Engineering Complex Phenotypes in Industrial Strains E-Book

Ranjan Patnaik

• Herausgeber: John Wiley & Sons
• Kategorie: Fachliteratur
• Sprache: Englisch

Seitenzahl: 488

Veröffentlichungsjahr: 2012

Beschreibung

This book highlights current trends and developments in the area of engineering strains. The book details the current and future tools used in the production of bulk chemicals and biofuels from renewable biomass using green technologies. Complex phenotypes are traits in a microbe that requires multiple genetic changes to be modulated simultaneously in the microorganism's DNA. Knowing what those genetic changes are for a given trait and how to make those changes in the most efficient way forms the motivation behind writing this book. This book explains the newer tools to develop and enable engineered strains at time scales much faster that the natural evolution process so that we can increase a cells' production of a certain substance, increase process productivity, and extend metabolic capability. This book provides a one stop platform of reference for practicing researchers in the field of industrial biotechnology. This book also explains that the commercial success of a process that uses microbial catalysts over platforms that use chemical catalysts and fossil fuels depends on the time it takes to engineer these microbes to perform the desired reaction under harsh manufacturing conditions and at rates that meets the criteria for economic feasibility.

Leseprobe

Table of Contents

COVER

TITLE PAGE

COPYRIGHT PAGE

FOREWORD

PREFACE

CONTRIBUTORS

1 CLASSICAL STRAIN IMPROVEMENT

1.0 INTRODUCTION

1.1 THE APPROACH DEFINED

1.2 MUTAGENESIS

1.3 GENOTYPIC LANDSCAPES

1.4 SCREENING

1.5 CONCLUSIONS

2 TRACER-BASED ANALYSIS OF METABOLIC FLUX NETWORKS

2.0 INTRODUCTION

2.1 SETTING UP A STOICHIOMETRIC NETWORK MODEL

2.2 SMALL-SCALE MODELS VERSUS GENOME SCALE MODELS

2.3 NETWORK ANALYSIS: MAXIMUM THEORETICAL YIELD

2.4 (STOICHIOMETRIC) METABOLIC FLUX ANALYSIS

2.5 CARRYING OUT A LABELING EXPERIMENT

2.6 MEASURING ISOTOPE LABELING PATTERNS

2.7 TRACER-BASED MFA

2.8 VALIDATING METABOLIC FLUX NETWORKS

2.9 CONCLUSIONS

ACKNOWLEDGMENTS

3 INTEGRATION OF “OMICS” DATA WITH GENOME-SCALE METABOLIC MODELS

3.0 INTRODUCTION

3.1 GENOME-SCALE METABOLIC NETWORKS

3.2 CONSTRAINT-BASED MODELING THEORY

3.3 CURRENT ANALYSIS OF OMICS DATA

3.4 NEW APPROACHES TO DEVELOPING MODEL CONSTRAINTS

3.5 USE OF GENE EXPRESSION DATA IN METABOLIC MODELS

3.6 USE OF METABOLOMICS DATA IN METABOLIC MODELS: TMFA EXAMPLE

3.7 INTEGRATION OF MULTIPLE OMICS DATA SETS

3.8 FUTURE DIRECTIONS AND APPLICATIONS TO STRAIN ENGINEERING

4 STRAIN IMPROVEMENT VIA EVOLUTIONARY ENGINEERING

4.0 INTRODUCTION

4.1 METHODOLOGIES FOR EVOLUTIONARY ENGINEERING

4.2 EXAMPLES OF EVOLUTIONARY ENGINEERING

4.3 CONCLUSIONS AND FUTURE PROSPECTS

ACKNOWLEDGMENTS

5 RAPID FERMENTATION PROCESS DEVELOPMENT AND OPTIMIZATION

5.0 INTRODUCTION

5.1 OVERVIEW OF CLASSICAL FERMENTATION PROCESS DEVELOPMENT METHODOLOGY

5.2 FERMENTATION PROCESS DEVELOPMENT AND OPTIMIZATION

5.3 RAPID PROCESS DEVELOPMENT AND OPTIMIZATION USING CONVENTIONAL FERMENTATION SYSTEM

5.4 STRAIN EVALUATION AND PROCESS OPTIMIZATION UNDER SCALE-DOWN CONDITIONS

5.5 CONTROL AND SENSOR TECHNOLOGIES FOR MINIBIOREACTOR

5.6 COMMERCIAL HIGH-THROUGHPUT FERMENTATION SYSTEMS

5.7 TRENDS IN DEVELOPMENT OF HIGH THE GREATA-THROUGHPUT MINIBIOREACTOR SYSTEM

5.8 CASE STUDIES OF FERMENTATION PROCESS DEVELOPMENT AND OPTIMIZATION USING HIGH-THROUGHPUT MINIBIOREACTORS

5.9 CONCLUSIONS AND THE PATH FORWARD

6 THE CLAVULANIC ACID STRAIN IMPROVEMENT PROGRAM AT DSM ANTI-INFECTIVES

6.0 INTRODUCTION

6.1 THE BIOSYNTHETIC PATHWAY TO CLAVULANIC ACID

6.2 THE STRATEGY FOR IMPROVEMENT OF MULTIPLE COMPLEX PHENOTYPES

6.3 RESULTS AND DISCUSSION

6.4 FUTURE PERSPECTIVES

ACKNOWLEDGMENTS

7 METABOLIC ENGINEERING OF RECOMBINANT E. COLI FOR THE PRODUCTION OF 3-HYDROXYPROPIONATE

7.0 INTRODUCTION TO BIOSYNTHESIS OF 3-HYDROXYPROPIONIC ACID

7.1 ORGANIC ACID TOXICITY

7.2 UNDERSTANDING 3-HP TOXICITY

7.3 STRAIN DESIGN

7.4 COMBINING 3-HP TOLERANCE AND 3-HP PRODUCTION

7.5 SUMMARY

8 COMPLEX SYSTEM ENGINEERING: A CASE STUDY FOR AN UNSEQUENCED MICROALGA

8.0 HISTORICAL PERSPECTIVE

8.1 ANALYSIS OF ALGAL BIOMASS COMPOSITION

8.2 DEVELOPMENT OF HYPOTHESIS-DRIVEN STRAIN IMPROVEMENT STRATEGIES

8.3 IMPLEMENTATION OF BIOLOGICAL TOOLS I— DEVELOPMENT OF A TRANSFORMATION SYSTEM

8.4 IMPLEMENTATION OF BIOLOGICAL TOOLS II—DEVELOPMENT OF A SELF-LYSING, OIL-PRODUCING ALGA FOR BIOFUELS PRODUCTION

8.5 CONCLUDING REMARKS

ACKNOWLEDGMENTS

9 MEIOTIC RECOMBINATION-BASED GENOME SHUFFLING OF SACCHAROMYCES CEREVISIAE AND SCHEFFEROMYCES STIPTIS FOR INCREASED INHIBITOR TOLERANCE TO LIGNOCELLULOSIC SUBSTRATE TOXICITY

9.0 INTRODUCTION

9.1 METHODOLOGY

9.2 RESULTS AND DISCUSSION OF STRAIN DEVELOPMENT

9.3 CONCLUSIONS AND FUTURE DIRECTIONS

INDEX

Copyright © 2013 by John Wiley & Sons, Inc. All rights reserved.

