A novel plant E3 ligase stabilizes Escherichia coli heat shock factor σ32

The heat shock response is crucial for organisms against heat-damaged proteins and maintaining homeostasis at a high temperature. Heterologous expression of eukaryotic molecular chaperones protects Escherichia coli from heat stress. Here we report that expression of the plant E3 ligase BnTR1 significantly increases the thermotolerance of E. coli. Different from eukaryotic chaperones, BnTR1 expression induces the accumulation of heat shock factor σ32 and heat shock proteins. The active site of BnTR1 in E. coli is the zinc fingers of the RING domain, which interacts with DnaK resulting in stabilizing σ32. Our findings indicate the expression of BnTR1 confers thermoprotective effects on E. coli cells, and it may provide useful clues to engineer thermophilic bacterial strains.

experiments revealed that BnTR1 expression induced the accumulation of heat shock factor σ 32 . However, unlike molecular chaperones such as sHSPs, the RING domain of BnTR1 was the active site for its function in E. coli. We found that two zinc fingers in the RING domain were able to interact with DnaK and σ 32 , respectively, resulting in σ 32 stabilization. Together, our findings reveal that heterologous expression of BnTR1 provides thermoprotective effects on E. coli cells, and it may yield useful insights into the development of engineered thermophilic bacteria.

Heterologous expression of BnTR1 enhances Escherichia coli thermotolerance and up-regulates HSPs.
Our previous study demonstrated that BnTR1 plays a key role in conferring thermal resistance among multiple plant species 28 . Surprisingly, we observed a similar trend when we expressed BnTR1 in E. coli. There was little change in growth rates between pET and pET-BnTR1 cells at the normal temperature (Fig. 1a), while transformed cells expressing BnTR1 showed superior growth over cells expressing the empty vector alone upon temperature up-shift. After 10 hours of heat stress, pET-BnTR1 cell growth was significantly greater than the total pET cell growth (Fig. 1a). Noticeably, after 1 hour of exposure at 48.8 °C, 67% of pET-BnTR1 cells survived, while only 22% of cells with the empty vector survived (Fig. 1b). Hence, heterologous expression of BnTR1 provided E. coli cells with tolerance against heat stress without affecting growth under normal culture conditions.
To further assess the impact of BnTR1 expression, we performed microarray analyses to explore the transcriptional changes of E. coli cells when cultured at 37 °C or 42 °C. Principal component analysis (PCA) was first applied to determine the distance between the transcriptomes (Fig. 1c). The first principal component (PC1) holding the largest variance (64%) distinctly clustered pET-BnTR1 cells and pET cells into two groups. We also noted that the second principal component (PC2) contributed 11% variance and slightly separated the samples by culture temperatures. These data demonstrated that changes to the transcriptome were primarily due to BnTR1 expression.
Next, to achieve a robust list of differentially expressed genes (DEGs), we used five independent statistical methods with stringent thresholds (Supplementary Fig. S1a). In consequence, we found that BnTR1 altered the expression levels of 112 and 122 genes at 37 °C and 42 °C, respectively (Supplementary Tables S1 and S2). Intriguingly, nearly half (44 up-regulated and 17 down-regulated) of all DEGs were detected under both normal and heat stress conditions ( Supplementary Fig. S1b), suggesting that BnTR1 expression induced conserved transcriptional changes at different temperatures. In particular, many bacterial HSPs were significantly up-regulated upon BnTR1 expression (Fig. 1d,e). Specifically, expression of the DnaK/DnaJ and GroEL/GroES chaperone teams, which function to re-fold and stabilize denatured proteins 29 , increased 16-and 24-fold in pET-BnTR1 cells compared with cells expressing the empty vector (Supplementary Table S3). Furthermore, the levels of proteases, such as HslU/HslV, which function in protein degradation, were approximately 14-fold higher in pET-BnTR1 cells (Fig. 1d). To explore the physiological functions of DEGs, we performed gene ontology (GO) analysis. Consistent with the increase in HSPs, the most significantly changed GO terms were closely related to "response to heat" and "protein folding" processes ( Fig. 1f). To determine whether the transcriptome changes were due to the stress of over-expression proteins, we set a control group with E. coli cells expressing PUB18, which is a U-box E3 ligase from Arabidopsis thaliana 30 . We did not observer significant changes of the HSP gene dnaK in PUB18 expressing E. coli cells ( Supplementary Fig. S2). Taken together, these data suggest that the heterologous expression of BnTR1 in E. coli specifically up-regulated bacterial HSPs. BnTR1 expression induces σ 32 accumulation. Because the σ 32 is the central player in regulating HSP transcription 8 , we next investigated the changes of the σ 32 level in E. coli cells expressing BnTR1. Interestingly, more than half of the common up-regulated DEGs, together with HSP genes, were directly regulated by σ 32 (Fig. 2a and Supplementary Table S3). Three σ 32 regulons, dnaK, groEL, and ipbA, were selected and validated their up-regulation using quantitative RT-PCR (qRT-PCR). Notably, rpoH (the gene encoding σ 32 ) remained at its basal transcriptional level, which was confirmed by both microarray and qRT-PCR (Fig. 2a,b). These data indicate that BnTR1 may not participate in the transcriptional regulation of σ 32 .
