# 4TH APPLIED SYNTHETIC BIOLOGY IN EUROPE

EDITED BY : Jean Marie François, Fayza Daboussi and Jussi Jantti PUBLISHED IN : Frontiers in Bioengineering and Biotechnology

#### Frontiers eBook Copyright Statement

The copyright in the text of individual articles in this eBook is the property of their respective authors or their respective institutions or funders. The copyright in graphics and images within each article may be subject to copyright of other parties. In both cases this is subject to a license granted to Frontiers. The compilation of articles constituting this eBook is the property of Frontiers.

Each article within this eBook, and the eBook itself, are published under the most recent version of the Creative Commons CC-BY licence. The version current at the date of publication of this eBook is CC-BY 4.0. If the CC-BY licence is updated, the licence granted by Frontiers is automatically updated to the new version.

When exercising any right under the CC-BY licence, Frontiers must be attributed as the original publisher of the article or eBook, as applicable.

Authors have the responsibility of ensuring that any graphics or other materials which are the property of others may be included in the CC-BY licence, but this should be checked before relying on the CC-BY licence to reproduce those materials. Any copyright notices relating to those materials must be complied with.

Copyright and source acknowledgement notices may not be removed and must be displayed in any copy, derivative work or partial copy which includes the elements in question.

All copyright, and all rights therein, are protected by national and international copyright laws. The above represents a summary only. For further information please read Frontiers' Conditions for Website Use and Copyright Statement, and the applicable CC-BY licence.

ISSN 1664-8714 ISBN 978-2-88963-812-3 DOI 10.3389/978-2-88963-812-3

#### About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

#### Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

#### Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

#### What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# 4TH APPLIED SYNTHETIC BIOLOGY IN EUROPE

Topic Editors:

Jean Marie François, Institut Biotechnologique de Toulouse (INSA), France Fayza Daboussi, Institut National de la Recherche Agronomique (INRA), France Jussi Jantti, VTT Technical Research Centre of Finland Ltd, Finland

Citation: François, J. M., Daboussi, F., Jantti, J., eds. (2020). 4th Applied Synthetic Biology in Europe. Lausanne: Frontiers Media SA. doi: 10.3389/978-2-88963-812-3

# Table of Contents


# Editorial: 4th Applied Synthetic Biology in Europe

Jean Marie François 1,2 \*, Jussi Jantti <sup>3</sup> and Fayza Daboussi 1,2

<sup>1</sup> Toulouse Biotechnology Institute, Université de Toulouse, CNRS, INRA, INSA, Toulouse, France, <sup>2</sup> Toulouse White Biotechnology Center, UMS-INSA-INRA-CNRS, Ramonville St Agnes, France, <sup>3</sup> VTT Technical Research Centre of Finland, Espoo, Finland

Keywords: synthetic biology, genome engineering, metabolic engineering, evolution engineering, cell-free systems, bionanoengineering

**Editorial on the Research Topic**

#### **4th Applied Synthetic Biology in Europe**

The congress of Applied Synthetic Biology in Europe was the 4th of this kind to be organized under the auspices of the European Federation of Biotechnology in Toulouse from 24 to 27 October 2018. It gathered about 130 participants from all over Europe as well as some participants from Asia, Brazil, and USA. Five sessions encompassing main disciplines in Synthetic Biology were on the program, and we acknowledged receiving several contributions by many of the participants to this meeting in each of these sessions. The sessions that received the most contributions were those dealing with topics in synthetic pathways for bio (chemical) manufacturing and synthetic biology for biotechnology applications. Relevant in these reports were concerns about pathways design and reconstruction, sometimes guided by computational analysis to propose non-natural metabolic routes for maximal yield and the choice of the more appropriate microorganism as potential industrial producer. In this regard, the production of trans-cinnamate from glucose was achieved by rational engineering of L-phenylalanine accumulation in Pseudomonas taiwensis since this bacteria is much more tolerant than E. coli to this non-natural product (Otto et al.). However, toxicity of this product remains a serious bottleneck in order to reach economically viable titers and yields of t-cinnamate from sugars, which might be overcome by application of adaptive laboratory evolution. As a well-established microbial platform for fatty acid accumulation (Blazeck et al., 2015), the oleaginous Yarrowia lipolytica was engineered for de novo biosynthesis of odd fatty acid from inexpensive carbon source (Park et al.). The critical point in this work was to enable an endogenous supply of propionyl-CoA as the key precursor for the synthesis of these fatty acids. These authors brought this capability to the yeast cell by the expression of a modular pathway with seven enzymatic steps that employed the threonine biosynthetic route and its deamination into α-ketobutyrate. The industrial potential of this strategy will await further improvement in the specificity and activity of the decarboxylating enzyme on α-ketobutyrate as well as enhancement of carbon fluxes to propionyl-coA through metabolic engineering tools. Two other papers in this session were dedicated to implement non-natural pathways to produce biobased products from C5 and C6 sugars at the highest yield. Bator et al. compared in silico and in vivo the synthetic capabilities of three metabolic routes, namely the xylose isomerase, the Dahms and the Weimberg pathways for xylose metabolization into a wide variety of valuable products in Pseudomonas putida KT2240. The computational analysis with flux balance analysis (FBA)

#### Edited and reviewed by:

Pablo Ivan Nikel, Novo Nordisk Foundation Center for Biosustainability (DTU Biosustain), Denmark

> \*Correspondence: Jean Marie François fran\_jm@insa-toulouse.fr

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

> Received: 08 April 2020 Accepted: 15 April 2020 Published: 06 May 2020

#### Citation:

François JM, Jantti J and Daboussi F (2020) Editorial: 4th Applied Synthetic Biology in Europe. Front. Bioeng. Biotechnol. 8:431. doi: 10.3389/fbioe.2020.00431

indicated highest yield of 12 out of the 14 value products by the isomerase pathway, which was confirmed in vivo for the production of mono-rhamnolipids and pyocyanin. While reinforcing the usefulness of FBA to guide pathway design, the limitation of these in silico methods is that they cannot predict best productivity and titer. Ultimately, the work showed that higher production rate and titer were obtained with the Weimberg pathway for these high value products. Product yield higher than those calculated from the stoichiometry of natural pathways can be obtained by pathway reconstruction in which loss of carbon as CO<sup>2</sup> is avoided, as recently shown for the construction of the "NOG" pathway (Bogorad et al., 2013) and reviewed by François et al.. Along this line, Lachaux et al. devised a cyclic pathway enabling optimal conversion of C5 and C6 carbon into glycolic acid by reallocating the function of two E.coli kdsD and fsaA genes encoding arabinose-5-P isomerase and fructose-6-P aldolase, respectively, in combination with the non-oxidative pentose pathway. Altogether, these contributions illustrate the trend of integrating systems biology, synthetic biology, and evolutionary engineering with traditional metabolic engineering to expedite the development of relevant microbial cell factories for the production of bio-based products from renewable carbon sources, as the cornerstone of the emerging Bio-Economy. However, there is still a gap to be filled in to make these strains industrially competitive and the attendant fermentation processes economically viable. Notably, better knowledge of the management of the cellular energy and redox balance by metabolic engineering as reviewed in Kalnenieks et al. for the ethanologenic bacterium Zymomonas mobilis and development of analytical tools for rapid screening of products yield, such as sensing and monitoring production of benzoic derivatives by mean of a biosensor (Castaño-Cerezo et al.) are important to make rapid progress in this applied field of Synthetic Biology.

Generation and investigation of biological functions can be in in vitro systems using cell-free systems that rests on the use of cell free protein synthesis (CFPS). Pioneered by Nirenberg and Matthaei (1961) to elucidate the genetic code, cell-free systems are gaining more and more credit in the field of Synthetic Biology as a prototype platform to engineer biological parts at levels of protein, metabolism and cell (Shimizu et al., 2006; Hodgman and Jewett, 2012; Moore et al., 2017). Cell free synthetic biology was particularly emphasized in the session "Biosciences and Bionanoengineering' with several contributions. Spice et al. reported on the synthesis and assembly of hepatitis B virus-like particles (VLP) using cell-free system prepared from Pichia pastoris, which can represent an alternative approach to the production of VLP for therapeutic purposes. On the other hand, Gabant and Borrero exploited cell-free protein synthesis system as a rapid method to expedite production and validation of a large set of bacteriocin, which are cataloged into a database, termed "PARAGEN1.0." Cell free system is also suggested as an alternative strategy for engineering and production of organophosphorus hydrolysing enzymes to be used to decontaminate highly toxic agricultural pesticides and chemical warfare (Thakur et al.). In addition, these authors make a thorough overview on the practice that are currently used for decontaminating environments and seawater from toxic organophosphorus compounds, emphasizing on the need to strongly enhance catalytic activity and stability of these organophosphorus hydrolysing enzymes by various bionanoengineering approaches.

Genome editing, genome engineering, and synthetic genomics are among the most disruptive technology at the heart of Synthetic Biology. A contribution to this session was a paper by Millacura et al. who report on a cell-based logic system termed ParAlleL that allows to construct complex logic circuits enable to respond to a combination of inputs. They demonstrate their concept through the construction of three simple logic circuits, a 1-bit full adder, a 1-bit subtractor, and a numeric display. Experimentally, the concept was shown by transforming E. coli with three out of six plasmids to make them able to survive only when precisely three antibiotics are provided. This represents an interesting way of constructing circuits from cellular populations in which logic is constructed and circuitry functionality displayed simply by the surviving cells in the presence of antibiotics. Very challenging in genetic and genome engineering is to build efficient and robust gene expression leading to high protein production, for instance. Eichmann et al. tentatively solved the problem of expression and production of so-called three difficult-to-express proteins by systematic evaluation of ribosome binding sites, N-terminal affinity tags and periplasmic translocation sequences to which was fused green fluorescent protein (GFP) to identify high producers by fluorescence activated cell screening (FACS). Large combinatorial libraries of genes were expressed and the outcome was measured in a platform monitoring expression automatically. Though yields were still low, this work unraveled that TAT-dependent signal sequence from E. coli YahJ for expression is recognized in the fast-growing bacterium Vibrio natriegens, which could be used as an industrial platform for production of recombinant proteins.

This editorial can be concluded by referring to the article by El Karoui et al. which reports on a workshop that brought together British academics and industrialists at the University of Edinburgh to discuss the future of synthetic biology. If processes that underlie all living systems can be manipulated, modified or reshaped using the tools of Synthetic Biology, then in return we need to take the public with us on this journey to create a productive dialogue about the work we can do and the impact it can have on our world and our way of life.

#### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

**6**

# REFERENCES


Shimizu, Y., Kuruma, Y., Ying, B. W., Umekage, S., and Ueda, T. (2006). Cellfree translation systems for protein engineering. FEBS J. 273, 4133–4140. doi: 10.1111/j.1742-4658.2006.05431.x

**Conflict of Interest:** JJ was employed by company VTT Technical Research Centre of Finland.

The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 François, Jantti and Daboussi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Rational Engineering of Phenylalanine Accumulation in *Pseudomonas taiwanensis* to Enable High-Yield Production of *Trans*-Cinnamate

Maike Otto<sup>1</sup> , Benedikt Wynands <sup>1</sup> , Christoph Lenzen<sup>2</sup> , Melanie Filbig<sup>2</sup> , Lars M. Blank <sup>2</sup> and Nick Wierckx <sup>1</sup> \*

#### *Edited by:*

Jean Marie François, UMS3582 Toulouse White Biotechnology (TWB), France

#### *Reviewed by:*

Jian-Ming Liu, Technical University of Denmark, Denmark John A. Morgan, Purdue University, United States

> *\*Correspondence:* Nick Wierckx n.wierckx@fz-juelich.de

#### *Specialty section:*

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

*Received:* 11 September 2019 *Accepted:* 23 October 2019 *Published:* 20 November 2019

#### *Citation:*

Otto M, Wynands B, Lenzen C, Filbig M, Blank LM and Wierckx N (2019) Rational Engineering of Phenylalanine Accumulation in Pseudomonas taiwanensis to Enable High-Yield Production of Trans-Cinnamate. Front. Bioeng. Biotechnol. 7:312. doi: 10.3389/fbioe.2019.00312 1 Institute of Bio- and Geosciences (IBG-1: Biotechnology), Forschungszentrum Jülich GmbH, Jülich, Germany, <sup>2</sup> Institute of Applied Microbiology, Rheinisch-Westfälische Technische Hochschule (RWTH) Aachen University, Aachen, Germany

Microbial biocatalysis represents a promising alternative for the production of a variety of aromatic chemicals, where microorganisms are engineered to convert a renewable feedstock under mild production conditions into a valuable chemical building block. This study describes the rational engineering of the solvent-tolerant bacterium Pseudomonas taiwanensis VLB120 toward accumulation of L-phenylalanine and its conversion into the chemical building block t-cinnamate. We recently reported rational engineering of Pseudomonas toward L-tyrosine accumulation by the insertion of genetic modifications that allow both enhanced flux and prevent aromatics degradation. Building on this knowledge, three genes encoding for enzymes involved in the degradation of L-phenylalanine were deleted to allow accumulation of 2.6 mM of L-phenylalanine from 20 mM glucose. The amino acid was subsequently converted into the aromatic model compound t-cinnamate by the expression of a phenylalanine ammonia-lyase (PAL) from Arabidopsis thaliana. The engineered strains produced t-cinnamate with yields of 23 and 39% Cmol Cmol−<sup>1</sup> from glucose and glycerol, respectively. Yields were improved up to 48% Cmol Cmol−<sup>1</sup> from glycerol when two enzymes involved in the shikimate pathway were additionally overexpressed, however with negative impact on strain performance and reproducibility. Production titers were increased in fed-batch fermentations, in which 33.5 mM t-cinnamate were produced solely from glycerol, in a mineral medium without additional complex supplements. The aspect of product toxicity was targeted by the utilization of a streamlined, genome-reduced strain, which improves upon the already high tolerance of P. taiwanensis VLB120 toward t-cinnamate.

Keywords: *Pseudomonas*, metabolic engineering, *trans*-cinnamic acid, L-phenylalanine, rational engineering, glycerol, glucose

# INTRODUCTION

Trans-cinnamate is an aromatic compound naturally occurring in plants, where it serves as central intermediate for the biosynthesis of a large number of substances, including coumarins, flavonoids, and phenylpropanoids (Chemler and Koffas, 2008; Vogt, 2010). It is widely used in industry for flavoring, pharmaceuticals, and cosmetics (Fausta et al., 1999; De et al., 2011; Vargas-Tah and Gosset, 2015) and can serve as precursor for bio-based drop-in chemicals such as styrene (McKenna and Nielsen, 2011) or for value-added chemicals such as the stilbene pinosylvin (van Summeren-Wesenhagen and Marienhagen, 2015). Commercial production currently happens from petroleum-based, non-renewable feedstocks in processes that demand high amounts of energy and release toxic by-products (Bruckner, 2010; Tietze et al., 2015). This comes along with increasing apprehension on global climate change and depleting aromatic fossil resources. Hence, there are both environmental and economic drivers for alternative synthesis routes.

Microbial production by whole-cell biocatalysis represents a less environmentally demanding alternative to the common and well-established petrochemical processes (Hatti-Kaul et al., 2007; Becker and Wittmann, 2012; Cho et al., 2015; Kallscheuer et al., 2019). Herein, microbes can be supplied with renewable feedstocks (e.g., glucose, glycerol, or lignin, Kohlstedt et al., 2018; Johnson et al., 2019), and metabolic conversion enables the synthesis of diverse products under mild production conditions, either via native enzymes (Hosseinpour Tehrani et al., 2019) or by heterologous expression of foreign genes (Kallscheuer et al., 2019). In plants, t-cinnamate is formed through deamination of the amino acid L-phenylalanine by phenylalanine ammonialyase (PAL) (Cochrane et al., 2004; Huang et al., 2010), and can be further converted into the cis-isoform by photoisomerization (Salum and Erra-Balsells, 2013). While only the trans-isoform is involved in biosynthesis pathways in plants (Salum and Erra-Balsells, 2013), cis-cinnamate displays higher anti-bacterial activities (Chen et al., 2011). The heterologous synthesis of this enzyme in various microorganisms, including Escherichia coli (Vargas-Tah et al., 2015; Bang et al., 2018) or Streptomyces lividans (Noda et al., 2011) enabled t-cinnamate production. Product toxicity is a limiting factor for the efficiency of many microbial production process with these hosts (McKenna and Nielsen, 2011). From a biochemical engineering perspective, it is important to "begin with the end in mind" when developing microbial production strains (Straathof et al., 2019), and product toxicity is one important aspect for this.

Bacteria of the genus Pseudomonas are considered as promising alternative host to produce aromatics such as t-cinnamate. Their robust growth behavior and metabolic versatility recently enabled the synthesis of many different industrially relevant compounds, including a variety of aromatics (Nijkamp et al., 2007; Kuepper et al., 2015; Wynands et al., 2018), rhamnolipids (Tiso et al., 2017), terpenes (Mi et al., 2014), or prodiginines (Domröse et al., 2015). This is in particular due to their high stress tolerance, which has been extensively investigated in the past (Kusumawardhani et al., 2018). Pseudomonads are able to thrive under both endogenous and exogenous oxidative stress, enabled by their particular central carbon metabolism architecture (Chavarría et al., 2013; Nikel et al., 2015). Furthermore, they are equipped with a variety of native tolerance mechanisms that allow growth in the presence of highly toxic compounds (Sardessai and Bhosle, 2002; Segura et al., 2012). For this, Pseudomonads are able to adapt the composition of the inner and outer membrane to lower the permeability for substrates and they can increase their production of energy to fuel energy-consuming tolerance mechanisms (Isken and de Bont, 1998; Ramos et al., 2002; Belda et al., 2016). Some strains of Pseudomonas additionally express solvent efflux pumps that can actively extrude toxic compounds from the inner membrane (Kieboom et al., 1998) and degrade solvents such as toluene (Ramos et al., 2002) or styrene (Velasco et al., 1998). Additional transporters confer higher resistance to further aromatic molecules like p-hydroxybenzoate or pcoumarate (Verhoef et al., 2010; Calero et al., 2018). The native capabilities of Pseudomonas species thus form a strong basis for a biocatalytic process for t-cinnamate synthesis by expanding process options due to reduced product toxicity.

To further exploit its potential as industrial production strain, we recently enhanced the bioprocess features of the solventtolerant strain Pseudomonas taiwanensis VLB120 by successive feature reduction (Wynands et al., 2019). The deletion of redundant genomic elements such as proviral segments, genes for biofilm formation and flagella expression and the megaplasmid pSTY resulted in strains with 15% enhanced growth rates and increased biomass yields, thereby improving the overall performance of the strain under bioprocess conditions. In addition, genes encoding for the efflux pump TtgGHI located on the pSTY plasmid were re-integrated into the chromosome of the genome-reduced chassis (GRC) strains to maintain tolerance enabled by this efflux pump. These modifications, in combination with inherent tolerance traits, make the engineered strains promising hosts to efficiently synthesize aromatic compounds like t-cinnamate, as well as potential hydrophobic derivatives such as styrene or stilbenes.

The potential of Pseudomonas species for t-cinnamate production has been demonstrated in a strain of P. putida S12 by the expression of PAL from Rhodosporidium toruloides (Nijkamp et al., 2005). Here, enhanced precursor supply was realized by mutagenesis and subsequent selection on the toxic analog m-fluoro-phenylalanine, leading to the production of 5 mM tcinnamate with a yield of 6.7% Cmol Cmol−<sup>1</sup> . However, although genomic analysis of this strain provided leads to enhance the flux through the shikimate pathway in P. taiwanensis VLB120 (Wynands et al., 2018), the genetic rearrangements enabling the enhanced flux specifically toward L-phenylalanine have not been determined. We recently reported a study on phenol production from L-tyrosine in P. taiwanensis VLB120, where 22 mutations in 50 different strains delivered a comprehensive insight on the rational engineering of L-tyrosine accumulation (Wynands et al., 2018). In the most productive strain, five genes involved in the degradation of aromatic intermediates and the gene pykA were deleted to increase the precursor supply for the synthesis of aromatic amino acids. The flux through the shikimate pathway was further enhanced by genomic modifications of genes at their native chromosomal locus resulting in a strain able to accumulate up to 2.8 mM of L-tyrosine from 20 mM glucose.

The aim of this study was to build on this knowledge to redirect the enhanced flux to L-tyrosine in P. taiwanensis GRC3 151pykA-tap (Wynands, 2018) toward <sup>L</sup>-phenylalanine, given the close relation of their biosynthesis pathways, and thereby expanding the potential product range. Furthermore, the superior tolerance of the streamlined P. taiwanensis GRC toward high concentrations of t-cinnamate was assessed. Routes involved in L-phenylalanine degradation were deleted, leading to L-phenylalanine accumulation. Titers were further increased by an additional overexpression of feedback-resistant versions of the bottleneck enzymes AroG and PheA. Heterologous expression of a PAL from Arabidopsis thaliana in the engineered strains then enabled the production of t-cinnamate. In a completely minimal medium, the plasmid-free chassis strains produced up to 6.3 mM of t-cinnamate from glycerol, which corresponds to a yield of 48% Cmol Cmol−<sup>1</sup> . Cultivation in fed-batch fermentations resulted in titers of around 33.5 mM, which is the highest reported titer of microbially produced t-cinnamate in a cultivation medium without complex supplements. The engineered strains can further serve as efficient platform to produce various products from L-phenylalanine and t-cinnamate, including styrene, benzoate, or plant polyphenols such as pinosylvin.

# MATERIALS AND METHODS

#### Bacterial Strains, Plasmids, and Cultivation Conditions

Strains and plasmids used in this study can be found in **Tables 1**, **2**. For cloning purposes, E. coli and Pseudomonas cells were cultivated at 37 or 30◦C, respectively, either in liquid LB medium containing 5 g L−<sup>1</sup> sodium chloride or on solid LB agar plates [additionally containing 1.5% (w/v) agar]. After tri- or four-parental mating procedures, Pseudomonads were isolated on cetrimide agar (Sigma Aldrich) plates supplemented with 10 mL L−<sup>1</sup> glycerol. Kanamycin (50 µg mL−<sup>1</sup> ) or gentamicin (20 µg mL−<sup>1</sup> ) was added to cultures or plates when necessary.

During production and toxicity experiments in shake flasks and well plates, liquid cultures of P. taiwanensis were grown in mineral salt medium (MSM) adapted from Hartmans et al. (1989) at pH 7.0 without the addition of antibiotics. The medium's standard phosphate buffer capacity was increased to 5-fold (111.5 mM K2HPO<sup>4</sup> and 68 mM NaH2PO4). Glucose or glycerol were added as sole carbon source at indicated concentrations. Main cultures were inoculated at an OD<sup>600</sup> of 0.2, from seed cultures grown in MSM containing glucose as carbon source. Production experiments were performed in 500 mL Erlenmeyer flasks with a culture volume of 50 mL, which were cultivated in a rotary shaker with a frequency of 200 rpm and a throw of 50 mm. Toxicity assays were performed in 96-round low-well plates with a filling volume of 200 µL in the System Duetz <sup>R</sup> cultivation system (EnzyScreen, Leiden, Netherlands) with a shaking frequency of 225 rpm. OD<sup>600</sup> values were calculated from the measured green values using a calibration. Phenylalanine accumulation was analyzed in 24-square deep-well plates with a filling volume of 5 mL and a shaking frequency of 225 rpm, and growth was monitored in the Growth profiler <sup>R</sup> system.

# Plasmid Construction and Genomic Modification

For plasmid construction, DNA fragments were PCR amplified using the Q5 high-fidelity polymerase (New England Biolabs, New Ipswich, USA) with corresponding overhangs to enable subsequent Gibson assembly (Gibson et al., 2009) with the HiFi DNA assembly master mix (New England Biolabs, New Ipswich, USA). Primers were ordered from Eurofins Genomics (Ebersberg, Germany). The gene sequences for aroGfbr (from E. coli K12 W3110) and AtPAL (from A. thaliana) were codonoptimized for P. taiwanensis VLB120 using the online tool OPTIMIZER (Puigbo et al., 2007) and ordered as synthetic DNA fragments from Thermo Fisher Scientific (Waltham, USA). The gene pheAT310I was amplified from genomic DNA of strain P. putida S12palM12. pEMG-based plasmids were transformed into E. coli DH5α λpir cells, pBG-based plasmids were transformed into E. coli PIR2. Correct plasmid assembly was verified by Sanger sequencing performed by Eurofins Genomics (Ebersberg, Germany). Integration at the attTn7-site was achieved by patch-mating of the E. coli donor strain holding the respective pBG-plasmid, the helper strain E. coli HB101 pRK2013, DH5α λpir pTNS1 providing the required transposase and the recipient Pseudomonas taiwanensis strain as described by Wynands et al. (2018). After the mating procedure, Pseudomonas were isolated on Cetrimide agar containing gentamicin and correct integration was confirmed by colony PCR using the OneTaq Quick-Load Master Mix (New England Biolabs, New Ipswich, USA).

Genomic deletions and point mutations were realized using the I-SceI-based method developed by Martínez-García and de Lorenzo (2011) using a streamlined protocol adapted by Wynands et al. (2018). Successful deletions were verified by colony PCR using the OneTaq quick-Load Master Mix (New England Biolabs, New Ipswich, USA).

## Fed-Batch in Controlled Bioreactors

Fermentations were carried out in fed-batch operation mode in DASbox <sup>R</sup> mini-bioreactors using the DASware control software (Eppendorf, Hamburg, Germany). The reactors were set up of 385 ml glass vessels, two Rushton-type impellers driven by direct overhead drives, feeding lines for acid, base, and carbon source, a temperature sensor, an EasyFerm Plus K8 120 pH-sensor (Hamilton Company, Reno, NV, USA) and an InPro <sup>R</sup> 6800 series O<sup>2</sup> sensor (Mettler-Toledo International Inc., Columbus, OH, USA). The starting volume was 100 mL and the temperature was maintained at 30◦C. Gas supply was provided via headspace with a starting gas flow rate of 6 sL h−<sup>1</sup> . The initial agitation frequency was 500 rpm. The dO<sup>2</sup> was controlled with a cascade, keeping the concentration of diluted oxygen at 35% by first increasing the agitation speed up to 1,200 rpm, then increasing the oxygen concentration in the air supply to a maximum of 80% and subsequently increasing the gas flow rate. pH 7 was maintained by automatic addition of 5% NH<sup>3</sup> or 1 M H2SO4.

#### TABLE 1 | Plasmids.


Cells were grown in MSM according to Hartmans et al. (1989), where the addition of mineral salts stock was increased by 2-fold (MgCl2.6H2O 0.2 g L−<sup>1</sup> , ZnSO4.7H2O 0.004 g L−<sup>1</sup> , CaCl2.2H2O 0.002 g L−<sup>1</sup> , FeSO4.7H2O 0.01 g L−<sup>1</sup> , Na2MoO4.2H2O 0.0004 g L −1 , CuSO4.5H2O 0.0004 g L−<sup>1</sup> , CoCl2.6H2O 0.0008 g L−<sup>1</sup> , MnCL2.2H2O 0.002 g L−<sup>1</sup> ). The initial batch medium contained either 20 mM of glucose or 40 mM of glycerol as sole carbon source. The reactors were operated in a dO2-controlled fed-batch mode. When glucose or glycerol was depleted, the dO<sup>2</sup> signal rapidly increased as a result of metabolic arrest. A dO<sup>2</sup> signal >70 triggered a feed pump, resulting in a feed shot of 5 mM of glucose or 10 mM of glycerol per trigger initiation.

#### Analytical Methods

Optical densities of liquid cultures were measured at 600 nm using an Ultrospec 10 Cell Density Meter (GE Healthcare, Illinois, USA).

Samples taken from the cultures were centrifuged at 13,000 rpm for 2–5 min and the supernatant was analyzed by HPLC. Aromatics quantification was performed using a Beckman System Gold 126 Solvent Module with a 168 diode 201 array detector (Beckman Coulter, Brea, USA) and an ISAspher 100-5 C18 BDS reversed-phase 202 column (ISERA, Düren, Germany) at 30◦C and a flow rate of 0.8 mL min−<sup>1</sup> . Elution took place with a gradient starting at 90% H2O containing 0.1% (v/v) TFA and 10% methanol. This ratio was held for 2 min, followed by gradual increase to 100% methanol over the course of 8 min. After 2 min at 100% methanol, initial ratios were reached again within 1 min and held constant for further 2 min. UV detection was conducted at a wavelength of 245 nm. L-phenylalanine and L-tyrosine quantification shown in **Figure 3** was performed with the same column and elution setup as mentioned before, in a Dionex Ultimate 3000 HPLC system, where detection took place with a Corona Veo charged aerosol detector (Thermo Fisher Scientific, Waltham, MA, USA). Glucose, gluconate and glycerol analysis was performed using a Beckman System Gold 126 Solvent Module with a System Gold 166 UV-detector (Beckman Coulter, Brea, USA) and Smartline RI Detector 2300 (KNAUER, Berlin, Germany) on a MetabAAC column (ISERA, Düren, Germany). Elution took place with 5 mM H2SO<sup>4</sup> at an isocratic flow of 0.5 mL min−<sup>1</sup> and a temperature of 30◦C for 20 min. Glucose and glycerol were analyzed using the RI detector, gluconate concentrations were determined with the UV detector at a wavelength of 210 nm. Due to co-elution of glucose and gluconate, glucose concentrations were determined by subtraction of the gluconate concentration.

#### RESULTS AND DISCUSSION

## Enhanced *Trans*-Cinnamate Tolerance of Streamlined *P. taiwanensis* Chassis Strains

Product toxicity vastly influences the efficiency of a microbial production process. Cellular stress induced by toxic compounds leads to decreased production rates or complete growth arrest, thereby limiting feasible product titers and yields (Sardessai and Bhosle, 2002; McKenna and Nielsen, 2011; Kusumawardhani et al., 2018). Compared to hydrophobic

#### TABLE 2 | Bacterial strains.


aromatics such as styrene, t-cinnamate is mildly toxic at neutral pH and it exhibits a different mode of action. Increasing concentrations of t-cinnamate lead to enhanced osmotic stress in cell cultures, but will also cause the inhibition of specific enzymes and disturb cellular processes, resulting in growth defects (Olasupo et al., 2003; Guzman, 2014). Growth of E. coli, for example, is heavily impaired at concentrations above 2.7 mM (400 mg L−<sup>1</sup> ) and completely inhibited at >5.4 mM (800 mg L−<sup>1</sup> ) when t-cinnamate is added at the beginning of cultivation (Olasupo et al., 2003; McKenna and Nielsen, 2011). This necessitates the addition of expensive complex supplements such as yeast extract or casamino acids or a biotransformation approach with late PAL induction at high cell densities to allow for sufficient accumulation of the precursor L-phenylalanine (Noda et al., 2011; Bang et al., 2018).

These problems can be avoided by the utilization of robust, natively stress-resistant host strains such as Pseudomonas taiwanensis VLB120. As shown in **Figure 1A**, the wildtype can grow in the presence of 30 mM (4.4 g L−<sup>1</sup> ) t-cinnamate in MSM with a rate of 0.24 ± 0.01 h−<sup>1</sup> . P. taiwanensis VLB120 is equipped with particular mechanisms that enable this superior tolerance toward a variety of toxic compounds. While RNDtype efflux pumps mainly act on hydrophobic aromatic solvents and a number of antibiotics (Terán et al., 2003; Köhler et al., 2013; Volmer et al., 2014), an ABC transporter (Ttg2ABC) was recently identified as crucial for p-coumarate tolerance, the hydroxylated derivative of t-cinnamate (Calero et al., 2018). Furthermore, chaperone upregulation is observed as response to protein misfolding (Segura et al., 2012).

To further exploit the native potential of P. taiwanensis VLB120 as microbial cell factory, we recently engineered a

genome-reduced variant of this strain which exhibited higher growth rates and enhanced biomass formation (Wynands et al., 2019). The benefit of genome reduction for potential aromatics production strains is underlined by the tolerance test in the presence of 30 mM t-cinnamate (**Figure 1A**). The streamlined strains GRC1 and GRC3 had a reduced lag phase in the presence of high t-cinnamate concentrations, while the growth rate remains similar to the wildtype. In addition, the GRC strains reached higher final OD<sup>600</sup> values compared to the wildtype. This is particularly important to improve the efficiency of a microbial production process by reducing the overall process time and lowering the amount of substrate needed to generated and maintain the biomass. The only exception is strain P. taiwanensis GRC2, which grew slower and with a longer lag phase than the wildtype. In contrast to GRC1 and 3, GRC2 constitutively expresses ttgGHI genes that encode a solvent-efflux pump. Although this constitutive expression greatly increases the fitness of GRC2 in the presence of hydrophobic solvents such as toluene, it causes a fitness reduction in the absence of solvents (Wynands et al., 2019). This drawback is absent in the other two strains, where the whole ttg operon is absent (GRC1), or the regulatory genes ttgVW are included to induce the pump in response to solvents (GRC3). The similar performance of GRC1 and GRC3 in the presence of t-cinnamate indicates that this aromatic acid does not induce expression of the ttgGHI, nor does the TtgGHI pump contribute to cinnamate tolerance. Given this similar performance, and in light of the potential to use a t-cinnamate-producing strain as platform for a variety of further industrially relevant hydrophobic aromatics, the strain GRC3 was chosen as host to maintain the possibility of solvent efflux by TtgGHI. As shown in **Figure 1B**, P. taiwanensis GRC3 can grow in the presence of t-cinnamate concentrations of up to 50 mM (7.4 g L−<sup>1</sup> ), which is 10-fold higher than MIC for E. coli. The native tolerance potential of P. taiwanensis VLB120 in combination with improved bioprocess features obtained by streamlining of this strain hence set an ideal starting point for an efficient microbial production process of t-cinnamate.

# Rational Engineering of L-phenylalanine Production

t-cinnamate is the deamination product of L-phenylalanine and the enhanced supply of this precursor in microbial chassis strains is thus crucial to allow efficient production. A variety of approaches toward L-phenylalanine overproduction were reported over the last years, delivering highly productive strains of e.g., E. coli (63 g L−<sup>1</sup> from glucose) (Ding et al., 2016) or Corynebacterium glutamicum (16 g L−<sup>1</sup> from glucose) (Zhang et al., 2015). Biosynthesis of L-phenylalanine and L-tyrosine both starts with the conversion of chorismate into prephenate, which then branches via phenylpyruvate to L-phenylalanine or via 4-hydroxyphenylpyruvate to L-tyrosine. Popular strategies aim to deregulate a 3-deoxy-D-arabino-heptolusonate-7-phosphate (DAHP) synthase (e.g., AroG, AroF, or AroH) or the chorismate mutase/prephenate dehydratase (PheA) activity and to increase the availability of the precursors phosphoenolpyruvate (PEP) and erythrose-4-phosphate (E4P) (Rodriguez et al., 2014; Ding et al., 2016; Huccetogullari et al., 2019). Pseudomonads have the catabolic potential to degrade a high number of aromatic compounds. This ability is a key feature applied for many biotechnological aspects such as the metabolism of complex substrates (Xu et al., 2018), but it also increases the difficulty of manufacturing aromatic amino acid accumulation in this species. We recently reported a study describing the rational engineering of P. taiwanensis VLB120 to produce phenol via the aromatic amino acid L-tyrosine (Wynands et al., 2018). The deletion of five genes (pobA, hpd, quiC, quiC1, quiC2) associated to the degradation of shikimate pathway-derived compounds resulted in strains unable to catabolize L-tyrosine, L-phenylalanine, p-hydroxybenzoate, and 3-dehydroshikimate to ensure aromatics accumulation. The flux toward L-tyrosine was subsequently enhanced by the insertion of point mutations into the native locus of genes in the shikimate pathway, resulting in enzymes with amino acid substitutions (TrpEP290S, AroF-1 P148L, PheAT310I). The deletion of pykA (encoding for a pyruvate kinase) additionally reduced the flux of PEP toward

t-cinnamate. Red arrows and annotations indicate gene deletions, purple arrows represent enzymatic downregulation, green annotations highlight point mutations introduced into the native gene locus, and blue tags and arrows represent the overexpression of heterologous genes. DAHP, dihydroxyacetone phosphate; KDPG, 2-keto-3-deoxy-6-phosphogluconate; F6P, fructose-6-phosphate; G3P, glyceraldehyde-3-phosphate; 3PG, 3-phosphoglycerat; S7P, seduheptulose-7-phosphate; PEP, phosphoenolpyruvate; PYR, pyruvate; E4P, erythrose-4-phosphate; DAHP, 3-deoxy-D-arabinoheptulosonate-7-phosphate; DHQ, 3-dehydroquinate; DHS, 3-dehydroshikimate; SA, shikimate; S3P, shikimate-3-phosphate; EPSP, 5-enolpyruvyl-shikimate-5-phosphate; CHO, chorismate; 4-HB, 4-hydroxybenzoate; PCA, protocatechuate; ANT, anthranilate; TRP, tryptophan; PRE, prephenate; PP, phenylpyruvate; HPP, 4-hydroxyphenylpyruvate; ARO, arogenate; TYR, tyrosine; PHE, phenylalanine; HMG, homogentisate; PAA, 2-phenylacetamide; PEA, phenylethylamine; PykA/PykF, pyruvate kinase isozymes; QuiC/QuiC1/QuiC2, 3-dehydroshikimate dehydratase isozymes; PobA, 4-hydroxybenzoate 3-monooxygenase; Hpd, 4-hydroxyphenylpyruvate dioxygenase; AroF-1P148L/AroGfbr, DAHP synthase isozymes; TrpEP290S, anthranilate synthase (component I); PheAT310I, bi-functional chorismate mutase/prephenate dehydratase; PhhAB, phenylalanine 4-monooxygenase; KatG, catalase-peroxidase; PVLB\_10925, aromatic-L-amino-acid decarboxylase; PAL, phenylalanine ammonia-lyase.

pyruvate, thereby increasing the precursor pool for the shikimate pathway. An introduction of these mutations in the genomereduced strains of P. taiwanensis VLB120 led to the efficient production of phenol from L-tyrosine (Wynands et al., 2019). As shown in **Figure 3**, the strain P. taiwanensis GRC3 151pykAtap accumulates 2.85 ± 0.02 mM of tyrosine from 20 mM of glucose, which is comparable to the non-genome-reduced strain. Three genes involved in phenylalanine catabolism were subsequently deleted from the chromosome of P. taiwanensis GRC3 151pykA-tap: (i) phhAB encoding for phenylalanine-4-monooxygenase involved in the conversion of phenylalanine to tyrosine (Herrera et al., 2010), (ii) PVLB\_10925, putatively encoding an aromatic-L-amino-acid decarboxylase responsible for the decarboxylation into phenylethylamine, and (iii) katG, putatively coding for a catalase-peroxidase which converts phenylalanine into phenyl acetamide (**Figure 2**).

Pseudomonads possess a variety of transporters that allow export and import of aromatic amino acids. Transcriptome analysis of a Pseudomonas putida strain with increased flux toward tyrosine revealed that upon increased intracellular aromatic amino acid levels, amino acid exporters are upregulated, while uptake systems are downregulated (Wierckx et al., 2008). Aromatics accumulation shown in **Figure 3** displays extracellular concentrations, indicating enhanced efflux of both tyrosine and phenylalanine.

The deletion of phhAB led to the accumulation of 2.22 ± 0.02 mM of phenylalanine and 0.29 ± 0.01 mM tyrosine after 50 h (**Figure 3**). This indicates that about 88% of L-tyrosine

accumulating due to the enhanced flux toward prephenate in strain P. taiwanensis GRC3 151pykA-tap is stemming from <sup>L</sup>phenylalanine. In most microorganisms, L-phenylalanine cannot be converted into L-tyrosine in which case it is solely synthesized from prephenate (Guroff and Ito, 1965). In Pseudomonas however, the pathway via L-phenylalanine seems to be a major route for L-tyrosine formation under the given conditions (Wierckx et al., 2009). As described for other Pseudomonas species, Xanthomonads and Alcalignes (Ahmad et al., 1990), this diversity of aromatic amino acid metabolism account for the flexibility of the organism to cope with a variety of end product analogs (Fiske et al., 1983). The concentration of L-phenylalanine in the culture of P. taiwanensis GRC3 151pykA-tap1phhAB was slightly reduced after 70 h, indicating that other pathways involved in L-phenylalanine degradation were still active in this strain. The subsequent deletion of gene PVLB\_10925 prevented this degradation, resulting instead in a further increase of the final titer to 2.62 ± 0.00 mM L-phenylalanine. The deletion of katG had no influence on phenylalanine accumulation and the strain shows a similar production pattern as the strain without a deletion. Indeed, as the progenitor strain P. taiwanensis GRC3 151pykA-tap was already not able to grow on <sup>L</sup>phenylalanine and L-tyrosine (Wynands et al., 2018), this confirms that catabolic pathways are either not active (katG) or not connected to the central carbon metabolism (PVLB\_10925) under the applied conditions, even though this strain contains the genetic inventory for the full degradation of the resulting 2 phenylacetamide and phenylethylamine. This is underlined by similar observations in an L-phenylalanine overproducing strain of P. putida DOT-T1E (Molina-Santiago et al., 2016).

L-Tyrosine and L-phenylalanine production could be increased further by the additional overexpression of AroGfbr , a feedback inhibition resistant versions of the DAHP synthase from E. coli K12 W3110 (Kikuchi et al., 1997) and PheAT310I from P. putida S12palM12 (Nijkamp et al., 2005). The coding genes were integrated chromosomally at the attTn7-site and expressed under the control of the constitutive promoter P14g (Zobel et al., 2015). Overexpression in the tyrosine-accumulating strain P. taiwanensis GRC3 151pykA-tap led to tyrosine titers of about 3.5 ± 0.11 mM after 70 h. Transient L-phenylalanine accumulation was observed with this strain, further underlining the high flux through PhhAB. Upon overexpression of aroGfbr and pheAT310I in strains with deletions of phhAB, katG and PVLB\_10925, the resulting strain GRC3 181pykA-tap attTn7::P14d-aroGfbr -pheAAT31I (18 = eight deletions of pathways involved in aromatics degradation) accumulated 3.0 ± 0.07 mM of L-phenylalanine after 70 h. Also, L-tyrosine production was reduced, likely due to PheAT310I increasing the flux of prephenate toward L-phenylalanine.

#### Production of *t*-cinnamate From Glucose and Glycerol in Shake Flasks

The L-phenylalanine-accumulating strain P. taiwanensis GRC3 181pykA-tap was subsequently used as host for the production of t-cinnamate. The deamination of L-phenylalanine into t-cinnamate was achieved using the PAL2 enzyme from A.

thaliana. While four genes encoding PAL have been identified in A. thaliana (Cochrane et al., 2004), PAL2 displayed the highest specific activity when expressed in E. coli (McKenna and Nielsen, 2011). In addition, it has no activity on L-tyrosine as a substrate, in contrast to a variety of other yeast PALs (Cui et al., 2008). As overexpression of aroGfbr and pheAT310I improved L-phenylalanine yields by 15% (**Figure 3**), combinatorial expression with PAL on the Tn7 transposon under the control of P14g was applied for t-cinnamate production in the strain P. taiwanensis GRC3 181pykA-tap attTn7::P14gAtPALaroGfbr -pheAT310. The strain was cultivated in shake flasks in MSM containing either 20 mM of glucose or 40 mM of glycerol as sole carbon source. Glycerol is a promising alternative as feedstock for microbial production processes as abundant by-product of the bio-diesel production (Zambanini et al., 2016). When cultivated on glucose, the strain accumulated 3.80 ± 0.20 mM t-cinnamate, corresponding to a yield of 25.9 ± 0.1% Cmol Cmol−<sup>1</sup> , with a volumetric productivity of 0.10 ± 0.00 mM h −1 (**Figure 4**). From glycerol, the t-cinnamate titer was further increased to 6.33 ± 0.12 mM, corresponding to a yield of 47.5 ± 0.9% Cmol Cmol−<sup>1</sup> . As observed for the production of phenol and p-hydroxybenzoate (Wynands et al., 2018; Lenzen et al., 2019), metabolism of glycerol is highly beneficial for enhanced titers and yields of compounds derived from aromatic amino acids, likely due to metabolic rearrangements favoring the supply of PEP and E4P (Nikel et al., 2014; Poblete-Castro et al., 2019). The volumetric productivity on glycerol is slightly lower compared to glucose (0.09 ± 0.01 mM h−<sup>1</sup> ), as a result of the reduced growth rate. While this strain has a very high yield, its reliable application as a production host was problematic. When the experiment described above was repeated, the cultures displayed varying growth behavior in terms of lagphase, growth rates, and productivity (**Figure S1**), even though the cultivation procedure remained identical. Re-transformation of the Tn7-transposon bearing the overexpression constructs and culturing of single colonies yielded the same unstable phenotype. Sequencing of the attTn7-regions of theses strains

glucose (solid line) or 40 mM of glycerol (dotted line). Error bars represent the standard error of the mean (n = 3).

revealed no mutations in this sequence and verified correct integration. Likely, a combination of the extreme drain on carbon posed by the high t-cinnamate yield, the burden of heterologous overexpression of genes, and the production of a mildly toxic compound provide a strong selection for suppressor mutations. This impairs the reproducibility and renders the strain P. taiwanensis GRC3 181pykA-tap attTn7::P14gAtPALaroGfbr -pheAT310 unsuitable as reliable production host in its current form.

In contrast, P. taiwanensis GRC3 181pykA-tap attTn7::P14<sup>f</sup> AtPAL, harboring only the PAL for L-phenylalanine deamination under the control of the weaker P14f promoter, showed reliable growth and production patterns. This strain still carries feedback-resistant aroF-1P148<sup>L</sup> , pheAT310<sup>I</sup> , but only in their native context. The lack of in trans overexpression of aroGfbr and pheAT310I reduced t-cinnamate titers and yields slightly, with 3.34 ± 0.07 mM produced from glucose and 5.25 ± 0.22 mM from glycerol, which corresponds to yields of 22.8 ± 0.5 and 38.9 ± 1.6% Cmol Cmol−<sup>1</sup> , respectively. To the best of our knowledge, these are still the highest yields of t-cinnamate produced in a microbial process using a mineral medium without the addition of complex supplements. This strain furthermore reached higher final biomass and exhibited improved volumetric productivity on glucose (0.13 ± 0.00 mM h −1 ) and glycerol (0.11 ± 0.00 mM h−<sup>1</sup> ) compared to the strain additionally expressing aroGfbr and pheAT310I. In the course of the cultivations, no accumulation of L-phenylalanine was observed. This indicates that PAL activity is not inhibited by the concentrations of t-cinnamate reached in these cultures, in contrast to observations in other hosts and with PALs from different organisms (Nijkamp et al., 2005; McKenna and Nielsen, 2011; Molina-Santiago et al., 2016). Furthermore, no L-tyrosine accumulation was observed, in contrast to the phenylalanine-producing equivalent that does not express PAL (**Figure 3**), likely due to a reduction of product inhibition of the upstream metabolic enzymes as a result of efficient conversion to t-cinnamate.

growth (black lines) and t-cinnamate accumulation (blue lines) during fermentation of P. taiwanensis GRC3 181pykA-tap attTn7::P14gAtPAL- aroGfbr-pheAT<sup>310</sup> on glucose (A) and glycerol (B). The glycerol figure (B) displays data of a single reactor. The bottom graphs represent growth (black lines) and t-cinnamate accumulation (green lines) during fermentation of P. taiwanensis GRC3 181pykA-tap attTn7::P14fAtPAL on glucose (C) and glycerol (D). The error bars represent the standard error (n = 2).

#### Production in Controlled Bioreactors

In order to assess higher-level t-cinnamate production under more industrially relevant conditions, the strains were cultivated in fed-batch fermentations in controlled bioreactors. The fermenters were operated in a dO2-stat fed-batch mode where carbon depletion and the resulting increase of the dO<sup>2</sup> due to metabolic arrest triggered the initiation of a pulse feed of either 5 mM glucose or 10 mM glycerol (Johnson et al., 2016, 2019). This feeding protocol allows controlled addition of carbon source throughout the whole fermentation process, independent from time-varying carbon demands of the cell, thereby preventing excess carbon surplus (Johnson et al., 2016). The resulting feeding trends of single reactors are exemplarily shown in **Figure S3**. Cultivation took place at pH 7 in a mineral medium without the addition of complex supplements.

Under these conditions, the strain P. taiwanensis GRC3 181pykA-tap attTn7::P14gAtPAL-aroGfbr -pheAT310 accumulated 17.2 ± 0.28 mM of t-cinnamate from glucose, corresponding to a yield of 11.2 ± 0.6% (Cmol Cmol−<sup>1</sup> ) and a volumetric productivity of 0.19 ± 0.20 mM h−<sup>1</sup> . As shown in **Figure 5A**, the culture reaches a final OD<sup>600</sup> of 12.7 after around 65 h. At this point, a decrease in biomass concentration was observed. Furthermore, there was no significant increase in t-cinnamate titers from this point on, in spite of the fact that the base strain GRC3 tolerated higher concentrations of t-cinnamate up to 50 mM (**Figure 1**). It is possible that several factors impact the lowered growth performance of the production strains in the presence of t-cinnamate. One factor is the overexpression of heterologous genes which leads to lowered growth rates and impaired fitness as observed in shake flask experiments (**Figure 4**). Especially the overexpression of AroG and PheA appears to lead to increased cellular stress, which in turn also lowers the growth performance in the presence of toxic compounds. Furthermore, experiments regarding styrene toxicity have demonstrated a crucial difference in tolerance depending on whether a compound is added exogenously to the medium or produced intracellularly (Lian et al., 2016).

Fed-batch fermentations using glycerol as carbon source further underline complications using the phenotypically unstable strain P. taiwanensis GRC3 181pykA-tap attTn7::P14gAtPAL-aroGfbr -pheAT310 (**Figure 5B**; **Figure S2**). During fermentations, cultures in duplicate reactors showed different growth behavior while stemming from the same preculture. The cultures showed differences in lag-phase, growth rate and productivity. The final titers remained comparable, likely as a result of product inhibition as described before. **Figure 5B** shows a single fermentation on glycerol as an example. Here, the strain produced up to 19.8 mM t-cinnamate, with a yield


TABLE 3 | Comparison of t-cinnamate titer, yield, and volumetric productivity of strains P. taiwanensis GRC3 181pykA-tap attTn7::P14gAtPAL- aroGfbr-pheAT310 and GRC3 181pykA-tap attTn7::P14fAtPAL in shake flask cultivations and fed-batch fermentations.

of 47.8% Cmol and a volumetric productivity of 0.13 mM h−<sup>1</sup> . The extended lag phase observed during glycerol assimilation (Poblete-Castro et al., 2019) might even increase the occurrence of suppressor mutations by induction of the SOS response.

As observed for shake flask cultivations, reproducibility and stable production was restored in a strain overexpressing solely PAL for L-phenylalanine conversion. The strain P. taiwanensis GRC3 181pykA-tap attTn7::P14<sup>f</sup> AtPAL produced up to 33.2 ± 2.4 mM t-cinnamate from glucose (**Figure 5C**) and 33.5 ± 2.7 mM from glycerol (**Figure 5D**) in a fed-batch fermentation. While the productivity throughout the whole cultivation of P. taiwanensis GRC3 181pykA-tap attTn7::P14<sup>f</sup> AtPAL on glucose was 0.30 ± 0.02 mM h−<sup>1</sup> , increasing concentrations of tcinnamate impaired the strain's fitness and productivity. The volumetric productivity within the first 25 h is 0.55 ± 0.08 mM h −1 , which is reduced after 38 h to 0.16 ± 0.04 mM h−<sup>1</sup> . At the same time, a drop in OD<sup>600</sup> was observed at concentrations above 20 mM t-cinnamate. As production continued and cells were still viable at this point as indicated by the ongoing substrate consumption, this drop in OD<sup>600</sup> values hints toward an increased cumulative burden on the cell. The drop in OD<sup>600</sup> was not caused by biofilm formation, likely due to the deletion of the biofilm-associated lap genes in the GRC3 strain. Membrane adaptation is a key mechanisms of Pseudomonas to cope with environmental stress. Membrane active substances and certain environmental conditions cause adaptations such as cis- to trans-isomerization of fatty acids to increase membrane rigidity and lead to the formation of outer membrane vesicles to facilitate biofilm formation (Eberlein et al., 2018). In addition, many porins as well as import and export pumps are differentially expressed under stress, altering membrane permeability (Volkers et al., 2006; Ramos et al., 2015). The importance of membrane structure on similar compounds such as p-coumarate has recently been demonstrated (Calero et al., 2018) and can likely be partially be transferred to t-cinnamate tolerance.

As enzyme inhibition of different PALs by t-cinnamate has been observed in various studies (Nijkamp et al., 2007; McKenna and Nielsen, 2011), the decrease in productivity could also be linked to this effect. However, no L-phenylalanine accumulation was observed over the course of the fermentations, indicating that the precursor was completely converted by the PAL. Furthermore, no L-tyrosine accumulation was observed during fermentations.

In contrast to the strain overexpressing AroGfbr and PheAT310I, experiments with strain P. taiwanensis GCR3 181pykA-tap attTn7::P14<sup>f</sup> AtPAL are highly reproducible. While an additional expression of AroGfbr and PheAT310I led to higher yields in shake flasks, solely PAL expression in the L-phenylalanine-overproducing chassis strains enables both higher titers and volumetric productivities in fed-batch fermentations (**Table 3**). To the best of our knowledge, these are the highest t-cinnamate yields reported for a microbial production process. While titers of 6.5 g L−<sup>1</sup> (43.9 mM) have been reported in fed-batch fermentations of t-cinnamate-producing strains of E. coli (Bang et al., 2018), these processes required the addition of yeast extract in the initial batch medium and casamino acid in the feeding solution. The utilization of these engineered P. taiwanensis GRC strains enables growth without the addition of complex supplements. However, tolerance mechanisms for t-cinnamate remain to be further investigated and exploited to enhance titers and thus the feasibility of microbial t-cinnamate synthesis.

#### CONCLUSION

In this study, we describe the rational engineering of P. taiwanensis VLB120 towards efficient t-cinnamate production. The plasmid-free strain bearing no auxotrophies synthesized t-cinnamate from glucose or glycerol, with yields of up to 48% Cmol Cmol−<sup>1</sup> in shake flask cultivations. Titers were increased up to 33.35 mM in fed-batch fermentations using glycerol as sole carbon source. As product titers achieved in this study impair fitness and productivity of the chassis strains, the native tolerance features of Pseudomonas allowing enhanced t-cinnamate tolerance require further investigation and enhancement to increase process efficiency. One promising target is the ABC transporter Ttg2ABC, an extrusion pump involved in p-coumarate tolerance (Calero et al., 2018). An overexpression of this pump might deliver enhanced tolerance toward t-cinnamate, its regulation however remains to be investigated. Fine-tuning of additional aroG and pheA overexpression, e.g., by using weaker or inducible promoters, could furthermore avoid the observed growth defects while still maintaining the high yields achieved with this setup. Overall, the results underline the high potential of Pseudomonas species to produce chemical building blocks using aromatic amino acids as precursors. The establishment of efficient microbial production of the model compound t-cinnamate will in future serve as foundation to expand the product collection of this versatile species, ranging from bulk chemicals such as styrene (Lee et al., 2019) to specialty compounds such as stilbenes (van Summeren-Wesenhagen and Marienhagen, 2015).

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

### REFERENCES


# AUTHOR CONTRIBUTIONS

NW conceived and supervised the study with the help of LB. MO and MF performed the experiments with support of BW and CL. BW provided the GRC strains with deletions for L-tyrosine accumulation and deletion vectors. MO wrote the manuscript with the help of BW and NW. All authors read and approved the manuscript.

# FUNDING

MO, BW, CL, and NW were funded by the German Research Foundation (DFG) through the Emmy Noether program (WI 4255/1-1).

#### ACKNOWLEDGMENTS

We kindly acknowledge Eppendorf AG for assistance regarding scripting of automated feed protocols.

## SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00312/full#supplementary-material


increased specific styrene epoxidation activity. Appl. Environ. Microbiol. 80, 6539–6548. doi: 10.1128/AEM.01940-14


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Otto, Wynands, Lenzen, Filbig, Blank and Wierckx. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Comparison of Three Xylose Pathways in *Pseudomonas putida* KT2440 for the Synthesis of Valuable Products

Isabel Bator <sup>1</sup> , Andreas Wittgens 2,3,4, Frank Rosenau2,3,4, Till Tiso<sup>1</sup> and Lars M. Blank <sup>1</sup> \*

1 iAMB - Institute of Applied Microbiology, ABBt – Aachen Biology and Biotechnology, RWTH Aachen University, Aachen, Germany, <sup>2</sup> Institute for Pharmaceutical Biotechnology, Ulm-University, Ulm, Germany, <sup>3</sup> Ulm Center for Peptide Pharmaceuticals, Ulm, Germany, <sup>4</sup> Max-Planck-Institute for Polymer Research Mainz, Synthesis of Macromolecules, Mainz, Germany

#### *Edited by:*

Jean Marie François, UMS3582 Toulouse White Biotechnology (TWB), France

#### *Reviewed by:*

Antoine Danchin, INSERM U1016 Institut Cochin, France Mingfeng Cao, University of Illinois at Urbana-Champaign, United States

> *\*Correspondence:* Lars M. Blank lars.blank@rwth-aachen.de

#### *Specialty section:*

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

*Received:* 04 September 2019 *Accepted:* 23 December 2019 *Published:* 17 January 2020

#### *Citation:*

Bator I, Wittgens A, Rosenau F, Tiso T and Blank LM (2020) Comparison of Three Xylose Pathways in Pseudomonas putida KT2440 for the Synthesis of Valuable Products. Front. Bioeng. Biotechnol. 7:480. doi: 10.3389/fbioe.2019.00480

Pseudomonas putida KT2440 is a well-established chassis in industrial biotechnology. To increase the substrate spectrum, we implemented three alternative xylose utilization pathways, namely the Isomerase, Weimberg, and Dahms pathways. The synthetic operons contain genes from Escherichia coli and Pseudomonas taiwanensis. For isolating the Dahms pathway in P. putida KT2440 two genes (PP\_2836 and PP\_4283), encoding an endogenous enzyme of the Weimberg pathway and a regulator for glycolaldehyde degradation, were deleted. Before and after adaptive laboratory evolution, these strains were characterized in terms of growth and synthesis of mono-rhamnolipids and pyocyanin. The engineered strain using the Weimberg pathway reached the highest maximal growth rate of 0.30 h−<sup>1</sup> . After adaptive laboratory evolution the lag phase was reduced significantly. The highest titers of 720 mg L−<sup>1</sup> mono-rhamnolipids and 30 mg L−<sup>1</sup> pyocyanin were reached by the evolved strain using the Weimberg or an engineered strain using the Isomerase pathway, respectively. The different stoichiometries of the three xylose utilization pathways may allow engineering of tailored chassis for valuable bioproduct synthesis.

Keywords: *Pseudomonas putida*, xylose, metabolic engineering, rhamnolipid, phenazine, pyocyanin, flux balance analysis, heterologous production

# INTRODUCTION

For the establishment of a circular bioeconomy, the chemical industry has to overcome the massive and ever-increasing use of fossil resources and the concomitant production of environmental pollutions including greenhouse gases. The alternative is CO<sup>2</sup> as carbon source, either directly or fixed via chemocatalysis or plants (Olah et al., 2009; Goeppert et al., 2012). CO2-fixation though cannot only proceed under natural conditions, there are also many synthetic approaches to convert CO<sup>2</sup> to valuable products. The conversion of CO<sup>2</sup> mainly results in the production of biomass in microbes and is well-studied. Hence, it immediately becomes clear that the knowledge about CO2-fixation pathways can be utilized to redirect the metabolic flow into the production of chemicals using the synthetic biology arsenal. On the one hand cell-free systems are used to fix CO<sup>2</sup> via multienzyme cascades (Schwander et al., 2016; Satagopan et al., 2017), on the other hand engineered autotrophic microbes are used for the production of chemicals based on CO<sup>2</sup> (Lan and Liao, 2011; Angermayr et al., 2012). If the usage of non-autotrophic microbes is desired for the production of multi-carbon compounds via CO2-fixation, either existing pathways (e.g., the Calvin cycle) or newly designed pathways have to be introduced into the cells (Guadalupe-Medina et al., 2013; Antonovsky et al., 2016; Bouzon et al., 2017; Schada von Borzyskowski et al., 2018), with the outcome of alternative stoichiometries (Liebal et al., 2018). CO2-fixation via plants is a common choice as lignocellulosic biomass is abundant and does not compete immediately with food applications. Lignocellulosic biomass contains about 25–50% hemicellulose (Saha, 2003). In addition to glucose, the pentose xylose is a dominant building block in hemicellulose. For example, bagasse fibers consist of ∼23% xylose (Lee, 1997). While wheat straw and corn stover are industrially used as carbon source for bioethanol production (Larsen et al., 2008; Maas et al., 2008; Zhao et al., 2018), exemplifying the feasibility of the alternatives, most other microbes than yeast are still on the experimental scale.

The well-characterized chassis Pseudomonas putida KT2440 is known for its versatile metabolism, the richness of cofactors, and high solvent tolerance (Ramos et al., 1995; Nelson et al., 2002; Blank et al., 2010; Tiso et al., 2014). Further, the genome is sequenced and well-annotated (Nelson et al., 2002; Belda et al., 2016; Winsor et al., 2016; Nogales et al., 2019). The structure of the central carbon metabolism and the fluxes on different carbon sources are well-studied for P. putida KT2440 (Sudarsan et al., 2014). Further, many tools for genomic engineering are available (Martínez-García and de Lorenzo, 2011; Poblete-Castro et al., 2012; Nikel and de Lorenzo, 2018). Besides, its non-pathogenicity contributes to it being a chassis relevant for production of valuable products (Nelson et al., 2002; Tiso et al., 2014; Loeschcke and Thies, 2015; Wittgens and Rosenau, 2018).

