Identification of metabolite-protein interactions among enzymes of the Calvin Cycle in a CO2-fixing bacterium

(1)

EXAMENSARBETE INOM BIOTEKNIK, AVANCERAD NIVÅ, 30 HP

STOCKHOLM, SVERIGE 2020

Identification of metabolite-protein interactions among enzymes of the Calvin Cycle in a CO2- fixing bacterium

EMIL SPORRE

KTH

SKOLAN FÖR KEMI, BIOTEKNOLOGI OCH HÄLSA

(2)

1. Abstract

The Calvin – Benson cycle is the most widespread metabolic pathway capable of fixing CO2 in nature and a target of very high interest to metabolic engineers worldwide. In this study, 12 metabolites (ATP, AMP, NADP, NADPH, 2PG, 3PGA, FBP, RuBP, PEP, AKG, Ac-CoA and phenylalanine) were tested for protein – metabolite interactions against the proteome of Cupriavidus necator (previously Ralstonia eutropha) in the hopes of finding potential examples of allosteric regulation of the Calvin – Benson cycle. This is accomplished through the use of the LiP-SMap method, a recently developed shotgun proteomics method described by Piazza et al. capable of testing a metabolite of interest for interactions with the entire proteome of an organism at once. A functional protocol was developed and 234 protein – metabolite interactions between ATP and the proteome of C. necator are identified, 103 of which are potentially novel. Due to time constraints and setbacks in the lab, significant results were not produced for the other 11 metabolites tested.

C. necator is an industrially relevant chemolithoautotroph that can be engineered to produce many valuable products and is capable of growth on CO2 and hydrogen gas. The bacteria were grown in continuous cultures after which the proteome was extracted while retaining its native state.

Subsequently, the proteome was incubated with a metabolite of interest and subjected to limited, non-specific proteolysis. The resulting peptide mix was analyzed by liquid chromatography coupled tandem mass spectrometry (LC – MS/MS).

Keywords: Cupriavidus necator, chemolithoautotroph, limited proteolysis, Calvin Cycle, carbon fixation, proteomics, allosteric regulation, metabolite-protein interactions.

(3)

2. Sammanfattning

Calvin-Benson-cykeln är den mest utbredda metaboliska processen i naturen med vilken det är möjligt att fixera CO2 och en måltavla av högsta intresse för bioteknologer världen över.

I den här studien testades 12 metaboliter (ATP, AMP, NADP, NADPH, 2PG, 3PGA, FBP, RuBP, PEP, AKG, Ac-CoA and phenylalanine) för interaktioner mot proteomet från Cupriavidus necator (tidigare Ralstonia eutropha) i hopp om att hitta potentiella exempel på allosterisk reglering av

Calvin-Benson-cykeln. Detta uppnåddes genom användning av LiP-SMap-metoden, en nyligen utvecklad proteomikmetod beskriven av Piazza et al. kapabel av att testa en metabolit av intresse mot en organisms hela proteom simultant. Ett funktionellt protokoll utvecklades och 234

interaktioner mellan ATP och proteomet av C. necator identifierades, varav 103 potentiellt är

nyupptäckta. På grund av tidsbrist och motgångar i labbet producerades inga signifikanta resultat för de resterande 11 metaboliterna som testades.

C. necator är en industriellt relevant kemolitoautotrof som kan växa på CO2 och vätgas, samt

manipuleras till att producera många värdefulla produkter. Bakterierna odlades i kemostater varefter proteomet extraherades i sitt naturliga tillstånd. Sedan inkuberades proteomet med en metabolit av intresse och utsattes för begränsad, icke-specifik proteolys. Den resulterande peptidblandningen analyserades via tandem masspektrometri kopplad till vätskekromatografi (LC – MS/MS).

(4)

3. Introduction

Over the last century several changes have been observed in the climate system. Global ice bodies have lost mass steadily, global temperatures show a linearly increasing trend and the oceans have become more acidic. These changes will increase in severity over the coming decades and new climate-related risks are likely to appear. Key predicted risks include food and water insecurity, loss of biodiversity and increased frequency of severe storm surges, flooding, and extreme weather events. The main driver behind climate change is anthropogenic greenhouse gas (GHG) emissions.

They have increased dramatically since the industrial revolution and highest rate of change has been observed since the turn of the millennium. 78 % of the total increase in GHG emissions between 1970 and 2010 was due to CO2 emissions from industrial processes and combustion of fossil fuel. As such, climate change poses a serious global threat and CO2 emissions is the main force behind it (1).

In addition to the threat of climate change, scarcity of oil and dependency on foreign nations for supply are major reasons to look for an alternative source of fuels and chemicals. One such

alternative is to use microorganisms to convert biomass into the desired products, effectively making them part of the carbon cycle thus ensuring they do not contribute to atmospheric accumulation of CO2. However, the vast majority of fuels and chemicals produced commercially by microbial means utilize biomass as feedstock, usually edible crops such as corn, sugarcane or rapeseed. The use of such biomass comes with two inherent problems: the need for large areas of arable land and the logistical costs of handling large amounts of feedstock.

The environmental gains from using products from renewable biomass are severely diminished when the cultivation of said biomass necessitate the clearing of forests or other types of land that naturally sequester CO2. One study claims that without taking land use changes into account, an attempt to keep atmospheric CO2 levels below 450 ppm would result in the destruction of all the world’s forests and savannahs by 2065, releasing large amounts of CO2 annually in the process (2). The use of dedicated energy crops and residuals from agriculture and forestry as biomass feedstock is a heavily researched alternative that mitigate these effects. However, there are many concerns regarding the environmental sustainability of such methods, including limited amount of available land (3), energy-intensive pretreatments and agriculture-related emissions (4).

In addition, many cases of microbial production have difficulty competing commercially with cheaper petroleum-based products. For example, as of 2016 the share of biofuels in the transport sector was only 3.4 % despite large scale commercial production and a heavy medial and political presence over several decades (5). A significant part of the production cost for microbial products using biomass as feedstock is the cultivation, harvesting, transporting, storing and pre-treating the feedstock (6)(7). To avoid the above described problems of conventional microbial production, much research is being done regarding autotrophic microbial production that effectively removes the need for biomass as feedstock entirely, replacing it with CO2. However, large scale production requires large amounts of CO2, which creates a need for either a nearby source of CO2, such as a fossil-based power plant, or a transport infrastructure capable of moving the required amount of CO2 from such a source to the production location.

