Abstract
The bumblebee Bombus terrestris is commonly infected by a trypanosomatid gut parasite Crithidia bombi. This system shows a striking degree of genetic specificity where host genotypes are susceptible to different genotypes of parasite. To a degree, variation in host gene expression underlies these differences, however, the effects of standing genetic variation has not yet been explored. Here we report on an extensive experiment where workers of twenty colonies of B. terrestris were each infected by one of twenty strains of C. bombi. To elucidate the host’s genetic bases of susceptibility to infection (measured as infection intensity), we used a low-coverage (~2 x) genome-wide association study (GWAS), based on angsd, and a standard high-coverage (~15x) GWAS (with a reduced set from a 8 x 8 interaction matrix, selected from the full set of twenty). The results from the low-coverage approach remained ambiguous. The high-coverage approach suggested potentially relevant genetic variation in cell surface and adhesion processes. In particular, mucin, a surface mucoglycoprotein, potentially affecting parasite binding to the host gut epithelia, emerged as a candidate. Sequencing the gut microbial community of the same bees showed that the abundance of bacterial taxa, such as Gilliamella, Snodgrassella, or Lactobacillus, differed between ’susceptible’ and ’resistant’ microbiota, in line with earlier studies. Our study suggests that the constitutive microbiota and binding processes at the cell surface are candidates to affect infection intensity after the first response (captured by gene expression) has run its course. We also note that a low-coverage approach may not be powerful enough to analyse such complex traits. Furthermore, testing large interactions matrices (as with the full 20 x 20 combinations) for the effect of interaction terms on infection intensity seems to blur the specific host x parasite interaction effects, likely because the outcome of an infection is a highly non-linear process dominated by variation in individually different pathways of host defence (immune) responses.
Introduction
Genetics and genetic variation affect infection outcomes in almost any study of host-parasite systems so far. For example, genetic sequences encode epitopes on the surface of parasites, which are recognized by receptor molecules that are encoded by host genes [1]. Extensive databases, such as the Immune Epitope Database (IEDB) [2], archive the enormous diversity of parasitic sequences; vice versa, host genetic variation in putative defence genes, e.g., receptor molecules, is likewise accessible in data repositories (e.g., the pattern-recognition data base, PRDDB [3] and various others, accessible via Immunological Databases and Tools). For non-model organisms, the knowledge is far less sophisticated than can be found in these databases. In fact, while homology offers powerful insights into shared function, to understand the genetic determinants of host-parasite interactions requires directly querying these systems with associative approaches, such as with genome-wide association studies (GWAS).
The genetic basis for host resistance or tolerance [4] can be very different. For example, the resistance of the water flea, Daphnia magna, against the bacterial pathogen Pasteuria ramosa, depends on a single genetic locus (a QTL) that explains over 50% of observed variation. By contrast, its resistance against the microsporidian, Hamiltosporidium tvaerminensis, follows a quantitative genetic pattern where multiple loci with weak effects add up to the overall variation (explaining a total of 38%; [5–7]). Hence, even when genes are identified that encode important elements of the host defence, their contribution to observed variation in resistance at the population level can be quite different. Furthermore, the actual molecular and physiological mechanisms underlying these interactions often are complex and diverse. Accordingly, the answers from associative studies will not always be simple and robust, but inevitably reflect the complexity of the processes themselves. Nevertheless, the identification of genes associated with a resistance phenotype, such as infection intensity, is helpful as a start for understanding the mechanistic bases of host-parasite interactions.
The system studied here, bumblebees (Bombus terrestris), infected by the common trypanosomatid parasite, Crithidia bombi, has been studied for some time, and considerable knowledge on the ecology, evolution, or genetics has been accumulated. B. terrestris is a social insect with colonies headed by a single, and singly-mated queen, that reach sizes of a few dozen to perhaps one hundred workers. During the active season, the majority of workers are infected by one to several ’strains’ (genotypes) of this parasite [8, 9]. Earlier studies have provided evidence that, both, the host and the parasite genotype (the ’strain’) are important determinants of infection outcome. For example, the resulting infection intensity varies with the paternal genetic line even among sister workers living within the same colony [10]. Furthermore, when a cocktail of different strains is infected, these will be ’filtered’ out—according to bee and parasite genotype—allowing only a few strains to transmit further [11, 12]. The study of host gene expression upon exposure to C. bombi reveals that the host responds by activating genes known to be involved in defence (e.g., anti-microbial peptides, [13]), and host gene expression patterns carry the signature of the specific host-parasite interaction [14].
Here, we employ a genome-wide association study (GWAS) approach to elucidate the genomic basis for the observed host-parasite interaction in this system. For this purpose, we used a cross-infection (fully factorial) design with worker bees of 20 host colonies exposed to 20 different strains of the parasite, amounting to a total 1,200 bees in the experiment (Table 1). We use infection intensity (i.e., the number of parasite cells per bee) as our measure of host resistance and ask what genes or genetic regions associate with this variable. We report results from two different methods: a high-coverage genetic analysis using a selected subset in an 8 x 8 (host colony x parasite strain) matrix, and a low-coverage approach for the full 20 x 20 interaction matrix design. The latter approach was motivated by the development of new analytical tools that can handle many samples each having a low coverage of their genomic sequence [15]. This presents an attractive cost-effective alternative by using many rather than few samples, with the inevitable compromise in terms of sequence coverage. Here we take the opportunity to provide data and compare these two design philosophies.
Table 1. Experimental design.
| Strain | ||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Colony | A | B | C | D | E | F | G | H | I | J | K | L | M | N | O | P | Q | R | S | T |
| Code | 8.068 | 8.075 | 8.076 | 8.091 | 8.161 | 10.027 | 10.132 | 10.175 | 10.290 | 10.486 | 12.246 | 12.444 | 12.248 | 12.448 | 12.450 | 14.065 | 14.149 | 14.172 | 14.255 | 14.330 |
| 15 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 20 | 3 | 3 | 2 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 25 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 2 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 33 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 2 |
| 39 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 47 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 1 |
| 59 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 81 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 82 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 90 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 2 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 91 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 95 | 2 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 124 | 3 | 3 | 3 | 3 | 1 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 2 | 3 | 3 | 3 |
| 141 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 152 | 3 | 2 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 183 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 194 | 3 | 3 | 3 | 3 | 2 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 2 | 3 | 3 |
| 225 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 |
| 269 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 2 | 3 | 2 |
| 319 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 3 | 2 |
The cross-infection matrix shows the number of replicates (bees) that were exposed to infection for a given combination of Colony (rows; project code, D of colony: 15,…, 319) and Strain (columns; project code of parasite strain: A,…, T). Standard code of the parasite strains (second row) indicates the label used for the long-term studies in this system. In a few cases, only two instead of three replicates could be evaluated. Total sample size is N = 1,198 bees for the entire matrix (corresponding to the low-coverage study), and n = 192 bees for the high-coverage study (8x8 matrix), indicated by the shaded cells. For choice of the subset, see text.
Previous studies also have clarified that the bees’ microbiota varies among individuals and colonies and provides protection against infections by C. bombi to varying degrees [16]; it also adds to the specific interactions in this host-parasite system [17]. Further studies show that the structure and presence of certain bacteria in the microbiota, notably Gilliamella and Snodgrassella, affect the degree of protection [18–23]. We therefore sequenced the bacterial gut metagenome for each bee (its bacterial microbiota) in the high-coverage (8 x 8) matrix and ask what microbial differences predict infection outcome and how these microbial communities correlate with host genomic variation. We finally discuss methodological lessons from this work and the problems associated with a diffuse genetic architecture of host resistance and the relevance of the interaction subsets in terms of variable defence processes.
Material and methods
General methods
Spring queens of Bombus terrestris were collected in March and April 2014 in northern Switzerland and raised to colony in the lab following our standard protocols (Table S1 in S12 File). All queens were tested for the possible presence of Crithida infections, using qPCR with primers (CriRTF2, CriRTR2) amplifying a 56-bp fragment of the 18S rRNA gene, see [12]; only uninfected queens and their colony were included in the experiment. The qPCR-procedure also measured infection intensity (see below). The parasites used for the experimental infections were taken from our strain collection of C. bombi clones derived from single-cell isolates from bees sampled in the field, and as used before [24, 25] (Table S2 in S12 File). Here, these clones are called ’strains’. In practice, strains are defined by their multi-locus genotype based on polymorphic microsatellites at five loci (e.g. [11]).
Infection protocol and assessment of success
As routinely used and already described in earlier studies [11, 26], randomly selected workers (representing a snapshot of the colony resistance overall) were collected from the colonies and randomly assigned to treatment, i.e. exposure to one of the strains. For infection, the workers were starved for 2 h, before offering them a droplet of sugar water with the inoculum. For each colony x strain combination, five to six bees were exposed to the same inoculum for infection. Among those, we picked three at random to be used for the analyses. The infective dose was set at 10,000 cells per bee. The bees were sacrificed on day 7 post-exposure, the resulting infection intensity and body (wing) size measured. Both, infective dose and post-exposure time, were chosen based on the results of earlier studies assessing the effects of dose [9] and the time-dependent success of further transmission [11]. Our experiments were carried out over four weeks, with the host x strain combinations matched over this period (i.e., Monday workers were collected from colonies 1 to 5 and exposed to all parasite strains, then the corresponding scheme with Tuesday workers and colonies 6 to10, etc.) In each case, the bee’s abdomen was separated and prepared as samples for storage and for the later DNA extraction (see below). As already mentioned above, workers from twenty colonies were exposed to twenty parasite strains in a cross-infection design (Table 1). Infection intensity was assessed by quantitative PCR, using a master mix (with 212.0 μl) EvaGreen mixed with 42.4 μl Primer Cri-RT forward/reverse (5 μM each) and 593.6 μl H2O prepared for 96 microtiter plate; 8 μl mix plus 2 μl DNA were used per well. This method of quantifying infection intensity was already validated in an earlier study in the same system by means of dilution series [12]. Validation for the current study is shown in S12 File (assessing infection intensiy).
DNA isolation and library preparation
For each bee, the separated abdomen was put individually in a tube and snap-frozen in liquid nitrogen. The tube was then knocked on the bench to break open the abdomen. Subsequently, the sample was transferred to a new tube provided with 20 small glass beads (0.1–0.15 mm in diameter; for homogenization and later centrifugation) and stored at -20°C until processing. We extracted the DNA from aseptically dissected whole guts, by first adding 1 ml Buffer ATL with proteinase K to each sample (manufactured by Qiagen). The buffer was previously prepared as mix of 90 ml ATL with 10 ml proteinase K to yield a volume of 100 ml per plate. After buffer was added, we homogenized the samples using a bead mill homogenizer (Bead Ruptor), cooled with liquid nitrogen to keep temperature between 15° and 25°C, for 4 x 45 s with a 20 s pause in between. The samples were then centrifuged at 12,000 g for 3 min. We subsequently took 200 μl of liquid from the middle of the tubes and transferred these extracts into a 96-well analysis plate (’blue bottom’) from Qiagen’s DNeasy Blood and Tissue Kit. The plates were processed according to the manufacturer’s instructions by incubating them at 56°C for 2 h in a plate shaker at 250 rpm, followed by adding 410 μl of Buffer AL with EtOH (mixed from 40 ml AL with 41.8 ml EtOH). Finally, the samples were eluted with 200 μl Buffer AE for further processing. For the low-coverage study, we reduced the volumes of the Illumina Nextera to reduce costs, and have followed the protocol given in [27]. High-coverage library preparation and sequencing was done at the Functional Genomics Centre Zurich (FGCZ) on a HiSeq 4000, using the respective Illumina Truseq kit.
Bioinformatics
Low-coverage study (20 x 20 matrix)
Low-coverage libraries were sequenced (paired-end 150 nt, PE150) at an expected 2–3 x coverage on a NovaSeq instrument at NovoGene (Oxford and Beijing). Since the quality of the NovaSeq data were already very high, we did not further quality-filter the data prior to mapping. Sequences were mapped against the reference genome from Ensemble Metazoa; assembly ’Bter_1.0’, using BWA (v 0.0.17). Low-read, low-quality reads (e.g., strongly clipped reads, or those containing adapters) were removed (MAQ10), using sambamba (v0.6.6), and PCR-duplicates removed using Picard Tools (v 2.18.27). Overlapping reads were clipped using bamUtil (v 1.0.14). We then locally realigned indels using gatk (v 3.8). Samples with less than 0.5 x-coverage (a total of 10 samples) were removed (see also Fig A1 in S12 File, for the distribution of coverage). We have only used placed scaffold for the following analyses.
