Each probe was tested under http://www.arb-silva.de/search/testprobe/against the curated SILVA “SSU Ref” or “LSU Parc” databases, for deposited sequences having up to three mismatches (Quast et al., 2013). The #114 database was released 2 months after #113 web release (Supplementary Table S1). The results of the ARB-SILVA “testprobe” analyses were filtered under Microsoft Office Excel^®, deleting double entries (repeated accession number indicating different matching sites with the same number of mismatches). To look for potential unwanted matches with other rRNA types, probes directed against the SSU were tested against the “LSU parc” database, and LSU probes against the “SSU ref” database. Potential unwanted matches with known mRNA sequences were searched in the NCBI “nucleotide collection nr/nt” (GenBank + EMBL + DDBJ + PDB + RefSeq) database using the BLASTN 2.2.27+program (Johnson et al., 2008)¹. Genomic sequences were not counted as matches, nor were mRNA sequences of organisms unexpected in marine plankton sample (e.g., mRNA sequence of the hen Gallus gallus). The matching list was therefore screened for sequences belonging to cultured strains, to test the probe specificity (Table 1, “Control strain”: (–)pCtrl).

Analytical and cell-culture grade biochemicals, Sartorius and Whatman Poly-Carbonate Track-Etched filtration membranes were purchased from Sigma–Aldrich (PCTE, Saint-Quentin Fallavier, France) and Dominique Dutscher (Brumat, France). Culture media for Eukarya were from the National Center for Marine Algae and Microbiota (NCMA, East Boothbay, Maine, United States) or the Roscoff Culture Collection (RCC, Roscoff, France). The natural seawater picoplanktonic sample (0.2–3.0 μm size-fraction) was collected on a 0.2 μm-pore 47mm PCTE filter after pre-filtration on 3.0 μm- and 10.0 μm-pore PCTE filters of 250 mL of water collected at the deep chlorophyll maximum 100 m deep from the open ocean South Pacific Ocean Time-series (SPOT) observatory near New Caledonia (october 2014). Triplicates 1/16th filter portions were hybridized with bacterial probes under different conditions described below. HRP-labeled oligonucleotide probes were from Thermo Fischer Scientific GmbH (Ulm, Germany), the TSA Plus Fluorescein Evaluation Kit from Perkin Elmer SAS (Courtaboeuf, France) and the Citifluor AF1 from Biovalley (Montpellier, France).

Cultured strains (Supplementary Table S2) were selected to belong either to the group targeted by the probe (positive probe control abbreviated as (+)pCtrl) or to an outgroup with the closest rRNA sequence having one to three mismatches with the probe (negative probe control abbreviated as (–)pCtrl, Table 1). (–)pCtrl strains were available within culture collections for all the tested probes except for Pela01, for which the closest outgroup strain available had four mismatches. Information on control strains used in this study is summarized in Supplementary Table S2. Haloarcheon msnc14(3) was grown on a liquid medium, while bacterial media were supplemented with 15 g L^-1 agar, and prepared in distilled water according to DMSZ instructions (Supporting Information). Cells were harvested in early stationary growth phase. Cultures were fixed for 15 min at room temperature with 1% buffered paraformaldehyde (PFA, w:vol) final concentration, and clumps of cells were disaggregated by vortexing when needed, prior to immobilization on 0.2 μm-pores 47 mm PCTE membranes, embedding in low-gelling point 0.4% agarose and storing in absolute ethanol at –80°C (detailed in Supporting Information).

