Illuminating the dark metabolome of Pseudo-nitzschia-microbiome associations

Abstract

The exchange of metabolites mediates algal and bacterial interactions that maintain ecosystem function. Yet, while thousands of metabolites are produced, only a few molecules have been identified in these associations. Using the ubiquitous microalgae Pseudo-nitzschia sp., as a model, we employed an untargeted metabolomics strategy to assign structural characteristics to the metabolites that distinguished specific diatom-microbiome associations. We cultured five species of Pseudo-nitzschia, including two species that produced the toxin domoic acid, and examined their microbiomes and metabolomes. A total of 4826 molecular features were detected by tandem mass spectrometry. Only 229 of these could be annotated using available mass spectral libraries, but by applying new in silico annotation tools, characterization was expanded to 2710 features. The metabolomes of the Pseudo-nitzschia-microbiome associations were distinct and distinguished by structurally diverse nitrogen compounds, ranging from simple amines and amides to cyclic compounds such as imidazoles, pyrrolidines and lactams. By illuminating the dark metabolomes, this study expands our capacity to discover new chemical targets that facilitate microbial partnerships and uncovers the chemical diversity that underpins algae-bacteria interactions.

Fig. 1.

Workflows demonstrating (A) the experimental design and sampling procedure and (B) computational tools used to analyze and annotate the mass spectra. Five different Pseudo-nitzschia species cultures were grown in biological duplicates and harvested at the end of the exponential growth phase. Cell pellets were used for 16S rRNA sequencing and microscopy samples for enumeration of attached and free-living bacteria were sampled during the growth phase. Metabolomic samples were taken in technical duplicates (50mL) for the whole metabolome (acidified to pH2 and sonicated) and for the dissolved metabolome (filtered and acidified to pH2). All samples were solid phase extracted and analyzed by non-targeted high-resolution liquid chromatography tandem mass spectrometry (HR LC-MS/MS) in data dependent acquisition mode. (B) MZmine 2 was used for feature extraction and calculation of feature intensities (XIC). GNPS is used for feature-based molecular networking and searching the public and commercial standard libraries for spectral matches. ClassyFire assigns chemical classification to the features with spectral matches. For each feature SIRIUS ranks molecular formula (MF) candidates based on MS1 isotope patterns and MS/MS spectra analysis; ZODIAC re-scores MF candidates of individual features by considering all features in the dataset to increase the number of correct MF annotations. The machine learning-based tool CANOPUS predicts probabilities of over 2,497 compound classes of each feature.

Fig. 2.

PCoA plots showing the dissimilarities between (A) microbiomes and (B) metabolomes, color coded by the five Pseudo-nitzschia species. The relative abundance of 16S data was angular transformed and shows clear distinction between Pseudo-nitzschia species (PERMANOVA p < 0.001). MS1 features are TIC and chlorophyll a relativized and normalized (angular transformation). While filled circles correspond to the whole metabolome samples, empty circles represent dissolved metabolome samples. PC1 and PC2 reveal the separation between Pseudo-nitzschia species and dissolved versus whole metabolome respectively as indicated by the arrows. PERMANOVA analysis proved distinct Pseudo-nitzschia metabolomes as either dissolved metabolites only, the whole metabolomes only, or the dissolved and whole metabolome together, as well as dissolved vs. whole metabolome (all p < 0.001). (D) depicts the distribution of the 16S microbial community on the class and family level for each Pseudo-nitzschia species by taking the mean abundance of culture replicates.

Fig. 3.

Two-way cluster analysis showing the distribution of distinguishing features driving the separation between Pseudo-nitzschia species cultures. Area under the peak of MS1 features are TIC and chlorophyll a relativized and normalized (angular transformation). Z-score across dissolved (DM) and whole metabolome (WM) samples were calculated and are depicted in gray (see legend). Clusters were colored according to which Pseudo-nitzschia cultures the features were most abundant in. Features are labeled with putative compound names of their spectral library matches (library IDs, bold) and their chemical classifications (library IDs and analogs; using ClassyFire). Identification levels following the convention proposed by Sumner et al. (2007) are denoted by superscript following the compound name (see Supplemental Information 1.3.4 for definition). A more detailed table including adduct, measured precursor mass-to-charge ratio, the mass difference to the standard and all other gathered information about the given spectral match can be found in Table S3. Mirror plots of the spectral matches are shown in Fig. S11. Red and gray squares indicate if the annotated compound is nitrogen-containing or cyclic, respectively.

Fig. 4.

Two-way cluster analysis showing the distribution of mean compound class probabilities for abundant distinguishing features (bold) and relative abundance of 16S data (italic). For compound classes dissolved metabolite samples were used and the mean of technical duplicates was calculated. Values depicted in gray were calculated z-scores for biological duplicates of the five Pseudo-nitzschia cultures. Clusters were colored according to the Pseudo-nitzschia species where values were the highest. The red and gray squares indicate if compound classes are nitrogen-containing or cyclic by definition. The colored dots indicate bacteria class.

Fig. 5.

Molecular subnetworks of DA, pantothenic acid and two unknown molecular families. Every node represents a feature labeled with the mass-to-charge ratio (m/z); the distribution of relative abundance across Pseudo-nitzschia cultures is visualized by pie charts. Features related to each other (cosine > 0.7) are connected by edges and result in a subnetwork. Subnetwork (A) has two library ID matches (DA and methyl-DA) and one analog match (methyl-DA). The other nodes represent adducts such as [2M−H]⁺, covalently bound dimers and other DA derivatives. Compound class probabilities for the features are represented as bar graphs next to the nodes. Probabilities were high for azacyclic compounds (cyclic compounds with at least one nitrogen atom), pyrrolidine carboxylic acids and kainoids, which can all be confirmed by the known structure of DA. The pantothenic acid subnetwork (C) includes a library ID match with the adduct [M+H−H₂O], an analog match, which is likely to be pantothenic acid with an acetate addition, and two other related features. The probabilities for azacyclic and benzenoids are close to zero, but probabilities are high for alcohols and polyols and N-acyl amides, which is coherent with the known structure. Subnetwork (B) and (D) represent unknown features only. While no spectral matches are available, the combination of in-silico annotation tools allowed the annotation of MFs (Table S6) and compound classes. Here, probabilities are high for the compound classes azacyclic compounds and benzenoids for both, but we were able to identify different substructures of the unknown subnetwork.

Fig. 6.

Illuminating the dark metabolome of Pseudo-nitzschia-microbiome associations. Global network demonstrating the spatial distribution of (A) spectral matches (library IDs and analogs) and (B) in-silico annotation using the SIRIUS workflow (ZODIAC MFs, CANOPUS compound class predictions and CSI:FingerID structures). While GNPS library IDs are available for 4.7% of features, analog search increases this number to 12%. The spectral matches cover some subnetworks extensively, but most subnetworks have no annotations at all. Since subnetworks are structurally similar, we can propagate chemical classifications when spectral matches are present in a subnetwork, which means 25% of features can be classified (not shown). The SIRIUS workflow is able to annotate spectra absent from the spectral libraries and using the conservative cut-off of ZODIAC score >0.98, 56% of features were annotated. These features cover the global molecular network more comprehensively and if chemical classes are propagated within subnetworks, we have structure information about 75% of features in the dataset.