Strains used in this study were derived from the Synechocystis GT-W substrain, ΔpilD and ΔpilA1/ΔpilA2 (ΔpilA1/2) mutant strains were described in Linhartová et al. (2014) and the sigF^– mutant in Barker et al. (2006). Inactivation of the pilA2 gene by insertion of a Spectinomycin resistance gene has been performed by Bhaya et al. (2000). For this work, the identical mutation was fully segregated in the GT-W substrain (pilA2^– strain; for this work “–”superscript is used for insertion mutants). The ΔpilD/pilA2^– mutant was described in Linhartová et al. (2014). Photoautotrophic suppressor mutants of the ΔpilD strain were generated by cultivation on a BG-11 plate with 5 mM glucose for 2 weeks at normal light intensity (40 μmol of photons m^–2 s^–1) at 28°C. For mixotrophic conditions (Glc⁺), the BG-11 medium was supplemented with 5 mM glucose. Liquid cultures were grown in Erlenmeyer flasks under normal light at 28°C on a rotary shaker (120 rpm). Mixotrophic cultures were diluted daily with fresh Glc⁺ medium and collected at OD_{730 nm} of ∼ 0.4. To study the effects of photoautotrophic conditions (Glc^–), the cells grown in Glc⁺ conditions were harvested at 6000 rpm, washed with BG-11 medium, diluted into BG-11 medium without glucose and then incubated at normal light for 48 h. To determine optical density of aggregated strains, cell aggregates were disintegrated using an IKA TP 18/10 Ultra-Turrax homogenizer (IKA Werke, Germany).

DNA for genomic sequencing was isolated as described by Ermakova-Gerdes and Vermaas (1999), with slight modifications. Briefly, the cells were washed by saturated NaI solution and lysed by lysozyme and SDS. The lysate was treated with proteinase K, extracted with phenol, phenol-chloroform (1:1) and chloroform and treated with RNase. The DNA was precipitated with sodium acetate and ethanol, air-dried and resuspended in water. Synechocystis re-sequencing was performed commercially at the Gene Profiling Facility, Princess Margaret Hospital, Toronto. 100 bp paired-end sequencing was performed on an Illumina HiSeq 2000 platform. The average sequencing depth was around 70. Reads were mapped to the Synechocystis GT-S sequence (NC_017277) and the variants were called using default settings of the Geneious 11.0 software.¹ Only variants with a higher than 60% frequency were considered.

Total RNA was isolated from cells (40 mL cultures at OD_{750 nm} ∼ 0.5) using a hot phenol method according to Tyystjärvi et al. (2001), except that cell pellets were resuspended in 0.3 M sucrose and 10 mM sodium acetate (pH 4.5). Samples containing 3.75 μg total RNA were denatured with glyoxal and separated on a 1.2% agarose gel in phosphate buffer and transferred to Hybond-N membrane (Tyystjärvi et al., 2001). The pilA1 and 16S rRNA genes were amplified from genomic DNA using pilA1-f/pilA1-r and rrn-f/rrn-r primers, respectively (Supplementary Table 1). DNA fragments were labeled with α-dCTP 10 mCi/mL (Perkin Elmer) using the Prime-a-Gene Labeling System (Promega, United States) according to the manufacturer’s instructions. The probes were purified using Illustra ProbeQuant G-50 Micro Columns (GE Healthcare). Membranes were prehybridized in 6 × SSC, 1 × Denhardt’s, 0.1% SDS, 100 μL/mL herring sperm DNA at 67°C for 1 h. A denatured pilA probe was added and hybridized at 67°C overnight. Membranes were rinsed twice with 3 × SSC, 0.1% SDS and then washed twice for 10 min at 60°C in 3 × SSC, 0.1% SDS and autoradiographed. The pilA probe was removed by washing the membrane three times with 0.5% SDS at 95°C for 10 min and then reprobed with the 16S rRNA probe.

About 100 mL of cells at an OD_{730 nm} of ∼0.4 were harvested by centrifugation at 6,000 × g for 10 min at 4°C and pelleted cells were first washed with and then resuspended in buffer A (25 mM MES/NaOH, pH 6.5, 10 mM MgCl₂, 10 mM CaCl₂, and 25% glycerol). Cells were mixed with 100–200 μm diameter glass beads in 1:1 ratio (1 volume of dense cell solution with 1 volume of glass beads) and broken (6 × 40 s) using a MiniBeadbeater 16 (Biospec, United States). To separate soluble and membrane fractions, samples were centrifuged at 30,000 × g for 15 min at 4°C. Pelleted membranes were washed once with an excess of buffer A and then resuspended in 250 μL of buffer A. Chlorophyll (Chl) concentration in the isolated membranes was measured spectrophotometrically according to Porra et al. (1989).

