Escherichia coli Top10 (ThermoFischer scientific, Carlsbad, CA, USA) was used for all cloning purposes and was grown in lysogeny broth (LB) or on LB agar (LB broth with 1.5% w/v agar) at 37°C. Saccharomyces cerevisiae AH109 and S. cerevisiae Y187 (BD Bioscience) were used in the yeast two-hybrid screen. After transformation, all yeast strains were grown at 30°C on Synthetic Defined (SD) medium (Roucourt et al., 2009), with omission of specific amino acids, dependent on the desired selection, as shown below.

Open reading frames (ORFs) containing the various domains and inter and intraconnective regions of the kalimantacin assembly line were amplified from the genomic DNA of P. fluorescens BCCM_ID9359 using Phusion^® High Fidelity DNA polymerase (ThermoFischer scientific). An overview of the primers and the length of the corresponding fragments can be found in Supplementary Table S1. The PCR fragments were inserted in the pCR^TM8/GW/TOPO^® vector (ThermoFischer scientific) by A-overhang ligation. Subsequently, transfer of coding fragments from the TOPO vector to the Gateway^TM compatible bait (pGBT9) and prey (pGAD424) vectors (Clontech) was realized using Gateway^® LR Clonase^TM Enzyme Mix, following the manufacturer’s protocol. All constructs were verified by Sanger sequencing (GATC Biotech).

Saccharomyces cerevisiae AH109 (Mata) and S. cerevisiae Y187 (Matα) were transformed with bait and prey vectors, respectively. Transformation of the constructs was performed on 96-well scale, using the protocol of Rajagopala and Uetz (2011). Both yeast strains are auxotrophic for tryptophan, leucine, histidine and adenine. Selection for yeast cells containing the bait vector was performed on SD media lacking Trp, while media without Leu were used for prey selection. Autoactivation of bait constructs was verified by an assay using empty prey vector and prey vector with an unrelated gene from Pseudomonas aeruginosa PAO1, mvaT. In this assay, the concentration of 3-amino-1,2,4-triazole (3-AT) necessary to reduce leaky expression of reporter genes was optimized at 3 mM.

The actual Gal4p-based yeast two-hybrid screening was performed by an eight-clone pooled, array-based mating screening. First, the bait clones were pooled in groups of eight in such way that every clone was represented in two pools. These pools were mated with the entire prey array, resulting in diploid yeast cells containing one bait and one prey vector. The diploid yeast cells were spotted on SD media lacking histidine and adenine to screen for activation of HIS3 and ADE2. 3-AT and 5-bromo-4-chloro-3-indoyl-α-D-galactopyranoside (X-α-gal) were independently added to the selective media, minimizing leaky expression of HIS3 and enabling detection of α-galactosidase activity by expression of MEL1. The reciprocal screening, prey pool – bait array, was also performed.

Positive interactions were independently confirmed by co-transformation of S. cerevisiae AH109 with bait and prey constructs followed by spotting on selective media in twofold dilution series. Finally, the level of the detected protein interactions was quantified using an α-galactosidase assay (Clontech Laboratories, 2009).

Yeast two-hybrid screening is a very sensitive and powerful method for detection of protein–protein interactions. Its ability to screen large libraries and even visualize transient interactions makes this technique particularly suited for the analysis of PKS and NRPS systems. However, as an intrinsic limitation of the Y2H approach, expression of bacterial proteins in yeast cells can result in the absence of post-translational modifications present in a natural context, which will impose limitations to the results obtained in the screen. Literature shows that N- and C-terminal fragments of PKS or NRPS domains, often described as linkers and docking regions, are involved in specific interactions connecting modules and domains (Broadhurst et al., 2003; Tang et al., 2007; Buchholz et al., 2009; Cheng et al., 2009). In view of this, delineation of the fragments in this analysis was set up in such way that each domain was flanked by the connector region between two adjacent domains. As such, each flanking region was represented at least two times in the high-throughput screening, as illustrated in Figure 1. After amplification, 63 fragments were obtained representing the entire kalimantacin biosynthesis cluster, including tailoring domains BatA-BatM.

First, the 63 fragments were inserted in the pCR^TM8/GW/TOPO^® vector, followed by Gateway transfer to both yeast two-hybrid vectors: bait vector pGBT9 and prey vector pGAD424. Each fragment was tested both as bait and prey protein, thus raising the reliability of the screen. To verify the efficiency of this pooled array approach, a positive control was included in the screen, using interacting proteins Luz24_gp4 (pGBT9) and MvaT (pGAD424) (Wagemans et al., 2015).

