Improving phenylalanine

production by Escherichia coli

using comprehensive
metabolomics information

Mariët J. van der Werf

TNO: Netherlands Organization for Applied

Scientific Research
• Independent research organisation (NGO)
• Founded in 1932
• >5400 Employees
• Turnover (2005) 562 MEuro
• 2/3 = market turnover
• broad knowledge and technology base
• International client base
• offices in Detroit, Boston, Tokyo

TNO improves the competiveness of companies and assists

governments with policy matters
 TNO: Industrial Biotechnology
• Track record of >25 years
• Seven out of 10 leading companies in industrial biotechnology are our customers
• Broad technology base
• From established technologies to the newest technologies
• Worldwide forefront position on specific topics

Feedstock Fungal Metabolomics

Engineering Biotechnology

Improving the economics of

microbial production processes

• The costs of large-scale microbial production

processes are primarily determined by substrate
costs
• Use cheaper substrates/feed stocks
• Yield improvement
• Strain improvement
• Medium (fermentation) optimization
 Microbial Strain Improvement
• The genetic possibilities are almost infinite to the
metabolic engineer

Target selection: Which gene(s) to pick?

Target selection
– current state-of-the-art
• Rational selection
• Target the genes that you think to be important
• A lot of ‘educated guess’ and ‘gut feeling’
• Metabolic (flux) models
• Construct a (limited) metabolic model from substrate to
product
• Predict in silico which genes should be targeted
• Reactions that are not known to exist, or that are not put
into the model, are not considered
• Metabolomics
 Target selection by metabolomics
Genome
Transcriptome
Proteome

• Why metabolomics? Metabolome

• Closest to the phenotype Phenotype

• Data analysis/biostatistics
• Translation of differences in metabolomes into phenotypic
differences

Analyze all
Biostatistics
metabolites Identify and
understand key
Identify correlations biological processes
Phenotype In multiple data sets

Question driven metabolomics approach

• Results in an un-biased, global view of the
biological processes involved
• Starting point for generating hypothesis
• Hypothesis-free approach
• ‘Fishing expedition’
• Key to successful
application is
EXPERIMENTAL DESIGN
• Optimization of the information
content of the metabolomics data
set with respect to the question
under study
 TNOs Metabolomics
Rapid sampling and

platform collection of quenched

sample(s)
(Diluted)
Extracelluar fluid
Extracellular
Concentrated metabolites
cell suspension
• 6 complementary analytical Sample for determination
of biomass concentration
methods Add IS biomass normalization

• Quantitative Extraction
• Allows the detection of 95-97% of Add IS for volume
Intracellular + interstitial normalization
the microbial metabolites metabolites

• Has been extensively validated in

order to allow the analysis of
‘snapshot’ samples
• Has been successfully applied
for the analysis of metabolomes
Split sample for analysis by the
different analytical methods
Add ISs for every analytical method
Further work-up of the sub-samples
as required for the different methods
Add IS for normalization injection volume
Add ISs for derivatization efficiency

of many (micro-) organisms Add ISs for retention time normalization

Analyze the sub-samples

0 NL: 2.55E6
505.83
400,000 100
Base Peak m/z=
220.00-1200.00 F: ITMS - c
90
ESI Full ms [
807.87 120.00-1200.00] MS
80
fullms_esiposneg120_1200
300,000 70
_nr16

425.84
60

200,000

100,000
50

40

30

20 238.86
864.24 860.25
741.72
524.63

338.80 338.80
521.82

765.09
…..
447.81 816.82
10
844.29 811.21

Koek et al (2006) Anal. Chem. 78:1272 0 0

0 5 10 15 20
Time (min)
25 30 35 40

10 15 20 25 30 35
Coulier et al (2006) Anal. Chem. 78:6573

Demonstration project

• Are targets identified by the combined

metabolomics/MVDA approach really important for
strain improvement?

• Phenylalanine production by Escherichia coli

• Patented strain that had already been optimized
by > 8 steps of rational design
 Experimental design
• Generation of samples under (growth) conditions that
result in large differences in phenylalanine production
• Controlled batch fermentations
• 28 metabolome samples analyzed

Exp9: Exp7: Exp5: Exp10: Exp3:

wild-type Succinate 3x pH=7 2%

Strain: Carbon Phosphate Oxygen Sampling times:

Reference pH: 16, 24, 40
Phe- Source: concentration: tension:
fermentation overproducer glucose 1x 6.5 set at 30% and 48 hours

Exp6: Exp4: Exp2:

1/3 x 300 rpm 32 hours

Analysis of the metabolomes

TIC: DSM-X90.D
Abundance

4500000

4000000

3500000

3000000

2500000

2000000

1500000

1000000

500000

Time--> 8.00 12.00 16.00 20.00 24.00 28.00 32.00 36.00 40.00 44.00 48.00 52.00 56.00
Data analysis
- “Finding the needle in the haystack”

• Differential expression
• 100’s of targets
• Assumption: the larger the response, the higher the
biological relevance
• But: is a constitutive gene that is 20% upregulated less
relevant than an inducible gene that is 1000-fold
induced?
• Multivariate data analysis tools

