It should provide a nice head-start to know how the ggplot package of R works, or alternatively, a nice click and interactive interface to make nice plots.

In the last few weeks, I have been working on my first public Shiny app. I have been working a bit with R lately and was interested in trying its internet interface. The objective was to decress the entry barrier to R and ggplot to some of my colleagues while keeping it interesting, so I decided to create an app that showed side-by-side a plot with the important code sections relevant to make it plot possible.

After a long debate about it, I decided to tweet and open-source it. The app is now uploaded here, and the I published the source code on Github here.

The app was well-received, even by the publishers of the original idea, and the overall feedback has been positive, so I am sure that updates and new apps are coming in the near future :)

Please let me know if you use it, if you find it useful, if you find something is missing, or any other kind of feedback.

This is an excerpt of the talk I gave last week on the Leiden University IBL "In The Spotlight" Biology Seminars. Zebrafish is emerging as a suitable research animal for modelling human diseases, but what is the translational potential of its research to humans?

Introduction

In 2011, I started my PhD working with zebrafish to study innate immunity and inflammation at the Immunology and Genomics group of the Institute of Marine Research in Vigo, Spain. While I was there, a paper published in PNAS ("Genomic responses in mouse models poorly mimic human inflammatory diseases", by Seok J et al.; link) had huge media coverage and went viral through news oulets over the world. The article criticized the lack of correlation between human and mouse gene regulation after an acute inflammatory stimuli, and it sparked a nice discussion of the appropriateness of animal models in research (a direct reply: "Genomic responses in mouse models greatly mimic human inflammatory diseases", by Takao K & Miyakawa T; link).

My opinion on the matter is complex and may warrant its own post in the future, but on this talk I wanted to approach a question that directly affected my PhD: If mice, the cornerstone of biomedical research are being challenged as a suitable model for human inflammatory diseases... what the hell are we even trying to do with the zebrafish, which is even less evolutionary related? How dare we even propose that as a model?

If the use of murine models for studying human inflammation processes is questioned, how relevant can a more evolutionarily distant species such as zebrafish be for modelling human diseases? (Forn-Cuni G et al., 2017)

Turns out, I just finished publishing my first 1st author paper -a characterization in zebrafish of a central protein in the inflammatory response, the C3 gene of the complement system-, and it had a major impact in how I approached zebrafish research further on.

Immune-related Gene Duplication in Zebrafish

The C3 is a central protein in the complement system cascade. It interacts with pathogens, opsonizes them for easier phagocytosis and can lead to the downstream formation of pores in the pathogen membrane through the Membrane Attack Complex. It was already known that in zebrafish there were 3 independent c3 genes instead of one since a while ago, but when my supervisors turned their eye to them and searched on the genome, they found 8 instead. The duplication of immune-related genes is a story that keeps repeating when working with zebrafish.

So I started to wonder... Why are so many duplications of immune genes in zebrafish? Which is the origin of all these genes, and how do they relate to the ones in humans? Are they regulated similarly to humans? Do they even have the same functions?

Teleost fish experienced an additional round of whole genome duplication than tetrapods, the sometimes wrongly called the Fish Specific Genome Duplication (salmonids had fourth one). That explains why most of the genes we work on zebrafish that may be under strong evolutionary pressure (as the ones directly interacting with pathogens) are duplicated. It is so common that we have official nomenclature rules for naming these genes.

However, the origin of duplicated genes in zebrafish does not stop here. According to Lu et al, 2012 (link), zebrafish also had an increased capacity of creating and retaining tandem gene duplications through its lineage evolution. Furthermore, when they analysed these genes that were duplicated in tandem they found that they were enriched in immune functions. Consequently, it is not unusual to find multiple copies of immune-related genes in the zebrafish genome.

The problem arises when these gene copies diversify to get new specific functions, sometimes specific for different tissues or pathogens, sometimes they can evolve faster and gain new functions, and sometimes they can gain regulatory or inhibitory functions. Referencing the previous paper, two of the c3 genes that we characterized, c3b.1 and c3b.2, have opposite, regulatory functions than the typical c3 genes during inflammation, probably because of competitive inhibition or by being scavenger receptors, who knows.

Immune-related genes that interact with pathogens are often duplicated in zebrafish -because of the additional Whole Genome Duplication and the teleost tendency of duplicating genes in tandem-, and they have evolved to new regulations and functions.

While zebrafish may share 70% of genes with mammals, immune-related genes are vastly diversified and most of them don't have 1-to-1 orthologs. In that context, how do we compare the functions and regulation during inflammation of genes with no clear evolutionary conservation? And therefore, if most of the proteins interacting with pathogens, activating and involved in the inflammatory response are not evolutionary conserved, how good of a model the zebrafish can actually be for these processes?

Comparison of Gene Regulation During Acute Inflammation

To find out, we compared the transcriptomic response to an acute inflammatory stimulus between zebrafish and mammals... with a little twist. In most vertebrates, bacterial LPS is recognized by the innate immunity receptor complex TLR4/CD14/MD2. However, fish have been classically regarded as "LPS-resistant", mostly due to the fact that most species lack MD2, and the TLR4 ohnolog present in some of them is not evolutionarily conserved and does not recognize LPS. Currently, the LPS sensing mechanism in fish is unknown. Despite that, stimulating zebrafish with LPS produces an acute inflammatory response, and we wanted to characterize it.

As expected, this analysis was not void of technical (and philosophical) difficulties. How can we compare the zebrafish gene regulation to mice or humans when only about 10% of the genes regulated in the inflammatory response have homology without question? And if there are two paralog genes with different expression patterns, which one do we consider for the comparison? Bluntly, if we apply the same 1-to-1 correlation method that we use in murine models to compare genes between zebrafish and humans, we are going to have a bad time.

