Methods to elucidate the genetic actors of seed development and germination

In A. thaliana, the distinct stages and tissues of seed development and germination are sequentially regulated by complex genetic networks (reviewed in Carrera-Castaño et al. (Reference Carrera-Castaño, Calleja-Cabrera, Pernas, Gómez and Oñate-Sánchez2020), Baud et al. (Reference Baud, Corso, Debeaujon, Dubreucq, Job, Marion-Poll, Miquel, North, Rajjou and Lepiniec2022), and Verma et al. (Reference Verma, Attuluri and Robert2022)). In particular, WUSCHEL-RELATED HOMEOBOX (WOX) transcription factors are key regulators of early embryo morphogenesis controlling body axis establishment (Haecker et al., Reference Haecker, Groß-Hardt, Geiges, Sarkar, Breuninger, Herrmann and Laux2004). Later, in seed development, LAFL (LEAFY COTYLEDON1, ABSCISIC ACID INSENSITIVE3, FUSCA3, and LEAFY COTYLEDON2) transcription factors trigger seed maturation, while the MBW (MYB, bHLH and WDR) protein complex is crucial for the biosynthesis and accumulation of flavonoid compounds in the seed coat (reviewed in Lepiniec et al. (Reference Lepiniec, Devic, Roscoe, Bouyer, Zhou, Boulard, Baud and Dubreucq2018), Baud et al. (Reference Baud, Corso, Debeaujon, Dubreucq, Job, Marion-Poll, Miquel, North, Rajjou and Lepiniec2022), and Zumajo-Cardona et al. (Reference Zumajo-Cardona, Gabrieli, Anire, Albertini, Ezquer and Colombo2023)). ABA-related genes are known to be central regulators of desiccation tolerance and primary dormancy (reviewed in Nambara and Marion-Poll (Reference Nambara and Marion-Poll2003) and Chauffour et al. (Reference Chauffour, Bailly, Perreau, Cueff, Suzuki, Collet, Frey, Clément, Soubigou-Taconnat, Balliau, Krieger-Liszkay, Rajjou and Marion-Poll2019)). Additionally, DELAY OF GERMINATION 1 (DOG1) has been identified as a key regulator of seed dormancy acting through both ABA-dependent and ABA-independent pathways (Bentsink et al., Reference Bentsink, Jowett, Hanhart and Koornneef2006; Née et al., Reference Née, Kramer, Nakabayashi, Yuan, Xiang, Miatton, Finkemeier and Soppe2017). Lastly, germination is controlled by the GA pathway, and in particular, DELLA proteins that act as negative regulators of seed germination (Tyler et al., Reference Tyler, Thomas, Hu, Dill, Alonso, Ecker and Sun2004). These genetic regulators can either be expressed in a specific or in several seed cell types (Belmonte et al., Reference Belmonte, Kirkbride, Stone, Pelletier, Bui, Yeung, Hashimoto, Fei, Harada, Munoz, Le, Drews, Brady, Goldberg and Harada2013; Peirats-Llobet et al., Reference Peirats-Llobet, Yi, Liew, Berkowitz, Narsai, Lewsey and Whelan2023). Hence, it is crucial to study the spatial and temporal expression dynamics of these actors to better understand the regulation of the processes of the successive stages of seed development and germination. This is a challenging undertaking as seeds are small and progressively accumulate protective layers, such as flavonoids in the seed coat. Methods, such as RNA in situ hybridization (ISH) or promoter–reporter fusion assays, have been used to study the spatiotemporal activity of seed gene networks (Haecker et al., Reference Haecker, Groß-Hardt, Geiges, Sarkar, Breuninger, Herrmann and Laux2004; Francoz et al., Reference Francoz, Ranocha, Pernot, Le Ru, Pacquit, Dunand and Burlat2016; Palovaara et al., Reference Palovaara, Saiga, Wendrich, van‘t Wout Hofland, van Schayck, Hater, Mutte, Sjollema, Boekschoten, Hooiveld and Weijers2017). Both these methods allow localization of the expression of a gene of interest; by using a labelled complementary RNA probe for ISH or by cloning the promoter of this gene in front of a reporter gene, usually β-glucuronidase (GUS) or green fluorescent protein (GFP) gene, for promoter–reporter fusion assays. However, these methods only offer low throughput and consequently only permit the study of a limited number of genes. In addition, these methods require a prior knowledge of studied genes and do not allow novel gene discovery.

