Transcriptome dynamics of the Myxococcus xanthus multicellular developmental program

The bacterium Myxococcus xanthus exhibits a complex multicellular life cycle. In the presence of nutrients, cells prey cooperatively. Upon starvation, they enter a developmental cycle wherein cells aggregate to produce macroscopic fruiting bodies filled with resistant myxospores. We used RNA-Seq technology to examine the transcriptome of the 96 hr developmental program. These data revealed that 1415 genes were sequentially expressed in 10 discrete modules, with expression peaking during aggregation, in the transition from aggregation to sporulation, or during sporulation. Analysis of genes expressed at each specific time point provided insights as to how starving cells obtain energy and precursors necessary for assembly of fruiting bodies and into developmental production of secondary metabolites. This study offers the first global view of developmental transcriptional profiles and provides important tools and resources for future studies.


Introduction
Myxococcus xanthus is a soil-dwelling d-proteobacterium that exhibits a complex multicellular life cycle with two phases: growth and starvation-induced development (Muñoz-Dorado et al., 2016). When nutrients are available, cells divide to produce a community known as swarm. Swarms are predatory (although not obligate) and can digest prokaryotic and eukaryotic microorganisms (Pérez et al., 2016). Upon starvation, cells in the swarm enter a developmental program, during which they migrate into aggregation centers and climb on top of each other to build macroscopic structures termed fruiting bodies. To form fruiting bodies, starving cells glide on solid surfaces by using two motility systems, known as A-(adventurous) and S-(social) motility, which allow individual cell movement or group movement that requires cell-cell contact, respectively (Mauriello et al., 2010;Nan et al., 2014;Islam and Mignot, 2015;Chang et al., 2016;Schumacher and Søgaard-Andersen, 2017). After completion of aggregation (24 hr post-starvation), cells differentiate into environmentally resistant myxospores, which are embedded in a complex extracellular matrix (Figure 1). Each fruiting body contains » 10 5 -10 6 myxospores. Interestingly, only » 10% of the starving population become myxospores (O'Connor and Zusman, 1991a) as most cells (around 60%) undergo programmed cell death, most likely to provide the rest of the population enough nutrients to successfully build fruiting bodies (Wireman and Dworkin, 1977;Nariya and Inouye, 2008). The remaining cells differentiate into a persister-like state, termed peripheral rods (PR), which surround the fruiting bodies (O'Connor and Zusman, 1991a;O'Connor and Zusman, 1991b; O'Connor and correlation >0.98), with the exception of 24 hr samples (R 2 correlation = 0.80), which may be related to lack of synchrony between cells in the transition from aggregation to sporulation. The median of both values was utilized for further analysis ( Table 1 and Table 1-source data 1).
As a first data validation step, we determined the expression profiles of two different developmental genes using b-galactosidase transcriptional reporters [spiA::Tn5-lacZ (strain DK4322) and fmgE::Tn5-lacZ (strain DK4294)] (Kroos et al., 1986) from cells developed under the same conditions used in this study. Comparison of these b-galactosidase activities to the RNA-Seq data indicated the patterns were similar and in agreement with those obtained with other strategies (Figure 2). Moreover, the expression profiles of many genes that have been previously characterized from b-galactosidase reporter activity, qRT-PCR, or microarray analyses were compared with those obtained with our data. This analysis revealed a general agreement (Figure 2-source data 1).