Moreover, this small fraction of genes that are conserved -which have correlation values from zebrafish to mammals similar than to mice to humans- are the typical already known inflammatory markers, so there is not much new information that we can learn from it.

1-to-1 gene correlation between zebrafish and humans is difficult to establish, and due to evolutionary pressure, most genes are not functional homologues. But the cellular systems and pathways themselves during the inflammatory response are strikingly well conserved.

But, on the other side, when comparing the enrichment of the whole transcriptomic response, the similarities are obvious. When sensing inflammatory stimuli, zebrafish activate the same response framework than us: immune response activators (nfkb, ap1 components, ...); immune response inhibitors; pro-apoptotic and anti-apoptotic genes; inflammatory mediators (as cytokines and chemokines); IFN-stimulated, antigen presentation, tissue invasion, autophagy-related genes, etc. The whole framework is the same, it's just that the specific tools are different and specialized.

Discussion

The zebrafish provides a vast tool set to study innate immunity and inflammation. But it is a relatively recent addition to the biomedical research community and we have tons of work to do in order to learn the specific nuances of its immune responses and interaction with pathogens. This is a crucial step, not only to better understand our model and its limitations, but also because it can lead to potential therapeutic breakthroughs on its own.

But, in the meantime, from my point of view, if we want to achieve translational impact of our research to the human inflammatory disease and the clinic, we should forget about specific genes, which function may -or may not- be conserved and may be better suited to study in other models, and stick to what our model does well: reproduce whole pathways, systems, and cellular dynamics. In a way, move away from specific significant differentially expressed genes into other, more systems biology oriented, approaches.

Recently I had to annotate a rnaseq analysis that was only mapped to Ensembl IDs for an enrichment analysis in GSEA. That meant getting the maximum number of zebrafish Ensembl gene IDs to its human homolog/ortholog (you know the drill with zebrafish) Gene Symbols. So I spent some time analysing different mapping ways from to lose the least amount of information possible while still preserving the variability of zebrafish duplicates.

Some options that I tried:

1. Use the GSEA-provided zebrafish.chip
2. Simply get all genes to caps and use only the direct matches between zebrafish and human genes
3. Get homologs through:
   3.1. Biodbnet
   3.2. The R AnnotationDBi package (org.Dr.eg.db in this case)
   3.3. biomaRt
4. A bunch of other crazy stuff

In the end, I found that using a two-step process would yield the best results for the time spent:

1. Map the Ensembl IDs to ZFIN IDs
2. Use the ZFIN database as a chip for the conversion

1. From zebrafish Ensembl Gene IDs to ZFIN IDs

Thankfully, the mapping from Ensembl IDs to ZFIN IDs is pretty easy and straightforward. Here are two R functions to get the job done through the biomaRt and org.Dr.eg.db packages:

# Function txdbAnnot.
# Input: a list of Ensembl Gene IDs.
# Return: an annotated dataframe.
txdbAnnot <- function(listOfIDS,
                      attributes = attributes,
                      keys = keys) {
  library("org.Dr.eg.db")
  # print("Annotating using org.Dr.eg.db")
  txdbAnnotDF <- select(org.Dr.eg.db, 
                      keys=listOfIDS, 
                      columns=attributes, 
                      keytype=keys,
                      multiVals="first")
  return(txdbAnnotDF)
}

bioMartConversion <- function(listOfIDS, 
                              attributes = attributes,
                              filters = filters) {
  library("biomaRt")
  # print("Annotating using BioMaRt")
  zfishMart <- useMart("ensembl", dataset="drerio_gene_ensembl")
  geneAnnot <- getBM(attributes = attributes, 
                     filters = filters,
                     values = listOfIDS, 
                     mart = zfishMart)
  return(geneAnnot)
}

# Function deleDuplicatesDataFrame.
# Input: a dataframe and a column with duplicates.
# Output: the dataframe without the duplicates.
deleteDuplicatesDataFrame <- function(df, col) {
  dup.idx <- which(duplicated(df[col]))
  return(df[-dup.idx,])
}

And then annotate using one of the methods:

# You can use the call to annotate with other IDs as
# attributes <- c("ENTREZID", "SYMBOL", "ZFIN", "GENENAME")
zfinTxdb <- txdbAnnot(listOfIDS, 
                      attributes = "ZFIN",
                      keys = "ENSEMBL")
# Clean up the duplicates
zfinTxdb <- deleteDuplicatesDataFrame(zfinTxdb, "ZFIN")

or

# You can use the call to annotate with other IDs as
# attributes <- c("ensembl_gene_id", "entrezgene", "external_gene_name", 
              "zfin_id_id", "description")
zfinBiomaRt <- bioMartConversion(listOfIDS, 
                                 attributes = c("ensembl_gene_id","zfin_id_id"),
                                 filters = "ensembl_gene_id")
# Clean up the duplicates
zfinBiomaRt <- deleteDuplicatesDataFrame(zfinBiomaRt, "ensembl_gene_id")

I personally prefer using AnnotationDBI because it's faster.

2. The ZFIN.chip for GSEA

ZFIN provides data dumps for most of its data, and one really useful table is the Human and Zebrafish Orthology table. You can find it at the downloads page.

Just create a mapping chip without duplicates using the columns ZFIN ID as Probe, Human gene Symbol as Gene Symbol, and the zebrafish gene name as Gene Title. That way the 1-to-many orthologs can be treated as different (technically, as different probes for the same gene) but we don't lose info or pick just one randomly.

That's it. I know that it's not perfect, but this is the way in which I am getting the most insightful results on my research.

What are your thoughts? Get in touch and discuss it!