More recently, comprehensive techniques, such as bulk RNA-seq, have also been performed on seeds (Le et al., Reference Le, Cheng, Bui, Wagmaister, Henry, Pelletier, Kwong, Belmonte, Kirkbride, Horvath, Drews, Fischer, Okamuro, Harada and Goldberg2010; Belmonte et al., Reference Belmonte, Kirkbride, Stone, Pelletier, Bui, Yeung, Hashimoto, Fei, Harada, Munoz, Le, Drews, Brady, Goldberg and Harada2013; Dekkers et al., Reference Dekkers, Pearce, van Bolderen-Veldkamp, Marshall, Widera, Gilbert, Drost, Bassel, Müller, King, Wood, Grosse, Quint, Krasnogor, Leubner-Metzger, Holdsworth and Bentsink2013; Khan et al., Reference Khan, Millar, Girard and Belmonte2014; Narsai et al., Reference Narsai, Gouil, Secco, Srivastava, Karpievitch, Liew, Lister, Lewsey and Whelan2017; Hofmann et al., Reference Hofmann, Schon and Nodine2019). Bulk RNA-seq allows the study of global gene expression in the whole seed or in a seed tissue at different stages. Related to this, individual seed tissues can be isolated either manually or by performing laser-capture microdissection which uses laser cutting to extract areas of interest from a biological sample (Dekkers et al., Reference Dekkers, Pearce, van Bolderen-Veldkamp, Marshall, Widera, Gilbert, Drost, Bassel, Müller, King, Wood, Grosse, Quint, Krasnogor, Leubner-Metzger, Holdsworth and Bentsink2013; Betts et al., Reference Betts, Berkowitz, Liu, Collins, Skadhauge, Dockter, Burton, Whelan and Fincher2017; Liew et al., Reference Liew, Wang, Peirats-Llobet, Berkowitz, Whelan and Lewsey2020). However, bulk RNA-seq does not allow the capture of cell type heterogeneity and transcript signatures within the studied sample. Hence, this technique provides average transcriptome information across a number of cells, and its spatial resolution is limited by the precision of dissection. Additionally, the isolation of seed tissues by laser-capture microdissection is a challenging and time-consuming process that does not allow high throughput.

In the last few years, single-cell transcriptomics methods (scRNA-seq and snRNA-seq) have been developed and applied to several plant organs, including A. thaliana roots (Denyer et al., Reference Denyer, Ma, Klesen, Scacchi, Nieselt and Timmermans2019; Shulse et al., Reference Shulse, Cole, Ciobanu, Lin, Yoshinaga, Gouran, Turco, Zhu, O'Malley, Brady and Dickel2019) and leaves (Kim et al., Reference Kim, Symeonidi, Pang, Denyer, Weidauer, Bezrutczyk, Miras, Zöllner, Hartwig, Wudick, Lercher, Chen, Timmermans and Frommer2021; Lopez-Anido et al., Reference Lopez-Anido, Vatén, Smoot, Sharma, Guo, Gong, Anleu Gil, Weimer and Bergmann2021), as well as maize and sorghum roots (Guillotin et al., Reference Guillotin, Rahni, Passalacqua, Mohammed, Xu, Raju, Ramírez, Jackson, Groen, Gillis and Birnbaum2023). Single-cell technologies allow us to study the transcriptomic profile of isolated cells and its dynamics during plant development or in response to environment changes.

Several recent studies have demonstrated how scRNA-seq and snRNA-seq can be applied to A. thaliana seeds (Kao et al., Reference Kao, Schon, Mosiolek, Enugutti and Nodine2021; Picard et al., Reference Picard, Povilus, Williams and Gehring2021; Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023; Liew et al., Reference Liew, You, Auroux, Oliva, Peirats-Llobet, Ng, Tamiru-Oli, Berkowitz, Hong, Haslem, Stuart, Ritchie, Bassel, Lister, Whelan, Gouil and Lewsey2024). Single-cell transcriptomics is a very promising approach for analysis of the many cell types and cell states in A. thaliana seeds, with the potential to be applied at different stages of seed development and germination. For example, snRNA-seq has successfully been performed on A. thaliana embryos at the globular stage (Kao et al., Reference Kao, Schon, Mosiolek, Enugutti and Nodine2021). This was done by isolating embryo-specific nuclei using fluorescent activated cell sorting (FACS) in a transgenic line expressing nuclear-localized GFP under the control of the embryo-specific WOX2 promoter. This enabled analysis of spatial genetic regulation during early A. thaliana embryogenesis. At this stage, the different embryonic cell types specialize and establish a basic body plan. Single-nucleus transcriptomics allowed gene expression programmes and transcription factor signatures to be associated with these differentiating cell types. Interestingly, differences in the expression of epigenetic regulators were observed across embryonic cell types. For instance, genes involved in 24-nucleotide small interfering RNAs (siRNAs) biogenesis were preferentially expressed in the suspensor which connects maternal tissues with the embryo, while transcripts involved in siRNA binding and siRNA-dependent gene silencing were enriched in the embryo, and more specifically in shoot meristem initials (Gutzat et al., Reference Gutzat, Rembart, Nussbaumer, Hofmann, Pisupati, Bradamante, Daubel, Gaidora, Lettner, Donà, Nordborg, Nodine and Mittelsten Scheid2020). This suggests a signalling coordination between seed cell types, with a different site for siRNA production (the suspensor) and response (shoot meristem initials). It has been previously proposed that this mechanism might be involved in the repression of transposon mobility, thus limiting their mutagenic potential in the shoot meristem (reviewed in Mosher and Melnyk (Reference Mosher and Melnyk2010)). Hence, this study illustrates how single-cell technology can be used to investigate transcriptional programmes as well as cell communications, and, in particular, signalling pathways involving several cell types, in seeds.

