© 1999 Robert A. Freitas Jr. All Rights Reserved.

Robert A. Freitas Jr., Nanomedicine, Volume I: Basic Capabilities, Landes Bioscience, Georgetown, TX, 1999

10.2.3.1 Biochemical Computers

The Systron-Donner Analog Computer familiar to engineering students in the 1960s and 1970s used potentiometers and high-gain DC amplifiers, called operational amplifiers, to allow the creation of electrical circuits comprised of adders, multipliers, and integrators arranged in feedback loops capable of solving complex time-dependent differential equations modeling actual physical systems. In 1998, Motorola and others were producing programmable analog chips containing arrays of operational amplifiers and other analog components. Related control systems are commonplace in human physiology.⁷¹

Biochemical feedback loops exemplified by enzyme activity in the living cell exhibit similar analog computational functionality.^{2695,3509,3543,3544} To start with, the net flow of carbon through an enzyme-catalyzed reaction may be influenced by

1. the absolute quantity of enzyme present,

2. the pool of nonenzyme reactants and products available, and

3. the catalytic efficiency of the enzyme.⁹⁹⁶

All three influences may themselves be subject to recursive enzymatic control. For instance, the absolute quantity of enzyme present is determined by competing rates of synthesis and degradation; both of these processes are controlled by other enzymes. Product synthesis may be triggered by an inducer, or may be repressed by an end product. Local concentrations of coenzymes or metal ions can regulate catalytic efficiency. Inhibited activity of an enzyme in a biosynthetic pathway by an end product of that pathway, such as the much-studied inhibition of aspartate transcarbamoylase by cytidine triphosphate (CTP),^3689-3691 is called feedback inhibition.^3689-3693 Typically the feedback inhibitor, acting as a negative allosteric effector, binds to the sensitive enzyme at an allosteric site that is spatially distinct from the catalytic site. Further fine control is provided by additional multiple feedback loops, exhibiting such features as cumulative feedback inhibition (e.g., the inhibitory effect of two or more end products on a single regulatory enzyme is strictly additive),^3694-3696 multivalent feedback inhibition (e.g., complete inhibition occurs only when two or more end products both are present in excess),^3697,3698 and cooperative feedback inhibition (e.g., single end product present in excess inhibits the enzyme, but inhibitory effect of two or more end products present in excess is "multiplicative").^3699,3700 Similarly, parallel chemical computers consisting of open, bistable coupled reaction systems³⁵⁴² capable of pattern storage and pattern recognition have been proposed and simulated to model a neural network,¹⁹¹¹ and experiments have been performed with a system of 16 coupled bistable chemical oscillators using the chlorite-iodide reaction.¹⁹¹²

Networks of interacting enzymatic reactivities can also be used to create biochemical digital computers.²⁸⁵¹ An enzymatic Turing machine was first proposed by C.H. Bennett,²⁹⁶ and Liberman³⁴⁷¹ described the living cell as an analog-digital molecular stochastic parallel computer. More recently, T. Knight of the MIT Artificial Intelligence Laboratory suggested that normal cellular activities can be coopted to implement digital logic gates, a process he calls "cellular gate technology."^1962,3544 In this scheme, protein concentrations are used as binary signals. The signal is "1" if the marker protein is being synthesized, thus reaching some detectable threshold equilibrium value where the rates of synthesis and degradation are equal. The signal is "0" if the synthesis of the marker protein is being inhibited by some other protein, and thus the concentration is trending toward zero.

This principle may be used to make logical circuit elements. For example, an inverter system consists of two different proteins, one of which is inhibiting the synthesis of the other. A NOR gate requires a system in which either of two proteins can inhibit the synthesis of a third protein; multivalent feedback inhibition in which two proteins are both required to inhibit production of a third makes a NAND gate. This is sufficient to implement any Boolean function, given enough gates and signals. Two NORs crossed in a loop makes a bistable multivibrator, or flip-flop, a fundamental component in data registers, shift registers, rectangular pulse generators, or pulse delay elements. The logic circuitry (the arrangement of "gates" and "wires") comprises a set of designed DNA sequences that determine which proteins are being synthesized and which combinations of proteins can inhibit the synthesis of others.

