The Stringent Stress Response Controls Proteases and Global Regulators under Optimal Growth Conditions in Pseudomonas aeruginosa

Microorganisms need to adapt rapidly to survive harsh environmental changes. Here, we showed the broad influence of the highly studied bacterial stringent stress response under nonstressful conditions that indicate its general physiological importance and might reflect the readiness of bacteria to respond to and activate acute stress responses. Using RNA-Seq to investigate the transcriptional network of Pseudomonas aeruginosa cells revealed that >30% of all genes changed expression in a stringent response mutant under optimal growth conditions. This included genes regulated by global transcriptional regulators and novel downstream effectors. Our results help to understand the importance of this stress regulator in bacterial lifestyle under relatively unstressed conditions. As such, it draws attention to the consequences of targeting this ubiquitous bacterial signaling molecule.


Introduction 51
To deal with stress and/or harmful environmental conditions, microbes can adopt versatile 52 adaptive lifestyles. To enable such lifestyle changes to occur rapidly, bacteria have evolved 53 complex hierarchical regulatory networks to trigger diverse molecular responses that alter gene 54 expression and protein activity. Global regulatory systems enable the coordination of downstream 55 effectors that help recognize and appropriately respond to new environments. In particular 56 microbial life depends on the ability to rapidly switch from favorable conditions such as rapid 57 growth in nutrient rich media, to recognize and counteract external threats, and switch into a 58 survival mode (1). Here we wondered whether such stress adaptations might also operate under 59 optimal, rapid-growth conditions that are not usually considered "stressful". 60 As long as sufficient and appropriate nutrients are provided and toxic agents are absent, 61 bacteria continue to replicate, although in normal culture they eventually stop growing (e.g. in 62 Escherichia coli at around two billion bacteria per ml). On a cellular and molecular level, the 63 processes that they undergo during rapid growth are, however, likely quite stressful with rapid 64 replication, protein synthesis, cell division and reorganization of the cell (2). For example, there is 65 a disconnect between bacterial division every 20-40 minutes under optimal conditions and 66 replication and segregation of the chromosomal DNA, which is 1000 times the length of the cell 67 and thus highly condensed, that requires 60-90 minutes (2). Furthermore, as the density of bacteria 68 increases they start to experience depletion of one or more essential nutrients/growth requirements, 69 and/or the formation of inhibitory products such as organic acids, which eventually leads to the 70 stationary phase (3). It is known that maintenance of bacteria in stationary phase is guided by the 71 alternative stress/starvation sigma factor (σS) (4, 5), and the stringent stress response (3) which 72 indicates that cessation of growth in broth culture occurs under stressful circumstances. However, 73 it is worth asking about the mechanistic impacts of such factors during rapid, apparently 74 uninhibited growth. 75 One major mechanism for dealing with stress is through the stringent stress response 76 intermediated by the second messenger guanosine tetra-phosphate (ppGpp) (6, 7). The activation 77 of the stringent stress response during amino acid starvation is due to the accumulation of 78 uncharged, deacylated tRNA molecules in the cytosol that enter the ribosome A site and ultimately 79 cause ribosome stalling (8). In most Gram-negative bacteria, two enzymes, RelA and SpoT, 80 mediate ppGpp homeostasis. Recently, Winther et al. (9) showed that RelA binds to empty tRNA 81 serine-hydroxamate (SHX) to mimic amino acid starvation, or stationary phase-growth of P. 144 aeruginosa cultures. While a very similar distribution of RelA-YFP was observed after stress 145 induction, there was a significantly higher coefficient of variation ( Figure 1C) consistent with 146 increased fluorescence distribution as spots within the cell (as opposed to equally distributed 147 signal). This was consistent with an interpretation that RelA was bound to the ribosome but 148 dissociated to some extent during stress. 