Molecular Recording: Transcriptional Data Collection into the Genome

Abstract

Advances in regenerative medicine depend upon understanding the complex transcriptional choreography that guides cellular development. Transcriptional molecular recorders, tools that record different transcriptional events into the genome of cells, hold promise to elucidate both the intensity and timing of transcriptional activity at single-cell resolution without requiring destructive multi-time point assays. These technologies are dependent on DNA writers, which translate transcriptional signals into stable genomic mutations that encode the duration, intensity, and order of transcriptional events. In this review, we highlight recent progress towards more informative and multiplexable transcriptional recording through the use of three different types of DNA writing—recombineering, Cas1-Cas2 acquisition, and prime editing—and the architecture of the genomic data generated.

Graphical abstract

The journey from the zygote to fully differentiated cell relies on a series of temporally choreographed transcriptional events. Understanding the precise order of events that yields one cell type versus another is critical to advance our knowledge of basic developmental biology. Moreover, this knowledge has practical ramifications for regenerative medicine, as the sequence of events that unfolds in a developing cell may be mimicked in vitro to produce replacement parts for degenerative diseases. Unfortunately, conventional transcription assays like RNA-seq and in situ hybridization are not well suited to understand long and complex processes like development because they require destruction of cells in the midst of the event for analysis. To reassemble the resulting transcriptional snapshots into a continuous process requires analytical assumptions that are not always true [1].

An emerging set of technologies aims to solve this problem by recording biological data into a molecular record (DNA, RNA, or protein) that remains inside the cell during the process of interest. Since the data collection is non-destructive, the overarching biological process plays out from beginning to end, after which the data can be collected by imaging or sequencing. Much work remains to be completed to realize the lofty goal of recording all biological data into molecular records. However, a suite of useful molecular parts is emerging. Over short timescales (minutes to hours), the age of individual transcripts can be encoded onto the transcripts themselves using RNA deaminases, and the order of selected transcriptional events can be recorded as fluorescent tags incorporated into elongating protein polymers [2–4]. Over longer timescales (hours to days), DNA is the recording medium of choice. DNA-based dynamic lineage recorders (reviewed in [5,6]) use CRISPR nucleases to diversify sites in the genome to encode the relationship of cells over multiple generations. In this review, we will focus on a particular suite of molecular recording technologies: those that aim to record the order of transcriptional events over long timescales by writing a DNA-based molecular record.

The potential impact of transcriptional molecular recording in DNA is illustrated by Cre and FLP recombinase-based reporters—ubiquitous tools within developmental biology—that can be considered simple transcriptional molecular recorders. These systems link a transcriptional event to the expression of a recombinase that makes a permanent genomic modification to a cell. Thus, the occurrence of an event is stably recorded in DNA and can be read-out at a later point in time, often by using the genomic modification to turn on a fluorescent protein. These reporter lines have been invaluable to identify specific cell populations that rely on a given transcription factor to define their cell fate [7,8].

Yet, despite the clear value of these early recorders, they are limited to recording only as many independent events as the number of fluorescent proteins that can be resolved simultaneously. Moreover, they only encode the occurrence of an event, but not when it happened. By omitting the fluorescent readout and focusing instead on the mark made to the genome as the data itself, the number of distinct signals recorded can be further extended. Moreover, if these marks to a genome are organized sequentially, event order can also be determined to yield a richer understanding of complex cellular processes. We will focus on three molecular strategies for such recordings using distinct molecular components: (1) recombinases; (2) reverse transcriptases (RTs) and CRISPR integrases; and (3) RTs and CRISPR nucleases. Despite the differences in encoding, we will argue in the concluding section that a common data structure is emerging across all strategies.

Recombinases

Recombinase-based molecular recorders rely on the ability of DNA recombinases to flip or delete a DNA sequence surrounded by two recognition sites as a genetic mark or output. Additional layers of complexity can be added by combining multiple recombinases, promoters, and terminators to create circuits capable of responding with different genetic outputs depending on the number, order, and mixture of distinct inputs [9–13] (Figure 1a). This approach enables the development of bacterial sentinel cells that can be used as biosensors to analyze human samples, for instance detecting too much glucose in urine, a sign of diabetes [14]. Furthermore, multiple recombinase circuits have been validated in mammalian cells [15,16,13].

Figure 1.

Three different DNA writers enable transcriptional molecular recording.

Transcriptional molecular recorders rely on the activity of a DNA writer to record different transcriptional events. (a) Site-specific recombinases are expressed from a promoter of interest and modify DNA between two recognition sites. Each edit from an orthogonal recombinase indicates a transcriptional event, but the order of events cannot be reconstructed. (b) Cellular or barcoded RNAs are expressed from a promoter of interest and reverse-transcribed into DNA pre-spacers that can be acquired and integrated into a CRISPR array via Cas1-Cas2. Since each spacer is always integrated next to the leader sequence, the order of events can be inferred. (c) Barcoded pegRNAs are expressed from promoters of interest. The pegRNA directs a prime editor to the pegRNA binding sequence (PBS) in a pre-engineered array, where it incorporates a barcode corresponding to a transcriptional event, obfuscates the previous PBS, and adds a new PBS to which the next prime editor can bind. This design allows the order of events to be inferred.