To detect the amount of PilA1 protein by immunoblotting, membrane proteins containing 1 μg of Chl were solubilized with 2% SDS and 1% dithiothreitol for 30 min at room temperature and separated on a 12–20% or 16–20% linear gradient SDS-PAGE gel containing 7 M urea (Komenda et al., 2019). The proteins were transferred to a PVDF membrane and the PilA1 protein was detected with a specific primary antibody raised against amino acids 147-160 of the PilA1 (prepared commercially by GenScript, United States). The antibody raised against the recombinant fragment R117-S384 of Synechocystis YidC was kindly provided by Prof. Jörg Nickelsen (Ludwig-Maximilians-University, Munich, Germany). The antibody against Synechocystis Ycf48 was kindly provided by Professor Peter Nixon (Imperial College, United Kingdom). The secondary antibody was conjugated with horseradish peroxidase (Sigma, Germany) and the signal was detected using Immobilon Crescendo substrate (Millipore, United States). Membrane proteins (each sample contained 1 or 4 μg of Chl if not indicated otherwise) were solubilized with 2% SDS and 1% dithiothreitol for 30 min at room temperature and separated on a 16–20% linear gradient SDS-PAGE gel containing 7 M urea.

Analysis of membrane proteins (contained 4 μg of Chl) under native conditions was performed by clear-native (CN) PAGE as described by Komenda et al. (2019). Individual components of protein complexes were resolved by incubating the gel stripe from the first dimension in 2% (w/v) SDS and 1% (w/v) dithiothreitol for 30 min at room temperature, and proteins were separated along the second dimension by a gradient SDS-PAGE described above. Proteins were in-gel stained by SYPRO Orange (Sigma, Germany).

Co-immunoprecipitation was performed essentially as described in Linhartová et al. (2014). Briefly, the anti-SecY antibody was incubated overnight with membrane proteins isolated from the WT, ΔpilD mutant and rev2 and rev3 strains grown with glucose, and then the antibody was immobilized on Protein A—Sepharose (Sigma, Germany). The resin was extensively washed and the remaining proteins were eluted in 1% SDS/dithiothreitol and 0.05% Bromophenol Blue in 25 mM Tris/sucrose buffer and separated by 12–20% linear gradient SDS-PAGE. The gel was blotted and probed with selected antibodies. The primary antibody against the Synechocystis SecY and SecE proteins were raised in rabbits using the synthetic peptide fragments containing amino acids 4-14 and 1-19, respectively.

For protein labeling, cell cultures containing 75 μg of Chl mL^–1 were incubated with a mixture of [³⁵S]-Met and [³⁵S]-Cys (Hartmann Analytics) at moderate light intensity (40 μmol of photons m^–2 s^–1) at 28°C; further details of the labeling procedure are described in Dobáková et al. (2009). Two-dimensional (2D) protein separation was performed as described above. The 2D gel was stained with SYPRO Orange, blotted onto a PVDF membrane and exposed to a phosphorimager plate (GE Healthcare, Austria) overnight and scanned by Storm 860 (GE Healthcare, Austria).

The model of the thylakoid membrane, fully hydrated, with ions and equilibrated for 200 ns, has been described previously (Daskalakis, 2018). The structure of pPilA1 protein was modeled using iTASSER (Yang et al., 2015) with a C-score of = −1.3. Molecular dynamics (MD) simulations using the thylakoid membrane were performed using GROMACS version 2018.2 (SoftwareX, 2015, 1–2, 19–25), the AMBER03 forcefield (Duan et al., 2003) and the SPC water model. The protein was inserted into the membrane using Schrodinger’s Maestro software, and the remaining preparation steps (solvent box generation and ion placement) were carried out with GROMACS. The system was minimized using steep descent until the maximum force in the system was below 400 kJ/mol/nm. Next, NVT equilibration was performed at 303 K for 20 ns followed by NPT equilibration at 303 K and 1 atm for 40 ns. Finally, 500 ns production simulations were performed with frames saved every 1 ns. In both equilibration phases, as well as the subsequent production simulation, the time step was 2 fs, since h-bonds were constrained using the LINCS algorithm. The temperature was controlled using a V-rescale thermostat in the NVT equilibration and Nose-Hoover thermostat in NPT and production phases. The pressure was controlled with a Parrinello-Rahman barostat. Electrostatics were taken into account using PME with 1.2 nm electrostatic and van der Waals cut-offs.