149 SpoT-GFP localized at the cell pole in P. aeruginosa 150 We utilized a SpoT-GFP fusion protein to visualize its subcellular localization and found that 151 under normal growth conditions, the fusion protein localized at the cell pole and the septal ring of 152 elongated cells ( Figure 1B). Bacterial cells are well organized factories where cellular asymmetry 153 and compartmentalization plays a vital role in many cellular processes (34). In recent years, the 154 importance of localization of certain proteins at the pole of rod-shaped bacteria has become 155 increasingly evident, with such proteins being crucial for fundamental cellular and regulatory 156 processes (including chromosome segregation, cell cycle, chemotaxis, adhesion, and motility), as 157 well as virulence (34,35). Intriguingly the manifestation of pili or flagella at a specific pole allows 158 bacteria to quickly move through the environment and associate with e.g. mucosal surfaces, which 159 is interesting given the polar localization of SpoT and the role of ppGpp in such processes (36, 160 37). The localization at the division septum was consistent with results in Caulobacter, where 161 SpoT was found to be involved in blocking the initiation of DNA replication and regulation of 162 DnaA, a conserved replication initiator protein (38), although we did not investigate this potential 163 function here. 164 Intriguingly, upon stress induction by either carbon-or iron (dipyridyl treatment) limitation, 165 the median signal intensity of the SpoT-GFP complex decreased by 5.7-fold and 4-fold, 166 respectively, and the signal became more uniformly distributed. The decreased coefficient of 167 variation correlated with irregular fluorescence observed across the cells ( Figure 1B, D). This 168 observation might relate to the ppGpp hydrolyzing activity of SpoT, which presumably maintains 169 ppGpp homeostasis; indeed, inside the cell, ppGpp molecules are broadly distributed and bind to 170 ribosomes that are found around the nucleoid and at the cell poles (30). 171 Transcriptional changes dependent on ppGpp occurred during normal growth 172 Since both RelA and SpoT were obviously present during logarithmic growth in nutrient-rich 173 conditions (SI Appendix, Fig S1)  analysis. This demonstrated the upregulation of six pathways (including biosynthetic and 192 metabolic processes, as well as the type 3 secretion system and cell surface signaling pathways) 193 and downregulation of six pathways (including cell transport, type 2 and 6 secretion systems, and 194 acetyltransferase activity) in GO ( Fig. 2A), while KEGG analysis indicated the downregulation of 195 six pathways (including chemotaxis, quorum sensing, and amino acid and fatty acid metabolism) 196 ( Fig. 2B). Upregulation, in the stringent response mutant, of the Type 3 Secretion System was 197 interesting since this system requires bacterial cells to contact with the host to directly inject its 198 substrates that include cytotoxins (40). In contrast, several genes encoding Type 2 secretion 199 systems (PA3095-3103, Xcp genes that mediate exoprotease secretion and PA0677-PA0689, Hxc 200 genes, and effector LapA) and Type 6 secretion systems (T6SS) (Figure 2A secreting effector proteins that can inhibit or kill them through a contact-delivery system (41). The 204 H2 and H3 systems are important P. aeruginosa pathogenesis mediators (42). Although GO 205 analysis indicated that the T6SS were overall downregulated in the stringent response mutant 206 (Figure 2A), the H1 system was actually upregulated. This agrees with others who showed that 207 H1-T6SS is repressed by when quorum sensing is highly active, whereas such circumstances 208 activate H2 and H3 (42). Indeed, regulation of the T6SSs might be under quorum sensing control 209 since quorum sensing pathways were downregulated in our KEGG analysis (Fig. 2B). 210 Quorum sensing is a complex network that cells use for cell-to-cell communication. The 211 quorum sensing network follows a hierarchy with LasR at the top directly influencing 212 transcriptional regulators such as qscR, vqsR, and rhlR, which regulate another set of regulators 213 such two component response systems (e.g., gacAS, pprAB), which in turn further regulate quorum 214 sensing (43, 44). The stringent response is known to activate quorum sensing in Pseudomonas 215 through increased expression of lasR and rhlR when relA was overexpressed, indicating that 216 quorum sensing can be activated independently of cell density (45), which is in accordance with 217 our data. 218 We also observed that differential expression often involved co-expressed neighbouring genes. 