 4b). These clusters showed high expression of Oryza sativa homeobox 1 (OSH1) and OSH15 (Fig. 4c), genes associated with shoot meristem identity^49,50. As shown in Fig. 4b, rhizome nodes were observed to generate roots. Correspondingly, quiescent center-like/root meristem clusters (clusters 1 and 2) were detected (Fig. 4b), with strong expression of homologs of the well-known A. thaliana root identity Plethora (PLT) genes, including PLT1 and PLT2 (ref. 51; Fig. 4c).

Collectively, these findings demonstrate that FXcryo-Cell is a valuable tool for studying difficult-to-digest plant samples. The improved method can provide fresh insights into the complexity of cell types of both tiller and rhizome nodes in rice.

Application of cryoFX-Cell to field samples

Compared to greenhouse-grown samples, applying scRNA-seq to field-collected plant materials provides a more accurate representation of plant responses to complex and dynamic environmental conditions at the cellular level. As previously mentioned, traditional protoplasting-based scRNA-seq methods require immediate processing of freshly collected plant samples, significantly limiting their use in field-based research. The optimized cryoFX-Cell method enables the use of cryopreserved plant samples for single-cell preparation, offering an ideal solution for field sample analysis.

To evaluate this approach, emerging crown roots were collected from 45-day-old field-grown maize (insets in Fig. 4d). The samples were rapidly frozen in dry ice and stored at −80 °C for long-term preservation (Fig. 4d). Following a 50-day storage period, a cryoFX-Cell experiment was conducted. The results yielded 9,127 captured cells, with a median gene count of 1,857 per cell (Supplementary Table 1). Cell clusters were annotated based on established marker genes and published references^52–54 (Fig. 4d, Extended Data Fig. 6c and Supplementary Table 1). Notably, consistent with previous findings that primary and crown roots have similar anatomy⁵⁵, the vast majority of maize primary root cell types identified using traditional scRNA-seq methods were also recovered⁵³ (Fig. 4d and Extended Data Fig. 6d,e).

Application of FXcryo-Cell to probe acute wounding responses at single-cell resolution

As sessile organisms, plants have evolved complex mechanisms to cope with dynamic environmental challenges. In response to wounding, calcium and jasmonic acid signaling pathways are rapidly activated at the site of damage, initiating tissue repair and organ regeneration^56–59. To date, studies have primarily relied on bulk samples^56,60, leaving it unclear whether different plant cell types exhibit distinct responses to wounding. To address this question, transcriptomic changes at the single-cell level must be examined. Traditional plant scRNA-seq methods require protoplast preparation, which may inadvertently introduce wounding effects into control samples, complicating the acquisition of ground-truth data from unwounded plants. Fixation preserves the original state of plants at the time of sampling, effectively mitigating this issue.

To evaluate FXcryo-Cell in this context, the third-to-last true leaves of 24-day-old A. thaliana plants were wounded. Intact (control) and wounded leaves were harvested 2 h after injury, fixed and washed twice with 0.1× PBS, rapidly frozen in liquid nitrogen and then cryopreserved at −80 °C. After 5 days, the samples underwent FXcryo-Cell processing (Fig. 3a). The intact leaf sample captured 7,815 cells, with a median of 3,746 transcribed genes per cell, while the wounded sample captured 8,979 cells, with a median of 4,012 transcribed genes per cell (Supplementary Table 1).

To compare FXcryo-Cell with traditional protoplasting-based scRNA-seq, previously published enzymatic digestion scRNA-seq data from the Chory laboratory⁶¹ were retrieved. Batch effect correction was performed using the Harmony algorithm, integrating these datasets with those generated by FXcryo-Cell. As shown in Fig. 5a, the integrated cell atlas demonstrated strong concordance across the three datasets. Cell clusters were annotated based on established A. thaliana leaf scRNA-seq studies and known marker genes (Fig. 5b, Extended Data Fig. 8a and Supplementary Table 1).

Fig. 5 |. Probing wound response in A. thaliana leaves at single-cell resolution using FXcryo-Cell.

a, Joint analysis of intact (green) and wounded (blue) A. thaliana leaf cell atlases generated using FXcryo-Cell, alongside the intact A. thaliana leaf cell atlas (red) obtained from traditional (trad.) protoplasting-based scRNA-seq⁶¹. The Harmony algorithm was applied to correct batch effects. Different colors indicate distinct scRNA-seq datasets. b, Annotation of the integrated cell atlas from a, based on published cell-type marker genes (Extended Data Fig. 8a and Supplementary Table 1). Different colors represent distinct cell types. c, Comparative analysis of the three atlases shown in a. Different colors represent distinct scRNA-seq datasets. Wound-responsive epidermal and mesophyll cells are circled. Note that these cells were barely detected in the intact leaf atlas generated by FXcryo-Cell (green), but evident in that generated by traditional protoplasting-based scRNA-seq (red). d, UMAP plots displaying expression levels of wound-response genes across the three datasets from a. Wound-response genes extracted from Gene Ontology (GO) terms (GO:0009611 and Supplementary Table 2) were used as the gene set and analyzed using AUCell. e, Wound responsiveness across cell types. Wound-response genes extracted from GO terms (GO:0009611 and Supplementary Table 2) were the gene set. AUCell scores for each cell type were calculated and plotted. Data are presented as a box plot depicting the mean (center line), interquartile range (box bounds), and minima and maxima (whiskers). f, Upset plot presenting shared and unique wound-induced genes in each cell type. The number of genes for each cell type is presented on the top of the colored bars. Total number of wound-induced genes in each cell type is presented on the right. g, GO term analysis of unique and shared wound-induced genes in each cell type. Enrichment was tested using a one-sided hypergeometric (Fisher’s exact) test as implemented in the ‘enrichGO’ function of the R package clusterProfiler. The P values were adjusted for multiple comparisons using the Benjamini–Hochberg method. h, UMAP plots presenting representative cell-type-specific wound-induced genes. Only cells corresponding to each specific cell type are presented. The upper and lower rows indicate intact and wounded A. thaliana leaf cell atlases produced using FXcryo-Cell (c), respectively.

Compared to intact leaves, wounded leaves exhibited two additional cell clusters corresponding to epidermal and mesophyll cells (Fig. 5c). Many wound-induced genes were detected exclusively in wounded leaves⁶⁰ (Extended Data Fig. 8b). Genes associated with plant wounding responses (GO:0009611; Supplementary Table 2) were extracted and analyzed using AUCell⁶², confirming that these unique cell populations were strongly linked to wound-response pathways (Fig. 5d). Moreover, within the same cell type, newly emerged cells in wounded leaves processed using FXcryo-Cell, as well as those in intact samples processed via the traditional protoplasting method⁶¹, exhibited high enrichment of stress-response-related genes relative to other cells within the same cell cluster (Fig. 5c and Supplementary Table 3). Notably, scRNA-seq data obtained using traditional protoplasting exhibited greater similarity to FXcryo-Cell data from wounded leaves (Fig. 5c,d and Extended Data Fig. 8b), suggesting that stress-related gene expression patterns in existing published plant cell atlases may not fully reflect true biological states, potentially including cell clusters that do not naturally exist.

Importantly, comparison atlases generated from intact and wounded leaves using FXcryo-Cell enabled us to investigate wound responses in plants at the single-cell resolution under fixed-cell conditions. Interestingly, we found that different cell types did not respond uniformly to wounding: dividing cells exhibited the lowest wound responses, while epidermal, mesophyll, phloem and bundle sheath cells displayed a stronger response (Fig. 5e and Supplementary Table 2). Notably, while different cell types had common wound-response genes (Fig. 5f,g, Extended Data Fig. 8c and Supplementary Table 2), they also possessed their own unique wound-response genes (Fig. 5f–h, Extended Data Fig. 8c–e and Supplementary Table 2), with this phenomenon being particularly pronounced in mesophyll cells. Gene Ontology term analysis indicated that these cell-type-specific wound-response genes may be associated with various biological processes (Fig. 5g, Extended Data Fig. 8d and Supplementary Table 2). For instance, genes in mesophyll cells were enriched for terms linked to abscisic acid metabolic processes, programmed cell death and response to jasmonic acid. Conversely, epidermal cells were associated with terms linked to calcium ion transport (Fig. 5g). These findings indicate that plant wound perception is more complex than previously envisioned, with different cell types responding to wound signals differently. The biological significance underlying these differences requires further exploration.