However, P. putida KT2440 lacks the capability to metabolize xylose. Microbial assimilation of xylose has so far been observed by three different metabolic pathways: An isomerase and two oxidative pathways called Weimberg and Dahms. The Isomerase pathway is characterized by the presence of a xylose isomerase and xylulokinase. The resulting intermediate xylulose is subsequently phosphorylated to xylulose-5-phosphate and introduced into the pentose phosphate pathway. The Isomerase pathway is originally found in prokaryotes, such as Escherichia coli and Bacillus subtilis (David and Weismeyer, 1970; Wilhelm and Hollenberg, 1985). The oxidative pathways start with the oxidation of xylose to xylonate. This step is often catalyzed by a periplasmic dehydrogenase. A xylonate dehydratase then converts xylonate into 2-keto-3-deoxy-xylonate. The resulting intermediate is dehydrated to α-ketoglutaric semialdehyde in case of the Weimberg pathway. In case of the Dahms pathway, α-ketoglutaric semialdehyde is split by an aldolase into pyruvate and glycolaldehyde. While pyruvate is directly converted to acetyl-CoA in the central carbon metabolism, glycolaldehyde is further metabolized in several steps to 2-phosphoglycerate (Franden et al., 2018). 2-Phosphoglycerate is then converted in glycolysis. In the Weimberg pathway, an additional oxidation step from α-ketoglutaric semialdehyde to 2-oxoglutarate is present. The latter metabolite is converted in the tricarboxylic acid (TCA) cycle (**Figure 1**). The Weimberg pathway was reported for example to be present in P. fragi, Haloferax volcanii, and Caulobacter crescentus (Weimberg, 1961; Stephens et al., 2007; Johnsen et al., 2009). The genes of the latter species were already heterologously expressed in different organisms to establish the Weimberg pathway (Meijnen et al., 2009; Radek et al., 2014; Rossoni et al., 2018). Recently, also P. taiwanensis VLB120 was found to be a native xylose-consumer using the Weimberg pathway (Köhler et al., 2015). The Dahms pathway was first found to be present in an unclassified Pseudomonas strain (Dahms, 1974). Some of the genes of C. crescentus were also heterologously expressed to establish the Dahms pathway in E. coli, because E. coli harbors a gene coding for an aldolase, which catalyzes the last step of the Dahms pathway (Choi et al., 2017; Cabulong et al., 2018).

P. putida KT2440 has already been engineered to utilize xylose via the Isomerase pathway (Le Meur et al., 2012; Dvorák and de Lorenzo, 2018). Furthermore, P. putida S12 was equipped successfully with xylose metabolism genes to establish the Isomerase and the Weimberg pathway (Meijnen et al., 2008, 2009). In the studies mentioned above, growth via the Isomerase pathway was rather slow. This phenomenon occurred because most of the substrate was oxidized to the dead-end product xylonate. Further, xylose transport was reported to be inefficient in P. putida KT2440. By implementing adaptive laboratory evolution (ALE) growth on xylose could be improved (Meijnen et al., 2008). Using a different approach, Dvorák and de Lorenzo additionally inserted the proton-coupled symporter XylE, also achieving improved growth (Dvorák and de Lorenzo, 2018).

So far, metabolic engineering approaches were limited to metabolic alterations in order to synthesize a target compound based on a previously determined substrate. We here propose to exploit the metabolic potential of microbes for the production of chemicals by electing for the substrate of choice the most suited degradation pathway. Pseudomonads are suitable hosts for the heterologous production of several products, e.g., surfactants, aromatics, terpenoids, and phenazines besides others (Wittgens et al., 2011, 2018; Loeschcke and Thies, 2015; Schmitz et al., 2015b; Tiso et al., 2017; Wynands et al., 2018). However, the usage of xylose for the synthesis of valuable chemicals with P. putida KT2440 has not been reported so far.

In this work, we aimed at engineering a chassis, which can be used for the production of desired secondary metabolites using xylose. First, we investigated the theoretical potential for the synthesis of valuable products using the different xylose pathways via flux balance analysis (FBA). Second, we implemented the three different xylose pathways into P. putida KT2440 and optimized growth by ALE. Third, we used these strains for the synthesis of mono-rhamnolipids and pyocyanin, a derivative of phenazines. We demonstrate efficient growth on xylose via the oxidative pathways. The results show that using xylose pathways having different stoichiometries leads to differences in substrate consumption, growth, and production rates, as well as in biomass and product yields. We propose to use different xylose degradation pathways in dependence of product needs.

# MATERIALS AND METHODS

## Bacterial Strains, Media, and Growth Conditions

For strain maintenance and cloning experiments, the strains P. putida KT2440 (DSM6125, ATCC47054), E. coli DH5α (New England Biolabs, Ipswich, MA, USA), and E. coli PIR2 (ThermoFisher Scientific, Waltham, MA, USA) were routinely cultivated in LB medium containing 10 g L−<sup>1</sup> peptone, 5 g L−<sup>1</sup> yeast extract, and 10 g L−<sup>1</sup> NaCl. P. putida was cultivated at 30◦C and E. coli at 37◦C. If required, 50 µg mL−<sup>1</sup> kanamycin or 30 µg mL−<sup>1</sup> gentamycin were added to the medium to avoid loss of plasmid. After mating procedures, P. putida strains were selected on cetrimide agar (Sigma-Aldrich, St. Louis, MO, USA). Growth and production experiments were performed using M9 minimal medium with a final composition (per L) of 8.5 g Na2HPO4x2H2O, 3 g KH2PO4, 0.5 g NaCl, 1 g NH4Cl, 2 mM MgSO4, 4.87 mg FeSO4x7H2O, 4.12 mg CaCl2x2H2O, 1.5 mg MnCl2x4H2O, 1.87 mg ZnSO4x7H2O, 0.3 mg H3BO3, 0.25 mg Na2MoO4x2H2O, 0.15 mg CuCl2x2H2O, 0.84 mg Na2EDTAx2H2O (Sambrook and Russell, 2001), and 10 g glucose for pre-cultures or 10 g xylose for main cultures. Growth experiments were performed in 500 mL shake flasks with 10% filling volume at 200 rpm and in 24-deep well-plates (SystemDuetz; Enzyscreen B.V., Heemstede, The Netherlands) with 1 mL filling volume at 300 rpm.

#### <sup>13</sup>C-Labeling Experiments

Strains for the isotope labeling experiments were grown under the conditions stated above. The medium contained 50% of 1-13C labeled xylose (99% purity, Sigma-Aldrich, St. Louis, MO, USA). When the culture reached pseudo-steady state, samples with defined amounts of biomass (0.3 mg) were taken, hydrolyzed, and derivatized according to Schmitz et al. (2015a). The MS data were processed with iMS2Flux (Poskar et al., 2012), which enables the correction of the data for the presence of naturally occurring isotopes and the determination of the fractional labeling of selected amino acids. The fractional labeling indicates the fraction of <sup>13</sup>C-labeled carbon atoms.

#### Plasmid and Strain Construction

The genes for xylose utilization were amplified from genomic DNA of E. coli DH5α and P. taiwanensis VLB120 (isolated with High Pure PCR Template Preparation Kit, Roche Holding, Basel, Switzerland). Plasmid construction of pBT-Isomerase, pBT-Weimberg, and pBT-Dahms was planned with NEBuilder Assembly online tool and performed with NEBuilder HiFi DNA Assembly (New England Biolabs, Ipswich, MA, USA) (Gibson et al., 2009). The expression vector pBT including the constitutive Ptac promoter (Koopman et al., 2010) was linearized with XbaI (New England Biolabs, Ipswich, MA, USA) prior to assembly. DNA fragments for Gibson Assembly were amplified using Q5 High-Fidelity DNA Polymerase (New England Biolabs, Ipswich, MA, USA) according to the manual. All used primers with their nucleotide sequences are listed in the supplemental information (**Supplementary Table 1**). For pBT-Isomerase, the operon xylAB (b3565 and b3564 from E. coli DH5α) was amplified using primers IB-27 and IB-28 and cloned into the linearized pBT vector by Gibson Assembly. To generate pBT-Weimberg, an operon from P. taiwanensis VLB120 was used. The first two genes (PVLB\_18555 and PVLB\_18550), originally on antisense strand, were amplified using primers IB-63 and IB-64 and the following two genes (PVLB\_18560 and PVLB\_18565) were amplified using primers IB-65 and IB-66. Afterwards, the linearized pBT vector and both amplified fragments were assembled in one reaction. After that, pBT-Weimberg was used as template for pBT-Dahms. The backbone including PVLB\_18555 and PVLB\_18550 was amplified from pBT-Weimberg with primers IB-67 and IB-74. Amplification of gene PVLB\_18565 from P. taiwanensis VLB120 was performed using primers IB-68 and IB-124 and the gene for the aldolase yagE (b0268 from E. coli DH5α) was amplified using the primers IB-118 and IB-125 to create a synthetic operon. All fragments were assembled via Gibson reactions. In a final step to obtain pBT-Dahms, the putative dehydratase PVLB\_18550 was deleted from the assembled vector. Therefore, the vector was amplified with primers IB-93 and IB-94. Subsequently, the amplified vector was self-ligated using T4 Polynucleotide Kinase, T4 DNA Ligase, and T4 DNA Ligase Reaction Buffer (all New England Biolabs GmbH, Ipswich, MA, USA) according to the manual.

The resulting plasmids, pBT-Isomerase, pBT-Weimberg, and pBT-Dahms, were transferred individually into chemically competent E. coli PIR2 cells using heat shock according to Hanahan (1983). Transformants were selected on LB agar with kanamycin. Positive colonies were verified via colony PCR using OneTaq 2x Master Mix with Standard Buffer (New England Biolabs, Ipswich, MA, USA). The efficiency of colony PCR was increased by lysis of cell material with alkaline polyethylene glycol according to Chomczynski and Rymaszewski (2006). Plasmids were validated by Sanger sequencing performed by Eurofins (Brussels, Belgium). Subsequently, plasmids were isolated with Monarch Plasmid Miniprep Kit (New England Biolabs, Ipswich, MA, USA) and transferred via electroporation in competent P. putida cells according to Choi et al. (2006). Electroporation was performed using a GenePulser Xcell (Biorad, Hercules, CA, USA) using a cuvette with a 2 mm gap and the settings 2.5 kV, 200 , and 25 µF. Plasmids pBT-Isomerase and pBT-Weimberg were transferred individually in P. putida KT2440 and P. putida KT2440 1gcd. Plasmid pBT-Dahms was transferred individually in P. putida KT2440, P. putida KT2440 1PP\_2836, and P. putida KT2440 1gclR 1PP\_2836. Harboring of the plasmids was verified via colony PCR using primers IB-5 and IB-35 (as above).

Deletion mutants were obtained using the I-SceI-based system developed by Martínez-García and de Lorenzo (2011). Briefly, 700 bp upstream and downstream flanking regions of the target sites, named TS1 and TS2 regions, were amplified from the genomic DNA of P. putida KT2440 and cloned into the nonreplicative pEMG vector by Gibson Assembly. The resulting plasmids, pEMG-gcd and pEMG-PP\_2836, were transferred individually into chemically competent E. coli PIR2 cells using heat shock as described above. Afterwards, the plasmids were verified by Sanger sequencing and transferred from the E. coli PIR2 strains into the required Pseudomonasstrains via triparental mating according to Wynands et al. (2018). The I-SceI-encoding plasmid pSW-2 was used and no 3-methylbenzoate for induction of I-SceI expression was needed according to Wynands et al. (2018). Positive colonies with loss of kanamycin resistance were verified for targeted deletion by colony PCR using OneTaq 2x Master Mix with Standard Buffer. To obtain marker-free and pure clones, the recombinant strains were cured of pSW-2 plasmid by re-inoculation in LB medium without gentamycin and verified again by colony PCR. In this study, the gene for glucose dehydrogenase (PP\_1444) was deleted in the wild type, resulting in P. putida KT2440 1gcd. The gene for the putative 2-keto-3-deoxy-xylonate dehydratase (PP\_2836) was deleted in the wild type and P. putida KT2440 1gclR (Li et al., 2019), resulting in P. putida KT2440 1PP\_2836 and P. putida KT2440 1gclR 1PP\_2836.

To demonstrate the production of secondary metabolites, two target molecules, rhamnolipids and phenazines, were chosen. For the production of mono-rhamnolipids, the mini-Tn7 delivery transposon vector (pBG) developed by Zobel et al. (2015), which inserts a synthesis module in a single genomic locus of the chromosome, was used. Therefore, plasmid pBG14ffg with a stronger constitutive, synthetic promoter than pBG14g (Sebastian Köbbing, RWTH Aachen University, personal communication) was used as backbone and amplified with primers SK06 and SK07. The genes enabling rhamnolipid production rhlAB from P. aeruginosa were amplified with primers SK08 and SK09 and cloned into the amplified backbone by Gibson Assembly. The resulting mini-Tn7 vector pSK02 was integrated into the genome of xylose consuming strains via homologous recombination according to Zobel et al. (2015). The integration of the rhamnolipid synthesis module into the attTn7 site was verified by colony PCR as described above. Mono-rhamnolipid producing clones were identified using cetrimide-blood agar plates (7.5% (v/v) sheep blood; Fiebig-Naehrstofftechnik, Idstein-Niederauroff, Germany). If rhamnolipid synthesis occurred, a halo around the colony was visible, because of their hemolytic activity. The 10 clones showing the largest halos were selected for experiments in minimal medium with 10 g L−<sup>1</sup> xylose to quantitatively examine rhamnolipid production. Subsequently, the best three producers of each strain were subjected to replicate experiments and out of these the best was chosen for further characterization and is described in this study.

Phenazine production was achieved by plasmid-based expression. Plasmid pJNN\_phzA1-G1,M,S (modified from Schmitz et al., 2015b) was transferred into competent xylose consuming strains via electroporation according to Choi et al. (2006). Harboring of the plasmid was confirmed via colony PCR as described above. Based on Schmitz et al. (2015b), five clones were used to inoculate 1 mL minimal medium containing antibiotics and 10 g L−<sup>1</sup> xylose at a starting OD<sup>600</sup> of 0.1 and induced with 0.1 mM salicylate to induce phenazine synthesis. If phenazine synthesis occurred, a blue color, which is typical for pyocyanin production, was observed. The best three producers of each strain were subjected to replicate experiments and out of these the best was chosen for further characterization and is described in this study.

All used strains in this study are listed in **Table 1**.

# Adaptive Laboratory Evolution

For adaptation to xylose, the different xylose-consuming strains were grown in M9 minimal medium containing 10 g L−<sup>1</sup> xylose. OD<sup>600</sup> was measured daily and the cells were sequentially transferred to fresh medium with a starting OD<sup>600</sup> of 0.1. The sub-culturing was carried out 30 times for the isomerase-strain, 17 times for the Weimberg-strain, and 26 times for the Dahmsstrain. The inhomogeneous culture was streaked out on LBagar to obtain single isolates, which were subsequently tested for adaption to xylose in a 96-well-plate in a Growth Profiler 960 (Enzyscreen B.V., Heemstede, The Netherlands).

#### Analytical Methods

#### Analysis of Bacterial Growth

The optical density at 600 nm (OD600) was measured using an Ultrospec 10 cell density meter (Amersham Biosciences, UK). A correlation between OD<sup>600</sup> and cell dry weight (CDW) was established. An OD<sup>600</sup> of 1.0 corresponds with a cell dry weight of 369 mg L−<sup>1</sup> .

#### Analysis of Xylose and Xylonate

Xylose and xylonate concentrations in the supernatant were analyzed in a Beckmann Coulter System Gold High Performance Liquid Chromatography (HPLC) (Beckmann Coulter, Brea, CA, USA) with a Metab-AAC 300 × 7.8 mm separation column (particle size: 10µm, ISERA GmbH, Düren, Germany), a UV detector 166 (Beckmann Coulter, Brea, CA, USA) at 210 nm and a refractory index detector (RI 2300, Knauer GmbH, Berlin, Germany). Elution was performed with 5 mM H2SO<sup>4</sup> at a flow rate of 0.5 ml min−<sup>1</sup> at 40◦C.

TABLE 1 | Bacterial strains used in this study.


(Continued)

#### TABLE 1 | Continued


#### Analysis of Rhamnolipids

Reversed-phase chromatography was performed for analyzing mono-rhamnolipid concentrations based on a method developed earlier (Behrens et al., 2016; Tiso et al., 2016). For sample preparation, the supernatant was mixed 1:1 with acetonitrile and stored at 4◦C overnight. Subsequently, the mixture was centrifuged at 16,500 g for 2 min. All samples were filtered with Phenex RC syringe filters (0.2µm, Ø 4 mm, Phenomenex, Torrance, USA). The HPLC system Ultimate 3000 with a Corona Veo Charged Aerosol Detector (Thermo Fisher Scientific, Waltham, MA, USA) was used. For separation, a NUCLEODUR C18 Gravity 150 × 4.6 mm column (particle size: 3 mm, Macherey-Nagel GmbH & Co. KG, Düren, Germany) was used. The flow rate was set to 1 ml min−<sup>1</sup> and the column oven temperature was set to 40◦C. Acetonitrile (A) and 0.2% (v/v) formic acid in ultra-pure water (B) were used as running buffers. The method started with a ratio of 70% buffer A: 30% buffer B and a linear gradient was applied to reach a ratio of 80:20% in 8 min. The acetonitrile fraction was increased linearly from 80 to 100% between 9 and 10 min and decreased linearly to 70% between 11 and 12.5 min. The measurement was stopped after 15 min.

#### Analysis of Phenazines

The concentration of phenazines was determined in a spectrophotometer Cary 60 (Agilent, Santa Clara, CA, USA) by analyzing the blue phenazine derivative pyocyanin. Samples were centrifuged at 16,500 g for 2 min. Afterwards, the supernatant was vortexed and pyocyanin was measured at 691 nm. For quantification, Lambert-Beer's law with an extinction coefficient of 4.31 mM−<sup>1</sup> cm−<sup>1</sup> for pyocyanin was used (Filloux and Ramos, 2014).

#### Analysis of Proteinogenic Amino Acids

<sup>13</sup>C-labeling pattern of proteinogenic amino acids were determined by gas chromatography coupled with mass spectrometry (GC-MS) as described in Schmitz et al. (2015a). A Trace GC Ultra coupled to an ISQ single quadrupole mass spectrometer with a TriPlus RSH autosampler (all Thermo Fisher Scientific, Waltham, MA, USA) was used. For separation, a TraceGOLD TG-5SilMS capillary column (length: 30 mm, inner diameter: 0.25 mm, film thickness: 0.25µm, Thermo Fisher Scientific, Waltham, MA, USA) was used. A sample volume of 2 µL was injected at 270◦C with a split ratio of 1:50. The flow rate of the carrier gas helium was set to 1 mL min−<sup>1</sup> and the oven temperature was kept constant for 1 min at 140◦C. The temperature was increased with a gradient of 10◦C min−<sup>1</sup> to 310◦C and then kept constant for 1 min. The temperature of the transfer line and the ion source were both set to 280◦C. Ionization was performed by electron impact ionization at −70 eV. All raw data were analyzed using Xcalibur (Thermo Fisher Scientific, Waltham, MA, USA).

#### Flux Balance Analysis

For the prediction of product yields, the genome-scale model of P. putida KT2440, iJN1411, was used (Nogales et al., 2017). All simulations were carried out in MATLAB (version R2017b, the Mathworks, Inc., Natick, MA, USA) using the COBRA toolbox (Schellenberger et al., 2011), with the linear programming solver of Gurobi<sup>1</sup> . First, the existing model network was extended by the xylose utilization pathways as well as the biosynthesis routes to the desired products, e.g., mono-rhamnolipids, pyocyanin, and ethylene glycol, using reaction information from KEGG (Ogata et al., 1999). Subsequently, the evaluation of the respective product yields for the usage of the three xylose utilization pathways and their combination was performed. Here, the uptake of xylose for all investigated cases was set to 10 mmol g−<sup>1</sup> CDW h −1 and the uptake of other carbon sources (e.g., glucose) was set to zero. The production of the target molecule was used as objective function in the extended model.

#### RESULTS

#### Alternative Xylose Utilization Pathways for Maximal Product Yield

With FBA one can compute the maximal product yield in different scenarios (e.g., growth, maintenance coefficients, aeration). Here, the genome-scale model P. putida KT2440 iJN1411 (Nogales et al., 2017) was modified by introducing the alternative xylose utilization pathways. In addition, various synthesis reactions for valuable products according to Werpy and Petersen (2004) and chemicals we are working on were introduced into iJN1411.

The maximal product yields of 12 of the 14 metabolites computed on xylose were produced by the Isomerase pathway (**Table 2**). The exceptions of maximal yields of metabolites are such products that are directly synthesized from intermediates of the Weimberg and Dahms pathway. For example, the intermediate 2-oxoglutarate of the Weimberg pathway is an interesting metabolite, because it is introduced into the TCA cycle without carbon loss. The direct conversion of 2-oxoglutarate into glutamate (or proline, OH-proline, etc.) would thus benefit from the Weimberg pathway. This was also computed by FBA with a maximal product yield of 1 mmol mmol−<sup>1</sup> for the Weimberg pathway, and only 0.83 and 0.75 mmol mmol−<sup>1</sup> for the Isomerase and the Dahms pathway, respectively. The intermediate glycolaldehyde, using the Dahms pathway for xylose degradation, itself is a relevant chemical for industrial applications and a precursor for ethylene glycol synthesis. Ethylene glycol can neither be synthesized via the Isomerase nor the Weimberg pathway, while the maximal product yield of the Dahms pathway is 1 mmol mmol−<sup>1</sup> . Consequently, the Dahms pathway is a favorable pathway to produce glycols out of xylose. In case of the Isomerase pathway, xylulose-5-phosphate is part of the pentose phosphate pathway and the intermediate erythrose-4-phosphate is a starting metabolite for aromatics synthesis via the shikimate pathway. This explains that the Isomerase pathway has the highest yield for the production of the here synthesized aromatic compound pyocyanin (0.27 mmol mmol−<sup>1</sup> ). Moreover, the highest maximal mono-rhamnolipid yield is also possible via the Isomerase pathway (0.14 mmol mmol−<sup>1</sup> ) and lower yields for the Weimberg and the Dahms pathway (0.08 and 0.12 mmol mmol−<sup>1</sup> ) were computed. This is due to CO<sup>2</sup> production using the TCA cycle TABLE 2 | FBA computed maximal product yield of valuable products for three different xylose utilization pathways (Isomerase, Weimberg, and Dahms) and the combination of the Weimberg (W) and the Dahms (D) pathway.


<sup>a</sup>Due to constraints in the model, the maximal yield for glycerol cannot be computed for the combination of the Weimberg and the Dahms pathway.

For the combined pathways, also the percentage of the used pathway is indicated. The bold values indicate the highest maximal product yield.

for the Weimberg pathway and converting glycolaldehyde in several steps to 2-phosphoglycerate for the Dahms pathway, also releasing one CO2. In addition, the maximal product yields for all possible combinations of the xylose pathways (Isomerase + Weimberg, Isomerase + Dahms, Weimberg + Dahms, and Isomerase + Weimberg + Dahms) were computed. In general, it was observed that always one of the pathways in the combinations is favored and is thus exclusively used. Interestingly, and contrary to the above, the combination of the Weimberg and the Dahms pathways resulted in higher maximal product yields for nine of the 14 metabolites (**Table 2**) compared to the single pathway strains. Thus, in **Table 2** only the combined activities of the latter pathways are also listed. In summary, the in silico study indicates that the strain equipped with the Isomerase pathway is best-suited for a range of chemicals including the production of the here chosen metabolites mono-rhamnolipids and pyocyanin. The Weimberg and Dahms pathways are favored in niche applications (**Table 2**), when xylose is the sole carbon and energy source.

#### The Oxidative Pathways Enable Efficient Xylose Utilization

A genome analysis using BLAST showed that P. putida KT2440 possesses genes (Altschul et al., 1990), which might encode enzymes needed in xylose utilization, but a complete pathway seems to be absent, indeed correlating with no-growth on this sugar (**Figure 2A**). All genes required for the Isomerase,

<sup>1</sup>www.gurobi.com

Weimberg, and Dahms pathways were compared to the genome of P. putida KT2440. No homologs for the genes xylA and xylB from E. coli for the Isomerase pathway are present. Additionally, no homologs for the genes PVLB\_18565 from P. taiwanensis and yagE from E. coli for the Weimberg pathway and the Dahms pathway, respectively, are present. In contrast, homologous genes of the Weimberg pathway from P. taiwanensis are present (homologs to PVLB\_18550, PVLB\_18555, and PVLB\_18560 with sequence identities of 68–72%). HPLC analysis revealed that xylose was consumed by non-growing P. putida KT2440, however xylose was converted into the dead-end product xylonate (**Figure 2A**). Xylonate synthesis proceeds likely via the gene products of gcd (PP\_1444) and gnl (PP\_1170) that

are besides for glucose also active for xylose and its lactone derivative (**Figure 1**).

In order to engineer P. putida for the utilization of xylose, P. putida was equipped with a plasmid for the Isomerase pathway and designated P. putida KT2440 pIso. After inoculation, strain P. putida KT2440 pIso showed little growth (**Figure 2B**) with a maximal growth rate of 0.02 h−<sup>1</sup> and a substrate consumption rate of 0.47 mmol L−<sup>1</sup> h −1 (**Table 3**). In addition, pBT-Isomerase was introduced into P. putida KT2440 1gcd. This strain is unable to produce the dead-end product xylonate and thus more substrate should be available for biomass formation, as it was reported previously in a study with P. putida S12 (Meijnen et al., 2008). The resulting P. putida KT2440 1gcd pIso showed no

TABLE 3 | Growth characteristics of wild type, recombinant, and evolved P. putida KT2440 strains on xylose.


<sup>a</sup>The growth rate was determined during the exponential phase.

<sup>b</sup>The substrate consumption rate was determined until xylose is depleted. For the wild type, the whole cultivation time was used as reference.

<sup>c</sup>For the wild type, the substrate consumption rate is considered as substrate conversion rate, because it is not able to grow on xylose.

Values are the arithmetic mean of biological triplicates. The deviation indicates the standard deviation from the mean (n = 3).

improved growth compared to P. putida KT2440 pIso (data not shown). Nevertheless, these results suggest that the cloned genes xylAB from E. coli are functionally expressed in P. putida KT2440 and allow slow growth of the recombinant strain.

In order to establish the oxidative Weimberg pathway in P. putida KT2440, the plasmid pBT-Weimberg was transferred into P. putida KT2440 resulting in P. putida KT2440 pWeim. Interestingly, P. putida KT2440 pWeim showed efficient growth with a maximal growth rate of 0.30 h−<sup>1</sup> and a strong accumulation of xylonate (**Figure 2C**). This led to a long lag phase of ∼24 h (**Table 3**). We define the lag phase as a phase before growth was observed due to homogenous adaptation. The growth of a small phenotypic subpopulation on the new carbon source due to heterogenous adaptation described by Kotte et al. (2014) is not considered. The observed lag phase indicates a bottleneck in the conversion of xylonate to 2-keto-3-deoxy-xylonate or in the transport from the periplasm to the cytoplasm. Further, the substrate consumption rate was about 4-fold higher (1.94 mmol L−<sup>1</sup> h −1 ) in comparison to P. putida KT2440 pIso (**Table 3**). These results show that the xylose operon from P. taiwanensis is functionally expressed in P. putida KT2440 and results in efficient growth. To verify that the glucose dehydrogenase (Gcd) features a side-activity for xylose in P. putida KT2440, the plasmid pBT-Weimberg was introduced into P. putida KT2440 1gcd yielding P. putida KT2440 1gcd pWeim. After inoculation, no growth was observed (data not shown). This behavior confirms the assumption that Gcd is active for xylose because xylose needs to be converted to xylonate to establish the Weimberg pathway (**Figure 1**).

Moreover, the expression of genes encoding the Dahms pathway was targeted. Therefore, the plasmid for the Dahms pathway was introduced into wild type yielding P. putida KT2440 pDahms. As described before, genome analysis revealed that a homologous gene of PVLB\_18560 is present (PP\_2836). By comparison with genes expressing xylose-metabolizing enzymes from C. crescentus, PVLB\_18560 is assumed to encode a dehydratase, which converts 2-keto-3-deoxy-xylonate to αketoglutaric semialdehyde (**Figure 1**). As a result, it is assumed that the product of the homologous gene PP\_2836 in P. putida KT2440 is likely to catalyze this reaction. To verify this hypothesis, the plasmid pBT-Dahms was also transferred into P. putida KT2440 1PP\_2836 yielding P. putida KT2440 1PP\_2836 pDahms. P. putida KT2440 pDahms reached an OD<sup>600</sup> of 8.2 after 74 h, while P. putida KT2440 1PP\_2836 pDahms stopped growing after 48 h (final OD<sup>600</sup> of 2.0). The latter strain showed cell clumping after 74 h, which implies stress. We speculated that low optical density and cell-clumping are caused by an accumulation of the toxic intermediate glycolaldehyde. These findings strongly indicate that PP\_2836 encodes a 2-keto-3-deoxy-xylonate dehydratase. In order to overcome the toxic effect of glycolaldehyde, strain P. putida KT2440 1gclR was used. The glyoxylate carboligase oxidase pathway is active in this mutant, thus avoiding glycolaldehyde accumulation (Li et al., 2019). The plasmid pBT-Dahms was introduced into P. putida KT2440 1gclR 1PP\_2836 resulting in P. putida KT244011 pDahms. In comparison to P. putida KT2440 1PP\_2836 pDahms, strain P. putida KT244011 pDahms indeed grew to higher cell densities, which led to the conclusion that the genes encoding the enzymes for the Dahms pathway are successfully expressed and the prevention of glycolaldehyde accumulation leads to efficient growth. The growth of P. putida KT244011 pDahms was further examined and a maximal growth rate of 0.21 h−<sup>1</sup> and a substrate consumption rate of 1.46 mmol L−<sup>1</sup> h <sup>−</sup><sup>1</sup> was observed (**Table 3**). Further, accumulation of xylonate occurred as in the case of the Weimberg strain. Thereby, a lag phase of ∼34 h results (**Figure 2D**).

As stated above, it was observed that not only a long lag phase occurs when the Isomerase pathway is used, but also when the oxidative pathways are used. Regulatory mechanisms are assumed as a cause, but in the case of the oxidative pathways, the used route for metabolization is different. There, the xylose gets oxidized to xylonate via xylono-1,4-lactone by the enzymes encoded by gcd and gnl. It is assumed that these enzymes are active for xylose, but with a lower specificity and thus, the conversion is slower than for glucose and a long lag phase is the result. Further, accumulation of xylonate was observed, which indicates a bottleneck in xylonate conversion or transport.

## Adaptive Laboratory Evolution Improves Growth on Xylose

It was shown that all xylose metabolism plasmids are functionally expressed resulting in weak growth for the Isomerase pathway and efficient growth for the Weimberg and Dahms pathways, respectively. However, the lag phase of all strains was relatively long. To optimize growth, ALE was applied. For each of the strains, P. putida KT2440 pIso, P. putida KT2440 pWeim, and P. putida KT244011 pDahms, two parallel cultures were sequentially transferred to fresh minimal medium with xylose as the sole carbon source (**Figure 3A**). The performance of P. putida KT2440 pWeim and P. putida KT244011 pDahms was improved after 17 and 26 transfers corresponding to ∼115 and 160 generations, respectively. The number of generations was calculated based on the starting OD<sup>600</sup> and the end OD<sup>600</sup> of every transfer. Afterwards, 24 isolates were obtained from the culture broth and up to six from each culture were randomly selected and tested for growth on xylose in comparison to the initial strains. In **Figure 3B** five isolates in comparison to the initial strain P. putida KT2440 pWeim are shown. The adaptation toward xylose is not similar for all isolates, as some isolates still had a longer lag phase, suggesting that an isolation step to obtain the best growing strain is required. Hereafter, the best isolates were designated as P. putida KT2440 pWeim2 and P. putida KT244011 pDahms2. The adaptation of P. putida KT2440 pIso to xylose was problematic because the strain did not grow anymore after several transfers.

The colonies showing the best performance (P. putida KT2440 pWeim2 and P. putida KT244011 pDahms2) were characterized in detail regarding their growth behavior. Both showed improved growth compared to the initial strains. The lag phase of P. putida KT2440 pWeim2 was shortened from ∼24 to 3 h (**Figure 2E**). However, the maximal growth rate of the strain was 0.21 h−<sup>1</sup> . Hence, the lag phase was 1.5-fold reduced compared to the initial strain (**Table 3**). For P. putida KT244011 pDahms2, the lag phase was shortened from ∼34 to 3 h (**Figure 2F**). The growth rate remained the same with a value of 0.21 h−<sup>1</sup> (**Table 3**). Further, the substrate consumption rate was 3 mmol L−<sup>1</sup> h −1 for both evolved strains, which is about 1.6- and 2-fold higher than the consumption rates of their initial strains.

To identify possible mutations, which are responsible for the reduced lag phase and xylonate accumulation, the plasmids of P. putida KT2440 pWeim2 and P. putida KT244011 pDahms2 were isolated and sequenced. Mutations in the promoter, xylose genes, replication initiator protein, origin of replication, and antibiotic resistance of the plasmids carried by strains P. putida KT2440 pWeim2 and P. putida KT244011 pDahms2 could be excluded. Hence, mutations in coding regions in the genome are expected and will be analyzed in the future.

#### The Deletion of PP\_2836 Enables the Exclusive Usage of the Dahms Pathway

To demonstrate that the isolated Weimberg or Dahms pathway are present in the engineered strains, growth experiments with 1-13C labeled xylose as substrate were performed. Afterwards, the labeling patterns of the proteinogenic amino acids were analyzed. Since xylose is metabolized via different routes using the Weimberg or the Dahms pathway, the labeling pattern is welldistinguishable (**Figure 4**). Xylose is converted in several steps to 2-oxoglutarate via the Weimberg pathway, which is a direct precursor of glutamate, glutamine, and proline. This would lead to 1-13C labeled glutamate, glutamine and proline. In the central carbon metabolism, 2-oxoglutarate is further processed via the TCA cycle and the labeling is removed via CO<sup>2</sup> generation. Furthermore, in the case of the Dahms pathway, xylose is converted into pyruvate and glycolaldehyde. Since pyruvate is a precursor for the amino acids alanine, lysine, and valine, they are fractionally labeled. Pyruvate is also converted to acetyl-CoA in the central carbon metabolism, removing the labeled carbon atom by CO<sup>2</sup> generation.

For these experiments, P. putida KT2440 pWeim und P. putida KT244011 pDahms were used. As seen before, deletion of the gene PP\_2836 entails a completely different growth behavior, which indicates that the deletion of PP\_2836 is necessary to establish the Dahms pathway. It is assumed that xylose is utilized via both pathways simultaneously (Weimberg and Dahms) in a strain that still harbors PP\_2836 and is deficient in gclR (to activate degradation of the intermediate glycolaldehyde, see **Figure 1**). This assumption was experimentally tested by introducing the plasmid pBT-Dahms into P. putida 1gclR. The resulting P. putida 1gclR pDahms was also used in the labeling experiments.

Significant distinctions of the fractional labeling with labeled xylose were observed for all three strains (**Figure 5**). For the Weimberg-strain, alanine, lysine, and valine were almost not

bars indicate deviation from the mean (n = 3).

labeled at all while glutamate and proline showed a labeling about 0.1. This meets the expectations since 50% of 1-13C-labeled xylose was used (one atom out of ten carbon atoms from two xylose molecules should be labeled). For the Dahms-strain, the fractional labeling of the amino acids, which are derived from pyruvate (alanine, lysine, and valine), is about 0.13, 0.06, and 0.07, respectively, and for glutamate and proline below 0.05. Theoretically, a fractional labeling of about 0.17 for alanine, 0.08 for lysine, and 0.1 for valine would be assumed because one out of six, ten, and twelve carbon atoms from two xylose molecules should be labeled, respectively. The observed labeling is a little bit lower than expected, but the distribution meets the expectation since again 50% of 1-13C-labeled xylose was used. For more quantitative statements, metabolic flux analysis has to be performed in order to involve fluxes through the central carbon metabolism. Further, as described above the labeling is removed when CO<sup>2</sup> is generated. However, via anaplerotic reactions the labeled CO<sup>2</sup> can be reincorporated and ultimately result in labeled glutamate and proline. Nevertheless, both strains showed a distinct labeling pattern, which indicates that the Weimberg and Dahms pathways are truly used in isolation.

Moreover, for P. putida KT2440 1gclR pDahms it was determined that the fractional labeling of glutamate and proline was about 0.1 as in the case of the Weimberg-strain and that the labeling of alanine, lysine, and valine was in between the values of the Weimberg- and the Dahms-strain. These results show that the fractional labeling is a combination of the single

medium containing 50% 1-13C-xylose. Error bars indicate deviation from the mean of three technical replicates from one growth experiment (n = 3). Ala, alanine; Glx, glutamate and deaminated glutamine; Lys, lysine; Pro, proline; Val, valine.

fractional labelings of the Weimberg and the Dahms pathway. Thus, it is suggested that the Weimberg and the Dahms pathway are active simultaneously in this strain. Consequently, the above stated declaration that PP\_2836 encodes a dehydratase is verified. The deletion of this gene is necessary to obtain a strain, which solely uses the Dahms pathway.

# Variations in Xylose Metabolization Networks Lead to Specialized Microbial Cell Factories

Different stoichiometries of the pathways have an influence on the production yield of specific metabolites. In order to show this dependency, two secondary metabolites (rhamnolipids and phenazines) were chosen for heterologous synthesis in efficient xylose metabolizers. Exemplarily, strains with the individual pathways (Isomerase, Weimberg, and Dahms) were chosen and the production of the initial strains and the adapted strains was investigated. For the production of rhamnolipids, the rhamnolipid synthesis module was integrated as single copy into the genome of all xylose utilizing strains. P. putida KT2440 EM42 1gcd pSEVA2213\_xylABE (Dvorák and de Lorenzo, 2018) was used as optimized strain using the Isomerase pathway because no evolved P. putida strain using the Isomerase pathway was generated. The rhamnolipid producers were designated P. putida KT2440 pIso\_RL, P. putida KT2440 pWeim\_RL, P. putida KT244011 pDahms\_RL, P. putida KT2440 pIso2\_RL, P. putida KT2440 pWeim2\_RL, and P. putida KT244011 pDahms2\_RL.

P. putida KT2440 pIso\_RL, utilizing the Isomerase pathway for xylose degradation, showed only little rhamnolipid production (47 mg L−<sup>1</sup> ) after 96 h, because most of the xylose was converted to the dead-end product xylonate. While P. putida KT2440 pIso2\_RL produced seven times more monorhamnolipids after 54 h (**Figure 6A**), although the cells were clumped and did not reach high optical densities. On the contrary, the strains with the Weimberg and Dahms pathways reached higher rhamnolipid titers. P. putida KT2440 pWeim\_RL produced 13 times more mono-rhamnolipids than P. putida KT2440 pIso\_RL in less time (54 h) and the evolved strain with the rhamnolipid production module P. putida KT2440 pWeim2\_RL produced even more mono-rhamnolipids after 30 h (720 mg L−<sup>1</sup> ). Further, P. putida KT244011 pDahms\_RL and P. putida KT244011 pDahms2\_RL produced six times more mono-rhamnolipids after 54 h (290 mg L−<sup>1</sup> ) and 13 times more mono-rhamnolipids after 30 h (620 mg L−<sup>1</sup> ), respectively. Besides, differences in the substrate consumption were observed. While P. putida KT2440 pWeim2\_RL and P. putida KT244011 pDahms\_RL consumed all of the provided xylose, xylose and xylonate in the cultures of the other rhamnolipid-producing strains was detected after growth stopped (**Supplementary Figure 1**). Hence, the ratio of the amount of product synthesized and the amount of substrate consumed (yield) is considered. P. putida KT2440 pIso2\_RL had the highest yield (78 mg g−<sup>1</sup> ), while P. putida KT2440 pWeim2\_RL and P. putida KT244011 pDahms2\_RL reached comparable yields, however, 1.4-fold lower than P. putida KT2440 pIso2\_RL (**Figure 6C**). As expected, the evolved and optimized strains showed a higher rhamnolipid concentration and yield than their initial xylose utilizing strains (**Figures 6A,C**).

The synthesis of phenazines was achieved by transferring plasmid pJNN\_phzA1-G1,S,M into all initial xylose utilizing strains and the optimized or evolved strains. The pyocyanin producers were named P. putida KT2440 pIso\_PZ, P. putida KT2440 pWeim\_PZ, P. putida KT244011 pDahms\_PZ, P. putida KT2440 pIso2\_PZ, P. putida KT2440 pWeim2\_PZ, and P. putida KT244011 pDahms2\_PZ. All strains produced pyocyanin in a similar range (∼15–20 mg L−<sup>1</sup> , **Figure 6B**), but in a different period, except the strains using the Isomerase pathway. P. putida KT2440 pIso\_PZ produced around three times less (6 mg L−<sup>1</sup> ) pyocyanin and P. putida KT2440 pIso2\_PZ produced roughly two times more (30 mg L−<sup>1</sup> ) pyocyanin in comparison to the evolved strain using the Dahms pathway. Pyocyanin was harvested at the same time points as monorhamnolipids. Furthermore, as observed in rhamnolipid production experiments, differences in substrate consumption were recognized. P. putida KT2440 pIso\_PZ converted even more xylose to xylonate than in the rhamnolipid production experiments, which could then not be used for pyocyanin production. Further, less xylose was consumed than in the rhamnolipid production experiment (**Supplementary Figure 1**). Consequently, the yield was determined and P. putida KT2440 pIso2\_PZ showed the best yield (20 mg g−<sup>1</sup> ) (**Figure 6D**). Furthermore, the evolved strains reached higher yields than their initial strains (P. putida KT2440 pWeim2\_PZ: 3.2 mg g −1 , P. putida KT244011 pDahms2\_PZ: 2.5 mg g−<sup>1</sup> ). It was confirmed that the evolved and optimized strains could reach higher product titers and yields than their initial strains, also in case of pyocyanin production (**Figures 6B,D**). This can be explained by the adaptation toward the novel substrate xylose for P. putida KT2440 pWeim2 and P. putida KT244011 pDahms2. In case of P. putida KT2440 pIso2 the production of xylonate is prevented by deletion of the glucose dehydrogenase in comparison to P. putida KT2440 pIso and consequently more carbon is available for the production of pyocyanin. Conspicuously, growth was slower for the phenazine-producing strains compared to the rhamnolipid-producing strains and consequently the total consumption of xylose was lower. This might be due to the metabolic burden through the usage of kanamycin to avoid loss of the plasmid harboring the phenazine synthesis genes during cultivation. Further, product toxicity might be a cause for the xylose leftovers. In accordance to the computed data in 3.1., P. putida KT2440 pIso2 harboring the rhamnolipid or phenazine genes showed in both scenarios the best yield (**Figure 6**).

#### DISCUSSION

In this study, we integrated three bacterial xylose utilization pathways—Isomerase, Weimberg, and Dahms—in P. putida KT2440, to compare in silico and in vivo the synthesis capacities of these alternative degradation pathways. While other studies focus on heterologous production with P. putida KT2440, we aimed at showing the production of various products (i.e., monorhamnolipids and pyocyanin) from xylose. Further, we wanted to emphasize that computational analysis can guide strain design, reaching higher yields.

FBA indicates strong preferences for alternative pathways depending on the product of choice due to different stoichiometries. It was determined that mainly the Isomerase

pathway is preferred, however the Weimberg and Dahms pathways are favored in niche applications, when for example intermediates of the pathways are precursor for the products. To show this dependency, the xylose pathways and synthesis pathways were integrated in P. putida KT2440. In the course of isolating the Dahms pathway, two genes (PP\_2836 and PP\_4283) were deleted. After successful pathway implementation, the synthesis capacities for mono-rhamnolipids and pyocyanin were investigated. The resulting data matched the computed data. The approach of in silico metabolic network design driven by the product of choice, adds another degree of freedom for metabolic engineering.

The computed maximal product yield for both synthesized metabolites, mono-rhamnolipids and pyocyanin, was produced by the Isomerase pathway. In both cases, this was confirmed in vivo (**Figure 6**). While the highest rhamnolipid titer (700 mg L−<sup>1</sup> ) was reached by the evolved strain using the Weimberg pathway, the highest yield (78 mg g−<sup>1</sup> ) was reached by the strain using the Isomerase pathway. The highest pyocyanin titer (30 mg L−<sup>1</sup> ) and yield (20 mg g−<sup>1</sup> ) were also reached by the strain using the Isomerase pathway. Taken together, it was shown that the reached yields are dependent on the metabolization route and hence the stoichiometries of the pathway, which can be determined in silico.

Many studies deal with the heterologous production by P. putida KT2440. Production of biosurfactants, such as hydroxyalkanoyloxy alkanoates (HAA), mono-rhamnolipids, and di-rhamnolipids (Tiso et al., 2017; Wittgens et al., 2017, 2018), terpenoids (zeaxanthin and β-carotene) (Beuttler et al., 2011; Loeschcke et al., 2013), amino acid-derived compounds (e.g., phenazines) (Schmitz et al., 2015b), polyketides/nonribosomal peptides (e.g., flaviolin, prodigiosin) (Gross et al., 2006; Loeschcke et al., 2013; Domröse et al., 2015), and Nmethylglutamate (Mindt et al., 2018) was demonstrated. These examples show how diverse the production spectrum in P. putida KT2440 can be. To create sustainable production processes in times of high environmental pollution, one searches for alternative substrates (Vanholme et al., 2013). The concept of using xylose or other C<sup>5</sup> sugars for platform chemicals was already considered before (Werpy and Petersen, 2004). For example, HAA and terpenoids are benefitting from the Isomerase pathway as acetyl-CoA is the precursor (**Table 2**). Prodigiosin synthesis involves the precursors pyruvate, proline, and malonyl-CoA (Williamson et al., 2006), favoring a combination of the Isomerase and the Weimberg pathway. The Weimberg pathway would be beneficial for synthesis of N-methylglutamate. The attempt of using xylose as renewable source for the production of metabolites was already considered in several studies, where P. taiwanensis VLB120 was used for the synthesis of monorhamnolipids, phenol, and 4-hydroxybenzoate (Tiso et al., 2017; Wynands et al., 2018; Lenzen et al., 2019). P. taiwanensis VLB120 utilizes xylose via the Weimberg pathway natively (Köhler et al., 2015), but engineering of the strain and enabling xylose utilization via the Isomerase pathway could enhance product yield on substrate due to the superior stoichiometry.

While the integration of the xylose pathways worked in vivo, we observed strong differences between the oxidative and the Isomerase pathway operations. Pseudomonades using the three xylose pathways natively were reported (Hochster, 1955; Dahms, 1974; Köhler et al., 2015), indicating that these pathways are compatible with the Pseudomonades metabolic network. Notably, we did not find any report suggesting the presence of two of the xylose pathways. The implementation of the Isomerase pathway in P. putida was shown before (Meijnen et al., 2008; Le Meur et al., 2012; Dvorák and de Lorenzo, 2018; Wang et al., 2019). In accordance to our results, only weak growth was observed in two studies (Meijnen et al., 2008; Dvorák and de Lorenzo, 2018), which was improved by rational or nonrational engineering. In another study, growth on xylose via the Weimberg pathway in P. putida was shown to be immediately efficient (Meijnen et al., 2009), matching our results. As already considered by Wang et al. (2019), xylose metabolization seems to be metabolically demanding for P. putida KT2440 using the isomerase route and might be the reason for the discrepancies. But how can this difference between the Isomerase pathway and the oxidative pathways be explained? The consideration of other carbon sources indicates that the usage of oxidative pathways is likely to have an advantage for Pseudomonas species. The metabolization of the C<sup>5</sup> sugar arabinose proceeds in E. coli via the Isomerase pathway (Laikova, 2001), whereas in other Pseudomonas species the utilization of arabinose happens via oxidative steps to form the intermediate pentonic acid (Lockwood and Nelson, 1946; Weimberg and Doudoroff, 1955). In the case of galacturonic acid the utilization starts with an isomerase reaction in E. coli (Ashwell et al., 1960). Whereas, in Pseudomonas species the first enzyme is a dehydrogenase, which catalyzes an oxidation reaction (Kilgore and Starr, 1959; Richard and Hilditch, 2009). There might be two explanations why Pseudomonas favors oxidative pathways. First, the conversion of the substrate into an intermediate acid prevents other microbes from using this substrate, which is an advantage in terms of survival. Consequently, the environment is acidified due to the resulting intermediate acid, which creates a further advantage for propagation. P. putida KT2440 is a soil bacterium and is able to cope with extreme conditions, such as nutrient limitation, temperature shifts, and pH changes, in contrast to enterobacteria that occur in nutrient-rich niches (Martins Dos Santos et al., 2004; Reva et al., 2006). Second, the energy metabolism is more flexible, because the dehydrogenases have different redox cofactors. This can be explained using glucose as an example. In general, the redox metabolism is balanced and therefore, the rates of reduction and oxidation of the redox cofactors have to be highly similar (Blank et al., 2010). The electrons, which are released during oxidation of glucose and gluconate, are used to reduce PQQ and FAD<sup>+</sup> and are feeding directly the electron transport chain. There, PQQ is directly reoxidized by transferring the electrons to ubiquinone in the inner membrane (Ebert et al., 2011; Tiso et al., 2014). The transport of glucose over the membrane costs two ATP per molecule, while only one proton and one sodium ion are necessary for the transport of gluconate and ketogluconate, respectively. Therefore, P. putida KT2440 saves energy using the oxidative pathway and does not require additional cofactor regeneration systems.

However, a prolonged lag phase was observed for all three engineered strains. P. putida KT2440 pIso had the longest lag phase with 100 h, P. putida KT2440 pWeim had a lag phase of 24 h, and P. putida KT244011 pDahms had a lag phase of 34 h (**Table 3**). To identify possible bottlenecks, the plasmids were sequenced after ALE. Interestingly, no mutations in the xylose utilization genes and other encoding areas of the vectors could be detected. Different studies showed that mutations in the replication initiation protein or in the antibiotic resistance cassette could enhance growth by lowering the plasmid copy number, and subsequently reducing the need of resources for the synthesis of kanamycin resistance (Jakob et al., 2013; Mi et al., 2016). A second possible bottleneck could be the transport of xylonate from the periplasm to the cytoplasm. P. putida KT2440 harbors the transporter GntT (PP\_3417), enabling the transport of gluconate from the periplasm into the cytoplasm (Porco et al., 1997). Such a transporter is characterized for E. coli. A conceivable possibility would be, that this transporter might be active for xylonate, but due to its high-affinity toward gluconate it might operate slower for xylonate. Another option could be that a slow, low-affinity transporter is used instead. To improve the growth performance and reduce the lag phase, ALE was implemented. Interestingly, the lag phase was reduced significantly, while the growth rates did not increase and in one case even decreased during ALE. This can be explained by positive selection for a reduced lag phase and negative selection for the growth rate. The fast-adapting cells does not have stringently the highest growth rate. After reduction of the lag phase, it can be selected for increased growth rates now. While laboratory evolution is an easy method to increase the overall fitness of a population in laboratory scale, the design of the ALE approach for adaptation of the population is not trivial. Conditions such as time point of transfer, passage size, and growth phase can alter in ALE studies (Charusanti et al., 2010; LaCroix et al., 2015). Batch cultivation and continuous (chemostat) cultivation are the mostly used ALE techniques (Dragosits and Mattanovich, 2013; Gresham and Dunham, 2014; LaCroix et al., 2017). Of these, regularly transferred batch cultures are more popular because effort and costs are rather low. However, this method has several limitations due to varying conditions (LaCroix et al., 2017). Further, this method is slower than an automated ALE because the transfer is usually done on a daily basis. While in automated ALE processes, several parameters, including the optical density and growth rate, are monitored online. If the growth rate increases over the course of the ALE, the passage frequency can automatically be increased. Additionally, cultures probably reach the stationary phase in batch cultivations, which then results in an improved survival in stationary phase or decreased lag phase (Wiser and Lenski, 2015). Thus, it is not surprising that the batch cultivations for adaptation used in this study resulted in a shortage of the lag phase.

The here presented alternative pathways for xylose utilization open another degree of freedom in the design and metabolic engineering of the production strain. Dependent on the product of interest, the experimenter can compute the best network design considering three different stoichiometries for xylose use. The general applicability of this approach is outlined here, while as an outlook the detailed single strain optimization and the coconsumption of carbon sources in silico and in vivo should be worked on.

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/**Supplementary Material**.

#### AUTHOR CONTRIBUTIONS

IB performed all molecular engineering and characterization experiments, analyzed the data, prepared figures, conducted the in silico experiments, and wrote the manuscript. TT provided guidance on P. putida biotechnology and edited the manuscript. TT, AW, and FR discussed the data and critically revised the manuscript. LB advised on all experiments, analyzed and discussed data, and edited the manuscript.

#### FUNDING

The scientific activities of the Bioeconomy Science Center were financially supported by the Ministry of Culture and Science

#### REFERENCES


within the framework of the NRW Strategieprojekt BioSC (No. 313/323-400-00213). The GC-MS/MS was funded by the German Research Foundation DFG (Förderkennzeichen: INST 2221018-1 FUGG). Furthermore, the authors are grateful to the Fachagentur Nachwachsende Rohstoffe e.V. (FNR; Förderkennzeichen: 22013314), the Ministry of Science, Research and the Arts of Baden-Württemberg (MWK; Förderkennzeichen: 7533-10-5-186A and 7533-10-5-190) and the EU project Horizon 2020 AD GUT (ID: 686271) for providing financial support.

#### ACKNOWLEDGMENTS

We thank Prof. de Lorenzo [Systems and Synthetic Biology Program, Centro Nacional de Biotecnología (CNB-CSIC)] for providing P. putida EM42 1gcd pSEVA2213\_xylABE, Dr. Simone Schmitz (RWTH Aachen University) and Prof. Agler-Rosenbaum (Leibniz Institute for Natural Product Research and Infection Biology, Hans-Knöll-Institute) for providing plasmid pJNN\_phzA1-G1,M,S, Sebastian Köbbing (RWTH Aachen University) for providing plasmid pBG14ffg, and Wing-Jin Li (RWTH Aachen University) for providing P. putida KT2440 1gclR. We kindly acknowledge Hans M. Böhne and Sebastian Kruth for their experimental help.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00480/full#supplementary-material


production of poly(d-lactate-co-glycolate) and poly(d-lactate-co-glycolate-cod-2-hydroxybutyrate) in Escherichia coli. Microb. Biotechnol. 10, 1353–1364. doi: 10.1111/1751-7915.12721


using the carbon-efficient Weimberg pathway. Microbiol. 164, 287–298. doi: 10.1099/mic.0.000611


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Bator, Wittgens, Rosenau, Tiso and Blank. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# *De novo* Biosynthesis of Odd-Chain Fatty Acids in *Yarrowia lipolytica* Enabled by Modular Pathway Engineering

#### Young-kyoung Park <sup>1</sup> \*, Rodrigo Ledesma-Amaro<sup>2</sup> \* and Jean-Marc Nicaud<sup>1</sup>

<sup>1</sup> Université Paris-Saclay, INRAE, AgroParisTech, Micalis Institute, Jouy-en-Josas, France, <sup>2</sup> Imperial College Centre for Synthetic Biology and Department of Bioengineering, Imperial College London, London, United Kingdom

Microbial oils are regarded as promising alternatives to fossil fuels as concerns over environmental issues and energy production systems continue to mount. Odd-chain fatty acids (FAs) are a type of valuable lipid with various applications: they can serve as biomarkers, intermediates in the production of flavor and fragrance compounds, fuels, and plasticizers. Microorganisms naturally produce FAs, but such FAs are primarily even-chain; only negligible amounts of odd-chain FAs are generated. As a result, studies using microorganisms to produce odd-chain FAs have had limited success. Here, our objective was to biosynthesize odd-chain FAs de novo in Yarrowia lipolytica using inexpensive carbon sources, namely glucose, without any propionate supplementation. To achieve this goal, we constructed a modular metabolic pathway containing seven genes. In the engineered strain expressing this pathway, the percentage of odd-chain FAs out of total FAs was higher than in the control strain (3.86 vs. 0.84%). When this pathway was transferred into an obese strain, which had been engineered to accumulate large amounts of lipids, odd-chain fatty acid production was 7.2 times greater than in the control (0.05 vs. 0.36 g/L). This study shows that metabolic engineering research is making progress toward obtaining efficient cell factories that produce odd-chain FAs.

Keywords: odd-chain fatty acids, propionyl-CoA, *Yarrowia lipolytica*, metabolic engineering, Golden Gate assembly, synthetic biology

# INTRODUCTION

Microbial oils (lipids and fatty acid-derived products) are regarded as promising alternatives to fossil fuels in the face of growing concerns over environmental issues and energy production. To lessen the cost of producing microbial oils, considerable effort has been dedicated to enhancing production yield (Dulermo and Nicaud, 2011; Tai and Stephanopoulos, 2013; Qiao et al., 2015); using low-cost carbon substrates (Papanikolaou et al., 2013; Lazar et al., 2014; Guo et al., 2015; Ledesma-Amaro and Nicaud, 2016); and targeting high-value lipids (Xue et al., 2013; Xie et al., 2015). Odd-chain fatty acids (FAs), a type of valuable lipid, are a product with potential because they can be used in a variety of applications. Notably, research has revealed that odd-chain FAs with chain lengths of 15 and 17 carbons may have functional importance for nutrition and medical field. For example, cis-9-heptadecenoic acid has anti-inflammatory effects and can help treat psoriasis, allergies, and autoimmune diseases (Degwert et al., 1998). Pentadecanoic acid and heptadecanoic

#### *Edited by:*

Fayza Daboussi, Institut National de la Recherche Agronomique (INRA), France

#### *Reviewed by:*

Jiazhang Lian, Zhejiang University, China Carol Sze Ki Lin, City University of Hong Kong, Hong Kong

#### *\*Correspondence:*

Young-kyoung Park youngkyoung.park@inra.fr Rodrigo Ledesma-Amaro r.ledesma-amaro@imperial.ac.uk

#### *Specialty section:*

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

*Received:* 31 August 2019 *Accepted:* 27 December 2019 *Published:* 22 January 2020

#### *Citation:*

Park Y, Ledesma-Amaro R and Nicaud J-M (2020) De novo Biosynthesis of Odd-Chain Fatty Acids in Yarrowia lipolytica Enabled by Modular Pathway Engineering. Front. Bioeng. Biotechnol. 7:484. doi: 10.3389/fbioe.2019.00484 acid can be used as biomarkers of food intake in dietary assessments, the risk of coronary heart disease (CHD), and the risk of type II diabetes mellitus (Forouhi et al., 2014; Jenkins et al., 2015; Pedersen et al., 2016; Pfeuffer and Jaudszus, 2016). The chemical properties and potential biological activities of odd-chain FAs are now being more extensively studied (Rezanka and Sigler, 2009), so it is possible that novel nutritional and pharmaceutical applications could soon be discovered. In addition, odd-chain FAs and their derivatives are precursors for manufacturing substances such as pesticides, flavor and fragrance compounds, hydraulic fluids, plasticizers, coatings, and other industrial chemicals (Fitton and Goa, 1991; Avis, 2000; Clausen et al., 2010; Köckritz et al., 2010). Despite the broad range of applications for FAs, studies aiming to produce odd-chain FAs using microorganisms have had limited success because microorganisms produce a much greater proportion of evenchain FAs than odd-chain FAs.

Generally, de novo fatty acid synthesis in microorganisms begins with the condensation of acetyl-CoA and malonyl-CoA (**Figure 1A**). Then, the elongation step occurs: longchain FAs are synthesized in a reaction catalyzed by fatty acid synthase (FAS). The resulting acyl-CoA products are esterified to generate lysophosphatidic acid (LPA), then phosphatidic acid (PA), and finally diacylglycerol (DAG) before forming triacylglycerol (TAG), the compound in which the lipids are stored. For odd-chain FAs, propionyl-CoA could be converted from propionate is a primer for the fatty acid synthesis. The condensation of both propionyl-CoA and malonyl-CoA results in the formation of 3-oxovaleryl-ACP, which is the launching point for odd-chain FA synthesis. This five-carbon compound goes through elongation, where two carbons are added in each cycle. Then odd-chain FAs can be synthesized as described in **Figure 1B**. However, most microorganisms require the presence of propionate in the medium to produce odd-chain FAs. Wu and San have shown that E. coli can produce odd-chain FAs—namely undecanoic acid (C11:0), tridecanoic acid (C13:0), and pentadecanoic acid (C15:0)—if grown with propionatesupplemented medium (Wu and San, 2014). They introduced a propionyl-CoA synthetase (prpE) from Salmonella enterica and acyl-ACP thioesterases (TEs) from Umbellularia californica and Ricinus communis. Additionally, propionate supplementation has allowed various yeasts (e.g., Yarrowia lipolytica, Rhodotorula glutinis, Cryptococcus curvatus, and Kluyveromyces polysporus) to produce odd-chain FAs (Zheng et al., 2012; Kolouchová et al., 2015). More recently, in Y. lipolytica, metabolic engineering and the optimization of propionate feeding helped boost the production of odd-chain FAs, namely heptadecenoic acid (C17:1) (Park et al., 2018).

Therefore, to date, most studies have been focused on processes that involve propionate supplementation. However, due to the high cost (Poirier et al., 1995; Aldor et al., 2002) and toxic effects of propionate (Fontanille et al., 2012; Park and Nicaud, 2019), it is crucial to find alternative pathways for generating propionyl-CoA to be able to produce odd-chain FAs on large scales. There have been a few studies in which odd-chain FAs have been produced using glucose. Tseng and Prather have shown that, in E. coli, the production of very short odd-chain FAs (i.e., propionate, trans-2-pentenoate, and valerate) can be increased through the upregulation of threonine biosynthesis (Tseng and Prather, 2012). Another study demonstrated that the overexpression of threonine biosynthesis in E. coli resulted in increased levels of odd-chain FAs (mainly C15:0): from 0.006 to 0.246 g/L. This study further modified the experimental strain by replacing the native β-ketoacyl-ACP synthase (encoded by FabH) with one from Bacillus subtilis (encoded by FabHI) so that there was a biochemical preference for propionyl-CoA over acetyl-CoA (Lee et al., 2013). However, to date, no one has reported the de novo production of odd-chain FAs in Y. lipolytica or in any other yeasts without propionate supplementation. Consequently, we need more research on metabolic engineering approaches for producing odd-chain FAs in this group of microorganisms.