Genotype (SNP-) likelihoods were then calculated with angsd (v 0.925) [15] and, subsequently, the covariance matrix was estimated using PCangsd (v 0.98) [28]. We added an extra sample (e.g., DNA from a single, haploid male) to each library pool as an internal control. We also added a couple of replicates of individuals used in the GWAS. However, these samples were for controls and checking of procedures only; they were not used for the actual analyses. Among all other replicates, we chose the three samples with highest coverage for further analyses. Individuals that were not grouped according to population were removed as outliers, resulting in n = 1,170 individuals (i.e., a few missing due to technical difficulties). In a further step, the phenotypic data (infection intensity, using sqrt-transformed intensities to normalize variances) was corrected for the parasite strain effect with the lm-function in R based on approximation to normality of data. Based on the first results of the high- and low coverage study (that did not yield any host x parasite interaction effect), we focussed on the effect of host genotypic background (i.e., the ’colony’ term) alone. This interprets the parasite strain effect as the ecological background (see Discussion for justification of this rationale). For the GWAS, we then used 19 dimensions of the covariance matrix to account for population structure [29] (this number of dimensions proved to be the most effective), and the corrected phenotypic data (i.e., the residuals from the strain effect), using the score test of angsd. We have considered only variants with a minor frequency of more than 5% and used a total of 1.9 million sites. The genome-wide significance level was set to P = 0.1 after Bonferroni correction. Manhattan Plots and qqplots were generated with the R-package GWASTools [30]. Only the top twenty SNPs with the highest significance for an association with the phenotype are discussed here.
High-coverage study (8 x 8 matrix)
There are a total of over 125,000 possible combinations to form an 8 x 8-matrix from of our full (20 x 20) design. A randomly picked set of 100,000 such matrices were statistically analysed for the effect of the host x parasite interaction term on infection intensity. We chose one of these picks with the strong effect of the interaction terms for our 8 x 8 subset, with the idea to maximize the chances to find a corresponding signal with GWAS (see Supporting Information, Methods, for further details). As with the low-coverage data, and based on the first results, we removed the parasite strain effect for the further analyses and applied sqrt-transformed intensities to normalize variances. High-coverage libraries were sequenced (paired-end 150 nt, PE150) at circa 15 x-coverage on a Illumina HiSeq4000 instrument at the FGCZ. Raw reads were trimmed, using Trimmomatic (v 0.35), and then mapped against the reference genome (Bter_1.0), using BWA (v 0.0.17). Like the low-coverage study, low-quality reads were removed (MAQ10) and PCR-duplicates marked, using Picard Tools (v 2.0.1). The SNPs were then called using freebayes (v 1.0.2) adopting the default settings. The resulting raw variant table subsequently was filtered, using vctools (v1.0.1) and vcflib (v 1.0.1). We retained a SNP-mapping quality > 20, a mean depth > 10, minimum depth > 3, and a minor allele frequency of at least 5%. SNPs with too high coverage (more than 30 x), complex SNPs, and indels were removed. We allow for a maximum of missing sites of 5% per locus. Furthermore, all SNPs on unplaced scaffold were ignored. To correct for highly-linked SNPs, we applied a LD-pruning procedure, using plink (v 2), with window size of 200 bp, step size 50 bp, and a correlation threshold of r2 > 0.75. Eventually, we ended up with 504,111 high-quality SNPs across the 192 samples from the 8 x 8 matrix.
GWAS was then done on these corrected phenotypic data (as before, residuals from the parasite strain effect) whilst additionally correcting for the host relatedness effect due to similar genotypes with a given colony, using gemma (v 0.98.4) [31]. Again, we used a genome-wide significance level of P = 0.1 after Bonferroni correction, and Manhattan Plots and qqplots with the R-package GWASTools.
Analysis of the microbiota
This was done for the 8 x 8 high coverage matrix. The preparation of samples and DNA extraction for microbiota sequencing followed the protocols described above and, for example, in [32]. Samples were prepared for 16S Amplicon Sequencing. The libraries were sequenced on an Illumina MiSeq using a 600-cycle sequencing kit with paired-end 300- cycle option (PE300), using universal primers for the V3/V4 region. Because of technical difficulties with the readout of machine, we pooled reads from 4 different runs from the same Illumina MiSeq batch. In total, 48,415,850 paired-end raw reads that passed Illumina internal filtering (PF). Microbiota raw data are deposited with ENA, see also S10 File. We divided the data processing into steps according to the computing resource requirements. A detailed description of the AmpSeq workflow can be found in the file ’4-DataPrep_ Workflow_Report’ in Appendix C of S12 File. In short, the paired-end reads were filtered (e.g., PhiX and low complexity), 3’-end trimmed, overlapping pairs were merged and the primer regions were removed. In a next step, the data was subjected to a size range, GC content range and quality filtering. The loss of data during each of these processing steps is also listed in the file. The cleaned data were then (a) clustered using UPARSE 2 (for OTUs) with a 97%-identity criterion, which produced 772 OTUs; and (b) error-corrected and zero-radius clustered (z) using UNOISE3 (for zOTUs) [33], which produced 997 zOTUs. The results were similar, but we use the latter for the analyses reported here, based on the widely accepted advantages of this latter method in terms of improved OTU identification [34–37].
We used SINTAX with the 16S reference SILVA (v128) to predict the taxonomic association for the clusters. We consider SILVA as a conservative, but reliable database. Because the critical zOTUs (see below) are of prime importance, we additionally have blasted those items against the NCBI-database. The latter database offers more taxonomic resolution, but its entries are also considered to be more error-prone.
Results
We first report our findings from the ’low-coverage’ study, where all individuals in the entire experiment were included (n = 1,170 bees) but with lower sequencing depth. We refer to this part as the 20 x 20 (full) matrix (colonies x parasite strains) study. In the subsequent sections, the ’high-coverage’ studies are covered, which refers to bees sequenced at a widely accepted coverage level, e.g. [38]. This sample was also screened for the microbiota. We refer to this part as the 8 x 8 matrix study.
Genetic associations from the low-coverage study (20 x 20 matrix)
Appendix A in supporting information
The overall variation of infection intensities per bee, across colonies or parasite strains, is shown in Fig S1 of S12 File. All of these samples were sequenced with an estimated coverage of 3.4x. After dropping all unplaced scaffolds, this value dropped to 2.1x, presumably because sequences were not covered at random and the unplaced scaffold contained many repetitive sequences. Using square-root-transformed infection intensities corrected for strain effects (see Methods; labelled as ’lmm_sqrt_res’ in the Appendix), we could use 1,782,441 SNPs to test for any association with infection intensity. None of these SNPs reached significance at the genome-wide level (Fig A2 in S12 File). However, to have at least some insights, the top twenty SNPs (i.e., with the highest levels of association) are shown in Table 2.
Table 2. GWAS from low-coverage study (20 x 20 matrix).
| Rank | Code | SNPc | LGd | Position | Gene | Annotatione | Characteristic |
|---|---|---|---|---|---|---|---|
| 1 | CM001178.1 | G/A | B10 | 4206573 | LOC100647628 | kinesin-like protein | intronic |
| 2 | CM001179.1 | T/A | B09 | 9599615 | LOC100650970 | nephrin | intronic |
| 3 | CM001177.1 | A/G | B11 | 5529605 | LOC100645564 | peripheral plasma membrane protein; splice variants | intronic |
| 4 | CM001169.1 | C/T | B01 | 1931967 | LOC100651512 | B-cell receptor CD22 | intronic |
| 5 | CM001179.1 | G/A | B11 | 13429222 | N/A | N/A | N/A |
| 6 | CM001171.1 | G/T | B03 | 13298216 | LOC105667178 | uncharacterized | intronic |
| 7 | CM001175.1 | G/A | B07 | 8603884 | LOC100644761 | phosphoinositide-dependent protein kinase | intronic |
| 8 | CM001181.1 | G/A | B13 | 3266810 | N/A | N/A | N/A |
| 9 | CM001171.1 | C/T | B03 | 10561994 | N/A | N/A | N/A |
| 10 | CM001180.1 | T/G | B12 | 1072389 | LOC100643982 | disks large 1 tumor suppressor protein | intronic |
| 11 | CM001175.1 | G/A | B07 | 6138344 | N/A | N/A | N/A |
| 12 | CM001181.1 | T/A | B13 | 3899059 | N/A | N/A | N/A |
| 13 | CM001175.1 | A/G | B07 | 6104364 | N/A | N/A | N/A |
| 14 | CM001181.1 | G/T | B13 | 648719 | LOC100643707 | transcription factor HES1 | intronic |
| 15 | CM001169.1 | A/G | B01 | 16216739 | N/A | N/A | N/A |
| 16 | CM001177.1 | C/T | B09 | 11148524 | LOC100644210 | dopamine D2-like receptor | intronic |
| 17 | CM001181.1 | G/A | B13 | 702677 | LOC100643707 | transcription factor HES1 | intronic |
| 18 | CM001179.1 | G/A | B11 | 5045960 | LOC105666228 | histone-lysine N-methyltransferase SETMAR-like (gene 6179) | exon |
| 19 | CM00176.1 | G/C | B08 | 2507890 | LOC100647422 | basement membrane-specific heparan sulfate proteoglycan protein | intronic |
| 20 | CM001169.1 | G/A | B01 | 3251298 | N/A | N/A | N/A |
Top 20 SNPs that associate with infection intensitya,b. Recovery of transcripts, genes, and automated annotations.
a Square-root transformed data, corrected for colony relatedness structure and strain effect. See file ’2-TopSNP_LowCoverage’ (Appendix A in S12 File).
b None reaches genome-wide significance level (cutoff P = 0.1) after Bonferroni correction.
c Major/minor allele.
d Linkage group.
e based on assembly Bter_1.0.
The top twenty SNPs are distributed in annotated and uncharacterized regions (’NA’, 8 out of 20 SNPs). Among the annotated ones are kinesin-like proteins, membrane proteins, or genes assigned with names similar to B-cell receptors (cells that do not exist in insects). Furthermore, the top SNPs are spread over several chromosomes, but are more common on chromosomes B09 and B11 (Bombus has n = 18 chromosomes). All but one of the specified top SNPs are intronic; some are close but many more distant from the next exon. None of these SNPs are therefore expected to be found in a transcribed product and to have a direct effect on a functioning protein. Nevertheless, their presence may affect the transcription process in some way (see Discussion). The exception to this pattern is the SNP at rank 18 that is located within a predicted exon (Table 2). A SNP in LOC10064v3982 (rank 10), annotated as disks 1 large 1 tumor suppressor protein is remarkable as it shows up again in the high coverage study below. The annotations (if any) for the top SNPs of this analysis are hard to reconcile with obvious functions in defences against infections. We return to these findings in the general discussion.
Genetic associations from the high-coverage study (8x 8 matrix)
Appendix B in supporting information
For this section, the 8 x 8 matrix was analysed. There was considerable variation in infection intensities per bee among colonies, parasite strains, and their interaction (Fig 1). As described in Methods, infection intensities were square-root-transformed and values corrected for strain effects (label: ’lmm_sqrt_res’) based on their approximation to normality of data. In addition, we pruned the SNPs for linkage disequilibria among pairs to correct for over-representation of non-informative couplings. A total of 504,111 SNPs were available for analysis.
Fig 1. Boxplot showing the interactions in the chosen envelope of 8 x 8 colonies vs. parasite strains.
Shown are the infection intensities (y-axis) and, on the x-axis, the 8 blocks (colonies, nrs. 15, 33, 59, 82, 91, 141, 225, 319) with the 8 parasite strains within each block (nrs. 8.075, 8.091, 10.290, 10.486, 12.444, 12.450, 14,149, 14.172); see also Table 1 in text. Each combination is indicated by labels with a string such as 8.075.15, indicating the combination of strain ’8.075’ with colony ’15’. Colours only chosen to aid clarity. The overall interaction term has a significance of P = 7.97 10−7.
None of the twenty SNPs with the highest association levels (Table 3) reached the threshold level for a genome-wide association, although those on chromosome B12 came close (Fig 2; numerical value of parameter ’p_lrt’ in Suppl Info, file 3-TopSNP_HighCoverage). Hence, we tentatively discuss SNPs presumably associated with infection intensity. Interestingly, among the top twenty SNPs, two annotated genomic regions stand out: mucin-12, followed by disks large 1 tumor suppressor protein, both, on chromosome B12 (Table 3). It is quite striking that 12 out of the top twenty SNPs are in one of these two regions, with 10 entries alone in the vicinity of mucin-12. The genomic region of mucin-12 is shown in more detail in Fig 2. A closer look shows that all SNPs of Table 3 are intronic, that is, do not fall into a genomic region translated into a protein.