To compare the performance of FXcryo-Cell and snRNA-seq, both intact and wounded leaves were subjected to snRNA-seq, and the resulting datasets were compared with those generated using the traditional protoplasting method and FXcryo-Cell. As shown in Extended Data Fig. 9a,b, the snRNA-seq data were largely integrated with the cell atlases generated by the other two methods. As expected, snRNA-seq yielded a lower median number of detected genes per nucleus (n = 900 for intact; n = 922 for wounded), compared to the protoplasting method and FXcryo-Cell (n = 3,710 for intact; n = 4,031 for wounded; Supplementary Table 1). Notably, the snRNA-seq data from intact leaves exhibited slight upregulation of wound-response genes, part of the protoplasting-induced gene set, suggesting that nuclear isolation during snRNA-seq preparation can also elicit a mild protoplasting response (that is, part of the protoplasting-affecting genes; Extended Data Fig. 9c). Consistently, minor differences in cluster identity were observed between snRNA-seq and FXcryo-Cell for epidermal and mesophyll cells in intact leaves (Extended Data Fig. 9a). These results collectively support the superior performance of FXcryo-Cell in capturing acute wounding responses in plants.

Taken together, FXcryo-Cell provides unprecedented advantages for investigating acute plant responses to external stimuli. By preserving the plant’s original state at the time of sampling, this method is particularly beneficial for experiments requiring rapid processing, especially in challenging or time-sensitive sampling conditions. In the future, FXcryo-Cell could be applied to plant–pathogen interactions and responses to diverse abiotic stresses.

Discussion

We have demonstrated that FX-Cell and its derivatives FXcryo-Cell and cryoFX-Cell significantly expand the utility of scRNA-seq in plants. As with any new methods, potential limitations must be considered. The primary advantage of these techniques lies in their ability to enhance the release of single cells from plant tissues. Nonetheless, certain contexts present challenges.

First, some plant cells have large fluid-filled vacuoles and are very fragile after fixation; for instance, we found maize leaf mesophyll cells do not hold up well to our method. For such high-water-content cells, alternative approaches may be more appropriate. We recommend initially testing this method using commercially available enzymes to assess how well the cells of interest are successfully released before proceeding to digestion with RNase-depletion enzymes (see Methods for detailed recommendations).

Second, the large size and irregular morphology of many plant cells (10–100 μm) compared to animal cells (10–30 μm) may lead to clogging in microfluidic chips used for droplet-based scRNA-seq. In contrast, protoplast-derived plant cells assume a spherical shape, making them more compatible with the microfluidic channels of such platforms. Therefore, for plant tissues and organs with large cells, a well-based scRNA-seq platform such as Rhapsody (BD) may provide a more suitable alternative. Additionally, bio-sorters such as the Hana Namocell are suitable to cells up to 40 μm in diameter, such as maize meiocytes³¹, which are compatible with most plant cell types.

Third, FX-Cell, FXcryo-Cell and cryoFX-Cell do not guarantee successful single-cell preparation for all plant tissues and organs. This limitation likely arises from the presence of specialized cell wall components in certain plants. Further optimization of digestion enzyme combinations and formulations may enhance dissociation efficiency in the future. However, due to the denaturing effect of fixation on proteins, FX-Cell is not compatible with workflows requiring fluorescence-activated cell sorting^63–65.

Taken together, FX-Cell does not resolve all bottlenecks in plant single-cell transcriptomics. In contrast, snRNA-seq retains certain advantages: it is broadly applicable across all plants, tissues and organs owing to its reliance on mechanical dissociation and the uniformity nuclear size^18,66–69. Moreover, when samples are fixed properly, snRNA-seq can prevent transcriptional alterations during nuclear suspension preparation, thereby preserving cell identity. As a result, cell types were generally well represented⁷⁰, and snRNA-seq can be reliably used to examine plant responses to environmental stimuli at single-cell resolution^19,71. Thus, a combined application of FXcryo-Cell and snRNA-seq will provide a more complete and accurate cell atlas.

Over the past two decades, scRNA-seq has revolutionized the understanding of animal cell identity, development and evolution, yet its application in plants has been slower to develop due to the extensive optimization required for protoplast isolation. FX-Cell has the potential to overcome these barriers, facilitating discoveries across diverse plant species and tissues. Ultimately, these methods could lay the foundation for the development of Plant Cell Atlases for every target species^26,27.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41592-025-02900-2.

Methods

Plant materials and growth conditions

For maize anther, root tip, shoot apical meristem, leaf and ear samples (Fig. 1a–c and Extended Data Fig. 1a,c), Zea mays (inbred line W23 bz2) individuals were grown under greenhouse conditions in Stanford, USA (14 h light/10 h dark). Daily irrigation and fertilization were maintained for robust growth. Beginning 5 to 6 weeks after planting, individual plants were felled ~20 cm above ground level for anther dissection between 8:00 and 9:00. The collected plants were taken to the laboratory within 10 min where the tassels were dissected out of the stem and leaf whorl. For the maize crown root sample (Fig. 4d), Z. mays (inbred line B73) were germinated in the greenhouse, and the 2-week-old seedlings were transplanted to the field in Shanghai. The crown roots were harvested from 45-day-old plants.

The seeds of O. sativa (ZH11) were sown on filter paper soaked in water in petri dishes. After 5 days at 28 °C (10 h light/14 h dark), the root tips (Extended Data Fig. 3c) and tiller nodes (insets in Fig. 4a) were dissected and digested. O. longistaminata (wild rice) seedlings were grown in a greenhouse at 28 °C (12 h light/12 h dark). The rhizome nodes were harvested for digestion (insets in Fig. 4b). S. martensii seedlings were purchased from the market and cultivated in the greenhouse at 21 °C (16 h light/8 h dark). The shoot apices were harvested for digestion. The seeds of A. thaliana (Col-0) plants were grown in the greenhouse at 21 °C (16 h light/8 h dark), and the third-to-last true leaves of 24-day-old plants were harvested for digestion (intact sample). For wounding treatment, the leaves were wounded by a syringe needle and harvested for digestion (wounded sample). A. trichopoda and N. colorata were grown under tropical (27 °C) greenhouse conditions in Stanford (12 h light/12 h dark).

Chemicals and reagents

The sources for all the chemicals and reagents used in this study are given in Supplementary Table 1.

Maize anther dissection and protoplasts/cells preparation

A Leica M60 dissecting scope and stage micrometer (Thermo Fisher Scientific) were used to isolate 2.0-mm anthers from the upper florets of spikelets along the central spike of the tassel. Protoplast/cell release of fixed and fresh maize anthers was compared in a variety of conditions. Three 2.0-mm anthers were pooled per replicate with five replicates per condition. Fresh anthers were digested at 30 °C for 90 min or 16 h in the enzyme mix (1.25% wt/vol cellulase-RS, 0.5% pectolyase Y-23, 0.5% macerozyme-R10, 0.5% hemicellulose) from Nelms and Walbot³¹. Fixed samples were left in ice-cold Farmer’s solution (3:1 100% ethanol:glacial acetic acid) for 2 h, washed twice in ice-cold 0.1× PBS (Sigma-Aldrich) for 5 min, then digested at 30 °C or 50 °C for 90 min in 20 mM 4-morpholineethanesulfonic acid (MES), pH 5.7, with a 1:10 dilution of RNase-depleted reduced enzyme mix stock (enzymes were stored in glycerol stocks, see below; stocks were normalized so that a 1:10 dilution has the same A280 as a 1.25% wt/vol cellulase-RS and 0.4% wt/vol macerozyme-R10 solution). The protoplasts/cells from the digested, fixed anthers were dissociated via shear force between two microscope slides with thin tape as a spacer. For each replicate, the number of protoplasts/cells was estimated using a hemocytometer and then averaged. Images of the dissociated cells were taken on a Nikon Diaphot inverted microscope with a mounted Nikon D40 camera.

