Introduction

Sensing and integrating various transmembrane signals is a key aspect of cellular decision making. For example, activation of CD8+ cells requires co-activation of TCR and CD28 molecules, meanwhile, this activation can be inhibited by the PD-1 pathway(1). By abstracting this biological process, we can get: the activation of CD8+ cell = activated TCR AND (activated CD28 NIMPLY activated PD-1) (Fig. 1a). Programming cells with predictable complex transmembrane signal inputs – customized intracellular signal outputs logic relationships are significant for expanding the widespread applications of mammalian cells, such as cellular immunotherapy(2-5), tissue patterning(6, 7) (Fig. 1b).

Figure 1: The logical integration of complex transmembrane signals has crucial biological implications.
(a) Immune cells recognize target cells by integration of multiple transmembrane logic signals. Immune cells need to be co-stimulated by TCR and CD28 to be activated. Tumor cells can induce PD-1 activation of immune cells by expressing PD-L1 molecules. PD-1 can, in turn, interrupt the progression of the immune response by terminating the co-stimulatory signal of CD28. Thus, tumor cells cannot be recognized by immune cells and cause immune escape. (b) Engineering the cells to perform logical operations on multiple transmembrane signals to produce a variety of customizable outputs could enable cells to gain novel application potential.

The transmembrane signal transduction is obstructed by the cell membrane, a unique feature that makes designing multiple transmembrane signal sensing and integrating systems a huge challenge. As we have seen, synthetic biologists have achieved great success in the integration of non-transmembrane signals. Although researchers already built partially(8-12) or completely(13-19) binary logic gates in both prokaryotes(8, 9, 11, 12, 15, 17, 18) and eukaryotes(10, 13, 14, 16, 19), all of these mentioned design limited by they can only integrate signals that are confined within the cell membrane(11, 12, 16, 19), or are the small molecules which could freely penetrate the cell membrane(8-10, 13-15, 17, 18). Therefore, developing a complete system that enables it to sense and integrate complex transmembrane signals is a current core challenge for synthetic biology.

We designed the ENABLE (Engineered, Across-membrane, Binary Logic in Eukaryotes) system and achieved the first complete transmembrane binary Boolean logic in mammals. The three-layer modular design (Receptor, Amplifier, and Combiner) of the ENABLE system gives the system great expandability, which not only promising a host of application potentials but also provides a design paradigm for the future transmembrane logic decision system. At the same time, the ENABLE system is not only capable of running single-cell-based centralized logic calculations but also enables distributed logic calculations based on single-cell-single-cell contacts through spatial wiring by the surface antigens. These characteristics give the ENABLE system the possibility to perform the sophisticated cellular computation.

Result

Engineer SynNotch enables it to receive extracellular signals and generate orthogonal intracellular signals

In order to be able to implement a custom multiplexed transmembrane signal input/output relationship, the first condition is that engineering modular receptor to enable it to recognize extracellular signals and transduce them into customized intracellular signals. To this end, a variety of techniques have been developed, such as Tango(20), CAR(21), GEMS(22), MESA(23-25), SynNotch(4-7), etc. High programmability of the extracellular and intracellular domains of SynNotch, as well as adaptation to contact-dependent signaling, fully meets our needs. Thus we ultimately apply SynNtoch technology as a receiving port for extracellular signals of transmembrane logic gates.

Synthetic Notch (SynNotch) has a minimal core regulatory region of Notch receptor(6). By wiring it to a chimeric extracellular domain (such as a single-chain antibody) and a chimeric intracellular domain (such as a transcription factor), SynNotch can recognize customized surface ligand signals and produces customized intracellular outputs. In the mechanical force activation model(26), when SynNotch recognizes a homologous ligand on an adjacent cell, its minimal core regulatory region will undergo a series of cleavage and eventually release the intracellular domain into the nucleus, and drives the expression of the user-specified downstream circuits (Fig. 2a).

