Biomolecular computing systems principles, progress and pote

REVIEWS Yaakov Benenson

Biomolecular computing systems: principles, progress and potential 生物分子计算体系:原理、进展和潜力

Abstract 摘要

The task of information processing, or computation, can be performed by natural and man-made ‘devices’. Man-made computers are made from silicon chips, whereas natural ‘computers’, such as the brain, use cells and molecules. Computation also occurs on a much smaller scale in regulatory and signalling pathways in individual cells and even within single biomolecules.

信息处理或计算,可以通过自然的和人造的“装置”实现。人造电脑由硅芯片构成,而自然的“计算机”,如大脑,由细胞和分子构成。计算也可以在更小规模的单个细胞调控和传递信号的通路中进行,甚至可以在单个的生物分子中进行。 Indeed, much of what we recognize as life results from the remarkable capacity of biological building blocks to compute in highly sophisticated ways. Rational design and engineering of biological computing systems can greatly enhance our ability to study and to control biological systems. Potential applications include tissue engineering and regeneration and medical treatments. This Review introduces key concepts and discusses recent progress that has been made in biomolecular computing.

事实上，我们将大部分工作看作是来自于生物模块的非凡能力，能够以一种高度复杂的方式进行计算。生物计算系统的合理的设计和建造可以大大提高我们研究和控制生物系统的能力。其潜在的应用包括组织工程、组织再生和医学治疗等。

本文介绍了生物分子计算的主要概念，并讨论了生物分子计算领域最近的进展。 Any delicately structured system will have difficulty surviving in the world if it is left to its own devices, as it will be subject to malfunction of its complex internal organization and to environmental perturbations. The survival of man-made objects such as aeroplanes or natural objects such as birds crucially depends on internal control mechanisms that counter such damage. These mechanisms have the same stereotypical structure (FIG. 1a): sensors that collect information (input) from inside and outside the system; computers or processors that interpret this information to determine potential damage and to decide on the response; and actuators (output) that carry the response out. In animals, this description applies equally well to the brain, to the immune and homeostasis systems and to the regulatory pathways in individual cells. 任何微妙的结构化系统如果总是依赖于自己的装置，那么他将很难在世界上生存，因为它会受制于自己的复杂内部组织和外界环境的扰动。人造物体像飞机，或自然物体像鸟的生存的关键取决于内部控制机制对于这种扰动的应对。这些机制具有共同的典型结构（图1a）：传感器从体系内部和外部收集信息（输入）；计算机或处理器处理这些信息，确定其潜在的危害，并决定作出响应；执行器（输出）实施响应。在动物中，这样的过程同样适用于大脑、免疫系统和体内平衡系统，包括单个细胞的调控通路。

图 1-a

Beyond the control that living systems require in order to survive, humans have always striven to control and manipulate biological objects to achieve their own goals. Recently, such goals have included engineering of metabolism for bioproduction; controlling the immune system during organ transplantation; manipulating cell differentiation and spatial organization for tissue engineering; and, at the pinnacle of these efforts, controlling myriad aspects of human physiology to treat or cure disease. Here I argue that rational design of biological computation is needed to rapidly advance our ability to exercise such control.

除了生命系统为了存活所需的控制，人们一直想要控制和操纵生物对象来实现自己的目标。最近，这些目标已经包含了生物生产的代谢工程；器官移植过程中免疫系统的控制；组织工程学的细胞分化和空间结构的操纵；这些努力的巅峰，就是控制人体生理学的种种方面来治疗治愈疾病。在这里，我们认为，为了迅速推进人类行使这种控制的能力，生物计算的合理设计是很必要的。

Computation is evident even in single-input–single-output systems, such as a conditional transgene in a mouse or an inducible expression vector, because the specific way in which the inducer and the transgene expression are linked implies a certain mathematical relationship (mapping) between them. However, more intricate multiple-input–multiple-output computations that are able to address complex internal and environmental states and to trigger multiple processes in response are needed to expand the applications of biomolecular engineering.

在单输入单输出系统中的计算是显而易见的，如老鼠体内的条件转基因，或者可诱导的表达载体，因为诱导物和转基因表达之间的特定方式的联系，蕴含着一定的数学关系（映射）。然而，更复杂的多输入多输出计算，能够处理复杂的体内和环境状态，触发多个进程的响应，来扩大生物分子工程的应用。

Accordingly, this Review focuses on the efforts to engineer complex computational modules from molecular and biological building blocks and to furnish these modules with sensors and actuators. First, I introduce some of the potential applications and then explain the theory of molecular computing. I then discuss the progress in biomolecular implementation of different models of computation and highlight key challenges posed by biological settings. I emphasize generally applicable design principles that would allow quick, flexible construction of systems with functions as diverse as drug screening, environmental monitoring or disease diagnosis.

因此，本文的重点在利用分子和生物模块设计复杂的计算模型，并将这些模块和传感器、执行器进行布置。首先介绍一些潜在的应用，然后阐述分子计算理论。接着讨论不同生物分子计算模型的实现的进展如何，着重讨论生物设置构成的关

键性挑战。强调一般通用的设计原则，快速、灵活的系统结构，具备药物筛选、环境监测、疾病诊断等多种多样的功能。

What can be done? 可以做什么呢？

Historically, the concept of a molecular computer dates decades back. First experiments encoded mathematical objects, such as connected graphs, with nucleic acids. These molecules were then biochemically manipulated to generate DNA- or RNA-encoded solutions to small instances of computationally hard problems. Later on, work commenced on autonomous systems, with the explicit vision of their eventual use for biological control. 历史上，分子计算机的概念可以追溯到数十年前。第一个编码数学对象的试验，如核酸构成的连通图。这些分子通过生化操作产生DNA或RNA编码，来解决小实例的计算困难的问题。后来，开始了自治系统的工作，其最终的愿景是用于生物控制。