Published by John Wiley & Sons, Inc., Hoboken, New Jersey.

Published simultaneously in Canada.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, [phone], fax [phone], or on the web at www.copyright.com. Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, [phone], fax [phone], or online at http://www.wiley.com/go/permissions.

Limit of Liability/Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose. No warranty may be created or extended by sales representatives or written sales materials. The advice and strategies contained herein may not be suitable for your situation. You should consult with a professional where appropriate. Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.

For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at [phone], outside the United States at [phone] or fax [phone].

Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic formats. For more information about Wiley products, visit our web site at www.wiley.com.

Library of Congress Cataloging-in-Publication Data:

Engineering complex phenotypes in industrial strains / edited by Ranjan Patnaik.

p. cm.

 Includes bibliographical references and index.

ISBN 978-0-470-61075-6 (hardback)

 1. Industrial microorganisms. 2. Genetic engineering. I. Patnaik, Ranjan, 1969–

 QR53.E54 2012

 579'.163–dc23

2012015254

FOREWORD

The increasing demands for renewable chemicals, materials, and fuels, as well as the continuing evolution of capabilities in biology, chemistry, and engineering, are giving rise to significant efforts in using biotechnological approaches in new process configurations. These approaches are particularly well suited to conversions of carbohydrate and other biological starting compounds into useful materials, as enzymes and microbes naturally transform these substances. Building on a fair history of industrial use of microbes in the production of high-value, low-volume materials, such as pharmacologically active compounds, vitamins, and amino acids, we are now extending these approaches to the production of higher volume/lower value chemicals, such as monomers for making polymers, lubricants, and fuels. As we progress up this volume curve, the demands on the bioprocess become more and more stringent, and highly integrative approaches among disciplines are required to produce the biocatalysts and associated processes necessary for commercially viable outcomes.

Coincident with this evolution, a number of books and monographs have appeared on the subject of metabolic engineering and systems biology, and the primary literature is becoming more and more detailed. With this backdrop, this book does not attempt to be an authoritative reference on tools and techniques, but rather focuses on the strategies and approaches that enable commercial biocatalyst design. It should be of use to graduate students and early career professionals in the field, or to other generalists and professionals from related disciplines who are eager to grasp the basic tenets of engineering biocatalysts. In addition, it may well be of value in providing corporate managers and government officials with insights into the requirements for successful program outcomes.

This book gives an overview of current approaches, with examples drawn from academia and industry and covering biocatalysts ranging from Escherichia coli and Steptomyces to yeast and microalgae. The vitality of the field is exemplified by the relatively young ages of the contributors, who are shaping the field with their novel approaches, and the inclusion of case studies adds a realistic dimension to the exposition.

JOHN PIERCELondon

PREFACE

This book highlights current trends and developments in the area of engineering industrial strains for the production of bulk chemicals and biofuels from renewable biomass. The commercialization of bioprocesses derived from the use of superior engineered strains often requires the balance between unknowns and trade-off between multiple complex traits of the biocatalyst. Complex phenotypes are traits in a microbe that require more than one genetic change (multigenic) to be modulated simultaneously in the microorganism’s genome for full expression. Knowing what those genetic changes are for a given trait and how to manipulate those targets in the most efficient way, forms the motivation for writing this book. The chapters address tools and methodologies developed for engineering such complex traits or phenotypes in industrial strains. Emphasis is on the multidisciplinary (metabolic engineering, screening, fermentation, downstream) nature of the approach or strategy that is used during the course of developing such a commercial biocatalyst. Keeping in perspective the multidisciplinary nature of activity and the interests of a broader range of readers, the topics included in the chapters are not meant to be fully exhaustive in their respective areas; rather, the emphasis is on comparison and integration of different tools and objectives. Chapters 1–5 summarize broadly the current tools and technologies available for engineering a complex phenotype in an industrial strain with brief reference to examples, while Chapters 6–9 highlight in detail the application of such tools and methodologies in the form of case studies.

Chapter 1 summarizes the age-old proven approach for engineering in­­dustrial strains using mutagenesis, followed by screening or selection, often termed classical strain engineering (CSI). Discussions of the applicability of CSI for engineering complex traits provide information on its suitability and limitations. Chapter 2 describes the current state of the art in the use of 13C tracer-based analysis and metabolic flux analysis for engineering complex pathways. Chapter 3 describes the utility of genome-scale models by integration of “omics” technology and physiological data to address engineering of complex traits.

The probability of commercial success of a bioprocess that uses microbial catalysts and renewable feedstocks, as compared with platforms that use chemical catalysts and fossil fuel-derived feedstocks, greatly depends on the time it takes to engineer these microbes to perform the desired reaction under harsh manufactur­ing conditions at rates, titers, and yields that meet the criteria for economic feasibility. Chapter 4 addresses new evolutionary strain engineering approaches that are superior to CSI in developing complex traits rapidly. Transitioning from laboratory-scale demonstration to commercial-scale operation is not only time-consuming but also expensive, especially with the uncertainties associated with scalability of complex traits. Chapter 5 describes an integrative platform for rapid fermentation process development and strain evaluation that not only minimizes the number of false positives from a strain engineering program but also provides a cost-effective approach to optimize fermentation conditions.

Chapter 6 is a case study on the use of CSI (Chapter 1) and improved strain screening strategies (Chapter 5) at Dutch State Mines for engineering Streptomyces clavuligerus for commercial production of anti-infectives. Chapter 7 is a case study on the use of evolutionary approaches (Chapter 4) at Opx Biotechnologies for improving tolerance of Escherichia coli to 3-hydroxypropionoic acid. Chapter 8 is a complete strain engineering case study from the National Renewable Energy Laboratory in an unsequenced microalga, Chlorella vulgaris, for production of biofuels. The authors have highlighted integration of improved analytics and strain screening approaches (Chapters 1 and 5) with “omics” technology (Chapter 3) for addressing needed improvements in multiple complex traits. Chapter 9 demonstrates the feasibility of using genome-shuffling approaches (Chapter 4) in Saccharomyces cerevisiae and Schef­feromyces stiptis for improving tolerance to inhibitors in lignocellulosic substrates.

Scientists, engineers, and project managers who are leaders in their respective areas of research and drawn from diverse fields of science and engineering have contributed to the above chapters. The book has attempted to capture the thought processes on which they so often rely during the initiation and development of a commercial biocatalyst project. I hope the readers find the content of the book to be intellectually satisfying.

I would like to thank the editors at John Wiley & Sons for being patient and for their cooperation during the course of this project.