3.1 Autotrophic microbial production

There are many aspects of autotrophic microorganisms that are desirable for production of chemicals and fuels. Many can grow to high densities and have a higher yield per area when compared to terrestrial crops (8). They can be grown on non-arable land and have simple nutrient requirements, mainly water, CO2 and a source of energy such as sunlight or hydrogen gas. In addition, several autotrophic organisms are well studied and have been manipulated genetically to

(6)

2 produce a range of valuable products including several hydrocarbons, fatty acids, biodegradable polymers, proteins, pharmaceuticals and polysaccharides (9).

The most prevalent and well-studied autotrophic organisms associated with biofuel production are cyanobacteria and microalgae, both photoautotrophs. While these hosts hold great promise, they are currently unable to compete commercially with petroleum-based products, or even conventional heterotrophic hosts (10). There are many challenges concerning autotrophic production, two of the largest are cultivating sufficiently high-density cultures and harvesting them in an efficient manner (9).

An inherent problem when using photoautotrophs is the difficulty of providing light evenly to all cells in the culture. When such a culture attains a desirable high cell density, light will have trouble penetrating to the central region of the culture. As such, peripheral cells will be exposed to higher levels of light than central cells and while it is certainly possible to achieve high-density cell cultures of such microorganisms, it requires advanced and expensive technical solutions such as complex photobioreactors. An alternative is to use more low-tech open systems which is easier and cheaper to scale up but results in lower cell density. Due to the small cell size and dilute nature of the cell suspensions, harvesting and dewatering is another major challenge, representing up to 30 % of the total biomass production cost (10).

While several technical solutions are under development to tackle these problems, another option is to use an autotrophic bacterium that does not perform photosynthesis. Cupriavidus necator

(previously Ralstonia eutropha) is a facultative chemolithoautotroph capable of both heterotrophic and autotrophic growth. It utilizes hydrogen gas as an energy source instead of light and is capable of very high cell densities (over 200 g/L). As such, denser cultures can be achieved with cheaper

conventional bioreactors by adding hydrogen gas. In addition, C. necator is known to direct most of its metabolic flux to synthesis of polyhydroxybutyrate (PHB) under certain nutrient limiting

conditions. This tendency has been exploited by metabolic engineers to force carbon flux toward a desired product (11). This makes C. necator an industrially interesting bacterium with great potential for production of chemicals and fuels, which is why this bacterium was chosen for this study.

3.2 The Calvin-Benson cycle

Almost all life on Earth depends on the assimilation of atmospheric CO2 through the Calvin-Benson cycle by autotrophic organisms such as plants, cyanobacteria and microalgae. Carbon fixation through the cycle is initiated by the enzyme ribulose-1,5-bisphosphate oxygenase/decarboxylase (Rubisco) which catalyzes a reaction turning CO2 and ribulose-1,5-bisphosphate (RuBP) into 3- phosphoglycerate (3PGA). 3PGA is then converted into glyceraldehyde-3-phosphate (G3P) which can then be used to build other compounds. However, for every six G3P made, five must be used to regenerate RuBP to keep fixing CO2. In addition, the Rubisco enzyme is an extremely slow catalyst that is afflicted by a side reaction using atmospheric O2 to create 2-phosphoglycolate (2PG) that must then be salvaged through photorespiration (12)(13).

As such, the Calvin-Benson cycle is not particularly efficient, and much effort has gone into increasing the carbon flux through the cycle. Transcriptional and redox regulation of the cycle is well

understood as well as some carbon concentrating mechanisms. Allosteric regulation, on the other hand, is a relatively unexplored field that could have large effects on the flux through the cycle.

Identifying protein-metabolite interactions in the cycle is the first step towards manipulating the allosteric regulation of the cycle which could lead to significantly increased flux through the cycle.

This would naturally be of great use to metabolic engineers working with autotrophic production worldwide, whether it concerns biofuels or other valuable products.

(7)

3

3.3 Metabolites of interest

During this study, 12 metabolites were tested for protein-metabolite interactions against the proteome of C. necator. These were chosen because of they have the potential to be regulators of enzymes in the Calvin-Benson cycle. Each metabolite was tested in two concentrations, one low and one high, based on observed cellular concentrations in metabolomics experiments

(14)(15)(16)(17)(18), feasible concentration ranges from models (19) and other experiments

(20)(21)(22). In many cases, the concentrations used are significantly higher than the ones observed in metabolomics experiments. This is because the activity of allosteric regulation depends on the concentration. A metabolite might not exercise detectable allosteric regulation until it has accumulated to a sufficiently high concentration. As such, the decision was made to increase the concentration of metabolites to represent the conditions during which they would be active allosterically. In addition, the fast turnover times of metabolites in the cell mean that it is very easy for metabolite pools to decay in between the harvest and quenching steps of metabolomics experiments. The metabolites of the Calvin-Benson cycle turn over particularly fast, some with turnover times of less than 0.1s. Thus, a delay of only a fraction of a second between harvest and quenching may cause a metabolite pool to decay, leading to an underestimated concentration (23).

Here follows a short motivation as to why each metabolite was chosen and the concentrations used:

Adenosine tri- and monophosphate (ATP, AMP)

A The adenosine compounds are energy-carriers and thus reflect the energy state of the cell. Since the Calvin-Benson cycle is energy demanding, it would be logical for it to be inactivated when energy is scarce. This could be mediated through any of these molecules. Also, since these molecules are known to interact with many proteins in the metabolism, their inclusion can be valuable for data validation. In addition, AMP has been shown to inhibit several enzymes in the Calvin-Benson cycle in cyanobacteria, C. necator and Thiobacillus neapolitanus (24)(25)(26). The high and low

concentrations (mM) used for ATP and AMP were 32 / 2 and 5 / 0.5 respectively.

