Engineering Yeast for Industrial Biotechnology


Good afternoon, you’re all very welcome to
this afternoon’s webinar. My name is John Morrissey, I’m coordinater of the CHASSY project,
which is hosting this webinar for you today. Thank you all very much for taking the time
to join. So the CHASSY project, just to say a couple of words of introduction, it’s an
EU-funded project, which is aiming to develop new yeast chassis for industrial biotechnology
applications. So if you’re interested in learning more about
the project, you can follow the project and ask questions on Twitter, @ChassyProject,
via our website, chassy.eu, or our LinkedIn group. Indeed, if you’ve got questions from
this webinar that don’t get answered during the webinar, in case we don’t get time, you
can follow up by asking either on LinkedIn or on Twitter. So this particular project
has a range of partners, who are listed on your screen here. But I’m very pleased today
to be joined by two of my colleagues, Jack Pronk from the Technical University of Delft
and Jens Nielsen from Chalmers University in Sweden. Jack is professor of industrial
microbiology and biotechnology at the Technical University of Delft, he’s an expert in microbial
physiology and synthetic biology and he’s going to talk to us about some of that. Jens
Nielsen is Professor of Systems Biology at Chalmers University. He’s also the Chief Scientific
Officer of Novo Nordisk in Denmark and he’s going to talk to us about systems biology
and metabolic engineering. Both of these speakers have a wealth of experience in working across
the academia / industry interface and have very good insights, I think, on how we can
advance yeast biotechnology and how we can develop new prospects and new directions.
So, I think that’s enough of an introduction from me. In terms of the format, Jack Pronk
is going to give his presentation first. It’s about 20 minutes. He’ll be followed by Jens
Nielsen for about 20 mintues, and then we have about 20 minutes left for questions.
If you have questions, you can ask them as we go along. If you look at the bottom of
your screen (you might need to use your cursor), you will see a Q&A function. If you just click
on that Q&A, you’ll get a window and you’ll be able to ask a question. As the presentation
is going on, I’ll be collating the questions. And then, when we get to the end, after both
presentations, I’ll direct the questions to the appropriate speaker. So I think, with
that, I’m going to stop sharing my screen, I’m going to hope the technology doesn’t let
us down, and I’m going to ask Jack to start his presentation And there we go.
There we go, Jack. Good afternoon, morning, or evening, depending
on where you are. Thanks John for the introduction. What I’d like to do is to briefly share some
of the experiences of the Delft group in implementing some technologies for engineering yeast genomes
and yeast strains that have really transformed our academic research and that I think also
have the potential to transform work in industry. I guess most of you in the audience will be
aware that yeast are already very important platforms for production in the biotech industry
– they’re robust, there’s a very well-developed large-scale fermentation technology for yeasts,
and there’s a tremendous body of knowledge on, in particular Saccharomyces cerevisiae,
but increasingly also on so-called ‘non-conventional’ non-Saccharomyces yeasts. Developments in
the field have recently been driven by extremely fast developments in strain engineering technology
and this already benefits the implementation of novel product pathways in yeast, but also,
and that’s mainly what I’ll be talking about today, in improvement of existing processes
that use yeast cells. Now, as a scaffold or as an example that I
would like to use today, I’ll discuss the impact of genome editing technologies and
also of evolutionary engineering in a very well established application of Saccharomyces
cerevisiae and that’s the production of fuel ethanol. Either so-called ‘first generation’
ethanol production, currently still the largestsingle process in terms of product volume in industrial
biotech, and also ‘second-generation’ ethanol production where we use agricultural residues
or dedicated energy crops as the raw material. Now, talking about genome editing in this
day and age is not possible without saying a few words about what’s known as the ‘CRISPR
revolution’. CRISPR is a very simple, a very precise, and also a very efficient means of
editing genomes and what it enables researchers to do is to, in a very targeted manner, introduce
double strand breaks in the DNA. This cartoon over here is not from a scientific journal
or biochemical textbook, it’s from The Economist, which really illustrates the societal impact
that CRISPR has and is expected to have in the future. It very nicely illustrates what
the system does. It consists of an endonuclease protein that cuts DNA, but this cut is very
much targeted by a small piece of RNA, the so-called guide-RNA. And if you want to see
this system in action, let’s see if I can get this movie to start… or not. Well, anyhow,
this is a very charming movie, which we won’t get to see right now, apparently, which shows
how this CRISPR protein, the orange blob, is moving along the DNA until it finds its
