Selection of Bacterial Mutants in Late Infections: When Vector Transmission Trades Off against Growth Advantage in Stationary Phase

Pathogens can evolve inside their host, and the importance of this mutation-fueled process is increasingly recognized. A disease outcome may indeed depend in part on pathogen adaptations that emerge during infection. It is therefore important to document these adaptations and the conditions that drive them. In our study, we took advantage of the possibility to monitor within-host evolution in the insect pathogen X. nematophila. We demonstrated that selection occurring in aged infection favors lrp defective mutants, because these metabolic mutants benefit from a growth advantage in stationary phase (GASP). We also demonstrated that these mutants have reduced virulence and impaired transmission, modifying the infection outcome. Beyond the specific case of X. nematophila, we propose that metabolic mutants are to be found in other bacterial pathogens that stay for many generations inside their host.

accumulating where it fuels the evolution of pathogens inside the host, in particular in human diseases (3,4). While mutation generally occurs throughout the whole genome, there are other mechanisms that impact a restricted set of genes. These mechanisms can be epigenetic alterations, where clonal populations of bacteria modify their phenotype by changing their regulatory state (5,6). These can also be genetic alterations, as in phase variation where a specific high-rate mutation mechanism produces reversible changes in one or a few genes (7,8). Phase variation is described as the basic mechanism that makes antigenic variation a successful instrument for some pathogens to escape their host immune system (9).
The insect bacterial pathogen Xenorhabdus is a promising model to study the adaptive nature of mechanisms that control phenotypic variation inside an infection. This pathogen kills insects and proliferates in their cadavers for 1 or 2 weeks, before it is transmitted by the nematode vector Steinernema (10). During this long period, Xenorhabdus maintains high densities inside the insect and, therefore, potentially accumulates phenotypic variation. As the complete life cycle of Xenorhabdus can be experimentally reproduced (11), it is possible to quantify how this variation impacts each of its different stages.
Xenorhabdus isolated from the wild typically are in a form described as primary, but in culture media, they convert to another, secondary form when reaching long-term stationary phase (12). In his seminal paper, Akhurst (13) showed that secondary forms of Xenorhabdus also appear during infection and that this occurs at a rate that varies greatly among Xenorhabdus strains. Although the phenotypic differences between the two forms can also vary depending on strain and species, only cells of the primary form are able to bind bromothymol blue dye, are motile, agglutinate red blood cells, and produce fimbriae, hemolysins, proteases, antimicrobials, and crystalline inclusion bodies (12,(14)(15)(16)(17). Interestingly, genes for which expression differs between primary and secondary forms have also been shown to play a role in the interaction between Xenorhabdus nematophila, its nematode vector, and its insect target (18). This strongly suggests that the emergence of secondary forms should impact X. nematophila interactions with their invertebrate hosts. This could also explain why nematodes have so far been reported to carry primary forms, although previous experiments suggest that secondary forms of X. nematophila are capable of both killing insects (16) and being transmitted by the nematode Steinernema carpocapsae (19).
The alternation between primary and secondary forms has so far been interpreted as a case of phase variation (12), and later termed phenotypic variation (17). As the phenotype of secondary forms matches that of lrp defective mutants (20)(21)(22)(23)(24), it has also been proposed that the Lrp master regulator might be involved in the production of secondary variants in X. nematophila (20). However, the mechanism driving phenotypic variation in X. nematophila is unknown.
The goals of this study were to better understand (i) the molecular mechanisms that are responsible for the production of phenotypic variants and (ii) the impact of these variants on transmission by the vector. To do so, we investigated a large collection of X. nematophila isolates with various phenotypic forms. We found that secondary forms are not phase variants but rather plain lrp mutants. We also found that at least a third phenotypic form exists in X. nematophila which is not a lrp mutant. All these variants have a growth advantage in stationary phase (GASP [25]) which probably explains why they reach higher loads than primary forms during late infection. We then quantified how these variants are transmitted by the nematode vector S. carpocapsae and found indications that isolates that reach the highest densities in insects are the least transmitted by nematodes. We therefore propose that X. nematophila experiences a trade-off between traits that are favored during late infection and traits that increase transmission.
