Decoding and recoding plant development

Abstract

The development of multicellular organisms has been studied for centuries, yet many critical events and mechanisms of regulation remain challenging to observe directly. Early research focused on detailed observational and comparative studies. Molecular biology has generated insights into regulatory mechanisms, but only for a limited number of species. Now, synthetic biology is bringing these two approaches together, and by adding the possibility of sculpting novel morphologies, opening another path to understanding biology. Here, we review a variety of recently invented techniques that use CRISPR/Cas9 and phage integrases to trace the differentiation of cells over various timescales, as well as to decode the molecular states of cells in high spatiotemporal resolution. Most of these tools have been implemented in animals. The time is ripe for plant biologists to adopt and expand these approaches. Here, we describe how these tools could be used to monitor development in diverse plant species, as well as how they could guide efforts to recode programs of interest.

One-sentence summary: Recent advances in tracking cell lineage and molecular states could inspire new strategies to understand and engineer plant development.

Introduction

The development of most multicellular organisms begins with a single cell. To have such complexity arise from the simplest unit of life requires a careful coordination of cell divisions accompanied by a progressive partitioning of cell fate among the daughters (Figure 1A). Although these dynamic multicellular behaviors were among the first to entice biologists, many questions remain unanswered. The earliest approaches to studying development relied on observations with the naked eye (Boyes et al., 2001). These foundational works revealed the remarkable variety of natural forms. With advances in microscopy and cell culture techniques, scientists were able to study development at the cellular level (Van Lijsebettens and Van Montagu, 2005; Ovečka et al., 2018), leading to critical insights like the conservation of recognizable stages of development in different organisms. In the present day, researchers can watch development unfold in real-time with a variety of methods that monitor cell biological and molecular events (Figure 1B). In plants, single-cell RNA sequencing is increasingly widespread, and allows for computational assembly of gene expression over a developmental trajectory (Ryu et al., 2019). For high-quality spatial information, in situ expression can be tracked using fluorescent reporters (Ckurshumova et al., 2011) or fluorescent probes binding RNA (Daigle and Ellenberg, 2007). Combining both temporal and spatial readouts in real-time has remained a major challenge. The recording of gene expression dynamics is especially difficult over very short timescales where transcription, translation and potentially, proper folding of reporters are too slow to capture events, as well as for long timescales where months or years of observation might be needed.

Figure 1.

Decoding plant development by cell-lineage tracing and molecular state decoding. A, Multicellular organisms arise from a single cell through a complex series of divisions and differentiations. A cell lineage corresponds to the cells that originated from the division of one cell. Retracing the cell-lineage tree of one organism or of the development of one specific organ is key to understanding the developmental process. B, Knowing the molecular state of cells over a defined time scale is key to understanding development. These molecular states include details such as which genes are expressed, what is the state of the gene expression network, etc. A putative gene expression network is represented in the figure in gray, with the genes/proteins/molecules active at each stage of development highlighted in black, thick black for high activity and thin black for reduced activity.

Recently, several advances in synthetic biology have enabled real-time tracking of cell lineages with simultaneous decoding of gene expression in mammalian cells. These methods rely on a variety of DNA-based editing tools used as recording systems that should be adaptable for plants. DNA-based recording systems can allow observation at timescales that are not accessible otherwise, as information is encoded in the genome and can be retrieved at any time afterward. These DNA-based editing tools can also be used to engineer development pathways by engineering expression of genes in a defined order with user-specified timing and location. Several recent reviews include comprehensive histories of cell-lineage tracking (Woodworth et al., 2017; Burgess, 2018; Peng et al., 2020), molecular recording (Ishiguro et al., 2019; McKenna and Gagnon, 2019), and recoding and re-engineering development (Ollé-Vila et al., 2016; Ebrahimkhani and Ebisuya, 2019). In this review, while focusing on these three aspects, we will highlight and compare a few of these techniques as a general introduction to the field, and propose directions for this research to study plant development.