Of the several microbial hosts used in such systems, the oleaginous yeast Y. lipolytica is the most studied and has been engineered to produce large amounts of lipids and lipid derivatives, such as ricinoleic acid (Beopoulos et al., 2014), conjugated linoleic acids (Imatoukene et al., 2017), cyclopropane FAs (Czerwiec et al., 2019), and cocoa butter-like oils (Papanikolaou et al., 2003). Y. lipolytica can naturally grow in a broad range of substrates and has been further engineered to use even more substrates (Ledesma-Amaro et al., 2016). In addition,

many different synthetic biology tools have been created for and applied in Y. lipolytica, such as Gibson assembly, Golden Gate assembly, TALEN editing, and CRISPR/Cas9 editing [please see the recent review by Larroude et al. (2018) for more details]. These tools used in tandem with genetic and metabolic engineering strategies have boosted the capacities of Y. lipolytica, making the yeast into a promising host for biotechnological production processes.

The objective of this study was to biosynthesize odd-chain FAs de novo from glucose without propionate supplementation. We constructed a modular metabolic pathway for synthesizing propionyl-CoA from oxaloacetate in Y. lipolytica and confirmed that accumulation of odd-chain FAs was increased. We also investigated whether it could be competitive to produce oddchain FAs from propionyl-CoA via the methylcitrate pathway using the engineered strain. Additionally, we overexpressed the pyruvate dehydrogenase (PDH) complex in the cytosol to see if it could improve the conversion of α-ketobutyrate to propionyl-CoA. This work demonstrates that our metabolic engineering strategy for directing metabolic fluxes through specific pathways can enhance odd-chain FA production.

# MATERIALS AND METHODS

### Strains and Media

Media and growth conditions for E. coli were previously described by Sambrook and Russell (2001); those for Y. lipolytica were previously described by Barth and Gaillardin (1997). Rich medium (YPD) and minimal glucose medium (YNB) were prepared as described elsewhere (Milckova et al., 2004). The YNB contained 0.17% (w/v) yeast nitrogen base (without amino acids and ammonium sulfate, YNBww), 0.5% (w/v) NH4Cl, 50 mM KH2PO4-Na2HPO<sup>4</sup> (pH 6.8), and 2% (w/v) glucose. To complement strain auxotrophies, 0.1 g/L of uracil or leucine was added as necessary. To screen for hygromycin resistance, 250µg/ml of hygromycin was added to the YPD. Solid media were prepared by adding 1.5% (w/v) agar.

# Construction of Plasmids and Strains (*E. coli* and *Y. lipolytica*)

We used standard molecular genetic techniques (Sambrook and Russell, 2001). Restriction enzymes were obtained from New England Biolabs (Ipswich, MA, USA). PCR amplification was performed in an Eppendorf 2720 Thermal Cycler with either Q5 High-Fidelity DNA Polymerase (New England Biolabs) or GoTaq DNA polymerases (Progema, WI, USA). PCR fragments were purified using a PCR Purification Kit (Macherey-Nagel, Duren, Germany), and plasmids were purified with a Plasmid Miniprep Kit (Macherey-Nagel).

The plasmids used in this study were constructed using Golden Gate assembly, as described in Celinska et al. (2017). The genes in the A, T, and H module were obtained via PCR using the genomic DNA of Y. lipolytica W29. Internal BsaI recognition sites were removed via PCR using the primers listed in **Supplementary Table 1**. The plasmids for each module included the Zeta sequence, the URA3 ex marker, and gene expression cassettes containing the TEF1 promoter and LIP2 terminator.

For the cytosolic PDH complex, all the genes were synthesized and cloned in the plasmid pUC57 by GenScript Biotech (New Jersey, US). Cytosolic PDX1 was cloned into the expression plasmid (JME2563) using the BamHI and AvrII restriction sites. The other four genes were cloned into two plasmids (JME4774 and JME4775) using Golden Gate assembly.

To disrupt PHD1, the cassettes were constructed to include a promoter (pPHD1), a marker (URA3 or LEU2), and a terminator (TPHD1), which allowed the ORF gene to be removed via homologous recombination, as described in Papanikolaou et al. (2013).

Gene expression and disruption cassettes were prepared by NotI digestion and transformed into Y. lipolytica strains using the lithium acetate method, as described previously (Barth and Gaillardin, 1997). Gene integration and disruption were verified via colony PCR using the primers listed in **Supplementary Table 1**. The replicative plasmid harboring the Cre gene (JME547; **Table 1**) was used for marker rescue (Fickers et al., 2003). After transformation with the Cre expression plasmid, the loss of the marker gene was verified on YNB with/without uracil. The loss of the replicative plasmid was checked using replica plating on YPD with/without hygromycin after culturing on YPD for 24 h. To construct the prototrophic strain, a LEU2 fragment from plasmid JMP2563 was transformed. All the strains and plasmids used in this study are listed in **Table 1**.

# Culture Conditions for the Lipid Biosynthesis Experiments

The lipid biosynthesis experiments were carried out in minimal media, and the cultures were prepared as follows: an initial preculture was established by inoculating 10 mL of YPD medium in 50 mL Erlenmeyer flasks. Then, the pre-culture was incubated overnight at 28◦C and 180 rpm. The resulting cell suspension was washed with sterile distilled water and used to inoculate 50 mL of minimal medium YNBD6 containing 0.17% (w/v) yeast nitrogen base (without amino acids and ammonium sulfate, YNBww, Difco), 0.15% (w/v) NH4Cl, 50 mM KH2PO4-Na2HPO<sup>4</sup> buffer (pH 6.8), and 6% (w/v) glucose. This medium had been placed in 250 mL Erlenmeyer flasks. The cultures were then incubated at 28◦C and 180 rpm.

# Lipid Determination

Lipids were extracted from 10 to 20 mg of freeze-dried cells and converted into FA methyl esters (FAMEs) using the procedure described by Browse et al. (1986). The FAMEs were then analyzed using gas chromatography (GC), which was carried out with a Varian 3900 instrument equipped with a flame ionization detector and a Varian FactorFour vf-23ms column, where the bleed specification at 260◦C is 3 pA (30 m, 0.25 mm, 0.25µm). The FAMEs were identified via comparisons with commercial standards (FAME32, Supelco) and quantified using the internal standard method, which involves the addition of 100 µg of commercial dodecanoic acid (Sigma-Aldrich). Commercial oddchain FAs (Odd Carbon Straight Chains Kit containing 9 FAs, OC9, Supelco) were converted into their FAMEs using the same method employed with the yeast samples. They were then TABLE 1 | The plasmids and strains used in this study.


(Continued)

TABLE 1 | Continued


identified using GC and compared with the odd-chain FAs from the yeast samples.

To determine dry cell weight (DCW), 2 mL of the culture was taken from the flasks, washed, and lysophilized in a pre-weighed tube. The differences in mass corresponded to the mg of cells found in 2 mL of culture.

#### RESULTS

#### Modular Pathway Engineering Was Used for Odd-Chain Fatty Acid Synthesis

Propionyl-CoA is a key primer in the synthesis of odd-chain FAs. It can be synthesized using β-oxidation from direct precursors, propionate, or long-chain FAs. It can also be created from other metabolites via several metabolic pathways, such as the citramalate/2-ketobutyrate pathway, the aspartate/2 ketobutyrate pathway, the methylmalonyl-CoA pathway, the 3-hydroxypropionate pathway, and the isoleucine or valine degradation pathway (**Supplementary Figure 1**; Han et al., 2013; Lee et al., 2013). Here, we tested if the overexpression of the α-ketobutyrate pathway—which produces threonine as an intermediate—could increase levels of propionyl-CoA in Y. lipolytica. As shown in **Figure 2A**, the pathway eventually forms the amino acids aspartate, homoserine, and threonine from oxaloacetate. Then, threonine is deaminated to generate α-ketobutyrate, a reaction catalyzed by threonine dehydratase. Alpha-ketobutyrate is directly or sequentially converted into propionyl-CoA by the pyruvate dehydrogenase (PDH) complex or pyruvate oxidase, respectively. The upregulation of threonine has previously been used to boost propionyl-CoA availability in E. coli. Lee et al. showed that levels of oddchain FAs could be increased by introducing the threonine biosynthesis pathway (which creates α-ketobutyrate from aspartate semialdehyde), especially when mutated homoserine dehydrogenase (thrA<sup>∗</sup> , reduced feedback inhibition) was also expressed (Lee et al., 2013). The percentage of oddchain FAs out of total FAs increased from <1 to 18% by overexpressing the threonine pathway in E. coli. The predominant odd-chain FA produced was pentadecanoic acid (C15:0).

In this study, we enhanced the extended threonine biosynthesis pathway (the aspartate/α-ketobutyrate pathway) from oxaloacetate to α-ketobutyrate—by overexpressing seven genes (**Figure 2A**). There were three modules (**Figures 2A,C**): the aspartate synthesis module (module A), which included AAT2; the homoserine synthesis module (module H), which included HOM3, HOM2, and HOM6; and the threonine synthesis module (module T), which included THR1, THR4, and ILV1. While threonine and α-ketobutyrate (module T) are synthesized in the mitochondria in S. cerevisiae, the same may not be true in Y. lipolytica. While the locations of the relevant enzymes are as yet unknown in Y. lipolytica, predictive analyses (**Supplementary Table 4**) suggest enzyme location may differ between S. cerevisiae and Y. lipolytica. Because the module T enzymes were predicted to occur in the cytoplasm in Y. lipolytica, we used the original sequence of each gene in this study, as described in **Supplementary Table 3**. However, more research is needed to confirm enzyme locations in Y. lipolytica. The genes in each module were cloned into one plasmid using Golden Gate assembly (**Figure 2B**). They were expressed under the constitutive promoter pTEF1, and the expression cassette of each module was randomly integrated into the genome. Each module (A, T, and H) in the pathway was overexpressed in Y. lipolytica both individually and in tandem. The strain with the full modular pathway (ATH) was constructed by removing and reusing the URA3 marker (**Supplementary Figure 2**). We verified gene integration using colony PCR with the primer set of promoters and the ORF gene (**Supplementary Table 1**).

#### The Engineered Strain Could Produce Odd-Chain Fatty Acids Using Glucose as Its Sole Carbon Source

To determine whether the modular metabolic pathway was effective in producing odd-chain FAs, we evaluated the performance of the engineered strains overexpressing the individual modules and the entire pathway. The strains

were cultivated in YNBD6 medium under nitrogen limitation conditions (C/N = 60), which have been found to positively influence lipid synthesis (Beopoulos et al., 2012; Ledesma-Amaro et al., 2016). For the WT-A strain, which overexpressed AAT2, and the WT-T strain, which overexpressed THR1, THR4, and ILV1, total lipid content (%, g/g DCW) was lower than in the wild-type (WT) strain; the percentage of odd-chain FAs out of total FAs was similar (**Table 2**). For the WT-H strain, which overexpressed HOM3, HOM2, and HOM6, this percentage was slightly greater (1.91%) than that in the WT strain (0.84%) (**Table 2** and **Figure 3A)**. For the WT-ATH strain, which overexpressed the entire pathway, oddchain FA synthesis was significantly greater, odd-chain FA content (%, g/g DCW) was 3.8 times higher, and odd-chain FA titers (g/L) were 3.6 times higher than in the WT strain; the percentage of odd-chain FAs out of total FAs was 3.86%, which was 4.6 times higher than the value seen in the WT strain. These results indicate that the engineered aspartate/αketobutyrate pathway can supply propionyl-CoA; they also show that the full pathway is needed for effective odd-chain FA synthesis.

WT-ATH primarily produced C17:1 FAs (**Figure 3B**). This profile resembles that of an engineered Y. lipolytica strain that received propionate supplementation (Park et al., 2018). This finding implies that enhancing carbon flux through the α-ketobutyrate pathway can boost propionyl-CoA availability and odd-chain FA synthesis the same way that propionate supplementation can.

### Odd-Chain Fatty Acid Production Was Significantly Improved in a Lipid-Accumulating Strain

Once we had determined that the strain overexpressing the full modular pathway produced more odd-chain FAs, we wanted to boost their accumulation. Previously, we had engineered a Y. lipolytica strain, JMY3501, to accumulate large amounts of lipids (Lazar et al., 2014). We eliminated lipid degradation and


TABLE 2 | Fatty acid (FA) production in the wild-type (WT) strain and the engineered strains after growth on YNBD6 medium for 120 h.

The values represent the means and standard deviations for two independent experiments. DCW, dry cell weight.

odd-chain FAs relative to total FAs in the wild-type (WT) and engineered strains. (B) Odd-chain FA profiles (percentage of each FA out of total FAs) of the WT strain and engineered strains. WT, control strain; WT-A, strain expressing module A; WT-T, strain expressing module T; WT-H, strain expressing module H; WT-ATH, strain expressing the full pathway (all three modules). The results represent the means and standard deviations for two independent experiments.

remobilization in this obese strain by deleting the POX (POX1- 6) genes (Beopoulos et al., 2008) as well as the TGL4 gene, which encodes a triglyceride lipase (Dulermo et al., 2013). In addition, we overexpressed DGA2, which encodes the major acyl-CoA:diacylglycerol acyltransferase (Beopoulos et al., 2012), and GPD1, which encodes glycerol-3-phosphate dehydrogenase, in order to push and pull TAG biosynthesis (Dulermo and Nicaud, 2011; Tai and Stephanopoulos, 2013). Therefore, using JMY3501, we built a new obese strain that overexpressed our full modular pathway. It was called the obese-ATH strain. We then studied lipid production under the same conditions as before. As expected, total lipid accumulation was 2.29 fold greater in the obese-ATH strain than in the WT-ATH strain (**Tables 2**, **3**). Interestingly, the obese-ATH strain also accumulated more odd-chain FAs: the percentage of odd-chain FAs out of total FAs was 5.64% in the obese-ATH strain vs. 3.86% in the WT-ATH strain. The obese-ATH strain produced 0.36 g/L of odd-chain FAs, which is 7.2 times greater than the amount produced by the regular obese strain. The obese-ATH strain and the WT-ATH strain differed in their evenchain FA profiles (**Table 4**). The obese-ATH had slightly higher levels of C16:0 and slightly lower levels of C18:1, a common pattern seen in strains with the obese background regardless of the carbon source (Lazar et al., 2014; Ledesma-Amaro et al., 2016).

## The Disruption of *PHD1* in the Methylcitrate Cycle Did Not Increase Odd-Chain Fatty Acid Production

In a previous study, we inactivated PHD1, the gene that encodes 2-methylcitrate dehydratase, which catalyzes the conversion of 2-methyl citrate to 2-methyl-cis-aconitate in the methylcitrate cycle; we showed that the resulting higher levels of propionyl-CoA could be used to synthesize greater amounts of odd-chain FAs (Park et al., 2018). To investigate whether the inhibition of the methylcitrate cycle—via the deletion of PHD1—could further improve the accumulation of odd-chain FAs, we disrupted the PHD1 gene in both the WT-ATH strain and the obese-ATH strain. The two phd11 strains displayed higher total lipid content compared to their relative controls (**Supplementary Table 2**), a result that was also demonstrated in Papanikolaou et al. (2013); however, they also displayed lower percentages of odd-chain FAs out of total FAs (**Figure 4**). This latter negative effect was significantly more pronounced in the obese-ATH phd11 strain than in the WT-ATH phd11 strain. For the obese-ATH phd11 strain, this figure dropped by 50%, and levels of odd-chain FAs were 67% of those seen in the relative control (0.24 vs. 0.36 g/L). These results suggest that disrupting the methylcitrate cycle does not provide the benefits seen previously (Park et al., 2018) when strains are already overexpressing the aspartate/αketobutyrate pathway.

TABLE 3 | Fatty acid (FA) production in the obese strain and the obese-ATH strain after growth on YNBD6 medium for 120 h.


The values represent the means and standard deviations for two independent experiments. DCW, dry cell weight.

TABLE 4 | Comparison of the lipid profiles (% of each FA) of the WT-ATH strain and the obese-ATH strain.


The values represent the means and standard deviations for two independent experiments.

The action of 2-methylcitrate dehydratase is restricted to the mitochondria. Consequently, disabling this enzyme does not directly improve cytosolic levels of propionyl-CoA. In the previous studies showing that the disruption of PHD1 boosted odd-chain FA synthesis (Papanikolaou et al., 2013; Park et al., 2018), there was also propionate supplementation. Taken together, these findings suggest that, when the methylcitrate cycle is disrupted, the overexpression of the aspartate/α-ketobutyrate pathway has a weaker effect on propionyl-CoA levels than does propionate supplementation.

#### Overexpression of the Cytosolic Pyruvate Dehydrogenase (PDH) Complex Improved Lipid Synthesis but Reduced Odd-Chain Fatty Acid Production

The PDH complex consists of three main catalytic components: E1 (pyruvate dehydrogenase, encoded by PDA1 and PDB1), E2 (dihydrolipoamide acetyltransferase, encoded by LAT1), and E3 (dihydrolipoamide dehydrogenase, encoded by LPD1). There is a fourth component (protein X, encoded by PDX1) that binds to and positions E3 relative to E2. In Y. lipolytica, the PDH complex is found in mitochondria and catalyzes the pathway from pyruvate to acetyl-CoA. A few studies have examined the functional expression of the PDH complex in Y. lipolytica. One study attempted to overexpress the direct pathway from pyruvate to acetyl-CoA in a coordinated manner (Markham et al., 2018), and another study showed that the individual overexpression of PDA1 (shared subunits with α-ketoglutarate dehydrogenase) improved α-ketoglutarate production (Guo et al., 2014).

Here, we wanted to redirect the aspartate/α-ketobutyrate pathway to produce propionyl-CoA via the PDH complex, which has already been shown to be possible in E. coli (Danchin et al., 1984; Tseng and Prather, 2012; Lee et al., 2013). The PDH complex was built so it could be associated with the full modular pathway (ATH) and could provide a pool of propionyl-CoA in the cytosol where lipid synthesis takes place. The mitochondrial targeting sequences (MTSs) of each gene were predicted using MitoProt (Claros and Vincens, 1996), and the gene sequences used in this study are described in **Supplementary Table 3**. We then created the obese-ATHP strain by overexpressing the cytosolic PDH subunits in the obese-ATH strain. Next, to explore lipid accumulation dynamics, the strain was grown in YNBD6 with lipoic acid, which is required for cytosolic PDH activity in yeast (Kozak et al., 2014). Compared to the control strain, the obese-ATHP strain had significantly lower levels of oddchain FAs (a 3.8 times lower percentage of odd-chain FAs out of total FAs); however, total lipid levels were higher (**Table 5**). The increase in lipid production might have resulted from the increased levels of cytosolic acetyl-CoA resulting from the overexpression of the PDH complex. A similar strategy was utilized in Saccharomyces cerevisiae, where PDH was introduced into the cytosol and was found to increase levels of acetyl-CoA (Kozak et al., 2014; Lian et al., 2014) and of the target compound of interest. Since acetyl-CoA is a key precursor in the production of both even- and odd-chain FAs, increasing levels of acetyl-CoA promotes lipid synthesis in general. However, the substantially lower levels of odd-chain FAs in the obese-ATHP strain implies that the PDH complex shows greater specificity for pyruvate than for α-ketobutyrate. The higher Km value of α-ketobutyrate


The values represent the means and standard deviations for two independent experiments. DCW, dry cell weight.

compared to that of pyruvate has been seen elsewhere, such as in E. coli (Bisswanger, 1981), Neurospora crassa (Harding et al., 1970), and mammalian cells (Bremer, 1969). Therefore, it is important to explore enzyme engineering strategies that modify substrate specificity or that introduce other enzymes that can convert acetyl-CoA to propionyl-CoA with a view to further improving odd-chain FA production via a threoninebased upregulation strategy.

#### DISCUSSION

In this study, we developed a synthetic biological strategy for the de novo production of odd-chain FAs in Y. lipolytica. It is important to note that the wild-type Y. lipolytica strain produces only negligible amounts of odd-chain FAs even though it has an excellent capacity to accumulate large quantities of lipids. Several studies have shown that propionate supplementation can increase the production of odd-chain FAs (Fontanille et al., 2012; Kolouchová et al., 2015; Park et al., 2018). However, research has yet to explore the de novo production of odd-chain FAs from sugars in Y. lipolytica.

The overexpression of the aspartate/α-ketobutyrate pathway (from oxaloacetate to homoserine and threonine) via Golden Gate assembly resulted in higher levels of odd-chain FAs being produced from glucose. The best strain generated a level of oddchain FAs, 0.36 g/L in flask that is the highest to date to be achieved in Y. lipolytica without propionate supplementation. Furthermore, it is comparable to the levels seen in a previous study of ours that employed propionate supplementation, where odd-chain FAs titers were 0.14 and 0.57 g/L in the wild-type strain and the obese strain, respectively (Park et al., 2018). To further increase the amount of propionyl-CoA produced via the overexpression of threonine synthesis, we constructed a cytosolic pyruvate dehydrogenase (PDH) complex. Because of the lower specificity of the PDH complex for α-ketobutyrate vs. pyruvate, the engineered strain generated lower levels of propionyl-CoA than did the relative control; however, the increased levels of acetyl-CoA in the engineered strain led to larger amounts of total FAs. This study is the first to describe the functional expression of the native PDH complex in the cytosol in Y. lipolytica, an approach that could also be employed to produce acetyl-CoA-derived compounds, such as polyhydroxybutyrates, isoprenoids, sterols, polyketides, polyphenols, alkanes, and alkenes (Nielsen, 2014).

Before the approach described here can be applied at industrial scales, certain issues must be resolved. It is necessary to perform research in which combinatorial pathway analysis (Lütke-Eversloh and Stephanopoulos, 2008), targetedproteomics analysis (Redding-Johanson et al., 2011), and genome-scale metabolic network modeling (Xu et al., 2011) are used to identify bottlenecks and enhance metabolic fluxes to propionyl-CoA and odd-chain FAs. The results of the PDH overexpression experiment suggest that acetyl-CoA is a competitive precursor to propionyl-CoA in odd-chain FA synthesis and is also a precursor in lipid synthesis. Therefore, better balancing the pools of acetyl-CoA and propionyl-CoA could be key in further increasing odd-chain FA content. One way to improve odd-chain FA synthesis is to introduce enzymes—such as CoA transferase—that redirect CoA moieties from acetyl-CoA to propionyl-CoA (Yang et al., 2012) or that have greater specificity for 3-oxovaleryl-ACP than for acetoacetyl-CoA (Slater et al., 1998). Overall, this work has shown that applying synthetic biological engineering strategies in Y. lipolytica to improve odd-chain FA production could be useful in a wide range of pharmaceutical and industrial contexts.

#### DATA AVAILABILITY STATEMENT

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher.

#### AUTHOR CONTRIBUTIONS

YP, RL-A, and J-MN planned the study. YP designed and carried out the experiments and drafted the manuscript. All the authors revised and approved the final manuscript.

#### FUNDING

YP was the recipient of a Ph.D. scholarship from the Kwanjeong Educational Foundation (KEF).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00484/full#supplementary-material

# REFERENCES


lipolytica encoded by TGL3 and TGL4 genes: new insights into the role of intracellular lipases and lipid body organisation. Biochim. Biophys. Acta. 1831, 1486–1495. doi: 10.1016/j.bbalip.2013.07.001


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Park, Ledesma-Amaro and Nicaud. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Metabolic Engineering of Bacterial Respiration: High vs. Low P/O and the Case of Zymomonas mobilis

Uldis Kalnenieks\*, Elina Balodite and Reinis Rutkis

Institute of Microbiology and Biotechnology, University of Latvia, Riga, Latvia

Respiratory chain plays a pivotal role in the energy and redox balance of aerobic bacteria. By engineering respiration, it is possible to alter the efficiency of energy generation and intracellular redox state, and thus affect the key bioprocess parameters: cell yield, productivity and stress resistance. Here we summarize the current metabolic engineering and synthetic biology approaches to bacterial respiratory metabolism, with a special focus on the respiratory chain of the ethanologenic bacterium Zymomonas mobilis. Electron transport in Z. mobilis can serve as a model system of bacterial respiration with low oxidative phosphorylation efficiency. Its application for redox balancing and relevance for improvement of stress tolerance are analyzed.

#### Edited by:

Jean Marie François, UMS3582 Toulouse White Biotechnology (TWB), France

#### Reviewed by:

Jian-Ming Liu, Technical University of Denmark, Denmark Ralf Takors, University of Stuttgart, Germany

> \*Correspondence: Uldis Kalnenieks kalnen@lanet.lv

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 30 August 2019 Accepted: 28 October 2019 Published: 12 November 2019

#### Citation:

Kalnenieks U, Balodite E and Rutkis R (2019) Metabolic Engineering of Bacterial Respiration: High vs. Low P/O and the Case of Zymomonas mobilis. Front. Bioeng. Biotechnol. 7:327.

doi: 10.3389/fbioe.2019.00327

Keywords: respiratory chain, metabolic engineering, energy coupling, redox balance, stress resistence, Zymomonas mobilis

# INTRODUCTION

During the last few decades a vast body of evidence on the structure, function and regulation of microbial electron transport chains has accumulated, enabled by the rapid advancement of omics techniques and molecular cloning tools. Research has focussed on deciphering of particular electron pathways in several model microorganisms, establishing their physiological role, and understanding the principles of their regulation. Transcriptional regulation in Escherichia coli in response to variation of available electron acceptors perhaps is the best known success story (for reviews, see Spiro and Guest, 1991; Unden and Bongaerts, 1997; Green and Paget, 2004). Here we briefly discuss how this wealth of theoretical knowledge can be used for industrial applications in the present era of metabolic engineering and synthetic biology. Generally speaking, microbial aerobic electron transport chains represent a promising, yet at the same time, challenging target for rational metabolic engineering, since their activities transcend several key cellular processes, like the catabolic ATP yield, maintenance of intracellular redox balance, production of reactive oxygen species (ROS), and protection against oxidative stress. In the present minireview we attempted to systematize the strategies of respiratory chain engineering, based on their impact on the cellular energy and reducing equivalent balance. We did not review the biotechnological applications of bacterial anaerobic respiration, electron transport chains of extremophiles, as well as bioelectrochemistry—interaction of bacterial electron transport with electrodes. Instead of trying to embrace all aspects of microbial electron transport, we here focussed on a few current biotechnology workhorses, showing how the metabolic engineering of their respiration had yielded predictable (or sometimes, not so predictable) effects on their growth and production. Some emphasis was put on low energy-coupling electron transport and the bacterium Zymomonas mobilis as a representative of the energetically uncoupled respiration.

## RESPIRATORY ENGINEERING TO IMPROVE FERMENTATIVE CATABOLISM

Microbial respiration represents a major sink of reducing equivalents and a source of energy (ATP) for the cellular metabolism. However, if the target compound of a bioprocess is produced by a redox-balanced fermentative pathway, then respiration, by interfering with the cellular redox and energy balance, causes a decrease of product yield, to mention Pasteur effect as a classical example (for a review, see Barnett and Entian, 2005). NADH supply of reactions, producing the desired target compound of fermentative catabolism, often represents a bottleneck (Zhao et al., 2017), especially under aerobic or microaerobic conditions (Bennett and San, 2017). Removing oxygen from the growth medium seems a straightforward approach for turning off the unwanted respiration and fueling more NADH into the fermentative catabolism. However, prolonged maintenance of a strict anoxic condition may be a challenging task. It may complicate, for example, long-term directed evolution experiments, like those aimed at engineering of the E. coli fermentative metabolism (Portnoy et al., 2008), where complete absence of oxygen is essential. On the other hand, many yeast species performing ethanol fermentation require limited presence of oxygen (Visser et al., 1990), often within a certain narrow concentration range for optimum ethanol yield (Acevedo et al., 2017). From the technical point of view, keeping low, steady, micro-aerobic oxygen levels in bioreactors (especially for larger volumes) may be difficult, due to the sensitivity of oxygen mass transfer coefficient to the agitation rate, insufficient responsiveness of the dissolved oxygen probe and spatial in homogeneities (Wu et al., 2015a).

Decreasing the activity of electron transport chain itself represents a viable alternative way to downregulate respiration, so that the cells would behave similar to that of anoxic or microaerobic environment even in the presence of oxygen excess (Bennett and San, 2017). Such a metabolic engineering strategy of limiting the use of oxygen by the cell itself would be useful also under some co-cultivation settings, e.g., when an end-product of fermentative catabolism of one microorganism species is used as a substrate for aerobic metabolism of the second, co-cultivated species. Decreasing of respiration may be achieved by respiratory inhibitors, keeping in mind that inhibitor effects may serve as a proof of principle, yet are hardly applicable to industrial bioprocesses. Thus, Acevedo et al. (2017), using rotenone, the inhibitor of the mitochondrial NADH dehydrogenase complex I, demonstrated an increase of aerobic ethanol yield from xylose in the yeast Scheffersomyces stipitis. Cyanide at submillimolar concentrations was shown to improve growth and ethanol yield in aerobic culture of ethanologenic bacterium Zymomonas mobilis (Kalnenieks et al., 2000).

Not surprisingly, construction and study of respiratory knockout mutants nowadays has become the dominating approach to the downregulation of microbial respiration, and has brought positive results. In E. coli, removal of all its terminal oxidases, followed by adaptive evolution, yielded strains able to produce lactate as a fermentation product from glucose and to undergo mixed-acid fermentation during aerobic growth (Portnoy et al., 2008). For Corynebacterium glutamicum, another important biotechnological producer, a respiratory chain mutant lacking both the cytochrome bd branch and the cytochrome bc1 aa<sup>3</sup> branch was constructed (Koch-Koerfges et al., 2013), and its ability of fermentative growth without aerobic respiration was demonstrated. The mutant retained only 2% of the wild type oxygen consumption rate and displayed fermentative catabolism with L-lactate as major and acetate and succinate as minor products. In aerobically grown Klebsiella pneumoniae knocking out of its NADH dehydrogenase complex I (nuo) improved 2,3-butanediol production (Zhang et al., 2018). In the ethanol-producing, facultatively anaerobic bacterium Z. mobilis inactivation of its Type II respiratory NADH dehydrogenase (ndh) largely improved growth and ethanol yield under aerobic conditions (Kalnenieks et al., 2008; Hayashi et al., 2011, 2012; see below).

Some metabolic engineering studies have attempted finetuning of the respiration rate in E. coli. In order to mimic microaerobic metabolism of E. coli under condition of excessive aeration, Zhu et al. (2011) manipulated respiration rate of cells by externally adding various amount of coenzyme Q<sup>1</sup> to a strain lacking the ubiCA genes necessary for quinone biosynthesis. They demonstrated that synthesis of fermentation products, like lactate and ethanol, varied inversely with the added amount of Q1. In the further study (Wu et al., 2015a), a nearly theoretical yield of lactate production was achieved under fully aerobic conditions by introducing a tunable pathway competing with biosynthesis of the precursors of coenzyme Q8. This regulatory strategy was named by authors "metabolic transistor" (Wu et al., 2015b), allowing to control a large flux (electron transport) by a small change in the availability of one of the electron transport components (Q8). An approach somewhat similar to that of Zhu et al. (2011) has been applied to Lactococcus lactis, in which the activity of respiratory chain naturally depends on externally added hemin, the precursor of the cytochrome heme cofactors (Liu et al., 2017). The authors constructed a multiple dehydrogenase knock-out strain of L. lactis with its respiratory chain as the only remaining NADH sink, and showed how the external control of respiration rate by hemin affected its growth and the degree of reduction of the synthesized products (acetoin vs. butanediol).

Bacterial respiratory chains can be engineered to control NADH/NAD<sup>+</sup> ratio and reducing equivalent fluxes also under anaerobic conditions. For enhancing NADH-dependent fermentative production in some strict anaerobes, the activity of electron transport chain could be used to generate NADH at the expense of other reduced cofactors, e.g., ferredoxin. This approach recently was demonstrated in Clostridium thermocellum (Lo et al., 2017), and in the extreme thermophile Caldicellulosiruptor bescii (Williams-Rhaesa et al., 2018), in which conversion of pyruvate to acetyl-CoA proceeds largely via pyruvate:ferredoxin oxidoreductase. By overexpressing Rnf, the membrane bound energy-conserving ferredoxin:NAD<sup>+</sup> oxidoreductase, NADH flux to the NADH-dependent ethanol synthesis could be improved, leading to a substantial increase of ethanol production in both these bacteria, and a concomitant decrease of production of unwanted side-products.

Notably, some recent works on modification of respiratory NADH dehydrogenase activity in aerobic or facultatively anaerobic bacteria reported unexpected, NADH dehydrogenasedependent metabolic shifts under strictly anaerobic conditions, which at present seem difficult to explain. Under aerobic conditions, manipulating the activity of either the NADH dehydrogenase complex I (nuo) or the Type II NADH dehydrogenase (ndh) was shown to affect the intracellular NADH/NAD<sup>+</sup> ratio, and accordingly, the catabolic product spectrum in a fairly predictable manner, as reported for several bacteria, e.g., Bacillus subtilis (Gyan et al., 2006), E. coli (Liu Q. et al., 2014), K. pneumoniae (Zhang et al., 2018), and Streptococcus agalactiae (Lencina et al., 2018). Yet, for example in C. glutamicum, overexpression of the ndh gene under oxygen-deprived conditions also substantially decreased the NADH/NAD<sup>+</sup> ratio, at the same time increasing the glucose consumption rate (Tsuge et al., 2015). Here "oxygen-deprived conditions" referred to <0.01 ppm oxygen, and importantly, no alternative terminal electron acceptors were present in the medium. Similarly, Hayashi et al. (2015) found a strong dependence of NADH/NAD<sup>+</sup> ratio on the Ndh activity in complemented Z. mobilis ndh mutant strains under strictly anaerobic conditions. No electron acceptors alternative to oxygen are known for this bacterium. Steinsiek et al. (2014) noted that inactivation of the NADH dehydrogenase complex I (nuo) in E. coli affected not only its aerobic metabolism, but induced significant metabolic response also under anaerobic conditions, and without alternative external electron acceptors—again, conditions where the respiratory chain should not be active. They observed genes coding for TCA cycle enzymes being upregulated, and higher amounts of succinate being synthesized by the mutant. At the same time, they noted a substantial shift of the ArcA phosphorylation pattern in the nuo mutant relative to the parent strain. Earlier Perrenoud and Sauer (2005) demonstrated that ArcA-dependent transcriptional regulation controls TCA cycle in E. coli not only under (micro)aerobic, but also under fully anaerobic conditions. Yun et al. (2005) reported higher lactate and formate yields in E. coli ndh mutant and ndh nuo double mutant under strictly anaerobic conditions. Apparently, more research is needed to understand these anaerobic effects and their possible relation to the global regulatory mechanisms linking respiratory chain to the central metabolism. At the same time, it is important to take them into account for practical respiratory engineering.

#### IMPROVING RESPIRATORY ENERGY-COUPLING

The classical work of Bauchop and Elsden (1960) provided evidence that on average one mole of ATP could fuel synthesis of slightly above 10gdrywt of bacterial biomass. With a few exceptions, this value (YATP) was found to be roughly similar for different microorganisms (Stouthamer, 1977). The apparent YATP is far from being a true "biological constant" for all growth conditions, since it depends on cellular maintenance energy requirements and energy-spilling reactions (Russell and Cook, 1995). Yet, primarily it is the variation of energy generation efficiency, but not that of YATP, that underlies the observed vast differences of microbial growth yields. Regarding aerobic catabolism, the efficiency of energy generation is determined by the stoichiometry of oxidative phosphorylation (Calhoun et al., 1993). The stoichiometry of oxidative phosphorylation (P/O) is the molar ratio between the generated ATP macroergic phosphate bonds and the oxygen atoms reduced to water by the respiratory chain (Hinkle, 2005). The mechanistic (maximum) P/O value for each particular case can be estimated from the known proton motive stoichiometries of the respiratory complexes and H+-ATP synthase, keeping in mind that the real values are lower due to proton leaks and slips in the energycoupling mechanisms. Typically, the bacterial electron transport chains are branched, and for electrons traveling from NADH via the respiratory complexes to oxygen, their contribution to energy coupling strongly depends on the particular pathway that the electrons take. For example, in E. coli electron transport chain (for a review, see Unden and Bongaerts, 1997) the stoichiometry of proton translocation per one transported electron (H+/e) may vary between 1 and 4 (accordingly, H+/O varies between 2 and 8). Assuming the proton/ATP stoichiometry of the H+- ATP synthase close to 3 (Kashket, 1982), the theoretical lower and upper limits for P/O in E. coli appear to be 0.67 and 2.67, respectively. In reality, E. coli and most other bacteria with branched respiratory chains tend to employ their energetically most efficient and less efficient electron transport branches simultaneously, resulting in a submaximal net energy coupling efficiency, which is hard to measure directly under in vivo condition. A metabolic network model with a growth ratedependent biomass composition, used for simulation of E. coli chemostat cultivations on several carbon substrates at different dilution rates, has yielded an estimate of P/O close to 1.5 (Taymaz-Nikerel et al., 2010). Thermodynamic analysis suggests that the submaximal P/O values, typically in the range of 1.0–1.5, indicate a tradeoff between metabolic energy conservation and energy dissipation, ultimately directed toward maximization of the microbial growth rate (Westerhoff et al., 1983; Molenaar et al., 2009; Werner et al., 2010).

From a practical point of view, increasing of P/O represents a metabolic engineering strategy for improving microbial biomass yield (**Figure 1**). For increase of P/O, the electron transport branches with low coupling efficiency have to be inactivated, and/or those with high efficiency have to be overexpressed. Biomass yield (and in some cases, e.g., in lactic acid bacteria, also multiple stress tolerance; Zotta et al., 2013) can thus be improved, which is desirable for cost-efficient biomass production of starter cultures, for the net productivity of bioconversions, and for the production of biomass-related compounds such as proteins, storage polymers and vitamins (Minohara et al., 2002; Brooijmans et al., 2007; Pedersen et al., 2012; Richhardt et al., 2013; Liu Q. et al., 2014).

Apart from biomass, high P/O ratio has been shown to push the synthesis of various products via pathways that consume (or at least, do not produce) ATP (de Kok et al., 2012). This takes

place because the energetically efficient respiration minimizes the amount of carbon substrate that has to be oxidized to generate ATP for needs of growth and maintenance. Given that the carbon source is not limiting and its uptake is rapid enough, the high P/O may facilitate production of incompletely oxidized compounds, either target products or, sometimes, undesired inhibitory metabolites. This phenomenon is known as "overflow metabolism" (Neijssel and Tempest, 1976; **Figure 1**). Enhancing product synthesis by elevating P/O has been a strategy of quite a few metabolic engineering projects. In most cases, elevation of P/O is achieved by knocking out the energetically least efficient terminal oxidase (typically cytochrome bd), and thus redirecting electron flow to energetically more efficient terminal oxidase(s). Poly-γ-glutamic acid (γ-PGA), a multifunctional biopolymer with various applications, is an example of a product, for which ATP supply is essential for its biosynthesis. By elimination of the cytochrome bd branch in Bacillus licheniformis respiratory chain, Cai et al. (2018) reached an increase of γ-PGA yield by almost 20%. Efficient supply of ATP is critical also for production of riboflavin. In a fed-batch culture of riboflavin-producing B. subtilis a 30% higher riboflavin titer was accumulated by the cyd mutant compared to the control strain (Zamboni et al., 2003), apparently because of redirection of the electron flow to the proton-pumping aa<sup>3</sup> terminal oxidase. Similar approach was used for improving of N-acetylglucosamine production in B. subtilis (Liu Y. et al., 2014) and lysine yield in C. glutamicum (Kabus et al., 2007). Liu Q. et al. (2014) demonstrated that inactivation of the cytochrome bd-II terminal oxidase and/or the type II (energy non-coupling) respiratory NADH dehydrogenase results in a several-fold increase of poly(3-hydroxybutyrate) accumulation in an E. coli recombinant producer strain.

However, under some natural conditions, and also in numerous bioprocess setups it may be preferable for bacteria to have respiration with low rather than high P/O.

## INCREASING RESPIRATION RATE WHILE DECREASING ENERGY-COUPLING EFFICIENCY

Low P/O is preferable when oxidation of NAD(P)H is needed for a rapid regeneration of NAD(P)<sup>+</sup> and/or consumption of oxygen, without producing of extra ATP (**Figure 1**). Thus, the free-living nitrogen-fixing bacteria Azotobacter vinelandii demonstrate the utility of rapid, low-P/O respiration as one of possible means to protect their nitrogenase complex from dioxygen by keeping the intracellular oxygen concentration low (Poole and Hill, 1997). It was shown that the activity of the low energy-coupling cytochrome bd terminal oxidase (Kelly et al., 1990) in combination with an energy non-coupling NADH:UQ oxidoreductase (Bertsova et al., 2001) is critical for A. vinelandii ability to fix nitrogen while growing under conditions of air saturation. In bioprocesses, energetically uncoupled respiration is useful for avoiding production of unneeded biomass and overflow metabolites, while enabling conversion of reduced substrate into more oxidized product(s). The obligately aerobic alpha-proteobacterium Gluconbacter oxydans belonging to the family of acetic acid bacteria is a relevant example. Due to its numerous periplasmic membrane-bound dehydrogenases that incompletely oxidize sugars, sugar alcohols, and other compounds, G. oxydans has been used for decades as a biocatalyst for bioconversions (Deppenmeier et al., 2002). It oxidizes a variety of substrates and transfers the electrons via ubiquinone to the quinol oxidases bo<sup>3</sup> and bd. Although the energetically more efficient, proton-pumping bo<sup>3</sup> represents its major terminal oxidase activity (Richhardt et al., 2013), due to some reason the proton pumping stoichiometry of the G. oxydans respiratory chain is more than two-fold lower than that of aerobically cultivated E. coli bearing analogous major terminal oxidase.

Seeking to relate the overflow metabolism phenomenon in the model microorganisms E. coli and S. cerevisiae to their respiratory capacity, Vemuri et al. (2006a, 2007) demonstrated that overexpression of plant non-coupling alternative terminal oxidases, and/or the energy non-coupling, soluble water-forming NADH oxidase from Streptococcus mutans reduces the overflow metabolism, decreases the biomass yield, and shifts catabolism toward more oxidized products. In E. coli overexpression of streptococcal NADH oxidase was shown to raise the TCA cycle flux and to elevate the NADPH/NADP<sup>+</sup> ratio (Holm et al., 2010). In combination with the deletion of arcA gene it dramatically reduced acetate accumulation and hence improved capacity of recombinant E. coli to overexpress heterologous proteins (Vemuri et al., 2006b). Expression of the soluble water-forming NADH oxidases (nox) of Streptococci or Lactococci, which do not contribute to energy coupling, has been widely applied in metabolic engineering projects for enhancing respiration, lowering its energy-coupling efficiency, and decreasing the intracellular NADH/NAD<sup>+</sup> ratio. Overexpression of noxE from L. lactis in Saccharomyces cerevisiae has been shown to shift catabolism toward more oxidized products, like acetaldehyde, acetate, and acetoin (Heux et al., 2006). NoxE overexpression prevents accumulation of reduced byproducts, such as glycerol and xylitol, yet improves ethanol yield during xylose fermentation by engineered S. cerevisiae (Zhang et al., 2017), as well as improves oxidative and osmotic stress tolerance of S. cerevisiae aerobic batch culture on glucose (Shi et al., 2016). The same oxidase has been overexpressed in E. coli (Balagurunathan et al., 2018) and in L. lactis (Bongers et al., 2005) to increase acetaldehyde yield at the expense of more reduced products of sugar catabolism. Bacterial soluble NADH oxidases are recently used also for in vitro metabolic pathways, reconstituted with isolated enzymes, as the means for recycling NADH (Nowak et al., 2015; Taniguchi et al., 2017).

In a broader sense, the metabolic engineering strategy for uncoupling of energy metabolism is not confined merely to manipulating the electron transport modules. A decrease of the energy-coupling efficiency can be achieved also by targeting the proton-dependent ATPase in order to disrupt oxidative phosphorylation, or to stimulate ATP hydrolysis. Following this approach, one can achieve energetical uncoupling in combination with overflow metabolism. Indeed, several studies have reported a decrease of biomass yield, growth rate, and intracellular ATP/ADP ratio, an increase of glycolytic and respiratory rates, and a rise of acetate production (overflow) in ATPase-deficient E. coli mutants (Jensen and Michelsen, 1992; Noda et al., 2006). Noda et al. (2006) observed activation of the low energy-coupling modules in the electron transport chain of such mutant: the type II NADH dehydrogenase and cytochrome bd terminal oxidase were upregulated. The reason for that might be the need for rapid regeneration of NAD<sup>+</sup> to keep the rate of glycolysis high, while at the same time investing less into proton motive force generation, since it no more could drive ATP synthesis. A similar shift in respiratory chain composition and an onset of acetate overflow were reported by Schuhmacher et al. (2014) in a wild type E. coli, grown under phosphate limiting condition. Notably, these physiological effects of phosphate limitation, resembling the ATPase knock-out phenotype, correspond to earlierin vitro findings with membrane preparations (Turina et al., 2004; D'Alessandro et al., 2008); and reconstituted liposome systems (D'Alessandro et al., 2011), indicating that H+-ATPases of E. coli and Rhodobacter capsulatus at low phosphate and ADP concentrations decouple ATP hydrolysis from transmembrane proton translocation. Causey et al. (2003) employed the acetate overflow effect for metabolic engineering of an acetate-producing E. coli. By inactivating the H+-ATP synthase along with disruption of the TCA cycle and elimination of native fermentative pathways, they obtained a strain, which synthesized acetate as its major product, reaching 86% of the theoretical maximum yield. Apart from E. coli, a somewhat similar response to inactivation of ATP synthase has been documented in other biotechnologically important bacteria, like C. glutamicum and Bacillus subtillis (Santana et al., 1994; Koch-Koerfges et al., 2012).

Alternatively, instead of disrupting oxidative phosphorylation, one can decrease the energetic efficiency of metabolism just by stimulating futile hydrolysis of ATP. In their pioneering work (Koebmann et al., 2002a,b) overexpressed the F<sup>1</sup> part of H+-ATPase, which could be used as an universal module for metabolic engineering of intracellular ATP hydrolysis without interfering with any particular metabolic reaction. In E. coli and L. lactis overexpression of F<sup>1</sup> led to a decrease of growth rate and intracellular energy level, and (depending on the strain and culture growth phase) also stimulated glycolytic flux to a various degree, decoupling it from anabolism. Using this approach Liu et al. (2016) demonstrated that futile hydrolysis of ATP could enhance product synthesis via pathways that generate ATP. By overexpressing of F<sup>1</sup> in L. lactis strain engineered for acetoin production they managed to improve the product yield and productivity at the expense of biomass formation. As noted by Holm et al. (2010), the metabolic effects of overexpression of the F<sup>1</sup> part of H+-ATPase and those of the soluble water-forming NADH oxidase had much in common. In both cases for E. coli the result was an elevated glucose consumption and reduced growth rate and biomass yield, although the acetate overflow was seen only with F1, but not with Nox.

Finally, since majority of bacterial electron transport chains are branched (for a review, see Poole and Cook, 2000), in principle it should be possible to decrease P/O also by inactivation of the energetically efficient electron transport branch(es), while leaving the less efficient, low-coupling ones active. However, in several cases inactivation of the energetically efficient routes has resulted in a drop of the bulk activity of mutant respiratory chain, actually causing poor growth and an elevation of overflow metabolism. Reported examples are the E. coli strains with linear electron transport chains (Steinsiek et al., 2014), or aa<sup>3</sup> knock-out mutant of B. subtilis (Zamboni and Sauer, 2003), bearing the least energetically efficient electron transport components, yet accumulating more acetate than the respective wild types. Hence, turning a natural bacterial respiratory chain with an average P/O value into a low-P/O and high-rate one often implies complex genetic modifications and fine-tuning, including knocking out the high-coupling modules and concerted overexpression of low-coupling respiratory chain constituents. Obviously, the best alternative to such complex genetic engineering might be employing bacterial respiratory chains with a naturally low P/O, whenever possible. L. lactis is one of such bacteria, bearing a low energy-coupling, inducible respiratory chain, suitable for oxidizing surplus NADH, while little affecting cellular ATP production (Liu et al., 2017). Notably, although NoxE from this bacterium has been expressed in several microorganisms to elevate their respiration rate (see above), its own inducible respiratory chain proved to be more efficient than NoxE for regeneration of its own intracellular NAD<sup>+</sup> pool and for its oxidative stress protection. The ethanol-producing alphaproteobacterium Z. mobilis is another prominent example of aerobic metabolism with a naturally uncoupled respiration.

## Zymomonas mobilis: AN EXAMPLE OF UNCOUPLED ENERGY METABOLISM

Zymomonas mobilis is a facultatively anaerobic bacterium with a very rapid and efficient homoethanol fermentation pathway (Rogers et al., 1982), involving the Entner-Doudoroff (E-D) glycolysis, pyruvate decarboxylase, and two alcohol dehydrogenase isoenzymes (Sprenger, 1996). Under anaerobic or microaerobic conditions up to 98% of substrate carbon is incorporated in ethanol. Metabolic engineering of Z. mobilis has largely focussed on biofuel production from variety of renewable substrates, as well as on increasing its stress-resistance (Zhang et al., 1995; Mohagheghi et al., 2015; Yang et al., 2016; Charoensuk et al., 2017; for recent reviews, see Wang et al., 2018; Xia et al., 2019). Fermentative catabolism in Z. mobilis is loosely linked to its anabolism and growth, and proceeds with a high rate, both during the exponential growth phase, or under non- or slowlygrowing conditions (Belaich and Senez, 1965). With respect to its metabolism, Z. mobilis somewhat resembles lactic acid bacteria (Sauer et al., 2017), also having high carbon substrate uptake rates with low biomass formation in combination with simple central metabolic pathways, leading to a limited number of metabolites. However, the aerobic energy metabolism in Z. mobilis seems to stand further away from the "bacterial mainstream," than that of lactic acid bacteria.

In contrast to the inducible electron transport of lactic acid bacteria, which requires hemin to be present in growth media (Brooijmans et al., 2007; Koebmann et al., 2008), Z. mobilis bears a constitutive respiratory chain. It consists of the Type II NADH dehydrogenase (ndh), coenzyme Q10, and the cytochrome bd terminal oxidase (cydAB) as the established major electron carriers (**Figure 2**), together with respiratory FAD-dependent D-lactate dehydrogenase (Ldh), PQQ-dependent glucose dehydrogenase (Gdh), a putative cytochrome c peroxidase (PerC), and some other minor or still unidentified constituents (Strohdeicher et al., 1990; Kalnenieks et al., 1998; Sootsuwan et al., 2008). The cytochrome bd seems to be the only terminal oxidase, which is encoded in the Z. mobilis genome (Seo et al., 2005; Yang et al., 2009), and has been detected spetroscopically (Kalnenieks et al., 1998; Sootsuwan et al., 2008). Z. mobilis, like G. oxydans (Hanke et al., 2012) and some Acetobacter species, possesses a "dead-end" cytochrome bc<sup>1</sup> branch, apparently lacking a terminal cytochrome c oxidase, hence its function in electron transport remains obscure. The cytochrome c peroxidase seems more likely to function as a quinol peroxidase (Balodite et al., 2014), than to terminate the bc<sup>1</sup> electron transport branch, as suggested by Charoensuk et al. (2011). Nevertheless, Z. mobilis mutant strain with inactivated cytochrome b subunit of the bc<sup>1</sup> complex shows a distinct respiratory phenotype (Strazdina et al., 2012), best seen in mutant membrane preparations as an increase of antimycinresistance of oxygen consumption and alteration of cytochrome bd reduction kinetics with NADH. In G. oxydans its bc1-deficient strain shows a growth defect under acidic conditions, while in the wild type the qcrABC genes are upregulated under oxygen limitation (Hanke et al., 2012). Further research might reveal novel roles for the bc<sup>1</sup> complex in these alpha-proteobacteria and perhaps, put forward new ideas for their respiratory metabolic engineering.

For some reasons, the energetic efficiency of Zymomonas respiration in aerated culture is significantly lower than reported for other bacteria, e.g., L. lactis, with a similar dominating electron transport pathway from Ndh to CydAB (Brooijmans et al., 2007; Rutkis et al., 2014, 2016), which makes it quite unique among the producer microorganisms. Unlike such prominent facultatively anaerobic workhorses of biotechnology as E. coli and B. subtilis, aeration does not improve Z. mobilis biomass yield (Belaich and Senez, 1965; for a review, see Kalnenieks, 2006). Study of respiratory knock-out mutants (Strazdina et al., 2012; Rutkis et al., 2014) did not indicate presence of any putative low energy-coupling electron transport branch, to which one might ascribe the overall low energetic efficiency. Oxidative phosphorylation has been reported in non-growing cells and cell membrane vesicles, and cells are capable of ATP synthesis with artificially induced proton gradient (Kalnenieks et al., 1993). Yet, the proton-dependent ATPase in Z. mobilis under aerobic growth conditions operates in the direction of ATP hydrolysis (Rutkis et al., 2014). This is possibly because of some

intrinsic deficiency of proton motive force maintenance, or due to an unidentified proton leakage pathway in growing cells. Being an active sink of reducing equivalents, respiration lowers the ethanol yield, since it withdraws NADH from the alcohol dehydrogenase reaction. Hence, respiration causes accumulation of acetaldehyde, the inhibitory metabolic precursor of ethanol, which contributes to the poor aerobic growth of this bacterium, at the same time being a valuable target product of Z. mobilis aerobic bioprocess (Wecker and Zall, 1987). From the practical point of view, respiratory chain of Z. mobilis is well-suited for oxidation of NADH (and NADPH, since its Ndh has a good affinity for both cofactors) without stimulation of biomass growth. Aerobic acetaldehyde production by Z. mobilis so far is the best reported application of its respiratory potential. Recently, acetaldehyde-producing strains were developed, that reach 70% of the theoretical maximum acetaldehyde yield (Kalnenieks et al., 2019), as well as strains with the acetaldehyde-generating pyruvate decarboxylase reaction relocated to periplasm, in order to mitigate toxic effect of acetaldehyde on cell interior (Balodite et al., 2019).

It is not obvious, what are the physiological function(s) of Z. mobilis respiration in nature. Possibly, the respiratory chain might be accelerating glucose consumption by lowering the intracellular NADH/NAD<sup>+</sup> ratio (Rutkis et al., 2016), like it has been shown previously for C. glutamicum (Tsuge et al., 2015). Rutkis et al. demonstrated that an active respiratory chain is advantageous for reaching higher initial glucose uptake rate in non-growing cells, thus improving their survival, when glucose is available in small amounts and for restricted periods of time (most likely, that is what happens during the natural life cycle of this bacterium).

Respiratory chain apparently plays some role in the protection of Z. mobilis against oxic, high temperature, inhibitor and salt stresses. Cytochrome c peroxidase (perC) knock-out mutant of the thremotolerant strain TISTR 548 shows impaired growth at elevated temperature (37 and 39◦ C) (Charoensuk et al., 2011). Hypersensitivity to hydrogen peroxide (although, not as pronounced as in catalase-deficient strain) is another phenotypic trait of Z. mobilis perC mutants (Charoensuk et al., 2011; Balodite et al., 2014). In their recent work with TISTR 548, Charoensuk et al. (2017) presented results of transposon mutagenesis, followed by the mutant screening for the loss of thermotolerance. They identified 26 knock-out mutations, deleterious for thermotolerance, most of which were related to biosynthesis and maintenance of cell membranes and envelope. Yet, among the mutated genes there was the flavoprotein WrbA, a new type (type IV) of NAD(P)H:quinone oxidoreductase (Patridge and Ferry, 2006), thought to function as a protector against oxidative stress. Martien et al. (2019) noted upregulation of all respiratory chain components following exposure of a strictly anaerobic culture of Z.mobilis Zm4 to vigorous aeration, and suggested that the activity of PerC and cytochrome bd limits the toxic effects of oxygen. Interestingly, Ndh activity was shown to confer advantage for Z. mobilis aerobic growth and flocculation on minimal media (Jones-Burrage et al., 2019). Upregulation of ndh and some other oxidoreductase genes were observed under lignocellulose inhibitor stress (Chang et al., 2018). Hayashi et al. (2015) demonstrated positive impact of Ndh activity on Z. mobilis salt stress resistance. Likewise, activity of respiratory chain is positively correlated with acetic acid tolerance in Acetobacter pasteurianus (Wu et al., 2017), with multiple stress tolerance of Lactobacillus plantarum (Zotta et al., 2013), and with resistance of E. coli to critically high growth temperatures (Murata et al., 2018). Details of the mechanisms behind these general stress-protective effects, however, largely remain obscure and require further study.

Paradoxically, turning off Z. mobilis respiration also appears to be beneficial for its aerobic growth. The Z. mobilis ndh knock-out mutant shows improved aerobic growth and ethanol synthesis on rich growth medium (Kalnenieks et al., 2008; Hayashi et al., 2011, 2012), which partly can be explained by the decrease of acetaldehyde accumulation. At the same time, inactivation of the ndh gene elicits an adaptive response at the level of the electron transport chain and several oxidative stress protection systems (Strazdina et al., 2012). In particular, the mutant bears higher activity of respiratory D-lactate dehydrogenase (Ldh). Hayashi et al. (2011, 2012) analyzed a set of spontaneous Z. mobilis respiratory mutants with Ndh-deficiency, and observed their improved aerobic growth and ethanol synthesis at elevated temperature (39◦ C). Recently Strazdina et al. (2018) demonstrated that the elevated Ldh activity of the ndh strain is crucial for its temperature-resistance. Inactivation of ldh against the Ndh-deficient background renders the non-respiring double mutant more temperature sensitive than the ndh strain, bringing it back to the level of the wild type. The protective function of the respiratory Ldh at elevated temperature needs further examination, in order to establish what specific electron pathway(s) starting from Ldh might be involved in Z. mobilis thermotolerance.

# CONCLUSIONS AND OUTLOOK

Engineering of respiratory chain represents a potent lever to modify metabolism of biotechnological producer microorganisms. Rational metabolic engineering of culture growth and product synthesis, based on targeted modification of bacterial respiratory chain branches has been proven successful in numerous cases. At the same time, engineering of electron pathways can elicit complex responses, particularly concerning physiology of stress resistance, which may be difficult to explain within the present picture of electron transport. Modification of respiratory chain components can cause unpredicted effects on growth, catabolism and intracellular redox balance. Some effects, observed under strictly anaerobic conditions, possibly indicate gaps in our understanding of anaerobic electron transport pathways. Apparently, in order to uncover the full potential of respiratory metabolic engineering, more basic research on the structure and regulatory interactions between electron transport and the rest of microbial metabolism is needed. For broadening our knowledge of the alternative physiological roles of respiration, bacteria like Z. mobilis, which do not use their respiration for energy generation, might be helpful as experimental model systems.

## AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### REFERENCES


#### FUNDING

The present work was funded by the Latvian ERDF project 1.1.1.1/16/A/185 and by Latvian Council of Science project lzp-2018/2-0123.


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Kalnenieks, Balodite and Rutkis. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Synthesis and Assembly of Hepatitis B Virus-Like Particles in a Pichia pastoris Cell-Free System

#### Alex J. Spice1,2, Rochelle Aw1,2, Daniel G. Bracewell<sup>3</sup> and Karen M. Polizzi1,2 \*

<sup>1</sup> Department of Chemical Engineering, Imperial College London, London, United Kingdom, <sup>2</sup> The Imperial College Centre for Synthetic Biology, Imperial College London, London, United Kingdom, <sup>3</sup> Department of Biochemical Engineering, University College London, London, United Kingdom

Virus-like particles (VLPs) are supramolecular protein assemblies with the potential for unique and exciting applications in synthetic biology and medicine. Despite the attention VLPs have gained thus far, considerable limitations still persist in their production. Poorly scalable manufacturing technologies and inconsistent product architectures continue to restrict the full potential of VLPs. Cell-free protein synthesis (CFPS) offers an alternative approach to VLP production and has already proven to be successful, albeit using extracts from a limited number of organisms. Using a recently developed Pichia pastorisbased CFPS system, we have demonstrated the production of the model Hepatitis B core antigen VLP as a proof-of-concept. The VLPs produced in the CFPS system were found to have comparable characteristics to those previously produced in vivo and in vitro. Additionally, we have developed a facile and rapid synthesis, assembly and purification methodology that could be applied as a rapid prototyping platform for vaccine development or synthetic biology applications. Overall the CFPS methodology allows far greater throughput, which will expedite the screening of optimal assembly conditions for more robust and stable VLPs. This approach could therefore support the characterization of larger sample sets to improve vaccine development efficiency.

