An oxybenzone exposition study on gilt-head bream


This vignette contains a case study of the effects of environmental contamination on gilt-head bream (Sparus aurata) [@ziarrusta2018bream]. Fish were exposed over 14 days to oxybenzone and changes were sought in their brain, liver and plasma using untargeted metabolomics. Samples were processed using Ultra-performance liquid chromatography mass-spectrometry (UHPLC-qOrbitrap MS) in positive and negative modes with both C18 and HILIC separation.

The mortality of exposed fish was not altered, as well as the brain-related metabolites. However, liver and plasma showed perturbations, proving that adverse effects beyond the well-studied hormonal activity were present.

The enrichment procedure implemented in FELLA [@picart2017null] was used in the study for a deeper understanding of the dysregulated metabolites in both tissues.

Building the database

At the time of publication, the KEGG database [@kanehisa2016kegg] --upon which FELLA is based-- did not have pathway annotations for the Sparus aurata organism. It is common, however, to use the zebrafish (Danio rerio) pathways as a good approximation. KEGG provides pathway annotations for it under the organismal code dre, which will be used to build the FELLA.DATA object.



# Filter the dre01100 overview pathway, as in the article
graph <- buildGraphFromKEGGREST(
    organism = "dre", 
    filter.path = c("01100"))

tmpdir <- paste0(tempdir(), "/my_database")
# Make sure the database does not exist from a former vignette build
# Otherwise the vignette will rise an error 
# because FELLA will not overwrite an existing database
unlink(tmpdir, recursive = TRUE)  
    keggdata.graph = graph, 
    databaseDir = tmpdir, 
    internalDir = FALSE, 
    matrices = "none", 
    normality = "diffusion", 
    niter = 100)

We load the FELLA.DATA object to run both analyses: <- loadKEGGdata(
    databaseDir = tmpdir, 
    internalDir = FALSE, 
    loadMatrix = "none"

Given the 11-month temporal gap between the study and this vignette, small changes to the amount of nodes in each category are expected (see section 2.4 Data handling and statistical analyses from the study). Please see the Note on reproducibility to understand why.

Note on reproducibility {#notereproducibility}

We want to emphasise that each time this vignette is built, FELLA constructs its FELLA.DATA object using the most recent version of the KEGG database. KEGG is frequently updated and therefore small changes can take place in the knowledge graph between different releases. The discussion on our findings was written at the date specified in the vignette header and using the KEGG release in the Reproducibility section.

Enrichment analysis on liver tissue

Defining the input and running the enrichment {#defenrichliver}

Table 1 from the main body in [@ziarrusta2018bream] contains 5 KEGG identifiers associated to metabolic changes in liver tissue and 12 in plasma. Our first enrichment analysis with FELLA will be based on the liver-derived metabolites. Also note that we use the faster approx = "normality" approach, whereas the original article uses approx = "simulation" with niter = 15000 This is not only intended to keep the bulding time of this vignette as low as possible, but also to demonstrate that the findings using both statistical approaches are consistent.

cpd.liver <- c(

analysis.liver <- enrich(
    compounds = cpd.liver, 
    data =, 
    method = "diffusion", 
    approx = "normality")

All the metabolites are successfully mapped:

analysis.liver %>% 
    getInput %>% 
    getName(data =

Below is a plot of the reported sub-network using the default parameters. The five metabolites are present and lie within the same connected component.

    method = "diffusion", 
    data =, 
    nlimit = 250,  
    plotLegend = FALSE)

We will examine the igraph object with the reported sub-network and some of its reported entities in tabular format:

g.liver <-  generateResultsGraph(
    object = analysis.liver, 
    data =, 
    method = "diffusion")

tab.liver <- generateResultsTable(
    object = analysis.liver, 
    data =, 
    method = "diffusion")

The reported sub-network contains around 100 nodes and can be manually inquired:


Examining the pathways

Figure 2 from the original study frames the five metabolites in the input around Phenylalanine metabolism. We can verify that FELLA finds such pathway and two closely related suggestions: Tyrosine metabolism and Phenylalanine, tyrosine and tryptophan biosynthesis.

path.fig2 <- "dre00360" # Phenylalanine metabolism
path.fig2 %in% V(g.liver)$name

These are the reported pathways:

tab.liver[tab.liver$Entry.type == "pathway", ]

Examining the metabolites {#exammetplasma}

Figure 2 also gathers two types of metabolites: metabolites in the input (inside shaded frames) and other contextual metabolites (no frames) that link the input metabolites.