Nicotineamide adenine dinucleotide phosphate (NADP, NADPH)

The ratio of NADP to NADPH reflects the redox ratio of the cell which is another logical basis for regulation of the cycle, since it consumes six NADPH for each G3P that exits the cycle. In addition, NADPH has been shown to mediate light-induced regulation of the cycle and to allosterically regulate carbon concentrating mechanisms in cyanobacteria (27)(28)(29). The high and low concentrations (mM) used for NADP and NADPH were 5 / 0.5.

2-phosphoglycolate, Ribulose-1,5-bisphosphate and 3-phosphoglycerate (2PG, 3PGA, RuBP)

2PG and 3PGA are the products of the Rubisco enzymes oxygenase and carboxylase activity,

respectively. Since there are several examples of products inhibiting the enzymes that produce them in the metabolism, such as glucose-6-phosphate and hexokinase in glycolysis, there is precedence for either of these metabolites to inhibit Rubisco. Since 2PG is also the first step of a costly and

unwanted metabolic pathway it could have regulatory properties to minimize flux through it. In addition, 2PG has been proposed to signal carbon limitation in the cell and has been shown to allosterically regulate carbon concentration mechanisms in cyanobacteria (30). 3PGA has also been shown to inhibit phosphoribulokinase in spinach (31). Similarly, many enzymes are activated by their own substrate, which indicates that RuBP could be a potential activator of Rubisco. It has also been shown to modulate the formation of carbon concentrating mechanisms in cyanobacteria (28). The

(8)

4 high and low concentrations (mM) used for 2PG, 3PGA and RuBP were

4 / 0.4, 5 / 0.5 and 10 / 0.5 respectively.

Acetyl coenzyme A, phospoenolpyruvate and α – ketoglutarate (AcCoA, PEP, AKG) These molecules are all key metabolites in other pathways in the central carbon metabolism, which makes them solid candidates for signaling the states of other pathways, through which they can mediate how ready the cell is for anabolism. In addition, AKG has been shown to signal nitrogen limitation and modulate formation of CCM in cyano bacteria (28)(30). PEP has been shown to inhibit phosphoribulokinase in C. necator (24) and many other enzymes in various bacterial species (25)(32) (33)(34). The high and low concentrations (mM) used for AcCoA, PEP and AKG were 2 / 0.05, 10 / 0.5 and 10 / 0.5 respectively.

Fructose-1,5-bisphosphate (FBP)

Similarly, FBP could signal the state of the Calvin-Benson cycle to other pathways in which it is involved. It has also been pointed out as a potential source of metabolic oscillations in the

photosynthetic reaction system (35). The high and low concentrations (mM) used for FBP was 2 and 0.05.

Phenylalanine (Phe)

Phenylalanine has the potential to signal whether the cell is ready to grow through protein synthesis.

For example, access to other nutrients could limit the cells capacity to grow in which case the Calvin- Benson cycle could be slowed to prevent excessive carbon uptake. The high and low concentrations (mM) used for phenylalanine was 0.1 and 0.01.

3.4 Limited proteolysis small-molecule mapping (LiP-SMap)

The LiP-SMap method was first described by Piazza et al. in 2018 and the workflow in this study has been directly adapted from their original paper (36). After cultivation and harvest of the bacteria, the proteome will be extracted while retaining the native state of the proteins. Two sets of proteome extract will then be prepared, one will be treated with a metabolite of interest and one will be left untreated. Both of these will then be subjected to non-specific, limited proteolysis by the enzyme proteinase K. In the untreated version, the proteome will be digested non-specifically which gives rise to a certain digestion profile of peptides. In the treated version, the metabolite of interest will interact with proteins and result in an altered digestion profile. The interaction between a metabolite and a protein can block a certain peptide for digestion or expose a previously hidden peptide by changing the conformation of the protein. Both peptide mixes will then be completely digested with LysC and trypsin and analyzed by liquid chromatography-coupled tandem mass spectrometry (LC – MS/MS). When the data generated from the two samples is compared, some peptides will be more abundant in one or the other because the metabolite of interest has interfered with proteolysis.

Peptides with a statistically significant difference in detected intensity will then be annotated as having an interaction with the metabolite of interest. It is crucial that the native state of proteins is maintained through proteome extraction, metabolite treatment and limited proteolysis for the result to accurately represent interactions inside the cell.

(9)

5

4. Materials and methods

4.1 Cultivation and harvest

C. necator strain DSMZ 428 was grown in two subsequent overnight shake flask cultures with

minimal medium whose composition is described in appendix table 1. Four chemostat reactors were prepared with minimal media and 3.5 g/L fructose before being inoculated with the overnight cultures to a starting OD600 of 0.05. OD600 and OD720 measurements were made every 15 minutes.

Initially they grew as batch cultures until growth ceased, after which a feed of 8 g/L formic acid was added at a rate of 0.01 mL/min. After the OD had leveled out the feed was increased by another 0.01 mL/min to a maximum of 0.04 mL/min. The cells were harvested after the cultures had had a stable OD for a minimum of five doubling times on the formic acid feed. The OD600 at harvest varied between 1.2 and 1.5. The cultures were harvested through centrifugation and washed thrice with cold lysis buffer. The cells were then resuspended in lysis buffer and flash frozen in liquid nitrogen for storage at -80 °C. Two cultivations were made with three aliquots made per biological replicate for a total of 24 samples.

4.2 Lysis and protein quantification

Frozen samples were brought out of -80 °C storage and thawed on ice. They were then lysed mechanically through rigorous shaking by a FastPrep-24 5G lysis machine over six cycles of 45 seconds, 6.5 m/s with 30 seconds on ice between cycles. The lysate was centrifuged for 5 min at 21,000g and 4 °C. The resulting supernatant was run through a Zeba Spin Desalting column for 2 min at 1,500g and 4 °C. The protein concentration was then evaluated through a Bradford assay. Seven samples of equal volume were then created containing 100 μg protein. Six of the samples were treated with metabolite stock solutions while the final tube was used as a blank. All samples were incubated at room temperature for at least 10 minutes.