Gene expression profiles organize into 10 developmental groups (DGs)
To identify developmentally regulated transcripts with similar expression patterns, genes containing measured RPKM (reads per kilobase pair of transcript per million mapped reads) values for all time points were further analyzed. First, all genes with <50 reads and/or high replicate variability between the two replicate datasets (R 2 correlation <0.7) were removed (Table 1-source data 2). 1557/7229 (21.5 %) genes passed quality criteria filters. Some of these genes (142; 9.1%) were not significantly (>2 fold) up-or down regulated during the developmental program, suggesting these genes could be considered constitutively expressed. 1415 significantly regulated genes (90.9% of the genes passing quality control) were then analyzed for clusters of similar expression patterns. Briefly, initial evaluation of several clustering methods (see Materials and methods for details) revealed kmeans clustering with 6-12 clusters produced the best clustering of genes with similar expression profiles. Refinement of kmeans clusters by visual inspection indicated 10 DGs best explained the number of unique gene profiles ( Figure 3A, Figure 3-figure supplement 1, and Figure 3-source data 1). Clusters were organized with peak expression profiles corresponding to progression through the developmental program ( Figure 3A). The relative expression profiles presented here in the heat maps are a log 2 normalized RPKM value relative to the mean of the entire RNA-Seq trajectory for a given gene. This relative expression profile is log 2 (RPKMi time-point x/RPKMi average), where i is a given gene, x is a given time-point, and RPKM average is the average RPKM value of all time-points. Although the sensitivity of the RNA-Seq technology allows detection of genes with low expression levels, some genes were excluded in this analysis after removing those with less than 50 reads. This is especially important for genes that are expressed during growth at levels not detectable with this technology. Only 13 genes were found in this situation ( Figure 3-source data 2), which have not been included in the analyses shown below. DGs 1 and 2 contained genes that were immediately down-regulated relative to growth conditions, DGs 3-5 correlated with the aggregation phase, DGs 6-7 correlated with the transition from aggregation to sporulation, and DGs 8-10 correlated with the sporulation phase ( Figure 3A). The number of genes attributed to each cluster ranged from 5.5% (DG3) to 14.1% (DG8) (Figure 3-figure supplement 2). Similar proportions of genes exhibited peak expressions in growth (21.8%), aggregation (23.8%), and transition from aggregation to sporulation (18.8%) phases. The final phase accounted for peak expressions of 35.6% of genes, consistent with the significant morphological and physiological rewiring that must occur during spore differentiation and preparation for extended quiescence.
When the number of genes by RPKM values that account for 50% of all mRNA expression at each time point was compared, it was observed that this number was low (~300) during growth, and at 48 hr it steadily increases to 762 by the end of sporulation (Figure 3-figure supplement 3). This suggests that the transcriptome is becoming distributed across a broader number of genes during development. The following source data is available for figure 2: Source data 1. Comparison of expression profiles of several developmental genes described in the literature and included in the DGs with the RPKM profiles from this study. DOI: https://doi.org/10.7554/eLife.50374.007 Source data 1. RPKM values for developmentally controlled genes and distribution of the genes in the ten developmental groups. DOI: https://doi.org/10.7554/eLife.50374.013 Source data 2. Genes with no reads at time 0 hr. DOI: https://doi.org/10.7554/eLife.50374.014 Source data 3. Previously reported developmental genes and identification in the 10 developmental groups shown in Figure 3A. DOI: https://doi.org/10.7554/eLife.50374.015 Source data 4. Genes involved in development that are not included in the DGs shown in Figure 3A   As a first step in analysis of the DGs, genes that have been previously described to affect M. xanthus development were identified. Of the~280 characterized genes, 167 were included in the DGs. Most showed peak expression patterns at a time point that matched with the developmental phase where they have been reported to function, some of which have a well-defined role on development (Figure 3-source data 3 and Figure 3-figure supplement 4). A notable exception to this are genes that are thought to function in very early stages during development, but are included in DG10. Examples of these genes are sasS, which is involved in A signaling (Yang and Kaplan, 1997); csgA, which encodes the C signal (Shimkets and Rafiee, 1990); and romR, which is involved in polarity control of motility (Leonardy et al., 2007). However, DG10 genes also showed high expression at 6 hr, consistent with early activities and raising the possibility of additional functions at the latest stage of development. Finally, not all known developmental genes appeared in the 10 DGs defined here (Figure 3-source data 4). Of the 108 missing genes, 28 genes encoded TCSs and 20 serine/threonine protein kinases (STPKs).

Genes identified through chemical induction of sporulation mainly map to DG7 and DG8
A core sporulation transcriptome was previously defined using an artificial method for inducing spore differentiation (Müller et al., 2010). Using this method, myxospores can be induced in cells growing in rich broth culture by addition of chemicals such as 0.5 M glycerol (Dworkin and Gibson, 1964). Chemical-induced sporulation bypasses the requirement for starvation, motility (aggregation), and alternate cell fates . Comparison of these two transcriptome sets determined that 1388 genes passed quality criteria in both transcriptome studies ( Figure 3B, Figure 3source datas 5 and 6, Figure 3C, and Figure 3-source data 7). 92% (218/237) of genes significantly up-regulated (>2 fold) in the sporulation transcriptome were also significantly up-regulated in the developmental transcriptome. These genes were significantly over-represented in DGs 7 and 8, with low probability (p) that this association is due to random chance (p=1Â10 À12 and p=2Â10 À11 , respectively). DGs 7 and 8 peak expression at the transition to sporulation and sporulation phases, respectively. Co-regulated genes in these groups include those predicted to be involved in generation of sugar precursors for spore coat (see below). Likewise, genetic loci involved in spore coat synthesis (exo) and surface polysaccharide arrangement (nfs) (Müller et al., 2012;Holkenbrink et al., 2014) fell in DG8.
