Enhancement of Microbial Biodesulfurization via Genetic Engineering and Adaptive Evolution

In previous work from our laboratories a synthetic gene encoding a peptide (“Sulpeptide 1” or “S1”) with a high proportion of methionine and cysteine residues had been designed to act as a sulfur sink and was inserted into the dsz (desulfurization) operon of Rhodococcus erythropolis IGTS8. In the work described here this construct (dszAS1BC) and the intact dsz operon (dszABC) cloned into vector pRESX under control of the (Rhodococcus) kstD promoter were transformed into the desulfurization-negative strain CW25 of Rhodococcus qingshengii. The resulting strains (CW25[pRESX-dszABC] and CW25[pRESX-dszAS1BC]) were subjected to adaptive selection by repeated passages at log phase (up to 100 times) in minimal medium with dibenzothiophene (DBT) as sole sulfur source. For both strains DBT metabolism peaked early in the selection process and then decreased, eventually averaging four times that of the initial transformed cells; the maximum specific activity achieved by CW25[pRESX-dszAS1BC] exceeded that of CW25[pRESX-dszABC]. Growth rates increased by 7-fold (CW25[pRESX-dszABC]) and 13-fold (CW25[pRESX-dszAS1BC]) and these increases were stable. The adaptations of CW25[pRESX-dszAS1BC] were correlated with a 3-5X increase in plasmid copy numbers from those of the initial transformed cells; whole genome sequencing indicated that during its selection processes no mutations occurred to any of the dsz, S1, or other genes and promoters involved in sulfur metabolism, stress response, or DNA methylation, and that the effect of the sulfur sink produced by S1 is likely very small compared to the cells’ overall cysteine and methionine requirements. Nevertheless, a combination of genetic engineering using sulfur sinks and increasing Dsz capability with adaptive selection may be a viable strategy to increase biodesulfurization ability.