cut site, and then it cuts the DNA. Now this is interesting because loose ends in the DNA
are very very nice places to explore for genetic engineers. They’re fantastic sites to introduce
specific mutations, to insert novel genes, and to eliminate existing genes. One of the
fantastic possibilities that this system offers is multiplexing. So not just introducing a
single mutation at the time in one site, one locus on the genome, but to introduce multiple
mutations at the same time. I will illustrate this in a few minutes. A note of course in
here, although the science is tremendously well resolved, this does not hold for the
intellectual property rights. There are still legal fights going on right now and the advice,
especially to small and medium enterprises entering this field would be to inform yourself
really well about the current legal status before committing to an experiment with this
technology. Now it’s very simple to narrow down genome
editing to just CRISPR and, in fact, that would be an over-simplification. There have
been a number of other techniques that have contributed a lot to the current acceleration
in the field. One of those that’s particularly important also for yeast engineering is recombination-
assembly of DNA fragments. Instead of old-fashioned methods in which we used restriction enzymes
and ligases to cut and paste together DNA fragments, we increasingly use recombination,
either in vivo or in vitro, to assemble large numbers of fragments, for example dozens at
a time, with an unprecedented precision and reproducability. Also, custom DNA synthesis,
which enables us not only to make novel and optimised gene sequences, but also to synthesise
linker fragments that we can use for this assembly has really sped up the field. Also
the costs are decreasing although the decline, for us, could be a bit faster still. Then,
as the constructs we’re making with genome editing become more and more complicated and
technically involved, quality control becomes more and more essential. Whole genome resequencing,
whole genome DNA resequencing, has become indispensible as a means for checking the
constructs we make. This has become possible because whole genome sequencing has become
a lot cheaper than it was a few years ago. I’ll come back to that later.
So to illustrate how things have sped up, I’d like to go back to some of the, the most
ancient research in our group on pentose fermentation by Saccharomyces cerevisiae. Fifteen years
ago, we were still talking about the ‘pentose challenge’; the inability of Saccharomyces
cerevisiae to convert 2 sugars, xylose and arabinose, that are very abundant in these
second generation feedstocks for ethanol production. Now we spent a number of years, in our group,
and similar work has been done elsewhere, to engineer Saccharomyces cerevisiae strains
by the expression of isomerase-based pathways through metabolism of xylose and also for
the metabolism of araginose. Zooming in on xylose first, this required already quite
a substantial set of genetic modifications – about 7 or 8 of them which, over a period
of a numer of years, we sequentially introduced one-by-one into the yeast genome and found
that this really helped to get the yeast to ferment xylose. Similar work was done on araginose
and in 2010 this provided us with academic strains – strains doing their job under academic
conditions – that fermented all three sugars – glucose, xylose, and araginose.
Going back to the xylose part of the work for a little bit, the strain engineering work
cost us about 3 years, of course there was some planning in between, but just building
the strains, checking them and moving on to the next step, covered a period of about 3
years. And a few years ago, Maarten Verhoeven, a
PhD student in the group, decided to give this another try, but now using the set of
techniques that I just discussed. So, CRISPR-based genome editing, in vivo assembly, and checking
the constructs with a whole genome resequencing. And essentially all the modifications and
even a little bit more than what we did in the original strains, was introduced without
having any plasmids floating around, everything neatly integrated in one locus in the genome,
and this took 6 days instead of 3 years, which, I think, is a nice illustration of how these
developments speed up the field. And, of course, the strains could grow on xylose after these
modifications. Now this is a very simple example of how relatively
simple strategies can be accelerated but of course the much more interesting thing is
that we can now access much more complicated engineering strategies, and there’s already
a couple of very very beautiful examples of that in the literature with expressing complicated
pathways like, for example, opioid synthesis in yeast by the Smolke group at Stanford in
the United States. They have shown that the expression of a large number of plant genes
enable Saccharomyces cerevisiae to make opioids. This sort of daring metabolic engineering
strategies has become a lot more accessible now with the availability of the tools that
I just mentioned. We can also use these technologies to improve
the chassis, the core machinery, of the yeast cells for very well established products.
