Bacterial cross-feeding can promote gene retention by lowering gene expression costs

Abstract

Gene loss is expected in microbial communities when the benefit of obtaining a biosynthetic precursor from a neighbor via cross-feeding outweighs the cost of retaining a biosynthetic gene. However, gene cost primarily comes from expression, and many biosynthetic genes are only expressed when needed. Thus, one can conversely expect cross-feeding to repress biosynthetic gene expression and promote gene retention by lowering gene cost. Here we examined long-term bacterial cocultures pairing Escherichia coli and Rhodopseudomonas palustris for evidence of gene loss or retention in response to cross-feeding of non-essential adenine. Although R. palustris continued to externalize adenine in long-term cultures, E. coli did not accumulate mutations in purine synthesis genes, even after 700 generations. E. coli purine synthesis gene expression was low in coculture, suggesting that gene repression removed selective pressure for gene loss. In support of this explanation, R. palustris also had low transcript levels for iron-scavenging siderophore genes in coculture, likely because E. coli facilitated iron acquisition by R. palustris. R. palustris siderophore gene mutations were correspondingly rare in long-term cocultures but were prevalent in monocultures where transcript levels were high. Our data suggests that cross-feeding does not always drive gene loss, but can instead promote gene retention by repressing costly expression.

Graphical abstract

INTRODUCTION

Changes to the genetic inventory of a given microbe can be influenced by living in a community. An example of this influence is described in the Black Queen Hypothesis (BQH). Based on the card game Hearts, where players avoid holding the costly Queen of Spades, the BQH posits that gene loss (herein we do not distinguish loss-of-function mutations from gene deletions for convenience) will occur when acquiring a resource from neighbor is less costly than producing that resource (1). The BQH involves a producer that generates a public resource, and a beneficiary that uses the resource and is thus subject to beneficial gene loss. The BQH is analogous to the emergence of cheaters that lose genes and benefit by exploiting a public resource. However, unlike a cheater, a beneficiary is less likely to be the same species as the producer and is less likely to harm the producer (1, 2). The BQH, or analogous processes, are now commonly used to explain the prevalence of auxotrophs, which cannot synthesize one or more essential nutrients (3, 4). Indeed, several groups have observed that auxotrophs carrying biosynthetic gene deletions have a fitness advantage over wild-type cells when corresponding biosynthetic precursors are provided (5–9).

A confounding aspect of BQH prediction of gene loss is that nutrient availability can also repress gene expression and thus lower the cost of a gene (3, 10, 11); most of a gene’s cost comes from its expression (12–16), and thus repressing gene expression could have similar cost-savings as a loss-of-function mutation. Thus, one can arrive at an opposite prediction wherein cross-feeding could promote gene retention by repressing costly gene expression. This prediction is well-documented from studies of cheaters. For example, cheaters that had lost genes for iron-scavenging siderophores emerged when iron was scare and siderophore genes were expressed, but gene loss was not observed when iron was plentiful and siderophore genes were repressed (11, 17, 18). Thus, nutrient availability alone might not be a good predictor for gene loss and the emergence of beneficiaries.

Previously, we developed cocultures pairing fermentative Escherichia coli and phototrophic Rhodopseudomonas palustris, wherein each species is dependent on the other for essential nutrients; E. coli ferments glucose into organic acids that provide essential carbon to R. palustris while R. palustris converts N₂ gas into NH₄⁺, providing essential nitrogen to E. coli (19). More recently, we discovered that R. palustris also externalizes adenine at levels that can support an E. coli purine auxotroph (20, 21). There is also suggestive evidence that E. coli facilitates R. palustris iron acquisition, perhaps through cross-feeding of E. coli siderophores (22). Based on the BQH alone, one might predict that the public availability of adenine and siderophores would enrich for E. coli adenine auxotrophs and R. palustris siderophore mutants over time.

Here we examined long-term monocultures and cocultures for evidence of gene loss due to the availability of non-essential adenine and possible siderophores from a partner. In each case gene retention was correlated with low expression, suggesting that repression of gene expression can decrease the frequency of gene loss.

RESULTS

Gene loss is not beneficial when adenine is provided to E. coli monocultures.