Autonomous computations were implemented with nucleic acids and protein building blocks reconstituted in simple buffers and occasionally in cell-free extracts. These ‘biochemical’ approaches allowed rapid progress towards increasing complexity and proof-of-concept experiments, demonstrating the potential of information processing with molecules. Parallel efforts have attempted to embed genetically encoded ‘biological’ computing systems in live objects, mainly single cells, using recombinant DNA technology and synthetic biology. Both approaches complement and cross-pollinate each other, constituting a common scientific endeavour

自治计算在缓冲液和偶尔的体外提取物中，通过核酸和蛋白质模块实现重组。这些生化方法面对复杂度和概念验证实验的不断增加，飞速发展，验证了分子进行信息处理的潜力。另一个探究的方向是，利用重组DNA技术和合成生物学，试图将基因编码的生物计算体系嵌入活体对象，主要是单细胞。这两种方法相辅相成，构成了统一的科学奋进号。

Molecular computers could in principle perform diverse monitoring, manipulation and control tasks. A common task is generating a certain physiological effect in a cell conditioned on specific intra- and extracellular cues. The simplest case would reflect the cues’ state at a given time: for example, ‘when gene A is active AND gene B is inactive AND gene C is mutated → generate the effect’ (FIG. 1b).

分子计算机在原则上可以执行多种监控，操作和控制任务。一个常见的工作是在特定的细胞内部或外部的诱因下产生一定的生理效应。最简单的例子是，在给定的时间内反映诱因的状态：例如，当基因A是积极的,基因B是消极的，基因C产生突变→生成效应（图1b）

图1-b

The mapping from the inputs to the effect could be made increasingly elaborate by allowing different sets of conditions to generate the same out-

come or by explicitly considering time, such as ‘gene C got mutated after gene B became inactive and gene A became active’. Furthermore, the mapping could include quantitative thresholds for ‘active’ or ‘inactive’ states, or it could integrate activities gradually. Depending on the application, the effect itself could be ‘all or none’ or gradual. Numerous areas will benefit from these tools, including cell reprogramming, disease diagnosis and treatment, environmental monitoring, tissue engineering and real-time measurements in live cells. A recent example includes a prototype gene circuit that integrates six disease markers to identify and to destroy human cancer cells in culture.

从输入到输出的映射可以通过不同的条件产生相同的输出，或者通过明确的考察时间来逐步精细，如基因C的突变发生在基因B转为消极和A转为积极之后。此外，映射可能包括“积极”态或“消极”态的量化阈值，或者逐步整合活性。根据不同的应用，输出本身可以实现“all”“none”或渐变。许多领域将受益于这些工具，包括细胞重编程，疾病诊断治疗，环境监测，组织工程学和活细胞实时监测。一个最近的例子是一个原型基因电路，集成了六个疾病标签，来识别和消灭培养皿中的癌细胞。

More complex tasks can also be envisaged. For example, a computer could receive a biopolymer such as DNA or RNA as its input and, depending on certain intracellular conditions, could modify or generate a new polymer whose sequence reflects these conditions in complex ways (FIG. 1c). It could be a DNA sequence that records the number of cell divisions, the cell’s history of exposure to toxins or a protein-coding sequence that is altered in response to environmental stress to encode a polypeptide that best copes with this stress. 可以设想更复杂的任务。例如，一个计算机可以接收生物聚合物，如DNA或RNA

作为它的输入，根据细胞内固定的条件，修改或生成一个新的聚合物，其序列以一种复杂的方式反映了这些条件（图1C）。它可能是一个DNA序列，记录细胞分裂的数量，细胞暴露毒素的历史，或者是一个蛋白质编码序列，通过变异来应对环境压力, 通过编码形成一个多肽是应对这种压力最好的处理。

图 1-c

Similar programmed alteration could also occur in regulatory DNA to change gene regulatory programs. These examples constitute a drastic intervention, leading to irreversible genomic modification. Apart from application potential, the capacity to do so is of basic importance because, in the extreme case, this is tantamount to programmable genome alteration by the cell itself. The move from replicating to ‘computing’ the genetic code would greatly enrich our understanding of life.

类似的程序化变更也可以发生在调控DNA中，来改变基因调控程序。这些例子构成一个强烈的干预，导致了不可逆的基因组修饰。除了应用方面的潜力，实现这种做法的能力十分重要，因为在极端情况下，这相当于基于细胞自身的可编程的基因组变更。从复制到“计算”，基因编码将大大丰富我们对生命的理解。 With these examples in mind, I now consider the details of molecular computer design.

通过这些例子,我们现在来细细分析分子计算的细节设计。

The theory of molecular computing 分子计算的理论

Molecular computing theory incorporates elements of systems biology, chemical reaction networks and control theory in order to derive molecular-based solutions of information-processing tasks in a systematic fashion. The derivation (FIG. 1d) can be done at different resolutions: an abstract network diagram; a coarse functional description for individual network nodes (for example, ‘repressor’); a detailed structural assignment; and, ultimately, an experimental setup comprising the species, their amounts and the laboratory protocols.

分子计算理论结合了生物学体系、化学反应网络及控制理论的元素，以获得一个系统的以分子为基础的信息处理的解决方案。这个词源（图1d）可以由不同的决议得到：一个抽象的网络图；对于单个网络节点的一个粗糙的功能性描述（例如，“抑制剂”）；一个详细的结构分配；或者，最终实现一个由物种及相应数量和实验室协议构成的实验装置。

图 1-d

Computer science is instrumental in this process because it deals with systematic problem solving by means of models of computation. Thus, recognizing that a specific task naturally falls under a particular model helps to generate coarse-grained network diagrams. (Their successful translation into biochemical processes, however, depends on ingenuity more than anything else.) A number of biologically relevant models are described below.