And this example from our own group I’d like to explain in a bit more detail. Five years
ago we published a strategy for improving the yield of ethanol on feedstock in first
generation bioethanol processes – a very well established process that has been around for
years. We hypothesised and subsequently also showed that expression of autotrophic enzymes,
let’s say, carbon dioxide fixing enzymes, phophoribolukinase and ribulose 1-5 biphosphate
carboxylase enabled a higher ethanol yield in Saccharomyces cerevisiae. This was a nice
proof of principle study, but the strong impact on ethanol yield – a 13% increase of yield,
which is very relevant for a bulk product like ethanol, was achieved in chemostat conditions
and we were unable at the time to demonstrate the same improvement of performance in fast-growing
batch cultures. So in the subsequent years, we devised a strategy to optimise those strains
so that they would also work under industrially-relevant conditions. This involved quite a large set
of genetic modifications. We had to alter the regulation of one of the genes to avoid
toxic effects. We had to overexpress genes in the pentose phosphate pathway to improve
supply of this precursor molecule, the substrate for phosphoribulokinase. We had to integrate
multiple copies of the Rubisco gene, express two chaperonins, and we also had to alter
the native glycerol pathway. Now all in all, this could have presented quite a hurdle in
terms of strain engineering only 3 or 4 years ago. However Ioannis Papapetridis, who did
this work in our laboratory as a PhD student, took less than a month to construct this strain,
with all the modifications chromosomally integrated using the toolset that I just briefly described,
and this strain now had a wild type growth rate in anaerobic batch cultures on glucose,
and so, the anticipated increase in ethanol yield without a need to go to low growth rates
or chemostat cultures. So, CRISPR assisted yeast genome enables an
extreme acceleration of, let’s say, ‘conventional’ metabolic engineering, but it also enables
us now to access much more complex strain designs than were hitherto practical. There
are also options for combinatorial design, so generate multiple pathway configurations
at the same times and then give screening for the most optimal configurations. This
toolbox continues to expand at a very fast rate. Also, for example, Jean-Marc Daran and
others in our group have worked on this intensively, the introduction of other CRISPR variants
like CPF1 instead of Cas9. These tools are also increasingly applicable in non-Saccharomyces
yeasts and also here, Jean-Marc and colleagues have recently published about a tool that
can be used in multiple yeasts at the same time and automation is becoming a very strong
trend here. These techniques can be standardised to such an extent that they can be automated
by robots at a very large scale. I’d like to change themes here now a little
bit an switch to another, complimentary approach to engineering microbial genomes, and yeast
genomes in particular. This is called evolutionary engineering, or sometimes Adaptive Laboratory
Evolution (ALE). What we do in evolutionary engineering is to apply a selective pressure
on a growing culture that specifically favours mutants that have a trait of interest, let’s
say tolerance to a toxic compound, fast growth on a particular substrate, or whatever else
we’re interested in. We keep this culture growing for a long time under this selective
pressure until the culture has been taken over with mutants of improved performance,
which we can then isolate and characterise. The advantage of this approach over targeted
engineering is that it does not require prior detailed knowledge on molecular mechanisms
that underly a trait. It can be very easily automated as well by having computer-controlled
batch reactors, sequencing batch reactors, continuous activation reactors and all by
the use of robotics. Now there’s all sorts of industrially-relevant
traits to which we can apply evolutionary engineering. They all have one thing in common
– the selective pressure should act on single cells and lead to an increased growth rate
or increased survival. This offers a lot of possibilities and once
we’ve applied these principles and isolated evolved strains, indicated by the red, you
can analyse their genome. Until cheap genome resequencing became available, this approach,
to a large extent, remained a ‘black box’ approach, but whole genome resequencing has
really opened up this black box and enabled us to very rapidly identify mutations in the
DNA that may have contributed to increased performances and to check which of these mutations
are indeed causal, we can use again the genome editing toolbox to introduce them into non-evolved
strains, analyse their performance and find out which ones are really contributing. This
generates 2 types of outputs, first it teaches us how the system works, we learn from this,
extend our knowledge, but it also generates portable genetic elements; we do not have
to repeat these experiments all the time when we move to different strain backgrounds, from
laboratory strains to industrial strains, for example. We can simple transfer the mutations
to different backgrounds. As an example, I’ll show some work very briefly
on acetid acid tolerance. Acetic acid is an integral component of plant biomass and therefore
also an inevitable component of plant biomass hydrolysates, the substrates for second generation
bioethanol production. In yeast strains, in particular xylose fermenting yeast strains,
do not like acetic acid at all. This is illustrated here for glucose growing cultures. Unless
cells are pre-adapted to acetic acid by growing them at a lower permissive concentration of
the inhibitor, they stop growing above a certain concentration, and only after pre-adaptation
can they continue to grow at higher concentrations. Now I won’t go into this picture in detail,
it would take too long, but basically what we’ve done is to set up a dynamic selection
protocol by the alternated cycles of growth in the presence of the inhibitor, whose concentrations
were increased over time with cycles in which there was no acetic acid present. This strongly
favoured cells that did not require this pre-adaptation. We ran this experiment a number of times in
parallel and indeed, the strains that were isolated in this way all showed much higher
degree of acetic acid tolerance in the absence of pre-adaptation than in the case of the
non-evolved parental strain. Now, the question is: how do you get as fast as possible to
the underlying mutations so that you can check them in other backgrounds?