Compared to group 1, group 2 variants cannot adsorb dye, are not motile, do not secrete antibiotics, and have weak or no hemolytic and lipolytic activities (13). Among our 34 isolates, 14 had such characteristics ( Fig. 1A and Table 1). Remarkably, these group 2 variants also had smaller cells than those of group 1 (Fig. 1A), which adds a new characteristic to their set of phenotypes.
The remaining 20 isolates were heterogeneous: they were all motile and had large cells, but 15 of these isolates always formed blue colonies on NBTA medium, while 5 did form both blue and red colonies (Fig. 1A). A clustering analysis restricted to these three quantitative phenotypes (motility, cell size, and proportion of red colonies) confirmed that four isolates formed a third distinct group of variants (isolates G3#9, G3#22, G3#48, and G3#27 [ Fig. 1B]). We added isolate G3#47 to this group 3, although it clusters with the first group, because it formed red colonies. Group 3 variants repeatedly combined  23   25   18  19  39  41  37  45  32  24  25  40  36  28  29  31  27  48  9  22  23  26  21  42  47  51  30  38  12  35  44  56  59  46  43  For each isolate, average log forward scatter (FSC), a proxy for cell size, and average motility halo diameter, a measure of cell motility, are represented. The red part of the small pie charts indicates the proportion of red colonies observed for each isolate on NBTA culture medium, and the pie chart diameter increases with average measure of antibiotic activity. The boxed numbers identify the five isolates that constitute the third group. Other numbers identify isolates that we will use most often as representative of their group (G1#23 and G2#25 for group 1 and 2, respectively) or that we discuss specifically in text (G2#31). (B) Result of a hierarchical clustering analysis (using a unweighted-pair group method using average linkages [UPGMA] method) based on log FSC, motility, and proportion of red colonies. The asterisk indicates variant #47, which we added to group 3 but clusters with group 1. Boldface numbers identify isolates that we will use most often as representative of their group.  (20), just like the group 2 isolates of our collection. It is therefore possible that the group 2 variants are lrp mutants. To test this hypothesis, we sequenced lrp and its promoter region for the 34 isolates of our collection. All group 1 and group 3 isolates had an lrp sequence identical to that of the F1 reference genome (26). Conversely, 13 out of the 14 group 2 isolates had one nonsynonymous mutation. Isolate G2#31 (Fig. 1A) was the sole group 2 isolate with no mutation. Interestingly, G2#31 also stands apart as the group 2 isolate with the lowest in vitro growth rate (see Additional file 2 at https://figshare.com/s/4cf02d0aa870dd20d5ab). To discard the possibility that there are other mutations that could be responsible for the phenotype, we performed complementation by inserting a functional copy of lrp in the chromosome of G2#25, which has a single nucleotide polymorphism (SNP) in lrp codon 120. The complemented variant with group 2 phenotypes showed a full restoration of the phenotypes that are typical of group 1 variants (see Additional file 3 at the above URL).
To validate that these results were not only restricted to our collection, we initiated 24 independent in vitro cultures from the group 1 variant G1#23, which we plated every day on NBTA medium. Beyond being an experimental replicate, this experiment allowed us to monitor the changes of phenotypes over time (at 1, 5, and 7 days). As expected, all clones initially formed blue colonies and were motile ( Fig. 2A). After 5 days of incubation, we were able to sample bacteria forming red colonies ( Fig. 2A). They did not differ from other isolates in terms of cell size, but they had significantly lower motility and antibiotic activity (by Kruskal-Wallis rank sum test, P ϭ 1.41eϪ04 and P ϭ 1.08eϪ05 for motility and antibiotic activity, respectively, using measurements that were first averaged by culture). After 7 days of incubation ( Fig. 2A), bacteria forming red colonies were small nonmotile cells, while those forming blue colonies were large motile cells (Kruskal-Wallis rank sum test, P Ͻ 2.2eϪ16 for both cell size and motility averaged by culture). Over the 7 days of culture, the frequency of red variants increased from 0% to about 20%.