计算机科学是在这个过程中是有帮助的，因为它通过计算模型的方式处理系统的问题。因此，我们自然的认识到一个特定的任务是设计一个特定的模型，来产生粗略的网络图。（能够成功转化到生化过程，然而，相比于其他的，能够实现更多的是取决于独创性。）。下面介绍了一些生物方面的模型。

Logic circuits 逻辑电路

Many objectives can be expressed as logic functions. For example, a reporter system that is capable of producing GFP if the cell is in mitosis under osmotic stress needs to compute the logic function ‘GFP = [mitotic marker] AND [osmotic stress marker]’. Logic circuits — a model for computing logic functions — are therefore a useful inspiration for building such reporters. Because any mapping of ‘all or none’ cues to an ‘all or none’ outcome is a logic function, diverse monitoring and control applications in cells can be enabled by systematic approaches to constructing molecular logic circuits.

许多目标可以表示为逻辑功能。例如，一个reporter系统，如果细胞在渗透应力下进行有丝分裂能够产生GFP，需要计算逻辑函数GFP = [有丝分裂标记]AND[渗透应力标记]”。逻辑电路，用于计算逻辑函数的一个模型，是建立一个这样的reporter的有用灵感。因为任何“all 或者 none”的映射对应于一个“all 或

者 none”的输出，构成了一个逻辑功能，细胞中不同的监测和控制应用程序可以通过系统的方法来构建分子逻辑电路。

These circuits, which are also known as digital or Boolean circuits, belong to ‘circuit-based’ models in which data pass through ‘wires’ between small computational units (known as ‘gates’) that perform simple operations (FIG. 2a,b). Circuit models resemble coupled chemical reactions, in which the concentrations of individual species are interpreted as their values, and individual interactions are compared to circuit wires.

这些电路，被称为数字电路或布尔电路，属于“基于电路”的模型，数据通过小计算单元（称为门）之间的线，进行简单的操作（图2a,b）。电路模型与对应的化学反应相似，其中单个物质的浓度看做他们的值，单个物质之间的相互作用相当于电路导线。

图 2-a

图 2-b

Notably, specially designed circuits and their molecular implementations can act as memory, which is a key component of many useful computations. Digital-circuit-inspired approaches to molecular computation include enzymatic cascade logic(FIG. 2c), a universal set of enzyme-based gates (FIG. 2d), transcription-based universal gates such as NAND and NOR (FIG. 2e) and normal-form circuits (FIG. 2f). Compartmentalizing gates in separate microbial strains and connecting them by diffusing small molecules has been proposed and tested as a way of controlling cell communities.

值得注意的是，专门设计的电路及其分子实现可以作为内存，是许多有用计算的关键组成部分。数字电路启发的分子计算的方法包括酶促级联逻辑（图2c），一个基于酶的门的通用集（图2d），基于转录的门如与非门和或非门（图2e）和标准形式的电路（图2f）。已经提出在独立的微生物菌株上相隔开的门，用扩散小分子连接他们，经测试可以作为一种控制细胞群落的方法。

图 2-c

图 2-d

图 2-e

图2-f

Logic circuits are an approximation of the chemical process because concentrations are continuous. Analogue circuits contain gates that compute simple non-Boolean functions such as addition, multiplication and integration. Biomolecular analogue circuits have been examined theoretically. Ideas about analogue biological computing were also explored in the context of artificial neural networks composed of neuron-like gates that weight, add and ‘digitize’ the inputs. Early theoretical works on their molecular implementation inspired quite different experimental systems, showcasing elements such as an ‘adder’ gate, whose output level is a sum of its inputs’ levels.

逻辑电路是化学过程的一种近似，因为浓度是连续的。模拟电路包含非布尔运算的门，像加法运算、乘法运算和集成。生物分子的模拟电路已经通过理论验证。模拟生物计算的思想也已经在人工神经网络的背景下探究过，包含像神经元一样的门，量化、加入并且数字化输入。分子实现的早期理论启发了完全不同的实验系统，展示了 “加法器”等原理，它的输出是输入的总和。

State machines 状态机 A second large class of models comprises state machines that manipulate discrete data units — symbols — that are stored on tapes (FIG. 3). The com-puter processes the symbols according to specific rules. Even though these models have traditionally been used to prove statements about computability, parallels to transcription and translation make them plausible inspiration for practical biomolecular computations. Successful experimental implementation has been achieved so far for simpler, read-only models, such as finite automata. Theoretical blueprints of molecular Turing machines suggest

specific ways of encoding the symbols, the states and state transitions. A crucial challenge that is evident in those designs is symbol writing: modifying a small portion of the tape in a particular location (somewhat like RNA editing) at each computation step. Thus, these ideas are still awaiting experimental implementation.

第二大类模型，包含状态机，处理存储在磁带上的离散数据单元-符号（图3）。根据特定的规则处理这些符号。虽然这些模型通常被用来证明语句的可计算性，通过转录和翻译，使他们对于实际的生物分子计算来说，可以成为合理的设计。至今已经成功的试验验证了简单的只读模式，如有限自动机。理论上的分子图灵机的设计建议采用独特的方式对符号、状态和状态转换进行编码。在这些设计中一个显而易见的关键性挑战是符号的书写：每个计算步骤在一个特定的位置修改磁带的一小部分（有点像RNA修订）。因此，这些想法仍在等待实验验证。

图 3

In a finite automaton, the tape symbols can be replaced with a temporal sequence of short events, resulting in ‘reactive’ cell control systems that wait for their inputs while preserving the memory of the past. A molecular implementation was proposed in which the environmental trigger activates a recombinase that in turn modifies a specific DNA sequence (such as a gene promoter). This modified DNA is identified with a new system state because it remains stable and because it determines which DNA modification (that is, the next ‘state’) is caused by the next trigger (that is, the next ‘symbol’).