Well, as I said, whole genome resequencing is really a large part of the answer here.
And this is what we did. We resequenced multiple strains that had become acetic acid tolerant
without the need for pre-adaptation and we found between 5 and 21 different mutations
per evolved strain. Six genes were involved in multiple strains. Now instead of going
straight to reverse engineering of these mutations, we first applied a trick that works beautifully
well in Saccharomyces cerevisiae, and that’s classical genetics. So we mated the evolved
strains with a non-evolved acetic acid sensitive strain and then sporolated the resulting diploid,
and then double checked which of the mutations we found segregated with acetic acid tolerance,
so we checked acetic acid tolerance of the segregant and saw which of the mutations were
reproducibly associated with high tolerance. This gave us 4 genes, 4 mutations, different
mutations in different strains but in the same 4 genes and indeed when we then reintroduced
combinations of these genes we were able to show that the resulting strains approached
the tolerance of the evolved strains. This combinations with classical genetics makes
the technique extremely powerful and reliable. So, evolutionary engineering is a complimentary
and powerful approach for improving yeast performance. Genome sequencing enables fast
identification of mutants, and classical genetics can be a big help in sifting causal mutations.
Also here, CRISPR-assisted genome editing facilitates reverse engineering. Other important
trends that I don’t have time to discuss here are automation, definitely of the whole evolution
process, and a new trend is to also apply this very powerful method to phenotypes and
genotypes that do not by themselves confer a selective advantage to yeast cells. So for
example, yeast cells do not have a particular advantage in making products that cost ATP,
synthesis costs ATP, but we may be able engineer around that by building synthetic regulatory
circuits that confer a selective advantage to product-forming microbes. This approach,
I think we’ll hear about a lot in the coming years.
so coming to the end of this brief presentation, I’d like to thank the partners in CHASSY,
and I’d particularly like to thank this gentleman for his inspired and inspiring lead in the
project. I’d like to thank my colleagues and sponsors of our work in Delft, and in particular
my colleague Jean-Marc Daran, who’s leading our work in the CRISPR area, he’s also our
conscience in molecular biology, and with whom I enjoy working in CHASSY. Thank you
very much for your attention. Thank you
Jack for
a very nice presentation and to John also for organising and introducing this. So I’m
going to talk about systems biology, particularly with focus on the yeast metabolism and how
we can advance using mathematical models to begin to find metabolic engineering targets.
So before I get into some of the modelling work and quantitative omics characterisation,
let’s just rephrase and remind ourselves that yeast is not only a very important industrial
host organism, but it’s also a very important model organism for studying eukaryol biology.
So yeast has been used to, for example, unravel many of these interactions as illustrated
in this slide here. We have AMPK, which is in yeast called SNF1 which is a major energy
sensor in all eukaryol cells. How it interacts with the TORC Complex, which is a nitrogen
sensor basically, and these two are kind of controlling very much cell proliferation,
but also many important metabolic pathways. Much around the regulation of these two kinases
has been resolved using yeast as a model organism. But of course our focus here is the use of
yeast as a cell factory, and it is used extensively. We heard some very nice stories from Jack
about engineering yeast for improved bioethanol production, but also a number of other products
have been produced using yeast, and are being produced today using yeast as a cell factory.
And even more are in the pipeline or being developed. Many of the reasons for this is
that yeast is particularly robust, it’s extremely well characterized, it’s genetically trackable
and it’s generally regarded as safe for production of many different products. So this is one
of the reasons that we have, in my lab, focussed very much on using yeast as what we call a
platform organism. So we’ve been looking into how you can engineer
yeast to produce a whole range of different chemicals. These range from fatty acids and
fatty acid-derived chemicals to commodity chemicals like 3 hydroxy propionic acid and
coumaric acid, which are important building blocks in the chemical industry, to more fine
chemicals like sesquilterpenes and ornitine, resveratrol, but also more speciality products
like antibiotics and recombinant proteins. And in connection with that, we are developing
a number of synthetic biology tools and systems biology tools to advance this engineering
of yeast. Here I wanted mainly to focus on work that we’ve been doing on systems biology.
And the reason why we are currently focussing a lot on developing systems biology tools
is that, as we’ve heard from Jack’s presentation, there’s been a lot of advancement in terms
of ‘build’ in recent years, particularly CRISPR/Cas technology has really advanced our ability
to make multiple genetic modifications even in one go through multiplexing, but we’ve
also seen a number of tools that allows us to test – many biosensors have been developed
that allow us to, for example, use high throughput methods for testing and screening of strains.