We then sequenced lrp in clones sampled after 8 days of incubation and found nonsynonymous mutations in 15 out of the 18 sampled red colonies. Red variants sampled at day 5 were also lrp mutants, although in a smaller proportion (8 out of 14 sequenced clones). We then averaged the measurements performed on these clones by culture and found that red variants with and without lrp mutations had similarly low antibiotic activity (Wilcoxon test, P ϭ 0.26) but that red variants with no lrp mutation were more motile (Wilcoxon test, P ϭ 2.66eϪ3 and P ϭ 1.65eϪ3 at day 5 and day 8, respectively) and had larger cells than red variants with lrp mutations at day 8 (Wilcoxon test, P ϭ 1.65eϪ3). They therefore match our definition of group 3 (Table 1). Furthermore, their frequency decreased from day 5 to day 8, which suggests that group 3 variants appeared earlier and were replaced by group 2 variants. This is confirmed by another replicate experiment, where we found red colonies appearing early to be significantly more motile than those appearing late (see Additional file 4 at https:// figshare.com/s/4cf02d0aa870dd20d5ab).
lrp mutations are diverse and not reversible. Phenotypic variants in Xenorhabdus were considered phase variants (15). If this theory is correct, mutations in lrp should be the product of a specific molecular mechanism (6). Overall, we identified 25 distinct nonsynonymous mutations in lrp (Fig. 2B). Of these 25 mutations, 15 most probably caused profound alterations to the translated protein: five IS5 insertions and three large duplications totally modified the lrp sequence, three indels caused a frameshift, and one indel and three SNPs caused nonsense mutations. The remaining 10 mutations are SNPs that changed a single amino acid in Lrp. The mutations we found in group 2 For SNPs that cause a substitution, the change in amino acid is given. Tn1, Tn2, and Tn3 are three transposons. Numbers in parentheses give the number of independent replicate cultures where a mutation has been observed (one if not indicated). The mutations we found are grouped here in four categories: three mutations were tandem duplications of a 26-to 116-bp-long fraction of the lrp sequence; five mutations were insertion sequences (ISs) with Tn1, Tn2, and Tn3 being three different transposons which belong to the group of IS5 insertion sequences, and share the same two insertion points in lrp; four mutations were insertions or deletions of a few bases (indels); 13 mutations were single nucleotide polymorphism (SNPs). (C) Cell size and motility of cells sampled from prolonged culture of eight different group 2 isolates (one sense mutation, one nonsense mutation, and one frameshift in either the HTH domain or RAM coregulator response domain, and two transposon insertions). The colors of the symbols indicate the color of colonies on NBTA medium; gray crosses indicate that the color was neither blue nor red. Eight days after cultures were started, we observed cells that have recovered part of the functions typical of group 1 variants. Many of these reversions, however, are not complete: some revertants have large cells but remain nonmotile; other are motile but have small cells or are still red. (Inset) Number of cultures (out of the 12 we performed for each isolate) where blue or hemolytic colonies were observed. HTH and RAM identify isolates that have a mutation in either of the two main active domains of the Lrp protein. As in Fig. 1C, Tn1 and Tn2 are two distinct insertion sequences, FS stands for frameshift, SNP indicates sense point mutations, and the asterisks indicate nonsense mutations.
Persistence in Cadavers Trades Off against Transmission ® variants are therefore highly diverse, which makes them unlikely to result from a single specific mutation mechanism.