Interestingly, DGs 9 and 10, with peak expression profiles corresponding to the latest developmental time points, were not well represented in the up-regulated sporulation transcriptome. DG9 genes were significantly over-represented in the down-regulated sporulation transcriptome (p=0007). The DG9 cluster represents genes that were expressed during growth in rich media, down-regulated in response to starvation, and later up-regulated during the final phases of development ( Figure 3A). Genes in this cluster appear to be involved in transport, respiration, and transcriptional regulation, and may be required during rapid growth and perhaps in the final maturation phases of sporulation. It is likely that the chemical-induced sporulation transcriptome may not have included very late induced sporulation genes as RNA was harvested until four-hours after induction. Chemical-induced spores are heat and sonication resistant at this stage, but final maturation may continue past this point. Finally, DG10 contained genes that were over-represented (p=7Â10 À10 ) in the not-regulated sporulation transcriptome gene set. We speculate this pool of genes may be present in PRs, a cell fate enriched late in the developmental program and not present during chemical induction of sporulation. Genes in this DG encode several STPKs, the bEBP Nla26, and several proteins involved in secondary metabolite (SM) biosynthesis (see below).
Good correlation was observed in genes that were expressed during growth and down-regulated in both transcriptome sets, with genes from DGs 1, 2, 5, and nine being over-represented in the down-regulated sporulation transcriptome set ( Figure 3C). Of the large pool of genes that were not significantly regulated in the sporulation transcriptome (754), a relatively small number (12%) were also not significantly up-or down-regulated in the developmental transcriptome. This pool of genes likely represents constitutively expressed genes that serve as good normalization markers ( Figure 3source data 5) and includes housekeeping genes such as the transcription termination factor Rho (MXAN_RS11995), the cell cytoskeletal protein MreB (MXAN_RS32880) (Müller et al., 2012;Treuner-Lange et al., 2015;Fu et al., 2018), the gliding motility and sporulation protein AglU (MXAN_RS14565) (White and Hartzell, 2000), and the CheA homolog DifE (MXAN_RS32400), which is required for exopolysaccharide production and social motility (Yang et al., 2000).

Analysis of developmental regulated genes
All genes included in the 10 DGs were individually analyzed to find out which processes were affected during development, which may explain the different events that occur during aggregation and sporulation. Here, we have focused on six different processes.

A-and S-motility genes exhibit different developmental expression profiles
The two phases of the M. xanthus lifecycle (predatory growth and development) depend on A-and S-motility engines and their associated regulatory proteins (Pérez et al., 2014;Islam and Mignot, 2015;Mercier and Mignot, 2016;Schumacher and Søgaard-Andersen, 2017). Our data have revealed that many of the motility genes were developmentally regulated and that the expression profiles of the distinct A-and S-machinery genes clearly differed during development (Figure 4). A-motility genes were up-regulated during early development, while S-motility genes first decreased at 6 hr, and then returned to growth levels during aggregation (except for pilA), suggesting that A motility is preferentially used by cells during aggregation. During sporulation, expression of genes encoding both motility engines decreased (Figure 4), although some of them peaked again at this stage. The peak of some A-motility genes during sporulation is consistent with the repurposing of certain A-motility proteins to function in spore coat assembly (Wartel et al., 2013). At the end of development, the expression levels of the motility genes remain high. However, it should be reminded that although mature fruiting bodies are static structures, the PRs surrounding them are motile (O'Connor and Zusman, 1991b).