Probe specificity on control strains was tested first using the formamide concentrations recently validated [if possible in TSA-FISH conditions according to Biegala et al. (2002) and Biegala and Raimbault (2008), Table 1 and Figure 3]. If the (–)pCtrl displayed a positive signal, formamide concentration (i.e., stringency) was increased by 5% increments until past the melting curve inflexion point, which gave a negative signal with the (–)pCtrl, and a positive signal with the (+)pCtrl (Table 1). Hybridization steps are briefly mentioned below, when detailed procedure is provided in Supplementary Materials. Cells were fixed, dehydrated and perforated, when needed, for the HRP-probe to pass through the cell wall, and hybridized with a HRP-coupled oligonucleotidic probe (Table 1). Probe hybridization was revealed by a TSA reaction using fluorescein-labeled tyramide (FITC, green fluorescence) and cellular DNA was DAPI-stained (blue fluorescence). The stringency of the hybridization conditions was optimized by adjusting the concentrations of formamide in the hybridization buffer (and of salt when necessary) in the washing buffer. Increasing stringency contributed to detach the oligonucleotide from the (–)pCtrl. We started with concentrations of 40 and 50% formamide for eukaryotic and prokaryotic probes, respectively.

Images were acquired at 100× magnification (40× for L. reticulosa) with an epifluorescence ECLIPSE 50i microscope (Nikon) equipped with excitation and emission dichroïc filters for DAPI and FITC detection, and a digital camera (see details in Supplementary Materials, QICAM 12-bit color cooled, QImaging). Pictures time exposure was an essential parameter to be defined in order to conclude on the success of a hybridization experiment. Time exposure was thus set up in order to reach saturation (i.e., the best picture, Figures 1, 3) on the (+)pCtrl and then kept constant for the (–)pCtrl and other strain controls abbreviated as (+)sCtrl and (–)sCtrl (Supplementary Figure S1). However, we permitted a decrease in time exposure on the (+)sCtrl when the emission fluorescent signal was over-saturating preventing the presentation of results (Supplementary Figure S1), and some (–)pCtrl pictures were over-exposed to illustrate slight positive unspecific signals (Figure 1, arrows). FITC exposure times ranged from 600 ms for most probes to 1.2 s for Gam42a, 1.5 s for Pela01, and 2 s for Roseo536R. Camera colors were preserved, and no image processing was done on FITC pictures, except for the abovementioned over-exposed (–)pCtrl pictures (Figure 1), where brightness and contrast were increased as detailed in the figure caption, to allow proper visualization. In contrast DAPI exposure and image processing were adapted and processed to allow best picture to be presented (Figure 1). DAPI exposure ranged from 200 ms to 2.8 s, and DAPI pictures were processed where needed using Microsoft Office Picture Manager^® to optimize the brightness, contrast and Gamma parameters (Figures 1, 3, and Supplementary Figure S1).

Differences in cell concentrations measured on natural seawater samples were analyzed by the non-parametric pairwise Mann–Whitney test on raw data (i.e., individual replicate values), due to the low number of replicates, using the Statistica 6 software. Data are reported as averages and standard deviation (SD) and statistical difference was accepted at p ≤ 0.05.

Among the six studied prokaryote probes and competitor, only EubII and EubIII remained fully specific, meaning they were 100% complementary to sequences affiliated to Eubacterial, targeted sequences (none sequences in outgroup column at zero mismatches, Supplementary Table S1). Eub338 alone targeted 93.3% of the SSU Ref #114 Silva database sequences affiliated to Bacteria with a perfect match or one terminal mismatch, missing 99.9% of the sequences classified as Verrucomicrobiales and Planctomycetales (Supplementary Figure S2). When EubII and EubIII target sequences (88% of Verrucomicrobiales and 49% of Planctomycetales affiliated sequences, respectively), were added to those of Eub338, the coverage of Eubacteria affiliated sequences increased up to 96%. The Eub338+EubII+EubIII mix was specific to the Eubacteria domain with 586338 hits, while only two outgroup hits affiliated to salt-water Archaea were targeted by Eub338. Other probe mix outgroup hits in the database harbored either one central mismatch with Eub338 and EubIII (“30c”, Supplementary Table S1), or two mismatches with EubII. Except for two of these outgroup sequences being affiliated to coastal air fungi, the other sequences were 18S rRNA sequences amplified from terrestrial samples (9 sequences), or affiliated to marine species harboring plastids with target 16S rRNA (chlorarachniophytes, 23%) or over 20 μm in size (70%). The majority of closest outgroup sequences had two to three mismatches with the probe mix: they were affiliated with Archaea (“28tc”, “211ttc”, Supplementary Table S1) and Eukarya (“745cc” including 46 sequences affiliated to uncultured marine Ciliophora).