MD simulations with the POPC membrane were run using Desmond 2015.4 (Bowers et al., 2006), the OPLS2005 force field, and the SPC water model. The protein was inserted into the membrane using Schrodinger’s Maestro software, and the remaining preparation steps were performed with Desmond. The equilibration set-up was the default protocol for Desmond, consisting of (1) Brownian Dynamics simulation in NVT at 10 K with restraints on solute heavy atoms for 100 ps, (2) NVT simulation at 10 K with restraints on solute heavy atoms for 12 ps, (3) NPT simulation at 10 K with restraints on solute heavy atoms for 12 ps, (4) NPT simulation at 300 K and 1 atm with restraints on solute heavy atoms for 12 ps, (5) NPT simulations at the same conditions with no restraints for 24 ps, and (6) production NPT simulation at the same conditions with no restraints for 500 ns.

To generate suppressor mutations restoring the photoautotrophy of the Synechocystis ΔpilD deletion mutant, we grew ΔpilD cells on BG11 agar plates without glucose (Glc^–) at normal light conditions (40 μmol of photons m^–2 s^–1). However, cells bleached completely in a few days without forming any suppressor colonies forcing us to try a second strategy. In contrast to photoautotrophic conditions, ΔpilD cells grow well on agar plates supplemented with 5 mM glucose (Glc⁺ conditions) for few days but after several days the cell pigmentation turned pale (Supplementary Figure 1) and the majority of cells died in 2 weeks. Notably, many green ΔpilD suppressor colonies appeared after 2 weeks on Glc⁺ plates (Supplementary Figure 1) and virtually all of them were able to grow photoautotrophically when tested on Glc^– plates. We selected four colonies (rev1-4), that differed markedly in size and pigmentation, for an in-depth analysis.

To identify suppressor mutations in rev1-4 strains, we performed their whole genome re-sequencing using an Illumina HiSeq platform. Obtained reads were mapped on the GT-S reference genome (Tajima et al., 2011) and, after subtraction of mutations that had been identified previously to occur in our WT laboratory substrain (GT-W; Tichý et al., 2016), we revealed mutations specific for the ΔpilD revertants (Figure 1 and Table 1). The point mutation in rev1 resulted in the amino-acid substitution H45P in the SigF sigma factor (Figure 1A) and the mutation in rev2 led to the S3G substitution in a signal peptide of the major pilin, PilA1 (Figure 1B). The mutation in rev3 is located in the pilA1-pilA2 intergenic region (Figure 1C), specifically in the predicted terminator of the pilA1 gene (Hess, W., personal communication). The rev4 strain contained two mutations leading to the R211C substitution in SigF (Figure 1A) and to the E95K substitution in the RNAP γ subunit.

Particular mutations in the pilA1 gene might directly reduce the PilA1 content, and thus lead to a formation of ΔpilD suppressor lines. Suppressor mutations in the sigF gene make also sense as the RNAP-SigF holoenzyme is responsible for the transcription of the specific regulon comprising the pilA1 gene (Asayama and Imamura, 2008) and potentially also other pilA genes in Synechocystis (Flores et al., 2019; see section “Discussion”).

The rev1 and rev4 strains (suppressor mutations in the sigF gene) grew as well as WT on BG-11 plates both mixotrophically (Glc⁺ plates) and photoautotrophically (Glc^– plates; Figures 2A,B). Rev2 and rev3 strains bearing suppressor mutations in the pilA1 gene region, grew into a much thinner layer than WT, and the rev2 strain partially bleached after 3 weeks on Glc⁺ plates (Figure 2A). Under Glc^– conditions, rev2 and rev3 strains proliferated only for a few first days and the rev3 mutant visibly faded showing a poorer phenotype than rev2 (Figure 2B), opposite to that seen in mixotrophic conditions. Rev1 and rev4 strains grew well also in photoautotrophic liquid culture (Glc^–; Figure 2C). Consistent with the growth restrictions on the agar plate, both rev2 and rev3 suffered from a growth defect in Glc^– liquid cultures (Figure 2C). Growth of the rev2 strain was slow throughout the experiment (Figure 2C) and its Chl level remained constant (Supplementary Figure 1B), whereas the initial proliferation of rev3 was similar to WT, but at higher culture density the growth of rev3 was arrested (Figure 2C), which was accompanied by a sharp decrease in Chl content (Supplementary Figure 1B).