In anaerobic cocultures where R. palustris provided essential nitrogen (NH₄⁺) to E. coli and E. coli provided essential carbon (organic acids) to R. palustris (19), we learned that R. palustris also externalizes enough adenine to support an E. coli ΔpurH purine auxotroph (20, 21). Herein, we recognized adenine availability as an opportunity to look for evidence of the BQH; de novo purine synthesis costs more ATP than the purine salvage pathway, and thus could lead to purine synthesis gene loss.

Fitness benefits have been observed for auxotrophic mutants, when the corresponding nutrient is available either as a supplement or via cross-feeding (5–8). Thus, we first tested whether loss of purine biosynthesis would be advantageous in the presence of adenine by comparing the growth of E. coli ΔpurH versus wild-type (WT) PFM2 parent cultures. We focused on E. coli PFM2 instead of the MG1655 (used previously by our lab) because PFM2 does not carry the rph mutation that leads to slower growth in the presence of adenine (23, 24), which can be alleviated by a ΔpurH mutant (Chuang and McKinlay in prep). Despite the availability of adenine, the ΔpurH mutant did not grow faster than the PFM2 parent in the presence of adenine (Fig 1A, B).

Figure 1. Adenine availability does not provide a fitness advantage to an engineered E. coli PFM2 purine auxotroph.

A. Growth of WT E. coli PFM2 and its ΔpurH mutant in monoculture with and without 35 μM adenine (ade). B. Monoculture growth rates for WT PFM2 vs a corresponding ΔpurH mutant with and without 35 μM adenine. ‘a,’ indicates statistically similar comparisons (p > 0.01) as determined using a one-way ANOVA with Tukey’s multiple comparisons test. C. Invasion-from-rare assay competing a ΔpurH mutant against its PFM2 parent in coculture with R. palustris CGA676 under N₂-fixing conditions where R. palustris excretes NH₄⁺ and adenine. The orange line is the best fit from a linear regression analysis with 95% CI shaded in purple. Corresponding equations and x-intercepts (upper, lower bounds for 95% CI) are shown. Change in frequency = (E. coli ΔpurH / (E. coli WT + E. coli ΔpurH))_final – (E. coli ΔpurH / (E. coli WT + E. coli ΔpurH))_initial. A-C. Each data point represents a measurement from a single biological replicate.

We also assessed the fitness impact of a ΔpurH mutation by competing the mutant against the PFM2 parent in coculture with NH₄⁺- and adenine-excreting R. palustris NifA* (CGA676). We used a range of initial frequencies so that we could simultaneously test for coexistence by mutual invasibility, where an equilibrium frequency can be extrapolated from the x-intercept by linear regression analysis (25–27). However, there was poor linear correlation in the invasion-from-rare assay (Fig 1C). The same was true when a ΔpurH::km^R mutant was used (Fig S1). However, the ΔpurH mutant tended to decrease in frequency relative to parent in cocultures, where adenine is available (Fig 1C and S1). Taken together, our results suggest that a ΔpurH mutant does not have a competitive advantage over PFM2 in the presence of adenine. Thus, at least in this engineered context, we do not expect that adenine availability to benefit E. coli PFM2 purine auxotrophs.

Adenine auxotrophs are rare in long-term cocultures.

Although the engineered ΔpurH mutant did not have an obvious fitness advantage, it is possible that (i) there was a subtle difference in growth rate that could be enriched over many generations and (ii) other kinds of purine auxotrophy mutations could lead to a fitness advantage. To account for these possibilities, we looked for the evidence of purine auxotrophs in long-term monocultures and cocultures of E. coli PFM2 and NH₄⁺- and adenine-excreting R. palustris (CGA676; NifA*) that we had maintained for other reasons (Fig 2A; Table S1–5). Adenine is amply available even in evolved cocultures (Chuang and McKinlay in prep). Focusing on the latest common timepoint between monoculture vs coculture treatments (generation 650), we did not observe obvious mutations in purine synthesis genes in PFM2 in cocultures or monocultures (Fig 2B; Fig S2). When considering other time points, only one purine biosynthesis gene mutation across was observed in one monoculture line at generation 204 (purT at 54.3%, line 15; Table S2).

Fig 2. E. coli adenine auxotrophs are not prevalent in long-term cocultures.