Table 3. GWAS from high-coverage study (8 x 8matrix).
| Rank | Code | SNPc | LGd | Position | Gene | Annotatione | Characteristic |
|---|---|---|---|---|---|---|---|
| 1 | CM001180.1 | A/G | B12 | 754245 | LOC100643741 | mucin-12 | intronic |
| 2 | CM001180.1 | A/C | B12 | 630576 | LOC100643741 | mucin-12 | intronic |
| 3 | CM001180.1 | A/G | B12 | 199640 | LOC100642650 | elongation of very long chain fatty acids protein AAEL008004 | intronic |
| 4 | CM001180.1 | T/C | B12 | 624565 | LOC100643741 | mucin-12 | intronic |
| 5 | CM001180.1 | T/C | B12 | 678846 | LOC100643741 | mucin-12 | intronic |
| 6 | CM001180.1 | G/A | B12 | 907647 | LOC100643982 | disks large 1 tumor suppressor protein | intronic |
| 7 | CM001180.1 | C/G | B12 | 708084 | LOC100643741 | mucin-12 | intronic |
| 8 | CM001180.1 | T/C | B12 | 815048 | LOC100643741 | mucin-12 | intronic |
| 9 | CM001180.1 | A/G | B14 | 1274461 | LOC100645175 | transcription factor 12 | intronic |
| 10 | CM001184.1 | A/G | B16 | 3562604 | N/A | N/A | N/A |
| 11 | CM001180.1 | G/T | B12 | 676081 | LOC100643741 | mucin-12 | intronic |
| 12 | CM001180.1 | A/G | B12 | 815680 | LOC100643741 | mucin-12 | intronic |
| 13 | CM001180.1 | A/G | B12 | 1382663 | LOC100643982 | disks large 1 tumor suppressor protein | intronic |
| 14 | CM001171.1 | T/G | B03 | 11841846 | N/A | N/A | N/A |
| 15 | CM001178.1 | C/T | B10 | 9502791 | LOC100643455 | beta-1 syntrophin | intronic |
| 16 | CM001180.1 | A/G | B12 | 703215 | LOC100643741 | mucin-12 | intronic |
| 17 | CM001180.1 | G/A | B12 | 642886 | LOC100643741 | mucin-12 | intronic |
| 18 | CM001172.1 | C/T | B04 | 640996 | N/A | N/A | N/A |
| 19 | CM001177.1 | A/G | B09 | 6824389 | LOC100648443 | lachesin-like transcript | intronic |
| 20 | CM001183.1 | T/C | B15 | 4658505 | N/A | N/A | N/A |
Top 20 SNPs that associate with infection intensitya,b. Recovery of transcripts, genes, and automated annotations.
a None reaches genome-wide significance level (cutoff P = 0.1) after Bonferroni correction. See file ’3-TopSNP_HighCoverage’ (Appendix B in S12 File).
b square-root transformed data, corrected for colony relatedness structure and strain effect, and using LD-pruned data; criterion 0.75 (i.e. pairs of SNPs dropped for LD > 0.75).
c Major/minor allele.
d Linkage group.
e based on assembly Bter_1.0.
Fig 2. GWAS for the high-coverage study (8 x 8 matrix).
(a) Genomic region annotated as mucin-12 (LOC100643471). The vertical red lines indicate the position of SNPs associating with infection intensity. Red numbers above line refer to rank listed in Table 3. All SNPs are intronic and refer to four transcripts: XM_003399573.2, XM_012314481.1, XM_012314481.2, and XM_012314481.3. The region shown here extends from pos. 532,098 to 867,238. The blue band refers to contigs, the green band to the predicted gene, mucin-12. (b) Manhattan plot for the high-coverage study; data as described in text. Top panel: Probability (p) of association with infection intensity for SNPs along the genome (with chromosome B01 to B18 in different colours). No SNP reached the genome-wide threshold (dashed line corresponds to basic cutoff-value of P = 0.1 after Bonferroni correction). Bottom panel: comparing expected to observed probabilities. The calculations used sqrt-transformed infection intensities (case ’lmm_sqrt_res’ above, pruned for LD > 0.75), and using p_lrt as measure.
Role of the microbiota
Appendix C in supporting information
For the bees of the high coverage study (8 x 8-matrix, a total of n = 192 bees), we screened their gut microbiota using 16S RNA-amplicon sequencing. The reads were clustered into zOTUs (zero-radius Operational Taxonomic Units; for a technical report see files ’4_DataPrep_Workflow_Report’, and ’C1_Amplicon_Technical_Report’, Suppl Info). A list of the metagenomic data deposited with ENA (accession PRJEB52013) is given in S11 File. From the total of 192 possible samples (three replicates each), two samples were excluded for technical problems, one additional sample was included as negative control (i.e., a ’blank’ sample, with no bee material added), leaving a total of n = 191 samples. The blank control produced reads of zOTUs. However, compared to an average of 204,062.7 total reads per sample from the experimental bees, the control only produced 8,589 reads in total, with very low reads for all zOTUs with a slight exception: zOTU26 annotated as Bifidobacteriaceae (raw counts: n = 1,009, compared to n = 125,103 reads in the experimental bees). We conclude, therefore, that the results for the experimental bees were not significantly distorted by procedure-related errors [39]. The nucleotide sequences of all zOTUs can be found in file ’9_16S_ZOTU.fa’ of Appendix D in S12 File.
In all, each of the 8 x 8 combinations is thus represented by three bees (replicates), except for the two cases where only two replicates could be used (colony 15 exposed to strain B, and colony 319 to Q). Library sizes per sample varied, with a significant effect of the depth of coverage (number of reads) on the number of zOTUs per sample (R = 0.089, F1,188 = 19.53, p < 0.0001) and on zOTU diversity (rarefaction plots, Fig C1 in S12 File). The number of reads did not vary with colony or strain, but zOTU- diversity increased with sequencing depth (Fig C2 in S12 File) and varied among colonies (Fig C3 in S12 File). For the entire set, and after processing the raw data, we retrieved 997 zOTUs. Of those, only three could be identified to the species level, but half (52%) to the genus, and the majority (80% and more) to the level of family and above. Note that although we use a stringent confidence threshold of 85%, these assignments are based on homology and must therefore be treated with some caution.
We analysed the pairwise distances of all samples with MDS (Multi-Dimensional Scaling) to identify the clustering of samples based on microbial community composition and abundance. We also extracted the corresponding ordination axes for further analysis. Among the various distance measures used for this purpose, ’weighted Unifrac distance’ (wuf) and ’Bray-Curtis distance’ (bc) were closest to normality. Finally, wuf was chosen for further discussion, but the conclusions would remain the same with bc. Fig 3 shows the results of the MDS-analysis. We find no significant difference in the zOTU-composition among colonies, with most zOTUs belonging to the Beta- and Gamma-Proteobacteria, and to a lesser degree also to the Alphaproteobacteria.
Fig 3. Ordination (multidimensional scaling, MDS) of microbiota samples by colony.
Shown are the two primary axes (Axis.1, Axis.2) of a Principal Component Analysis (PCoA), based on weighted unifrac distances (wuf) between zOTUs. Numbers in brackets indicate percentage of total variation explained by the axis. Colours indicate colony (legend on the right). Samples (n = 7) with less than 10,000 reads were not considered. Colony 141 (C141, in blue) seems somewhat apart from the others, but no significant differences exist among colonies.
Rarefaction plots showed that for almost all samples, the zOTU richness reached a plateau with low to moderate library sizes, with only a few samples not saturating within the limits of the coverage (Fig C1 in S12 File). Because of the heterogeneity among samples, we additionally used an improved method (see [40, 41]) to derive species richness and diversity for further analyses. With this method, the three replicates per colony x strain combination served to estimate a mean and variance in these measures and for this combination. In all, we wanted to know whether the richness or diversity of zOTUs correlates with infection intensity in a given bee. For this analysis, we used the R-package ’breakaway’ [42] and ’divNet’ [43], and worked with a reduced set of 772 OTUs from the UNIPARSE procedure (see Methods). The results were as follows.
Species (zOTU) richness. Richness (alpha diversity) estimates with the Chao1-index are shown in Fig C2a in S12 File. A few cases (e.g., colony 141 infected by strain L) stand out, but there is no overall significant variation among samples. Diversity per sample was estimated with the Shannon-Wiener Index (package ’divNet’ [43], Fig C2b in S12 File). Alpha diversity varied across samples but not at a significant level. But a significant variation existed among the microbiota from different colonies (Fig C4 in S12 File). A negative relationship between infection intensity and microbiota diversity may exist but there is no clear statistical support from our data (Fig C5 in S12 File).
Classification by microbiota composition
Appendix D in supporting information
In a further step, we also grouped the samples by their microbiota composition, motivated by the pattern shown in Fig 3 (i.e., a dominant spread of the data along Axis.1), and by previous studies showing that the microbiota affects infection by Crithidia [16, 19]. We utilized the pairwise taxonomic distances (wuf) between the microbiotas of the samples, leaving a total of 183 individuals in the set. As Fig 3 indicates, Axis 1 of the multi-dimensional scaling adds most to the separation of cases, explaining roughly two thirds of the total variation in the composition of the microbiota. We therefore classified samples according to whether their Axis 1 is equal or higher than zero (’positive’) or smaller than zero (’negative’). In fact, using a classification by both axes, Axis 1 and Axis 2, did not add more information (see Fig D1 in S12 File).
We were interested to see whether there are host genomic associations with microbiota composition in our data. Here again we analysed using GWAS but where the phenotype is not infection intensity but the value of Axis 1 or Axis 2, respectively. Files ’5-TopSNP_MBiota_Ax1’ (a total of 502, 953 SNPs) and ’6-TopSNP_MBiota_Ax2’ (Appendix C in S12 File) summarize the results (showing the 20 SNPs with the highest association levels). As these analyses show, SNPs that have the highest association levels with microbiota composition are either located in uncharacterized genomic regions or have annotations not directly associated with immune defences, such as adenylate cyclase, EH domain-binding proteins (typically involved in vesicle trafficking), mannosidases, or transporters. These genomic associations to Axis 1 and Axis 2 are difficult to understand and we cannot see a direct connection between microbiota composition and genomic regions of these SNPs.
The situation became clearer by looking at the GWAS-associations of, again, infection intensity as the phenotype, but for individuals grouped according to microbiota composition along Axis 1 (which covers 65.1% of the total variation, Fig 3). In fact, Fig 4 shows that infection intensity significantly differs between two groups, which we call ’resistant’ and ’susceptible’, respectively. In our case, samples with a microbiota composition that is characterized by having values of Axis 1 ≥ 0 (’positive’, ’resistant’, n = 118 individuals) were less infected compared to the ’negative group’ (with values Axis 1 < 0; ’permissive’, n = 65 individuals). The grouping itself is somewhat arbitrary by selecting the wuf distance measure, but virtually the same results would emerge from using bc. More importantly, Fig 4 demonstrates again that infection intensity is not independent of the microbiota and its taxonomic structure. For the two separate GWAS, the twenty SNPs with the strongest association level to infection intensity are shown in Tables 4 and 5.
Fig 4. Infection intensity with respect to zOTU composition group.
Samples were separated by multi-dimensional scaling (see Fig 3) and grouped by the value of their first component (Axis 1 < 0, or Axis 1 ≥ 0). The ’negative’ group (’permissive’) had significantly higher infection intensities than the ’positive’ (’resistant’) group (Walsh’s t = 6.12, p ≪ 0.0001, df = 95.6). zOTU-distances based on wuf.
Table 4. Top 20 SNPs associated with infection intensity in samples having Axis.1 < 0.
| Rank | Location | SNPc | LGd | Position | Gene | Annotatione | Characteristic |
|---|---|---|---|---|---|---|---|
| 1 | CM001182.1 | G/A | B14 | 5686065 | LOC100642385 | nucleosome assembly protein | intronic |
| 2 | CM001171.1 | T/C | B03 | 11840228 | N/A | N/A | non-coding |
| 3 | CM001171.1 | A/G | B03 | 12218710 | LOC100648315 | phosphatase and actin regulator | intronic |
| 4 | CM001171.1 | A/G | B03 | 12383573 | LOC100648013 | putative serine/threonine-protein kinase S6KL | intronic |
| 5 | CM001171.1 | T/C | B03 | 12748227 | N/A | N/A | N/A |
| 6 | CM001171.1 | G/A | B03 | 13158328 | LOC100650315 | division abnormally delayed protein | intronic |
| 7 | CM001179.1 | T/A | B11 | 9688234 | LOC100650970 | nephrin | intronic |
| 8 | CM001180.1 | T/A | B12 | 1233258 | LOC100643982 | disks large 1 tumor suppressor protein | intronic |
| 9 | CM001178.1 | T/C | B10 | 8653227 | N/A | N/A | N/A |
| 10 | CM001179.1 | G/A | B11 | 6727668 | LOC100649546 | uncharacterized | intronic |
| 11 | CM001183.1 | T/C | B15 | 7927222 | LOC100651177 | serine protease stubble | intronic |
| 12 | CM001179.1 | G/A | B11 | 9857303 | LOC100650970 | nephrin | intronic |
| 13 | CM001183.1 | G/A | B15 | 7905861 | LOC100651422 | uncharacterized | intronic |
| 14 | CM001183.1 | C/A | B15 | 8246283 | N/A | N/A | N/A |
| 15 | CM001183.1 | A/G | B15 | 8914728 | LOC100646518 | uncharacterized | intronic |
| 16 | CM001175.1 | T/C | B07 | 8825700 | LOC100644761 | 3-phosphoinositide-dependent protein kinas | non-translated region of 1 transcript |
| 17 | CM001170.1 | T/C | B02 | 5742310 | LOC100645882 | protein dead ringer homolog | intronic |
| 18 | CM001183.1 | G/A | B15 | 7466300 | LOC100644824 | max dimerization protein 1-like | intronic |
| 19 | CM001171.1 | T/G | B03 | 14156904 | N/A | N/A | N/A |
| 20 | CM001177.1 | G/A | B09 | 5694646 | LOC100645564 | peripheral plasma membrane protein CASK | intronic |
’Permissive’, grouped by multi-dimensional scaling, based on wuf a. Phenotype is infection intensity (sqrt-transformed values, corrected for colony relatedness structure and strain effect)b. High-coverage study (8 x 8 matrix).
a compare Fig 2; None reaches genome-wide significance level (cutoff P = 0.1) after Bonferroni correction. Fom file: 7-TopSNP_ MBiota_negative (Appendix D in S12 File).
b using LD-pruned data; criterion 0.75 (i.e., pairs of SNPs dropped for LD > 0.75).
c Major/minor allele.
d Linkage group.
e based on assembly Bter_1.0.