If group 2 variants were phase variants, lrp mutations should also be reversible (6). We thus tested the reversibility of group 2 isolates by monitoring prolonged static LB cultures of eight group 2 isolates with distinct lrp mutations. After 8 days of incubation, we observed cells capable of forming blue colonies for most of tested group 2 isolates (Fig. 2C). The sequencing of lrp revealed that none of these phenotypic reversions was associated with a genetic reversion of the initial mutation. Accordingly, most of these phenotypic reversions were only partial, with blue colonies being composed of either small or weakly motile cells (Fig. 2C). This confirms that the lrp mutations in group 2 isolates are not produced by a phase variation mechanism. Interestingly, the probability of phenotypic reversion differed among the isolates tested (Fig. 2C, inset, glm with binomial error, Chi2 ϭ 31.38, df ϭ 7, P ϭ 5.28eϪ05) which suggests that the way phenotypes are restored in lrp mutants depends on the precise nature of the mutation. Finally, we never observed revertants in cultures of the group 3 variants G3#9, G3#22, and G3#48. Group 2 and group 3 variants are under positive selection during prolonged in vitro culture. Variants of X. nematophila reach high frequency in prolonged in vitro cultures, which suggests they are under strong positive selection. To test this, we first estimated bacterial survival during stationary phase (Fig. 3A). All three variants tested had the same survival during early stationary phase (Wilcoxon test, P Ͼ 0.07), but G1#23 has a 10-fold-lower survival in late stationary phase compared to early stationary phase (Wilcoxon test, P ϭ 5eϪ4), while G2#25 and G3#9 maintained high survival (Wilcoxon test, P Ͼ 0.09). Group 2 and group 3 variants therefore seem to resist better than group 1 variants to the stressful conditions of late stationary phase, explaining why group 2 and group 3 variants reach higher densities than group 1 variants in prolonged culture (see Additional file 2 at https://figshare.com/s/4cf02d0aa870dd20d5ab). Bacterial survival during early and late stationary phase. We estimated the proportion of living bacteria as the ratio between the total number of cells in a LB culture (estimated with a Thoma cell counting chamber) and the number that can form colonies on NBTA. We contrast here the proportion of living bacteria measured in early or late stationary phase to that measured during exponential phase. Asterisks indicate significant deviations from zero, as tested by a Wilcoxon test. Survival does not significantly vary between early stationary phase and exponential phase. During late stationary phase, G1#23 experiences a 10-fold decrease in survival, while G2#25 and G3#9 survival does not vary significantly. (B) Log competitive index (CI) of G1#23, G2#25, and G3#9 inoculated in a F1V1 (non-GFP variant from group 1) culture. For each culture, CI is estimated as the variation in proportion of GFP-labeled variants. Asterisks indicate significant deviation from zero, as tested by a Wilcoxon test. As expected, the group 1 GFP-labeled variant G1#23 shows a marginal decrease in frequency when inoculated in the non GFP-labeled group 1 F1V1. Conversely, G2#25 and G3#9 both increase in frequency.
We then investigated competitive ability of the same three isolates (Fig. 3B). For this purpose, we inoculated G1#23, G2#25, and G3#9 in a LB culture of F1V1, a non-GFPlabeled group 1 variant. We then quantified the variation of frequency of green fluorescent protein (GFP)-labeled variants (competitive index [CI]): over 5 days of incubation, G1#23 decreased slightly in frequency (Wilcoxon test, P ϭ 0.016), while G2#25 and G3#9 both increased in frequency relative to the nonfluorescent competitor (Wilcoxon test, P ϭ 4.88eϪ4 in both cases). G3#9 had an intermediate competitive advantage (Wilcoxon test, P ϭ 1eϪ4) which correlates with the previous observation that group 3 variants had intermediate phenotypes. Altogether, our results suggest that phenotypes of group 2 and group 3 variants have a growth advantage in stationary phase (GASP [25]) and are therefore under positive selection in aged cultures.