RANJAN PATNAIK

CONTRIBUTORS

Hal S. Alper, Ph.D., Department of Chemical Engineering, University of Texas, Austin, TX

Lawrence Chew, Ph.D., Pfenex Inc., San Diego, CA

Yat-Chen Chou, National Renewable Energy Laboratory, Golden, CO

Nathan Crook, Department of Chemical Engineering, University of Texas, Austin, TX

Michael Dauner, Ph.D., E. I. du Pont de Nemours and Company, Wilmington, DE

Bryon S. Donohoe, Ph.D., National Renewable Energy Laboratory, Golden, CO

Jing Du, Ph.D., Department of Chemical and Biomolecular Engineering, University of Illinois at Urbana-Champaign, Urbana, IL

Ryan T. Gill, Ph.D., Department of Chemical and Biological Engineering, University of Colorado-Boulder, Boulder, CO

Michael T. Guarnieri, Ph.D., National Renewable Energy Laboratory, Golden, CO

Marcus Hans, Ph.D., DSM Biotechnology Center, Delft, The Netherlands

Byoungjin Kim, Ph.D., Energy Biosciences Institute, University of Illinois at Urbana-Champaign, Urbana, IL

Eric P. Knoshaug, National Renewable Energy Laboratory, Golden, CO

Bert Koekman, Ph.D., DSM Biotechnology Center, Delft, The Netherlands

Lieve M.L. Laurens, Ph.D., National Renewable Energy Laboratory, Golden, CO

Matthew L. Lipscomb, Ph.D., OPX Biotechnologies, Inc., Boulder, CO

Tanya Warnecke Lipscomb, Ph.D., OPX Biotechnologies, Inc., Boulder, CO

Michael D. Lynch, Ph.D., OPX Biotechnologies, Inc., Boulder, CO

Vincent J.J. Martin, Ph.D., Department of Biology, Centre for Structural and Functional Genomics, Concordia University, Montréal, Québec, Canada

Robin Osterhout, Ph.D., Genomatica, Inc., San Diego, CA

Ranjan Patnaik, Ph.D., Head Biofuels R&D, DuPont India Private Ltd., Hyderabad, India

Priti Pharkya, Ph.D., Genomatica, Inc., San Diego, CA

Philip T. Pienkos, Ph.D., National Renewable Energy Laboratory, Golden, CO

John Pierce, Ph.D., Chief Bioscientist, BP, London, UK

Dominic Pinel, Department of Biology, Centre for Structural and Functional Genomics, Concordia University, Montréal, Québec, Canada

Jun Sun, Ph.D., E. I. du Pont de Nemours and Company, Newark, DE

Stephen Van Dien, Ph.D., Genomatica, Inc., San Diego, CA

Huimin Zhao, Ph.D., Department of Chemical and Biomolecular Engineering, University of Illinois at Urbana-Champaign, Urbana, IL

1

CLASSICAL STRAIN IMPROVEMENT

Nathan Crook and Hal S. Alper

1.0 INTRODUCTION

Improving complex phenotypes, which are typically multigenic in nature, has been a long-standing goal of the food and biotechnology industry well before the advent of recombinant DNA technology and the genomics revolution. For thousands of years, humans have (whether intentionally or not) placed selective pressure on plants, animals, and microorganisms, resulting in improvements to desired phenotypes. Clear evidence of these efforts can be seen from the dramatic morphological changes to food crops since domestication (1). These improvements have been predominantly achieved through a “classical” approach to strain engineering, whereby phenotypic improvements are made by screening and mutagenesis of strains that use methods naive of genome sequences or the resulting genetic changes. This approach is well suited for strain optimization in industrial microbiology, which commonly exploits complex phenotypes in organisms with poorly defined or monitored genetics. As a recognition of importance, Arnold Demain and Julian Davies begin their Handbook of Industrial Microbiology and Biotechnology with “Almost all industrial microbiology processes require the initial isolation of cultures from nature, followed by small-scale cultivations and optimization, before large-scale production can become a reality” (2). The classical approach is concerned with the central steps in this process—between isolation and large-scale production. Hence, the methods and techniques utilized in this approach amount to “unit operations,” that is, standard procedures that can be generically applied to any desired strain of interest.

A variety of approaches are used to force genetic (and hence phenotypic) diversity including naturally occurring genetic variation and genetic drift, mutagenesis, mating/sporulation, and/or selective pressures. These methods have garnered large successes across a wide range of host organisms owing mostly to the absence of required sophisticated genomic information or genetic tools (3). Thus, the classical approach can be applied to both model organisms (such as Escherichia coli and Saccharomyces cerevisiae) and newly isolated or adapted industrial strains. As a result, the classical approach has seen wide adoption in industrial fermentations due to its proven track record in alcohol and pharmaceutical production. Finally, strains developed in this manner are currently accorded non-genetically modified organism (GMO) status, removing significant barriers to their acceptance by both regulatory agencies and consumers. This chapter will highlight several of the approaches and successes that exemplify the classical approach for improving complex phenotypes of industrial cells as well as indicate its limitations and potential interfaces with emerging technology.

1.1 THE APPROACH DEFINED

The classical approach is characterized by the introduction of random mutations (either forced or natural) to a population of cells followed by screening and/or selection to isolate improved variants. The defining quality of classical strain engineering (as opposed to other evolutionary engineering methods) is genome-wide mutagenesis. This approach utilizes techniques that introduce variation across all regions of the genome, in contrast to other techniques that specifically target the mutations to single genes (or subsequences thereof). To date, this approach has been successful in improving complex phenotypes because of the global nature of classical methodologies (see Box 1.1 in this chapter and case study in Chapter 6). Complex phenotypes such as tolerance to environmental stress, altered morphology, and improved flocculation characteristics are often influenced by the interactions between multiple (often uncharacterized) genes. In contrast, without significant prior understanding, variants generated through mutagenesis of specific genomic subsections are unlikely to gain proper coverage of the genotype. Indeed, as will be discussed later, this approach has continuously yielded improved variants for a wide variety of complex biotechnological applications. The theory and techniques for the two major steps of classical strain improvement (CSI) (mutagenesis and screening) are the focus of this chapter, including practical recommendations for their implementation as well as brief discussion of examples of each method’s industrial application.