Protoplast/cell preparation from other tissues and species

Protoplast/cell release from maize root tips was compared in three conditions (Extended Data Fig. 1c): (1) fresh protoplasting as described⁵³, (2) fixation and digestion with the enzyme concentrations as described⁵³, or (3) fixation and digestion with reduced enzyme mix. Maize seeds were treated, germinated and grown as described⁵³. Seedling primary roots were cut 5 mm above the tip with a scalpel. Three root tips were pooled per replicate with five replicates per condition. Fresh root tips were pretreated, washed and digested in the enzyme mix as described⁵³, filtered, and washed. The fresh protoplasts were counted with a hemocytometer. The fixed samples were left in ice-cold Farmer’s solution for 2 h, washed twice with ice-cold 0.1× PBS for 5 min then digested at 50 °C with the enzyme mix as described⁵³ or reduced enzyme mix as described above (Extended Data Fig. 1c). Digested tissue was manually disrupted with pipetting, then the number of individual cells counted with a hemocytometer.

We expanded our sampling by comparing our method to a standard protoplasting protocol³¹ in three additional maize tissues (apical meristems, young leaves, young ears) and four non-model plant taxa and tissues (leaves from the basal angiosperms A. trichopoda and N. colorata (waterlilly), and leaf and stem tissue from the non-model Brassicaceae C. bursa-pastoris; Fig. 1c). For each tissue, comparably sized samples were either fixed in Farmer’s solution, washed twice 0.1× PBS and digested at 50 °C with reduced enzyme mix or directly digested at 30 °C for 90 min in the enzyme mix from Nelms and Walbot³¹. For each replicate, the number of protoplasts/cells was estimated using a hemocytometer and then averaged. Images of the dissociated cells were taken on a Nikon Diaphot inverted microscope with a mounted Nikon D40 camera.

For plant tissues that cannot not be digested enzymatically using conventional protoplasting protocols, we recommend preparing a digestion buffer that contains a combination of multiple digestion enzymes, including cellulase, pectinase, macerozyme, hemicellulase, laccase, driselase⁷², xylanase, mannanase and glucanase. The tissues are then digested with the FX-Cell method without RNase depletion. If single cells can be effectively released, we suggest systematically removing one enzyme at a time to assess the efficiency of the digestion, thereby determining which enzymes are essential. This approach could lead to the development of a simple and efficient enzyme formulation for FX-Cell, as excessive enzyme amounts can impair RNase depletion. Conversely, if single cells cannot be effectively released, the degradation of the cell wall may be a limiting step, necessitating the identification of specific enzymes to ensure effective digestion.

Single-cell preparation from rice tiller nodes, wild rice rhizome nodes and shoot apices of S. martensii

Plant samples were harvested in ice-cold Farmer’s solution and fixed for 2 h, followed by two washes in ice-cold 0.1× PBS for 5 min each. The plant samples were then shredded and digested at 50 °C for 90 min in either C + M cocktail solution (that is, reduced enzyme mix, 1.25% cellulase-RS, 0.4% macerozyme-R10, 0.4 M mannitol, 20 mM MES, pH 5.7, 10 mM KCl, 10 mM CaCl₂ and 0.1% BSA) or C + M + S + P cocktail solution (1% cellulase-RS, 0.5% macerozyme-R10, 1% snailase, 0.5% pectinase, 0.4 M mannitol, 20 mM MES, pH 5.7, 10 mM KCl, 10 mM CaCl₂ and 0.1% BSA). The samples were gently sucked and blown to fully release the cells. The cells were filtered twice using cell strainers (40-μm diameter; Falcon, 352340), spun at 300g for 3 min and washed three times with 0.1× PBS at room temperature (RT). The concentration of cells was measured using a hemocytometer (Fig. 1d and Extended Data Fig. 1b).

Preparation of GMP-Sepharose/agarose

To prepare the GMP-Sepharose, we proceeded as follows (see the step-by-step protocol in Supplementary Methods): Add 1.028 g NaIO₄ to 24 ml of ultrapure H₂O (that is, 0.2 M NaIO₄). After NaIO₄ is completely dissolved, add 1.952 g guanosine 5′-monophosphate disodium salt (5′-GMP), and incubate at RT, protected from light, with gentle shaking for 1 h. The resultant solution is called oxidized GMP solution.

Resuspend 25 ml settled volume of NH₂-Sepharose (Shanghai Puke, RF1052) by adding 50 ml of ultrapure H₂O, and then filter to remove the liquid with the Büchner funnel. Repeat this step twice more. Add 50 ml of 0.1 M borax (pH 9.0), thoroughly suspend NH₂-Sepharose and filter to remove the liquid with the Büchner funnel. Repeat this step two more times. Add 72 ml of 0.1 M borax (pH 9.0), fully resuspend NH₂-Sepharose and transfer it to a glass beaker. Wash the Büchner funnel with another 72 ml of 0.1 M borax (pH 9.0) once, and transfer the mixture to the same beaker. Take 3 ml of NH₂-Sepharose and add 0.5 ml of 0.2 M GMP (non-oxidized GMP). Mix thoroughly by shaking and place it at 4 °C (this sample serves as a negative control for evaluation of the coupling efficiency).

Add 200 μl of 50% glycerol to the aforementioned oxidized GMP solution and incubate at RT with gentle shaking for 30 min. Add 24 ml of this solution to 141 ml of NH₂-Sepharose, mix thoroughly to suspend NH₂-Sepharose, cover the beaker with plastic wrap and incubate at RT with gentle shaking for 2 h to 4 h. At this point, NH₂-Sepharose has initially coupled with GMP, and we refer to it as GMP-Sepharose thereafter.

Transfer the GMP-Sepharose to a Büchner funnel to remove the liquid. Add 50 ml of 0.1 M borax (pH 9.0), fully suspend GMP-Sepharose and filter to remove the liquid. Repeat this step two more times. Add 84 ml of 0.1 M borax (pH 9.0), thoroughly shake to resuspend GMP-Sepharose and transfer it to a glass beaker. Wash the Büchner funnel with another 84 ml of 0.1 M borax (pH 9.0) once, and transfer the mixture to the same beaker.

Slowly add 544 mg NaBH₄, incubate at 4 °C with gentle shaking for 1 h (cover with plastic wrap, poking a small hole to ensure ventilation) and remove the liquid with the Büchner funnel. Wash the GMP-Sepharose three times with 50 ml of 0.1 M borax (pH 9.0), once with 50 ml of H₂O, once with 1 M NaCl and finally remove the liquid with the Büchner funnel. Transfer GMP-Sepharose out and dilute it to 50 ml with 1 M NaCl for long-term storage. The same procedure can be used to prepare the GMP-Agarose.

Evaluation of the coupling efficiency of GMP-Sepharose

To evaluate the coupling efficiency of GMP-Sepharose, we proceeded as follows (see the step-by-step protocol in Supplementary Methods): Take 1 ml of GMP-Sepharose, remove the liquid by centrifugation, and do the same with the NH₂-Sepharose incubated with non-oxidized GMP. Transfer both to separate 50-ml centrifuge tubes, add 40 ml of H₂O, and incubate at 4 °C with rotation for 30 min, then centrifuge to collect the precipitate. Wash GMP-Sepharose and NH₂-Sepharose with H₂O twice, using 40 ml each time. Take approximately 100 μl settled volume of solid GMP-Sepharose and NH₂-Sepharose, add 0.5 ml of 1 M HCl, and heat in a boiling water bath for 1 h. Once the samples have cooled, use a Nanodrop 2000 to measure the DNA concentration, detecting the absorbance in the mode for DNA (using 1 M HCl as the blank). The high-temperature acid hydrolysis of GMP-Sepharose will produce guanine, which has the maximum absorbance at 248 nm. The GMP-coupled NH₂-Sepharose will show significantly higher absorbance at 248 nm compared to the control NH₂-Sepharose (Extended Data Fig. 1f).