Decoding development

Tracking the molecular state of cells during development

Cells must detect and process endogenous and exogenous signals to enact appropriate responses, such as differentiation. To fully understand the molecular trajectory underlying fate transitions, we need methods that can sense and relay information in a way that can be dynamically and quantitatively readout by an observer. In the past 20 years, -omic methods have enabled increasingly precise quantification of DNA, RNAs, proteins, small molecules, ions, and a large number of metabolic intermediates, as well as the associated modifications for many of these categories (Figure 2). These methods capture a snapshot of the molecular state of the organism at the bulk level. Cell/nuclei sorting and single-cell methods have pushed this effort forward, making it possible to assess variation in cellular response within a given cell type. They also allow the capture of the small subset of cells occupying transient stages within a given developmental program. In plants, single-cell RNA sequencing has allowed the characterization of several plant tissue types, and of several cell-fate transitions, such as endodermal differentiation and regeneration of the primary root meristem after injury (Jean-Baptiste et al., 2019; Mironova and Xu, 2019; Nelms and Walbot, 2019; Shulse et al., 2019; Liu et al., 2020; Gala et al., 2021; Meir et al., 2021). An important caveat to these methods is that they require the destruction and disruption of the sample, and thus cannot produce real-time readouts or preserve spatial relationships.

Figure 2.

Decoding the molecular state of development in a spatiotemporal manner. Molecular states in a developing plant vary throughout time and space. It is therefore essential to analyze and record development in a spatiotemporal manner. “Omics” methods enable analysis at a specific time of development, and can therefore be used as a snapshot of the cells at specific timepoints, giving a very detailed level of information. Microscopy methods can be used to follow the expression of a few genes or to follow a few proteins over time. New DNA-based methods can be used as dynamic recorders that can be turned on at the beginning of a mechanism, and then turned off and read. Dynamic recording methods can allow the following of multiple signals, and of genes being expressed, over a long period of time without disturbing development and with single-cell precision. All three methods can be or have been performed in any organ of a developing plant.

Fluorescent reporters in combination with high-throughput, high-precision microscopy have allowed the imaging of cells in a continuous manner in their native context (Figure 2). Live-cell imaging using GFP (Green Fluorescent Protein) tags on proteins or gene-expression reporters has also allowed researchers to track the expression of specific genes (i.e. cell-type specific reporter [Birnbaum et al., 2005]) and protein mechanisms (localization, aggregation, degradation) during cellular processes of interest. Synthetic and natural sensors have also been used to track the presence of specific small molecules or environmental changes, such as the concentration of a phytohormone in individual cells using FRET (Fluorescent Resonance Energy Transfer) sensors (Jones et al., 2014; Rizza et al., 2019; Herud-Sikimić et al., 2021). Hormones are key for plant development and precise tracking of their concentration in individual cells over time allows for a better understanding of development (i.e. auxin in root [Brunoud et al., 2012; Wend et al., 2013]). However, these systems can often be restricted in the amount of information they can produce since the number of fluorescence tags and proteins available are limited. They also require continuous acquisition to achieve recording of dynamics over time, causing extensive photobleaching that hinders long time-scale acquisition. Consequently, this continuous imaging can affect the recorded biological mechanism and is also quite technically challenging as most tissues are not amenable to live-imaging with current devices.

Recently, DNA-based recording systems that overcome some of these technical challenges have been developed, allowing the sensing and relaying of multiple signals simultaneously (Figure 2). By recording the molecular state of the cell at the DNA level, these methods allow for the storage of such information in one step and its subsequent reading in a second step. Two methods exist for this type of DNA-based recording, with one based on serine integrases from bacteriophage (Grindley et al., 2006; Merrick et al., 2018) and the second on CRISPR-Cas9 (Doudna and Charpentier, 2014). The basic principle of these two techniques is highlighted in Box 1. Most of these systems have been tested in bacteria (Sheth et al., 2017; Zúñiga et al., 2020) or in mammalian cells (Perli et al., 2016; Tang and Liu, 2018), but they could also be useful for tracking the molecular state of plants.

Among several methods based on CRISPR-Cas9, we will focus here on CAMERA (CRISPR-mediated analog multievent recording apparatus; Tang and Liu, 2018), which has been tested in mammalian cells as well as in bacteria (Figure 3). CAMERA is based on a cytidine deaminase fused to a catalytically inactive dCas9. This chimeric protein introduces single-base mutations at a locus whose sequence is complementary to a single-guide RNA (sgRNA). In different versions of CAMERA, the genome editor and the single and/or multiple sgRNAs are under the control of inducible promoters. In this way, the presence of an input signal, or a combination of multiple signals, is recorded via a change in the genome sequence. Depending on the duration of the input signal and/or the order of occurrence of multiple signals (mediating expression of different sgRNAs), the ratio and pattern of edited DNA will vary. For example, to track the order of occurrence of two different signals A and B, the sgRNA of input A is designed to target two loci, with one locus being targeted only if it was already mutated by an editing event responding to input B (Figure 3, lower part). The pattern of the signals can be retraced by sequencing the specific locus targeted by the sgRNA(s). Moreover, CAMERA allows analog recording, as the proportion of target loci modified is a function of the amount of input signal, and of the duration of induction. For example, 2, 20,and 200 ng mL⁻¹ of anhydrotetracycline induction lead to different editing proportions over 70 generations. Recording can be performed over a wide time scale, such as from 3 h to 3 d, and has been used to record a wide range of signals such as antibiotics, nutrients, viruses, and light in bacteria, and the induction of the Wnt signaling pathway in mammalian cells.