Table 5. Top 20 SNPs associated with infection intensity in samples having Axis.1 ≥ 0.
| Rank | Location | SNPc | LGd | Position | Gene | Annotatione | Characteristic |
|---|---|---|---|---|---|---|---|
| 1 | CM001182.1 | T/C | B14 | 10846214 | LOC100646390 | lachesin | intronic |
| 2 | CM001181.1 | T/C | B13 | 1802533 | N/A | N/A | N/A |
| 3 | CM00179.1 | G/A | B11 | 13129707 | LOC105666237 LOC100642498 |
uncharacterized | intronic |
| 4 | CM001182.1 | T/G | B14 | 9825985 | N/A | N/A | N/A |
| 5 | CM001184.1 | A/G | B16 | 3562604 | N/A | N/A | N/A |
| 6 | CM00180.1 | A/G | B12 | 815680 | LOC100643741 | mucin-12 | |
| 7 | CM001178.1 | A/G | B10 | 13330302 | LOC105666188 | CCR4-NOT transcription complex subunit 6-like | intronic |
| 8 | CM001182.1 | T/C | B14 | 11198348 | LOC100649003 | cyclin-dependent kinase-like | exon |
| 9 | CM001184.1 | G/T | B16 | 2728789 | N/A | N/A | N/A |
| 10 | CM001178.1 | G/A | B10 | 10428051 | N/A | N/A | N/A |
| 11 | CM00171.1 | T/G | B03 | 11841846 | N/A | N/A | N/A |
| 12 | CM001184.1 | G/T | B16 | 2967795 | N/A | N/A | N/A |
| 13 | CM001184.1 | G/A | B16 | 2991827 | N/A | N/A | N/A |
| 14 | CM001184.1 | G/A | B16 | 3277554 | LOC100642943 | nuclear RNA export factor 1-like | exon |
| 15 | CM00184.1 | G/T | B16 | 2553260 | LOC100648917 | sorting nexin-13-like protein | exon |
| 16 | CM001178.1 | T/C | B10 | 10745738 | LOC100647933 | leucine-rich repeat-containing protein | intronic |
| 17 | CM001178.1 | C/T | B10 | 10615741 | LOC100651334 | solute carrier organic anion transport family member 2A1 | intronic, close to exon |
| 18 | CM001172.1 | G/T | B04 | 6651056 | LOC100648473 | 2-acylglycerol O-acetyltransferase 1-like | in exon of 2 transcripts; intronic for 3 transcripts; rest in non-coding regions |
| 19 | CM001171.1 | G/A | B03 | 7486034 | LOC100650996 | lachesin in A. mellifera (Ig-Superfamily) | intronic |
| 20 | CM001171.1 | A/G | B03 | 7515414 | LOC100650996 | lachesin in A. mellifera (Ig-Superfamily) | intronic |
’Resistant’, grouped by multi-dimensional scaling, based on wuf a. Phenotype is infection intensity (sqrt-transformed values, corrected for colony relatedness structure and strain effect)b. High-coverage study (8 x 8 matrix).
a compare Fig 2; None reaches genome-wide significance level (cutoff P = 0.1) after Bonferroni correction. From file: 8-TopSNP_MBiota_positive (Appendix D in S12 File).
b using LD-pruned data; criterion 0.75 (i.e. pairs of SNPs dropped for LD > 0.75).
c Major/minor allele.
d Linkage group.
e based on Bter_1.0.
When comparing the two tables, it seems that among the twenty SNPs (out of a total of 420,567 SNPs in the set from 65 bees) with the highest association levels with the value of Axis 1, there are more SNPs located in genes with an annotation for the permissive group (’negative’ group, Table 4) than for the resistant group (’positive group’, with 450,319 SNPs from 116 bees; Table 5). In the former group, all SNPs are intronic. Among those, the SNP at rank 8 refers to disks large 1 tumor suppressor protein, which was identified in the high-coverage analysis before (Table 3). Remarkably, among the SNPs of the resistant group that are in regions with an annotation, 4 out of 12 are in an exon. These are those in LOC1000649003, LOC100642943, LOC100648917, and LOC100648473 (Table 5). When looking at the annotations, two cases stand out even though they are intronic. On one hand, the SNP at rank 6 (Table 5) is in the genomic region for mucin-12, already prominent in the high-coverage analysis above (LOC100643741, Table 3). On the other, several SNPs locate to lachesin, an Ig-superfamily protein that, for example, is associated with neurogenesis and antiviral responses in insects.
The zOTU-composition of resistant and permissive samples was found to be different for all diversity measures, except for the Inverse Simpson index (Fig D2 in S12 File). Resistant samples had lower zOTU-diversity in terms of Species Richness and Chao1-indices, whereas the opposite was observed when using the Shannon-Index, a finding similar to a previously reported case [22]. Similarly, the taxonomic zOTU-composition of the two groups showed differences at the level of Class (Fig 5), Phyla or Order (Figs D3 and D4 in S12 File).
Fig 5. zOTU-composition at the level of systematic class.
Shown are the samples in the ’permissive’ (negative; left panel) and the ’resistant’ (positive; right panel) group (see text). Bacilli/Betaproteobacteria and Gammaproteobacteria make a difference (see colour code). Distances for ordination based on weighted unifrac (wuf).
Summary indices, such as Chao1 or Shannon-Wiener, are obviously very crude. A closer inspection provides more information. For this purpose, we scrutinized the abundance of the zOTUs in the resistant (positive Axis 1) vs the permissive (negative Axis 1) group. The samples were filtered as described in Methods and to exclude rare taxa (i.e., with counts less than ten). Subsequently, the difference between groups was calculated as a log-fold change (using DESeq2) in abundance between the two groups, and samples filtered further by only accepting significant changes at an adjusted padj < 0.1. This procedure resulted in eleven ’critical’ zOTUs that differed between groups, described by the available taxonomic information (Table 6). In particular, Lactobacillus was more abundant in the permissive group, whereas Snodgrassella, Gilliamella, and the (candidate) genus Schmidhempelia was more abundant in resistant.
Table 6. Critical zOTUs in individual microbiotas of resistant and permissive groups.
| (a) zOTUs associated with Resistant | |||||||
| zOTU | Kingdom | Phylum | Class | Order | Family | Genus | Speciesa |
| ZOTU1 | Bacteria (1.00) | Proteobacteria (1.00) | Betaproteobacteria (1.00) | Neisseriales (1.00) | Neisseriaceae (1.00) | Snodgrassella (0.99) | Snodgrassella gandavensis*, S. communis*, S.alvi* |
| ZOTU2 | Bacteria (1.00) | Proteobacteria (1.00) | Gammaproteobacteria (1.00) | Orbales (1.00) | Orbaceae (1.00) | -- | Gilliamella bombi*, G. bombicola*, G. apis* |
| ZOTU4 | Bacteria (1.00) | Proteobacteria (1.00) | Gammaproteobacteria (1.00) | Orbales (1.00) | Orbaceae (1.00) | Gilliamella (0.75) | Gilliamella bombi*, G. bombicola*, G. apicola (?), G. mensalis* |
| ZOTU7 | Bacteria (1.00) | Proteobacteria (1.00) | Gammaproteobacteria (1.00) | Orbales (1.00) | Orbaceae (1.00) | Cand_Schmidhempelia (1.00) | Gilliamella bombi*, G. bombicola*, G. apicola* |
| ZOTU30 | Bacteria (1.00) | Proteobacteria (1.00) | Gammaproteobacteria (1.00) | Orbales (1.00) | Orbaceae | Gilliamella (0.99) | Gilliamella bombicola*, G. mensalis*, G. apicola* |
| (b) zOTUs associated with Permissive | |||||||
| zOTU | Kingdom | Phylum | Class | Order | Family | Genus | Speciesa |
| ZOTU3 | Bacteria (1.00) | Firmicutes (1.00) | Bacilli (1.00) | Lactobacillales (1.00) | Lactobacillaceae (1.00) | Lactobacillus (1.00) | Lactobacillus bombicola*, L. kullabergensis*b, L.kimbladi*2, L. huanngpiensis*, L. apis*, L. helsingborgensis*b |
| ZOTU6 | Bacteria (1.00) | Firmicutes (1.00) | Bacilli (1.00) | Lactobacillales (1.00) | Lactobacillaceae (1.00) | Lactobacillus (1.00) | Lactobacillus bombi, Bombilactobacillus bombis*, B.mellis*, Apilactobacillus kunkeei*b |
| ZOTU13 | Bacteria (1.00) | Firmicutes (1.00) | Bacilli (1.00) | Lactobacillales (1.00) | Lactobacillaceae (1.00) | Lactobacillus (1.00) | Lactobacillus apis* |
| ZOTU69 | unresolved | -- | -- | -- | -- | -- | |
The samples were processed and filtered as described in Methods and contain read counts > 10, and refer to an adjusted significance level of padj < 0.1. Taxonomic information is from the used libraries (see text). Numbers in parentheses are bootstrap confidence levels.
a additional matches from blast to the NCBI nucleotide database indicated by asterisk (*). Only matches with full (100%) sequence identity of the submitted string, and only those with an assigned regular (binomial) taxonomic name shown here.
b recently described species described from honeybees.
All of the zOTUs of Table 6 were additionally blasted against the NCBI nucleotide database. This resulted in a more detailed taxonomic resolution for all these zOTUs. At the same time, however, the results are muddled by matches to several taxa. A part of these multiple matches can be explained by our reads not covering the complete sequences of the NBI database. However, an equally important source of ambiguity are the burgeoning recent changes in taxonomic nomenclature. For example, among those critical for resistant phenoytpes, zOTU1 (via NCBI) now matches Snodgrassella gandavensis, S. communis and S. alvi. The first two of those are very recently proposed taxa, restricted to Snodgrassella in bumblebees [44]. zOTU2 and zOTU7 match Gilliamella bombis, G.bombicola, and G. apicola. Again, the former two are recently erected taxa for Gilliamella found in bumble bees [45]. Similar matches are found for zOTU4, including the new species G. mensalis [45]. We note that zOTU69 is a relatively short amplicon (304bp vs. other zOTUS with over 420 bp), and perhaps a product of false priming, as it relates (via blasting) to the host genome. Excluding this zOTU does not, however, change the conclusions.
For the permissive group, the taxon zOTU3 matches several Lactobacillus taxa identified with he NCBI blasts. Several of those have been proposed as new species associated with honeybees (L. kimbladi, L. kullabergensis, L. helsingborgensis) [46] showing complex phylogenetic relationships [47], and are therefore unlikely to be true symbionts of bumblebees. zOTU6 is variably identified with Lactobacillus bombi, Bombilactobacillus bombi, and B. mellis, whereby Lactobacillus is the basonym and Bombilactobacillus refers to a newly defined genus found in bumblebees [48], similar to the new genus Apilactobacillus in honeybees [49]. zOTU13 and zOTU30 match with Lactobacillus apis, which according to the revised nomenclature would refer to taxa found in honeybees only [47]. Matches are found variably on chromosomes 14 (B. terrestris), but also on chromosomes 6, 9, 13, 15, and 20 (in other Bombus species).
Varying the interaction matrix
Appendix E in supporting information
We started this study with the goal to uncover the effects of genomic variation on the prevalence or intensity of an infection (’the infection phenotype’) due to the host x parasite genotype interaction. For this system, such interaction effects have been amply demonstrated in previous studies, e.g. [11, 12, 50], including effects on gene expression, e.g. [14, 51]. Here, we use ’host’ as a token for the genotypic background of a given host colony, whereas the parasite is defined as a C. bombi clonal genotype, based on polymorphic microsatellite loci (note that the strains used here later were also fully sequenced [25]).
Contrary to these expectations, analysing the full (20 x 20)-matrix, showed that infection intensity was strongly affected by the main effects of colony and strain, but not by their interaction (ANOVA for transformed infection intensities. Colony effect: F19,784 = 20.795, P ≪ 0.001, Strain effect: F19,784 = 19.878, P ≪ 0.001, interaction: F361,784 = 0.981, P = 0.58). We therefore scrutinized the situation further by selecting sub-matrices of smaller dimensions, such as an 8 x 8-matrix (one of which was chosen for the high-coverage study) and then calculated the corresponding ANOVAs; we here call the respective matrix dimension (e.g., k = 8) the envelope size. As immediately obvious, there are many different possible sets of k = 8 out of N = 20 colonies; in fact, the exact number of possible sets is: permutations. The same is true for selecting other values of k out of N strains. These must then be combined with the set of selected colonies, which together rapidly generates very large numbers of possible combinations. For this reason, we used 1,000 random picks (permutations with no replacement) of Colony x Strain set combinations, wherever the possible number of permutations exceed this value, and calculated an ANOVA for each of them. The results are summarized in Fig 6.
Fig 6. Prevalence of significant effects.
Shown are the number of cases in relation to envelope size (k), and per 1,000 random permutations, where either the main effects for Colony or Strain, or the Interaction effects became significant (at level, pcrit = 0.01) in ANOVAs. Envelope size k defines the size of the analysed matrix (k x k). Infection intensity was transformed with exp = 1/3 to normalize variances. Here, all cases (incl. bees with zero infection) were used.
