Rewritable multi-event analog recording in bacterial and mammalian cells

Abstract

We present two CRISPR-mediated analog multi-event recording apparatus (CAMERA) systems that use base editors and Cas9 nucleases to record cellular events in bacteria and mammalian cells. The devices record signal amplitude or duration as changes in the ratio of mutually exclusive DNA sequences (CAMERA 1), or as single-base modifications (CAMERA 2). We achieved recording of multiple stimuli in bacteria or mammalian cells, including exposure to antibiotics, nutrients, viruses, light, and changes in Wnt signaling. When recording to multi-copy plasmids, reliable readout requires as few as 10-100 cells. The order of stimuli can be recorded through an overlapping guide RNA design and memories can be erased and re-recorded over multiple cycles. CAMERA systems serve as “cell data recorders” that write a history of endogenous or exogenous signaling events into permanent DNA sequence modifications in living cells.

Introduction

Recent technologies have enabled the study of the internal state of cells in exquisite detail, including the sequence of the genome, the status of epigenetic modifications, and the identity and abundance of cellular RNAs, proteins, metabolites that collectively determine cell state (1, 2). Far less developed are tools to reveal a cell’s history and how that history determines present and future cell states, despite the potential impact of such capabilities. Detailed information on cell states during division and differentiation, for example, could illuminate the process of aging, and recording the presence and duration of exposure to external or internal stresses could inform the emergence of cancer and other diseases. Recording a cell’s history in a highly multiplexable, durable, and minimally perturbative form has been a long-standing challenge of the life sciences (3, 4).

Transient recording of environmental signals has been achieved by manipulating transcription and translation in bacteria (5). Information recorded in this manner, however, cannot be passed on to future generations of cells and the recording process itself is delicate since many factors contribute to transcription and translation efficiencies. In contrast, recombinases can modify designated genomic sequences and the resulting information stored in DNA can be read even after cell death (6–9). Although individual signals of interest can be stably recorded using recombinase-based memory devices, orthogonal recombinases are required to record more than one bit of information. Cellular recording devices operated by recombinases have been successfully applied to record the presence or suggest the absence of stimuli, but their use to record signal strength, duration, or order is more challenging (3).

Based on their recording format, cellular memory devices can be categorized into digital and analog devices. In contrast to digital memory devices that store information in one of two distinct states (on or off), analog memory devices that leave permanent marks in DNA in a manner that reflects the strength or duration of endogenous or exogenous stimuli in theory could illuminate cellular history, reveal how a stimulus dictates downstream responses, and improve our ability to predict cell behavior (3). Recently Lu and coworkers reported synthetic cellular recorders integrating biological events (SCRIBE), an elegant memory device that translates exogenous signals into point mutations in a bacterial genome through Beta protein-assisted single-stranded DNA incorporation (10). Because the production of single-stranded DNA by the adapted retrovirus cassette is not efficient, SCRIBE requires the sampling of large populations of bacteria for both recording and readout (10).

To develop a memory device that is less dependent on a large cell population, we chose the clustered regularly interspaced short palindromic repeats (CRISPR)-Cas9 nuclease (11–14) and Cas9-derived base editors (15, 16) to serve as DNA writing modules. Both Cas9 nuclease and base editors make changes in cellular DNA in an efficient and programmable manner when complexed with guide RNAs (11, 15). If linked to stimuli or cell state changes, these DNA modifications in principle could serve as durable messages that reflect a cell’s history and could be read out using modern sequencing technologies, even after cell death. In this work, we present two CRISPR-mediated analog multi-event recording apparatus (CAMERA) systems and demonstrate their ability to record simultaneously multiple cell states, including exposure to antibiotics, nutrients, viruses, light, and a kinase inhibitor that alters endogenous Wnt signaling.

Results

A plasmid compensation system as an information carrier in bacteria

We chose the S. pyogenes Cas9 (SpCas9) nuclease as an initial DNA writing module because it functions robustly across many different cell types in vitro and in vivo (13, 17). SpCas9 makes double-stranded DNA breaks at loci that match the 20-base “spacer region” of a single guide RNA (sgRNA) and that are near an NGG protospacer-adjacent motif (PAM). In mammalian cells, the resulting double-stranded breaks can be repaired by nonhomologous end joining (NHEJ) and similar processes to introduce insertions and deletions (indels), or through homology-directed repair (HDR) by supplying a template strand. In bacteria, however, double-stranded DNA breaks frequently cause cell death or a loss of extrachromosomal DNA (18, 19). To translate DNA loss following double-stranded breaks into durable information, we designed a high-copy number plasmid compensation system to store DNA modification states. This strategy enables analog recording within each cell and thereby avoids dependence on large cell populations.

The plasmid compensation system includes a pair of nearly identical recording plasmids, R1 and R2, that differ only at a 3-nucleotide coding mutation in an EGFP gene (Fig. 1A). The EGFP gene encoded by R1 expresses full length-fluorescent protein, whereas the EGFP gene in R2 contains a premature stop codon and cannot produce fluorescent protein (Fig. 1A). Because the two plasmids are virtually identical, we hypothesized that their fitness cost to host cells is very similar and they should coexist in a stable ratio for long periods of time.

Fig. 1. Plasmid ratios controlled by Cas9 nuclease in CAMERA 1.

(A) Schematic representation of CAMERA 1 systems. Recording plasmids R1 and R2 are identical except a 3-nucleotide coding mutation in the EGFP gene. The expression of the Cas9:sgRNA complex is controlled by the signal of interest and results in R1 depletion in the bacteria that carry the recording plasmid pair. (B) Stability of the R1:R2 ratio in E. coli S1030 cells in the absence of the writing plasmid. (C) In vitro cleavage of the wild-type and mutated EGFP gene by Cas9 in the presence of sgRNA1. The designed spacer sequence targets the distinct region in EGFP so the Cas9:sgRNA complex cleaves R1 much faster than R2. (D) Recording the amplitude and duration of aTc by CAMERA 1.0. Values and error bars reflect the mean and s.d. of three replicate cultures derived from a single bacterial colony.