Figure 3.

DNA-based recording system: an example with CAMERA. The CAMERA technique is a DNA-based recording system that has been used to track the duration and order of occurrence of two to three signals. Recorder plasmids with the dCas9 targets were introduced into a cell alongside the base-editing dCas9 complex. The expression of sgRNAs specific to the recorder plasmids was placed under the control of the input signals (signal 1 in blue and signal 2 in orange). Signals were small molecules added into the growth media. The presence of the input signal leads to DNA editing of the recorder plasmid in the position corresponding to it, with the level of editing (such as the number of plasmids edited) being relative to the time of exposure of the signal. A, Version of CAMERA that records the duration of one signal. In the presence of the blue signal, sgRNA1 is expressed, leading to DNA editing. The duration of the signal is determined by the percentage of edited plasmid (analyzed by sequencing). A short signal duration (left) leads to a smaller fraction of edited plasmids when compared to a long signal duration (right). B, Version of CAMERA that records the order-of-occurrence of two signals (adapted from Tang and Liu, 2018). The initial DNA sequence of the target is written in gray, with nucleotides that are targeted for editing by the sgRNA of the different signals shown in black. Signal 1 leads to mutation of GGG to AAA, and the sgRNA induced by signal 2 will only hybridize with a site already edited by interaction with sgRNA1. If signal 1 occurs before signal 2, the G to C base-editing (in orange) also occurs (left side), and if signal 2 occurs before signal 1, the G to C base-editing cannot be obtained (right side).

CAMERA has allowed the recording of up to three signals at the same time but could theoretically be expanded using additional sgRNAs and promoters responding to different molecules or cellular mechanisms. However, the efficiency of DNA editing using CRISPR-Cas9 is highly dependent on the particular sgRNA, which will probably be the limiting factor to scale this system to a high number of inputs. Large numbers of sgRNAs also result in a competition for Cas9, which can further reduce the fidelity of the system (Huang et al., 2021). The reading of the output has been done by Sanger sequencing of the edited target site, resulting in the loss of spatial context. In the same work, versions of CAMERA have been engineered to have GFP expression as an output once DNA editing machinery repairs a mutated fluorescent protein into a functional fluorescent protein. While these versions allow preservation of the spatial context of signal recording by using fluorescence imaging, they are limited to a single output and up to two input signals. While this system has only been used in cell culture, it could in principle be used in plants to record extracellular and intracellular signals over a long time period. This could be quite interesting for following cells through progressively restricted cell fates during differentiation, for example, or after stimulation with a wound or change to the environment.

Integrase-based circuits have also been used for recording multiple signals. When compared to CRISPR-Cas9, integrase DNA editing is highly efficient, achieving over 90% efficiency in most single-switch systems (Yang et al., 2014; Zúñiga et al., 2020). Since integrase causes the inversion or excision of DNA at the position of its corresponding integrase sites, integrase targets can be designed to induce or repress gene expression through the inversion or excision of genes or promoters. Integrase-based circuits have been implemented in numerous organisms, including bacteria (Bonnet et al., 2013; Roquet et al., 2016; Guiziou et al., 2019), animals (Lakso et al., 1992; Pichel et al., 1993; Bischof et al., 2007; Weinberg et al., 2019) and plants (Hou et al., 2014), and allows the processing of multiple signals (Zúñiga et al., 2020). In plants, single inducible switches have been characterized in Nicotiana benthamiana (Bernabé-Orts et al., 2020). By placing integrase sites in an interdependent manner, researchers have designed circuits to record the order of occurrence of multiple signals. Although these circuits have been characterized in cell cultures (mammalian or bacteria) using synthetic inducible promoters as input signals, these systems should be adaptable for use in multicellular organisms using natural inputs (e.g. detecting small molecules such as hormones or the induction of gene expression during development). While the integrase switch is a binary mechanism, integrase expression can be tuned to respond to different signal levels, allowing the recording at a finer definition. Since the output of the circuit is produced through fluorescent reporters, single-cell tracking can be accomplished using microscopy, which preserves the spatial context of cells under investigation. This system could therefore be used to visualize gene expression and the order in which two or more genes are expressed. The state of the system can also be read at the DNA or RNA level. This should enable combining the tracking of order of occurrence of events with single-cell RNA sequencing, allowing the coupling of proteomic data with spatiotemporal gene expression information.