BOX 1.1: APPLICATION OF CSI IN SAKE FERMENTATION

The Japanese-brewed sake is produced from rice mash using Aspergillus oryzae to saccharify the rice and strains of sake yeast (genus Saccharomyces cerevisiae) to ferment the sugars to ethanol. The ideal process imposes a number of complex traits on the sake yeast, including high fermentation capacity over the 20- to 25-day process at low temperatures (typically 10°C), high ethanol tolerance (ethanol levels can approach 15–20%), minimal foaming, resistance to contaminating microbes, and the ability to create the correct proportion of flavor components including higher alcohols and esters (82). Many of these traits have been approached using methods of the classical approach including mutagenesis, selection, and cell mating. Specifically, UV and chemical mutagenesis have dominated as a means of retaining GRAS status for this yeast. Moreover, difficulty in sporulation has limited genetic dissection and a more rational approach until recently (83). Natural selection and isolation from hundreds of years of fermentation has resulted in the series of commonly used strains named the Kyokai series, with Kyokai no. 7 and Kyokai no. 9 as the main fermentation strains used industrially. Due to the superior brewing capacity of Kyokai no. 7, many attempts have been made to improve this strain through the classical approach as well as dissect the underlying genetic changes. Recently, it has been demonstrated that the breeding and selec­tion process of this strain resulted in heterozygosity of many alleles re­­sponsible for ethanol production and aromatic compound synthesis (84,85) as seen by sporulation analysis. Many attempts have been made to im­­prove the characteristics of Kyokai no. 7. Non-foaming mutants have been isolated from spontaneous clones as well as UV-induced mutants using selection methods such as cell agglutination and froth floatation (86). Improved strains have also been isolated through chemical mutagenesis (e.g., by EMS) to select for improved flavor profiles. In this case, mutant Kyokai no. 7 strains more resistant to cerulenin were thought to produce more ethyl caproate, an important flavor component. This approach was successful in improving this flavor component; however, the complete portfolio of complex phenotypes was not fully assayed (47). Finally, prevention of contaminants has been explored through mating sake yeast strains with strains exhibiting the killer phenotype (56), which would ward off contaminating yeasts. Collectively, these examples of complex pheno­type engineering highlight the difficulties of the process, specifically; it is often hard to create all traits at once. The evolution of the sake yeast demonstrates the power of the classical approach. More recent attempts have been made to use the rational or evolutionary approach for this strain; however, Kyokai no. 7 remains the industrial favorite for sake production.

1.2 MUTAGENESIS

A fundamental parameter dictating success in classical strain engineering is the frequency and type of mutation applied to the parent cells. Typically, this rate is determined by the dose and type of mutagen delivered. To test mutagen specificity and rate, it is common to generate an inactive (mutant) form of some easily assayable gene (e.g., LacZ in E. coli) that differs from the wild-type gene by a single base-pair change, and test the frequency of reversion. For example, Cupples et al. generated six variants of LacZ to show that many common mutagens (EMS, NTG, 2-aminopurine, and 5-azacytidine) are in fact quite specific for certain base-pair changes in E. coli (4). Hampsey undertook a similar approach in S. cerevisiae and found similarly that mutagens were highly specific. However, the mutation frequencies and specificities were significantly different from those observed in E. coli (5). Frameshift and deletion frequencies can also be detected through analysis of a cleverly mutated marker (6). Through analyses of reversion frequencies or high-throughput sequencing, a detailed picture of a treatment’s mutagenic profile may be ascertained. This detailed information can be then be used to compute several useful quantities, such as the average number of mutations per genome or the expected number of distinct variants among a mutated population. Knowledge of these frequencies and landscapes are especially useful when designing a selection program, as detection of rare variants (e.g., individuals possessing certain particular mutations and no more) will require many individuals to be screened, whereas more probable patterns of mutagenesis (e.g., if additional silent or neutral mutations are tolerable) will not. At the same time, more focused patterns of mutation inherently limit the search space.

1.2.1 Numerical Considerations in Screen Design

Although in general every possible base substitution will occur at a different frequency (and vary nonuniformly throughout the genome), it is instructive to neglect deletions or insertions and assume all base changes at each site are equiprobable (i.e., occur at the same frequency) to make use of the binomial distribution, to obtain approximate probabilities of any desired mutagenic outcome. If the probability of a single base being mutated to any other base is p, then the probability that a genome of size g has n mutations after mutagenesis is:

By using well-known properties of the binomial distribution, the average number of mutations per genome is gp with variance gp(1 − p). Random genetic drift results in mutation rates of 10−10 to 10−5, while forced mutagenesis can elicit rates upwards of 10−3 as described below, so this will restrict the range of p. It is apparent that if p is too low (that is, less than 1/g), there will be many variants with few or no mutations and a vanishingly small population of highly mutated individuals. Furthermore, the binomial coefficient indicates that libraries with low mutation rate (and thus a high population of slightly mutated individuals) are very likely to be redundant, that is, have many individuals of the same genotype. Thus, it is of interest to know the expected number of distinct variants in a mutant library to guide screen design. Patrick et al. developed a suite of algorithms to compute many quantities of interest for screening a mutant pool derived from a mutagenic procedure of arbitrary specificity, including the expected number of distinct mutants following mutagenesis (7,8). If the library is highly redundant, then screening of the entire mutated population may not be necessary to ensure complete coverage. As diversity increases, however, the required screening fraction will approach unity. Since complex phenotypes are controlled by the action of multiple genes, high mutation rates are often employed, generally resulting in high library diversity and a strong incentive to screen the entire mutated pool.

To choose the correct rate of mutagenesis and screening, it is important to know the rarity of the phenotype of interest. In the worst and most restrictive case, an improved phenotype will be acquired by mutants containing only a certain set of mutations. For example, consider a particular phenotype that only manifests itself when n-specific mutations are present and no more. In this case, one must determine the mutation rate which maximizes the fraction of n-mutant variants in the mutated population (using one of the tools mentioned earlier) and screen until a reasonably high probability of complete coverage is achieved. For a genome of g base pairs, we can take the derivative of the binomial distribution with respect to mutation rate and set it equal to zero:

Eliminating constants and taking the derivative, we have:

The obvious interesting candidate for a solution is:

Taking the second derivative of the binomial distribution yields:

Because

we can substitute our candidate solution into the remaining portion of the second derivative to determine its sign:

which is clearly negative for g > n. Hence, the likelihood of attaining n mutations in a genome of size g is maximized when the mutation rate is n/g. This maximum likelihood is:

It is generally necessary to screen more than the number of possible mutants to ensure coverage of the diversity. To obtain, on average, F fractional coverage of all n-mutant variants, it will be necessary to solve

for L, where a is the probability of selecting the correct n-mutant variant (1/V in this case, where V is the number of possible n-mutant variants [given by the binomial coefficient]) and L is the library size (7). For a small-sized genome (106 base pairs) and a phenotype requiring two specific mutations (hence at an optimal mutation rate of 2*10−6), L works out to be 5.5*1012 to obtain 95% coverage, on average, which is outside the scope of most screening programs (Assuming a standard yeast cell density of 107 per mL and an average cell sorting rate of 103 per second, screening this many individuals would require 550 L of culture (for growth-based selections) or 241 years of cell sorting [for fluorescence-based screens]!)

Luckily, most complex phenotypes can tolerate the existence of additional silent/neutral mutations. To account for a small number of allowable neutral mutations, let us assume that the desired n mutations may be found in any variant containing up to m > n total mutations, but no more. The analysis for this case proceeds in much the same way as before with one minor alteration resulting from the fact that a variant with m > n mutations contains

instances of n mutations. Therefore, maximizing the quantity

with respect to p will yield the mutation rate which maximizes the number of n-mutant combinations encountered in the randomized pool. This rate may be used to find the probability that a variant selected at random will have between n and m mutations:

Given that a variant has between n and m mutations, the probability that it contains the mutations of interest can be obtained by summing the probabilities of finding the mutations of interest at each particular mutational level:

where

is the probability that a variant with a mutation rate i is selected, and

is the probability of finding a particular combination of n mutations within that variant. r and a can then be substituted to the equation for F, which is solved for L as before. Continuing with the example stated above, if the search is expanded to allow desired mutations to occur in a background of up to 5 mutations, then the mutation rate can be increased to 4.2*10−6, requiring screening of 4.3*1011 individuals, which, though an order of magnitude less than in the previous case, is still rather unmanageable.