A. Long-term monoculture and coculture conditions. OAcs, organic acids. B. Enriched E. coli mutations in genes for nucleobase metabolism and regulation of nitrogen metabolism for long-term monocultures and cocultures. Each point represents a mutation frequency in a given gene for a given evolutionary line. C. Random evolved E. coli isolates (generation 650; n=4) were screened for adenine auxotrophy in liquid monoculture conditions with and without 50 μM adenine. Ancestral strain (Anc) and ΔpurH (ΔH) mutant values are also included (n=3). Each point represents a single measurement for a single isolate (replicate measurements were not made for any isolate). Each line connects measurements for the same isolate grown with and without adenine.

To account for non-obvious mutations that might lead to purine auxotrophy. We also screened four E. coli isolates from each monoculture and coculture line (40 monoculture isolates and 40 coculture isolates) for adenine auxotrophy in liquid cultures in 96-well plates (Fig 2C). Although several monoculture isolates could not grow in minimal media, suggesting that dependencies might have developed between E. coli subpopulations, only two monoculture isolates were possibly adenine auxotrophs (line 20; Fig 2C). We suspect that the other auxotrophs are due to mutations in rpoC, which were conditionally enriched in monoculture (Fig S2A). Certain mutations in E. coli rpoC are known to generate polyauxotrophies (28, 29).

No adenine auxotrophs were isolated from cocultures and all coculture isolates grew in defined media, suggesting that our coculture conditions might not favor emergence of other dependencies between subpopulations. Overall, we conclude that E. coli adenine auxotrophs are not a major subpopulation in any long term coculture despite the availability of adenine.

Cross-fed adenine represses purine synthesis gene expression.

One possible explanation for why E. coli purine auxotrophs did not emerge in coculture is because adenine availability repressed gene expression, thus lowering the purine biosynthesis gene costs. In support of this notion, we previously saw that E. coli MG1655 down-regulated several purine synthesis genes in coculture versus monoculture (22). To determine if the same is true for PFM2, we quantified purH transcripts by RT-qPCR. Coculture PFM2 purH levels were 20% of those in monoculture (Fig 3A). Adding adenine to monocultures resulted in a purH transcript level that was 12% of that observed without adenine, suggesting that the low expression in coculture was due to adenine availability (Fig 3A).

Fig 3. Purine biosynthesis genes are down-regulated in coculture.

RT-qPCR quantification of purH transcript levels in E. coli PFM2 (A; relative to hcaT) and R. palustris CGA676 (B; relative to fixJ). Adenine (+ ade) was added to a final concentration of 0.1 mM to ensure that it was not used up during culture growth. Values are the mean ± SD, n = 3–4. Statistically significant differences from the monoculture condition for each strain were determined using an unpaired two-tail t-test; ***, p < 0.001; ****, p < 0.0001.

R. palustris CGA676 also showed less purH transcript in coculture, 50% of that observed in monoculture (Fig 3B). It is unclear how E. coli would influence repression of R. palustris purH expression. However, the lower expression supports previous findings that E. coli does not stimulate R. palustris adenine production (20). Adenine availability and expression of purine biosynthesis genes curiously did not lead to purine biosynthesis gene mutations in R. palustris monocultures. It is possible that R. palustris lacks effective adenine uptake mechanisms. In a separate study, we found that adenine can be toxic to R. palustris, and purine externalizing strains, including the parent of CGA676 are more resistant to exogenous adenine (Chuang and McKinlay, in prep). Thus, CGA676 might be pre-dispositioned for adenine externalization rather than uptake.

R. palustris siderophore gene loss is correlated with expression.

We looked for other correlations of gene retention and low gene expression in our long-term cultures. A standout was R. palustris siderophore genes which accumulated mutations in monoculture but not coculture (Fig 4A, B; Fig S2). We previously saw that R. palustris down-regulated siderophore synthesis genes in coculture with E. coli MG1655, relative to monocultures (22). We verified that this trend was also true for R. palustris in coculture with PFM2; RT-qPCR analysis showed that the transcript levels for the siderophore synthesis gene RPA2390 was 20% in coculture compared to monoculture (Fig 4C). Adding soluble iron (ammonium ferric citrate) to monocultures decreased RPA2390 expression to 3% of that observed without added iron, suggesting that the low expression in coculture was a response to enhanced iron availability in coculture (Fig 4C).

Fig 4. Siderophore gene loss is prevalent in R. palustris monocultures where gene expression is high.