Tracking cell lineage during development

To fully decode developmental programs and assess variation within and between species, we need a complete map of cell ancestry for a given tissue or organ, in addition to a real-time readout of cell state. This process, called cell-lineage tracing, can be retrospective or prospective (Woodworth et al., 2017). For retrospective cell-lineage tracing, a cell’s family tree is determined retrospectively after development by looking at naturally occurring somatic mutations with DNA sequencing. The theory underlying this method is that mutations accumulate randomly over time. The more related cells are, the more mutations they will share. Retrospective cell-lineage tracing has been used to study human diseases, such as cancer, and human development (Navin et al., 2011; Lodato et al., 2015). As natural somatic mutations are relatively rare, cell lineages can be determined over long periods, but cannot be deconvoluted over a small number of cell divisions. For prospective cell-lineage tracing, a marker is introduced at the start of the development and is used to track descendants of the marked ancestral cell(s) (Figure 4). Many different types of prospective markers have been used, including dyes (Serbedzija et al., 1989), biological markers (Serna et al., 2002; Kurup et al., 2005; Pan et al., 2013), and unique DNA barcodes (Kebschull and Zador, 2018; Figure 4). Prospective cell lineage requires genetic modification, while retrospective techniques can be used in organisms/samples where genetic modification is not accessible.

Figure 4.

Prospective cell-lineage tracing. In prospective cell-lineage tracing, a marker is introduced into a founder cell of an organism early in development (dvp). The founder cell grows and divides, producing a subpopulation of cells that possesses the introduced marker. The cells of the developed organism can be screened for the introduced marker and relation to the original founder cell can be determined by the presence of the marker. Four typical methods of prospective lineage tracing include the introduction of a dye (in orange), a biological reporter gene (e.g. GFP), DNA barcode(s), or a dynamic DNA barcode. The DNA barcoding method introduces one or multiple synthetic sequences of DNA into the genome of the host, which can be sequenced from descendant cells. The dynamic DNA barcoding method introduces both a synthetic sequence of DNA and DNA editing machinery into the host cell. The DNA editing machinery will cause stochastic mutation of the synthetic sequence, which both act as a marker and allow further determination of lineage based upon similarity of mutations.

Prospective cell-lineage tracing has been the historical method of choice for studying plant cell lineage, as full genome sequencing of thousands of individual cells was not feasible until recently. Many past studies utilized biological markers in combination with Activator/Dissociation transposons (Dawe and Freeling, 1990; Kidner et al., 2000; Serna et al., 2002; Kurup et al., 2005; Dubrovsky, 2018). In this method, a transposable element blocks expression of a reporter gene, and the element is excised stochastically (Dawe and Freeling, 1990; Serna et al., 2002) or through heat shock (Kidner et al., 2000; Kurup et al., 2005; Dubrovsky, 2018) in one or few cells, leading to expression of the reporter gene in that cell and its descendants. Time-lapse microscopy experiments can then be performed to track the cells of a specific lineage. Other methods, such as prospective cell barcoding systems, used viruses or transposons to introduce unique labels into multiple cell lineages and track them in parallel. Taken together, along with traditional dyeing techniques, these systems have proven to be immensely useful in basic cell-lineage tracing.

One limitation of these methods is that they can group cells from common lineages, but do not report the order in which cells diverged or the time scale of these events. Dynamic prospective cell-lineage tracing can be achieved by adding markers that are mutated over time by gene editing, creating unique markers that can differentiate cells from the same population (Figure 4). These techniques allow recording of cell lineage in a short time scale but not over longer periods of time since a high mutation rate may cause similar mutations to occur by chance in unrelated groups, blurring the line between related and unrelated groups. Combining different techniques should allow the decoding of full cell lineages. The lack of cell migration in plants facilitates tracing cell lineages over much longer time scales than what is possible in animals (Liu et al., 2010).