219 This could be visualized by projecting differentially expressed genes onto the circular chromosome 220 ( Fig. 2C), which revealed 34 clusters of genes (total 289 genes, 17% of differentially expressed 221 genes) with multiple genes with a similar direction of expression that had a broad array of 222 functions. Indeed within these clusters, we observed differential expression of genes involved in 223 various secretion systems, transporters, quorum sensing, pyocin synthesis, vitamin B12 synthesis, 224 heme export and cytochrome C synthesis, ethanol oxidation, sulfur and carbon metabolism, 225 biofilm formation, fatty acid and LPS biosynthesis, polyamine transport, adherence, ribosomes, 226 and other biosynthetic gene synthesis systems (Dataset 2). Intriguingly only genes in 8 out the 34 227 clusters were organized in operons. 228

The stringent response is required for environmental adaptations 229
Since the ΔrelAspoT stringent response double mutant grew normally during the logarithmic 230 phase, we postulated that during normal growth the stringent response might control processes that 231 prepare cells for more stressful situations including environmental adaptations. Therefore, we 232 examined the RNA-Seq data with this in mind. 233 Once bacteria have invaded host tissues, they must immediately deal with stresses imposed by 234 host responses. One element assisting the initiation of colonization of P. aeruginosa and 235 counteracting host responses is direct cytotoxicity towards host cells that processes epithelial 236 surfaces, enables adhesion, and counteracts the action of phagocytic cells (46). Stringent response 237 mutants demonstrate decreased cytotoxicity towards human bronchial epithelial cells as well as 238 minimal hemolytic activity, cf. wild type (12). This was consistent with the finding here that many 239 prominent cytolytic proteases were under the control of the stringent response under rapid growth 240 conditions, as judged by their downregulation in the double mutant, including LasA elastase (-7.3 241 fold), LasB elastase (-7.8 fold), Protease IV (piv -3.2 fold), alkaline protease AprA (-9.6 fold) and 242 PA3535 (-2.1 fold); similarly the heat stable hemolysin/rhamnolipid (synthesized by rhlABC = -243 4.5, -8.3 and -8.7 fold respectively) was substantially down regulated (Dataset 1). This is also 244 generally consistent with the role of the stringent response in adjusting to amino acid deprivation 245 since extracellular proteases could digest proteins in the environment creating an additional source 246 of needed amino acids. Nevertheless protease upregulation by ppGpp under normal growth 247 conditions could be considered preparatory to more stressful circumstances. 248 Conversely, as mentioned above there was upregulation by about two-fold of the Type 3 249 Secretion System (T3SS) machinery and regulatory genes in several adjacent operons (PA1699-250 PA1725) (Fig. 2) as well as the effector/toxin exoT. Both of these observations could be related 251 in part to stringent regulation of quorum sensing genes under rapid growth conditions since we 252 observed down regulation in the ΔrelAspoT stringent response double mutant of rhlI (-3.0 fold), 253 rhlR (-2.6 fold), lasR (-1.6 fold), rsaL (-3.0 fold), pqsH (-2.0 fold) and pqsL (-3.6 fold; cf. pqsA-E 254 that were 3.9-4.6 fold upregulated). Exo-proteases and hemolysin are upregulated during quorum 255 sensing which is in turn upregulated by ppGpp, whereas T3SS genes are negatively controlled by 256 RhlI (47). 257 The ability to respond rapidly is necessary to accommodate sudden environmental changes 258 and the failure to adapt can have negative or even lethal consequences to the organism. Consistent 259 with a role in preparing for environmental adaptation, the stringent response was shown to be 260 required for swarming motility, rhamnolipid production, adherence, and pyoverdine and 261 pyocyanin production in strain PAO1 (SI Appendix Fig. S4), in accordance with previous studies 262 (12, 37, 48). P. aeruginosa cells that encounter semi-viscous conditions with a poor nitrogen 263 source are driven to swarm rapidly across surfaces (49). Intriguingly, while the stringent response 264 double mutant was unable to swarm, we found that, in rich medium liquid broth, where even the 265 wild type fails to swarm, genes required for swarming, such as LasB (50) and is therefore a key regulator required for adhesion (52). The tad locus (PA4302-PA4306, 281 Cluster 33), the type IVb pilin flp, and the usher-and chaperone-encoding cupE1 were >2-fold, 282 7.1 fold, and 2.8 fold downregulated in the stringent response mutant (Dataset 1-2, Fig. 2C). This 283 is in accordance with results for their