Measuring the protein levels further supported the assumption of post-transcriptional regulation of σ 32 . The BnTR1 expression rapidly trigged σ 32 accumulation within 30 min, and it persisted for at least 90 min (Fig. 2c). In accordance with the increased transcription, DnaK synthesis was concomitantly increased and reached its peak level after 1 hour (Fig. 2c), indicating that cellular σ 32 was in an active state. Because abnormal protein production in E. coli can also increase HSPs 31, 32 , we examined whether the increase in σ 32 was due to the aggregation of unfolded BnTR1. To minimize the basal BnTR1 protein level and to avoid its toxic effect on E. coli cells, BnTR1 with a tightly regulated pBAD promoter was generated and induced by L-arabinose under a low temperature in E. coli W3110 and C600 strains. Remarkably, σ 32 levels still increased dramatically even when BnTR1 was slightly induced at 30 °C (Fig. 2d). We also tested dosage effects of BnTR1 on the σ 32 accumulation by using different concentrations of inducer. When supplemented with L-arabinose, σ 32 levels increased in E. coli expressing BnTR1 compared with the control groups harbouring empty vectors. Moreover, when BnTR1 was induced, both the σ 32 level and the amount of BnTR1 were higher in E. coli cells treated with 1% L-arabinose than with 0.01% and 0.001% L-arabinose, respectively (Fig. 2e). In addition, cell lysate tests demonstrated that the majority of BnTR1 was concentrated in the supernatant and was barely detectable in the inclusion bodies (Fig. 2f).
The RING domain is the active site of BnTR1. As an E3 ligase, BnTR1 possesses a typical RING domain chelating two zinc atoms to form two zinc fingers 28 , which led us to explore whether the RING domain is essential for BnTR1 function in bacteria. The BnTR1 mutants, BnTR1ΔZn1 (BnTR1 C66S/C69S ), BnTR1ΔZn2 (BnTR1 C82S/ C84S ) and BnTR1ΔZn1/2 (BnTR1 C66S/C69S/C82S/C84S ), had no influence on cell growth at the normal temperature (Fig. 3a). As expected, wild-type BnTR1 kept the in vitro E3 ubiquitin ligase activity, and E. coli cells benefited from the expression of wild-type BnTR1 at 42 °C ( Fig. 3a and Supplementary Fig. S3). Interestingly, though BnTR1 mutants lost the ligase activity, they had different effects on E. coli cells. The growth rate of BnTR1ΔZn1 cells was similar but slightly lower than that of BnTR1 cells (Fig. 3a). In sharp contrast, the growth of BnTR1ΔZn2 and BnTR1ΔZn1/2 cells significantly decreased (Fig. 3a). The cell growth results indicated that mutations in the first zinc finger (Zn1) did not substantially affect BnTR1 activity, but the second zinc finger (Zn2) was more crucial.
