Unity in defence: honeybee workers exhibit conserved molecular responses to diverse pathogens

Abstract

Background: Organisms typically face infection by diverse pathogens, and hosts are thought to have developed specific responses to each type of pathogen they encounter. The advent of transcriptomics now makes it possible to test this hypothesis and compare host gene expression responses to multiple pathogens at a genome-wide scale. Here, we performed a meta-analysis of multiple published and new transcriptomes using a newly developed bioinformatics approach that filters genes based on their expression profile across datasets. Thereby, we identified common and unique molecular responses of a model host species, the honey bee (Apis mellifera), to its major pathogens and parasites: the Microsporidia Nosema apis and Nosema ceranae, RNA viruses, and the ectoparasitic mite Varroa destructor, which transmits viruses.

Results: We identified a common suite of genes and conserved molecular pathways that respond to all investigated pathogens, a result that suggests a commonality in response mechanisms to diverse pathogens. We found that genes differentially expressed after infection exhibit a higher evolutionary rate than non-differentially expressed genes. Using our new bioinformatics approach, we unveiled additional pathogen-specific responses of honey bees; we found that apoptosis appeared to be an important response following microsporidian infection, while genes from the immune signalling pathways, Toll and Imd, were differentially expressed after Varroa/virus infection. Finally, we applied our bioinformatics approach and generated a gene co-expression network to identify highly connected (hub) genes that may represent important mediators and regulators of anti-pathogen responses.

Conclusions: Our meta-analysis generated a comprehensive overview of the host metabolic and other biological processes that mediate interactions between insects and their pathogens. We identified key host genes and pathways that respond to phylogenetically diverse pathogens, representing an important source for future functional studies as well as offering new routes to identify or generate pathogen resilient honey bee stocks. The statistical and bioinformatics approaches that were developed for this study are broadly applicable to synthesize information across transcriptomic datasets. These approaches will likely have utility in addressing a variety of biological questions.

Keywords: Apis mellifera; Co-expression network; DWV; IAPV; Immunity; Meta-analysis; Nosema; RNA virus; Transcriptomics; Varroa destructor.

Fig. 1

Heat maps illustrating the expression levels (relative ranks) of the 344 significantly regulated genes across the 19 transcriptome datasets. Genes are categorized as 56 up-regulated genes (top left), 109 down-regulated genes (bottom left), and 179 differentially regulated (up and down) genes (right). Orange shows increased expression and blue decreased expression after pathogen infection. Top classification is N for Nosema infection, N/V for Nosema and RNA virus co-infection, V for virus, and M for Varroa mite (‘Varroa plus virus’). Numbers at the bottom correspond to dataset numbers in Table 2. Each row represents the differential expression of the same gene across all 19 datasets. In each category, genes are ordered following the arithmetic means of their ranks displayed in the right column of the heat map. Note the presence of genes showing decreased expression in some datasets although found as statistically up-regulated across datasets, and vice-versa

Fig. 2

Comparison of the evolutionary rate between genes showing significant differential expression and genes without significant differential expression across the 19 datasets. Relative evolutionary rates on the Y-axis are quantified from pairwise alignments of the protein sequences, and represent the average of inter-species protein sequence identities normalized to the average identity of all inter-species orthologs from OrthoDB [33]. The vertical black lines along the median and mean values of each category represent the standard deviation (thick lines) and the 95% confidence intervals (thin lines). Horizontally, the width of each violin box represents the density of the data values, i.e. the distribution of the data along the y axes, for each category

Fig. 3

Gene co-expression network. a Main module of the gene co-expression network, representing 3,589 interconnected genes. Red nodes show genes significantly regulated across the 19 transcriptome datasets, and black nodes show non-significantly regulated genes. Square nodes show the most connected (hub) genes. Grey edges illustrate positive correlation between two gene expression profiles while blue edges show negative correlations. A file available at https://idata.idiv.de/DDM/Data/ShowData/35 provides the possibility of navigating within the network. b Scatter plot representing the total number of connections (x-axis) over the number of connections to significantly regulated genes across the 19 transcriptome datasets for the most (top 5%, N = 209) connected genes (i.e. hub genes). Red triangles show significantly regulated hub genes, while black dots show non-significantly regulated hub genes. Two hub genes with high connectivity to significantly regulated genes are shown: a kynurenine/alpha-aminoadipate aminotransferase (LOC724239), and a L-lactate dehydrogenase (LOC411188). c Main module from the co-expression network of the immune genes of the honey bee. Coloured nodes represent immune genes from the Toll (purple), JAK/STAT (brown), apoptosis (green), RNAi (blue) and Imd (pink) pathways (see immune genes list in the Additional file 2: Table ST13). Oval nodes show genes with low connectivity, squares show genes with high connectivity (hub genes, with at least 34 connections). Genes significantly regulated across the 19 transcriptome datasets have a red outline. Black edges represent positive co-expression and blue edges are negative co-expression. In insets, the expression profiles across the 19 transcriptome datasets (black lines) of the four immune hub genes (i.e. highly connected immune genes), accompanied by expression profiles of genes with which they are connected. Orange profiles display similar profiles (positive connections, i.e. black lines in the network) and blue reflect opposite profiles (negative connections, i.e. blue lines in the network). The y-axis displays the relative ranks of differential expression level, from up-regulated (value towards 1) to down-regulated (value towards 0)

Fig. 4

Diagram of the canonical innate immune response of the honey bee. Gene names in colour-filled boxes show evidence of significant regulation after infection by Nosema (yellow), or infection by RNA viruses and/or infestation by Varroa mites (light blue) or all pathogens (grey). Orange lines surrounding a box show increased expression and blue surrounding lines indicate decreased expression after pathogen infection –mixed orange and blue lines show genes found differentially-regulated, either up- or down-regulated across the datasets. Notably, the AMP defensin-1 exhibited increased expression in most of the datasets, but a decreased expression in the abdominal tissues of honey bees infected by Nosema. Therefore, a mixed background and outline colour are displayed. Green surrounding lines show genes found non-significantly regulated in this analysis. Solid lines with arrows show gene interactions reported in the literature, and dotted arrows indicates new potential interactions inferred from our gene co-expression network analysis

Fig. 5

Methodological workflow of the directed rank product analysis (DiRank). This new method aims to identify genes with similar expression profile to a theoretical or observed profile of another gene. Gene expression values and profiles (geps) (shown in blue) and custom profile (cp) (shown in red), consisting of relative rank values, serve as input (yellow boxes). In rectangular matrices, gene expression values are reported in rows, while columns represent the transcriptome datasets. A custom profile can either be a user-defined profile or an existing gene expression profile. The directed rank product analysis aims to identify genes with a similar expression profile to the custom profile and to assign associated p-values. The custom profile is subtracted from each of the gene expression profiles and each difference (gep - cp) is transformed by 1 -| gep - cp|. Transformed gene expression values and corresponding profiles are shown in green in the grey box. These transformed gene expression values are then used as input data for a rank product analysis. As an example, the transformed gene expression values surrounded by an orange frame are ranked on top by the rank product analysis as the original gene expression profile was the most similar (before transformation) to the custom profile