Depletion RNase from digestion enzymes using GMP-Sepharose/agarose

To deplete RNase from digestion enzymes, we proceeded as follows: Resuspend the cell digestion enzymes in RNase binding buffer (RBB; 150 mM NaCl, 10 mM citrate, pH 7.0). For the samples in Fig. 1b,c,f, we used C + M cocktail. For other samples, we used the C + M + S + P cocktail. Spin at 5,000 rpm at 4 °C for 5 min, and collect the supernatant. Repeat this step once. Take 100 μl of this enzyme solution for later use and measure the protein concentration using the Protein A280 mode on a Nanodrop 2000.

We used a two-step method involving gravity chromatography column and FPLC column to purify the digestion enzymes (see the step-by-step protocol in Supplementary Methods): Load 4 ml of GMP-Sepharose onto a gravity chromatography column. Fill the remaining volume with 21 ml of GMP-Sepharose in an FPLC column (Aogma, XK-1030A). Equilibrate the GMP-Sepharose gravity chromatography column three times with 5 ml of RBB. Slowly add the enzyme solution to the GMP-Sepharose gravity chromatography column, collecting the effluent. After all the enzyme solution has passed through the resin, slowly add an additional 6 ml of RBB to the resin surface and collect the effluent. Transfer the effluent to a Sartorius 15 ml 10-kDa molecular weight cutoff polyethersulfone ultrafiltration tube (Sartorius, FUF151) and centrifuge at 4,000g and 4 °C to concentrate the protein. When the protein concentration approaches that of the original enzyme solution, transfer it to a 2-ml centrifuge tube. To regenerate the column, wash the GMP-Sepharose gravity chromatography column three times with 5 ml of ultrapure H₂O, three times with 5 ml of RNAse elution buffer (5 M NaCl, 1 mM 5′-GMP, 1 mM 2′(3′)-GMP), three times with 5 ml of 1 M NaCl. Add 4 ml of 1 M NaCl and store the GMP-Sepharose column at 4 °C for repeated use.

Equilibrate the GMP-Sepharose FPLC column with 45 ml of RBB. Load 1 ml the concentrated enzyme onto the GMP-Sepharose FPLC column. Collect fractions and pool those with A280 absorbance >0.1. Concentrate pooled enzymes with a Sartorius 15-ml 10-kDa molecular weight cutoff polyethersulfone ultrafiltration tube. When the protein concentration reaches that of the original enzyme solution, transfer it to a 2-ml centrifuge tube. To regenerate the GMP-Sepharose FPLC column, wash the column with 45 ml RNAse elution buffer and 45 ml 1 M NaCl. To further reduce RNase (Fig. 2d), the obtained enzyme solution can be purified again using a GMP-Sepharose FPLC column. Finally, we obtained a 10× RNase-depleted digestion enzyme stock solution, at the same concentration as the original solution.

RNase activity assays

To prepare the enzyme solution, the 10× original enzyme or 10× RNase-depleted enzyme was diluted to a 1× concentration using RBB. RNase inhibitors (1 μl μl⁻¹, Roche; New England Biolabs, M0314L), EDTA (1 mM), tRNA (1 μg μl⁻¹) and tri-GMP (1 mM) were then added, as illustrated in Fig. 2d. RBB buffer was used as a control. RNase activity was quantified using the RNAse Activity Fluorescence Detection Kit (Abcam, ab273299) with a white 96-well microplate at 24 °C (Fig. 2d and Extended Data Fig. 1e,g).

Assessment of RNA integrity

Maize anther RNA quality was tested in four conditions of cell preparation (Fig. 1f): (1) flash frozen, (2) fixed in Farmer’s solution then washed twice in 0.1× PBS then flash frozen, (3) fixed, washed and digested in commercial enzymes, and (4) fixed, washed and digested in RNase-depleted enzymes purified by GMP-Agarose. For each condition, 2.0-mm anthers were isolated from five separate plants with ten anthers pooled per plant. The flash frozen samples were homogenized via bead beating in a 2000 Geno/Grinder (SPEX CertiPrep) with baked 4-mm steel balls. The fixed samples were left in ice-cold Farmer’s solution for 2 h, washed twice in ice-cold 0.1× PBS for 5 min, then incubated at 50 °C for 90 min with RNase-depleted or commercial enzymes (1.25% wt/vol cellulase-RS and 0.4% wt/vol macerozyme-R10). The RNeasy Plant Mini Kit (Qiagen) was used to extract RNA from samples via the standard protocol. RNA was quality-checked on an Agilent 2100 BioAnalyzer with the RNA 6000 Nano assay (Agilent Technologies). The RIN for the five replicates of each condition were averaged and reported alongside the error.

Using a similar protocol, we examined the RNA qualities of the rice root tip samples that were fixed and digested with the C + M cocktail at different temperatures and times (Extended Data Fig. 1i,k), as well as the samples prepared by different methods (Extended Data Fig. 5c). Additionally, we assessed RNA integrity through RNA gel electrophoresis (Extended Data Fig. 1h,j).

Fixed maize anther cell preparation for CEL-Seq2

Anthers from four individuals of wild-type maize (inbred line W23 bz2) were dissected out. One of the three anthers per floret was used for imaging on a Nikon Diaphot inverted microscope with a Nikon D40 mounted camera at a magnification of ×10. The remaining two anthers per floret were fixed in ice-cold Farmer’s solution for 2 h, washed twice for 5 min in 0.1× PBS, and then one anther was digested at 50 °C for 90 min in the RNase-depleted enzyme mix, while the other anther was saved at −20 °C. Following digestion, shear force was applied to the anther between two microscope slides with thin tape on each end to prevent the anther from being fully crushed. The top microscope slide was slid back and forth 5–10 times and the sample checked under the dissecting scope to ensure separation of the fixed cells. The cells were washed from the slides into 1 ml of cold 0.1× PBS via pipette and stained with SYBR Green I nucleic acid gel stain (Invitrogen) for 20 min. The cells were then filtered through a 100 μm (if bound for the BioSorter) or 40 μm (if bound for the Hana) nylon cell strainer (Corning) into 50-ml Falcon tubes. The stained cells were then sorted into 384-well plates or 96-well plates, each well containing 0.8 μl Primer Master Mix (0.225% Triton X-100, 1.6 mM dNTP mix, 1.875 μM barcoded oligo(dT) CEL-Seq2 primers; Sigma-Aldrich, New England Biolabs) using a BioSorter (Union BioMetrica) or Hana Single Cell Dispenser (Namocell). Following cell sorting, the plates were spun at 400g and then stored at −80 °C.

CEL-Seq2 library preparation and data analysis

Single-cell libraries were prepared following the CEL-Seq2 protocol with modifications³⁶. Read filtering, mapping, initial processing and cell clustering for the CEL-Seq2 dataset were performed as described³⁶. To initially compare our dataset with that of known cell types, we assessed the similarity of our data with laser-capture microdissection (LCM) sequencing data of known cell types and whole anthers⁷³, which were also prepared from 2.0-mm W23 maize anthers using the same CEL-Seq2 library preparation. UMIs were normalized into transcripts per million (TPMs) and log transformed after adding a pseudocount of 100. We then subtracted the single-cell TPMs by the log-transformed TPMs of the whole anthers to produce ratio measurements. The LCM data had samples for tapetal, meiocyte and other somatic (middle layer, endothecium, epidermis) cell types and were similarly processed relative to the whole-anther data⁷⁴. We then calculated the cell-to-cell Pearson’s correlations of all single cells relative to each of the LCM samples.