在一个有限自动机中，磁带符号可以用短事件的瞬时序列代替，导致被动的单元

控制系统，在等待输入的同时保留了过去的记忆。提出了一个分子实现，环境触发激活重组酶，修饰特定的DNA序列（如基因启动子）。这种修饰的DNA被确定为一个新的系统状态，因为它是稳定的，并且决定了哪个DNA修饰（即下一个“状态”）是由下一个触发引起的（即下一个“符号”）。

Other models 其他模型

A number of new models of computations explicitly consider the biological or chemical medium as well as specific biochemical transformations. Among them are: splicing systems, which use recombinant DNA protocols to generate new sequence libraries in a programmable fashion; membrane computing, which is a model that requires compartmentalization and information exchange between compartments; computation in excitable media, which builds on processes such as oscillating Belousov–Zhabotinsky reactions and may have relevance to pattern establishment in embryogenesis; and computation based on gene recombination, which was inspired by the elaborate gene rearrange-ment in ciliates.

一些新的计算模型明确地考虑到生物或化学介质以及特定的生化变化。其中包括：剪接系统，即利用DNA重组协议以一种可编程的方式产生新的序列库；膜计算，一个隔室之间需要分割和信息交换的模型；激发介质计算，建立在Belousov–Zhabotinsky振荡反应等过程上，可能和胚胎发育的模式构建相关；基于基因重组的计算，灵感来自于纤毛虫上的复杂的基因重排。

The notions of distributed computing and amorphous computing are also instrumental for conceptualizing large stochastic networks of chemical processes and information processing by multiple interacting agents (for example, bacterial cells or organelles), respectively. These works are another

important source of inspiration for future experimental molecular computing. 分布式计算和无定形计算的概念对化学过程的大规模随机网络和多个相互作用的中介的信息处理是有帮助的，（例如，细菌细胞或细胞器）。这些工作是未来分子计算实验的一个重要的灵感来源。

Experimental logic circuits 逻辑电路的实验

Abstract logic networks are easy to sketch but are difficult to implement with molecules. Any logic circuit blueprint requires a set of real-world switches that comprise the basic gates and their networks. In computer engineering, transistors are universally used to implement switching schemes of almost unlimited complexity. Such truly universal building blocks may never become available for molecular systems because, unlike the circuit board, where all of the gates are physically localized and separated, molecular components diffuse and mix.

抽象的逻辑网络易于表示，但用分子实现就很难了。任何逻辑电路的设计需要一套现实世界的开关，由基本的门和网络结构构成。在计算机工程，普遍采用晶体管来解决无限复杂的开关切换方案。这种现实意义上的通用的构建模块可能永远不会成为可用的分子系统，不像电路板，基本所有的门都是物理上局部定位和分开的，而分子总是扩散和混合的。

Thus, barring compartmentalization of each component in its own membrane, gates and wires must be structurally distinct. Features required from an effective molecular switch or gate include: the existence of a robust digital regime (that is, input levels that produce either a very low or a very high (saturated) output); gate scalability, which is the capacity to receive an

increasing number of inputs without dramatic design alterations; and com-posability, which is the capacity to operate together with other gates in parallel and/or in cascades in a predictable manner. In theory, some biological building blocks possess all of these features. In practice, increasing the size of biochemical and biological circuits is challenging even when it is theoretically possible.

因此，除非每个组件都划分在自己的隔膜内，门和线必须在结构上有所区分。一个有效的分子开关或门所需的特点包括：一个强大的数字制度（即输入必须产生一个非常低或非常高（饱和）的输出）；门的可扩展性，可以在不进行较大的改动下，接收更多的输入；组合性，能够以一种可预见的方式并行的和其他的门进行与或运算的级联。在理论上，一些生物模块都拥有这些功能。尽管在理论是是可行的，但是在实践中，生物分子电路的规模增大是一项较大的挑战。 In biochemical circuits, interacting sets of single- and double-stranded DNA oligonucleotides seem to provide the answer because interactions between DNA strands follow Watson–Crick rules and thus can be predicted computationally with a high certainty. In cells, these DNA structures are unstable and the development of additional non-native, ‘orthogonal’ cellular processes is still in its infancy.

在生化电路中，相互作用的单、双链DNA寡核苷酸组 似乎提供了可行的方案，由于DNA链之间的沃森–克里克作用，可以预见到一个高精确度的计算。在细胞中，由于这些DNA结构是不稳定的，以及附加的非天然结构，正交的细胞过程仍处于起步阶段。

Thus, native building blocks are used as the basis for novel switches to ensure their operation in cells. Below we describe switch and circuit design based on DNA, followed by RNA and then protein building blocks. composed of multiple

subunits that formed functional enzymes only following certain input combinations, implementing complex logic cascades。

因此，天然的建筑模块作为全新开关的基础，确保了他们在细胞内的运作。下面我们描述了基于DNA的开关及电路设计，其次是RNA，然后是蛋白质的建筑模块。由多级子单位组成，形成的功能酶只能遵循一定的输入组合，从而实现复杂的逻辑通路。

A different approach in oligonucleotide-based logic is based on the observation that abstract reaction networks can be instantiated with DNA species interacting through strand migration or strand displacement without creating or breaking chemical bonds. From initial experimental computations with a four-gate circuit integrating six microRNA-like inputs in the test tube, these circuits have grown to include up to twelve gates, eight inputs and more than 100 building blocks.

通过报告分析，发现了一种基于寡核苷酸的不同的逻辑实现，抽象的反应网络可以不新建也不打破化学键，通过链迁移或 链置换和DNA分子相互作用。通过初始的四门电路的实验，在试管里集成六个类似微小核糖核酸的输入，这些电路已发展到多达十二个门，八个输入和100多个构建模块。

These most recent gates are based on artificial neurons with thresholds (FIG. 4), implementing multiple-input–multiple-output AND and OR logic. Molecular neurons were also used to build a circuit resembling a small Hopfield neural network with associative memory features, underlying their use in both digital and analogue computing networks.