The R1:R2 ratio serves as the information carrier that reflects the signal of interest in an analog mode. To convert the signal of interest into an R1:R2 ratio change, a Cas9:sgRNA pair induced by the stimulus cleaves plasmid R1 but not R2 (Fig. 1A). The resulting double-stranded break causes the loss of R1. Because the two recording plasmids share the same origin of replication that controls the total copy number of the plasmids in bacteria, the loss of R1 initiates the replication of the remaining plasmids and the gradual accumulation of R2. A high copy number plasmid (pUC) was chosen to maximize the analog recording range of the system (Fig. 1A).

To test the stability of the plasmid compensation recording system, E. coli strain S1030 (20) was co-transformed with R1 and R2 and two single colonies with different R1:R2 ratios were isolated. The colonies were separately grown in LB media at 37 °C and the culture was diluted 500- or 1,000-fold six times over 144 hours for a total dilution ratio of 10¹⁷ (Fig. 1B). The two starting colonies contained 29% R1 and 60% R1, and their R1:R2 ratio was very stably maintained throughout the growth and dilution process (Fig. 1B), ending at 29% R1 and 59% R1, respectively. These results indicate that R1:R2 ratio can serve as a stable analog information carrier across a range of plasmid ratios.

To assess the potential growth burden that the recording plasmid pair might impose on bacteria, we measured growth curves for the parental E. coli strain S1030 and two S1030 colonies containing R1 and R2 in different ratios (29% R1 or 60% R1, fig. S1). The colonies harboring R1 and R2 exhibited the same growth rate as the parental strain in the presence or absence of the selection antibiotic and all bacterial cultures reached the same final cell density, suggesting that the recording plasmids do not substantially impair bacterial fitness.

A CRISPR nuclease writing module enables CAMERA 1

We designed a writing module that cleaves R1 but not R2 at the three-nucleotide region that differs between R1 and R2. This region was chosen to be proximal to the PAM to maximize the selectivity of the writing module (21) (Fig. 1A). The EGFP gene fragments from both plasmids were incubated in vitro with the Cas9:sgRNA complex. The functional EGFP gene amplified from plasmid R1, but not the mutated EGFP gene encoded by plasmid R2, was cleaved into two fragments (Fig. 1C). These results establish that the writing module can distinguish plasmids R1 and R2 and introduce double-stranded breaks selectively in R1.

Next we moved the system into live bacteria to test if we could translate an exogenous signal into a durable change in the DNA content of the cell. We placed a TetO promoter that is inducible with anhydrotetracycline (aTc) upstream of the Cas9 gene, and placed a constitutive Lac promoter upstream of the R1-targeting sgRNA in writing plasmids W1.0.1-W1.0.3 (Fig. 1D and fig. S2), forming the CAMERA 1.0 system. Bacteria containing CAMERA 1.0 with a R1:R2 ratio of 58:42 were used to test aTc-stimulated recording. After being cultured in the presence or absence of aTc for 3 h and 6 h, the bacteria were harvested and analyzed for their R1 content by high-throughput sequencing (HTS). In the absence of aTc, R1 content remained steady (59%) after 3 h and was only slightly lower (56%) after 6 h (Fig. 1D). This basal level of R1 consumption can be attributed to low-level transcription of the uninduced TetO promoter. In contrast, R1 content responded strongly to the presence of aTc and decreased to 21% in 3 h, and to 4% after 6 h (Fig. 1D). Collectively, these results suggest that CAMERA 1.0 can sensitively detect and record the presence of an exogenous small molecule and the duration of exposure in an analog format.

Recording of multiple stimuli using CAMERA 1 derivatives

To enable recording of more than one stimulus, we installed the LacO promoter, which is suppressed by LacI and activated by IPTG, upstream of the sgRNA to generate CAMERA 1.1 (Fig. 2A). Both aTc and IPTG are required to initiate recording in CAMERA 1.1. We chose a bacterial colony carrying CAMERA 1.1 with a starting R1 content of 77% and applied different inducer combinations for 3 h (Fig. 2A). As expected, R1:R2 ratio remained stable in the absence of stimuli or in the presence of 0.5 mM IPTG only. A slightly lower R1 content (to 70%) was observed when the bacteria were treated with 100 ng/mL aTc only (Fig. 2A), consistent with the known leakiness of the LacO promoter in the absence of IPTG (22). Importantly, R1 content decreased to 37% when bacteria were cultured in the presence of both aTc and IPTG (Fig. 2A), indicating that both stimuli are required to promote substantial R1:R2 ratio changes, recapitulating an “AND” Boolean logic gate (23).

Fig. 2. Multi-event recording and resetting of CAMERA 1 systems.

(A) Construction of a “AND” Boolean logic gate using CAMERA 1.1. Both IPTG and aTc are required for initiation of the recording process. (B) Analog recording of IPTG concentration by CAMERA 1.1 as reported by EGFP fluorescence. (C) Repeated recording and erasing of CAMERA 1.2 by application of aTc and IPTG and kanamycin. S: starting state; E: erase (5-20 generations); R: record (5-10 generations). (D) Repeated recording and erasing of CAMERA 1.3 by inducing different writing complexes. (E) Dose-dependent recording and erasing using CAMERA 1.3. Values and error bars reflect mean and s.d. of three replicates.