Clusters were determined and visualized by UMAP with a resolution of 0.01 (Extended Data Fig. 2a). The data generated by the cells sorted by BioSorter or Hana were highly similar (Extended Data Fig. 2b). Correlation values of each cell with the LCM tapetal, meiocyte and other somatic cell types were mapped onto the UMAP (Extended Data Fig. 2c–e), as well as the percentages of transcripts from the plastid genome and mitochondrial genome (Extended Data Fig. 2f). The meiocyte cluster was manually separated from the endothecium cluster based on the LCM correlation data and meiocyte marker genes; it is likely that Monocle did not separate these clusters due to the scarcity of meiocyte cells despite the clear separation in the UMAP. The other somatic 1 cluster was subset and reclustered to identify and separate the epidermis cluster based on putative marker genes of the known biology of the cell type (Extended Data Fig. 2g–j).

snRNA-seq on rice root tips and Arabidopsis leaves

To generate snRNA-seq data for rice root tips and Arabidopsis leaves (Fig. 2 and Extended Data Fig. 9), we followed these steps: First, prepare a 1× NIB buffer from the 4× NIB buffer stock solution (Sigma-Aldrich, CELLYTPN1), and add 1× cocktail (Roche, 4693132001) along with 0.2 U μl⁻¹ RNase inhibitor (Thermo Fisher Scientific, N8080119). Homogenize rice root tips (1.0 cm in length) or Arabidopsis leaves (intact or wounded) using a blade in 200 μl of 1× NIB buffer. Transfer the homogenate to a 15-ml conical tube and adjust the volume to 2 ml by adding 1× NIB buffer. Incubate the mixture at 4 °C with rotation for 15 min. Next, filter the suspension through a 40-μm cell strainer into a 15-ml conical tube. Add 100 μl of 10% Triton X-100 and mix thoroughly. Centrifuge at 1,000 g at 4 °C for 10 min, carefully discarding the supernatant. Resuspend the pellet in 1 ml of 1× PBS (supplemented with 1× cocktail, 0.1% BSA and 0.2 U μl⁻¹ RNase inhibitor) and add DAPI (4′,6-diamidino-2-phenylindole) to achieve a final concentration of 1 μg ml⁻¹. Perform fluorescence-activated nucleus sorting, loading 200 μl of 1× PBS (supplemented with 1× cocktail, 0.1% BSA and 0.2 U μl⁻¹ RNase inhibitor) as the landing buffer. Sort 200,000 nuclei, then centrifuge at 1,000g at 4 °C for 10 min. Discard the supernatant, add 30 μl of 1× PBS (supplemented with 1× cocktail, 0.1% BSA and 0.2 U μl⁻¹ RNase inhibitor), and resuspend the nuclei. Count the number of nuclei using a microscope, and conduct snRNA-seq using the Chromium Next GEM Single Cell 3ʹ Reagent Kits v3.1 (10× Genomics, PN-1000268), following the manufacturer’s instructions. Libraries are sequenced in 150-base pair paired-end mode on a DNBSEQ-T7 sequencer (GenePlus).

FX-Cell on rice and Arabidopsis root tips

To generate cell atlases of rice and Arabidopsis root tips by FX-Cell (Fig. 2 and Extended Data Fig. 4a,b), we proceeded as follows (see step-by-step protocol in Supplementary Methods): Place the rice root tips (1 cm in length) or Arabidopsis root tips (0.5 cm in length) into Farmer’s solution in a 2-ml centrifuge tube and apply a vacuum until the material settles to the bottom of the tube. Incubate on ice for 30 min. Add pre-chilled 0.1× PBS to the fixed material, mix well, and incubate on ice for 5 min. Discard the supernatant. Repeat this step once. Prepare the cell wall digestion enzyme solution with the following components: 0.4 M mannitol, 20 mM MES (pH 5.7), 10 mM KCl, 10 mM CaCl₂, 0.1% BSA, 1× RNase-depleted C + M + S + P digestion enzymes (diluted from aforementioned 10× RNase-depleted digestion enzyme stock solution), 1 mg ml⁻¹ tRNA, and 1 mM tri-GMP (a mixture of 0.1 M 5′-GMP and 0.1 M 2′(3′)-GMP). Filter the enzyme solution using a 0.45-μm filter. Transfer the material to a small Petri dish (diameter 30 mm), add an appropriate amount of enzyme solution, and incubate with shaking at 40 °C and 80 rpm for 10 min. Disrupt the material and incubate with shaking at 40 °C and 80 rpm for 20 min. Filter the material through a 40-μm cell strainer into a 2-ml centrifuge tube, centrifuge at 4 °C and 300g for 3 min, and discard the supernatant. Resuspend the cells in 1 ml of 0.1× PBS, filter the cell suspension through a 40-μm cell strainer into a 2-ml centrifuge tube, centrifuge at 4 °C and 300g for 3 min, and discard the supernatant. Wash the cells with 1 ml of 0.1× PBS, centrifuge at 4 °C and 300g for 3 min and discard the supernatant. Repeat this step once. Resuspend the cells in 50 μl of 0.1× PBS and count the number of cells with a microscope. Use the Chromium Next GEM Single Cell 3ʹ Reagent Kits v3.1 (10× Genomics, PN-1000268) for preparation of scRNA-seq library, following the manufacturer’s instructions. Libraries are sequenced in a 150-base pair paired-end mode on a DNBSEQ-T7 sequencer (GenePlus).

FXcryo-Cell and cryoFX-Cell

To generate cell atlases of rice root tips, cultivated rice tiller nodes, wild rice rhizome nodes and wounded A. thaliana leaves by FXcryo-Cell (Figs. 3, 4a,b and 5), we proceeded as follows (see step-by-step protocol in Supplementary Methods): Place the plant material into ice-cold Farmer’s solution and apply a vacuum until the material settles to the bottom of the tube. Incubate on ice for 30 min. Add pre-chilled 0.1× PBS to the fixed material, mix well and incubate on ice for 5 min. Discard the supernatant. Repeat this step. Freeze the material in liquid nitrogen and transfer it to a −80 °C freezer for long-term storage. To prepare the cells, thaw the frozen material at RT for 5 min and digest it with the RNase-depleted C + M + S + P digestion enzyme solution at 40 °C for 30 min as described above. All other steps follow the FX-Cell protocol.

To generate cell atlases of the rice root tips (Fig. 3e) and the crown roots from field-grown maize plants (Fig. 4d) by cryoFX-Cell, we proceeded as follows: Collect and freeze the material in liquid nitrogen and transfer it to a −80 °C freezer for long-term storage. To prepare the cells, thaw the frozen material at RT for 5 min, fix with Farmer’s solution and digest with the RNase-depleted C + M + S + P digestion enzyme solution at 40 °C for 30 min as described above. All other steps follow the FX-Cell protocol.

seq-FISH assay

Fresh rice tiller nodes and wild rice rhizome nodes quickly harvested and rapidly frozen in liquid nitrogen-cooled isopentane for 1 min, before being embedded in optimal cutting temperature (OCT) compound on dry ice. Cryosections with a thickness of 16 μm were prepared using a Leica cryostat CM3050S (Leica, Germany) and mounted onto adhesive slides. seq-FISH was performed to detect spatial RNA expression following established methods^75,76. The procedure has been modified, with detailed reagent formulations provided below. All other steps remain unchanged from the Lin’s method⁷⁶.

In brief, the fresh frozen sections were first incubated at 37 °C for 5 min and then fixed in 4% (wt/vol) paraformaldehyde for an additional 5 min. Following fixation, the sections were washed twice with diethyl pyrocarbonate-treated 1× PBS. Next, they were treated with the HyperView Quench Buffer in LUMINIRIS GEM (LUMINIRIS, China) for photobleaching to minimize fluorescence background. The samples were dehydrated through a series of ethanol concentrations and subsequently washed three times with 1× PBS-T (PBS with 0.1% Tween 20), before being treated with 0.1 M HCl at 37 °C for 5 min. The sections were then incubated sequentially with custom target gene probes (Sangon, China) in a hybridization mix at 37 °C for 4 h. The sequences of the target genes are provided in Source Data.