Regardless of the significance criterion used (i.e., pcrit = 0.05 or pcrit = 0.01) and regardless of using all cases or only infected individuals, the pattern was the same: as the envelope size, k, increases (i.e., the dimension of the matrix becomes larger), the main effects became more prominent whereas the interaction effects became less likely to become statistically significant. Hence, contrary to the usual expectation that with higher degrees of freedom for the interaction term, it should be more likely to find a significant interaction effect, it was actually less likely. Similarly, ANOVA estimates interaction term after the main effects have been estimated, such that less of the total variance remains available for the estimates. But this artefact cannot explain the situation here, as we discuss below.
Discussion
Genetic basis of infection phenotype
The work reported here attempted to identify SNPs of the host, B. terrestris, that predispose an individual to infection by Crithidia bombi in a cross-infection matrix. Two GWAS-methods were used on the same material—analysing a full 20 x 20 cross-infection matrix at low coverage [15], and analysing a reduced 8 x 8 matrix at high coverage. As Tables 2 and 3 show, both methods identify candidate SNPs, but the particular SNPs vary considerably. In both cases, we noted that the analysis could not separately retrieve the main (by Colony, by Strain) and interaction (Colony x Strain) effects, which would have allowed us to identify SNPs that characterize the well-described genotypic matching with those reported earlier studies [14, 51]. We therefore ignored the effect of parasite strain in the reported analysis (i.e., leading to data corrected for colony structure and strain effects). This mimics a situation where colonies would be exposed to a variety of strains, which is, in fact, what would happen in the field situation. Thus, we are capturing genomic variation that determines general susceptibility or resistance to infection by C. bombi.
The low-coverage study identified a considerable number of SNPs in uncharacterized loci or in non-coding regions. Among all loci listed in Table 2, LOC100645564 (rank 3), annotated as a possible splice variant of a plasma protein, and disks large 1 tumor suppressor protein (rank 10), might be involved in immune defences, whereas all other loci refer to genes with no clear link to immunological function. For the high-coverage study (8 x 8 matrix), the twenty SNPs that have the highest association with infection intensity are shown in Table 3. Note that our interpretations must be subject to the proviso that none of these SNPs reached a genome-wide level of significance. The overall picture is nevertheless telling because most of the SNPs with the highest levels of association are in annotated genomic regions. Of those, the majority (9 out of 16) are located within the gene LOC100643741, whose product is annotated as mucin-12 (Fig 2). Another two SNPs are within LOC100645175, annotated as disks large 1 tumor suppressor protein, and one is similar to lachesin, an Ig-superfamily protein.
Mucin-12 is identified by homology with vertebrates (mammals); usually it encodes a membrane glycoprotein as a member of the mucin family. These are very ubiquitous proteins. In many organisms, mucins form a protective mucous layer on epithelial surfaces—including the gut—and also play a role in intra-cellular signalling [52–54]. In most insects, the gut is lined with the peritrophic membrane, which is thought to derive from mucus-forming mucins, and which acts as a protective barrier [52]. In Drosophila [55] or locusts [56], for example, mucins are also involved in development and organ morphogenesis, including the formation of the peritrophic membrane of the midgut. Moreover, in Drosophila, the build-up of gut mucus is affected by infection, and the renewal of the gut epithelium after an infection-induced damage is essential for further protection and tolerance to infection [57]. Mucins in the tsetse fly are involved in the production of the peritrophic matrix and play a role in protection against Trypanosoma infections [58].
In Trypanosoma cruzi (causing Chagas disease in humans), the infective stages selectively bind to gastric mucin, mediated by gp82, a stage-specific putative mucin-surface protein of the parasite itself [59–61]. Interestingly, gp63 is a surface protease of most trypanosomatids that is also involved in adhesion to the host cells [62], for example, in the closely related trypanosomatid, Leptomonas [63]. It is also considered a virulence factor of Crithidia mellificae and Lotmaria passim in honeybees [64], and is essential for the diversity of infections by T. cruzi [65]. The four gp63-variants found in C. bombi are very divergent and distinct from its close ally, C. expoeki.
Although there is no firm knowledge on the role of parasite surface proteases and host mucins in bumble bees, it is reasonable to assume that mucins are also important for the infection, establishment, and longer-term outcome of an infection by C. bombi. The sequence of B. terrestris mucin-12 contains ankyrin-repeats, which typically are involved in protein-protein interactions and are among the most common protein structures more generally. A closer look, however, suggests that this mucin-12 is probably not involved in the peritrophic matrix as it lacks mucin or chitin binding of the tsetse fly ortholog [66, 67]. Nevertheless, mucin-12 (now re-labelled as mucin-5AC in the NCBI database) may remain an interesting candidate in bumblebees, as Crithidia needs to attach with its flagellum to the gut epithelium to establish a successful infection [68], similar to some but not all trypanosomatids [69, 70].
Disks large 1 tumor suppressor protein (Table 3) is also a member of a common protein family and found in many organisms. A common function seems to be in embryonic neuronal development during embryonic stages. It also serves in cellular adhesion [71–73]. Again, nothing is known about its function in bees. However, with respect to infection by C. bombi, this gene remains an interesting candidate for the same reason (adherence to the gut wall) as mucin-12. Both, mucin-12 and disks 1 are furthermore known to have several splice variants (see the annotations for assembly Bter_1.0, available via Ensemble Metazoa) whose functions are not known. Interestingly, too, a genome-wide study of the adaptive evolution of bees found that disks large 1 tumor suppressor protein shows signs of positive selection in social bee clades, potentially indicating a role in defence against parasites or the evolution towards sociality [74]. Finally, lachesin that appears in Tables 3 and 5 is typically a cell surface recognition/adhesion immunoglobulin (Ig) protein involved in insect development [75], but also in wound repair [76], and presumably in immune defences [77].
Characteristically, the vast majority of SNPs found here are intronic, with few exceptions (see Tables 2–5). In some cases, intronic regions are known to affect the phenotype [78], or the affect the process of alternative splicing process [79]. Among those, non-coding RNAs may play an important role [80]. However, as yet, nothing is known about these processes in B. terrestris. Due to close proximity, the identified SNPs genetically are linked to the expressed sequences in these genomic regions, but nothing is known about these effects either. We also note that the analyses with the low-coverage approach may not be very suitable for a complex trait, or one that is based on differential gene expression rather than genetic variation in structural genes as may be the case in our study.
Microbiota
For the high-coverage study (8 x8 matrix) we screened the bees for their microbiota seven days post-parasite exposure. A total of 997 zOTUs entered the analyses. Not surprisingly, only half of them can only be identified at the genus level or above. To clarify the role of the microbiota further, we split the samples according to their zOTU composition using multidimensional scaling (see Fig 3), defining a ’resistant’ and a ’permissive’ group of bees (Fig 4). Regardless of the currently changing taxonomy, it seems clear from Table 6 that the bacterial genera Snodgrassella and Gilliamella are associated with a resistant phenotype, whereas susceptible phenotypes associate with Lacotbacillus and Bombilactobacillus, respectively. The taxa identified here overlap with those reported in earlier studies (Table 7). In particular, the more resistant phenotypes in our study (Table 6) as well as in previous studies (Table 7) are associated with Gilliamella [18, 19, 21], Snodgrassella [18]. These two taxa are representatives of a larger range of bacterial microbiota that shows activity against pathogens such as C. bombi but also against American foulbrood (Paenibacillus larvae) and others in honeybees [81]. Lactobacillus associates with higher infection (susceptible) in our study; the reverse is found by [21] or [23]. In addition, Bombiscardova has effects in other studies, being less abundant in resistant phenotypes (Table 7) [20]. Snodgrassella alvi is known to protect honeybees from infection by the opportunistic pathogen, Serratia marcescens [82].
Table 7. Studies on the protective effects of microbiota against Crithidia infections in bumble bees.
| Study type a | Finding b | Remarks | Source |
|---|---|---|---|
| Microbiota cross-transplanted. Bacteria by TRFLP-profiles. Assessed 7 d post-exposure. | Infection affected by donor microbiota. Major effects: Snodgrassella alvi, Gilliamella apicola, Lactobacillus. |
[17] | |
| Field-based experimental colonies. In situ OTU-composition and infection status. | High OTU diversity associated with higher infection intensity. Associated with lower infection: Gilliamella apicola, Snodgrassella alvi. |
Colonies under different food supply and immune-challenge treatments (with no effects). | [18] |
| in situ OTU-composition in bees from agricultural and semi-natural sites. Infections observed. Species: B. bimaculatus, B. griseocollis, B. imaptiens. | Higher richness of core bacteria associated with lower infection, but reverse for non-core bacteria. Associated with lower infection: Gilliamella; with higher infection: Acetobacteraceae, |
Core bacteria defined as taxa associated with Apis or Bombus. | [19] |
| 18 h post-exposure: (a) OTU composition. (b) Gene expression when microbiota is cross-transplanted (whole-genome transcripts). |
(a) Total of159 OTUs; 19 common OTUs account for 99% of reads.
|
Host colonies grouped as resistant or susceptible, depending on infection intensities after exposure. | [20] |
| in situ OTU-composition in bees from the wild and two habitats (urban, forest). Infections observed. | OTUs with suspected effect: Gilliamella, Snodgrassella. No association of OTU-diversity with infection. |
Urban microbiota dominated by core bacteria (i.e. those commonly associated with bees). Forest microbiota by environmental bacteria. | [101] |
| OTU-composition before and after (7 d post-exposure). | High pre-infection OTU diversity associated with higher infection intensity. OTUs with effects: Snodgrassella, Gilliamella, Lactobacillus, Orbus, Bombiscardovia. Post-infection OTU diversity not changed and not related to infection outcome. |
[22] | |
| Microbiota cross-transplanted from queens into naïve wild and commercial worker bees. Infection 7 d post-transplant, assessed 7 d post-exposure. Species: B. impatiens. |
High OTU diversity associated with lower infection intensity. Associated with lower infection: Apibacter, Lactobacillus Firm-5, Gilliamella. No association: Snodgrassella alvi. |
Subtle differences in microbiota affect infection. | [21] |
| Inhibition of Crithidia in culture, originating from wild bees in Vermont, Illinois. Bacteria isolated from European B. lapidarius. | Associated with lower infection: Lactobacillus (Firm-5) bombicola at low pH. | Resistance mediated by pH of the gut. pH affected by microbiota, especially Lactobacillus. | [23] |
a the cited studies used Bombus terrestris, except where mentioned. All studies by metagenomics, except where mentioned.
b likely taxonomic classification at the time of study (closest BLAST hits).
The corresponding GWAS-analyses for the two groups found many of the top 20 SNPs associated with the permissive group in regions with uncharacterized loci or non-coding regions (Table 4). For the resistant group, where 12 of 20 top SNPs were located in annotated regions, four SNPs located within an exon (Table 5). Interestingly, across analyses, the genomic region of mucin-12 (see Tables 3 and 5), disks large 1 tumor suppressor protein (Tables 2–4) and lachesin (Tables 3 and 5) appeared repeatedly. Nothing is known about the function of these genes in bumble bees. Yet, a possible significance of these genes for directly influencing infections by Crithidia, or indirectly by supporting microbial communities that protect against an infection is likely.
Published gene expression studies (Table 8) in this system (and elsewhere) tend to use early phases of infection to examine expression, typically 18 h to 48 h post-exposure. This time window covers the early responses against the infection, which, in bumble bees, extends over hours to a few days [83–85]. By contrast, GWAS typically addresses the ’static’ genetic architecture of the defence machinery, as no ongoing expression is measured. Moreover, in our study, GWAS is assessed by a phenotype indicating the long-term outcome of infection, i.e., infection intensity 7 d post-exposure. This time interval corresponds to the development of a full infection by C. bombi [86, 87] (doubling time of C. bombi in the host is estimated at 10 to 14 h [24]). So far, there seems to be no GWAS study that analyses long-term infection intensity as we do here. Rather, several GWAS studies in bumble bees address morphological traits, such as colouration [88, 89], or tongue length [90]. A direct comparison with other studies is therefore not yet feasible.