Group 2 and group 3 variants also appear inside insects. Akhurst, in the first description of Xenorhabdus variants (13), mentions that they appear in vitro but also in insects during infection. To test this in the case of our particular strain of X. nematophila, we monitored the appearance of group 2 and group 3 variants after injecting the group 1 isolate G1#23 in Galleria mellonella. As observed in vitro, red variants increased in frequency and reached 10% of the CFU on average 10 days after injection (Fig. 4A). Red colonies were initially slightly (although nonsignificantly) more motile than blue colo- Asterisks indicate that the difference in motility was significantly different from zero, as tested by a Wilcoxon test. Compared to blue isolates, red isolates are initially as motile (as expected for group 3 variants) but become less motile at day 3 and 6 (as expected for group 2 variants). (C) Log CFU per insect (5 days after injection; see Materials and Methods) as a function of log CFU per ml in agitated LB culture (after 91 h of incubation; see Materials and Methods). Each symbol on a graph corresponds to one of the 34 isolates of the collection. The color of the symbol indicates the group the variant belongs to, and the symbol size increases with increasing cell size (approximated by log FSC as in Fig. 1A). Bacterial density in insects strongly and positively correlates to density in vitro which suggests that traits favored in aged LB cultures are also favored during late infections.
Persistence in Cadavers Trades Off against Transmission ® nies, as expected for group 3 variants, but became significantly less motile at both days 3 and 6, as expected for group 2 variants (Fig. 4B). Lack of difference at day 10 comes mostly from a late increase in motility in red variants (e.g., average motility is 9.73 Ϯ 1.74 mm in red variants at day 3 and increases to 13.7 Ϯ 3.15 mm at day 10). Finally, we found that red variants isolated from infected G. mellonella were lrp mutants when they had group 2 phenotypes, but they carried no lrp mutation when they had group 3 phenotypes.
The increase in variant frequency in infections (Fig. 4A) suggests that the conditions that favor group 2 and 3 variants in vitro may also apply inside insects. To test this, we measured bacterial densities of the 34 isolates of our collection 5 days after injection in G. mellonella larvae. We found that the density in vivo positively correlated with the density variants reached in vitro ( Fig. 4C; Kendall correlation, ϭ 0.40, P ϭ 0.7eϪ4): the variants that best performed in vitro also reached high densities in insects.
Group 2 variants are less virulent and least transmitted. Transmission of Xenorhabdus relies in part on their capacity to kill the insect host. We found that all variants retained this capacity, although group 2 variants took slightly more time than others to kill G. mellonella (Fig. 5A and Table 2). We then calculated a proxy of bacterial transmission (R 0 ) for each of the 34 isolates of the collection. This quantity incorporates the parasitic and reproductive success of nematodes, the number of Xenorhabdus bacteria carried per injective juvenile (IJ) and IJ survival during dispersal; it predicts the number of new infections that can be initiated from a single infected insect (11). We found that transmissions were comparable in group 1 and group 3, while group 2 variants had a much lower transmission than the other groups (Fig. 5B, pairwise Wilcoxon test with Holm correction, P ϭ 0.14 and P ϭ 3.1eϪ07, respectively). Reduced transmission in group 2 variants was explained by lower parasitic success, reduced fecundity of nematodes, and an increase in the death rate of nematodes (Kruskal-Wallis test, P ϭ 1.11eϪ5, P ϭ 3.40eϪ5, and P ϭ 2.85eϪ4, respectively). For a complete and detailed analysis of the components of R 0 transmission, see Additional file 5 at https://figshare.com/s/4cf02d0aa870dd20d5ab. Isolates that reached the highest in vivo (and in vitro) loads were the least transmitted (Fig. 5B, Kendall's rank correlation, ϭ Ϫ0.43, P ϭ 1.946eϪ4), although they did reassociate with nematodes (Fig. 5C, Kendall's rank correlation, ϭ 0.26, P ϭ 0.032). These isolates were not transmitted mostly because they impaired nematode reproduction (Kendall's rank correlation between f and in vivo CFU, ϭ Ϫ0.39, P ϭ 8.55eϪ4).
As expected, switching occurred during infections: group 1 infections produced some IJs which carry red variants, while the majority of group 2 infections produced some IJs carrying blue variants (Fig. 5D). In contrast to what we observed in vitro, we found that group 3 variants did revert in the insect, with group 3 infections producing IJs that carry bacteria forming blue colonies. Overall, our data indicate that the proportion of group 2 and group 3 infections producing IJs carrying blue revertants is significantly higher than the proportion of IJs emitted from group 1 infections that carry red variants (Wilcoxon test, P ϭ 1.34eϪ3).