Similarly, a single-nucleus study was carried out on seeds 4 d after pollination and further selected endosperm-specific nuclei based on maternal read percentage (Picard et al., Reference Picard, Povilus, Williams and Gehring2021). Two different A. thaliana accessions (Col-0 and Cvi) were crossed to study imprinted genes in the endosperm cell subtypes. This revealed the expression patterns of imprinted genes across endosperm cell types, including, notably, that paternally imprinted genes (PEGs) were predominantly expressed in the chalazal endosperm, while maternally imprinted genes (MEGs) were expressed across all endosperm cell types. This study also better characterized transcriptomic profiles in specific endosperm cell types. In particular, genes involved in DNA replication were preferentially expressed in the chalazal endosperm, while no mitotic markers were found, suggesting that this region undergoes endoreduplication. This is consistent with previous observations of large and polyploid nuclei in this compartment (Baroux et al., Reference Baroux, Fransz and Grossniklaus2004). Additionally, different transcriptomic profiles were observed within the chalazal endosperm nuclei. The authors further demonstrated that these variations are associated with a subdifferentiation within the chalazal endosperm, which consists of two regions: the chalazal nodules, which delimits the chalazal endosperm, and the chalazal cyst, which interacts with the seed coat. By applying single-cell technology, this study provided the first transcriptomic profiles of specifically of these specific regions (Nguyen et al., Reference Nguyen, Brown and Lemmon2000).

scRNA-seq was also applied to germinating A. thaliana embryos, enabling the discovery that most embryonic cells go through a common transcriptional state at the beginning of germination (Liew et al., Reference Liew, You, Auroux, Oliva, Peirats-Llobet, Ng, Tamiru-Oli, Berkowitz, Hong, Haslem, Stuart, Ritchie, Bassel, Lister, Whelan, Gouil and Lewsey2024). This study enabled the observation that suites of cell-type-specific transcription factors were involved in subsequent transcriptional remodelling, contributing to the establishment of cell-type-specific patterns of gene expression. A subset of these transcription factors was also validated as an influencing germination rate. In addition, the study supported an earlier investigation, suggesting that the provasculature might play a key role in determining the break of seed dormancy (Topham et al., Reference Topham, Taylor, Yan, Nambara, Johnston and Bassel2017).

Recently, a large-scale snRNA-seq study was carried out on A. thaliana organs and tissues at different development stages, including siliques that contain developing seeds and two time points during germination (Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023). Seed cell types, and in particular embryonic ones, were identified in snRNA-seq carried out on siliques. This latter study illustrates the opportunity to study seed development by performing a single-cell analysis either on whole siliques, and to further select seed-specific cells, or on isolated developing seeds. Interestingly, siliques were found to be the organ with the greatest complexity among all A. thaliana organs and tissues analysed in this study, indicating the existence of multiple seed and silique cell types and cell states. Furthermore, snRNA-seq was conducted to characterize cell profiles of dissected maize endosperm, showing that this technique can also be transferred to seeds of other plant species including crops (Yuan et al., Reference Yuan, Huo, Zhang, Wang, Wang, Chang, Cai, Song, Galbraith, Zhang, Huang, Song and Ma2024).

The establishment of single-cell transcriptomics methods for plant tissues has been accompanied by the development of spatial transcriptomics methods, which allow in situ transcriptome profiling of sectioned samples (Xia et al., Reference Xia, Sun, Li, Li, Zhao, Chen, Qin, Chen, Chen, Liu, Yin, Mu, Wang, Xu, Li, Yuan, Qiao, Hao, Wang, Xie, Xu, Liu, Li, Chen, Liu, Yin, Yang, Wang, Gu and Xu2022). This technology is short-read sequencing based and offers comprehensive transcriptome analysis. The resolution of spatial transcriptomics depends on the technology used (e.g. 10 μm for Slide-seqV2; 55 μm for 10× Genomics Visium), which is sufficient to profile transcripts in individual cells or group of cells (Ståhl et al., Reference Ståhl, Salmén, Vickovic, Lundmark, Navarro, Magnusson, Giacomello, Asp, Westholm, Huss, Mollbrink, Linnarsson, Codeluppi, Borg, Pontén, Costea, Sahlén, Mulder, Bergmann, Lundeberg and Frisén2016; Stickels et al., Reference Stickels, Murray, Kumar, Li, Marshall, Di Bella, Arlotta, Macosko and Chen2021). Resolution is constantly improving, so in future, spatial transcriptomics might be used to profile the transcriptome at subcellular levels (Oliveira et al., Reference Oliveira, Romero, Chung, Williams, Gottscho, Gupta, Pilipauskas, Mohabbat, Raman, Sukovich, Patterson, Visium and Taylor2024). The advantage of spatial transcriptomics is that the technologies offer direct spatial information as there is no tissue disruption such as for single-cell studies.

Spatial transcriptomics has recently been performed on germinating barley seeds at six time points, elucidating the spatiotemporal regulation of over 14,000 genes expressed in the first 24 h after imbibition (Peirats-Llobet et al., Reference Peirats-Llobet, Yi, Liew, Berkowitz, Narsai, Lewsey and Whelan2023). This strategy has also been applied to developing wheat grains, as well as to nectarine and peach fruit including their seed (Zhao et al., Reference Zhao, Cao, Qin, Liu, Guan, Sun, Liu, Zhou, Yang, Liu, Hu, Ninkuu, Sun and Wang2023; Li et al., Reference Li, Wan, Wang, Li, Wu, Chen, Han and Chen2024). However, the resolution of spatial transcriptomics is still low compared with single-cell techniques, which is a major limitation when studying small tissues such as A. thaliana seeds (about 500 μm). Alternatively, imaging-based spatial transcriptomics methods have also been adapted for use on plants, such as Multiplexed Error-Robust Fluorescence in situ Hybridization (MERFISH; Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023; Nobori et al., Reference Nobori, Monell, Lee, Zhou, Nery and Ecker2023). These techniques allow an in situ profiling of hundreds of selected target genes at a subcellular resolution using fluorescent probes but require prior knowledge of these target genes. Overall, single-cell and spatial transcriptomics are two complementary approaches to perform transcriptomic profiling of cell types and states in a tissue of interest. These technologies can be integrated to provide a comprehensive spatial description with a precise resolution (Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023).