The challenge ahead is to build the tools needed to couple molecular/cellular state and cell division/expansion in real time at cellular resolution. Several promising technologies have been applied in animals and these seem well suited for moving into plant systems. For dynamic prospective cell-lineage tracing, CRISPR-Cas9 and recombinase DNA-editing strategy have been used. The CRISPR-Cas9 system makes use of its ability to edit or excise nucleotide pairs to generate incredibly diverse novel barcodes in vivo (Frieda et al., 2017; Kalhor et al., 2018; Chan et al., 2019). For example, CRISPR-Cas9 was used in mice to stochastically edit a defined site, resulting in over a thousand unique insertions/deletions across three cut sites (Chan et al., 2019). Recombinases have also been employed to generate barcode diversity in vivo by using stochastic inversion or excision events (Pei et al., 2017; Chow et al., 2020). One technique called Polylox proposed the use of 10 identical recombinase sites (instead of recombinase site pairs) to further amplify dynamic barcode diversity. Such a system would be capable of producing an astounding 1,866,868 potential end states (Pei et al., 2017).

These techniques demonstrate the incredible potential of the CRISPR-Cas9 and recombinase systems to produce diverse and dynamic DNA barcodes. However, these techniques require the lysis of cells and destruction of multicellular architecture to perform DNA or RNA sequencing and retrieve the novel barcodes. In the interest of maintaining these structures, some scientists have turned to the single-molecule fluorescence in situ hybridization (smFISH) technique as a less invasive readout method. In this technique, fluorescent-labeled oligo probes are introduced to fixed cells and can tag mRNA to report barcode edit state via fluorescence microscopy. Two systems that make extensive use of smFISH are memory by engineered mutagenesis with optical in situ readout (MEMOIR; Frieda et al., 2017) and intMEMOIR (Chow et al., 2020). MEMOIR operates through the random introduction of approximately 8 recording loci or scratchpads, each composed of 10 repeated units, into the host genome (Figure 5A). It uses CRISPR-Cas9 to stochastically target, cut, and delete the scratchpads resulting in 256 possible unique end states. After stochastic deletion of some scratchpads, smFISH can be used to tag all remaining scratchpads, then all existing barcodes. By overlaying the readouts and observing colocalization of barcodes and scratchpads, the state of the scratchpad (either deleted or intact) for each barcode can be observed and cell lineage and relation can be inferred without lysis of cells (Bayer et al., 2015; Frieda et al., 2017). intMEMOIR requires the introduction of 10 pairs of serine integrase sites side-by-side into a single integrase target unit (Figure 5B). Serine integrase stochastically targets the pairs, inverting or excising their DNA regions. Such a method results in 59,049 possible unique end states. Depending on the state of the integrase target unit (inverted, excised, or intact), a different mRNA is expressed (no mRNA for intact target, mRNA blue for inversion, or mRNA orange for excision; Figure 5B). As each mRNA from each target unit is different, the state of the system can be detected through multiple rounds of smFISH (Chow et al., 2020; Figure 5C). The main differences between intMEMOIR and MEMOIR are (1) using integrase, each unit can have three different states versus two with Cas9 system and (2) the efficiency of DNA editing is different between integrase and CRISPR-Cas9, allowing different rates of divergence between observable cell states.

Figure 5.

Examples of dynamic cell-lineage tracing with visual output. A, The MEMOIR technique uses a Cas9/sgRNA DNA editing mechanism. Cas9 proteins coupled with scratchpad-targeted sgRNAs cause the stochastic collapse of scratchpads. Multiple scratchpad barcode pairs are added to the cells, with identical scratchpads coupled to unique barcodes (corresponding to the different colored lines). After induction of the expression of Cas9, scratchpads will be deleted over time in a stochastic manner. By observing individual cells and noting which scratchpad–barcode pairs had their scratchpad deleted, cell lineage can be extrapolated. This observation is performed using smFISH. B, The intMEMOIR technique uses an integrase DNA editing mechanism. The basic integrase target unit is composed of two attP sites in different orientations (the gray filled triangle) and one attB site (the white triangle). The integrase can mediate either excision or inversion of the DNA between the two sites (edited sites are half filled), leading to three different states: no expression of barcode (blue and orange arrows with white filling), expression of barcode 1 (solid blue arrow), or expression of barcode 2 (solid orange arrow). As multiple occurrences of this integrase target unit are present in the cell, DNA editing will accumulate over time, and cell lineage can be recapitulated by analyzing the state of these multiple units. The smFISH technique is similarly used to determine the state of the intMEMOIR unit. C, intMEMOIR, like MEMOIR, enables the retracing of cell-lineage trees over a short period of time through stochastic and heritable DNA editing. The system is inducible, and after induction, the cells are grown for some time to amplify individual lineages (different colors) to simplify observation. A schematic is shown here to represent this experiment on a cellular culture.