Keywords: cell-free protein synthesis, virus-like particles, Pichia pastoris, synthetic biology, hepatitis B core antigen

# INTRODUCTION

Cell-free protein synthesis was developed as early as the 1960s, where it was critical in determining the genetic code (Nirenberg and Matthaei, 1961). Over the decades, advances in synthetic biology are driving a renaissance in CFPS and it is emerging as a transformative platform technology (Carlson et al., 2012). This pique in interest is mainly due to the capacity to produce protein in a rapid and facile manner, providing benefits to multifarious applications. These include the design of de novo metabolic pathways (Hodgman and Jewett, 2012), personalized medicine (Ogonah et al., 2017) and biosensing (Pardee et al., 2016). CFPS has also shown distinct advantages in producing difficult-to-express proteins, such as those toxic to the cell in vivo (Katzen et al., 2005). Finally, due

#### Edited by:

Jean Marie François, UMS3582 Toulouse White Biotechnology (TWB), France

#### Reviewed by:

Yoshihiro Shimizu, RIKEN, Japan Navin Khanna, International Centre for Genetic Engineering and Biotechnology, India

#### \*Correspondence:

Karen M. Polizzi k.polizzi@imperial.ac.uk

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 30 August 2019 Accepted: 28 January 2020 Published: 14 February 2020

#### Citation:

Spice AJ, Aw R, Bracewell DG and Polizzi KM (2020) Synthesis and Assembly of Hepatitis B Virus-Like Particles in a Pichia pastoris Cell-Free System. Front. Bioeng. Biotechnol. 8:72. doi: 10.3389/fbioe.2020.00072

**Abbreviations:** CFPS, cell-free protein synthesis; CIP, calf-intestinal alkaline phosphatase; CTD, C-terminal domain; DLS, dynamic light scattering; HBc, hepatitis B core antigen; MWCO, molecular weight cut-off; NTD, N-terminal domain; OD600, optical density at 600 nm; TEM, transmission electron microscopy; VLP, virus-like particle.

to the unparalleled access to the reaction environment, optimization and characterization can be achieved with an ease that far exceeds in vivo systems, creating opportunities for high throughput screening and rapid prototyping.

Cell-free extracts based on a variety of host organisms are continually being developed. Arguably, extracts could be prepared from any cultivatable cell-type. The most commonly utilized systems and those which are commercially available are from Escherichia coli (Chen and Zubay, 1983), Chinese hamster ovary cells (CHO) (Brödel et al., 2014), wheat-germ extract (WGE) (Anderson et al., 1983; Madin et al., 2000), rabbit reticulocyte lysate (RRL) (Jackson and Hunt, 1983) and insect (Sf9) cells (Stech et al., 2014). Developing new extracts is still of great interest, unlocking access to advantageous properties of the organisms from which they are derived. The available repertoire of developed CFPS extracts is still expanding and now includes HEK293 (Bradrick et al., 2013) Saccharomyces cerevisiae (Gan and Jewett, 2014), BY-2 tobacco cells (Buntru et al., 2014), Streptomyces venezuelae (Moore et al., 2017), Bacillus megaterium (Moore et al., 2018), Vibrio natriegens (Gan and Jewett, 2014) and most recently Pichia pastoris (Aw and Polizzi, 2019).

Pichia pastoris (syn. Komagataella spp.) has developed over the past three decades to become the most commonly utilized protein expression system after E. coli, for both lab and industrialscale protein production (Bill, 2014). The popularity of P. pastoris largely stems from high volumetric productivity, afforded by the ability of the organism to grow to high cell densities (Darby et al., 2012). In addition, P. pastoris has proven to be highly versatile, with thousands of proteins successfully synthesized to date (Ahmad et al., 2014). It has also become an important expression host for the production of VLP vaccines (Wang et al., 2016).

Virus-like particles are supramolecular protein assemblies, composed of capsid-forming proteins that self-assemble, mimicking the structure of the native viruses from which they are derived (Huang et al., 2017). VLPs are non-infectious and unable to replicate, as they lack a viral genome, removing the concerns about infectivity often associated with live-attenuated or inactivated viruses. This makes VLPs exciting potential vaccine candidates, as they also display strong immunogenicity and the induction of innate and adaptive immunity in animals and humans (Roldão et al., 2010; Crisci et al., 2012). The first VLP vaccine was commercialized in 1986 and was comprised of the hepatitis B virus surface antigen, demonstrating the efficiency and potential of VLP vaccines (Gerety, 1984). In recent years there has been a shift toward engineering increased complexity, with third generation VLPs finding applications in imaging (Manchester and Singh, 2006; Schwarz and Douglas, 2015), catalysis (Patterson et al., 2014, 2015; Maity et al., 2015) and template synthesis (Wnêk et al., 2013). Additionally, VLPs are ideal for the development of platforms for heterologous antigen display, drug encapsulation and chimeric and/or hybrid VLP vaccines (Hill et al., 2017). Advances in synthetic biology are aiding these efforts by widening the repertoire of potential antigens and improving the modularity of these platforms. Despite the attractive attributes, the use of VLPs as vaccines is still hampered by issues related to their production, such as particle integrity, yield, and purification, leading to a demand for new manufacturing approaches to address these challenges (Charlton Hume et al., 2019).

Due to the ability to quickly synthesize protein, CFPS could enable portable and expeditious manufacture of vaccines in response to emerging pathogenic threats or in resource-limited areas. Cell-free systems capable of VLP production may also find use as rapid prototyping platforms for potential VLP vaccine candidates prior to full-scale industrial manufacture or wider nanotechnology applications. Of particular interest is identifying the optimal VLP assembly conditions, which are easier to establish in vitro due to the open nature of the reaction environment. Greater understanding of VLP assembly may help overcome issues related to architectural heterogeneity commonly associated with inconsistent product generation. Additionally, in vivo expression of VLPs can cause cellular toxicity that inhibits growth, such as in the case of the toxic intermediate A2 protein, which was successfully produced and incorporated into Qβ VLPs in a cell-free system (Smith et al., 2012).

A number of CFPS systems have been utilized for the production of VLPs. Hepatitis B virus core antigen VLP has been produced in extracts derived from E. coli (Bundy et al., 2008) and RRL (Ludgate et al., 2016). Human papillomavirus L1 VLP has been produced in S. cerevisiae cell-free systems (Wang et al., 2008) and Human norovirus VLP has been produced in E. coli cell-free systems (Sheng et al., 2017). WGE has been used extensively for the production of VLPs derived from Hepatitis B virus (Lingappa, 1994), Hepatitis C virus (Klein et al., 2004), Human immunodeficiency virus (HIV) (Lingappa et al., 1997), and Simian immunodeficiency virus (SIV) (Dooher and Lingappa, 2004; Lingappa et al., 2005).

The history of P. pastoris as a robust and versatile expression host able to produce VLPs from diverse origins suggests the recently developed CFPS system could be ideally suited as a high-throughput VLP manufacturing and characterization platform. The platform could be specifically suitable for the rapid optimization and validation of DNA constructs and protein variants in vitro, circumventing some of the lengthy cloning procedures associated with strain development, prior to fullscale industrial manufacture in vivo. Additionally, recombinant protein production in the P. pastoris CFPS system is relatively untested. Probing the capabilities of the P. pastoris CFPS system for complex recombinant protein production is therefore a valuable measure of the future potential of the platform for synthesis of more complex biopharmaceuticals.

As a first step toward this goal, the wild-type Hepatitis B virus core antigen (HBc) VLP was selected as a proof-of-concept, as it is one of the best characterized model VLPs. Even though there are no clinically approved native non-chimeric HBc VLP vaccines, HBc VLPs are a popular scaffold molecule for the construction of prophylactic and therapeutic vaccine candidates, many of which are currently undergoing clinical trials (Kutscher et al., 2012). HBc VLPs are an excellent platform for foreign antigen presentation due to their high amenability to foreign insertions, and efficient self-assembly (Pumpens and Grens, 2002) and therefore have been developed for the display of foreign antigens from a vast array of pathogens (Charlton Hume et al., 2019). Some of the chimeric and/or hybrid VLPs have been approved

for clinical or veterinary use; for example, against influenza (De Filette et al., 2008; Fiers et al., 2009; Kazaks et al., 2017) and malaria (Schödel et al., 1994; Sällberg et al., 2002; Nardin et al., 2004). Unlike the HBs-Ag which is an enveloped VLP, the HBc VLP is non-enveloped and therefore structurally less complex. This relative simplicity makes it a good test case for production in a cell-free system. In this study we have developed a rapid method for the production and partial purification of HBc VLPs produced in a P. pastoris cell-free system.

# MATERIALS AND METHODS

# Media and Growth Conditions

Bacterial strains were cultured in Miller lysogeny broth (LB) medium (1% peptone from casein, 0.5% yeast extract, 1% NaCl) supplemented with 37 µg ml−<sup>1</sup> Kanamycin (Sigma−Aldrich, Dorset, United Kingdom). P. pastoris strains were cultured in a rich yeast peptone dextrose (YPD) medium (2% peptone from casein, 1% yeast extract, and 2% dextrose) and with 350 µg ml−<sup>1</sup> Geneticin (VWR, Lutterworth, United Kingdom) for selection. P. pastoris strains were cultured in baffled glass flasks or in 50 ml Falcon tubes at a volume of no more than 20% of the total volume of the vessel starting from an OD<sup>600</sup> of 0.1.

# Strains

Bacterial recombinant DNA manipulation was carried out in E. coli strain NEB 5−α (New England Biolabs (NEB), Hertfordshire, United Kingdom). P. pastoris (syn. Komagataella phaffi) strain FHL1 was described previously (Aw and Polizzi, 2019).

### Plasmid Construction

The CFPS expression plasmid was generated using the Gibson DNA assembly method as described previously (Gibson et al., 2009). The pET−28b vector (Merck UK Ltd., Hertfordshire, United Kingdom) was used as a backbone. The full-length wildtype (WT) HBc core antigen (adw serotype) codon-optimized for P. pastoris and the cricket paralysis virus (CrPV) internal ribosome entry site (IRES) sequences were synthesized by GeneArtTM (Thermo Fisher Scientific, Paisley, United Kingdom). In order to make the expression constructs, desired fragments were amplified with 30 bp of homology using primers purchased from Thermo Fisher Scientific using Phusion <sup>R</sup> High Fidelity DNA polymerase (New England Biolabs). A Kozak sequence (GAAACG) was included in the primers, directly upstream of the coding region. This sequence was selected based on previously displayed efficacy in the production of recombinant proteins in P. pastoris and is recommended in the ThermoFisher PichiaPinkTM Expression System manual. The PCR reaction products were gel extracted using the ZymocleanTM Gel DNA Recovery kit (Zymo Research Corporation, Irvine, CA, United States) prior to the assembly reaction. After confirmation of the correct plasmid assembly, a synthetic 50 bp polyA tail was generated using annealed primers and inserted into the vector by restriction cloning using XhoI and NotI (NEB). The plasmid material generated for CFPS reactions was extracted using the Qiagen Maxi Prep Kit (Qiagen, Crawley, United Kingdom).

## Crude Extract Preparation and Coupled Cell-Free Protein Synthesis

Pichia pastoris cells were grown and extracts were prepared as previously described (Aw and Polizzi, 2019). Briefly, overnight cultures of P. pastoris strain FHL1 were grown in 5 ml of YPD medium and then used to inoculate 200 ml of YPD medium to an OD<sup>600</sup> of 0.1. Cell growth was allowed to proceed to an OD<sup>600</sup> of 18–20 at 30◦C, 250 rpm. Cells were harvested, washed, homogenized and prepared using the previously described methodology (Aw and Polizzi, 2019). Cellfree protein synthesis reactions were performed by a coupled in vitro transcription/translation system at a volume of 50 µL at room temperature for 3 h with no shaking as previously detailed (Aw and Polizzi, 2019).

## In vitro Capsid Assembly and Purification

The protocol for capsid assembly and the assembly reaction mixture was adapted from Ludgate et al., 2016. The assembly reaction mixture contained 3 µL of the product of the coupled CFPS reaction per 10 µL total assembly reaction volume and consisted of 1x restriction buffer 3 (100 mM NaCl, 50 mM Tris–HCl, 10 mM MgCl2, 1 mM dithiothreitol (DTT), pH 7.9; (NEB) supplemented with 1x completeTM, mini, EDTAfree protease inhibitor cocktail (Roche, Basel, Switzerland) and 1 U µL <sup>−</sup><sup>1</sup> Murine RNase inhibitor (NEB). For reactions supplemented with exogenous phosphatase, 1 U µL <sup>−</sup><sup>1</sup> of calf intestinal alkaline phosphatase (CIP; NEB) was supplemented into the assembly reaction mixture. The reaction mixtures were incubated overnight (16 h) at 37◦C. At this stage, samples were then either concentrated or the purification process continued. For concentration, Vivaspin <sup>R</sup> 500, 100 kDa MWCO centrifugal concentrators (Sartorius, Göttingen, Germany) were used according to the manufacturer's instructions and samples were buffer exchanged three times with PBS to a final volume of 50 µL.

For purification, after overnight assembly, the reaction mixture was heated at 60◦C for 1 h to precipitate contaminating proteins followed by centrifugation (5000 × g, 10 min). After centrifugation, the supernatant was collected, and solid ammonium sulfate added to 40% saturation. The mixture was incubated on a laboratory rotator for a minimum of 2 h or overnight at 4◦C. The mixture was centrifuged (10,000 × g, 30 min) and the resulting protein pellet resuspended in a minimal volume of phosphate-buffered saline (PBS).

## SDS-PAGE Gel Analysis

Sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS– PAGE) was performed using 15% Tris–HCl SDS-PAGE gels. Protein samples were denatured by boiling for 15 min in reducing SDS sample buffer (0.0625 M Tris–HCl, pH 6.8, 2.3% (w/v) SDS, 10% (w/v) glycerol, 0.01% Bromophenol blue, 5% (v/v) β-mercaptoethanol). Molecular weight was estimated by using a pre-stained protein ladder (Pageruler, 10–170 kDa, Thermo Fisher Scientific). SDS-PAGE gels were stained using SimplyBlueTM SafeStain (Invitrogen, Carlsbad, CA, United States) according to the manufacturer's instructions.

# Dynamic Light Scattering (DLS) Measurements

fbioe-08-00072 February 13, 2020 Time: 15:18 # 4

Dynamic light scattering measurements were performed using a Malvern Zetasizer Nano ZS instrument (Malvern Instruments Ltd., Worcestershire, United Kingdom). The software utilized to collect and analyze the data was Zetasizer Series 7.13 from Malvern. Forty µL of each sample was analyzed in a 45 µL lowvolume glass cuvette. All measurements were conducted at a controlled temperature of 25◦C and for each sample 15 runs of 70 s were performed with three repetitions.

# Transmission Electron Microscopy (TEM) Analysis

A 3.4 µL sample of purified VLP solution between 0.02–0.05 mg ml−<sup>1</sup> was applied to a carbon coated copper/Formvar grid and negatively stained with 2% w/v uranyl acetate, pH 4. Images were taken by a TEMCAM XF416 (TVIPS, Oslo, Norway) camera in a CM200 (Philips, Amsterdam, Netherlands) electron microscope at an acceleration voltage of 200 kV. ImageJ software (version 1.51) (Maryland, United States) was used to process the TEM images and add scale bars.

# RESULTS AND DISCUSSION

## Synthesis, Assembly and Crude Purification of Hepatitis B Viral Capsids

The full-length wild-type HBc VLP was selected as a proof-ofconcept due to extensive previous characterization and its use as a chassis molecule for novel vaccine designs (Pumpens and Grens, 2001). The icosahedral HBc VLP is composed of multiple copies of a single 21 kDa protein, the HBc protein. The WT HBc monomer can be divided into three distinct regions: an NTD (position 1–140) responsible for capsid assembly, a CTD (positions 149–183 or 185, depending on the strain) responsible for nucleic acid binding, and a linker region (position 140–149). The assembly domain contains the major immunodominant region, comprised of two α-helices forming the spikes on the particles. The full-length HBc forms dimers and the capsid itself is dimorphic, assembling into both the T = 3 (90 dimers) and T = 4 (120 dimers) icosahedral VLPs with diameters of 32 nm and 35 nm, respectively (Crowther, 1994; Roseman et al., 2012). In recent years, HBc VLPs have been used as a versatile chassis for both nanotechnology and vaccine applications (Chackerian, 2007; Ludwig and Wagner, 2007) primarily due to particle stability and integrity.

The full-length HBc gene (amino acids 1–183, codonoptimized for P. pastoris) was cloned into the pET−28b vector (Merck United Kingdom Ltd.) using the Gibson Assembly method (Gibson et al., 2009). Pichia pastoris CFPS reactions were performed to synthesize the HBc in a reaction volume of 50 µL at room temperature without shaking for 3 h.

During our early experiments, we found that the HBc VLPs were unable to self-assemble directly in the CFPS reaction (data not shown). Previous studies have shown that HBc VLP self-assembly in vitro is well known to be highly dependent on monomer concentration (Lingappa et al., 2005; Ludgate et al., 2016). When the concentration of HBc dimer exceeds a critical concentration, nearly all free dimer associates into capsid (Katen and Zlotnick, 2009). The HBc concentration required for assembly in vitro is high (40–80 µM) at physiological salt concentrations (Wingfield et al., 1995; Tan et al., 2013), although spontaneous self-assembly of the HBc VLP has also been shown to be triggered by increasing the salt concentration (Bundy et al., 2008). Based on the previously reported yield of luciferase produced in our system (∼4.5 µM) (Aw and Polizzi, 2019), we postulated that lack of assembly was due to low dimer concentration prohibiting capsid self-assembly. In a report describing HBc VLP production in an RRL-based CFPS system, it was shown that full-length HBc VLP assembly can be achieved at lower dimer concentrations by modulating the phosphorylation state of the CTD (Ludgate et al., 2016). We therefore aimed to recreate the capsid assembly conditions from this study to ascertain whether the HBc synthesized in our system could be induced to self-assemble into capsids, even at concentrations below the self-assembly threshold. To achieve this, we designed a synthesis, assembly and purification protocol for determining the optimum particle assembly conditions (**Figure 1**).

In order to provide adequate time for capsid assembly, the completed CFPS reactions containing HBc were diluted into the assembly reaction mixture and incubated overnight at 37◦C for 16 h. Exogenous CIP, as a broad spectrum phosphatase that can also act on protein, was supplemented into selected samples to dephosphorylate the CTD of the HBc in order to modulate capsid assembly (Ludgate et al., 2016). Since the CTD undergoes non-specific RNA binding, which is modulated by its phosphorylation state, inclusion of an RNase inhibitor in the assembly reaction mixture was also essential to preserve RNA integrity (Ludgate et al., 2016). It is highly probable therefore that dephosphorylation of the CTD promotes non-specific interactions with nucleic acids, facilitating capsid assembly. It has been previously shown that a positively charged CTD interacts with negatively charged nucleic acids to enable capsid assembly at 5 nM (Klein et al., 2004) and the removal of the phosphate may contribute to increasing the local positive charge.

After incubation, centrifugation was conducted to remove aggregates and the supernatant was applied to a 100 kDa MWCO centrifugal concentrator and buffered-exchanged three times against PBS. A high MWCO membrane is frequently used in HBc VLP production as a concentration and rudimentary purification step (Kazaks et al., 2017; Wetzel et al., 2018).

Samples were taken from the retentate of the concentrator and the resultant flow-through for SDS-PAGE analysis (**Figure 2**). The HBc monomer has a molecular weight of 21 kDa, but no band of this size was observed across all samples. A band was, however, observed at approximately 40 kDa for the CIPtreated samples (green arrow) which could be indicative of the HBc dimer (42 kDa). Presence of a band at the expected size of the dimer in the CIP-treated samples suggested HBc VLP

assembly, though further confirmation was required. It is highly likely that the HBc monomer is not visible in the flowthrough of the untreated HBc samples due to dilution. Based on previous concentrations of luciferase produced and the molecular weight of HBc, the concentration is insufficient to meet the criteria for SDS-PAGE detection with SimplyBlueTM SafeStain (Invitrogen). It is worth noting that the other prominent band found in the CIP-treated samples at approximately 70 kDa is the CIP monomer (red arrow), which is added at very high concentrations.

## Characterization by DLS

To validate HBc VLP assembly, retentate samples taken after centrifugal concentration were subjected to morphological analysis. Samples of 40 µL were obtained from the concentration step and transferred to a 45 µL low-volume glass cuvette. Using DLS, the diffusion of the particles moving under Brownian motion was measured and the particles sizes of VLPs calculated. Both the CIP-treated and untreated HBc samples gave a positive result from the DLS characterization (**Figure 3**). The untreated HBc sample (**Figure 3A**) displayed a Z-average particle size of 36.89 nm. Considering our initial hypothesis that VLPs would be unable to assemble as a result of low dimer concentration, this was an unexpected result. It is possible that the concentration of monomer produced in our CFPS system is sufficient for capsid self-assembly, but only when subjected to the conditions provided by the assembly buffer. The beneficial conditions promoting capsid assembly could be the increased salt concentration of the assembly buffer, which was previously shown to trigger capsid self-assembly (Bundy et al., 2008), the inclusion of RNase inhibitor, which preserves RNA integrity thereby facilitating co-localization of free dimer, or a combination of the two. The CIP-treated HBc sample (**Figure 3B**) displayed a Z-average particle size of 37.97 nm. Both observed Z-average particle sizes are close to the reported size of the T = 4 HBc VLP (∼35 nm) (Crowther, 1994; Zlotnick et al., 1996). The T = 4 capsid has been reported as the most abundant product of WT HBc produced by in vitro self-assembly (∼95%) (Zlotnick et al., 1999).

#### Characterization by TEM

In order to verify that the products of our rapid synthesis, assembly and purification protocol contained correctly formed

HBc VLPs, TEM with 2% uranyl acetate negative staining was used (**Figure 4**). The average diameter for the untreated HBc VLPs was 29.3 nm (SD = 1.4 nm, n = 13) and from the CIP-treated HBc VLPs the average diameter was 28.9 nm (SD = 1.8 nm, n = 19). The average size of particles from both samples correlate with previously reported diameters of 30.7 nm from HBc VLPs produced using in vitro methods (Bundy et al., 2008) and 30 nm from in vivo methods (Zlotnick et al., 1996). Interestingly,

supplemented with CIP (CIP-treated assembly). Scale-bar size: 100 nm.

a greater number of fully assembled particles were visible in the CIP-treated sample relative to the untreated sample. This could explain why a faint band with a size corresponding to the dimer was visible in the SDS-PAGE analysis for the CIPtreated sample (**Figure 2**, Lane 2), but no band corresponding to the expected size of the monomer or dimer was observed for the untreated samples (HBc) (**Figure 2**, Lane 1). As previously discussed, dephosphorylation of the CTD likely boosts capsid assembly by promoting non-specific interactions with nucleic acids, facilitating capsid self-assembly. Despite this, there was no great observable difference in average particle size or morphology with both samples still displaying a proportion of partially assembled or disassembled particles, in addition to other smaller protein contaminants. It is possible the smaller contaminants are HBc monomer, CIP or host-cell protein carried through the purification. The presence of host-cell contaminants correlates with the SDS-PAGE analysis (**Figure 2**) suggesting further development of the purification protocol was needed.

# Heat-Treatment and Ammonium Sulfate Purification of Assembled HBc VLPs

Once the successful synthesis and assembly of HBc VLPs was demonstrated, the next aim was to develop an improved methodology for VLP purification (**Figure 5**). We decided to develop this process using the CIP-treated samples, as the TEM characterization indicated an increased average number of

particles relative to the untreated samples. In order to keep in line with creating a simple and rapid purification process, we strived to avoid more time-consuming purification steps, such as chromatography, and those that are costly and poorly scalable for production purposes, such as ultracentrifugation. It has been reported that an efficient method of removing host-cell protein contaminants is the inclusion of a heat treatment step which is both fast and simple (Ng et al., 2006; Lee et al., 2015; Li et al., 2018). As fully-assembled HBc VLP has been shown to be highly thermostable (Ng et al., 2006; Lee and Tan, 2008), heat-labile hostcell proteins can be denatured and precipitated from solution. Toward this end, we removed the centrifugal concentrator step and included a 1 h incubation at 60◦C after the initial assembly reaction in order to remove protein contaminants and improperly assembled HBc VLPs. After heating, the reaction mixture was subjected to centrifugation and the supernatant collected for ammonium sulfate precipitation.

Ammonium sulfate precipitation is a well-established method for separation of HBc VLPs and is widely used as a preparatory purification step (Palenzuela et al., 2002). Particles were precipitated by addition of solid ammonium sulfate to 40% saturation (Yang, 2005) followed by incubation at 4◦C on a rotator for a minimum of 2 h. The reaction was then centrifuged, and the resultant precipitate resuspended in a minimal volume of PBS, thereby purifying and concentrating the HBc VLPs.

To validate the efficacy of our new purification method, samples were taken from the precipitate and supernatant from each step in the purification process for SDS-PAGE analysis (**Figure 6**). The initial heat treatment step, included to denature and precipitate less thermostable host-cell proteins, appears to remove a large number of contaminants as shown in Lanes 5 and 6 which correspond to the resuspended precipitates from the heat treatment of the HBc assembly reaction containing CIP and negative control. Heat treatment alone does result in sufficient purification to detect the presence of a band at approximately 20 kDa in the supernatant of the HBc assembly reaction (Lane 7), which correlates to the expected size of the HBc monomer (21 kDa). There are still a number of other protein contaminants present, including a band at 70 kDa indicative of CIP. CIP cannot be heat inactivated for removal unless heated to 80◦C, which would cause dissociation of the assembled HBc VLPs

(Li et al., 2018). Despite the presence of CIP, we note that much of the CIP is removed during the improved purification process. For example, in **Figure 7** the CIP band in lane 7 (representing CIP remaining in the sample after heat treatment alone) is much more intense relative to that of the HBc monomer than the CIP band in lane 3 (representing CIP remaining after both heat treatment and ammonium sulfate precipitation) relative to the HBc monomer.

To improve the purity of the VLPs, a number of possible strategies could be employed. The concentration of CIP used in this study was selected based on that previously reported (Ludgate et al., 2016); however, a lower concentration may still

be sufficient for CTD dephosphorylation to promote assembly and would reduce the amount of CIP contaminant present in the sample after purification of the HBc VLPs. Alternatively, a heat-labile alkaline phosphatase such as one derived from Shrimp (Nilsen et al., 2001) could be used instead of CIP (Mehta et al., 2009; Crawley et al., 2011), enabling its irreversible removal from the sample via heat inactivation at 65◦C. This would enable removal of the phosphatase alongside the heat-labile host-cell proteins during the heat inactivation step of the purification procedure. Finally, a hexahistidine, FLAG, or other affinity purification tag could be fused to recombinantly produced CIP allowing its removal with a single chromatography step.

After adding the additional ammonium sulfate precipitation step, there are two prominent bands remaining in the HBc assembly reaction containing CIP sample (Lane 3) at the expected band size of the HBc monomer (blue arrow) and CIP protein (red arrow). It is possible that some of the CIP protein remains bound to HBc during the purification process or it may become encapsulated as the VLP forms, leading to its persistence. For example, it was recently observed that an influenza VLP contained an internal component density postulated to be encapsulated cellular components (McCraw et al., 2018). The CIP protein is also visible in the negative control sample containing CIP (Lane 4), along with other higher molecular weight contaminants.

The presence of a single band indicative of the HBc monomer in **Figure 6** compared to the observation of a band indicative of the HBc dimer in **Figure 2** could be due to a combination of factors. Firstly, there is large disparity in the total protein concentration of the samples with those in **Figure 2** having a much greater total protein concentration compared to those in **Figure 6**. It is possible, therefore, that the amount of SDS and/or beta-mercaptoethanol present in the sample to solubilize and fully denature the amount of protein was insufficient. Secondly, the HBc dimer in **Figure 2** could be due to re-dimerization of the HBc monomer between sample preparation and loading. It has previously been reported that disulfide re-oxidation can occur during SDS-PAGE (Kumar et al., 1993; Grigorian, 2005) and both the HBc monomer and dimer have been previously observed to co-occur (Storni and Bachmann, 2004).

The total protein concentration of the purified HBc VLP sample was obtained by measuring protein absorbance at 280 nm (A280) and found to be 68 µg ml−<sup>1</sup> . Densitometry analysis conducted using the ImageJ software was used to estimate the yield of HBc VLP in our system at 6.4 µg ml−<sup>1</sup> . This is a slight improvement on the reported yield of HBc VLP of 4 µg ml−<sup>1</sup> previously produced in a WGE system (Lingappa et al., 2005). The yield is currently considerably lower in our system compared to production of HBc VLP in vivo, with reported yields of assembled HBc VLP in P. pastoris up to 3 mg per 1 g wet cell weight (Freivalds et al., 2011).

#### Characterization of Purified HBc VLPs by TEM

We next sought to determine whether the product of our improved purification protocol contained correctly formed HBc VLP. To achieve this, TEM with 2% uranyl acetate negative staining was performed (**Figure 7A**). A number of fully formed HBc VLPs can be observed with an overall increase in the uniformity of their morphology compared to **Figure 4**. This indicates removal of partially assembled or disassembled HBc VLPs in the developed purification scheme relative to the initial TEM analysis using the crude purification method (**Figure 4**). A number of misfolded or aggregated proteins still persist in the sample, however. Based on the SDS-PAGE analysis in **Figure 6**, we postulate that these are likely to be aggregates of CIP, as only fully formed VLPs remain in the supernatant after heat treatment (Ng et al., 2006). An additional TEM grid was prepared at a lower sample concentration of 5 µg ml−<sup>1</sup> . Staining was followed by three washes in filtered distilled water for 1 min each in order to attain single particle imaging for accurate size estimation (**Figure 7B**). The wash steps were included in the staining procedure to remove excess uranyl acetate stain and led to a reduction in stain permeation of the HBc VLPs (**Figure 7B**). The observed diameter for the displayed particle was approximately 29.5 nm. The reported size correlates with previously reported diameters from HBc VLP produced using in vitro methods (Bundy et al., 2008), in vivo methods (Zlotnick et al., 1996) and our previously shown assembled particle diameter of 29.4 nm using the crude purification method.

#### CONCLUSION

Cell-free protein synthesis is an incredibly powerful and versatile tool in synthetic biology, which can be applied in myriad ways. We have generated a rapid HBc VLP synthesis, assembly and purification scheme demonstrating the potential of the P. pastoris CFPS system for the production of complex supramolecular assemblies. This approach could enable ultra-scale-down rapid prototyping of protein variants for novel vaccine technologies.

Additionally, by exploiting the open nature of the reaction environment, the conditions for assembly can be optimized, alleviating the current challenges of structural heterogeneity in VLP manufacture. We plan to expand on this work, further testing the capabilities of the system by producing VLPs of increased complexity. In future, we envision increased effectiveness in response to emerging pathogenic threats, through high-throughput vaccine development approaches.

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the article/supplementary files.

#### AUTHOR CONTRIBUTIONS

AS designed and performed the experiments. RA and KP helped to design the experiments. KP and DB helped to conceive the

#### REFERENCES


study. All authors helped to draft the manuscript and read and approved the final manuscript.

#### FUNDING

This project was funded in part with UK aid from the UK government, and the UK Department of Health and Social Care is gratefully acknowledged for funding the Future Vaccine Manufacturing Research Hub through the Engineering and Physical Sciences Research Council (EPSRC, grant number: EP/R013764/1). AS would also like to thank the EPSRC Centre for Innovative Manufacturing in Emergent Macromolecular Therapies (EP/I033270/1) for Ph.D. studentship funding.

#### ACKNOWLEDGMENTS

The authors would like to thank Mr. Ciaran McKeown and Mr. Paul Simpson for assistance with DLS and TEM, respectively.

by Providing a Ligand for an allosteric binding site in the α-kinase domain. J. Biol. Chemi. 286, 2607–2616. doi: 10.1074/jbc.M110.177014




analysis, and in vitro capsid assembly. Biochemistry 34, 4919–4932. doi: 10. 1021/bi00015a003


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Spice, Aw, Bracewell and Polizzi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Synthetic Biology Applied to Carbon Conservative and Carbon Dioxide Recycling Pathways

#### Jean Marie François 1,2 \*, Cléa Lachaux <sup>2</sup> and Nicolas Morin1,2

<sup>1</sup> Toulouse Biotechnology Institute (TBI), Université de Toulouse, CNRS, INRA, INSA, Toulouse, France, <sup>2</sup> Toulouse White Biotechnology Center (TWB), Ramonville-Saint-Agne, France

The global warming conjugated with our reliance to petrol derived processes and products have raised strong concern about the future of our planet, asking urgently to find sustainable substitute solutions to decrease this reliance and annihilate this climate change mainly due to excess of CO<sup>2</sup> emission. In this regard, the exploitation of microorganisms as microbial cell factories able to convert non-edible but renewable carbon sources into biofuels and commodity chemicals appears as an attractive solution. However, there is still a long way to go to make this solution economically viable and to introduce the use of microorganisms as one of the motor of the forthcoming bio-based economy. In this review, we address a scientific issue that must be challenged in order to improve the value of microbial organisms as cell factories. This issue is related to the capability of microbial systems to optimize carbon conservation during their metabolic processes. This initiative, which can be addressed nowadays using the advances in Synthetic Biology, should lead to an increase in products yield per carbon assimilated which is a key performance indice in biotechnological processes, as well as to indirectly contribute to a reduction of CO<sup>2</sup> emission.

Keywords: microbial physiology, metabolic engineering, synthetic biology, carbon dioxide, bio-based products, chemicals

# INTRODUCTION

The global warming is caused by the anthropogenic emission of greenhouse gases (GHGS), among which emission of carbon dioxide (CO2) is broadly accepted by the scientific community as the most contributing factor to this climate change. While Nature has been orchestrated with a natural carbon cycle, the industrial revolution that begun in the early 19th century resulted in a rise of CO<sup>2</sup> emission that slowly but definitively exceeded the natural geochemical carbon cycle. According to the Keeling curve, which reports the daily carbon dioxide measurements at Mauna Loa observatory, the CO<sup>2</sup> emission has dramatically increased over the last 50 years, mostly due to deforestation and burning of fossil fuels (https://scripps.ucsd.edu/programs/keelingcurve/). These issues are now becoming critical for the human being, which urged the scientific community to find solutions to decrease our reliance on fossil fuels that should eventually change our life style. Biofuels and more generally biorefinery have emerged as promising solution whose purpose is to exploit microorganisms as cell factories to convert non-edible but renewable carbon sources such as lignocellulosic sugars into bioethanol as alternative energy (Duwe et al., 2019; Rosales-Calderon and Arantes, 2019) or into commodity chemicals that can replace those obtained from petrochemistry (Clark et al., 2012; Chen and Dou, 2016; Singhvi and Gokhale, 2019). In spite

#### *Edited by:*

Jens O. Krömer, Helmholtz Centre for Environmental Research (UFZ), Germany

#### *Reviewed by:*

S. Venkata Mohan, Indian Institute of Chemical Technology (CSIR), India Robert Mans, Delft University of Technology, Netherlands

#### *\*Correspondence:*

Jean Marie François fran\_jm@insa-toulouse.fr; jean.marie-francois@inra.fr

#### *Specialty section:*

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

*Received:* 27 August 2019 *Accepted:* 11 December 2019 *Published:* 10 January 2020

#### *Citation:*

François JM, Lachaux C and Morin N (2020) Synthetic Biology Applied to Carbon Conservative and Carbon Dioxide Recycling Pathways. Front. Bioeng. Biotechnol. 7:446. doi: 10.3389/fbioe.2019.00446 of the apparent sustainability of this solution, the use of chemoorganotrophic microbes such as yeast or E. coli present a caveat, as CO<sup>2</sup> is released during the process of carbon metabolism (**Figure 1**). A wonderful alternative would be to exploit autotrophic microorganisms such as acetogenic anaerobic bacteria or microalgae (Schiel-Bengelsdorf and Durre, 2012; Scaife et al., 2015) as they have the capability to capture atmospheric CO2. However, these biological systems are at the moment industrially inefficient, due to their slow growth, poor productivity and low energy conversion yield (Claassens, 2017). The purpose of this mini-review is to expose and discuss original strategies that have emerged recently and that have employed synthetic biology tools to rewire the carbon metabolism of heterotrophic microorganisms to achieve maximal carbon conservation during their metabolism. Basically, this endeavor can be reached by either redesigning carbon metabolism network to optimize carbon conservation or recapturing carbon loss using CO<sup>2</sup> fixing or carboxylating systems. Autotrophic CO<sup>2</sup> fixation by reductive pentose phosphate cycle (rPP also known as Calvin-Benson cycle) as well as other natural metabolic pathways that perform carbon fixation including reductive TCA cycle, 3-hydroxypropionate cycle, 3-hydroxypropionate/4 hydroxybutyrate cycle, and dicarboxylate/4-hydroxybutyrate cycle are not considered in this mini-review, since they have been extensively reviewed in previous excellent papers (Berg et al., 2010; Fuchs and Berg, 2014; Erb and Zarzycki, 2018).

## REWIRING CENTRAL CARBON METABOLISM OF HETEROTROPHIC ORGANISMS FOR CARBON CONSERVATION

### Complete Carbon Conservation of Sugar Metabolism by the Non-oxidative Glycolysis Pathway

Glycolysis or the Embden-Meyerhof-Parnas pathway (EMP) is a fundamental metabolic pathway in most living systems that decomposes sugars into pyruvate and recovers energy of this breakdown into ATP and reducing equivalents NADH. To fuel the cell with some essential anabolic precursors, pyruvate has to be decarboxylated into acetyl-CoA. This decarboxylation step releases carbon dioxide in the environment, resulting in 33% loss of carbon yield. This wasted CO<sup>2</sup> may have major impact on the overall economy of bio-based products derived from fermentable carbon sources. This 33% carbon loss due to the decarboxylation of pyruvate has been challenged recently by Liao's group (Bogorad et al., 2013) who constructed a synthetic pathway termed "the non-oxidative glycolysis (NOG)" that can overcome this carbon loss, leading to a conversion of one mole of glucose to 3 moles of acetyl-moieties (**Figure 1** and **Table 1** for the stoichiometric equation). The logic of NOG relies on the phosphorylating cleavage of sugar phosphates



<sup>\*</sup>Gibbs energy value of the reaction was calculated using eQuilibrator (at http://equilibrator. weizmann.ac.il/).

by a phosphoketolase (Fxpk), which, combined with a carbon rearrangement cycles, creates a cyclic pathway with F6P as the input molecule (**Figure 1**). The irreversibility of this pathway is ensured by the phosphoketolase reaction that cleaves Xu5P into glyceraldehyde-3-P (GAP) and AcP and by the fructose-1,6-bisphosphatase reaction (Fbp) which recycles back F6P from FBP. The carbon rearrangement also involves transaldolase (Tal) and transketolase (Tkt) and the net result of this FBP-dependent cycle is the irreversible formation of three acetyl-phosphate (AcP) molecules from one F6P molecule. As phosphoketolase also displays phosphorylating cleavage activity on F6P (Tittmann, 2014), cleaving the latter molecule into AcP and E4P, a sedoheptulose bisphosphate (SBP)-dependent network can be conceived yielding to the same solution. In this case, this cycle will not need transaldolase (tal gene) but will require a SBP-aldolase to condense DHAP and E4P into SBP, and a sedoheptulose phosphatase (Sbp) to recycle back S7P. Such cycle can potentially exist since Sbp is essential in the Calvin-Benson cycle, and has been identified in yeast as being implicated in ribogenesis (Clasquin et al., 2011) and in methylotrophic bacteria where it participates in the ribulose 5-monophosphate (RuMP) cycle (Stolzenberger et al., 2013).

In their Nature report, Liao and coworkers succeeded in demonstrating the in vitro and in vivo functioning of NOG (Bogorad et al., 2013). The in vitro system required the core of eight enzymes, which were either purchased when commercially available or produced by expression in E. coli. This was the case for phosphoketolase which is encoded by fxpk gene in Bifidobacterium adolescensis (Yin et al., 2005) and for some enzymes of the pentose phosphate pathway (PPP). In accordance with the pathway model, F6P initially added at 10 mM was readily and completely converted into 30 mM AcP. To validate the in vivo pathway, they chose to use xylose as the initial carbon substrate, as the phosphotransferase-dependent uptake and phosphorylation of glucose is not compatible with NOG and also because xylose does not cause repression of the fbp encoding fructose-1,6-bisphosphatase (Sedivy et al., 1984), which is needed together with the overexpression of fxpk. Other enzymes of the NOG pathway were natively expressed on a high copy plasmid, including pta and ackA which codes for phosphate acetyl transferase and acetate kinase, respectively, and which convert AcP into acetate. In addition, competitive fermentative pathways were disabled (i.e., deletion of ldhA, pflB, adhE, and frdBC to prevent production of lactate, formate, ethanol, and succinate) and the experiment was carried out under anaerobic condition to avoid further oxidation of acetate in the TCA cycle. Under this condition, 2.2 moles of acetate per mole of xylose were produced by the engineered E. coli equipped with the NOG pathway. In theory, the maximal yield that can be obtained by this conversion process should be 2.5. This yield termed the thermodynamic feasibility "Y<sup>E</sup> " can be calculated as the ratio between the degree of reduction of the substrate γ<sup>s</sup> to that of the product γp. Y<sup>E</sup> is independent of the stoichiometry of the pathway and provides yield limits of the thermochemical process whatsoever the followed pathway or the catalyst used (Dugar and Stephanopoulos, 2011). In this case, Y<sup>E</sup> of acetate from xylose (γ<sup>s</sup> = 20) is 2.5 since γ<sup>p</sup> of acetate = 6, whereas the stoichiometric yield "Yst" is 1,66. Therefore, the efficiency of the NOG pathway which can be calculated as the ratio of Y<sup>E</sup> to Yst is 50% better than the classical EMP pathway. The fact that the in vivo conversion of xylose into acetate by the engineered E. coli expressing the NOG pathway only reached 88% of Y<sup>E</sup> was due to some by-products formation, notably succinate.

According to the stoichiometric equation (**Table 1**), each mole of acetyl-CoA produced from glucose by NOG can be converted into acetate, which is accompanied by the production of one ATP. Therefore, a net balance of 2 ATP and 3 acetate molecules can be obtained from glucose by NOG. However, this pathway is not compatible for growth as it can generate neither the 12 metabolic building blocks nor the reducing power needed for synthetizing all cellular constituents. These features may explain why NOG has not been retained as a viable pathway during evolution. Nevertheless, in a second seminal paper published in PNAS, Liao's group succeeded in evolving a E. coli strain that could grow while relying on NOG for carbon catabolism (Lin et al., 2018). To achieve this challenge, they blocked glycolysis at the level of glyceraldehyde-3-P dehydrogenase and phosphoglycerate kinase by deleting the corresponding genes as well as the methylglyoxal and the Entner-Doudoroff pathways to prevent bypass of glucose into pyruvate. The glyoxylate shunt and gluconeogenesis was then upregulated to enable conversion of acetyl-CoA to metabolic intermediates required for growth, whereas the reducing power and ATP were derived from TCA, which could be done only under aerobic condition. Since these genetic interventions could potentially generate futile cycles, notably at the level of F6P/FBP and pyruvate/ acetate, pfkA gene encoding the major phosphofructokinase and poxB encoding pyruvate oxidase were deleted. The final strain termed PHL13 contained 10 genes deletion and the 2 overexpressed genes pck and fxpk encoding PEP carboxykinase and phosphoketolase. This engineered strain was still unable to grow on glucose unless acetate was present. It was therefore subjected to evolution by serial transfer to selective media and meanwhile, a quest for the limiting enzymes in the NOG cycle was carried out by devising a clever whole-pathway assay in crude extract of the engineered strains. Practically, the assay was to measure the rate of AcP production from F6P in a crude extract augmented with a mix of all purified NOG enzymes and determine the effect of removing a particular enzyme from this mix on AcP production. This study identified phosphoketolase as the most limiting enzyme followed by transketolase and transaldolase. Therefore, these genes were expressed on a high copy plasmid that also contained genes encoding the glucose permease (glf) from Zymomonas mobilis, the glucokinase (glk), and a AMPinsensitive PEP-activated FBPase (glpX) from E. coli (Donahue et al., 2000). Strain PHL13 that carried out this constructed plasmid was then subjected to further adaptive laboratory evolution (Mattanovich et al., 2014; Jang et al., 2019) to get rid of its dependence to acetate and rely only on glucose for growth. After further improvement of transketolase activity by the whole-pathway assay, these authors could isolate a slow growth colony NOG21 on glucose minimal medium, which was further evolved to yield a faster growing colony NOG26. Use of <sup>13</sup>C glucose labeled on C3 and C4 confirmed that the growth was accomplished by NOG since acetate was labeled on its C1 and C2, as expected. Genome sequencing of this evolved strain unexpectedly unraveled a reduced expression of fxpk and pck. This result could be explained by a fitness response of the strain to the growth condition in order to avoid "kinetic traps" that would cause imbalance of the metabolic flux, as for instance by draining out all oxaloacetate from the TCA cycle or Xu5P from the NOG cycle. Other mutations were identified in the metabolic network such as a deletion of pykF encoding the major pyruvate kinase or mutation in pts genes, which likely reduced the PEP dependent phosphorylation of glucose, leaving thus more PEP for gluconeogenesis. All these genomic modifications likely resulted in a fine-tuning of pathway regulation that would be difficult to predict a priori.

From a biotechnological perspective, the potential of NOG is very limited unless this pathway is coupled with another to make compounds other than acetate. An example would be to establish a NOG- based reductive fermentation in which additional reducing agents such as molecular hydrogen is provided together with sugars. In this condition, 3 moles ethanol can be produced per mole of glucose which would increase by 50% the maximal yield from natural sugar fermentation pathway. The challenge here will be to express hydrogenase to allow the input of additional reducing equivalents. Another practical use of NOG shall be to combine with the Calvin-Benson-Bassham (CBB) cycle (Calvin, 1962) for the biosynthesis of acetyl-coA derived products, such as 1-butanol and fatty acids by autotrophic organisms (Liu et al., 2011). This metabolic rewiring could represent a 50% increase of CO<sup>2</sup> into acetyl-CoA over the native pathway, concomitantly reducing the requirement of RuBisCO activity to reach the same carbon yield (Tcherkez et al., 2006).

## The Pentose-Bifido-Glycolysis Cycling Pathway to Solve the Lack of Reducing Power in NOG

A serious drawback of the NOG pathway is its inability to generate reducing power that is essential for the biosynthesis of added-value molecules from acetyl-CoA such as isoprene, fatty acids or polyhydroxybutyrate. A first approach to overcome this limitation was proposed by Opgenorth et al. (2016) who designed a synthetic in vitro pathway termed "Pentose-Bifido-Glycolysis" cycle (PBG) for the synthetic conversion of glucose into PHB via acetyl-CoA. This PBG cycle is divided in three metabolic parts. The first part is the phosphorylation of glucose into G6P which enters the cycle through the oxidative branch of the PPP and breaks down into Xu5P while producing NADPH. In a second phase, Xu5P is split into AcP and GAP by the phosphoketolase Fxpk. While AcP can be converted into acetyl-CoA by the CoA-phosphotransferase enzyme (Pta), GAP is recycled in a third phase into G6P. This occurs by the condensation of GAP and DHAP into FBP by the aldolase Fba, followed by a process involving a unique phosphofructokinase that catalyzes the ATP-dependent formation of F6P from FBP. At variance to the traditional phosphofructo-1-kinase reaction (Uyeda, 1979), this E. coli PFK encoded by pfkB was found to work reversibly and thus to regenerate ATP needed for glucose phosphorylation. This last phase enables to complete the cycle in ATP neutral manner. Accordingly, for each glucose entering the PBG cycle, two moles of acetyl-CoA, 2 moles of CO<sup>2</sup> and 4 moles of NADPH is produced at zero cost of ATP (**Table 1**). However, there are two caveats with this synthetic pathway design. First the phosphoketolase Fxpk can also split F6P into E4P and AcP, which could abrogate the functioning of the cycle. Secondly, there is large excess of NADPH, as the synthesis of one PHB monomer only requires one NADPH, which therefore could lead to a feedback inhibition of the pathway. To solve the first issue, these authors (Opgenorth et al., 2016) complemented the pathway with PPP enzymes, namely transketolase (Tkt), transaldolase (Tal) and ribose-5-P isomerase (Rpi) in order to recycle E4P into the cycle (**Figure 2B**). The second problem was disentangled by introducing a "NAD(P)H purge valve" to regulate the build-up of NADPH. The outcome of this sophisticated purge valve is to uncouple carbon flux from NADPH production enabling a non-stoichiometric pathway production of NADPH. The construction of this purge valve requires the presence of NAD+-dependent glucose-6-P dehydrogenase (Zwf) and NAD+-dependent phosphogluconate dehydrogenase (Gnd) enzymes together with their native NADP<sup>+</sup> dependent counterparts and a H2O-forming NADH oxidase (NoxE). While NADPH is reoxidized during PHB production, high level of NADPH would not penalize the cycle since it is taken over by the NAD+-dependent enzymes which continue to provide the precursor acetyl-CoA while NADH is readily reoxidized by NoxE. This original cell-free system production of bio-based chemicals from glucose turned out to be quite efficient as the authors reported a maximum productivity of the PBG cycle of 0.7 g PHB/L/h with a yield of 90%. There is great hope in developing such a process at an industrial scale since cell-free system can overcome common problems afflicting in vivo biological systems, such as building-up toxic intermediates, low productivities due to competing pathways and undesirable byproducts (Dudley et al., 2015). However, cell-free systems require to solve major technical and economic problems including production, purification, and stability of enzymes or their potential recycling.

Another approach to disentangle the failure of the NOG pathway to generate redox power was proposed by Wang et al. (2019) through the in vivo construction of an "EP-Bifido" pathway. This synthetic pathway is very similar to PGB as it combines EMP and PPP pathway with the expression of phosphoketolase enzymes acting on F6P and Xu5P, but it does not employ PFK reaction to regenerate the ATP that is needed to phosphorylate glucose (**Figure 2B)**. As indicated in **Figure 1**, the major difference between NOG and the EP-Bifido is that the latter implicates the oxidative portion of the PPP, which leads to the production of NADPH. However, this metabolic process occurs at the expense of one CO2. The remaining reaction steps share those of the NOG with notably Fbp that provides the second driving force and recycles back F6P into the cycle. In theory, this pathway design should yield 2.5 acetyl CoA per mole glucose with two moles of NAD(P)H at the cost of 1 ATP (see **Table 1**). However, it may have the capacity to achieve an even higher yield of acetyl-CoA, as this will depend on favoring G6P into PPP with respect to EMP. In theory, 2.66 moles of acetyl-CoA could be produced from 1 glucose if 66% of G6P goes through PPP. To create this EP-Bifido pathway and expect to attain this optimal carbon yield, the following genetic modifications were carried out in E. coli. On the one hand, zwf and gnd were overexpressed to direct glucose into the PPP together with fxpk encoding F6P/Xu5P phosphoketolase from Bifidus adolescentis. On the other hand, the engineered strain was deleted for pfkA encoding the major phosphofructokinase, edd/eda encoding enzymes of the Entner-Doudoroff and ackA encoding acetate kinase to avoid, respectively, ATP wasteful by FBP/F6Pcycle, the production of pyruvate that could bypass the EP-Bifido pathway and the loss of AcP into acetate. Altogether, the engineered E. coli strain equipped with the EF-Bifido pathway grew 3 times slower than the wild type strain, did not produce any acetate and released 50 to 70% less CO<sup>2</sup> than the wild type strain. The performance of the EF-Bifido pathway in acyl-CoA derived polydroxybutyrate (PHB), fatty acids or mevalonate was moreover demonstrated by showing that the yield of these products was improved by 145, 56, and 48% respectively in engineered strains as compared to the control strains (Wang et al., 2019), arguing for a biotechnological relevance of reengineering the carbon metabolism of E. coli with this pathway.

## The Glycoptimus Pathway to Convert Sugars Into Glycolic Acid Without Carbon Loss

While the refactoring of central carbon metabolic pathway was devoted to maximal acetyl-moieties from glucose, the glycoptimus pathway deals with a rewiring of the carbon metabolism to achieve maximal yield of glycolic acid (GA) from C5 and C6 sugars without carbon loss (**Figure 1**). According to thermodynamic calculation (Dugar and Stephanopoulos, 2011), the Y <sup>E</sup> of glycolic acid (γ<sup>p</sup> = 6) from glucose (γs = 24) or from pentose (γ<sup>p</sup> = 20) should be 4 and 3.3, respectively, whereas the maximal yield Y<sup>S</sup> based on the pathway stoichiometry is 2 from glucose and 1.66 from pentose (Dolan and Welch, 2018). Using the natural pathway, no <13 genetic modifications were executed to achieve a production of 52 g/L glycolate from glucose at a yield of 50 % of the Yst (Soucaille, 2007). Further genetic modifications were proposed by Deng et al. (2018), by combining Dahms pathway with the glyoxylate shunt, enabling a yield close to 90% of the stoichiometric yield. The problem to reach the thermodynamic Y<sup>E</sup> lies at two levels. A first one is to avoid the loss of CO<sup>2</sup> at the pyruvate decarboxylation step, while the second is more challenging as it requires to capture one carbon mole as CO2. Since direct CO<sup>2</sup> fixation by heterotrophic system is still very complicated as experienced by Milo and coworkers in E. coli (Antonovsky et al., 2016), we rewired the carbon central metabolism of E. coli in order to overcome the loss of CO<sup>2</sup> during the conversion of sugars into glycolic acid. Theoretically, the proposed pathway should lead to a maximal yield of 3 GA per glucose (**Table 1**) or 2.5 GA per pentose, which is 50% higher than obtained by the natural pathway (**Figure 1**). This pathway design implicates the overexpression of kdsD and fsaA of E. coli which codes for an arabinose-5P (Ara5P) isomerase and a class I aldolase, respectively. Even though these two genes are endogenous to E. coli, this pathway is not natural in this bacteria because the Kdsd enzyme is naturally implicated in the synthesis of 2-keto-3-deoxy-octulosonate (KDO), a constituent of the outer member of cell wall lipopolysaccharide (Lim and Cohen, 1966), whereas Fsa has been initially reported as a F6P aldolase that catalyzes the aldol cleavage of F6P into dihydroxyacetone (DHA) and glyceraldehyde-3-P (GAP) (Schurmann and Sprenger, 2001). The bacterium E. coli also harbors an ortholog of fsaA termed talC or fsaB (Reizer et al., 1995) but the physiological function of these two genes is still completely unknown. In particular they are not expressed under standard—LB- culture condition (Schurmann and Sprenger, 2001). Nevertheless, Clapes' team demonstrated the remarkable originality of Fsa enzyme as a unique biocatalyst for asymmetric cross-aldol addition of glycoladehyde. They also reported that Fsa can cleave Ara5P into glycolaldehyde and GAP with a 10 fold higher affinity than F6P (Garrabou et al., 2009), which was confirmed in Lachaux et al. (2019). Therefore, the consecutive action of Kdsd and Fsa leads to the conversion of pentose phosphate intermediate Ru5P into glycolaldehyde and GAP (**Figure 1**). While glycolaldehyde can be oxidized into glycolic acid by an aldehyde dehydrogenase encoded by aldA (Caballero et al., 1983), GAP is recycled into Ru5P through the non-oxidative pentose phosphate pathway (**Figure 1**). This cycling scheme may result in theory in the production of 3 moles of GA per mole of glucose, only if the lower part of the glycolysis at the level of GAP is blocked. It also cost one ATP but generates 3 moles of NADH (**Table 1**) that needs to be reoxidized to allow the continuous functioning of the cycle. The reoxidation of NADH requires an active respiratory chain which is coupled to the synthesis of ATP. This ATP provision will be useful to regenerate ATP used in glucose phosphorylation, for uptake of sugars as well as for cellular maintenance. The in vitro and in vivo function of the glycoptimus pathway was validated although the yield of GA from glucose and xylose did not exceed 30% of the maximal yield expected. Possible explanation of this poor efficiency of the pathway could be that (i) the blockage of the lower part of glycolysis by deletion of gapA encoding GADPH was detrimental for cell viability; (ii) all competitive pathways using the same intermediaries have not yet been dismantled in the engineered strains; and (iii) the carbon flux in the kdsD-fsaA pathway is not optimized yet (Lachaux et al., 2019). Finally, the reoxidation of NADH is likely to generate more ATP than needed for the functioning of the glycoptimus cycle, which may result in feedback inhibition due to accumulation of intermediates and eventually depletion of cofactors. Solving these various issues together with the development of an appropriate fermentation practice should lead to an economically viable process of glycolic acid synthesis.

#### The Methanol Condensation Cycle for Methanol Assimilation Without Carbon Loss

Abundant carbon sources other than sugars, such as methanol can be also exploited for the biosynthesis of added-value molecules. However, assimilation of these alternative resources also poses problem with carbon conversation. As an example, the ribulose-5- monophosphate pathway (RuMP) that is present in methylotrophic bacteria and in some yeasts such as Pichia pastoris (Yurimoto et al., 2002; Muller et al., 2015; Russmayer et al., 2015) is one of the three natural methanol assimilatory pathways known so far (Zhang et al., 2017) which can produce one acetyl-CoA from three formaldehydes at the expense of the decarboxylation of pyruvate. Liao's team (Bogorad et al., 2014) showed that the decarboxylation step could be avoided by combining the RuMP with phosphoketolase enzymes, leading to the production of acetyl-CoA from 2 formaldehydes through a methanol condensation cycle (MCC) as depicted in **Figure 2**. However, this MCC can only work if the thermodynamically unfavorable NAD+-dependent methanol dehydrogenase is coupled to the reoxidation of NADH into reduced products such as ethanol, n-butanol or even higher alcohols (Bogorad et al., 2014). Remarkably, the MCC reaction is completely redox balanced and independent of ATP, and thus does not involve the phosphofructokinase enzyme of the RuMP, as well as the triose phosphate isomerase (Tpi), fructose 1,6-bisphosphate aldolase (Fba) and fructose-1,6-bisphosphatase (Fbp) of the NOG pathway. This property has been exploited to demonstrate the in vitro functioning of MCC to the production of ethanol or n-butanol. Therefore, a cell-free system could be potentially exploitable for larger scale production of higher alcohols with maximal yield and at high productivity if conditions for enzymes and intermediates stability are assured. In addition, this in vitro study pointed out the role of phosphoketolase in the robustness of the MCC, showing that its optimal running requires well-balanced levels of this enzyme with respect to the others, as a low or excessive phosphoketolase activity could trigger kinetic traps that significantly diminish rate of end-product production, due to either accumulation or depletion of intermediates metabolites in the cycle.

# CARBON CONSERVATION SYNTHETIC PATHWAYS BUILDING ON CARBOXYLATING REACTIONS

## Reversal of the Glyoxylate Shunt to Build C2 Compounds With Maximal Carbon Conservation

The TCA or citric acid cycle elucidated by Krebs and Johnsson in 1937 (see Krebs and Johnson, 1980 for an historical perspective or Akram, 2014 for a recent review of TCA role in intermediary metabolism) is considered as the central metabolic hub of the cell whose amphibolic nature provides intermediates for numerous metabolic functions. This cycle generates energy and reducing power needed for anabolic activities of the cell at the expense of complete oxidation of pyruvate into carbon dioxide. The glyoxylate shunt, discovered by Krebs and Kornberg in 1957 (see Dolan and Welch, 2018) for an excellent review on this pathway) avoids the decarboxylation steps of the TCA cycle and allows acetyl-CoA to be converted into a C4 carbon without carbon loss. This shunt is therefore key for organisms that grow on C2 to C4 carbon such as acetate, glycerol and malate, or on ketogenic amino acids (i.e., aspartate, glutamate). The glyoxylate shunt takes carbon away from the TCA cycle at the level of isocitrate, just before the commencement of the first decarboxylation step. This requires two specific enzymes that are the hallmark of this cycle: isocitrate lyase that cleaves isocitrate into succinate (C4) and glyoxylate (C2), and malate synthase, which condenses glyoxylate with acetyl-CoA to malate. Due to the large negative Gibbs value of the malate synthase reaction (1rG'<sup>0</sup> = −44 kJ/mol), the glyoxylate cycle can only run in the acetyl-CoA condensation. However, running the cycle clockwise directly could be beneficial to supply acetyl-CoA from sugars by bypassing the decarboxylation step of pyruvate. As illustrated in **Figure 3**, a reversal of the glyoxylate shunt (rGS) in which the thermodynamically favorable PEP carboxylating reaction is implemented could generate 2 acetyl-CoA from TCA intermediates. This challenge was raised by the Liao' group according to the following strategy (Mainguet et al., 2013). At first, they demonstrated that isocitrate lyase

from 2 methanol. Here, it is represented the metabolic pathway leading to the production of n-butanol which requires 4 methanol. Enzymes abbreviations are: Medh, NAD+-dependent methanol dehydrogenase; Hps, hexulose-6P synthase; Phi, phosphohexuloseisomerase; Tkt, transketolase, Fxkp, F6P/Xu5P phosphoketolase; Tal, transaldolase; Rpe, ribulose-5-P epimerase; Rpi, ribose-5-isomerase. Abbreviation for metabolites as in Figure 1.