Introduction
Biological removal of organic sulfur from petroleum has been investigated for more than twenty-five years as a possible cost effective and environmentally benign alternative to chemical removal strategies [1]. While the removal of sulfur from organic compounds in petroleum can be accomplished by chemical engineering processes such as hydrodesulfurization [2], a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 oxidative desulfurization [3,4], and extractive desulfurization [5], the removal of sulfur to the increasingly stringent levels required by environmental regulations is challenging. This is especially true of dibenzothiophene (DBT) and its alkylated derivatives (the most common organic sulfur compounds in petroleum, particularly in heavy fractions) which are recalcitrant to chemical desulfurization [2,6,7]. The specificity of biochemical reactions and the ability to achieve desulfurization at ambient temperatures and pressures, and without the need to supply hydrogen, has resulted in the continued interest in biodesulfurization [8]. The first desulfurization pathway characterized (the "4S pathway" mediated by enzymes DszA, DszB and DszC, encoded by the dszABC operon) is that from Rhodococcus erythropolis strain IGTS8 [1,[9][10][11][12]. The 4S pathway removes the sulfur from DBT while maintaining its caloric value. Currently available desulfurization biocatalysts, however, are not suitable for practical applications, as these would require large increases in enzyme activity [8] as well as more thermostable enzymes and hosts [13,14].
Analogs of the IGTS8 dszABC operon encoding enzymes that function above 50˚C have been characterized [15][16][17][18][19][20], and a great deal of work has been done to use modern genetic technology to enhance enzyme levels [21,22]. However, prior research concerning the biodesulfurization trait has demonstrated that attempts to overexpress the genes that encode the biochemical pathway for the sulfur-specific cleavage of carbon-sulfur bonds, dszABC, failed to yield dramatic improvements in desulfurization activity, and seem to have reached a ceiling that is far below the required level for an economical oil biodesulfurization process [8]. Moreover, the involvement of genes other than the dszABC genes in the optimal expression of the desulfurization trait has been demonstrated, suggesting that yet to be identified genes may contribute to the optimal expression of the biodesulfurization trait [23]. As a new approach, a synthetic gene S1 encoding a methionine and cysteine rich peptide (Sulpeptide 1) to act as a sulfur sink was designed and inserted into the dszABC operon (producing dszAS1BC), in a manner that would force its expression along with that of the dszA, B, and C genes. The hope was to produce a loop in which increasing expression of S1 would lead to increased dszABC expression which would lead to even greater S1 expression and so on [24].
After transforming recombinant plasmids containing either the dszABC or dszAS1BC constructs into desulfurization-negative Rhodococcus opacus, both recombinants were subjected to adaptive selection, in which cells were grown in chemically defined medium (CDM) with DBT as the sole sulfur source. Cultures were transferred into fresh CDM-DBT when growth reached the midpoint of log phase, so as to select for the fastest growing (and presumably fastest DBT metabolizing) strains. The idea was to combine this adaptive selection with the forced expression of S1 (to continually increase sulfur demand) in an effort to breach the ceiling for the expression of the Dsz enzymes. The extent of the first selection experiment (40 passages) served as proof of concept that this approach might succeed; adaptive selection in passages 30 through 40 led to a large increase in desulfurization activity, although the presence of the sulpeptide was correlated with a much smaller increase [24].
The experiments reported here constitute a more extensive and comprehensive continuation of this work. The desulfurization negative microbial host Rhodococcus qingshengii CW25 was selected for use in this study rather than a desulfurization competent host such as Rhodococcus erythropolis IGTS8 because the use of a dsz-negative culture allowed for the selection of the dsz trait in transformation experiments (specifically, we wished to compare strains both with and without expression of the sulpeptide, and the sulpeptide containing operon could only be constructed in vitro), and to better enable the detection of non-dsz genes that may contribute to the functioning of the biodesulfurization trait. We reasoned that mutations in nondsz genes that influence the biodesulfurization trait, and may already be present in R. erythropolis IGTS8, would be more readily detected in the non-desulfurizing culture R. qingshengii CW25. We chose not to alter the expression of genes such as dszD that are known to influence the desulfurization trait [14] because we sought to use adaptive evolution to learn how to prioritize genetic/biochemical targets for future research.
Our results indicate that improved efficiency of the utilization of sulfur and increases in the copy numbers of the dsz operon, without any other mutation, are sufficient to account for the increases in desulfurization that occurred during selection. The presence of the sulpeptide gene, however, does not drive the reaction faster, as anticipated, but rather enhances growth of the engineered bacteria, and appears to be responsible for augmenting their overall biodesulfurization efficiency. Although it was designed with a signal sequence to mediate its secretion from the cell, the sulpeptide protein may act as a sulfur reserve that prevents bacterial growth slowdowns when sulfur in the environment becomes scarce, which suggests that the use of sulfur sinks in combination with multiple copies of the dsz operon is an engineering strategy that may help the overall efficiency of biodesulfurization.

Bacterial strains, plasmids and media
Rhodococcus qingshengii CW25, which is desulfurization negative, was used as a host for expression of the desulfurization operon (dszABC) from Rhodococcus erythropolis IGTS8; strain CW25 has been deposited as NRRL B-65406 in the ARS Culture Collection (NRRL) of the USDA in Peoria, Illinois (USA). We had previously constructed plasmids pRESX-dszABC and pRESX-dszAS1BC [24]; they contain the native dszABC operon or its sulpeptide-containing derivative, respectively, cloned into the E. coli-Rhodococcus shuttle vector pRESX [25] under control of the kstD promoter (PkstD).
For most experiments testing desulfurization ability, cells were grown in CDM medium, prepared as described by Wang et al. [26], but without vitamins or yeast extract; kanamycin was also omitted from this medium to eliminate any sulfur contributed by its sulfate salt. To determine the minimal concentration of sulfur source required for maximum growth rate, CDM medium without kanamycin was made with either sulfate (ranging from 0-1 mM) or DBT (ranging from 0-0.5 mM) serving as the sole sources of sulfur; glucose (0.06 M) was the carbon source and growth was monitored by OD 600nm . The solubility of DBT in water is about 6 μM [27], so that DBT at minimal concentrations of 0.05 mM and above (see also below) replenished the soluble DBT as it was utilized by the cells.
Transformation of pRESX-dszABC and pRESX-dszAS1BC into strain CW25 CW25 cells were grown in LB medium at 30˚C, 200 rpm to an OD600nm of 0.4-0.5. Ampicillin was then added to the culture to a final concentration of 200 μg/ml (to weaken the cell wall) followed by 2 more hours at 30˚C, 200 rpm. Cells were harvested by centrifugation, washed 4 times with ice cold 0.3 M sucrose, and resuspended in 1 ml of ice cold 0.5 M sucrose.
30 μl of cells prepared in this way were mixed with 500 ng of either pRESX-dszABC or pRESX-dszAS1BC, transferred to an ice-cold electroporation cuvette (1 mm gap), and subjected to a single pulse at 1500 V, 200 μs. This was followed immediately by dilution with 1 ml of pre-warmed SOC medium [28] followed by incubation for 6 hours at 30˚C, 200 rpm. Cells were harvested by centrifugation, resuspended in 50 μl of SOC, and spread on LB plates containing 500 μg/ml kanamycin. Random colonies from these plates were selected and confirmed by colony PCR using primers specific for PkstD (to identify pRESX) and S1 (diagnostic for the dszAS1BC construct).