The property that allows strain engineering programs to be feasible is the additivity of the effects of mutations; that is, even if a particular combination of 12 point mutations is optimal, a couple of them, even in isolation, will be beneficial. This allows engineering to proceed in several single mutation steps as opposed to a single multiple-mutation bound. Because any given single mutation is much more probable than a particular double mutation, the probability of isolating improved variants is greatly increased. Even if a phenotype could only be improved by a single base-pair change in the absence of any others, the number of mutants that must be screened is 8.2*106, which is attainable from a fraction of a milliliter of culture under growth selection or from less than an hour of cell sorting (9).

These probabilities guide strain selection. In the following sections, multiple mutagenesis techniques and screening strategies will be discussed. Particular attention should be paid to mutation rate and throughput, respectively, so that screening programs are designed and carried out efficiently.

1.2.2 Random Genetic Drift

Natural mutations due to errors in replication take place at frequencies between 10−5 and 10−10 , depending on the strain and organism. Given this frequency, it is not surprising that large-scale selections are required to isolate any improved mutant. These frequencies are supported by a meta-analysis of phenotype occurrence frequencies for the basic yeast S. cerevisiae (10). Phenotype reversions requiring single base-pair changes including amino acid auxotrophy reversion and resistances occurred, on average, at a frequency of 10−8. On the basis of this low mutation rate alone, it would appear that random genetic drift may be most suited for the optimization of phenotypes under the control of nonepistatic factors, despite the prevalence of epistatic interactions in nature (11). However, single point mutations are not the only genetic change to take place in evolving cells. Specifically, Lenski et al. found that the majority of the genotypic changes observed through the course of a long-term natural evolution experiment in E. coli resulted from transpositions and rearrangements as opposed to single base-pair substitutions (12). These large-scale genetic changes have a much higher probability of generating mutants on distant peaks than do point mutations. The variety of possible genetic changes resulting from a natural evolution program points to its versatility in optimizing a wide variety of complex phenotypes. However, generation of mutants with this method requires a significant amount of time, during which individuals are subjected to growth-based selective pressures. If the phenotype of interest is at odds with growth, then this mutagenic procedure may not be optimal with respect to library size and screening. Natural mutagenesis, however, does lend itself very well to growth phenotypes, as no additional effort on the part of the strain engineer is required to generate mutants and compare them against the fittest variant.

1.2.2.1 Tracking Evolution through Neutral Phenotypes 

Since natural evolution experiments are often accomplished in continuous liquid culture, new variants are constantly being generated and compared against the fittest variant. Therefore, evolution does not proceed in rounds or stages like most forced evolution experiments, and so it is unclear when fitness increase has ceased or when a population has stopped evolving. To overcome this limitation, one can make use of neutral markers to detect a mutation event. Neutral markers are genotypes that confer no alteration in growth rate yet whose phenotype is easily detectable. Common neutral phenotypes include resistance to phage T5 or reversion to lactose fermentation (when the appropriate selective pressure is not present in the screening medium, of course). In a strain normally deficient in either of these phenotypes, it is expected that over the course of evolution these neutral mutations will become fixed in the evolving population at a slow but steady rate. Due to the low probability of attaining the neutral marker and the fact that it does not pose a selective pressure on those who carry it, it is assumed that neutral markers are never predominant in the culture and that over time the prevalence of this phenotype will attain a steady state as the rate of incidence becomes balanced by mutational losses. When a variant of high competitive fitness is generated (which in all likelihood does not possess the neutral marker), it steadily outcompetes the existing population, driving the proportion of the neutral marker down. After this new variant becomes predominant in the culture, the fraction of mutants containing the neutral marker again increases at a slow but steady rate as before. As a result, these sharp declines in the presence of a neutral marker signal the appearance of a new adaptive mutation, and the jagged graph of the neutral marker over time is called a periodic selection curve (see Figure 1.1). However, if the adaptive mutation happens to occur on a variant containing the neutral marker, the fraction of the neutral marker will approach unity in the selective medium, although it confers no selective advantage. This phenomenon is known as “hitchhiking,” and it is predicted to occur often in evolving systems (13,14). In addition, the appearance of an additional adaptive mutation does not necessarily imply any significant improvement in phenotype as desired by the strain engineer, as fixation of mutations may simply result in a competitive advantage quite unrelated to the phenotype of interest. Therefore, it is necessary to assay for improvement as selection proceeds, to ensure that progress is still occurring.

FIGURE 1.1. Periodic selection in an asexual population. The numbers of successive adaptive clones and the number of clones possessing the neutral marker are tracked over time. (A) The prevalence of the neutral marker increases in the wild-type population at a rate determined by natural mutation frequency. (B) The first adaptive mutant (with a large selective advantage) appears in the neutral marker-null population and quickly outcompetes the original clone, causing a sharp decline in the number of cells containing the neutral marker. (C) The prevalence of the neutral marker increases in the first adaptive mutant at a rate determined by natural mutation frequency. (D) The second adaptive mutant (with a relatively small selective advantage) appears in the neutral marker-null population and gradually outcompetes the original clone, causing a slower decline in the number of cells containing the neutral marker than in time interval (B).

Reproduced with permission of Annual Reviews, Inc., from Reference (14); permission conveyed through Copyright Clearance Center, Inc.

1.2.2.2 Genetic Determinants of Mutation Rate 

Although a wide range of chemical and physical agents can serve to mutate a microbial population (discussed later), a variety of more natural factors can contribute to an increase in the mutation rate of naturally evolving populations, including ploidy, genotype, and environmental conditions. Diploid strains have been shown to more quickly fix adaptive mutations than haploids. Since diploid strains make copies of genes at twice the rate of haploid strains, adaptive mutations are generated at twice the rate. Assuming adaptive mutations are dominant, this increased generation of adaptive mutations should result in an increase in the rate of fixation of adaptive mutations. Indeed, Paquin and Adams showed that diploid strains accumulated advantageous mutations at 1.6 times the rate of haploid strains (15). This implies that diploid strains may achieve maximum phenotypic increase much faster than haploid strains, allowing further selection programs to be undertaken.

In addition to diploid strains, a number of additional “mutator” genotypes are known to increase the mutation rate in bacteria (16,17). These genotypes may encode for enzymes that are naturally mutagenic or may confer mutagenic activity upon an existing enzyme. Although the presence of a gene conferring a high mutation rate would appear detrimental, it has been theorized that a gene conferring a 1000-fold increase in mutation rate in a particular individual can cause a population to increase in fitness quite quickly while remaining in but a small fraction of individuals (18). The ability of this genotype to confer a selective advantage without becoming ubiquitous can be understood through the high rate of reversion of the mutator genotype, due to its high mutation rate. This implies that mutator strains may be an excellent starting point for a variety of evolution experiments, as isolates from the resulting culture are likely to be genetically stable. However, it should be noted that most known mutator genes only achieve a 100-fold increase in mutation frequency, at which level they have been theorized to attain a much larger fraction of the population (5–10%) (16,17). In addition, the mutator phenotype may be amplified by certain chemicals such as thymidine, allowing for increased control over evolutionary rate over the course of the experiment. Mutator genes are also unique in that they may have a very specific mutational spectrum. In E. coli, mutY increases the frequency of GC->TA transversions, mutT results in TA->GC transversions, and the mutD5 mutation appears to increase the rate of mutation in a nonspecific manner. The specificity of mutator genotypes allows a great deal of control over the spectrum of mutants generated, possibly enabling preservation of a counter-selected genotype that is necessary for the application of interest.