Table 8. Synopsis of results from various expression studies on Crithidia infections in bumble bees.
| Study type a | Finding, significant effects upon infection b | Remarks c | Source |
|---|---|---|---|
| Candidate gene expression. 24 h / 48 h post-exposure. |
Upregulation: abaecin, hymenoptaecin, defensin. | Clear effect of colony. GxG interaction (but not for abaecin). | [51] |
| Candidate gene expression. 10 d post-exposure. |
Upregulation: hemomucin, relish. No effect: MyD88, TEP |
4 candidate genes. | [102] |
| Candidate gene expression. Time series (1,2,3,4,5,24,26, 48 h) post-exposure. |
Upregulation (5 genes): IK2, peroxidase, calcineurin, Tamo-like, plexin A. No effect: vitellogenin, HDLBP, fat-spondin, cathespin, serpin, aconitase, Sur8, Hsp90, Ran-type molecule. |
14 candidate genes. Focus on temporal course of expression (from 1… 24 h post-infection). | [85] |
| Candidate gene expression. 18 h post-exposure. |
Upregulation: abaecin, lysozyme, serpin 27a. No effect (in most genes): hemomucin, relish, basket, defensin, hymenoptaecin, peroxidase, transferrin, ferritin, TEP A. |
28 candidate genes. Bees from two localities; slight differences found. Effects of colony pervasive. | [103] |
| Candidate gene expression. 18 h post-exposure. |
Upregulation: BGRP1, hopscotch, abaecin, hymenoptaecin, ferritin, lysozyme 3, peroxiredoxin 5, cytochrome P450. Downregulation: TEP A. |
26 candidate genes. Upregulation in many receptor and effector genes, partly in signalling and metabolic genes. Strong effect of parasite clone (strain). Effect of colony and colony x strain interaction (GxG). Expression of genes correlated. | [26] |
| Whole-genome transcripts. 18 h post-exposure. |
Effect of GxG exposure: apidaecin, abaecin, estersae FF4, limkain-b1, LRR GPCR4, SPN3. Effect of infecting strain: putative AMP, exonuclease, synacptic vesc. glycoprotein, MELK, cytochrome P450, apidemin 2, maelstrom, Spatzle 4, four uncharacterized loci. |
Focus on expression effects under GxG Interaction | [14] |
| Whole-genome transcripts. 24h / 48 h post-exposure. |
Differential expression (up/down) according to broad function for: peritrophic membrane (fibrillin, chitinases, peritrophin), receptors (Dscam variants, GNBPs,), serpins (proteases, inhibitors), signalling (Spaetzle, myD88, dorsal,…), effectors (defensin, hymenoptaecin, apidaecin, hemolectin, TPX4, eater,…), metabolism (cytochrom P450), and with various functions (royal jelly protein,…). | A total of 489 genes affected. 591 genes affected specifically under a GxG interaction. 24 h post-infection: 165 transcripts upregulated, 324 downregulated. Several genes are alternatively spliced and expressed. |
[93] |
a The cited studies used Bombus terrestris.
b For specification of genes, see source.
c GxG interaction refers to host genotype x parasite genotype differential expression.
Interestingly, gene expression studies of the Bombus-Crithidia system typically identify changes in the expression of anti-microbial peptides, such as abaecin, apidaecin, hymenoptaecin, or defensin, but also genes expressed for the recognition of parasites (GNBP, Dscam), for immune signalling (Spaetzle, myD88), or transcription factors and activators (relish, dorsal; Table 8). As mentioned above, these genes are involved in the early response to infection when the expression is also measured. By contrast, our ’static’ GWAS uncovers a more permanent genetic trace, where mucin-12 ranks prominently.
For the longer-term outcome of an infection, the microbiota is an essential determinant. Among the critical taxa identified here (Table 6), Snodgrassella and Gilliamella line the ileum as a biofilm [91] and are suspected to competitively inhibit the attachment of trypanosomatids to the gut wall [92]. It is therefore conceivable that the mucin-microbiota-Crithidia nexus is the key to the establishment and control of infection in Bombus after the early responses (characterized by the expression anti-microbial peptides, reactive oxygen species, or changes in metabolism) have run their course. Indeed, the possible functions of the most prominent genes found here seems to revolve around cell surface and binding properties. Adhesion to host cells is a key requirement for C. bombi to successfully infect [68]. Similarly, the parasite-genotype x host-genotype interaction signature can be retrieved for gene expression in the early responses [14, 51, 93], but was no longer present in our static genetic analyses by GWAS. At the same time, the rapidly expressed anti-microbial peptides (AMPs) are often evolutionarily conservative and evolve slower than other immune genes [94]. However, they can be expressed differently and hence used as a tailor-made cocktail of compounds, whose effects on infecting parasites, such as C. bombi, potentiates and depends on the quantitative combination of the single AMPs [95, 96]. The importance of this flexible strategy is plausible by the need for a quick and effective response as early as possible in the course of an infection [97]. This differential use of AMPs is what is likely visible in the early responses, and emerges in form of a variable expression profile (Table 8). By contrast, the main process in the longer-term responses may be binding of parasite cells to host cell surfaces, where the more constitutive protective mucus layer and the established microbiota become the key players. At the time being, this remains a hypothetical conjecture but opens interesting perspectives.
The interaction matrix
We did not find a parasite-genotype x host-genotype interaction effect when statistically analysing the infection phenotype (i.e., infection intensity) in the full 20 x 20 matrix, whereas the main effects of colony (host genotypic background) and parasite strain were highly significant (Table E1 in S12 File). We propose that the pattern of losing an interaction effect as more colony x strain combinations are included in a matrix (Fig 6) likely reflects a biological effect rather than a statistical artefact. In a basic scenario, when there is a high degree of specificity in the host-parasite interaction, the interaction effect should indeed become more prominent as more combinations are tested, i.e., when the interaction matrix has a larger envelope (more host-genotypes by more parasite-genotypes). However, this is only true when the specificities are ’homogeneous in kind’, that is, they function in the same way; in the extreme case, when different colonies only succumb to one, but different parasite strain each. However, there are at least two elements that change this basic scenario.
Firstly, the interaction of a particular colony with a particular parasite strain is specific in the sense that, primarily, the trajectory of the developing infection in host varies among hosts, among infecting parasite strains and their combination. This can be conceptually illustrated by the infection trajectory in the host’s ’disease space’—which, for example, may be defined by the axes of host health status and infection intensity [98]. As a result, the outcome—that is, the infection intensity as measured here—might be extremely variable due to, for instance, different defence mechanisms becoming activated along the specific trajectory. To put it figuratively, in a given host-parasite combination A, the infection may result in upregulated host metabolic activity, whereas in a different host-parasite combination B, infection may instead alter the expression of anti-microbial peptides. This amounts to the scenario that multiple or alternative genetic pathways are recruited in different host x parasite combinations. Such phenomena have been documented in our system [14, 26]. As a result, the trajectories of health and disease may unfold in very different ways and, therefore, generate very non-linear, non-homogeneous infection outcomes (with their associated genetic architecture) despite an otherwise ordered matrix of host-parasite interactions. Secondly, in the Bombus-Crithidia system, the situation is also different from the basic scenario because the colony specificities act more like a quantitative filter, i.e., colonies are susceptible to different range of parasite strains; for example, a colony may be susceptible to few or to many different strains on top of being susceptible to some more than to others. Note that interaction effects still exist—a particular colony will interact with a particular parasite strain in specific ways, different from any other combination. It is just that the above mentioned deviations from the basic scenario eventually will undermine the statistical interaction effect as more colonies and parasite strains are included in the analysed interaction matrix.
Conclusion
Any host colony is typically exposed to a large number of parasite strains over its life, e.g. [99, 100], which is the relevant level of selection in its environment (although we put aside the common and important situation of simultaneous infections with multiple parasite genotypes [24]). In this situation, averaged over the life span of colonies, the specific host-parasite interaction effects add to eventual fitness and, for example, support a role of rapidly recruited anti-microbial peptides (e.g., as revealed in expression studies). At the same time, the basic overall genetic architecture of a colony is associated with a general susceptibility to infection or the maintenance of the microbiota (and can be identified by GWAS). Based on our results, we propose that, for this effect, parasite adhesion to host cell surfaces may be the key process.
Supporting information
The file contains data from all of the N = 1,200 bees that were tested.
(XLSX)
The file contains the top 20 SNPs associated with infection intensity in the 20 x 20 matrix.
(XLSX)
The file contains the top 20 SNPs associated with infection intensity in the 8 x 8 matrix, using sqrt-transformed phenotype data, corrected for the strain effect, and pruned from linkage disequilibria LD > 0.75.
(XLSX)
The file contains the detailed report on the Amplicon sequencing procedure.
(PDF)
The file contains data from the top 20 SNPs associated with Axis.1 of the Multi-Dimensional Scaling (MDS) of zOTUs. Values for Axis.1 were LD-pruned with LD > 0.75.
(XLSX)
The file contains data from the top 20 SNPs associated with Axis.2 of MDS-ordination of OTUs. Values for Axis.2 were LD-pruned with LD > 0.75.
(XLSX)
The file contains data from the top 20 SNPs associated with cases where Axis.1 < 0 (’permissive’ group, from MDS-ordination of zOTUs. Values for Axis.1 corrected for structure, sqrt-transformed, and LD-pruned.
(XLSX)
The file contains data from the top 20 SNPs associated with cases where Axis.1 ≥ 0 (’resistant group’), from MDS-ordination of zOTUs. Values for Axis.1 corrected for structure, sqrt-transformed, and LD-pruned.
(XLSX)
The file contains the nucleotide sequences of all identified zOTUs in the project.
(FA)
This file contains the technical report for the sequencing of the bacterial metagenome.
(TXT)
This file contains a list of all samples from which the bacterial metagenome was typed and deposited with ENA (accession PRJEB52013). We used universal primers for the V3/V4 region of 16S RNA. A key to sample numbering is given.
(PDF)
File contains several sections. Methods: assessing infection intensity, selecting the 8x8 sub-matrix. Basic data: explanations, lists of colonies and parasite strains. Variation of infection intensity across colonies and strains. Explanations and supporting results for data in Appendices A to E.
(PDF)
Acknowledgments
We thank Katia Pfrunder-Cardoso for help in experiments and in the molecular work. We thank the staff of the Genetic Diversity Centre (GDC) at ETH Zurich for help with the genetic analyses. Parts of the computing was done with the high-performance cluster Euler at ETH Zürich. We thank Claire Rose and Alvaro Acosta-Serrano (Liverpool School of Tropical Medicine) for their expert assistance in comparing B. terrestris’ mucin-12 to mucins found in the tsetse fly peritrophic matrix.
Data Availability
All raw data, including the metagenomic data for the microbiota analyses, are deposited in the EMBL-EBI European Nucleotide Archive (ENA) repository (http://www.ebi.ac.uk/ena), under the study accession PRJEB52013.