Gene expression patterns are consistent with use of glycogen and lipid bodies as energy sources During growth, M. xanthus does not appear to consume sugars as carbon or energy sources (Watson and Dworkin, 1968;Bretscher and Kaiser, 1978). Instead, pyruvate, amino acids, and lipids are efficiently utilized which directly enter the tricarboxylic acid (TCA) cycle . It has been long debated as to whether M. xanthus utilizes a fully functioning glycolytic pathway (Curtis and Shimkets, 2008). It was speculated that the pathway may be utilized primarily in the gluconeogenic direction to produce sugar precursors necessary for spore coat production (Youderian et al., 1999;Chavira et al., 2007;Getsin et al., 2013). It is unknown how these pathways contribute to energy production during development when starving cells must synthesize energy currencies (i.e. ATP) over a period of at least three days.
Analysis of the DGs revealed that most genes involved in energy generation (pyruvate dehydrogenase complex, TCA cycle and oxidative phosphorylation) were found in DG2 ( Figure 5A). Therefore, although they are down-regulated at 6 hr, they are later up-regulated at lower levels than during growth. In contrast, many genes encoding enzymes of the glycolytic/gluconeogenic pathway were up-regulated during development, with most reaching maximum expression levels at the completion of aggregation (24-48 hr) ( Figure 5B). This up-regulated group included genes encoding homologs of glucokinase (glkC) and phosphofructokinase (pfkA), which are specific for the glycolytic pathway. These observations are consistent with a transcriptional rewiring of metabolic pathways during the developmental program, perhaps to take advantage of changing carbon/energy sources. For instance, developmental up-regulation of the glycolytic pathway genes may allow developing cells to obtain energy from sugars released from cells undergoing developmental lysis or from glycogen. Glycogen accumulates during late stationary phase/early development, and then disappears prior to sporulation (Nariya and Inouye, 2003). Enzymes predicted to be involved in synthesis of glycogen, such as GlgC, were found in DG4, while enzymes involved in utilization of glycogen, such as trehalose synthase, GlgP, and MalQ appeared to be constitutively expressed. The observation that these competing pathways show overlapping expression profiles suggests that regulation of glycogen production/consumption is likely regulated post-transcriptionally, as has been demonstrated by phosphorylation of PfkA by Pkn4 (Nariya and Inouye, 2003).
Lipid bodies also accumulate in cells prior to sporulation. They mainly consist of triacylglycerides derived from membrane phospholipids as cells shorten in length, which are later used as an energy source (Hoiczyk et al., 2009;Bhat et al., 2014). However, the profiles of genes involved in both straight-and branched-chain primer synthesis and elongation of fatty acids observed in this study ( Figure 6A and B) suggest that some level of lipid synthesis occurs during development. Moreover, genes involved in the alternative pathway to produce isovaleryl-CoA  were induced ( Figure 6A and B). These lipids might be either incorporated into lipid bodies, be responsible for changes in lipid composition of the membranes of myxospores and/or PRs, and/or be used as precursors for SMs production. On the other hand, genes involved in lipid degradation reached maximum expression at 24-48 hr (b-oxidation) or even at later times (by other pathways) ( Figure 6C). These fatty acid degradation enzyme profiles suggest that lipids are preferably consumed during sporulation.
These data offer an overall picture of the central metabolism during development, which reinforces the notion that macromolecules recycled from growth phase or released from lysing cells can be directly used to yield energy, but are also used to synthesize glycogen and lipids that are stored for later consumption.
Amino acid and sugar precursors required for developmental macromolecule synthesis may be released by protein and polysaccharide turnover and gluconeogenesis In addition to energy, starving cells need a source of sugar precursors to synthesize developmentally specific polysaccharides required for motility (Li et al., 2003), fruiting body encasement (Lux and Shi, 2005), spore coat synthesis (Kottel et al., 1975;Holkenbrink et al., 2014), and spore resistance (McBride and Zusman, 1989). It has been suggested that these sugars are derived from gluconeogenesis (Youderian et al., 1999). Consistently, our data have revealed that genes encoding enzymes specific for gluconeogenesis, such as phosphoenolpyruvate carboxykinase was in DG2, and GlpX (fructose-1,6-bisphosphatase) were present during growth and throughout development ( Table 1-source data 2). Thus, these observations, as well as those presented above ( Figure 5B), indicate that gluconeogenesis likely contributes to sugar precursor production at various stages during development. Moreover, the observation that four glycosyl hydrolases were specifically up-regulated during development ( Figure 7A) suggests the cells may recycle vegetative polysaccharides or scavenge polysaccharides released from cells induced to lyse. These released free monomers could be synthesized into specific developmental polysaccharides by the series of glycosyl transferases that are also developmentally up-regulated ( Figure 7B).