1.2.2.3 Applications of Random Genetic Drift 

Continuous culturing and serial transfers have been successfully used to select for fast-growing strains generated through a natural evolution program (19–22). Included in this list are improvements of basal-level growth rate as well as improvements in growth rate on alternative sugars such as xylose. Accumulated mutations in a yeast strain selected on xylose over time resulted in greatly altered xylose transport kinetics, doubling Vmax (15.8 to 32 mmol/[g dry weight]/h) and reducing Km by 25% (132 to 99 mM) (22). A second highlighted example involves a study on the bacterium E. coli, where 10,000 generations were studied via serial culturing (12). The resulting strains exhibited a 50% improvement in fitness as well changes in other complex phenotype such as cell size and morphology. Moreover, this study highlighted that the mutations regulating these phenotypes were indeed quite rare and diverse. In some cases, this change could be accomplished by point mutations; however, genomic rearrangements were also seen. Furthermore, most of the change occurred during the first 2000 generations, with improvements slowed over the last 8000. This highlights the importance of screening high levels of mutants, a prime difficulty with natural selection-based mutations. Both of these examples are highly relevant because growth improvement is a highly complex process. Not only does metabolism need to be regulated and carried out more efficiently, but a number of additional factors such as substrate uptake, metabolite tolerance, and reproductive machinery also need to be optimized in a fast-growing strain. This breadth obtainable by classical strain engineering would be unfeasible in more directed approaches. As a final example, Wiebe et al. used a glucose-limited chemostat to select for mutants of Fusarium graminearium with delayed onset of colonial morphology, further illustrating the power of natural evolution to enrich for highly complex phenotypes (23). Delayed onset of a particular phenotype requires alteration of a wide range of regulatory factors, especially for a trait that is carried out by a plethora of cellular machinery. Furthermore, because many factors controlling morphology are unknown, directed approaches would be ineffective at generating highly improved variants.

Industrially, natural genetic drift is always under way in large-scale fer­mentations. As an example, naturally improved strains of yeasts (both Saccharomyces and Pichia sp.) have been isolated from a sulfite liquor fermentation plant. These strains demonstrated the complex phenotype improvement of increased tolerance to acetic acid and enhanced galactose fermentation capacity (24). Samples from ongoing fermentations, especially long-term culturing, will present a diverse genetic population. The continuous sampling and analysis for these cultures can give rise to novel, complex phenotypes. However, other methods such as forced mutagenesis can improve the frequency of improvements as well as the prospect for success.

1.2.3 Forced Mutagenesis

Mutagenesis by ultraviolet (UV) or chemical treatment is a widely used approach for obtaining point mutations to create auxotrophic markers and improve strains. Cellular exposure to UV radiation can disrupt DNA structure, leading to a dose-specific occurrence of mutations. In addition, certain chemicals such as ethyl methane sulphonate (EMS) and nitrosomethykguanidine (NTG) have been known to cause DNA damage. It should be cautioned that all of the agents described in this section are mutagenic and hence carci­nogenic and thus extreme care must be used during handling to prevent damaging exposure. In addition, certain chemicals carry orthogonal risks. For instance, NTG is explosive (25). Therefore, the benefits of increased mutation rate must be weighed against increased safety costs when working with these compounds.

The attractiveness of chemical and physical mutagens is the increased mutational capabilities compared with natural variation. Mutation frequencies are often measured as a function of auxotrophic marker development or gene mutation reversion. While basal-level, natural drift mutations can result in average reversion frequencies of nearly 1 in 10−8 (as described above), induced mutations by methods such as UV radiation can increase this value upwards of 10−3 to 10−5, depending on the organism used and the intensity/duration of irradiation (26–28). Similar results and mutation frequencies can be seen with chemical mutagenesis using reagents such as EMS and NTG. Given these rates, it is still expected that the frequency of obtaining auxotrophic mutants in diploids by direct mutagenesis should be rare without prior selection. However, it has been demonstrated that auxotrophic mutants due to forced mutagenesis of diploid industrial strains can occur at frequencies of 10−4, illustrating that the mechanism of mutations is still unknown (29). Therefore, improved mutants will occur at rates higher than those suggested by the probabilities. In terms of fitness landscapes, this higher mutation rate allows a further exploration of genotypic space. Thus, generating mutations with this method may yield variants located on more distant peaks, possibly at higher levels of fitness. When the phenotype of interest is influenced by a significant number of epistatic interactions, the resulting landscape will be more rugged, making forced mutagenesis more desirable for isolating improved variants than natural evolution.

1.2.3.1 Optimal Mutagen Dose 

Mutation rate and cell survival are both strongly affected by mutagen dose. Thus, it is necessary to determine optimum mutagen dose. If the mutation rate is too low, variants with mutations (and especially improved phenotypes) will be rare compared with unmutated cells, making detection difficult even in high-throughput screens. In contrast, if the mutation rate is too high, the effects of deleterious mutations will swamp those of beneficial mutations, yielding poorly performing (or even nonviable) variants. Since the goal of the strain engineer is to maximize the number of beneficial mutations per variant, plots akin to Figure 1.2 are often constructed to evaluate the effects of different types or doses of mutagen on the trait of interest. A crude measure of phenotype on a small number of variants is preferred to minimize the resources spent at this preparatory stage. Curves similar to those in Figure 1.2 are often seen, depending on the phenotype of interest. Reversions to prototrophy or resistance mutations are normally monotonic, whereas titer-increasing or decreasing mutations normally follow an inverted U curve (30–32). It is important to note that as titer is improved, the likelihood of finding further beneficial mutations is reduced, making the statistic of population variance as important as the average for selection of optimal mutagen dose (33). Indeed, Lenski, et al. found a hyperbolic decline in fitness increase over time in a population of E. coli undergoing natural selection, suggesting that a small number of mutations of large effect were fixed in the population during early times whereas a larger number of mutations of smaller effect were fixed in subsequent generations, assuming a constant mutation rate per generation (34). Hence, the optimal mutagen dose is likely to change as superior variants are isolated. The dose that results in the highest enrichment in desirable variants is then applied and a large number of variants are screened more accurately.

FIGURE 1.2. Common dose–response curves. Determining the optimal mutagen dose is critical for success in identifying altered mutants. This optimal level is depen­dent on the type/doses of mutagen and on the trait of interest. In general, three types of response curves are seen: reversion to prototrophy or resistance mutations normally follow the monotonic curve, whereas titer-increasing or decreasing muta­tions normally have an inverted-U shape. If the culture contains a subpopulation of radiation-sensitive individuals, biphasic behavior may be observed.