Encoding intensity and duration of events with recombinases

Early recombinase recorders were constructed to record whether a transcriptional event occurs rather than an event’s relevant analog characteristics, such as duration or intensity. To overcome this limitation, several recombinase-based recorders were developed to more closely mimic an analog recorder, by increasing the total number of recording cells or the number of recording plasmids per cell [17,18]. While the presence of a genomic mark in a single bacterium or plasmid can only encode the occurrence of a signal, the number or percentage of such events in a larger population of loci can reflect the duration or intensity of a signal. A recurring critique of recombinase-based recorders has been that the number of independently recorded events is limited to the number of orthogonal recombinases [19]. However, a recent tweak on the approach used catalytically inactive Cas9 to direct a single recombinase to integrate distinct sequences into an expanding genomic array, depending on the gRNA expressed [20]. This design enabled duration and intensity recordings of multiple transcriptional events simultaneously, but since recombinases do not inherently have a writing direction, additional molecular components will be required to reliably reconstruct event order.

RTs and CRISPR Integrases

An alternative DNA writer with inherent directionality uses the CRISPR integrases Cas1 and Cas2. These integrases act as part of a bacterial immune system that acquires phage DNA sequences into a genomic repository as an immunological memory of that phage. This genomic repository, called the CRISPR array, consists of a short leader sequence followed by a series of unique fragments of DNA from foreign invaders, called spacers, separated from each other by repetitive DNA sequences called repeats. During acquisition, a complex of Cas1 and Cas2 can integrate spacers into the CRISPR array next to the leader sequence [21,22]. Since the Cas1-Cas2 complex always inserts its newly caught spacer next to the leader, the recorded spacers are also captured in chronological order, where the oldest events are those furthest away from the leader. Early technological work in this area showed that the Cas1-Cas2 acquisition system can be hijacked to store electroporated synthetic oligonucleotides as spacers, even if these spacers contained information unrelated to the immune system—such as data encoding a video [23,24].

Capturing transcriptional events via Cas1-Cas2 acquisition using reverse transcription

Unfortunately, most natural CRISPR systems integrate DNA, rather than the RNA that would be required for a transcriptional recorder. One workaround to this problem is to use a biological signal to modulate the copy number of a plasmid containing pre-spacer sequences [25]. This approach was later modified by replacing the biological signal with an electrical signal to encode data in bacteria for industries in need of secure data [26]. Alternatively, other recent strategies instead convert RNA into DNA using an RT (Figure 1b).

Components of a CRISPR system from M. mediterranea (MMB-1), including a natural RT-Cas1 fusion, were shown to acquire spacers derived from donor RNA in MMB-1 [27]. RT-Cas1 has now been translated into a recording technology [28–30••]. Using an F. saccharivorans RT (FsRT)-Cas1 fusion which can promiscuously reverse-transcribe most RNA transcripts into potential spacers, this technology captures a diverse set of RNA-derived spacers into a population of CRISPR arrays and thus extends the global transcriptomic snapshot provided by RNA-seq to a global transcriptomic history. Proof-of-concept work demonstrated differentiable transcriptomic histories of bacteria that were or were not transiently exposed to a herbicide [28].

In addition to using Cas1-Cas2-based transcriptional recorders to develop a global transcriptomic history, we showed that they can also be used to infer the ordering of two specific promoters of interest [31•]. This approach replaces the promiscuous FsRT with a retron RT, which only reverse transcribes its own non-coding RNA (ncRNA). Plasmids were engineered to express two inducible promoters of interest, each linked to a different barcoded retron ncRNA. The ordering of the barcoded spacers in individual genomic arrays can be used to determine the order in which different promoters were previously induced.

Porting Cas1-Cas2-based molecular recorders to mammalian cells

Although certain elements relevant to transcriptional recording, such as the retron RT, are able to reverse-transcribe RNA and even mediate editing in eukaryotic cells, including yeast and human cells [32–35], Cas1 and Cas2 have so far not been shown to be functional in eukaryotic cells. This limitation to porting Cas1-Cas2-based recording systems into mammalian cells, which may rely on host factors like E. coli immune host factor [36,37], must still be solved to increase the impact of this technology.

RTs and CRISPR Nucleases

A CRISPR component that has had no trouble being ported into eukaryotic cells is Cas9. Previous event recorders have already capitalized on Cas9 or another CRISPR nuclease, Cas12a, to link specific biological stimuli to edits in DNA, but these technologies do not capture the chronological order of the events [38–42] (see also [43,44] for in-depth reviews of how CRISPR nucleases have been used as event recorders). However, by combining Cas9’s flexible operation in multiple cell types with the architectural strengths of a CRISPR array and an RT to convert RNA signals into DNA, new recording systems based on prime editing [45] have been created in mammalian cells (Figure 1c).

