We retrieved from the RefSeq database at NCBI the genomes of B. quintana str Toulouse (accession number NC_005955.1) and Bartonella henselae str. BM1374163 (accession number NZ_HG965802.1), the representative strain of this species. B. henselae is the most closely related species to B. quintana within the genus (Alsmark et al., 2004). We used OrthoVenn3 (Sun et al., 2023) to visualize and compare both genomes to solve some annotation problems.

The metabolic network reconstruction and refinement of the model were performed in several steps. An initial annotation of the genome sequence with RAST (Rapid Annotation using Subsystem Technology; Aziz et al., 2008; Brettin et al., 2015) was made and provided to ModelSEED (Overbeek et al., 2014; Seaver et al., 2021) to obtain a draft metabolic network with a general biomass reaction that needs to be curated. Many databases were used to ensure accuracy in the gap-filling of this preliminary metabolic network: KEGG (Ogata et al., 1999; Kanehisa et al., 2016; Kanehisa et al., 2017), MetaCyc and BioCyc (Caspi et al., 2018) have extensive information on genes, reactions, and pathways; BLAST (Altschul et al., 1990; Altschul et al., 1997) allows to search for orthologs and pseudogenes; BRENDA (Jeske et al., 2019) and ExPASy (Gasteiger et al., 2003) enable the search for enzymatic activities absent in our model organism. The addition or removal of reactions and genes in the model was based on genomic information, taking into account all complete annotated protein-coding genes for which there was evidence of a given activity in this organism or another one of the same genus. Enzymes with a broad substrate spectrum capable of compensating for the absence of other proteins were also considered. All reactions that are part of unconnected modules were removed from the model to avoid inconsistencies.

Modifications were made to the draft metabolic model using the COBRApy module in Python (Ebrahim et al., 2013). We used MEMOTE to determine the quality of the obtained GEM (Lieven et al., 2020). Cytoscape (Shannon et al., 2003) was used to interactively visualize the model, perform various topological analyses of the network, and create high-quality illustrations to display the results.

The biomass composition is a crucial factor in genome-scale metabolic models, significantly influencing their predictive accuracy and effectiveness. Flux balance analysis was performed to determine the optimal metabolic flux distributions, maximizing the biomass reaction as an objective function under steady-state assumption (Orth et al., 2010). Since there is no biomass composition information available for the genus Bartonella, we compared the biomass reaction prediction for the model with the well-documented biomass composition of E. coli model iJO1366 as a reference to adjust the biomass composition of our model. The components present in iJO1366 and not in the Bartonella biomass reaction were added with the corresponding stoichiometry; after manual curation of the network, compounds that could not be produced by the model were removed from the biomass equation. The stoichiometric values for dNTPs were calculated using the BOFdat software (Lachance et al., 2019). Due to the lack of kinetic data, we performed two different analyses: one without any constraint to the model input and output, and another with the only restriction of an oxygen intake flux of 15 mmol/gDW/h^-1, which is the maximum flux of oxygen established for E. coli (Varma and Palsson, 1994).

In addition, we performed a genomic analysis of the transporter components involved in the model. We also generated single and double in silico knockouts of genes associated with transport to determine the changes in growth yield. Both FBA and in silico mutation simulations were performed using COBRApy.

B. quintana str Toulouse (CIP 1033739) was provided by the Biological Resource Center of Institut Pasteur (CRBIP). All experiments started by streaking B. quintana from 100 μL of one- or two-passage frozen stock onto chocolate agar plates, made with Columbia blood agar base (Oxoid) supplemented with 5% defibrillated ovine blood (ThermoFischer Scientific). 2-oxoglutarate (500 mg/L, K), thiamine (1 mg/L, T) and folate (0.01 mg/L, F) were added when indicated. All cultures were plated in triplicate and incubated at 35°C in a 5% CO₂ chamber, and the cells were harvested after 7 days with brain-heart infusion (BHI) broth.

