SecMet-FISH: labeling, visualization, and enumeration of secondary metabolite producing microorganisms

Abstract Our understanding of the role of secondary metabolites in microbial communities is challenged by intrinsic limitations of culturing bacteria under laboratory conditions and hence cultivation independent approaches are needed. Here, we present a protocol termed Secondary Metabolite FISH (SecMet-FISH), combining advantages of gene-targeted fluorescence in situ hybridization (geneFISH) with in-solution methods (in-solution FISH) to detect and quantify cells based on their genetic capacity to produce secondary metabolites. The approach capitalizes on the conserved nature of biosynthetic gene clusters (BGCs) encoding adenylation (AD) and ketosynthase (KS) domains, and thus selectively targets the genetic basis of non-ribosomal peptide and polyketide biosynthesis. The concept relies on the generation of amplicon pools using degenerate primers broadly targeting AD and KS domains followed by fluorescent labeling, detection, and quantification. Initially, we obtained AD and KS amplicons from Pseuodoalteromonas rubra, which allowed us to successfully label and visualize BGCs within P. rubra cells, demonstrating the feasibility of SecMet-FISH. Next, we adapted the protocol and optimized it for hybridization in both Gram-negative and Gram-positive bacterial cell suspensions, enabling high-throughput single cell analysis by flow cytometry. Ultimately, we used SecMet-FISH to successfully distinguish secondary metabolite producers from non-producers in a five-member synthetic community.


Introduction
The significance of c hemicall y mediated inter actions among micr oor ganisms is well established, yet the importance of the diversity of microbial secondary metabolites in shaping the assembly and dynamics of micr obial comm unities r emains poorl y understood.This is in large part due to the intrinsic limitations of culturing environmental bacteria under laboratory conditions, which result in a significant gap between the number of bacterial lineages we can culture and analyze under manmade conditions, and the number we can detect in nature by culture independent methods (Steen et al. 2019 ).Mor eov er, complexity and spatial organization is crucial for understanding how secondary metabolite producers are distributed and how they influence microbial community assembly, composition, and function; elements which are excluded in the analyses of laboratory monocultures.
Metagenomic sequencing of environmental samples has facilitated culture independent discovery of novel taxa and functional genes (Rinke et al. 2013 ).Ho w e v er, suc h a ppr oac hes also omit spatial organization and microscale variability.Moreover, this appr oac h has the inherent disadvantage of sequencing all DNA, be it informative or not, which means that sequences of interest are diluted within data of secondary inter est, especiall y in highl y complex natur al micr obial comm unities (Zaheer et al. 2018, Libis et al. 2019, Robinson et al. 2021 ).While metagenomic assembly and binning methods ha ve impro ved dramatically in the last few years (Wu et al. 2016, Nurk et al. 2017, Nissen et al. 2021 ), ty-ing functional genes to specific taxa remains an additional challenge when r ele v ant phylogenetic markers and functional genes of inter est ar e distributed acr oss contigs in incomplete assemblies .Moreo ver, genetic regions with high levels of repetition and tandem repeats (such as in biosynthetic gene clusters, or BGCs) are notorious breakpoints in genome assemblies, as exemplified by an extreme level of fragmentation of metagenomically-derived BGCs.By contr ast, de v elopment of fluor escence in situ hybridization (FISH) based a ppr oac hes using pol ynucleotide pr obes has the adv anta ge of dir ect visualization r ather than computational inference through the labelling of bacterial cells based on the presence or absence of specific genes (Moraru et al. 2010, Barrero-Canosa et al. 2017 ).This technique is routinely used to couple bacterial phylogeny to functionality and investigate where these cells are localized in situ (Ansorge et al. 2019 , Richards andMattes 2021 ).Ho w e v er, curr ent a ppr oac hes ar e c hallenged with limited signal intensity of single-copy target sequences and the fact that cells ar e first c hemicall y fixed and immobilized on a surface, whic h pr ohibits selective extraction and downstream analysis.To circumvent this , con ventional FISH methods e.g.targeting bacterial ribosomes have been adapted for 'in-solution' use, enabling high thr oughput anal yses of cell populations b y flo w c ytometry (Flo w-FISH; Yilmaz et al. 2010, Haroon et al. 2013 , Fr een-v an Heer en 2021 ) as well as enrichment of targeted phylogenetic groups by fluor escence-activ ated cell sorting (FACS; Podar et al. 2007 ).Combined with metagenomics of FACS-enriched bacterial communi-ties, or single cell sequencing, Flow-FISH has provided genomic sequences of r ar e and elusiv e comm unity members (Podar et al. 2007, Grieb et al. 2020 ).T hese methods ha ve , ho w ever, focused on selection of micr oor ganisms based on their phylogeny, or gene expr ession, r ather than their genetic potential to produce specific metabolites.
Combining the potential of direct-geneFISH with that of insolution FISH methodology, we aimed to establish an a ppr oac h that allows for fluorescent labeling, visualization and in-solution enumeration of microorganisms based on their biosynthetic potential.We envisioned that detection, quantification, and potentiall y fluor escence-guided extr action (e.g.FACS or optical tr a pping) of bacteria rich in biosynthetic genes would be a valuable tool for understanding the role of secondary metabolite producing micr oor ganisms in the envir onment, but also for genome mining and drug discovery (Robinson et al. 2021, Geers et al. 2022 ), as secondary metabolites play a significant role in modern medicine, serving as the inspiration for small-molecule ther a peutics.To accomplish this, we aimed to le v er a ge the conserved and modular nature of the adenylation (AD) and ketosynthase (KS) domains, both involved in the biosynthesis of secondary metabolites, including compounds with e.g.antibiotic activities (Walsh 2016 ).These gene domains are readily amplified with PCR through the use of established degenerate primers (Piel 2002, Ayuso-Sacido and Genilloud 2005, Geers et al. 2022 ), and their modular organization ensures a series of target sequences for hybridization, allowing for signal intensities sufficient for detection across instrumentation with different sensitivities.First, we set out to show that a mix of PCR amplicon sequences, rather than one specific sequence, can successfully be used to synthesize polynucleotide probes and label target cells.Next, we aimed to adapt the protocol for labelling of Gr am-positiv e and Gr am-negativ e bacterial cells in suspension to enable high-throughput detection and quantification using flow cytometry .Lastly , we assessed whether the fluorescence signal intensities generated by the SecMet-FISH approach are sufficient for selective identification of secondary metabolite producers in a synthetic community.