The fast depletion of PSII in the ΔpilD mutant after its shift from Glc⁺ to Glc^– conditions has been reported previously (Linhartová et al., 2014). We, therefore, analyzed the content of PSII in revertants 48 h after changing the growth media from Glc⁺ to Glc^– conditions. Indeed, ΔpilD mutant cells lost most of the PSII complexes and particularly the dimeric PSII form disappeared almost completely (Figure 2D). In contrast, all four suppressor mutants were capable of maintaining dimeric PSII although only rev1 and rev4 had the level comparable to WT (Figure 2D). As shown by a more detailed analysis of membrane proteins by 2D clear-native/SDS electrophoresis (CN/SDS-PAGE), rev3 cells were visibly defective in the accumulation of membrane proteins as the total levels of many of them were reduced (Supplementary Figure 2).

A striking characteristic of Synechocystis ΔpilD and ΔpilA1/2 mutants is very intense cell aggregation under Glc⁺ growth conditions (Figure 3). This type of aggregation persists a vigorous shaking at 120 rpm and thus differs from cell flocculation characteristic of large flocs formed by cells settled down at the bottom of the cultivation flask (Conradi et al., 2019). Interestingly, the rev1 and rev4 strains, harboring mutations in the sigF gene, exhibited almost no tendency to form such aggregates (Figure 3) and the sigF deletion strain also did not aggregate (Figure 3). Aggregation of rev2 cells was weaker than that of the ΔpilD mutant and a portion of cells remained fully dispersed, whereas the rev3 aggregated comparably to the parental strain (Figure 3). None of these strains aggregated under photoautotrophic conditions (Figure 3).

We conclude that suppressor mutations in the rev1 and rev4 strains almost completely restored the viability of the ΔpilD mutant in our standard laboratory conditions. The remaining two strains (rev2, rev3) were only partially complemented. Although the rev3 strain proliferated better than rev2 if both supplemented with glucose, it appeared phenotypically closest to the original ΔpilD mutant based on the poor accumulation of membrane proteins (Supplementary Figure 2), a compact cell aggregation (Figure 3), the growth inhibition, and loss of Chl during the prolonged cultivation on plates (Figure 2B) as well as in liquid cultures (Figure 2C).

As all suppressor mutations could potentially affect the expression of the pilA1 gene, we quantified the pilA1 transcripts by Northern blotting both in Glc^– and Glc⁺ conditions. Note, that the glucose itself had a strong suppressor effect on the prepilin toxicity (Figures 2A,B). The pilA1 transcript was found to be monocistronic (Figures 4A,B), consistent with previous reports (He and Vermaas, 1999; Asayama and Imamura, 2008), and the pilA1 mRNA was 4 to 5 times more abundant in the ΔpilD mutant both in Glc^– and Glc⁺ than in WT (Figures 4C,D). As expected no pilA1 transcript was detected in the sigF^– mutant (Figure 4A).

In the rev1 strain, the H45P mutation in the SigF protein reduced the pilA1 mRNA content to a similar level as in WT both in Glc^– and Glc⁺ conditions (Figure 4). On the other hand, in the rev4 strain (mutations in SigF and the γ subunit of RNAP) the pilA1 transcript level dropped to the WT level only in Glc^– but remained similar as in the ΔpilD strain in mixotrophic conditions (Figure 4). The amount of pilA1 mRNA in rev2 (mutation in the pilA1 gene) was high (Figure 4), indicating that the suppressor effect of this revertant line is not due to the weakened pilA1 expression. In the rev3 strain, the point mutation in the pilA1-pilA2 intergenic region lowered the amount of pilA1 transcript to the WT level only in Glc⁺ conditions. Interestingly, we also detected a longer pilA1 transcript (∼1,800 bp) in rev3 and specifically under the photoautotrophic conditions (Figures 4A,B); this result was observed in three biological replicates (Supplementary Figure 3). The length of this transcript corresponds to pilA1-pilA2-sll1696 polycistronic mRNA suggesting that the rev3 mutation might weaken the pilA1 terminator.