Previous work by Morsut et al. showed that the natural surface antigen CD19 and the non-natural surface antigen EGFP have good orthogonality(7). We applied this and designed two surface antigens, surCD19 and surEGFP. By immunostaining, we were able to clearly detect surCD19 and surEGFP expressed on 293T cells (Fig. 2b). Thus these two surface antigens can serve as ideal dual antigens for our transmembrane binary Boolean logic inputs.

We selected αCD19 (anti-CD19)(6), LaG17 (anti-EGFP with low affinity), LaG16-2 (anti-EGFP with high affinity)(27) as the extracellular domain of SynNotch, mouse Notch1 core regulatory region as the core transmembrane region of SynNotch. Two transcription factors, tTAA (TetR-VP48) and GV2 (Gal4DBD-VP64), which are orthogonal to each other, are used as the intracellular domain of SynNotch (Fig. 2c). We found that the combination of different extracellular domains and intracellular domains has differentiated activation characteristics (Fig. 2d) in transient transfection experiments. In all combinations, αCD19-mN1c-GV2 was not able to be activated efficiently. LaG16-2-mN1c-GV2 and αCD19-mN1c-tTAA have relatively high activation and low background expression. LaG16-2 was able to be efficiently activated only when paired with surEGFP but not surCD19. surCD19 has a similar property (Fig. 2e). The excellent orthogonality between LaG16-2 and αCD19 ensures that we will not undergo signal crosstalk when using dual surface antigens as signal inputs in the future. In the following experiments, we used the LaG16-2-mN1c-GV2 and αCD19-mN1c-tTAA which has good performance, as actual transmembrane signal transduction elements. We used mCherry-tagged LaG16-2-mN1c-GV2 and EGFP-tagged αCD19-mN1c-tTAA to construct a stable cell line with co-expressing of dual SynNtochs for subsequent experiments (Fig. S).

Meanwhile, as we mentioned above, some types of SynNotch have high background expression. This phenomenon aroused our great interest. So we designed a systematic experiment to carry out the fundamental mechanism of SynNotch activation. We designed a systematic experiment to fundamentally study the mechanism of SynNotch activation and hope to optimize it. We specifically covered the details of this section in a SynNotch oriented study in this project.

Figure 2: Engineering SynNotch can receive orthogonal extracellular signals and generate orthogonal intracellular signals.
(a) SynNotch has customizable extracellular and intracellular domains, as well as a core regulatory region. After the extracellular domain of the single-chain antibody recognizes the corresponding surface antigen, the core regulatory region undergoes two cleavages to finally release the intracellular segment of the transcription factor into the nucleus to regulate the downstream circuits. (b) Use surEGFP and surCD19 as ideal antigens. Top left, surEGFP schematic. Green, surEGFP autofluorescence; purple, surEGFP was stained with anti-GFP Rabbit and anti-Rb AF647. Purple shows a clear contour line on the schematic. Bottom left, surCD19 schematic. The extracellular region of surCD19 was labeled with HA tag and stained with anti-HA AF488. Scale bar, 5 μm. (c) Schematic diagram of the SynNotchs we constructed. Different extracellular scFvs and different intracellular transcription factors were used to serve as extracellular or intracellular domains of SynNotch. (d) SynNotch with different extracellular domain and intracellular domains has differentiated expression. SynNotch loaded Receiver cells were stimulated with surAg loaded Sender cells or Mock cells carrying no surAg. the percentage of the cells which highly expressing EGFP (EGFP++%) was analyzed by flow cytometry after 24 hours of co-culture. The error bar indicates SD (n=3). (e) LaG16-2 and αCD19 were able to recognize surEGFP and surCD19 orthogonally. LaG16-2-mN1c-tTAA or αCD19-mN1c-tTAA was activated by surEGFP or surCD19, respectively, and TRE3GV-d2EGFP was used as a downstream reporter. The error bar indicates SD (n=3).

Design three-layer logic processing paradigm for dual transmembrane signals.