DNA extractions and PCR amplification of biosynthetic domains
DNA from P. rubra , S. coelicolor and the synthetic community (Syn-Com) samples were extracted using the NucleoSpin Tissue kit (740952.250Mac hery-Na gel, German y).PCRs wer e performed for two applications: 1) to amplify the conserved DNA sequences encoding the AD and KS domains, as well as the Non-Poly 350 sequence, for downstr eam fluor ophor e attac hment to synthesize SecMet-FISH probes (see below), and 2) for amplicon sequencing of the conserved biosynthetic AD and KS domains.All PCR amplifications were synthesized with HotStarTaq pol ymer ase (Qiagen) in 50 μL reaction volumes.Detailed information on primer sequences , thermal cycling conditions , and r ea gent and template concentr ations ar e summarized in Table S1 .
Amplification of the AD and KS domains from P. rubra and S. coelicolor gDN A w as ac hie v ed dir ectl y in a one-ste p PCR, exce pt for the AD amplification from P. rubra .Due to an ad ditional, putati ve unspecific product (as checked by agarose gel analysis), a two-step PCR w as emplo y ed for AD amplification from P. rubra .PCR product of the first round (38 cycles) was separated on an 1.5% a gar ose gel after which the ≈700 bp band was excised and gel purified using the GFX PCR DNA purification kit (Illustra, USA) and eluted in 25 μL elution buffer (10 mM Tris, 1 mM EDTA).Gel-purified products were diluted to a concentration of 1-10 pg/ μL and used as template for the second PCR round consisting of 35 cycles.For amplicon sequencing purposes, 8 bp-indexed primers were used in this second round PCR.