(ICL) is reversible in vivo by showing that a E. coli strain made auxotroph for glutamate (glu<sup>−</sup> strain) by deletion of gltA and prpC encoding citrate synthase was able to grow on glucose mineral medium supplemented with glyoxylate and succinate. In contrast, this glu<sup>−</sup> strain was unable to recover growth when glyoxylate was replaced by malate, even if dctA gene encoding a non-glucose repressible malate importer from B. subtilis had been overexpressed. These results confirmed that malate synthase reaction was not reversible in vivo. To get around this obstacle, they wished to express an ATP-dependent malate thiokinase (Mtk) encoded by mtkA and mtkB from Methylobacterium extorquens. In this methanotrophic bacterium, this enzyme contributes to C1 assimilation through the serine cycle (Fei et al., 2014). This reaction step is then followed by the cleavage of malyl-CoA into glyoxylate and acetyl-CoA by a malylcoA lyase (Mcl) naturally present in microorganisms that use the 3-hydroxypropionate cycle for autotrophic carbon dioxide fixation (Berg et al., 2007). Unfortunately, the expression of mtkAB and mcl from Methylobacterium extorquens expressed in a glu<sup>−</sup> strain did not rescue growth on malate and succinate likely because Mtk was not functional. They overcame this problem by screening various putative malate thiokinase and discovered that sucCD-2 from Methylococcus capsulatus annotated as encoding a succinyl-CoA synthetase (Ward et al., 2004) had very good

Mtk activity. Expression of sucCD-2 gene with mcl restored the capacity of growth of a glu<sup>−</sup> strain on glucose supplemented with malate and succinate. The second important action was to recycle OAA from isocitrate that is accompanied by the release of a second molecule of acetyl-coA. This involves to reverse the aconitase and the citrate synthase reaction. While the first one is readily reversible, the conversion of citrate to OAA by citrate synthase is strongly unfavorable (1rG'◦∼ + 34 KJ/mol). It can nevertheless be obtained by expressing an ATP-dependent citrate lyase encoded by acl which is present in several eukaryotic cells and in some archeabacteria (Fatland et al., 2000; Verschueren et al., 2019). The in vivo functionality of the Acl reaction was tested in E. coli strain made auxotroph to aspartate (asp<sup>−</sup> strain) by deleting genes encoding all enzymes that produce its precursor OAA (ie ppc, mdh, mqo), as well as citrate synthase and citrate lyase (gltA and citE). As expected, the asp<sup>−</sup> strain recovered growth on a glucose minimal medium when supplemented with citrate. Although the thermodynamically unfavorable reactions were solved to potentially create a reverse glyoxylate cycle, it was also necessary to remove isocitrate dehydrogenase (icd) to avoid siphoning isocitrate obtained by condensation of glyoxylate and succinate into α-ketoglutarate. Then, the assembly of the complete pathway in asp<sup>−</sup> strain which was otherwise deleted for mdh, mqo, ppc, icd, and aceB consisted in the overexpression of BsdctA, McsuccD-2, Rsmcl, and Ctacl genes. This strain was shown to grow on glucose supplemented with malate and succinate, demonstrating the in vivo function of this rGS. Until now, the complete demonstration of the "in vivo" reversal of the glyoxylate cycle has not been fully demonstrated. It will require to connect OAA to malate which can be easily carried out by the reversible NAD+-dependent malate dehydrogenase encoded by mdh. However, cycling malate shall require also expression of a fumarase (Fum) and a fumarate reductase (Frd), both enzymes being preferentially active under anaerobic conditions (Lin and Iuchi, 1991; Tseng, 1997).

In theory, the integration of rGS with the central carbon metabolic pathway via the PEP carboxylase encoded by ppc could lead to the conversion of one mole glucose into 3 moles of acetyl-CoA, thus achieving a 50% yield increase over the native pathway. Contrary to the NOG pathway that converts glucose to 3 AcCOA at the expense of one ATP, the conversion of glucose into AcCOA by the rGS generates one extra ATP. In addition, while NOG is redox neutral, the rGS pathway results in the net production of one NADH and one reduced quinone per glucose (**Table 1**). This indicates that rGS is more energetically efficient than NOG which could be an advantage with respect to the production of acetyl-CoA derived products such as isoprenoids, fatty acids, long chain alcohols as they required reduced power and ATP (Tabata and Hashimoto, 2004; Kondo et al., 2012; Robles-Rodriguez et al., 2018).

# The Malyl-CoA—Glycerate Cycling (MCG)

Like the rGS pathway, the synthetic Malyl-CoA-Glycerate (MCG) cycling pathway allows the conversion of glucose into acetyl-CoA. As depicted in **Figure 3**, this synthetic pathway relies on 6 natural reaction steps that are catalyzed by endogenous enzymes in E. coli, completed by the heterologous expression of mtkAB and mcL from methylotrophic bacteria. The importance of these two reaction steps in the in vivo functionality of the MCG was demonstrated by showing that the growth on glucose of an acetyl-COA auxotroph E. coli strain (1aceE 1poxB 1plfB) can be rescued upon the expression of mtkAB and mcl genes. Like NOG and rGS, the theoretical molar yield of acetyl-COA per glucose consumed is 3 (**Table 1**). Another interesting feature of the MCG is to be able to assimilate C2 compound such as glycolate or glyoxylate (**Figure 3**). Hence, in photosynthetic organisms, expression of MCG could improve carbon fixation in combination with the CBB cycle by re-assimilating glycolate produced by photorespiration. Also, the conversion of glucose into acetyl-CoA by the MCG is redox neutral and costs one ATP for glucose phosphorylation, unless the glycolytic PEP carboxykinase as it exists in the rumen bacterium Actinobacillus succinogenes (Kim et al., 2004; Leduc et al., 2005) is used instead of PPC, which would result in the production of one extra ATP.

#### EXPLOITING CARBOXYLATING ENZYMES FOR CO<sup>2</sup> FIXATION IN HETEROTROPHIC ORGANISMS

#### Carboxylating Enzymes as a Key Reaction in Heterotrophic Organisms for CO<sup>2</sup> Fixation

In an elegant paper published in 2010, Bar-Even et al. (2010) developed a constrained-based modeling approach that explored all possibilities that could be devised from a repertoire of ∼5,000 known metabolic enzymes reported in the KEGG database (https://www.genome.jp/kegg/pathway.html) to generate carbon fixation pathways that are alternative to the 5 known to date. Criteria used in their modeling were (i) the enzyme's specific activity is equal or superior to that of RuBisCO which is used as the benchmark activity; (ii) a minimal energy cost of the pathway, which corresponds to the cost in NAD(P)H, ferredoxins, FADH2, and ATP in the production of one mole of products from CO2; (iii) thermodynamic feasibility of the pathway under a plausible physiological range of metabolites concentrations; (iv) topology of the pathway, which incorporates the number of enzymatic reactions the carbon fixation pathway can make as an independent unit and (v) the compatibility or integration of this synthetic pathway with the endogenous metabolic network. According to these criteria, they identified several cycling pathways containing four to six enzymatic steps leading to the production of the C2 compound glyoxylate. Although most of them were unrealistic because thermodynamically unfeasible or using oxygen-sensitive ferredoxin-oxidoreductase enzyme, these authors found that kinetically efficient pathways absolutely require a carboxylating enzyme with high activity and affinity toward CO<sup>2</sup> or HCO<sup>−</sup> <sup>3</sup> whose reaction must be physiologically irreversible. PEP carboxylase (ppc gene) and pyruvate carboxylase (pyc gene) came at first in their listing, followed by acetyl-CoA and propionyl-CoA carboxylase. Accordingly, they designed a pathway family termed Malonyl-CoA-Oxaloacetate-Glyoxylate (MOG) cycle employing either PEP carboxylase or pyruvate carboxylase as the sole carboxylating enzyme and that has a higher carboxylating activity than the reductive pentose pathway (rPP). This MOG pathway resembles the natural C4 cycle in which the recuperation of CO<sup>2</sup> arising from malate decarboxylation to pyruvate by RuBisCO is replaced by a PEP or pyruvate carboxylating enzyme. The net product of MOG is the C2-carbon glyoxylate, which can be converted into GAP by the bacterial-like glycerate pathway (Eisenhut et al., 2006; Igamberdiev and Kleczkowski, 2018) (**Figure 4**). Interestingly, this pathway turns out to be thermodynamically feasible and more efficient than the rPP, although it has not been found in Nature yet. Moreover, combined with the plant natural C4 cycle, it could allow to overcome the futile CO<sup>2</sup> cycling that is taking place in the bundle sheath cell and which is due to malate decarboxylation coupled to reassimilation of CO<sup>2</sup> by RuBisCO. In addition, this coupling could result in an extra CO<sup>2</sup> fixation with the release of glyoxylate. Therefore, it can be anticipated that the expression of MOG in autotrophic cells may provide great advantages from a biotechnological point of view, with faster growth and increased crop yields because of additional carbon dioxide input and absence of competing oxygenation reaction that reduces carbon fixation by plants. Whether the implementation of this MOG pathway in a heterotrophic cell like E. coli is biotechnologically relevant is less obvious. Indeed, this implementation will require the expression of heterologous genes encoding six out of the nine reactions of this cyclic pathway (**Figure 4**), namely the malate thiokinase (mtkAB), the malyl-CoA lyase (mcl), the methylmalonyl-COA carboxytransferase (mct), the malonyl-CoA reductase (mcr), the β-alanine-pyruvate transaminase (bapta), and the alanine 2,3 aminomutase (aam). More importantly, the stoichiometry evaluation of this route from glucose indicated that it can produce 2 moles of glyoxylate and 2 moles of acetate, with the requirement of two mole of CO2. However, while globally thermodynamically favorable and redox balanced, this reaction can occur at the expense of 3 ATP, which can be obtained by consumption of acetate, and hence resulting in the reemission of 2 moles of CO2. Therefore, this route has likely no meaning to be expressed in a chemoorganotrophic system.

# The Modified Serine Cycle Pathway in *E. coli*

In methylotrophic bacteria, the serine cycle is a natural pathway to assimilate C1 carbon such as methanol or methane into acetyl-CoA intermediate without loss of carbon (Smejkalova et al., 2010). A modified serine cycle pathway has been designed in E. coli to allow co-assimilation of C1-carbon such as methanol or formate with bicarbonate to generate acetyl-COA as a precursor for several bio-based products (Yu and Liao, 2018). This synthetic pathway encompasses nine enzymatic reactions for the cycle in which CO<sup>2</sup> is captured by the reaction catalyzed by PEP carboxylase, and four additional reaction enabling assimilation into 5–10 methylene tetrahydrofolate (CH<sup>2</sup> = H4F). The methyl unit is afterwards transferred on serine cycle by a Hmtf, hydroxymethyl transferase.

reaction involving glycine and catalyzed by serine hydroxymethyl transferase (Shmt). From these 13 enzymatic steps, seven (highlighted in red in **Figure 5**) had to be implemented using heterologous genes from various organisms, which included S. cerevisiae for the glyoxylate:alanine transaminase (Agt) encoded by AGX1, serine dehydratase (Sdh) from Cupriavidus necator encoded by SdaA, malate thiokinase (Mkt), and malyl-CoA lyase(Mcl) from M.extorquens. For methanol oxidation and its conversion to 5–10 methylene tetrahydrofolate, a variant of the Cupriavidus necator methanol dehydrogenase (Medh) with higher specific activity was used together with formate-H4F ligase (Fhlfl) and 5,10-methylene tetrahydrofolase synthase (Mthfs) from Moorella thermoacetica. The NAD+-linked formadehyde dehydrogenase (Faldh) from Pseudomonas putida was employed to oxidize formaldehyde to formate. All these genes were carried on 3 different plasmids and transformed into E. colistrain HY106, which was deleted for genes to avoid byproducts formation and shortage of intermediates in the cycle. Accordingly, aceB and glcB both encoding malate synthase were deleted to avoid competitive reverse reaction catalyzed by Mkt and Mcl. The gcl gene encoding glyoxylate carboxyligase was deleted to prevent shortage of glyoxylate into tartronate semialdehyde, as well as gcvP to avoid loss of glycine by decarboxylation. Genes encoding lactate dehydrogenase (ldhA) and fumarate reductase (frABCD) were also deleted in this strain to avoid shortage of pyruvate into D-lactate and malate into succinate. As indicated in **Table 1**, co-assimilation of methanol with CO<sup>2</sup> needs 3 moles ATP and 1 mole reduced equivalent per mole of acetyl-CoA produced. Hence, it cannot proceed without addition of another

carbon source to provide this energy and cofactors. Thus, the in vivo function of the complete cycle was demonstrated using C <sup>13</sup>-methanol together with xylose to supply energy and PEP. In addition the pyruvate to malate flux was increased upon overexpression of pyc and mdh encoding pyruvate carboxylase and malate dehydrogenase in the engineered E. coli equipped with the Mser pathway. The engineered strain could produce acetate from which 33% came from methanol. When the same experiment was carried out with both C<sup>13</sup> methanol and C<sup>13</sup> bicarbonate, the amount of acetate produced was the same but the labeling of both C1 and C2- carbon of acetyl-moieties was enriched by 2-fold, indicating that bicarbonate has been also incorporated into acetyl moieties. To conclude, the construction of this synthetic serine cycle in E. coli enables the assimilation of C1 compounds such as methanol to increase the production of C2-compounds as precursor of bio-based products such as PHB, isoprenoid, fatty acids, etc. In addition, this cycle, combined with the glyoxylate shunt, is capable to support growth from pyruvate that is formed from two C1-carbon (formate or methanol) and one bicarbonate, unless the mole excess of NAD(P)H is reoxidized by respiratory chain coupled with ATP.

## The HOB Pathway as an Alternative for C1-Carbon Metabolism Through CO<sup>2</sup> Fixation by PEP Carboxylase

Another illustration about the potential function of carboxylating enzyme to incorporate CO<sup>2</sup> into biomass comes from the work of Bouzon et al. (2017). These authors built a synthetic pathway that is alternative to the ubiquitous one carbon pathway transferring C1-moities (i.e., formyl, hydroxymethyl, or methyl group) from tetrahydrofolate for the synthesis of purine nucleotides, thymidylate, methionine, or coenzyme A (**Figure 4**). These C1-moities are obtained from the pivotal intermediate 5,10-methylene tetrathydrofolate (CH2=H4F) produced by the condensation of tetrahydrofolate with formaldehyde that originates from serine or glycine. To replace these amino acids as C1-donor, these authors devised a cyclic pathway in which the C1 compound (in this case formaldehyde) is obtained from the aldolytic cleavage of the non-natural intermediate 2-keto-4-hydroxybutyrate (HOB). As shown in **Figure 4**, the complete cycle encompasses 8 enzymatic steps in which the C1- source is HCO<sup>−</sup> 3 that is captured by PEP carboxylase to yield OAA. Through a set of 4 natural enzymatic step, OAA is converted into homoserine which gives rise to HOB through a transaminase reaction. The best candidate for this non-natural transaminase reaction was obtained by employing an automated evolutionary technology originally developed by Marlière et al. for evolving E. coli possessing DNA in which thymine was replaced by the synthetic building block 5-chlorouracil (Marliere et al., 2011). The evolved transaminase was found to correspond to a variant of the alanine: pyruvate transaminase encoded by alaC having amino change at position 142 (Ala to Pro) and 245 (Tyr to Glu). The HOB is then cleaved into pyruvate and formaldehyde which is readily transferred into H4F by the promiscuous hydroxymethytransferase (Mtf) encoded by panB. The cycle is then closed by converting back pyruvate into PEP, which implicates PEP synthase. Overall, the functioning of this cycle costs two moles of NADPH and three moles of ATP per mole of CO<sup>2</sup> incorporated (**Table 1**). To implement this synthetic pathway that completely rewires the C1-canonical metabolism by imposing HOB as the essential metabolic intermediate, the authors banked on an evolutionary trajectory of a E. coli strain lacking all enzymes necessary for the natural C1- transfer moieties using an original automated evolutionary technology that is now commercialized as "Heurisko" by the company Altar (http://www.altar.bio/). The authors estimated that the CO<sup>2</sup> assimilated in the biomass of E. coli expressing this HOB pathway is about 2 times more than in wild type strain (Bouzon et al., 2017). However, being quite energy consuming, the relevance of this pathway in the production of biotechnological molecules methionine is questionable.

## EXPRESSION OF THE CALVIN-BENSON CYCLE ENZYMES FOR *IN-SITU* CO<sup>2</sup> REINTEGRATION INTO THE CENTRAL METABOLIC NETWORK

Under anaerobic condition, sugar fermentation by the yeast Saccharomyces cerevisiae into ethanol generated excess of NADH, which must be reoxidized to warrant redox-cofactor balancing (Van Dijken and Scheffers, 1986). This reoxidation diverts between 4 and 10% of the sugar into glycerol, which has a significant impact on the economy of large scale yeast-based bioethanol production. To reduce or even eliminate this byproduct and in the same time, to increase the ethanol yield, the group of Jack Pronk in TU Delft proposed to use CO<sup>2</sup> as electron acceptor for the reoxidation of this excess of NADH. A theoretical analysis indicates that the replacement of glycerol by CO<sup>2</sup> can increase the ethanol yield from sugar by 14% if the CO<sup>2</sup> is incorporated through phosphoribulokinase (PRK) and ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO) onto ribulose 5-P (Ru5P) (**Figure 5**). In their design, Ru5P must be produced from the non-oxidative pentose phosphate route in order to reoxidize all biosynthetic NADH and not NADPH through a transhydrogenase-type conversion NADPH-> NADH (Guadalupe-Medina et al., 2013). This clever idea was demonstrated in two elegant papers published in Biotechnology for Biofuels (Guadalupe-Medina et al., 2013; Papapetridis et al., 2018). In the first paper, the authors expressed the codon-optimized prokaryotic form II RuBisCO-encoding cbbM gene from T. denitrificans in a centromeric plasmid under the strong TDH3 promoter and showed that RuBisCO was functional only upon co-expression of E. coli groEL/groES encoding chaperones whereas surprisingly T denitrificans chaperones encoded by cbb02/cbbQ2 were ineffective. The PRK from Spinachia oleracea was integrated together with the E. coli groEL/groES gene in the yeast genome at the CAN1 locus under the galactose-inducible GAL1 promoter. Remarkably, expression of this minimal Calvin cycle in this engineered yeast cultivated under anaerobic chemostat condition showed a 68% reduction of glycerol production accompanied by a 10% increase of ethanol yield. Saturating the culture with CO<sup>2</sup> resulted in a 90% decrease of glycerol, which could be explained in part by the low affinity of form II RuBisCO for CO<sup>2</sup> (KC02 = 0.26 mM). In a second paper, the authors carried out several targeted metabolic engineering to optimize the fermentation kinetic, improving ethanol yield and reducing glycerol production. To achieve this goal, nine copies of the cbbM overexpression cassette along with a single expression cassette of E. coli groEL/groES chaperones were integrated at the SGA1 locus using CRISPR-Cas9 tool followed by the integration at X-2 locus of the PRK1 under the inducible anaerobic DAN1 promoter. The resulting engineered strain expressing a more robust Calvin-Benson cycle showed significant (31%) reduction of glycerol under glucose anaerobic condition. However, the contribution of the PRK/RuBisCO pathway to NADH reoxidation decreased with increased growth rate, with a concurrent higher contribution of the glycerol pathway. Therefore, a complementary strategy was to abrogate this pathway by removal of glycerol-3-P dehydrogenase encoded by GPD1 and GPD2. While GPD1 is essential for growth of Saccharomyces cerevisiae under osmotic stress (Albertyn et al., 1994), loss of GPD2 strongly affects growth under anaerobic condition (Ansell et al., 1997). Accordingly, the deletion of GPD2 in the engineered strain bearing the Calvin cycle resulted in a 62 % decrease of glycerol and 13% increase of ethanol yield as compared to a reference strain. An additional engineering strategy was dedicated to overexpress genes encoding the transketolase and transaldolase with the aim to increase the availability of Ru5P. Indeed, this intermediate is not solely the substrate of PRK, but it is also indirectly implicated in the

biosynthesis of nucleic acid, aromatic amino acids and vitamins. Thus, its depletion through PRK could be in part responsible for the poor growth of the strain expressing the Calvin cycle. In accordance with this idea, overexpression of non-oxidative PPP genes RPEI, TKL1, TKL2, TAL1, RKI1 in the engineered strain also deleted for GPD2 resulted in a specific growth rate that was virtually the same as the reference strain under anaerobic batch condition on glucose. Furthermore, this engineered strain achieved the best fermentation performance with an increased ethanol yield of 15% and a 90% decrease of glycerol production (Papapetridis et al., 2018).

A comparable strategy has been applied by two other research groups which have equipped a xylose-utilizing Saccharomyces cerevisiae with the two Calvin-cycle enzymes PRK and RuBisCO to enhance xylose fermentation with in situ CO<sup>2</sup> fixation (Li et al., 2017; Xia et al., 2017). As S. cerevisiae cannot naturally ferment xylose, fungal pathway consisting of a xylose reductase (XR) and xylulose dehydrogenase (XDH) or the bacterial xylose isomerase (XI) have been heterologously expressed, together with upregulation of the native pentose phosphate pathway (PPP) to achieve efficient and rapid xylose fermentation (Matsushika et al., 2009; Young et al., 2010; Oreb et al., 2012). While the XR/XDH pathway appears to be more effective for xylose fermentation rate and ethanol productivity than the XI pathway (Karhumaa et al., 2007), it leads to cofactor imbalance with a surplus of NADH that is regenerated at the cost of by-products formation such as glycerol and xylitol. Expression of the PRK-RuBisCO module should exploit this NADH surplus for recycling CO<sup>2</sup> that is generated during the fermentation. This strategy was developed by Xia et al. (2017) which used a previously evolutionary engineered S. cerevisiae SR8 strain that was able to efficiently coferment xylose and glucose. This engineered strain harbored the XR/XDH together with an upregulated PPP and was deleted for PHO13 and ALD6 as beneficial targets for xylose fermentation (Kim et al., 2013). Similarly to the work of Guadalupe-Medina et al. (2013), two copies of the cbbM gene encoding form-II RuBisCO from R. rubrum under the strong TDH3 promoter and groEL/groES cassette under TEF1 promoter were integrated in the yeast genome (notably at ALD6 and PHO13 locus, respectively) leading to a stable platform SR8C<sup>+</sup> strain. Only RuBisCO activity was detected in yeast strain co-expressing cbbM and groEL/ES. Then, while the overexpression of PRK from Spinacia oleracea alone was toxic for growth on xylose but not on glucose, likely because of ATP depletion or consecutive hyperaccumulation of RuBP, expression of the PRK-RuBisCO module rescued complete growth on xylose, which was accompanied by a 24% reduction of xylitol and 10% increase of ethanol yield while glycerol yield was similar to the control strain. Moreover, it was demonstrated that this engineered S. cerevisiae carrying the Calvin-cycle enzymes released 7% less CO<sup>2</sup> than the control strain during anaerobic xylose fermentation. In a more recent publication, Li et al. (2017) engineered a xylose-utilizing S. cerevisiae for co-utilization of maltose, xylose and CO2. The rationale behind this strategy was to provide metabolite precursors, energy and reduced cofactor useful for xylose reduction by XR from maltose, which otherwise does not repress xylose fermentation. In addition, a mXR variant that preferred NADH over NADPH was co-expressed together with XR and XDH on a high copy plasmid pRS425. At variance to previous works, genes encoding form-I RuBisCO of Ralstonia eutropha or form-II of R. rubrum, PRK from Spinacia oleracea or Ralstonia eutrophia together with E. coli groEL and groES encoding chaperone were cloned in a low copy plasmid YcpLac33 and transformed into the xylose-utilizing yeast. They reported that these engineered strains grew better on a mix maltosexylose medium than control strain. They exhibited higher ethanol productivity and yield, with the highest increase of ethanol yield (+ 15%) and sugar consumption rate (+ 63%) with the engineered strain expressing heterotrophic form-I RuBisCO. Using <sup>13</sup>C-labeled CO<sup>2</sup> and a metabolic flux index MFI <sup>h</sup>−C02 (Gong et al., 2015) for relative quantification of flux ratio between the CO2-fixing by-pass pathway and the central carbon metabolic pathway, an incorporation of 8% CO<sup>2</sup> into Ru5P was estimated at a rate of 390 ± 50 mg CO2/L/h, which is a rate that is in the range of CO<sup>2</sup> fixation by natural autotrophic microbes (Gong et al., 2015). Altogether, these results indicated that the beneficial effect of implementing the CO2-fixation pathway on growth and fermentation largely overcome the additional cost of ATP that is required when this pathway is expressed.

The RuBisCO-based pathway was also investigated in E. coli initially as a screen to find out variants of RuBisCO enzyme with higher catalytic activity and better affinity to CO<sup>2</sup> (Parikh et al., 2006). It was then exploited for in situ CO<sup>2</sup> assimilation in bioproduction (Zhuang and Li, 2013; Gong et al., 2015). For this purpose, the RuBisCO-encoding genes rbcl-rbcX-rbcS from the cyanobacterium Synechococcus sp PC7002 and PRKencoding gene from Synechococcus elongatus were expressed on a high copy plasmid pET30a and arabinose was provided to the medium as an extra carbon source to enhance C5 phosphorylated sugar required for RuBisCO enzyme. Under this condition, the authors reported that the expression of RuBisCO alone promoted faster and complete consumption of arabinose, whereas the overexpression of PRK alone caused some growth retardation, which was explained by accumulation of RuBP. It was shown that the engineered strain expressing the Calvin-cycle was able to assimilate under anaerobic growth condition CO<sup>2</sup> at a rate comparable to that of the autotrophic cyanobacteria and microalgae (Gong et al., 2015). However, to overcome the co-cultivation condition, the same authors decided to enhance the non-oxidative pentose phosphate pathway by either overexpressing transketolase (tktAB) or by deleting glucose-6- P dehydrogenase (zwf) encoding gene. These genetic actions resulted in a significant decrease of CO<sup>2</sup> yield per glucose consumed. Genes encoding D-lactate dehydrogenase (ldh) and fumarate reductase (frd) were further deteled in this zwf mutant in order to favor recycling of in situ CO<sup>2</sup> into fermentation products ethanol or acetate. However, while this reduced CO<sup>2</sup> production proved the Calvin cycle pathway to work in vivo, and the expression of this cycle did not impede the growth on glucose, only 70% of carbon initially fed in the medium was recovered at the end of the fermentation. Moreover, there was no significant increase in C2-related compounds (Li et al., 2015; Yang et al., 2016). It turned out that this fermentation profile was explained by a huge accumulation of pyruvate that likely resulted from an excessive glycolytic flux at the sugar uptake and phosphorylation step leading to accumulation of NADH and concurrent reduction of pyruvate-formate lyase activity. This activity was inhibited by two concurrent actions, namely a direct inhibition by NADH and a downregulation of pflB encoding this enzyme due to the repression of its transcriptional activator encoded by arcA (Yang et al., 2016). Intriguingly, these transcriptional effects were already observed upon the simple overexpression of RuBisCO, suggesting a direct impact of the protein on the central carbon metabolism. To overcome these inhibitory problems, pyruvate decarboxylase and alcohol dehydrogenase encoding genes from Zymomonas mobilis were overexpressed in the 1zwf 1ldh 1frbd mutant strain equipped with the Calvin-cycle enzymes PRK and RuBisCO. This genetic intervention resulted in a yield of C-2 compounds from glucose 20% higher than that of theoretical fermentation value, which was due to a direct in situ CO<sup>2</sup> recycling into these products by the RuBisCO system. In addition, a rate of 53 mg/L/h of CO<sup>2</sup> consumption was evaluated, which was in the range of rate measured for CO<sup>2</sup> fixation by microalgae (Gonzalez Lopez et al., 2009; Ho et al., 2012), arguing that mixotrophic fermentation can be a competitive alternative for reducing loss of CO<sup>2</sup> and converting it into biobased products.

# CONCLUSIONS AND OUTLOOK

The carbon management during the metabolic conversion of renewable carbon sources such as sugars by heterotrophic organisms is an attractive, yet partial, solution to combat the acute problem of continuous CO<sup>2</sup> emission associated with human activities. Since carbon dioxide is an intrinsic byproduct of carbon metabolism ensuring the irreversibility of metabolic pathways in which it is involved, the rewiring of carbon metabolism aiming to circumvent this carbon loss must take into account this cost. However, in most of the case that were presented here, this cost can be overridden by the benefit of a higher carbon yield achieved by these refractory or synthetic pathways. This higher carbon yield is the second advantage of this carbon management since it can in principle increase the practical performance of a product per carbon consumed, which is a decisive parameter for the economic evaluation of a biotechnology process. Nonetheless, the synthetic pathways aiming at reducing carbon loss or assimilate CO<sup>2</sup> are still in their infancy. Substantial work is required to evaluate the potential for industrial application of some of them, such as NOG, MCG, rGS or glycoptimus, whereas others such as the MOG, MSer, and HOB will likely remain as elegant intellectual models illustrating the power of Synthetic Biology in its ability to reformat metabolic pathways.

While heterotrophic organisms are not naturally able to assimilate carbon dioxide, an attractive and complementary strategy to those developed elsewhere to capture CO<sup>2</sup> is to equip these organisms with the two key Calvin cycle enzymes in order to capture CO<sup>2</sup> produced during sugar fermentation, and meanwhile use it as an electron acceptor for NADH reoxidation. This dual function has been developed in yeast and E. coli and was shown to increase C2- products such as ethanol (Papapetridis et al., 2018; Tseng et al., 2018). The high concentration of carbon dioxide that prevails in industrial fermentations and their usual anaerobic to micro-aeration conditions are factors that should support the proper functioning of this capture system since RuBisCO has low affinity to CO<sup>2</sup> and its carboxylation activity is antagonized by oxygen. Therefore, this engineering strategy could be implemented without major investment at the industrial scale owing to the fact that GMO's legislation is modified.

Another promising strategy that is already attracting much interest is to exploit the high catalytic capacity of carboxylating enzymes such as the PEP carboxylase to capture "CO2" in vivo. In addition of engineering specific pathways allowing this capture such as rGS or MCG, a great advantage of using

#### REFERENCES


heterotrophic organisms such as yeast or E. coli for CO<sup>2</sup> fixation is their fast growth rates and capability to reach high cell density in bioreactors, both criteria contributing to CO<sup>2</sup> assimilation rate that are higher than that of auxotrophic organisms. Therefore, engineering heterotrophic microbes for CO<sup>2</sup> fixation can be a promising avenue to both reduce CO<sup>2</sup> emission and in the meantime increase carbon yield as the fixed CO<sup>2</sup> can be easily integrated into the central carbon metabolism. To be even more efficient, the next step will be to employ other energy source than sugar for CO<sup>2</sup> fixation.

#### AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

#### ACKNOWLEDGMENTS

The authors wish to acknowledge Agence Nationale de la Recherche (ANR), Ministry of Research and Education (MENRT), and TWB for technical and financial support.

functions of the same protein in Escherichia coli. J. Biol. Chem. 258, 7788–7792.


phosphate pathway to achieve low carbon dioxide emission fermentation. Bioresour. Technol. 187, 189–197. doi: 10.1016/j.biortech.2015.03.090


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 François, Lachaux and Morin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# A New Synthetic Pathway for the Bioproduction of Glycolic Acid From Lignocellulosic Sugars Aimed at Maximal Carbon Conservation

#### Edited by:

Matthias Georg Steiger, Vienna University of Technology, Austria

#### Reviewed by:

Peter Richard, VTT Technical Research Centre of Finland Ltd, Finland Peter J. Schaap, Wageningen University & Research, Netherlands Kris Niño Gomez Valdehuesa, Myongji University, South Korea

> \*Correspondence: Jean Marie François fran\_jm@insta-toulouse.fr

#### †Present address:

Cláudio J. R. Frazao, Thomas Walther, TU Dresden, Institute of Natural Materials Technology, Dresden, Germany

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 15 August 2019 Accepted: 12 November 2019 Published: 27 November 2019

#### Citation:

Lachaux C, Frazao CJR, Krauβer F, Morin N, Walther T and François JM (2019) A New Synthetic Pathway for the Bioproduction of Glycolic Acid From Lignocellulosic Sugars Aimed at Maximal Carbon Conservation. Front. Bioeng. Biotechnol. 7:359. doi: 10.3389/fbioe.2019.00359 Cléa Lachaux 1,2, Cláudio J. R. Frazao1†, Franziska Krauβer <sup>1</sup> , Nicolas Morin1,2 , Thomas Walther 1,2† and Jean Marie François 1,2 \*

<sup>1</sup> Toulouse Biotechnology Institute (TBI), Université de Toulouse, CNRS, INRA, INSA, Toulouse, France, <sup>2</sup> TWB, Toulouse, France

Glycolic acid is a two-carbon α-hydroxy acid with many applications in industrial sectors including packaging, fine chemistry, cosmetics, and pharmaceutics. Currently, glycolic acid is chemically manufactured from fossil resources. This chemical mode of production is raising some concerns regarding its use in health for personal care. Microbial production of GA stands as a remarkable challenge to meet these concerns, while responding to the increasing demand to produce bio-sourced products from renewable carbon resources. We here report on the design and expression of a novel non-natural pathway of glycolic acid in E. coli. The originality of this new pathway, termed "glycoptimus" relies on two pillars. On the one hand, it requires the overexpression of three naturally occurring E. coli genes, namely kdsD encoding a D-arabinose-5-P isomerase, fsaA encoding a class 1 aldolase that cleaves D-arabinose-5-P into glyceraldehyde-3-P and glycolaldehyde, and aldA coding for an aldehyde dehydrogenase that oxidizes glycoladehyde in glycolate. These three genes constitute the "glycoptimus module." On the other hand, the expression of these genes together with a reshaping of the central carbon metabolism should enable a production of glycolic acid from pentose and hexose at a molar ratio of 2.5 and 3, respectively, which corresponds to 50% increase as compared to the existing pathways. We demonstrated the 'in vivo' potentiality of this pathway using an E. coli strain, which constitutively expressed the glycoptimus module and whose carbon flow in glycolysis was blocked at the level of glyceraldehyde-3-P dehydrogenase reaction step. This engineered strain was cultivated on a permissive medium containing malate and D-glucose. Upon exhaustion of malate, addition of either D-glucose, D-xylose or L-arabinose led to the production of glycolic acid reaching about 30% of the maximum molar yield. Further improvements at the level of enzymes, strains and bioprocess engineering are awaited to increase yield and titer, rendering the microbial production of glycolic acid affordable for a cost-effective industrial process.

Keywords: synthetic biology, metabolic engineering, glycolic acid, aldolase, white biotechnology

# INTRODUCTION

Glycolic acid (GA) is a two-carbon α-hydroxy acid (HOCH2COOH) with dual properties of both alcohol and moderately acid (pKa 3.83). It is the simplest organic acid, which finds multiple applications in the cosmetic industry to improve skin texture, in pharmaceutic industries to treat skin diseases, in textile industry as a dyeing and tanning agent, in food industry as flavor and preservative as well as for cleaning and sanitizer agent in household and industry (https:// www.grandviewresearch.com/industry-analysis/glycolic-acid-

industry). Polymerization of glycolic acid alone or with other acids monomer such as lactic acid yields thermoplastic resins with excellent gas barrier properties. These polymers have the capability of being hydrolysed in aqueous environments gradually and controllably, making them good candidates for packaging materials or dissolvable sutures useful for biomedical applications (Fredenberg et al., 2011; Gädda et al., 2014). This large panel of glycolic acid applications accounts for the fact that the demand for this organic acid is constantly growing from US\$ ∼300 million in 2017 to US\$ ∼406 million in 2023, exhibiting a CAGR of 6.83% during this forecast period (https:// www.researchandmarkets.com/reports/4542547/glycolic-acidmarket-forecasts-from-2018-to-2023). Even though glycolic acid can be extracted from plants such as sugarcane, pineapple, and sugar beets, it is chiefly chemically manufactured from fossil resources by carbonylation of formaldehyde at high-pressure temperature (Drent et al., 2001). Alternatively, it can be produced from the enzymatic conversion of glycolonitrile using microbial nitrilases (He et al., 2010) or by bioconversion of ethylene glycol using Gluconobacter oxydans as the biocatalyst (Kataoka et al., 2001; Wei et al., 2009). However, these chemo-enzymatic methods rely on known irritant and carcinogenic chemicals, making their use problematic for some applications, especially within the personal care products industry.

In addition to these health concerns, another strong impetus to develop alternative and sustainable solutions for glycolic acid production comes from environmental and societal needs to reduce our dependence on fossil-based products and to promote bio-production from renewable carbon sources using microbial cell factories (Aguilar et al., 2013). GA represents a good opportunity for such an advancement, motivated by the fact that there are no natural microbial producers to produce at high yield this platform molecule from sugars (Salusjarvi et al., 2019). Consequently, biotechnological production of this simple organic acid from renewable resources has received a substantial interest in the recent years leading to the engineering of four different routes as depicted in **Figure 1** (and reviewed in Salusjarvi et al., 2019). The glyoxylate shunt (GS) is the natural pathway, the physiological function of which is to bypass the oxidative decarboxylation of TCA cycle, thereby conserving carbon skeletons for biomass (Dolan and Welch, 2018). This bypass starts at the level of isocitrate, which is aldolytically cleaved into succinate and glyoxylate by isocitrate lyase (ICL) encoded by aceA in E. coli. The first attempt for the bioproduction of glycolic acid from D-glucose via the glyoxylate shunt in Escherichia coli has been patented by METEX (Soucaille, 2007). To achieve a production of ca 57 g/L at 45% of the theoretical yield, 13 genetic modifications have been implemented, including the overexpression of the NADPH glyoxylate reductase (GLR) encoded by ycdW/ghrA, attenuation of isocitrate dehydrogenase (IDH) and deletion of side pathways that prevent lactate and acetate production as well as to the oxidation of glycolate. Further genetic modifications brought about by Deng et al. (2018) resulted in an engineered strain able to produce 65 g/L at 90% of the theoretical yield. D-xylose and ethanol have been used as carbon source for GA production via engineering of GS pathway in Saccharomyces cerevisiae and Kluyveromyces lactis (Koivistoinen et al., 2013), whereas acetate was the carbon substrate for glycolate production by a engineered C. glutamicum strain for TCA cycle and GS pathway and using D-glucose for growth (Zahoor et al., 2014). However, the exploitation of GS for glycolic production presents at least two major limitations. The first one deals with the fact that the pathway has been optimized to use only D-glucose as the carbon source, either by releasing glucose repression of the glyoxylate shunt genes in E.coli (Gui et al., 1996) or by expressing these genes under non-repressible glucose promoter in the yeast S. cerevisiae (Koivistoinen et al., 2013). The second problem is related to the NADPH preference of the glyoxylate reductase enzymes, which generates redox imbalance in the cell. In E. coli, this problem is even reinforced if ICDH activity is attenuated since it results in a reduction of NADPH availability.

Three -non-natural- GA pathways (**Figure 1**) have been constructed and expressed in bacteria and yeast that could in part overcome the problems stated above. Firstly, the production of GA by these routes bypasses the central carbon metabolism. As such, it requires fewer enzymatic reaction steps than with the natural GS: only three enzymatic steps are required for the X1P pathway (Cam et al., 2016) while four for the R1P (Pereira et al., 2016a) and for the Dahms pathway (Cabulong et al., 2018). Lowering the number of reaction steps might have a positive effect on the production yield as it reduces the energy cost for protein synthesis and metabolic burden and reduce loss of intermediates by competitive pathways (Bilgin and Wagner, 2012). Second, GA is produced from glycolaldehyde by a NAD+-dependent (glycol)aldehyde dehydrogenase. However, only pentose sugars can be assimilated by these non-natural pathways. In addition to this limitation, only two carbons of the C5-intermediates are utilized to produce GA whereas the three remained carbons are diverted into biomass via DHAP. To overcome in part this carbon loss, X1P and R1P pathways have been combined with GS bypass, enabling assimilation of both hexose and pentose sugars (Alkim et al., 2016; Pereira et al., 2016b). Conversion of C5 sugar to GA with high yield has been also proposed by linking part of the GS pathway with two reverse glyoxylate pathway (RGP) enzymes malate thiokinase and malylcoA lyase through Dahms pathway to recuperate pyruvate and recycle malate into glyoxylate (Cabulong et al., 2018).

The combination of the natural glyoxylate shunt with nonnatural GA producing pathways is very promising, as for instance, a conversion rate of D-xylose into GA at 92% of the theoretical yield was obtained by the Dahms-GS-RGP (Cabulong et al., 2018). However, these pathways still present

some metabolic hurdles such as redox imbalance between NADH and NADPH. In addition, they do not reach the maximum energy yield (Y<sup>E</sup> ) which is the maximum amount of product that can be formed from a substrate (Dugar and Stephanopoulos, 2011). This Y<sup>E</sup> value is pathway independent and is determined by the ratio γs/γ<sup>p</sup> where γ<sup>s</sup> and γ<sup>p</sup> are the reduction degrees of the substrate and the product, respectively. Accordingly, it can be calculated that Y<sup>E</sup> of glycolic acid (γ<sup>p</sup> = 6) from Dglucose (γ<sup>S</sup> = 24) and from pentose (γ<sup>S</sup> = 20) would be 4 and 3.3, respectively. However, this yield can only be reached if the biological system is capable of uptaking a carbon mole as CO2. Alternatively, if the loss of CO<sup>2</sup> at the level of pyruvate is prevented (see **Figure 1**), the theoretical yield of GA from D-glucose and pentose would be 3 and 2.5 moles/mole, which is 50% higher than that obtained by natural and non-natural

xylC (xylonolactone lactonase); xylD (xylonate dehydratase); yagE (KDX aldolase); aldA (glycoaldehyde dehydrogenase).

pathways (**Table 1**). Given this postulate, the purpose of this communication was to design and validate a novel -non-natural-GA pathway enabling an efficient assimilation of pentose and hexose derived from lignocellulosic biomass into GA to reach these yields and that in the meantime solving most of the problems raised by the existing ones and notably to overcome redox imbalance.

#### MATERIALS AND METHODS

#### Chemicals and Reagents

All chemicals and solvents were purchased from Sigma-Aldrich unless otherwise stated. Restriction endonucleases and DNAmodifying enzymes were from New England Biolabs. DNA plasmid were extracted using GeneJET Plasmid Miniprep Kit



(Thermo Scientific). DNA sequencing was subcontracted to Eurofins SAS (Ebersberg, Germany).

#### Plasmid Construction

Vectors pZA23, pZA33, pZE23, and pZS23 from Expressys <sup>R</sup> were used as they are inducible by IPTG and harbor a lightened structure of the lacI gene which reduces its size (2 358 to 3 764 bp) and they are modulable (easy to change replication origin, resistance marker and vector promoter by restriction/ligation). In this study, the promoter PA1lac0-1 in pZA33 has been replaced by the constitutive promoter proC and by the inducible promoter Ptac, generating pZA37 and pZA36, respectively. Likewise, the promoter PA1lac0-1 of pZS23 has been replaced by proD generating pZS28.

Gene cloning was carried out using NEBuilder HIFI DNA Assembly Master Mix (NEB E2621). This method enables several fragments to be assembled in a single step. The commercial mixture provided by New England Biolabs contains (a) an exonuclease, which creates 3′ single strand ends, which facilitates assembly of the fragments, which share a sequence complementarity; (b) a polymerase, which fills the empty spaces after the fragments have been assembled; and (c) a ligase, which links fragments together. The E. coli kdsD, fsaA and aldA genes were amplified from the genome DNA extracted from E. coli K12 MG1655 by PCR using primers described in **Table S1**. Fragments (kdsD + fsaA or aldA) were then inserted by HiFi assembly <sup>R</sup> into linearized beforehand pZ with primers hybridizing on either side of the MCS. All the plasmids have been verified by sequencing. The resulting plasmids bearing the kdsD, fsaA, and aldA genes are reported In **Table 2**.

#### Strain Construction and Transformation

The E. coli strains used in this work are listed in **Table 3**. Gene deletion (i.e., glcD, fucA, mgsA,pfkA, ptsG) was made by transduction using the phage P1vir. The preparation of the lysates P1vir and the transduction procedures were carried out as described in Bremer et al. (1984) with slight modifications. Strains (donor strain) from KEIO collection (Murakami et al., TABLE 2 | Plasmids used or constructed in this study.


2007) bearing a single deletion and a kanamycin antibioticresistance cassette was inoculated (200 µl of an overnight preculture made in LB) in 5 ml of LB containing 0.2% D-glucose and 5 mM CaCl<sup>2</sup> for 30 min at 37◦C. Then, 100 µl of P1vir lysate (∼ 5 × 10<sup>8</sup> phages/ml) was added to each donor culture and incubated at 37◦C for 2 to 3 h until the culture was clear and the cells were completely lysed (Baba et al., 2006). The lysates were recovered by filtration using 25 mm sterile syringe filters with a 0.2µm support membrane (Pall) and preserved at 4◦C. To delete the gene of interest, the receiving strain was infected with P1vir bearing the donor gene deletion cassette having a kanamycin resistance. For this purpose, the receiving strain was previously cultivated in 5 ml LB medium at 37◦C, collected by centrifugation at 1 500 g for 10 min and re-suspended in 1.5 ml of 10 mM MgSO<sup>4</sup> and 5 mM CaCl2. P1vir Lysate bearing the gene deletion cassette from the donor strain was added (0.1 ml) to the receiving strain suspension and incubated for 30 min at 37◦C. Then, 0.1 ml of 1 M sodium citrate was added, then 1 mL LB, and this cellular suspension was incubated of 1 h at 37◦C, 200 rpm before being spread on a solid LB medium with the appropriate antibiotic. Colonies were screened by PCR to isolate successful transduction events. Removal of the antibiotic cassette was carried out by transformed of the bacteria strains with pCP20 plasmid bearing the FLP recombinase, followed by PCR checking.

The competent non-commercial strains were prepared according to the protocol of Chung et al. (1989) with minor modifications as followed. A pre-culture was made overnight in LB overnight. Fresh LB culture was then inoculate with cells at a DO<sup>600</sup> of 0.1. When DO<sup>600</sup> reaches about 0.5, 2 ml of culture was collected and to the resulting pellet was resuspended in 300 µl TSS buffer [2.5%(w/v) PEG 3350, 1 M MgCl2, 5%(vol/vol) DMSO]. After 10 min on ice, plasmid of interest was added to the cell suspension, which was further incubated for 30 min on ice. This step was followed by a heat shock at 42◦C for 90 s. The transformed cells were put on ice for 10 min then 400 µl LB was added and the culture was incubated at 200 rpm for 1 h at 30◦C. After centrifugation at 8 000 rpm for 2.5 min, the cell pellet was


resuspended in 600 µl of LB and 150 µl were spread on a solid LB plates with the appropriate antibiotic.

#### Cultures Conditions

For molecular biology techniques, the bacteria strains were cultured in the LB medium (10 g/L trypton, 5 g/L yeast extracts and 5 g/L NaCl), and 5 g/L agar was added for solid medium in petri dishes. For GA production, the mineral medium M9 was used. Unless otherwise stated, it contained per liter: 18 g Na2HPO<sup>∗</sup> 4 12H2O, 3 g KH2PO4, 0.5 g NaCl, 2 g NH4Cl, 0.5 g MgSO<sup>∗</sup> 4 7H2O, 0.015 CaCl<sup>∗</sup> 2 2H2O, 1 ml of 0.06 M FeCl<sup>3</sup> from a 1000 × stock solution in concentrated HCl, 2 ml of 10 mM thiamine HCl stock solution, 20 g MOPS, and 1 ml of trace element solution (1000 × solution of 0.5 g Na2EDTA<sup>∗</sup> 2H2O, 0.18 g CoCl<sup>∗</sup> 2 6H2O, 0.18 g ZnSO<sup>∗</sup> 4 7H2O, 0.4 g Na2MoO<sup>∗</sup> 4 2H2O, 0.1 g H3BO3, 0.12 g MnSO<sup>∗</sup> <sup>4</sup>H2O, 0.12 g CuCl<sup>2</sup> × H2O). The carbon source D-glucose, L-arabinose or D-xylose was added at a final concentration of 10 g/l, the pH was adjusted to 7 with acid 3-(N-morpholino) propanesulphonic (MOPS) and then filter sterilized through 0.2µm membranes. The antibiotics ampicillin, kanamycin and chloramphenicol were added when required at concentrations of 100, 50, and 25 mg/L, respectively. For the E. coli strains defective in transketolase activity (1tktA 1tktB), M9 medium was supplemented with 500µM L-phenylalanine, 250µM L-tyrosine, 200µM L-tryptophan, 6µM p-aminobenzoate, 6µM p-hydroxydenzoate, and 280µM shikimate and trace of LB (20% <sup>V</sup>/V). For the strains defective in glyceraldehyde-3-phosphate dehydrogenase activity (1gapA), the M9 medium was completed with 0.4 g/L malic acid adjusted at pH 7 with KOH. When required, the appropriate antibiotics was added to the medium at 100µg/mL for ampicillin, 50µg/mL for kanamycin, or 25µg/mL for chloramphenicol. The bacteria cultures were placed in rotatory shaker at 200 rpm and at 37◦C. Growth was monitored by measuring absorbance at 600 nm with a spectrophotometer (Biochrom Libra S11).

### Enzymes Production, Purification, and Assays

The kdsD, fsaA, and aldA genes were amplified using primers in **Table S1** and cloned in the expression vector pET28a (Novagen). The E. coli strain BL21 (DE3) transformed with the plasmid bearing these genes were inoculated from a pre-culture made in LB-kanamycin (50 mg/ L) in 200 mL of LB-Kanamycin at 600 nm (DO600) of 0.1 at 16◦C in a rotary shaker at 200 rpm. When DO<sup>600</sup> reached 0.6–0.8, the expression of the protein of interest was induced by addition of IPTG at 1 mM final concentration 1 mM IPTG. After 16 h at 16◦C, the culture was collected by centrifugation at 4,800 rpm for 15 min at 4◦C. The cell pellet was re-suspended in 1.5 mL washing buffer (50 mM HEPES, pH 7.5; 0.3 M NaCl) and sonicated four times (30 s each) at 30% power on ice. The HIS-tagged proteins were purified using cobalt resin according to the protocol described in commercial kit (Clontech). Purification of the protein was verified by SDS-PAGE electrophoresis and protein concentration was measured by Bradford method (Bradford, 1976)

Enzymes assays were made in a Tris-Cl, 100 mM pH 7.5/10 mM MgCl<sup>2</sup> buffer at 37◦C. Unless otherwise stated, KdsD activity was measured in a coupled assay in the presence of 3 mM NAD<sup>+</sup> and 5 mM D-Ribu-5P with 10µg/ml of each of FsaA and AldA. FSA was measured by coupling GAP produced from cleavage of 3 mM Ara5P with the oxidation of 0.2 mM NADH at 340 nm in the presence of 1 U/ml of triose isomerase and glycerol-3-P dehydrogenase. AldA was measured in the same buffer by reduction of NAD<sup>+</sup> (3 mM) at 340 nm in the presence of 5 mM glycolaldehyde.

# Analytical Methods

Extracellular metabolites were determined by high performance liquid chromatography (HPLC) with an Ultimate 3000 chromatograph (Dionex, Sunnyvale, USA). The HPLC system was equipped with a cation exchange column (Aminex, HPX87H; 300 × 7.8 mm, 9µm, BioRad), an automatic injector (WPS-3000RS, Dionex), an IR detector (RID 10A, Shimadzu) and a UV detector (SPD-20A, Shimadzu). The sample injection volume was 20 µL, and the compounds were separated in an Aminex HPX-87H column protected by a Micro-Guard Cation H pre-column (BioRad, USA). The separation was performed at 35◦C with 1.25 mM H2SO<sup>4</sup> at 0.5 mL min−<sup>1</sup> as mobile phase. All samples were centrifuged (2 min at 10,000 g) and syringefiltered (0.2µm), and the resulting supernatant kept at −20◦C until analysis.

# Genome-Scale Modeling and Flux Balance Analysis

The iJO1366 genome-scale E. coli model (Orth et al., 2011) has been adapted to simulate GA production through the glycoptimus pathway. Notably, a modified fsaA reaction was added to the model to allow the conversion of arabinose-5P into glyceraldehyde-3P and glycolaldehyde. Parsimonious flux balance analyses (pFBA) were performed using the OptFlux software (Rocha et al., 2010), using either D-glucose, L-arabinose or D-xylose as a carbon source. For each simulation, the uptake rate of the carbon source was arbitrarily set at 10 mmol.g−<sup>1</sup> CDW.h−<sup>1</sup> . Non-growth associated maintenance was set at 3.15 mmolATP.g−<sup>1</sup> CDW.h−<sup>1</sup> , and reactions involved in the electron transport system were constrained to simulate a realistic P/O ratio, as previously described (Orth et al., 2011). pFBA simulations were performed using GA export as the objective function to maximize.

# Statistical Methods

Statistical analyses were conducted in Microsoft Excel <sup>R</sup> using the Analysis ToolPak package. A two-tailed unpaired t-test was to use to compare fluorescence induction levels, in which an alpha level of p < 0.05 was set for significance.

# RESULTS AND DISCUSSION

## Pathway Design and Modeling

The pathway design that could maximally convert the lignocellulose-derived sugars D-glucose, D-xylose and Larabinose into GA with, in theory, no loss of CO<sup>2</sup> is illustrated in **Figure 2**. This route relies on two key enzymatic steps that are sequentially catalyzed by an isomerase encoded by kdsD/yrbH (Meredith and Woodard, 2003), which interconverts D-Ribu-5P and D-Ara-5P, and an aldolase encoded by fsaA (Schurmann and Sprenger, 2001) that cleaves this intermediate into glycolaldehyde and GAP. The latter is shuttled back to D-Ribu-5P via the pentose phosphate pathway. Overall, this cyclic pathway should in theory produces 3 moles GA per mole of consumed D-glucose and 2.5 moles GA per mole of consumed pentose sugar, as confirmed by parsimonious flux balance analyses (**Table S2** and **Figure S1**). The proposed pathway is thermodynamically favorable as deduced from the sum of 1rG'<sup>0</sup> of each reaction using Equilibrator (http://equilibrator. weizmann.ac.il/) (**Figure S2**). The large negative Gibbs energy results from the phosphorylation of the sugar and the oxidation of glycolaldehyde by the glycoladehyde dehydrogenase encoded by aldA (**Figure 2**). In addition, this metabolic pathway which we termed "glycoptimus" is redox balanced since NADH produced can be regenerated to NAD<sup>+</sup> via oxidative phosphorylation, leading to ATP production. This energy provision is sufficient to sustain GA production, based on the assumption that (i) 2 moles ATP are produced per mole of NADH and (ii) 2 moles of ATP are required for the active transport and phosphorylation

FIGURE 2 | Scheme of the glycoptimus pathway. Key steps are D-arabinose-5-phosphate isomerase (KdsD) and fructose-6-phosphate aldolase (FsaA) shown as red arrows. In green is shown the oxidation of glycoladehyde to glycolate by glycoladehyde dehydrogenase AldA. Glk, glucokinase; Pgi, phosphoglucoseiomserase, Tal, transaldolase; Tkt, transketolase; Rpi, ribose-5-P isomerase; Rpe, ribulose-5-P epimerase; Fba, fructose1,6-bisphosphate aldolase; Tpi, triose isomerase; Fbp, fructose1,6-bisphosphate phosphatase; xylA, xylose isomerase; xyB, xylulose 5-P kinase; araA, arabinose isomerase; araB, L-ribulose-5-P kinase. Stoichiometry of the glycoptimus reaction with sugar is given below.

of each mole of sugar. As flux balance analyses equilibrate redox and energetic balances, modeling of the glycoptimus pathway highlighted that optimal flux toward GA production should be accompanied with a reduction of the P/O ratio. Indeed, the model adjusts the production of ATP to the needs of the system (i.e., sugar uptake and non-growth associated maintenance). Excess of redox cofactors is therefore equilibrated by the introduction of futile cycles within the model. These results highlight the fact that an in vivo strategy toward a maximal GA production yield might require further process development (e.g., use of micro-aerobic conditions) to adapt oxidative phosphorylation and redox balance during the production phase. Alternatively, the introduction of additional reactions to oxidize the excess of NADH produced by AldA could be considered.

# Kinetic Properties of Key Enzyme Encoded by kdsD and fsaA and in vitro Validation

Our rational to design this pathway arose from the fact that E. coli genome is endowed with kdsD/yrbH gene that encodes an Ara5P isomerase (Meredith and Woodard, 2003). The physiological function of this enzyme is to generate D-Ara-5P from D-Ribu-5P as the first precursor in the biosynthesis of 3-deoxy- (D)-manno-octulosonate (KDO), a sugar moiety located in the lipopolysaccharide layer of most Gram-negative bacteria. We have revised the kinetic properties of the three key enzymes in the proposed pathway, namely KdsD, FsaA, and AldA, all being expressed in E. coli BL21 (DE3) strain as HIS-tagged fusion proteins. The enzyme KdsD was confirmed to be highly specific toward D-Ara-5P and D-Ribu-5P. However, the determined catalytic constant (kcat) and Michaelis-Menten affinity constant (KM) values for the D-Ribu-5P substrate were found to be distinct of those previously reported by Meredith and Woodard (2003). In particular, a four-fold higher K<sup>M</sup> (D-Ribu-5P, ∼1.3 mM) and a 10-fold lower kcat (D-Ribu-5P, 24 s−<sup>1</sup> ) were obtained in our study (**Table 4)**. These differences could be ascribed to the presence of the HIS-tag at the N-terminus of the protein, which could interfere with the catalytic activity of this enzyme. For the aldolytic cleavage of D-Ara-5P into glycolaldehyde et GAP, the aldolase encoded by fsaA was retained based on the work of Garrabou et al. (2009) which reported a higher specificity of this enzyme to D-Ara-5P than to F6P; the latter being cleaved into dihydroxyacetone and GAP (Schurmann and Sprenger, 2001). While the physiological function of this class I aldolase in E. coli is still unclear, this enzyme is profitably employed as a biocatalyst for stereoselective C-C bond formation (Clapes et al., 2010). The K<sup>M</sup> and kcat determined on D-Ara-5P using purified His-tagged FsaA enzyme (named FsaA-HIS) was in the range of those obtained by Garrabou et al. (2009) (**Table 4**). Since KdsD and FsaA both catalyse reversible reactions, the feasibility of this pathway is ensured by coupling the NAD+-dependent glycolaldehyde dehydrogenase to the oxidative phosphorylation chain (resulting in NAD<sup>+</sup> regeneration coupled to ATP synthesis). We thus determined the kinetic properties of the purified AldA (**Table 4**). While the K<sup>M</sup> for glycoladehyde was similar to that reported in the original work of Caballero et al. (1983), the kcat was 10-fold lower, which may be due to difference in enzyme purification and condition of enzymatic assay. Nonetheless, determination of catalytic efficiencies (expressed as kcat/KM) indicated that FsaA is most likely the rate-limiting step in the metabolic route yielding GA from D-Ribu5-P (**Table 4**).

We then validated the functioning of the glycoptimus route in vitro by incubating D-Ribu5-P with purified KdsD, FsaA, and AldA (**Figure 3**). Pathway validity was assessed by following the increase of absorbance at 340 nm corresponding to NADH production coupled with the stoichiometric formation of GA as confirmed by HPLC analyses. In the absence of either KdsD or FsaA, no NADH formation was observed as expected.

# Growth-Based Screening of the Glycoptimus Pathway

The most direct way to show that a synthetic pathway works in vivo is to express the genes of that pathway in a plasmid under the dependence of an inducible promoter. Upon promoter induction with a suitable inducer, the production of the compound of interest can be monitored. However, it is also wiser to associate the implementation of a synthetic pathway with a phenotypical trait that is fully dependent of the in vivo functioning of the pathway. With this procedure, effects of plasmids copy number, promoter strength, RBS, position of the genes in the operon can be investigated in a combinatorial manner to find out the expression modules leading to the best production pathway. With respect to our glycoptimus pathway, we took advantage of the previous finding that an E. coli strain defective in transketolase encoded by tktA and tktB is unable to grow on D-xylose or L-arabinose (Josephson and Fraenkel, 1969, 1974). Expression of kdsD-fsaA-aldA module pathway shall restore cell growth in the aforementioned sugar substrates. With this approach (see **Figure 2**), the assimilation of D-xylose or L-arabinose will result in the formation of glycoladehyde and GAP. While the former can be oxidized into GA, GAP will be used for cell growth (see **Figure S3**). The herein proposed screening strategy shall in principle meet two criteria: (i) growth rescue of a 1tktA 1tktB mutant on D-xylose (or L- arabinose), and (ii) GA production correlated to growth proficiency of the mutant strain bearing the pathway. To avoid degradation of GA product, the glcD gene encoding the glycolate oxidase (Pellicer et al., 1996) has been also deleted from the template screening strain.

Accordingly, we initially constructed a single kdsD-fsaA-aldA operon under the control of the IPTG-inducible Plac promoter expressed from medium- and high-copy number plasmids from the pZ collection. Unexpectedly, none of these constructs rescued growth of the MG16551tktA 1tktB 1glcD on D-xylose. Upon sequencing of these plasmids, we systematically found a deletion in the Plac promotor, which was due to a repeat motif of this promotor present in these pZ plasmids. This event has been recognized earlier and explained by a replication-slipping mechanism, which often occurs due to the presence of short repeat motifs in these plasmids (Kawe et al., 2009). Replacement of Plac by Ptac was also unsuccessful, resulting instead in Tn10 transposon insertion in the kdsD gene. We therefore changed our strategy by expressing aldA separately from kdsD-fsaA. We found that splitting the operon in two different but compatible plasmids enabled growth rescue of the 1tktA 1tktB 1glcD strain. Encouraged by this observation, we constructed 24 different expression systems in which combination of 3 types of vectors carrying three different but compatible replication origin and 4 different promoters (2 inducible Ptac and Plac and 2 constitutive proC and prod, Davis et al., 2011) were attempted, while genes order of kdsD-fsaA and their original RBS sequences remained unchanged. These 24 different expression systems **(**see details on plasmid constructs for this purpose in **Table S3** and **Figure S4**) were transformed into MG16551tktA 1tktB 1glcD strain. After an overnight pre-culture in M9 medium with 10 g/l of Dglucose supplemented with 0.1% (w/v) LB and 0.1% (w/v) yeast extract, the cells were collected by centrifugation, washed once with sterile water and resuspended at an initial 0D<sup>600</sup> of 0.5 in M9 containing 10 g/L D-xylose complemented with the necessary auxotrophic requirements. After 46 h of cell cultivation during which growth was monitored (notice that growth rate of the transformants was rather low in the range of 0.03 to 0.05 h−1, data not shown), the supernatant was collected to



of purified KdsD, FsaA, and AldA enzyme.

measure GA produced and D-xylose consumed by HPLC. The results supported in part our assertion as the amount of GA produced was roughly correlated (R <sup>2</sup> ∼ 0. 53) with growth (**Figure 4A**). A slightly better correlation was obtained between GA produced and D-xylose consumed (R <sup>2</sup> ∼ 0.72, see **Figure S5**), which may indicate that GA was produced in part independently to growth. Also, it was unanticipated to find a significant production of GA (4.7 mM) in strains bearing empty plasmids (namely strain Screen00 and Screen14), which otherwise hardly grew on D-xylose. A careful analysis of the exometabolome in these transformants showed the presence of D-xylulose whose accumulation was the highest in the control strains Screen00 and Screen14. Moreover, an inverse correlation could be drawn by plotting GA yield per D-xylose vs. D-xylulose yield per Dxylose (see **Figure S5B**). While these data support the notion that GA production is coupled to growth, it still does not explain where GA comes from in the non-growing strains. There is at least two possibilities, which are likely additive. On the one hand, GA could arise from the aldolytic cleavage of D-xylulose by FsaA. We indeed found a weak activity of this enzyme on D-xylulose, which exhibited a very weak affinity on this substrate (**Figure S6**). Alternatively or complementary to this possibility, part of the hyperaccumulated phosphorylated sugars that are found in the non-growing strain (up to 15 mM, data not shown) can be diverted to GA using the endogenous activity of KdsD, FsaA, or its homolog FsaB and AldA (data no shown).