The expression of membrane proteins on the cell membrane is limited. For example, cells stably expressing chimeric antigen receptor (CAR) on the membrane are expected to have an expression level in the order of 50,000 molecules/cell(28). It can be compared that in a transient experiment using cationic liposomes, the number of plasmids entering each cell’s nucleus can be as high as 50,000 molecules(29). Therefore, it is critical to design a circuit so that it can accommodate the limited transmembrane signal. Model analysis shows that the use of transcription cascade can effectively amplify the signal.

Based on this, we designed a three-layer transmembrane logic circuit paradigm and limited the transcription units inside the cell to three. The first layer is the "Receptor" layer on the cell membrane, the function of which is to transduce the dual orthogonal antigen signal into the cell by SynNotch to the signal carried by two activating-form transcription factors. The second layer is the "Amplifier" layer within the cell membrane. It contains up to two transcription units. TRE3GV (7xTetO-minCMV) and URE2G (4xUAS-minCMV) are paired activating-form promoters of tTAA and GV2, acting as two channels, receiving signals from two SynNotchs of the "Receptor" layer, respectively. The "Amplifier" layer amplifies the signal from the cell membrane and converts it into a composite signal type carried by engineered zinc finger-based transcription factors. The last layer is the "Combiner" layer within the cell membrane. This layer contains only one transcription unit that integrates the composite signals from the two channels of the "Amplifier" layer and performs the final logical decision output (Fig. 3).

In order to be able to receive dual transmembrane signals and produce 16 complete binary Boolean logic, we designed a set of interactive grammar for the interaction between the "Amplifier" layer and the "Combiner" layer within the cell membrane. This grammar consists mainly of the following elements, a transcription system based on the activating-form, silencing-from or NIMPLY-form promoters (the three are collectively referred to as a synthetic transcription factor-promoter pairs based transcription system); intein-based protein in vivo fusion systems, proteolytic enzyme-based protein in vivo destruction systems (the two are collectively referred to as a protein fusion/destruction-based transcription factor modification system).

Figure 3: The multiple transmembrane signals are processed using a three-layer logic processing paradigm.
On the first layer, the "Receptor" layer, SynNotch receives extracellular signals and then transduces into intracellular signals. The second layer, the "Amplifier" layer, contains two channels, receives signals from the "Receptor" layer, amplifies them and converts them into a variety of signal types based on transcription factors. The third layer, the "Combiner" layer, receives the dual-channel signal from the Amplifier layer and integrates to produce a logical output.

Construct a synthetic transcription factor-promoter pairs based transcription system.

When constructing an artificial gene circuit, it is necessary to consider the interference of the host's endogenous signal. By using a cross-species promoter or synthetic promoter, and ensuring the designed gene circuit to be orthogonal to the host endogenous gene is the key to maintaining system robustness(30). We constructed three mammalian adapted DNA binding domains (DBD) by by using synthetic Zinc finger (SynZF) (31, 32) of ZF21.16, ZF42.10, and ZF43.8. We improved the design paradigm about synthetic transcription factors (SynTF) – synthetic promoter (SynPro) pairs proposed by iGEM 2017 Fudan, constructed three mammalian adapted activating-form transcription factor (aTF) – activating-form promoter (aPro) pairs and three silencing-form transcription factor (sTF) – silencing-form promoter (sPro) pairs to serve as candidate downstream elements of SynNotch. Using a dual-fluorescence dual-plasmid test system, we confirmed that ZF21.16-, ZF42.10-, ZF43.8-VP64 have good activation characteristics (Fig. 4b), ZF21.16-, ZF42.10 -, ZF43.8-KRAB has good inhibition properties (Fig. 4e). At the same time, ZF21.16, ZF42.10, ZF43.8 are orthogonal to each other and to both TetR and Gal4, which are the DBD of selected aTF for SynNotch. By using an identical intermediate transcription factor as a common wiring molecule for the two upstream transcription factors, we can generate an implicit OR gate (Fig. 4f, g) and an implicit NOR gate (Fig. 4h, i).