The above-mentioned studies on seed development and germination show the potential of single-cell transcriptomics to enhance our understanding of seed biology. Single-cell technology enables the establishment of gene expression atlas of seed tissues and stages that can be further utilized to investigate transcriptional programmes. These first results also illustrate how scRNA-seq can both precisely define very specific cell types (such as the chalazal endosperm; Picard et al., Reference Picard, Povilus, Williams and Gehring2021) and unveil common transcriptional states (during germination; Liew et al., Reference Liew, You, Auroux, Oliva, Peirats-Llobet, Ng, Tamiru-Oli, Berkowitz, Hong, Haslem, Stuart, Ritchie, Bassel, Lister, Whelan, Gouil and Lewsey2024). However, at this time, no single-cell transcriptomic study has been carried out to comprehensively cover the whole seed during the different stages of its development. Further single-cell studies will provide more insights into the spatiotemporal regulations of gene expression in A. thaliana seeds, as well as in other plant species, and the diversity and interaction among seed cell types. Additionally, complementing single cell with spatial transcriptomics will refine the spatial dimension in these analyses, and increase our comprehension of seed structure. These will certainly be facilitated by the development and the decreasing costs of single cell and spatial technologies in the coming years, enabling an increase in the number of studies in seed research. In the next section, we propose a workflow for scRNA-seq to facilitate future single-cell analyses of seeds.

Workflow for a single-cell transcriptomics study

The workflow of single-cell studies starts with sample preparation during which cells (scRNA-seq), or nuclei (snRNA-seq) are extracted from the tissue of interest (Fig. 2; reviewed in Shaw et al. (Reference Shaw, Tian and Xu2021)). In A. thaliana, protoplast extraction has commonly been used to isolate cells by removing cell walls, but nucleus extraction is increasing in popularity (Yoo et al., Reference Yoo, Cho and Sheen2007; Wang et al., Reference Wang, Zhao, Xiang, Duan, Chen, Guo and Sahu2023). Extraction of protoplasts or nuclei is followed by their separation using microfluidic technologies on droplet-based platforms such as 10x Genomics or DropSeq (Macosko et al., Reference Macosko, Basu, Satija, Nemesh, Shekhar, Goldman, Tirosh, Bialas, Kamitaki, Martersteck, Trombetta, Weitz, Sanes, Shalek, Regev and McCarroll2015; Zheng et al., Reference Zheng, Terry, Belgrader, Ryvkin, Bent, Wilson, Ziraldo, Wheeler, McDermott, Zhu, Gregory, Shuga, Montesclaros, Underwood, Masquelier, Nishimura, Schnall-Levin, Wyatt, Hindson, Bharadwaj, Wong, Ness, Beppu, Deeg, McFarland, Loeb, Valente, Ericson, Stevens, Radich, Mikkelsen, Hindson and Bielas2017). Individual protoplasts or nuclei are encapsulated within droplets, and the transcripts within these droplets are then utilized for library preparation. This process involves reverse transcription, labelling with unique molecular identifiers and cell-specific barcodes, and subsequent amplification. Barcoding is a crucial step that enables the association of reads with their corresponding transcripts and cells of origin in subsequent analyses. The library is then sequenced, and raw reads generated for each cell are processed in a bioinformatic pipeline.

Figure 2. Workflow for single-cell transcriptomics. Cells or nuclei are first extracted from a tissue of interest and isolated by droplet encapsulation. This is coupled with cell- or nucleus-specific tagging of the transcripts with the use of barcoded beads. From these tagged transcripts, a library is prepared and sequenced. In the last step, cells are clustered based on their transcriptomic profile and this clustering serves as a basis for subsequent data analyses.

The bioinformatic pipeline starts with read mapping to the relevant species’ genome and data preprocessing. Cell quality control is performed during preprocessing. Low-quality cells are removed, which include damaged cells as well as droplets containing multiple cells or no cells (empty droplets). Further processing steps are then usually performed with the R toolkit Seurat (Hao et al., Reference Hao, Stuart, Kowalski, Choudhary, Hoffman, Hartman, Srivastava, Molla, Madad, Fernandez-Granda and Satija2024). Seurat has been specifically developed for single-cell transcriptomics processing and analyses. This toolkit takes as input a raw count matrix in which the numbers of transcripts detected in each cell are stored. Several processing steps, such as data normalization and scaling, are usually performed on the raw counts. Then, Seurat can be used to cluster cells using their transcriptomic signatures, based on dimensionality reduction using methods such as principal component analysis (PCA) or uniform manifold approximation and projection (UMAP; Jolliffe, Reference Jolliffe and Lovric2011; McInnes et al., Reference McInnes, Healy and Melville2020). Obtained clusters are next annotated with the help of known marker genes expressed in specific cell types or states. These annotated clusters can be used for subsequent analyses. It is for instance possible with Seurat to extract information on specific genes or cell types. Other tools can also be used to infer cell-type-specific gene co-expressions and regulatory networks from single-cell transcriptomics (Bravo González-Blas et al., Reference Bravo González-Blas, De Winter, Hulselmans, Hecker, Matetovici, Christiaens, Poovathingal, Wouters, Aibar and Aerts2023; Su et al., Reference Su, Xu, Shan, Cai, Zhao and Zhang2023). Alternatively, a pseudotime trajectory analysis can be carried out on single-cell data with tools such as Monocle 3.0 (Cao et al., Reference Cao, Spielmann, Qiu, Huang, Ibrahim, Hill, Zhang, Mundlos, Christiansen, Steemers, Trapnell and Shendure2019). Pseudotime studies arrange cells based on their transcriptional states and consequently infer a developmental progression. Additional studies can also be performed on single-cell data such as gene ontology enrichment analysis or transcription factor predictions (Lin et al., Reference Lin, Ding and Bar-Joseph2020; Liew et al., Reference Liew, You, Auroux, Oliva, Peirats-Llobet, Ng, Tamiru-Oli, Berkowitz, Hong, Haslem, Stuart, Ritchie, Bassel, Lister, Whelan, Gouil and Lewsey2024).