From the correlation of GA produced to D-xylose consumed, it turned out that strains termed Screen09 and Screen23 were the best producers with a product yield of 0.48 mol GA/mol of D-xylose (**Figure 4B**). Interestingly, in both strains, the kdsDfsaA operon and aldA gene were carried out in a mediumand a low-copy plasmid, respectively. While the expression system of these genes was IPTG inducible in strain Screen09, kdsD-fsaA, and aldA were driven by the constitutive promoter proD and proC, respectively. We also validated that these two strains could produce GA from L-arabinose. As shown in **Figure 5**, the production of GA on D-xylose and L-arabinose was significantly better in strain Screen09 and Screen23 than in the control strain Screen00, which showed that the module pathway driven either by an IPTG inducible (strain Screen09) or

flask. Growth was monitored at 600 nm (A) and after 150 h, when OD<sup>600</sup> was stable, which is after 150 h, sample was taken to measure GA in the supernatant. (B)

by constitutive promoters (strain Screen23) was operational. In addition, although the titer of GA on D-xylose was higher than on L-arabinose, the yield was actually comparable (i.e., 0.5 mole / mole of sugar), indicating that D-xylose was better assimilated than L-arabinose.

Results shown are the mean of two independent experiments.

# Further Engineering of the Central Carbon Metabolism and Effect on GA Production

While the results reported above proved that the kdsD-fsaA + aldA module pathway is functional in vivo, we wanted to go one-step further by refactoring the central carbon metabolism to drive more carbon for GA production. Therefore, our starting E. coli strain was W3CG, which is defective in the lower part of glycolysis due to a Tn10 transposon insertion into gapA encoding glyceraldehyde-3P dehydrogenase (Ganter and Pluckthun, 1990). The homologous gene gapB was found to be inactive in this strain (Tsuruno et al., 2015). Hence, this strain requires a gluconeogenic substrate such as malate in addition to D-glucose to grow. Our metabolic engineering for maximal GA production implied to delete competing but non-essential pathways for growth, as they could siphon intermediates out of the glycoptimus pathway (genetic intervention are schematically depicted in **Figure S7**). Therefore, the oxidation of glycolate into glyoxylate was abrogated by deletion of glcD encoding glycolate dehydrogenase (Pellicer et al., 1996). Meanwhile, arcA, which encodes the DNA-transcriptional regulator of the twocomponent system ArcAB implicated in redox state signaling (Malpica et al., 2006) was disrupted since it was reported to repress aldA (Pellicer et al., 1999). Deletion of mgsA encoding methylglyoxal synthase (Saadat and Harrison, 1998) and fucA encoding a fuculose aldolase was also carried out in the recipient strain to reduce siphoning of DHAP/GAP out of the glycoptimus pathway. We also deleted pfkA encoding PFK-1 isoform I, which contributes to 90% of the phosphofructokinase activity (Kotlarz et al., 1975) to avoid wasteful recycling of F6P into F1,6P2. Finally, the sugar/H<sup>+</sup> transporter encoded by galP was overexpressed

by swapping its own promoter by the strong constitutive proD promoter to favor uptake of D-glucose and D-xylose (Henderson, 1990).

This engineered strain termed Glyco00 was transformed with the constitutive expression system 23 corresponding to plasmid pKF6 (kdsD-fsaA under proC promoter in a medium copy plasmid) and pA7 (aldA under proD promoter in a low copy plasmid) to yield strain Glyco23. Both the untransformed and transformed strains were firstly cultivated in mineral medium M9 in the presence of malate (5 g/L) and D-xylose (1 g/L), with a growth rate in the range of 0.07 −0.11 h−<sup>1</sup> . After complete consumption of malate, D-xylose, L-arabinose or Dglucose were added at 5 g/L. GA production was monitored for 24 h. It can be seen in **Figure 6** that strain Glyco23 was able to produce GA from hexose and pentose sugars. However, the production yield was only 0.2, 0.6, and 0.68 mole GA per mole of D-glucose, D-xylose and L-arabinose, respectively (**Figure 6B**). This very low yield is not in accordance with our model. In addition, only a small amount of sugar has been consumed during the 48 h-incubation period, suggesting that the expression of the glycoptimus pathway has impaired the energetic status of the cell. Ongoing work is underway to clarify this problem. It is interesting to notice that the control strain Glyco00 did not accumulate GA, which supports the hypothesis that the production of GA seen in screen00 (see **Figure 4**) might originate from D-xylulose and/or phosphorylated sugars that were accumulated, what did not occur in Glyco00 because the pentose phosphate pathway was operational.

Several hypothesis can be drawn to optimize the carbon flux toward GA production with additional metabolic engineering.

FIGURE 6 | Titer and yield of GA in strain Glyco and Glyco23 on D-xylose and L-arabinose. Strain Glyco00 corresponded to E. coli WC3G 1gapA 1glcD 1arcA 1mgsA 1fucA 1pkf proD-galP, transformed with empty vector pZA36 and pZS27 whereas pKF6 and pA7, was used to yield Glyco23 strain. The strains were preliminary cultivated in 50 ml of mineral medium M9 containing 5 g/L malate and 1 g/L D-xylose at 37◦C in 250 ml shake flask at 37◦C. When malate was completely consumed, 5 g/L of either D-xylose or L-arabinose was added and the culture was further incubated for 24 h. (A) Titer and (B) yield were determined after 24 of cultures. The results are the mean ± SD of two independent experiments.

According to our model, the glucose-6P dehydrogenase encoded by zwf, which catalyzes the entrance of D-glucose into the pentose phosphate pathway should be deleted, as well as of eda-edd of the Entner-Doudoroff, as both reaction can bypass deletion of gapA and generate pyruvate which will then feed the TCA cycle. Additionally, sugar transport can be a valuable target for optimization. Notably, the phosphoenolpyruvate-dependent phosphotransferase (PTS) system does not induce a carbon loss per se, but it causes some deviation of PEP that contributes to biomass production. Loss of PTS function may have two significant advantages: better assimilate pentose sugars and make possible to uncouple growth and production over time. A bi-phasic process can further be developed to improve the yield. The first phase will be dedicated to the production of biomass from an affordable carbon source in C2, C3, or C4 (ex: malate) to fuel TCA cycle. Meanwhile, the second phase of the fermentation will be dedicated to the production of glycolic acid from lignocellulosic sugars. As highlighted by the flux balance analyses, this second phase could also benefit from further process development, notably by using micro-aerobic conditions to adapt oxidative phosphorylation and redox balance during the production phase. By combining further metabolic engineering and process development, it should be feasible to reach GA production yield close to theoretical maximum yield of 3 moles of GA per mole of C6 and 2.5 moles per mole of C5.

#### CONCLUSION

A new non-natural pathway for GA production has been conceived, implemented and in vivo validated. This new pathway termed glycoptimus relies on two pillars. On the one hand, it required the overexpression of three naturally occurring E. coli genes, namely kdsD, fsaA and aldA, whose physiological role are still unclear except for kdsD. This synthetic pathway combined with the refactoring of the central carbon metabolism to minimize the carbon loss as CO<sup>2</sup> at the level of pyruvate should allow to yield theoretically 2,5, and 3 moles of GA from lignocellulosic pentose and hexose, respectively. We successfully demonstrated that this pathway was operational in vivo, leading to the production of GA from D-glucose, D-xylose and Larabinose, albeit at yield that was only at 20–30% of the theoretical ones. Nevertheless, these results argued for a great potential of this microbial process in term of industrial feasibility which will require further optimisation including among others, additional metabolic and strain engineering as well as efficient coupling of NADH reoxidation and energy requirement for sugars uptake and phosphorylation. In addition to ensure stability and robustness of this synthetic pathway, bioprocessengineering optimisation will be determinant notably because a two-stage process in which a production phase uncoupled from growth is likely the most favorable condition to be conducted to achieve high yield and titer of GA with engineered strains equipped with the glycoptimus pathway.

#### REFERENCES


#### DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

#### AUTHOR CONTRIBUTIONS

CL, CF, and FK performed the experiments. NM carried out the modeling. TW and JF provided guidance for the experimental setups and helped interpreting the results. JF and CL wrote the paper, which has been improved by TW, CF, and NM. All authors approved the final version.

### FUNDING

This study was supported in part by the' Agence Nationale de la Recherche - France (Synpathic project, ANR14-CE06-0024) to JF. CL was financed by MENERT (Ministery of Research and Education) during this work.

#### ACKNOWLEDGMENTS

We are grateful to Miss Lucie Spina for HPLC analysis.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00359/full#supplementary-material


glycolate oxidase and the glc regulator protein. J. Bacteriol. 178, 2051–2059. doi: 10.1128/jb.178.7.2051-2059.1996


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Lachaux, Frazao, Krauβer, Morin, Walther and François. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Development of a Biosensor for Detection of Benzoic Acid Derivatives in Saccharomyces cerevisiae

Sara Castaño-Cerezo<sup>1</sup> \*, Mathieu Fournié<sup>1</sup> , Philippe Urban<sup>1</sup> , Jean-Loup Faulon2,3 and Gilles Truan<sup>1</sup>

<sup>1</sup> TBI, Université de Toulouse, CNRS, INRA, INSA, Toulouse, France, <sup>2</sup> Micalis Institute, INRA, AgroParisTech, Université Paris-Saclay, Jouy-en-Josas, France, <sup>3</sup> Chemistry School, Manchester Institute of Biotechnology, The University of Manchester, Manchester, United Kingdom

#### Edited by:

Jean Marie François, UMS3582 Toulouse White Biotechnology (TWB), France

#### Reviewed by:

Muralikannan Maruthamuthu, Purdue University, United States M. Kalim Akhtar, United Arab Emirates University, United Arab Emirates

> \*Correspondence: Sara Castaño-Cerezo castanoc@insa-toulouse.fr

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 01 August 2019 Accepted: 13 November 2019 Published: 07 January 2020

#### Citation:

Castaño-Cerezo S, Fournié M, Urban P, Faulon J-L and Truan G (2020) Development of a Biosensor for Detection of Benzoic Acid Derivatives in Saccharomyces cerevisiae. Front. Bioeng. Biotechnol. 7:372. doi: 10.3389/fbioe.2019.00372 4-hydroxybenzoic acid (pHBA) is an important industrial precursor of muconic acid and liquid crystal polymers whose production is based on the petrochemical industry. In order to decrease our dependency on fossil fuels and improve sustainability, microbial engineering is a particularly appealing approach for replacing traditional chemical techniques. The optimization of microbial strains, however, is still highly constrained by the screening stage. Biosensors have helped to alleviate this problem by decreasing the screening time as well as enabling higher throughput. In this paper, we constructed a synthetic biosensor, named sBAD, consisting of a fusion of the pHBA-binding domain of HbaR from R. palustris, the LexA DNA binding domain at the N-terminus and the transactivation domain B112 at the C-terminus. The response of sBAD was tested in the presence of different benzoic acid derivatives, with cell fluorescence output measured by flow cytometry. The biosensor was found to be activated by the external addition of pHBA in the culture medium, in addition to other carboxylic acids including p-aminobenzoic acid (pABA), salicylic acid, anthranilic acid, aspirin, and benzoic acid. Furthermore, we were able to show that this biosensor could detect the in vivo production of pHBA in a genetically modified yeast strain. A good linearity was observed between the biosensor fluorescence and pHBA concentration. Thus, this biosensor would be well-suited as a high throughput screening tool to produce, via metabolic engineering, benzoic acid derivatives.

Keywords: p-hydroxybenzoic acid, p-aminobenzoic acid, biosensor, synthetic biology, yeast

# INTRODUCTION

Synthetic biology proposes cutting-edge methodologies for using natural resources (Jullesson et al., 2015; Smanski et al., 2016; Le Feuvre and Scrutton, 2018; Schindler et al., 2018), understanding basic cellular functions (Metzger et al., 2018; Toda et al., 2018) or reprogramming cell fate (Black and Gersbach, 2018). Biosynthesis of small organic molecules by microorganisms, instead of manufacturing them from petroleum-sourced synthesized chemicals, is one of the main objectives of metabolic engineering (Khalil and Collins, 2010; Markham and Alper, 2015; Li et al., 2018). While the demand for green (sustainable) chemistry is growing steadily, there are still few examples of successful industrial production of molecules from microorganisms. The clear bottleneck in metabolic engineering lays in the low throughput of analytical techniques used to determine products yields compared to the rate of the construction of new engineered microorganisms (Rogers and Church, 2016). To overcome this limitation, natural or synthetic biosensors are potential tools capable of correlating the concentration of a target chemical within the cell to an easily monitored output signal, such as fluorescence (Farmer and Liao, 2000; Schulman and Heyman, 2004; Dietrich et al., 2010; Liu et al., 2017; Mannan et al., 2017).

The yeast Saccharomyces cerevisiae is a widely used model organism and production host, due to a comprehensive collection of metabolic engineering tools and deep knowledge of its genetics accumulated over half a century. However, the development of synthetic biosensors in yeast somewhat lags behind those available for E. coli (Leavitt and Alper, 2015). One of the first transcription factor-based biosensors was exemplified with the NhaR system from Pseudomonas putida. This system could drive the transcription of β-galactosidase as a function of benzoic acid and 2-hydroxybenzoic acid concentrations in E. coli. The system was applied to screen active enzymes capable of converting benzaldehyde and 2-hydroxybenzoaldehyde to their carboxylic acid derivatives (van Sint Fiet et al., 2006). Since then, several successful applications of biosensors have been described in E. coli (Selvamani et al., 2017; Ganesh et al., 2019) and more recently in cell-free systems (Eggeling et al., 2015; Voyvodic et al., 2019). However, transferring genetic sensors between various organisms remains highly challenging. For example, the simple transfer of a tetracycline resistance gene circuit from yeast to mammalian cells required extensive optimizations such as the translation of the reporter, the DNA sequences of the heterologous proteins, the nuclear localization signal of the transcription factor and the design of the promoter (Nevozhay et al., 2013).

In yeast, one of the first designed sensing systems was used to monitor the intracellular S-adenosylmethionine concentrations (Umeyama et al., 2013). Transcriptional-based biosensors were also constructed to screen for muconic acid-producing yeast strains (Leavitt et al., 2017; Snoek et al., 2018). Other types are now implemented, using optogenetic regulation such as the blue light-activated EL222 from Erythrobacter litoralis that was recently reported to control, in yeast, the mitochondrial isobutanol-producing pathway (Zhao et al., 2018).

4-hydroxybenzoic acid (pHBA), a molecule produced from chorismate by chorismate lyase, is present in low amounts in bacterial cultures (Winter et al., 2014). pHBA is essential in all organisms for coenzyme Q synthesis (Tran and Clarke, 2007) but was found to inhibit yeast growth when added to the culture medium (Ando et al., 1986; Palmqvist et al., 1999; Larsson et al., 2000). pHBA-derived natural products form a large group of secondary metabolites that exhibit a wide variety of biological activities (Wang et al., 2018). Industrial uses of pHBA include manufacturing of liquid crystal polymers and thermoplastics used for space technologies (Rothschild, 2016). The alkyl esters of pHBA, namely parabens, are also widely used as preservatives in drugs, cosmetic products and food products. However, their toxicity is a major human health concern (Giulivo et al., 2016). Currently, pHBA is chemically synthesized from petroleumderived building blocks. Biotechnology offers an alternative for pHBA production, based on the shikimate pathway (Lee and Wendisch, 2017). Recently, researchers have engineered yeast to overproduce pHBA, increasing the flux to chorismate and expressing, in the modified strain, the chorismate lyase gene (UbiC) from E. coli (Averesch et al., 2017).

Therefore, pHBA is a molecule for which in vivo biological detection is critical, and a performant pHBA biosensor could be used either to build detection kits capable of operating in various biological environments or to screen the capacity of modified strains to overproduce pHBA. In this study, we report the construction of a synthetic transcription factor (sTF) capable of sensing pHBA, in vivo, in Saccharomyces cerevisiae. We tested the transcriptional activity against the native effector of R. palustris HbaR and a battery of metabolites with similar structure and determine the promiscuity and binding characteristics of each of them. Last, we exemplified the activity of our newly constructed biosensor in a pHBAproducing strain.

#### MATERIALS AND METHODS

#### Culture Conditions

The vectors described in this work were constructed in the E. coli DH5α strain using standard molecular biology protocols (Sambrook and Russell, 2001). The CEN.PK 2C-1 S. cerevisiae strain (MATa; ura3-52; trp1-289; leu2-3, 112; his3:1; MAL2-8C; SUC2) was used throughout the entire study.

For the screening experiments, yeast strains were grown in synthetic medium containing 20 g/L of glucose, 6.9 g/L of yeast nitrogen base with or without pABA and Complete Supplement Mixture (CSM) drop-out minus uracil and histidine (Formedium). The different inducers were dissolved in the medium and pH was then adjusted to 5.5. All culture media were sterilized by filtration. For pHBA production, the synthetic medium without pABA and with CSM drop-out minus His, Ura, and Leu was supplemented with 76 mg/L tyrosine, phenylalanine and tryptophan.

Yeast strains were cultured as follows. A fresh yeast colony from a SD minus Ura and His agar plate containing pABA was resuspended in water to an OD600nm of 0.2 and 5 µL of this solution was used to inoculate 200 µL of media. Cells were grown at 30◦C in a MixMate (Eppendorf) with 900 rpm agitation for 20 h. For the pHBA production experiments, a fresh colony was resuspended in water to an OD600nm of 0.2 and 50 mL of culture medium was inoculated with 400 µL of this solution. Cells were grown at 30◦C with an agitation of 200 rpm in an INFORS incubator (25 mm orbital) for 65 h.

**Abbreviations:** bp, base pair; pHBA, 4-hydroxybenzoic acid; pABA, 4 aminobenzoic acid; 2HBA, 2-hydroxybenzoic acid (salicylic acid); 3HBA, 3 hydroxybenzoic acid; 2NBA, 2-aminobenzoic acid (anthranilic acid); 3NBA, 3-aminobenzoic acid; 2,5DHBA, 2,5 dihydroxybenzoic acid (gentisic acid); 3,4DHBA, 3,4 dihydroxybenzoic or protocatechuic acid; CSM, Complete Supplement Mixture; SD, Synthetic Defined medium; sTF, Synthetic transcription factor; sBAD, sensor of benzoic acid derivatives; YNB, Yeast Nitrogen Base.

#### Vectors and Strains Constructions

All plasmids and strains used in this study are listed in **Supplementary Tables 1, 2**. Yeast transformation was performed using the high efficiency protocol from Gietz (2014). The hbaR gene from Rhodopseudomonas palustris, the native ubiC gene from E. coli and the modified ARO4K226L gene from S. cerevisiae were codon-optimized for S. cerevisiae (Twist Bioscience). All DNA sequences can be found in **Supplementary Table 3**.

The pHBA sTF was constructed using the vector FRP880 as a backbone. FRP880 was digested with EcoRV to remove the estradiol-sensing domain and the HbaR-sensing domain was inserted by blunt-end cloning leading to the plasmid pSCC185. To create the ySCC185-F strain, we first integrated a DNA fragment expressing mKATE2 under the control of the TDH3 promoter into the locus 2 of chromosome X (Mikkelsen et al., 2012) and carrying the auxotrophic marker TRP1, yielding ySCC001. pSCC185 was then linearized with the restriction enzyme PacI, integrated in the HIS3 locus in ySCC001, yielding ySCC185. Finally, the linearized plasmid FRP795, containing a promoter with 8 LexA DNA binding domains and a CYC1 minimal promoter controlling the expression of the mCitrine fluorescent protein was integrated in ySCC185, yielding ySCC185-F. Integrations were verified by colony PCR and by functional analysis.

To build the pHBA-overproducing strain, we used the vector pENZ030 containing two homologous arms for the integration in the HO locus, a LEU2 auxotrophic marker and a bidirectional promoter formed by TDH3 and PGK and flanked by ADH1 and ADH2 terminators. The ubiC ORF was cloned downstream of the TDH3 promoter by cutting pENZ030 with XhoI. The ubiC gene was inserted in pENZ030 by isothermal assembly. The pHBA-producing strains were derived from ySCC185- F. First, the TRP3 gene was deleted using a hygromycin resistance cassette from the vector pUG75 flanked by a 38 bp homologous region of TRP3 yielding ySCC185-F-T. The ARO7 gene was then deleted and its promoter used to express the ARO4K229L mutant (Williams et al., 2015). To construct this, the previous strain was transformed with two cassettes. The first one contained the gene encoding the ARO4K229L mutant with two homologous regions of 40 bp, upstream for ARO7 and downstream for the second integration cassette. The second cassette contained the CYC1 terminator and G418 resistance. This DNA was amplified from the vector pMRI34 and the PCR containing homologous regions for the first cassette and the end of ARO7 gene yielding ySCC185-F-A. This last strain was used to express the E. coli ubiC gene, using a PCR amplified DNA fragment and the vector pENZ030- UbiC as a template, yielding ySCC185-UbiC. The control strain for the pHBA production experiments was derived from ySCC185-F-A with an integrated copy of the empty vector pENZ30 (ySCC185-30).

#### Flow Cytometry

Cells were incubated for 30 min with 70µg/mL cycloheximide (prepared in DMSO), further diluted to reach a cell count between 0.5 × 10<sup>6</sup> and 1.5 × 10<sup>6</sup> cells/mL and then immediately injected into a flow cytometer MACSQuant VYB (Miltenyi Biotec, Germany). Regions were determined as a function of the mCitrine and mKate2 fluorescence. Optimal laser and filter setups for the two dyes were as follows: 488 nm laser and 525/25 Band Pass B1-filter for mCitrine, and 561 nm laser and 615/10 Band Pass Y2-filter for mKate2. The expression profile of mKate2 was measured for each sample by the MACSQuant VYB flow cytometer with the MACSQuantify TM Software (Miltenyi Biotec, Germany). A filter was applied on FSC-A/SSC-A to select homogeneous cells regarding size, shape, and cellular complexity. The mean fluorescence value of mKate2 and mCitrine was calculated and exported.

#### Western Blot

Protein extracts were prepared following the protocol described by Zhang et al. (2011). Briefly, 1 OD600nm of pelleted cells were pre-treated with 100 µL of 2 M lithium acetate solution, left standing on ice for 5 min, followed by 5 min centrifugation at 5,000 g, 4◦C. The supernatant was discarded and 100 µL of a 0.4 M solution of NaOH added. After gentle resuspension and 5 min standing on ice, samples were centrifuged 5 min at 4 ◦C. Cell pellets were vigorously vortexed with 60 µL of a dye solution containing bromophenol-blue and supplemented with 5% β-mercaptoethanol. After denaturation at 99◦C for 10 min, 10 µL of each sample was deposited on a 10% SDS page gel. Semi-dry transfer was performed on a PVDF membrane (Merck Millipore, Darmstadt, Germany) using a Trans-Blot <sup>R</sup> SD Cell BioRad apparatus (15 V at 600 mA during 30 min). Five percent powdered milk in TBS was used as a blocking agent. Mouse primary antibody anti-EGFP (Thermoscientific), and secondary anti-mouse IgG coupled with alkaline phosphatase (Thermoscientific) were diluted following instructions from the provider. Antibody incubations were performed in 5% powdered milk in TBS during 1 h. Proteins were detected by the incubation of BCIP/NBT AP substrate buffer (Sigma-Aldrich, St. Louis, MO, USA).

#### pHBA HPLC Analysis

Supernatants of pHBA-producing strains were collected during 3 days of batch growth. Analysis of the pHBA production was carried out on a Waters Alliance HPLC system coupled with a 996 Waters PDA detector. Twenty microliter of the culture supernatant was injected on a Xterra column (100 × 2.0 mm and 3µm particle size) maintained at 45◦C during the analysis. The mobile phases consisted in a mixture of A (H2O with 0.1% formic acid) and B (acetonitrile with 0.1% formic acid). The flow was set at 1 mL/min with the following gradient: 0–0.1 min 100% A, 0.1–11 min 40% A and 60% B, 11–12.5 min 100% B, 12.5–13 min 100% A. Absorbance spectra from 210 to 400 nm were recorded. pHBA was quantified by its absorbance at 254 nm.

#### Statistical Analysis

Statistical analyses were performed using the GRAPHPAD PRISM 8.2 software.

# RESULTS AND DISCUSSION

## Design and Validation of the Synthetic Transcription-Based Biosensor

The architecture chosen for the pHBA sTF was inspired by the work performed by Stelling and co-workers, in which a collection of modular tripartite sTFs, activated by estradiol, were developed (Ottoz et al., 2014). We thus designed our sTF using LexA and B112 proteins as DNA binding and transactivation domains, respectively (**Figure 1**). We cloned the ORF of the HbaR ligand binding domain from R. palustris, known to be specifically activated by pHBA (Egland and Harwood, 2000), between the transactivation and DNA binding domains (pSCC185, see **Supplementary Table 2**). To report the transcriptional activation of our synthetic pHBA biosensor, designated as sHbaR, we used the previously constructed vector FRP795 containing the gene encoding for the mCitrine yellow fluorescent protein downstream of the minimal CYC1 promoter with 8 LexA DNA binding sites (Ottoz et al., 2014). The resulting strain, ySCC185-F, bears both the reporter gene and the sHbaR sTF.

To validate transcriptional activation by pHBA, the strain ySCC185-F was cultured in 96-well plates at 30◦C with or without pHBA (2 mM). mCitrine fluorescence was measured at late exponential phase (OD600nm = 4) to get the best sensitivity possible with a rather small volume of culture. As shown in **Figure 2**, the mCitrine fluorescence is 3-fold higher with pHBA compared with the control medium that did not contain the inducer. We also verified that the measured fluorescent signal is proportional to the quantity of mCitrine measured by Western Blot (**Supplementary Figure 1**). Notwithstanding the fact that a previous in vitro study described a sTF activated by pHBA (Yao et al., 2018), to our knowledge, sHbaR is the first biosensor that can be activated by pHBA in vivo, in S. cerevisiae.

#### pABA Also Activates the sHbaR Sensing System

Analysis of the first experiments demonstrated that the fluorescent signal measured with the strain ySCC185-F in the

absence of pHBA is greater than the signal measured without the reporter (**Supplementary Figure 2A)**. This could be due either to some leakage of the synthetic promoter in the absence of the sTF or promiscuous activation of sHbaR by molecule(s) present in the medium or in the cells. In fact, promiscuity of benzoic acid-responsive natural transcription factors has been reported before, evidencing that they can respond, to different degrees, to a variety of substituted benzoic acids (but not pHBA) (Xue et al., 2014). Therefore, we hypothesized that a compound present in the synthetic medium could be also an effector of sHbaR.

The most similar metabolite to pHBA found in our synthetic media is p-aminobenzoic acid (pABA), which, according to the supplier's medium composition, is present at a final concentration of 1.46µM. We thus analyzed if the low levels of pABA in this medium could activate sHbaR. As expected,

the fluorescence signal produced by the strain ySCC185-F is reduced by almost 30% in a synthetic medium devoid of pABA compared to the same one containing pABA (p-value 0.0014, **Figure 3**). This indicates that a rather small concentration of this metabolite in the growth medium indeed activates sHbaR in vivo. We also tested if tyrosine and phenylalanine that possess rather similar chemical structures to pHBA and that are present in the CSM at a concentration of 300µM each could also activate sHbaR. We grew ySCC185-F with and without these amino acids and observed no difference between both conditions (**Supplementary Figure 2B**). We also supplemented each of the two amino acids to a final concentration of 2 mM and could not observe any difference in the level of fluorescence compared with the control medium (**Supplementary Figure 2B**). Since the promoter-reporter construction that we have used showed absolutely no leakage of transcription in the absence of inducers in a previous study (Ottoz et al., 2014), we thus attributed the remaining fluorescence signal to the presence of other molecules capable of activating sHbaR.

To further understand the possible competition or synergy between pABA and pHBA, we grew the strain ySCC185-F in YNB (containing or not pABA) in the presence of externally added pABA or pHBA (2 mM) (**Figure 3**). When pABA is present in the medium, the addition of 2 mM of pHBA or pABA induced, respectively, a 3- or 4-fold fluorescence increase. The same trend is observed on a synthetic medium devoid of endogenous pABA, i.e., a 4-fold augmentation of fluorescence with 2 mM pHBA and 6-fold with 2 mM pABA. We thus hypothesize that there is no competition or synergy between these two metabolites and sHbaR and that pABA is simply a better effector for sHbaR than pHBA.

#### Promiscuity of the Synthetic sHbaR Sensing System

Since sHbaR seems to not completely distinguish between two different substituents at the para position of benzoic acid, we decided to change the name of our biosensor to sBAD (previously known as sHbaR), standing for "sensor of benzoic acid derivatives." We further investigated the promiscuity of sBAD with different benzoic acid-derived chemicals at 2 mM each, in the absence of pABA in the growth medium (**Figure 4**). Interestingly, benzoic acid itself leads to a 5-fold activation of sBAD, a value comparable to the one obtained with pHBA (∼4 fold) and pABA (∼6 fold). The other three molecules substituted in the ortho position (2HBA (salicylic acid), 2NBA (anthranilic acid), and aspirin) were able to produce a low, but significant (2-fold) response of sBAD. A general trend for the specificity is remarkably conserved between hydroxy- or amino-substituted benzoic acid derivatives: the para position is the preferred one, followed by the ortho position and the meta position, which is the least potent for sBAD activation. For the rest of the tested metabolites, only those mono-substituted with a hydroxyl or amino group at the ortho or para position of the benzoic ring provided a significant increase of fluorescence compared to the control. Surprisingly, a larger substituent at the ortho position (aspirin, **Figure 4**H) can also activate sBAD. On the contrary, all other di-substituted benzoic acid derivatives were incapable of activating sBAD. This result indicates that sBAD has a rather tight control of the number of substituents on the aromatic ring.

It is known that the positions and the chemical properties of benzoic acid substituents influence the pK<sup>a</sup> of the carboxylic group of the corresponding chemicals (**Supplementary Table 4**). A difference of 1.5 pH units indeed exists between the orthohydroxybenzoic acid (pK<sup>a</sup> = 3) and pHBA (pK<sup>a</sup> = 4.6). Thus, all tested metabolites have different pKas and their diffusion through the plasma membrane of yeast might be slightly different. Moreover, it is known that the rates of import of benzoic acid derivatives (p-coumaric acid and pHBA) in S. cerevisiae are different (Barnhart-Dailey et al., 2019). However, as pHBA, 2HBA, pABA, 2NBA, benzoic acid, and aspirin all activate sBAD in vivo under the conditions tested, we assume that they are transported in the cell in order to activate the intracellular sBAD and trigger the corresponding fluorescence signal. Alternatively, the diverse responses obtained could also originate from different binding affinities of the different compounds to the sTF. As no crystal structure of the native transcription factor HbaR from R. palustris with its cognate binder (pHBA) is available, we can only hypothesize that the delocalization of π-electrons of the aromatic benzene ring may partly control the recognition of the inducer by the binding domain of the sBAD biosensor.

## Dynamic Properties of the sBAD Sensing System

To obtain a more accurate resolution of the cellular responses and the operational dynamic range of the most active inducers, we quantified the binding affinity using dose-response curves with the yeast-expressed sBAD (**Figures 5A–D**). For the metabolites

2HBA and 2NBA, we could not determine the different parameters since the signal never reached saturation at the highest concentrations tested (**Supplementary Figure 3**). This was probably due to some growth defect of our strain cultured with 2HBA at a concentration over 2 mM and the difficulty in dissolving 2NBA at concentrations higher than 3 mM.

In our experimental conditions, pABA and benzoic acid induced sBAD at much lower concentrations than pHBA which is supposed to be the natural inducer of HbaR in R. palustris. The EC50 for pHBA is 4- and 15-fold higher than the ones measured for pABA and benzoic acid, respectively (**Table 1**, **Figure 5E**). It should be noted that both the EC50 and dynamic range values were slightly smaller with aspirin than with pHBA. The biggest dynamic range was found with pABA, which in any case showed the best performance as an effector compared to the three others tested activators (**Figure 5E**). Two hypotheses can explain our results. The first one relates to the three-dimensional structure of sBAD. In our design, the effector binding domain

TABLE 1 | Affinity and dynamic range parameters of sBAD with the inducers HBA, pABA, benzoic acid, and aspirin.


Reported values are calculated as the mean of three biological replicates ± SD.

was inserted between two other proteins, thus possibly changing the global fold of the HbaR sensing domain and consequently, the specificity or affinity of sBAD for alternate molecules such as pABA. The second hypothesis is that the native bacterial HbaR transcription factor could be promiscuous. For example, ligand promiscuity has been observed with the transcription factor XylR of Pseudomonas putida which regulates genes involved in the metabolism of aromatic compounds and can be activated by toluene, m-xylene, or benzene (Galvão and de Lorenzo, 2006). Even more interesting, the transcription factor BenM from Acinetobacter baylyi bears not one but two binding pockets in its structure, one for cis,cis-muconic acid and the second for benzoic acid (Craven et al., 2009). HbaR from R. palustris belongs to the CRP-FNR transcription factor family that do not display ligand promiscuity compared with the XylR-NtrC or LysR families (Egland and Harwood, 2000). It should be noted, however, that CRP from P. putida can be activated by cyclic AMP as well as cyclic GMP, albeit with a lower affinity (Arce-Rodríguez et al., 2012). All this information suggests that pHBA may not be the only physiological binder of HbaR in R. palustris.

#### In vivo Application of sBAD in pHBA-Overproducing Yeast

We next addressed whether sBAD would support real-time monitoring of in vivo pHBA metabolite production. To do so, we engineered the ySCC185-F strain to overproduce pHBA as previously described (Averesch et al., 2017). We first deleted the TRP3 gene and then substituted the wild-type version of ARO7 with the ARO4K229L gene. This strain named ySCC185-F-A

Frontiers in Bioengineering and Biotechnology | www.frontiersin.org

biological replicates.

can produce higher yields of chorismate compared to the ySCC185-F strain (Averesch et al., 2017). ySCC185-F-A was then further modified to express the ubiC (chorismate lyase) gene from E. coli under the constitutive promoter TDH3, leading to the strain ySCC185-UbiC. We then assessed the number of cells, fluorescence of mCitrine and mKate2 (reporter of the cell number) for 65 h of cultivation with the newly constructed strains ySCC185-UbiC and ySCC185-30 (control strain). Both strains showed similar specific growth rates, indicating that the achieved level of pHBA production does not alter cell physiology (**Supplementary Table 5**). After 65 h of cultivation, the strain ySCC185-UbiC produced 1.32 mM of pHBA while the control strain ySCC185-30 did not produce any detectable amounts of it (**Figure 6A**). The production yield achieved in our strain is half of that reported by Averesch et al. (2017). This difference might be attributed to the modifications included in our parental strain (ySCC185-F) or differences in the media composition/culture conditions. Nevertheless, the concentrations reached during the cultures are in the range of detection of our biosensor, making it useful for the proof of concept of its potential use in metabolic engineering.

We then analyzed the response of sBAD to the internal production of pHBA in a medium devoid of pABA. As previously seen with ySCC185-F, a residual fluorescence signal is observed in the control strain ySCC185-30 and this signal increases with time (**Figure 6B**). However, mCitrine fluorescence is always higher with ySCC185-UbiC than with ySCC185-30. The accumulation of mCitrine after 60 h in ySCC185-30 and ySCC185-UbiC is probably due to a continuous production of the fluorescent protein, even after cell growth arrest. In fact, even the pHBA production increases after 60 h of culture while the cell count does not change anymore after 40 h (**Supplementary Figure 4**). Remarkably, there is a good linear correlation (R <sup>2</sup> = 0.83) between the mCitrine fluorescence signal and pHBA amounts produced by ySCC185-F UbiC cells (**Figure 6C**). More importantly, if we plot the ratio of the mCitrine fluorescence values (representative of sBAD activity) and the mKate2 fluorescence values (representative of the cell count) against pHBA concentrations obtained in the production experiments, the linearity of the response is even greater (R <sup>2</sup> = 0.95, **Figure 6D**, **Supplementary Figure 5**).

Therefore, our synthetic biosensor sBAD is functional and accurate enough to detect and measure the in vivo production of pHBA and, as such, can be used to screen for strains producing high titers of pHBA (up to 10 mM) in batch cultures. Furthermore, the capacity of sBAD may not be restricted to screening pHBA overproduction in yeast but can also be applied to pABA, benzoic acid and, to a lesser extent, other orthomonosubstituted benzoic acid derivatives.

#### CONCLUSIONS

This work has presented the successful design, construction and characterization of an orthogonal benzoic acid derivative biosensor (sBAD) in S. cerevisiae using the binding domain of the bacterial HbaR from R. palustris linked to the LexA protein as a DNA binding domain and B112 as a transactivation domain. This biosensor system allowed us to easily monitor, in real-time, a fluorescence signal linearly correlated to extracellular concentrations of pHBA. The promiscuity of the biosensor in yeast was also analyzed and, quite unexpectedly, pABA and benzoic acid also produce responses that are even more pronounced than the one measured with pHBA, notably in terms of sensitivity and dynamic range (for pABA). The relatively strong effect of the substituent position (para > ortho > meta), rather than the chemical properties of the substituent itself, indicate that the physicochemical properties that govern the promiscuous recognition of benzoic acid derivatives by sBAD are non-trivial. As our output signal is easily measurable, and considering the strong linearity obtained between our reporter signals and pHBA concentrations, our biosensor opens the way to a more thorough study of its binding properties, for instance with random mutagenesis approaches. This novel biosensor for detecting benzoic acid derivatives can be considered a useful tool to improve the production of production of such derivatives by screening large populations of yeast mutants. Moreover, the strong recognition of pABA or benzoic acid by sBAD could also be used for medical sensing purposes in bodily or chemical fluids, allowing an easily measurable output signal (fluorescence) to be monitored over a range of fluid concentrations covering three orders of magnitude.

## DATA AVAILABILITY STATEMENT

The datasets generated for this study are available on request to the corresponding author.

# AUTHOR CONTRIBUTIONS

J-LF and GT conceived this project. SC-C designed the experiments, carried out experimental work, analyzed, and interpreted the data. MF contributed to the analysis of the data. SC-C, GT, and PU wrote the paper. All authors assisted in this process.

# FUNDING

This work was supported by the Toulouse White Biotechnology consortium (pre-competitive financial support METASENS). J-LF also acknowledges funding provided by BBSRC Grant No. BB/M017702/1.

# ACKNOWLEDGMENTS

We wish to thank the team of Molecular and Metabolic Engineering of Toulouse Biotechnology Institute for their support and discussions, especially to Ruben Impellizzeri. Also, Delphine Lestrade is acknowledged for her help in the use of the flow cytometer.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00372/full#supplementary-material

# REFERENCES


in Saccharomyces cerevisiae. Biotechnol. J. 12:1600687. doi: 10.1002/biot.201 600687


**Conflict of Interest:** The authors declare that this study received funding from Toulouse White Biotechnology. The funder was not involved in the study design, collection, analysis, interpretation of data, the writing of this article or the decision to submit it for publication.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2020 Castaño-Cerezo, Fournié, Urban, Faulon and Truan. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Enzymatic Bioremediation of Organophosphate Compounds—Progress and Remaining Challenges

#### Meghna Thakur <sup>1</sup> , Igor L. Medintz <sup>2</sup> and Scott A. Walper <sup>2</sup> \*

<sup>1</sup> College of Science, George Mason University, Fairfax, VA, United States, <sup>2</sup> Center for Bio/Molecular Sciences, U.S. Naval Research Laboratory, Washington, DC, United States

Organophosphate compounds are ubiquitously employed as agricultural pesticides and maintained as chemical warfare agents by several nations. These compounds are highly toxic, show environmental persistence and accumulation, and contribute to numerous cases of poisoning and death each year. While their use as weapons of mass destruction is rare, these never fully disappear into obscurity as they continue to be tools of fear and control by governments and terrorist organizations. Beyond weaponization, their wide-scale dissemination as agricultural products has led to environmental accumulation and intoxication of soil and water across the globe. Therefore, there is a dire need for rapid and safe agents for environmental bioremediation, personal decontamination, and as therapeutic detoxicants. Organophosphate hydrolyzing enzymes are emerging as appealing targets to satisfy decontamination needs owing to their ability to hydrolyze both pesticides and nerve agents using biologically-derived materials safe for both the environment and the individual. As the release of genetically modified organisms is not widely accepted practice, researchers are exploring alternative strategies of organophosphate bioremediation that focus on cell-free enzyme systems. In this review, we first discuss several of the more prevalent organophosphorus hydrolyzing enzymes along with research and engineering efforts that have led to an enhancement in their activity, substrate tolerance, and stability. In the later half we focus on advances achieved through research focusing on enhancing the catalytic activity and stability of phosphotriesterase, a model organophosphate hydrolase, using various approaches such as nanoparticle display, DNA scaffolding, and outer membrane vesicle encapsulation.

Keywords: organophosphate, enzyme, chemical warfare agent, bioremediation, catalysis, outer membrane vesicle, decontamination, phosphotriesterase

# INTRODUCTION

Biological diversity spread across the ecological niches of the world has allowed for the evolution of cellular systems that enable the survival of microbes, plants, and animal in these highly variable environments. This has been recognized at the phenotypic level since the time of Charles Darwin's voyage on the Beagle. Today we characterize Darwinian evolutions at the genetics level recognizing

#### Edited by:

Jean Marie François, UMS3582 Toulouse White Biotechnology (TWB), France

#### Reviewed by:

Georgios Skretas, National Hellenic Research Foundation, Greece Pasquale Stano, University of Salento, Italy

\*Correspondence: Scott A. Walper scott.walper@nrl.navy.mil

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 29 July 2019 Accepted: 09 October 2019 Published: 08 November 2019

#### Citation:

Thakur M, Medintz IL and Walper SA (2019) Enzymatic Bioremediation of Organophosphate Compounds—Progress and Remaining Challenges. Front. Bioeng. Biotechnol. 7:289. doi: 10.3389/fbioe.2019.00289

**110**

that such adaptability occurs through alterations in cellular physiology or mechanisms, many of which are the product of adapted or novel enzymatic pathways and the products they produce. Examples such as the bacteria found in arsenic contaminated aquifers or the fungi capable of growing within nuclear reactors highlight the ability of Nature to adapt to harsh environments and exploit available resources to survive (Dadachova and Casadevall, 2008; Gnanaprakasam et al., 2017; Gu et al., 2017). In addition to basic survival mechanisms, the endless quest for the raw materials necessary to sustain life has also allowed for organisms to develop metabolic pathways allowing them to scavenge, convert, and utilize an incredible range of both natural and man-made chemicals to satisfy these needs. The cellular tools of these conversions, enzymes, are capable of chemical conversions and transformations that occur incredibly fast, often at the rate of diffusion, and with a specificity and selectivity that cannot easily be rivaled using chemical catalysis. For this reason, enzymes are rapidly developing as invaluable tools in medicine as therapeutics, as agents of environmental decontamination, and as reagents that enable a diverse spectrum of commercial applications.

Historically, mankind exploited cellular processes such as fermentation for the preparation of food and drink. Today, advances and understanding in cellular and metabolic processes has allowed for an explosion in the use of purified enzymes within the food industry (Raveendran et al., 2018). Similarly, the use of enzyme-based therapeutics continues to rise as these biomolecules typically show reduced toxicity and immunogenicity compared to other chemically synthesized alternatives (Dean et al., 2017; Kumar and Abdulhameed, 2017; Yari et al., 2017). Enzymes for commercial and industrial applications is also on the rise as these biomolecules offer a sustainable path with reduced toxic waste products (Chapman et al., 2018). One area of particular interest in research and development is in the area of bioremediation and detoxification of chemicals.

Bioremediation at its most basic definition is the use of natural biological systems such as plants and microbes to degrade or consume environmental pollutants. The potential of bioremediation was recognized decades ago starting with efforts to treat wastewater using environmental isolates. The first documented success with environmental microbes was detailed in a 1975 report by Raymond et al. that describes the degradation of petroleum-derived hydrocarbons by microbial populations (Raymond et al., 1975; Dvorak et al., 2017). Expanding on these studies, researchers have sought to harness this phenomenon for other purposes, particularly environmental decontamination. Contaminants such as industrial waste products including polycyclic aromatic compounds, organic dyes, heavy metals, and polyhalogenated compounds are common contaminants in aquifers and soil as are petroleum-based hydrocarbons and pharmaceutical compounds. Fortunately, each of these have been targets of various bioremediation strategies (Singh and Walker, 2006; Yi and Crowley, 2007; Peng et al., 2008; Ben Mansour et al., 2012; Montgomery et al., 2013; Rodgers-Vieira et al., 2015; Dzionek et al., 2016; Brar et al., 2017; Dvorak et al., 2017; Ojuederie and Babalola, 2017; Wu et al., 2017; Liu et al., 2018). The adaptability and versatility of plants and microbes has enabled bioremediation as a "biologicallyfriendly" strategy to eliminate many of the harmful contaminants released by the everyday machinations of human society, see schematic overview in **Figure 1**. At present, bioremediation of environmental reservoirs is accomplished using primarily natural, non-engineered bacteria that have been isolated from contaminated sources and exhibit an ability to consume or convert the target contaminant as highlighted in several examples (Cologgi et al., 2011; Silar et al., 2011; Dubinsky et al., 2013; Prakash et al., 2013; Gustavsson et al., 2016). With the rise of synthetic biology and microbial engineering, future strategies may likely include engineered microbes containing novel metabolic pathways or tailored for enhanced fitness in austere environments. Despite the advantages these strategies may afford, regulations in the U.S. and across Europe limit or prohibit the use and release of genetically modified organisms into the environment. For this reason, many researchers are investigating the potential direct use of recombinant enzymes for the purpose of environmental decontamination. Often referred to as cell-free synthetic biology, these strategies offer a fine level of control where researchers can tailor reaction conditions such as enzyme concentrations, the availability of co-factors, ionic strength, and others. Of significant importance, unlike the engineered microbes from which they are derived, cell-free catalytic systems are non-replicative and are therefore less likely to be impacted by regulatory policies.

This review will focus on enzymatic systems for the bioremediation of a ubiquitous but highly dangerous group of man-made compounds collectively referred to as organophosphates (OPs). First described in the late nineteenth century, these compounds existed in relative obscurity until their potential as anti-biological agents were recognized during World War II. Since then they have rapidly evolved from chemical warfare agents (CWAs) to commercial pesticides used all over the world. As both agents of war and agricultural tools, these chemical compounds have left a lasting effect on the environment as they are contaminants of both soil and water capable of inducing illness and death to those that inadvertently encounter them. Fortunately, microbial species have rapidly evolved enzymatic processes to degrade these contaminants and researchers have developed strategies to utilize these biological catalysts for bioremediation of OP compounds. In this review we will detail many of the enzymes currently employed and studied for OP remediation describing the properties of each and efforts to develop them into products useful for environmental cleanup. Unfortunately, recombinant enzymes often suffer from issues that limit their wide-spread use as a viable material. However, as the tools and techniques of biological engineering continue to expand, the utility of enzymes as therapeutics and implements of decontamination of organophosphate compounds has been reinvigorated. This review will describe how new methods of biological engineering such as synthetic biology and cell-free synthetic biology are breathing new life into enzyme-based catalytic systems for organophosphate decontamination and neutralization. Here we pay particular attention to efforts to enhance the catalytic activity, substrate promiscuity, and

stability of the more commonly used OP hydrolyzing enzymes through enzyme engineering, assembly of scaffolded cell-free systems, and through the development of microbe-based systems of bioremediation that revolve around the re-engineering of bacterial genetic circuits and cellular systems. Each of these areas highlights how the engineering of biology at the protein, system, and organism level can be utilized to address issues such as environmental contamination that affect populations around the globe.

# ORGANOPHOSPHATES

Organophosphate compounds are typically phosphate esters formed from a reaction between alcohols and phosphoric acid resulting in a compound with the general structure of O=P(OR)3. OP compounds are found in both naturally occurring biomolecules and industrial products such as pesticides including insecticides and herbicides. The first of these compounds was chemically synthesized in the late nineteenth century by Phillipe de Clemont and Wladimir Moschnin, but it was not until much later, in 1932, that Willy Lange and his student Gerde von Krueger described the cholinergic effects of these compounds (Costa, 2018; Franjesevic et al., 2019). Over the next 20 years, British and German scientists produced a suite of OP compounds that could be used as highly potent nerve agents. Later, American scientists and others adapted these chemical syntheses to manufacture OP insecticides. Chemical structures for some of the more prevalent and toxic OP compounds are shown in **Figure 2.** Toxicity of these compounds is highly variable and dependent on the chemical composition, the route of entry or exposure, and whether exposure is acute or chronic (Munro et al., 1999; Costa, 2018). While not as lethal as their CWA counterparts, OP pesticides can also be quite dangerous capable of causing illness or death. For both pesticides and CWAs, the ease and low cost of production paired with the high degree of efficacy has ensured that OP compounds continue to find use in both beneficial and nefarious applications.

OP compounds were rapidly adopted as CWAs first by the Nazis and subsequently by the British during World War II to replace the chlorine-based weapons of the previous World War. The German efforts during this time period produced the G-series of agents including sarin, cyclosarin, and soman. In parallel, the British developed a series of reagents including diisopropylfluorophosphate (DFP), VX, and VR. While diverse in chemical composition and structure, CWAs and OP pesticides function through the disruption of the enzyme

FIGURE 2 | Structure of OP compounds. The diversity of OP compounds is highlighted here in the chemical structures of some of the better known CWA nerve agents (sarin, tabun, soman, cyclosarin, VX, and DFP) and pesticides (paraoxon, parathion, chlorpyrifos, trichlorfon, dichlorvos, malathion, TEPP).

acetylcholinesterase, a critical component of signal transduction in the neuromuscular junctions of both insects and animals, see schematic in **Figure 3**. Accumulation of acetycholine, a neurotransmitter, in the synaptic cleft leads to overstimulation of acetylcholine receptors thereby causing a cholinergic crisis characterized by anxiety, headache, convulsion, tremor and death (Jokanovic, 2018; Naughton and Terry, 2018; Sikary, 2019). Though developed during the Second World War these compounds did not see use until many years later in other conflict areas. Sarin saw widespread use by the Iraqi army during the Iran-Iraq conflicts of the 1980s resulting in the deaths of thousands despite international treaties condemning the use of CWAs. Later, in 1995, the Japanese doomsday cult Aum Shinrikyo deployed sarin gas on a Tokyo subway train killing 13 and injuring or affecting nearly 1,000 others. The same group also experimented with the production and use of the even more toxic organophosphate CWA, VX (Zurer, 1998). These compounds continue to find nefarious use today. In 2013 and again in 2017, the Syrian government bombarded cities within their own borders with sarin gas to quell a growing rebellion. OP nerve agents have also been used for assassinations as shown in the 2017 slaying of Kim Jong-nam a half-brother of the current North Korean leader and the attempt on the life of a former Russian spy, Sergei Skripal in 2018.

While governments and terrorist organizations have employed OPs as weapons of mass destruction and terror, these compounds have also seen widespread use in agriculture in both developed and developing countries. During the twentyfirst century, OPs were the most widely used pesticides in the United States with as many as 36 compounds approved for domestic or agricultural use as summarized in **Supplemental Table 1** (Roberts and Reigart, 2013). It is estimated that millions of tons of these compounds are produced and dispersed each year in the U.S. alone (Atwood and Paisley-Jones, 2017). The ubiquitous utilization of these compounds leads to a high rate of human exposure and OP poisoning. It is estimated that around 3,000,000 people are exposed to OP compounds per year accounting for more than 300,000 fatalities (Bird et al., 2019). Despite the well-documented acute and chronic toxic effects, OPs are seen as vital components of commercial agriculture and pest control (Knutson and Smith, 1999; Jokanovic, 2018; Lushchak et al., 2018; Mostafalou and Abdollahi, 2018). However, in developed countries there is growing concern for the long-term use of these compounds and their environmental accumulation and toxicity. As an example, chlorpyrifos, which was first registered with the U.S. Environmental Protection Agency (EPA) in 1965, has seen a gradual shift in policy regarding its use beginning in 1996 with phase out of home use and in some agricultural products such as tomatoes. Since then the EPA has continued to examine this OP and its environmental and health effects modifying policy for its use. Mostly recently these studies and others have resulted a 2018 decision by the U.S. Ninth Circuit Court of Appeals ordering an EPA-approved ban of the compound (https://www.epa.gov/ ingredients-used-pesticide-products/chlorpyrifos). Continuous studies of OP compounds and events such as the death of 25 school children in India from meals contaminated with monocrotophos, an agricultural pesticide, has led to increased public attention which in turn has fueled scientific research in methods of ensuring public safety through the development of new safer pesticides and methods of remediating OPs from the environment (Banerji et al., 2013; Than, 2013).

# DECONTAMINATION OF ORGANOPHOSPHATES

The reactive nature of organophosphate compounds has allowed for the development of numerous strategies to remove and/or neutralize these materials from the environment. Typically, decontamination of organophosphates, both pesticides and CWAs, is accomplished through (1) physical removal or dilution of the material through the use of absorbents or flushing contaminated areas, (2) chemical decontamination, or (3) bioremediation using microbial species or purified enzymes (Jacquet et al., 2016). Ideally, several of these methods would be used in conjunction to detoxify target environments, equipment, and infrastructure (see **Table 1**).

# Physical Decontamination

Physical decontamination of contaminated areas involves the use of incineration, sorbent materials, or simply burying or washing contaminated surfaces to minimize the risk of exposure. Unfortunately, this does not typically neutralize the material which can lead to secondary contamination particularly of soils and aquifers. Materials for this type of decontamination can span the gamut from nanoparticles to bulk use of minerals in material platforms. For example, to address aquifer contamination with OP pesticides Liu et al. developed graphenecoated silica nanoparticles capable of rapidly capturing nine different pesticides from spiked water samples (Liu et al., 2013). Other reactive materials such as cerium dioxide have shown success against both pesticides and CWAs. In laboratory trials, the research team of Lubos Vrtoch showed that cerium dioxide sorbent materials could be used to rapidly degrade both pesticides and CWAs such as VX (Janos et al., 2014). In these studies the group showed that methyl parathion, soman, and VX could be degraded in as little as 30 min after exposure using their cerium dioxide sorbent materials. Unfortunately, while cerium dioxide does have potential as a decontamination strategy studies have shown that exposure to these materials can cause immunostimulation in laboratory models and can lead to toxicity in some human cell lines (Gehlhaus et al., 2009; Mittal and Pandey, 2014).

# Chemical Decontamination

Chemical reagents facilitating the hydrolysis of pesticides and CWAs as a route of decontamination are by far the most prevalent and include hydrolysis, oxidation, and reduction mechanisms. At present, materials relying on chemical hydrolysis are most commonly used for large-scale remediation of contaminated areas (Jacquet et al., 2016). Common reagents such as sodium hydroxide and hypochlorite will react with many OP compounds though both of these compounds are corrosive and not suitable for personal use or with sensitive materials



(Dowling and Lemley, 1992; Tuorinsky et al., 2009; Kitamura et al., 2014). Commercial products such as Decontamination Solution 2 (DS2), BX-24, and Decontamination Formulation DF-200 are employed by U.S. and NATO forces and are highly effective against most organophosphate compounds. As with sodium hydroxide and hypochlorite both DS2 and DF-200 are composed of chemicals that limit their use for personnel or sensitive equipment due to their corrosive nature. Given warfighter and support personnel needs, personnel decontamination solutions have also been developed including the M291 Skin Decontamination Kit (SDK) and the Reactive Skin Decontamination Lotion (RSDL). These products have variable effectiveness toward organophosphate agents and neither are suitable for wound or eye decontamination. These materials and other emerging chemical decontamination strategies are thoroughly reviewed in Jacquet et al. (2016).

Decontamination agents such as DS2 and bleach are corrosive in nature and generate hazardous waste and are therefore not considered safe. This leads to a dire need for non-toxic, noncorrosive, and environmentally compatible decontamination strategies. The use of microbes and purified enzymes for decontamination has been an area that is being extensively researched and has met with success in terms of degradation of OP compounds in commercial applications.

#### Bioremediation

As will be discussed in subsequent sections, several enzymes have been identified in nature capable of degrading OP pesticides and CWAs (Dawson et al., 2008; Yair et al., 2008; Pizzul et al., 2009; Diao et al., 2013; Iyer et al., 2013; Geed et al., 2016; Brar et al., 2017). Several of these enzymes have been localized to the surface of microbes for evolution of substrate promiscuity and direct bioremediation of OP compounds (Richins et al., 1997; Cho et al., 2002; Zhang et al., 2004; Makkar, 2013; Alves et al., 2015, 2016, 2018; Bigley et al., 2015; Bigely et al., 2019). Cell-mediated bioremediation and biocatalysis is well-established and documented within the literature, however, application can be limited as the release of genetically-modified organisms is not currently considered a viable approach to bioremediation. Additionally, the unique role that OP compounds play as both commercial products and weapons of mass destruction necessitates remediation strategies that can also transition to therapeutic agents and personnel decontamination products.

Over the past several years, numerous successes have been realized with both the isolation of new enzymes capable of degrading OP compounds and the evolution of older enzymes to improve their utility in point-of-concern decontamination. In the subsequent sections we describe those enzymes most commonly referenced in the literature and highlight some of the new players in the field. For each of these enzymes, we also briefly discuss how scientists are attempting to further enable the use of these biological catalyst through protein engineering to increase substrate promiscuity and activity as well as other strategies to improve catalytic activity and stability.

# ENZYMES FOR DECONTAMINATION

As previously mentioned, OP-intoxication occurs through the irreversible binding of OP compounds to acetylcholinesterase, an enzyme found within the neuromuscular junction, inactivating it. While acetylcholinesterase is incapable of degrading these compounds, enzyme catalysts capable of degrading OP compounds have been identified in microbial species, in squid, and even in mammals. As will be detailed below, enzymatic degradation of OP compounds occurs through nucleophilic attack of the phosphorus center of the compound mediated by a pair of divalent metal ions, a water molecule, and reactive amino acids contained within the active site of the enzyme (Aubert et al., 2004; Wymore et al., 2014). These enzymes have varying specificities for OP compounds but all follow similar mechanisms of action.

An ideal decontaminant candidate must be able to rapidly detoxify the warfare agents at a molecular level and be able to decontaminate any surface such as paint, concrete, rubber seals, asphalt, metal, plastics, clothing, and skin. Additionally, these reagents should be environmentally friendly without lasting impact on soil, vegetation, and animal life or underground water sources. Use of OP degrading enzymes for environmental decontamination present significant advantages as they can rapidly hydrolyze the nerve agents, and are environmental friendly, non-flammable, non-corrosive and can be disposed

FIGURE 4 | Organophosphate hydrolyases. Representative protein structures from the Protein Data Bank (PDB) database. From left to right: Crystal structure phosphotriesterase OpdA from Agrobacterium radiobacter (PDB: 2D2J); crystal structure of zinc-containing phosphotriesterase (PTE) from Pseudomonas diminuta (PDB: 1HZY, Chain A); crystal structure of diisopropylfluorophosphatase (DFPase) from Loligo vulgaris (PDB: 1E1A); crystal structure of serum paraoxonase PON1 from Oryctolagus cuniculus (PDB: 1V04); crystal structure of phosphotriesterase-like lactonase SsoPox from Sulfolobus solfataricus (PDB: 2VC5, Chain A); crystal structure of organophosphate acid anhydrolase (OPAA) from Altermonas sp. (PDB: 3L7G, Chain A).

TABLE 2 | Summary of enzymes discussed.


OPs, Organophosphates; DEVX, Diethyl-VX; DMVX, Dimethyl-VX; DFP, Diisopropyl fluorophosphate; GB, sarin; GF, cyclosarin.

of safely and easily. Several enzymes have evolved in nature with the capability of degrading the OP compounds, see **Figure 4** for names and representative structure and **Table 2** for a summary details on each. Extensive research is being carried out to improve their efficacy and stability, thus enhancing their appeal as decontamination agents. Some of these advanced enzymes are being investigated for in vivo efficacy, and one of the enzyme based decontamination solutions is already in commercial use in Australia. We now briefly summarize some of the more prominent enzymes studied for OP remediation.

#### Diisopropylfluorophosphatase (DFPase)

Diisopropylfluorophosphatase (DFPase), isolated from the European squid Loligo vulgaris, catalyzes the hydrolysis of OP compounds such as diisopropylfluorophosphate (DFP) and a number of G-type nerve agents, including sarin (GB), soman (GD), cyclosarin (GF), and tabun (GA) (Hartleib and Ruterjans, 2001; Soares et al., 2018). Several structural studies on DFPase have revealed the presence of a six-bladed ß-propeller structure with two calcium ions, one required for catalysis and the other for providing structural integrity (Scharff et al., 2001; Blum and Chen, 2010; Blum et al., 2010). Several theoretical and computational studies in agreement with the experimental results have highlighted the existence of at least two different pathways for the degradation of different OP compounds which could eventually lead to improved strategies to engineer DFPase for more efficient degradation of OP compounds (Soares et al., 2018; Zhang et al., 2018). In addition, mutational studies have also been carried out and have led to an understanding of the mechanism of action of the enzyme on substrates (Blum et al., 2006). Remarkably, in a preliminary in vivo study, Melzer et al. found that addition of pegylated DFPase minimizes lethality in rats challenged with a subcutaneous 3xLD<sup>50</sup> dose of soman, thereby highlighting its potential for in vivo use (Melzer et al., 2012).