Based on transcriptomic data, Kopf et al. (2014) identified a Synechocystis transcription unit (TU1331) coding for a putative pilA1 anti-sense RNA with length ∼100 bp longer than the “sense” pilA1 mRNA. The rev3 mutation is located in the promoter region of the expected anti-sense gene between −10 and −35 elements (Figure 1C). However, we did not detect such an antisense transcript on the Northern blot, which suggests that its level is much lower than that of pilA1 (Figures 4A,B).

This transcript analysis revealed a complex regulation of the pilA1 gene regarding the trophic mode. Since the pPilA1 protein is connected with lethality under Glc^– conditions (Figure 2B), the lowered pilA1 transcript (to the level detected in WT; Figure 4D) in rev1 and rev4 after 48 h without glucose could explain the ability of these strains to grow photoautotrophically. On the other hand, the very strong expression of pilA1 in the rev2 and rev3 strains in Glc^– (Figure 4D) implies that they suppress the prepilin toxicity by a different mechanism(s).

The level of PilA1 protein in the isolated cellular membrane fraction was assessed using immunoblotting. The newly synthesized PilA1 protein (prepilin) is matured by the PilD protease and further glycosylated (Kim et al., 2009; Linhartová et al., 2014). In Synechocystis WT cells, we detected only the mature glycosylated PilA1 protein, whereas the prepilin or the non-glycosylated pilin were below the detection limits (Figure 5). It is notable that in WT cells the PilA1 content greatly increased after shifting cells from Glc⁺ to Glc^– conditions (Figure 5B and Supplementary Figure 4). This increase in the PilA1 level in WT has not been observed previously after 24 h of incubation under Glc^– conditions (Linhartová et al., 2014) indicating that the higher accumulation of PilA1 requires longer (48 h) incubation in the absence of glucose or it is stimulated with higher cell density.

Without processing protease in the ΔpilD strain, a high amount of prepilin accumulated both in Glc⁺ and Glc^– conditions, circa 50% of which was non-glycosylated pPilA1 (pPilA1ⁿ) and 50% glycosylated pPilA1 (pPilA1^g; Figure 5). Consistently with the low level of pilA1 mRNA, the rev1 cells contained almost no pPilA1ⁿ and less of pPilA1^g than the ΔpilD mutant regardless of the growth conditions used. Similarly, low pilA1 mRNA was accompanied by traces of pPilA1ⁿ and a low amount of pPilA1^g in rev4 in autotropic conditions. In mixotrophic conditions, the enhanced pilA1 transcripts of rev4 (Figure 4) were accompanied by a high amount of pPilA1^g although the amount of pPilA1ⁿ remained low. Thus, pilA mRNA and prepilin protein levels correlated in suppressor lines containing mutations in the SigF factor unlike in the other two suppressor strains. The high pilA1 transcript levels in rev2 strain did not lead to high pPilA1 content, rev2 contained less pPilA1 proteins than pilD mutant and especially pPilA1ⁿ content was low both in photoautotrophic and mixotrophic conditions (Figure 5). Unlike the other revertant strains, the rev3 strain contained more pPilA1^g than the ΔpilD mutant both in photoautotrophic and mixotrophic conditions, whereas the level of pPilA1ⁿ in rev3 was lower than in ΔpilD mutant only in photoautotrophic conditions (Figure 5). Thus the ratio of pPilA1 forms in rev3 shifted from half glycosylated and half non-glycosylated in Glc⁺ to ∼75% of the glycosylated prepilin over ∼25% of the non-glycosylated in Glc^–.

Accumulation of pPilA1 in the ΔpilD mutant in Glc^– was previously shown to cause degradation of the SecY translocase and the YidC insertase (Linhartová et al., 2014) as well the lumenal Ycf48 protein, which forms a complex with YidC and facilitates the insertion of Chl molecules into PSII subunits (Bučinská et al., 2018; Yu et al., 2018). This process was accompanied by abolished production of PSII subunits (Linhartová et al., 2014). The proteolytic degradation of YidC and Ycf48 in the ΔpilD mutant in Glc^– (Figure 5) is in agreement with the earlier results. Notably, the content and stability of YidC insertase and Ycf48 did not appear affected in any of the suppressor strains including rev3 that is deficient in the content of membrane proteins (Supplementary Figure 2).