One advantage of the CAMERA 1 design is that it records signals in an analog format that can capture more information than binary switches. To explore the analog recording capabilities of CAMERA 1.1, we treated the bacterial culture with different doses of IPTG ranging from 0 to 150 μM with a constant aTc input of 100 ng/mL for 3 h (Fig. 2B). The R1 content was followed by monitoring EGFP fluorescence and by DNA sequencing. EGFP expression was initiated by diluting the bacterial culture with fresh media lacking aTc or IPTG after the recording process was finished. As anticipated, the EGFP signal decreased as the concentration of IPTG increased, reflecting an increased depletion rate of R1, saturating at 30 μM of IPTG (Fig. 2B). Importantly, the relationship between EGFP loss and IPTG concentration at low dosages (≤ 10 μM) was predictable and linear (Fig. 2B), suggesting that R1:R2 ratio can be used to infer signal amplitude in a reliable manner. HTS of the bacterial culture confirmed these dose-dependent changes in R1:R2 ratio (fig. S3). Collectively, these findings establish that CAMERA 1.1 can record multiple stimuli of interest in an analog, dose-dependent, and durable manner.

Erasing and re-recording of CAMERA 1 systems

Memory devices are particularly versatile if they can be erased and rewritten as needed. Instead of using R1 and R2, the CAMERA 1.2 system contains two recording plasmids, R3 and R4, that each confer resistance to different antibiotics. Similar to R1, R3 can be targeted by a writing plasmid expressing Cas9 and an sgRNA to cause a shift in the R3:R4 ratio. To minimize the growth advantage between R3 and R4, we fused genes encoding two antibiotic resistance proteins, chloramphenicol acetyltransferase (Cat, which inactivates chloramphenicol), and aminoglycoside-3′-phosphotransferase (Aph3′, which targets kanamycin), and incorporated a single point mutation in either of the two domains. R3 expressed inactive Cat H195A (24) fused to wild-type Aph3′, while R4 expressed inactive Aph3′ D208A (25) fused to wild-type Cat (Fig. 2C). As both plasmids express two nearly identical proteins, their relative fitness cost in the absence of antibiotic should be minimal. In the presence of either antibiotic, R3 and R4 should confer different fitness benefits.

Bacteria containing a starting R3 content of 39% maintained a steady R3:R4 ratio in conditions lacking antibiotic and responded to the presence of chloramphenicol or kanamycin by shifting the plasmid ratio in a highly reproducible, dose-dependent manner favoring the plasmid with the corresponding functional resistance domain (fig. S4). These results indicate that the information stored in the ratio of R3 and R4 can be reset in either direction using exogenous small molecules. By successively exposing cells to media containing either kanamycin (to reset the R3:R4 ratio to a high level) or aTc + IPTG (to induce Cas9 + sgRNA production and cleave R3, lowering the R3:R4 ratio), we performed three successive rounds of erasing and recording using CAMERA 1.2, with strong response levels in each round (Fig. 2C), demonstrating that this system can be used repeatedly to record and erase exposure to stimuli.

We developed an alternative resetting mechanism in CAMERA 1.3 that is independent of antibiotic resistance by including a second sgRNA circuit. In addition to one guide RNA cassette (sgRNA1) present in writing plasmid W1.2 that targets R3, we incorporated a second guide RNA expression unit (sgRNA2) under the control of a rhamnose-inducible promoter (P_Rha) to generate writing plasmid W1.3. The Cas9:sgRNA2 complex targets plasmids R4. Similar to the recording process in which the expression of sgRNA1 controlled by IPTG results in the loss of R3, the transcription of sgRNA2, induced by rhamnose, should lead to the cleavage of R4, and thus restore plasmid R3 levels. Indeed, E. coli strain S1030 that carried 36% or 77% of R3 successfully went through multiple rounds of recording and erasing upon alternating exposure to rhamnose or IPTG (Fig. 2D, and fig S5). In addition, the strength of the stimulus (here, the concentration of rhamnose or IPTG) was reflected in the rate of R3:R4 change (Fig. 2E).

HTS analysis of the recording plasmids after the final round of resetting and recording reveal a minimal frequency (≤ 0.06%) of insertions and deletions (indels) (table S1), suggesting that Cas9-mediated DNA cleavage does not substantially induce random mutations in the plasmid compensation system in bacteria, and both the recording and erasing processes result in minimal loss of future recording or erasing function. Taken together, these results validate CAMERA 1.2 and 1.3 as rewritable, durable cellular memory devices with distinct resetting mechanisms.

Base editing mediates recording in CAMERA 2

We recently developed base editors, chimeric proteins consisting of a DNA base modification enzyme, a catalytically impaired Cas9 nickase, and a base excision repair inhibitor (15, 26–28). Base editors efficiently introduce single C•G to T•A mutations at guide RNA-programmed loci in a wide variety of eukaryotic cells and organisms (15, 16, 29–34). Predictable, durable point mutation of genomic or plasmid DNA by base editing has the potential to serve as an ideal information carrier in synthetic memory devices (Fig. 3A). To incorporate a base editor in CAMERA, we first characterized base editing in E. coli, as base editors have not been extensively used in prokaryotic cells. Since bacteria lack nick-directed mismatch repair exploited by the third-generation base editor (BE3), we used the second-generation base editor (BE2) that contains a cytidine deaminase fused to a catalytically dead Cas9 (dCas9), rather than to a Cas9 nickase, as the protein component of our writing complex (15).

Fig. 3. CAMERA 2 systems use base editing to record the amplitude and duration of exogenous signals.