Adaptive selection
From LB-kanamycin (500 μg/ml) plate stocks a colony of each transformant (CW25[pRESX-dszABC] and CW25[pRESX-dszAS1BC]; denoted as "P0" in each case) was inoculated into CDM-0.05 mM DBT with either ethanol (0.043 M) or glucose (0.06 M) as the sole carbon source and grown at 30˚C, 200 rpm. Kanamycin (sulfate) was omitted from this medium, with plasmid stability forced by the need to retain the dszABC operon. Cultures were monitored spectrophotometrically during incubation, and were passaged (subcultured) into fresh medium when growth reached the middle of log phase. Thus, this protocol selected for the cells able to grow the fastest with DBT as sole sulfur source. Each passage averaged about 4 generations.
Three individual selection experiments, with some variation in parameters, were performed ( Table 1). At intervals, cultures were plated onto LB-kanamycin (500 μg/ml) plates to confirm culture purity. In the first experiment, a single colony was selected at intervals for determination of plasmid stability using colony PCR as described above, grown and assayed for DszABC specific activity, and also used as the inoculum for the downstream passage. In the second and third experiments, three single colonies were selected randomly at each interval for PCR and specific activity assay, and, to lessen the possibility of variations resulting from random selection, the one with the highest specific activity was the inoculum for the downstream passage. Passage numbers are denoted with "P" and the passage number.

Specific activity ("resting cell") assay
Single colonies of CW25 transformants (or strain IGTS8 as positive control) were grown in kanamycin-free CDM-0.05 mM DBT with either glucose (0.06 M) or ethanol (0.043 M) as the carbon source. These media compositions were determined to be optimal for DszABC activity through a preliminary series of experiments testing various concentrations of glucose and ethanol, two basal media, and DBT compared to DMSO as sole sulfur source (see Results and S1 Fig). Cells were grown at 30˚C, 200 rpm to mid-log phase (OD 600nm of about 0.4-0.7); this pre-culture was inoculated into 50 ml of the same medium in a 250 ml flask with initial OD 600nm of no more than 0.05, and grown at 30˚C, 200 rpm to mid-log phase (OD 600nm of about 0.4-0.7).
After harvesting by centrifugation and washing once with CDM containing no DBT (or ethanol, glucose, or kanamycin), cells were resuspended in the same CDM, with OD 600nm adjusted to 1.0. 5 ml of the cell suspension was transferred to a 25 ml flask for the desulfurization resting cell activity assay (which measures conversion of DBT to 2-hydroxybiphenyl (2-HBP) by the Gibbs assay; [24]). 25 μl of 400 mM DBT (dissolved in ethanol) was added, to make a starting DBT concentration of 2 mM. Data points were taken at 0, 1, 2 and 4 hours.
Cells from the rest of each suspension were harvested, washed with, and then resuspended in, distilled water and then transferred to a preweighed aluminum dish for drying and dry cell

Desulfurization ability of growing cells
Pre-cultures were prepared as described above for specific activity measurements, and inoculated into 50 ml of fresh kanamycin-free CDM-0.05 mM DBT containing 0.043 M ethanol in 250 ml flasks with an initial OD 600nm of no more than 0.05. Growth conditions were 30˚C, 200 rpm. When the culture reached late log phase (OD 600nm of about 0.7-1.0), the time was recorded and 1 ml was taken for Gibbs assay of the 2-HBP accumulated in the growth medium.