The developmental program is mainly triggered by amino acid starvation, yet developmentally specific proteins need to be newly synthesized. A source of these amino acids could be released from lysed cells and turnover of proteins that are not required during development. Consistently, 60 proteases appeared in the 10 DGs. 16 of these were specific to growth and down-regulated after 6 hr of starvation, while the rest were sequentially up-regulated at different time points ( Figure 7C). All of these proteases likely release amino acids, but some of them may additionally function in regulatory processes, as has been reported for the protease PopC (DG7), which is thought to function in C-signal generation by cleavage of CsgA p25 to generate the p17 fragment (Lobedanz and Søgaard-Andersen, 2003;Rolbetzki et al., 2008). Genes involved in secondary metabolism are developmentally up-regulated M. xanthus produces multiple SMs, some of which (i.e. myxovirescine and myxoprincomide) facilitate predation (Xiao et al., 2011;Müller et al., 2016). However, we have found that although genes responsible for their biosynthesis are expressed during growth, their expression increases during development ( Figure 8A-D). The M. xanthus genome codes for 18 nonribosomal peptide synthetases (NRPS), 22 polyketide synthases (PKS), and six mixed PKS/NRPS genes located in regions predicted to be involved in SMs synthesis (Korp et al., 2016). Of these, 10 were developmentally upregulated, and 9/10 were found in DGs 8, 9 and 10 ( Figure 8C-E). Moreover, 129 genes assigned to the DGs were located in those SM regions (Korp et al., 2016), and 51 of these were also identified in DGs 8, 9, and 10 ( Figure 3-source data 1).
Together, these observations suggest that SMs may play previously unrecognized roles during development. For instance, they may be used to protect developing cells from other microbes in the soil, kill competitors to yield nutrients, or used as signaling molecules. In agreement with the data presented above concerning chemical-induced sporulation, an intriguing possibility is that PRs may produce SMs to defend spores inside the fruiting bodies or to release nutrients from prey to promote germination.

Translation may be developmentally rewired
It was previously reported that the protein composition of ribosome complexes purified from growing cells versus myxospores was different (Foster and Parish, 1973), but this intriguing observation has not been further pursued. Data from this study suggest a role for translation regulation in control of M. xanthus development. 76 genes involved in translation fit the criteria for inclusion in the DGs, which represents 5.4% of the genes included in them. Most of the genes involved in translation (i.e. encoding ribosomal proteins, initiation, elongation and termination factors, ribosome maturation and modification proteins, and several aminoacyl-tRNA ligases) are down-regulated by 6 hr, likely as the result of the stringent response (Starosta et al., 2014). Interestingly, most of them were subsequently up-regulated to a level similar to, or even higher than that observed during growth ( Figure 9A and B). Some genes, including several aminoacyl-tRNA ligases, were mainly expressed during sporulation ( Figure 9A and B).
The observation that many ribosomal proteins are differentially regulated during development ( Figure 9B) suggests that their relative ratio varies during development, yielding ribosomes with altered protein composition. It is remarkable the differences in expression profiles exhibited by duplicated genes of ribosomal proteins. M. xanthus encodes two paralogs for proteins S1, S4, S14, L28, and L33 (Table 1-source data 2). Only one of the two paralogous genes for ribosomal proteins S1 and S14 was found in the DGs (Figure 3-source data 1). Most notably, the two S4 paralogs were in the DGs, and exhibit similar RPKM values during growth. However, the RPKM for MXAN_RS16120 was 10-fold higher than that for MXAN_RS32955 at 48 hr of development ( Figure 9C). Neither of the two paralogs of the 50S subunit were found in the DGs (Table 1-source data 2). These data confirm the previous results of Foster and Parish (1973), and provide a new overall perspective on the changing composition of translational machinery during development.
While it is possible that some of the transcriptional changes observed in the translational machinery could be related to ribosomal hibernation in myxospores and/or PRs (Yoshida and Wada, 2014;Harms et al., 2016;Gohara and Yap, 2018), we speculate that regulation of translational machinery may play an important role in directing the developmental program. It is known that some bacteria build alternative ribosomes to improve fitness on different growth conditions by altering the core ribosomal protein stoichiometry and differential expression of paralogous ribosomal proteins (Foster and Parish, 1973;Nanamiya et al., 2004;Hensley et al., 2012;Prisic et al., 2015;Slavov et al., 2015). In addition, some ribosomal proteins also play extraribosomal functions, such as control of transcription or mRNA decay (Warner and McIntosh, 2009).