The autoactivation assay (Supplementary Table S2) revealed six bait proteins which already activate the three reporter genes (HIS3, ADE2 and MEL1) when combined with an empty prey vector or a prey containing the mvaT gene. As a consequence, these were excluded from the bait pools. One bait protein, KS1, demonstrated only limited activation of the reporter genes in this control experiment and was retained in the pools.

The remaining bait proteins and pGBT9 positive control were pooled in groups of eight such that every clone was represented in two pools. In total, 16 unique bait pools were created and each pool was separately tested against the entire prey array by mating and testing on selective media. Blue colonies grown on SD-WLHA + X-α-gal plates were considered a positive outcome. By combining the different results, pools interacting with the same prey could be identified. The common bait clone of these pools was determined as the possible interaction partner of that specific prey fragment.

Twelve possible protein–protein interactions were observed after mating the bait pools with the prey array. The reciprocal screen confirmed 9 of these interactions, but also discovered 16 new possible interacting proteins (Figure 2). The positive control was clearly present in both screens, validating the effectiveness of this pooled array screening method.

Each interaction was verified by co-transformation of S. cerevisiae AH109 with the corresponding bait and prey proteins. As a negative control, an unrelated protein Gpa1 was used. In addition, the self-activating bait protein KS1 was tested against MvaT. The screen confirmed 13 interacting pairs, while an additional four bait and prey couples only grew on SD-WLH plates. All validation data and the quantitative interaction data are combined in Table 1 and are individually discussed and interpreted in the next paragraphs.

The Y2H analysis shows an interaction between the ER BatK and both ACP8 and ACP10. The interaction of BatK with ACP10 can be observed in both directions, whereas ACP8 as bait interacts with the prey construct of BatK, but the reverse direction only results in limited growth on medium lacking histidine (Table 1). BatK reduces the double bond created in module 8 by the KR and DH domain (Mattheus et al., 2010a), hence interaction around module 8 is logical. Whether it binds to ACP8, ACP10 or both should be further investigated, but from our analysis it appears that this trans-acting enzyme not only recognizes the double bond of the growing intermediate but also attaches to the enzymatic machinery.

KS1 is one of the proteins which shows (limited) autoactivation in this yeast two-hybrid screen. Indeed, the construct in yeast results in a small white colony on the most stringent medium. Hence, results showing big blue colonies, clearly different from the result after the autoactivation test, can be taken into account (Table 1; Supplementary Table S2). This strategy resulted in seven potential interaction partners for KS1.

BatJ is predicted to be the trans-AT and delivers a malonyl building block to each module (Mattheus et al., 2010a). In this analysis, only an interaction with KS1 is observed. It has been suggested before that trans-acting AT domains would dock immediately downstream of the KS domain and it has been observed that the region behind the KS domains in many trans-AT systems consists of many residues resulting in unfolded proteins that only adopt a defined fold in the presence of other binding proteins (Cheng et al., 2009; Davison et al., 2014). The interaction of the KS1 fragment with BatJ supports this hypothesis, however, docking of BatJ to this region would be expected to occur in each module. Recently, different studies have shown that trans-AT domains can directly transfer their acyl group to the ACP domain, even in the absence of a KS domain or post-KS linker region and suggestions rise that the trans-AT domains do not even dock to the mega enzyme cluster (Aron et al., 2007; Musiol et al., 2011; Wong et al., 2011; Ye and Williams, 2014). This could explain the low amount of interactions seen for BatJ.

The interaction observed between KS1-KR1 and KS1 can be explained by the homology of the C-terminal fragment of the KS domain in trans-AT PKSs with the so-called KS-AT and post-AT linkers identified in the DEBS KS-AT di-domain structures (Tang et al., 2006, 2007; Davison et al., 2014). These linkers wrap back over the KS domain to stabilize the KS-AT structure. It is possible that the C-terminal fragment here shows comparable behavior and thus interacts with the KS domain. The dimerization of KS1 is discussed further.