4.3 Limited Proteolysis and complete digestion

After incubation, proteinase K was added simultaneously to all samples at a 1:100 enzyme to substrate mass ratio. The samples were then incubated at 25 °C for 5 min immediately followed by incubation of 96 °C for 3 min. Afterwards, sodium deoxycholate and DTT were added to all samples to final concentrations of 5 % and 20 mM respectively after which the samples were again incubated at 96 °C for 10 min. 30 mM chloroacetamide was added to each sample for alkylation, followed by 30 min of room temperature incubation in the dark. Finally, the samples were diluted 5X in PBS and digested by 0.2 μg/μL Trypsin/LysC mix at 37 °C, 400 RPM, for 18h ± 30 min overnight. Digestion was halted by addition of formic acid to reduce pH below 2 which caused sodium deoxycholate to precipitate. Samples were then centrifuged at 14,000g for 10 min after which the supernatant was removed and stored at -20 °C.

4.4 Peptide purification and LC – MS/MS analysis

Frozen peptide mixes were thawed to room temperature and loaded onto activated Waters SepPak tC18 cartridges. The columns were then washed with 0.1 % formic acid after which the samples were eluted with 50 % acetonitrile – 0.1 % formic acid. The resulting peptide mixes were dried with a vacuum centrifuge and stored in -20 °C for several weeks until the mass spectrometry machine was available. At that point, the peptides were resuspended in 0.1 % formic acid and analyzed by LC – MS/MS utilizing a Q-exactive HF Hybrid Quadrupole-Orbitrap Mass Spectrometer coupled with an UltiMate 3000 RSLCnano System with an EASY-Spray ion source.

2 μL of each sample was onto a C18 Acclaim PepMap 100 trap column (1 mm x 15 mm, 5μm, 100 Å) with a flow rate of 15 μL per min, using 3 % acetonitrile, 0.1 % formic acid and 96.9 % water as

(10)

6 solvent. The samples were then separated on an ES806A EASY-Spray C18 analytical column (150 μm x 15 cm, 2 μm, 100 Å) with a flow rate of 3.6 μL per minute for 50 minutes using a linear gradient from 1 to 32 % with 95 % acetonitrile, 0.1 % formic acid and 4.9 % water as secondary solvent. For the evaluation experiment, the columns were changed to C18 Acclaim PepMap 100 trap column (75 μm x 2 cm, 3 μm, 100 Å) and ES802 EASY-Spray PepMap RSLC C18 Column (75 μm x 25 cm, 2 μm, 100Å) respectively. The flow rate on the first column was also decreased to 7 μL per min and the elution time was reduced to 40 min.

After separation MS analysis was performed using one full scan (resolution 30,000 at 200 m/z, mass range 300 – 1200 m/z) followed by 30 MS2 DIA scans (resolution 30,000 at 200 m/z, mass range 350 – 1000 m/z) with an isolation window of 10 m/z. Precursor ion fragmentation was performed with high-energy collision-induced dissociation at an NCE of 26. The maximum injection times for the MS1 and MS2 were 105 ms and 55 ms respectively, and the automatic gain control was set to 3e6 and 1e6 respectively.

The EncyclopeDIA and Prosit workflows were used to generate a predicted library from a fasta file of C. necators proteome against which an EncyclopeDIA search was performed to generate a list of detected peptides (37)(38).

4.5 Data analysis

The output of the LC – MS/MS analysis was in the form of a list of peptides, their protein of origin and detected intensity. Each metabolite treatment was considered a separate experiment in which the intensity of every detected peptide was compared to the intensity of that same peptide in the blank sample. Any peptide that did not have at least three measurements in all three samples (high concentration, low concentration and blank) was excluded from the result. Statistical analysis was performed pair-wise with the MSstats package in R (39). A significant peptide was defined as one that has an absolute log2FC of at least 1 and a q-value of less than 0.1, meaning the False Discovery Rate (FDR) was set to 10 %. In the evaluation experiment the q-value threshold was lowered to 0.01, decreasing the FDR to 1 %. Every protein with significant peptides in the high concentration ATP comparison was looked up in the UniProt annotation database to check if the interaction was novel or previously recorded. Any mention of ATP under cofactor, protein name or molecular activity caused a protein to be defined as previously annotated.

The statistical viability of the data was investigated through p-value distributions, Principal Component Analysis (PCA), QQ-plots and Pearson correlation between biological replicates and treatments.

4.6 Small scale evaluation experiment

After the data produced by the above described protocol was discovered to not be statistically sound, a small-scale evaluation experiment was performed in which several parameters were changed. The method of statistical analysis was changed to group-wise through use of four technical replicates rather than one. Instead of overnight digestion by lysC/trypsin mix, the samples were pre- digested by lysC at 37 °C, 400 RPM, 3h after which they were digested by trypsin at 37 °C, 400 RPM, 16h overnight. In addition, four blank replicates and four replicates of 32 mM ATP were used instead of one blank coupled with several metabolites of varying concentrations. Also, 30 mM

chloroacetamide was replaced with 10 mM iodoacetamide, the concentration of DTT was lowered from 20 mM to 10 mM and the 5X dilution in PBS was replaced with a 10X dilution in ammonium bicarbonate (0.1M). To compensate for the larger volume caused by the increased dilution, 20 % formic acid was used to halt the overnight digestion. Also, all samples were prepared and analyzed in parallel, rather than on separate days, in order to reduce the technical variation between replicates.

(11)

7

5. Results

Each metabolite was tested in two concentrations and each test generated a list of peptides and their differential intensities. Though the experiment was performed for 12 metabolites, problems with access to the LC – MS/MS machine caused the results for 9 of the metabolites to be significantly delayed and they will not be presented here. Instead focus lies on the three first metabolites (ATP, NADP and PEP), the statistical analysis and the results from the evaluation experiment with ATP.