To determine the cell growth in the different media, we prepared plate cultures in triplicate, harvested all cells in each plate, and determined the optical density at 600 nm (OD₆₀₀) to estimate the initial concentration. Subsequently, we prepared a series of dilutions in liquid BHI, ranging from 10^–1 to 10^–7, and duplicate control chocolate agar plates were inoculated with dilutions from 10^–3 to 10^–7 for the cell count of the initial culture.

From the 10^–3 and 10^–4 dilutions obtained from these plates, triplicate inoculations of 100 μL were made for each condition under investigation, along with the control medium. After a 7-day incubation period, colonies were collected in liquid BHI, and the OD₆₀₀ was determined. The optical densities of all conditions were normalized, and 5 μL dots of dilutions ranging from 10^–3 to 10^–7 from each culture were spotted in a gridded distribution to assess the proportion of the cells that remain viable. The protocol is schematized in Figure 1.

Cells grown in two agar plates were harvested in 2 mL of PBS buffer and centrifuged at 4°C for 10 min at 3500 g. The pellet was resuspended in 500 μL of PBS and washed three times under the same conditions. The sample was resuspended in 300 μL of solubilization buffer (8 M urea, 2 M thiourea, 4% V/V CHAPS, 20 M Tris, Roche Complete Protease Inhibitor 1x) to release and solubilize the total cellular proteins. Finally, the sample was centrifuged at 4°C for 5 min at 10 000 g, and the supernatant was collected.

A comparative proteomic analysis of samples obtained in the different culture conditions was conducted using the SWATH mass spectrometry method (Collins et al., 2017; Ludwig et al., 2018). Mass spectrometry was performed at the Central Service for Experimental Research (SCSIE) of the University of Valencia. The proteomic data were analyzed in R with the Differential Enrichment analysis of the Proteomics data (DEP) package (Zhang et al., 2018).

The first draft metabolic model of B. quintana str Toulouse was obtained using the SEED server of ModelSEED, based on a newly obtained annotation by RAST. The genome of B. henselae str. BM1374163 (Alsmark et al., 2004) was used to search for orthologous genes to fix some wrong annotations in genes involved in metabolic functions, thus helping to close some gaps using COBRApy (Ebrahim et al., 2013). This Python module enables the addition, deletion, and modification of each model element. These two organisms shared 95.3% of protein-coding sequences and 79.2% of collinear genes (Supplementary Figure S1). These similarities allowed for a more accurate and efficient search for the addition of metabolic reactions and the rapid reconstruction of the draft model for B. quintana, enabling us further to refine it into a well-curated, specific model.

Biomass composition can substantially affect fundamental model predictions (Beck et al., 2018), and for this reason, it needs to be integrated into metabolic models, which was also performed with COBRApy. The list of compounds included in the modified biomass equation, as well as their coefficients, are provided in Supplementary Table S1.

The genes and reactions included in the model were manually curated to get a refined metabolic model (Figure 2; Supplementary Table S2), composed of 300 protein-coding genes associated with metabolism (representing the 23.1% of the total of genes annotated in the genome) involved in 370 reactions, 49 of which are exchange reactions and 44 of which are involved in transport. 355 metabolites are included in the model, 343 of which (96.6%) are included in at least two reactions and 139 (30,1%) in three or more reactions. The quality analysis performed with MEMOTE assigned it a score of 85% (Supplementary Material S1), similar to the score obtained for other GEMs of model organisms (van ‘t Hof et al., 2022). The unrestricted model (Bquint GEM) predicts a growth rate of 17.1 1/h and the oxygen-restricted model (Bquint GEM O2) predicts a growth rate of 4.2 1/h.

Related to bioenergetics and carbon sources, we observed that there are no transporters for simple sugars like glucose or fructose. Furthermore, this organism has an incomplete glycolysis pathway, a fact that has also been observed in B. henselae, lacking both hexokinase (EC 2.7.1.1) and phosphofructokinase (EC 2.7.1.11) (Canback et al., 2002; Chenoweth et al., 2004). The remaining pathway reactions are still present in the network because they are reversible and also act in gluconeogenesis. This second pathway provides the precursors for the pentose phosphate pathway, which produces the ribulose phosphate needed for nucleotide synthesis. The Krebs cycle is complete, and it is the primary producer of reductive power (NADH).