Cell doubling time determinations
Pre-cultures were prepared as described above, inoculated into 50 ml of fresh kanamycin-free CDM-0.05 mM DBT containing 0.043 M ethanol in 250 ml flasks, and grown at 30˚C, 200 rpm. The initial OD 600nm was 0.05. OD 600nm was read multiple times until the end of log phase. Data of OD 600nm were converted into log 10 OD 600nm values, which were plotted against incubation time (in hours), and the maximum slope (determined with at least 3 time points) used to calculate cell doubling time.
Genome assembly was completed utilizing four de novo assemblers: SPAdes 3.5 [29]; Ray 2.3.1 [30]; A5-miseq 20141120 [31]; and Geneious 7.1 [32]. For the Ray assemblies, kmer size was determined using Kmergenie 1.6663 [33]. For Geneious assemblies, paired-end reads were trimmed for quality with an Error Probability Limit of 0.001 and de novo assembled at Species names in parentheses in the heading of each section are those which supplied the sequences of the genes in that section, against which the genomes of passages P0, P4, P10, and P100 were compared. The sequences of dszA, dszB, dszC, and S1 are identical to those reported elsewhere [10,24]. E-values for the highest similarity blastp hit in these passages are displayed, except for genes annotated by Prokka, which have a minimum e-value of 1e-30. Genes without blastp matches above e-value 1e-10 are designated as "no hits". medium sensitivity using only paired reads. Contigs smaller than 1 kbp were discarded, and assemblies were compared using Quast 2.3 [34]. For each genome, the best assembly was selected: P0, SPAdes (24 contigs, 6,325,434 bp total length, 2,642,655 bp longest contig, 1,373,508 bp N 50 ); P4, Geneious (18 contigs, 6,259,093 bp total length, 1,119,966 bp longest contig, 748,535 bp N 50 ); P10, SPAdes (25 contigs, 6,402,155 bp total length, 1,640,192 bp longest contig, 696,102 bp N 50 ).
The P100 genome assembly was more completely assembled for submission to NCBI as the complete sequence of Rhodococcus qingshengii CW25 without the pRESX plasmid (BioProject PRJNA316739). Contigs from the best P100 assembly (Ray, 20 contigs, 6,417,168 bp total length, 1,640,764 bp longest contig, 704,582 bp N 50 ) were compared to the other assemblies (using Geneious), and merged in instances where at least two of three other assemblies spanned the gap. The original paired-end reads were then mapped to the merged assembly to verify even and complete coverage at each merger junction. The final P100 assembly contained 10 contigs, 6,399,510 bp total length, 2,298,790 bp longest contig and 1,403,103 bp N 50 , not including the pRESX-dszAS1BC plasmid.
Draft annotation of the four assemblies was completed using Prokka 1.11 [35] with a minimum e-value threshold of 1e-30. Final annotation of the P100 assembly of Rhodococcus qingshengii CW25 was completed using the NCBI Prokaryotic Genome Annotation Pipeline [36] (Genome Accession LVXC00000000). The original taxonomic classification of this sample was Rhodococcus erythropolis CW25; however, NCBI genome submission staff completed an average nucleotide identity (ANI) analysis and concluded this strain was likely misidentified. The strain was found to share 98.12% identity over 93% of the submitted genome sequence to the type genome of Rhodococcus qingshengii compared to a lower maximum match of 95.5% to the Rhodococcus erythropolis genomes present in the NCBI database (The GenBank Submissions Staff, personal communication, April 1, 2016).