A large interconnected regulatory network controls development
M. xanthus encodes a large repertoire of signaling/regulatory proteins presumed necessary to direct and coordinate its multicellular lifecycle in response to extra-and intra-cellular cues. Examples include one-component systems (OCS; transcriptional regulators that contain a sensing domain), TCS genes (sensor histidine kinase [HK] and response regulator [RR] proteins connected by phosphorelay), alternative sigma factors, and STPKs. Many of these proteins have been characterized in this bacterium, and in some cases, individual signal-transduction pathways and their exact role in controlling development have been defined Schramm et al., 2012;Muñoz-Dorado et al., 2014;Rajagopalan et al., 2014). However, the data presented here provide for the first time a view of the entire developmental cycle as an integrated system. We observed developmental up-or down-regulation of a larger number of regulatory genes than previously reported (Muñoz-Dorado et al., 2014;Rajagopalan et al., 2014). A significant number of these genes showed expression patterns that suggested they play important roles in modulation of aggregation and/or sporulation. 21 OCS genes were developmentally regulated, 9 of which were down-regulated. The remaining were differentially up-regulated during development ( Figure 10A), including the repressor LexA (Campoy et al., 2003), and the regulators SasN (Xu et al., 1998) and MrpC (Sun and Shi, 2001;Ueki and Inouye, 2003;McLaughlin et al., 2018).
Of the 272 encoded TCS genes, we found 47 included in the DGs: 23 HKs, 10 hybrid HKs (HyHKs, containing HK and RR modules in the same polypeptide), and 14 RRs. 6 TCS genes were shut down during development ( Figure 10B). Only 16 of 47 TCS genes have been previously characterized, thus pinpointing additional candidates for further characterization. Additionally, 5 of the 14 RRs belong to the group of bEBPs (plus MXAN_RS07605, which exhibits an architecture similar to bEBPs, but contains a GAF instead of a receiver domain). bEBPs function to activate expression from sigma 54 dependent promoters (Morett and Segovia, 1993). Besides the bEBPs, FruA is the only other RR found in the DGs that contains a DNA-binding domain. Out of the remaining 8 RRs, five were standalone receiver domains (CheY-like), two contain a putative diguanylate cyclase output domain, one of which has been characterized (Skotnicka et al., 2016), and one is RomR, which modulates motility (Leonardy et al., 2007). As many of the HKs and RRs that were developmentally regulated are orphans, the results presented here may help to identify cognate pairs.
Regarding other transcription factors, nine sigma factor genes were found to be developmentally regulated in a sequential fashion ( Figure 10C), including those encoding the major sigma factor RpoD (Inouye, 1990) and RpoN (sigma 54) (Keseler and Kaiser, 1997). The expression profiles of these two sigma factor genes clearly differ during development. While rpoD was down-regulated at 6 hr, up-regulated during aggregation, and then down-regulated during sporulation, rpoN is up-regulated throughout development ( Figure 10D). The expression profiles of rpoN and bEBP genes are consistent with previous results demonstrating that a bEBP cascade is triggered upon starvation (Giglio et al., 2011). Additionally, sigC, which encodes a group II sigma factor (Apelian and Inouye, 1993), was identified in DG7. The remaining six are predicted to encode extracytoplasmic functions (ECF) sigma factors, including rpoE1, involved in motility (Ward et al., 1998). It is noteworthy that the genes encoding the four subunits of the core RNA polymerase are developmentally regulated with profiles similar to that observed for the ribosomal proteins ( Figure 10E).