DISCUSSION
Xenorhabdus nematophila secondary variants are characterized by a well-known suite of phenotypic traits (inability to adsorb bromothymol blue, reduced motility, reduced antibiotic, hemolytic, lipolytic, and proteolytic activities [12][13][14]) to which we added in this work their smaller cell size and the fact that they better survive and reach higher densities than group 1 in prolonged culture. Earlier studies have demonstrated that group 2 variants share many of these traits with lrp defective mutants, and it has therefore been proposed that Lrp was controlling phenotypic variation in X. nematophila (20). Using a cohort of isolates, we demonstrate here that the switch to secondary forms in Xenorhabdus is caused by lrp mutations. But in contrast to what is expected in phase variation (5), these mutations were of diverse nature, and we never observed complete phenotypic reversions involving the restoration of a functional lrp sequence. Overall, our data therefore demonstrate that secondary variants of X. nematophila are not phase variants, but instead plain lrp mutants. The fact that we observed reversion at rates that vary among lrp mutants suggests that reversion is achieved through a variety of compensatory mechanisms, probably involving several Lrp regulated genes that are yet to be identified.
Surprisingly, we never observed synonymous mutations in lrp. This indicates that lrp is not a mutational hot spot: the lrp mutants we have found are probably part of the standing genetic diversity present at a low frequency inside the bacterial population, frequently detected in our experiment because they are under positive selection.
Lrp belongs to a family of global regulators that are known to respond to nutrient availability and regulate cell metabolism in case of food shortage (27,28). Previous work (18) and our in vitro measurements (Additional file 2 at https://figshare.com/s/ 4cf02d0aa870dd20d5ab) demonstrate that Xenorhabdus lrp mutants grow more slowly in rich culture medium, but we found that they survive better and reach 10-fold-higher loads than nonmutants during prolonged stationary phase. Most importantly, lrp Persistence in Cadavers Trades Off against Transmission ® mutants did outcompete group 1 variants in aged cultures, a phenotype described as growth advantage in stationary phase (GASP [25]). This interpretation is further strengthened by the fact that one of the first GASP mutants identified was an lrp mutant of Escherichia coli K-12 (29,30). We therefore propose that secondary variants in X. nematophila are under selection during late infection because they are GASP mutants.
We also documented here the existence of a third class of phenotypic variants, which do not carry lrp mutations and share phenotypic characteristics with both group 2 variants (red colonies, reduced antibiotic activity) and group 1 variants (large and motile cells, lipolytic activity). Interestingly, such a combination of traits has been reported in some variants that Cowles et al. (20) considered secondary forms but that might in fact be similar to our group 3 variants ( Table 2 in reference 20). These variants outcompete group 1 variants and thus display a GASP phenotype, although weaker than that we measured in group 2 variants. The genetic or epigenetic mechanism responsible for the emergence of this phenotype is yet to be identified.
Photorhabdus, the sister genus of Xenorhabdus, provides additional evidence that pathogens can produce a variety of forms during infection. Photorhabdus produce both secondary variants and M forms, another type of variant which is the only one capable of reassociating with the nematode vector (31). However, to our knowledge, Lrp is not involved in the production of any kind of variant in Photorhabdus. Although the two genera have very similar life cycles, Photorhabdus therefore seems to produce variants that differ from those described for Xenorhabdus.