A prime editor (PE) consists of a Cas9 nickase fused to an RT. The Cas9 nicks at a specific site encoded on its accompanying prime editing guide RNA (pegRNA). Afterwards, the RT reverse primes off the exposed genomic cut site to reverse-transcribe an edit-encoding extension on the pegRNA and create a precise edit in the target site [45]. To build a molecular recorder, the edit-encoding extension of the pegRNA can be engineered to include a sequence that encodes a barcode followed by a pegRNA-binding sequence that a future pegRNA needs to mediate the next edit. As a result, a series of barcodes, all encoded by unique pegRNAs expressed under different promoters or signals of interest, can be added in an ordered fashion to a chosen genomic locus, much like the architecture of a CRISPR array.

Three recent papers have used this prime editing-based strategy to develop a PE-based molecular recorder within mammalian cells that can accurately encode information [46•] and measure the strength and intensity of different activated signaling pathways, such as Wnt [47••]. Additionally, after sequentially transfecting cells with plasmids expressing pegRNA, PE-based recorders accurately captured the order of the nucleic acid delivery [46•,48•].

Although the use of prime editing overcomes the limitation of porting Cas1-Cas2-based molecular recorders to eukaryotic cells, current recorders using PEs must either pre-build an array of a defined length [46•] or use multiple pegRNA-binding sequences that switch back and forth [48•], creating arrays that are less open-ended than the CRISPR arrays that were the inspiration. Moreover, inefficiencies in barcode insertion still need to be overcome to capture ordered biological information within living cells with these systems.

Outlook & outstanding challenges

The quest to build an ideal transcriptional molecular recorder has spawned numerous molecular incarnations. Each DNA writer has its own unique strengths and weaknesses. Recombinases are comparatively efficient, with a better chance of recording rare or transient events, but are more difficult to scale and lack directionality. Cas1-Cas2-based approaches are nearly agnostic to the number of barcodes and use an open-ended array, which bodes well for scalability and recording duration, but are currently limited to bacteria. Prime editing-based approaches are multiplexable and deployable in eukaryotes, but have a less open-ended data array, and need improvements in efficiency to record event timing.

Nonetheless, the most modern and promising transcriptional recordings share a common recording infrastructure to store transcriptional information. This data structure can be described as a CRISPR-like array which continually expands in a single direction by adding barcodes corresponding to a transcriptional signal. Such an infrastructure is highly multiplexable, stores high-density data, and conveys clear event ordering.

However, such a data structure also carries some inherent limitations. First, arrays containing multiple signals of interest, and thus clear event ordering information, are very rare. Since the probability of integrating multiple spacers is multiplicative, the chances of creating an array with two or three signals of interest recorded is exponentially rarer than an array with one signal of interest. This issue is further compounded by the low integration efficiencies of current DNA writers like Cas1-Cas2 and PE. These multi-spacer arrays may still be captured by increasing sequencing depth or using methods like SENECA to specifically enrich and amplify expanded CRISPR arrays [28,29]. Nonetheless, the rarity does limit the ability of any expanding array-based recorder to approach single-cell resolution. One solution is to optimize recording components to increase efficiency so that arrays with multiple informative events are more frequent, but there are likely inherent limits on that optimization (e.g., the refractory time it takes to repair a genomic array after barcode integration).

Second, deconvolving signal intensity and duration is extremely difficult. Analyses of molecular recordings that use expanding arrays are influenced by both characteristics, since each of them result in an increased number of integrated spacers [31•,47••]. Instead, more nuanced methods or multi-dimension readouts are necessary to describe a signal that occurs earlier but at a lower intensity compared to a signal that occurs later but at a higher intensity. In future incarnations, including an additional timestamp within the spacer sequence could help differentiate signal duration from intensity.

Finally, developing an expanding array-based recorder will almost always require some amount of pre-engineering, both to insert signals into a genomic recording locus and to drive the machinery that performs the DNA writing. The recording may also place a burden on its host that could perturb the transcriptomic changes that researchers aim to capture. This limitation may prevent transcriptional recording technologies from being used directly in the cells of interest. Rather, engineered sentinel cells may be a more tractable approach to understanding certain aspects of human health [14,49,30••].

Extensive research into different molecular components has mediated impressive gains in the scalability and transferability of transcriptional recorders, and continued optimization will likely result in greater efficiency and resolution. However, given an expanding array-based data structure’s inherent constraints, investigating alternative architectures for biological data storage is a promising method to further revolutionize transcriptional molecular recording. Additional progress in incorporating direct- or random-access memory to biological data storage could, for example, reveal novel ways to approach single-cell resolution and unambiguous information content. After overcoming these key barriers, transcriptional molecular recorders are poised to become a widespread and invaluable tool in understanding gene expression during complex events like development.

Highlights.

• Molecular recorders capture transcriptional activity within the genome of cells

• Transcriptional activity is recorded by linking a signal of interest to a DNA writer

• Recombinases can be used to capture event timing, but scale poorly

• Cas1-Cas2 DNA writers enable scalable recordings of event timing, but have been limited to bacteria

• Prime editing translates Cas1-Cas2’s strengths to mammalian cells, but is not yet efficient enough to capture data in single cells