# Paraoxonase (PON1)

The paraoxonase enzyme produced in the human liver is a calcium dependent enzyme able to hydrolyze aryl esters, lactones, and OPs (Rajkovic et al., 2011). Structurally, it is also described as a six-bladed propeller structure that utilize a calcium ion within its active site similar to DFPase (Gold et al., 2000; Rajkovic et al., 2011; Mackness and Mackness, 2015). Although PON1 shares identity with DFPase in the active site and tertiary structure, it shows a different substrate preference and is unable to hydrolyze bulky OPs such as DFP but does show some activity for VX (Bajaj et al., 2013; Purg et al., 2017). In addition to its activity against organophosphates, paraoxonases have been extensively studied for their anti-inflammatory, anti-oxidative, anti-atherogenic, antimicrobial, anti-diabetic, and OP-detoxifying properties. Human paraoxonase-1 (PON1) can readily hydrolyze paraoxon, a commercial OP pesticide; however, its activity toward G- and V- type nerve agents is limited to the less toxic enantiomers of these agents. As a result, the enzyme has been subjected to substantial experimental and computational characterization to enhance its catalytic efficiency and enantioselectivity (Gupta et al., 2011; Goldsmith et al., 2012; Le et al., 2015).

As the only human-derived enzyme capable of degrading OP compounds, PON1 has the greatest potential as a therapeutic to counter OP poisoning. Unfortunately, PON1 suffers from a lack of stability unless bound to a high density lipoprotein (HDL) and is further limited (as this complex) by environmental conditions and by several common biomolecules that have been shown to interact with and limit the activity of the HDL-PON1 complex (Ferretti et al., 2001; Jaouad et al., 2003; Rozenberg and Aviram, 2006; Gaidukov et al., 2009). To improve upon the potential of PON1 as therapeutic, Aharoni et al. generated a mutant version of the PON1 that exhibited a 40-fold increase in OPdegrading activity compared to the parental enzyme (Aharoni et al., 2004). This recombinant PON1 was later combined with a HDL particle (referred to as BL-3050) and tested in vivo in mice by the same research team (Gaidukov et al., 2009). In these studies the recombinant PON1 showed a longer halflife in mouse models and provided a significant improvement in survival rates for treated mice (87.5%) vs. non-treated controls (37.5%).

#### Organophosphate Hydrolase (OpdA)

One of the most efficient OP-degrading enzymes, organophosphate hydrolase (OpdA), was isolated form Agrobacterium radiobacter by Horne et al. in 2002 (Horne et al., 2002). The OpdA enzyme adopts an (α/β)<sup>8</sup> barrel structure with a heterobinuclear Fe–Zn metal center and shows an enhancement in specific activity when supplemented with cobalt (Jackson et al., 2008). The enzyme can hydrolyze a wide variety of OP pesticides and has been shown to inactivate G-type nerve agents such as tabun (GA), sarin (GB), soman (GD), and ethylsarine (GE) with varying efficiencies (Dawson et al., 2008). Several mutants of OpdA with improved activity against these nerve agents have also been produced as described by Yang et al. (2003). In addition to in vitro studies, the in vivo therapeutic potential of OpdA has also been demonstrated. For instance, separate research groups showed that administration of OpdA in rats and monkeys improved survival after poisoning with highly toxic OP pesticides (Bird et al., 2008; Jackson et al., 2014). In addition to its therapeutic applications, OpdA is the only enzyme that is currently commercially employed as a bioremedier. The Australian company Orica Ltd, has marketed the OpdA-containing product LandguardTM OP-A for use in pesticide decontamination of water sources (Anderson et al., 2011; Scott et al., 2011). Studies starting in 2004 have shown that LandguardTM is able to significantly reduce OP levels in agricultural wastewater validating recombinant enzymes as a viable method of decontaminating contaminated sources at a significant scale.

# Phosphotriesterase (PTE)

Phosphotriesterase (PTE), also termed organophosphate hydrolase (OPH) is a zinc-dependent bacterial enzyme that belongs to the amidohydrolase superfamily and was first identified in soil bacteria that hydrolyzed the pesticide parathion (Dumas et al., 1989a,b). The tertiary structure of the protein is an (α/β)8-barrel or TIM-barrel, in which the binuclear active site is located at the C-terminus of the protein (Vanhooke et al., 1996; Benning et al., 2001). The most commonly utilized PTE enzyme was derived from Brevundimonas diminuta (previously Pseudomonas diminuta) although homologs have been identified in other species including Sulfolobus solfataricus (discussed below) and Deinococcus radiodurans which are typically referred to as phosphotriesterase-like lactonases (PLLs) (Dumas et al., 1989a; Merone et al., 2005; Hawwa et al., 2009). The microbial substrate of PTE and PLLs has not been identified, though it likely varies between species. It is believed and has been shown in some experimental studies that these enzymes likely arose from bacterial lactonases (Chow et al., 2009; Afriat-Jurnou et al., 2012; Elias and Tawfik, 2012). PTE shows high catalytic activity toward a number of pesticides but only moderate activity toward nerve agents such as sarin, cyclosarin, and VX. The environmental persistence of V-agents such as VX has led several research teams such as those of Frank Raushel of Texas A&M University and others to invest heavily in mutational studies of this enzyme to improve activity toward VX and other substrates (Briseno-Roa et al., 2011; Tsai et al., 2012; Bigley et al., 2013, 2015; Naqvi et al., 2014; Bigely et al., 2019). Recently, Khersonsky et al. developed a robust automated method which led to generation of PTE with a remarkable >1,000-fold improvement in nerve-agent hydrolysis (Khersonsky et al., 2018). Moving forward in that direction, Bigely et al. (2019), screened a 28,800 member six-site PTE mutant library against multiple V-agent analogs, and identified multiple variants with >1500-fold increase in kcat/K<sup>M</sup> for the hydrolysis of VX (Bigely et al., 2019).

Of the OP-degrading enzymes currently studied, PTE has some of the fastest catalytic rates and has shown the greatest promise for engineering substrate activity as described above. Early studies with direct injection of PTE as a countermeasure for OP poisoning showed some success though the half-life of enzyme was relatively low (100 min) necessitating the development of an extracorporeal reactor system to purify the blood (Masson et al., 1998). A later review by Raushel summarized some of the progress with PTE as a therapeutic in mouse models including efforts to use PTE as a pretreatment to afford some level of protection (Ghanem and Raushel, 2005). With the mutational improvement of the enzyme's stability and substrate range it is not difficult to envision PTE as a viable medical countermeasure for OP poisoning.

#### SsoPox

A hyperthermophilic PLL, SsoPox, was isolated from the Archaeon Sulfolobus solfataricus (Merone et al., 2005; Hiblot et al., 2012). SsoPox exhibits high activity toward acyl-homoserine lactones and oxo-lactones but a low phosphotriesterase activity. Similar to OP hydrolases of the amidohydrolase superfamily, the folded structure of SsoPox is an (α/β)<sup>8</sup> barrel in which the active site resides at the C-terminal section of the structure (Hiblot et al., 2013). While SsoPox show reduced activity toward many pesticides compared to PTE and others, it is an extremely rugged enzyme exhibiting activity at temperatures up to 100◦C (melting temperature is 104◦C) and in the presence of several detergents and other denaturing agents (Hiblot et al., 2012). These properties make SsoPox a promising candidate for the development of field-deployable reagents for bioremediation. To further improve upon the enzyme, several groups have employed a structure-based design approach to increase the phosphotriesterase activity and improve the active site recognition for an increased range of OP substrates (Merone et al., 2010; Jacquet et al., 2017). Recently, Vitola et al. developed a biocatalytic membrane reactor (BMR) by covalently immobilizing a triple mutant of the SsoPox on polymeric membranes, leading not only to high paraoxon degradation but also long-term stability of the free enzyme (Vitola et al., 2019).

# Organophosphate Acid Anhydrolase (OPAA)

The organophosphate acid anhydrolase enzymes are bacterial prolidases isolated from several species of Alteromonas bacteria (deFrank and Cheng, 1991). The Alteromonas OPAAs are dipeptidases that typically cleave dipeptide bonds in which the C-terminal residue is a proline (Cheng et al., 1999). Although not involved in the metabolism of OPs, OPAAs also show activity against a wide range of these compounds cleaving P-F, P-O, P-CN, and P-S bonds (Cheng et al., 1999). Prolidases are structurally distinct from other bacterial hydrolases such as PTE and OpdA and therefore show different OP substrate specificities and activities (Xiao et al., 2017). As with many of the other enzymes discussed here, OPAA has undergone mutagenic studies to enhance its catalytic activity and substrate specificity. Bae et al. showed that a series of mutations in OPAA led to significant increase in activity toward sarin and soman (Bae et al., 2018). Additionally, in these studies, the authors were able to isolate a double mutant version of OPAA that was able to show enantioselectivity for one enantiomer of sarin. The Alteromonas OPAA, similar to other OP enzymes, have greatest activity at biologically-relevant temperature (25– 37◦C) which can limit their utility in field-based applications. As has been observed with other proteins, prolidases isolated from extremophiles such as members of the Pyrococcus genus often show improved thermostability as shown by Theriot et al. (2011). Studying both the wild-type and mutant versions of the Pyrococcus horikoshii enzyme, the authors showed that these new OPAA enzymes had improved shelf-life and thermal stability compared to Alteromonas OPAAs making them ideal candidates for field distribution.

## Other Enzymes

In addition to the above described enzymes, several proteins have been shown to possess a moonlighting OP hydrolyzing activity. For instance, senescence marker protein (SMP30) isolated from mouse liver cytosol, is capable of hydrolyzing DFP and other OPs such as sarin, soman, and tabun (Scott and Bahnson, 2011; Dutta et al., 2019). Although SMP30 can hydrolyze similar substrates like DFPase, it is not a calcium binding protein, and shows activity in the presence of Mg2<sup>+</sup> and Mn2<sup>+</sup> (Dutta et al., 2019). Bacterial lactonases have also shown potential as OP-degrading enzymes due to their high degree of substrate promiscuity (Draganov, 2010). Like SsoPox, other PLLs have been described in the literature as showing some potential as tools for OP bioremediation. In separate publications, Zhang et al. describe a PLL from Geobacillus kaustophilus referred to as GkaP (Zhang et al., 2012, 2015). This enzyme typically cleaves the 6 membered ring structures of lactones but does show some activity toward ethyl-paraoxon. Other enzymes capable of degrading the substrate methyl-parathion have also been described in the literature (Rani and Lalithakumari, 1994; Zhongli et al., 2001; Yang et al., 2008). These enzymes typically utilize a binuclear metal active site but show little structural similarity to phosphotriesterases (Dong et al., 2005).

Many of the enzymes presented here have a long history in the development of therapeutics, sensors, and agents for environmental and personal decontamination. As the tools of genomics and systems, synthetic, and molecular biology continue to evolve it is anticipated that more enzymes capable of degrading OP compounds will be identified in nature and evolved in the laboratory to improve their applicability. In the subsequent sections we discuss the efforts of researchers to both improve and implement these enzymes for field use. As a discussion encompassing all of the aforementioned enzymes would prove prohibitively long, we focus on efforts targeting the PTE or OPH enzymes only.

### RELEVANCE AND POTENTIAL FOR DEPLOYMENT

OP compounds fill two distinct and highly divergent applications in human society. At the most benign level, the compounds are used to benefit agricultural production and to control the spread of arthropod disease. At the opposing end of this spectrum, OP compounds are some of the most dangerous weapons produced by man. Though the potential for human harm is dramatically different, even widely used pesticides can be incredibly dangerous following prolonged or high concentrations of exposure. Therefore, the need to develop reliable methods to both detect and eliminate these compounds is important and relevant to both civilian and military populations. While successes have been achieved with engineering bacterial systems for decontamination and detection through mechanisms such as cell-surface localization of enzyme, the release of engineered organisms is not currently accepted by nations around the world. Cell-free synthetic biology, which can rely on cell extracts or purified recombinant enzymes, offers an avenue for utilizing the elegance of biological systems for detection and bioremediation in a non-replicating system. As will be described in the subsequent sections, enzyme-based cell-free systems have seen success as sensors, tools of bioremediation, and as potential therapeutics.

#### Sensors

While the use of pesticides is vital to commercial agriculture, the long half-lives and soil and water retention time of many of these organophosphate compounds necessitates environmental monitoring (Uchimiya et al., 2012; Hossain et al., 2014; Fosu-Mensah et al., 2016; Pan et al., 2018). Contamination of water sources by agricultural run-off leads to elevated concentrations of these compounds in large-scale waterways such as rivers and deltas as well as soil (Pedersen et al., 2006; Babu et al., 2011). Often, OP compounds are photodegradable or easily hydrolyzed in aerobic soils, however, as shown in a recent publication by Pan et al. and the map of the Yangtze River Delta region shown in **Figure 5,** elevated concentrations of OP compounds can easily be measured in environmental samples (Pan et al., 2018). While these effects are largely felt at the local level, the globalized agricultural economy enables contaminated food products to be distributed world-wide.

The need for OP sensors touches both civilian and military walks of life. PTE-based sensors have not transitioned beyond the laboratory, however, several research groups have shown that surface-immobilized PTE is viable in sensor development (Schoning et al., 2003; Simonian et al., 2004; Andrianova et al., 2016; Hondred et al., 2017, 2018). Typically, these sensors are electrochemical in nature monitoring the formation of hydrolysis products. Schöning et al. explored a number of parameters such as immobilization strategies and reaction conditions while developing a sensor which measures small changes in pH resulting from the hydrolysis of OP compounds by PTE (Schoning et al., 2003). As another example, Andrianova et al. describe an ion-selective fieldeffect transistor (ISFET) that utilized immobilized PTE to detect paraoxon and methyl-paraoxon (Andrianova et al., 2016). In these studies the researchers highlighted the utility of their ISFET system in prolonged and continuous monitoring of water samples without loss of function which is relevant to the development of distributable water system sensors.

Enzyme concentration and activity in sensors is vitally important to achieving the high levels of sensitivity desired in OP detection. Often this can be accomplished through the integration of nanoparticles into the sensor platform as a scaffold for enzyme assembly. In two separate studies Hondred et al. developed a printed, graphene-based sensor to detect paraoxon and other pesticides highlighting the use of both enzymes and electrochemical detection (Hondred et al., 2017, 2018). The researchers used a variety of manufacturing techniques including the laser annealing of platinum nanoparticles (NPs) to increase enzyme loading and the overall sensitivity of the sensor as shown in **Figure 6**. In these studies they were able to show a detection limit of 3 nM for the paraoxon substrate. The benefits of combining nanostructures and enzymes in OP sensors has been realized and exploited by several other research groups whose efforts are summarized in a recent review by Sheng et al. (Xiong et al., 2018).

# Improving the Utility of PTE

As discussed above for each of the OP hydrolases, protein engineering methods have led to significant improvement in the activity of the enzyme and substrate/enantio- specificity. Beyond direct modification of the amino acid sequence, immobilization of enzymes to solid supports further improves their applicability as there are often inherent benefits of improved stability and enhanced catalytic activity associated with immobilization (Mateo et al., 2007; Mohamad et al., 2015; Proschel et al., 2015; Hoarau et al., 2017). As an enzyme with significant potential as both a bioremedier and therapeutic, PTE has been attached to a range of different materials to improve its activity, longterm stability, and; therefore, viability as a deployable reagent. To illustrate the potential for increased functionality of PTE when immobilized to a solid surface, Raynes et al. cross-linked the enzyme to amyloid fibrils generated from insulin and crystallin and observed a marked improvement in thermal stability (Raynes et al., 2011). Similarly, covalently conjugating PTE to negatively charged gold nanoparticles not only decreased Km value and increased Vmax and Kcat values, but also achieved stability within a wider range of pH (2–12) and temperature (25–90◦C) (Karami et al., 2016). This phenomenon of improved enzymatic and biophysical properties is not unique to PTE as highlighted in numerous review articles and recent publications (Chen M. et al., 2017; Kreuzer et al., 2017; Arsalan and Younus, 2018; Wang et al., 2018).

The enhancement of PTE activity and other enzymes at nanoparticle interfaces has been a focus of research efforts in the Medintz laboratories for several years. Focusing on semiconductor quantum dots (QDs) as the model scaffold, the team developed strategies for QD capping and PTE immobilization that led to the formation of highly active enzyme structures (Susumu et al., 2014). A subsequent study described in Breger et al., showed that the rate and efficiency of enzyme catalysis was directly affected by the size of the QD NP as shown in **Figure 7** (Breger et al., 2015a). While PTE is a highly active enzyme functioning at nearly the rate of diffusion, the team

two QD NPs. Significant changes in the rate of initial velocity for PTE were observed when the enzyme was immobilized to 525 nm (B) and 625 nm (C) QDs. The rate of catalysis for free enzyme (A) was compared to each PTE-QD structure at varying ratios of nanoparticle to enzyme. In each instance the rate of reaction was markedly improved compared to the free enzyme control. Reproduced with permission (Breger et al., 2015a), Copyright American Chemical Society 2015.

FIGURE 8 | Schematic of an Engineered PTE Triple Helix Structure—A bacterial collagen-like domain was appended to the end of PTE to force the assembly of a trimer structure with a rigid tail section that could be assembled to QDs and other NPs for the purpose of studying kinetics of this non-natural structure. Reproduced from Breger et al. (2015b) with permission from the Royal Society of Chemistry.

observed an ∼4-fold increase in the initial rate of the reaction and an increase in efficiency ∼2-fold greater than the free enzyme. Further experiments showed that this enhancement was not the result of a reduction in activation energy rather a change in the rate of dissociation of product from the enzyme active site as similarly described for the free enzyme by Caldwell et al. (1991). Expanding upon these successes, the team assembled a trimeric PTE structure by attaching a collagen-like domain to the C-terminus of this enzyme (Breger et al., 2015b). The goal here was to create a material that could be spun into a fiber for subsequent integration into textiles such as clothing. Efforts focused on both increasing the potential packing density of the enzyme as well as examining the effect on enzyme activity as the spatial positioning of the enzyme increased in relation to the QD surface as seen in the schematic of **Figure 8**. Though the overall activity of the PTE-trimer was reduced compared to monomeric PTE, the QD-assembled, trimeric structures did show enhanced activity compared to free trimer when assembled to the NPs. The authors postulated that the observed enhancement could be the result of substrate localization due to the expanded solvation shell, the increased avidity due to the trimeric structure, or the controlled orientation of the enzyme itself.

As has been shown with QDS and other abiotic scaffolds, immobilization of enzymes to solid supports is beneficial in sensor development and, in many cases, efforts to improve the kinetic and biophysical properties of the enzyme. To ensure optimum benefit from immobilization it is imperative to ensure enzyme function by carefully selecting immobilization strategies that (1) ensure the enzyme structure is not perturbed and (2) the active site of the enzyme is not occluded in any way. DNA scaffolds and DNA nanostructures can be computationally designed based on Watson-Crick base pairing to ensure consistent and reproducible display of enzymes (Proschel et al., 2015; Quin et al., 2017; Qiu et al., 2018; Sun et al., 2019). Using QDs as their solid support and PTE as their model enzyme, Breger et al. designed a strategy to conjugate the two components using a DNA linker comprised of three modified single-stranded DNA (ssDNA) oligomers (Breger et al., 2017). The goal here was to establish novel methods of QDfunctionalization that potentially reduced biomolecule fouling of the QD surface while simultaneously validating a method that could be used for subsequent spatial positioning studies. Conveniently, QDs provide a simple method of biomolecular immobilization via metal affinity interaction between the shell of the QD and hexahistidine (His6) residues at the enzyme's termini which are included for purification over nickel chelate media. The PTE was conjugated to a second ssDNA molecule and then tethered to the QD surface using a third ssDNA molecule that was complementary to the other two sequences. In these studies, the authors observed some variability in the rate enhancement between conditions but were able to conclusively show that overall enzyme activity was improved above that of free enzyme using this method of NP assembly. In a subsequent study described by Samanta et al., the research team altered their strategy using a pyramidal DNA structure as a scaffold to arrange PTE-coated QDs on the surface of the DNA cage in hopes of establishing a new hybrid scaffolding system to enhance enzyme activity (Samanta et al., 2018). Here the authors designed ssDNA oligos, each terminating with a trimeric histidine repeat that would assemble into 3-way junctions. These subunits were further combined to form a pyramidal DNA cage. The free histidine repeats were then used to assemble QDs to the cage surface through metal affinity between the QD and the terminal histidine amino acids. Subsequent attachment of PTE was accomplished as previously described using the Hi<sup>6</sup> tag at the Cterminus of the enzyme see **Figure 9.** As in previous efforts by the Medintz research team, variability in enzyme enhancement was observed based on the size of the QD and on the ratio of enzyme and the DNA cages. The optimized conditions in these studies showed a nearly 12.5-fold enhancement in the rates of catalysis for the DNA cage, QD-assembled enzyme compared to both free enzyme controls and PTE assembled to QD alone.

# Enabling PTE Deployment

Enzymes are notoriously limited by issues of stability. Once removed from the ideal biological conditions of the cell, these intricately folded structures often unravel or collapse into nonfunctional amino acid chains or globules. While loss of function is expected with all biomolecules, researchers and commercial entities alike struggle to develop methods to limit loss of enzyme activity through storage formulation, direct protein engineering, and other approaches. The efficiency of remediation, rapid rates of catalysis, and environmental and biological compatibility of OP hydrolases such as of PTE make these ideal candidates for the development of commercial products. As mentioned above attaching enzymes to solid supports has been shown to improve enzyme stability and longevity. These strategies, however, are not always amenable to the development of therapeutic agents nor the wide-spread distribution of the materials due to cost of production and the potential for secondary immune response or environmental contamination. In contrast, the encapsulation of enzymes and other biomolecules has a long history in the development of therapeutics and, as will be discussed below, a future in the development of enzyme systems for bioremediation.

While the human PON1 enzyme shows catalytic activity against both V- and G-type agents, the rates of catalysis are not high enough for this enzyme to be an ideal therapeutic or prophylactic (Rajkovic et al., 2011; Le et al., 2015). In contrast, PTE catalysis of many OP compounds occurs at the rate of diffusion (Ghanem and Raushel, 2005). In a foundational publication, Pei et al. developed a method of PTE encapsulation targeting implementation of the enzyme as a medical counter measure (Pei et al., 1994). Here the authors permeablized murine erythrocytes which were subsequently filled with recombinant PTE using a hypotonic solution then resealed. This method yielded between 30 and 77% encapsulation depending upon the methods employed. While the PTE-filled erythrocytes were not tested in vivo the authors did show that hydrolysis of paraoxon could be achieved using this system and that enzyme encapsulation did not impair enzyme function. Budai et al. shifted away from natural carriers and characterized enzyme activity in synthetic stealth liposomes showing efficient hydrolysis of paraoxon in in vitro studies (Budai et al., 2009). In these foundational studies, the research team examined a range

of lipid monomers and assessed overall stability of liposomeenzyme construct over 4 days in guinea pig plasma. The authors showed that the encapsulated enzyme was able to significantly out-perform the free enzyme in both plasma and controls in buffered solutions. Liposomal encapsulation has also been used for the colocalization of enzyme and other therapeutics such as atropine as described for OPAA by Petrikovics (Petrikovics et al., 2000). These studies suggest that encapsulation of enzyme could be used as pathways for in vivo delivery of detoxifying therapeutics, or as discussed below, vehicles for field deployment.

In addition to enzyme stability, the cost of production can often be limiting to the realization of enzyme-based tools for bioremediation. In efforts to circumvent both of these complications simultaneously, researchers at the U.S. Naval Research Laboratory explored the direct production and encapsulation of PTE in proteoliposomes produced by Escherichia coli. Throughout their life cycle, bacteria release fragments of their outermost membrane containing proteins, nucleic acids, and other biomolecules which have been implicated in a range of functions from pathogenesis to community communication (Deatherage et al., 2009; Schwechheimer et al., 2013). As mimics of the parental bacterium's membrane, outer membrane vesicles (OMVs) are comprised of a protein decorated lipid bilayer that affords protection to cargo proteins from environmental proteases and nucleases. To exploit these benefits, Alves et al. utilized a protein-protein ligation system to anchor PTE to the inner wall of the E. coli outer membrane enabling protein loading into nascent OMVs, see **Figure 10** (Alves et al., 2015). Purified directly from the bacterial culture media, these particles showed high catalytic activity toward paraoxon. In subsequent studies, the research group showed that these biological NPs conveyed significant improvements to the stability of the encapsulated PTE allowing it to undergo iterative cycles of freezing and thawing and lyophilization as shown in **Figure 11** (Alves et al., 2016). Additionally, enzyme activity was preserved under elevated temperatures, in high salt solutions that mimic seawater, and over a range of pH conditions (Alves et al., 2016, 2018). These PTE-filled OMVs were also tested in paraoxonspiked environmental water samples, on surfaces such as painted metal coupons produced to mimic military vehicles, and on glass surfaces to demonstrate their utility in bioremediation (Alves et al., 2018). Finally, OMVs for PTE production and utilization also allowed for a simple and cost-effective method of large-scale enzyme production and dissemination. Alves et al. showed that their PTE-filled OMVs could also be easily purified from bulk growth media using an engineered His<sup>6</sup> affinity tag expressed on the surface of the vesicle (Alves et al., 2017). As bacteria such as the E. coli used here can be grown in fermenters at industrial scale, this one step purification protocol significantly reduces the burden associated with manufacture and downstream processing often associated with commercial enzyme production.

Taken as a whole and shown here with PTE, OMVs offer a highly adaptable vehicle for the design, development, and implementation of green biological tools for bioremediation. The ease of production using fermentation currently employed by large- and small-scale breweries shows the feasibility of this form of biomanufacturing. Additionally, the improvements to storage and enzyme stability has been observed with other OPdegrading enzymes such as DFPase and others (unpublished data from the Walper laboratory). Finally, OMVs offer a highly versatile platform to engineer and develop new bioremediation tools. The Walper laboratory has focused on E. coli as their proof-of-concept platform but other environmental species could similarly be engineered to develop bioremediation tools. As non-replicating biological particles, upstream modification of membrane channels, surface features, or others to improve functionality would likely not fall under the same regulations for environmental release currently encountered for microbial

FIGURE 10 | Protein-protein ligation strategy for outer membrane vesicle (OMV) loading. A recombinant membrane protein (OmpA) presenting a SpyTag label was expressed in E. coli as an outer membrane anchor for a phosphotriesterase (PTE)—SpyCatcher fusion protein. Interaction between the SpyTag/SpyCatcher partners allowed for the PTE enzyme to be pulled in to nascent OMVs during formation.

FIGURE 11 | Encapsulation of PTE within bacterial OMVs—The directed packaging of PTE within bacterial OMVs conveys improved stability to the enzymes allowing it to maintain activity over several common storage conditions such as freezing at −80◦C and lyophilization and rehydration. Unlike some synthetic liposome formulations, bacterial OMVs maintain their morphology following lyophilization and rehydration (middle panel, b). Encapsulation within the vesicles also allowed the enzyme to be stored for up to 14 days over a range of conditions while maintain a high level of enzyme activity compared to controls. This figure is a combination of data sets presented in Alves et al. (2015, 2016). Both figure reproduced with permission.

species. In conclusion, the composition of bacterial membrane vesicles enables researchers to devise numerous strategies for both their functionalization and purification which will undoubtedly lead to the development of new technologies exploiting these materials as sensors and tools for environmental decontamination (Chen Q. et al., 2017; Jan, 2017; Su et al., 2017).

## THE FUTURE OF ENZYMATIC SYSTEMS FOR BIOREMEDIATION AND DECONTAMINATION

The purpose of this review to was to illuminate the reader to both the need for, and successes with enzymes as tools for the bioremediation and decontamination of OP compounds. Whether used as pesticides or agents of war, these compounds have demonstrated potential to cause both acute and longterm harm to individuals and ecosystems alike. Unfortunately, with their wide-scale use in both developed and developing countries it is impossible to foresee a future in which these compounds are not released upon the world at a scale of millions of tons per annum. While chemical methods, and physical methods to some degree, can be employed for decontamination, neither of these pathways is sustainable. Therefore, it is imperative to establish green and sustainable bioremediation tools.

As we have shown here, scientists continue to make significant advances in developing enzymes as tools for environmental and personnel decontamination. In this review we very briefly describe six of the more prominent enzymes capable of degrading organophosphate compounds, however, with increasing genomic sequencing and cataloging of environmental microbes researchers will undoubtedly tease out new enzymes and novel pathways for the bioremediation of OPs and other environmental contaminants. As shown here, using the tools of molecular biology and protein design researchers have been able to significantly modify these enzymes vastly increasing the substrate binding ability and catalytic activity. These foundational studies will continue to evolve to improve both current and future OP hydrolases.

The large-scale use of enzymes as commercial products, therapeutics, and others has historically been impeded by the costs and difficulties associated with manufacture and stabilization. While these factors have historically proved as arguing points, companies such as Novozyme and others continue to demonstrate that the commercialization of enzymes in attainable and profitable. Additionally, the rapid proliferation of synthetic biology-based research efforts have led to the formation of companies such as Gingko Bioworks that focus largely on improving biomanufacturing capabilities through protein selection, gene optimization, and large-scale fermentation models. With significant financial backing in these companies and others, it is undeniable that there is a growing recognition in the potential of biological systems.

Finally, we have shown here with examples such as QD scaffolds and OMV encapsulation that downstream processing of enzyme systems offer potential mechanisms of significantly improving the applicability of enzymes as field deployable materials. In the proof-of-concept studies shown here we see significant improvements in both catalysis and stability using assembly and encapsulation platforms. These systems also offer the potential for multienzyme assemblies to move beyond the simple hydrolysis of substrate "A" to product "B," but rather from "A" to "D" through multi-enzyme pathways that demonstrate increased catalytic activity due to substrate channeling or proximity. These multi-enzyme pathways will be necessary to completely eliminate organophosphate compounds which are often nearly as toxic as the first hydrolysis product (Council, 1996).

In conclusion, while significant gains have been made in developing enzymes as candidate biomaterials the road ahead is still a long one. Further demonstrations in relevant environments and conditions are required to validate the potential of the materials for both personal and large-scale decontamination alike. This will require active collaborations between academy, government, and industry; each of which possesses a unique set of skills that can contribute to the realization of enzymes as tools for bioremediation and in decontamination. Unfortunately, even if all of these pieces are combined and the potential of enzymes realized, it is changes in societal and institutional perceptions of these dangerous compounds and their long-term impacts that are most needed and likely the furthest off.

# AUTHOR CONTRIBUTIONS

All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication.

# FUNDING

This work was made possible in part with funding from the Office of Naval Research (ONR), ONR Global, U.S. Naval Research Laboratory (NRL), and NRL Nanoscience Institute base funding. IM also acknowledges the National Institute of Food and Agriculture, U.S. Department of Agriculture, under award number 2016-67021-25038. SW acknowledges additional funding support from the Applied Research for Advancement of S&T Priorities Synthetic Biology for Militaryrelevant Environments sponsored by the Office of the Secretary of Defense.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00289/full#supplementary-material

# REFERENCES


warfare agent VR. Biochemistry 54, 5502–5512. doi: 10.1021/acs.biochem. 5b00629


K-10 in the amidohydrolase superfamily. Biochemistry 48, 4344–4353. doi: 10.1021/bi9004045


nanoparticle bioconjugates for pesticide detection. Analyst 142, 3261–3271. doi: 10.1039/C6AN02575G


organophosphates degradation. J. Hazard. Mater. 365, 789–795. doi: 10.1016/j.jhazmat.2018.11.063


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Thakur, Medintz and Walper. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Selection of High Producers From Combinatorial Libraries for the Production of Recombinant Proteins in *Escherichia coli* and *Vibrio natriegens*

Joel Eichmann1,2, Markus Oberpaul <sup>3</sup> , Tobias Weidner <sup>1</sup> , Doreen Gerlach<sup>3</sup> and Peter Czermak 1,2,3 \*

#### *Edited by:*

*Jussi Jantti, VTT Technical Research Centre of Finland Ltd, Finland*

#### *Reviewed by:*

*Dae-Hee Lee, Korea Research Institute of Bioscience and Biotechnology (KRIBB), South Korea Antoine Danchin, INSERM U1016 Institut Cochin, France*

> *\*Correspondence: Peter Czermak peter.czermak@lse.thm.de*

#### *Specialty section:*

*This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology*

*Received: 15 July 2019 Accepted: 20 September 2019 Published: 04 October 2019*

#### *Citation:*

*Eichmann J, Oberpaul M, Weidner T, Gerlach D and Czermak P (2019) Selection of High Producers From Combinatorial Libraries for the Production of Recombinant Proteins in Escherichia coli and Vibrio natriegens. Front. Bioeng. Biotechnol. 7:254. doi: 10.3389/fbioe.2019.00254* *1 Institute of Bioprocess Engineering and Pharmaceutical Technology, University of Applied Sciences Mittelhessen, Giessen, Germany, <sup>2</sup> Faculty of Biology and Chemistry, Justus-Liebig University of Giessen, Giessen, Germany, <sup>3</sup> Branch for Bioresources, Fraunhofer Institute for Molecular Biology and Applied Ecology, Giessen, Germany*

The optimization of recombinant protein production in bacteria is an important stage of process development, especially for difficult-to-express proteins that are particularly sensitive or recalcitrant. The optimal expression level must be neither too low, which would limit yields, nor too high, which would promote the formation of insoluble inclusion bodies. Expression can be optimized by testing different combinations of elements such as ribosome binding sites and N-terminal affinity tags, but the rate of protein synthesis is strongly dependent on mRNA secondary structures so the combined effects of these elements must be taken into account. This substantially increases the complexity of high-throughput expression screening. To address this limitation, we generated libraries of constructs systematically combining different ribosome binding sites, N-terminal affinity tags, and periplasmic translocation sequences representing two secretion pathways. Each construct also contained a green fluorescent protein (GFP) tag to allow the identification of high producers and a thrombin cleavage site enabling the removal of fusion tags. To achieve proof of principle, we generated libraries of 200 different combinations of elements for the expression of an antimicrobial peptide (AMPs), an antifungal peptide, and the enzyme urate oxidase (uricase) in *Escherichia coli* and *Vibrio natriegens*. High producers for all three difficult-to-express products were enriched by fluorescence-activated cell sorting. Our results indicated that the *E. coli* ssYahJ secretion signal is recognized in *V. natriegens* and efficiently mediates translocation to the periplasm. Our combinatorial library approach therefore allows the cross-species direct selection of high-producer clones for difficult-to-express proteins by systematically evaluating the combined impact of multiple construct elements.

Keywords: Golden Gate, MoClo, protein secretion, genetic engineering, high-throughput screening, antimicrobial peptide, uricase

# INTRODUCTION

Escherichia coli is one of the most common production hosts for recombinant proteins and many expression construct components are available to improve yields and facilitate protein recovery, including a variety of promoters and fusion proteins (Moore et al., 2016; Schreiber et al., 2017). Although, many recombinant proteins are strongly expressed with short production times and can be recovered as soluble products, others are difficult to express either because they are toxic to the host, resulting in low yields, or because they form insoluble inclusion bodies due to incomplete or incorrect folding (Carrió and Villaverde, 2002; Saïda, 2007). These challenges are often encountered when expressing antimicrobial peptides (AMPs), many of which require the formation of disulfide bonds for efficient folding and biological activity. In some cases inclusion bodies are produced deliberately, but the purification and refolding of insoluble proteins requires elaborate downstream processing and often limits the yield and functionality of the product (Hoffmann et al., 2018).

The folding of recombinant proteins in bacteria can also be improved by targeting the periplasm, a space between the outer and inner membrane that contains more chaperones than the cytoplasm, thus promoting folding and resulting in higher yields and less complex downstream processing steps. The oxidizing environment of the periplasm also favors the formation of disulfide bonds (Choi and Lee, 2004; Balasundaram et al., 2009). Translocation of proteins to the periplasm has been a subject to research since the very beginnings of recombinant protein expression technology (Gray et al., 1985; Oliveira et al., 1999). There are two major protein secretion pathways in bacteria. The general secretory (Sec) pathway exports unfolded proteins during or after translation (Tsirigotaki et al., 2017) whereas the twinarginine translocation (Tat) pathway exports folded proteins. Secretion to the periplasm is achieved by adding an N-terminal signal peptide, which is cleaved off during translocation (Berks et al., 2000). Overexpression of the Tat machinery has been shown to increase the rate of recombinant protein secretion (Browning et al., 2017).

The optimization of recombinant protein production requires the appropriate combination of regulatory elements to control expression and fusion tags for purification. Translation initiation occurs at a ∼35 bp sequence upstream of the start codon, containing the Shine-Dalgarno sequence. This so called ribosome binding site strongly effects the mRNA secondary structure (Dreyfus, 1988; Gold, 1988). Expression level screening is often based on combinatorial libraries of short constitutive promoters and/or ribosome binding sites (Coussement et al., 2014; Mahr et al., 2016). Multiple platforms for fusion tag screening have also been developed for bacteria, yeasts, and animal cells. However, all require the laborious cloning of large numbers of expression constructs (Abdulrahman et al., 2009; Sinah et al., 2012; Steinmetz and Auldridge, 2017).

One of the major drawbacks of high-throughput screening is that the expression levels and fusion tags are generally considered independently (one factor at a time), for example by using the same ribosome binding site for all constructs and testing a range of fusion tags. This is disadvantageous because the rate of protein synthesis strongly depends on mRNA secondary structures, so the combination of elements used in an expression construct can have a profound effect on its performance (Punginelli et al., 2004; Espah Borujeni and Salis, 2016). To overcome this challenge, we implemented a platform to generate combinatorial libraries of ribosome binding sites, secretion signals, and fusion tags. The library included five ribosome binding sites systematically combined with the Tatspecific signal peptides ssTorA and ssNapA, as well as the ssDmsA, ssYahJ, and ssYcdB peptides that are recognized in both the Tat and Sec pathways. All signal peptides have been previously demonstrated to mediate periplasmic translocation in E. coli (Tullman-Ercek et al., 2007; Fisher et al., 2008). These elements were further combined with four different Nterminal affinity tags for purification: glutathione-S-transferase (GST), maltose-binding protein (MBP), the small ubiquitin-like modifier (SUMO), and thioredoxin. All four tags have been shown to improve the solubility of recombinant proteins (Young et al., 2012), and affinity chromatography methods are available based, respectively, on immobilized glutathione (Schäfer et al., 2015), amylose (Reuten et al., 2016; Han et al., 2018), SUMOspecific antibodies (Butt et al., 2005), and the formation of reversible disulfide bonds (Mambetisaeva et al., 1997; McNiff et al., 2016). All four fusion tags were also combined with an additional N-terminal His<sup>6</sup> tag for purification by immobilized metal affinity chromatography (IMAC) (Loughran et al., 2017), resulting in eight different purification tag versions in the library. By combing the five ribosome binding sites, five secretion signals, and eight purification tags, the library contained 200 combinations of elements in total.

Libraries were assembled for three difficult-to-express proteins. The first product was the insect metalloproteinase inhibitor (IMPI) from the greater wax moth Galleria mellonella (Wedde et al., 1998). This AMP is 69 amino acids in length and contains five disulfide bonds, a major challenge for bacterial expression systems. The production of soluble IMPI has been achieved in redox-engineered E. coli strains with an oxidizing cytoplasm that promotes disulfide bond formation (Joachim et al., 2019). Moreover, IMPI has been produced with high yields in the form of inclusion bodies, and was successfully resolubilized using the Cry4AaCter pull-down tag (Hoffmann et al., 2019). However, IMPI has yet to be expressed as a soluble product by targeting the periplasm. The second product was the antifungal peptide lucimycin from the common green bottle fly Lucilia sericata. This peptide is 77 amino acids in length and has been expressed as a soluble product in E. coli, but with low yields (Pöppel et al., 2014; Schreiber et al., 2017). The third product was the enzyme urate oxidase (uricase) also from L. sericata. This enzyme is localized in the Malpighian tubes (Baumann et al., 2017) and catalyzes the degradation of uric acid to 5-hydroxyisourate, which is then converted to allantoin (Ramazzina et al., 2006). It has previously been expressed in E. coli albeit in the form of inclusion bodies (Baumann et al., 2017). All three targets were expressed as fusion proteins containing a secretion signal, fusion tag, a green fluorescent protein (GFP) fusion partner to allow the identification of high-producer clones by fluorescence activated cell screening (FACS), and a thrombin recognition site to cleave the fusion partners from the target protein.

Although, the expression libraries were designed for E. coli strain BL21, we also used them to optimize recombinant protein expression in Vmax Express, a commercial strain of the fastgrowing γ-proteobacterium Vibrio natriegens engineered for protein production by integrating a T7 expression cassette (Weinstock et al., 2016). This bacterium can utilize inexpensive carbon sources and has a high growth rate, making it ideal for biotechnology applications (Hoffart et al., 2017). It was recently demonstrated to be an alternative host for recombinant protein production (Schleicher et al., 2018; Becker et al., 2019). Here, we report the successful transfer of multiple expression construct elements from E. coli to V. natriegens, allowing the high-throughput screening of expression libraries and the identification of clones producing high yields of soluble recombinant proteins.

# MATERIALS AND METHODS

### Strains and Growth Conditions

Escherichia coli NEB 10-beta (NEB, Ipswich, Massachusetts, USA) was used for cloning and plasmid amplification. E. coli BL21(DE3) (Merck, Darmstadt, Germany) and V. natriegens Vmax Express (SGI-DNA, La Jolla, California, USA) served as expression hosts. Unless stated otherwise, E. coli was grown at 37◦C in LB Miller broth (LB) (Carl Roth, Karlsruhe, Germany) and Terrific Broth (TB) (Carl Roth) supplemented with 4 g/L glycerol. V. natriegens was cultivated in LB and TB supplemented with V2 salts (204 mM NaCl, 4.2 mM KCl, 23.14 mM MgCl2) at 30◦C. The following antibiotics were used for plasmid selection: ampicillin (100µg/mL), kanamycin (35µg/mL) and spectinomycin (E. coli 65µg/mL, V. natriegens 50µg/mL). Unless stated otherwise, the bacteria were cultivated in 300 or 500-mL shaking flasks with four baffles, filled with 30 or 50 mL medium, respectively, in an orbital shaker at 250 rpm. The medium was inoculated to an OD<sup>600</sup> of 0.1 from glycerol stocks. Protein expression was induced at an OD<sup>600</sup> of 1.0 by adding IPTG to a final concentration of 1 mM. Glycerol stocks were prepared by diluting an overnight culture with medium to reach an OD<sup>600</sup> of 0.1 and cultivating as above until the OD<sup>600</sup> reached 1.0. At this point, the cells were centrifuged (5,000 × g, 15 min, 4◦C) and the pellet was resuspended in ice-cold medium supplemented with 15% (v/v) glycerol. The volume was adjusted to OD<sup>600</sup> = 5.0 and the vials were stored at −80◦C.

# Library Cloning

Basic parts > 100 bp were synthesized (BioCat, Heidelberg, Germany) and subcloned in pUC57-Kan. Small basic parts were introduced into pUC57-Kan by PCR, using the Q5 Site-Directed Mutagenesis Kit (NEB). For assembly using the MoClo process, 40 fmol of each plasmid was mixed with 10 U T4 DNA Ligase (Promega, Mannheim, Germany), 2 µL of the corresponding buffer and 10 U BsaI or BbsI (NEB). The reaction mix was adjusted to a total volume of 20 µL with double-distilled water and incubated in a PCR cycler (PEQLAB, Erlangen, Germany) at 37◦C for 30 min, followed by 20 cycles at 37◦C (2 min) for digestion and 16◦C (5 min) for ligation. After a final restriction step at 50◦C for 10 min, the enzymes were heat-inactivated at 65◦C for 5 min, and 10 µL of this MoClo reaction mix was introduced into competent E. coli NEB 10-beta cells as described below.

# Preparation and Transformation of Chemically Competent *E. coli*

We inoculated 300 mL LB medium with 300 µL of an overnight culture of E. coli NEB 10-beta or E. coli BL21(DE3) cells grown at 37◦C to OD<sup>600</sup> = 0.9. The culture was transferred to sterile 50-mL centrifugation tubes and chilled on ice for 15 min. After centrifugation (5,000 × g, 10 min, 4◦C), each pellet was resuspended in 12.5 mL 100 mM CaCl2. The cells from three tubes were pooled and stored on ice for 30 min before centrifugation as above. Each pellet was resuspended in 5 mL ice-cold 100 mM CaCl<sup>2</sup> supplemented with 15% (v/v) glycerol. Aliquots of 250 µL competent cells were transferred to cryogenic vials, frozen on dry ice and stored at −80◦C.

For transformation, 80 µL of the competent cell suspension was mixed with 10 µL MoClo reaction mix and incubated on ice for 15 min before a 42◦C heat shock for 1 min, followed by incubation on ice for 1 min. The cells were then mixed with 1 mL LB and incubated in a ThermoMixer (Eppendorf, Hamburg, Germany) at 37◦C shaking at 1,000 rpm for 1 h. The cells were plated on LB agar supplemented with the corresponding antibiotic, and incubated at 37◦C overnight.

# Transformation of Chemically Competent *V. natriegens*

A vial containing 50 µL of chemically competent V. natriegens Vmax Express cells was thawed on ice and mixed with 200–400 ng plasmid library DNA. The cells were incubated for 30 min on ice before a 42◦C heat shock for 1 min, followed by incubation on ice for 2 min. The cells were then mixed with 1 mL pre-warmed LB supplemented with V2 salts. The cells were allowed to recover in a ThermoMixer at 30◦C shaking at 1,000 rpm for 2 h before streaking them on pre-warmed selection plates and incubating at 30◦C overnight.

#### Fluorescence Microscopy

Microscope slides with cavities were filled with boiling 1% (v/v) agarose in TAE buffer and covered with a cover glass. When the agarose had cooled, the cover glass was removed and 2 µL of culture medium was added, before applying a new cover glass. Fluorescence microscopy was carried out using a Leica DMI6000 instrument (Leica Microsystems, Wetzlar, Germany) fitted with an HCX PL FLUOTAR phase contrast objective (100 x, numerical aperture 1.3), L5 filter cube (excitation filter BP 480/40 nm, dichromatic mirror 505 nm, suppression filter BP 527/30 nm), and PhotoFluor II light source (Chroma, Bellows Falls, Vermont, USA) at 470 nm. Images were captured with a DFC360FX camera (Leica Microsystems). Brightness and contrast adjustments and image cropping were carried out using Fiji software (Schindelin et al., 2012).

#### Fluorescence-Activated Cell Sorting

FACS was carried out using a BD FACSCalibur device (BD Bioscience, San Jose, California, USA), including standard laser and filter equipment. Phosphate buffered saline (PBS) was used as the sheath fluid (Thermo Fisher Scientific, Waltham, Massachusetts, USA). E. coli and V. natriegens cells grown as described above were sorted by FACS 4 h post-induction. For sample preparation, a small quantity of culture was suspended in PBS in a 5-mL test tube (VWR, Radnor, Pennsylvania, USA). The cell density was adjusted with PBS to achieve 2,000 ± 200 events per s. Forward-scatter (FSC) and side scatter (SSC) characteristics were applied to distinguish cells from background noise. The threshold was set on FSC to reduce background noise. Events of interest were gated, excited at 488 nm and analyzed for their fluorescence intensity using the 530/30 nm band-pass filter (FL1- H). A gate was applied to the top 5–10% of events with the highest fluorescence intensity, resulting in the sorting of 50,000 cells in exclusion mode. Subsequently, the cell suspension was concentrated from ∼150 to 1 mL using a 0.2-µm nylon filter membrane (Merck), transferred to a flask with fresh medium and incubated overnight. The following day, serial dilutions of the culture were streaked on LB agar plates. Additionally, shaking flasks with fresh medium were inoculated from the overnight cultures at an OD<sup>600</sup> of 0.1, grown until the OD<sup>600</sup> reached 1.0 and induced with a final concentration of 1 mM IPTG. Four hours post-induction, the fluorescence intensity of the cells was analyzed using the same settings described above.

#### Selection of High Producers

Each well of a 96-deep-well plate (VWR) was filled with 500 µL TB. Single clones picked from the agar plate previously streaked with sorted cells were used to inoculate the wells, and the cultures were grown overnight at 30◦C shaking at 400 rpm. The following day, a new 96-deep-well plate was filled with 460 µL TB and inoculated with 20 µL of the overnight cultures. After cultivation for 3 h as above, recombinant protein expression was induced

TABLE 1 | Organisms and expression construct combinations found in the selected clones.


*The frequency indicates the number of clones carrying a specific combination. High producers highlighted in bold were identified as most productive clones and used for the production of the model proteins. The ribosome binding sites and secretion signal sequences are provided in the Supplementary Material.*

by adding 20 µL of IPTG to each well (final concentration 1 mM). Four hours post-induction, the plates were centrifuged (5,000 × g, 10 min, 4◦C) and the pellets were resuspended in 25 µL BugBuster Master Mix (Merck) and incubated at room temperature for 20 min. Insoluble fractions were pelleted by centrifugation (5,000 × g, 20 min, 4◦C) and the supernatants containing the soluble fractions were diluted in PBS. For the 12 clones showing the highest fluorescence in the soluble protein fraction, the production procedure was repeated in 50 mL TB in 500-mL shake flasks. Plasmids from these cultures were isolated using the NucleoSpin Plasmid EasyPure kit (Macherey-Nagel, Düren, Germany) according to the manufacturer's instructions before sequencing (Microsynth Sequlab, Göttingen, Germany).

#### Osmotic Shock Procedure

Pellets from 2-mL cultures (see section Strains and growth conditions) were resuspended in 1 mL osmotic shock buffer 1 (20 mM Tris-HCl, 0.25 mM EDTA, 200 g/L sucrose, pH 8.0) and incubated on ice for 10 min. After centrifugation (16,000 × g, 10 min, 4◦C) the pellet was resuspended in 1 mL osmotic shock buffer 2 (20 mM Tris-HCl, 0.25 mM EDTA, pH 8.0) and incubated as above. After further centrifugation as above, the supernatant containing the periplasmic protein fraction was transferred to a fresh reaction tube and stored at 4◦C. The remaining pellet was resuspended in 100 µL BugBuster Master Mix and incubated at room temperature for 20 min. The soluble and insoluble fractions were then separated by centrifugation (16,000 × g, 20 min, 4◦C) and 90 µL of the supernatant containing the cytoplasmic protein fraction was transferred into a fresh tube, mixed with 810 µL PBS and stored at 4◦C. The remaining supernatant was discarded, and the pellet containing the insoluble and membrane protein fraction was resuspended in 1 mL PBS.

#### Protein Analysis

The fluorescence intensity of each 10-µL sample of soluble protein from the periplasmic/cytoplasmic fraction was determined by mixing with 90 µL PBS, transferring to the wells of a 96-well microtiter plate and measuring the fluorescence using a Synergy HT plate reader (BioTek Instruments, Winooski, Vermont, USA) against GFP standards (Abnova, Taipei City, Taiwan), which were used to generate a standard curve. For SDS-PAGE and western blot analysis, 5–10 µL of each sample (soluble or insoluble protein in PBS) was mixed with 3.75 µL of Laemmli buffer supplemented with β-mercaptoethanol and brought to a total volume of 15 µL with double-distilled water. The sample was denatured by incubation at 95◦C for 5 min and 10 µL was loaded on a 4–20% Citereon TGX Stain-Free Precast Gel (Bio-Rad, Hercules, California, USA). After running for 25 min at 250 V, gel images were captured using the ChemiDoc Imaging System (Bio-Rad). Western blots were performed using the Trans-Blot Turbo Transfer Pack and Transfer System (Bio-Rad) according to the manufacturer's instructions. Membranes were blocked with 5% bovine serum albumin (BSA) for 1 h and washed three times (5 min each) with 0.1% Tween-20 in PBS before incubation at room temperature for 2 h with the primary rat-anti-GFP 3H9 antibody (ChromoTek, Planegg-Martinsried, Germany) diluted 1:5,000 in PBS containing 0.05% Tween-20. After washing as above, the membranes were incubated at room temperature for 2 h with the horseradish peroxidase (HRP)-conjugated secondary goat anti-rat IgG antibody (Jackson ImmunoResearch Europe, Ely, UK) and Precision Protein StrepTactin-HRP Conjugate (Bio-Rad), each diluted 1:10,000. HRP-mediated luminescence was visualized using Clarity Western ECL Substrate (Bio-Rad) according to the manufacturer's protocol. Uricase activity was determined in the soluble fractions using the Amplex Red Uricase Assay Kit (Thermo Fisher Scientific) according to the manufacturer's instructions.

# RESULTS AND DISCUSSION

#### Expression Library Cloning Strategy

The combinatorial expression libraries were assembled in two consecutive cloning steps, starting from basic parts (the ribosomal binding sites, secretion tag sequences, and target protein sequences). Each of the basic parts was flanked by BsaI sites and housed in the cloning vector pUC57-Kan. Because the 4-bp BsaI overhang ends would result in frame shifts, two additional bases were introduced between the 5′ overhang and the coding sequences. For each of the five secretion tags, a synthetic ribosomal binding site was designed using the RBS Calculator (Salis et al., 2009; Espah Borujeni et al., 2014). Based on the promoter sequence and the coding sequence of the secretion tag, the ribosomal binding site was generated with the restriction of base numbers applying only to (N)25CAGG or (N)30CAGG sequences, where CAGG represents the BsaI overhang. The sites with the highest predicted translation initiation rates (TIRs) were chosen for the library (**Table S1**). Given that translational initiation depends not only on the ribosomal binding site but also on the subsequent coding sequence, the five ribosomal binding sites would lead to five different expression levels for each secretion tag. The TIR was predicted using a reverse calculation for each ribosomal binding site + secretion tag combination (**Table 1**).

The first cloning step was the assembly of basic parts in Level 1 MoClo vectors (**Figure 1A**). We used the MoClo vectors from the Waldminhaus laboratory, which carry the ccdB gene in addition to lacZα (Schindler et al., 2016). The ccdB gene

FIGURE 1 | Assembly of combinatorial expression libraries. (A) Schematic illustration of the MoClo assembly. Basic parts are housed in pUC57-Kan, flanked by BsaI-sites. The T7 promoter is housed in pMA350. The thrombin cleavage site, green fluorescent protein (GFP), one of the proteins of interest (POI), and the terminator (Ter) are assembled in pMA352. The five ribosome binding sites (RBS) and five secretion signals are pooled, and assembled with one of the eight affinity tags (A1–A8) as Pre-libraries in pMA351. The eight affinity tags are: A1, SUMO; A2, His6-SUMO; A3, MBP; A4, His6-MBP; A5, GST; A6, His6-GST; A7, Trx; A8, His6-Trx. In Level 1, plasmids and inserts are flanked by BbsI sites. For final library assembly, the Pre-libraries are pooled and assembled with the Prefix and Suffix. (B) BsaI digestion of plasmids from a pooled library, and plasmids from single clones. The upper band represents the vector backbone, and the lower bands represent the assembled expression cassettes. Inserts vary in size due to the differing sizes of the affinity tags.

encodes an inhibitor of DNA gyrase and thus kills E. coli cells taking up undigested or re-ligated plasmids (Bernard, 1996). This eliminates the background of undigested plasmids and increases the percentage of positive clones. Basic parts were assembled in three Level 1 vectors by Golden Gate cloning using BsaI and T4 ligase (Engler et al., 2009). The IPTG-inducible T7 promoter part, including the lacI repressor gene, was cloned in the first vector, pMA350, and is hereafter called Prefix. Pre-libraries were assembled in the second vector, pMA351. In this step, the coding sequence for one of the eight affinity tags was mixed with the five ribosomal binding sites and five secretion tags, resulting in 25 combinations. Eight Pre-libraries were prepared, one each for the eight affinity tags in the final library design. In the third vector, pMA352, we assembled the coding sequences for GFP, the thrombin cleavage site, and the protein of interest, as well as a transcriptional terminator. This building block is hereafter called Suffix and three versions were assembled, one for each target protein. Individual Prefix and Suffix clones were picked and verified by Sanger sequencing. Furthermore, all clones in each Pre-library (n = 110–900) were pooled and analyzed by Sanger sequencing, identifying the overlapping signals of the variable library elements.

The final libraries were assembled in vector pMA62, carrying a spectinomycin resistance gene and the IncP replication system (Werner et al., 2012; Schindler et al., 2016). Preliminary experiments revealed that this vector achieves efficient transformation and protein expression in V. natriegens. Equimolar amounts of all Pre-libraries were mixed in one tube with the Prefix, one of the three Suffix clones and the end-linker pMA671 for Golden Gate assembly using BbsI and T4 ligase. End-linkers with suitable overhangs serve as adapters whenever fewer than seven inserts are assembled in a MoClo reaction. Final libraries for each target protein were generated individually, and theoretically contained 200 combinations of ribosomal binding sites, secretion signals and affinity tags. For each library, 2,000–4,000 clones were pooled. Given a cloning efficiency of 80–100% at each step, there was a >99% probability that the final libraries covered all possible combinations. BsaI restriction analysis of the pooled libraries resulted in a pattern of five bands (**Figure 1B**). Given that all the ribosomal binding sites and secretion tags were similar in size, whereas the affinity tag sequences ranged from ∼300 to 1,100 bp, each of the four lower bands represent the combinations with one of the affinity tags. The upper band represents the vector backbone. BsaI test digestions of plasmids from individual clones revealed the upper band but only one the four possible lower bands.

The two-stage MoClo assembly (**Figure 1A**) was preferred to standard Golden Gate assembly, where all basic parts are assembled in a single reaction, because as the number of parts increases the efficiency of Golden Gate assembly declines and

FIGURE 2 | Microscopic images of pooled cultures carrying a combinatorial expression library. Arrows indicate cells with homogeneous GFP, indicating a soluble product. Stars show GFP foci, indicating insoluble inclusion bodies. Images represent the expression of the library for IMPI production in *V. natriegens* (A–C) and *E. coli* (D–F) 4 h post-induction.

more errors are introduced (Potapov et al., 2018). When building the libraries, none of the assembly steps involved more than four parts, and accordingly we generated large numbers of colonies at low error rates. During Level 1 and Level M cloning, sequencing revealed 87, 5% of randomly picked clones to be positive (42/48). Furthermore, when multiple parts differing in size are used to generate combinatorial libraries, there is a bias toward smaller parts. By assembling the Prefix, Pre-libraries and Suffix clones, the size differences among the basic parts were minimized. In the final libraries, we found no indication that inserts from smaller Pre-libraries were overrepresented in a manner that would lead to biased screening. Furthermore, the preparation of separate Prefix, Pre-libraries and Suffix clones makes the selection of false positives during FACS less likely. In Golden Gate cloning, single parts might not be assembled, resulting in an incomplete construct, but only whole parts would be lost. During final assembly, a small but significant number of incomplete constructs will be generated. However, the incomplete construct would either lack the promoter (missing Prefix), the ribosomal binding site and start codon (missing Pre-library) or the GFP (missing Suffix). Due to the design of the library assembly method, none of these incomplete constructs would produce a strong GFP signal. Thus, clones carrying misassembled plasmids would be excluded by FACS. For future screening with other products, only the Suffix needs to be cloned. All remaining parts can be reused. The assembly of multiple Pre-libraries means that some variants can be left out if not required and others can be added. Altogether, the design and assembly of the libraries provides a high level of modularity and flexibility.

# The Selection of High Producers

The final libraries for each target protein were introduced into the E. coli BL21(DE3) and V. natriegens Vmax Express production strains, and the libraries in each case were separately pooled and stored as glycerol stocks. At the beginning of the selection pipeline, shaker flasks containing growth medium were inoculated from glycerol stocks to an OD<sup>600</sup> of ∼0.1. Recombinant protein expression was induced by adding IPTG when the culture reached an OD<sup>600</sup> of ∼1.0.

Fluorescence microscopy revealed the diversity of protein expression levels in the libraries (**Figure 2**). We assigned the cells to three groups. The first and largest of the groups comprised cells showing no fluorescence, indicating that the fusion protein was not expressed (or expressed at levels below the detection threshold) or was expressed but inactive, perhaps due to incomplete folding. The second group comprised cells showing homogeneous fluorescence (white arrows in **Figures 2B,E**), indicating the expression of a soluble recombinant protein. The intensity of fluorescence varied widely in this group, including cells with very strong signals representing high producers.

The final category comprised cells with punctate fluorescence (white stars in **Figures 2B,E**), indicating the presence of inclusion bodies.

For FACS-based selection, the cells were resuspended in PBS 4 h post-induction. Two gates were applied, the first to separate cells from background events and the second (applied to the fluorescence channel) to separate the 5–10% of cells with the highest fluorescence intensity. We sorted 50,000 cells in each of the three libraries. At a sorting rate of 30–100 events per s, these high-producer cells were separated into a large volume of PBS (100–150 mL). We therefore concentrated the suspension to 1 mL by removing the buffer using a sterile filter and a vacuum pump, and transferred the cells to a new flask for overnight incubation.

The following day, the growth and induction procedures were repeated with a culture inoculated to an OD<sup>600</sup> of 0.1 from the sorted overnight culture. **Figure 3** shows the fluorescence intensity of cells before and after sorting. Most of the pre-sorting cells showed little or no fluorescence, although the V. natriegens population expressing lucimycin was a notable exception. After one round of cultivation and sorting, the fluorescence intensity of the V. natriegens cells had increased by approximately one order of magnitude for most of the cells producing IMPI and uricase, and by approximately two orders of magnitude for the cells producing lucimycin. The fluorescence of the pooled E. coli cultures also increased after sorting, although not to the extent observed with V. natriegens. In most cases, the fluorescence intensity remained similar to that in the original culture, although small subpopulations showed strong increases in fluorescence (**Figure 3B**). After two to four rounds of growth and sorting, the cells carried plasmids representing a single combination of elements from the original library. Production with this plasmid resulted in large amounts of insoluble recombinant protein (data not shown). To exploit the diversity of high producers, we therefore decided to analyze the 5–10% of cells with the highest fluorescence intensity after only one round of screening, at the cost of a higher background of less productive clones.