We showed that Xenorhabdus variants of group 2 and 3 better resist the conditions of in vitro late stationary phase compared to group 1 variants. We found indications that the combination of traits that make them more competitive under these conditions is also advantageous in the insect: the variants that reach the highest densities in aged LB cultures also reach the highest loads in late infections. This explains why variants of group 2 and 3 repeatedly appear in insects. This also asks the question of their adaptive value in the natural situation where they interact with their nematode vector. In fact, lrp mutants are poorly transmitted, which cannot be explained by a deficiency in reassociation with nematodes, as we found that IJs emitted from group 2 infections carried as many bacteria as those from group 1 infections. This observation contradicts the prediction of Cao et al. (22) but supports previous findings by Sicard et al. (19). Low transmission of group 2 comes instead from a sharp decrease in nematode reproduction, which is in agreement with earlier experimental results (20,22) and corresponds with the well-established detrimental effect of secondary variants in mass production of S. carpocapsae (32). We also found that IJs emitted from group 2 infections have lower survival during dispersal compared to group 1 infections, which is probably yet another indication that infections initiated with lrp mutants constitute an unfavorable environment for S. carpocapsae, which is consistent with the study of Cao et al. (22). These observations may be compared to those of Morran et al. (33) who have shown that strong positive selection of virulence in X. nematophila negatively affects its transmission by S. carpocapsae. Finally, we found that a high proportion of IJs produced from group 2 or group 3 infections carry group 1 variants. This might be because reversion occurs in insects at a higher rate than in culture or because dispersing nematodes preferentially associate with blue variants when mixed with red ones. In any case, our data suggest that group 2 and group 3 variants are counterselected during the transmission phase of Xenorhabdus. Chapuis et al. (11) have demonstrated that death rate of IJs increases with the number of X. nematophila cells they carry. As a result, survival of IJs during dispersal trades off against their capacity to initiate a new infection. Here, we show that variants that reach higher loads in late infections are the least transmitted by nematodes (Fig. 4C). This can be understood as yet another trade-off, distinct from that demonstrated by Chapuis et al. (11). Traits favored during infection because they permit high bacterial loads are disfavored during transmission because they decrease nematode reproduction. In E. coli, lrp mutants are thought to be better adapted to late stationary phase in part because they scavenge some of the amino acids they need, instead of producing them (29,30). In X. nematophila, lrp also controls the production of exoenzymes which are known to support nematode reproduction (20)(21)(22)(23)(24), and this advantage would come at the cost of a reduction in transmission.
Here, we have investigated the diversity of phenotypic forms that appear in prolonged cultures of X. nematophila. We found that these forms can be classified in several distinct groups, which have in common that they have GASP phenotypes and thus increase in frequency both in vitro and during late infections, inside the insect host. We finally demonstrate that the variants that reach their highest loads, both in vitro and in insects, are those that are the least transmitted, because they negatively impact the reproduction of their nematode vectors (Fig. 6). Similar situations could probably arise in other pathogens that stay for many generations inside their host (34,35). GASP are detected and rapidly increase in frequency because they better resist the conditions of stationary phase than primary variants (in blue). When nematodes start dispersing around 10 to 15 dpi, the population of X. nematophila within the host cadaver may comprise a high proportion of GASP variants. In principle, GASP variants, could therefore contribute to transmission. However, our data suggest that nematodes that carry GASP variants have a lower probability to succeed in infecting new insects. mutants, which can impact transmission, may therefore influence the evolution of pathogens that form long-lasting infections.

MATERIALS AND METHODS
X. nematophila isolates. Xenorhabdus nematophila isolates were obtained from static cultures of the green fluorescent protein (GFP)-labeled strain F1D3 (19). Samples of 10 independent LB cultures of strain F1D3 were streaked onto NBTA plates (13) after 3, 7, and 13 days of incubation at 28°C. Primary forms form blue colonies on NBTA, while secondary variants form red colonies (12,14,15). Each time a red colony was observed, a blue colony from the same petri dish was also sampled. Overall, we obtained 34 distinct isolates (see Additional file 1 at https://figshare.com/s/4cf02d0aa870dd20d5ab) which we stored in 20% glycerol at -80°C.
Measuring colony phenotypes. For each isolate, we measured four phenotypic traits that differ among Xenorhabdus variants. Swimming motility was measured as the diameter of a halo formed by motile bacteria on 0.35% agar culture medium (14,15). Antibiotic activity was quantified by measuring the diameter of inhibition halos, using Micrococcus luteus as a target strain (36). Extracellular lipolytic activity was assessed by the presence of precipitated material surrounding the colony cultured on Tween 20 agar (36). Hemolytic activity was quantified by the presence of a clearing surrounding bacteria grown on standard sheep blood agar plates (37).