A, B. R. palustris CGA676 mutation frequencies in siderophore synthesis genes for each monoculture (A) or coculture line (B) where mutations were observed (no mutations were observed in lines 21, 25, and 27). Repeat colors in a given graph indicate different mutations in the same gene. C, D. RT-qPCR quantification of siderophore synthesis genes in R. palustris (C; RPA2390 relative to fixJ) and in E. coli (D; entF relative to hcaT). Values are the mean ± SD, n = 3. Statistically significant differences from the monoculture (mono) condition for each strain were determined using an unpaired two-tail t-test; *, p < 0.05; **, p < 0.01; co, coculture, Fe; ammonium ferric citrate.

R. palustris might down-regulate siderophore production in coculture if E. coli facilitates iron acquisition. Others have speculated that R. palustris can use siderophores from other bacteria, because its genome encodes multiple transporters but only one siderophore synthesis cluster for petrobactin-like siderophores (30, 31); E. coli produces enterobactin siderophores (32). In agreement with this speculation, E. coli siderophore gene expression seemed to respond to iron loss to R. palustris; the entF transcript level was 3.3-fold higher in coculture than in monoculture (Fig. 4D). These observations support a notion that R. palustris can use E. coli siderophores, however the actual molecular nature of this relationship is pending a thorough investigation.

Despite the increase in E. coli entF expression in coculture, E. coli siderophore mutants were rare and inconsistent across coculture lines and were not observed in long-term monocultures (Table S2, S4). It is possible that this expression range is not enough to pose a fitness cost, perhaps due to a low iron requirement under fermentative conditions (e.g., little use for iron-containing cytochromes). In agreement with this notion, transposon insertions in E. coli MG1655 siderophore genes had mainly neutral effects on fitness in monocultures and cocultures (21).

Discussion

In long-term cocultures, the availability of cross-fed adenine, and possibly siderophores, did not result in corresponding gene loss in a recipient. Our data suggest that the externalized resource repressed recipient gene expression, thereby lowering gene cost and selective pressure for gene loss. Conversely, siderophore loss-of-function mutations accumulated in R. palustris monocultures where gene expression was high. Although there are likely situations where cross-feeding promotes gene loss, there are also likely situations where cross-feeding instead promotes gene retention through repressing gene expression.

Our results do not counter the BQH, which posits that gene loss will occur if the benefit of acquiring resources from a neighbor outweighs the cost of gene retention (1). If the cost of gene retention is low, then gene loss would only have neutral fitness effects, rather than beneficial effects. Others have suggested that low gene expression is not enough to promote gene retention because the maintenance and activity of sensor and regulatory proteins can still drive of gene loss (3). However, the fitness benefits of mutants lacking biosynthetic genes suggests that benefits can be realized while regulatory circuits remain intact (5–9). The association between gene cost and gene expression is well-documented (10–16). Our data, and data from others, suggest that there are cases where resource availability can lower gene cost simply by repressing gene expression. For example, repression of an antibiotic resistance gene in E. coli eliminated the fitness cost of that gene (10). In another example closer to the conditions of the current study, iron-limitation that prompted high siderophore production by P. aeruginosa led to a higher frequency of siderophore-deficient mutants compared to iron-rich conditions where siderophore production was low (11). The P. aeruginosa results were presented in the context of cheating, where gene loss is associated with exploitation that harms the producer, unlike BQH beneficiaries that should not harm the producer (1). Yet, the conditions leading to the emergence of cheaters and BQH beneficiaries are analogous (1). As such, many lessons from literature on cheaters can apply to BQH scenarios.

Insights into BQH gene-loss can also come from resource-rich monocultures, with the caveat that nutrient access might be well above that from a cross-feeding partner. For example, repression of gene expression in long-term nutrient-rich E. coli monocultures could explain why the loss of some catabolic genes was not observed in parallel lines and occurred without fitness effects (33). Repression of gene expression will shift the cost of a gene towards neutral. Gene loss with neutral fitness effects can still occur, but enrichment of those mutations would require hitchhiking with a beneficial mutation or passage through a severe population bottleneck. Cost-neutral gene loss, as opposed to beneficial gene loss, might also require time, and possibly spatial structure, that were not part of our experimental design. Should cost-neutral auxotrophic mutants emerge in nature, the chances of rescue via cross-feeding from a neighbor could be high (34, 35).