We have previously speculated that the pPilA1ⁿ form of prepilin, rather than the glycosylated pPilA1^g, is responsible for the deleterious effect on the membrane-protein synthesis (Linhartová et al., 2014). In accordance with that, lower amounts of the pPilA1ⁿ (but not pPilA1^g) than in pilD mutant was measured in all suppressor lines (Figure 5).

To better understand prepilin toxicity, we assessed the synthesis of membrane proteins in cells grown under Glc^– conditions by radioactive pulse labeling (20 min) using a [³⁵S]-Met/Cys mixture. Labeled membrane proteins were separated on SDS-PAGE (Coomassie and SYPRO Orange stained gels are shown in Supplementary Figures 5, 6) and autoradiographed (Figure 6). As shown previously, the synthesis of membrane proteins in the ΔpilD strain under Glc^– conditions is extremely weak (Linhartová et al., 2014) and we, therefore, compared the protein labeling of rev1-4 strains with WT (all after 48 h in Glc^–) and with the ΔpilD (Glc⁺) sample. Intriguingly, even in a (fresh) Glc⁺ culture, the overall labeling of membrane proteins in ΔpilD was much less intensive than in WT (Glc^–; Figure 6A). On the contrary, in rev1, rev2, and rev4 strains the total synthesis of membrane proteins was similar to that in WT. Consistent with the lowered amount of membrane proteins in rev3 in Glc^– conditions (Supplementary Figure 2), the total protein labeling in this strain was < 50% of that in WT including low synthesis of D1 and D2 core subunits of PSII (Figure 6B).

The radioactive signal on pPilA1^g was much weaker in all rev strains than in the ΔpilD (Glc⁺) sample (Figure 6A). Given this low intensity, we cannot exclude a contribution of other weakly labeled proteins with gel mobility similar to pPilA1^g. Moreover, the pPilA1ⁿ comigrates with an unknown intensively labeled protein (marked by an asterisk in Figure 6A) and therefore cannot be resolved on 1D-gel. To obtain a better resolution of labeled pPilA1^g and pPilA1ⁿ, the identical samples as in Figure 6A were separated by 2D CN/SDS-PAGE and blotted. The radioactivity of the blotted proteins was visualized by phosphorimaging and the identity of pPilA1^g and pPilA1^g spots then verified by immunoblotting (Supplementary Figure 7).

We found that the synthesis of prepilin in rev1 and rev4 was slow as we were not able to detect the radiolabeled pPilA1^g or pPilA1ⁿ in these strains (Figure 6B). Thus the low amount of both forms of pPilA1 protein detected by immunoblotting probably slowly accumulate and at least a fraction of pPilA1 is long-lived without the PilD protein in rev1 and rev4 (Figure 5). In contrast to rev1 and rev4, the pulse labeled forms of prepilin were well detectable in rev2 and rev3. In rev3, about a half of the labeled prepilins was non-glycosylated, whereas in rev2 pPilA1ⁿ was the prevailing form (Figure 6B). It should be noted that in the previous study we were not able to detect PilA1ⁿ in WT even by radiolabeling (Linhartová et al., 2014). Therefore, the glycosylation of PilA1 in WT must be very fast, but results from rev strains indicate that the glycosylation of prepilin occurs slowly.

Two-dimensional separation of radiolabeled proteins allowed us to inspect the process of PSII biogenesis. The labeled D1 and D2 subunits incorporated into the new PSII complexes were much weaker in rev3 than in other rev strains (Figure 6B). However, the process of PSII assembly was not restored in rev2 either. Although the synthesis of D1 and D2 subunits in rev2 appeared as intensive as in WT (Figure 6A), the signal of labeled D1 and D2 in monomeric and dimeric PSII was lower than in rev1 or rev4. The partially blocked biogenesis of PSII correlated with the lowered level of PSII in rev2 and rev3 (Figure 2D).