Genome sequence analysis
The direct comparison of protein sequences in the four passages was accomplished via BLAST protein homology searches [37]. Proteins involved in various sulfur metabolism pathways in other species [38] were searched against the four passage genomes with a minimum e-value threshold of 1e-10 to identify any potential homologs in Rhodococcus qingshengii CW25 ( Table 2).
The coverage of each of the four assemblies and their respective plasmids was determined by read mapping in Geneious and the relative plasmid copy numbers approximated from the plasmid:genome coverage ratio. To detect potential mutations accumulating throughout the generations, the sequence reads from each passage were filtered using a minimum quality score of 32 and a minimum length post-trimming of 150 nt with sickle (https://github.com/ najoshi/sickle), mapped on the curated P100 assembly with BWA-SW 0.7.13-r1126 [39], converted from SAM to BAM format with SAMtools 1.3.1 [40], and single nucleotide polymorphisms (SNPs) and insertions-deletions (indels) calculated using VarScan 2.4.2 [41] (Table 3), as implemented in the Pombert lab SSRG pipeline (https://github.com/PombertLab).

Sulfur demand and optimal medium for CW25 transformants
A series of experiments were run to determine the optimal medium for growth and desulfurization activity for the CW25 transformants. Experiments run with varying concentrations of either DBT or sulfate as sole sulfur source showed maximum growth rates at 0.05 mM sulfate or DBT. Higher concentrations of DBT were inhibitory to growth (the actual inhibition perhaps because of higher concentrations of 2-HBP produced [42]), while higher sulfate concentrations neither increased nor decreased growth rates. The maximum growth rate with sulfate was about two to three times that for DBT (Fig 2).
Further experiments measuring desulfurization specific activity of the CW25 transformants identified 0.043 M ethanol as superior to 0.06 M glucose (by about 2-fold) as the carbon source in minimal CDM-DBT. DMSO was inferior to DBT as a sulfur source (by about 5-fold), and 0.1 mM DBT was only slightly superior to 0.05 mM DBT. In addition, as a base minimal medium, CDM was superior to BSM [10] (about 60% better; S1 Fig). Thus, the medium chosen for assessing DszABC specific activity was CDM-0.05 mM DBT containing 0.043 M ethanol as carbon source.
Growth and specific activity measurements P0 (the initial transformants; see above), and passages P4, P20, P40, P60 and P100 of both CW25[pRESX-dszABC] and CW25[pRESX-dszAS1BC] were obtained from the first selection experiment and used as templates in colony PCR using PkstD and S1 primers to confirm the stabilities of pRESX-dszABC and pRESX-dszAS1BC. The presence of PkstD in all passages for both strains (Fig 1A), and S1 in all passages of only CW25[pRESX-dszAS1BC] (Fig 1B) confirmed the stability of both plasmids throughout the entire experiment.
As expected from our selection for faster growth, the growth rates of both transformants increased markedly, especially during early passages (Fig 3). Following P20 the growth rate of CW25[pRESX-dszAS1BC] was greater than that of CW25[pRESX-dszABC]. Overall, from P0 through P100, the growth rates of CW25[pRESX-dszABC] and CW25[pRESX-dszAS1BC] in medium with DBT as sole sulfur source increased by 6.8-fold and 12.8-fold, respectively.
The same passages, along with IGTS8 as a positive control, were assayed for desulfurization specific activity (Fig 4). The specific activities of P0 CW25[pRESX-dszABC] and P0 CW25 [pRESX-dszAS1BC] were 2.4 units and 3.1 units, respectively, both much lower than IGTS8  (Fig 4). The specific activity assay was repeated with growth in CDM-0.05 mM DBT with 0.06 M glucose as carbon source with similar results, except that, as expected (see above), activities were generally lower than those of ethanol grown cells (S1 Fig).
Two additional selection experiments of 30 passages each were also completed (details in Table 1), and in each case plasmid stability throughout was confirmed by PCR (S2, S3 and S4   Fig 5A and 5B. Although there are some differences from the results of the first selection experiment, the overall trends are consistent with them, specifically a gradual decline in specific activity after an early peak. The second experiment also shows that the early specific activity advantage of CW25[pRESX-dszAS1BC] over CW25[pRESX-dszABC] is substantial through at least 20 passages.