The model includes 11 of the 20 proteogenic amino acids synthesis pathways (Ala, Asn, Asp, Gln, Glu, Gly, Lys, Phe, Pro, Ser, Tyr) and has specific transporters for another 6 (Arg, His, Ile, Leu, Thr, Val). There are no particular transporters for the remaining amino acids (Cys, Met, Trp), but they could be taken up through the ABC system for the import of dipeptides and oligopeptides. Subsequently, these compounds can be hydrolyzed to yield individual amino acids.

Regarding vitamins and cofactors, most of them have incomplete biosynthetic pathways. Pantothenate (vitamin B5), the precursor of coenzyme A, is the only one with a complete path, and only a few key enzymes are missing for the biosynthesis of NAD, pyridoxal phosphate, and ubiquinone. In the case of NAD, the salvage pathway is entirely functional, so this bacterium is capable of recycling intermediates for its synthesis. Any of these vitamins and cofactors have a specific transporter to incorporate them or any of the intermediates of the pathways.

In order to elucidate the essential compounds for the growth of B. quintana that could be incorporated from the culture medium, a comprehensive analysis of all reactions facilitating the transport of molecules from the extracellular environment was undertaken. In the absence of kinetics data, this analysis is sequence-based and aimed at improving the functional annotation of transporters encoded in the genome. Consequently, nine additional transport reactions that were absent from the original annotation were incorporated into the model. Among the added reactions, the branched-chain amino acid (BCAA) transporter and the heme group transporter stand out, as these compounds are pretty abundant in blood. The resulting model comprises 57 genes associated with transport that encode the proteins that constitute 25 transporters able to perform a total of 42 transport reactions. The proteins involved in the electron transport chain are not included. We have computed the flux variation in in silico mutants for each of these transporters, both individually and in combinations of two, to measure their impact on the growth of the organism. The outcomes have been categorized as essential, moderate, and non-essential based on the level of conservation in biomass production (<10%, 10%–80%, and >80%, respectively; Supplementary Table S3). We observed that 18 out of the 25 genes are essential, as their absence completely prevents bacterial growth, notably including transporters for most amino acids and ions. When a double mutant analysis was performed, there were no significant differences except for potassium transport. That is due to the presence of redundant transporters that share substrates and can complement each other. Thus, only in the absence of both does a significant deficit for the transported molecule occur in the cell, preventing its growth.

Based on the in silico knockout analyses, we searched for important compounds for cell growth that are present in low concentrations in commercial media and could be added to improve growth. We calculated the exchange fluxes in the model, which are linked to the transport reactions of molecules found in the medium, to incorporate them into the cell.

As a next step in the model reconstruction, we streamlined the overall flux of the models towards biomass production, obtaining a list of thirty-seven essential compounds from the environment required for the growth of the models (Table 1). Notably, in the absence of glucose or other basic sugar transporters, most influxes are associated with amino acids, as well as intermediates of the Krebs cycle. A high influx of 2-oxoglutarate and glutamine is observed, both of which are related to transamination reactions and nitrogen metabolism. It is noteworthy that glutamine and the BCAAs (valine, leucine, and isoleucine) have a relatively high concentration in the blood compared to other amino acids (Brody, 1999). The main differences we observed between Bquint GEM and Bquint GEM O2 are related to amino acid metabolism. When oxygen intake is restricted, aspartate, serine and D-alanine intake rises, while glutamate and glycine decreases. In the case of glutamate, the observed decrease is compensated because it participates in multiple transamination reactions, allowing its biosynthesis from aspartate and glutamine, which could explain the increase in aspartate intake. Regarding glycine, it plays a role in the one-carbon pool metabolism and it is synthesized from serine and folate derivatives. Lastly, D-alanine serves as a key precursor in bacterial cell wall biosynthesis and can be converted into L-alanine by alanine racemase (EC 5.1.1.1).