Following hybridization, the sections were incubated with ligation mixture containing 0.5 U μl⁻¹ SplintR Ligase, 1× SplintR reaction buffer, 50% glycerol, 0.2 μg μl⁻¹ bovine serum albumin and 1 U μl⁻¹ RNase Inhibitor (Roche, 3335399001) at 37 °C for 30 min. Subsequently, circularization was performed using a solution composed of 0.1 U μl⁻¹ T4 DNA ligase, 0.5 μM splint oligonucleotide (Sangon, China, TCTTAAACAGCTTGATACCG), 0.2 μg μl⁻¹ BSA, 1 U μl⁻¹ RNase Inhibitor and 1× T4 DNA ligase reaction buffer, with incubation at 37 °C for 30 min. Finally, rolling-circle amplification was carried out in an amplification mix containing 1× phi29 DNA polymerase buffer, 5% glycerol, 1 mM dNTPs, 1 U μl⁻¹ phi29 DNA polymerase and 0.2 μg μl⁻¹ BSA, at 30 °C overnight. After each incubation step, the slides were washed three times with 1× PBS-T at RT.

For fluor-probe hybridization, the sections were incubated with hybridization buffer containing bridge probes and fluorescent probes at 30 °C for 30 min. After washing three times with 1× PBS-T, the sections were mounted with SlowFade Gold Antifade Mountant with DAPI. Imaging was performed using a spinning-disk confocal microscope (Spin SR, Evident) equipped with sCMOS camera (Fusion, HAMAMATSU). After imaging, coverslips were removed by vertically incubating slides in 1× PBS. Fluorescence signals were stripped using removal buffer at 37 °C for 5 min, followed by three 1× PBS-T washes. Subsequent hybridization cycles used bridge/fluorescence probes for the next round of targets, repeating hybridization, washing with PBS-T, mounting with SlowFade Gold Antifade Mountant with DAPI and confocal imaging to acquire sequential seq-FISH signals.

Genome assembly and annotation of O. longistaminata

Initial genome assembly was performed using PacBio HiFi long-read data with hifiasm (v0.19.5) under default parameters. Chromosome-scale assembly was then achieved using RagTag (v2.1.0) for scaffolding⁷⁷. The mitochondrial genome was assembled using oatk (v1.3.2)⁷⁸ with specific reference files for mitochondrial and plastid assembly, using parameters ‘--m angiosperm_mito.fam’ and ‘--p angiosperm_pltd.fam’, respectively. The assembled mitochondrial sequences were aligned to the nuclear genome using minimap2 (v2.24)⁷⁹ with parameters ‘--eqx -a -x asm20’. Subsequently, nuclear-mitochondrial DNA (NUMT) regions with sequence identity ≥95% were filtered out. The final genome assembly integrated the validated organellar and nuclear genomes.

Repeat annotation was conducted in two phases: First, a nonredundant transposable element library was constructed using EDTA (v2.0.0) with default plant parameters⁸⁰. The genome was then masked for repeats using the EDTA-derived transposable element library. Gene structure annotation was completed using egap (v1.0.3) with the following configurations: (1) Taxonomic constraints (taxid = 4,528, specific to O. sativa); (2) Protein sequence evidence from the IRGSP-1.0 reference genome (obtained from RAP-DB, https://rapdb.dna.affrc.go.jp/); (3) Tissue-specific RNA-seq data encompassing aerial (leaf/stem) and underground (root/rhizome) developmental stages of the wild-type plants. As shown in Extended Data Fig. 10, the genomes of O. sativa and O. longistaminata exhibited good synteny.

Analysis of the scRNA-seq data generated by the 10× Genomics platform

Raw sequencing data were processed according to the manufacturer’s recommended pipeline (https://www.10xgenomics.com/cn/support/software/cell-ranger/latest). We used Cell Ranger (v7.0.0) to perform all preprocessing steps, including index building and gene expression quantification. The reference genome (TAIR10) and annotation (Araport11) for A. thaliana were obtained from TAIR (https://www.arabidopsis.org/). The reference genome (IRGSP-1.0) and annotation (2021–05-10) for O. sativa were obtained from rap-db (https://rapdb.dna.affrc.go.jp/index.html). The reference genome (B73 v5) and annotation (Zm00001eb.1) for maize were obtained from maizeGDB (https://www.maizegdb.org/). The genome assembly and annotation of O. longistaminata were generated in this study (see above). We implemented a comprehensive analytical workflow using Seurat v5 (https://satijalab.org/seurat/)⁸¹. This workflow included quality control, data normalization, dimensionality reduction, clustering, data integration and differential expression analysis. While maintaining consistent quality-control parameters within species, we applied species-specific filtering criteria across different species. Detailed quality-control metrics are provided in Supplementary Table 1. Following quality-control steps, we performed data normalization using the ‘NormalizeData’ function and scaled the expression values with the ‘ScaleData’ function to enable cross-cell comparisons. Dimensionality reduction and clustering were achieved by successive applications of ‘RunPCA’, ‘RunUMAP’, ‘FindNeighbors’ and ‘FindClusters’ functions. Differentially expressed genes were identified using the ‘FindAllMarkers’ and ‘FindMarkers’ functions.

Integration of scRNA-seq datasets

To mitigate the impact of cell number discrepancies in published datasets during data integration, we used stratified sampling for the published datasets of rice root tip³⁷ and Arabidopsis leaf⁶¹, retaining 51% and 38% of cells, respectively (Supplementary Table 4). For integrating the published datasets of rice root tips³⁷, maize root tips⁵³ and Arabidopsis leaves⁶¹ with those generated by FX-Cell, cryoFX-Cell or FXcryo-Cell, we used the ‘IntegrateLayers’ function in Seurat⁸¹. For merging the rice root tip cell atlases generated by FX-Cell, FXcryo-Cell and cryoFX-Cell, we applied the ‘merge’ function in Seurat⁸¹. Subsequent dimensionality reduction and clustering analyses followed the aforementioned workflow.

Cell-type annotation

Cell-type annotation for all single-cell datasets in this study was based on cell-type marker genes derived from published literature (Supplementary Table 1). For wild rice datasets, the orthologous genes of O. sativa and A. thaliana in wild rice were identified by BlastP, which facilitated cell-type annotation through the use of established marker genes from these two species.

GO enrichment analysis and AUCell score calculation

GO enrichment analysis for Arabidopsis was conducted using clusterProfiler (v4.4.1)⁸² combined with the org.At.tair.db (v3.15.1) annotation package, with all parameters set to their default values. We retrieved comprehensive GO term information for the target species from the GO database (https://geneontology.org) to construct pathway-specific gene sets. Cell-specific pathway (GO:0009611) activity scores were then calculated using the ‘AUCell_run’ function implemented in the AUCell package (v1.18.1)⁶², followed by downstream visualization analysis using complex heatmap^83,84.

Overlap coefficient analysis

To evaluate the consistency between our FX-Cell datasets and published datasets, we calculated the Szymkiewicz–Simpson coefficient (overlap coefficient)⁸⁵ for marker gene sets identified by ‘FindAllMarkers’ (default parameters) in the integrated rice root tip data:

$$\mathrm{o}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{a}\mathrm{p}(\mathrm{m}\mathrm{F}\mathrm{X},\mathrm{m}\mathrm{P}\mathrm{u}\mathrm{b})=\frac{\left|\mathrm{m}\mathrm{F}\mathrm{X}\bigcap \mathrm{m}\mathrm{P}\mathrm{u}\mathrm{b}\right|}{\mathrm{m}\mathrm{i}\mathrm{n}\left(\right|\mathrm{m}\mathrm{F}\mathrm{X}|,|\mathrm{m}\mathrm{P}\mathrm{u}\mathrm{b}\left|\right)}$$

where mFX represents markers from FX-Cell data and mPub denotes markers from published data. An overlap coefficient ≥0.4 was considered to indicate significant similarity between gene sets.