To identify the distinct functions of the two zinc fingers, the levels of HSPs and σ 32 were measured. HSPs in BnTR1ΔZn1/2 and BnTR1ΔZn2 cells dropped to basal levels; however, mutation of Zn1 had little effect on HSPs as the HSP levels were similar in BnTR1ΔZn1 and wild-type BnTR1 cells. We observed a positive correlation between σ 32 levels and HSP synthesis. The expression of wild-type BnTR1 and BnTR1ΔZn1 induced the accumulation of σ 32 over 10 to 30 min, whereas no significant changes of σ 32 levels were detected in BnTR1ΔZn1/2 and BnTR1ΔZn2 cells (Fig. 3b). It is important to note that the cells expressing mutant BnTR1 were cultured under non-stress conditions (30 °C). These results were in strong agreement with the growth rate experiments, suggesting that zinc fingers play an important role in the function of BnTR1; in particular, Zn2 was indispensable for the up-regulation of HSPs and σ 32 .
To further understand BnTR1 activity in E. coli, we next explored the distribution of its homologues. By mapping BnTR1 homologues to the reconstructed phylogenetic tree, we found that BnTR1 homologues emerged in ferns (Selaginella moellendorffii) but are missing in mosses (Physcomitrella patens), green algae, and red algae. Interestingly, unlike full-length BnTR1, the RING domain was widely spread among the Plantae and was even detected in yeast (Saccharomyces cerevisiae) ( Fig. 3c and Supplementary Fig. S4a). Considering the importance of the RING domain, we next conducted a domain search in prokaryotic genomes. In E. coli, DnaJ chaperone contained a zinc-finger domain next to the J-domain 33 . Moreover, other types of zinc fingers, such as the B-box and AN1, were integrated with the J-domain in specific bacterial and archaeal species 34 (Fig. 3d), raising the possibility that BnTR1 and DnaJ are analogous in terms of their structural topology or enzymatic activity. The comparison of the structures of BnTR1 and DnaJ by molecular modelling indicated that the eight residues (cysteine/ histidine) of BnTR1 coordinated the zinc ions in a cross-braced shape ( Supplementary Fig. S4b). By contrast, the two zinc fingers in DnaJ constituted a right angle and formed a V-shaped molecule 33 (Supplementary Fig. S4c).
The dissimilar topology led us to extend the comparison of zinc fingers in vivo. A complementary experiment using the dnaJ mutant E. coli strain MF634 revealed that cells expressing wild-type BnTR1 formed very few clones at 43 °C, indicating that BnTR1 alone was unable to compensate for the loss of DnaJ (Supplementary Fig. S5). Because the J-domain and glycine/phenylalanine-rich (G/F) region are crucial for DnaJ chaperone functions 35 , a chimeric protein made by concatenating BnTR1 with the J-domain and G/F region of DnaJ was engineered and named JdBnTR1 (Fig. 3e). Surprisingly, JdBnTR1 rescued the defective cells at the high temperature, but mutations in both zinc fingers (JdBnTR1ΔZn1/2) were incapable of rescuing the growth defect at 43 °C (Fig. 3e). Taken together, these results suggest that the up-regulation of HSPs and σ 32 was not a result of over-expression of foreign proteins or possible BnTR1 solubility problems, but rather the physiological function of BnTR1 in the post-transcriptional regulation of σ 32 .
BnTR1 stabilizes σ 32 in vivo through its RING domain. Because σ 32 activity is negatively controlled by the KJE chaperone team 15,36 , it is unknown whether BnTR1 could hinder KJE-dependent regulation of σ 32 . The assay of KJE refolding of denatured luciferase substrate was employed to test the function of two zinc fingers of BnTR1. BnTR1 and its mutant forms sharply decreased the reactivation activity of the KJE chaperon team in 10 minutes (Fig. 4a). However, along with increasing the reaction time, BnTR1 mutants displayed different effects. After 20 min, BnTR1ΔZn2 and BnTR1ΔZn1/2 showed distinguishably higher levels of reactivated luciferase than the wild-type BnTR1 and BnTR1ΔZn1. To confirm our observations, series of BnTR1 mutant concentrations were supplied into the reaction. We confirmed the inhibitory effects, which were even more significant when we used a low concentration (0.0125 μm) of BnTR1 mutant (Fig. 4b). Based on these results, BnTR1 could act as an intruder in the KJE refolding system in vitro and confirmed the distinct functions of two zinc fingers.