Amplicon sequencing and analysis
Amplicons of the biosynthetic AD and KS domains were sequenced b y Nov ogene (Cambridge , UK) on an Illumina No vaseq 6000.All fastq files wer e dem ultiplexed using cutadapt v1.18 (Martin 2011 ) and quality filtered with fastp v0.20 (Chen et al. 2018 ), using default settings.Briefly, sequence pairs with more than 40% bases < 15 Phr ed-scor e wer e r emov ed, and all sequences wer e 3 trimmed by quality, i.e. bases in a sliding window having a mean Phr ed-scor e < 15 were trimmed.
For analysis of the AD and KS domains amplified from the model strain P. rubra , demultiplexed reads were mapped to the r efer ence genome (NCBI accession: GCA_005886805.2 and GCA_008931305.1) with bowtie2 v2.3.5 (Langmead and Salzberg 2012 ) using default options except for the maximum fragment length which was set to 1000 bp.Mapped regions were analyzed using the mpileup function of samtools 1.6 (Li et al. 2009 ) and further processed and visualized with custom R scripts.In order to quantify the le v el of in vitro PCR-amplification and sequencing of AD and KS domains from P. rubra, a custom database of BGCs and their locations in this genome was built using antiSMASH v6.0 (Blin et al. 2021 ).The genomic coordinates of these BGCs were then cr oss-r efer enced with the bowtie2 ma ppings, and domains wer e consider ed successfull y amplified if the number of ma pped reads at their genomic location amounted to more than 1% of the total mapped reads.

Polynucleotide probe synthesis for SecMet-FISH
Gene probes were synthesized by chemical labelling with the Alexa Fluor 488, 594 and 647 fluor ophor es using the Ulysis Nucleic Acid Labeling kit (Life Technologies, USA).First, AD and KS amplicons from the same target sample were pooled and cleaned up with the AMPure XP magnetic beads with an elution volume of 25 μL in labelling buffer from the labeling kit.Subsequently, 1 μg of pooled amplicon DN A w as labelled follo wing the protocol described dir ect-geneFISH (Barr er o-Canosa et al. 2017 ).For eac h fluor ophor e, a non-sense contr ol pol ynucleotide pr obe NonPol yPr350 (Moraru et al. 2010 ) was synthesized using the same procedures, and used as a negativ e contr ol pr obe in all SecMet-FISH experiments.

SecMet-FISH in solution
To start the in-solution SecMet-FISH pr otocol tar geting Gr amnegative bacteria, fixed cells were pelleted by centrifugation in PCR tubes for 2 min at 10 000 × g and washed twice with PBS.Cells were resuspended in 50 μL hybridization buffer consisting of 40% (v/v) formamide, 900 mM NaCl, 20 mM Tris-HCl pH = 7.4, 0.1 mg/mL DNA from salmon sperm, 0.01% (w/v) SDS and 10% (w/v) dextran sulfate and sonicated for 1-2 min to aid resuspending the cell pellet.Polynucleotide probes were added to a final concentration of 250 pg/ μL for hybridization.Double stranded target DNA and probe DNA were denatured for 20 min at 80 • C and hybridization r eactions wer e incubated for 16 h at 46 • C. Following hybridization, cells were pelleted and washed with pre-warmed 48 • C washing buffer (46 mM NaCl, 20 mM Tris-HCl pH 7.4, 5 mM EDTA pH 8.0 and 0.01% (w/v) SDS), pelleted again and incubated at 48 • C for 20 min in pre-warmed washing buffer.Finally, cells w ere centrifuged, w ashed one time with PBS and resuspended in 50-100 μL of PBS.
For the adjustment of the protocol to target the Gr am-positiv e S. coelicolor , the following modifications were made.PBS-washed cells wer e tr eated with 0.5 mg/mL l ysozyme in 0.1 M Tris-HCl pH 7.4 for 30 min at 37 • C and washed twice with PBS.Lysozyme treated cell suspensions were not subjected to sonication for resuspension of the pellet.Denaturation was performed at 75 • C for 20 min, and the hybridization buffer contained 50% (v/v) formamide, whereas the washing buffer NaCl concentration was adjusted to 18 mM.

Microscopy analysis
Hybridized cell suspensions were stained with 2 μg/mL (final concentration) DAPI, incubated for 10 min in the dark, after which 3 μL of cell suspension was pipetted on top of a coverslip, mixed with 3 μL Vectashield mounting medium and cov er ed by a 5 ×5 ×1 (length x width x height) mm a gar ose (1% w/v in milliQ water) pad.Mounted samples were observed with an inverted epifluor escence Nik on Ti2 micr oscope equipped with a Prime BSI Scientific CMOS (Teledyne Photometrics) camera, using a 60 × oil immersion plan a poc hr omatic objectiv e. Alexa488 pr obe signal was recorded with a 470 nm lamp and the GFP-3035D filterset using an exposure time of 200 ms; Alexa594 probe signal was recorded with a 555 nm lamp and the mCherry-C filterset using an exposure time of 200 ms; DAPI signal was recorded with a 395 nm lamp and the DAPI-5060C filterset using an exposure time of 50 ms.Sets of images of the same sample type, including the non-sense negativ e contr ol pr obe labelled with Alexa594, wer e pr ocessed in Im-ageJ using the same procedure and settings.
To quantify the degree of association between DAPI and Alexa594 probe signal in in-solution SecMet-FISH, images of triplicate P. rubra samples labelled with DAPI and the Alexa594labelled SecMet pr obes wer e anal yzed in Ima geJ, using the 'Count cells' function.Here, all instances of DAPI and/or Alexa594 signal was summarized in a 2 ×2 contingency table according to cooccurrence.

