3.2. Relations between Microarray Signal, Cell Counts and Detection of Toxins
The insertion of a taxonomic hierarchy file in the GPR-Analyzer [20] gave us the advantage to distinguish false positives among the species-specific probes in the microarray analysis and exclude them prior to data interpretation. Briefly, for a species to be present, the entire taxonomic hierarchy leading to that species must also be present. The slopes of culture calibration curves of each species incorporated into the GPR-Analyzer allow for the transformation of microarray signals into cell abundances.

3.2.1. Pseudo-nitzschia and ASP Toxins
Pseudo-nitzschia was observed throughout the sampling period and is the only potentially toxic phytoplankton genus that formed a dominant bloom according to the cell counts. The microarray detected three of five Pseudo-nitzschia genus-level probes (PSN + some Frags_25_dT, PSN + FRAGS02-25new_dT and PsnGS02_25_dT) throughout the sampling period (Figure 3(a)). The other two generic-level probes (PsnGS01_25_dT, and PSN no pungens_25_dT) were excluded because the S/N ratio was not always above two. These two are not as strong as the other three probes, which are positioned at the top of the hierarchy file and thus do not cause the hierarchy test to fail. Weaker probes are always placed inside stronger probes to prevent such failure of true positives. Domoic acid (DA) was detected with ELISA [23] (Table 5) in both October samples (4A and 6A) but not in sample 3A (22.08.2011) where 400 cells·L−1 of P. multistriata were counted. This result suggests that the threshold for detecting DA with ELISA is somewhere between 400 and 1,700 cells·L−1 for the species P. multistriata. Furthermore, the Multi SPR gave no signal even though the last October sample had 40,600 cells·L−1. In general, it was found that the ELISA was more sensitive to lower amounts of toxin than the Multi SPR [23]. Because it is quite arduous to identify Pseudo-nitzschia multistriata to the species-level with light microscopy, and because some of our species-specific probes are still being optimized, we focused our comparison on P. multistriata (i.e., the “sigmoid” group) with two genus-level probes and three species-level probes on the array. The October bloom of 40,600 cells·L−1 of Pseudo-nitzschia multistriata matched the microarray with positive hits (S/N ratio above 2) of the two genus-level probes (PSN + some Frags_25_dT and PSN+FRAGS02-25new_dT) and the three species-level probes (PmulausD01_25_dT, PmulacalD02_25_dT, and PmulaD03_25_dT) (Figure 3(b)). The probe PcalfrauD04_25_dT (now interpreted to be a genus-level probe because it cross-reacted with all Pseudo-nitzschia spp. tested) showed consistent high signals for all Pseudo-nitzschia spp. in calibration curves (data not shown) and field samples.
Figure 3  Microarray signals of (a) the Pseudo-nitzschia spp. Genus-level probes (PSN + some Frags_25_dT, PSN + FRAGS02-new_dT and PsnGS02_25_dT) and (b) P. multistriata species-level probes (PmulausD01_25_dT, PmulacalD02_25_dT, PmulaD03_25_dT) normalized against Dunaliella tertiolecta (DunGS02_25_dT) for the field samples taken in Arcachon Bay, France and compared to cell counts. The graphs show only probes that yielded a signal above the detection limit (signal/noise ratio > 2), except for PmulaD03_25_dT, which is only in sample 6A above the S/N ratio. The sampling dates (24.07.2011, 08.08.2011, 22.08.2011, 04.10.201 and 20.10.2011) correspond to the sampling names: 1A, 2A, 3A, 4A and 6A. Cell counts are depicted in log10 on the secondary y-axis and as columns.
microarrays-02-00001-t005_Table 5 Table 5  Toxins measured by Multi SPR and ELISA during the sampling period in Arcachon Bay, France, adapted from [23].