When tested on pure cultured strains, EubII was already specific to (+)pCtrl with zero mismatch (Figure 1) and one mismatch (Verrucomicrobium spinosum DSM 4136, data not shown) at 50% formamide, and Eub338 was specific to strains with (+)pCtrl at 60% formamide (Table 1 and Figure 1). However, it was necessary to increase the concentration to 65% to obtain a complete negative signal with EubIII on the control cells displaying two to three mismatches. This result imposed the use of 65% formamide as specific conditions for the Eub338+EubII+EubIII mix). At 65% formamide, EubII detached from target sequences with one central mismatch (Verrucomicrobium spinosum DSM 4136, not shown). However, 99% of these sequences were targeted by EubIII. When tested on natural environment samples, these changes in formamide concentration (65%) induced significantly (Mann–Whitney p = 0.025, N = 15) higher counts than conditions previously described in the literature (55%, Figure 2).

The Gam42a probe matched with 84% of the 23S rRNA sequences affiliated to gamma-Proteobacteria in the LSU Parc #114 database (Supplementary Table S1). Out of the 6639 sequences matching perfectly with the probe 0.5% were outgroup hits, half of which being affiliated to alpha- and beta-Proteobacteria. Sequences with one central mismatch were at 12% classified as target gamma-Proteobacteria, but they also included many sequences affiliated to non-target alpha- and beta-Proteobacteria. The unlabelled Bet42a competitor matched with 91% of the outgroup hits with one central mismatch, but also prevented the detection of 8% of target gamma-proteobacteria harboring one central mismatch.

Concerning the alpha-proteobacteria Roseobacter clade, 91% of the identified member sequences were detected using both Ros537 and Roseo536R probes (Supplementary Table S1). These probes also targeted 2.9% of sequences classified as non-Roseobacter Bacteria, mostly (73%) other Rhodobacterales alpha-Proteobacteria, which may be found in marine samples. The RoseoC536R competitor masked sequences with one central mismatch, including 99.7% of the sequences identified as outgroup hits, but also 6% of the identified Roseobacter clade sequences. For both group-specific probes Gam42a and Roseo536R, specificity was reached without competitors (Figure 1). The original stringency of 50% formamide was confirmed for Gam42a, when it had to be increased up to 60% for Roseo536R (Figure 1). When tested in the natural environment significantly higher counts were obtained with Roseo536R used at 60% formamide compared to the former use of 50% formamide and competitors (Mann-Whitney p = 0.049, N = 6), and the addition of Bet42a competitor did not affect the proportion of cells detected by Gam42a (Mann–Whitney p = 0.154, N = 9, Figure 2). All the cells hybridized by the group-specific probes Gam42a and Roseo536R were also detected by the mix Eub338+EubII+EubIII in the conditions validated in the present study.