corresponding regulator PprB, which was 4.7 fold 284 downregulated. Induction of the stringent response led to the expression of pprB and tadG ( Table  285 1) suggesting that the stringent response regulates pilus assembly via PprB. Adaptation and 286 attachment are the first steps before bacterial biofilm formation which is also regulated by the 287 stringent response (74). Consistent with this, we found that 31% (228 out of 734) of genes that are 288 known to be required for biofilm formation in Pseudomonas (53) were either up-(92 genes) or 289 downregulated (136 genes) in the ΔrelAspoT mutant. Intriguingly, 11 biofilm regulators were 290 also identified with 4 upregulated (including efflux pump repressor mexR, T3SS assembly 291 regulator pcrH and repressor ptrB, and anti-sigma factor vreR) and 7 downregulated (including the 292 two-component sensory protein pprA, the chloramphenicol resistance activator cmrA, and 293 mycobactin siderophore uptake regulator femR). Overall these data support the proposal that the 294 stringent response regulates rapid adaptation to environmental changes. 295 The metalloprotease AprA, a novel downstream effector of the stringent stress response, was 296 required for full virulence 297 Since the stringent response has been shown to influence P. aeruginosa infection in multiple 298 models (12, 14, 37, 48, 54, 55) and strongly regulated cytolytic proteases under normal growth 299 conditions, we further investigated the importance of one of these, AprA, in a high-density murine 300 skin infection model. The alkaline protease aprA (PA1249) was 4.1 fold upregulated upon SHX 301 induction and 9.6 fold downregulated in the mutant. The deletion of the aprA gene did not affect 302 lesion sizes ( Figure 3B) or bacterial counts ( Figure 3C) in the abscess. However, after three days 303 there was a significant enhancement of survival (77%) of the aprA mutant cf. the wild-type PAO1 304 (~33% survival of mice) ( Figure 3D). Therefore, AprA acts as a novel downstream effector of the 305 stringent response and is required for full virulence under high-density infections; previous data 306 has implicated this protein in the regulation of virulence and destruction of host defence systems 307 (56). 308 Based on this result, we further examined protease-encoding genes using the combined lists 309 of peptidases/proteases from MEROPS (57) and the Pseudomonas Genome Database (58). It was 310 found that 37.6% (73 of 194 genes; p-value <0.001) of genes were significantly dysregulated in 311 the stringent response mutant ( Figure 3A). We then focused on genes that were dysregulated in 312 the double mutant, according to RNA-Seq, and showed an inverse correlation, using qRT-PCR, 313 when ppGpp production was induced by SHX or relA overexpressed (Table 1). 314 The zinc-dependent protease PA0277 (5.2 fold downregulated by SHX induction, 9.9-fold 315 upregulated in the mutant) is directly controlled by the post-transcriptional regulator RsmA (59), 316 and the expression of rsmA is directly activated by AlgR (60) and indirectly via GacA through 317 RsmY and RsmZ (61). All of these were further controlled by the stringent stress response under 318 normal conditions. The immunomodulating metalloprotease impA (PA0572) was 2.8 fold 319 upregulated by SHX and 9.1 fold downregulated in the mutant. ImpA is important during infection 320 and protects P. aeruginosa from neutrophil attack, by cleaving the P-selectin glycoprotein ligand-321 1 on neutrophils as well as targetting CD43 and CD44 involved in leukocyte homing (62). ImpA 322 contains a LasR-regulated Xcp-dependent signal sequence (63), but SHX induction did not 323 influence the expression of lasR. Therefore, impA might be under direct control of the stringent 324 response. The putative caseinolytic peptidase clpP2 (PA3326) was 2.4 fold upregulated by SHX, 325 and 10.9 fold downregulated in the stringent response mutant; it is known to be involved in 326 motility, biofilm formation, pigmentation and iron scavenging (64). We thus investigated impact 327 of these proteases in the high-density murine skin infection model. There was no effect when 328 mutants in PA0277, ImpA, or ClpP2 were tested (data not shown). 329

Stringent regulation of global transcriptional regulators 330