Overall desulfurization ability
To examine overall desulfurization ability, that due to the combination of growth and specific activity, we examined the rate of accumulation of 2-HBP in the culture broth after growth (in CDM-0.05 mM DBT containing 0.043 M ethanol) of various passages from the first experiment for both strains. Most likely due to the increased growth rates of later passages, the results differ from those of the resting cell specific activity assay. For both transformants, there was a large increase in 2-HBP accumulated per ml of culture to a level that was stable through P100 (36-fold for CW25[pRESX-dszABC] and 33-fold for CW25[pRESX-dszAS1BC] compared to P0 in each case) (Fig 6). Comparison of these increases with the somewhat smaller increases in specific activity (Fig 4), indicates that the overall increase in DBT metabolized is, not surprisingly, a combination of both better growth and increased specific activity of the later passages. Both transformants eventually performed similarly to IGTS8, the S1 expressing strain doing slightly better than the non S1 expressing strain.

Whole genome sequencing
The genomes of CW25[pRESX-dszAS1BC] passages P0, P4 and P100 from the first experiment and passage P10 from the third experiment were assembled and genes encoding key enzymes in DBT metabolism were identified and compared for the presence of mutations (Table 2). In no case, however, were any mutations found when the starting strain (P0) was compared with P4, P10, and P100. This lack of mutations was not restricted to the Dsz genes; no modification was found in genes involved in the downstream metabolism of sulfites produced by the 4S pathway, in Dam methylases or in the sulfite exporter (TauE/SafE), which would allow the cells to discard sulfite in addition to using it. Homologues to the putative sulfite oxidoreductase identified in R. erythropolis RP4 [23], sulfite dehydrogenase, or the hemoprotein subunit (CysI) of sulfite reductase were not found in strain CW25. In no case were any of the dsz genes transferred from plasmid to genome.
The complete genomes of P0, P4, and P100 from the first selection experiment and P10 from the third selection experiment were also compared to identify all the changes in the genomes that had occurred from P0 to the later passages. The total SNP and indel quantities across all generations were inconsequential, with the largest variation still being less than 0.01 SNPs/kb, and none of the changes occurred in any of the genes we examined. Furthermore, most of the SNPs inferred appear to result from false positives, as all but 3 of the 37 SNPs between P0 and P100 (Table 3) were also found when P100 was mapped against itself. The slightly larger SNP counts inferred for P100 against itself (and P10 against P100) were likely due to an increase in false positives caused by variations in read lengths (post-trimming) and coverage (see genome coverage data in Table 3). The sequence coverage data indicate an increase in plasmid copy number (and thus dszAS1BC copy number) as adaptive selection proceeded that fairly closely parallels the increase in desulfurization specific activity ( Table 3). Details of the genes in which SNPs were found when the complete genomes of P0 and P100 from the first selection experiment were compared are shown in Table 4. None of the indels found between P0 and P100 were located in coding regions.

Discussion
The need for cleaner sources of energy requires improvements on many fronts. In the petroleum industry, tighter environmental restrictions dictate for cleaner fuels with ever decreasing amounts of sulfur compounds. While current chemical processes have their own advantages, biodesulfurization may help complement these approaches in a cost effective and environmentally friendly way. To do so, however, biodesulfurizing bacteria will require enzymatic efficiency and thermostability at the temperatures required for petroleum processing [8]. Here, we aimed to improve upon enzymatic efficiency by a combination of genetic engineering and adaptive evolution. This approach resulted in two separate yet integrated effects, an increase in the specific activity of DBT metabolism and a larger increase in growth rate when DBT was the sole sulfur source.