As the Zn2 of BnTR1 is more important for its inhibitory effect, we hypothesized that this region physically interacts with the DnaK chaperone. Therefore, 6His-BnTR1 (N terminus) and BnTR1 mutants were purified and used as prey in co-immunoprecipitation (Co-IP) experiments. Indeed, DnaK and BnTR1 formed a stable complex, which remained intact when Zn1 was mutated (BnTR1ΔZn1) (Fig. 4c). As expected, the DnaK-BnTR1ΔZn2 or DnaK-BnTR1ΔZn1/2 complex was barely detectable, suggesting that DnaK interacted with the Zn2 of BnTR1 in vivo. We next used a bacterial two-hybrid system to confirm their interaction. We identified a clear signal for the DnaK-BnTR1 and DnaK-BnTR1ΔZn1 interactions (Fig. 4d).
It is known that σ 32 is a substrate of DnaK and the co-chaperone DnaJ 36 . We observed different growth rates (Fig. 3a) and inhibition effects (Fig. 4a) between BnTR1ΔZn1 and BnTR1ΔZn1/2, suggesting that Zn1 may play a minor role in the stabilization of σ 32 . Surprisingly, BnTR1 and σ 32 formed a rather tight complex, and mutation of Zn2 had no effect on their interaction (Fig. 4e). The interaction between BnTR1 and σ 32 was dependent on Zn1, as mutation in this region completely disrupted the interaction, which was demonstrated by Co-IP and bacterial two-hybrid experiments (Fig. 4e,f).

Discussion
In this study, we report heterologous expression BnTR1, a plant E3 ligase containing a RING domain, could effectively protect E. coli cells from heat stress. BnTR1 expression in E. coli dramatically increased the level of heat shock factor σ 32 , even at low temperatures, resulting in the significant up-regulation of HSPs. It has been well established that HSP expression rapidly increases following temperature upshift to protect E. coli cells from heat-damaged proteins 1, 3, 4 . In our study, several HSPs were induced by BnTR1 expression, including DnaK, GroEL, ClpB and HtpG, which are molecular chaperones with functions in deterring unfolded protein aggregation and assisting in their refolding 29,37 . Other HSPs, such as HslU/hslV and ClpP, are proteases that function to degrade and dissolve heat-denatured proteins 38 . Thus, the increased heat resistance largely depends on the cumulative effects of multiple molecular chaperones and proteases, which are essential to refold and degrade heat-damaged proteins. Therefore, cells expressing BnTR1 seem less affected and damaged by heat stress. Consistent with this idea is a recent report that over-expression of the GroEL/GroES chaperones increase the maximum growth temperature of wild-type E. coli to 47.5 °C 39 . Heat shock factor σ 32 is known to be the key regulator of E. coli HSPs 5 . Although BnTR1 triggered considerable activation of σ 32 , which is sufficient to induce HSPs, σ 32 transcription remained relatively constant. Thus, the apparent linkage between BnTR1 and σ 32 is at the post-transcriptional level.
It could be argued that the enhanced HSP levels are not due to a specific BnTR1 function, as previous studies have shown that overexpression of abnormal proteins increases HSP levels 31,32 . The simplest explanation for the induction of HSPs is the accumulation of unfolded BnTR1. However, this simple model seems inconsistent with our current data. These reported abnormal proteins are misfolded or unfolded with aberrant high-order structure, and are found in inclusion fractions 31,32 , while the majority of BnTR1 protein was soluble. In addition, BnTR1 in E. coli remained active rather than being unfolded. Thus, the thermotolerance of E. coli was not due to the stress of the over-expression of exogenous proteins.