Although plots akin to Figure 1.1 are useful for single-round selection programs, selecting the optimal mutagen dose based on the maximum observed phenotypic increase may cause problems for prolonged selection experiments. Although this dose will maximize the single-round phenotypic increase, subsequent rounds of improvement will have to contend with any deleterious mutations that will have occurred, possibly limiting maximal improvement in phenotype. For cases where improved titer is important, it is generally accepted that low mutation rates are better than high, even though high rates will yield a more immediate benefit (35). The rationale behind this choice is that the small number of mutations selected in a low mutation rate program will have a much higher probability of being beneficial than the high number obtained in a more error-prone program. Therefore, any subsequent beneficial mutations will not have their effects attenuated by the presence of deleterious mutations. Only after low levels of mutagen fail to yield improved variants is it desirable to increase mutagen concentration, thus expanding the evolutionary search to reveal epistatic peaks in the fitness landscape. By alternating low and high mutation rates in this manner, the prevalence of deleterious hitchhikers may be minimized.

1.2.3.2 Determination of Mutagen Specificity and Frequency 

A wide variety of chemical mutagens have been used to introduce DNA damage. Not only do individual chemicals produce different mutation profiles as mentioned above, but the environmental context and strain in which these mutagens are applied can also have a large effect on the changes observed (36,37). Chemical mutagens have been found to delete large (∼1 kbp) sections of an organism’s genome as well as generate mutations at the single base-pair level (38). Furthermore, the advent of high-throughput sequencing technology allowed the identification of mutagen-specific “hotspots” in E. coli, emphasizing the nonrandom nature of the induced changes (39). In addition, it has been noted that NTG acts upon the DNA replication forks, causing the resulting mutations to be tightly clustered (40). Therefore, it is highly recommended to change mutagens as a strain improvement program proceeds, not only to avoid development of resistance, but also to allow fuller sampling of genomic sequence space. Alternatively, it is possible to apply multiple mutagens in the same dose; however, the mutagens must act on different DNA repair pathways in order for this approach to be beneficial (41). In general, unless mutagenesis rates and specificities have been previously characterized for the strain of interest, characteristics of mutagens may be known only approximately, especially if the organism’s cellular repair pathways are unusual.

1.2.3.3 Mechanisms of Mutagenesis 

Most of the mutagens introduced in this section serve to make DNA repair machinery more error prone, in addition to damaging DNA directly. Therefore, it is plausible that over the course of a selection program a mutation that confers resistance to a particular mutagen will arise. In this situation, no change in phenotype will be observed upon mutagenesis. To distinguish this case from cases where further phenotypic improvement is unlikely, some easily observable reversion phenotype may be used to confirm efficacy of the mutagenic treatment. In cases where resistance has developed it will be necessary to attempt different types of mutagens in order to introduce mutations via an alternate repair pathway (35).

UV light has been extensively studied in terms of its mutagenic frequency, specificity, and mechanism, in large part due to its ease of implementation (27,42–44). Cells may be mutated simply by exposing them to UV light for some length of time (analogous to the dose of a chemical mutagen). In the case of E. coli, it is thought that UV radiation causes DNA damage but that these initial lesions are not converted into base-pair changes until activation of the SOS repair pathway, a global response to DNA damage. For this reason, mutation frequency initially increases according to the square of UV dose, confirming that two distinct DNA lesions are required for mutagenesis to occur: one to induce the SOS repair pathway, and the second to cause a detectable phenotypic change. After this initial stage, mutations increase linearly with exposure as lesions continue to accumulate after SOS induction. A regime of higher order response to radiation indicates the appearance of mutations dependent on the presence of two DNA lesions in proximity (43). Finally, the mutation rate reaches a steady value as lethal mutations accumulate. Treatment with UV radiation is analogous to chemical mutagenesis in that UV has its own mutagenic specificity and frequency. However, studies have shown UV to be slightly broader in its action than other mutagens (4).

1.2.3.4 Effects of Environment 

Since each of the mutagens previously described require the action of cellular repair machinery, “recovery” of mutated cells in rich medium has been found to increase the mutation rate realized in the surviving cells. Not only does this treatment allow sufficient time for SOS repair to convert DNA lesions into base-pair changes, it also allows mutated proteins to be synthesized, which is important if screening occurs directly after mutagenesis. Also, certain additives to this recovery medium may promote or inhibit recovery of mutated cells. For example, addition of caffeine and acriflavine following UV mutagenesis will increase the mutation rate (35), whereas addition of manganese II, purine nucleosides, and inhibitors of protein synthesis will decrease the mutation rate (45). Interestingly, 8-methoxypsoralen is antimutagenic when present before UV irradiation, but is mutagenic if introduced after UV, illustrating the complexity of the repair mechanisms involved (46). It should be noted, however, that any treatment that increases the mutation rate will also increase lethality; therefore, it should be ascertained whether such treatments actually increase the number of mutants per survivor before mutation-amplifying additives are introduced (35).

1.2.3.5 Applications of Forced Mutagenesis 

Forced mutagens have seen wide use in development of complex phenotypes. For example, UV mutagenesis was used to generate auxotrophic mutants of sake yeast (29). This is significant because it allows this yeast to be used in breeding programs and in metabolic engineering efforts utilizing molecular cloning techniques. The ability of mutagens to introduce variation in a wide variety of organisms is a major benefit to this approach, as standard genetic manipulation techniques are only established for a handful of (possibly industrially suboptimal) strains. Furthermore, EMS was used to generate sake yeast mutants with improved flavor profiles, clearly indicating the ability of forced mutagenesis to improve industrially relevant complex phenotypes whose molecular basis may be largely unknown (47,48). In addition, a forced mutagenesis/selection scheme was used to screen for improved microalgae capable of producing L-ascorbic acid. By screening over 105 mutants, a greater than 50-fold improvement in specific productivity was achieved (49). These results illustrate the large size of libraries necessary to achieve metabolic phenotypes. Many processes (including the penicillin production process (50)) also rely on this method to continuously enhance strains. As an example of this process improvement, penicillin titers are over 40,000-fold higher in improved strains than the original isolated wild-type strain (51). Improved antibiotic production is a prime example of a complex phenotype, as strains must evolve not only the enzymes responsible for antibiotic synthesis but also any factors involved in nutrient transport and chemical tolerance. The organisms normally responsible for high production of a compound of interest are often genetically uncharacterized; thus, the genome-blind nature of the classical approach becomes an asset. This approach of mutagenesis and screening has even been used to improve the activity of baker’s yeast for bread making purposes (52). As a result of the ease of operation and selection, this method continues to be used to generate complex phenotypes in industrial cells.