Flow c ytometr y analysis
Flo w c ytometry w as emplo y ed for analysis of SecMet-FISH in solution on a MACSQuant VYB flow cytometer (Miltenyi Biotec, German y).Cytogr ams wer e r ecorded using SSC as trigger and cells were discriminated from noise by gating DAPI-stained cells in the V1 channel with a 405 nm laser and 450/50 nm filter.SecMet-FISH signals wer e r ecorded in the B1 channel with a 488 nm laser and 525/50 nm filter for Alexa488 labelled probes, the Y2 channel with a 561 nm laser and 615/20 nm filter for Alexa594 labelled probes and the Y3 channel with a 561 nm laser and 661/20 nm filter for Alexa647 labelled probes.Flow cytometry data of hybridized P. rubra samples were processed, analyzed and plotted in R software using the flowViz pac ka ge (Sarkar et al. 2008 ).First, raw .fcsdata files were imported and data points with a negativ e v alue for either the DAPI or A594 channel were removed (less than 1% of all the data).Then, data was transformed using the formula log(x + 1) and subsequently plotted.

Synthetic community experiment
Synthetic community (SynCom) samples were created by mixing late exponential phase cultures of P. mariniglutinosa , A. fischeri , V. anguillarum , E. coli and P. rubra in ratios of 1:1:1:1:1 (20% P. rubra abundance sample) and 12:12:12:12:1 (2% P. rubra abundance) based on OD600 values of the cultures.After mixing, a subsample of the SynCom samples was dir ectl y fixed in a 1% paraformaldehyde solution for one hour at room temper atur e, washed twice with PBS and stored at -20 • C in 50% EtOH-PBS until use for SecMet-FISH.The remainder of the sample was used for DNA extraction and subsequent PCR and probe synthesis.

Fluorescence activated cell sorting (FACS)
Cell sorting was ac hie v ed on a SONY MA900 cell sorter through a 70 μm sorting chip with sheath fluid (ClearSort TM , PBS) and using the semi-yield sorting mode .T he cell sorter was calibrated and optically aligned using polystyrene Automatic Setup Beads (Sony) prior to each sorting experiment.The backscatter channel was used as trigger and signals wer e r ecorded using a 405 nm laser and 450/50 nm filter for DAPI signal and a 561 nm laser and 617/30 nm filter for Alexa594 signal.Before sorting, a subsample of the target sample was recorded to set a gate for sorting: first, a parent gate was drawn to distinguish noise from cells using the DAPI channel, and a secondary gate was set in the DAPI vs Alexa594 plot to select the top 10% e v ents in the Alexa594 channel.This gate was subsequently used to sort cells from the sample.After sorting, cells were collected by centrifugation (10 000 × g for 2 min.)and resuspended in a total volume of 100 μL and stored at −80 • C until DNA extraction.Four samples were included in the analysis: two with the non-sense probe, and two with the P. rubra AD/KS probes.

DNA extraction and 16S rRNA gene PCR on sorted populations
DNA from FACS sorted cells was extracted using the microvolume DNA extraction method using the physical l ysis pr otocol (Br amucci et al. 2022 ).In brief, a cell suspension of 100 μL was lysed using a lysis buffer consisting of 0.17 M KOH and 0.013 M dithiothreitol, pH 12, for 10 min.at room temperature, follo w ed b y a freeze-thaw cycle at −80 • C and 5 min.incubation at 55 • C. Lysates wer e neutr alized using 2.5 M Tris-HCl buffer, pH 5.0 and DNA was ca ptur ed using Agencourt AMPur e XP ma gnetic beads (Bec kman Coulter, USA) in a 1:1.6 ratio with an extended DNA binding incubation time of 10 min.
As for biosynthetic domain PCRs, amplifications were done using HotStarTaq pol ymer ase (Qia gen) in 50 μL r eaction volumes.Detailed information on primer sequences, thermal cycling conditions, and r ea gent and template concentr ations ar e summarized in Table S1 .The PCR amplicons were run on a 1% a gar ose gel alongside a positive control ( P. rubra gDN A) and tw o negative controls (H 2 0 and PBS).