Challenges for single-cell transcriptomics in seeds

One of the crucial steps of single-cell transcriptomics is sample preparation, which requires careful method design. In particular, a choice between protoplast (scRNA-seq) and nucleus (snRNA-seq) isolation has to be made beforehand. scRNA-seq and snRNA-seq produce comparable results, with each method having its own strengths and drawbacks (Guillotin et al., Reference Guillotin, Rahni, Passalacqua, Mohammed, Xu, Raju, Ramírez, Jackson, Groen, Gillis and Birnbaum2023; Wang et al., Reference Wang, Zhao, Xiang, Duan, Chen, Guo and Sahu2023). The primary drawback with scRNA-seq is that it presents a bias in capturing some cell types, particularly because of the diverse sizes of plant cells and the variable efficiency of getting protoplasts from different cell types. Additionally, protoplast extraction normally takes a longer time than nuclei extraction and utilizes enzymatic digestion which can trigger changes in gene expression due to stress responses (Birnbaum et al., Reference Birnbaum, Shasha, Wang, Jung, Lambert, Galbraith and Benfey2003). On the other hand, snRNA-seq only captures nuclear mRNAs, leading to lower transcript detection due to the inability to capture cytoplasmic mRNAs. These drawbacks should be taken into consideration when designing a single-cell workflow.

So far, in seeds, snRNA-seq has proven its efficiency to profile all cell types of seed development and germination in A. thaliana (Kao et al., Reference Kao, Schon, Mosiolek, Enugutti and Nodine2021; Picard et al., Reference Picard, Povilus, Williams and Gehring2021; Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023). Additionally, snRNA-seq seems more appropriate for early seed development to capture isolated nuclei from the syncytial endosperm (Picard et al., Reference Picard, Povilus, Williams and Gehring2021). However, scRNA-seq could be more relevant for late seed development and germination, as transcripts are progressively stored in cell cytoplasm until germination (Nakabayashi et al., Reference Nakabayashi, Okamoto, Koshiba, Kamiya and Nambara2005). It has furthermore been demonstrated that transcription is not essential for radicle elongation during germination, emphasizing the significant role of stored mRNAs (Rajjou et al., Reference Rajjou, Gallardo, Debeaujon, Vandekerckhove, Job and Job2004). The selective translation of a subset of stored mRNAs is a critical regulatory mechanism that allows seeds to rapidly germinate, independent of de novo transcription (Galland et al., Reference Galland, Huguet, Arc, Cueff, Job and Rajjou2014; Bai et al., Reference Bai, van der Horst, Cordewener, America, Hanson and Bentsink2020; reviewed in Sano et al. (Reference Sano, Rajjou and North2020)). The specific targeting of certain mRNAs for translation appears to involve both sequence and structural features. However, there is currently a gap in research regarding the spatial regulation of this translational selectivity at the cellular level within seeds. Protoplast isolation from seed tissues can nonetheless be challenging because of seed coat protective layers hindering a homogenous digestion. So far, scRNA-seq has only been used to profile cells of the germinating embryo, and the ability to extract protoplasts from the seed coat or endosperm is still to be demonstrated. Therefore, if the protoplast method is used, a protocol must be optimized that efficiently captures all the different seed cell types. By comparison, spatial transcriptomics has been efficiently utilized to capture all seed development and germination cell types of various plant species (Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023; Peirats-Llobet et al., Reference Peirats-Llobet, Yi, Liew, Berkowitz, Narsai, Lewsey and Whelan2023; Li et al., Reference Li, Wan, Wang, Li, Wu, Chen, Han and Chen2024). However, it is worth noting that this technique is still very costly and is usually applied to a limited number of sections, meaning that the data represent only a portion of a seed.