Among the five eukaryotic probes and the six mixes analyzed, the Euk1209, Pela01 probes and the mixes Chlo02+NChlo01, Euk1209+Chlo02, Euk1209+NChlo01, Euk1209+NChlo01+Chlo02 were fully specific (Supplementary Table S1). However, the Euk1209 probe covered the eukaryotic domain only partially (Supplementary Figure S3) as it missed 4223 sequences classified as eukaryotes with one central mismatch, including 277 sequences of marine planktonic species (Supplementary Table S1). The two outgroup hits with two central mismatches were affiliated to uncultured Bacteria and Archaea. Euk1209 also had three central mismatches with many outgroup aquatic bacterial sequences and Archaea including the cultivated Haloarchaeon msnc14, used as the closest (–)pCtrl strain (Figure 3). The non-chlorophyte NChlo01 probe complemented the Euk1209 targets with 3432 hits, including 531 sequences affiliated to planktonic organisms. The NChlo01 probe was, however, not specific to the non-chlorophyte divisions, matching perfectly with 111 chlorophytes, and it had one central mismatch with 45 sequences of mostly marine uncultured organisms identified as Archaea. Chlo01 probe, complementary to NChlo01 matched with zero mismatches with three Archaea outgroup sequences. When used as a mix with Euk1209+NChlo01, Chlo01 decreased the specificity of the mix by adding three outgroup hits with zero mismatches (Supplementary Table S1). Chlo01 was also far from being specific to the Chlorophyta division when used on its own as it showed perfect match with 2173 and 211 sequences affiliated to non-chlorophyte Alveolata and Heterokonta, respectively. In contrast, Chlo02 matched with 81% of the sequences identified as Chlorophyta (Supplementary Table S1) and matched with only seven sequences linked to small planktonic non-chlorophyte eukaryotes. Yet, the hybridization of many outgroup sequences from planktonic organisms having one central mismatch with Chlo02 may be an additional source of false positives. However, when added to the Euk1209+NChlo01 mix, the Chlo02 probe added 109 target hits to the mix without adding any outgroup hits to the mix contrary to Chlo01 probe.

The pelagophyte class includes the Pelagomonadales and Sarcinochrysidales, with 100 and 92% of the sequences affiliated to these “orders”, respectively, matching with the Pela01 probe. Only two sequences of uncultured putative pelagophytes presented one and two central mismatches with the probe (Supplementary Table S1). Outgroup sequences, with two and three mismatches, were identified at 88% as non-target pluricellular rhodophytes and opisthokonts unlikely to be found in marine plankton samples, and at 11% as large marine Radiolaria (>100 μm). The remaining 1% were affiliated to uncultured picoplanktonic freshwater Archaea, Amoebozoa, or sporocyte-forming eukaryotes. Finally, Pras04 detected almost all (95%) the sequences classified as Mamiellophyceae, but also three non-target sequences affiliated to marine organisms (dinoflagellate, stramenopile and chrysophyte, Supplementary Table S1). It had one central mismatch with 17 target hits, and with many outgroups hits which may be found in marine planktonic samples (e.g., 416 sequences of stramenopiles, 219 sequences affiliated to non-Mamiellophyceae chlorophytes). At 40% formamide, all eukaryotic probes were specific to their (+)pCtrl, giving negative hybridization signals with (–)pCtrl having one to four central mismatches (Figure 3 and Table 1). A slight brownish color was seen in the (–)pCtrl P. subviridis which show some autofluorescence of this photosynthetic strain.

The Eub338+EubII+EubIII probe mix has been developed to extend the specificity of Eub338 to Planctomycetales and Verrucomicrobiales, which are abundant in fresh- and sea-waters. The mix has been used previously for the analysis of environmental samples in CARD-FISH at 35 or 55% formamide in the hybridization buffer (Wilhartitz et al., 2007; Thiele et al., 2012). In the present study, the positive Eub338 TSA-FISH signal obtained with the eubacterium Verrucomicrobium spinosum raises concerns about the specificity of these hybridization conditions. This strain indeed harbors a 16S rRNA sequence having two mismatches with Eub338, similarly as 28 archaeal sequences (“28 tc”, Supplementary Table S1). To reduce the risk for false positives it is necessary to increase stringency by increasing formamide concentration in hybridization buffer from 55 to 60% when using Eub338 probe at one mismatch specificity. Although some archaeal cell wall deprived of lysozyme target site, may not allow HRP-labeled probes to penetrate the cell and target 16S rRNA (Pernthaler et al., 2002), still, some Euryarchaeota, which belong to Archaeal kingdom, require no perforation for TSA-FISH assay (Schönhuber et al., 1997).