What is really limiting our advancements today is very much about the design, but also acquiring
and collecting the knowledge that we are obtaining through going through this design, build,
test cycle. And so that’s where mathematical models can of course assist with and hopefully
drive this field forward. And so my hope is, of course, that we can
eventually be in a situation like we are in any other engineering discipline, where computer
models are used as an integrated part of the design. The challenge we are clearly facing
in biology is that we don’t have all the knowledge still, but I would argue that also using an
extensive modelling framework would actually assist in discovery and when the models are
failing would give us new knowledge about what needs to be incorporated in order to
really get increased knowledge about our cell factory.
So if we turn specifically to yeast metabolism, the yeast cell has about 2000 metabolic reactions,
they are associated with about 900 genes, as we know now, this is about 15% of the yeast
genome that is allocated to metabolism. As you will see later, it’s a larger fraction
of the proteome, in terms of mass, that is allocated for metabolism. But this 15% seems
to be rather conserved across many species, that this is what a cell is allocating for
metabolic functions. If we look at the metabolism and how it is connected, then we see a pattern
where it’s definitely now we are taught biochemistry. The way we are taught biochemistry is by going
through individual pathways one by one, starting often with the Embden-Meyerhof-Parnas pathway,
or glycolysis. We move on to the TCA cycle, maybe branch out to the pentose-phosphate
pathway, and so on and so on. And so we learn pathways and this is our view of how metabolism
is operating – that it’s segmented into these individual pathways. But as you can see here,
this is definitely not the case. What is shown here is an interaction graph of all the enzymes
and how they are connected with each other by sharing different metabolites. The metabolites
are coloured according to which compartment in the cell they are present in – either the
cytoplasm, here we have mitchondria, here we have endoplasmic reticulum and so on. And
you can see that it’s extremely connected, the metabolic network here. So clearly if
we have a pertubation in one of these enzymes, it will really migrate out through this whole
metabolic network here, and it’s therefore very hard to make simple predictions for rational
design without using models to assist in this. But there is another interesting constraint
that we are often forgetting, and that is that the total proteome in the cell is really
constrained also. In Saccharomyces cerevisiae this is constrained of about half of the cell
mass, more precisely, about 45% of the cell mass is allocated to proteins and what is
shown here is how that proteome mass is distributed into different cellular functions. We have
a whole large chunk here to the ribosomal system and translation machinery in general.
We have something for chaperones, we have something for glycolysis – you can see it’s
also a big chunk. Amino acid biosynthesis, this is here for yeast grown at a minimum
medium. Here for oxidative phospholation, and so on and so on.
Then what you can see is that this is a reference condition, when yeast is grown in a chemostat
at dilution rate .1, but if we turn up the temperature, yeast gets stressed, what happens
is that it starts to activate ethanol fermentation, over here we have complete respiration, and
we can see that there’s now allocated more proteome mass to glycolysis. We can zoom in
and look at individual proteins and we can see, for example, that Pgk1 is increasing.
You can also see that Pdc pyruvate decarboxylase, which is a main enzyme for conversion of pyruvate
into ethanol, is up-regulated, and so on. So this view here gives us this very clear
illustration that when a certain group, or one individual goes up, something else has
to go down. This is something that we often tend to forget when we do metabolic engineering.
We’re often just overexpressing heterologous enzymes, but that would actually have a trade-off
in the cell because something else has to go down within the proteome. And therefore,
again, this is another area where we can begin to use models to actually compensate for this
and use them to describe what are the consequences of expressing heterologous pathways.
So this is the question – how can we perform integrative analysis of this kind of data?
And here, we are using this concept of genome-scale metabolic models. These are comprehensive
network models where we are compiling information about all the individual reactions. So we
have here for example the stoichiometry, we are compiling information about cofactors
used in these reactions, the enzymes that are catalyzing this could be for example here,
isoenzymes, so we kind of have an ‘or’ relationship here – can be either E1 or E2, but can also
be protein complexes that are enzymes, so here we have an ‘and’ relationship, we need
both P4 and P5 to form this enzyme that catalyzes this reaction. So all this information we
can compile, when we have done that, genome-wide in yeast, we get the network that I showed
you before, and this network clearly provides information about the connectivity of all
the different reactions. But we can also have a mathematical representation of this, and
this is what we can use going forward. Now what we are particularly interested in
is to begin to expand the footprint of these classical models that are using this concept
of flux balancing to begin to incorporate this constraint of proteome allocation as
I have talked about, adn this can best be illustrated if we look at a very simple pathway,
so 3 steps, we could talk about a low efficient pathway where we have low Kcat values, versus
a high efficient pathway where we have high Kcat values. If these two pathways, with everything
else being equal are going to operate at the same flux, we would need more proteome mass
allocated to this pathway than to this one. And as I said, that proteome mass doesn’t
come for free. Expressing this pathway here versus this one, this one would give larger
trade-offs in terms of cell growth, for example, than others. So it’s very important to consider
particularly the Kcat information, but actually also the protein size, because that is also
something that counts in of course in the proteome allocation.