Another crucial challenge of single-cell transcriptomics is the cell clustering step and correct annotation of obtained clusters, which is essential to perform downstream analyses. The annotation relies on previously selected marker genes, which usually come from existing publications where specific cells or tissues have been analysed, including prior single-cell studies. Liew et al. (Reference Liew, You, Auroux, Oliva, Peirats-Llobet, Ng, Tamiru-Oli, Berkowitz, Hong, Haslem, Stuart, Ritchie, Bassel, Lister, Whelan, Gouil and Lewsey2024) curated such a list of embryonic marker genes for seed germination. However, as no single-cell study has been previously carried out across the entirety of seed development, there is no list of marker genes demonstrated to be spatially associated with the different cell types and stages of A. thaliana seed development. Thus, we have compiled a list of seed development marker genes to facilitate cluster annotation of future single-cell studies (Fig. 1B; Supplementary Table S1). To do so, genes demonstrated to be specifically associated with seed cell types and stages were selected. Data arise from RNA ISH or promoter–reporter fusion assays, as well as previous single-cell studies of seed tissues and germination (Francoz et al., Reference Francoz, Ranocha, Pernot, Le Ru, Pacquit, Dunand and Burlat2016; Kao et al., Reference Kao, Schon, Mosiolek, Enugutti and Nodine2021; Picard et al., Reference Picard, Povilus, Williams and Gehring2021; Liew et al., Reference Liew, You, Auroux, Oliva, Peirats-Llobet, Ng, Tamiru-Oli, Berkowitz, Hong, Haslem, Stuart, Ritchie, Bassel, Lister, Whelan, Gouil and Lewsey2024). We obtained a list of 369 genes associated with one or a limited number of seed cell types and with various developmental stages (Supplementary Table S1). Seed markers obtained from bulk RNA-seq were not included as they are likely less precise, but these could potentially be used to cross-validate cluster identity (e.g. datasets from Le et al. (Reference Le, Cheng, Bui, Wagmaister, Henry, Pelletier, Kwong, Belmonte, Kirkbride, Horvath, Drews, Fischer, Okamuro, Harada and Goldberg2010) and Belmonte et al. (Reference Belmonte, Kirkbride, Stone, Pelletier, Bui, Yeung, Hashimoto, Fei, Harada, Munoz, Le, Drews, Brady, Goldberg and Harada2013)). The list of seed markers might also be complemented by the addition of marker transcripts obtained in other single-cell studies that precisely annotate specific cell types. For instance, to better characterize developing vascular tissues in seed embryos, vasculature markers curated by Kim et al. (Reference Kim, Symeonidi, Pang, Denyer, Weidauer, Bezrutczyk, Miras, Zöllner, Hartwig, Wudick, Lercher, Chen, Timmermans and Frommer2021) could be implemented.

Integration of single-cell multi-omics techniques

In addition to single-cell transcriptomics, several other modalities of single-cell technology have been recently developed in A. thaliana and other plant species. For instance, these include single-cell epigenomics, metabolomics and proteomics (Farmer et al., Reference Farmer, Thibivilliers, Ryu, Schiefelbein and Libault2021; Marand et al., Reference Marand, Chen, Gallavotti and Schmitz2021; Li et al., Reference Li, Wood, Vu, Hamilton, Rodriguez Lopez, Payne, Serna Guerrero, Gase, Yamamoto, Vaillancourt, Caputi, O'Connor and Buell2023; Montes et al., Reference Montes, Zhang, Nolan and Walley2024). Additionally, spatial metabolomics has also been applied to plants, enabling precise profiling of feature distribution in tissues (Yamamoto et al., Reference Yamamoto, Takahashi, Mizuno, Anegawa, Ishizaki, Fukaki, Ohnishi, Yamazaki, Masujima and Mimura2016). Spatial ‘omics now displays cellular resolution, thus providing complementary information to single-cell technologies. However, single-cell and spatial methods have so far been mostly restricted to plant vegetative tissues.

Epigenomic analysis has been carried out at single-cell resolution in plant species. In particular, chromatin accessibility has been mapped in isolated nuclei from A. thaliana roots using a single-nucleus Assay for Transposase-Accessible Chromatin sequencing (snATAC-seq; Farmer et al., Reference Farmer, Thibivilliers, Ryu, Schiefelbein and Libault2021). To perform snATAC-seq, nuclei are first extracted from a tissue of interest and open chromatin is cleaved and tagged by a transposase. Nuclei are next isolated and transposed DNA fragments are barcoded, with similar methods as for single-cell transcriptomics. Sequencing and analysis of these DNA fragments enables nucleus-specific profiling of the open chromatin in the studied tissue. Hence, snATAC-seq can be used to understand the regulatory landscape of individual cells as well as to identify cell types based on their chromatin accessibility profiles. Single-nucleus epigenomics has not been performed yet in seeds but such analyses would be likely to provide significant new biological insights because of the uniquely dynamic nature of chromatin conformation in seeds across development and germination. For example, it is known that there is allele-specific silencing involved in endosperm parental imprinting (Picard et al., Reference Picard, Povilus, Williams and Gehring2021). Nuclear condensation also occurs during seed maturation and is followed by a global epigenomic remodelling when germination takes place (van Zanten et al., Reference van Zanten, Koini, Geyer, Liu, Brambilla, Bartels, Koornneef, Fransz and Soppe2011; Narsai et al., Reference Narsai, Gouil, Secco, Srivastava, Karpievitch, Liew, Lister, Lewsey and Whelan2017). These different processes involve chromatin structural changes, and single-cell epigenomics would provide further insights into the relationships between chromatin configuration and reactivation of the genome at the cellular level.