1.2.4 Strain Mating

Strain mating represents an effective tool for generating a population with a high number of non-detrimental mutations. One of the main limitations of random mutagenesis is the high probability that the changes induced in a daughter cell will be detrimental, and this probability increases as the mutation rate increases. However, high mutation rates are required to escape local optima on the fitness landscape. Strain mating allows recombination to occur between two divergent (yet functional) genotypes, generating a library of highly mutated individuals. However, because meiotic recombination operates at the level of the gene, as opposed to the base pair, any mutations introduced are likely to be in the genomic context in which they were found in one of the parents, decreasing the likelihood that such mutations are lethal. In essence, this technique allows exploration of distant regions of the fitness landscape without the associated high probability of failure, thus allowing generation of mutants located specifically on regions of high fitness. It is obvious that strain mating will have its maximum effect when applied to two highly divergent members of the same species, allowing a high rate of mutation with a low probability of lethality. This technique allows the possibility of running multiple mutagenesis programs in parallel (utilizing perhaps different mutagenic techniques and screening strategies) and mating the most successful individuals from each program, especially if mutagenesis has been conducted to minimize the occurrence of deleterious mutations. In fact, it has been theorized that the accumulation of deleterious mutations causes evolving populations to gradually reach a maximum fitness. If this is the case, then strain mating should greatly improve the potential for phenotypic increase (53). It should be cautioned, however, that if significant epistatic interactions exist between genes, there will be a high likelihood of disrupting them upon mating, possibly leading to inferior individuals (54).

Protoplast fusion is a distinct method of strain mating that does not involve recombination. Instead, the cell walls of two individuals are digested away and their genetic material is combined to form a new individual with both sets of chromosomes (55). This technique allows the characteristics of both organisms to be combined (forming a heterokaryon) without the risk of recombination loss. Heterokaryons are often verified by nutrient complementation. Therefore, it is imperative that each parent be auxotrophic for a different compound. Removing the cell wall while preserving the cell membrane is a delicate process. Therefore, reliable isolation of heterokaryons is dependent on a number of factors, including protoplast isolation from exponentially growing cells, maintenance of isotonicity in the protoplast media, and the addition of polyethylene glycol as a fusogenic agent. Since both genomes are isolated from viable individuals, the probability of deleterious interactions is small. Further, since enzymatic deficiencies are recessive, any lack of functionality in one parent will be complemented by the genome of the other. An added benefit of generating a polyploid strain is the doubling of the effective mutation rate for each gene, allowing evolution of improved phenotypes to proceed at a faster pace, as mentioned earlier.

Strain mating can be used to combine two distinct functionalities into one organism. For example, a common problem in fermentations is the evolution of a “killer” phenotype, whereby a nonproductive individual gains the ability to secrete a toxic compound, thus outcompeting the organisms of desirable phenotype and resulting in a failure of the fermentation. However, strain mating can be used to generate a productive “killer” phenotype, whereby the resulting population is able to both secrete the product of interest and kill any contaminants that may be introduced (56). Indeed, Bortol et al. were able to fuse strains of S. cerevisiae possessing the “killer” phenotype with traditional baker’s yeast, producing competitive variants that retained the ability to make dough rise (57). Clearly, strain mating has enormous potential for generating mutants improved in a variety of complex phenotypes.

1.3 GENOTYPIC LANDSCAPES

If phenotype and genotype are graphed such that related genotypes are close together, the resulting landscape is ripe with series of peaks and valleys, with peaks representing genotypes of high fitness and valleys representing genotypes of low fitness. Natural selection dictates that individuals residing on higher peaks are more likely to reproduce, and upon reproduction, a new generation arises at some genotypic distance away from their parents, depending on the mutation rate. As evolution proceeds, the population will tend toward peaks and away from valleys if the selection pressure is toward improved phenotype. As a result, understanding aspects of this landscape helps develop proper mutagenesis and selection strategies. For example, what magnitude of fitness differential is observed when moving from a peak to a valley? Are there many different peaks, or just one? Do there exist mountain ridges connecting each peak, or are each separated by deep chasms? A good understanding of the topography of this fitness landscape will allow prediction of the evolutionary trajectories of a population under selection. For the strain engineer, this understanding will allow comparison of different selection programs in terms of their ability to generate a mutant residing on the tallest peak of this evolutionary landscape.

One important quality of fitness landscapes is their “ruggedness” as developed by Kauffman (58). This quality indicates the correlation in phenotype observed between related genotypes. In the limit of no ruggedness (i.e., perfect correlation), it can be shown that there is only one peak in the fitness landscape and that this peak is accessible from any genotype by progressing through successively more fit one-mutant neighbors. However, in the limit of maximal ruggedness (no correlation between related genotypes), the landscape is essentially random, with many local optima and a very small chance of encountering the global optimum by progressing through successively more fit one-mutant neighbors. This quality is therefore extremely important for evaluating a priori which mutagenic and selection treatments are likely to yield improved mutants. Treatments that result in small genotypic changes (such as single base-pair changes) may only be able to proceed in small steps through the fitness landscape. If selection is operated such that the fittest mutant is selected for subsequent mutagenesis and selection, then repeated rounds of generation and screening are only able to yield a local maxima. However, as mutagenic treatments become more severe, the possibility of generating a mutant on a more distant peak becomes higher at the expense of mutant generation on the current peak. Naturally, more severe mutagenic treatments become more desirable as the number of peaks in the evolutionary landscape increases. Alternatively, selection regimes that preserve a nonzero fraction of suboptimal mutants may also have an advantage in detecting more distant peaks. However, the costs associated with running many selec­tion programs in parallel may prove too great (59). It has been shown that phenotypes that exhibit a high degree of epistasicity resemble more rugged fitness landscapes, whereas phenotypes under the control of genotypes whose effects are perfectly additive resemble the gradual “Fujiyama” type of landscape (58).

1.4 SCREENING

The success of classical strain engineering is due in large part to the ability of researchers to search through a large number of variants to isolate a few improved individuals: a process called screening. As more mutants are screened, the probability of isolating an improved variant increases linearly (assuming the number of mutants generated is large compared with the number of mutants screened). Therefore, significant effort has been spent to develop improved techniques and technologies to allow larger numbers of variants to be assayed per unit time. Screens can be classified into one of two broad categories: rational screens and random screens (35). Rational screens are defined by their exploitation of knowledge about the system of interest, whereas random screens are of more general applicability. Ideally, the quantity being measured during the screen will correspond exactly to the phenotype of interest, but in cases where this is difficult to measure several orthogonal correlates of phenotype may be assayed to decrease the rate of false positives. It is important to keep in mind that the optimal screening strategy will depend in large part on the phenotype of interest and any prior knowledge of the system. In addition, resource limitations may restrict which screens may be performed as well as the number of mutants that may be assayed. The error rate of a screen is also of critical importance, as poorly designed growth screens may yield false positives and noisy assays will necessitate screening replicates to increase confidence. Finally, the importance of selecting individuals that exhibit true phenotypic improvements cannot be overemphasized. The powerful techniques of classical strain engineering often generate individuals that may perform well in a particular screen yet do not produce the phenotype of interest. Although detection of these “screening artifacts” can be largely eliminated through careful experimental design, further characterization of isolates at the conditions of interest is often required.

1.4.1 Rational Screens