Statistical analysis
To determine if there was a significant association between the DAPI and Alexa594 AD/KS probe signals, a χ 2 -test was conducted on the data acquired from ImageJ (587 data points).Additionally, the percentage of associated signals was calculated.
To test for significant differences between the Alexa594 signal values of P. rubra incubated with AD/KS probes and non-sense probes in flow cytometry, a two-sample t-test was used on the log(x + 1) transformed flow cytometry data.Differences between the gating percentages of the SynCom sample groups were tested for statistical significance using a one-way analysis of variance (ANOVA) after a variance stabilizing log transformation of the data, follo w ed b y post-hoc Tuk e y's test for pairwise group comparisons.

Results
The genetic potential for production of non-ribosomal peptides and polyketides was chosen as the target feature for the development of SecMet-FISH (Fig. 1 ).We reasoned that the conserved and re petiti ve nature of the biosynthesis machinery of these compound classes would enable sufficient labelling of their genomic loci to allow for detection and quantification using fluorescence microscop y and flo w c ytometry.The Gr am-negativ e marine bacterium Pseudoalteromonas rubra S4059 (hereafter P. rubra ) was selected as model and target strain, as it is an excellent producer of secondary metabolites and contains multiple non-ribosomal peptide synthetase (NRPS) and polyketide synthase (PKS) BGCs in its genome (Vynne et al. 2011, Paulsen et al. 2019 ).To assess the applicability of using a mix of domain amplicons rather than an exact genomic locus as in direct-geneFISH, we first PCR amplified AD and KS domains from P. rubra using degenerate primers targeting conserved sequences of these biosynthetic domains .T he lengths of the resulting amplicons wer e ar ound 700 bp, as expected, and ther efor e similar in size to the pol ynucleotide pr obe used as proof-of-principle for dir ect-geneFISH (Barr er o-Canosa et al. 2017 ).Sequence analyses of the PCR products revealed that a small fraction of the AD and half of the KS domains present in the genome of P. rubra (detected with antiSMASH) was successfully amplified: four out of 70 AD and five out of 10 KS do-mains were retrieved from the sequencing reads .T he vast majority (99%) of the AD sequence reads mapped to one NRPS cluster (BGC#3, 4 mapped AD domains) and 99% of the KS sequencing reads mapped to three PKS/NRPS hybrid clusters (BGC#10, 3 ma pped KS domains; BGC#11, 1 ma pped KS domain and BGC#15, 1 mapped KS domain), demonstrating negligible amplification of off-targets (Fig. 2 ).Hence, at least nine specific loci, distributed over four BGCs, were identified as potential target hybridization sites for SecMet-FISH in P. rubra assuming stringent specificity.
To visualize said loci with fluorescence microscop y, w e generated SecMet-FISH probes by labelling the AD/KS amplicon mix with the Alexa488 dye and hybridized them in P. rubra cells immobilized on a membrane filter along with labelled non-sense probes as negative controls (Fig. 3 ).Using fluorescence microscopy, we observed the expected morphology of P. rubra from the DAPIstain and observed no Alexa488 signal when adding the nonsense probes (Fig. 3 B).Of m uc h mor e inter est, ho w e v er, wer e the clear and discrete Alexa488 fluorescence signals observed when the amplicon probes were hybridized (Fig. 3 A).These results demonstrate the ability to fluor escentl y label specific NRPS and PKS/NRPS hybrid BGCs, and more generally, that a mix of polynucleotide pr obes tar geting conserv ed domain sequences scattered throughout the genome can successfully be used as targets for SecMet-FISH.