Recently, single-cell metabolomics has also been developed in plants, allowing the quantification of metabolites at single-cell resolution with high throughput (Li et al., Reference Li, Wood, Vu, Hamilton, Rodriguez Lopez, Payne, Serna Guerrero, Gase, Yamamoto, Vaillancourt, Caputi, O'Connor and Buell2023). This method relies on protoplast isolation and separation by droplet microfluidics, followed by mass spectrometry (MS). Isolated protoplasts are clustered by PCA, and cluster identity can be determined based on specific metabolomic signatures. Single-cell metabolomics could, therefore, also be helpful to profile metabolite accumulations in different seed cell types. Indeed, it is known that seed tissues are associated with specific metabolite profiles, and single-cell technologies would allow a more precise characterization of their distribution (Zhang et al., Reference Zhang, Ghatak, Bazargani, Bajaj, Varshney, Chaturvedi, Jiang and Weckwerth2021). Similarly, single-cell proteomics has also been established in A. thaliana roots based on comparable technologies, enabling the identification of 3,763 proteins in 756 root cells (Montes et al., Reference Montes, Zhang, Nolan and Walley2024).

Another technology to profile metabolite distribution in plant tissues is spatial metabolomics. Spatial metabolomics relies on MS imaging, which employs either a laser or an electrospray to ionize molecules at specific spots of sectioned tissue. Several technologies are used to perform ionization, the most popular one being matrix-assisted laser desorption/absorption (MALDI). It is then possible to obtain a spatial mapping of the primary or specialized metabolites, or plant hormones, of interest with a resolution up to a few micrometres which is close to single-cell resolution (Kompauer et al., Reference Kompauer, Heiles and Spengler2017). MALDI MS was, for example, used to quantify ABA in immature common bean seeds, revealing that it mainly localizes in the embryo (Enomoto et al., Reference Enomoto, Sensu, Yumoto, Yokota and Yamane2018). However, MS imaging provides low throughput since it allows the selection of a limited number of ions and is then giving information on a reduced set of metabolites. In addition, the small size of A. thaliana seeds represents a limitation to conduct MS imaging with current technologies. Overall, there is a growing interest of single-cell and spatial ‘omics in plant research, but only a few studies have applied them so far on seeds. One major challenge of seed research would be to adapt these methods to developing and germinating seeds of various species, including A. thaliana.

In addition to performing single-cell and spatial analyses individually, these technologies can also be combined in integrated multi-omic studies (reviewed in Vandereyken et al. (Reference Vandereyken, Sifrim, Thienpont and Voet2023)). This approach provides a broad overview of cellular processes in a biological object. Multi-omics integration can either refer to the parallel use of several technologies on a unique sample or to the combined analysis of datasets obtained from multiple ‘omics on different samples from the same tissue. For instance, it is possible to perform simultaneous snRNA-seq and snATAC-seq on the same isolated nuclei (Farmer et al., Reference Farmer, Thibivilliers, Ryu, Schiefelbein and Libault2021). Alternatively, single-cell and spatial ‘omics can also be carried out separately on isolated cells and a section of the tissue of interest, with the subsequent integration of resulting datasets (Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023).

While the integration of single-cell multi-omics techniques is very promising to study biological processes and characterize metabolic pathways, very few publications in plant sciences have presented such single-cell multi-omics due to its challenging nature. Combined snRNA-seq and snATAC-seq have been carried out in A. thaliana root and leaf as well as on rice and sorghum shoot (Farmer et al., Reference Farmer, Thibivilliers, Ryu, Schiefelbein and Libault2021; Nobori et al., Reference Nobori, Monell, Lee, Zhou, Nery and Ecker2023; Swift et al., Reference Swift, Luginbuehl, Schreier, Donald, Lee, Nery, Ecker and Hibberd2023; Liu et al., Reference Liu, Majeed, Kumar, Pajerowska-Mukhtar and Mukhtar2024). The integrated scRNA-seq and single-cell metabolomics were performed for the first time in the medicinal plant Catharanthus roseus and allowed a better characterization of the monoterpene indole alkaloid pathway, including the identification and validation of new genes involved in the biosynthesis of these compounds (Li et al., Reference Li, Wood, Vu, Hamilton, Rodriguez Lopez, Payne, Serna Guerrero, Gase, Yamamoto, Vaillancourt, Caputi, O'Connor and Buell2023). Lastly, the integration of single-cell with spatial ‘omics has so far been mostly limited to transcriptomics (Lee et al., Reference Lee, Nobori, Illouz-Eliaz, Xu, Jow, Nery and Ecker2023). Such single-cell multi-omic analyses would be very helpful to better understand A. thaliana seed development and germination and to characterize non-genetic regulators. In particular, the phytohormones ABA and GA are, respectively, known to regulate seed development and to trigger germination (reviewed in Shu et al. (Reference Shu, Liu, Xie and He2016)). An integrated multi-omic study performed using bulk sample analyses allowed the successive roles of ABA to be deciphered during A. thaliana seed development and germination (Chauffour et al., Reference Chauffour, Bailly, Perreau, Cueff, Suzuki, Collet, Frey, Clément, Soubigou-Taconnat, Balliau, Krieger-Liszkay, Rajjou and Marion-Poll2019). Thus, transitioning from bulk analyses to single-cell and spatial technologies would offer enhanced resolution and help to fine-tune the spatiotemporal roles of ABA and GA during A. thaliana seed development and germination. More generally, the integrated multi-omic studies of the seed at single-cell resolution would allow deeper comprehension of the different roles of seed tissues and their subdomains during seed development and germination. This would enable precise profiling of gene expression, chromatin accessibility and metabolite accumulation in the various cell types and states of seeds.