Funding Statement
The work was benefitted from grants by the Swiss National Science Foundation (grant 31003A_146467) and the ERC (grant 268853 RESIST) to PSH. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
References
- 1.Coico R, Sunshine G. Immunology: a short course—7th ed. Hoboken, New Jersey: John Wiley & Sons; 2015. [Google Scholar]
- 2.Dhanda SK, Mahajan S, Paul S, Yan Z, Kim H, Jespersen MC, et al. IEDB-AR: immune epitope database—analysis resource in 2019. Nucleic Acids Res. 2019;47(W1): W502–W506. doi: 10.1093/nar/gkz452 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Kaur D, Patiyal S, Sharma N, Usmani SS, Raghava GPS. PRRDB 2.0: a comprehensive database of pattern-recognition receptors and their ligands. Database. 2019;2019: baz076. doi: 10.1093/database/baz076 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Schneider D, Ayres J. Two ways to survive infection: what resistance and tolerance can teach us about treating infectious diseases. Nat Rev Immunol. 2008;8: 889–895. doi: 10.1038/nri2432 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Routtu J, Ebert D. Genetic architecture of resistance in Daphnia hosts against two species of host-specific parasites. Heredity. 2015;114: 241–248. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Hall MD, Routtu J, Ebert D. Dissecting the genetic architecture of a stepwise infection process. Mol Ecol. 2019;28(17): 3942–3957. doi: 10.1111/mec.15166 [DOI] [PubMed] [Google Scholar]
- 7.Bento G, Routtu J, Fields PD, Bourgois Y, Du Pasquier L, Ebert D. The genetic basis of resistance and matching-allele interactions of a host-parasite system: The Daphnia magna-Pasteuria ramosa model. PLoS Genet. 2017;13(2): e1006596. doi: 10.1371/journal.pgen.1006596 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Tognazzo M, Schmid-Hempel R, Schmid-Hempel P. Probing mixed-genotype infections II: High multiplicity in natural infections of the trypanosomatid, Crithidia bombi, in its host, Bombus spp. PLoS One. 2012;7(11): e49137. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Schmid-Hempel P, Schmid-Hempel R, Wilfert L. Pollinator diseases—the case of bumblebees. In: Wilson K, Fenton A, Tompkins D, editors. Wildlife Disease Ecology. Cambridge: Cambridge University Press; 2019. p. 3–31. [Google Scholar]
- 10.Baer B, Schmid-Hempel P. Bumblebee workers from different sire groups vary in susceptibility to parasite infection. Ecol Lett. 2003;6: 106–110. [Google Scholar]
- 11.Schmid-Hempel P, Puhr K, Kruger N, Reber C, Schmid-Hempel R. Dynamic and genetic consequences of variation in horizontal transmission for a microparasitic infection. Evolution. 1999;53(2): 426–434. doi: 10.1111/j.1558-5646.1999.tb03778.x [DOI] [PubMed] [Google Scholar]
- 12.Ulrich Y, Sadd B, Schmid-Hempel P. Strain filtering and transmission of a mixed infection in a social insect. J Evol Biol. 2011;24(2): 354–362. doi: 10.1111/j.1420-9101.2010.02172.x [DOI] [PubMed] [Google Scholar]
- 13.Deshwal S, Mallon EB. Antimicrobial peptides play a functional role in bumblebee anti-trypanosome defense. Dev Comp Immunol. 2014;42(2): 240–243. doi: 10.1016/j.dci.2013.09.004 [DOI] [PubMed] [Google Scholar]
- 14.Barribeau SM, Sadd BM, du Plessis L, Schmid-Hempel P. Gene expression differences underlying genotype-by-genotype specificity in a host-parasite system. Proc Natl Acad Sci USA. 2014;111: 3496–3501. doi: 10.1073/pnas.1318628111 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Korneliussen TS, Albrechtsen A, Nielsen R. ANGSD: Analysis of next generation sequencing data. BMC Bioinformatics. 2014;15(1): 356. doi: 10.1186/s12859-014-0356-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Koch H, Schmid-Hempel P. Socially transmitted gut microbiota protect bumble bees against an intestinal parasite. Proc Natl Acad Sci USA. 2011;108(48): 19288–19292. doi: 10.1073/pnas.1110474108 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Koch H, Schmid-Hempel P. Gut microbiota instead of host genotype drive the specificity in the interaction of a natural host-parasite system. Ecol Lett. 2012;15: 1095–1103. doi: 10.1111/j.1461-0248.2012.01831.x [DOI] [PubMed] [Google Scholar]
- 18.Koch H, Cisarovsky G, Schmid-Hempel P. Ecological effects on gut bacterial communities in wild bumble bee colonies. J Anim Ecol. 2012. doi: 10.1111/j.1365-2656.2012.02004.x [DOI] [PubMed] [Google Scholar]
- 19.Cariveau DP, Powell JE, Koch H, Winfree R, Moran NA. Variation in gut microbial communities and its association with pathogen infection in wild bumble bees (Bombus). The ISME Journal. 2014;8(12): 2369–2379. doi: 10.1038/ismej.2014.68 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Näpflin K, Schmid-Hempel P. Immune response and gut microbial community structure in bumblebees after microbiota transplants. Proc R Soc Lond B. 2016;283: 20160312. doi: 10.1098/rspb.2016.0312 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Mockler BK., Kwong WK., Moran NA., Koch H, Drake H, L. Microbiome structure influences infection by the parasite Crithidia bombi in bumble bees. Appl Environ Microbiol. 2018;84(7): e02335–02317. doi: 10.1128/AEM.02335-17 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Näpflin K, Schmid-Hempel P. High gut microbiota diversity provides lower resistance against infection by an intestinal parasite in bumblebees. Am Nat. 2018;192(2): 131–141. doi: 10.1086/698013 [DOI] [PubMed] [Google Scholar]
- 23.Palmer-Young EC, Raffel TR, McFrederick QS. pH-mediated inhibition of a bumble bee parasite by an intestinal symbiont. Parasitology. 2019;146(3): 380–388. Epub 2018/09/24. doi: 10.1017/S0031182018001555 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Salathé R, Tognazzo M, Schmid-Hempel R, Schmid-Hempel P. Probing mixed-genotype infections I: Extraction and cloning of infections from hosts of the trypanosomatid Crithidia bombi. PLoS One. 2012;7(11): e49046. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Gerasimov E, Zemp N, Schmid-Hempel R, Schmid-Hempel P, Yurchenko V. Genomic variation among strains of Crithidia bombi and C. expoeki. mSphere. 2019;4(5): e00482–00419. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Barribeau SM, Schmid-Hempel P. Qualitatively different immune response of the bumblebee host, Bombus terrestris, to infection by different genotypes of the trypanosome gut parasite, Crithidia bombi. Infect, Genet Evol. 2013;20: 249–256. [DOI] [PubMed] [Google Scholar]
- 27.Therkildsen NO, Palumbi SR. Practical low-coverage genomewide sequencing of hundreds of individually barcoded samples for population and evolutionary genomics in nonmodel species. Mol Ecol Resour. 2017;17(2): 194–208. doi: 10.1111/1755-0998.12593 [DOI] [PubMed] [Google Scholar]
- 28.Meisner J, Albrechtsen A. Inferring population structure and admixture proportions in low-depth NGS data. Genetics. 2018;210(2): 719–731. doi: 10.1534/genetics.118.301336 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29.Kang HM, Zaitlen NA, Wade CM, Kirby A, Heckerman D, Daly MJ, et al. Efficient control of population structure in model organism association mapping. Genetics. 2008;178(3): 1709–1723. doi: 10.1534/genetics.107.080101 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Gogarten SM, Bhangale T, Conomos MP, Laurie CA, McHugh CP, Painter I, et al. GWASTools: an R/Bioconductor package for quality control and analysis of genome-wide association studies. Bioinformatics. 2012;28(24): 3329–3331. doi: 10.1093/bioinformatics/bts610 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Zhou X, Stephens M. Genome-wide efficient mixed-model analysis for association studies. Nat Genet. 2012;44(7): 821–824. doi: 10.1038/ng.2310 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Näpflin K, Schmid-Hempel P. Host effects on microbiota community assembly. J Anim Ecol. 2017;87: 331–340. doi: 10.1111/1365-2656.12768 [DOI] [PubMed] [Google Scholar]
- 33.Edgar RC, Flyvbjerg H. Error filtering, pair assembly and error correction for next-generation sequencing reads. Bioinformatics. 2015;31(21): 3476–3482. doi: 10.1093/bioinformatics/btv401 [DOI] [PubMed] [Google Scholar]
- 34.Edgar RC. UNOISE2: improved error-correction for Illumina 16S and ITS amplicon sequencing. bioRxiv. 2016. doi: 10.1101/081257 [DOI] [Google Scholar]
- 35.Needham DM, Sachdeva R, Fuhrman JA. Ecological dynamics and co-occurrence among marine phytoplankton, bacteria and myoviruses shows microdiversity matters. The ISME Journal. 2017;11(7): 1614–1629. doi: 10.1038/ismej.2017.29 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36.Xue Z, Kable Mary E, Marco Maria L. Impact of DNA sequencing and analysis methods on 16S rRNA genebacterial xommunity analysis of dairy products. mSphere. 2018;3(5): e00410–00418. doi: 10.1128/mSphere.00410-18 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Prodan A, Tremaroli V, Brolin H, Zwinderman AH, Nieuwdorp M, Levin E. Comparing bioinformatic pipelines for microbial 16S rRNA amplicon sequencing. PLoS One. 2020;15(1): e0227434. doi: 10.1371/journal.pone.0227434 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Song K, Li L, Zhang G. Coverage recommendation for genotyping analysis of highly heterologous species using next-generation sequencing technology. Sci Rep. 2016;6(1): 35736. doi: 10.1038/srep35736 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39.Salter SJ, Cox MJ, Turek EM, Calus ST, Cookson WO, Moffatt MF, et al. Reagent and laboratory contamination can critically impact sequence-based microbiome analyses. BMC Biol. 2014;12(1): 87. doi: 10.1186/s12915-014-0087-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Bunge J, Willis A, Walsh F. Estimating the number of species in microbial diversity studies. Annu Rev Stat Appl. 2014;1(1): 427–445. [Google Scholar]
- 41.Willis AD. Rarefaction, alpha diversity, and statistics. Front Microbiol. 2019;10(2407). doi: 10.3389/fmicb.2019.02407 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Willis A, Bunge J. Estimating diversity via frequency ratios. Biometrics. 2015;71(4): 1042–1049. doi: 10.1111/biom.12332 [DOI] [PubMed] [Google Scholar]
- 43.Willis AD, Martin BD. DivNet: Estimating diversity in networked communities. bioRxiv. 2018: 305045. doi: 10.1101/305045 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Cornet L, Cleenwerck I, Praet J, Leonard Raphaël R, Vereecken Nicolas J, Michez D, et al. Phylogenomic analyses of Snodgrassella isolates from honeybees and bumblebees reveal taxonomic and functional diversity. mSystems. 2022;7(3): e01500–01521. doi: 10.1128/msystems.01500-21 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 45.Praet J, Cnockaert M, Meeus I, Smagghe G, Vandamme P. Gilliamella intestini sp. nov., Gilliamella bombicola sp. nov., Gilliamella bombi sp. nov. and Gilliamella mensalis sp. nov.: Four novel Gilliamella species isolated from the bumblebee gut. Syst Appl Microbiol. 2017;40(4): 199–204. doi: 10.1016/j.syapm.2017.03.003 [DOI] [PubMed] [Google Scholar]
- 46.Olofsson TC, Alsterfjord M, Nilson B, Butler È, Vásquez A. Lactobacillus apinorum sp. nov., Lactobacillus mellifer sp. nov., Lactobacillus mellis sp. nov., Lactobacillus melliventris sp. nov., Lactobacillus kimbladii sp. nov., Lactobacillus helsingborgensis sp. nov. and Lactobacillus kullabergensis sp. nov., isolated from the honey stomach of the honeybee Apis mellifera. Int J Syst Evol Microbiol. 2014;64(Pt 9): 3109–3119. Epub 2014/06/20. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 47.Heo J, Kim S-J, Kim J-S, Hong S-B, Kwon S-W. Comparative genomics of Lactobacillus species as bee symbionts and description of Lactobacillus bombintestini sp. nov., isolated from the gut of Bombus ignitus. J Microbiol. 2020;58(6): 445–455. doi: 10.1007/s12275-020-9596-3 [DOI] [PubMed] [Google Scholar]
- 48.Kang J-P, Huo Y, Hoang V-A, Yang D-U, Yang D-C, Kang S-C. Bombilactobacillus apium sp. nov., isolated from the gut of honeybee (Apis cerana). Arch Microbiol. 2021;203(5): 2193–2198. doi: 10.1007/s00203-021-02249-y [DOI] [PubMed] [Google Scholar]
- 49.Zheng J, Wittouck S, Salvetti E, Franz CMAP, Harris HMB, Mattarelli P, et al. A taxonomic note on the genus Lactobacillus: Description of 23 novel genera, emended description of the genus Lactobacillus Beijerinck 1901, and union of Lactobacillaceae and Leuconostocaceae. Int J Syst Evol Microbiol. 2020;70(4): 2782–2858. doi: 10.1099/ijsem.0.004107 [DOI] [PubMed] [Google Scholar]
- 50.Schmid-Hempel P, Reber Funk C. The distribution of genotypes of the trypanosome parasite, Crithidia bombi, in populations of its host, Bombus terrestris. Parasitology. 2004;129: 147–158. [DOI] [PubMed] [Google Scholar]
- 51.Riddell CR, Admas S, Schmid-Hempel P, Mallon EB. Differential expression of immune defences underlies specific host-parasite interactions in insects. PLoS One. 2009;4(10): e7621. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Dias RO, Cardoso C, Pimentel AC, Damasceno TF, Ferreira C, Terra WR. The roles of mucus-forming mucins, peritrophins and peritrophins with mucin domains in the insect midgut. Insect Mol Biol. 2018;27(1): 46–60. doi: 10.1111/imb.12340 [DOI] [PubMed] [Google Scholar]
- 53.Jing T, Wang F, Qi F, Wang Z. Insect anal droplets contain diverse proteins related to gut homeostasis. BMC Genomics. 2018;19(1): 784. doi: 10.1186/s12864-018-5182-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Pinzón Martín S, Seeberger PH, Varón Silva D. Mucins and pathogenic mucin-like molecules are immunomodulators during infection and targets for diagnostics and vaccines. Front Chem. 2019;7(710). doi: 10.3389/fchem.2019.00710 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 55.Syed ZA, Härd T, Uv A, van Dijk-Härd IF. A potential role for Drosophila mucins in development and physiology. PLoS One. 2008;3(8): e3041. doi: 10.1371/journal.pone.0003041 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 56.Zhao X, Zhang J, Yang J, Niu N, Zhang J, Yang Q. Mucin family genes are essential for the growth and development of the migratory locust, Locusta migratoria. Insect Biochem Mol Biol. 2020;123: 103404. doi: 10.1016/j.ibmb.2020.103404 [DOI] [PubMed] [Google Scholar]