Although both MEMOIR and intMEMOIR nicely preserve multicellular architecture, they also have some shortcomings that stem from their low number of potential end states. The greatest barrier to cell-lineage tracing is when two unrelated cells generate the same mutation due to random chance. As such, recording techniques must often choose between extended recording times or heightened recording resolution. A higher mutation rate will allow researchers to differentiate between very closely related cells since each cell develops new mutations but will decrease the recording time since similar mutations appear more rapidly in unrelated cells. Conversely, a low mutation rate allows researchers to record cell lineage for long periods of time, but will also lower resolution since closely related cells do not develop new mutations as they differentiate from each other. These issues are exacerbated in systems with a lower number of different possible end states, as the chance of random mutation producing the same end state in a 256-end state system is fairly high when compared to a system with 1,866,868 potential end states. Experiments done in the foundational studies for MEMOIR and intMEMOIR were limited to 16 and 36 h, respectively (Frieda et al., 2017; Chow et al., 2020). These recording times pale in comparison to the 8-d recording time demonstrated by the CRISPR technique developed by Chan et al. (2019). It should also be noted that the smFISH readout technique is complex and requires a substantial investment of resources and protocol optimization. Fluorescent protein reporters may be a more appropriate alternative form of readout, at least to launch these methods in plants. This alternative option will limit the total number of loci that can be followed to the number of distinct fluorescent proteins that can be analyzed. Therefore, in the long run, an smFISH-type method will be needed to trace a larger number of cells over longer periods.

The use of different systems depends greatly on the situation that researchers find themselves facing. In systems where destruction of multicellular structure is not a concern, other CRISPR systems or recombinase systems prove to be excellent candidates given their incredible ability to generate barcode diversity.

Many cell-lineage tracing studies are now combined with single-cell RNA sequencing (Bowling et al., 2020; He et al., 2020; Wagner and Klein, 2020), directly linking cell lineage, cell fate, localization, and/or gene expression. The rapid development of technologies in this area leaves us optimistic that the systems biology vision of synthesizing information about development across the molecular, cellular, and organismal scale is likely within reach.

Recoding development

Engineering development or environmental response

To better understand the mechanisms underlying developmental processes, biologists have often used approaches that perturb the system, thereby revealing critical regulators or events. For example, forward or reverse genetic approaches rely on mutations to connect genes to specific phenotypes. Current gene editing technologies have made this approach even more straightforward to apply to a range of organisms, even those that had not historically served as model or reference organisms. If we combine what is already known about plant development with the ever-improving tools for engineering biology, we should be able to go further in our perturbations. We should be able to recode development to better understand it. Recoding development permits us to test our current knowledge and to test which parts of the process are essential. In addition, it allows the engineering of biological systems of interest for humans. Recoding can be done in a top-down approach starting from an already existing organism, or from a bottom-up approach by engineering development from single cells (Figure 6A).

Figure 6.

Recoding life. A, Different plants adapt differently to different stimuli. This figure serves as a schematic example of a weed being well adapted to an environment while the crop plant is not growing well due to poor root development. By engineering the development of the root system of the crop plant, we could obtain crops capable of growing well in this harsh environment while producing high-quality fruit in a sustainable manner, without the need for additional water or supplements. B, Single cells have been engineered to develop into different shapes and patterns according to the genetic program implemented. Programs 1 and 2 correspond to different genetic programs and the cell colors correspond to different molecular states, such as expression of a fluorescent protein or production of defense hormones. C, By better understanding development, we could synthetically recapitulate evolution by engineering cellular communication and clonal formation in unicellular organisms.

Engineering of signaling pathways can be performed at different scales: at the level of the sensor, of the signal processing, or of the output (Leydon et al., 2020). A highly successful proof-of-principle of this approach has been the engineering of the receptor for abscisic acid (ABA) to detect another chemical (Park et al., 2015). The application of this new chemical will trigger the ABA pathway in transgenic plants, leading to greater tolerance to drought. It is also possible to engineer new synthetic signaling pathways that respond to existing hormones. One example is the hormone-activated Cas9-based repressors (HACRs), which can bring any gene of interest under the control of the endogenous hormones auxin, gibberellins, or jasmonates (Khakhar et al., 2018). This can have dramatic impacts on developmental programs, as was demonstrated with the targeting of the auxin transporter PIN-FORMED1 to alter shoot architecture. Whether engineering new signals or new responses, these approaches allow researchers to perturb developmental and environmental response pathways to understand the native systems and to fulfill design specifications to solve applied problems.