Several computational tools have been developed to perform single-cell and spatial multi-omic data integration (reviewed in Vandereyken et al. (Reference Vandereyken, Sifrim, Thienpont and Voet2023)). To give some examples, the integration of snRNA-seq and snATAC-seq can be carried out with Seurat (Hao et al., Reference Hao, Stuart, Kowalski, Choudhary, Hoffman, Hartman, Srivastava, Molla, Madad, Fernandez-Granda and Satija2024). The R package spacexr can be used to integrate single-cell and spatial transcriptomic data (Cable et al., Reference Cable, Murray, Zou, Goeva, Macosko, Chen and Irizarry2022; Li et al., Reference Li, Zhang, Guo, Xu, Li, Fang, Hu, Zhang, Yao, Tang, Liu, Zhao, Lin, Cheng, Chen, Xue and Qu2022). So far, there is no computational tool specifically designed to integrate single-cell transcriptomics with spatial or single-cell metabolomics. However, it is still possible to extract information from transcriptomic and metabolic data annotated as similar seed cell types, and subsequently perform integration on averaged profiles by using tools such as the mixOmics R package (Welham et al., Reference Welham, Déjean, Lê Cao and Burger2023). Overall, the integrated single multi-omics is a new field in plant science, for which new tools are being developed and are needed.

Perspectives

Single-cell studies are an emerging trend in plant research, providing comprehensive insights into plant cell types and states. In contrast, intensive research in model animal species over the last 15 years has led to the establishment of protocols, analysis pipelines and well-defined cell-type marker genes for single-cell analyses (reviewed in Jovic et al. (Reference Jovic, Liang, Zeng, Lin, Xu and Luo2022)). In plants, the extent of application of single-cell studies has varied depending on the growth stage or organ. A. thaliana roots and leaves have been quite widely studied through single-cell ‘omics, but by comparison A. thaliana seed has been analysed much less extensively. This is likely because of the specific composition of seeds, which makes extracting protoplasts or nuclei from them difficult. In this review, we have summarized what has been achieved so far and proposed tools to study seeds at the single-cell level. Further developing single-cell and spatial technologies for seeds could help to answer key questions about seed development and germination: what are the different seed cell types and states, and to what extent do they communicate between each other? What is the role of cell specialization in producing self-sufficient and resilient seeds that will give rise to the next generation? How the various seed cell types individually contribute to germination? And what are the cellular mechanisms underlying genome reorganization during germination?

However, the multiplication of the number of single-cell studies poses challenges. First, there is a need to standardize single-cell techniques to be able to compare the different studies performed on the same or on different A. thaliana tissues. This idea was included in the recently proposed Plant Cell Atlas, which aims to comprehensively identify and map all plant cell types (Plant Cell Atlas Consortium et al., Reference Jha, Borowsky, Cole, Fahlgren, Farmer, Huang, Karia, Libault, Provart, Rice, Saura-Sanchez, Agarwal, Ahkami, Anderton, Briggs, Brophy, Denolf, Di Costanzo, Exposito-Alonso, Giacomello, Gomez-Cano, Kaufmann, Ko, Kumar, Malkovskiy, Nakayama, Obata, Otegui, Palfalvi, Quezada-Rodríguez, Singh, Uhrig, Waese, Van Wijk, Wright, Ehrhardt, Birnbaum and Rhee2021). Standardization is, however, complex as each plant tissue presents its own specificities and challenges when single-cell experiments are conducted, due to the distinct biochemistry and composition of different cell types. Recently, Grones et al. (Reference Grones, Eekhout, Shi, Neumann, Berg, Ke, Shahan, Cox, Gomez-Cano, Nelissen, Lohmann, Giacomello, Martin, Cole, Wang, Kaufmann, Raissig, Palfalvi, Greb, Libault and De Rybel2024) proposed guidelines to set up experimental and computational workflows for single-cell transcriptomics. These recommendations could serve as a basis in the single-cell standardization process. One key, but difficult, objective in standardization is to obtain similar cell clusters from separate studies of the same tissue so that independent studies may be compared. We believe that using established lists of marker genes specific to A. thaliana cell types will contribute towards single-cell standardization. Such lists will likely be first curated manually, as done previously for different plant tissues and as we propose here for seed development (Supplementary Table S1). Those lists should also be evolvable by additions of new marker genes provided by future studies or machine learning (Yan et al., Reference Yan, Lee, Song, Li, Schiefelbein, Zhao and Li2022). Finally, single-cell studies provide huge amounts of new data on A. thaliana cell types, for which new methods of computational analysis, documenting and storage are needed. Thus, another great challenge of single-cell ‘omics is to extract relevant information from the datasets that are generated. Such an enterprise will be facilitated by the increasing number of studies that are being performed and the development of new tools in plants.

Overall, the application of single-cell ‘omics to seeds will allow the community to gain a deeper understanding of seed development and germination at the cellular level. Single-cell studies on seeds have so far been mainly limited to A. thaliana and have focused on transcriptomics. Therefore, technical development is required for other plant species and other modalities, which might present new challenges. Application of single-cell ‘omics to crop species could be particularly useful to enrich our knowledge on crucial agronomical traits, such as grain yield, content and vigour as well as germination efficiency.