First of all, we can check that all the input metabolites appear in the suggested sub-network. While it's expected that most of the input metabolites appear as relevant, it is an important property of our method, in order to elaborate a sensible biological justification of the experimental differences.

cpd.liver %in% V(g.liver)$name

On the other hand, one of the two contextual metabolites is also suggested by FELLA, proving its usefulness to fill the gaps between the input metabolites.

cpd.fig2 <- c(
    "C00079", # Phenylalanine
    "C00082"  # Tyrosine
cpd.fig2 %in% V(g.liver)$name

Enrichment analysis on plasma

Defining the input and running the enrichment

As shown in section Defining the input and running the enrichment, 12 KEGG identifiers (one ID is repeated) are related to the experimental changes observed in plasma, which are the starting point of the enrichment:

cpd.plasma <- c(
) %>% unique

analysis.plasma <- enrich(
    compounds = cpd.plasma, 
    data =, 
    method = "diffusion", 
    approx = "normality")

The totality of the 11 unique metabolites map to the FELLA.DATA object:

analysis.plasma %>% 
    getInput %>% 
    getName(data =

Again, the reported sub-network consists of a large connected component encompassing most input metabolites:

    method = "diffusion", 
    data =, 
    nlimit = 250,  
    plotLegend = FALSE)

We will export the results as a network and as a table:

g.plasma <-  generateResultsGraph(
    object = analysis.plasma, 
    data =, 
    method = "diffusion")

tab.plasma <- generateResultsTable(
    object = analysis.plasma, 
    data =, 
    method = "diffusion")

The reported sub-network is a bit larger than the one from liver, containing roughly 120 nodes:


Examining the pathways

Figure 3 from the original study is a holistic view of the affected metabolites found in plasma, based on literature and on an analysis with FELLA. The 11 metabolites are depicted within their core metabolic pathways. We will check whether FELLA is able to highlight them, by first showing the reported metabolic pathways:

tab.plasma[tab.plasma$Entry.type == "pathway", ]

And then comparing against the ones in Figure 3:

path.fig3 <- c(
    "dre00591", # Linoleic acid metabolism
    "dre01040", # Biosynthesis of unsaturated fatty acids
    "dre00592", # alpha-Linolenic acid metabolism
    "dre00564", # Glycerophospholipid metabolism
    "dre00480", # Glutathione metabolism
    "dre00260"  # Glycine, serine and threonine metabolism
path.fig3 %in% V(g.plasma)$name

All of them but Glutathione metabolism are recovered, showing how FELLA can help gaining perspective on the input metabolites.

Examining the metabolites

As in the analogous section for liver, we will quantify how many input metabolites, drawn within a shaded frame in Figure 3, are reported in the sub-network:

cpd.plasma %in% V(g.plasma)$name

From the 11 highlighted metabolites, only one is not reported by FELLA: 5-Oxo-L-proline.

Conversely, two out of the three contextual metabolites from the same figure are reported:

cpd.fig3 <- c(
    "C01595", # Linoleic acid
    "C00157", # Phosphatidylcholine
    "C00037"  # Glycine
cpd.fig3 %in% V(g.plasma)$name

As Figure 3 shows, the addition of linoleic acid and phosphatidylcholine, backed up by FELLA, helps connecting almost all the metabolites found in blood.

FELLA misses glycine and, in fact, stays consistent with the pathway (Glutathione metabolism) and the input metabolite (5-Oxo-L-proline) that it left out from Figure 3. The fact that FELLA does not suggest such pathway seems to happen at several molecular levels and therefore none of its metabolites are pinpointed.

Even if the glutathione pathway was not reported, FELLA can greatly ease the creation of elaborated contextual figures, such as Figure 3, by suggesting the intermediate metabolites and the metabolic pathways that link the input compounds.


In this vignette, we apply FELLA to an untargeted metabolic study of gilt-head bream exposed to an environmental contaminat (oxybenzome). This study is an example of how FELLA can be useful for (1) organisms not limited to Homo sapiens, and (2) conditions not limited to a specific disease.

On one hand, FELLA helps creating complex contextual interpretations of the data, such as the comprehensive Figure 3 from the original article [@ziarrusta2018bream]. This material would be challenging to build through regular over-representation analysis of the input metabolites. On the other hand, metabolites and pathways suggested by FELLA were also mentioned in the literature and supported the main findings in the study. In particular, it helped identify key processes such as phenylalanine metabolism, alpha-linoleic acid metabolism and serine metabolism, which ultimately pointed to alterations in oxidative stress.


This is the result of running sessionInfo()


KEGG version:


Date of generation:


Image of the workspace (for submission):

tempfile(pattern = "vignette_dre_", fileext = ".RData") %T>% 
    message("Saving workspace to ", .) %>% 
    save.image(compress = "xz")

References {#references}

Try the FELLA package in your browser

Any scripts or data that you put into this service are public.

FELLA documentation built on Nov. 8, 2020, 6:57 p.m.