3.2.2. Dinophysis and Prorocentrum and DSP Toxins
The non-toxic species Dinophysis tripos was counted in sample 1A (20 cells·L−1) and the toxic species D. caudata in sample 6A (30 cells·L−1, Table 4). No other Dinophysis species was identified by using light microscopy. Only the top genus-level probe in the hierarchy for Dinophysis (DphyGS03_25_dT) was detected with the microarray in sample 6A, but no species-specific probes were detected with the microarray in sample 6A. This result suggests that the microarray threshold for D. caudata species probe is above 30 cells. 
Cells from the potentially toxic genus Prorocentrum (group of P. minimum, balticum, and cordatum) were counted in samples 2A and 4A (both with 400 cells·L−1, Table 4). In addition, two planktonic usually considered harmless species, P. micans (sample 2A) and P. triestinum (sample 3A, 4A and 6A), were also identified by light microscopy with abundances ≤800 cells·L−1 (Table S1). No Prorocentrum species were counted in sample 1A. None of the planktonic clade-level probe for Prorocentrum ProroFBS02_25_dT and the species-specific probes for P. minimum (PminiD01_25_dT) and P. micans (PmicaD02_25_dT) of the microarray detected the presence of these taxa. It is likely that they require higher cell numbers to achieve a signal. With the third generation of the MIDTAL microarray new probes for Prorocentrum (two clade-level and six species-level probes) were tested, but without the poly dT_15 spacer region to raise the probes higher above the surface because they were still under testing for specificity. The new planktonic Prorocentrum probe ProroFPS01 was detected in samples 1A, 2A and 6A whereas the benthic Prorocentrum probe ProroFBS01 was detected in samples 4A and 6A (Figure 4(a)). New species-specific probes were made for the benthic species P. belizeanum, maculosum, rathymum and mexicanum. The probe PbeliS01 specific for P. belizeanum was detected with the microarray in samples 4A and 6A and the probe PrathD01 specific for P. rathymum and mexicanum was detected in sample 6A (Figure 4(a)). As for both samples, the higher probe ProroFBS01 (benthic Prorocentrum) was detected; the species-specific probes are not false positives and point out the limitation of microscopic cell counting. The specificity of theses Prorocentrum species has only been tested against a limited number of species and it is also likely that these probes are cross-reacting to another species present in the sample. P. rathymum is found in Malaysia and in the Mediterranean and P. mexicanum has a Caribbean distribution. More work is needed to clarify the taxon that is reacting with this probe. One way to achieve this is to use the probe as a FISH probe and sort the labeled cells or look at them in the microscope.
Okadaic acid was detected by ELISA in all samples except for the first (sample 1A), and the Multi SPR gave no signal at all. We presume that because Prorocentrum was more abundant than Dinophysis; its species is the source of this toxin.
Figure 4  (a) Normalized signal of Prorocentrum-level probes (ProroFPS01 and ProroFBS01) and the species-level probes PrathD01 and PbeliS01. (b) Normalized signal of the Alexandrium genus-level probe AlexGD01_25_dT.

3.2.3. Alexandrium and PSP Toxins
Two cells of the genus Alexandrium (ca. 20 cells·L−1) were counted only in sample 1A, whereas the microarray detected it throughout the sampling period (Figure 4(b)) with the highest signal at the end of October (sample 6A). Furthermore, PSP toxins were detected with ELISA in late August (sample 3A) and the remaining sampling period, as well as with the Multi SPR in sample 6A (Table 5, see [23] for more discussion on toxin found in these samples). If toxin probes are efficient and therefore PSP toxins are indeed present, there are two different ways to explain the absence of Alexandrium in cell counts: either Alexandrium cells have effectively been missed with the microscope, or there are other PSP-containing microorganisms in the water that are not identified. Neither A. ostenfeldii, A. minutum nor A. tamarense probes were detected by the microarray and their calibration curves for each specific probe have a detection limit of 200 cells [24]. Thus, we are unsure as to which species could be contributing to the PSP toxin profile. It could be Gymnodinium catenatum (see below) or another member of the genus Alexandrium. A. pseudogonyaulax could be a potentially missed Alexandrium species. There are no A. pseudogonyaulax-specific probes on the microarray. There are also many species that are not well investigated for toxin production. However, our data underlines the importance of including additional genus- and species-level probes for Alexandrium, in order to capture the full variability found in this genus. In any case, the detection of Alexandrium and its PSP toxins shows the advantage of the combination of the two methods (species and toxins) to detect harmful species, as well as to detect new invasive species as climate changes and tropical species move into temperate regions.

3.2.4. Heterosigma akashiwo
The heterokont Heterosigma akashiwo was identified by microscopic cell counts in sample 4A (600 cells·L−1) and 6A (400 cells·L−1). The microarray detected this taxon with the species-specific probe LSHaka054425b_dT in all samples except 3A. In addition, two more species-specific probes gave positive signals in sample 4A (LSHaka0268A25_dT and LSHaka0358A24_dT), and four in sample 6A (LSHaka0268A25_dT, LSHaka0544A25c_dT, LSHaka0329A25_dT, and SSHaka0200A25_dT). This species can be difficult to identify, especially once preserved in Lugol’s. Two species-specific probes were designed from the 18S region (SSHaka) and six more from the 28S region (LSHaka) for H. akashiwo (Table 2, [25]). Their calibration curves show the sensitivity of each probe and point out a low affinity with the H. akashiwo RNA. Some probes showed no sensitivity below 5 or even 25 ng of RNA, i.e., more than 700 cells are required to get a S/N ratio above two. This means that, in our case, we had around 230 cells·L−1 of H. akashiwo because not all probes were detected.