The large number of differentially expressed genes was apparently controlled in a hierarchical 331 process. Thus no fewer than 132 regulators were differentially expressed in the ΔrelAspoT double 332 mutant, with 79 being downregulated and 53 upregulated. More highly dysregulated and 333 prominent regulators included alternative sigma factor and regulator of the general stress response 334 and quorum sensing rpoS (-3.8-fold), quorum sensing regulators rhlIR (-3.0 fold, -2.6 fold) and 335 The potential that regulators such as these were mediating the very substantial effects of RelA 344 and SpoT during logarithmic growth phase, prompted us to further investigate global regulation in 345 the ppGpp-deficient strain. The 1,669 DE genes were used to search for transcriptional regulators 346 ( Figure 4A) that influenced their expression by filtering to retain the first individual differentially 347 expressed in an operon; this reduced the list to 1,201 unique genes. Based on the location of the 348 start of the gene, we then extracted the region 250-bp upstream of the transcriptional initiation 349 sites for these genes (65) and searched for potential transcription factor binding site motifs. Motif-350 based sequence analysis (MEME) identified four significant motifs ( Figure 4B) that were further 351 processed using the motif comparison tool Tomtom, which revealed several known global 352 regulators from CollecTF, namely: VqsM, AlgR, AmrZ, PvdS, LasR, and OxyR ( Figure 4B). In 353 the next step, the regulon of each regulator extracted from CollecTF (66) was further tested for 354 statistical enrichment in the original list of differentially expressed genes, which revealed four 355 significantly dysregulated regulons: AlgR (p < 0.02), OxyR (p < 0.02), LasR (p < 0.01) and AmrZ 356 (p < 0.01); of these only LasR was itself modestly 1.57 fold downregulated. The relevance of this 357 analysis is indicated by the fact that global regulators like these control many downstream targets 358 affected by the stringent response, as typified by LasR, a mediator of the 3-oxo-C12-359 acylhomoserinelactone mediated quorum sensing response (45). 360

Global transcriptional regulators responded to relA overexpression but not SHX induction 361
To further explore regulation by the stringent-response mediator, we overexpressed the relA 362 gene in WT as well as induced the production of ppGpp with serine hydroxamate (SHX) ( Table  363 1). Intriguingly, the expression of several transcriptional regulators including argR, lasR vqsM, 364 amrZ, algR, and to some extent oxyR were only increased when relA was overexpressed but not 365 when the stringent response was induced with SHX (Table 1) without NH4Cl, and supplementation of 6.25 µM or 12.5 µM arginine. Statistical analysis was 503 performed using a paired one-sided t-test where each time point was compared to the starting 504 timepoint to identify a significant increase in promoter activity. 505

Cloning of the inducible relA and spoT constructs 506
The relA gene was PCR-amplified from PAO1 genomic DNA with primers 507 relA_fwd(Xba)/relA_rev(Hind). The amplified product was gel-purified and cloned into 508 pHERD20T via XbaI/HindIII to allow expression from the pBAD promoter, and subsequently 509 sequenced. The spoT gene was PCR-amplified from PAO1 genomic DNA with primers 510 spoT_fwd/spoT_rev(Hind). The amplified product was gel-purified, digested with HindIII and 511 cloned into pHERD20T via SmaI/HindIII to allow expression from pBAD and was subsequently 512 sequenced. Each plasmid was transformed into P. aeruginosa PAO1 wild-type as previously 513 described (82). 514

Cloning of the RelA-YFP and SpoT-GFP translational reporter fusions 515
The relA gene was PCR-amplified from PAO1 genomic DNA with primers 516 relA_fwd/relA_rev. Both primers had an EcoRI restriction site incorporated. The reverse primer 517 lacked the RelA stop codon. The amplified product was cloned onto pBAD24.yfp via EcoRI and 518 orientation verified via restriction digest and sequencing. The resulting pBAD24.relA-linker-yfp 519 plasmid was digested with BamHI, treated with S1 nuclease, and further digested with HindIII. 520 The resulting fragment was transferred onto pHERD20T via HindIII/SmaI. 521 The spoT gene was PCR-amplified from PAO1 genomic DNA with primers 522 spoT_fwd/spoT_rev-link-gfp_fwd. The reverse primer lacked the SpoT stop codon and had an in-523 frame linker sequence (atggtgtctatcactaaagatcaaatc) fused to the forward sequence of egfp. The 524 egfp gene was amplified from pBBR1.TIR.egfp.t0 with primers gfp-fwd-link-spoT-rev/egfp-rev 525 whereby the gfp-fwd primer was the reverse complement primer of the spoT-rev primer to allow 526 the fusion of both fragments in a second PCR with spoT_fwd and gfp_rev primers. The amplified 527 product was gel-purified, digested with HindIII and cloned into pHERD20T via SmaI/HindIII 528 before sequencing. Reporter fusions were verified as described below. 529  Fig S4 and (37)). The PAO1 wild-type 537 strain as well as the ΔrelA/spoT mutant were transformed with an empty pHERD20T vector control 538 and RelA-YFP and SpoT-GFP fusions. All strains were scraped from overnight grown plates and 539 suspended in sterile demineralized water to an OD600 of 0.025. Ten µl of a bacterial cell suspension 540 was applied onto a swarming agar plate and incubated at 37°C for 18 h. Experiments were repeated 541 at least three times. The motility complementation further verified an intact fusion protein (SI 542 Appendix, Fig S6). 543