The extent to which gene repression can promote gene retention will also depend on the level of expression strength of repression; as noted above, relatively low expression might explain the lack of E. coli siderophore gene mutations. Gene expression can be noisy, and that noise can be costly and subject to selective optimization in a stable nutrient regime (36, 37). Noisy gene expression might explain why some engineered auxotrophs have greater fitness over a prototrophic parent (5–9). In long-term E. coli monocultures, loss-of-function mutations for maltose catabolism were arguably associated with weak gene repression in the presence of glucose (38). During growth on glucose, ancestral expression of the maltoporin was half that compared to during growth on maltose. Several evolved isolates carried mutations that prevented growth on maltose. However, in an evolved isolate that retained the ability to grow on maltose, maltoporin expression with glucose was 5% of that with maltose (38), suggesting that stronger repression contributed to gene retention.

The level of nutrient availability also likely affects the level of gene repression. In other words, there might be a resource concentration range over which BQH gene loss would occur. For example, Acinetobacter baylyi histidine autotrophs had a fitness advantage over a prototrophic parent when there was ≤ 50 μM histidine, but the advantage was reversed as histidine levels reached 100–200 μM (6), possibly due to stronger gene repression in the parent, though this is speculation on our part. One might also consider that the responses of a prototrophic strain versus an otherwise clonal auxotroph to an extracellular nutrient can vary widely; whereas an auxotroph would benefit from the nutrient, the prototroph might experience beneficial, neutral or even detrimental fitness effects (6). The extent to which loss of a biosynthetic gene can have compounding fitness effects by affecting expression of other genes deserves investigation.

Our work presents a caveat to the popular expectation of cross-feeding as a driver of gene loss. Cross-feeding can instead promote gene retention when resource availability is high enough to repress gene expression and thus lower the cost of a gene. The likelihood of gene loss then becomes the same as any other gene with a neutral cost.

MATERIALS AND METHODS

Bacterial strains.

R. palustris CGA676 is derived from CGA0092 (39) and carries a nifA* mutation that causes NH₄⁺ excretion under N₂-fixing conditions (19, 40). E. coli PFM2 (41, 42) ΔpurH::km^R was made via lambda Red recombination (43) using constructs amplified from KEIO mutants (44) using primers YCC29 (GCG CAA ACG TTT TCG TTA CAA TGC) and YCC30 (TGC ATT ACC CGG AGC AAC). FLP-mediated excision was used to remove the kanamycin resistance cassette (km^R) to generate the ΔpurH strain (43).

Growth conditions.

Anoxic media in test tubes were prepared by bubbling N₂ through 10 ml of media in 27-ml anaerobic test tubes, then sealing with rubber stoppers and aluminum crimps prior to autoclaving. Monocultures and cocultures were grown horizontally at 30°C with light from a 45 W halogen bulb (430 lumens) and shaking at 150 rpm in minimal M9-derived coculture medium (MDC) (19) with either (i) E. coli monocultures: 25 mM glucose, 10 mM NH₄Cl, and cation solution (100X stock: 100 mM MgSO₄ and 10 mM CaCl₂); R. palustris monocultures: 20 mM sodium acetate and 10 mM NH₄Cl; or cocultures: 25 mM glucose and cation solution. Cultures with plasmid-carrying strains were also supplemented with 100 μg/ml gentamycin or 25 μg/ml chloramphenicol as appropriate. Starter cultures were inoculated with single colonies. R. palustris starter cultures were grown in MDC with 20 mM acetate and 10 mM NH₄Cl. E. coli starter cultures were grown aerobically in lysogeny broth, with 30 μg/ml kanamycin (km) when appropriate. E. coli starter cultures were washed twice in 1 ml MDC prior to inoculating test cultures or bioassays. Cocultures were inoculated with 0.1 ml each of R. palustris and E. coli to an initial optical density (OD₆₆₀) of ~0.003 each. Cultures in 96-well plates were treated similarly except that oxic stock solutions were used to prepare 0.2 ml volumes in each well. Anoxic conditions were achieved by sealing plates inside a BD GasPak EZ large incubation container with 2 anaerobic sachets.

Experimental evolution.