最新的门基于带有阈值的人工神经元（图4），实现多输入多输出的与逻辑和或逻辑。分子神经元被用来建立一个具有联想记忆功能的类似小型Hopfield神经网络的电路，其基础应用是数字计算网络和模拟计算网络。

图 4

Gene-based biological switches 基于基因的生物转换开关

Modification of the gene sequence (in the promoter and/or the coding region) not only affects a biological process, but it does so in a heritable fashion. Engineering modifications in the gene sequence can be most readily achieved with site-specific recombinases that are commonly used for genomic manipulation, such as Cre. The modification may include excision, insertion or inversion of a gene fragment, and it can be reversible or irreversible, depending on the recombinase and the arrangement of the recognition sites. 基因序列的修改（在启动子和编码区）不仅影响生化过程，而且它是以一种可遗

传的方式完成这项工作。基因序列的工程改造通过位点特异性是最容易实现的，通常用于基因组的处理，如肌酐。改造工作包括一个基因片段的切除，嵌入和倒置，这些工作可以是可逆，也可以是不可逆的，取决于重组酶和识别位点的排列。 Recombination becomes a truly digital biomolecular process when the cell contains exactly one copy of the genetic substrate and the process is irreversible. Interestingly, bona fide logic circuits with recombinases have yet to be shown. Instead, a number of experimental systems implemented the state machine computations, as we will see below. Another potentially interesting tool for heritable DNA modification is transposable elements: ‘selfish’ genetic elements that encode their own integration machinery.

当细胞恰好包含一个基因基板的拷贝时，重组成为一个真正的数字生物分子进程，其过程是不可逆的。有趣的是，真正由重组酶构成的逻辑电路还未出现。相反，我们下面将看到一些实验系统实现的状态机计算。另一个可能出现的可遗传性DNA修饰的工具是转位因子：只编码自身集成机构的自私的遗传因子。 Epigenetic switching is heritable modification of a gene’s state that occurs without changes in the nucleotide sequence — for example, by DNA methylation or histone modification. The change in state is usually rather binary, if slow. Moreover, the triggers causing the change and the biomolecular mechanisms are still poorly understood.

后生的开关是一个不改变核苷酸序列的基因状态修正的遗传性修饰。例如，通过DNA甲基化和组蛋白的修饰。状态的变化，如果慢的话通常相当于二进制。然而，对于触发器造成的变化及其分子机制仍知之甚少。

Thus, epigenetics have rarely been used in engineered circuits. However, a recent report showed a designed transcription factor that binds specific modi-fied histones and upregulates gene expression at adjacent loci using a VP64

transactivator domain (modified from herpes simplex virus). Large engineered circuits that incorporate rationally designed epigenetic switches might be useful for tissue engineering and cell-based therapy applications.

因此，实验胚胎学很少被运用在电路设计中。然而，最近的一份报告显示了一个结合特定的组蛋白修饰的转录因子的设计，使用反式激活因子域VP64在临近位点上调基因表达（由单纯疱疹病毒进行改动）。包含合理的后生开关的设计的大型工程电路，可能对组织工程和细胞治疗的应用有所帮助。

RNA-based biological switches and circuits 基于RNA的生物转换开关和电路

RNA-centred switches (FIG. 5a) have gained attention in recent years owing to their small size, modularity and susceptibility to diverse inputs such as metabolites, RNA and proteins. A typical bifunctional cis-acting riboswitch controls its host gene by transcriptional termination, ribosome-binding site blockage or self-cleavage; ligand binding modulates the gene control function. Engineered riboswitches have been integrated in tandem to implement various two-input gates.

RNA集中开关（图5a）由于他们的小尺寸，模块性和对不同输入物如代谢物、RNA和蛋白质的敏感性，已经在近几年得到了关注，。一个典型的双官能团的顺式作用的核糖开关，通过转录终止、核糖体结合位点堵塞和自我裂解控制其宿主基因；配体结合调节基因控制。核糖开关的设计被一前一后的集成，实现各种双输入门。

图 5-a

Recent work has shown a two-input NOR gate and a three-level cascade that is controlled by the interaction of an antisense-like RNA input with a riboswitch. Another study described a novel class of RNA aptamer switches that, when placed in intronic location, regulate alternative splicing following binding of cognate endogenous protein inputs。

近期的工作表明，双输入或门非和三级级联是通过带有核糖开关的反义RNA开关输入的互动所控制。另一项研究描述了一种新的RNA适配子开关，当放置在内含子的位置，通过绑定同源的内源蛋白输入选择性剪接调节。

Trans-acting RNA switches include small RNAs (sRNAs) in bacteria and microRNAs (miRNAs) in higher eukaryotes. miRNAs have been extensively studied as the basis for complex logic in mammalian cells, because multiple miRNAs that control the same gene implement NOR logic. The logic can be

‘upgraded’ by making miRNA activity the outcome of proportional or inverse sensing of external inputs. For example, if genetically encoded synthetic miRNAs are activated or repressed by transcription factors, any logic formula can in principle be computed with these transcription factor inputs.

反式作用RNA开关包括高等真核生物中细菌和微分子RNA中的小分子RNA（sRNA）。miRNAs被广泛研究作为在哺乳动物细胞中复杂的逻辑基础，因为多个miRNAs控制相同的基因，实现或非逻辑。逻辑可以通过让miRNA的活性比例输出或逆传感外部输入改善。例如，如果基因编码合成的miRNA通过转录因子激活或抑制，原则上任何逻辑公式都可以用这些转录因子输入进行计算。

Recently, a hybrid system was constructed that interrogates six endogenous miRNA markers to identify HeLa cancer cells. The circuit computes an AND logic with these markers, detecting a HeLa-specific expression profile and triggering an apoptotic gene in response (FIG. 5b). In contrast to miRNAs, sRNAs have yet to be extensively used in circuit design, despite the fact that natural circuits with complex sRNA-based logic have been discovered and simple synthetic switches have been built.