So what we are doing in our modelling concept is that we are beginning to consider the enzyme
costs. So traditionally we had this flux balancing concept, where we balance fluxes around each
metabolite. So flux in equals fluxes out, as illustrated here, but we are now expanding
this to consider costs of building the enzymes, and if it’s a large enzyme, that means that
there is a large cost with that, but also kinetic constraints imposed by the enzyme,
so is the Kcat high or low is also considered. So hereby we get penalty for using long pathways
that have inefficient, and I would also say, large, enzymes.
And so this is what we call ‘Gecko’ modelling approach. And so you can see here the way
the formulation goes is that we are capping and constraining the upper rate through each
individual reaction by the Kcat multiplied by the enzyme concentration. And here we can
use then quantitative proteomics for example to insert avalues here and we collect Kcat
from databases. Or, what we can do is also let the model calculate distribution of the
proteome and just say that the sum of all the proteins should be within a certain fraction
that is allocated for enzymes in the model. And this we can have, for example, from proteomics
data also. What we can get with this type of modelling is a really much improved predictive
performance. I’m not going to go into details about it here, but also what we can do is
we can get insight into enzyme distribution into specific pathways and I would rather
talk a little bit more about that. So when we introduced this modelling concept,
we were interested to see if we could explain the Crabtree effect, which has for a long
time been partly explained by regulatory constraints or kinetic constraints, but we’re interested
to see can it simply just be explained by this proteome allocation process. So just
very briefly about the Crabtree effect, when yeast is exposed to high glucose concentration,
it prefers to convert most of the glucose into ethanol by fermentation, rather than
having full respiration where you of course get much more ATP out of the process per moles
of glucose. And one could speculate and say why would you not want to capture more ATP
per moles of glucose? Well there is additional coast about activing respiration and TCA cycle
because you have many more enzymes here, and some of these enzymes are even large and complex
and you also have the whole respiratory system. So, yes, you get more ATP, but it’s also more
expensive to build this respiratory system than the fermentation route. And exactly when
we do take this into consideration, as I said, the protein mass, the sizes of the proteins,
and the catalytic efficiency, it turns out that this shift, when the cells have to grow
faster and the demand for ATP increases, it simply is better to shift and actually use
the fermentation route because you can get more ATP per protein mass on the return of
that. And that can now completely explain this turn-on of ethanol production as you’ll
see here when you go up in growth rate, you get a little bit of acetate production, you
get an increase in CO2 production, a decrease in oxygen consumption, and an increase in
glucose consumption. And you can even begin to, from the model, also see how different
pathways are changing in flux. Here oxidative phosphorylation is decreasing above a certain
critical dilution rate. And as I said, this enzyme-constrained model
as we call it will give you a much better prediction of experimental growth rates compared
with just flux balance analysis that simply misses to predict growth on a number of these
different carbon sources primarily because these are all fermentative carbon sources,
so to speak, that result in the crabtree effect. The current yeast model with proteome allocation
still has some places where it deviates, for example her on trehalose where it overpredicts
significantly. This is probably because of additional regulation or constraints associated
with trehalose metabolism that we don’t have in our metabolic model.
So, we also were interested to go further and see if we could then combine this multi-omics
analysis with the modelling to get insight into what is really controlling the growth,
and so there we looked again into this classical series of experiments where we are increasing
the dilution rate of yeasts, so basically the growth rate, we would then have an increased
uptake of glucose, as you can see here, and up to the critical dilution rate, and when
we’ve surpassed that, there’s an increase in slope here, where you get onset of ethanol
production, you get an increase in CO2 production, and you get a levelling off of oxygen consumption.
And for each dilution rate which was carried out in triplicate, we measured a number of
omics. RNAseq for transcriptome, proteomics, phosphoproteomics, metabolomics, and we also
quantified fluxes using the model. And all these data here were done in absolute quantitative
levels, so we get copies per cell. So here you see the distribution of the proteomics
of the proteome across these different dilution rates and what you see here now clearly is
that you have this increase in allocation of proteome for translation. So we see this
clearly, the allocation of more and more proteome for tranlating and actually synthesising protein.
Interestingly enough, we saw an increase in glycolytic flux, but we see actually a decrease
in allocation of proteome for the central carbon metabolism, which is basically glycolysis.