5.1 Initial comparison results

The initial experiment involving ATP, NADP and PEP yielded generally insignificant results. With the q-value threshold of 0.01 used in the original paper there were no significant hits. Even with a lowered threshold of 0.1, meaning a full 10 % of significant results are false positives, only the high concentration of ATP generated results worth mentioning. In addition, an absolute log2-fold change threshold of 1 was used. The comparison yielded 349 significant peptides (shown in detail in table A1) of which 70 originated from 40 proteins that were annotated in the UniProt database as having interactions with ATP. In addition, 33 peptides originated from 23 ribosomal proteins that were not annotated as having interactions with ATP. These represented the largest group of any kind. No significant peptides originated from a protein in the Calvin-Benson cycle, despite both

phosphoribulokinase and phosphoglycerate kinase explicitly using ATP as part of the reaction they catalyze. The average number of peptides detected per pr otein and average peptide length were calculated to 5.5 and 17, respectively.

Figure 1: Volcano plots of the LiP-SMap experiments of ATP, NADP and PEP with the high concentrations in the top panels and the low concentrations in the bottom panels. Each peptide is represented by one point and are separated by log2FC as compared to the untreated sample along the x-axis and by q-value along the y-axis. The q-value threshold is 0.1 and the log2FC threshold is 1. Peptides that clear those thresholds are deemed significant and are shown in red while all other peptides are shown in black.

(12)

8

5.2 Statistical analysis

As the observed data did not match the data produced in the original paper by Piazza et al., several statistical tests were made to understand the cause of the largely insignificant data.

5.2.1 P-value distributions

P-value distributions were generated for every comparison made. An even spread of staple-height indicates that all tests adhere to the null hypothesis, i.e. that none of the peptides change in abundance as a result of the metabolite treatment. A left-skewed distribution on top of a uniform distribution represent the proportion of non-null outcomes. All experiments, excepting the low concentration of ATP, have a left-skewed distribution which indicates that a portion of the tests had non-null outcomes. However, only the high concentration of ATP and the low concentration of PEP had tests that cleared the set thresholds to be considered significant. These are also the only distributions that are continuously declining and clearly level out. The lack of these features in the remaining experiments indicate the presence of conservative p-values, i.e. p-values that are

overestimated. Conservative p-values could occur if the assumptions made by the statistical analysis are not fulfilled.

Figure 2: p-value distributions of each metabolite comparison with peptide counts on the y-axis and p-values on the x-axis.

High concentrations are depicted in the top panels and low concentrations in the bottom panels.

(13)

9

5.2.2 Q-Q Plots

The comparison script from MSstats used in the experiment assumes that the outcome variable (log2-transformed intensity) is normally distributed. To test whether this held true, Q-Q plots were generated to assess the distribution of the data. By plotting the sample data against a theoretical data set that follows a normal distribution, one can see how well the distribution of the sample correlates to a normal distribution. Ideally, the plot would show a roughly straight line, indicating that the data sets followed similar distributions. Instead, the plots for all comparisons show a distinct s-shape which indicates a large number of extreme values.

Figure 3: Q-Q plots for each metabolite and concentration with the sample data on the y-axis and a theoretical data set following normal distribution along the x-axis.

(14)

10

Figure 4: Principal Component Analysis

A: Plot of the 1^st principal component against the 2^nd principal component with clusters defined by metabolite treatments.

B: Plot of the 1^st principal component against the 2^nd principal component with clusters defined by replicate.

5.2.3 Principal component analysis

To assess the variance within treatment groups, PCA was performed with each metabolite and concentration as one group. In an ideal scenario, each metabolite treatment (metabolite – concentration combination) would have had similar results, forming tight and distinct clusters.

However, the variation between replicates was greater than the variation between metabolite treatments, as indicated by the fact that the individual replicates form more distinct clusters than the metabolite treatments in the below graphs.

(15)

11

Figure 5: Pearson correlations

A: The top panel shows the Pearson correlations between each replicate of the same metabolite treatment while the bottom panel shows the Pearson correlations between metabolite treatments within replicates. The replicates are named “Metabolite_X_X” where the first number represents the concentration (1 being high and 2 being low) and the second represents the replicate number.

In each plot, every combination is represented by a specific color.

5.2.4 Pearson correlation by treatment

To further assess the apparent high variance within treatment groups, the Pearson correlation of each possible pairwise combination of replicates was calculated. Ideally, the replicates with the same metabolite – concentration combination would correlate highly with one another, while less so with other replicates. However, correlation between replicates of the same metabolite treatments were generally very low while also lower than between replicates of different combinations, as can be seen in the graphs below.

(16)

12

5.3 Evaluation experiment comparison results

The evaluation experiment with the updated protocol yielded 234 (visualized in figure 6A and shown in detail in table A2) significant peptides with a q-value threshold of 0.01 and an absolute log2-fold change threshold of 1. 46 of the significant peptides were annotated as having interactions with ATP in the UniProt database and 86 of the significant peptides that were not annotated originated from 35 ribosomal proteins. Phosphoribulokinase and phosphoglycerate kinase were still not detected as interacting with ATP. However, fructose-1,6-bisphosphatase, which is an enzyme involved in the Calvin-Benson cycle, was detected as having a significant interaction. The number of significant peptides and their q-values were very similar to the results of Piazza et al., as opposed to the initial experiment. The Q – Q plot (figure 6C) has a clear linear shape, indicating that the input data (the log2-intensities of the detected peptides) is normally distributed, fulfilling the assumption required by the statistical analysis. These findings are echoed in the p-value distribution (figure 6B) which is clearly left-skewed and declines continuously and sharply before clearly leveling out, indicating a low number of conservative p-values. The Pearson correlation between replicates (figure 6D) were notably higher in this experiment as compared to the initial experiment, though as there is only one metabolite treatment, it is not possible to compare across treatments.

Figure 6: Evaluation experiment results

A: Volcano plot of the LiP-SMap evaluation experiment. Each peptide is represented by one point with significant peptides in red and the remainder in black. They are separated by log2FC on the x-axis and by q-value on the y-axis.

B: The P-value distribution of the evaluation experiment with peptide counts on the x-axis and p-values on the y-axis.

C: Q – Q plot with sample data on the y-axis and a theoretical dataset following a normal distribution on the x-axis.

D: Bar plot of the Pearson correlation between each replicate in the experiment.

(17)

13

6. Discussion

The purpose of this study was to identify interactions between metabolites of interest and proteins in the Calvin-Benson cycle. This was partially accomplished in the form of the evaluation experiment, but the main focus of the study shifted partway through towards developing a protocol that

produced satisfactory results. This was also accomplished through the evaluation, though there are some experiments left to run before that process is complete.