The next three protein fragments observed to interact with KS1 are ACP9b, KS10-ACP10 and TE. As they are located at the end of the assembly line, these observed interactions may be false positives. For instance, it is possible for KS1 to interact with the post-KS10 linker region (KS10-ACP10), but since in vivo, the post-KS1 linker is covalently attached to the KS1 fragment, the chance of binding this fragment is much higher than the binding to the post-KS10 region. The interaction between KS1 and TE is potentially interesting. Since in vitro tests show that the TE-domain does not recognize the ACP domain, but the substrate attached to it, it might now be interesting to specifically look to the interaction between the TE domain and KS domains (Tran et al., 2008, 2010). Still, even if there is interaction between TE domains and KS domains, it is to be expected that this interaction occurs in the last module and not in the first of the assembly line.

The last protein found to interact with KS1 is MvaT. This result is irrelevant for the analysis, since MvaT is an unrelated protein from Pseudomonas aeruginosa PAO1. False positive results are inherent to protein interaction techniques, e.g., as a result of artificially combining proteins that can never meet in biological conditions. This result emphasizes the need for focused interaction studies to confirm results obtained in this study by other interaction techniques as discussed in the conclusions and perspectives.

In the kalimantacin assembly line BatE, BatF, KS1 and KS10 appear to at least dimerize. BatE is part of the β-methyl incorporation cassette, probably acting with module 5 and modules 9, 10, and 11 (Mattheus et al., 2010a). To our knowledge, no structural data of such cassettes have been described previously. As it is known that a conserved tyrosine residue in helix III of beta-branching ACP domains seems to attract the β-branching cassette and looks to be important for interaction with the cassette, it could be interesting to further explore the structure of such cassettes and to use this information in genetic engineering approaches (Haines et al., 2013).

BatF is the carbamoyl transferase that activates kalimantacin at the end of the biosynthesis pathway by transferring a carbamoyl group to the molecule (Mattheus et al., 2010a). The dimerization observed in this screen could indicate that O-carbamoyl transferases are composed as multimers, as has been seen for all N-carbamoyl transferase structures identified to date (Shi et al., 2015).

Different structures of PKS modules have been published and it is unclear whether these differences are due to the PKS system analyzed, the method used for structural elucidation or because incomplete PKS modules fold differently than the complete module. However, a common feature in all reported structures is the dimerization of the KS domain (Weissman, 2015). For the kalimantacin cluster, only KS1 and KS10 came back positive for dimerization in this Y2H screen. Although we would expect all KS domains to dimerize, the results are consistent with previous research. It was suggested before that additional dimerization domains in the PKS module help the KS dimerization, since every PKS module has at least one dimerization domain C-terminal of the KS domain (Dutta et al., 2014).

According to our analysis, modules 9 and 10 interact with each other at two different sites. KS10-ACP10 interacts with the ACP9b fragment, and KS9left with KS10-ACP10. Although modules 9 and 10 are adjacent to each other and interactions are expected, it is difficult to hypothesize on a biological reason for the observed results.

Apart from false positive results, it is also possible that interactions that do occur in natural conditions do not appear in the results. These false negative results can, for instance, result from instability of the selected protein fragment, incorrect folding or steric hindrance of the fusion with the activation or DNA binding domain, intrinsically associated with yeast two-hybrid screens (Rajagopala and Uetz, 2009).

The kalimantacin core assembly line consists of three polypeptides. For efficient production of kalimantacin, it would be expected that these proteins interact in a very efficient way. In our results, no interaction between the different proteins is observed. Research on so-called docking domains in the N-terminal and C-terminal region of PKS proteins proved that the dissociation constants of such interactions are very low and thus the proteins interact only transiently (Buchholz et al., 2009; Whicher et al., 2013). Possibly, the interactions are not strong enough to be detected in this pooled assay.

ACP and PCP domains are core proteins in PKS and NPRS clusters, respectively. They carry the new building block and the growing intermediate and are thought to be responsible for the transport of all compounds to the different active sites in the module. Hence, ACP domains are expected to interact with a lot of different proteins. In our analysis, the amount of interactions involving ACP domains was limited. As ACP domains adopt different structures depending on the attachment of the phosphopantetheinyl arm, it can be important that they are in their active holo-form for interaction with their partner proteins (Evans et al., 2009). In yeast cells, it is possible that the ACP domains of the kalimantacin cluster do not carry this arm, since not all phosphopantetheinyl transferases show a broad specificity for ACP domains (Lambalot et al., 1996). It might be interesting to repeat the assay described here with a yeast strain heterologously expressing the broad spectrum Sfp phosphopantetheinyl transferase (Quadri et al., 1998).