3.2.5. Species Unfound by Cell Counts but Identified with Microarray and Hierarchy File

Fish Killing Species
Lugol’s-fixed cells of Pseudochattonella are difficult to identify by light microscopy because the cell shape changes and the discharge of mucocysts gives them a warty appearance [26]. It is possible to distinguish the two sister species P. farcimen and P. verruculosa molecularly, because the two differ in several bases in the large ribosomal subunit [26]. In sample 6A, all genus-level probes of Pseudochattonella (PschGS01_25_dT, PschGS04_25_dT, PschGS05_25_dT) and the two species-level probes PfarD01_25_dT (P. farcimen) and PverD01_25_dT (P. verruculosa) were detected with a signal-to-noise above 2 (Figure 5(a)). The integrated calculation of cells L−1 in the GPR-Analyzer [20] revealed for Pseudochattonella farcimen 19,463 cells·L−1 and for Pseudochattonella verruculosa 48,428 cells·L−1, which is likely to be overestimated. Indeed, this species has only been identified in fjords and open waters of the North Sea, Skagerrak, and Kattegat, whose temperatures are below 10 °C [27]. During the summer–fall season, waters in Arcachon Bay are typically >25 °C [28]. Our results suggest perhaps another very closely related species as yet undetected could be in Arcachon Bay if the distribution of this species is exclusively in cold temperate waters. If this probe continues to show positive results, the probe could be used as a FISH probe to retrieve the cells giving the signal on the microarray for further investigations. Cells hybridized by the probe could be sorted by flow cytometry and investigated morphologically or molecularly. Once identified, the cells could later be brought into culture and their toxicity tested with bioassays.
Figure 5  (a) Normalized signal intensity of the genus-level probes (PschGS01_25_dT, PschGS04_25_dT, PschGS05_25_dT) of Pseudochattonella and the two species-level probes PfarD01_25_dT (Pseudochattonella farcimen) and PverD01_25_dT (Pseudochattonella verruculosa) for sample 6A (20.10.2011) only. (b) Normalized signal intensity of the class-level probes (PrymS03_25_dT, PrymS01_25_dT) and the clade-level probe (Clade01old_25_dT) of Prymnesium spp. (c) Normalized signal intensity of the genus-level probe of Karlodinium spp. (KargeD01_25_Dt).   No Prymnesiophyta were identified by cell counts, but the higher group probe for Prymnesiophyta (PrymS01_25_dT), the class level for Prymnesiophyceae (PrymS03_25_dT) and the clade-level probe for Prymnesium clade B1 (Clade01old_25_dT) were detected throughout the sampling period (Figure 5(b)). The second clade-level probe for Prymnesium clade B1 (Clade01new25_dT) was detected in samples 3A, 4A and 6A. Furthermore, the species-level probe for P. polylepis (CpolyS01_25_dT) was detected in sample 6A and in sample 4A the species-level probe for P. parvum (PparvD01_25_dT). This indicates the potential for a fish-killing event in Arcachon Bay under the appropriate conditions for growth. Although the Prymnesiophyta group is taken into account within the harmful phytoplankton monitoring program, the small size of this genus (<10 µm) as well as the smaller volume of water used for Utermöhl sedimentation and observation (100 mL maximum) than the volume of filtered seawater for RNA extraction, avoid any faithful microscopy identification and counting. The microarray can detect Prymnesium above 5 ng, which is equivalent to 3,800 cells for P. polylepis and 8,800 cells for P. parvum [29]. In our case (3 L filtered) it means 1,100 and 2,500 cells·L−1, respectively, which are high enough to be counted in a 10- or 100-mL sedimented subsample.
The genus Karlodinium (KargeD01_25_dT) was first detected in sample 3A and then with decreasing signals onwards (Figure 5(c)). Karlodinium veneficum is a high-biomass producer and the collapse of a bloom leads to the production of a surface scum that is visible as an oily, brownish discoloration of the water and kills fish and other gill-breathing animals [30]. However, no signals were detected for the six species-specific probes of K. veneficum present on the microarray. Based on their calibration curves (data not shown), the detection limit for four of the six probes is around 247 cells. The species-specific level probes are more sensitive than the genus-level probe. Therefore, we can exclude this species as a potential candidate being present in the bay. This is another example of how the microarray can detect potentially toxic species that are not counted or identified as being potentially toxic.

Azaspiracid Shellfish Poisoning (AZP) Toxins Producer
Azadinium spp. (AzaGS01_25_dT) was detected in sample 4A but only in two out of five spots on two different microarray slides. This may not be a genuine signal, but this species has only recently described [31] and it is also a relatively arduous species to identify based on light microscopy. Not all monitoring agencies are able to adjust their cell counts routinely to account for this toxic species. At least three more toxic species have been recently isolated and described [32,33,34].

Other PSP Toxins
One species-level probe out of four for Gymnodinium catenatum (LSGcat0270A24_dT) was detected in samples 1A, 4A and 6A. In samples 4A and 6A, the microarray also detected another species-level probe for G. catenatum (SSGcat0826A27_dT). The signals were not very high (S/N ratio between 2.2 and 4.9). G. catenatum is known to cause PSP and could therefore contribute, besides Alexandrium, to its detection via the ELISA and Multi SPR.