6.1 Initial experiment

The initial experiment yielded no significant results at a q-value threshold of 0.01 which in theory could mean two things, either that there were no significant interactions or errors were present in the experiment, by design or through execution. Part of the reason why ATP was chosen as a metabolite was that it was known to interact with many proteins in the proteome and that the data generated could be compared to that of Piazza et al. As such, it was immediately apparent that the results were unlikely to be true and that the fault likely rested within the design of the experiment and analysis. The statistical analysis established that the variance between replicates of the same treatment was unexpectedly high. The observed variance could then be of either biological or technical nature and it was impossible to distinguish between the two since there was only one technical replicate for each biological replicate. However, it was deemed unlikely that the bacteria would express significantly different proteomes since the cultivations were done in parallel in identical environments with only minor variations in timing and amounts of feed. More likely then, that each replicate had somehow been treated differently during the experimental procedure.

In the initial experiment the statistical analysis was done through pair-wise comparisons, meaning that while there were four replicates of each metabolite treatment, they were analyzed one by one against singular separate blanks. This design proved inadequate to handle the unexpectedly high technical variance observed. Indeed, since the technical variation was established to be much higher than the variation between different metabolite treatments it follows that most of the observed differences in peptide abundances were caused by technical variation rather than the metabolite treatments. In addition, only one technical replicate was generated for each biological replicate which increased the risk of extreme values, given that the technical variation was so high. Increased numbers of extreme values led to data that was not normally distributed, which caused increased numbers of conservative p-values. Subsequently, the applied FDR method could not cope and returned significantly higher q-values, rendering most of the results insignificant.

Other than general imprecision (variations in pipetted volumes and measured times etc.), a possible source of variation between technical replicates could be that the peptide mixes were not digested equally by trypsin and LysC, giving rise changes in peptide abundances which are not related to the metabolite treatment.

As such, several steps were taken to reduce technical variation for the evaluation experiment. To improve the statistical analysis, four technical replicates were generated from the biological replicate and compared group-wise to four blank samples. In addition, all samples for the evaluation

experiment were prepared and analyzed in parallel, possibly removing some variation due to imprecision. A pre-digestion step with lysC was added before the complete digestion by trypsin and the protein concentration of the samples was reduced by 75 % to increase the specificity of the enzymes. Additionally, PBS was exchanged for ammonium bicarbonate which has more appropriate pH for trypsin. Also, the concentration DTT was lowered and CAA was exchanged for IAA to improve the alkylation process, potentially improving the performance of the LC – MS/MS.

(18)

14 These actions proved effective as the Q – Q plot for the evaluation experiment (figure 6C) indicated a proper normal distribution which in turn likely led to fewer conservative p-values, as observed in the p-value distribution of the evaluation experiment (figure 6B). In addition, the higher Pearson

correlation between technical replicates in the evaluation experiment (figure 6D) indicates that the protocol adjustments were helpful in reducing technical variation.

6.2 Evaluation experiment

The evaluation experiment exhibited much improved results with a clear normal distribution across all replicates, lower q-values, and higher, more consistent correlation between replicates. However, it is still unclear whether the original variance came about through biological or technical variance as the evaluation experiment negated both through using only the single biological replicate and preparing all technical samples in parallel.

Over one-third of the significant peptides originated in ribosomal proteins even though none of them were annotated as having interactions with ATP. This could be caused by the binding of ATP to Mg²⁺

ions in the sample solutions. The positive ion neutralizes the negative charge of ATP and is required for ATP to be biologically active. The Mg²⁺ ions are also critical to the stability of the ribosomes and if the concentration of ATP is high enough, there might not remain sufficient unbound ions in solution to maintain ribosome stability. Escherichia coli has been shown to lower the concentration of ATP in the cells under Mg²⁺ limitation in a bid to sustain translation (40). As such, a probable explanation is that ATP causes a lack of Mg²⁺ ions, which in turn causes the ribosomal proteins to destabilize, exposing new peptides to proteolysis, giving rise to a different peptide distribution. This is further indicated by the fact that 83 of the 85 peptides had negative log2-fold changes. Negative log2-fold changes are generated when peptides that are not cut in the blank sample are cut in the sample treated with a metabolite, which is generally because new peptides were exposed due to

conformational changes. In contrast, positive log2-fold changes are caused by the metabolite binding to a protein and physically blocking proteolysis or by peptides being hidden from proteolysis through conformational changes. In short, the large percentage of negative log2-fold changes among the peptides originating from ribosomal proteins indicates that they were all subject to conformational changes. This chimes well with the theory that they were destabilized through lack of Mg²⁺ ions caused by a high concentration of ATP.

The results also showed interactions with 46 proteins that were annotated as having interactions with ATP in the UniProt database. Significantly more than 46 proteins are known to interact with ATP which calls into question why these proteins are missing from the results. One possible explanation is that proteins may require certain circumstances to be catalytically active. In essence, proteins might not interact with a specific metabolite if other co-factors are missing, effectively preventing any significant hits through the LiP-SMap method. These co-factors could be activators, free ions, or additional substrates. Since ATP is seldom the sole substrate of a protein, the absence of the other substrates might prevent ATP from interacting with the protein, depending on the order of which the substrates bind to the protein. One example of this could be found in phosphoribulokinase, an enzyme in the Calvin-Benson cycle using ATP as a substrate that did not register as a significant protein. Phosphoribulokinase has been shown to depend on NADPH for its activity in cyanobacteria, and the lack of said co-factor might have prevented the protein from interacting with ATP during this experiment (27). It is worth noting that the need for multiple co-factors might not be as big an issue in the case of allosteric regulation.

Another, perhaps more severe problem that might cause true interactions to remain undetected is the low peptide coverage observed in the experiment. The average peptide length was roughly 17 amino acids, and the average number of peptides detected for each protein was roughly 5.5. At best,

(19)

15 that represents slightly less than a hundred amino acids. Since the average protein length in C.

necator is 326, this represents less than one third of the protein sequence. Hence, it is possible that many interactions remain undetected simply because the relevant peptide is not detected. Perhaps the coverage can be improved simply by increasing the number of replicates. However, it is crucial to note that none of the above problems invalidate the significant interactions that were detected.