最近，构造了一种混合系统，通过审查六个内源性miRNA标记，来识别HeLa癌细胞。该电路通过这些标记实现了与逻辑计算，检测HeLa细胞特异的外形表达，响应触发细胞凋亡基因（图5b）。与miRNAs不同的是，sRNA已经被广泛应用于电路设计，尽管事实上，基于复杂的sRNA逻辑的自然电路已被发现，简单的合成开关也已建成。

图 5-b

Protein-based biochemical circuits 基于蛋白质的生物分子电路 In biochemical systems, protein enzymes have taken the central role because they can be used to implement a ‘metabolic logic’, in which the inputs and the outputs are enzyme substrates and products. Examples include a network of coupled enzymes with four substrate inputs that computed a four-input logic function and a system that implemented a set of universal NOR and NAND gates. Extensive theoretical analysis has suggested ways of coping with noise and uncertainty in enzymatic circuits. These results point to new ways of controlling metabolism by computational integration of key intermediates. In a

different effort, self-assembling peptides were extensively investigated as building blocks for logic networks. Translating these approaches to cells may open new avenues in protein-based biological computing.

生化系统中，蛋白质酶已经担任了核心的角色，因为它们可以被用来实现一个“代谢逻辑”，他的输入和输出都是酶的底物和产物。例子包括具有四个基板输入的耦合酶素的网络，用于计算四输入的逻辑函数和实现一套通用的或非、与非门。广泛的理论分析提出了用酶电路处理噪声和不确定度。这些结果指出了用计算合成关键中间体控制新陈代谢的新途径。在不同的工作中，自组装肽被广泛研究作为逻辑网络的构建模块。将这些途径转译到细胞可能开辟基于蛋白质的生物计算的新渠道。

Protein-based biological switches and circuits 基于蛋白质的生物转换开关和电路

Most engineered protein circuitry in cells uses transcription factors. Because they are very well understood, transcription factors have long been the subjects of rational design and are currently the workhorses of circuit engineering. In addition to a large repertoire of native transcription factors, de novo switches can be engineered from recombinant transcription factors that combine protein domains from different organisms and their target promoters. 大多细胞蛋白电路设计使用转录因子。因为他们很好理解，转录因子一直是合理设计的核心，是目前电路工程的骨干。除了大部分的本地转录因子，更使开关可以从重组转录因子中设计，从不同的生物体和他们的目标启动子中结合蛋白结构域。

Transcriptional activation and repression are usually interpreted as ‘buffer’ gates (equivalency) and NOT gates, respectively. In addition, circuits

combining a number of transcriptional elements have been shown to implement analogue bandpass behaviour in response to a diffusing chemical, with the output produced at intermediate, but not at low or high, input levels. In multiple-input systems, two-input promoter logic can comprise OR, NOR and ‘AND NOT’ operations but rarely AND. This is because AND would require a hard-to-engineer interaction between two transcription factors.

转录激活和抑制通常被分别看作是“缓冲”门和非门。此外，结合一部分转录元件的电路已被证明，通过对扩散化学的响应模拟带通滤波，在不高不低的输入电平下，输出中间产物。在多输入系统，两个输入启动子的逻辑可以包括或，或非，与非运算，但很少出现与运算。这是因为与运算需要两个转录因子之间的相互作用，这对工程师来说都是很难的工作。

However, a further increase in complexity to three or more inputs in a rational fashion has proved to be refractory, apart from in a few examples that are based on rational design and promoter library screening. Therefore, most experimental transcriptional circuits use one- or two-input universal gates in parallel and in cascades to increase the total number of inputs (FIG. 6a). 然而，除了在少数的基于合理的设计和启动子文库筛选的例子中，合理的将复杂性扩展到三个或更多输入已被证明是难治性的。因此，大多数实验的转录电路通过一个或两个输入的通用门的并联和级联来增加总输入数（图6a）。

图 6-a

Transcription factors can serve as inputs to the logic circuits or as sensors of other molecules or agents that modulate their activity. So far, in most reported biological computing circuits, they have been used as transducers of small-molecule concentrations. In addition, light sensing was used in a bacterial circuit to implement an edge detection task and in mammalian cells to activate a therapeutic gene product. These examples point to a potential use

of optogenetics tools in computing networks.

转录因子可以作为逻辑电路或其它分子传感器或调节活性中介的输入。到目前，大部分的生物计算电路中，它们已被用来作为小分子的浓度传感器。此外，光传感技术用于细菌电路来实现边缘检测，和哺乳动物细胞中来激活基因治疗产品。这些例子指出了光遗传学工具在计算网络潜在用途。

Logical integration of endogenously expressed transcription factor activities for cell surveillance and control purposes still remains a major challenge. In a recent development, endogenous transcriptional activity was sensed indirectly by endogenous promoters driving separate components of a yeast two-hybrid activator with the purpose of identifying and destroying cultured cancer cells in which both promoters are active at the same time.

逻辑集成的内源性表达的转录因子用于细胞监视和控制的目的仍然是一个重大的挑战。在最近的发展中，为了识别和摧毁培养的癌细胞，他们的启动子活跃在同一时间，通过内源启动子驱动酵母双杂交剂的单独部分，间接检测内源性转录的活性。

Protein-based switches that do not involve transcription factors are ubiquitous in nature, but their engineering has lagged behind. Studies have shown that signalling processes such as mating and osmolarity pathways in yeast and two-component signalling in bacteria can be rewired. In recent work, the mating pathway in type a yeast cells was manipulated genetically by placing some of the pathway’s components under chemical induction, thereby implementing two-input gates such as [pheromone-α] AND [inducer] → GFP (FIG. 6b).