Founder monocultures of E. coli PFM2 and R. palustris NifA* CGA676 were grown from a single colony in anoxic MDC with either 25 mM glucose, cation solution, and 3 mM NH₄Cl for PFM2 or 20 mM sodium acetate for CGA676. A single founder monoculture was then used to inoculate 10 monocultures and 10 cocultures. All cultures were grown horizontally without shaking at 30°C with light in MDC. PFM2 monocultures were supplemented with 10 mM glucose, cation solution, 25 mM NaCl, and 2.3 mM NH₄Cl. CGA676 monocultures were supplemented with 25 mM glucose, cation solution, 8 mM disodium succinate, 7.3 mM sodium acetate, 0.25 mM sodium formate, 1.4 mM sodium lactate, and 6.3 mM ethanol. Cocultures were supplemented with 25 mM glucose, cation solution, and 25 mM NaCl. Every 7 days, cultures were vortexed and 0.25 ml was transferred to fresh medium. About every 5 transfers, stocks were frozen in 25% glycerol at −80^oC and separate cell pellets from 1 ml samples were frozen for gDNA extraction.

Analytical procedures.

Cell densities were measured via turbidity (OD₆₆₀) using a Genesys 20 visible spectrophotometer (Thermo-Fisher).

Competition assays.

Competition assays were conducted in an invasion-from-rare format to consider coexistence by the mutual invasion criterion, where each population can increase when rare (25, 27). Cocultures were started from various initial frequencies (targeting 0.01 – 0.99) of each E. coli strain (WT vs ΔpurH or WT vs ΔpurH::km^R) for a total initial cell density of ~10⁶ colony forming units (CFU) / ml. R. palustris CGA676 was inoculated to an initial density of ~10⁶ CFU / ml. Frequencies were determined upon inoculation and after 5 days. WT and ΔpurH mutants were distinguished by counting CFUs on M9 agar with cations and 25 mM glucose and on LB agar and then determining ΔpurH populations from the difference. When ΔpurH::km^R mutants were used LB agar included km to allow for direct determination of population size. Change in frequency = (E. coli ΔpurH / (E. coli WT + E. coli ΔpurH))_final – (E. coli ΔpurH / (E. coli WT + E. coli ΔpurH))_initial (26).

Genome sequencing and mutation analysis.

gDNA was purified from cells using a Qiagen DNeasy Blood and Tissue kit following the manufacturer’s instructions. Lysis was facilitated after resuspension by adding proteinase K (50 μg/ml final), and incubating at 56°C for 10 min. RNaseA (4 μl, Promega) was then added and the lysate was incubated for 2 min before proceeding.

DNA fragment libraries were made using a NextFlex Bioo Rapid DNA kit and libraries were sequenced using Illumina NextSeq 500 150×150 paired-end runs by the IU Center for Genomics and Bioinformatics. Paired-end reads were pre-processed for quality with cutadapt 3.4 (45) with the following options: -a AGATCGGAAGAGC -A AGATCGGAAGAGC ; -q 15,10; -u 6. Mutations were called using breseq v. 0.32.0 on polymorphism mode (46). E. coli monoculture population sequences were mapped to the MG1655 genome (accession NC_000913); R. palustris monoculture population sequences were mapped to a concatenated reference genome consisting of the CGA009 chromosome (accession BX571963), and its plasmid pRPA (accession BX571964). Co-culture sequences were mapped to a concatenation of the E. coli and R. palustris reference genomes. Polymorphisms that co-occurred in both the monoculture and co-culture datasets were filtered and maintained as a subset to enrich for the most informative variants representing treatment differences (Table S1–4). This filtering step also removed sequence differences between the reference sequences and those of the experimental strains used. Variants were prioritized as mutations of interest if they were detectable at the final two sequencing timepoints and co-occurred across multiple populations in the same locus. All mutations can be found in Tables S1–4. Locus tag conversions can be found in Table S5.

Reverse transcription quantitative real-time PCR (RT-qPCR).

Cultures received 100 μM adenine or ammonium ferric [iron(III)] citrate as indicated. Cultures were harvested in exponential phase at 0.6–0.8 OD₆₆₀ except for E. coli monocultures (+/− adenine experiment), which were harvested at 0.3–0.4 OD₆₆₀ to avoid adenine depletion. Cultures were chilled on ice and pelleted by centrifugation. Lysis, RNA purification, and cDNA generation were performed exactly as described (20). Standard curves were generated using gDNA. Transcripts were quantified as described (20) using the appropriate primers (Table S6) with a Mastercycler ep realplex real-time PCR system (Eppendorf). Data was analyzed by realplex software using Noiseband. Specificities were validated by melting curves and by the presence of a single band on an agarose gel.

Statistical analyses.

Statistical analyses of growth and RT-qPCR data were performed using Graphpad Prism (v10).