Together, these observations suggest that a high number of regulators directs the developmental program of M. xanthus, with some acting simultaneously and others sequentially to perfectly modulate the different events that occur through development. In other model bacteria such as Caulobacter crescentus, more than 19% of the genes have been found to be developmentally regulated (Laub et al., 2000), a similar percentage to that found in M. xanthus, 57% of which are under direct control of the five master regulators identified in this bacterium (Zhou et al., 2015). In Bacillus subtilis, out of 4100 genes, 520 were dependent on Spo0A, but not on s F , while 66 were dependent upon both regulatory proteins (Fawcett et al., 2000). And in the case of Streptomyces coelicolor, 1901 genes (24% of the ORFs) exhibit differences when the substrate mycelium differentiates to a multinucleated mycelium, with a large number of transcriptional regulators involved (Yagüe et al., 2013). Although the number of regulators included in the DGs is larger in M. xanthus than in other model bacteria, it remains to be elucidated how many of them directly modulate development.  Toward elucidation of a complete developmental gene-regulatory circuit M. xanthus is an extraordinary bacterium with a large coding potential and a complex lifecycle. The developmental cycle consists of two consecutive events: aggregation and sporulation, during which cells segregate into three different cell fates. Moreover, this process lasts over three days and is triggered by starvation. During this extended period, starving cells must glide to build fruiting bodies and synthesize numerous macromolecules exclusive to fruiting bodies and myxospores. Consistent with this complexity, the transcriptomic analyses presented here revealed that 1415 genes are developmentally regulated with a high degree of confidence, exhibiting expression times that peak at either growth, aggregation and/or sporulation. Analysis of individual genes and the processes in which they participate has shed some light about how cells regulate the expression of motility genes to allow the cells to reach the aggregation centers, or how they rewire metabolism to both obtain energy and monomers to build new macromolecules or to synthesize SMs. Moreover, these data have also revealed that the translational and transcriptional machinery is deeply altered to modulate the different events of development, offering new insights that require to be experimentally pursued to determine the function of all these regulators.
Although the role of translation in the regulation of the developmental cycle has not yet been addressed, a large number of transcriptional regulators that function during development have been identified (Figure 3-figure supplement 4, and Figure 11-source data 1). The data presented here corroborate the results obtained by other myxobacteriologists. For instance, our analyses are in good agreement with an established model in which starvation activates four genetic regulatory networks: bEBP, Mrp, FruA, and Nla24 modules, with the latter being activated by cyclic-di-GMP (Kroos, 2017). However, the high number of transcriptional regulators and signaling proteins found in this study to be developmentally regulated at the mRNA level is much higher than expected. As shown in Figure 11, many uncharacterized regulators peak at either aggregation or sporulation (Figure 11-source data 1). Those that are expressed at the same time point may be interconnected to properly modulate specific events and guarantee the proper sequential expression of genes. The analysis of the expression levels of the different regulators has revealed that fruA and mrpC exhibit maximum RPKM values of 7046 and 4415, respectively, while none of the others reach 1000 (except for two sigma factors) ( Figure 11-source data 1). This may explain why FruA and MrpC are considered master regulators of development, while many of the rest have not been identified as playing crucial roles in the lifecycle of M. xanthus. Undoubtedly, the developmental mRNA expression profiling presented here will act as a blueprint for the complete elucidation of the M. xanthus developmental regulatory program. Now that the changes in gene expression are measured, identifying the regulatory inputs of each promoter will be critical to understand the complete genetic circuitry controlling development. Preparation of cells for RNA-Seq experiment M. xanthus strain DK1622 (Kaiser, 1979;Goldman et al., 2006) was used in this study. Cells were grown in CTT liquid medium (Hodgkin and Kaiser, 1977) at 30˚C with vigorous shaking (300 rpm) to 3.0 Â 10 8 cells/ml (optical density at 600 nm [OD 600 ] of 1), and then harvested and resuspended in TM buffer (10 mM Tris-HCl [pH 7.6]; 1 mM MgSO 4 ) to a calculated density of 4.5 Â 10 9 cells/ml (OD 600 of 15). For each time replicate, 200 ml aliquots of concentrated cell suspension were spotted onto thirteen separate CF agar plates (Hagen et al., 1978). Two replicates of cells were harvested from plates at 6, 12, 24, 48, 72 and 96 hr (samples WT_6, WT_12, WT_24, WT_48, WT_72 and WT_96, respectively), and the obtained pellets were transferred immediately into 0.5 ml of RNA Protect Bacteria Reagent (Qiagen). Cells were then incubated at room temperature for 5 min, harvested by centrifugation at 5000 g for 10 min (4˚C), and stored at À80˚C after removal of the supernatant. For the t = 0 samples (sample WT_0), two replicates of 30 ml of the original liquid culture (OD 600 of 1) were harvested by centrifugation as above, resuspended in RNA Protect Bacteria Reagent, and processed in the same manner.