A further increase of formamide concentration up to 65% was found necessary to prevent the hybridization of EubIII with uncultured marine Alveolata (Ciliophores) harboring two mismatches with the probe. When tested on Eub338 and EubII probes specificity 65% of fomamide did not change positive signals on probe target from 60% formamide condition. Therefore, we recommend using the Eub338+EubII+EubIII probe mix at a formamide concentration of 65%. In the natural environment, such as the one used in this study (Figure 2), using 65% formamide resulted in higher cell counts than when 55% was used. This result was unexpected, since when stringency conditions are increased they are usually expected to reduce the amount of false positive species targeted by the probe, thus reducing the overall cell concentration.

Together with the Gammaproteobacteria, the Alphaproteobacteria can account for approximately 30 to 60% of the bacterial diversity in the world ocean (e.g., Pommier et al., 2007). Their respective proportions may result from environmental physico-chemical parameters, and influence local biogeochemistry (Falkowski et al., 2008). Since Gam42a is the only probe currently available to target the largest Gammaproteobacteria diversity, we validated the hybridization conditions to avoid its binding to sequences with one central mismatch. We found no use for the competitor Bet42a since Gam42a with or without its competitor show the same amount of target and outgroup at zero mismatches (Supplementary Table S1) and Gam42a on its own was specific at 50% formamide to its target and did not bind H. muralis with one central mismatches (Figure 1). Accordingly, with these results Gam42a alone or with its competitor targeted comparable cell numbers in a natural sample (Figure 2). However, the 55% formamide concentration used by Manti et al. (2012) was found to be too stringent, decreasing the positive signal intensity (data not shown). The relative specificity of Gam42a has not declined since its design (Manz et al., 1992; Amann and Fuchs, 2008; present study), but the classification of Gammaproteobacteria is experiencing some changes with the development of multi-gene phylogeny (Williams and Kelly, 2013). We found Gam42a to match for instance with 19 sequences of 23S rRNA affiliated to Acidithiobacillales, which were very recently excluded from the class Gammaproteobacteria (Williams and Kelly, 2013). This miss-identification may be problematic in marine environmental studies, since the Acidithiobacillales include strains isolated from seawater (Kamimura et al., 2003). This might call for the design of a new 16S rRNA probe in future studies.

Among Alphaproteobacteria, the Roseobacter clade is ubiquitously distributed from coastal to open ocean and from the surface down to the seafloor (Buchan et al., 2005). These members of the Rhodobacteraceae family were found to be key-players in the global carbon and sulfur cycles (Wagner-Döbler and Biebl, 2006). However, they are not easily brought to culture, stressing the need for specific culture-independent methods to evaluate their importance (Eilers et al., 2001). Within five years the 16S rRNA sequences from Roseobacter clade (Brinkmeyer et al., 2000), increased by one order of magnitude from 124 to 1497 sequences (Buchan et al., 2005), and reached 5440 (SSU Ref #114), calling for a re-evaluation of the probes in silico specificity. In addition, many “unidentified” sequences affiliated to uncultured Rhodobacteraceae have been discovered, which potentially belong to the Roseobacter clade. The specific stringency for Roseo536R defined by our study (60% formamide, without competitor) cannot avoid the hybridization of 1–3.4% of false positives. Using Roseo536R at 60% formamide prevents binding to many non-Roseobacter Alphaproteobacteria with one central mismatch. It also allows to have a much higher fluorescence on it target than when using former conditions (50% formamide + competitor). This observation may directly explain the significantly higher cell counts obtained in the natural environment with the new conditions than with former ones (Figure 2). This result might indicate that at 50% formamide, unspecific binding of the competitor on Roseo536R targets could prevent binding of the probe, and we therefore suggest that increasing the stringency is a better solution than using the competitor. Unfortunately, we found that it would miss 16% one mismatch sequences of uncultured Roseobacter members. Thus, similarly as Gam42a the design of FISH probes should be refined in further studies by targeting more specific groups.