The workflow for the selection of high producers from libraries is shown in **Figure 4**. The day after sorting, dilutions of cells were plated on agar. Single clones were cultivated in 96-deep-well plates and protein production was induced by adding IPTG. Four hours after induction, cells were pelleted by centrifugation, and the supernatant was discarded. After lysis, the fluorescence intensity of the soluble protein fraction was measured. The top clones based on these readings were grown in shaker flasks and induced by adding IPTG. The soluble and insoluble protein fractions were analyzed by SDS-PAGE and western blot (**Figures 5**–**7**). Plasmids isolated from the top candidates were dispatched for sequencing to identify the basic elements.

#### Production of Model Peptides and Proteins

The combinations of elements resulting in the highest yield for each of the three target proteins are summarized in **Table 1**. The highest yields of IMPI were achieved in E. coli using the

FIGURE 4 | High producer selection pipeline. Combinatorial libraries were cloned and introduced into the production strains. Cultures carrying the pooled libraries were grown and expression was induced. Cells showing the highest GFP levels were selected by FACS, resulting in a cell suspension in PBS. The cell suspension was concentrated by filtration, transferred into a flask with fresh medium and grown overnight. The resulting culture was plated on agar before transferring 96 single colonies to a 96-deep-well plate. Expression was induced for 4 h. Cells were pelleted, the soluble protein fractions were extracted, and the GFP signal was quantified. The clones showing the most intense fluorescence were cultivated in flasks and subsequently analyzed by SDS-PAGE, western blot, and Sanger sequencing. Image created with BioRender.

SUMO tag. The combination ssDmsA-MBP was present in many clones at the end of the selection pipeline but this resulted in lower yields than the SUMO fusion. Interestingly, the best combination for both lucimycin and uricase was ssYahJ-His6-Trx in V. natriegens. For uricase, all clones reaching the final selection step contained this combination. For lucimycin, the combination ssDmsA-MBP was again present in many clones at the end of the selection pipeline but it achieved lower yields than ssYahJ-His6- Trx. Only insoluble product was observed for the production of IMPI in V. natriegens and uricase in E. coli. The high-producer expression systems highlighted in **Table 1** were therefore used for the production of the three model proteins.

#### Production of IMPI

IMPI was produced in E. coli as a fusion to the SUMO tag and the product was exported to the periplasm mediated by the ssDmsA secretion signal. The soluble protein fractions from the periplasm and cytoplasm were separated by osmotic shock (**Figure 5A**). Most of the protein was located in the cytoplasm 2 and 4 h postinduction (**Figure 5B**). The protein yield in the cytoplasm did not increase further from 4 to 6 h post-induction. In contrast, the amount of protein in the periplasm doubled in this time. Six hours post-induction, about half of the soluble recombinant protein was located in the periplasm. At all times, a substantial quantity of insoluble product was detected (**Figure 5A**). IMPI

FIGURE 5 | IMPI production in *E. coli*. (A) Western blot (anti-GFP antibody) of soluble ssDmsA-SUMO-GFP-IMPI in the periplasmic (PP), cytoplasmic (CP), and insoluble and membrane (I) fraction 4 and 6 h post-induction. The black arrow indicates the expected band at 52.8 kDa. (B) Fusion protein quantification in the periplasmic and cytoplasmic fractions 2, 4, and 6 h post-induction. Error bars indicate the standard deviation.

contains five disulfide bonds, so translocation is likely to be accomplished by the Sec pathway, whereas the IMPI located in the cytoplasm is probably in an unfolded or incompletely folded state. Large amounts of insoluble protein result in the formation of inclusion bodies, as shown in **Figure 2**. The relatively low yield of 5–6 mg/L total soluble product must be regarded in the context of cell growth. Six hours post-induction, the cultures had reached an OD<sup>600</sup> of only 7.5–8.0 but were still in the exponential growth phase. Given the large quantity of insoluble product, the yield of this expression system could be increased substantially by optimizing the cultivation and production process.

#### Production of Lucimycin

The antifungal peptide lucimycin was produced in V. natriegens with a thioredoxin tag to improve solubility and a His<sup>6</sup> tag for purification by IMAC. The combination of the ssYahJ secretion signal and the selected ribosome binding site resulted in a low TIR compared to other combinations, but nevertheless achieved the highest yields (**Figure 6A**). Six hours post-induction the concentration of the soluble product was ∼20 mg/L in total, about two thirds of which was found in the periplasmic fraction following the osmotic shock (**Figure 6B**). In the periplasmic fraction, a second band is visible slightly below the expected band

FIGURE 6 | Lucimycin production in *V. natriegens*. (A) Western blot (anti-GFP antibody) of soluble ssYahJ-His-Trx-GFP-lucimycin in the periplasmic (PP), cytoplasmic (CP), and insoluble and membrane (I) fraction 4 and 6 h post-induction. The upper black arrow indicates the expected band at 53.2 kDa, the lower arrow indicates the putative mature protein after the cleavage of the secretion signal. (B) Fusion protein quantification in the periplasmic and cytoplasmic fractions 2, 4, and 6 h post-induction. Error bars indicate the standard deviation.

at 53.2 kDa in the western blot (**Figure 6A**). The Sec and Tat translocation machineries cleave N-terminal secretion signals at the C-terminal Ala-x-Ala motif (Freudl, 2018). The removal of the secretion signal could explain the presence of the lower band. The product yield increased continually during production, and the quantity of insoluble was moderate. As for the production of IMPI in E. coli, the optimization of the production process could significantly increase the yields of lucimycin.

#### Production of Uricase

We were unable to identify any E. coli clones that achieved the production of substantial quantities of the model enzyme uricase as a soluble protein. However, moderate yields were observed when the enzyme was expressed in V. natriegens using the same ssYahJ-His6-Trx combination that achieved the highest yields of lucimycin, albeit with a different ribosome binding site and thus a higher TIR. Even so, the production and localization of the enzyme differed from lucimycin (**Figure 7**). The uricase fusion protein accumulated mostly in the cytoplasmic fraction, but only 2 h post-induction substantial uricase activity was also detected in the periplasmic fraction (**Figure 7C**). The highest yield and activity were detected 4 h post-induction, and the amount of insoluble product increased over time (**Figures 7A,B**). Two unexpected bands were observed in the western blot after 4 and 6 h. The dominant band slightly below the 75 kDa marker is likely to reflect non-specific antibody binding. Similar patterns were detected in all V. natriegens samples (see **Figure 6A**). The lower band between the 37 and 50 kDa markers may reflect the incomplete translation of the fusion protein, resulting in a truncated product, including GFP and N-terminal fusion tags, but missing the majority of the uricase. This would explain the increasing yield of uricase in the periplasmic fraction over time even though the activity in this fraction decreases. Translation is often interrupted by the presence of rare codons. The sequences were originally codon optimized for E. coli, and further analysis revealed some codons that are rarely used in V. natriegens. The codon CCC for proline is one example (Lee et al., 2016), and it is found at the beginning of the uricase fusion protein sequence. Replacing rare codons to match V. natriegens codon preferences would therefore be a starting point for the improvement of uricase production.

# CONCLUSIONS

One of the major challenges during expression screening is the formation of insoluble inclusion bodies. These often contain large quantities of active protein, leading to strong GFP signals that cannot be distinguished from the homogeneous fluorescence of soluble proteins during FACS. This issue could be addressed by including a biosensor for inclusion body formation. The transcriptomic response to inclusion body formation has been investigated, and offers multiple starting points (Baig et al., 2014). The expression of a second fluorescent marker in the presence of inclusion bodies would allow such cells to be excluded, resulting in the specific gating of cells expressing soluble recombinant proteins.

We confirmed that V. natriegens is a suitable host for recombinant protein production, especially in the case of uricase. Although, uricases from other species have been produced successfully in E. coli (Nakagawa et al., 1995; Shaaban et al., 2015), this is not the case for L. sericata uricase and we were similarly unable to identify any E. coli clones that produced this enzyme in significant amounts. In contrast, V. natriegens produced large amounts of the enzyme in a soluble form.

We found that many of the functional elements in E. coli expression constructs could also be used in V. natriegens, including the ssYahJ secretion signal. Most of the lucimycin product was detected in the V. natriegens periplasmic fraction, and the presence of an additional smaller product indicated the cleavage of the secretion signal (**Figure 6A**). In E. coli, proteins fused to thioredoxin can be released from the cytoplasm by osmotic shock (Ajouz et al., 1998), but it is unclear whether this also occurs in V. natriegens. We expected to detect most of the Trx-uricase fusion protein in the periplasm (**Figure 7**) so it is unclear why the same fusion protein combination leads to the periplasmic localization of lucimycin. The abundance of insoluble uricase indicates incomplete or incorrect folding of the large fusion protein. Given that only folded proteins are translocated by the Tat pathway, an incompletely folded protein would be expected to remain in the cytoplasm. Together with the finding that ssYahJ was the major secretion tag selected in V. natriegens and other high producers (data not shown), we propose that ssYahJ is recognized as a secretion signal in V. natriegens. To validate these results, the ssYahJ tag needs to be systematically investigated in V. natriegens in the absence of thioredoxin.

Taken together, the screening resulted in the identification of high producers for three difficult-to-express products in E. coli and V. natriegens. These will serve as a starting point to improve yields by optimizing the production process, including medium development and scaled-up fermentation in bioreactors. We found that the MoClo system allowed the efficient construction of combinatorial libraries containing both regulatory and coding elements. The high modularity of this screening platform will facilitate the identification of additional

#### REFERENCES


elements and proteins of interest, and can serve as a blueprint for novel combinatorial library screening methods in microbial and eukaryotic expression systems.

#### DATA AVAILABILITY STATEMENT

All datasets generated for this study are included in the manuscript/**Supplementary Files**.

#### AUTHOR CONTRIBUTIONS

JE conceived, designed and performed all experiments, wrote the manuscript, and coordinated its preparation. MO established the FACS and cell concentration methods, and helped during all sorting experiments. DG and TW helped to draft and revise the manuscript. PC helped to draft and revise the manuscript, and supervised the research. All authors have given their approval for this final version of the manuscript.

#### FUNDING

This work was financially supported by the Hessian Ministry of Education and Art within the Hessen initiative for supporting economic and scientific excellence (LOEWE). Part of the work was funded by the German Ministry of Education and Research (BMBF) via the project 4-IN.

#### ACKNOWLEDGMENTS

We wish to thank Torsten Waldminghaus (Synmikro, Marburg, Germany) for providing the MoClo vectors, Jens Glaeser (Evotec International GmbH, Göttingen, Germany) for providing the FACS equipment, and technical and scientific advice, Sebastian Hausner and Jan Zitzmann for helping with cloning and data analysis, respectively, and Florian Scholz for excellent technical advice. We are grateful to Howard Salis (Pennsylvania State University, PA, USA) for help with the RBS Calculator. The authors thank Richard M. Twyman for manuscript editing.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00254/full#supplementary-material

Baig, F., Fernando, L. P., Salazar, M. A., Powell, R. R., Bruce, T. F., and Harcum, S. W. (2014). Dynamic transcriptional response of Escherichia coli to inclusion body formation. Biotechnol. Bioeng. 111, 980–999. doi: 10.1002/bit.25169


optimize the production of antimicrobial peptides. Microb. Cell Fact. 16:29. doi: 10.1186/s12934-017-0637-5


from the hemolymph of greater wax moth larvae, Galleria mellonella. Eur. J. Biochem. 255, 535–543. doi: 10.1046/j.1432-1327.1998.2550535.x


**Conflict of Interest:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Eichmann, Oberpaul, Weidner, Gerlach and Czermak. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# PARAGEN 1.0: A Standardized Synthetic Gene Library for Fast Cell-Free Bacteriocin Synthesis

Philippe Gabant\* and Juan Borrero†

Syngulon, Seraing, Belgium

#### Edited by:

Jean Marie François, UMS3582 Toulouse White Biotechnology (TWB), France

#### Reviewed by:

Dzung B. Diep, Norwegian University of Life Sciences, Norway Dae-Hee Lee, Korea Research Institute of Bioscience and Biotechnology (KRIBB), South Korea

#### \*Correspondence:

Philippe Gabant pgabant@syngulon.com

#### †Present address:

Juan Borrero, Sección Departamental de Nutrición y Ciencia de los Alimentos (Facultad de Veterinaria), Universidad Complutense de Madrid, Madrid, Spain

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 27 June 2019 Accepted: 22 August 2019 Published: 06 September 2019

#### Citation:

Gabant P and Borrero J (2019) PARAGEN 1.0: A Standardized Synthetic Gene Library for Fast Cell-Free Bacteriocin Synthesis. Front. Bioeng. Biotechnol. 7:213. doi: 10.3389/fbioe.2019.00213 The continuous emergence of microbial resistance to our antibiotic arsenal is widely becoming recognized as an imminent threat to global human health. Bacteriocins are antimicrobial peptides currently under consideration as real alternatives or complements to common antibiotics. These peptides have been much studied, novel bacteriocins are regularly reported and several genomic databases on these peptides are currently updated. Despite this, to our knowledge, a physical collection of bacteriocins that would allow testing and comparing them for different applications does not exist. Rapid advances in synthetic biology in combination with cell-free protein synthesis technologies offer great potential for fast protein production. Based on the amino acid sequences of the mature peptide available in different databases, we have built a bacteriocin gene library, called PARAGEN 1.0, containing all the genetic elements required for in vitro cell-free peptide synthesis. Using PARAGEN 1.0 and a commercial kit for cell-free protein synthesis we have produced 164 different bacteriocins. Of the bacteriocins synthesized, 54% have shown antimicrobial activity against at least one of the indicator strains tested, including Gram-positive and Gram-negative bacteria representing commonly used lab strains, industrially relevant microorganisms, and known pathogens. This bacteriocin collection represents a streamlined pipeline for selection, production, and screening of bacteriocins as well as a reservoir of ready-to-use antimicrobials against virtually any class of relevant bacteria.

#### Keywords: antimicrobial peptides, bacteriocins, cell-free extracts, in vitro synthesis, synthetic biology

The remarkable health benefits that have been achieved with antibiotics are being questioned by the rapid increase of resistant bacteria. World Health Organization's new Global Antimicrobial Surveillance System (GLASS) study shows an antibiotic resistance occurrence among 500,000 people with suspected bacterial infections throughout 22 countries (WHO, 2015). Hence a growing list of infections like pneumonia, tuberculosis and foodborne diseases are becoming sometimes impossible to treat as antibiotics are not so effective (WHO, 2015). The antibiotic resistance crisis has been attributed to an excessive as well as abusive use of antibiotics, and also as the consequence in the decrease of new drug development by the pharmaceutical industry due to reduced economic incentives and demanding regulatory requirements (Cooper and Shlaes, 2011; White, 2011). Moreover, it has been shown that the consumption of broad-spectrum antibiotics may damage the human commensal microbiota which has major roles in the host's health (Blaser, 2011; Willing et al., 2011; Lange et al., 2016). Consequently, the development of new antimicrobial agents that may be used in different settings (clinic, food industry, animal feed, etc.) has become crucial.

Antimicrobial peptides (AMPs) have been described as "evolutionarily ancient weapons against microbial infections" (Zasloff, 2002). AMPs are produced by all organisms, including human beings, and play an essential role in innate immunity. They are key elements of the innate immune system which deliver immediately effective and non-specific defenses against infections. Among the large variety of AMPs, there is a subgroup known as bacteriocins which are small peptides, produced by bacteria and ribosomally synthesized. Bacteriocins, which were discovered in 1925 (Gratia and Fredericq, 1946), exhibit strong activity against their target strains. This activity has often been found in the nanomolar range, making them in some cases more potent than their antibiotic analogs (Mathur et al., 2013; Ming et al., 2015). Another interesting feature is that, compared to antibiotics, most bacteriocins are only active against a narrow spectrum of bacteria close related to the producer. A few years ago, this narrow inhibition spectrum was considered as a major bottleneck, however today we know that big disruptions in the gut microbiota caused by broad spectrum antibiotics are coupled to immunological, metabolic, and neurological disorders in the host. Therefore, it is believed that targeting specific bacteria instead of the whole microbiome is more efficient and less intrusive (Mills et al., 2017).

Bacteriocins are an heterogenous group with a wide range of sizes, structures, modes of action, activity spectra, and target cell receptors. Their classification undergoes continuous modifications because of new developments regarding their structures and modes of action but they may be broadly divided into class I (post-translationally modified) class II (heat-stable, <10 kDa, unmodified) and class III (thermo-labile, >10 kDa) groups (Álvarez-Sieiro et al., 2016). Along with categorizing bacteriocins based on their modifications, they may be subdivided. The unmodified (class II) bacteriocins can be divided into four groups according to the classification by Álvarez-Sieiro et al. (2016). These subclasses are peptides that contain a YGNGV motif (in which N represents any amino acid; the class IIa peptides); twopeptide bacteriocins (class IIb peptides); leaderless bacteriocins (class IIc peptides); and unmodified, linear, non-pediocin-like, single-peptide bacteriocins that do not belong to other subclasses (class IId peptides).

Since the early 2000s, the successfulness of bacteriocins against unwelcome microorganisms has opened unumbered opportunities for innovative research, and currently, research on these AMPs is rapidly gaining importance. Bacteriocins on their own have shown a great potential mainly in food preservation and biomedical applications, and as a matter of fact, over 60% of the bacteriocin-related patents granted are related to these two areas. However, the use of bacteriocins in other alternative fields is growing, and there are many examples of bacteriocins for veterinary use, cosmetic use and bacteriocins applied on production–purification systems along with recombinant proteins or molecular modifications in the producer strains (López-Cuellar et al., 2016).

Since 2008, pharmaceutical companies have only introduced four new antibiotics, compared to the 16 approved from 1983 to 1987 and, for more than 40 years there have been no releases of new classes of antibiotics to treat Gram-negative bacilli. Meanwhile, the number of new bacteriocins is rapidly increasing and it is believed that more than 99% of bacteria can produce at least one bacteriocin, although most of which are still not identified (Riley and Wertz, 2002). Only from year 2004 to 2015, there have been published 429 scientific articles and 245 granted patents related to bacteriocins and these numbers are increasing to almost one bacteriocin related publication per day (López-Cuellar et al., 2016). In an effort to centralize all this information, there are several bacteriocin sequence data repository websites (http://bactibase.hammamilab.org/main.php, http://bagel4. molgenrug.nl/) regularly updated with the most relevant features of the new bacteriocins described. Moreover, in the last years there has been an increasing number of web servers hosting bioinformatic software acting as bacteriocin genome mining tools allowing the research community to identify bacteriocins and other antimicrobial peptides in genome/metagenome sequences (http://bactibase.hammamilab.org/main.php, http:// bagel4.molgenrug.nl/, https://antismash.secondarymetabolites. org/#!/start). Despite a clear interest in exploring the full potential of these peptides and the enormous amount of information on bacteriocins provided by researcher groups from all over the world, application of their potential remains to be explored. As with other molecules, the more information there is, the more difficult it is to choose the right candidate for a specific application.

We have identified that the availability of a physical collection of bacteriocin genes designed in a standardized format allowing a rapid production of peptides would boost innovation in the AMPs research community. This new resource should also allow for the first time to test and compare different bacteriocins, make combinations to analyse and characterize potential synergistic effects between unrelated bacteriocins as well as compare biochemical properties. Academic and industrial researchers would be able then to choose and select the right bacteriocin from this collection, tailored for each specific application.

Bacteriocin production and purification is not an easy task and is considered as one of the main bottlenecks for their use at industrial scale (Deraz et al., 2007; Jozala et al., 2008, 2015; Bali et al., 2014). Bacteriocins traditionally have been purified from their native producing strains, however this consists in a time-consuming and laborious procedure which often results in low yields of bacteriocin (Rodríguez et al., 2003). On top of that, collecting all bacteriocin producing strains described and purifying their active peptides is an almost impossible task. Other methods such as the heterologous production of bacteriocins in other microorganisms (bacteria or yeast) and in plants has been shown as a useful technology for producing higher concentrations of these peptides (Sánchez et al., 2008; Borrero et al., 2011, 2018). Nevertheless, only a few bacteriocins have been successfully produced by heterologous hosts since this approach is laborious and time consuming. Moreover, the risk that the bacteriocin will be produced at a yield not reaching the standard request of industrial processes is still quite high. There are alternatives to the use of microorganisms as peptide producers and with the recent advances in the field of Synthetic Biology, it is now possible to synthesize peptides both by chemical synthesis and by using cell-free extracts for ex-vivo protein production (Hols et al., 2019). Since its first application in deciphering genetic codes (Nirenberg and Matthaei, 1961) cellfree protein synthesis (CFPS) has become a key method for the obtention of recombinant proteins in order to match the rising demands for simple, inexpensive, and efficient protein production (Carlson et al., 2012; Hodgman and Jewett, 2012). The recent growing attention to crude extract CFPS is motivated mainly, but not only, by the unique benefit that barrier-free cell-free systems enable direct and fast access to enzymes and reaction conditions.

The peptidic nature of bacteriocins and the fact that they are ribosomally synthesized (Field et al., 2012) make them ideal candidates for production by CFPS. With the cell-free system there is no need of signal peptides or leader sequences attached to the bacteriocins and of dedicated transporter systems to secrete the mature and active peptide, as the bacteriocin is synthesized outside the cell. Moreover, with this technology the potential toxicity to the host is also excluded and there is no need to coproduce the immunity protein. This major technological locket could be overcome in most cases by working with cell-free extracts which is by design an abiotic technology. With these advantages in mind we have constructed a physical collection of bacteriocins: PARAGEN 1.0: a standardized synthetic gene library allowing to produce a broad set of bacteriocins in parallel.

A schematic description of the methodology used is shown in **Figure 1**. A list of class II bacteriocins was obtained from the two major genomic bacteriocin databases: BAGEL4 (van Heel et al., 2018) and BACTIBASE (Hammami et al., 2010). The aminoacidic sequence of the mature/active peptide of each bacteriocin was determined based on the information provided by these databases and, in some cases, going to the original publications and/or sequences found at the NCBI database (Database resources of the National Center for Biotechnology Information, 2015). All aminoacidic sequences were reverse-translated and codon optimized for Escherichia coli (www.bioinformatics.org/ sms2/rev\_trans.html). The nucleotide sequences were included in a vector backbone containing the T7 promoter region, a start codon (ATG) a stop codon (TAA) and a T7 terminator region. Recombinant vectors were constructed and amplified in E. coli DH10B standard strain and used as templates for cell-free protein synthesis using PURExpress <sup>R</sup> in vitro Protein Synthesis Kit (New England Biolabs). All bacteriocins synthesized were tested on petri dishes for antimicrobial activity against a set of 12 different indicators including Gram-positive and Gram-negative bacteria (**Table 1**). Bacteriocins showing a halo of inhibition against at least one of the indicators were considered as positive **(Supplementary Material)**. A total of 128 bacteriocins were synthesized, 23 belonging to the class IIa (18%), 15 belonging to the class IIb (12%), 15 belonging to the class IIc (12%), 50 belonging to the class IId (39%), and 25 belonging to the class III (20%) (**Figure 2**). Regarding their antimicrobial activity, 74% of the class IIa, 67% of the class IIb, 100% of the class IIc, 38% of the class IId and 32% of the class III bacteriocins showed activity against at least one of the indicators tested. The remaining bacteriocins (59 out of 128) did not show activity against any of the indicators tested **(Supplementary Material)**. However, we cannot conclude that these 59 bacteriocins are inactive as, most likely, they are just not active against the 12 indicators used in this work. Hence, a further analysis with a larger number of targets should be done. Other tests such as protease and pH stability of these peptides are currently being done in the lab.

TABLE 1 | Strains used as indicators to test the antimicrobial activity of the bacteriocins from PARAGEN 1.0.


As shown in this work, synthetic production of bacteriocins with cell-free systems is a viable alternative to bacterial production as it allows very rapidly the synthesis and screening of a large numbers of bacteriocins in the same experiment. Further efforts need to be done in order to reduce the cost of this system, optimize and increase peptide production. There is also a need on finding a method to quantify the production of active bacteriocin in order to fully compare antimicrobial activities. Several authors have already used cell-free as platform for the production for both single proteins but also for proteins produced through complex metabolic pathways (Koch et al., 2018). Some class I bacteriocins such as microcin J25 have been modified and produced in their active conformation ex-vivo (Yan et al., 2012), therefore we strongly believe that the production of class I bacteriocins should be possible soon with this technology.

To our knowledge, PARAGEN 1.0 is the largest collection of active native bacteriocins ready to be tested and it keeps growing since we are constantly adding new peptides to it. We are already working in a second library of non-natural bacteriocins (PARAGEN 2.0) consisting on chimeras of different bacteriocins or variants with single or multiple amino acid modifications. The possibilities that cell-free systems offer for bacteriocin production are almost unlimited and we believe that PARAGEN might be a useful tool for gaining knowledge of the precise modes of action of bacteriocins and more generally antimicrobial peptides, allowing the production of more active and more stable variants.

#### REFERENCES

Álvarez-Sieiro, P., Montalbán-López, M., Mu, D., and Kuipers, O. P. (2016). Bacteriocins of lactic acid bacteria: extending the family. This would contribute to speed up the screening for natural antimicrobial activities and the design of tailor-made bacteriocins for their applications in food industries and in medicine.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript/**Supplementary Files**.

#### AUTHOR CONTRIBUTIONS

PG and JB have contributed equally to the design and implementation of the research, to the analysis of the results and to the writing of the manuscript.

#### FUNDING

The construction of the PARAGEN 1.0 collection was funded by a Research Grant from the Waloon Region, Recherche Industrielle: PARAGEN convention n◦ 7627.

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00213/full#supplementary-material

Appl. Microbiol. Biotechnol, 100, 2939–2951. doi: 10.1007/s00253-016- 7343-9

Bali, V., Panesar, P. S., and Bera, M. B. (2014). Trends in utilization of agroindustrial byproducts for production of bacteriocins and their biopreservative applications. Crit. Rev. Biotechnol. 8551, 1–11. doi: 10.3109/07388551.2014.947916


Biotechnol. Biotechnol. Equip. 30, 1039–1050. doi: 10.1080/13102818.2016. 1232605


**Conflict of Interest Statement:** PG and JB were employed by company Syngulon S.A.

Copyright © 2019 Gabant and Borrero. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Future Trends in Synthetic Biology—A Report

Meriem El Karoui <sup>1</sup> \*, Monica Hoyos-Flight <sup>2</sup> and Liz Fletcher <sup>1</sup>

<sup>1</sup> SynthSys-Centre for Synthetic and Systems Biology, School of Biological Sciences, University of Edinburgh, Edinburgh, United Kingdom, <sup>2</sup> Innogen Institute, School of Social and Political Sciences, University of Edinburgh, Edinburgh, United Kingdom

Leading researchers working on synthetic biology and its applications gathered at the University of Edinburgh in May 2018 to discuss the latest challenges and opportunities in the field. In addition to the potential socio-economic benefits of synthetic biology, they also examined the ethics and security risks arising from the development of these technologies. Speakers from industry, academia and not-for-profit organizations presented their vision for the future of the field and provided guidance to funding and regulatory bodies to ensure that synthetic biology research is carried out responsibly and can realize its full potential. This report aims to capture the collective views and recommendations that emerged from the discussions that took place. The meeting was held under the Chatham House Rule (i.e., a private invite-only meeting where comments can be freely used but not attributed) to promote open discussion; the findings and quotes included in the report are therefore not attributed to individuals. The goal of the meeting was to identify research priorities and bottlenecks. It also provided the opportunity to discuss how best to manage risk and earn public acceptance of this emerging and disruptive technology.

Keywords: synthetic biology, biosystem, future trends and developments, biodesign automation, responsible research and innovation (RRI)

# INTRODUCTION

Synthetic Biology offers innovative approaches for engineering new biological systems or redesigning existing ones for useful purposes (see **Figure 1**). It has been described as a disruptive technology at the heart of the so-called Bioeconomy, capable of delivering new solutions to global healthcare, agriculture, manufacturing, and environmental challenges (Cameron et al., 2014; Bueso and Tangney, 2017; French, 2019). However, despite successes in the production of some high value chemicals and drugs, there is a perception that synthetic biology is still not yet delivering on its promise.

Moreover, there are some concerns from governments that synthetic biology expands the pool of agents of concern, which increases the need to develop detection, identification and monitoring systems, and proactively build countermeasures against chemical and biological threats (Wang and Zhang, 2019). The participation of representatives from various government organizations

#### Edited by:

Jean Marie François, UMR5504 Laboratoire d'Ingénierie des Systèmes Biologiques et des Procédés (LISBP), France

#### Reviewed by:

Vijai Singh, Indrashil University, India Chris John Myers, The University of Utah, United States

> \*Correspondence: Meriem El Karoui Meriem.Elkaroui@ed.ac.uk

#### Specialty section:

This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology

Received: 28 May 2019 Accepted: 08 July 2019 Published: 07 August 2019

#### Citation:

El Karoui M, Hoyos-Flight M and Fletcher L (2019) Future Trends in Synthetic Biology—A Report. Front. Bioeng. Biotechnol. 7:175. doi: 10.3389/fbioe.2019.00175

at this meeting is testament to their commitment to maintaining an active dialogue with the synthetic biology community. In this way, they aim to keep abreast of the changing nature of threats and provide the best advice to government about investment in science and technology and the introduction or amendment of regulatory processes.

The cost of DNA sequencing and synthesis have decreased dramatically (Carlson, 2014; Kosuri and Church, 2014) and we have access to more genetic information and more powerful genetic engineering capabilities than ever before. Critical investments in infrastructure are bearing fruit and, as is described below, synthetic biology is increasingly becoming, at least part of, the solution to many of our present and future needs in medicine, food and energy production, remediation, manufacturing, and national security. So what is the potential of synthetic biology and what challenges does it still face to realize this?

## ADVANCES IN SYNTHETIC BIOLOGY: THE STATE OF PLAY

#### Small Molecules: Production on Demand a Reality

Despite the lack of predictability in biology, and current technical constraints that limit data collection and analyses, we can now produce small molecules on demand using synthetic biology approaches.

Probably the most impressive examples come from the Foundry at the Broad Institute of MIT and Harvard. When the Defense Advanced Research Projects Agency (DARPA) put the MIT-Broad Institute Foundry's design capabilities to the test, its researchers were able to deliver 6 out of 10 molecules of interest to the US Department of Defense in 90 days. This "pressure test" confirms the potential of synthetic biology to address shortages of key compounds quickly (Casini et al., 2018).

Indeed, many labs can now design and construct relatively complex gene networks capable of generating a wide variety of "designer" molecules in a range of host cells; however, this is often a slow iterative process of trial and error.

As yet, very few small molecules in medicine are manufactured using a synthetic biology process; it remains very difficult to engineer microbes to carry out processes that Nature did not intend. This is to be expected: the performance of microbes is "good enough" from an evolutionary perspective. Microbes evolved to address the specific needs and challenges of their natural environments not those of industrial fermenters and bioreactors. Gene Transfer from one system to another may sound easy but in practice is hard work and rarely generates sufficient reward (i.e., increased yield) to justify the investment made. The application of automation and artificial intelligence (e.g., in designing and building plasmids) may help to reduce the time and cost—and improve return on investment—in the future (Zhang et al., 2018).

#### "Scale up is product specific – we need more synthetic biology in the production process"

Plants make alternative production platforms. Improvements in mining plant genomes and the development of effective transient expression systems have enabled large-scale production of, for example, vaccines in tobacco plants in just a few weeks (Dirisala et al., 2017; Emmanuel et al., 2018). Directing the production of synthetic biological materials to plant chloroplasts also shows promise (Boehm and Bock, 2019).

The photosynthetic reducing power generated in plant chloroplasts can be harnessed for the light-driven synthesis of bioactive molecules such as dhurrin, which protects plants against insects (Gnanasekaran et al., 2016).

However, underlying all these platforms is a knowledge gap in our understanding in how nature works. This makes it very hard to apply the design/build/test/learn cycles used in conventional engineering to the production of synthetic biological materials whatever the production platform (yeast, bacteria, plants, or human cells) if the platform itself is not well-understood (Sauro et al., 2006).

What we need now are instruments able to measure and characterize outputs, assisted by progress in robotics and automation, and the application of machine learning approaches to analyse the data generated. This will help us to generate more robust models of biological systems, so we can improve experimental design for future engineering strategies.

"We can do 'build'. 'Test' is the challenge when we want to learn from the iterative design process"

#### Healthcare: Reimagining Medicine

Synthetic biology is driving significant advances in biomedicine, which will lead to transformational improvements in healthcare. Already, patients are benefiting from so-called CAR (for chimeric antigen receptor) technology, which engineers the immune cells (T-cells) of the patient to recognize and attack cancer cells (June et al., 2018).

Genetically engineered viruses are being used to correct defective genes in patients with inherited diseases such as Severe Combined Immune Deficiency (SCID) or epidermolysis bullosa (Dunbar et al., 2018).

The ability to reprogramme somatic cells from patients into induced pluripotent stem cells is furthering our understanding of their disease, reducing the use of animals in research, and paving the way for the development of personalized medicines and cell therapies. In principle at least, we could engineer a patient's own cells to multiply, differentiate into different cell types and even self-assemble into new tissues, or even organs, to repair those damaged through disease or injury (Davies and Cachat, 2016; Satoshi et al., 2018).

Work on new vectors that are able to deliver large genetic loads to target tissues is helping to produce more efficient therapeutics and vaccines that will have fewer side effects and a smaller risk of resistance. Furthermore, optimizing antibody or vaccine production, or example, so that they are in an edible format (e.g., plant based), could greatly reduce the cost and increase the speed of vaccine production in an epidemic.

"We have the tools but need the creativity to make stuff that can't be made without synthetic biology"

In the next few years, genetically engineering pigs to be virus resistant and have human-like immune profiles could make xenotransplantation a clinical reality (Burkard et al., 2018). Engineering the microbiome is expected to lead to the development of synthetic probiotics (Dou and Bennett, 2018).

The synthetic biology initiative known as Human Genome Project-write (HGP-write) has set its sights even higher, rallying scientists to build entire human chromosomes (Boeke et al., 2016). Concerns have been raised about the ethics of creating "synthetic humans" and indeed the scientific and commercial value of such a project. More recently, HGP-write champions have proposed a more focused project to build a virus-resistant chromosome, making at least 400,000 changes to the human genome to remove DNA sequences that viruses use to hijack cells and replicate (Dolgin, 2018).

One of the many exciting opportunities that synthetic biology offers medicine is in the production of theranostic cell lines that can sense a disease state and produce an appropriate therapeutic response (Teixeira and Fussenegger, 2019). Several obstacles need to be overcome to achieve this goal: first, to expand the range of molecules that can be recognized by cellular "sensors" as inputs; and second, to better understand the genetic control factors that regulate gene expression in space and time so we can engineer better activator systems.

#### "At present we need pills because we can't swallow a chemistry kit"

Metabolomics is shedding light on many disease biomarkers. Because some biomarkers are shared between seemingly unrelated diseases, an accurate diagnosis will require the detection of multiple markers to provide a more unique "disease fingerprint." Work in whole cell and cell free systems to develop sensors of multiple disease-specific biomarkers could assist in earlier detection of disease and prognostic monitoring.

To expand the range of biologically detectable molecules, it is possible to design metabolic pathways that transform currently undetectable molecules of interest (e.g., hippuric acid, the prostate cancer biomarker) into molecules for which sensors already exist (in this case benzoic acid) (Libis et al., 2016).

Cybergenetics is an emerging field that is developing experimental tools for the computer control of cellular processes at the gene level in real time. Cybergenetic control can be achieved by interfacing living cells with a digital computer that switches on or off the embedded "genetic switch" using light (optogenetics) or chemicals (Gabriele et al., 2018; Maysam et al., 2019). Such systems could help to maintain cellular homeostasis by monitoring the state of the body and triggering an appropriate response upon the detection of dysregulation; for example, they could trigger the release of insulin when blood glucose levels rise as detected by a wireless diagnostic tool (Ye et al., 2011).

## Advanced Materials: Inspired by Nature, Improved by Synthetic Biology

Synthetic biology offers the opportunity to create responsive and multifunctional materials (Le Feuvre and Scrutton, 2018). The integration of biochemical components from living systems with inorganic components can lead to new materials that are able to sense the environment (or internal signals) and change their properties. These features could be particularly useful for improving protective clothing or building materials.

An issue when using microbes to produce composite materials is regulating the assembly of these materials to achieve specific desired properties. By understanding how microbes communicate with each other, it is possible to make them work better together and combine them with other production systems so that the properties of materials can be tailored for particular functions.

Interestingly, rather than modifying, or improving existing protein-based materials, an alternative approach involves using computational techniques to design completely novel proteins that self-assemble into predicted shapes (Ljubetic et al., ˇ 2017). Such "programmable" proteins open up even further opportunities for synthetic biology not only for materials science but also for medicine and chemistry.

# TACKLING THE CHALLENGES

Regardless of the research areas involved, there are some common challenges for the community to address.

## Design With the End in Mind

There was consensus that while the production of small molecules using synthetically engineered cells at the bench is becoming more tractable, these processes often do not translate well into mass production. Scalability needs to be incorporated into the initial design process by including features, for example, that reduce toxicity of the molecule to the production host or "chassis" and/or by introducing modifications that favor its extraction. Careful consideration of the right chassis could also greatly improve yields and significantly reduce cost of production.

"Nature has developed its own ways to concentrate and solubilise chemicals... we are not learning from this"

# Expanding the Host Repertoire

The number of microbes that are currently in use for the production of synthetic biological materials is only a tiny fraction of the total diversity that exists in Nature; only ten microbes are "domesticated" for industrial use. To identify the most appropriate chassis, it is worth turning to nature to identify species that have unique metabolic networks suited to host particular types of chemical reactions. For example, the soil bacterium Pseudomonas putida, which has adapted to harsh environmental conditions, is ideally suited to host redoxintensive reactions (Pablo and de Lorenzo, 2018). A treasure trove of, as yet, unexplored natural products, with novel and beneficial properties, exist in the plant kingdom. In addition, as noted above, plants are naturally excellent production hosts and it is now possible to "plug and play" combinations of plant pathways to generate novel molecules (Sainsbury and Lomonossoff, 2014; Evangelos and O'connor, 2016).

#### Developing a Universal Production System

To circumvent the problem of the impact of different host chassis on synthetic gene circuits, researchers would benefit from a universal synthetic expression system that permits the testing of new constructs. This would aid in the identification of optimal production platforms and decrease the need for organism-specific technologies. Combining this technology with DNA editing techniques, such as CRISPR/Cas9, will make the establishment of host production platforms and the generation of complex biosynthetic products easier and faster. One articulation of this could be in cell-free formats, in which the essential cellular machinery is reconstituted in vitro and used as a manufacturing platform (Villarreal and Tan, 2017; Koch et al., 2018).

"We need to reduce the burden of synthetic networks on endogenous circuits"

<sup>&</sup>quot;The tools are there, we just don't know what we want to make"

#### Move to Cell-Free Environments

Cell-free environments, interfaced with semiconductors, offer a powerful route for flexible and controllable production systems.

For example, nanoparticles made of semiconductor materials or quantum dots can be used to enhance enzyme activity in a cell free environment with a minimal set of ingredients. Multistep enzymatic pathways can be tethered to nanoparticle surfaces and, by avoiding the diffusion effect that takes place in cells, reaction rates can be increased 100-fold (Wang et al., 2017). This approach can be used to access nonnatural materials and circumvents the potential issue of toxicity in cells (as well as the regulatory issues of genetically modified cells).

Research on silicon chips containing immobilized genes and cell lysates allows detailed examination of gene expression in space and time. When the compartments in which DNA-driven reactions take place are linked, with materials flowing into and diffusing between the compartments, it is possible to recreate oscillating protein expression patterns and protein gradients akin to those observed in cells (Karzbrun et al., 2014). Gene expression in these "artificial cells" can be controlled with electrodes that prevent the assembly of proteins by ribosomes. By standardizing outputs, this strategy is improving the predictability of engineered genetic circuits.

## Systems Modeling, Standards, and Metrology

Whatever the system being explored, a robust model of that biological system is needed if predictable modifications of genetic circuits are to be designed and implemented with any confidence. To create such models, large sets of data arising from measuring multiple parameters of cell behavior under different conditions are essential (Fletcher et al., 2016).

There is great expectation that improvements in highthroughput data measurement and collection systems will generate exactly the large data sets needed. These can be analyzed using artificial intelligence or a machine learning approach to optimize the design of synthetic biological products and move away from the inefficient trial and error process (Decoene et al., 2018).

Finally, agreement on standards of design, assembly, data transfer, data measurement and regulatory rules, as well as on the language that is used, will help to improve the interdisciplinary and international collaborations that are required to drive the field forward. This is challenging for a community with such diverse interests and perspectives, and where data sharing, curating and quality control is not common practice.

However, without some form of agreed standards, many of the products and processes of synthetic biology will not translate well to industrial settings dependent on reproducible processes and beholden to exacting regulatory requirements.

Typically, academic researchers are driven by a need to understand the complexity of nature (and publish their work in high impact journals). Standardization and scaling up production are important, but have less of an academic pull than discovering a new product. Access to funding, industrial partnerships, and academic recognition are examples of potential incentives for carrying out this type of research.

"Standards restrict flexibility but enable interoperability"

# TACKLING RISK

Synthetic biology is an example of a dual-use technology: it promises numerous beneficial applications, but it can also cause harm. This has led to fears that it could, intentionally or unintentionally, harm humans or damage the environment. For example, there is huge value in our ability to engineer viruses to be more effective and specific shuttles for gene therapies of devastating inherited disorders; however, engineering viruses may also lead to the creation of even more deadly pathogens by those intent on harm.

"Synthetic biology should be regarded as an extension of earlier developments and technologies"

Some would argue that synthetic biology poses an existential risk and needs to be treated with extreme caution. However, many new technological advances across the decades have met similar concerns. The uncertainty and remote possibility of such risks could hamper the development of useful technology. Scientists, their host institutions and funding bodies should (and indeed already do) consider whether the research planned could be misused. Measures that reduce the likelihood of misuse and its consequences should be implemented and clearly communicated. The synthetic biology community needs to be aware of, and respond to, these challenges by engaging in horizon scanning exercises as well as open dialogue with regulatory bodies and the media.

"Don't avoid risk – manage it"

Being more open about risks, and how they are controlled, provides an opportunity to shift discourse toward the benefits of synthetic biology in addressing urgent global needs, such as the production of biofuels, food security and more effective medicines, and potentially improve public acceptance.

"The questions should not be 'what's the next big thing for synthetic biology' but 'where is the greatest unmet need'."

Despite the efforts by individual countries to establish synthetic biology research roadmaps, broader, international agreement on common standards (and red lines) across the field may help establish trust and to advance the best pre-competitive research into useful applications.

Meeting participants highlighted the importance of training in responsible research conduct and ethics. Given students' future role as science ambassadors and influencers, their training should not only convey skills and knowledge but also awareness and critical thinking about the prospects and potential for dual use of synthetic biology. All researchers must remain vigilant regardless of the many pressures and distractions of running a successful research lab; they may not have specialist training in identifying the risks of misuse but they are the people best placed to maintain informed oversight of risks.

One example of current synthetic biology research with potential dual use is gene drive technology, which can be used to propagate a particular suite of genes throughout a population. The benefits of using gene drive technology include the eradication of disease-carrying insect populations and the elimination of invading pest species but it has raised concerns about the unintended ecological impacts of reducing or eliminating a population (Callaway, 2018; Collins, 2018).

Similar release concerns surround research that is harnessing the ability of pathogens to target particular tissues in the body or particular chemicals in the environment, which could greatly aid efforts to deliver targeted therapies or clean-up contaminated sites. To date, such large-scale release for environmental bioremediation interventions has not been possible.

"We need to mind the gap between R&D scale up and communications . . . . One bad blog can kill a commercial product"

There was consensus that the need for regulation over this community remains important. Regulation needs to keep up to speed with the emerging technologies and should focus on the product rather than the process used to create it (Tait et al., 2017). Unsuitable regulatory frameworks (as well as unfavorable public perception) could discourage private sector investment in synthetic biology.

Open and balanced two-way communication between researchers, funders, companies and governments and the public will be vital. Consumers and activists may have no interest in the difference between making a chemical (e.g., a flavor or fragrance) synthetically and one made using genetically modified bacteria but they may instinctively distrust the latter.

## CONCLUSIONS AND RECOMMENDATIONS

The short talks presented at this 2-day meeting suggest that synthetic biology is at the cusp of many major breakthroughs and that it is perhaps timely to re-define the meaning of "success" in synthetic biology.

There are many hurdles to overcome but the potential for synthetic biology to deliver solutions to many global challenges improving healthcare, limiting environmental damage, and creating a wide variety of more sustainable processes—is great.

Meeting participants suggested that as a community they should support the measures listed below to help synthetic biology move beyond the proof-of-concept stage and to ensure that potential risks are minimized and dialog with the public can be optimized.

• Larger and longer investment in better big data management and processing (Artificial Intelligence and Machine Learning) systems, and in fundamental research on biosystems modeling, chassis development, and genome mining.


The workshop highlighted just how much more we have to learn from Nature itself. Synthetic biology is giving us insights of the criteria and processes that underpin all living systems; in turn, we can take this insight and design and use it to build a "better biology." However, we need to take the public with us on this journey, create meaningful and considered dialogue about the work we may do and the impact it might have on our world.

# AUTHOR'S NOTE

The meeting was hosted by the UK Center for Mammalian Synthetic Biology (the Center), based at the University of Edinburgh. The Center is building expertise in cell engineering tool generation, whole-cell modeling, computer-assisted design and construction of DNA and high-throughput phenotyping to enable synthetic biology for medicine and healthcare. The Centre's research will not only advance basic understanding of mammalian biology and pathology but also generate products and services for near-term commercial exploitation by the pharmaceutical and drug testing industries, such as diagnostics, novel therapeutics, protein-based drugs, and regenerative medicines. The Center is funded through the Research Council's UK "Synthetic Biology for Growth" programme and by the Biological and Biotechnology Sciences Research Council (BBSRC), the Engineering and Physical Sciences Research Council and the Medical Research Council.

# AUTHOR CONTRIBUTIONS

This is a meeting report and was drafted by MH-F with support from ME and LF.

#### ACKNOWLEDGMENTS

We are grateful to all colleagues that contributed to discussions at the workshop held in Edinburgh in May 2018 and whose

#### REFERENCES


views helped inform this report. We also gratefully acknowledge sponsorship for this workshop from the Office of Naval Research Global and the UK Centre for Mammalian Synthetic Biology (BBSRC Grant BB/M018040/1).


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 El Karoui, Hoyos-Flight and Fletcher. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# ParAlleL: A Novel Population-Based Approach to Biological Logic Gates

#### Felipe A. Millacura\*, Brendan Largey and Christopher E. French\*

*School of Biological Sciences, Institute of Quantitative Biology, Biochemistry and Biotechnology, University of Edinburgh, Edinburgh, United Kingdom*

*In vivo* logic gates have proven difficult to combine into larger devices. Our cell-based logic system, ParAlleL, decomposes a large circuit into a collection of small subcircuits working in parallel, each subcircuit responding to a different combination of inputs. A final global output is then generated by a combination of the responses. Using ParAlleL, for the first time a completely functional 3-bit full adder and full subtractor were generated using *Escherichia coli* cells, as well as a calculator-style display that shows a numeric result, from 0 to 7, when the proper 3 bit binary inputs are introduced into the system. ParAlleL demonstrates the use of a parallel approach for the design of cell-based logic gates that facilitates the generation and analysis of complex processes, without the need for complex genetic engineering.

#### Edited by:

*Jussi Jantti, VTT Technical Research Centre of Finland Ltd, Finland*

#### Reviewed by:

*Thomas Dandekar, University of Wuerzburg, Germany Chris John Myers, The University of Utah, United States*

\*Correspondence:

*Felipe A. Millacura S1647595@sms.ed.ac.uk Christopher E. French c.french@ed.ac.uk*

#### Specialty section:

*This article was submitted to Synthetic Biology, a section of the journal Frontiers in Bioengineering and Biotechnology*

Received: *17 December 2018* Accepted: *27 February 2019* Published: *21 March 2019*

#### Citation:

*Millacura FA, Largey B and French CE (2019) ParAlleL: A Novel Population-Based Approach to Biological Logic Gates. Front. Bioeng. Biotechnol. 7:46. doi: 10.3389/fbioe.2019.00046* Keywords: Escherichia coli, 3-bits, full adder, full subtractor, calculator-like display, parallel approach

# INTRODUCTION

A major challenge in the field of synthetic biology is the construction of complex logic circuits that analyze variables as in electronics; where a single circuit accepts one or more binary inputs to generate one or more binary outputs. A cell-based logic network consists of engineered cells producing an output macromolecule only if the corresponding pattern of inputs is present. The mechanism of analysis is commonly based on the use of transcriptional regulators, transcription factors, polymerases, receptors, or recombinases (Brenner et al., 2018). Some examples of genetic circuits mimicking computational behavior are toggle switches, oscillators, boolean logic gates, feedback controllers, and multiplexers. Although there are genetic circuits that simulate computational behavior, the complex engineering of their biological chassis is affected by gene expression noise, mutation, cell death, undefined, and changing extracellular environments and improper interactions with the cellular context (Andrianantoandro et al., 2006). Furthermore, complex genetic engineering is necessary when multiple input variables are analyzed, limiting the processing capacity of the system.

Biological multiplexers analyze one or more signals over a common transmission line using interconnected transcription factors, recombinases, antisense RNA, or CRISPR-like technology (Nielsen and Voigt, 2014; Roquet et al., 2016; Brenner et al., 2018). However, complex genetic engineering is needed for wiring the basic computational units, becoming inefficient for moving beyond simple NOT or AND logic gates or for scaling to 3 bit logic circuits. The complexity of the genetic engineering required can be reduced by using distributed logic circuits, where the computation is distributed among several physically separated cellular consortia that each sense only one signal and respond by secreting a communication molecule (Regot et al., 2011). As a circuit responds to one signal, but not another, due to spatial distribution, a change in the state of the system can be triggered as response, making synthetic learning possible (Macia et al., 2017; Shipman et al., 2017). Even though the consortium approach makes Boolean circuit design simpler, it still shows a slow response and considerable complexity since each cell needs to recognize, synthesize and secrete a wiring molecule (Macia et al., 2016).

Here we propose an alternative logic architecture, which decomposes a large circuit into a collection of small subcircuits acting in parallel (hereafter ParAlleL). Rather than having a single type of agent (such as a genetically engineered cell) doing the computation, ParAlleL has separate types of agent that each react to a different combination of inputs. A final output is then generated by combination of the responses, making all kinds of binary operation possible. As an example, here we show the implementation of this concept using cells resistant to different combinations of antibiotics, with the response indicated by growth. This is used to demonstrate a completely functional 3 bit full adder and full subtractor, as well as a calculator-style display that shows digits from 0 to 7 based on three binary input bits.

## METHODOLOGY

#### Reagents and Stock Solution Preparations

Antibiotic stock solutions were prepared as follows: 100 mg/ml carbenicilin disodium salt (Sigma-Aldrich #C1389), 50 mg/ml kanamycin sulfate (PanReac Applichem #A1493), 20 mg/ml chloramphenicol (Acros Organics #22792), 10 mg/ml tetracycline hydrochloryde (Duchefa Biochemie #T0150), 10 mg/ml gentamicin sulfate (Melford #G0124), and 50 mg/ml spectinomycin.HCl (LKT Labs #S6018). Developing solution contained 0.1 %w/v bromothymol blue (Sigma-Aldrich #114421) and 400 mM Trizma base pH7.5 (Sigma-Aldrich #T1503).

## Generation of Subcircuit Cells

E. coli JM109 was transformed with 200–300 pg of plasmid pSB4A5 (AmpR) or pSB4C5 (ChlR) (Registry of Standard Biological Parts) and selected on 100µg/ml carbenicilin (Am) or 20µg/ml chloramphenicol (Ch), respectively. Cells carrying the first bit plasmid were made chemically competent (Chung et al., 1989) and transformed with 200–300 pg of the 2nd bit plasmid, pSB1T3 (TetR) or pSB1K3 (KanR) (Registry of Standard Biological Parts). Selection was performed with the first antibiotic (Am or Ch) and the addition of 10 µg/ml Tetracycline (Tc) or 50µg/ml kanamycin (Km), obtaining the two-bit combinations Km/Am (KA), Tc/Am (TA), Km/Ch (KC), and Tc/Ch (TC). This set of strains is sufficient to implement all two-bit binary operations.

The third bit layer was generated by transforming these four strains with pSEVA631 (GenR) (Silva-Rocha et al., 2012; GenBank JX560348) or pMO9075 (SpmR) (Keller et al., 2011). Resulting strains were selected on the 2-bit antibiotic combinations plus 10µg/ml gentamicin (Gm) or 50µg/ml spectinomycin (Sm). This gave 8 strains, designated ATG (Am/Tc/Gm), AKG (Am/Km/Gm), ATS (Am/Tc/Sm), AKS (Am/Km/Sm), CTG (Ch/Tc/Gm), CTS (Ch/Tc/Sm), CKG (Ch/Km/Gm), CKS (Ch/Km/Sm) based on their resistance markers. This set of strains is sufficient to implement all three-bit binary operations. Plasmid specifications are listed in **Figure S2**, **Tables S1, S3**, with further information about these antibiotics in **Table S2**. Plasmid sequences are available in different formats at https://doi.org/10.7488/ds/2497.

#### Three-Bit Logic Operations

Tests were performed in 96-well microplates by inoculating cells (1:100) in LB broth (100 µL) supplemented with 1%w/v <sup>D</sup>(+) glucose (Fisher Chemical #G0500). Plates were incubated for 18 h at 37◦C without shaking and then developed by addition of the developing solution (0.1%w/v bromothymol blue in 400 mM

specified 1 bit, 2 bit, or 3 bit inputs (top of each column). Cells were inoculated (1:100) in LB supplemented with 1% w/v glucose. After 18 h of incubation at 37◦C, plates were developed by addition of 0.05 volumes of the developing solution.

Tris, pH7.5) in a ratio 1:20. Images were obtained using a Kodak ESPC315 Flatbed scanner. Design of the calculator-like display, full adder, and subtractor are shown in **Figures 2**, **3** and in Supplementary material (**Figure S3**). Raw figures were deposited at https://doi.org/10.7488/ds/2497.

#### RESULTS

In the distributed logic system of ParAlleL each input bit has two forms, ZERO and ONE, each of which is essential to certain output agents and inhibitory to others. Thus each agent reacts only to a certain combination of input bits, allowing generation of any arbitrary pattern of outputs for any pattern of inputs. In the implementation shown here, each input bit comes in two forms, each being an antibiotic lethal to sensitive strains. In this case, bit A is represented by ampicillin for zero, chloramphenicol for one, bit B by kanamycin for zero, tetracycline for one, and bit C by gentamicin for zero, spectinomycin for one. Thus, four strains are needed to implement any operation with two input bits, and eight strains for three input bits. In contrast to other cellbased logic schemes, only very minimal genetic engineering is required, essentially transformation with 3 different antibiotic resistance markers.

Cells show a global response concordant with the behavior expected for a 1 bit, 2 bit, or 3 bit system (**Figure 1**). For instance, when the input 101 (chloramphenicol, tetracycline and spectinomycin) is added to the system growth is only observed in the corresponding CTS cells, which carry the proper resistance markers. The response time of the system is around 12 h (**Figure S1**) but plates were developed at 18 h to avoid false negatives or positives.

In order to further test the ParAlleL system, a digital calculator-like display was designed (**Figure 2A**). In this case, multiple subcircuit cells are mixed in one well and the global response displays a number from 0 to 7 when the proper binary input is applied. Numbers represent the total eight possible values encoded within 3-bit binary inputs.

Input configuration versatility was proven by representing bit A in this case by gentamicin for zero, spectinomycin for one, bit B by kanamycin for zero, tetracycline for one, and bit C by ampicillin for zero, chloramphenicol for one. For instance, once input 110 represented by Ch/Km/Gm is added in the system, the number 6 is displayed (**Figure 2**).

Finally, a full adder and a full subtractor were designed. A full-adder adds three binary inputs, often denoted as A, B, and Cin, generating a Sum result (S) and a Carry-out (Cout). A full subtractor, on the other hand, has a minuend (X), a subtrahend (Y) and an additional Borrow-in (Bin) as inputs. The subtraction operation produces a difference (D) and a Borrow (Bout) (**Figure 3**).

In order to generate the full adder and subtractor, multiple subcircuit cells were mixed and distributed in two different wells (**Figure S3**). One well represents the solution (S) or difference (D) and a second one the carry (Cout) or borrow (Bout), for the adder and subtractor, respectively (**Figure 3**). A yellow color represents growth and a positive output 1, a blue color represents no growth and a binary 0 output instead.

#### CONCLUSIONS/DISCUSSION

Subcircuits that solve complex calculations in parallel have been extensively used for computation in order to reduce the total computation time. Translating this approach to biological systems would allow us to analyze complex processes, currently difficult in synthetic biology, as multiple simple sub-circuits.

In our proof of concept, we present a biological information processing system, ParAlleL, capable of exploiting the parallelism in mixed bacterial cultures. ParAlleL decomposes the analysis of 2 and 3 bit complex inputs, into 4 and 8 sub circuits, respectively (**Figure 1**). Each sub-circuit corresponds to a different E. coli strain carrying a different combination of antibiotic resistance markers (**Table S1**). As an example, in the 3 bit system the input 000 is represented by the antibiotics ampicillin, tetracycline and gentamicin (**Figure 1**). When this input is entered into the system, all cells that are not encoded for responding to 000 will die, but cells carrying the proper plasmid combination, pSB4A5, pSB1T3, and pSEVA621 will not (**Figures 1**, **2**), therefore, a live/dead response (output) is achieved in all sub circuits, the output of each well being one (growth) or zero (failure to grow) (**Figure 2** and **Figure S2**).

ParAlleL uses cellular consortia instead of a single type of cell. A similar approach has been developed by Macia et al. (2016, 2017) using eukaryotic cells, and even showing the possibility of generating transient memory. However, that approach requires a sophisticated design as it relies on a secreted intermediate molecule (hormone-like) that must be kept at the right production level, and that should be previously activated by X (Repressor) and Y (SsrA-tagged protein) degradation. Furthermore, since the output of the circuit is distributed among different consortia, the concentration of the secreted molecule can differ according to the number of cells simultaneously producing it. This kind of multicellular approach and others based on single cells require sophisticated wiring design (Silva-Rocha and de Lorenzo, 2008; Siuti et al., 2013; Macia et al., 2016, 2017). By contrast, ParAlleL requires very minimal genetic modification and little tuning to obtain reliable outputs (**Figures 2**, **3**).

The implementation of ParAlleL presented here is simple, but its further development to useful applications presents a number of challenges. Firstly, expansion to 4 bits and beyond would require further well-behaved and non-cross-reacting antibiotic resistance markers, and would probably lead to even greater disparities in growth rate than those observed in the three-bit system (**Figure S1**). This could be addressed, and the flexibility and usefulness of the system increased, by moving away from direct use of antibiotics to a system using tightly controlled inducible promoters, each controlling a lethal "death gene," such asccdB, with either the presence or absence of the inducer leading to lethality. In this way the system could be made to respond to different combinations of useful inputs, for example in the construction of multiplexed biosensors.

However, to achieve the full potential of ParAlleL, it will be necessary to generate layered systems in which the output from one layer serves as input to another layer. This might be accomplished via quorum sensing, but the implementation would be rather complex and limited. A more attractive option is to transfer the same concept, using a set of agents which each responds to a single combination of inputs, to an alternative system. For example, the concept could be implemented in a cellfree system, in which inputs may be present as small molecules interacting with transcription factors, or as DNA or RNA oligonucleotides. Outputs from the first layer, in the form of DNA or RNA molecules generated by DNA replication (e.g., PCR) or transcription, could then serve as inputs to a second layer. For example, in a PCR-based approach, one of a set of templates would be amplified based on which primer oligonucleotides were added; the output PCR product could then be processed to generate another set of oligonucleotides, or used directly to initiate priming on a second set of templates via formation of three-way junctions. Alternatively, in a transcription-based system, output RNAs from a first layer computation could act as guide RNAs directing binding of CRISPR-based transcription

#### REFERENCES


factors to further templates to generate a new set of guide RNAs, eventually leading to production of an output RNA such as a mRNA leading to translation of a visually detectable signal.

Another interesting aspect of ParAlleL is that, since the final output is the result of a population-based calculation, these systems may show a level of "Byzantine fault tolerance," allowing reliable outcomes even in the face of levels of noise which are unavoidable in biological systems. This would represent a new level of robustness in biological computation systems.

#### DATA AVAILABILITY

All datasets generated for this study are included in the manuscript and/or the supplementary files. More Data is available at https://doi.org/10.7488/ds/2497.

#### AUTHOR CONTRIBUTIONS

FM and CF conceived and designed the experiments. FM and BL performed the experiments. FM and CF analyzed and validated the data. FM and CF contributed reagents, materials, and analysis tools. FM and CF wrote—reviewed & edited the manuscript.

#### FUNDING

This work was supported by CONICYT/BC-PhD 721 70403 (FM).

#### ACKNOWLEDGMENTS

The authors acknowledge the valuable assistance of Dr. Louise Horsfall for supplying the pMO9075 plasmid and Dr. Aitor De las Heras for supplying the pSEVA631 plasmid. A preprint version of the manuscript has been published on bioRxiv (Millacura et al., 2018).

#### SUPPLEMENTARY MATERIAL

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fbioe. 2019.00046/full#supplementary-material


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2019 Millacura, Largey and French. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.