Indeed, the evaluation experiment indicated 103 novel interactions, though several of these might not be novel as the annotations in the UniProt database are not entirely consistent.

In conclusion, a working protocol has been developed despite initial setbacks and hundreds of metabolite-protein interactions have been identified for ATP, many of which are novel. While there is yet more development to be done, the foundation has been made for future experiments to

efficiently identify protein-metabolite interactions.

7. Future perspectives

In the short term, a test will have to be performed to assess whether the observed variance is caused by biological or technical variance. This will most likely be done by running an experiment similar to the described evaluation experiment, but with two biological replicates and thus twice the number of samples. Preferably, ATP will be used again so that the results can be compared to the ones

generated in this study. The degree to which the same peptides are detected again could help determine whether increasing the number of replicates is likely to improve protein coverage. After these issues have been investigated and the workflow adjusted accordingly, it will be possible to efficiently run LiP-SMap experiments for many metabolites against many different organisms.

In the long term, such research could lead to the mapping of allosteric regulation of the

Calvin-Benson cycle. This in turn could lead to manipulation of said regulation for any number of desired effects, for example increased carbon flux through the cycle. Such findings could prove very valuable for metabolic engineers worldwide and potentially offer significant aid in the fight against climate change.

8. Acknowledgements

This project could not have been completed without the aid of my two supervisors at KTH, Paul Hudson and Jan Karlsen. I would like to extend a huge special thanks to David Kotol of the Edfors group for running all of our samples on their LC – MS/MS apparatus and for helping us out so readily when we needed it. I would also like to thank everyone in the Hudson and Jonas groups for all the help and the welcoming atmosphere.

(20)

16

9. References

1. Core Writing Team, Pachauri RK, Meyer LA. Climate Change 2014: Synthesis Report.

Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. IPCC, Geneva, Switzerland. 2014. 151 p.

2. Searchinger TD, Hamburg SP, Melillo J, Chameides W, Havlik P, Kammen DM, et al. Fixing a Critical Climate Accounting Error. Science (80- ). 2009;326(October):527–8.

3. Field CB, Campbell JE, Lobell DB. Biomass energy: the scale of the potential resource. Trends Ecol Evol. 2008;23(2):65–72.

4. Luo L, Van Der Voet E, Huppes G, Udo De Haes HA. Allocation issues in LCA methodology: A case study of corn stover-based fuel ethanol. Int J Life Cycle Assess. 2009;14(6):529–39.

5. Oh YK, Hwang KR, Kim C, Kim JR, Lee JS. Recent developments and key barriers to advanced biofuels: A short review. Bioresour Technol [Internet]. 2018;257(December 2017):320–33.

Available from: https://doi.org/10.1016/j.biortech.2018.02.089

6. Rentizelas AA, Tolis AJ, Tatsiopoulos IP. Logistics issues of biomass: The storage problem and the multi-biomass supply chain. Renew Sustain Energy Rev. 2009;13:887–94.

7. Baral NR, Kavvada O, Mendez-Perez D, Mukhopadhyay A, Lee TS, Simmons BA, et al. Techno- economic analysis and life-cycle greenhouse gas mitigation cost of five routes to bio-jet fuel blendstocks. Energy Environ Sci. 2019;12(3):807–24.

8. Chisti Y. Biodiesel from microalgae. Biotechnol Adv [Internet]. 2007;25(3):294–306. Available from: http://dx.doi.org/10.1016/j.biotechadv.2007.02.001

9. Parmar A, Kumar N, Pandey A, Gnansounou E, Madamwar D. Cyanobacteria and microalgae : A positive prospect for biofuels. Bioresour Technol [Internet]. 2011;102(22):10163–72.

Available from: http://dx.doi.org/10.1016/j.biortech.2011.08.030

10. Rastogi RP, Pandey A, Larroche C, Madamwar D. Algal Green Energy – R&D and technological perspectives for biodiesel production. Renew Sustain Energy Rev [Internet]. 2018;82(April 2017):2946–69. Available from: https://doi.org/10.1016/j.rser.2017.10.038

11. Bi C, Su P, Müller J, Yeh Y, Chhabra SR, Beller HR, et al. Development of a broad-host synthetic biology toolbox for ralstonia eutropha and its application to engineering hydrocarbon biofuel production. Microb Cell Fact. 2013;12:107.

12. Andersson I, Backlund A. Structure and function of Rubisco. Plant Physiol Biochem.

2008;46(3):275–91.

13. Raines CA. The Calvin cycle revisited. Photosynth Res. 2003;75(1):1–10.

14. Hasunuma T, Kikuyama F, Matsuda M, Aikawa S, Izumi Y. Dynamic metabolic profiling of cyanobacterial glycogen biosynthesis under conditions of nitrate depletion. J Exp Bot.

2013;64(10):2943–54.

15. Dempo Y, Ohta E, Nakayama Y, Bamba T, Fukusaki E. Molar-Based Targeted Metabolic Profiling of Cyanobacterial Strains with Potential for Biological Production. Metabolites.

2014;4:499–516.

16. Yoshikawa K, Hirasawa T, Ogawa K, Hidaka Y, Nakajima T, Furusawa C, et al. Integrated transcriptomic and metabolomic analysis of the central metabolism of Synechocystis sp . PCC

(21)

17 6803 under different trophic conditions. Biotechnol J. 2013;8(May):571–80.

17. Shastri AA, Morgan JA. A transient isotopic labeling methodology for 13 C metabolic flux analysis of photoautotrophic microorganisms. Phytochemistry. 2007;68:2302–12.

18. Bennett BD, Kimball EH, Gao M, Osterhout R, Dien SJ Van, Rabinowitz JD. Absolute metabolite concentrations and implied enzyme active site occupancy in Escherichia coli. Nat Chem Biol.

2009;5(8).