The heterologous expression of eukaryotic molecular chaperones enables E. coli cells to resist heat stress at lethal temperatures [21][22][23][24]27 . This protective effect is closely related to chaperone activity in maintaining proteins in a folded state or preventing unfolded proteins from aggregation [25][26][27] . However, this is not the case for BnTR1, as purified BnTR1 exhibited no chaperone function in refolding denatured luciferases with KJE in vitro; instead, the efficiency of BnTR1 largely depended on its RING domain. Thus, we propose an alternative model. The KJE chaperone team inhibits the σ 32 activity by directly binding to σ 32 in vivo and in vitro [14][15][16] . BnTR1 interacts with DnaK to form a stable complex mainly through the Zn2, which may inhibit the negative effect of KJE on σ 32 , resulting in the accumulation of active σ 32 . Moreover, although mutation in Zn1 had little effect on σ 32 and HSP levels, our in vivo interaction studies suggest that Zn1 selectively interacts with σ 32 . It is not clear whether Zn1 and/or Zn2 could influence pathways associated with SRP and the protease FtsH, key regulators for σ 32 membrane localization and degradation [17][18][19][20] . Thus, we suggest that the increased amount of active σ 32 is largely attributed to the Zn2 of BnTR1 (Fig. 5).
BnTR1 is identified as a key player in the heat stress response of B. napus, and constitutive BnTR1 expression significantly increases the heat tolerance of multiple plant species 28 . Interestingly, we observed that heterologous expression of BnTR1 also conferred pronounced thermoprotective effects on E. coli cells. In plant the function of BnTR1 tightly relies on its E3 ligase activity, but BnTR1 may work in a different way in E. coli. All BnTR1 mutants used in this study lost their in vitro E3 ligase activity, whereas the effect of BnTR1ΔZn1 in E. coli was slightly affected. Although no significant homologues of BnTR1 have been detected in prokaryotes, we noted that the E. coli chaperone DnaJ possesses a domain containing two zinc fingers. Importantly, analogous to BnTR1, the two DnaJ zinc fingers play different roles such that one is important for DnaK-independent chaperone activity, while the other is essential for the interaction with DnaK 40, 41 . A rather surprising finding was that the expression of the JdBnTR1 chimeric protein (a fusion of the J-domain and BnTR1) rescued the dnaJ mutant strains at a high temperature. Thus, it suggested that BnTR1 and other proteins possessing similar zinc figure domains may be useful resources in the genetic engineering of thermophilic bacteria for industrial application.

Methods
Further details of bacterial viability assays, two-hybrid assays, recombinant protein expression and purification, in vitro assay of E3 ubiquitin ligase activity, Co-IP assays, and RNA extraction and qRT-PCR can be found in the Supplementary Methods. Strains, plasmids and growth condition. The E. coli strains and plasmids were commercially obtained from the American Type Culture Collection (ATCC) and the Coli Genetic Stock Center (CGSC) (Supplementary Table S4). The strains and their transformants were grown aerobically in Luria-Bertani (LB) medium (tryptone 10 g L −1 , NaCl 10 g L −1 and yeast extract 5 g L −1 , pH 7.4) and were supplemented with ampicillin (100 μg mL −1 ) and kanamycin (50 μg mL −1 ) when necessary. To validate the in vitro interaction of DnaK and σ 32 with BnTR1, Figure 5. Proposed model of σ 32 accumulation in E. coli cells expressing BnTR1. In wild-type E. coli cells, the KJE chaperone team interacts with σ 32 to inhibit its activity, and then facilities σ 32 degradation by the protease FtsH. In E. coli cells expressing BnTR1, the Zn2 in its RING domain of BnTR1 may interact with DnaK, thereby stabilizing σ 32 . the expression of recombinant proteins from E. coli Rosetta (DE3) was induced at the mid-logarithmic phase (OD 600nm = 0.6) using 0.5 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) for 30 min.