Next, we explored the compatibility of SecMet-FISH with suspended cells in solution, instead of immobilized cells on a surface, with the aim of combining this method with flo w c ytometry.Protocols for in-solution FISH and direct geneFISH were combined (see Materials and Methods) using paraformaldehyde-fixed P. rubra cells.A fraction of the population sho w ed w eak, dotlike fluorescent signals ( Fig. S1 ).In an attempt to obtain stronger signals, the protocol was optimized, and additional fluor ophor es (Alexa594 and Alexa647) were tested.We found that a decreased DNA denaturing temper atur e (80 • C), a longer hybridization time (16 h) and an increased probe concentration (250 pg/ μL) resulted in increased signal intensities as observed by microscopy (Fig. 4 A and Fig. S1 ).Importantl y, SecMet-FISH signals wer e absent when the probes were applied to E. coli cells analogous to their lack of AD and KS domains ( Fig. S2 ).To further assess if SecMet-FISH allows for robust quantification of cells based their genetic potential to produce secondary metabolites , i.e .PKs and NRPs , we anal yzed the degr ee of association between DAPI signal and AD/KS probe signal on triplicate P. rubra samples ( Fig. S3 ).The distribution of 587 data points (signals from DAPI and Alexa594 channels) sho w ed that on r ar e occasions, pr obe signal was observed dissociated from DAPI signal (2.9% of Alexa594 signals), whic h likel y r eflects hybridization to extracellular DNA.Moreover 450 out of 573 DAPI signals were associated with at least one Alexa594 signal, suggesting a labelling efficiency of 79% and a significant association between the two signals ( χ 2 , P < 0.001).
Furthermore, cell populations hybridized with the AD/KS pr obes wer e distinguishable fr om populations hybridized with the non-sense control probes by flow cytometry (t-test, P < 0.001; Fig. 4 B, D), corr obor ating that the a ppr oac h enabled specific binding to target BGCs in solution.Fluor ophor especific gates, designed such that only 1% of the cell population in the control sample would be ca ptur ed in the gate, r esulted in positiv e labelling percenta ges of 18, 47 and 92 for the Alexa 647, 488 and 594-labelled probes, r espectiv el y.
In order to assess if SecMet-FISH could enable identification, en umeration, and se paration of target cells from a mixed population, we created two distinct SynCom samples, each contain-  ing five bacterial strains including P. rubra at abundances of 2% and 20%, r espectiv el y.SynCom subsamples wer e used as starting material for DNA extraction, PCR amplification of AD and KS domains and probe synthesis using the Alexa594 fluorophore.Subsequently, SynCom samples were subjected to SecMet-FISH in solution, and hybridized subsamples wer e anal yzed b y flo w c ytometry.Flo w c ytometry anal ysis r e v ealed that hybridization with the AD/KS pr obes r esulted in larger percentages of cells captured in the high Alexa594 signal gate compared to hybridization with the negativ e contr ol pr obes (Tuk e y's test, 2% P. rubra SynCom: P = 0.043; 20% P. rubra Syncom: p = 0.011; Fig. 5 ), which indicates that the targeted labelling of AD/KS domains in the genomes allow for the successful discrimination between community members based on BGC content.We did, ho w e v er, not observ e a clear difference in positiv el y gated cell fr actions between the 20% and 2% P. rubra SynCom samples labelled with AD/KS probes (Tuk e y's test, P = 0.97).
To determine if DNA from labelled, and hence paraformaldehyde treated, and sorted cells where of sufficient quality for downstr eam anal ysis, we sorted cells out using FACS, ex- tracted DNA and performed 16S rRNA gene PCRs.We observ ed str ong bands for all four samples, suggesting that paraformaldehyde fixation does not crosslink DNA to a degree that pr ohibits downstr eam anal yses of SecMet-FISH sorted cells ( Fig. S4 ).
FISH pr otocols can v ary consider abl y based on the tar get bacterial cells, most notably due to differences in the cell wall structure between Gram-positive and Gram-negative bacteria.Tow ar ds a broader application potential of SecMet-FISH, we applied and optimized our protocol targeting Gram-positive bacteria, for which we used the secondary metabolite producing model str ain Streptom yces coelicolor A3.Ada ptations for hybridization targeting S. coelicolor were inclusion of a l ysozyme tr eatment for cell wall permeabilization, adjustment of the denaturation temper atur e to 75 • C and adjustment of the formamide concentration in the hybridization buffer to 50% (v/v).Epifluorescence micr oscopy anal ysis of S. coelicolor cells hybridized with the specific AD/KS probe mix revealed multiple SecMet-FISH signals dispersed over the filamentous cells (Fig. 6 ).Similar looking signals were not observed for cells hybridized with the non-sense control pr obe, demonstr ating that the SecMet-FISH method was successfull y ada pted for specific labelling of Gr am-positiv e bacteria as well.