(A) Schematic representation of CAMERA 2 systems. The writing plasmid expresses the writing complex consisting of BE2 and sgRNAs. The recording plasmid is targeted by the writing complex and generates memory in the form of C•G to T•A substitutions at guide RNA-specified loci. (B) Recording the concentration of aTc and the treatment duration in analog mode using CAMERA 2.0. (C) Recording the concentration of IPTG in the presence or absence of aTc and the treatment duration in analog mode using CAMERA 2.1. (D) The rate of base editing at position 166 of the EGFP gene in CAMERA 2.0 reflects the schedule of exposure to the inducer. (E) CAMERA 2.1 records the total time of exposure to IPTG, regardless of treatment pattern. (F) Recording four exogenous stimuli using CAMERA 2.4. The presence of each signal, individually or in combinations, was recorded by base editing at each of several specified positions in the EGFP gene. We constructed two mathematical models to simulate the behavior of CAMERA 2.4. A model that accounts for promoter leakage and competition between multiple guide RNAs for BE2 results in a more “digital” CAMERA 2.4 in which the absolute editing level at each position more readily reveals the presence or absence of the corresponding stimulus (fig. S7). Values and error bars reflect the mean and s.d. of three replicates.

In writing plasmid 2.0 (W2.0), BE2 expression is induced by aTc and sgRNA1 is constitutively transcribed (Fig. 3A). To test whether CAMERA 2.0, constructed using W2.0 and recording plasmid R1, can faithfully record the amplitude and duration of an exogenous signal, the bacterial culture was treated with aTc at different concentrations and diluted repeatedly to ensure constant expression of the writing complex. When complexed with BE2, sgRNA1 introduces a C•G to T•A mutation at position 166 of the EGFP gene in recording plasmid R1. As anticipated, base editing occurred in an analog mode and the total percentage of modified base increased with bacterial passage number in a highly linear and remarkably reproducible relationship (Fig. 3B). This observation indicates that base editing with BE2 in bacteria is robust and cumulative, reflecting the duration of exposure to the stimulus that induces expression of the writing complex (Fig. 3B). Moreover, the rate of editing can be controlled in a dose-dependent manner (Fig. 3B). By the end of the experiment (68 passages), 66% editing was observed with 200 ng/mL aTc and no significant decrease in editing rate was observed as the recording proceeded, suggesting that given enough time, base editing could approach 100% in bacteria (Fig. 3B).

Importantly, editing at the target locus accumulated at a slow but constant rate when aTc was present at a low concentration of 2 ng/mL (Fig. 3B). Under these low induction conditions, only 12% of the total recording range (G•C to A•T conversion at position 166 of the EGFP gene) was consumed by bacterial generation 68 (Fig. 3B), suggesting that CAMERA 2.0 can function as a molecular clock that records over hundreds of generations. Collectively, these findings establish CAMERA 2.0 is a highly responsive analog memory device that uses base editing to faithfully record the amplitude of an exogenous signal over a known time scale, or the duration of a signal of known strength, in the form of single nucleotide changes.

Recording of multiple stimuli using CAMERA 2 systems

We hypothesized that multiplexed recording could be achieved by CAMERA using multiple responsive guide RNA expression cassettes. To test this possibility, we constructed additional base editor writing plasmids W2.1, W2.2 and W2.3 by replacing the Lac promoter of the guide RNA in writing plasmid W2.0 with promoters regulated by IPTG, arabinose, and rhamnose, respectively, to generate devices CAMERA 2.1, 2.2, and 2.3 (Fig. 3C, and fig. S6). Similar to CAMERA 2.0, writing promoted by the BE2:sgRNA1 complex in CAMERA 2.1 occurred in a highly reproducible, predictable, and dose-dependent manner (Fig 3C). The leaky transcription of the TetO promoter enabled very slow but steady recording in the absence of aTc, whereas the recording space was consumed at a much faster speed in the presence of both IPTG and aTc (Fig. 3C).

To test whether the information recorded in CAMERA can be used to deduce the total exposure time of the device to a stimulus, we passed bacteria carrying CAMERA 2.0 for 40 generations and treated either the first 20 generations or the second 20 generations with 100 ng/mL aTc (Fig. 3D). The accumulation rate of editing at position 166 of the EGFP gene was strongly determined by exposure duration, and the presence or absence of aTc within a certain time window could be determined by comparing the editing rate of the sample with those of control samples that were always exposed to, or always shielded from, the stimulus (Fig. 3D). Similarly, bacteria carrying CAMERA 2.1 were treated with 0.5 mM IPTG for either the first half or the second half of the total incubation time (Fig. 3E). The editing rate strongly correlated with the presence of IPTG and the total accumulated editing frequencies in the two groups were nearly identical by the end of the experiment, suggesting that the information recorded by CAMERA 2.1 faithfully reflected the duration of exposure to the signal, regardless of when the exposure took place (Fig. 3E). Collectively, these observations suggest that the extent of base editing at a known time point can be used to deduce the dose of the stimulus, and that the stimulus duration can be calculated from the total base editing conversion if the stimulus dose is known.

The presence of both aTc and a second stimulus is required for CAMERA 2.1, 2.2, and 2.3 to initiate recording, a process that mimics the behavior of an “AND” gate. Indeed, in the absence of stimuli, CAMERA 2.2 showed no detectable activity, with ≤ 0.1% C•G to T•A editing at position 186 of the EGFP gene (fig. S6). Neither arabinose nor aTc by itself increased editing significantly (fig. S6). However, the presence of both inducers resulted in 9.0% C•G to T•A conversion after 24 h (fig. S6), suggesting that CAMERA 2.2 functions as a tightly regulated “AND” gate. Similarly, both rhamnose and aTc were required to initiate recording at position 186 of the EGFP gene by CAMERA 2.3 (fig. S6). We tested the recording efficiency at different concentrations of rhamnose in the present of 200 ng/mL aTc and confirmed that C•G to T•A conversion at position 195 correlated well with the dose of rhamnose (fig. S6), again demonstrating that signal intensity can be faithfully recorded and stored by CAMERA 2 systems.

One advantage of adapting CRISPR technologies to build synthetic memory devices is that multiple stimuli in theory can be recorded using multiple guide RNA units. To test whether CAMERA can simultaneously record multiple independent signals, we integrated all three small molecule-responsive guide RNA expression circuits from writing plasmids W2.1-2.3 into writing plasmid W2.4. Bacteria carrying CAMERA 2.4 were treated with different combinations of the four small-molecule inducers and indeed, editing at the designated EGFP positions could be used to predict the presence of the corresponding writing complexes and hence their corresponding stimuli (Fig. 3F and fig. S7). Thus the fidelity of the device is not compromised even in more complicated environments in which more than two stimuli are provided (Fig. 3F and fig. S7), suggesting that CAMERA 2 is a versatile and multiplexable memory device.

Recording event order using CAMERA 2 systems

Memory devices that are capable of recording the order of biological events are of great interest (3), as the order of changes in a cell’s environment or in the state of a cell can strongly determine cell fate (35). Murray and coworkers recently described a two-input temporal logic gate constructed using integrases to record the order and timing of inputs, but the limited number of possible output states (GFP, RFP, or neither) necessitated the sharing of the same output among five different combinations of ordered inputs, complicating the assignment of multiple cell states (36). We hypothesized that CAMERA 2 systems could record events that occur in a specific order using a different mechanism by overlapping two base editing targets such that base editing of DNA target 1 mediated by writing complex 1 (BE2:sgRNA5) is required before DNA target 2 can be recognized by writing complex 2 (BE2:sgRNA6). To test this possibility, we constructed CAMERA 2.5 in which the order of exposure to two small molecule inducers, arabinose and rhamnose, could be recorded (Fig. 4A). The three arabinose-induced C•G to T•A modifications resulting from base editing by writing complex 1 are located within target site 2 near its PAM. Rhamnose-induced sgRNA6 targets target site 2 only after modification by writing complex 1, but should not edit this site before base editing by writing complex 1 has taken place (Fig. 4A). Thus base editing at position 216 should only be initiated if rhamnose (stimulus 2) was provided after arabinose (stimulus 1), but not if the order of stimuli is reversed.

Fig. 4. CAMERA 2 systems can record the order of stimuli and a wide range of environmental signals.

(A) Schematic representation of CAMERA 2.5 that records stimuli in an order-dependent manner. (B) CAMERA 2.6 records the presence of arabinose at position 205-207 in format of C•G to T•A mutations. (C) The ratio of base editing at position 216:position 129 in CAMERA 2.5 indicates the order of exposure to two stimuli. A position 216:129 base editing ratio above 0.1 was only observed when the bacteria were treated first with arabinose and then with rhamnose, but not if arabinose exposure follows rhamnose exposure. (D) Phage infection recording by CAMERA 2.6. (E) Light exposure recording with CAMERA 2.7 in bulk culture and in small numbers of cells. Light exposure duration can be recorded faithfully in bulk culture as well as in samples of only 100 or 10 cells. Values and error bars in bar graphs reflect the mean and s.d. of three replicates. Dots and error bars in dot plots in (E) represent the mean and s.d. of 15 replicates of randomly sorted sets of many, 100, or 10 cells.

By using an additional target site of sgRNA 6 spanning positions 116-135 of a modified EGFP gene, CAMERA 2.5 is further equipped with the ability to independently record two stimuli (Fig. 4A). While editing at positions 205-207 and 129 record exposure to arabinose and rhamnose (Fig. 4B and fig. S8), respectively, the ratio of base editing at position 216 compared to position 129, both promoted by writing complex 2, reflects the order of application of the two stimuli (Fig. 4C). The activating treatment order of arabinose followed by rhamnose resulted in a position 216:129 base editing ratio of 0.54. When the treatment order was reversed such that rhamnose exposure precedes arabinose, this ratio was 6.8-fold lower (0.08) (Fig. 4C). Together, these results indicate that CAMERA 2.5 can record cellular events in a strongly order-dependent manner.

Recording of phage infection and light using CAMERA 2.0 derivatives

We applied the CAMERA 2.0 architecture in bacteria to sense viral infection of host cells by bacteriophage, and exposure to light. A phage shock promoter (PSP) driving sgRNA1 transcription was included in CAMERA 2.6 (Fig. 4D) (37, 38). Without phage infection, 9% base editing was observed at EGFP position 166 (Fig. 4D), consistent with previous reports of background transcriptional activity of PSP in the absence of phage (39). Base editing at position 166 increased 4.7-fold to 42% following infection with phage (Fig. 4D). Similarly, using a light-responsive expression system based on light-inhibited expression of the cI repressor gene, CAMERA 2.7 could record the presence of light with a 59-fold increase in recording site editing efficiency (Fig. 4E) (40). These results collectively demonstrate that the CAMERA 2 is capable of recording as single-nucleotide changes in bacterial DNA a wide range of signals, including exposure to antibiotics, nutrients, viruses, and light.

In principle, the recording process carried out by CAMERA systems should not require a large population of cells because the recording plasmid is present in hundreds of copies in each cell. To test the possibility of recording and reading CAMERA data in small cell populations, we characterized how light exposure was recorded by CAMERA 2.7 in a handful of cells as well as at the single-cell level (Fig. 4E and fig. S9). As expected, CAMERA 2.7 reliably recorded bacterial exposure to light in bulk cultures, with editing at EGFP position 166 in ~10⁶ cells increasing in a linear fashion with light exposure duration (from 1.2% to 57% editing over three days, Fig. 4E). Importantly, reliable recording and signal readout were also achieved using only 100-cell or 10-cell samples throughout the three-day recording process, although larger variations were observed with fewer cells, as expected (Fig. 4E). Even measuring 15 single-cell signals yielded similar average light duration-dependent editing efficiencies as those from bulk cultures (fig. S9). These data demonstrate that CAMERA can support analog-like recording even in small populations of only 10-100 cells.

CAMERA 2m systems record cell states in mammalian cells

Finally, we tested a CAMERA 2 variant in human HEK293T cells and chose an established human safe harbor gene, CCR5 (41), as the recording locus (Fig. 5A). We designed three individual sgRNAs that target different regions of the CCR5 gene (CAMERA 2m.0, Fig. 5B and fig. S10). Total C•G to T•A editing of 37%, 46%, or 19% was obtained at target positions A, B, or C of the CCR5 gene when using corresponding guide RNAs A, B, or C with BE3 (fig. S10). The C•G to T•A conversion frequency at each site increased a minimum of 270-fold compared with controls lacking the corresponding guide RNA. Robust multiplexed recording was also achieved using the three sgRNAs in all possible combinations and less than 0.07% editing was observed at any site for which the corresponding guide RNA was not supplied, suggesting that base editing cross-talk between guide RNAs is minimal in these human cells.

Fig. 5. CAMERA 2m recordings in mammalian cells.

(A) Schematic representation of CAMERA 2m in mammalian cells. (B) CAMERA 2m.0 functions in a multiplexed manner in mammalian cells by targeting the human safe harbor gene CCR5. (C) CAMERA 2m.1 records the presence of doxycycline in HEK293T cells through a doxycycline-controlled transcriptional activator. (D) CAMERA 2m.2 records the presence of doxycycline and IPTG in a multiplexed manner. Expression of sgRNA A and sgRNA B is repressed by LacI and TetR in the absence of stimuli and can be turned on at the addition of IPTG and doxycycline, respectively. (E) CAMERA 2m.3 responds to Wnt signaling and records the presence of a Wnt signaling stimulus at a target genomic safe harbor locus. Values and error bars reflect mean editing and the s.d. of three replicates.

By placing BE3 expression under doxycycline-induced control, the presence of the drug was recorded in the CCR5 loci with a base editing frequency 60-fold higher than in cells that never encountered doxycycline (CAMERA 2m.1, Fig. 5C). Further, by placing sgRNA expression under TetR and LacI suppressed promoters, CAMERA 2m.1 recorded the presence of both doxycycline and IPTG at different positions in the CCR5 loci (CAMERA 2m.2, Fig. 5D), confirming that CAMERA 2m can be multiplexed to record combinations of inputs in human cells.

The Wnt signaling pathway plays a crucial role in embryonic development and aberrant Wnt signaling is associated with a variety of diseases in humans (42). We sought to record Wnt signaling using CAMERA 2m in human cells. To achieve this goal, we placed the expression of BE3 under a (LEF-TCF)₇ promoter (43) that responds to Wnt signaling to initiate downstream gene expression in CAMERA 2m.3. Cells transfected with CAMERA 2m.3 were treated with LiCl, a GSK3 inhibitor that has been demonstrated to activate Wnt signaling (Fig. 5E) (44). We included a (LEF-TCF)₇-BE3-P2A-Luc construct that expresses a firefly luciferase protein together with BE3 so that we could monitor Wnt simultaneously by luminescence and by high-throughput sequencing of the CCR5 recording locus. As expected, cells transfected with (LEF-TCF)₇-BE3-P2A-Luc exhibited a large (140-fold) increase in Wnt signaling-driven luciferase expression upon LiCl treatment (fig. S11). This increase in Wnt signaling was permanently recorded by a 53-fold increase in base editing at the CCR5 locus upon LiCl treatment (Fig. 5E). These results demonstrate that Wnt signaling, a major endogenous mammalian signaling pathway, can be recorded by CAMERA 2m in human cells.

Discussion

We developed synthetic memory devices that record events of interest in live cells using two distinct CRISPR-mediated DNA modification mechanisms: Cas9 nuclease-catalyzed double-stranded DNA cleavage, and base editor-mediated point mutation. CAMERA records the amplitude of stimuli over a known time scale, or the duration of stimuli with a known amplitude, permanently in DNA of live cells. The analog nature of both recording systems allows the continuous monitoring of signals of interest and provides more information compared to canonical digital memory devices.

In CAMERA 1 systems (table S2), information is recorded in a form of plasmid R1:R2 ratio. Because R1 but not R2 expresses a functional fluorescent protein, information stored in CAMERA 1 systems can be read out transiently by monitoring post-recording cellular fluorescence in addition to the permanent readout by HTS. During the development of CAMERA 1, we decreased the RBS strength of Cas9 by four orders of magnitude (see Supplementary Text) to slow down the recording process, indicating that these systems can respond quickly and are highly sensitive. This exceptional sensitivity may enable recording of very weak environmental signals that would otherwise be difficult to detect using other methods. Endy and coworkers achieved resetting using recombinase-based synthetic memory devices (6). In this work, we developed two different strategies for CAMERA resetting that also enable repeated cycles of erasing and rewriting.

CAMERA 2 systems (table S2) translate stimuli of interest into single-nucleotide modifications. The devices can be multiplexed by stacking multiple responsive sgRNA units and we demonstrated that four exogenous signals could be recorded using CAMERA 2.4. Importantly, using a racheted overlapping protospacer design, CAMERA 2.5 can record events in an order-dependent manner, a capability that is difficult to achieve using other synthetic memory devices. By including environment-responsive circuits, virus infection and light exposure have also been faithfully recorded using CAMERA 2.6 and 2.7. The cI repressor translation mechanism integrated in CAMERA 2.7 is particularly versatile and can be used to translate a variety of environmental signals. Similar repression systems have been applied to monitor genetic (mutational inactivation of the cI gene) and epigenetic (proteolytic inactivation of cI) changes in E. coli by Radman and coworkers (45). We also demonstrated that CAMERA recording to high-copy plasmids maintains its reliability even in samples containing only 10-100 cells. The mammalian cell compatibility of base editing enables CAMERA 2m systems to function in human cells, including its use to record both exposure to external stimuli and flux through an endogenous signaling pathway.