In addition to traditional activating- and silencing-form promoters, we have also designed a novel NIMPLY-type promoter by adding multiple response elements corresponding to the activating-form transcription factor (aTF REs) and response elements corresponding to the silencing-form transcription factors (sTF REs) at the 5' and 3' ends of minCMV, respectively (Fig. 4j). Thus, the sTF can simultaneously apply (1) transcriptional inhibitors recruited by KRAB can inhibit promoter expression(33); (2) due to RE Located downstream of the promoter, when a sTF binds to the sTF REs, it can cause steric hindrance and enhance the ability of transcriptional inhibition by inhibiting the forward movement of RNA polymerase. This type of promoter exhibits the logical selectivity of NIMPLY type for aTF and sTF. NIMPLY-form promoter can be expressed only in the presence of an aTF and the absence of a sTF (Fig. 4j). 8xZF21.16-minCMV-2xZF43.8 can only be activated in the presence of ZF21.16-VP64 and in the absence of ZF43.8-KRAB, while 8xZF43.8-minCMV-2xZF21.16 can only be activated in the presence of ZF43.8-VP64 and in the absence of ZF21.16-KRAB (Fig. 4k). Through gradient transfection experiments, we also confirmed that the NIMPLY-type promoter exhibits the logic of NIMPLY in the ratio of different aTFs and sTFs (Fig. 4m). In addition, as previously reported cases of aPros(32) and sPros(16) with intensity tunability, NIMPLY-form promoter can also be tuned by using a different number of aRE sites corresponding to the. When the number of sTF REs is fixed and the number of aTF REs is increased, the maximum activation value is on the rise (Fig. 4l).

Figure 4: synthetic transcription factor-promoter pairs based transcription system.
(a), (d), (j). Schematic diagram of the working mechanism of the IDENTITY-, NOT-, NIMPLY-form promoter. An aTF/sTF is constructed by wiring the DNA binding domain (DBD) to a transcriptional activation domain (AD, such as VP64) or a transcriptional silence domain (SD, such as KRAB) via a linker. The IDENTITY-form promoter is just an activating-form promoter. Since it is expressed only in the presence of an aTF, the aPro is expressed as IDENTITY aTF logic. aPro’s structure contains multiple aTF corresponding response elements (aTF REs) inserted in the 5’ terminus of a minimal promoter domain (minPro, such as minimal CMV promoter). The NOT-form promoter is just an sPro. Since it is expressed only in the absence of an sTF, the sPro is expressed as NOT sTF logic. sPro’ structure contains multiple sTF corresponding response elements (sTF REs) inserted in the 3’ terminus of a constitutive promoter (conPro, such as CMV promoter). (b). synthetic aPros can be activated well in the presence of its corresponding aTFs (n = 3, error bar, SD). (c) Orthogonality testing of different DBDs. (e). Synthetic sPro are sufficiently inhibited in the presence of their corresponding sTFs (n = 3, error bar, SD). The dashed line indicates the intensity of expression of the CMV promoter under the same test conditions. Relative to inserting sTF REs in the 3’ terminus of conPro, a 5’-terminus structure can reduce interference with the basal expression of conPro (n = 3, error bar, SD). (f) Schematic diagram of implicit OR. By adding an intermediate layer, signals from aTF1, aTF2 are respectively received using two orthogonally aPros (aPro1, aPro2), then generate the aTF3 as a common wiring molecule. (g) implicit tTAA OR GV2 gate. Using tTAA, GV2 as the input signals. The TRE3GV, URE2G promoters receive signals from tTAA and GV2, respectively. ZF21.16-VP64 was used as aZF3, and its downstream 8xZF21.16-minCMV promoter controls the expression of d2EGFP (n = 3, error bar, SD). (h) Schematic diagram of the implicit NOR gate. Unlike implicit OR gate, here we use a sTF as a common wiring molecule. (i) implicit tTAA NOR GV2 gate. sTF1 is ZF21.16-KRAB, and its downstream 8xZF21.16-CMV controls d2EGFP expression (n=3, error bar, SD). (k). The NIMPLY-form promoter is highly expressed in the presence of aTF and in the absence of sTF. When aTF and sTF coexist, sTF plays a major role. (I) The maximum activation intensity of the NIMPLY-form can be tuned by changing the number of repeats of the aTF REs. (m). The NIMPLY-form promoter shows NIMPLY logic selectivity at different aTF, sTF levels.

Strengths of TALE-NOR gates

• They optimized their NOR gates by placing the operators upstream from the promoter as it produces minimal variability in reporter expression and retains efficient repression.
• They use the SAME type of NOR gates for all their circuit functions and allows for the construction of complex circuits with precision and interoperability.
• The implementation of TALE-based cellular circuits could be easily extended to other cellular chassis, such as yeast or bacteria.
• TALEs can accommodate only one or two mismatches out of the 18-nucleotide-recognition site, supporting the potential for large orthogonality.

Integrate TALE-NOR based logic gates with ENABLE

Using NAND gate as an example

Figure legend in the original study: Implementation of the two-input Boolean logic NAND function constructed from combinations of designed TALE repressor-based NOR gates.
The logic connectivity of biological NOR gates and experimental results from the firefly luciferase and confocal microscopy are shown, with the truth table below the luciferase reporter results. Their NAND gate, constructed from two NOR gates, will produce an output if at least one of the two signals is off. Only those TALE binding sites that were functionally relevant for the circuit are illustrated. The appropriate combination of processing logic and input plasmids were introduced into HEK293T cells for each combination of input values. Values are the mean of n = 4 ± s.d. For all functions, the statistical significance of the separation between the high and low states is significant at the level ***P < 0.001. Microscopic images are representative of five separate observations. Scale bars, 125 μm. nRLU, normalized relative light units.

ENABLE^TALE-NOR

ENABLE^TALE-NOR will use our Receptors to pass the signal across the membrane and will use our transcriptional modules to amplify the production of TALE-NOR transcription factors. Different combinations of TALE-NOR will behave as our Combinator to execute binary logic functions.

HOW TO IMPLANT ENABLE^T2

In summary

• The authors used RNA-binding proteins and their RNA target units to create a set of synthetic transcription-translation control devices that can be rewired in a plug-and-play manner.
• The transcription factors are triggered in the sense that they are activated only when they interact with specific inputs.
• They use RNA-binding proteins that inhibit the translation of transcripts containing specific RNA target units.
• They used a half-subtractor and half-adder to perform fundamental arithmetic operations (addition and subtraction) of two bits. Their half-subtractor was made by a combination of the XOR gate and the NIMPLY gate. Their half-adder was made by a combination of the XOR gate and the AND gate. For more details, please refer to Figure 4.
• They have achieved two-input-two-output integration.

Side-by-side comparison

Strengths of transcription-translation control devices

• These transcription-translation control devices are rational and has a predictable plug-and-play characteristic. Individual components can be readily rewired to perform computing activities, such as NOT, NAND, and NIMPLY.
• Their single-cell biocomputers are scalable to tissue structures and are straightforward to wire with host metabolism to have therapeutic impact.

Integrate T2 into ENABLE

Using NIMPLY gate as an example

Figure legend in the original study: Design synthetic NIMPLY gates in human cells.
(a) A ANDNOT B logic gate. By combining the two input signals erythromycin and phloretin in accordance with the truth table, transfected HEK-293 cells are programmed to produce d2EYFP exclusively in the presence of erythromycin and not phloretin as shown by fluorescence microscopy and FACS analysis. b.t., below the threshold of 104 fluorescence units. Error bars represent s.d.; n = 3.

ENABLE transcription-translation control devices

ENABLE^{transcription-translation} (ENABLE^T2) will use our Receptors to pass the signal across the membrane and will use our transcriptional module to amplify the production of trigger-controlled transcription factors, including both RNA-binding proteins and their RNA target units. Different combinations of trigger-controlled transcription factors will behave as our combinator to execute binary logic functions.