For biosynthesis it’s a little bit different, and so what you see really is some overall
changes here. As we zoom into specific biological processes,
we see, as was very clear from the previous map, this very strong linear correlation between
ribosomal system allocation of protein to ribosomes with specific growth rate, it’s
increasing linearly here with a remarkably strong Pearson. We see for amino acid biosynthesis,
we also see increase in growth. For nucleotide biosynthesis, we also see increases. It’s
a very low slope here, but there is an increase. Whereas when we look into glycolysis, as I
said, that is decreasing and for lipid biosynthesis there also is a decreasing, particularly at
the higher specific growth rate, which we believe is due also to changes in cell volume,
which will also change the lipid content per cell per gram dry weight. We were also looking
into what happens if we compare proteome fraction allocated to different processes below the
critical dilution rate, so that means from below a purely respitorial condition, to above.
And you can see translation is here, that’s because of the linear curve. But you can see
that there are, for example, mitochondria is changing, whereas for example in glycolysis
is also changing according to this picture here. But for many of the other standard processes,
you can see here transcription, stress, and so on, it’s actually a constant fraction that
is allocated at these two conditions. So we therefore zoom in on glycolysis specifically,
because we saw some clear differences. We see this increase in flux, but we see a decrease
in proteome allocation for glycolysis, so we were interested to see how that translated
into individual enzymes. Here we’re looking into a number of the individual enzymes. You
can see they have decreasing levels with the increasing dilution rate, whereas flux is
increasing. There is one deviation from this, this is hexokinase 2. Here, we have a very
strong correlation between flux, protein level, and messenger RNA level. Interestingly enough,
we across here a very strong corellation between absolute messenger RNA level and protein level,
which is something we see across from all enzymes in the cell. But clearly this points
out that there must be additional or different types of regulation that controls the activity
in these enzymes using glycolysis. And it turns out that it’s really very much related
to the phospho-proteomics allocation. So what we see is that the number of the phospho peptides,
for example Glk is one, or Pfk1 or Pfk2 here, we see decreased phosphorylation, and this
is indicated by the red lines, of several of these phosphopeptides across dilution rates.
And actually there is a very strong negative linear correlation between the abundance of
the phosphopeptides and fluxes, see the black line here. So this clearly shows that we get
a dephosphorylation of these enzymes and hereby they are likely going to be more active and
therefore that can compensate for the fact that the protein abundance is actually decreasing
for many of these enzymes. So in conclusion this allows us to do this
global mapping of flux control. So we can go in and say that, for example, for the EMP
pathway or glycolysis, we have some element of enzyme saturation. I didn’t talk too much
about that, but we can see that from the metabolomics data. We have a very strong regulation by
protein phosphorylation, we do have some elements of allosteric regulation. We do have transcriptional
regulation for one key enzyme, hexokinase 2, but basically not for any of the other
enzymes, whereas for really many other processes in the cell, and particularly for amino acid
biosynthesis and translation, we have very strong transcriptional regulation. Whereas
we see also for many other processes, there’s less contribution of phosphorylation, enzyme
saturation seems to be a more important factor. So in conclusion we can say that constraining
the proteome allocation in the cell really allows us to have much improved prediction
by these genome scale metabolic models, really that the growth rate is to a large extent
controlled by translational machinery, and that biosynthesis is mainly controlled by
protein level, but central carbon metabolism has really evolved to have this excess capacity
at low growth rates. So therefore it has additional regulatroy features inserted and now we can
also discuss why the cell has evolved to be operating like that.
So let me acknowledge the people who’ve been doing much of the work here, Benjamin and
Avlant have been doing it together with Cheng. They’ve been doing much of the modelling here.
Ibrahim also advancing some of the modelling work. Petri has been working very much on
the quanititative proteomics. And then visiting Professor Jianye was working on the analysis
of this growth rate curve their together with Yu. Verena Siewers has been assisting in coordinating
many of these projects and our collaborator who did the proteomics, and the funders and
you all for your attention. Thank you. Thanks a lot Jens, that was very informative,
and also Jack for his informative talk. So if people have questions, you can put them
in the panel. I’ve got a couple of questions in already, so I’m just going to direct those
to people. I’ve a question here for Jack. The question is whether the evolutionary improved
trait for acetic acid tolerance is stable. If this strain is stored for longer at -80
or grown on YPD for a longer time, do you maintain the trait?
Well after freezing, definitely, and thawing the phenotype is still there. We did not check
for very prolonged growth under non-selective conditions, but after just a few cycles of
batch cultivation, the cells are definitely still as tolerant as directly after the evolution.