Desulfurization activity
Intuitively, an enzymatic reaction can be sped up in three ways: by using a faster enzyme, by using more of an enzyme, and/or by preventing the reverse reaction from occurring. Here the improvements observed to the overall biodesulfurization efficiencies of our engineered strains did not result in the use of faster enzymes, as none of the enzymes involved in this process showed any mutation after the maximum number of generations we performed. However, four hundred generations can be considered a short time scale and evolution often occurs by long periods of slow changes interrupted sporadically by quick rapid adaptive bursts, which clearly was not the case here. What our multiple passages selected for instead is the presence of greater quantities of the enzymes of the desulfurization pathway. The 3.5-7.2X relative increase between the activities of CW25[pRESX-dszAS1BC] P0, and P100 of the first selection experiment and P10 of the second selection experiment fairly closely mirrors that of the relative plasmid copy numbers of those passages (3.5-4.7 X) ( Table 3). Since none of the dsz genes were transferred to the genome and no mutations were found in any chromosomal genes involved in plasmid replication, this increase in copy number clearly originates from the plasmid itself.
After the initial increases in specific activities in the early passages, all three replicates stabilized at roughly the same levels, suggesting that four plasmids per bacterium may be the maximum sustainable for both the S1-bearing and S1-less strains under the conditions used in this study. The high specific activity levels of the sulpeptide expressing strain in early passages of the first and second experiments were actually greater than those of IGTS8. Although  transient, these increases could possibly be related to physiological changes occurring during the selection process (see also below). When planning this experiment, we hoped to use the sulpeptide as a sulfur sink that would enhance the metabolism of DBT by accumulating sulfur released by the 4S pathway. However, except for the early passages in selection experiments 1 and 2 (Figs 4 and 5A) in which the specific activity of the sulpeptide-plus transformant was substantially greater than that of the sulpeptide-minus transformant, the increases in specific activity of the two strains eventually stabilized at levels that were generally similar to each other. Possible causes for this could be that the enzymes are already performing at maximum efficiency or that the reverse reaction is so slow that the removal of sulfite produced by the 4S pathway by the sulpeptide would have a negligible effect on the net reaction.
Another possibility is that the sulfur sink produced by the sulpeptide may be too small to have a substantial effect on sulfur demand. Specifically, the 14 methionines and cysteines encoded in S1 constitute but a tiny fraction compared to the total of 43,754 that are encoded in proteins from the main chromosome, and unless the plasmid expression level surpasses that of the genome by two or more orders of magnitude, the net effect from S1 alone is likely to be negligible (although the combined effect of all the proteins encoded on the plasmid might have an impact at a much lower threshold). Attempts to quantify S1 production using SDS gel electrophoresis were unsuccessful because of the small size of S1 and, perhaps, low levels of its synthesis. In any case, from these results we can conclude that increasing the number of dsz genes in the bacterium either by using a plasmid as shown here or by integration into the main chromosome yields greater desulfurization benefits and should be a key engineering focus.