The GO enrichment analysis was performed for each cluster’s marker genes using clusterProfiler (v4.4.1). The overlap coefficient for GO terms was calculated as:

$$\mathrm{o}\mathrm{v}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{a}\mathrm{p}(\mathrm{g}\mathrm{o}\mathrm{F}\mathrm{X},\mathrm{g}\mathrm{o}\mathrm{P}\mathrm{u}\mathrm{b})=\frac{|\mathrm{g}\mathrm{o}\mathrm{F}\mathrm{X}\cap \mathrm{g}\mathrm{o}\mathrm{P}\mathrm{u}\mathrm{b}|}{\mathrm{m}\mathrm{i}\mathrm{n}\left(\right|\mathrm{g}\mathrm{o}\mathrm{F}\mathrm{X}|,|\mathrm{g}\mathrm{o}\mathrm{P}\mathrm{u}\mathrm{b}\left|\right)}$$

where goFX and goPub represent GO terms from FX-Cell data and published data, respectively.

Statistical analysis and reproducibility

Student’s t-test was performed to determine the statistical significance between different samples in quantification and phenotypes. All statistical results and graphs were generated by GraphPad Prism 8 (https://www.graphpad.com/). The numbers of biological replicates and types of statistical analyses are given in the figure legends.

Extended Data

Extended Data Fig. 1 |. Supporting data for the generation of the fixation and digestion at high temperatures protocol.

a, Representative image showing the released cells from different tissues and species. Varying plant tissues were digested with the fixation and digestion at high temperatures protocol. See quantification results in Fig. 1c. b, Images showing the cells prepared from fixed rice tiller nodes, wild rice rhizome nodes, and shoot apices of Selaginella martensii. The tissues were subjected to fixation and digestion at high temperatures using a combination of two enzymes (that is, reduced enzyme mix, C + M) or four enzymes (C + M + S + P). Red arrows indicate single cells. Please note that the number of released cells was greatly increased when tissues were digested by a combination of four enzymes. Scale bars, 50 μm. See quantification results in Fig. 1d. c, Quantification of released protoplasts or cells from maize root tips. We compared the optimized protoplasting protocol from Ortiz-Ramirez et al.⁵³ and our fixation-based protocol. Maize root tips were digested following Ortiz-Ramírez et al. and released protoplasts were quantified via hemocytometer (grey bar, 1). Maize root tips were fixed then digested at 50 °C for 90 min using the Ortiz-Ramírez et al. (2018) protoplast enzyme mix (blue bar, 2) or reduced enzyme mix (blue bar, 3). The released cells were similarly quantified. Data are the mean ± s.e.m. from five independent biological replicates. Different letters denote statistically significant variation (two-sided Student’s t-test, p < 0.05, with no adjustment for multiple comparisons). The p-values are provided in Source Data. d, The addition of vanadyl ribonucleoside complexes (VRC) leads to production of black precipitates (red arrow) in the digestion solution. e, Assessment of RNase enzymatic activity in the digestion buffer. We evaluated the effectiveness of commercial RNase inhibitors (Roche, Cat. No. 3335399001; New England Biolabs, Cat No. M0314L), EDTA, triGMP, and tRNA on the RNase activity of the original digestion enzyme blends. “purif.” indicates that RNase was removed from the digestion solution via affinity chromatography. “Blank” utilizes the enzyme-dissolving RNase binding buffer (RBB) as a control. Data are the mean ± s.d. from three independent experiments. f, Evaluation of the coupling efficiency of GMP onto NH₂-Sepharose. The uncoupled NH₂-Sepharose exhibited an absorbance of 5.6 Abs at 248 nm, while the NH₂-Sepharose successfully coupled with GMP (GMP-Sepharose) displayed an absorbance of 10.1 Abs at 248 nm. g, Assessment of GMP-Sepharose column regeneration efficiency. We compared the performance (that is, RNase enzymatic activity in purified digestion enzyme blends) of columns that underwent regeneration (second purification with regeneration) versus those that did not undergo regeneration (second purification without regeneration). Data are the mean ± s.d. from three independent experiments. h, j, Assessment of RNA integrity through electrophoresis analysis. Rice roots were digested at various temperatures (h) and times (j). RNAs were extracted and subjected to electrophoresis analysis. Please note that RNA integrity decreased with increased time and temperature. On the far left is the DNA ladder. i, k, Assessment of RNA integrity through RIN analysis. Rice roots were digested at various temperatures (i) and times (k). RNAs were extracted and subjected to RIN analysis. Please note that RNA integrity decreased with increased time and temperature. Data are the mean ± s.d. from three independent experiments. Different letters denote statistically significant variation (two-sided Student’s t-test, p < 0.05, with no adjustment for multiple comparisons). The p-values are provided in Source Data.

Extended Data Fig. 2 |. Supporting data for the generation of the maize anther cell atlas by the fixation and digestion at high temperatures protocol and CEL-seq2.

a, UMAP clustering of 307 cells from 2.0 mm maize anthers. Six distinct clusters are shown in different colors. Maize anthers were fixed and digested with RNAse-depleted enzymes at 50 °C for 90 min. The resultant cells were sorted and isolated using either BioSorter (Union Biometrica) or Hana (Namocell) machines into 96-well plates, and scRNA-seq libraries were prepared using a modified CEL-Seq2 library preparation protocol. b, Comparison of different cell isolation method (Biosorter vs. Hana). Each dot represents a single cell. c–e, Correlation values of each cell with LCM tapetal (c), meiocyte (d), and other somatic cell types (middle layer, endothecium, and epidermis) (e) data. f, Percentage of total UMIs originating from the plastid for each cell. Please note that Murphy et al. uncovered that the endothecium contains chloroplasts unlike the other anther cell layers⁷⁴. g–j, UMAP plots showing expression patterns of tapetal (g), meiocyte (h), putative endothecium (i), and putative epidermis (j) marker genes. The corresponding cell clusters are labeled by dashed circles.

Extended Data Fig. 3 |. Supporting data for the annotation of the integrated rice root tip cell atlas.

a, Expression patterns of published cell type marker genes in the integrated rice root tip cell atlas generated by FX-Cell, FXcryo-Cell, and cryoFX-Cell (Fig. 3g; Supplementary Table 1). The diameter of the points represents the proportion of cells expressing a specific gene within a given cluster, while the color of the points indicates the relative expression level of the gene. Ep, epidermis; Rh, root hairs; Rc, root cap; Ex, exodermis; Vc, vascular cylinder; Xy, xylem; Ph, phloem; Pe, pericycle; En, endodermis; Me, meristem-like cells; Co, cortex. b, Cluster similarity analysis. The rice root tip cell clusters generated by traditional scRNA-seq and FX-Cell were compared. The overlap ratios of marker genes (left) and GO signaling pathways (right) of each cluster are given. An overlap coefficient greater than or equal to 0.4 (dashed line) is considered indicative of high similarity. c, Representative images showing the harvested rice root tips (white dashed boxes) for FX-Cell (Fig. 2e). The rice seeds were geminated and cultured on plates. Please note that, compared to the sample used by Zhang et al.³⁷, the sample used for FX-Cell developed fewer root hairs (magnified images on the right side) due to the increased water availability in the plates. Scale bars, 0.5 cm. Three independent experiments were performed.

Extended Data Fig. 4 |. Supporting data for the generation of a high-quality Arabidopsis root tip cell atlas by FX-Cell.

a, b, The Arabidopsis root tip cell atlas generated by FX-Cell. The resulting atlas was integrated with two published datasets from Zhang et al.⁴² (a) and Shahan et al.⁴¹ (b). The grayscale UMAP in (b) illustrate the coverage of the root cell atlas generated by FX-Cell. c, Expression of known Arabidopsis root cell-type markers (Fig. 1E in Shahan et al., 2022)⁴¹ in the integrated root tip cell atlas (b). The diameter of the points represents the proportion of cells expressing a specific gene within a given cluster, while the color of the points indicates the relative expression level of the gene. Col, Columella; QC, quiescent center; RC, root cap; LRC, lateral root cap; Xy, xylem; St, stele; Pr, protophloem; En, endodermis; Cor, cortex; GT, ground tissue; At, atrichoblast; Tr, trichoblast. d, UMAP plots showing the expression of known Arabidopsis root cell-type markers (Fig. 1E in Shahan et al., 2022)⁴¹ in the integrated root tip cell atlases (a, b). e, f, Upset plots showing shared and unique protoplasting genes in each root cell type in A. thaliana. The number of upregulated (e) or downregulated (f) genes for each cell type is presented on the top of the colored bars. Total number of protoplasting genes in each cell type is presented on the right. g, UMAP plots showing the original A. thaliana root atlas (left) and the atlases with depleted protoplasting genes (middle and right). The protoplasting genes were sourced from comparative bulk RNA-seq (intact tissue vs. protoplasts, middle) or the FX-Cell dataset (original atlas vs. FX-Cell atlas, right). h, Venn plots showing the overlap of cluster marker genes among the original A. thaliana root atlas and the atlases with depleted protoplasting genes shown in (g). Please note that cluster marker genes among these three atlases are largely overlapped.