Multivariate data analysis

• Unbiased data interpretation

• No a-priori knowledge about the biological system
required
• Strength of the correlation of the biomolecule with
the phenotype of interest forms the basis for
ranking the targets
PLS - Ranking of the important
metabolites [P] = b A + b B + b C + ...........
1 2 3

Top 15
Rank Identity
1 4-Hydroxybenzylalcohol
2 Chorismate
3 Glycine
4 Ethylmalate (?)
5 3,5-Dihydroxypentanoate (?)
6 Tyrosine
7 2-Amino-4-hydroxy-6-(erythro-1,2,3-trohydroxypropyl)-
dihydropteridine triphosphate (folate intermediate) (?)
8 Phenyllactate
9 Dipeptide with a glycine ?
10 N-acetylglutamate
11 Unknown- 15.85
12 N-acetyl-aspartate
13 2,3-Dihydroxybenzoate
14 Unknown – 17.04
15 Unknown – 14.56

Phenylalanine production by Escherichia coli

 Relation identified metabolite
- Phenylalanine titer
4-hydroxybenzylalcohol
4-Hydroxybenzylalcohol Chorismate
Peakarea metabolite
9000 0,035
8000 0,030
7000
0,025

concentration
6000

Metabolite
5000 0,020
4000 0,015
3000 muatant
mutant 0,010
2000 WT
1000 WT 0,005
0 0,000
0,00 0,50 1,00 1,50 2,00 2,50 0 0,5 1 1,5 2 2,5
Phe titer (g/l) Phe titer (g/l)

Glycine N-Acetylglutamate
1,E+05 350000

peakarea metabolite
1,E+05 300000
Peakarea metabolite

mutant
1,E+05 250000 WT
8,E+04 200000
6,E+04 150000
4,E+04 100000
2,E+04 50000
0,E+00
0
0,000 0,500 1,000 1,500 2,000 2,500 0 0,5 1 1,5 2 2,5
Phe titer (g/l) Phe titer (g/l)

Classification of relevant metabolites

Phenylalanine biosynthesis
Chorismate
Erythrose-4-phosphate
Side routes of phenylalanine biosynthesis
4-Hydroxybenzylalcohol
Tyrosine
2-Amino-4-hydroxy-6-(erythro-1,2,3-trohydroxypropyl)-
dihydropteridine triphosphate (folate intermediate)
Phenyllactate
2,3-Dihydroxybenzoate
Seemingly unrelated metabolites
Glycine
Ethylmalate
3,5-Dihydroxypentanoate
Dipeptide with a glycine?
N-Acetyl-glutamate
Unknown -15.85
N-Acetyl-aspartate
Unknown - 17.04
Unknown – 14.56
 Erythrose-4-phosphate
Leads identified in -

Chorismate conc.
0,035
mutant

the phenylalanine 0,030

0,025

0,020
WT

0,015

biosynthesis route 0,010

0,005

0,000
0,0 1,0 2,0 3,0 4,0

Phenylalanine yield

All positive Chorismate

correlations - Ubiquinone-8
2,3-dihydroxy- Prephenate
benzoate
- -

Enterobactin
•Intermediates: L-Phenylalanine L-Tyrosine
2-Amino-4-hydroxy-6-
⇒ Overexpress intermediate (erythro-1,2,3-trihydroxypropyl)-
dihydropteridine triphosphate
converting enzyme (gene)
L-Tryptophan

•Intermediates of side routes: Folate 4-hydroxybenzylalcohol

⇒Knock-out side-route
Thiamine

Validation of leads
Relative (specific) phenylalanine concentration (%)

180
Relative Phenylalanine concentration
160
Relative Specific Phenylalanine concentration
140

ve ment
120
impro
100
ult s:50% in
ics res ia l stra
80

olom d ust r
ta b in
Me of an
60

40

20

0
D

P
L

WT
+

+
+

+
+

+
N

-E

-T
N

-E

-T
N

-E

-T
 Medium improvement
- Validation of leads
120
Relative Product Concentration (%)

100
e ment
improv s
lt s: 12% roc es
80
resu duction p
m ics
ta bolo
t ria l pro
M e ndus
n i
of a
60

40
Reference + CI0355 + CI0353 + CI0254 - CI0241 + CI0247 + CI0243 + CI0121

Conclusions
• Proven that leads identified by a question driven
metabolomics/MVDA approach are relevant for
strain and medium improvement
• 12-50% improvement
• Multivariate data analysis tools are powerful tools to
extract relevant information from large data sets
• Unbiased identification and ranking of targets
• Metabolomics/MVDA works like a navigator
• Helps you find the quickest way and make the largest
steps
• Opens up the black box of cellular metabolism
Acknowledgements

• Microbiology
• Karin Overkamp, Nicole van Luijk, Roelie Bijl, Annette Maathuis,
Alwin Albers, Marloes van Beek, Machtelt Braaksma, Robert van
den Berg
• Analytical Chemistry
• Thomas Hankemeier, Maud Koek, Bas Muilwijk, Leon Coulier,
Richard Bas, Leo van Stee
• Biostatistics
• Sabina Bijlsma, Carina Rubingh, Bianca van der Vat, Jack Vogels,
Renger Jellema, Age Smilde, Albert Tas
• DSM