RNA extraction
To isolate RNA, frozen pellets were thawed and resuspended in 1 ml of 3 mg/ml lysozyme (Roche) and 0.4 mg/ml proteinase K (Ambion) prepared in TE buffer (10 mM Tris-HCl; 1 mM ethylenediaminetetraacetic acid [EDTA], pH 8.0) for cell lysis. Samples were incubated 10 min at room temperature. RNA extraction was carried out using the RNeasy Midi Kit (Qiagen) and each sample was eluted in 300 ml of RNase-free water. The concentration of RNA was measured using a NanoDrop ND-2000 spectrophotometer (NanoDrop Technologies, USA). To remove DNA, each RNA sample was supplemented with 1 unit of DNase I (from the DNA Amplification Grade Kit of Sigma) per mg of RNA and incubated at room temperature for 10 min. The reaction was stopped by adding the stop solution included in the kit and incubating 10 min at 70˚C. The obtained RNA was precipitated with 1/10 vol of 3 M sodium acetate and 3 volumes of ethanol, and resuspended in 50 ml of RNasefree water. The quality of the total RNA was verified by agarose gel electrophoresis, and the concentration was determined using NanoDrop as indicated above. Sequencing and transcriptomic data analysis

Double stranded copy DNA synthesis
The cDNA from two biological replicates of each condition (see above) was used for sequencing using the Illumina HiSeq2000 (100 bp paired-end read) sequencing platform (GATC Biotech, Germany). Sequence reads were pre-processed to remove low-quality bases. Next, reads were mapped against M. xanthus DK1622 ribosomal operon sequences using BWA software with the default parameters ). Remaining reads were subsequently mapped to the genome sequence with the default parameters and using the pair-end strategy. SAMtools ) was used to convert resulting data into BAM format. Artemis v.16.0.0 (Rutherford et al., 2000) was used for the visualization of the sequence reads against the M. xanthus genome. Once the transcripts were mapped to the genome, the average median value for each condition was used in further analyses ( Table 1 and Table 1-source data 1 and 2).

Developmental gene analysis
Genes with fewer than 50 reads in a given time point were removed from analysis. RPKM values of the remaining genes were then compared across the developmental time-points. Developmental expression was characterized by a > 2 fold change in RPKM values across the time course and a > 0.7 R 2 correlation coefficient of the two time course replicates. Log 2 fold-change calculations were performed using the Cluster 3 Software Package (de Hoon et al., 2004). Genes passing these criteria are presented in Figure 3-source data 1. Randomization of the time point RPKM values within each replicate data set yielded a false discovery rate of 3.97% based on five randomized simulations that scrambled the order of the time points across all genes in the two datasets. Genes passing the developmental expression criteria were hierarchically clustered using pearson correlation, spearman rank, euclidian distance, and kmeans clustering. By visually inspecting the clusters we found that kmeans clustering gave the best clustering of genes with similar expression profiles.

Assay of b-galactosidase activity
For quantitative determination of b-galactosidase activity during development, strains containing lacZ fusions (Figure 2) were cultured and spotted onto CF plates as described above. Cell extracts were obtained at different times by sonication and assayed for activity as previously reported (Moraleda-Muñoz et al., 2003). The amount of protein in the supernatants was determined by using the Bio-Rad Protein Assay (Bio-Rad, Inc) with bovine serum albumin as a standard. Specific activity is expressed as nmol of o-nitrophenol (ONP) produced per min and mg of protein. The results are the average and associated standard deviation from three independent biological replicates.

Microscopy
To observe swarm and fruiting bodies, 10 ml of cell suspension prepared as mentioned above were spotted onto CTT (for growth) or CF (for development) agar plates and incubated at 30˚C. Observation was on an Olympus SZX7 dissecting microscope. For scanning electron microscopy, samples obtained from CF and CTT agar plates were fixed with glutaraldehyde vapors for 24 hr at room temperature and then postfixed in aqueous 1% osmium tetroxide for 1 hr at 4˚C, washed three times in buffer, and poststained for 2 hr in buffered 0.5% uranyl acetate. Dehydration was accomplished by a graded series of ethanol. Samples were then critical-point dried and sputter coated with gold. Photographs were taken in a Zeiss DSM950 scanning electron microscope.
The following dataset was generated: Author (