Growth and confocal microscopy of protein-fusions 544
Cultures harboring the RelA-YFP construct were grown in BM2 (at 250 rpm) to an OD600 of 545 0.3 prior to induction with 5% L-arabinose. After 30 minutes of induction, cultures were split and 546 one received with 1 mM SHX for another 30 minutes. Stationary phase cultures at an OD of 547 2.5±0.3 were induced with 5% L-arabinose for 30 minutes. Cultures harboring SpoT-GFP were 548 grown to an OD600 of 0.3 prior induction with 5% L-arabinose for 30 min. Cultures were washed 549 one time with PBS (8000 rpm, 5 min) before resuspending in BM2 with or without carbon source 550 for 30 min. Iron starvation was induced with 1 mM dipyridyl for 30 minutes. 551 To visualize the fluorescent protein localization in cells, the fluorescent proteins were cross-552 linked in order to achieve minimal cell movement and preserve the physiological state. Cells were 553 cross-linked with 3.7% formaldehyde at room temperature for 1 hour. Cells were washed twice 554 with PBS by centrifugation (8000 rpm, 5 min) and the subsequent pellet resuspended in 1 ml of 555 PBS. Cells were visualized on a Zeiss microscope. 556 of British Columbia's Sequencing and Bioinformatics Consortium (generating single end reads 578 1×100 bp). The read quality and alignment of sequencing samples was carried out as previously 579 described (85). Briefly, FastQC v0.11.6 and MultiQC v1.6 were used for quality, STAR v2.6 for 580 the alignment of transcriptomic reads to the PAO1 reference genome (obtained from the 581 Pseudomonas Genome Database (58)) and read counts generated using HTSeq v0.11.2. Library 582 sizes had a minimum of 1.2 million, median of 3.1 million, and maximum of 7.6 million uniquely 583 mapped reads. Differentially expressed (DE) genes between the double mutant and wild type were 584 determined using DESeq2 v1.24.0 with thresholds of adjusted p-value ≤ 0.05 and absolute fold 585 change ≥ 1.5. 586

Visualization of RNA-Seq data 587
Relative expression of DE genes were plotted using the circular visualization software package 588 CIRCOS (86). Since we found stretches of genes that were dysregulated in the same direction, we 589 used an arbitrary cut-off of five consecutive genes to extract gene information for these regions. 590 The rationale for using at least five genes in a row was that four genes almost revealed 100 clusters 591 and six in a row also showed 34 clusters. 592

Regulatory elements and regulator enrichment 593
Operon information for P. aeruginosa PAO1 was downloaded from DOOR2 (87) and gene 594 annotations obtained from the Pseudomonas Genome Database (58). The following methodology 595 was then applied to all 1669 DE genes identified in the RNA-Seq experiment: The transcriptional 596 start site (TSS) of each gene was obtained from Gill et al. (65). For any TSS type denoted as 597 antisense, the strand for that gene was switched. In case that there was no TSS available, the start 598 codon as listed in the genome annotations was used. Then the list of 1669 genes filtered to include 599 the first gene from a given operon (or single genes for those not in an operon), yielding 1201 genes. 600 Once the starting location had been determined, the R package BSGenome v1.48.0 (88) was used 601 to extract the 250-bp upstream region for each of the 1201 genes. These sequences were then 602 submitted to MEME (89) for identification of novel motifs with a significance threshold of E-603 value ≤ 0.05. The program was set to find up to five motifs while all other settings were left at 604 their default. These five de novo motifs found by MEME were then submitted to Tomtom (90) to 605 identify potential matches to characterized motifs and their corresponding regulators within the 606 CollecTF database (66). Matches identified by Tomtom were considered significant with q-value 607 ≤ 0.5. 608 The list of all PAO1 regulators and their controlled genes (regulon) was downloaded from 609