Engineering from scratch

Engineering multicellular organisms from scratch is one “blue-sky” goal of synthetic biology. As a first step to building a plant from the ground up, researchers are working on building a synthetic chloroplast (Agapakis et al., 2011; Miller et al., 2020). In one example, Erb et al. engineered a synthetic metabolic pathway for converting light into chemical energy (Miller et al., 2020). By encapsulating the enzymes of the pathways in microdroplets, they created an artificial photosynthetic system mimicking chloroplast. This is a first step to obtaining a replicable unicellular photosynthetic system.

Cell–cell communication is a key element to engineer a multicellular organism. Synthetic cell–cell communication in mammalian cells has been engineered using synNOTCH receptors (Morsut et al., 2016), target T-cell signaling (Sadelain et al., 2013), and orthogonal pairs of cytokine and receptors (Sockolosky et al., 2018; Silva et al., 2019). Cell–cell communication combined with synthetic circuits allowing pattern formation leads to the engineering of multicellular behaviors, such as cellular patterns and structures (Toda et al., 2018; Figure 6B). Patterns have mainly been engineered following the same principle of corepression and activation, typically using two to three different repressors and/or activators (Basu et al., 2005; Santos‐Moreno and Schaerli, 2019). While this work has been done in nonplant organisms, phytohormones, such as auxin and cytokinin, have been used in yeast to build synthetic cell–cell communication channels (Chen and Weiss, 2005; Khakhar et al., 2016; Yang et al., 2019; Ma et al., 2020). In plants, the use of phytohormones to engineer cell–cell communication will lead to cross talk with existing signaling pathways. Signal networks orthogonal to natural phytohormones will have to be engineered. As a first step, synthetic phytohormone receptors connected to natural signaling pathways (concave TRANSPORT INHIBITOR RESPONSE 1, ccvTIR1 responding to convex INDOLE-3-ACETIC ACID INDUCIBLE, cvxIAA [Uchida et al., 2018], and PYRABACTIN RESISTANCE(PYR)^MANDI responding to mandipropamid [Park et al., 2015]), and synthetic receptors responding to natural phytohormones and with synthetic output (HACRs; Khakhar et al., 2018) have been engineered.

In addition to cell–cell communication, morphology of cellular shape has been engineered: artificial scaffolds, 3D bioprinting, and complex media formulations are currently used and being developed (Entcheva et al., 2004; Dey and Ozbolat, 2020). A model for controlling the formation of multicellular masses into arbitrary shapes has recently been published and is based on the design of recombinase-based logic circuits (Appleton et al., 2019). The theoretical principle is that one cell with one genetic circuit can, through genetic recombination, lead to different types of genetic circuits with different phenotypes. Therefore, a lineage tree is preencoded in the genetic circuit, and one cell will lead to a cellular population, which over time will have different encoded cell cycle states, such as division, asymmetrical division, or cycle arrest. While still being at the stage of a model, designs based on DNA encoding genetic circuits with memory have strong potential for engineering development at the molecular level.

Engineering to retrace evolution

By better understanding how multicellular organisms pass from a single cell to a well-organized multicellular system, we could recapitulate evolution by engineering key innovations in plants (Figure 6C). For animals, much work has been done in choanoflagellates as they are considered to be at the origin of multicellular organisms (Brunet and King, 2017). Cell–cell communication and specification could be engineered using synthetic genetic circuits in choanoflagellates to better understand the steps needed to become a multicellular organism. In a similar vein, for unicellular eukaryotes, multicellular yeast have been engineered by disrupting the transcription factor ACTIVATOR OF CUP1 EXPRESSION (ACE2), which prevents separation of mother–daughter cells, leading to “snowflake yeast” (Ratcliff et al., 2015). This work has shown that simple gene disruption can lead to important evolutionary changes (here the engineering of clonally developing clusters), and can be the first step to synthetically engineering multicellularity and recapitulating evolution.

Similar work could be performed in the plant kingdom. Indeed, different examples of independently evolved, complex multicellularity developed clonally (Knoll, 2011). In plants, multicellularity occurred separately at different times during evolution, making the evolution of plants as a multicellular organism even more interesting to study, as comparative studies can be performed (Umen, 2014). Charophyte algae are considered to be the closest algal relatives of land plants and would be a model of interest to engineer multicellularity. In addition to multicellularity, macroscopic complexity in green algae has been reached through diverse strategies, such as what is observed in giant uninucleate algae (i.e. Acetabularia). The engineering of a unicellular green algae to form various complex macroscopic forms would be an excellent place to start understanding the evolution of existing and novel plant forms.