In the most optimistic case, we assume operation near concentrations of c_prot ~ 10^-4 gm/cm³ (similar to most complement factors and some clotting factors in human blood; Appendix B), marker proteins and gating enzymes of mass MW_prot ~ 10 kilodaltons, and a minimum of N_prot ~ 100 protein molecules per gate, giving a typical logic element size of V_gate ~ N_prot MW_prot / N_A c_prot ~ 0.02 micron³/gate where N_A is Avogadro's number. Flip-flop registers may store ~50 bits/micron³. In theory, the DNA specifications for this computer could be grafted onto the genome of an animal cell (~20 micron³ = up to 1000 gate volumes) or a bacterial cell (~2 micron³ = up to 100 gate volumes), taking care to avoid any unwanted interactions with natural biochemical pathways. Minimum diffusion-limited gating time across a biochemical gate of characteristic dimension DX ~ V_gate^1/3 is given by Eqn. 9.80 as t = DX² / 2 D ~ 0.3 millisec assuming protein marker diffusivity D ~ 10^-10 m²/sec (Table 3.3), consistent with ~KHz maximum operating frequencies when DNA is localized near the gates. For nucleoplasmic DNA with cytoplasmic computation, diffusion-limited maximum switching frequency falls to ~0.1-1 Hz. For comparison with natural systems, genes controlling the fibroblast cell proliferation program have been observed to switch states in ~900 sec,²⁶⁸³ implying computation at ~0.001 Hz.

Other classes of biochemically-modulated biomolecular switches have been investigated. For example, Matthews¹⁹²⁹ incorporated an artificial molecular onoff switch into an enzyme by adding a pair of thiol groups into the active site of native lysozyme, replacing two amino acids on opposite side of the active site with cysteine, an amino acid with a thiol-containing side chain. Thiols may form covalent bridges (-SS-) or broken bridges (-SH HS-) under suitable chemical conditions. Cycling the solution composition causes the cysteines to form a bridge across the active site (inactivating the enzyme) or to break the bridge (activating the enzyme); the process is completely reversible. Another experiment¹⁹³⁰ involved genetic modification to the bacterial phage lambda, a virus with two behavioral states called lytic (replication-active) and lysogenic (replication-inactive). The NIH researchers installed a synthetic molecular switch¹⁹⁷⁶ in lambda that allows the virus to be biochemically toggled between its lysogenic and lytic phases by the introduction of active HIV-1 protease. The genetic circuit that controls the phage lambda lysis-lysogeny decision has been flowcharted in detail using Boolean logic operators.²²³ The immunological synapse³⁴⁵³ -- the specialized junction between a T cell and an antigen-presenting cell -- acts as a kind of biochemical logic gate with a ~0.06 Hz switching rate.^3453-3455

In 1998, the DNA molecule also was being actively investigated for biochemical computers. DNA-based Turing machines were first proposed by C.H. Bennett and R. Landauer.^296,1894 In 1994, L. Adelman demonstrated the first DNA computer^1895,1896 using fragments of DNA to compute the solution to a simple graph theory problem. Adleman used short DNA sequences to represent vertices of a network, or graph. Combinations of these sequences were then synthesized randomly by massively parallel chemical reactions in aqueous solution using a combination of the copying and combination reactions applied to the artificial DNA strands, comprising all possible random paths through the graph, after which a sequence representing the designed result was biochemically extracted from the stew.

Initially it was thought that DNA computing would be limited to the solution of combinatorial problems,¹⁸⁹⁷ but subsequent research has shown that the approach can be applied to a much wider class of digital computations.^1898-1902 The well-known problem of combinatorial explosion (e.g., even a small protein of 300 amino acids has a sequence space of 10³⁹⁰) leading to huge-volume DNA libraries during a complex computation may be avoided using a recursive selection approach¹⁸⁹⁹ crudely analogous to genetic algorithms. Some work has been done using plasmids and restriction endonucleases to implement a DNA-based Turing machine, using only commercially available restriction enzymes and ligases for every operation with states represented as sequences of bases around a plasmid,¹⁸⁸⁶ and Warren Smith has another Turing machine design using the chemistry of guide RNAs in Trypanosome kinetoplasts.³⁷⁰¹ Practical challenges include finding fast, efficient, and low-noise input and output techniques.^1902,2331

DNA computation is in theory energetically efficient, requiring only ~12 kT per ligation operation at room temperature, with massively parallel operations although each ligation reaction requires ~1 sec to complete.¹⁸⁹⁶ DNA data representation allows up to ~1 bit/nm³ storage density.¹⁸⁹⁵ However, a practical mass storage device would probably require 10-100 nucleotide-long words to avoid ambiguities during the recall process,¹⁹⁰³ thus reducing maximum density to 10⁷-10⁸ bits/micron³, though of course some of this reduction could be offset by reducing read rate requirements (e.g., by using multiple read heads with majority logic).

Last updated on 24 February 2003