Discussion
The de v elopment of catal yzed r e porter de position (CARD)-geneFISH (Moraru et al. 2010 ) and more recently direct-geneFISH (Barr er o-Canosa et al. 2017 ) has opened up the possibility for detection and localization of microbial cells with specific functional gene profiles in environmental samples, as well as the coupling of this potential to specific microbial taxa.One limitation in the applicability of this method for high-throughput detection and quantification, e.g. using flow cytometry, is that hybridization of labelled probes to their target sequences occur inside immobilized cells rendering them out of r eac h for downstr eam anal yses beyond direct imaging.In addition, the limited signal from single-copy target sequences poses a significant challenge for the utility of contemporary geneFISH approaches with flow cytometry or FACS.In this w ork, w e capitalized on the de v elopments within in-solution Flow-FISH a ppr oac hes (Yilmaz et al. 2010, Haroon et al. 2013, Freen-van-Heeren 2021 ) as well as the conserved and re petiti ve nature of NRPS and PKS BGCs to increase the targeted loci, enabling detection and quantification of cells labelled based on their biosynthetic gene content using both fluorescence microscop y and flo w c ytometry.Combining SecMet-FISH with optical tr a pping or FACS may thus allow for e.g.tar geted meta genomics or single-cell genomics on micr obial comm unity members rich in BGCs, as well as retrieval of genetic material for subsequent cloning and expression (Grieb et al. 2020 ) in the future .Moreo ver, SecMet-FISH is applicable with immobilized cells and thus allows for the generation of data describing the spatial distribution of secondary metabolite producers in complex microbial communities.
As a proof-of-concept, we used the biosynthetic capabilities of P. rubra as a target functional genetic trait.Specifically, we successfully labelled at least four BGCs encoding the production of non-ribosomal peptide and polyketide/non-ribosomal peptide hybrid secondary metabolites, through the conserved AD and KS domains within the BGCs.Despite the use of degenerate primer pairs that have been used extensiv el y for pr ofiling of the biosynthetic potential of various environmental niches (Piel 2002, Ayuso-Sacido and Genilloud 2005, Charlop-Po w ers et al. 2014, 2016, Lemetr e et al. 2017, Bec h et al. 2020, Geers et al. 2022, 2023 ), sequence analysis of the amplicons sho w ed that only about 5% and 50% of AD and KS domains, r espectiv el y, wer e amplified fr om P. rubra .In accordance with these findings, it has been shown before, for example through in silico PCR, that degenerate primer sets targeting AD and KS domains often amplify less than 40% of the targets (Geers et al. 2022 ).Under the conserv ativ e assumption that the generated probes only bind to the specific target sequence (specificity of 100%), our a ppr oac h is unable to label and ca ptur e the full diversity of AD and KS domains, and the cells har- boring them.This also implies that community members harboring more BGCs than the nine targets amplified here , e .g. uncultured Acidobacteriota and Verrucomicrobiota (Crits-Christoph et al. 2018, 2022, Waschulin et al. 2022 ), could produce a stronger signal than obtained in the pr esent pr oof-of-concept setup.Importantly, the nine genetic target loci amplified from the genome of P. rubra pr ov ed to be sufficient for signal detection by epifluor escence micr oscop y and flo w c ytometry, although the labelling efficienc y w as only 79% accor ding to the signal association analysis.
A major distinction between conv entional dir ect-geneFISH and SecMet-FISH is that washing steps in-solution necessitate repetitive centrifugation and resuspension steps.We speculate that this, in combination with the high temper atur e incubations during denaturation of the target DNA is sufficient to make the Gramnegative P. rubra cells permeable to the pol ynucleotide pr obes.Indeed, we observed cell lysis when cells were treated with lysozyme and/or incubated at temper atur es abov e 80 • C. Reports on bacterial cell wall damage due to centrifugation support this hypothesis (Peterson et al. 2012 ).Optimizations of the protocol included an increase in probe concentration and hybridization time.The necessity for these adaptations might be explained by the effectiv e pr obe concentr ation: assuming equal amplification of n unique and equally distributed AD and KS domains, the ex-pected value of each unique probe will be 1/ n , which here implies that the effective probe concentration decreases as the complexity of the system incr eases.Additionall y, the abundance of each domain is likely to follow the same exponential la ws go verning the abundances of their hosts, resulting in extreme skewness of domains as well.As hybridization kinetics are dependent on (effective) probe concentration (Wetmur and Fresco 1991 ), increased hybridization time and probe concentrations are needed to compensate for a decreased effective probe concentration.We observed the best signal detection using the Alexa594 fluor ophor e in the proof-of-concept experiments, while Alexa647 labelled probes resulted in poor signal detection.This corroborates pr e vious observ ations (Barr er o-Canosa et al. 2017 ), and highlights the importance of fluor ophor e sensitivity in polynucleotide probe hybridizations.
In our SynCom experiment we attempted to use SecMet-FISH to label and differentiate the tar get str ain P. rubra from a mixed sample containing five bacterial strains.Flo w c ytometry analysis sho w ed significant differentiation betw een labelled and nonlabelled cells, showing that SecMet-FISH in combination with flow cytometry is an efficient way of differentiating bacterial cells based on their biosynthetic gene r epertoir e .Hence , the complementation of SecMet-FISH with FACS could facilitate selective extraction and enrichment of bacteria from a complex micro-bial community based on their genetic potential for secondary metabolite production.With this aim, a different fluorescent labelling strategy has recently been proposed for the enrichment of secondary metabolite pr oducers fr om tunicate and nudibr anc h microbiomes (Kim et al. 2021, Džunková et al. 2023 ).In this appr oac h, the carrier protein involved in non-ribosomal peptide and polyketide biosynthesis was directly labelled using a fluorescent analog molecule as probe (Kim et al. 2021 ).The main difference between this a ppr oac h and SecMet-FISH is the r equir ement for expression of the biosynthetic pathway in order to ac hie v e fluorescent labelling of the target cells.Although labelling of functional biosynthetic enzymes has the adv anta ge of gaining ecological insights into the in situ production of secondary metabolites, the a ppr oac h is likel y to miss a substantial fraction of community members with a broad secondary metabolite r epertoir e, as secondary metabolism is a tightly regulated process (Van Wezel andMcDo w all 2011 , Santamaria et al. 2022 ).