Growth rate and cumulative 2-HBP production during growth
Because we selected for the fastest growing cells in each passage, it is not surprising that the growth rates of both strains increased greatly as the number of passages increased. During the first selection experiment, about 90% of the increases occurred in the first 20 passages, suggesting that the maximum growth rates for these conditions had been approached by that time (Fig 3); these rates were similar to that of IGTS8, even though occurring in a strain with no inherent DBT metabolizing ability. The overall pattern of the increase in growth rates was similar for both strains and at least qualitatively paralleled the patterns of both 4S pathway specific activity (Fig 4) and overall DBT metabolism (Fig 6) (see also below). This is as expected since growth is dependent on sulfur obtained from DBT via the 4S pathway While the presence of S1 did not affect the eventual specific desulfurization activity, its addition did have a positive effect on the growth rate (about twice the overall improvement and 17% faster maximal growth compared to the strain lacking sulpeptide (Fig 3)). The net effect was a small advantage in overall biodesulfurization ability of the sulpeptide-bearing strain (Fig 6). In any case, the selection process itself, apart from the influence of S1, was successful in producing large and stable improvements in overall desulfurization.
The maximum growth rates of both strains with DBT as sole sulfur source (at P100 of the first selection experiment) were greater (by about two to three times) than those of the respective P0 strains grown with excess sulfate as sole sulfur source. However, what causes this higher growth rate is unclear. In the sulpeptide-containing strain, only three insertion/deletions (indels) and 37 single nucleotide polymorphisms (SNPs) were found after about 400 generations between the P0 and P100 isolates, none of which were located in sulfur-related genes or promoters. While SNPs located in the 23S ribosomal RNA gene (A3852_29860; Table 4) could in theory affect a wide spectrum of functions via its role in translation, and thus potentially explain the observed changes in growth rate (a more efficient ribosome, even if only slightly, would certainly reap tangible benefits), the inferred mutations were not genuine and resulted from erroneous read mapping. The 23S ribosomal RNA gene is present in multiple copies in the genome, and SNPs inferred when duplicate or paralogous genes are involved tend to be overestimated due to the inherent limitations of read mapping-based polymorphism inferences [43].
On the other hand, the single mutation observed in the transcriptional regulator of the GntR family (A3852_29435) appears genuine, and the non-synonymous change from a hydrophilic residue (Ser-215) to a hydrophobic one (Gly) in P100 could have implications on its overall fitness, and thus potentially affect the expression of genes regulated by this factor. Nonsynonymous changes between P0 and P100 were also observed in genes coding for cytochrome C and a non-ribosomal peptide synthetase (Table 4), but the effects that such mutations could have on the fitness of the bacterium is unclear.
Nevertheless, considering that both the sulpeptide-bearing and sulpeptide-minus strains independently achieved greater growth rates at similar thresholds, under the same conditions, we do not believe that genetic mutations are behind this adaptation. It would be very unlikely for independent strains to achieve the same level of fitness via random mutagenesis, especially over such a small evolutionary time scale. Changes in gene expression not due to mutation (as observed in long term adaptation studies of Yersinia pestis [44]) are targets for future investigation (for example by proteomic studies); such changes may occur via heritable epigenetic changes [45], changes in regulatory circuits [46], or combinations of these mechanisms. Heritable epigenetic changes, sometimes induced by environmental factors, have been documented in a number of cases, mediated by the Dam methylase or similar enzymes [45]. If such mechanisms are involved in our system, however, they cannot be due to mutations to the Dam methylase gene in strain CW25, as no changes were found in its sequence in any of the four passages examined.
It is also possible that the changes in growth and metabolism we saw in the various passages may be due to physiological adaptations [46,47]. Such adaptations could be maintained by changes in the levels of transcriptional regulators, responding to environmental conditions, such as nutrient limitation. In Gram-negative E. coli, such changes are mediated by the RpoS sigma factor [46,47], while in Gram positive C. glutamicum and M. tuberculosis the SigB sigma factor plays a similar role [48,49]. In the experiments described here, limiting the sulfur source to a sparingly soluble molecule like DBT could be a signal of nutrient limitation. Any adaptations due to changes in functions of the alternative sigma factor SigB in Rhodococcus CW25, however, also cannot be due to mutations in the sigB gene, which did not occur in any of the later passages examined.
Taken together the results suggest that further work to increase biodesulfurization will have to take a global approach, considering a combination of genetic, epigenetic, and physiological changes. In addition, a better experimental strategy might be to first select a culture for maximum growth rate using sulfate, and then use this physiologically tuned culture as the starting point for selection for maximum growth rate with DBT. In this way changes in overall metabolic fitness optimization would have less chance of overshadowing changes that are specific to DBT metabolism.

Conclusions
Engineering of desulfurization competent Rhodococcus qingshengii to express a peptide designed to act as a sulfur sink (Sulpeptide 1), combined with adaptive evolution over the course of about 400 generations was successful in substantially increasing both their metabolism of DBT and their growth rate in medium with DBT as the sole sulfur source. The increased growth rate on DBT eventually exceeded (by more than 2-fold) the growth rate of unadapted strains grown with sulfate as sole sulfur source. The effect of the expression of Sulpeptide 1 contributed less to these increases than did the adaptive evolution.
The increases in growth rate and DBT metabolism were not correlated to any mutations in genes of obvious connection to DBT metabolism or sulfur metabolism in general, although the eventual increase in DBT metabolism was closely correlated to an increase in copy number of the genes encoding DBT metabolism. It may be that the important changes that occurred during the adaptation may include epigenetic and stable physiological ones. These results suggest that future research should first optimize growth on sulfate under ideal conditions and only then should the dsz genes be introduced and adaptive evolution experiments be performed.