CPSC 545/445 (Autumn 2003) - Class 21:  Microbial Engineering and DNA-based Computing
Module 7, Part 2

--
7.2 microbial engineering (Knight & Sussman Jr., 1997):

idea:
construct digital logic signals and gates within a living cell

purpose:
- engineer chemical behaviour of cells for use of sensors and effectors
- control fabrication processes on molecular scale

digital abstraction allows to ignore fine details of complex phenomenon

build on available biological mechanisms (modified version of naturally
occurring mechanisms)

here: build logic gates;
	represent signals by concentrations of DNA binding proteins
	implement non-linear amplification by in vivo 
		DNA-directed protein synthesis

basics:
- enzymes effectively switch chemical reactions on and off
- mRNA and proteins are continually degraded in the cell
  	mRNA: rapid degradation (30s-20min in E.coli); 
	proteins: deg rate is sequence dependent
- transcription repressors: typically bind to sites overlapping RNA polymerase
	binding sites - use repressor DNA binding proteins for building
	logical gate functionality

- digital logic gates essentially need high-quality nonlinearity:
	near 0, 1: low gain; in between: high gain
- biological realisation of nonlinear behaviour:
  - protein dimers as active form; in equilibrium, concentration of dimer
	= concentration of monomor ^ 2
    multimers realise higher power nonlin.
  - cooperative binding of proteins to substrate:
    first of several prot binding reactions at low rate;
    presence of bound protein enhances binding affinity
    -> subseq binding reactions occur more rapidly
  
example for naturally occurring genetic switch:
  lytic/lysogenous switch in cells infected with lambda phage
  (Mark Ptashne: A Genetic Switch: Phage lambda and Higher Organisms";
	2nd edition, Cell Press, Cambridge, 1992)

implementation of logical inverter:

input, output: DNA binding proteins A, B
(could also be other enzymes, e.g., effecting or detecting
	motion, chemoluminescense, chemical reactions, ...)

A, B both act as repressors for transcription of B_G, the gene coding for B.

B_G has three states: B_G A  and B_G B = inactive (repressed); B_G = active

-> kinetic equations of chemical model reactions give desired
	inverter behaviour (static transfer curve similar to NMOS inverter;
	simulations of dynamic behaviour give reasonable response curve
	with and without load; successful ring oscillator simulation)

implementation issues:

- gates are slow: 
  electronic gate: delay ~10 ps (pico sec) 
  biological gate: delay governed by speed of protein manufacturing -> minutes!
	-> mHz rates!

- complexity limitations:
  - need sufficient number of distinct DNA binding proteins
	that are not used elsewhere within host cell control
	-> maybe tens-hundreds of naturally existing proteins
  - finite volume of cell -> limited concentrations of proteins

- cell cycle coordination
  - total number of copies of given gate within cell varies over time
	(DNA replication prior to cell division)

- choice of host organism
  - simplest possible organism
	e.g., Mycoplasma capricolum -> genome is sequenced, but difficult 
	to culture, limited lab experience 
  - well-studied model prokaryote:
	E.coli: extensive experience & wide range of tools
		and techniques available
  - eventually: maximally simplified cell (engineered)


applications:
- medical: control of biological processes
- manufacturing novel materials and structures at molecular scale
	(nanotechnology); particularly: computing devices
	use slow biological systems as machine shops;
	protein = machine tools; DNA = control tape


---
7.3 DNA-based Biomolecular Computation

Basic idea: encode information in DNA strands,
	perform computation by means of standard biochemical techniques


Solution-based approach:
Adleman's Hamiltonian Path Experiment (Adleman, 1994)

Hamiltonian Path Problem (HPP): 
Given a directed graph G=(V,E) and vertices v,v', 
decide whether there is a (linear) path from v to v' 
that visits every vertex in V exactly once (using edges from E).

Note: HPP is NP-complete (closely related to Travelling Salesperson Problem)

Here: simple instance w/ 7 vertices v_1...v_6, 14 edges

Randomised Algorithm for HPP:

1) Generate random paths through G
2) Keep only paths from v to v'
3) Keep only paths with exactly n vertices
4) Keep only paths that enter each vertex in V at least once
5) If set of paths != empty, output "yes", otherwise, output "no".


DNA-based implementation:

Encode each vertex i by randomly generated 20mer O_i
Encode each edge (i,j) by 20mer O_{i->j} obtained by 
	concatenating 3' 10mer of O_i and 5' 10mer of O_j
	(except O_{0->j} = O_0 + 5' 10mer of O_j,
		O_{i->6} = 3' 10mer of O_i + O_6)

Step 1: Combine O_{i->j} oligos and complements of O_i oligos, ligate
	-> random paths in G (double-stranded DNA)
	Note: complement hybridisation ensures only correct paths
		are formed, i.e., paths that use edges in E

Step 2: PCR amplification, using primers O_0 and compl(O_6)
	-> only paths from v_0 to v_6 are amplified

Step 3: Electrophoretic separation, excise 140bp band
	-> only paths of length n=7 are retained
	(multiple PCR + gel purification were used to enhance purity)

Step 4: Affinity purification using biotin-avidin magnetic bead system,
	using 5 stages with magnetic beads with conjugated 
	oligos compl(O_1) ... compl(O_5)
	-> only paths that visit all of v_1 ... v_5 are retained
	(note: as a result of Step 2, all paths considered here
	visit v_o, v_6)

Step 5: PCR amplification with primers O_0 and compl(O_6), 
	-> obtain product = "yes" answer, no product = "no" answer

(Note: can use graduated PCR with primers O_6 and compl(O_1) ... compl(O_6), 
	electrophoretic separation to read out the actual path)


Note:
- main advantage over conventional computation:
  massively parallel, molecular scale
- entire procedure required 7 person-days of lab work
  (most time-consuming step: Step 4)
- number of different oligos is linear in size of graph
- quantity needed in step 1 = exponential in size of graph
  (but this can be somewhat improved using a better algorithm for HPP)



---
resources:


microbial engineering:

http://www.ai.mit.edu/people/tk/tk.html
-> webpage of tom knight, mit 

http://www.ai.mit.edu/people/tk/ce/microbial-engineering.html
-> tom knight's links on micr engin.


Thomas F. Knight, Jr. and Gerald Jay Sussman: Cellular Gate Technology
MIT Artificial Intelligence Laboratory, July 1997
[cellgates.ps]
-> http://www.ai.mit.edu/people/tk/ce/cellgates.ps

http://www.swiss.ai.mit.edu/~rweiss/bio-programming/index.htm


DNA-based computation:

L.Adleman: Molecular Computation of Solutions to Combinatorial Problems.
	Science 266(5187):1021-1024

DNA Computer. http://people.cornell.edu/pages/zf25/dna_computer.html

Lagally Research Group. http://mrgcvd.engr.wisc.edu/lagallygroup/research/DNA/DNA_Computation.htm