Conclusion
We report the development of SecMet-FISH, a method that enables the detection and distinct enumeration of bacteria carrying the genetic capacity to synthesize polyketide and non-ribosomal peptide metabolites.By (i) labelling multiple loci at once through synthesis of polynucleotide probes after degenerate PCR amplification of conserved sequences and (ii) hybridization within cells that are suspended in solution, SecMet-FISH offers the possibility for high-throughput discrimination and analysis of secondary metabolite producers by flow cytometry of labelled cells.SecMet-FISH can be adjusted to efficiently label BGCs within Gramnegativ e and Gr am-positiv e cells.Mor eov er, the a ppr oac h may be further extended to hybridization of other functional domains or genes or it may be combined with fluorescence assisted tr a pping or sorting for the specific acquisition of genetic material from secondary metabolite producers from environmental samples in the future.

Figure 2 .
Figure 2. Sequence read depth of the PCR amplified adenylation (AD, blue data points) and ketosynthase (KS, red data points) domains mapped to the genome of P. rubra S4059.The grey panel displays the genomic position of non-ribosomal peptide synthetase (NRPS) biosynthetic gene clusters (BGCs) (blue bars), NRPS/polyketide (PKS) hybrid BGCs (purple bars) and other BGCs (black bars).The blue panel provides a higher resolution display of the mapping of AD-amplicon reads to the NRPS genes of BGC #3.

Figure 4 .F igure 5 .
Figure 4. Labeling of suspended P. rubra cells SecMet-FISH in solution.Epifluorescence microscopy (A, A´) of P. rubra cells after hybridization with Alexa594 labelled AD/KS (A, A´) and non-sense control (C, C´) polynucleotide probes.Corresponding cytograms of samples hybridized with Alexa594 labelled AD/KS (B) and non-sense control (D) polynucleotide probes recorded with flow cytometry.Cells were counterstained with DAPI DNA stain.Dashed squares in A and C are displayed in close-up in A´and C´.Scale bar = 25 μm.

Figure 6 .
Figure 6.Epifluor escence micr oscopy of suspended S. coelicolor cells after hybridization with Alexa594 labelled AD/KS (A, A´) and non-sense control (B and B´) pol ynucleotide pr obes using in-solution SecMet-FISH.Cells were counterstained with DAPI DNA stain.Dashed squares in A and B are displayed in close-up in A´and B´.Scale bar = 25 μm.