Incorporating the recently developed ABE that mediates A•T to G•C base editing (27) could expand the versatility of CAMERA 2 systems by adding an additional dimension of recording that can directly reverse the edits introduced by BE3. CRISPR technology has been applied in mammalian cells for molecular recording of exogenous signals and mapping cell lineage using genomically integrated circuits (46–48). The Cas1/Cas2 DNA sequence capture system has been applied for analog recording in bacteria using large populations of cells (49). Wang and coworkers further developed this recording mechanism and reported an elegant “biological tape recorder” using Cas1/Cas2 and copy-inducible plasmids to follow exogenous signals in a multiplexed format over time while this work was under review (50). Compared to these devices, CAMERA systems are not dependent on genomic integration of barcoded scratchpads or protospacer arrays that could result in unpredictable changes in the genome and undesired cellular perturbations.

As a synthetic memory device that uses novel recording mechanisms, CAMERA has its own limitations. Due to the sensitivity of the writing process, background recording in the absence of the stimulus can be observed when using less tightly controlled induction systems. As a result, the sensitivity of CAMERA may need to be tuned for different applications. For example, a weaker RBS can be used to express base editors and Cas9 nucleases when background recording is undesired. Moreover, when recording to genomic loci, CAMERA 2 systems cannot achieve single-cell read-out of analog information and will typically require the analysis of a population of cells.

We only demonstrated the construction of a simple AND logic gate in this work. Additional research is needed to explore more complicated logic gates (23) using CAMERA. The shortage of orthogonal inducible expression cassettes also limits the application of CAMERA in more complex setups. This limitation might be addressed as more inducible transcriptional and translational regulation circuits are developed.

These limitations notwithstanding, the use of base editors in CAMERA 2 systems minimizes stochastic indels and translocations that arise from double-stranded DNA breaks. The capability of recording many endogenous signaling pathways of interest in a minimally perturbative and highly multiplexable manner offer substantial benefits for mammalian cell recording. The small sample size of 10-100 cells that CAMERA requires to achieve faithful analog recording in bacteria may prove especially useful for applications in which limited cellular material is available. We envision CAMERA being used for applications such as recording the presence of low-abundance extracellular and intracellular signals, mapping the linage of specific cell types, and constructing complex cell-state maps.

Materials and Methods

Cloning and plasmids

Oligonucleotides were ordered from Integrated DNA Technologies. PCR fragments for plasmid construction were amplified using PhuU polymerase (ThermoFisher Scientific) and assembled by USER enzyme mix (New England Biolabs) according to the manufacturer’s instructions. All DNA cloning was performed with NEB Turbo cells (New England Biolabs). Plasmids used in this work (see table S3 for plasmid design specifics) are available from Addgene. Primers used for high-throughput sequencing are listed in table S6.

Strains and chemicals

All bacterial CAMERA devices developed in this work were tested with E. coli strain S1030 (20) with the exception of CAMERA 2.6, which was characterized in E. coli strain S2063. The complete genotypes of S1030 and S2063 are listed in table S4. Unless otherwise noted, antibiotics were used at the following concentrations: carbenicillin (100 mg/ml), kanamycin (50 mg/ml), chloramphenicol (25 mg/ml), and spectinomycin (100 mg/ml). All chemicals were purchased from Sigma-Aldrich and Fisher Scientific.

Characterization of CAMERA 1.1 in E. coli S1030

E. coli S1030 were transformed with a mixture of 500 ng R1, 500 ng R2 and 100 ng W1.1 and plated on LB agar containing carbenicillin and spectinomycin. A total of eight colonies were picked, grown to dense cultures and analyzed for their R1 content. The bacterial culture carrying CAMERA 1.1 with 77% R1 and 23% R2 was selected for further test and split into three individual cultures. The bacterial cultures were inoculated 1:500 (v/v) into fresh LB media containing 1) no inducer, 2) 100 ng/mL aTc, 3) 500 μM IPTG, and 4) 100 ng/mL aTc and 500 μM IPTG. The treated bacteria were allowed to grow at 37 °C with shaking for 3 h and the R1/R2 ratio was analyzed by amplifying the EGFP fragment and sequencing using HTS.

To characterize the analog behavior of CAMERA 1.1, the starting cultures were inoculated 1:100 (v/v) into fresh LB media containing 0, 2, 5, 10, 20, 30, 40, 60, 80, 100, or 150 μM IPTG in the presence of 50 ng/mL aTc. The treated bacteria were allowed to grow at 37 °C with shaking for 4 h and the inducers were removed by diluting the culture in a 1:250 ratio with fresh LB and culturing overnight. The resulting R1/R2 ratio in the bacterial culture was analyzed by amplifying the EGFP gene and sequencing in a high throughput manner. To induce the EGFP expression as a transient readout, the bacterial culture was diluted again in a 1:125 ratio with fresh LB containing 0.25 mM arabinose. EGFP fluorescence was measured after 4 h of induction using a TECAN Infinite M1000 Pro plate reader with excitation/emission wavelength set to 485/530 nm.

Recording and erasing of CAMERA 1.2

E. coli S1030 were transformed with 500 ng of R3 and 500 ng of R4. The transformed bacteria were plated on LB agar containing 50 μg/mL kanamycin and 25 μg/mL chloramphenicol to select for the presence of both plasmids. A total of eight colonies were picked, grown in fresh LB and analyzed for their R3 content. The bacteria containing 38% R3 and 62% R4 were selected to test whether antibiotic treatment could promote the R3:R4 ratio shift. The selected bacterial culture was split into two individual cultures and diluted 1:30 into fresh LB media containing 0.4, 0.8, 1.2, or 1.6 mg/mL kanamycin or 100 μg/mL chloramphenicol. The process was repeated one more time before the resulting bacteria were analyzed for their R3 content.