Measuring cell size. Cell size was first estimated by flow cytometry analysis in three replicate in vitro experiments. In each experiment, exponential-phase cultures of each isolate were fixed with a 0.2% solution (vol/vol) of formaldehyde and analyzed with a FACSCalibur flow cytometer (Becton Dickinson), equipped with an argon air-cooled laser providing 15 mW at 488 nm and the standard filter set-up. Bacterial cells were discriminated from particles by applying a polygonal gate to green (530 Ϯ 15 nm) and red (585 Ϯ 21 nm) log-transformed fluorescence measurements and a rectangular gate on logtransformed side scatter (SSC) and forward scatter (FSC), using tools provided in R Bioconductor (38). We used log-transformed FSC as a proxy for cell size.
In other experiments, when cytometry was not applicable, we sampled colonies, put them on cover slides, and took pictures with an Olympus BX51 microscope with 400ϫ magnification. Images were analyzed using ImageJ (39), and the size of a cell was approximated by the maximum Feret diameter of the cell contour. We measured a minimum of 10 cells per sample and used the average Feret diameter (in microns) as a cell size estimate.
Measuring bacterial density and survival. In vitro bacterial density during stationary phase was measured on agitated LB cultures incubated at 28°C for 91 h. Five replicate experiments were performed, and appropriate dilutions were streaked on NBTA plates with 50 g kanamycin per ml to estimate density. In vivo bacterial density was estimated by injecting approximatively 2000 cells (i.e., 20 l of cultures diluted to 1:1,000) in the last instar larvae of the lepidopteran Galleria mellonella as previously described (19). Insect cadavers were homogenized in 100 l LB after 5 days of incubation at 28°C, and appropriate dilutions were streaked on NBTA to estimate density.
We estimated bacterial survival on three isolates (G1#23, G2#25, and G3#9) for which we ran 12 independent agitated LB cultures at 28°C. Samples were taken from each culture in late exponential phase, early stationary phase, and late stationary phase. For each sample, we estimated the total number of cells using a Thoma cell counting chamber and the number of cultivable cells by plating appropriate dilutions of samples on NBTA plates and counting the number of CFU. We then used the proportion of cultivable cells as a proxy for the proportion of viable cells and studied how it varied from exponential phase to early or late stationary phase.
Measuring competitive ability. We measured competitive ability of three fluorescent isolates (G1#23, G2#25, and G3#9) by letting them compete in LB cultures with a nonfluorescent primary variant (F1V1). Twelve independent late-stationary-phase cultures of each of the four isolates were used to inoculate these cultures. Before inoculation, cultures of group 2 and 3 variants were 10-fold diluted so that all groups of variants have similar densities. The two competing variants (125 l each) were then mixed, incubated for 120 h at 28°C, and plated on NBTA. Pictures of these plates were taken using an Olympus Axiozoom (x7) under fluorescent light (535 nm), so that fluorescent and nonfluorescent colonies could be distinguished. The densities of GFP and non-GFP cells were estimated at the onset of the experiment and after 5 days of incubation. From these estimates, we could compute a competitive index (CI) as the rate of increase in frequency of GFP bacteria over 5 days of incubation.
Measuring virulence toward insects. We measured virulence as the time to kill larvae of G. mellonella. This was done for five representative isolates of each group (isolates G1#21, G1#23, G1#42, G1#44, and G1#51 for group 1, isolates G2#25, G2#29, G2#36, G2#39, and G2#40 for group 2, and all members of group 3) and for three injected doses (corresponding to 20 l of a 1:1,000, 1:100, or 1:10 diluted culture). We conducted two replicate experiments and injected each dose four times. Appropriate dilutions of each injected culture were plated on NBTA to estimate the injected dose. We used the automated procedure described by Parthuisot et al. (40) to measure time of insect death, and we analyzed data with a Cox proportional hazard model, with variation among replicate experiments and variation among isolates within each group considered a Gaussian random effect. This analysis has been performed using the coxme library (41).