When comparing these 37 essential compounds with the composition of Columbia agar base and the blood used in the standard preparation of culture media for blood-feeding microorganisms, we observed that there are four compounds absent in this medium: 2-oxoglutarate, thiamine, folate, and β-alanine. In the case of 2-oxoglutarate, its input flux is the second highest after oxygen, which reflects its relevance in the growth of this organism. Additionally, this compound is a crucial element in sugar and amino acid metabolism. The other three compounds have much smaller fluxes but participate in very important reactions in bacterial physiology.

We selected three out of these four compounds for the experimental validation of an improved growth media for B. quintana, to simplify the validation protocol. β-alanine was excluded because preliminary tests showed no changes in Bartonella’s growth performance when it was added individually (data not shown).

Seven different types of supplemented chocolate agar plates were produced, by adding the three selected compounds separately (2-oxogulutarate, K; folate, F; thiamine, T), in paired combinations (FK, FT, and KT), altogether (FKT), and the protocol described in Figure 1 was followed to check for growth differences between them. OD₆₀₀ measurements were performed after 7 days of incubation. The experiment was repeated six times. Growth problems were observed in the plates inoculated from the 10^–4 dilution of the original culture, with instances where none of the replicates in an experimental condition showed bacterial growth. That led us to exclude these data from the statistical analysis due to the lack of sufficient measurements.

When comparing the data from plates inoculated from the 10^–3 dilution to control chocolate agar plates, no significant differences were observed in any of the seven types of plates (Figure 3A). During the analysis of these data, some replicates had to be discarded because the initial cultures used for comparison had an OD₆₀₀ above 2. In these cases, the dilutions reached saturation and generated inconsistencies between technical replicates from each medium. All data, including that from the discarded experiments, are provided in Supplementary Table S4.

Cell viability was also measured to compare the outcome in chocolate agar and supplemented media. A cell count was performed after 7 days to do so, starting with an adjusted OD₆₀₀ for all media (Figure 3B). Surprisingly, the viability in media with a single added compound (F, K, T) barely differs from the viability in control media. It is similar or slightly lower in the media with two compounds added (FK, FT, and KT), and, in the case of the medium with all three compounds added (FKT), there is a decrease of nearly two orders of magnitude in the number of cells.

We also analyzed if there were differences in protein levels among B. quintana cultures grown in the different media used in this study. We identified 820 proteins by mass spectrometry. A principal component analysis (PCA) of the top 500 proteins (Figure 4A) from our samples revealed differences among replicates of the same condition, which correlated with our previous growth experiments, where the variance among biological replicates was high, making it challenging to observe significant differences between the samples. Nevertheless, 16 of the identified proteins appeared to be differentially expressed compared to the control (Table 2). Proteins with significant changes are defined by user-defined cut-offs based on the adjusted p-value, which should be lower than 0.05, and the absolute fold-change, which should be larger than 0.5. The cut-off of 0.5 is roughly corresponding to the standard deviation of the fold changes in the FKT vs. C comparison.

A clustering analysis of these differentially expressed proteins (Figure 4B) revealed that all media containing 2-oxoglutarate grouped (cluster K). This group includes proteins related to cell wall modification and tRNA modifications, as well as changes in proteins involved in the cell cycle, cell adhesion, and morphology. There are also some stress-related proteins and antioxidant synthesis. There are not many differences between the control (C) and those media supplemented with folate (F) and thiamine (T). A slight increase in proteins involved in purine biosynthesis is observed, likely related to ATP and GTP synthesis (PurN), as well as proteins associated with RNA metabolism (MiaA, MnmE). Proteins related to cell wall formation (Q649_00507, Q649_00526, Q649_00219) are underexpressed. In contrast, those related to cell wall synthesis and turnover (Q649_01321, Q649_01278) are overexpressed in cluster K. Triple compound medium (FKT) only differs from cluster K due to a highly underexpressed protein, an NPLC/P60 endopeptidase related to cell growth and morphology that causes cell aggregation (Duchêne et al., 2019).