Conclusion

This is an exciting moment to be studying plant development. Synthetic biology and systems biology are rapidly adding new tools that will allow us to decode the molecular networks that give rise to complex forms. Recently, cell-lineage tracing methods have been combined with single-cell -omics to decode cell divisions, cell fate transitions, and transcriptional cell states during development on a single cell-level and in a temporal manner. While most tools have still been used only in animals, these approaches should be readily adaptable to plant systems and provide important new insights. We believe that the implementation of these techniques will allow a finer decoding of dynamic systems in plants, enabling both single-cell precision and collection of spatiotemporally relevant information. In the medium term, these systems could be easily adapted to engineer development by using developmental genes as outputs or using sgRNA to tune the regulation of genes. Recording circuits could also be used in nonmodel organisms with longer life cycles, as these systems do not require constant observation. The promise of recoding plant development at-will to advance our fundamental understanding of plants, and to engineer crops with traits useful for sustainable food production, is on the horizon (Figure 7).

Figure 7.

Engineering life to understand it. Cell-lineage tracing, recording of molecular state, and recoding of development are used together to better understand development and to allow the engineering of development.

Advances

• New DNA-recording techniques can track the order of occurrence of signals (CAMERA) and trace cell lineage (MEMOIR and intMEMOIR).

• Single-cell “omics” techniques are shedding new light on cell fate specification during plant development.

• Combining single-cell “omics” and cell-lineage tracing methods in plants are needed to better understand multicell behaviors like morphogenesis.

• Significant advances have been made in engineering hormone signaling. These approaches need to be expanded to other signaling pathways to enable sophisticated engineering of new morphologies and/or functions that can meet current and future environmental challenges.

• Engineering of multicellular organisms from scratch is still in an early stage; however, advances in areas like engineering a synthetic chloroplast hold great promise.

Outstanding questions

• Will current technologies enable changes in gene expression in individual plant cells to be recorded and reconstructed over timespans that range from seconds to years?

• If such technologies are deployed, will they reveal new principles that enable multicell coordination? For example, will they shed light on how symmetry in a developing organ is maintained in the face of stochastic variation in gene expression?

• Could integrase- and/or Cas9-based technologies be used to expand the current repertoire of plant morphologies to better serve specific functions or environmental conditions?

Box 1 DNA editing mechanics of CRISPR-Cas9 and integrase systems

In this article, there are two main biological systems we will discuss in detail for the purpose of targeted DNA editing: the CRISPR-Cas9 system and the integrase systems.

The CRISPR-Cas9 system uses the CRISPR-associated endonuclease (Cas9) protein and single guide RNA (sgRNA) molecules to cause targeted breaks at defined sites in the DNA. First, specific gRNAs are produced by the cell, which will combine with and guide the Cas9 proteins to a DNA site complementary to the gRNA. The Cas9 will then cause a double-stranded break at this site. From here, DNA can be ligated into the newly formed gap, or a second break can be made elsewhere to cause excision of the intervening DNA sequence.

The integrase systems use attP and attB binding sites in conjunction with integrase enzymes to either invert or excise specific sequences of DNA. Two different types of integrase exist: serine and tyrosine integrases; here we focus on serine integrase. Each serine integrase recognizes two specific DNA integrase sites of roughly 40 base pairs called attB and attP sites. If the two sites are in opposite orientations in the DNA, the integrase will mediate inversion of the DNA between the two sites. If the two sites are in the same orientation, the integrase will mediate excision of the DNA. This process is irreversible with integrase alone, and different integrases recognize different integrase site pairs; therefore, integrases are orthogonal to each other.

These two systems provide the function of editing DNA at defined sites. The CRISPR-Cas9 system is highly flexible since numerous target sequences can be edited by simply modifying the sgRNA used, while integrases require one integrase per integrase site pair. Conversely, the efficiency of CRISPR-Cas9 editing is highly dependent on the specific sgRNA and is in general lower than the integrase system. Comparatively, integrase-editing systems are highly efficient, with switches operating at >90% efficiency. In addition, integrases allow access to multiple DNA and transcriptional states, which permit the engineering of systems with a wider range of outputs.

 

S.G. conceptualized the subject of the review, performed the bibliographical research, designed the artwork, and wrote the manuscript. J.C.C. performed the bibliographical research, contributed to the design of the artwork, and wrote the manuscript. J.L.N. consulted throughout the conceptualization and writing process, as well as providing editorial advice.

The author responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (https://academic.oup.com/plphys/pages/general-instructions) is: Jennifer L. Nemhauser (jn7@uw.edu).