The greatly reduced expression of pilA1 in rev1 and rev4 under Glc^– offers a straightforward explanation of how the SigF and RNAP mutations suppress the loss of pilD. The mechanism(s) of rev2 and rev3 mutations, however, remained enigmatic. The rev3 mutation is located in the predicted terminator of the pilA1 gene and might thus increase the expression of the downstream located pilA2 and sll1696 genes (Figure 1C); in WT these two genes are co-transcribed (Kopf et al., 2014). To see any potential effect of the disrupted pilA2-sll1696 locus on the stability/biogenesis of PilA1 we immunodetected the PilA1 protein in pilA2^– (Glc^–) and ΔpilD/pilA2^– (Glc⁺) strains. Interestingly, another form of PilA1 was detected in the pilA2^– strain, which would correspond to the non-glycosylated PilA1 (PilA1ⁿ; Figure 7A). Moreover, in the ΔpilD/pilA2^– strain, the glycosylated pPilA1^g was almost completely absent while the level of pPilA1ⁿ remained high (Figure 7B). These data suggest that the pilA2-sll1696 locus is required for the modification of (p)PilA1 and also support the model that, rather than the pPilA1^g, the pPilA1ⁿ inhibits protein synthesis; note that the disruption of pilA2 does not complement the ΔpilD mutant (Linhartová et al., 2014). Indeed, if compared to the original ΔpilD strain, the pPilA1ⁿ in rev3 (Glc^–) appears specifically decreased (Figure 5B; see also Figure 7C).

The S3G suppressor mutation in rev2 is located in the signal peptide that is cleaved out by the PilD protease during the pPilA1 maturation. In our previous work, the native pPilA co-immunoprecipitated with SecY showing an aberrant interaction between the prepilins and the translocon (Linhartová et al., 2014). To check the potential effect of the S3G mutation of pilA1 on the interaction with translocons, we immunoprecipitated SecY and detected the co-purified pPilA1 in rev2 and rev3 strains. In the ΔpilD mutant the pPilA1^g and a smaller amount of pPilA1ⁿ co-eluted with SecY but, importantly, the mutated pPilA1-S3G protein did not co-immuoprecipitated with SecY from the rev2 strain (Figure 7C). Although the pPilA1 was less abundant in rev2 membranes than in ΔpilD (Figure 7C), no detectable pPilA1 in the SecY precipitate suggested that the S3G mutation makes the association with translocons less likely. In contrast to rev2, the pPilA1^g protein from the rev3 strain co-immunoprecipitated with SecY (Figure 7C).

According to available structures of the PilD-related aspartyl proteases (3S0X; 4HYC), the charged prepilin signal peptide is cleaved by PilD after being exposed to the cytosolic/stromal surface of the cell membrane (Hu et al., 2011; Li et al., 2013). To understand how the replacement of Ser3 with Gly3 in the signal peptide prevents the aberrant association of the prepilin to SecY, we performed MD simulations of the pPilA1ⁿ protein in a model TM (Daskalakis, 2018). As a starting position, the signal peptide was exposed in the cytosol (Figure 8A). Interestingly, the charged signal peptide almost immediately (6 ns) attached to the membrane surface, forming several hydrogen bonds between hydroxyl groups of galactolipids and Met1, Ser3, and Arg17 residues (Supplementary Figure 8A). Thereafter, the signal peptide remained attached to lipid sugars through the entire 0.5 μs MD simulation, just becoming deeply buried into the membrane (Figure 8A and Supplementary Video 1).

Although the association of the signal peptide with the model TM appears very robust, the SecY translocase is known to associate specifically with anionic lipids, such as phosphatidylglycerol or cardiolipin (reviewed in Crane and Randall, 2017). However, the total content of phosphatidylglycerol in Synechocystis cells is low (∼10%; Kopečná et al., 2015), and such a low content was used in our model TM (Daskalakis, 2018). To evaluate the effect of phospholipids on the interaction between the signal peptide and the membrane, we run the same simulation in the model phosphatidylcholine (POPC) membrane bilayer. In this case, the formation of stable contact between the signal peptide and POPC phosphate groups took longer (∼0.25 μs; Supplementary Figure 8B), but the peptide remained firmly embedded in the membrane later on (Figure 8B and Supplementary Video 2). Again, the network of hydrogen bonds between the signal peptide and lipids included the hydroxyl group of S3 residue (Supplementary Figure 8B).

We performed simulations of the mutated pPilA1-S3G and compared the strength of interaction between the signal peptide and lipids with the simulation of WT pPilA1 (see Supplementary Figures 8C,D for initial positions). In both TM and POPC membranes, the number of hydrogen bonds per frame was higher for the WT protein (Figure 8C). Notably, the interaction of the WT pPilA with the POPC membrane surface specifically strengthened (8.6 H-bonds per frame) after the formation of a stable contact with phospholipids. In contrast, the interaction of the mutated signal peptide with POPC remained transient (2.8 H-bonds per frame; Figure 8C) as documented also by the final position of the N-terminus extended above the membrane surface (Figure 8D and Supplementary Videos 3, 4).