Extended Data Fig. 5 |. Supporting data for the comparison of rice root tip cell atlases generated by different scRNA-seq methods.

a, b, Images showing the cells derived from the leaves of A. thaliana (a) and shoot apices of S. martensii (b) with three different methods. “fix.” indicates the samples that were fixed and subsequently subjected to high-temperature enzymatic digestion; “fix. + cryo.” refers to samples that were fixed, frozen, cryopreserved, then thawed and subjected to high-temperature enzymatic digestion; “cryo. + fix.” indicates the samples that were directly frozen and cryopreserved, then thawed, fixed, and subjected to high-temperature enzymatic digestion. Scale bars, 50 μm. c, Assessment of RNA integrity through RIN analysis. Samples were fixed and cryopreserved as described in (a, b). Data are the mean ± s.d. from three independent biological replicates. Different letters denote statistically significant variation (two-sided Student’s t-test, p < 0.05, with no adjustment for multiple comparisons). The p-values are provided in Source Data. d, Heatmap showing the similarity among the cell clusters derived different scRNA-seq methods. The annotated cell types are given on the right. e, UMAP plots showing the expression patterns of other cell type marker genes of rice root tip cell atlases generated by FX-Cell, FXcryo-Cell, or cryoFX-Cell (Figs. 2e and 3d, e). f, Quantification of released cells and protoplasts from rice tiller nodes and wild rice rhizome nodes. Tissues were enzymatically digested using traditional protoplasting method or FX-Cell. Data represent the mean ± s.d. from three independent experiments. Different letters denote statistically significant variation (two-sided Student’s t-test, p < 0.05). The p-values are provided in Source Data.

Extended Data Fig. 6 |. Supporting data for the cell atlases of difficult-to-digest and field-grown samples.

a, Expression patterns of cell type marker genes in the rice tiller node cell atlas generated by FXcryo-Cell (Fig. 4a; Supplementary Table 1). The diameter of the dots represents the proportion of cells expressing a specific gene within a given cluster, while the color of the dots indicates the relative expression level of the gene. Dc, dividing cells; Me, meristem-like; Xy, xylem; Xy & Ca, xylem/cambium; Ca, cambium; Ph, phloem; Pe & Pa, pericycle & parenchyma; Ep, epidermis; En & Co, endodermis & cortex; Ex, exodermis; Rc, root cap. b, Expression patterns of cell type marker genes in the wild rice rhizome node cell atlas generated by FXcryo-Cell (Fig. 4b; Supplementary Table 1). The diameter of the dots represents the proportion of cells expressing a specific gene within a given cluster, while the color of the dots indicates the relative expression level of the gene. Dc, dividing cells; Xy, xylem; Te, tracheary elements; Ph & Cc, phloem & companion cells; Co, cortex; En & Ex & Pe, endodermis, exodermis, and pericycle; Ca, cambium; LRP, lateral root primordia; QC, root quiescent center; SMC, shoot meristem like. c, Heatmap showing the top 150 genes from each cluster in the maize crown root cell atlas generated by cryoFX-Cell (Fig. 4d). Cell types are annotated using published maize root marker genes (Supplementary Table 1). The representative genes in each cell type are given on the right. Normalized expression levels are shown. Co, cortex; Ci, cortex initials; En, endodermis; Dc, dividing cells; Pe, pericycle; Ph & Si, phloem & sieve; QC, root quiescent center; Rc, root cap; St, stele (E); St, stele (M); Xy, xylem. d, Joint analysis of the maize root cell atlas generated by cryoFX-Cell (Fig. 4d) and traditional scRNA-seq⁵³. Harmony algorithm was applied to mitigate batch effects. e, Annotation of the integrated maize root tip cell atlas shown in (b), based on the published cell type marker gene (Supplementary Table 1). Different colors represent different cell types.

Extended Data Fig. 7 |. Supporting data for the validation of cell type annotations.

Sequential fluorescence in situ hybridization (seq-FISH) images showing expression patterns of selected cell cluster-specific genes in rice tiller nodes (a) and wild rice rhizome nodes (b). The images display transverse sections of the tiller nodes, rhizome nodes, and roots. The expression of each gene is represented on the left by UMAP and on the right by seq-FISH experiments. The gene name and the corresponding cell cluster are indicated at the top. Each gene’s transcripts are shown in different colors, while nuclei are stained with DAPI (4’,6-diamidino-2-phenylindole, grey). Scale bars, 50 μm. Two independent experiments were performed.

Extended Data Fig. 8 |. Supporting data for annotation of A. thaliana leaf cell atlas and analysis of wound response genes.

a, Expression patterns of representative cell type marker genes in each cell cluster of the integrated Arabidopsis leaf cell atlas in Fig. 5b (Supplementary Table 1). The diameter of the dots indicates the proportion of cells expressing a specific gene within a given cluster, while the color of the dots represents the relative expression levels of the genes. BSC, bundle sheath cells; Dc, dividing cells; Ep, epidermis; Gc, guard cells; Hy: hydathodes; Mi, myrosin idioblasts; Ph, phloem; Cc, companion cells; Xy, xylem; Mp, mesophyll. b, UMAP plots showing the expression patterns of representative wound-induced genes⁶⁰ in the integrated cell atlas in Fig. 5b. c, Upset plot presenting shared and unique wound-repressed genes in each cell type. The number of genes for each cell type is presented on the top of the colored bars. Total number of wound-induced genes in each cell type is presented on the right. d, GO term analysis of unique wound-repressed genes in each cell type. Enrichment was tested using a one-sided hypergeometric (Fisher’s exact) test as implemented in the “enrichGO” function of the R package clusterProfiler. The p-values were adjusted for multiple comparisons using the BenjaminiHochberg method. e, UMAP plots presenting representative cell-type specific wound-repressed genes. Only cells corresponding to each specific cell type are presented. The upper and lower rows indicate intact and wounded A. thaliana leaf cell atlases produced using FXcryo-Cell (Fig. 5c), respectively.

Extended Data Fig. 9 |. Supporting data for comparison of FXcryo-Cell and snRNA-seq.

a, Comparative analysis of the atlases generated by traditional protoplasting-based scRNA-seq (red) snRNA-seq (grey), and FX-Cell (green). b, Annotation of integrated cell atlases shown in (a). c, UMAP plots displaying expression levels of wound-response genes across the three datasets from (a). Wound-response genes extracted from GO terms (GO: 0009611; Supplementary Table 2) were used as the gene set and analyzed using AUCell.

Extended Data Fig. 10 |. Supporting data for the genome assembly of O. longistaminata.

The genomic differences between two genome assemblies, Oryza sativa and Oryza longistaminata, are illustrated by plot using plotsv.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41592-025-02900-2.

Data availability

The scRNA-seq, snRNA-seq and RNA-seq data (BioProject PRJCA035988) were deposited in Beijing Institute of Genomics Data Center (http://bigd.big.ac.cn)^86,87. The chromosome-level genome assembly and annotation of O. longistaminata have been deposited in the figshare database via https://doi.org/10.6084/m9.figshare.28457807.v1 (ref. 88). Source data are provided with this paper.

Code availability

The code used for the data processing and analysis mentioned in Methods is available on GitHub via https://github.com/WangLab-CEMPS/Xin_Ming_FX_Cell/.