The 6His-dnaJ was amplified by using E. coli C600 as the template and ligated into pET-28a (digested by BamHI and Xhol). pBAD24-dnaJ was then constructed by subcloning of the Ncol/Xhol dnaJ fragment from pET-28a-dnaJ. For the construction of pBAD24-JdBnTR1, the J domain of dnaJ (312 bp) was amplified from pET-28a-dnaJ and then ligated into pBAD24-BnTR1 (digested by NdeI and BamHI). The gene encoding for σ 32 (rpoH) was obtained by amplifying its entire coding region (855 bp) from E. coli C600 and then cloned into pET-28a (digested by BamHI and XhoI) to get the pET-28a-rpoH. To create pET-28a-luc, the luciferase gene was amplified and ligated into pET-28a (digest by BamHI and XhoI). Successful transformants were analyzed by colony PCR and constructs containing correct inserts were sequenced to ensure the accuracy.
Luciferase refolding assays. Luciferase refolding was evaluated as described previously with modifi- Immunoblotting. Western blotting was employed to determine the translational level of σ 32 and heat shock proteins. Briefly, E. coli C600 and W3110 pBAD24 transformants (OD 600nm = 0.6) were induced with 0.1% L-arabinose for 30 minutes. Cultures were harvested and divided into two equal aliquots. One aliquot was immediately precipitated with 5% trichloracetic acid (TCA) for protein concentration quantification. The precipitate was collected by centrifugation and washed with 80% iced acetone. Pellets were dried under vacuum and re-suspended with double distilled water. The concentration of proteins was quantified using pierce BCA protein assay kit (Thermo). The other aliquot was collected and re-suspended in SDS-PAGE sample buffer for western blotting analysis.
Significant homologs of full-length BnTR1 was detected by BLASTP (E-value < 10 −17 ). The Kyoto Encyclopedia of Genes and Genomes (KEGG) database 46 was used to construct the distribution of J-domain, RING, AN1, and B-box protein domains (E-value < 10 −4 ).

Microarray procedures.
Cy3 fluorescently-labeled cRNA was prepared and hybridized to Agilent E. coli whole-genome gene expression microarrays (8*15 K) according to the single channel microarray-based protocol (Agilent Technologies Inc.). The biological repeats were randomly distributed onto two microarray slides. The array images were scanned by the G2565BA Microarray Scanner System, and raw data were then normalized (quantile method), merged and filtered by Feature Extraction Software (Agilent Technologies Inc.).
The control probes and probes without annotation were dropped out as they were considered to have non-transcriptome biological meanings. As the Agilent E. coli microarray contains three other types of probes (designed for E. coli O157:H7, CFT073 and EDL933), these probes were neglected before the downstream analysis. For the "sibling probe-set", one gene corresponding to multiple probes, the average value to represent the gene expression intensity 47 . A total of 4108 genes were finally selected, the intensity values were log2 transformed.
To identify robust DEGs, five parametric and non-parametric methods were independently applied. First, three traditional processing methods, the Student's t-test, the Mann-Whitney U-test, and fold-change calculation were applied. The DEGs in each comparison were identified with the false discovery rate (FDR) <0.05 for the t-test and U-test, together with fold-change cut-off values of 2.0-fold decrease and increase. Two more sophisticated methods in R/Bioconductor packages "Limma" 48 and "RankProd" 49 were used (Supplementary Table S6). The intersection of DEGs identified using the five methods were considered to be significantly upor down-regulated in BnTR1 strains at 37 °C and 42 °C, respectively. The σ 32 regulons were summarized from Nonaka et al. 50 .
The latest GO information (submission date: 7/1/2015) of E. coli was retrieved from Gene Ontology Consortium. The minimum number of genes in each GO term was set as 5. The Fisher's exact test was used to determine the significant GO terms with threshold of p-value < 0.05.
For microarray data validation, the transcriptional levels of genes were detected (Supplementary Table S7). RNA was extracted as described above, cDNA was generated using an iScript cDNA synthesis kit (Bio-Rad) and qPCR was performed using SYBR Green Supermix (Bio-Rad). The results were analysed, and the mRNA levels were normalized against that of 16 s rRNA using the ΔΔCT method 51, 52 .
Statistical analysis. All statistical analysis was conducted by the R software.
Data availability. The microarray data has been deposited in the Gene Expression Omnibus with the accession number GSE85807.