19. Asplund-samuelsson J, Janasch M, Hudson EP. Thermodynamic analysis of computed pathways integrated into the metabolic networks of E . coli and Synechocystis reveals

contrasting expansion potential. Metab Eng [Internet]. 2018;45(April 2017):223–36. Available from: https://doi.org/10.1016/j.ymben.2017.12.011

20. Marcus Y, Harel E, Kaplan A. Adaptation of the Cyanobacterium Anabaena variabilis to Low CO Concentration in Their Environment ’. Plant Physiol. 1983;71:208–10.

21. Bolten CJ, Wittmann ÆC. Appropriate sampling for intracellular amino acid analysis in five phylogenetically different yeasts. Biotechnol Lett. 2008;30:1993–2000.

22. Hauf W, Schmid K, Gerhardt ECM, Huergo LF, Forchhammer K. Interaction of the Nitrogen Regulatory Protein GlnB (PII) with Biotin Carboxyl Carrier Protein (BCCP) Controls Acetyl-CoA Levels in the Cyanobacterium Synechocystis sp. PCC 6803. Front Microbiol [Internet].

2016;7:1700. Available from: https://www.frontiersin.org/article/10.3389/fmicb.2016.01700 23. Stitt M, Sulpice R, Keurentjes J. Metabolic Networks: How to Identify Key Components in the

Regulation of Metabolism and Growth. Plant Physiol [Internet]. 2010 Feb 1;152(2):428 LP – 444. Available from: http://www.plantphysiol.org/content/152/2/428.abstract

24. Abdelal ATH, Schlegel HG. Purification and regulatory properties of phosphoribulokinase from Hydrogenomonas eutropha H 16. Biochem J. 1974;139(3):481–9.

25. Division E, Field M. Regulation in the Chemolithotroph Thiobacillus neapolitanus: Fructose- 1,6-Diphosphatase. Arch Microbiol. 1973;93:23–8.

26. Hackenberg C, Hakanpäa J, Cai F, Antonyuk S, Eigner C, Meissner S, et al. Structural and functional insights into the unique CBS–CP12 fusion protein family in cyanobacteria. Proc Natl Acad Sci U S A. 2018;115(27):7141–6.

27. Tamoi M, Miyazaki T, Fukamizo T, Shigeoka S. The Calvin cycle in cyanobacteria is regulated by CP12 via the NAD(H)/NADP(H) ratio under light/dark conditions. Plant J. 2005;42(4):504–13.

28. Burnap RL, Hagemann M, Kaplan A. Regulation of CO2concentrating mechanism in cyanobacteria. Life. 2015;5(1):348–71.

29. Daley SME, Kappell AD, Carrick MJ, Burnap RL. Regulation of the cyanobacterial CO2- concentrating mechanism involves internal sensing of NADP+ and α-ketogutarate levels by transcription factor CcmR. PLoS One. 2012;7(7):1–10.

30. Zhang CC, Zhou CZ, Burnap RL, Peng L. Carbon/Nitrogen Metabolic Balance: Lessons from Cyanobacteria. Trends Plant Sci [Internet]. 2018;23(12):1116–30. Available from:

https://doi.org/10.1016/j.tplants.2018.09.008

31. Gardemann A, Stitt M, Heldt HW. Regulation of spinach ribulose 5-phosphate kinase by 3- phosphoglycerate. FEBS Lett. 1982;137(2):213–6.

32. Opitz R, Schlegel HG. Allosteric inhibition by phosphoenolpyruvate of glucose-6-phosphate dehydrogenase from bacteria and its taxonomic importance. Biochem Syst Ecol.

(22)

18 1978;6(3):149–55.

33. Tlapaksimmons VL, Reinhart GD. Comparison of the Inhibition by Phospho(enol)Pyruvate and Phosphoglycolate of Phosphofructokinase from B. stearothermophilus. Vol. 308, Archives of Biochemistry and Biophysics. 1994. p. 226–30.

34. Glenn K, Smith KS. Allosteric regulation of Lactobacillus plantarum xylulose 5-

phosphate/fructose 6-phosphate phosphoketolase (Xfp). J Bacteriol. 2015;197(7):1157–63.

35. Petterson U-R. Identification of possible two-reactant sources of oscillations in the Calvin photosynthesis cycle and ancillary pathways. Eur J Biochem. 1991;198:613–9.

36. Piazza I, Kochanowski K, Cappelletti V, Fuhrer T, Noor E, Sauer U, et al. A Map of Protein- Metabolite Interactions Reveals Principles of Chemical Communication. Cell [Internet].

2018;172(1–2):358-372.e23. Available from: https://doi.org/10.1016/j.cell.2017.12.006 37. Searle BC, Pino LK, Egertson JD, Ting YS, Maccoss MJ, Maclean BX, et al. Chromatogram

libraries improve peptide detection and quantification by data independent acquisition mass spectrometry. Nat Commun. 2018;9.

38. Gessulat S, Schmidt T, Zolg DP, Samaras P, Schnatbaum K, Zerweck J, et al. Prosit: proteome- wide prediction of peptide tandem mass spectra by deep learning. Nat Methods [Internet].

2019;16(June). Available from: http://dx.doi.org/10.1038/s41592-019-0426-7

39. Choi M, Chang CY, Clough T, Broudy D, Killeen T, MacLean B, et al. MSstats: An R package for statistical analysis of quantitative mass spectrometry-based proteomic experiments.

Bioinformatics. 2014;30(17):2524–6.

40. Pontes MH, Yeom J, Groisman EA, Pontes MH, Yeom J, Groisman EA. Reducing Ribosome Biosynthesis Promotes Translation during Low Mg 2 + Stress Article. Mol Cell [Internet].

2016;64(3):480–92. Available from: http://dx.doi.org/10.1016/j.molcel.2016.05.008

(23)

19

10. Appendix

Table A1: All significant peptides detected for the high concentration of ATP in the initial experiment.

(24)

20

(25)

21

(26)

22

(27)

23 1

(28)

24

(29)

25

Table A2: All significant peptides detected for the evaluation experiment.

’

(30)

26

(31)

27

(32)

28

(33)

29 www.kth.se

Identification of metabolite-protein interactions among enzymes of the Calvin Cycle in a CO2-fixing bacterium