不涉及转录因子的基于蛋白的开关在自然界是无处不在，但他们的设计已经落后。

研究表明，信令流程如在酵母和细菌的双组分信号的交配和渗透性途径可以重新布线。在最近的工作中，通过在化学诱导下对一些组件进行基因操作实现酵母细胞的交配途径的排版，从而实现双输入门如[pheromone-α] AND [inducer] → GFP（图6b）。

图 6-b

‘Sender’ type-α strains were engineered to secrete pheromone-α following chemical stimulation, allowing co-cultured ‘sender’ and ‘receiver’ strains to generate complex logic circuits linking up to three chemical inputs to two fluorescent outputs. Despite these successes, we are far from having a kit of engineered enzymes to design de novo signalling circuits or proteins that are capable of integrating multiple inputs (similar to histones or p53 (REF. 97)). “发件人”type-α菌株在化学刺激下分泌pheromone-α，允许共同培养“发件人”和“接收”菌株产生复杂的逻辑电路，连接多达三个化学输入到两个荧光输出。尽管有了这些成就，我们离一个设计重始信号的工程酶工具包和能够集成多输入的蛋白质（类似于组蛋白和p53（参考文献97））还很远。

Another class of useful switches is based on protein interactions. One of the first studies in this direction showed a modular platform for controlling protein

function with two independent protein inputs, resulting in various logic behaviours. Yeast two- and three-hybrid systems can implement AND logic, and many ‘split’ protein systems — in which the original functional protein is split into two peptides that then interact through add-on binding moieties to reconstruct the function — were devised to achieve a similar two-input AND effect.

另一类有用的开关是基于蛋白质的相互作用。在这个方向上的第一个研究显示了，一个模块化平台用于有两个独立的蛋白质输入的蛋白质控制，结果得到了各种逻辑行为。酵母双杂交和三杂交系统可以实现与逻辑，许多“分裂”蛋白系统——原有的功能蛋白分为两个肽，然后通过扩展结合部分的互动来重建功能，被设计来实现得到类似双输入与门的效果。

Finally, protein degradation is an ‘off ’ switch that does not require transcription or translation machinery. Site-specific proteases from orthologous sources were shown to be useful in this respect. Yet scaling and generalizing these approaches remain major challenges that require improved tools for protein structure and interaction prediction.

最后，蛋白质降解是一个断开开关，不需要转录或翻译机制。来源于直系同源的位点专一的蛋白酶被证明在这方面是有用的。然而，缩放和推广这些方法仍然是主要的挑战，需要改进用于预测蛋白质结构和相互作用的工具。

To summarize, the approaches that have delivered the largest computations so far are based on simple, natural-like switching, such as transcription regulation with transcription factors and post-transcriptional regulation by RNAi. Transcriptional regulation is amenable to cascading, whereas in post-transcriptional regulation the size grows laterally in a normal-form-like

circuitry. Other approaches require a great deal of engineering of individual elements and thus have only been shown to work for a small number of inputs; their potential for scaling up remains the subject of future work.

总而言之，到目前为止交付最大计算量的都是基于简单的、倾向于自然的开关，如转录因子的转录调控和RNAi的转录后调控。转录调控可以实现级联，而在转录后调节中，正常电路的大小会横向生长。其他方法所需要的个体元素的设计量很大，因此只有输入很少的工作得到了验证；规模扩大的潜力仍然是未来工作的主旨。

Experimental state machines 状态机实验

Self-assembly as biochemical Turing computation 自组装生物图灵的计算 Self-assembly — a hallmark of complexity — is ubiquitous in biology. Computer science has shown that self-assembly processes can be described as computations and that any computation can be implemented with self-assembling building blocks. In particular, rectangular tiles with differently coloured edges that stick together when their edges match can encode a Turing machine program and can assemble into intricate two-dimensional pat-terns (Supplementary Information S1 (figure)). Molecular implementation of this concept was first proposed and implemented for a periodic pattern of planar DNA tiles (FIG. 7a) and later for an aperiodic assembly that computed four consecutive XOR operations.

自组装—复杂性的标志—在生物中无处不在。计算机科学研究表明，自组装的过程可以被描述为计算，任何计算也可以通过自组装建筑模块实现。特别是，不同边界着色的长方形瓷砖，当他们的边界匹配可以编码一个图灵程序，或者可以组装成复杂的二维模型时，他们会粘合在一起（补充信息S1（图））。这一概念的

分子实现首次被提出，并用这种概念实现了平面DNA瓷砖的周期方向图（图7a），后来的非周期性自组装计算了四个连续的异或运算。

图 7-a

Algorithmic two-dimensional assembly has been extensively studied using counting as a benchmark because incremental counting is encoded in a small-size Turing machine program. Multiple challenges, such as error propagation, had to be overcome for experimental realization. Recently, counting to 12 with minimal errors has been achieved.

算术的二维组件已经被广泛的视为一个基准，因为增量计数是用一个小型的图灵程序编码的。但仍存在多重挑战，如误差蔓延，为了试验验证必须克服。最近，

已经通过最小误差数到12。

This research dovetails with other developments in DNA nanotechnology that are geared less towards complex computations and more towards interesting structural and dynamical features. These have resulted in robust generation of two-dimensional and three-dimensional objects from hundreds of deoxyoligo-nucleotides and in molecular-size DNA ‘walkers’. These nanostructures are often viewed as the basis for engineered biological organelles and transport, respectively.

这项研究与倾向于更复杂的计算、更有趣的结构和动力学特征的DNA纳米技术发展相吻合。这些都导致了从数百个脱氧寡核苷酸得到的健壮的二维和三维对象和化学分子尺寸的行走者。这些纳米结构往往分别被视为设计生物细胞器和运输的基础。

Supporting this notion, RNA building blocks that contain enzyme-binding aptamers have recently been shown to self-assemble into organelle-like structures in bacteria and have been reported to boost the efficiency of an enzyme-catalysed hydrogen-producing process by proximity effects. In another example, a nano-sized DNA container carrying molecular cargo was programmed to bind to cells and to open and release the cargo only when the cells expressed two surface markers simultaneously, bringing computing back into the picture and pointing towards a novel logic-driven approach to selective drug delivery in vivo.

为了支持这一观点，含有酶结合寡核苷酸适配子的RNA建筑模块最近被验证，在细胞自组装形成细胞器结构，而且已经被报道，可以通过邻近效应推进酶促 氢生产过程的效率。另一个例子中，一个DNA分子纳米集装箱运载货物的程序被绑

定到细胞上，只有在细胞的两个表面的标志同时表达时，打开和释放的货物，实现想象中的计算，指出了全新的体内逻辑驱动的选择性药物输送方法。

Tape-based biochemical state machines 基于磁带的生化状态机

The analogy between the state machine tape and biopolymers inspired ideas about molecular state machines, although the controller and the transition rules do not have obvious biological counterparts. In a number of non-autonomous theoretical blueprints, manual biochemical manipulations were proposed as the means to implement transitions. A hypothetical autono-mous Turing machine made of RNA molecules and modifying enzymes was put forward as an example of an energy-efficient computer.

状态机的磁带和生物大分子之间的类比激发了分子状态机的想法，虽然控制器和转换规则还没有明显的生物副本。在一些非自治理论的设计中，提出了人工生物化学操作的方法来实现转换。提出一个假想的RNA分子和修饰酶构成的自治图灵机，是节能计算机的一个例子。

Later, protein translation inspired a blueprint for an autonomous Turing machine as an interactive control centre of cellular processes. Experimentally, state-to-state transitions were first implemented by programmable DNA exten-sion with DNA polymerase using a template strand with ‘transition rule’ sequences. Each rule contained a binding site for the 3? end of the DNA molecule (which is the current state) and the extension template (which is the new state), such that after multiple rounds of annealing and extensions the molecule contained the history of all intermediate steps.

稍后，蛋白翻译激发了一个自治图灵机的设计，作为细胞处理的交互控制中心。

通过实验，状态转换第一次被程序化DNA 扩展模版实现，使用带有转移规则序列的模板链的DNA 聚合酶（代表新的状态）。每个规则包含一个DNA分子3?端的结合位点（代表目前的状态）和扩展模板（代表新的状态），这样在多轮的退火扩展后，分子包含所有中间步骤的历史。

Later, a finite automaton was generated in which the transitions were executed by autonomous biochemical steps based on DNA sticky-end recognition, ligation and digestion. The machine was fully programmable within its specifications — an alphabet of two symbols and the possibility of two states. This state machine was later augmented with sensory input channels that allowed it to interrogate biological signals in their native format: namely, mRNA levels.

随后，一个有限自动机产生了，交接工作基于DNA粘端识别、结扎和消化的自动生化步骤。这台机器在他的性能规范里是完全可编程的，使用一个包含两种符号的字母表和两个状态的可能性。该状态机后来增强了感觉输入通道，允许在生物信号的原有格式审查分子：即mRNA水平。

The resulting system was one of the first fully functional biochemical prototypes of a sensor– computation–actuation cycle, producing a biologically active molecular output only when a certain combination of input conditions was detected in the computer’s environment (FIG. 7b). In addition, automaton building blocks were used in molecular-level programming language. 由此产生的系统是第一个功能型生化原型传感器–计算器–周期驱动，只有当一定的输入条件组合在计算机环境中能检测到时，才产生生物活性分子输出（图7B）。此外，在分子层次的编程语言中使用了自动机构建模块。

图 7-b

Event-triggered biological state machine computation 事件触发的生物状态机的计算

This concept has been explored using recombinases and invertases as the state switches (see above) and their substrate DNA as state indicator. It was shown that a hypothetical system with 10 invertases could encode ~107 states, and an experimental circuit with two invertases was tested in bacteria. Invertase is a reversible activity, resulting in an equilibrium between two DNA sequences of opposite orientation and a mixture of different states. This was overcome in another study that used a set of invertases to count the number of chemical pulses in yeast cells.

已经使用重组酶和转化酶作为状态切换（见上文），以基底DNA为状态指示器。结果表明，用10蔗糖酶综合系统可编码107多个状态，在细菌中对两个转化酶形成的实验电路进行检测。转化酶活性是可逆的，导致两个相反方向的序列中存在一个平衡和多个状态的混合。这在另一项研究中得到了克服，在酵母细胞中使用一套转化酶计数化学脉冲的个数。

In this system, an inversion was coupled to concurrent inactivation of the invertase gene and activation of another invertase, resulting in high transition efficiencies and the required sequence of individual transitions (FIG. 7c). An event-triggered string transducer was also demonstrated biochemically; a sophisticated system of DNA-based aptamers took on different internal configurations in response to chemical inputs.

在这个系统中，一个反向被耦合到并发的一个蔗糖转化酶的失活和另一个蔗糖转化酶的激活，得到一个高的转化效率和所需的单个转换序列（图7c）。一个事件触发的字符串传感器也被生化验证；一个复杂的基于DNA的核苷酸适配子的系统，针对不同的化学输入采取不同的内部配置。

图 7-c

Methodological challenges 方法论的挑战

Biochemical and biological computing circuits present unique characterization challenges. These circuits, after they have been implemented, are indistinguishable from any natural biochemical or biological network, and they

本文来源：https://www.bwwdw.com/article/lp53.html