Ok, thanks Jack. I’ve a question for Jens. Jens, the work you did is very related to
Saccharomyces cerevisiae. Do you think that work on how glycolysis is regulated at the
different levels is also going to apply to other yeasts, which are maybe not so optimised
for ethanol production? That’s a good question. I think we would probably
see different types of regulation of glycolysis, Saccharomyces cerevisiae of course has this
feature that it can have this extremely high flux through the glycolytic pathway, and therefore
it has probably evolved to have a relatively large allocation of proteome for that pathway.
What is interesting of course to see is that at low dilution rates, it’s even allocating
even excess capacity for this pathway here, but I think this will be different. This is
something we’ll have to look into, for example if we go to Kluyveromyces or Yarrowia as we’re
doing in CHASSY, to see how are the similarities when we look across dilution rates in terms
of proteome allocation. I think that’ll be super exciting.
OK thanks Jens. I’ve a long question here as well. I’m going to read this out. Either
Jack or Jens can answer this one. It says ‘Yeast is indeed a workhorse for ethanol production,
but when it comes to other products, for example organic acids like succinic acid, it lags
behind prokaryotic hosts in terms of yield and productivity.’ I suppose the first part
– ‘Why do you think this is?’ Shall I try to explain to that?
Yeah, sure Jack. There are beautiful examples of very high
yields of succinic acid production by prokaryotes. The work of Sang Yup Lee I think would be
a fantastic example of that. However the main difference between those prokaryotic strains
and the currently engineered strains, of Saccharomyces cerevisiae in particular, is that the bacterial
systems use succinic acid production as their sole metabolic pathway. Processes are anaerobic
and there’s direct selection for productivity and in a way also for yield. The field of
engineered yeast systems, still require respiration to generate ATP, so the dicarboxylic acid
production is not the sole source of ATP. In fact, the processes may even still require
some ATP. A key factor in that, definitely, is the export of the acids across the membrane,
which is an ATP-requiring process. In addition to that redox metabolism, a compartmentation
of redox metabolism, causes some challenges in yeast systems. Having said that, there
are commercial applications of yeast-based succinic acid production in particular, which,
although there is a concession or a compromise with respect to yield, do profit from the
fact that the acid can be made at low pH. Ok thanks Jack. I’ve a question for Jens.
Jens the question is ‘How can the GEMs specifically be used for industrial metabolic engineering
approaches?’ So this is a good question also. I think that
models are used quite extensively by many companies for a couple of different things.
One thing is that, before you start a metabolic engineering effort, the models can allow you
to calculate a theoretical maximum yield very easily. It’s good to know what is the upper
theoretical yield that you can get from glucose, particularly for many processes where glucose
cost is the major cost for producing the product of interest. And so that allows you to do
at least initially some back-of-the-envelope type of economic calculations of whether you
should pursue this process or can you even compete with existing production routes. Secondly,
the models are very useful in terms of evaluating difference in yields. So, what if I insert
this pathway, or I go via this route here, how would that impact yield?
The models have been less efficient so far, in terms of guiding specific targets and saying
ok I need to attenuate or I need to increase the expression of this particular enzyme,
but this is of course where I hope that our new modelling approach will incorporate information
about proteomics and Kcat values that we can actually begin to address these issues, but
this is of course still to be proven, whether that’s the case.
OK thanks Jens. Jack, I’ve a question myself. This didn’t come on the list, but I was at
a CRISPR workshop a couple of weeks ago and there was a lot of talk about off-target effects
of Cas9 in particular. Is this something that you’ve had much problems with, or experienced
much with yeast systems? In our events, it’s not been a major problem,
but of course an occassional point mutation somewhere would be less of a problem in a
yeast cell than in gene therapy. We’ve resequenced, Jean-Marc has resequenced quite a lot of strains
instructed with CRISPR-assisted genome editing and we have no indications for major problems
there in the sense of off-target cutting by Cas9. It might exist, I mean the sample is
significant in terms of number, but it doesn’t mean that there cannot be effects.
OK, thank you. I’m just looking to see, I don’t have any more questions on the list
at the moment, and we’ve come to the end of our allocated hour, so I would like to take
this opportunity to thank Jack and Jens for giving those presentations, which I think
were excellent – really informative. I’ll thank all the people who attended, asked questions
and listened. If you do have further questions, as I mentioned at the beginning, please feel
free to send those to us via the LinkedIn group or indeed via Twitter, and we’ll do
our best to respond to those. And you can keep updated on the project and on developments
in yeast biotechnology on the CHASSYwebsite (chassy.eu) so thank you all very much.

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *