Multi-layered computational gene networks by engineered tristate logics
通过工程化三态逻辑构建的多层计算基因网络
Graphical abstract 图形摘要
Keywords 关键词
Introduction 引言
Synthetic biology aims to gain precise control over living cells and tissues by approaching biological systems with a mindset of computer programmers and chip designers.1,2 Hence, human cells were engineered in the past decade to resemble “intelligent” biocomputers with gene expression profiles manipulated to show digitalized patterns of logic gates,3,4 band-pass filters,5,6 oscillators,7,8 analog-to-digital converters,9 or Boolean calculators,10,11,12 forming important basis for a variety of sophisticated solutions for numerous disciplines, including environmental sensing,13 regenerative medicine,6 drug delivery,7 in vivo diagnostics,14,15 cell-state classification,16,17,18 or cell-based therapies.19,20,21 Although, to reach the ultimate goal of programming cells with a similar precision, reliability, and control capacity as with electronic computers,10,11 identification of an efficient, systematic, and scalable coding language specified for biological contexts would be fundamental.
合成生物学旨在通过以计算机程序员和芯片设计师的思维方式接近生物系统，从而对活细胞和组织进行精确控制。因此，在过去十年中，人类细胞被工程化为类似“智能”生物计算机的形式，其基因表达谱被操控以显示数字化的逻辑门模式、带通滤波器、振荡器、模数转换器或布尔计算器，形成了多种复杂解决方案的重要基础，涵盖环境传感、再生医学、药物递送、体内诊断、细胞状态分类或基于细胞的疗法等多个学科。尽管如此，为了实现与电子计算机相似的精确性、可靠性和控制能力的细胞编程的最终目标，识别出一种高效、系统且可扩展的生物学上下文特定编码语言将是基础。
合成生物学旨在通过以计算机程序员和芯片设计师的思维方式接近生物系统，从而对活细胞和组织进行精确控制。因此，在过去十年中，人类细胞被工程化为类似“智能”生物计算机的形式，其基因表达谱被操控以显示数字化的逻辑门模式、带通滤波器、振荡器、模数转换器或布尔计算器，形成了多种复杂解决方案的重要基础，涵盖环境传感、再生医学、药物递送、体内诊断、细胞状态分类或基于细胞的疗法等多个学科。尽管如此，为了实现与电子计算机相似的精确性、可靠性和控制能力的细胞编程的最终目标，识别出一种高效、系统且可扩展的生物学上下文特定编码语言将是基础。
So far, programming strategies for multi-layered biocomputation largely follow traditional design principles that were developed for integrated circuits in digital electronics.22,23 In both electronics and biology, Boolean logic gates were the basic building blocks from which all sophisticated circuits and microprocessor-based systems can be theoretically built from scratch. For example, the “universal” NAND and NOR gates can be sequentially layered to produce any other Boolean logic gate,24,25 whereas AND, OR, and NIMPLY gates also constitute a functionally complete set of logic operators enabling custom assembly of any computational algorithm of interest in cells.2 In real practice, however, this approach could reach its limits when developing gene circuits of higher complexity.12 Unlike in electronic circuits, signal processing across gene switches and connection modules within different transcriptional and translational (TX/TL) networks is relatively slow,9,26 rendering circuit performance enormously susceptible to gate delays and/or potentially redundant coding algorithms.27 Furthermore, although every part of electronic circuits is inherently orthogonal to each other showing almost no signal interference when brought in unity, engineering of large-scale biological circuits cannot circumvent laborious redesign of mutually orthogonal or functionally normalized genetic components to ensure that there is no crosstalk between individual modules.2,28 As a result, half-adders and half-subtractors remain as the most complex network topologies that could be engineered in single mammalian cells using multi-layered gene regulation strategies.10,29 More complex biological calculations, such as full adder and full subtractors capable of processing 3-inputs 2-outputs calculus in single-cell populations, were hitherto only achieved at single layers of gene expression, i.e., involving a specific set of recombinases designed to process a same target DNA transcript according to pre-programmed combinatorics.12 Construction of integrated gene networks capable of processing multiple signals across multiple gene expression units and/or multiple stages of gene expression (e.g., interconnected TX/TL control) is still beyond the limits of current biocomputational capacities.11,30
到目前为止，多层生物计算的编程策略在很大程度上遵循了为数字电子中的集成电路开发的传统设计原则。在电子学和生物学中，布尔逻辑门是所有复杂电路和基于微处理器的系统理论上可以从零开始构建的基本构建块。例如，“通用”NAND 和 NOR 门可以按顺序叠加以产生任何其他布尔逻辑门，而 AND、OR 和 NIMPLY 门也构成了一个功能上完整的逻辑运算符集合，使得在细胞中能够自定义组装任何感兴趣的计算算法。然而，在实际操作中，当开发更高复杂度的基因电路时，这种方法可能会达到其极限。与电子电路不同，基因开关和不同转录和翻译（TX/TL）网络内的连接模块之间的信号处理相对较慢，这使得电路性能极易受到门延迟和/或潜在冗余编码算法的影响。 27 此外，尽管电子电路的每个部分本质上是正交的，彼此之间几乎没有信号干扰，但大规模生物电路的工程设计无法避免对相互正交或功能归一化的遗传组件进行繁琐的重新设计，以确保各个模块之间没有串扰。 2 28 因此，半加器和半减器仍然是可以在单个哺乳动物细胞中使用多层基因调控策略工程化的最复杂网络拓扑。 10 29 更复杂的生物计算，例如能够在单细胞群体中处理 3 输入 2 输出计算的全加器和全减器，迄今为止仅在单层基因表达中实现，即涉及一组特定的重组酶，旨在根据预编程的组合学处理相同的目标 DNA 转录本。 12 构建能够在多个基因表达单元和/或多个基因表达阶段处理多个信号的集成基因网络（例如。相互连接的 TX/TL 控制仍超出了当前生物计算能力的限制。 11 30
到目前为止，多层生物计算的编程策略在很大程度上遵循了为数字电子中的集成电路开发的传统设计原则。在电子学和生物学中，布尔逻辑门是所有复杂电路和基于微处理器的系统理论上可以从零开始构建的基本构建块。例如，“通用”NAND 和 NOR 门可以按顺序叠加以产生任何其他布尔逻辑门，而 AND、OR 和 NIMPLY 门也构成了一个功能上完整的逻辑运算符集合，使得在细胞中能够自定义组装任何感兴趣的计算算法。然而，在实际操作中，当开发更高复杂度的基因电路时，这种方法可能会达到其极限。与电子电路不同，基因开关和不同转录和翻译（TX/TL）网络内的连接模块之间的信号处理相对较慢，这使得电路性能极易受到门延迟和/或潜在冗余编码算法的影响。 27 此外，尽管电子电路的每个部分本质上是正交的，彼此之间几乎没有信号干扰，但大规模生物电路的工程设计无法避免对相互正交或功能归一化的遗传组件进行繁琐的重新设计，以确保各个模块之间没有串扰。 2 28 因此，半加器和半减器仍然是可以在单个哺乳动物细胞中使用多层基因调控策略工程化的最复杂网络拓扑。 10 29 更复杂的生物计算，例如能够在单细胞群体中处理 3 输入 2 输出计算的全加器和全减器，迄今为止仅在单层基因表达中实现，即涉及一组特定的重组酶，旨在根据预编程的组合学处理相同的目标 DNA 转录本。 12 构建能够在多个基因表达单元和/或多个基因表达阶段处理多个信号的集成基因网络（例如。相互连接的 TX/TL 控制仍超出了当前生物计算能力的限制。 11 30
Here, we present a tristate-based logic synthesis (TriLoS) approach as a conceptual alternative to map various combinational logics into multi-layered mammalian gene networks. TriLoS transforms the logic algorithm of a given (bio)computational task into its disjunctive normal form and displays each clause of the logic expression as genetic implementations of interconnected tristate buffers. In electronics, tristate buffers have a unique architecture in comprising an upstream switch that directly controls the connectivity of a downstream switch so that multiple modules can be flexibly connected to a same output wire in parallel to enable resource-efficient data transfer without sacrificing switching speed.31 Thus, engineering of genetic variants of tristate buffers could be particularly attractive for biological computation because any residual gene expression activity potentially produced from OFF-state modules can be physically dismounted by design so that an entire gene circuit must no longer exclusively consist of strictly orthogonal building blocks in each and every part. Thus, we first create digital buffer (BUF) and inverter (NOT) switches controlled by a same data input A and further show that BUF and NOT can be flexibly plugged or unplugged by a control input B to produce the four tristate buffers BUFIF1, NOTIF1, BUFIF0, and NOTIF0. Then, we present a notion where each tristate buffer shows logic similarity to conventional AND, NIMPLY, and NOR gates within a biological context, and that all 2-input 1-output Boolean logic gates can be mapped through a specific combination between tristate buffers and BUF/NOT switches. Using such tristate-based design strategy, separate channels for additional output signals can be readily incorporated into the computational switchboard by plugging another independent set of data-input-A-controlled BUF and NOT switches onto the output module of control input B. Likewise, addition of multiple input signals is achieved by creating upstream layers of tristate buffers consisting of a new control input C and, in turn, converting signal B into its new data input. Thus, using TriLoS, we were not only able to program multi-layered gene networks with extraordinary robustness, modularity, and arithmetic complexity, but also demonstrate a putative treatment paradigm allowing a same batch of implanted cell-based therapeutics to operate a disease-specific 3-input 2-output drug secretion regimen in vivo. In conclusion, this work provides a resource-efficient alternative to program integrated gene networks that no longer indiscriminately follows traditional design blueprints of electronic circuits and might uncover the therapeutic potential of mammalian biocomputers.
在这里，我们提出了一种基于三态逻辑综合（TriLoS）的方法，作为将各种组合逻辑映射到多层哺乳动物基因网络的概念替代方案。TriLoS 将给定（生物）计算任务的逻辑算法转换为其析取范式，并将逻辑表达式的每个子句显示为互连的三态缓冲器的遗传实现。在电子学中，三态缓冲器具有独特的架构，包括一个上游开关，直接控制下游开关的连接性，从而使多个模块能够灵活地并行连接到同一输出线，以实现资源高效的数据传输而不牺牲切换速度。因此，三态缓冲器的遗传变体的工程设计对于生物计算特别具有吸引力，因为任何可能由 OFF 状态模块产生的残余基因表达活动都可以通过设计物理拆除，从而使整个基因电路不再必须在每个部分严格由正交构建块组成。 因此，我们首先创建由相同数据输入 A 控制的数字缓冲器（BUF）和反相器（NOT）开关，并进一步展示 BUF 和 NOT 可以通过控制输入 B 灵活地插入或拔出，以产生四个三态缓冲器 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0。然后，我们提出一个概念，其中每个三态缓冲器在生物学背景下显示出与传统与门、非门和或门的逻辑相似性，并且所有 2 输入 1 输出的布尔逻辑门可以通过三态缓冲器与 BUF/NOT 开关之间的特定组合进行映射。采用这种基于三态的设计策略，可以通过将另一组独立的由数据输入 A 控制的 BUF 和 NOT 开关插入控制输入 B 的输出模块，轻松地将额外输出信号的独立通道纳入计算开关板。同样，通过创建由新的控制输入 C 组成的三态缓冲器的上游层，可以实现多个输入信号的添加，并进而将信号 B 转换为其新的数据输入。 因此，使用 TriLoS，我们不仅能够编程具有卓越鲁棒性、模块化和算术复杂性的多层基因网络，还能够展示一种假定的治疗范式，使同一批植入的细胞治疗能够在体内执行特定疾病的 3 输入 2 输出药物分泌方案。总之，这项工作提供了一种资源高效的替代方案，以编程集成基因网络，这些网络不再无差别地遵循传统电子电路的设计蓝图，并可能揭示哺乳动物生物计算机的治疗潜力。
在这里，我们提出了一种基于三态逻辑综合（TriLoS）的方法，作为将各种组合逻辑映射到多层哺乳动物基因网络的概念替代方案。TriLoS 将给定（生物）计算任务的逻辑算法转换为其析取范式，并将逻辑表达式的每个子句显示为互连的三态缓冲器的遗传实现。在电子学中，三态缓冲器具有独特的架构，包括一个上游开关，直接控制下游开关的连接性，从而使多个模块能够灵活地并行连接到同一输出线，以实现资源高效的数据传输而不牺牲切换速度。因此，三态缓冲器的遗传变体的工程设计对于生物计算特别具有吸引力，因为任何可能由 OFF 状态模块产生的残余基因表达活动都可以通过设计物理拆除，从而使整个基因电路不再必须在每个部分严格由正交构建块组成。 因此，我们首先创建由相同数据输入 A 控制的数字缓冲器（BUF）和反相器（NOT）开关，并进一步展示 BUF 和 NOT 可以通过控制输入 B 灵活地插入或拔出，以产生四个三态缓冲器 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0。然后，我们提出一个概念，其中每个三态缓冲器在生物学背景下显示出与传统与门、非门和或门的逻辑相似性，并且所有 2 输入 1 输出的布尔逻辑门可以通过三态缓冲器与 BUF/NOT 开关之间的特定组合进行映射。采用这种基于三态的设计策略，可以通过将另一组独立的由数据输入 A 控制的 BUF 和 NOT 开关插入控制输入 B 的输出模块，轻松地将额外输出信号的独立通道纳入计算开关板。同样，通过创建由新的控制输入 C 组成的三态缓冲器的上游层，可以实现多个输入信号的添加，并进而将信号 B 转换为其新的数据输入。 因此，使用 TriLoS，我们不仅能够编程具有卓越鲁棒性、模块化和算术复杂性的多层基因网络，还能够展示一种假定的治疗范式，使同一批植入的细胞治疗能够在体内执行特定疾病的 3 输入 2 输出药物分泌方案。总之，这项工作提供了一种资源高效的替代方案，以编程集成基因网络，这些网络不再无差别地遵循传统电子电路的设计蓝图，并可能揭示哺乳动物生物计算机的治疗潜力。
Results 结果
Design and construction of VA- and Gra-regulated tristate buffers
VA 和 Gra 调节的三态缓冲器的设计与构建
In both electronics and biology, minimization of gate delays throughout the entire circuit is fundamental for faster data transfers and achievement of high computational performance.31 To this end, digital buffer (BUF) and inverter switches (NOT) with an in-degree score of one have shorter inherent gate propagation delays than Boolean logic gates (e.g., AND or OR gates).32 Therefore, we hypothesize that gene circuits engineered to exclusively consist of interconnected BUF and NOT switches could break the ground to highly efficient biocomputation with minimal coding redundancy. For this purpose, the architecture of tristate buffers33 would enable flexible combination and usage of BUF and NOT switches by design, in which the connectivity of any binary switch regulated by an input A is strictly governed by an upstream switch through another signal B (Figure 1A). For implementation of such tristate buffers in mammalian cells, a pair of antagonistic gene switches operating at a same stage of gene expression and responding to a same input signal are preferred candidates for potential BUF and NOT switches (Figure 1B). To engineer such switches, hepatitis C virus (HCV)-derived NS3a protein is an attractive module of choice because its association with a de novo designed grazoprevir (Gra)/NS3a complex reader (GNCR1) protein and dissociation from an apo NS3a reader (ANR) peptide are both triggered by the small molecule drug Gra (Figure 1C).34 When we used this Gra-controlled triad NS3a/GNCR1/ANR to build a TL regulation system using a recently reported strategy based on synthetic TL initiation factors (STIFs) and STIF-specific target mRNA,35 Gra could indeed trigger TL initiation through the reconstitution of GNCR1- and NS3a-containing STIFs (Figure S1A), but the STIF regulators MCP-NS3a (where the Bacteriophage MS2-specific coat protein [MCP]36 defines site specificity to target gene mRNA) and ANR-NSP3 (where the rotaviral nonstructural protein 3 [NSP3] protein37 initiates translation at ANR-specific mRNA sites) failed to result in Gra-repressible translation (Figure S1B). To fix this issue, we replaced NS3a by a synthetic mutant NS3a(H1) selected for increased binding affinity to ANR.38 Results showed that both trigger-inducible (Figure S1A) and repressible (Figure S1B) target gene translation could be engineered to depend on the same data input Gra in mammalian cells (Figure 1C). Furthermore, the signal-to-noise ratio of corresponding BUF (Figure 1D) and NOT switches (Figure 1E) could be systematically fine-tuned by increasing the number of tandem NS3a(H1)-repeats added to MCP.
在电子学和生物学中，整个电路中门延迟的最小化对于更快的数据传输和实现高计算性能至关重要。为此，具有入度得分为一的数字缓冲器（BUF）和反相器开关（NOT）具有比布尔逻辑门（例如，AND 或 OR 门）更短的固有门传播延迟。因此，我们假设工程化的基因电路仅由相互连接的 BUF 和 NOT 开关组成，可以开辟高效生物计算的新领域，具有最小的编码冗余。为此，三态缓冲器的架构将通过设计使 BUF 和 NOT 开关的灵活组合和使用成为可能，其中由输入 A 调节的任何二进制开关的连通性严格受另一个信号 B 通过上游开关的控制（图 1A）。为了在哺乳动物细胞中实现这种三态缓冲器，一对在同一基因表达阶段操作并对同一输入信号作出响应的对抗性基因开关是潜在的 BUF 和 NOT 开关的优选候选者（图 1B）。 为了设计这样的开关，源自丙型肝炎病毒（HCV）的 NS3a 蛋白是一个理想的选择模块，因为它与新设计的 grazoprevir（Gra）/NS3a 复合物读取器（GNCR1）蛋白的结合以及与 apo NS3a 读取器（ANR）肽的解离均由小分子药物 Gra 触发（图 1C）。 34 当我们使用这个 Gra 控制的三元组 NS3a/GNCR1/ANR 来构建一个基于合成 TL 起始因子（STIFs）和 STIF 特异性靶 mRNA 的 TL 调控系统时， 35 Gra 确实能够通过重组含有 GNCR1 和 NS3a 的 STIFs 来触发 TL 起始（图 S1A），但 STIF 调节因子 MCP-NS3a（其中噬菌体 MS2 特异性外壳蛋白[MCP] 36 定义了靶基因 mRNA 的位点特异性）和 ANR-NSP3（其中轮状病毒非结构蛋白 3[NSP3] 37 在 ANR 特异性 mRNA 位点启动翻译）未能导致 Gra 可抑制的翻译（图 S1B）。为了解决这个问题，我们用一种合成突变体 NS3a(H1)替换了 NS3a，该突变体被选中以增加与 ANR 的结合亲和力。 结果显示，触发诱导型（图 S1A）和可抑制型（图 S1B）靶基因翻译均可以在哺乳动物细胞中被设计为依赖于相同的数据输入 Gra（图 1C）。此外，通过增加添加到 MCP 的串联 NS3a(H1) 重复的数量，可以系统性地微调相应 BUF（图 1D）和 NOT 开关（图 1E）的信噪比。
在电子学和生物学中，整个电路中门延迟的最小化对于更快的数据传输和实现高计算性能至关重要。为此，具有入度得分为一的数字缓冲器（BUF）和反相器开关（NOT）具有比布尔逻辑门（例如，AND 或 OR 门）更短的固有门传播延迟。因此，我们假设工程化的基因电路仅由相互连接的 BUF 和 NOT 开关组成，可以开辟高效生物计算的新领域，具有最小的编码冗余。为此，三态缓冲器的架构将通过设计使 BUF 和 NOT 开关的灵活组合和使用成为可能，其中由输入 A 调节的任何二进制开关的连通性严格受另一个信号 B 通过上游开关的控制（图 1A）。为了在哺乳动物细胞中实现这种三态缓冲器，一对在同一基因表达阶段操作并对同一输入信号作出响应的对抗性基因开关是潜在的 BUF 和 NOT 开关的优选候选者（图 1B）。 为了设计这样的开关，源自丙型肝炎病毒（HCV）的 NS3a 蛋白是一个理想的选择模块，因为它与新设计的 grazoprevir（Gra）/NS3a 复合物读取器（GNCR1）蛋白的结合以及与 apo NS3a 读取器（ANR）肽的解离均由小分子药物 Gra 触发（图 1C）。 34 当我们使用这个 Gra 控制的三元组 NS3a/GNCR1/ANR 来构建一个基于合成 TL 起始因子（STIFs）和 STIF 特异性靶 mRNA 的 TL 调控系统时， 35 Gra 确实能够通过重组含有 GNCR1 和 NS3a 的 STIFs 来触发 TL 起始（图 S1A），但 STIF 调节因子 MCP-NS3a（其中噬菌体 MS2 特异性外壳蛋白[MCP] 36 定义了靶基因 mRNA 的位点特异性）和 ANR-NSP3（其中轮状病毒非结构蛋白 3[NSP3] 37 在 ANR 特异性 mRNA 位点启动翻译）未能导致 Gra 可抑制的翻译（图 S1B）。为了解决这个问题，我们用一种合成突变体 NS3a(H1)替换了 NS3a，该突变体被选中以增加与 ANR 的结合亲和力。 结果显示，触发诱导型（图 S1A）和可抑制型（图 S1B）靶基因翻译均可以在哺乳动物细胞中被设计为依赖于相同的数据输入 Gra（图 1C）。此外，通过增加添加到 MCP 的串联 NS3a(H1) 重复的数量，可以系统性地微调相应 BUF（图 1D）和 NOT 开关（图 1E）的信噪比。
Figure 1 Molecular architecture and design principles of genetic tristate buffers
图 1 遗传三态缓冲器的分子结构和设计原则
图 1 遗传三态缓冲器的分子结构和设计原则
Figure S1 Design and validation of BUF and NOT switches controlled by grazoprevir or vanillic acid, related to Figure 1
图 S1 由 grazoprevir 或香草酸控制的 BUF 和 NOT 开关的设计与验证，相关于图 1
图 S1 由 grazoprevir 或香草酸控制的 BUF 和 NOT 开关的设计与验证，相关于图 1
To form tristate buffers (Figure 1A), expression of Gra-controlled BUF and NOT switches must be regulated by an upstream gene switch, which, in turn, is governed by a further control input B (Figure 1B). Such upstream gene switch can either produce an inverted (known as “active-LOW” control signal resulting from an “IF0” switch) or non-inverted output signal by default (known as “active-HIGH” control signal from an “IF1” switch). For implementation in mammalian cells, it is therefore essential that a same trigger signal controls two mutually independent gene switches (IF0 and IF1) in parallel (Figure 1F). For proof of concept, we chose vanillic acid (VA) as an exemplary control input B and showed that cells transfected with a VA-specific murine olfactory receptor (MOR9-1) coupling to Gαs protein6 and a synthetic protein kinase A (PKA)/ cyclic AMP response elements binding protein 1 (CREB1)-responsive promoter39 produces a VA-inducible gene switch showing active-HIGH expression profiles (Figure S1D), whereas VA-repressible gene switches with typical active-LOW logics were created on the basis of a synthetic vanillic acid-dependent repressor (VanR)-dependent mammalian transactivator (VanR-VP64) modulating gene expression from cognate VanO-containing promoters40 (Figure S1C). Notably, VA-triggered IF0 and IF1 switches regulating different output signals showed no crosstalk to each other when introduced into the same cells (Figure 1G), thus fulfilling the eligibility requirements for upstream gene switches in tristate-based gene circuits (Figure 1F).
为了形成三态缓冲器（图 1A），Gra 控制的 BUF 和 NOT 开关的表达必须由一个上游基因开关调节，而该开关又受到进一步控制输入 B 的调控（图 1B）。这样的上游基因开关可以默认产生一个反向输出信号（称为“主动低”控制信号，源自“IF0”开关）或非反向输出信号（称为“主动高”控制信号，源自“IF1”开关）。因此，在哺乳动物细胞中的实现至关重要的是，同一个触发信号在并行控制两个相互独立的基因开关（IF0 和 IF1）（图 1F）。 为了概念验证，我们选择了香草酸（VA）作为示例控制输入 B，并展示了转染了 VA 特异性小鼠嗅觉受体（MOR9-1）与 Gαs 蛋白耦合的细胞 6 ，以及合成的蛋白激酶 A（PKA）/环 AMP 反应元件结合蛋白 1（CREB1）响应启动子 39 ，产生了 VA 诱导的基因开关，显示出活跃的高表达特征（图 S1D），而基于合成的香草酸依赖性抑制子（VanR）调节来自相应 VanO 含有启动子的基因表达的 VA 抑制性基因开关则具有典型的活跃低逻辑（图 S1C）。值得注意的是，VA 触发的 IF0 和 IF1 开关调节不同的输出信号，当引入到同一细胞中时，彼此之间没有交叉干扰（图 1G），从而满足了三态基因电路中上游基因开关的资格要求（图 1F）。
为了形成三态缓冲器（图 1A），Gra 控制的 BUF 和 NOT 开关的表达必须由一个上游基因开关调节，而该开关又受到进一步控制输入 B 的调控（图 1B）。这样的上游基因开关可以默认产生一个反向输出信号（称为“主动低”控制信号，源自“IF0”开关）或非反向输出信号（称为“主动高”控制信号，源自“IF1”开关）。因此，在哺乳动物细胞中的实现至关重要的是，同一个触发信号在并行控制两个相互独立的基因开关（IF0 和 IF1）（图 1F）。 为了概念验证，我们选择了香草酸（VA）作为示例控制输入 B，并展示了转染了 VA 特异性小鼠嗅觉受体（MOR9-1）与 Gαs 蛋白耦合的细胞 6 ，以及合成的蛋白激酶 A（PKA）/环 AMP 反应元件结合蛋白 1（CREB1）响应启动子 39 ，产生了 VA 诱导的基因开关，显示出活跃的高表达特征（图 S1D），而基于合成的香草酸依赖性抑制子（VanR）调节来自相应 VanO 含有启动子的基因表达的 VA 抑制性基因开关则具有典型的活跃低逻辑（图 S1C）。值得注意的是，VA 触发的 IF0 和 IF1 开关调节不同的输出信号，当引入到同一细胞中时，彼此之间没有交叉干扰（图 1G），从而满足了三态基因电路中上游基因开关的资格要求（图 1F）。
Next, we connected VA-regulated IF0 and IF1 switches (upstream module) and Gra-regulated BUF and NOT switches (downstream module), thus yielding the four different types of tristate buffers BUFIF1, NOTIF1, BUFIF0, and NOTIF0 (Figure 2). In tristate buffers, the upstream gene switch is used to control whether the downstream gene switch shall be physically connected to the circuit at all. For example, a BUFIF1 circuit would only allow a BUF switch (output signal Y follows the value of A) to “exist” as long as the upstream gene switch is active (BUF: Y = 0 if A = 0 and Y = 1 if A = 1; IF1: if B is equal 1). In other cases (e.g., when B is equal 0), the entire BUF switch is electrically unplugged from the circuit so that the overall output signal Y would fall into a third high-impedance state Z (Figure 2A). Likewise, a NOTIF0 buffer only empowers a NOT switch when the upstream switch is inactive (NOT: Y = 1 if A = 0 and Y = 0 if A = 1; IF0: if B is equal 0) and produces a high-impedance signal in any other situation (Y = Z when B is equal 1; Figure 2A). Although it is elementary to distinguish these three energetically different output states (0, 1, and Z) in electronic circuits, we hypothesize that from the viewpoint of target gene expression, Hi-Z (electrically disconnected) and 0 states (electrically switched off) are functionally similar within a biological context. Following this notion (gene expression is considered OFF when Y is either 0 or Z and considered ON when Y = 1), we experimentally confirm that BUFIF1 shows logic similarity with a conventional AND gate, whereas NOTIF0 is logically similar to a conventional NOR gate. Likewise, NOTIF1 and BUFIF0 show typical gene expression signatures of both variants of NIMPLY (AND NOT) gates (Figure 2A). Because the tristate architecture ensures that the downstream gene switches BUF and NOT cannot co-exist by design (Figure 1B) and downstream switches only exist when being effectively used (Figure 2A), tristate buffers can be generally constructed using mutually orthogonal IF1 and IF0 switches controlling individual sets of BUF and NOT switches, which, in turn, must not be strictly orthogonal to each other (Figure 1F). This would reduce unwanted background activities, susceptibility to signal propagation delay, as well as overall engineering labor.
接下来，我们连接了 VA 调控的 IF0 和 IF1 开关（上游模块）以及 Gra 调控的 BUF 和 NOT 开关（下游模块），从而产生了四种不同类型的三态缓冲器 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0（图 2）。在三态缓冲器中，上游基因开关用于控制下游基因开关是否与电路物理连接。例如，BUFIF1 电路仅在上游基因开关处于激活状态时（BUF：如果 A=0 则 Y=0，如果 A=1 则 Y=1；IF1：如果 B 等于 1）允许 BUF 开关（输出信号 Y 跟随 A 的值）“存在”。在其他情况下（例如，当 B 等于 0 时），整个 BUF 开关会从电路中电气断开，从而使得整体输出信号 Y 进入第三种高阻抗状态 Z（图 2A）。同样，NOTIF0 缓冲器仅在上游开关不活跃时（NOT：如果 A=0 则 Y=1，如果 A=1 则 Y=0；IF0：如果 B 等于 0）使 NOT 开关有效，并在任何其他情况下产生高阻抗信号（当 B 等于 1 时 Y=Z；图 2A）。 尽管在电子电路中区分这三种能量不同的输出状态（0、1 和 Z）是基础的，但我们假设从目标基因表达的角度来看，Hi-Z（电气断开）和 0 状态（电气关闭）在生物学背景下是功能上相似的。根据这一概念（当 Y 为 0 或 Z 时，基因表达被视为关闭；当 Y=1 时，基因表达被视为开启），我们实验确认 BUFIF1 与传统的与门在逻辑上相似，而 NOTIF0 在逻辑上与传统的或非门相似。同样，NOTIF1 和 BUFIF0 显示出 NIMPLY（与非）门的两种变体的典型基因表达特征（图 2A）。由于三态架构确保下游基因开关 BUF 和 NOT 在设计上不能共存（图 1B），并且下游开关仅在有效使用时存在（图 2A），三态缓冲器可以一般地使用相互正交的 IF1 和 IF0 开关来控制各自的 BUF 和 NOT 开关，而这些开关之间不必严格正交（图 1F）。 这将减少不必要的背景活动、信号传播延迟的敏感性，以及整体工程劳动。
接下来，我们连接了 VA 调控的 IF0 和 IF1 开关（上游模块）以及 Gra 调控的 BUF 和 NOT 开关（下游模块），从而产生了四种不同类型的三态缓冲器 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0（图 2）。在三态缓冲器中，上游基因开关用于控制下游基因开关是否与电路物理连接。例如，BUFIF1 电路仅在上游基因开关处于激活状态时（BUF：如果 A=0 则 Y=0，如果 A=1 则 Y=1；IF1：如果 B 等于 1）允许 BUF 开关（输出信号 Y 跟随 A 的值）“存在”。在其他情况下（例如，当 B 等于 0 时），整个 BUF 开关会从电路中电气断开，从而使得整体输出信号 Y 进入第三种高阻抗状态 Z（图 2A）。同样，NOTIF0 缓冲器仅在上游开关不活跃时（NOT：如果 A=0 则 Y=1，如果 A=1 则 Y=0；IF0：如果 B 等于 0）使 NOT 开关有效，并在任何其他情况下产生高阻抗信号（当 B 等于 1 时 Y=Z；图 2A）。 尽管在电子电路中区分这三种能量不同的输出状态（0、1 和 Z）是基础的，但我们假设从目标基因表达的角度来看，Hi-Z（电气断开）和 0 状态（电气关闭）在生物学背景下是功能上相似的。根据这一概念（当 Y 为 0 或 Z 时，基因表达被视为关闭；当 Y=1 时，基因表达被视为开启），我们实验确认 BUFIF1 与传统的与门在逻辑上相似，而 NOTIF0 在逻辑上与传统的或非门相似。同样，NOTIF1 和 BUFIF0 显示出 NIMPLY（与非）门的两种变体的典型基因表达特征（图 2A）。由于三态架构确保下游基因开关 BUF 和 NOT 在设计上不能共存（图 1B），并且下游开关仅在有效使用时存在（图 2A），三态缓冲器可以一般地使用相互正交的 IF1 和 IF0 开关来控制各自的 BUF 和 NOT 开关，而这些开关之间不必严格正交（图 1F）。 这将减少不必要的背景活动、信号传播延迟的敏感性，以及整体工程劳动。
Figure 2 Modular assembly of Boolean logic circuits using tristate buffers and BUF/NOT switches
图 2 使用三态缓冲器和 BUF/NOT 开关的布尔逻辑电路的模块化组装
图 2 使用三态缓冲器和 BUF/NOT 开关的布尔逻辑电路的模块化组装
Boolean logic gates engineered to exclusively consist of interconnected tristate buffers
布尔逻辑门被设计为仅由相互连接的三态缓冲器组成
After we showed that the four tristate buffers BUFIF1, NOTIF1, BUFIF0, and NOTIF0 can be engineered to emulate AND, NOR, and NIMPLY logics in mammalian cells, we wondered whether all 2-input 1-output Boolean logic gates can be assembled exclusively based on interconnected BUF and NOT switches. For this purpose, we kept the notion of setting Z approximately equal to 0 in biological systems and created mathematical representations for tristate buffers and BUF/NOT switches (BUF,
在我们展示四个三态缓冲器 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0 可以被设计成在哺乳动物细胞中模拟 AND、NOR 和 NIMPLY 逻辑之后，我们开始思考是否所有的 2 输入 1 输出布尔逻辑门都可以仅基于互连的 BUF 和 NOT 开关组装。为此，我们保持在生物系统中将 Z 近似设为 0 的概念，并为三态缓冲器和 BUF/NOT 开关（BUF，
在我们展示四个三态缓冲器 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0 可以被设计成在哺乳动物细胞中模拟 AND、NOR 和 NIMPLY 逻辑之后，我们开始思考是否所有的 2 输入 1 输出布尔逻辑门都可以仅基于互连的 BUF 和 NOT 开关组装。为此，我们保持在生物系统中将 Z 近似设为 0 的概念，并为三态缓冲器和 BUF/NOT 开关（BUF，
Figure S2 Implementation of tristate-based XOR logics in different mammalian cell lines, related to Figure 2
图 S2 在不同哺乳动物细胞系中实现三态基 XOR 逻辑，与图 2 相关
图 S2 在不同哺乳动物细胞系中实现三态基 XOR 逻辑，与图 2 相关
Expanding TriLoS-based gene circuits for multi-input, multi-output biocomputation
扩展基于 TriLoS 的基因电路以实现多输入、多输出的生物计算
Whereas Boolean logic gates convert multiple input signals into a single output signal according to pre-programmed algorithms,43 calculators typically produce multiple output signals. To expand TriLoS-based gene circuits to return multiple outputs in parallel, independent sets of data-input-regulated BUF/NOT switches, each controlling a different output signal, are therefore required (Figure 3A). In our example of using Gra as the data input A, new Gra-dependent BUFn/NOTn switches controlling expression of different target genes n must be engineered. Thus, apart from the “first” set of Gra-regulated switches based on STIF-dependent translation (Figure 1; designated BUF1/NOT1), we created a second set of Gra-responsive gene switches BUF2/NOT2 by incorporating the mutually exclusive triad NS3a(H1)/GNCR1/ANR (Figure 1C) now into the framework of synthetic generalized extracellular molecule sensor (GEMS) receptors44 (Figure 3B). GEMS receptors typically comprise an antibody-derived extracellular ligand binding domain, an EpoR-derived transmembrane domain (GEMSTM), and an intracellular signal transduction domain mediating activation of different signaling pathways in human cells upon dimerization of the cell surface receptor.44 To create GEMS-based BUF/NOT switches regulated by Gra, we replaced the antibody-domain of conventional GEMS constructs by either NS3a(H1), GNCR1, or ANR (Figure S3). Each GEMS variant was subsequently tested for different intracellular signaling domains, such as an IL-6RBm (modified interleukin 6 receptor B; triggering Janus kinase and signal transducer and activator of transcription 3 [JAK/STAT3]-signaling45), FGFR1int (intracellular part of fibroblast growth factor receptor 1; triggering mitogen-activated protein kinase [MAPK]-signaling46), or VEGFR2int (intracellular part of vascular endothelial growth factor receptor 2; triggering nuclear factor of activated T-cells [NFAT]-signaling47), with experimental results showing that only GEMS receptors containing intracellular IL-6RBm allowed for Gra-triggered target gene expression (Figure S3). Thus, co-expression of NS3a(H1)-GEMSTM-IL-6RBm (designated GEMSNS3a(H1)), GNCR1-GEMSTM-IL-6RBm (designated GEMSGNCR1), and a reporter gene expression vector driven by synthetic STAT3-specific promoters yields the new Gra-inducible BUF2 switch, whereas the analogous Gra-repressible NOT2 switch comprises ANR-GEMSTM-IL-6RBm (designated GEMSANR) instead of GEMSGNCR1 (Figures 3B and S3).
由于布尔逻辑门根据预编程算法将多个输入信号转换为单个输出信号， 43 计算器通常会产生多个输出信号。为了扩展基于 TriLoS 的基因电路以并行返回多个输出，因此需要独立的数据输入调节 BUF/NOT 开关组，每组控制不同的输出信号（图 3A）。在我们使用 Gra 作为数据输入 A 的例子中，必须设计新的 Gra 依赖的 BUFn/NOTn 开关，以控制不同靶基因 n 的表达。因此，除了基于 STIF 依赖翻译的“第一”组 Gra 调节开关（图 1；指定为 BUF1/NOT1）外，我们通过将相互排斥的三元组 NS3a(H1)/GNCR1/ANR（图 1C）纳入合成广义细胞外分子传感器（GEMS）受体的框架中，创建了第二组 Gra 响应基因开关 BUF2/NOT2（图 3B）。 GEMS 受体通常由抗体衍生的细胞外配体结合域、EpoR 衍生的跨膜域（GEMSTM）和一个介导细胞表面受体二聚化后在人类细胞中激活不同信号通路的细胞内信号转导域组成。 44 为了创建由 Gra 调控的基于 GEMS 的 BUF/NOT 开关，我们用 NS3a(H1)、GNCR1 或 ANR 替换了传统 GEMS 构建体的抗体域（图 S3）。随后对每个 GEMS 变体进行了不同细胞内信号域的测试，例如 IL-6RBm（修饰的白细胞介素 6 受体 B；触发 Janus 激酶和转录激活因子 3 [JAK/STAT3]信号通路 45 ）、FGFR1int（成纤维生长因子受体 1 的细胞内部分；触发有丝分裂原激活蛋白激酶 [MAPK]信号通路 46 ）或 VEGFR2int（血管内皮生长因子受体 2 的细胞内部分；触发活化 T 细胞的核因子 [NFAT]信号通路 47 ），实验结果表明，只有含有细胞内 IL-6RBm 的 GEMS 受体允许 Gra 触发的靶基因表达（图 S3）。 因此，NS3a(H1)-GEMSTM-IL-6RBm（称为 GEMSNS3a）、GNCR1-GEMSTM-IL-6RBm（称为 GEMSGNCR1）和由合成 STAT3 特异性启动子驱动的报告基因表达载体的共同表达产生了新的 Gra 诱导 BUF2 开关，而类似的 Gra 抑制 NOT2 开关则包含 ANR-GEMSTM-IL-6RBm（称为 GEMSANR），而不是 GEMSGNCR1（图 3B 和 S3）。
由于布尔逻辑门根据预编程算法将多个输入信号转换为单个输出信号， 43 计算器通常会产生多个输出信号。为了扩展基于 TriLoS 的基因电路以并行返回多个输出，因此需要独立的数据输入调节 BUF/NOT 开关组，每组控制不同的输出信号（图 3A）。在我们使用 Gra 作为数据输入 A 的例子中，必须设计新的 Gra 依赖的 BUFn/NOTn 开关，以控制不同靶基因 n 的表达。因此，除了基于 STIF 依赖翻译的“第一”组 Gra 调节开关（图 1；指定为 BUF1/NOT1）外，我们通过将相互排斥的三元组 NS3a(H1)/GNCR1/ANR（图 1C）纳入合成广义细胞外分子传感器（GEMS）受体的框架中，创建了第二组 Gra 响应基因开关 BUF2/NOT2（图 3B）。 GEMS 受体通常由抗体衍生的细胞外配体结合域、EpoR 衍生的跨膜域（GEMSTM）和一个介导细胞表面受体二聚化后在人类细胞中激活不同信号通路的细胞内信号转导域组成。 44 为了创建由 Gra 调控的基于 GEMS 的 BUF/NOT 开关，我们用 NS3a(H1)、GNCR1 或 ANR 替换了传统 GEMS 构建体的抗体域（图 S3）。随后对每个 GEMS 变体进行了不同细胞内信号域的测试，例如 IL-6RBm（修饰的白细胞介素 6 受体 B；触发 Janus 激酶和转录激活因子 3 [JAK/STAT3]信号通路 45 ）、FGFR1int（成纤维生长因子受体 1 的细胞内部分；触发有丝分裂原激活蛋白激酶 [MAPK]信号通路 46 ）或 VEGFR2int（血管内皮生长因子受体 2 的细胞内部分；触发活化 T 细胞的核因子 [NFAT]信号通路 47 ），实验结果表明，只有含有细胞内 IL-6RBm 的 GEMS 受体允许 Gra 触发的靶基因表达（图 S3）。 因此，NS3a(H1)-GEMSTM-IL-6RBm（称为 GEMSNS3a）、GNCR1-GEMSTM-IL-6RBm（称为 GEMSGNCR1）和由合成 STAT3 特异性启动子驱动的报告基因表达载体的共同表达产生了新的 Gra 诱导 BUF2 开关，而类似的 Gra 抑制 NOT2 开关则包含 ANR-GEMSTM-IL-6RBm（称为 GEMSANR），而不是 GEMSGNCR1（图 3B 和 S3）。
Figure 3 Expansion of tristate-based gene circuits to produce multiple output channels
图 3 基于三态的基因电路扩展以产生多个输出通道
图 3 基于三态的基因电路扩展以产生多个输出通道
Figure S3 Design and validation of grazoprevir-inducible and grazoprevir-repressible GEMS variants, related to Figure 3
图 S3 grazoprevir 诱导和抑制 GEMS 变体的设计与验证，与图 3 相关
图 S3 grazoprevir 诱导和抑制 GEMS 变体的设计与验证，与图 3 相关
To integrate both sets of BUFn/NOTn switches into a same tristate-based gene circuit, it is essential that there is no signal crosstalk between each individual set of gene switches when regulating different output modules in parallel. Therefore, we co-expressed an NOT1 switch controlling secreted alkaline phosphatase (SEAP) as a first reporter gene with a BUF2 switch controlling secreted nano luciferase (Nluc) as a second reporter gene. We also co-expressed the NOT2 switch controlling Nluc with a BUF1 switch controlling SEAP in the same cells. Results showed that each individual switch operated in a highly autonomous manner when triggered by Gra, demonstrating robust and interference-free performance in mammalian cells (Figure 3C). Next, we plugged both sets of Gra-regulated BUF/NOT switches onto the output wire of the VA-controlled upstream module, thus establishing a tristate-based switchboard capable of processing 2-input 2-output gene expression logics (Figure 3A). Because both IF1 and BUF2/NOT2 utilize different intracellular signaling pathways in mammalian cells, we first tested whether there is potential signal crosstalk between respective key components. Experiments show that an IF1-switch controlling SEAP expression could operate in parallel to a BUF2-switch driving Nluc expression in same cells. This not only supports circuit performance for our study but also indicates potential orthogonality between intracellular cyclic AMP (cAMP) and STAT3-signaling from a cell biology perspective in general (Figure S4). Finally, we assembled the four tristate buffers BUF2IF1, NOT2IF1, BUF2IF0, and NOT2IF0 in a similar plug-and-play manner as previously shown with the STIF-based BUF1/NOT1 set, again producing VA- and Gra-dependent gene circuits with logic similarity to AND, NOR, and IMPLY gates using the new BUF2/NOT2 set (Figure 3D).
为了将两组 BUFn/NOTn 开关整合到同一个基于三态的基因电路中，确保在并行调节不同输出模块时，各个基因开关之间没有信号串扰是至关重要的。因此，我们共同表达了一个控制分泌性碱性磷酸酶（SEAP）的 NOT1 开关作为第一个报告基因，以及一个控制分泌性纳米荧光素酶（Nluc）的 BUF2 开关作为第二个报告基因。我们还在同一细胞中共同表达了控制 Nluc 的 NOT2 开关和控制 SEAP 的 BUF1 开关。结果表明，每个独立的开关在 Gra 的触发下以高度自主的方式运作，展示了在哺乳动物细胞中强健且无干扰的性能（图 3C）。接下来，我们将两组 Gra 调控的 BUF/NOT 开关连接到 VA 控制的上游模块的输出线上，从而建立了一个能够处理 2 输入 2 输出基因表达逻辑的基于三态的开关板（图 3A）。由于 IF1 和 BUF2/NOT2 在哺乳动物细胞中利用不同的细胞内信号通路，我们首先测试了各关键组分之间是否存在潜在的信号串扰。 实验表明，控制 SEAP 表达的 IF1 开关可以与驱动 Nluc 表达的 BUF2 开关在同一细胞中并行操作。这不仅支持了我们研究的电路性能，还从细胞生物学的角度表明细胞内环磷酸腺苷（cAMP）与 STAT3 信号传导之间可能存在正交性（图 S4）。最后，我们以类似于之前展示的基于 STIF 的 BUF1/NOT1 组的即插即用方式组装了四个三态缓冲器 BUF2IF1、NOT2IF1、BUF2IF0 和 NOT2IF0，再次产生了 VA-和 Gra 依赖的基因电路，其逻辑与 AND、NOR 和 IMPLY 门相似，使用新的 BUF2/NOT2 组（图 3D）。
为了将两组 BUFn/NOTn 开关整合到同一个基于三态的基因电路中，确保在并行调节不同输出模块时，各个基因开关之间没有信号串扰是至关重要的。因此，我们共同表达了一个控制分泌性碱性磷酸酶（SEAP）的 NOT1 开关作为第一个报告基因，以及一个控制分泌性纳米荧光素酶（Nluc）的 BUF2 开关作为第二个报告基因。我们还在同一细胞中共同表达了控制 Nluc 的 NOT2 开关和控制 SEAP 的 BUF1 开关。结果表明，每个独立的开关在 Gra 的触发下以高度自主的方式运作，展示了在哺乳动物细胞中强健且无干扰的性能（图 3C）。接下来，我们将两组 Gra 调控的 BUF/NOT 开关连接到 VA 控制的上游模块的输出线上，从而建立了一个能够处理 2 输入 2 输出基因表达逻辑的基于三态的开关板（图 3A）。由于 IF1 和 BUF2/NOT2 在哺乳动物细胞中利用不同的细胞内信号通路，我们首先测试了各关键组分之间是否存在潜在的信号串扰。 实验表明，控制 SEAP 表达的 IF1 开关可以与驱动 Nluc 表达的 BUF2 开关在同一细胞中并行操作。这不仅支持了我们研究的电路性能，还从细胞生物学的角度表明细胞内环磷酸腺苷（cAMP）与 STAT3 信号传导之间可能存在正交性（图 S4）。最后，我们以类似于之前展示的基于 STIF 的 BUF1/NOT1 组的即插即用方式组装了四个三态缓冲器 BUF2IF1、NOT2IF1、BUF2IF0 和 NOT2IF0，再次产生了 VA-和 Gra 依赖的基因电路，其逻辑与 AND、NOR 和 IMPLY 门相似，使用新的 BUF2/NOT2 组（图 3D）。
Figure S4 Orthogonality between VA-inducible cAMP-dependent SEAP expression and grazoprevir-inducible STAT3-dependent Nluc expression in mammalian cells, related to Figure 3
图 S4 在哺乳动物细胞中，VA 诱导的 cAMP 依赖性 SEAP 表达与 grazoprevir 诱导的 STAT3 依赖性 Nluc 表达之间的正交性，相关于图 3
图 S4 在哺乳动物细胞中，VA 诱导的 cAMP 依赖性 SEAP 表达与 grazoprevir 诱导的 STAT3 依赖性 Nluc 表达之间的正交性，相关于图 3
To allow TriLoS-based gene circuits to accept a third input signal C, a new upstream layer of tristate buffers must be created in which C is taken as its new control input (Figure 4A). At the same time, old control input B of the downstream buffer would also serve as the data input of this new upstream buffer (Figure 4A). Thus, C must govern a new set of IF0/IF1-switches that globally control expression of any set of BUF/NOT switches triggered by either B (e.g., VA) or A (e.g., Gra). As an exemplary gene switch operating upstream of the VA-regulated TX layer of gene expression, we chose Cre recombinase48 as the control input C. Cre catalyzes excision of any target DNA segment placed between parallel loxP signals, thus enabling user-defined control of sequence rearrangement within a given transcription unit (Figure 4B). To engineer Cre-dependent IF1 switches, a poly(A) signal flanked by parallel loxP sites is placed downstream of a constitutive promoter driving target gene expression. Thus, the TX terminator in poly(A) will prohibit target gene expression unless the poly(A) signal is removed through pre-programmed Cre-loxP interaction (Figure 4B). Likewise, placing the target gene directly between parallel loxP sites results in a typical IF0 switch triggered by Cre. In such configuration, gene expression is constitutively active until Cre produces an empty transcription unit through mediation of target gene removal (Figure 4B). Based on this scheme, we engineered Cre-driven IF0/IF1 switches with the capability to regulate different output signals when introduced into the same cells, which again fulfilled the basic orthogonality requirement for upstream layer gene switches (Figures 1F and 4B). Then, we assembled all configurations of tristate buffers BUFIF1, NOTIF1, BUFIF0, and NOTIF0 now using C as the new control input to monitor IF0/IF1 switches (Figure 4C). Therefore, BUFIF1 is either mapped through the terms
为了使基于 TriLoS 的基因电路能够接受第三个输入信号 C，必须创建一个新的上游三态缓冲区层，其中 C 作为新的控制输入（图 4A）。与此同时，下游缓冲区的旧控制输入 B 也将作为这个新上游缓冲区的数据输入（图 4A）。因此，C 必须控制一组新的 IF0/IF1 开关，这些开关全局控制由 B（例如，VA）或 A（例如，Gra）触发的任何 BUF/NOT 开关的表达。作为一个在 VA 调控的基因表达 TX 层上游操作的示例基因开关，我们选择了 Cre 重组酶 48 作为控制输入 C。Cre 催化位于平行 loxP 信号之间的任何目标 DNA 片段的切除，从而实现用户定义的在给定转录单位内的序列重排控制（图 4B）。为了设计 Cre 依赖的 IF1 开关，一个被平行 loxP 位点夹住的 poly(A)信号被放置在驱动目标基因表达的组成性启动子下游。因此，poly(A)中的 TX 终止子将禁止目标基因表达，除非通过预编程的 Cre-loxP 相互作用去除 poly(A)信号（图 4B）。 同样，将目标基因直接放置在平行的 loxP 位点之间会导致由 Cre 触发的典型 IF0 开关。在这种配置中，基因表达是持续活跃的，直到 Cre 通过介导目标基因的去除产生一个空的转录单位（图 4B）。基于这一方案，我们设计了 Cre 驱动的 IF0/IF1 开关，能够在引入相同细胞时调节不同的输出信号，这再次满足了上游层基因开关的基本正交性要求（图 1F 和 4B）。然后，我们组装了所有的三态缓冲器配置 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0，现在使用 C 作为新的控制输入来监测 IF0/IF1 开关（图 4C）。因此，BUFIF1 要么通过
为了使基于 TriLoS 的基因电路能够接受第三个输入信号 C，必须创建一个新的上游三态缓冲区层，其中 C 作为新的控制输入（图 4A）。与此同时，下游缓冲区的旧控制输入 B 也将作为这个新上游缓冲区的数据输入（图 4A）。因此，C 必须控制一组新的 IF0/IF1 开关，这些开关全局控制由 B（例如，VA）或 A（例如，Gra）触发的任何 BUF/NOT 开关的表达。作为一个在 VA 调控的基因表达 TX 层上游操作的示例基因开关，我们选择了 Cre 重组酶 48 作为控制输入 C。Cre 催化位于平行 loxP 信号之间的任何目标 DNA 片段的切除，从而实现用户定义的在给定转录单位内的序列重排控制（图 4B）。为了设计 Cre 依赖的 IF1 开关，一个被平行 loxP 位点夹住的 poly(A)信号被放置在驱动目标基因表达的组成性启动子下游。因此，poly(A)中的 TX 终止子将禁止目标基因表达，除非通过预编程的 Cre-loxP 相互作用去除 poly(A)信号（图 4B）。 同样，将目标基因直接放置在平行的 loxP 位点之间会导致由 Cre 触发的典型 IF0 开关。在这种配置中，基因表达是持续活跃的，直到 Cre 通过介导目标基因的去除产生一个空的转录单位（图 4B）。基于这一方案，我们设计了 Cre 驱动的 IF0/IF1 开关，能够在引入相同细胞时调节不同的输出信号，这再次满足了上游层基因开关的基本正交性要求（图 1F 和 4B）。然后，我们组装了所有的三态缓冲器配置 BUFIF1、NOTIF1、BUFIF0 和 NOTIF0，现在使用 C 作为新的控制输入来监测 IF0/IF1 开关（图 4C）。因此，BUFIF1 要么通过
Figure 4 Multi-input biocomputation through additional input layers
图 4 通过额外输入层的多输入生物计算
图 4 通过额外输入层的多输入生物计算
Multi-layered Boolean calculus in individual mammalian cell populations
个体哺乳动物细胞群体中的多层布尔演算
Next, we showcase how all the 16 tristate buffers and BUF/NOT switches (modules 1–6 of Figure 2 accounting for output no. 1; modules 7–10 of Figure 3D for output no. 2; modules 11–18 of Figures 4C and 4D for either output) can be flexibly combined to constitute the computational logics of various biomedical applications of interest. A classical application type is the construction of Boolean calculators, which typically use various pre-programmed computational algorithms to convert two (in the case of half-adders and half-subtractors) or three arbitrary input signals (in the case of full adders and full subtractors) into two optical output signals each representative for a different 2n digit known from electronic computers.43 For example, a half-adder returns the digits sum S (representative for the 20 digit) and carry Y (representative for the 21 digit) through binary addition of the two inputs A (e.g., Gra) and B (e.g., VA), which can be also described by the two mathematical equations “
接下来，我们展示如何将所有 16 个三态缓冲器和 BUF/NOT 开关（图 2 的模块 1-6 对应输出编号 1；图 3D 的模块 7-10 对应输出编号 2；图 4C 和 4D 的模块 11-18 对应任一输出）灵活组合，以构成各种感兴趣的生物医学应用的计算逻辑。一种经典的应用类型是布尔计算器的构建，通常使用各种预编程的计算算法将两个（在半加器和半减器的情况下）或三个任意输入信号（在全加器和全减器的情况下）转换为两个光学输出信号，每个信号代表电子计算机中已知的不同 2n 位数。例如，半加器通过对两个输入 A（例如，Gra）和 B（例如，VA）进行二进制加法，返回数字和 S（代表 20 位数）和进位 Y（代表 21 位数），这也可以用两个数学方程“
接下来，我们展示如何将所有 16 个三态缓冲器和 BUF/NOT 开关（图 2 的模块 1-6 对应输出编号 1；图 3D 的模块 7-10 对应输出编号 2；图 4C 和 4D 的模块 11-18 对应任一输出）灵活组合，以构成各种感兴趣的生物医学应用的计算逻辑。一种经典的应用类型是布尔计算器的构建，通常使用各种预编程的计算算法将两个（在半加器和半减器的情况下）或三个任意输入信号（在全加器和全减器的情况下）转换为两个光学输出信号，每个信号代表电子计算机中已知的不同 2n 位数。例如，半加器通过对两个输入 A（例如，Gra）和 B（例如，VA）进行二进制加法，返回数字和 S（代表 20 位数）和进位 Y（代表 21 位数），这也可以用两个数学方程“
Figure 5 Boolean calculators with the (bio)computational capacity to produce two-bit fluorescent output signals
图 5 具有（生物）计算能力的布尔计算器，可以产生两位荧光输出信号
图 5 具有（生物）计算能力的布尔计算器，可以产生两位荧光输出信号
To map the arithmetic logic of full adders and full subtractors (Figure 5C), we first used a Karnaugh map (K-map)-based logic minimization method49 to convert the algebraic expression of various implicit Boolean truth tables of interest into a most simplified mathematical equation (Figure S5; Table S2). Hence, both sum S and difference D representative for 20 digits in full adders and full subtractors are calculated by the term “
为了映射全加器和全减器的算术逻辑（图 5C），我们首先使用基于卡诺图（K 图）的逻辑最小化方法 49 将各种隐式布尔真值表的代数表达式转换为最简化的数学方程（图 S5；表 S2）。因此，代表全加器和全减器中 20 位的和 S 和差 D 通过术语“
为了映射全加器和全减器的算术逻辑（图 5C），我们首先使用基于卡诺图（K 图）的逻辑最小化方法 49 将各种隐式布尔真值表的代数表达式转换为最简化的数学方程（图 S5；表 S2）。因此，代表全加器和全减器中 20 位的和 S 和差 D 通过术语“
Figure S5 Logic abstraction and simplification of complex computational algorithms, related to Figure 5
图 S5 复杂计算算法的逻辑抽象与简化，相关于图 5
图 S5 复杂计算算法的逻辑抽象与简化，相关于图 5
A prototype 3-input 2-output therapeutic biocomputer for cell-based diabetes treatment in vivo
一种原型 3 输入 2 输出的治疗性生物计算机，用于体内细胞基础的糖尿病治疗
Apart from building cell-based Boolean calculators by seeking for incremental complexity, another major goal of biocomputation driven by multi-layered gene networks is the development of programmable gene- and cell-based therapies that could enable on-demand secretion of therapeutic proteins in vivo.50,51 For example, diabetes mellitus is a chronic, multifactorial, and intractable metabolic disorder that would require different therapeutic regimen during different pathology states.52,53 Each pathology state could be defined by a different input signal, which may precisely coordinate the secretion of different therapeutic output signals according to the actual disease onset. For most type 2 diabetes (T2D) patients in particular, either glucagon-like peptide 1 (GLP-1) or insulin (INS) are eligible treatments.52 However, GLP-1 may become the major therapeutic alternative with the progression of INS resistance, whereas INS administration is the sole treatment solution for INS-deficient type 1 diabetes (T1D) patients or for T2D patients suffering from late-stage β cells exhaustion.54 From a biocomputational perspective, such treatment regimen could be epitomized with a (over)simplified Boolean truth table (Figure 6A); whereas T2D patients can be generally treated by administration of one single drug (e.g., VA; input 1) triggering secretion of both INS and GLP-1, INS-resistant T2D patients may consider the option of using another drug (e.g., Gra; input 2) to trigger exclusive secretion of GLP-1 (Figure 6A). By contrast, T1D patients may prefer yet another regimen where INS (AND NOT GLP-1 secretion) is provided by the co-administration of both drugs (Gra AND VA; Figure 6A). According to K-map-based logic minimization, such hypothetical treatment logic can be reduced to the formula
除了通过寻求增量复杂性构建基于细胞的布尔计算器之外，由多层基因网络驱动的生物计算的另一个主要目标是开发可编程的基因和细胞治疗，这可以实现体内按需分泌治疗蛋白。例如，糖尿病是一种慢性、多因素且难治的代谢紊乱，在不同的病理状态下需要不同的治疗方案。每种病理状态可以通过不同的输入信号来定义，这可能精确协调根据实际疾病发作分泌不同的治疗输出信号。特别是对于大多数 2 型糖尿病（T2D）患者，胰高血糖素样肽 1（GLP-1）或胰岛素（INS）都是合适的治疗方案。然而，随着胰岛素抵抗的进展，GLP-1 可能成为主要的治疗替代方案，而对于胰岛素缺乏的 1 型糖尿病（T1D）患者或处于晚期β细胞耗竭的 T2D 患者，胰岛素给药是唯一的治疗方案。 从生物计算的角度来看，这种治疗方案可以用一个（过于）简化的布尔真值表来概括（图 6A）；而 2 型糖尿病（T2D）患者通常可以通过单一药物（例如，VA；输入 1）的给药来治疗，该药物触发胰岛素（INS）和胰高血糖素样肽-1（GLP-1）的分泌，胰岛素抵抗的 T2D 患者可能考虑使用另一种药物（例如，Gra；输入 2）来触发 GLP-1 的独占分泌（图 6A）。相比之下，1 型糖尿病（T1D）患者可能更倾向于另一种方案，其中通过同时给药两种药物（Gra 和 VA；图 6A）提供胰岛素（并且不分泌 GLP-1）。根据基于 K 图的逻辑最小化，这种假设的治疗逻辑可以简化为公式
除了通过寻求增量复杂性构建基于细胞的布尔计算器之外，由多层基因网络驱动的生物计算的另一个主要目标是开发可编程的基因和细胞治疗，这可以实现体内按需分泌治疗蛋白。例如，糖尿病是一种慢性、多因素且难治的代谢紊乱，在不同的病理状态下需要不同的治疗方案。每种病理状态可以通过不同的输入信号来定义，这可能精确协调根据实际疾病发作分泌不同的治疗输出信号。特别是对于大多数 2 型糖尿病（T2D）患者，胰高血糖素样肽 1（GLP-1）或胰岛素（INS）都是合适的治疗方案。然而，随着胰岛素抵抗的进展，GLP-1 可能成为主要的治疗替代方案，而对于胰岛素缺乏的 1 型糖尿病（T1D）患者或处于晚期β细胞耗竭的 T2D 患者，胰岛素给药是唯一的治疗方案。 从生物计算的角度来看，这种治疗方案可以用一个（过于）简化的布尔真值表来概括（图 6A）；而 2 型糖尿病（T2D）患者通常可以通过单一药物（例如，VA；输入 1）的给药来治疗，该药物触发胰岛素（INS）和胰高血糖素样肽-1（GLP-1）的分泌，胰岛素抵抗的 T2D 患者可能考虑使用另一种药物（例如，Gra；输入 2）来触发 GLP-1 的独占分泌（图 6A）。相比之下，1 型糖尿病（T1D）患者可能更倾向于另一种方案，其中通过同时给药两种药物（Gra 和 VA；图 6A）提供胰岛素（并且不分泌 GLP-1）。根据基于 K 图的逻辑最小化，这种假设的治疗逻辑可以简化为公式
Figure 6 Therapeutic biocomputer for cell-based diabetes treatment in vivo
图 6 体内细胞基础糖尿病治疗的治疗生物计算机
图 6 体内细胞基础糖尿病治疗的治疗生物计算机
A similar therapeutic algorithm can also be drawn with three different input signals (Figure 6C). In fact, a related treatment scenario in the future may first require ex vivo manufacturing cell-based implants harboring such custom-designed computational logics, followed by life-long implantation of the engineered cells into a patient to allow different environmental signals (e.g., drug intake) to flexibly shuttle between different drug secretion programs from a same implant device. In this regard, using a third trigger signal C to provide exclusive INS (AND NOT GLP-1) secretion logics would be of higher practical significance in direct comparison with using a mixture of two separate trigger signals A and B (as was in the case of Figure 6A). Because T1D and β cell exhaustion are irreversible complications, a GLP-1 production module may no longer be required from the moment on when a patient starts an INS-only regimen. Therefore, a more patient-compliant truth table should be as follows (Figure 6C); in the absence of a third trigger signal (C = 0; e.g., during any stage of T2D), one trigger signal A (e.g., VA) controls GLP-1 secretion, whereas another trigger signal B (e.g., Gra) controls INS secretion—allowing a patient to flexibly adjust whether only one (INS is produced if A = 0 and B = 1; GLP-1 is produced if A = 1 and B = 0) or both therapeutic proteins shall be secreted into the bloodstream (INS and GLP-1 are produced if A = 1 and B = 1). However, because GLP-1 may never be a treatment option again for this patient upon diagnosis of T1D, a one-time exposure to a third trigger signal C may be useful to irreversibly “eliminate” the VA-regulated GLP-1 production module (if C = 1; INS follows the value of B, and A falls into a high-impedance state Z), which allows Gra-regulated INS production to remain his/her sole treatment option. This not only avoids long-term administration of excessive numbers of drugs (as would be in the case of a 2-input 2-output option shown in Figure 6A) but may also remove certain components (e.g., VA) from a patient’s food restriction regimen.
一个类似的治疗算法也可以通过三种不同的输入信号绘制（图 6C）。事实上，未来相关的治疗场景可能首先需要体外制造基于细胞的植入物，这些植入物具有定制设计的计算逻辑，然后将工程化的细胞终身植入患者体内，以便不同的环境信号（例如，药物摄入）能够灵活地在同一植入设备的不同药物分泌程序之间切换。在这方面，使用第三个触发信号 C 来提供独占的胰岛素（AND NOT GLP-1）分泌逻辑，与使用两种独立触发信号 A 和 B 的混合（如图 6A 所示）相比，具有更高的实际意义。由于 1 型糖尿病和β细胞耗竭是不可逆的并发症，因此从患者开始仅使用胰岛素的方案时，GLP-1 生产模块可能不再需要。因此，更符合患者需求的真值表应如下所示（图 6C）；在没有第三个触发信号的情况下（C = 0；例如，在任何 2 型糖尿病阶段），一个触发信号 A（例如，VA）控制 GLP-1 分泌，而另一个触发信号 B（例如。Gra）控制胰岛素（INS）分泌——允许患者灵活调整是否仅分泌一种（当 A = 0 且 B = 1 时产生 INS；当 A = 1 且 B = 0 时产生 GLP-1）或两种治疗蛋白质（当 A = 1 且 B = 1 时产生 INS 和 GLP-1）进入血液。然而，由于在诊断为 1 型糖尿病（T1D）后，GLP-1 可能再也无法成为该患者的治疗选择，因此对第三个触发信号 C 的一次性暴露可能有助于不可逆地“消除”VA 调节的 GLP-1 生产模块（如果 C = 1；INS 遵循 B 的值，而 A 进入高阻抗状态 Z），这使得 Gra 调节的 INS 生产成为他/她唯一的治疗选择。这不仅避免了长期使用过量药物（如图 6A 所示的 2 输入 2 输出选项），而且可能还会将某些成分（例如 VA）从患者的饮食限制方案中去除。
一个类似的治疗算法也可以通过三种不同的输入信号绘制（图 6C）。事实上，未来相关的治疗场景可能首先需要体外制造基于细胞的植入物，这些植入物具有定制设计的计算逻辑，然后将工程化的细胞终身植入患者体内，以便不同的环境信号（例如，药物摄入）能够灵活地在同一植入设备的不同药物分泌程序之间切换。在这方面，使用第三个触发信号 C 来提供独占的胰岛素（AND NOT GLP-1）分泌逻辑，与使用两种独立触发信号 A 和 B 的混合（如图 6A 所示）相比，具有更高的实际意义。由于 1 型糖尿病和β细胞耗竭是不可逆的并发症，因此从患者开始仅使用胰岛素的方案时，GLP-1 生产模块可能不再需要。因此，更符合患者需求的真值表应如下所示（图 6C）；在没有第三个触发信号的情况下（C = 0；例如，在任何 2 型糖尿病阶段），一个触发信号 A（例如，VA）控制 GLP-1 分泌，而另一个触发信号 B（例如。Gra）控制胰岛素（INS）分泌——允许患者灵活调整是否仅分泌一种（当 A = 0 且 B = 1 时产生 INS；当 A = 1 且 B = 0 时产生 GLP-1）或两种治疗蛋白质（当 A = 1 且 B = 1 时产生 INS 和 GLP-1）进入血液。然而，由于在诊断为 1 型糖尿病（T1D）后，GLP-1 可能再也无法成为该患者的治疗选择，因此对第三个触发信号 C 的一次性暴露可能有助于不可逆地“消除”VA 调节的 GLP-1 生产模块（如果 C = 1；INS 遵循 B 的值，而 A 进入高阻抗状态 Z），这使得 Gra 调节的 INS 生产成为他/她唯一的治疗选择。这不仅避免了长期使用过量药物（如图 6A 所示的 2 输入 2 输出选项），而且可能还会将某些成分（例如 VA）从患者的饮食限制方案中去除。
To implement such 3-input 2-output treatment logics, the corresponding truth table can be broken down to the mathematical equations
为了实现这种 3 输入 2 输出的处理逻辑，相应的真值表可以分解为数学方程
为了实现这种 3 输入 2 输出的处理逻辑，相应的真值表可以分解为数学方程
Discussion 讨论
A central goal of synthetic biology and modern biomedicine is the ability to program cellular machines as novel therapeutic alternatives of the future.9 However, engineering of large-scale gene circuits in mammalian cells is markedly lagging behind the progress of counterpart systems that have been described in prokaryotes and yeast.23,27,58 Although engineering principles of such unicellular organisms are already progressing toward standardized and automated design strategies,27 multi-layered biocomputation in mammalian cells is still confined to empiric approaches that primarily result from arduous trial-and-error cycles.10,11 In recent years, Boolean calculators of highest computing capacity (including full-adders, full-subtractors, and even decoders) were elegantly achieved at single layers of gene expression,30,59,60,61 thus setting new standards for complex logic computation in mammalian cells. By contrast, engineering of multi-layered transcription-translation (TX/TL)-based gene networks capable of responding to user- or environmentally defined cues and processing those signals across multiple levels of gene expression is currently prohibited by the lack of rational and systematic design principles applicable for eukaryotic gene regulation.30,62,63 In fact, biological systems are inherently hierarchical in a way that most sophisticated cellular behavior might only be encoded through multi-level control across different gene expression layers, but conventional strategies for producing such higher-order circuits are not as modular as once anticipated.28 To address this problem, we show that tristate buffers are not only compatible with but also particularly suited for programming gene circuits in biological systems. First, tristate buffers are organized by upstream and downstream switch elements by definition, which favors (re)construction of network topologies in a biological relevant manner because gene expression is hierarchically regulated in nature. Hence, we provide modular engineering principles for multi-level gene regulation involving interdependent genomic, TX/TL control (Figures 1, 2, 3, and 4) and show how these modules can be flexibly combined to provide tailor-designed cellular algorithms for various biomedical applications, such as complex Boolean calculus (Figure 5) or cell-based therapies (Figure 6). Evidently, such design logics can be flexibly adopted and expanded to also include other levels of gene expression in mammalian cells in future work. Second, the high-impedance state Z of tristate buffers could have been a missing link for robust, resource-efficient, and interference-free biocomputation. Although energetically active 0-states are particularly difficult to engineer because basal or non-specific expression can never be fully avoided in a biological context, gene circuits with 0 (OFF state) replaced by the physically unplugged Z state (inactive expression) might be an attractive solution. For example, a conventional AND gate would require substantial engineering effort to keep the output signal in three out of four conditions silent. Using tristate buffers, two of these inactive conditions would be warranted by the high-impedance Z state (Figure 2); by temporarily dismounting genetic modules that are not being used by a particular operation, significant portions of background activities are therefore attenuated by design, which might ultimately maximize the efficiency of signal processing and transmission. Third, the logic architecture of tristate buffers facilitates flexible mapping of various mathematical terms for regulatory interactions. The four tristate buffers BUFIF1, BUFIF0, NOTIF1, and NOTIF0 representative for each combination between individual BUF/NOT- (downstream switches) and IF0/IF1-switches (upstream switches) depict a general scheme of how to implement specific logic interrelations that are defined by different input signals. For example, BUFIF1 buffers describe all positive regulations between two trigger-inducible gene switches each driven by a different signal, such as B regulating A (
合成生物学和现代生物医学的一个核心目标是能够将细胞机器编程为未来的新型治疗替代方案。 9 然而，在哺乳动物细胞中大规模基因电路的工程化明显滞后于在原核生物和酵母中描述的相应系统的进展。 23 27 58 尽管这些单细胞生物的工程原理已经朝着标准化和自动化设计策略发展， 27 但哺乳动物细胞中的多层生物计算仍然局限于主要源于艰苦的试错循环的经验方法。 10 11 近年来，最高计算能力的布尔计算器（包括全加器、全减器甚至解码器）在单层基因表达中优雅地实现， 30 59 60 61 从而为哺乳动物细胞中的复杂逻辑计算设定了新标准。 相较之下，基于多层转录-翻译（TX/TL）的基因网络的工程设计，能够响应用户或环境定义的信号并在多个基因表达层次上处理这些信号，目前受到缺乏适用于真核生物基因调控的合理和系统设计原则的限制。实际上，生物系统本质上是分层的，最复杂的细胞行为可能仅通过不同基因表达层次的多级控制进行编码，但传统的生产这种高阶电路的策略并不像预期的那样模块化。为了解决这个问题，我们展示了三态缓冲器不仅与生物系统中的基因电路编程兼容，而且特别适合。首先，三态缓冲器的组织由上游和下游开关元件定义，这有利于以生物相关的方式（重新）构建网络拓扑，因为基因表达在自然界中是分层调控的。 因此，我们提供了多层次基因调控的模块化工程原则，涉及相互依赖的基因组、转录/翻译控制（图 1、2、3 和 4），并展示了这些模块如何灵活组合，以提供为各种生物医学应用量身定制的细胞算法，例如复杂的布尔计算（图 5）或基于细胞的疗法（图 6）。显然，这种设计逻辑可以灵活采用并扩展，以在未来的工作中包括哺乳动物细胞中基因表达的其他层次。其次，三态缓冲器的高阻抗状态 Z 可能是实现稳健、资源高效和无干扰生物计算的缺失环节。尽管在生物环境中，基底或非特异性表达永远无法完全避免，使得能量活跃的 0 状态特别难以工程化，但用物理上未插入的 Z 状态（非活性表达）替代 0（关闭状态）的基因电路可能是一个有吸引力的解决方案。例如，传统的与门需要大量工程努力，以保持在四种条件中的三种输出信号静默。 使用三态缓冲器，这两种非激活状态将由高阻抗 Z 状态（图 2）所保证；通过暂时卸载不被特定操作使用的基因模块，因此设计上显著减弱了背景活动的部分，这可能最终最大化信号处理和传输的效率。第三，三态缓冲器的逻辑架构促进了对各种数学术语在调控相互作用中的灵活映射。四个三态缓冲器 BUFIF1、BUFIF0、NOTIF1 和 NOTIF0 代表了个别 BUF/NOT-（下游开关）和 IF0/IF1 开关（上游开关）之间的每种组合，描绘了如何实现由不同输入信号定义的特定逻辑关系的一般方案。例如，BUFIF1 缓冲器描述了两个由不同信号驱动的触发诱导基因开关之间的所有正调控，例如 B 调控 A（
合成生物学和现代生物医学的一个核心目标是能够将细胞机器编程为未来的新型治疗替代方案。 9 然而，在哺乳动物细胞中大规模基因电路的工程化明显滞后于在原核生物和酵母中描述的相应系统的进展。 23 27 58 尽管这些单细胞生物的工程原理已经朝着标准化和自动化设计策略发展， 27 但哺乳动物细胞中的多层生物计算仍然局限于主要源于艰苦的试错循环的经验方法。 10 11 近年来，最高计算能力的布尔计算器（包括全加器、全减器甚至解码器）在单层基因表达中优雅地实现， 30 59 60 61 从而为哺乳动物细胞中的复杂逻辑计算设定了新标准。 相较之下，基于多层转录-翻译（TX/TL）的基因网络的工程设计，能够响应用户或环境定义的信号并在多个基因表达层次上处理这些信号，目前受到缺乏适用于真核生物基因调控的合理和系统设计原则的限制。实际上，生物系统本质上是分层的，最复杂的细胞行为可能仅通过不同基因表达层次的多级控制进行编码，但传统的生产这种高阶电路的策略并不像预期的那样模块化。为了解决这个问题，我们展示了三态缓冲器不仅与生物系统中的基因电路编程兼容，而且特别适合。首先，三态缓冲器的组织由上游和下游开关元件定义，这有利于以生物相关的方式（重新）构建网络拓扑，因为基因表达在自然界中是分层调控的。 因此，我们提供了多层次基因调控的模块化工程原则，涉及相互依赖的基因组、转录/翻译控制（图 1、2、3 和 4），并展示了这些模块如何灵活组合，以提供为各种生物医学应用量身定制的细胞算法，例如复杂的布尔计算（图 5）或基于细胞的疗法（图 6）。显然，这种设计逻辑可以灵活采用并扩展，以在未来的工作中包括哺乳动物细胞中基因表达的其他层次。其次，三态缓冲器的高阻抗状态 Z 可能是实现稳健、资源高效和无干扰生物计算的缺失环节。尽管在生物环境中，基底或非特异性表达永远无法完全避免，使得能量活跃的 0 状态特别难以工程化，但用物理上未插入的 Z 状态（非活性表达）替代 0（关闭状态）的基因电路可能是一个有吸引力的解决方案。例如，传统的与门需要大量工程努力，以保持在四种条件中的三种输出信号静默。 使用三态缓冲器，这两种非激活状态将由高阻抗 Z 状态（图 2）所保证；通过暂时卸载不被特定操作使用的基因模块，因此设计上显著减弱了背景活动的部分，这可能最终最大化信号处理和传输的效率。第三，三态缓冲器的逻辑架构促进了对各种数学术语在调控相互作用中的灵活映射。四个三态缓冲器 BUFIF1、BUFIF0、NOTIF1 和 NOTIF0 代表了个别 BUF/NOT-（下游开关）和 IF0/IF1 开关（上游开关）之间的每种组合，描绘了如何实现由不同输入信号定义的特定逻辑关系的一般方案。例如，BUFIF1 缓冲器描述了两个由不同信号驱动的触发诱导基因开关之间的所有正调控，例如 B 调控 A（
Hence, this work may establish multi-layered gene networks as another dominant paradigm of biocomputation, suggesting that a concept where various biomedical solutions can be flexibly designed or programmed from scratch could soon become a real practice.43,64 Currently, there are two classes of biocomputer architectures30: (1) single-layered systems that capitalize on naturally evolved or de novo designed orthogonality and target specificities between selected sets of recombinases,12 proteases,3,65 or protein-protein interactions4 to generate pre-defined target gene configurations within a same layer of gene expression and (2) multi-layered systems involving interconnected gene regulation between multiple gene expression stages as shown in this work. Although single-layered systems are more compact and generally favorable to create regulation logics of the highest computational complexity, multi-layered regulation networks are designed to mimic natural systems and may be the architecture of choice when responsiveness to external control signals is the engineering goal.30,64 Although single-layered biocomputers are already moving toward real-world applications, such as providing programmable memory barcodes for mechanistic studies, such as lineage tracing,66,67 the application potential of multi-layered systems remained limited for several reasons. For example, due to the aforementioned lack of systematic and modular design principles, many logic modules of conventional multi-layered approaches must be individually redesigned to yield the expected profiles of gene expression, resulting in high system complexity and incremental computational burden. Following the blueprint adopted from digital electronics, in particular, a multi-layered XOR gate is typically produced through addition of two parallel NIMPLY gates, which, in turn, are built on interconnected AND and NOT gates.10,11 Because each logic operation has inherent gate propagation delays, implementation of such complex gene circuits harboring three logic layers (AND/NOT → NIMPLY → XOR) would already reach the current limit of available engineering space of single cells. Thus, half-adders and half-subtractors consisting of an XOR gate paired with either AND or NIMPLY were the most complex multi-layered gene networks that could be engineered within mammalian cells.10,29 Engineering of multi-layered gene circuits with the computational capacity of full-adders and full-subtractors, which may comprise at least two half-adders or three XOR gates, might remain elusive when sticking with conventional layering strategies. Here, using TriLoS, we show that (1) engineering of OR, IMPLY, and XOR logics using tristate buffers is no longer of incremental complexity but rather has the same simplicity levels (Figure 2), (2) half-adders and half-subtractors can now be designed to “only” consist of three tristate buffers (Figures 5A and 5B), and (3) even full adders and full subtractors can be readily engineered to comprise only two layers of interconnected tristate buffers (Figure 5D), where parallel line-ups of multiple buffers do not add gate delays to the entire layer. Because gene switches designed to remain in an inactive mode (high-Z state) are electrically disconnected by design, signals would go through at most two tristate buffers for each calculation. Such savings in the maximally available engineering space within single-cell populations were fundamental for breaking the ground to achieve 3-input 2-output computational logics using multi-layered gene circuits in single-cell populations.
因此，这项工作可能将多层基因网络确立为生物计算的另一种主导范式，暗示着一种概念，即各种生物医学解决方案可以灵活地从零开始设计或编程，可能很快会成为现实。目前，有两类生物计算机架构： (1) 单层系统，利用自然进化或新设计的正交性和选定重组酶、蛋白酶或蛋白质-蛋白质相互作用之间的特异性，生成同一基因表达层内的预定义目标基因配置； (2) 多层系统，涉及多个基因表达阶段之间的互联基因调控，如本工作所示。尽管单层系统更为紧凑，通常更有利于创建最高计算复杂度的调控逻辑，但多层调控网络旨在模拟自然系统，当响应外部控制信号是工程目标时，可能是首选架构。 尽管单层生物计算机已经朝着实际应用迈进，例如为机制研究提供可编程的记忆条形码，如谱系追踪，但多层系统的应用潜力由于几个原因仍然有限。例如，由于上述缺乏系统性和模块化设计原则，许多传统多层方法的逻辑模块必须单独重新设计，以产生预期的基因表达谱，导致系统复杂性高和计算负担增加。根据数字电子学采用的蓝图，特别是，多层 XOR 门通常通过添加两个并行的 NIMPLY 门来实现，而这两个门又是基于相互连接的 AND 和 NOT 门构建的。由于每个逻辑操作都有固有的门传播延迟，实施这种包含三个逻辑层（AND/NOT → NIMPLY → XOR）的复杂基因电路将已经达到单细胞可用工程空间的当前极限。 因此，由一个异或门与一个与门或非门配对组成的半加器和半减器是可以在哺乳动物细胞中工程化的最复杂的多层基因网络。 10 29 使用传统的分层策略，具有全加器和全减器计算能力的多层基因电路的工程化，可能仍然难以实现，这些电路可能至少包含两个半加器或三个异或门。在这里，使用 TriLoS，我们展示了（1）使用三态缓冲器工程化或、非门和异或逻辑不再是增量复杂性，而是具有相同的简单性水平（图 2），（2）半加器和半减器现在可以设计为“仅”由三个三态缓冲器组成（图 5A 和 5B），以及（3）甚至全加器和全减器也可以轻松工程化，仅由两层互连的三态缓冲器组成（图 5D），其中多个缓冲器的并行排列不会给整个层增加门延迟。 由于设计为保持在非活动模式（高-Z 状态）的基因开关在设计上是电气隔离的，因此每次计算信号最多会经过两个三态缓冲器。这种在单细胞群体中最大可用工程空间的节省对于利用多层基因电路在单细胞群体中实现 3 输入 2 输出计算逻辑的突破至关重要。
因此，这项工作可能将多层基因网络确立为生物计算的另一种主导范式，暗示着一种概念，即各种生物医学解决方案可以灵活地从零开始设计或编程，可能很快会成为现实。目前，有两类生物计算机架构： (1) 单层系统，利用自然进化或新设计的正交性和选定重组酶、蛋白酶或蛋白质-蛋白质相互作用之间的特异性，生成同一基因表达层内的预定义目标基因配置； (2) 多层系统，涉及多个基因表达阶段之间的互联基因调控，如本工作所示。尽管单层系统更为紧凑，通常更有利于创建最高计算复杂度的调控逻辑，但多层调控网络旨在模拟自然系统，当响应外部控制信号是工程目标时，可能是首选架构。 尽管单层生物计算机已经朝着实际应用迈进，例如为机制研究提供可编程的记忆条形码，如谱系追踪，但多层系统的应用潜力由于几个原因仍然有限。例如，由于上述缺乏系统性和模块化设计原则，许多传统多层方法的逻辑模块必须单独重新设计，以产生预期的基因表达谱，导致系统复杂性高和计算负担增加。根据数字电子学采用的蓝图，特别是，多层 XOR 门通常通过添加两个并行的 NIMPLY 门来实现，而这两个门又是基于相互连接的 AND 和 NOT 门构建的。由于每个逻辑操作都有固有的门传播延迟，实施这种包含三个逻辑层（AND/NOT → NIMPLY → XOR）的复杂基因电路将已经达到单细胞可用工程空间的当前极限。 因此，由一个异或门与一个与门或非门配对组成的半加器和半减器是可以在哺乳动物细胞中工程化的最复杂的多层基因网络。 10 29 使用传统的分层策略，具有全加器和全减器计算能力的多层基因电路的工程化，可能仍然难以实现，这些电路可能至少包含两个半加器或三个异或门。在这里，使用 TriLoS，我们展示了（1）使用三态缓冲器工程化或、非门和异或逻辑不再是增量复杂性，而是具有相同的简单性水平（图 2），（2）半加器和半减器现在可以设计为“仅”由三个三态缓冲器组成（图 5A 和 5B），以及（3）甚至全加器和全减器也可以轻松工程化，仅由两层互连的三态缓冲器组成（图 5D），其中多个缓冲器的并行排列不会给整个层增加门延迟。 由于设计为保持在非活动模式（高-Z 状态）的基因开关在设计上是电气隔离的，因此每次计算信号最多会经过两个三态缓冲器。这种在单细胞群体中最大可用工程空间的节省对于利用多层基因电路在单细胞群体中实现 3 输入 2 输出计算逻辑的突破至关重要。
Irrespective of whether Boolean calculators were assembled with single3,12,63 or multiple layers of gene expression,10 fluorescent reporter proteins are commonly produced as the output signal (Figure 5). Although fluorescent signals are highly practical and suitable to visualize a “digitalized” regulation profile, engineering of biocomputers with the capability to drive expression and secretion of therapeutic proteins would open the door to numerous hitherto unexplored biomedical applications. To this end, we not only show that INS production levels can be precisely titrated into a disease-specific efficacy window at a cellular level (Figure S2) and in mice (Figures 6D and 6E) but also introduce a futuristic treatment concept that interprets any given medical problem with a mindset of software programmers (Figures 6A and 6C). In this notion, the logics of a preferred drug secretion regimen may be systematically denoted as a disease-specific “therapeutic algorithm” either in form of a Boolean truth table or a simplified mathematical equation shown in Figure 6. This logic formula can then be flexibly mapped by choosing a suitable combination of tristate buffers (modules 1–4 and 7–18; Figures 2, 3, and 4) and BUF/NOT switches (modules 5–6; Figure 2) in a modular and genuinely plug-and-play manner to fulfill various regulatory tasks of interest (Figures 6A and 6C). Here, we used multi-stage diabetes mellitus as an exemplary medical condition that may require secretion of different therapeutic proteins (output signals) at different physiological contexts (input signals). We also describe two possible treatment scenarios allowing either two (Figure 6A) or three external input signals (Figure 6C) to coordinate INS (during T1D) and GLP-1 secretion (during T2D) from a same batch of implanted cell-based therapeutics in vivo. Though the clinical significance of the therapeutic algorithms proposed in this work may not be definitive from the viewpoint of diabetes treatment, we rather emphasize that a concept allowing patients, doctors, and scientists to co-develop a personalized implant device reminiscent of a “therapeutic biocomputer” with adjustable logic operations in vivo could be technically feasible. Thus, TriLoS not only enables multi-layered gene networks to advance toward high computational capacity at a molecular level (Figure 5) but was further instrumental to move multi-input, multi-output biocomputation strategies toward applications of modern gene- and cell-based therapies (Figure 6). Theoretically, a similar cell therapy approach could also be realized using a single-layered programming strategy, just like how TriLoS has now enabled multi-layered gene networks to also achieve complex Boolean calculus that was previously dominated by single-layered strategies. Thus, the next generation of biocomputation may combine the best of both classes to set new unexplored goals in modern biomedicine.30
Taken together, this work describes a modular blueprint that allows engineering of complex gene networks consisting of multiple interconnected switches that operate across different layers of gene expression. Electronic tristate buffers are typically found in most bus drivers and registers of integrated circuits such as microprocessors, random access memory (RAM), and peripheral memory devices.31 Although tristate buffers are technically suited for but not primarily used to map logic circuits in digital electronics, we show that the unique advantages of tristate architectures are almost tailor-made for biological engineering. In fact, using a Hill-type model of TX/TL feedback systems,41,42 we confirm that a tristate-based approach can (1) significantly reduce the total number of biochemical reactions defining the overall size of the circuit, (2) minimize the total number of logic operations accounting for individual gate delays, and possibly for these reasons, (3) eventually achieve a more complete set of logic synthesis when directly compared with a state-of-the-art TX/TL network used for multi-level mammalian gene regulation10 (Table S1). Compared with conventional strategies of sequential layering of Boolean logic gates, tristate buffers are organized in a massively parallel manner wired to a shared output signal. This topology would significantly reduce the net distance a signal must travel through the entire circuit before reaching the output (Figure 5C). Such reduction in the effective circuit size is of particular importance within a biological context because signal processing times of gene expression is typically too slow to tolerate ineffective coding or excessive gate delays. Although minimization of gate propagation delay might not be a limiting factor for the programming of electronic computers, we show that gene circuits built on the basis of tristate buffers (instead of Boolean logic gates) can lead to massive savings in engineering effort and cellular resources for biocomputation. Additionally, engineering of gene circuits using tristate buffers permits that not every genetic component within the whole circuit must be strictly orthogonal to each other. Due to the parallel switching principle, BUF and NOT switches within a same output channel can never co-exist at the same time (Figure 3A). Thus, gene switches that are antagonistic, but not orthogonal to each other, would be most suited to take up these modules. In our example, we show that although different sets of Gra-regulated BUFn/NOTn switches should ideally be orthogonal to each other (Figures 1C and 3C), there is no orthogonality requirement for BUF and NOT switches within each individual channel
Limitations of the study
Although TriLoS provided a long-sought design strategy that enables systematic assembly of multi-layered gene networks with high modularity and flexibility, the overall complexity achievable within single mammalian cells was not significantly expanded. Using TriLoS, full adder and full subtractor-like gene circuits were finally realized for multi-layered gene networks within single-cell populations (Figure 5), but this kind of Boolean calculus has already been described with biocomputers running on a single-layered genetic architecture.3,12,63 Also, our experimental data show that only a minor fraction of transfected cells produced the expected fluorescence profile of EGFP and mCherry expression (Figures S8 and S9), which was admittedly lower than shown in a previously reported recombinase-based switchboard.12 Thus, in terms of overall biocomputational complexity, multi-layered biocomputers may remain inferior to single-layered counterpart systems. Nevertheless, we believe that future direction of this field should seek for application-oriented biocomputer designs that combine the best of both classes and address important problems in biomedicine. To emphasize this point, we used diabetes mellitus as an exemplary disease that could be approached with a mindset of 3-input 2-output biocomputation. Following such hypothesis, any user-defined treatment strategy could be summarized into a Boolean-like truth table that defines the preferred logics between drug intake (input signals) and therapeutic response (output signals). Such truth table is then broken down into a most simplified mathematical expression using K-map and systematically matched by modular assembly of tristate buffers. Evidently, the therapeutic algorithms for diabetes proposed in this work are not medically proven but rather based on our common understanding of this multifactorial disease (Figure 6). Also, such a biocomputer-centered treatment strategy is neither a unique feature of TriLoS nor for multi-layered gene networks in general and could be also technically accomplished with other bioengineering strategies, such as single-layered biocomputational logics. Overall, the aim of this study was the establishment of a standardized programming language that is custom-designed for multi-layered gene networks, whereas the types of applications used in this work to showcase robustness and performance (Boolean calculators, Figure 5; cell-based therapy, Figure 6) can be theoretically also achieved with other (bio)computational designs.
Figure S8 Flow cytometric data, related to Figure 5E
Figure S9 Flow cytometric data, related to Figure 5F
STAR★Methods
Key resources table
| REAGENT or RESOURCE | SOURCE | IDENTIFIER |
|---|---|---|
| Chemicals, peptides, and recombinant proteins | ||
| KOD One PCR Master Mix | Toyobo Inc. (Osaka, Japan) | Cat# KMM-201 |
| T4 DNA Ligase | New England Biolabs (Beverly, MA) | Cat# M0202L |
| Seamless Cloning Kit | Beyotime Biotechnology (Shanghai, China) | Cat# D7010M |
| Dimethyl sulfoxide (DMSO) | Solarbio Life Sciences (Beijing, China) | Cat# D8371 |
| Grazoprevir | Selleck Chemicals (Houston, TX) | Cat# S3728 |
| Tamoxifen | MedChemExpress (Monmouth Junction, NJ) | Cat# HY-13757A |
| 4-Nitrophenyl phosphate disodium salt hexahydrate (pNPP) | Aladdin Biochemical Technology (Shanghai, China) | Cat# P109039 |
| Diethanolamine (DEA) | Macklin Inc. (Shanghai, China) | Cat# D807525 |
| Ethanol anhydrous (EtOH) | Macklin Inc. (Shanghai, China) | Cat# E809056 |
| Polyethyleneimine MAX (PEI) | Polysciences (Eppelheim, Germany) | Cat# 24765 |
| Isopropanol | Rhawn Chemicals (Shanghai, China) | Cat# R018247 |
| Vanillic Acid | Rhawn Chemicals (Shanghai, China) | Cat# R017640 |
| Glycerol | Sangon Biotech (Shanghai, China) | Cat# A501745 |
| L-Homoarginine hydrochloride | Sangon Biotech (Shanghai, China) | Cat# A602842 |
| Magnesium chloride hexahydrate (MgCl2) | Sangon Biotech (Shanghai, China) | Cat# A610328 |
| Sodium citrate tribasic dihydrate | Sangon Biotech (Shanghai, China) | Cat# A100101 |
| Calcium chloride anhydrous (CaCl2) | Sinopharm Chemical Reagent (Shanghai, China) | Cat# 10005861 |
| Sodium Chloride (NaCl) | Sinopharm Chemical Reagent (Shanghai, China) | Cat# 10019318 |
| Citric acid | Sinopharm Chemical Reagent (Shanghai, China) | Cat# 30196768 |
| Puromycin dihydrochloride | Thermo Fisher Scientific (Waltham, MA) | Cat# A1113803 |
| Zeocin selection reagent | Thermo Fisher Scientific (Waltham, MA) | Cat# R25001 |
| Blasticidin S HCl | Thermo Fisher Scientific (Waltham, MA) | Cat# R21001 |
| Streptozotocin STZ | Sigma-Aldrich (St. Louis, MO) | Cat# S0130 |
| Dulbecco’s modified Eagle’s medium (DMEM) | Thermo Fisher Scientific (Waltham, MA) | Cat# 12100046 |
| Gibco fetal bovine serum (FBS) | Thermo Fisher Scientific (Waltham, MA) | Cat# 10099141; Lot# 2177370 |
| Penicillin/streptomycin solution (PenStrep) | Beyotime Biotechnology (Shanghai, China) | Cat# ST488 |
| 0.05% Trypsin-EDTA | Sangon Biotech (Shanghai, China) | Cat# A610629-0050; Lot# F319BA0030 |
| Lipofectamine 3000 reagent | Thermo Fisher Scientific (Waltham, MA) | Cat# L3000015 |
| Critical commercial assays | ||
| Chemiluminescent SEAP assay | Abcam (Cambridge, UK) | Cat# AB133077 |
| Nano-Glo® Luciferase Assay System | Promega (Madison, WI) | Cat# N1120 |
| Mercodia Mouse Insulin ELISA kit | Mercodia (Uppsala, Sweden) | Cat# 10-1247-01 |
| Human IgG Fc ELISA kit | Immunology Consultants Laboratory (Portland, OR) | Cat# E-80G |
| Bioactive GLP-1 ELISA kit | Merck Millipore (Darmstadt, Germany) | Cat# EGLP-35K |
| Experimental models: Cell lines | ||
| HEK-293 | ATCC | Cat# CRL-3216 |
| N2A | ATCC | Cat# CRL-131 |
| RD | ATCC | Cat# CCL-136 |
| Experimental models: Organisms/strains | ||
| C57BL/6J mice | Vital River Laboratories (Beijing, China) | C57BL/6J |
| db/db mice | Shanghai Institutes for Biological Sciences Shanghai Laboratory Animal Center (SLACCAS) (Shanghai, China) | BKS.Cg-+Leprdb/+Leprdb/JclSlac |
| Recombinant DNA | ||
| pcDNA3.1(+) | Thermo Fisher Scientific (Waltham, MA) | Cat# V790-20 |
| pMD2.G | Addgene | Cat# 12259 |
| psPAX2 | Addgene | Cat# 12260 |
| pCMV-T7-SB100 | Addgene | Cat# 34879 |
| pCreERT | Addgene | Cat# 13777 |
| Software and algorithms | ||
| OPLENIC | Mayduly Science Equipment Co., Ltd | version x64 |
| FlowJo™ | Becton Dickinson | version 10 |
| GraphPad Prism | GraphPad Software, LLC | version 9.5.0 |
Resource availability
Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Mingqi Xie (xiemingqi@westlake.edu.cn).
Materials availability
All unique reagents generated in this study (plasmids generated in this work, stable cell lines etc.) are available upon reasonable request from the lead contact.
The data that support the findings of this study are available on reasonable request to the lead corresponding author (M.X.).
Data and code availability
•
All datasets generated and analyzed during the current study are contained within the manuscript and accompanying supplemental data figures and tables.
•
This paper does not report original code.
•
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
Experimental model and study participant details
Animal experiments were performed according to the protocol (Protocol ID: 20-009-XMQ) approved by the Institutional Animal Care and Use Committee (IACUC) of Westlake University and in accordance with the Animal Care Guidelines of the Ministry of Science and Technology of the People's Republic of China.
8-week-old male C57BL/6J mice were ordered through the Laboratory Animal Resources Center (LARC) of Westlake University.
8-week-old male BKS. Lepr (db/db) mice were purchased from Shanghai Institutes for Biological Sciences Shanghai Laboratory Animal Center (SLACCAS) (Shanghai, China).
The insulin-deficient T1D mouse model was generated by daily injection of freshly diluted STZ (50mg/kg in 200μL ice-cold sodium citrate buffer [0.1M citric acid and 0.1M sodium citrate in a 1:1.3 ratio, pH 4.5]) to fasted wild-type C57BL/6J mice for five consecutive days. Chronic fasting hyperglycemia (>15 mM) developed after 3 weeks. Glycemia of mice was measured with a commercial glucometer (Sinocare plus Code Glucometer; detection range: 1.1-33.3 mM) purchased at a local pharmacy.
For implantation, db/db mice (∼50g body weight per mouse) received 1.8x107 encapsulated cells suspended in 1.5mL DMEM by intraperitoneal injection. T1D and wild-type C57BL/6J mice (20-25g body weight per mouse) received 8x106 cells per animal. For blood sampling and quantification of target gene expression, whole blood was collected from the submandibular vein of mice and clotted by incubation at 4°C for 2 h, before serum was isolated by centrifugation for 8 min at 8000g.
Method details
Vector Design
References and descriptions for all expression vectors are provided in Table S3 containing sequence information of key features. Full plasmid sequences used to produce all figures will be shared via WeKwikGene (https://wekwikgene.wllsb.edu.cn). Restriction endonucleases were purchased from New England Biolabs. PCR-amplification reactions were performed using the KOD One PCR Master Mix. Ligation reactions were performed using T4 DNA Ligase. In-fusion cloning was performed with Seamless Cloning Kit. Constructs were verified by Sanger sequencing service of Tsingke Biotechnology (Beijing, China). Primers and customed designed cDNA constructs were synthesized by Genewiz Inc. (Suzhou, China). pMD2.G (Addgene plasmid # 12259) and psPAX2 (Addgene plasmid # 12260) were gifts from Didier Trono. pCMV(CAT)T7-SB100 (Addgene plasmid # 34879) was a gift from Zsuzsanna Izsvak. pCreERT (Addgene plasmid # 13777) was a gift from Connie Cepko.
Chemicals
Dimethyl sulfoxide (DMSO) was purchased from Solarbio Life Sciences (Beijing, China). Grazoprevir (stock solution 10mM in DMSO) was purchased from Selleck Chemicals (Houston, TX). Tamoxifen (stock solution 10mM in DMSO) was purchased from MedChemExpress (Monmouth Junction, NJ). 4-Nitrophenyl phosphate disodium salt hexahydrate (pNPP) was purchased from Aladdin Biochemical Technology (Shanghai, China). Diethanolamine (DEA) and Ethanol anhydrous (EtOH) were purchased from Macklin Inc. (Shanghai, China). Polyethyleneimine MAX (PEI; stock solution 1 mg/ml in ddH2O) was purchased from Polysciences (Eppelheim, Germany). Isopropanol and Vanillic Acid (stock solution 500mM in DMSO) were purchased from Rhawn Chemicals (Shanghai, China). Glycerol (stock solution 10% w/w in ddH2O), L-Homoarginine hydrochloride, Magnesium chloride hexahydrate (MgCl2) and Sodium citrate tribasic dihydrate were purchased from Sangon Biotech (Shanghai, China). Calcium chloride anhydrous (CaCl2), Sodium Chloride (NaCl; stock solution 5M in ddH2O) and citric acid were purchased from Sinopharm Chemical Reagent (Shanghai, China). Puromycin dihydrochloride, Zeocin selection reagent and Blasticidin S HCl were purchased from Thermo Fisher Scientific (Waltham, MA). Streptozotocin (STZ) was purchased from Sigma-Aldrich (St. Louis, MO).
Cell culture and transfection
Human embryonic kidney cells (HEK-293, ATCC: CRL-3216), Neuro-2A (N2A, ATCC: CRL-131), Rhabdomyosarcoma (RD, ATCC: CCL-136) and derived cell lines were cultivated in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% (v/v) fetal bovine serum (Gibco FBS, Australia; lot no. 2177370) and 1% (v/v) penicillin/streptomycin solution (PenStrep). All cells were cultured at 37°C in a humidified atmosphere of 5% CO2 in air. For passaging, cells of pre-confluent cultures were detached by incubation in 0.05% trypsin-EDTA (Sangon Biotech; lot no. F319BA0030) for 3 min at 37°C, collected in 10 ml of cell culture medium, centrifuged for 2 min at 1000rpm, and resuspended in fresh culture medium at 1.5 x 105 cells/mL, before seeding into new tissue culture plates (Wuxi NEST Biotechnology, Wuxi, China). Cell number and viability were quantified using an electric field multichannel cell counting device (Invitrogen, USA, cat no. AMQAX1000). Unless indicated otherwise, transfection of HEK-293 and N2A was performed at 12h after seeding 50,000 cells into each well of a 24-well plate using a PEI-based protocol at a PEI:DNA ratio of 5:1 (w/w) and in a transfection volume of 50μL serum free DMEM native per well. The transfection mixture was incubated for 15 min at 25°C before dropwise addition to cells. RD cells were transfected using Lipofectamine 3000 reagent (Thermo Fisher Scientific) according to the manufacturer’s instructions. Cell culture medium was exchanged into fresh medium (not containing transfection reagents) at 6h after transfection.
Lentivirus Production
Recombinant replication-deficient lentivirus particles were generated by transfecting 5x106 native HEK-293 cells (cultured in a 10cm dish; Wuxi NEST Biotechnology) with 5mg pMD2.G, 10mg psPAX2 and 10mg transfer plasmid carrying the desired gene expression cassette. At 48h after transfection, culture supernatants containing lentiviruses were collected and cells were cultivated for another 48h following medium exchange. Both harvested stocks were mixed and purified with a 0.45mm filter for experimental use or storage at -80°C.
Generation of stable cell lines
Monoclonal HEK-MOR9(C0) cells stably transgenic for constitutive MOR9-1 expression were constructed by transduction of 5x104 HEK-293 cells with supernatants containing lentiviral particles created with pLZ276 as transfer plasmid. Following selection with 100 μg/ml puromycin, single cell clones showing highest MOR9-1 expression were be picked and harvested. Monoclonal HEK-MOR9(C1) cells stably transgenic for constitutive MOR9-1 and Cre expression were constructed by transduction of 5x104 HEK-MOR9(C0) cells with supernatants containing lentiviral particles created with pLZ358 as transfer plasmid. Following selection with 10 μg/ml blasticidin, single cell clones showing highest Cre expression were be picked and harvested. HEKCreERT cells stably transgenic for constitutive ERT2CreERT2 expression was constructed by co-transfecting 5x104 HEK-293 cells with pDG172 and pCMV-T7-SB100 in a 10:1 (w/w) ratio. After selection with 1 μg/ml puromycin for 15 days, the monoclonal cell lines were harvested and selected for highest fold-change of tamoxifen-inducible Cre activity (Figure S10). HEKMOR9-1/NS3a(H1) cells stably transgenic for expression of the computational core unit of BUF1/NOT1-based tristate buffers (Figure 2E), was constructed by co-transfecting 5x104 HEK-293 cells with pDG136, pDG137 and pCMV-T7-SB100 in a 10:10:1 (w/w/w) ratio. After selection with 100μg/mL Zeocin selection reagent, 10% of the surviving population was subjected to FACS-mediated cell sorting using the MA900 Multi-Application Cell Sorter (Sony Biotechnology, San Jose, CA). Corresponding monoclonal cell lines were harvested after random picking and cultivation of 24 single cell clones.
Figure S10 Validation of tamoxifen-dependent gene expression, related to Figure 6
Quantification of target gene expression
Expression levels of human placental secreted alkaline phosphatase (SEAP) in culture supernatants were quantified based on p-nitrophenyl phosphate (pNPP)-derived light absorbance at 415 nm. SEAP concentrations in animal serum were quantified using a chemiluminescent SEAP assay (Abcam, Cambridge, UK) on a Fluoroskan FL plate reader (ThermoFisher Scientific). Nluc levels were profiled using the Nano-Glo® Luciferase Assay System (Promega, Madison, WI). Insulin levels in culture supernatants and animal serum were quantified with a Mercodia Mouse Insulin ELISA kit (assay range: 0.21 - 6.12 μg/L). GLP1-hFc levels in culture supernatants were quantified with a human IgG Fc ELISA kit (assay range: 15.6 - 500 ng/mL). GLP1-hFc levels in animal serum were quantified with a Bioactive GLP-1 ELISA kit (assay range: 2 - 100pM).
Fluorescent imaging
Fluorescence microscopy was performed with a Nikon ECLIPSE Ts2-FL Fluorescence Microscope (Nikon Instruments Inc., Melville, NY) with respective excitation and emission filter sets (EGFP: 488/509 nm; mCherry: 587/610 nm). Images were acquired by OPLENIC software (version x64,10.1.14643.20190511).
Flow cytometry
Cell populations were analyzed with a CytoFLEX LX Flow Cytometer (Beckman Coulter, Indianapolis, IN) equipped for EGFP- (488 nm laser, 525/40 emission filter) and mCherry-detection (561 nm laser, 610/20 emission filter) and set to exclude dead cells and cell doublets. 10,000 cells were recorded per data set and analyzed with FlowJo™ Software (v10; BD Biosciences). Weighted EGFP or mCherry expression levels were determined by setting an arbitrary threshold of 105 fluorescence units and multiplying the percentage of gated cells by their median fluorescence, resulting in a correlated representation between fluorescence intensity with cell number.
Microencapsulation of mammalian cells
Intraperitoneal implants were produced by encapsulating transfected mammalian cells into alginate-poly-(L-lysine)-alginate beads of 400μm in diameter using B-395 Pro Encapsulator (BÜCHI Labortechnik AG; Flawil, Switzerland) set to the following parameters: a 200μm nozzle with a vibration frequency of 1300 Hz, a 25ml syringe operated at a flow rate of 450 units and 1.50kV for bead dispersion.
Hill-type model of Biomolecular Feedback Systems
Pseudo-Reactions describe every significant transcriptional, translational and degradation/dilution event in a biological system.42 The rate for transcription (
For repressors:
For activators:
The rates of mRNA and protein degradation/dilution due to growth are assumed to be proportional to the reactants based on the mass action law.41 A complete list of representative Pseudo-Reactions summarized in Table S1 is shown in the supplemental information (Tables S4–S8).
Data analysis
Statistical analyses were performed using GraphPad Prism 9. Two-tailed unpaired Student t-tests were used to evaluate the statistical significance between groups.
Acknowledgments
We thank Baojun Wang for generous advice and members of Westlake University Biomedical Research Core Facilities (BRCF) for technical support. This work was supported by Ministry of Science and Technology MOST Project grants 2020YFA0909200 (M.X.) and 2023YFF1205400 (J.S. and H.W.); Westlake Laboratory of Life Sciences and Biomedicine HRHI grant 202209009 (M.X.); Westlake Education Foundation; Tencent Foundation; National Natural Science Foundation of China (32071429, M.X.; 82300262, Q.C.); and Westlake University Center of Synthetic Biology and Integrated Bioengineering (M.X.).
Author contributions
J.S., H.W., and M.X. conceived the project. X.Q. and H.W. designed the tristate-based gene circuits. J.S. and M.X. designed the experiments. J.S., L. Zhang, S.L., Y.S., Y.L., J.J., Y. Wu, Q.X., Q.C., S.X., Y. Wang, and T.G. performed the experimental work. J.S., X.Q., L. Zhang, L. Zhu, H.W., and M.X. analyzed the results. X.Q., H.W., and M.X. wrote the manuscript. J.S., H.W., L. Zhu, and M.X. supervised the project and are responsible for all data, figures, and text. All authors read, corrected, and approved the manuscript.
Declaration of interests
J.S., S.L., X.Q., H.W., and M.X. are inventors on PCT Patent applications PCT/CN2022/137596 and PCT/CN2023/124626, “Trigger-inducible mRNA circularization,” submitted by Westlake University.
Supplemental information (2)
Document S1. Tables S1, S2, and S4–S9 and supplemental references
Table S3. Plasmids designed and used in this study, related to STAR Methods section “Vector design”
References
Way, J.C. ∙ Collins, J.J. ∙ Keasling, J.D. ...
Integrating biological redesign: where synthetic biology came from and where it needs to go
Cell. 2014; 157:151-161Green, A.A. ∙ Kim, J. ∙ Ma, D. ...
Complex cellular logic computation using ribocomputing devices
Nature. 2017; 548:117-121Gao, X.J. ∙ Chong, L.S. ∙ Kim, M.S. ...
Programmable protein circuits in living cells
Science. 2018; 361:1252-1258Chen, Z. ∙ Kibler, R.D. ∙ Hunt, A. ...
De novo design of protein logic gates
Science. 2020; 368:78-84Greber, D. ∙ Fussenegger, M.
An engineered mammalian band-pass network
Nucleic Acids Res. 2010; 38, e174Saxena, P. ∙ Heng, B.C. ∙ Bai, P. ...
A programmable synthetic lineage-control network that differentiates human IPSCs into glucose-sensitive insulin-secreting beta-like cells
Nat. Commun. 2016; 7, 11247Din, M.O. ∙ Danino, T. ∙ Prindle, A. ...
Synchronized cycles of bacterial lysis for in vivo delivery
Nature. 2016; 536:81-85Tigges, M. ∙ Marquez-Lago, T.T. ∙ Stelling, J. ...
A tunable synthetic mammalian oscillator
Nature. 2009; 457:309-312Müller, M. ∙ Ausländer, S. ∙ Spinnler, A. ...
Designed cell consortia as fragrance-programmable analog-to-digital converters
Nat. Chem. Biol. 2017; 13:309-316Ausländer, S. ∙ Ausländer, D. ∙ Müller, M. ...
Programmable single-cell mammalian biocomputers
Nature. 2012; 487:123-127Ausländer, D. ∙ Ausländer, S. ∙ Pierrat, X. ...
Programmable full-adder computations in communicating three-dimensional cell cultures
Nat. Methods. 2018; 15:57-60Weinberg, B.H. ∙ Pham, N.T.H. ∙ Caraballo, L.D. ...
Large-scale design of robust genetic circuits with multiple inputs and outputs for mammalian cells
Nat. Biotechnol. 2017; 35:453-462Wan, X. ∙ Volpetti, F. ∙ Petrova, E. ...
Cascaded amplifying circuits enable ultrasensitive cellular sensors for toxic metals
Nat. Chem. Biol. 2019; 15:540-548Courbet, A. ∙ Endy, D. ∙ Renard, E. ...
Detection of pathological biomarkers in human clinical samples via amplifying genetic switches and logic gates
Sci. Transl. Med. 2015; 7, 289ra83Liu, Y. ∙ Zeng, Y. ∙ Liu, L. ...
Synthesizing AND gate genetic circuits based on CRISPR-Cas9 for identification of bladder cancer cells
Nat. Commun. 2014; 5, 5393Xie, Z. ∙ Wroblewska, L. ∙ Prochazka, L. ...
Multi-input RNAi-based logic circuit for identification of specific cancer cells
Science. 2011; 333:1307-1311Nissim, L. ∙ Wu, M.R. ∙ Pery, E. ...
Synthetic RNA-based immunomodulatory gene circuits for cancer immunotherapy
Cell. 2017; 171:1138-1150.e15Angelici, B. ∙ Shen, L. ∙ Schreiber, J. ...
An AAV gene therapy computes over multiple cellular inputs to enable precise targeting of multifocal hepatocellular carcinoma in mice
Sci. Transl. Med. 2021; 13, eabh4456Schukur, L. ∙ Geering, B. ∙ Charpin-El Hamri, G. ...
Implantable synthetic cytokine converter cells with AND-gate logic treat experimental psoriasis
Sci. Transl. Med. 2015; 7, 318ra201Roybal, K.T. ∙ Rupp, L.J. ∙ Morsut, L. ...
Precision tumor recognition by T cells with combinatorial antigen-sensing circuits
Cell. 2016; 164:770-779Cho, J.H. ∙ Collins, J.J. ∙ Wong, W.W.
Universal chimeric antigen receptors for multiplexed and logical control of T cell responses
Cell. 2018; 173:1426-1438.e11Gander, M.W. ∙ Vrana, J.D. ∙ Voje, W.E. ...
Digital logic circuits in yeast with CRISPR-dCas9 NOR gates
Nat. Commun. 2017; 8, 15459Andrews, L.B. ∙ Nielsen, A.A.K. ∙ Voigt, C.A.
Cellular checkpoint control using programmable sequential logic
Science. 2018; 361, eaap8987Tamsir, A. ∙ Tabor, J.J. ∙ Voigt, C.A.
Robust multicellular computing using genetically encoded NOR gates and chemical 'wires'
Nature. 2011; 469:212-215Gaber, R. ∙ Lebar, T. ∙ Majerle, A. ...
Designable DNA-binding domains enable construction of logic circuits in mammalian cells
Nat. Chem. Biol. 2014; 10:203-208Zhao, Z.W. ∙ White, M.D. ∙ Bissiere, S. ...
Quantitative imaging of mammalian transcriptional dynamics: from single cells to whole embryos
BMC Biol. 2016; 14, 115Chen, Y. ∙ Zhang, S. ∙ Young, E.M. ...
Genetic circuit design automation for yeast
Nat. Microbiol. 2020; 5:1349-1360Choi, S. ∙ Lee, G. ∙ Kim, J.
Cellular computational logic using toehold switches
Int. J. Mol. Sci. 2022; 23, 4265Wong, A. ∙ Wang, H. ∙ Poh, C.L. ...
Layering genetic circuits to build a single cell, bacterial half adder
BMC Biol. 2015; 13, 40Letendre, J.H. ∙ Weinberg, B.H. ∙ Mendes, M. ...
Programmable mixed-signal biocomputers in mammalian cells
Preprint atbioRxiv. 2022;Harris, S.L. ∙ Harris, D.
2 - Combinational logic design
Harris, S.L. ∙ Harris, D.
Digital Design and Computer Architecture
Morgan Kaufmann, 2022; 52-104
Crist, S.C.
A tri-state logic family
Proceedings of the Eighth International Symposium on Multiple-Valued Logic
IEEE Computer Society Press, 1978; 1-6
Foight, G.W. ∙ Wang, Z. ∙ Wei, C.T. ...
Multi-input chemical control of protein dimerization for programming graded cellular responses
Nat. Biotechnol. 2019; 37:1209-1216Shao, J. ∙ Li, S. ∙ Qiu, X. ...
Engineered poly (A)-surrogates for translational regulation and therapeutic biocomputation in mammalian cells
Cell Res. 2024; 1-16Stripecke, R. ∙ Hentze, M.W.
Bacteriophage and spliceosomal proteins function as position-dependent cis/trans repressors of mRNA translation in vitro
Nucleic Acids Res. 1992; 20:5555-5564Piron, M. ∙ Delaunay, T. ∙ Grosclaude, J. ...
Identification of the RNA-binding, dimerization, and eIF4GI-binding domains of rotavirus nonstructural protein NSP3
J. Virol. 1999; 73:5411-5421Cunningham-Bryant, D. ∙ Dieter, E.M. ∙ Foight, G.W. ...
A chemically disrupted proximity system for controlling dynamic cellular processes
J. Am. Chem. Soc. 2019; 141:3352-3355Kemmer, C. ∙ Fluri, D.A. ∙ Witschi, U. ...
A designer network coordinating bovine artificial insemination by ovulation-triggered release of implanted sperms
J. Control. Release. 2011; 150:23-29Gitzinger, M. ∙ Kemmer, C. ∙ Fluri, D.A. ...
The food additive vanillic acid controls transgene expression in mammalian cells and mice
Nucleic Acids Res. 2012; 40, e37Lugagne, J.-B. ∙ Sosa Carrillo, S.S. ∙ Kirch, M. ...
Balancing a genetic toggle switch by real-time feedback control and periodic forcing
Nat. Commun. 2017; 8, 1671Xie, M. ∙ Fussenegger, M.
Designing cell function: assembly of synthetic gene circuits for cell biology applications
Nat. Rev. Mol. Cell Biol. 2018; 19:507-525Scheller, L. ∙ Strittmatter, T. ∙ Fuchs, D. ...
Generalized extracellular molecule sensor platform for programming cellular behavior
Nat. Chem. Biol. 2018; 14:723-729Silver, J.S. ∙ Hunter, C.A.
gp130 at the nexus of inflammation, autoimmunity, and cancer
J. Leukoc. Biol. 2010; 88:1145-1156Reichhart, E. ∙ Ingles-Prieto, A. ∙ Tichy, A.M. ...
A phytochrome sensory domain permits receptor activation by red light
Angew. Chem. Int. Ed. Engl. 2016; 55:6339-6342Abhinand, C.S. ∙ Raju, R. ∙ Soumya, S.J. ...
VEGF-A/VEGFR2 signaling network in endothelial cells relevant to angiogenesis
J. Cell Commun. Signal. 2016; 10:347-354Sternberg, N. ∙ Hamilton, D.
Bacteriophage P1 site-specific recombination. I. Recombination between loxP sites
J. Mol. Biol. 1981; 150:467-486Karnaugh, M.
The map method for synthesis of combinational logic circuits
Trans. AIEE Part I: Comm. Electron. 1953; 72:593-599Mansouri, M. ∙ Ray, P.G. ∙ Franko, N. ...
Design of programmable post-translational switch control platform for on-demand protein secretion in mammalian cells
Nucleic Acids Res. 2023; 51, e1Monteys, A.M. ∙ Hundley, A.A. ∙ Ranum, P.T. ...
Regulated control of gene therapies by drug-induced splicing
Nature. 2021; 596:291-295Bailey, C.J. ∙ Tahrani, A.A. ∙ Barnett, A.H.
Future glucose-lowering drugs for type 2 diabetes
Lancet Diabetes Endocrinol. 2016; 4:350-359Zaykov, A.N. ∙ Mayer, J.P. ∙ DiMarchi, R.D.
Pursuit of a perfect insulin
Nat. Rev. Drug Discov. 2016; 15:425-439Prentki, M. ∙ Nolan, C.J.
Islet beta cell failure in type 2 diabetes
J. Clin. Invest. 2006; 116:1802-1812Matsuda, T. ∙ Cepko, C.L.
Controlled expression of transgenes introduced by in vivo electroporation
Proc. Natl. Acad. Sci. USA. 2007; 104:1027-1032Ausländer, S. ∙ Wieland, M. ∙ Fussenegger, M.
Smart medication through combination of synthetic biology and cell microencapsulation
Metab. Eng. 2012; 14:252-260Wieland, M. ∙ Fussenegger, M.
Reprogrammed cell delivery for personalized medicine
Adv. Drug Deliv. Rev. 2012; 64:1477-1487Anderson, J.C. ∙ Voigt, C.A. ∙ Arkin, A.P.
Environmental signal integration by a modular AND gate
Mol. Syst. Biol. 2007; 3, 133Parres-Gold, J. ∙ Levine, M. ∙ Emert, B. ...
Principles of computation by competitive protein dimerization networks
Preprint atbioRxiv. 2023;Klumpe, H.E. ∙ Garcia-Ojalvo, J. ∙ Elowitz, M.B. ...
The computational capabilities of many-to-many protein interaction networks
Cell Syst. 2023; 14:430-446Chen, Z. ∙ Linton, J.M. ∙ Zhu, R. ...
A synthetic protein-level neural network in mammalian cells
Preprint atbioRxiv. 2022;Nielsen, A.A.K. ∙ Der, B.S. ∙ Shin, J. ...
Genetic circuit design automation
Science. 2016; 352, aac7341Fink, T. ∙ Lonzarić, J. ∙ Praznik, A. ...
Design of fast proteolysis-based signaling and logic circuits in mammalian cells
Nat. Chem. Biol. 2019; 15:115-122Fink, T. ∙ Lonzarić, J. ∙ Praznik, A. ...
Design of fast proteolysis-based signaling and logic circuits in mammalian cells
Nat. Chem. Biol. 2019; 15:115-122Frieda, K.L. ∙ Linton, J.M. ∙ Hormoz, S. ...
Synthetic recording and in situ readout of lineage information in single cells
Nature. 2017; 541:107-111Barker, N. ∙ van Es, J.H. ∙ Kuipers, J. ...
Identification of stem cells in small intestine and colon by marker gene Lgr5
Nature. 2007; 449:1003-1007Kim, H. ∙ Bojar, D. ∙ Fussenegger, M.
A CRISPR/Cas9-based central processing unit to program complex logic computation in human cells
Proc. Natl. Acad. Sci. USA. 2019; 116:7214-7219Xie, M. ∙ Ye, H. ∙ Wang, H. ...
beta-cell-mimetic designer cells provide closed-loop glycemic control
Science. 2016; 354:1296-1301Figures (16) 图（16）
Article metrics 文章指标
-
-
3Captures 捕获1Mentions 提及
-
Supplemental information (2)
补充信息 (2)
PDF (353.42 KB)
Document S1. Tables S1, S2, and S4–S9 and supplemental references
文档 S1。表 S1、S2 和 S4–S9 及补充参考文献
文档 S1。表 S1、S2 和 S4–S9 及补充参考文献
Spreadsheet (43.10 KB) 电子表格 (43.10 KB)
Table S3. Plasmids designed and used in this study, related to STAR Methods section “Vector design”
表 S3. 本研究中设计和使用的质粒，相关于 STAR 方法部分“载体设计”
表 S3. 本研究中设计和使用的质粒，相关于 STAR 方法部分“载体设计”