The specificity of Euk1209 had not been re-evaluated after its adaptation to TSA-FISH to detect dinoflagellates (Biegala et al., 2002), haptophytes and prasinophytes (Not et al., 2002; Biegala et al., 2003). The negative signal we obtained with an outgroup archaeal strain provides the first evidence that the former conditions were specific and may avoid the false positive detection of outgroup Archaea and Bacteria with two central mismatches or more. The latter should, however, not be a problem, since they can easily be distinguished from most eukaryotic cells showing a compact nucleus after DAPI staining, and since Eukarya do not require perforation for the penetration of an HRP-probe as do most Archaea and Bacteria. Unspecific TSA-FISH hybridization of prokaryotes should thus not happen when labeling eukaryotic cells with Euk1209. Nevertheless, since some Archaea do not need perforation as mentioned above (Schönhuber et al., 1997), and given that the marine archaeal diversity is still poorly known, it was important to make sure that Euk1209 was specific. This is important for example in the context of automatic detection, or of dual-hybridizations to detect eukaryotic intracellular Bacteria (Biegala et al., 2005). Previous studies stressed that Euk1209 needs to be complemented with other probes for a more exhaustive detection of the phytoplanktonic picoeukaryotic community (Not et al., 2004). We thus checked the current complementarity and specificity of the large spectrum probes NChlo01, Chlo01 and Chlo02 that had been designed against phytoplanktonic groups (Supplementary Table S1). We conclude that the Euk1209+NChlo01+Chlo01 mix (Not et al., 2004) should be replaced by the more specific Euk1209+NChlo01+Chlo02 mix, enlarging Euk1209 phytoplanktonic spectrum mostly with stramenopiles, alveolates, and chlorophytes. Recently, this mix has been used for the first time in a freshwater eukaryotic microbial community study (Mangot et al., 2013), but its specificity remained to be tested by in silico analyses and TSA-FISH with control cultures. Before our study, Chlo02 was known to be specific at 40% formamide to sequences with less than two mismatches (Biegala et al., 2003) which was sufficient for its use in the mix, but NChlo01 had only been validated in the presence of unlabelled Chlo01 differing by one mismatch (Biegala et al., 2003). Here we show that the same TSA-FISH stringency conditions avoid the hybridization of undesired outgroup sequences for each probe of the new mix.

The specific detection of the chlorophyte division by Chlo02 is hindered by outgroup hits consisting of matching sequences of <100 μm freshwater planktonic zygnematales, which can, however, easily be distinguished by their unique shape, during filter examination. We made sure for the first time that the TSA-FISH stringency conditions defined for Chlo02 in earlier studies (Figure 3 and Supplementary Table S1) selected against the hybridization of many sequences with one central mismatch affiliated to non-target Rhizaria, alveolates, heterokonts, and haptophytes. Narrowing down to the class level detection, we observed that the TSA-FISH conditions for the Mamiellophyceae chlorophyte subgroup-specific probe Pras04, had been previously optimized against strains with rRNA sequences having two mismatches (Not et al., 2004). Our study shows that the same conditions allowed selecting against non-target non-Mamiellophyceae chlorophytes and stramenopiles with one central mismatch detected by the in silico analysis, but prevented the hybridization of 17 target sequences. Finally, the specificity of the Pela01 Pelagophyceae class-directed probe, previously validated against strains with more than five mismatches (Not et al., 2002; Biegala et al., 2003), could not be tested against outgroup hits with three central mismatches since they consisted in uncultured organisms.

We propose a schematic overview of the protocol (Figure 4) that was followed for each re-evaluation of general and group-specific probes in TSA-FISH assays. The specificity of published probes must be verified from time to time as nucleotide databases are growing exponentially which may comfort or refute their use in the natural environment:

The probe is then ready to be used on environmental samples.