To perform recording and device resetting using CAMERA 1.2, E. coli S1030 were transformed with 500 ng R3, 250 ng R4 and 100 ng W1.1 and plated on LB agar containing 25 μg/mL kanamycin, 10 μg/mL chloramphenicol and 100 μg/mL spectinomycin. A bacterial colony carrying CAMERA 1.2 with 36% of R3 and 64% of R4 was selected for further characterization and split into three independent cultures. To initiate the recording process, the bacterial culture was inoculated 1:30 into fresh LB media containing 50 ng/mL aTc and 100 μM IPTG, whereas to reset the device, a similar inoculation protocol was performed with fresh LB media containing 0.8 mg/mL kanamycin. The inoculated culture was allowed to grow at 37 °C with shaking for 12-24 h to saturation. The process was repeated until a desired R3:R4 ratio was obtained. The R3 content was characterized by HTS analysis of the EGFP fragment amplified from the bacterial culture.

Characterization of CAMERAs 2.0 and 2.1 in E. coli S1030

E. coli S1030 were transformed with R1 and W2.0 and plated on LB agar containing carbenicillin and spectinomycin. A single colony was picked and cultured at 37 °C with shaking to obtain a dense culture as the starting material of the recording experiments. The split bacterial cultures were diluted 500- or 1,000-fold into fresh LB media containing 0, 2, 20, or 200 ng/mL aTc and grown in a 96-deep-well plate at 37 °C with shaking for 16-24 h before being diluted again. The process was repeated until 68 generations of bacteria were produced. Editing promoted by the BE2:sgRNAa complex was characterized by amplifying the EGFP gene from the bacterial culture and analyzing the amplicon using HTS.

E. coli S1030 carrying CAMERA 2.1 were treated with 1) no inducer, 2) 1 mM IPTG, 3) 200 ng/mL aTc and 1, 0.1, or 0.01 mM IPTG. Similar culturing and characterizing protocol was adapted as that was used for CAMERA 2.0.

To confirm that CAMERA 2.0 could record the present duration of a stimulus, E. coli S1030 cultures carrying CAMERA 2.0 were diluted 1,000-fold into fresh LB media and treated with or without 100 ng/mL aTc. The bacteria were grown in a 24-deep-well plate at 37 °C with shaking for 12 h and diluted 1,000-fold again into fresh LB containing the same concentrations of aTc. In the 3rd dilution, bacteria that had not encountered the inducer were split into fresh LB media with or without 100 ng/mL aTc. The process was repeated once in the 4th dilution. Similarly, bacteria that had been treated with aTc were split and treated with or without aTc from generation 20 to 40. E. coli S1030 carrying CAMERA 2.1 were tested for IPTG sensing using a similar set up.

Recording in the genomic safe harbor gene CCR5 in human cells

HEK293T cells (GenTarget Inc.) were cultured in 48-well plates (collagen-coated, ∼40,000 cells seeded per well) in DMEM plus GlutaMAX (Life Technologies) with 10% FBS. Transfection was performed 24 h after plating when cells reached 60-70% confluence. To initiate recording in the human safe harbor gene CCR5, 800 ng of BE3 plasmid and 40 ng of guide RNA plasmid (CAMERA 2m.0, the guide RNA sequences are listed in table S5) were transfected in each well using 1.2 μl Lipofectamine 2000 (Life Technologies) following the manufacturer’s protocol. To multiplex recording using multiple guide RNAs, each guide RNA plasmid was applied at a dose of 40 ng together with 800 ng of BE3 plasmid. The transfected cells were incubated for additional 3 days before being harvested for genomic DNA extraction. Base editing was quantified by amplifying the CCR5 gene fragment from genomic DNA by PCR and analyzing by HTS.

Recording Wnt signaling in the CCR5 loci of human cells

To enable CAMERA 2m to record Wnt signaling, we installed a (TCF/LEF)₇ promoter upstream BE3 and BE3-P2A-Luc to generate CAMERA 2m.3 ((TCF/LEF)₇-BE3 and ((TCF/LEF)₇-BE3-P2A-Luc). TOPFlash ((TCF/LEF)₇-Luc) (43) was used as a transient readout of Wnt signaling. A control plasmid that encodes the Renilla luciferase was included to normalize transfection efficiency for luminescence readout.

HEK293T cells were cultured in 96-well plates (collagen-coated, ∼20,000 cells seeded per well) in DMEM plus GlutaMAX (Life Technologies) with 10% FBS. Transfection was performed 24 h after plating when cells reached 60-70% confluence. CAMERA 2m.3 were prepared in 5 μL of reduced serum media (Opti-MEM, Life Technologies) with 200 ng of ((TCF/LEF)₇-BE3 or ((TCF/LEF)₇-BE3-P2A-Luc plasmids, 20 ng of U6-sgRNA A plasmid and 10 ng of Renilla luciferase plasmid and transfected using 0.5 μl Lipofectamine 2000. TOP-Flash plasmid (200 ng) was transfected using a similar setup without including the guide RNA plasmid. A stock solution of 1 M LiCl was prepared in ddH₂O and added to the media to a final concentration of 50 mM 10 h after transfection.

Firefly luciferase and Renilla luciferase activities were measured 24 h after LiCl treatment. Luciferase substrates were purchased from Promega. To characterize Wnt-stimulated base editing, the transfected cells were incubated for 3 days before being harvested for genomic DNA extraction. Base editing was quantified by amplifying the CCR5 gene fragment from genomic DNA by PCR and analyzing by HTS.

One sentence summary.

Base editors and CRISPR nucleases were used to create “cell data recorders” that enable durable, analog recording of stimuli and cell states in bacterial and mammalian cells including exposure to antibiotics, nutrients, viruses, and light, as well as Wnt signaling.