- 57.Buchon N, Broderick NA, Lemaitre B. Gut homeostasis in a microbial world: insights from Drosophila melanogaster. Nat Rev Microbiol. 2013;11: 615–626. [DOI] [PubMed] [Google Scholar]
- 58.Kozak RP, Mondragon-Shem K, Williams C, Rose C, Perally S, Caljon G, et al. Tsetse salivary glycoproteins are modified with paucimannosidic N-glycans, are recognised by C-type lectins and bind to trypanosomes. PLoS Negl Trop Dis. 2021;15(2): e0009071. doi: 10.1371/journal.pntd.0009071 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 59.Neira I, Silva FA, Cortez M, Yoshida N. Involvement of Trypanosoma cruzi metacyclic trypomastigote surface molecule gp82 in adhesion to gastric mucin and invasion of epithelial cells. Infect Immun. 2003;71: 557–561. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 60.Staquicini DI, Martins RM, Macedo S, Sasso GRS, Atayde VD, Juliano MA, et al. Role of gp82 in the selective binding to gastric mucin during oral infection with Trypanosoma cruzi. PLoS Negl Trop Dis. 2010;4(3): e613. doi: 10.1371/journal.pntd.0000613 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 61.Bayer-Santos E, Cunha-e-Silva NL, Yoshida N, Franco da Silveira J. Expression and cellular trafficking of GP82 and GP90 glycoproteins during Trypanosoma cruzi metacyclogenesis. Parasites & Vectors. 2013;6(1): 127. doi: 10.1186/1756-3305-6-127 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 62.Yao C. Major surface protease of Trypanosomatids: one size fits all?. Infect Immun. 2010;78(1): 22–31. doi: 10.1128/IAI.00776-09 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 63.Pereira FM, Bernardo PS, Dias Junior PFF, Silva BA, Romanos MTV, d’Avila-Levy CM, et al. Differential influence of gp63-like molecules in three distinct Leptomonas species on the adhesion to insect cells. Parasitol Res. 2009;104: 347–353. [DOI] [PubMed] [Google Scholar]
- 64.Ravoet J, Schwarz RS, Descamps T, Yañez O, Tozkar CO, Martin-Hernandez R, et al. Differential diagnosis of the honey bee trypanosomatids Crithidia mellificae and Lotmaria passim. J Invertebr Pathol. 2015;130: 21–27. [DOI] [PubMed] [Google Scholar]
- 65.Llewellyn MS, Messenger LA, Luquetti AO, Garcia L, Torrico F, Tavares SBN, et al. Deep sequencing of the Trypanosoma cruzi GP63 surface proteases reveals diversity and diversifying selection among chronic and congenital Chagas disease patients. PLoS Negl Trop Dis. 2015;9(4): e0003458. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 66.Rose C, Belmonte R, Armstrong SD, Molyneux G, Haines LR, Lehane MJ, et al. An investigation into the protein composition of the Teneral Glossina morsitans morsitans peritrophic matrix. PLoS Negl Trop Dis. 2014;8(4): e2691. doi: 10.1371/journal.pntd.0002691 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 67.Rose C, Casas-Sánchez A, Dyer NA, Solórzano C, Beckett AJ, Middlehurst B, et al. Trypanosoma brucei colonizes the tsetse gut via an immature peritrophic matrix in the proventriculus. Nat Microbiol. 2020;5(7): 909–916. doi: 10.1038/s41564-020-0707-z [DOI] [PubMed] [Google Scholar]
- 68.Koch H, Woodward J, Langat MK, Brown MJF, Stevenson PC. Flagellum removal by a nectar metabolite inhibits infectivity of a bumblebee parasite. Curr Biol. 2019;29(20): 3494–3500.e3495. doi: 10.1016/j.cub.2019.08.037 [DOI] [PubMed] [Google Scholar]
- 69.Brooker BE. Flagellar attachment and detachment of Crithidia fasciculata to the gut wall of Anopheles gambiae. Protoplasma. 1971;73: 191–202. [DOI] [PubMed] [Google Scholar]
- 70.Killick-Kendrick R, Molyneux DH, Ashford RW. Leishmania in phlebotomid sandflies. I. Modifications of the flagellum associated with attachment to the mid-gut and oesophageal valve of the sandfly. Proc R Soc Lond B. 1974;187: 409–419. [DOI] [PubMed] [Google Scholar]
- 71.Woods DF, Bryant PJ. The discs-large tumor suppressor gene of Drosophila encodes a guanylate kinase homolog localized at septate junctions. Cell. 1991;66(3): 451–464. Epub 1991/08/09. doi: 10.1016/0092-8674(81)90009-x [DOI] [PubMed] [Google Scholar]
- 72.Bilder D, Li M, Perrimon N. Cooperative regulation of cell polarity and growth by Drosophila tumor suppressors. Science. 2000;289(5476): 113–116. doi: 10.1126/science.289.5476.113 [DOI] [PubMed] [Google Scholar]
- 73.Mendoza C, Olguín P, Lafferte G, Thomas U, Ebitsch S, Gundelfinger ED, et al. Novel isoforms of Dlg are fundamental for neuronal development in Drosophila. J Neurosci. 2003;23(6): 2093–2101. doi: 10.1523/jneurosci.23-06-02093.2003 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 74.Mee L, Barribeau SM. Immunity vs sociality: Adaptive evolution tests suggest social lifestyle exerts greater selection pressures than host-pathogen coevolution in the bees. bioRxiv. 2021: 2021.2009.2007.459261. doi: 10.1101/2021.09.07.459261 [DOI] [Google Scholar]
- 75.Boyan G, Ehrhardt E. Epithelial domains and the primordial antennal nervous system of the embryonic grasshopper Schistocerca gregaria. Invert Neurosci. 2020;20(2): 6. doi: 10.1007/s10158-020-0240-z [DOI] [PubMed] [Google Scholar]
- 76.Johnston PR, Rolff J. Immune- and wound-dependent differential gene expression in an ancient insect. Dev Comp Immunol. 2013: 10–14. doi: 10.1016/j.dci.2013.01.012 [DOI] [PubMed] [Google Scholar]
- 77.Javed MA, Coutu C, Theilmann DA, Erlandson MA, Hegedus DD. Proteomics analysis of Trichoplusia ni midgut epithelial cell brush border membrane vesicles. Insect Sci. 2019;26(3): 424–440. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 78.Seo S, Takayama K, Uno K, Ohi K, Hashimoto R, Nishizawa D, et al. Functional analysis of deep intronic SNP rs13438494 in intron 24 of PCLO gene. PLoS One. 2013;8(10): e76960. doi: 10.1371/journal.pone.0076960 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 79.Chiang H-L, Wu J-Y, Chen Y-T. Identification of functional single nucleotide polymorphisms in the branchpoint site. Human Genom. 2017;11(1): 27. doi: 10.1186/s40246-017-0122-6 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 80.Moure UAE, Tan T, Sha L, Lu X, Shao Z, Yang G, et al. Advances in the immune regulatory role of non-coding RNAs (mirRNAa and lncRNAs) in insect-pathogen interactions. Front Immunol. 2022;13: 856457. Epub 2022/04/26. doi: 10.3389/fimmu.2022.856457 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 81.Praet J, Parmentier A, Schmid-Hempel R, Meeus I, Smagghe G, Vandamme P. Large-scale cultivation of the bumblebee gut microbiota reveals an underestimated bacterial species diversity capable of pathogen inhibition. Environ Microbiol. 2018;20(1): 214–227. doi: 10.1111/1462-2920.13973 [DOI] [PubMed] [Google Scholar]
- 82.Horak RD, Leonard SP, Moran NA. Symbionts shape host innate immunity in honeybees. Proc R Soc Lond B. 2020;287(1933): 20201184. doi: 10.1098/rspb.2020.1184 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 83.Korner P, Schmid-Hempel P. In vivo dynamics of an immune response in the bumble bee Bombus terrestris. J Invertebr Pathol. 2004;87: 59–66. [DOI] [PubMed] [Google Scholar]
- 84.Erler S, Popp. M, Latttorff HMG. Dynamics of immune system gene expression upon bacterial challenge and wounding in a social insect (Bombus terrestris). PLoS One. 2011;6(3): e18126. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 85.Riddell CE, Sumner S, Adams S, Mallon EB. Pathways to immunity: temporal dynamics of the bumblebee (Bombus terrestris) immune response against a trypanosome gut parasite. Insect Mol Biol. 2011;20(4): 529–540. [DOI] [PubMed] [Google Scholar]
- 86.Schmid-Hempel P, Schmid-Hempel R. Transmission of a pathogen in Bombus terrestris, with a note on division of labour in social insects. Beh Ecol Sociobiol 1993;33: 319–327. [Google Scholar]
- 87.Otterstatter MC, Thomson JD. Within-host dynamics of an intestinal pathogen of bumble bees. Parasitology. 2006;133: 749–761. doi: 10.1017/S003118200600120X [DOI] [PubMed] [Google Scholar]
- 88.Hines HM, Rahman SR. Evolutionary genetics in insect phenotypic radiations: the value of a comparative genomic approach. Curr Opin Insect Sci. 2019;36: 90–95. doi: 10.1016/j.cois.2019.08.013 [DOI] [PubMed] [Google Scholar]
- 89.Rahman SR, Cnaani J, Kinch LN, Grishin NV, Hines HM. A combined RAD-Seq and WGS approach reveals the genomic basis of yellow color variation in bumble bee Bombus terrestris. Sci Rep. 2021;11(1): 7996. doi: 10.1038/s41598-021-87194-y [DOI] [PMC free article] [PubMed] [Google Scholar]
- 90.Christmas MJ, Jones JC, Olsson A, Wallerman O, Bunikis I, Kierczak M, et al. A genomic and morphometric analysis of alpine bumblebees: ongoing reductions in tongue length but no clear genetic component. bioRxiv. 2021: 2021.2003.2017.435598. doi: 10.1111/mec.16291 [DOI] [PubMed] [Google Scholar]
- 91.Engel P, Martinson VG, Moran NA. Functional diversity within the simple gut microbiota of the honey bee. Proc Natl Acad Sci USA. 2012;109(27): 11002–11007. doi: 10.1073/pnas.1202970109 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 92.Schwarz RS, Bauchan GR, Murphy CA, Ravoet J, de Graaf DC, Evans JD. Characterization of two species of Trypanosomatidae from the honey bee Apis mellifera: Crithidia mellificae Langridge and McGhee, and Lotmaria passim n. gen., n. sp. J Eukaryot Microbiol. 2015;62: 567–583. [DOI] [PubMed] [Google Scholar]
- 93.Riddell CE, Lobaton Garces JD, Adams S, Barribeau SM, Twell D, Mallon EB. Differential gene expression and alternative splicing in insect immune specificity. BMC Genomics. 2014;15: 1031. doi: 10.1186/1471-2164-15-1031 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 94.Rolff J, Schmid-Hempel P. Perspectives on the evolutionary ecology of arthropod antimicrobial peptides Phil Trans R Soc Lond B. 2016;371: 20150297. doi: 10.1098/rstb.2015.0297 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 95.Rahnamaeian M, Cytryńska M, Zdybicka-Barabas A, Dobslaff K, Wiesner J, Twyman RM, et al. Insect antimicrobial peptides show potentiating functional interactions against Gram-negative bacteria. Proc R Soc Lond B. 2015;282: 20150293. doi: 10.1098/rspb.2015.0293 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 96.Marxer M, Vollenweider V, Schmid-Hempel P. Insect antimicrobial peptides act synergistically to inhibit a trypanosome parasite. Phil Trans R Soc Lond B. 2016;371: 20150302. doi: 10.1098/rstb.2015.0302 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 97.Duneau D, Ferdy J-B, Revah J, Kondolf H, Ortiz GA, Lazzaro BP, et al. Stochastic variation in the initial phase of bacterial infection predicts the probability of survival in D. melanogaster. eLife. 2017;6: e28298. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 98.Torres BY, Oliveira JHM, Tate Ann Thomas, Rath P, Cumnock K, Schneider DS. Tracking resilience to infections by mapping disease space. PLoS Biol. 2016;14(4): e1002436. doi: 10.1371/journal.pbio.1002436 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 99.Salathé R, Schmid-Hempel P. Genotypic structure of a multi-host bumblebee parasite suggests a major role for ecological niche overlap. PLoS One. 2011;6(8): e22054. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 100.Ruiz-Gonzalez MX, Bryden J, Moret Y, Reber-Funk C, Schmid-Hempel P, Brown MJF. Dynamic transmission, host quality and population structure in a multi-host parasite of bumble bees. Evolution. 2012;66: 3052–3066. doi: 10.1111/j.1558-5646.2012.01655.x [DOI] [PubMed] [Google Scholar]
- 101.Bosmans L, Pozo MI, Verreth C, Crauwels S, Wilberts L, Sobhy IS, et al. Habitat-specific variation in gut microbial communities and pathogen prevalence in bumblebee queens (Bombus terrestris). PLoS One. 2018;13(10): e0204612. doi: 10.1371/journal.pone.0204612 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 102.Schlüns H, Sadd BM, Schmid-Hempel P, Crozier RH. Infection with the trypanosome Crithidia bombi and expression of immune-related genes in the bumblebee Bombus terrestris Dev Comp Immunol. 2010;34(7): 705–709. doi: 10.1016/j.dci.2010.02.002 [DOI] [PubMed] [Google Scholar]
- 103.Brunner FS, Schmid-Hempel P, Barribeau SM. Immune gene expression in Bombus terrestris: Signatures of infection despite strong variation among populations, colonies, and sister workers. PLoS One. 2013;8(7): e68181. doi: 10.1371/journal.pone.0068181 [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
The file contains data from all of the N = 1,200 bees that were tested.
(XLSX)
The file contains the top 20 SNPs associated with infection intensity in the 20 x 20 matrix.
(XLSX)
The file contains the top 20 SNPs associated with infection intensity in the 8 x 8 matrix, using sqrt-transformed phenotype data, corrected for the strain effect, and pruned from linkage disequilibria LD > 0.75.
(XLSX)
The file contains the detailed report on the Amplicon sequencing procedure.
(PDF)
The file contains data from the top 20 SNPs associated with Axis.1 of the Multi-Dimensional Scaling (MDS) of zOTUs. Values for Axis.1 were LD-pruned with LD > 0.75.
(XLSX)
The file contains data from the top 20 SNPs associated with Axis.2 of MDS-ordination of OTUs. Values for Axis.2 were LD-pruned with LD > 0.75.
(XLSX)
The file contains data from the top 20 SNPs associated with cases where Axis.1 < 0 (’permissive’ group, from MDS-ordination of zOTUs. Values for Axis.1 corrected for structure, sqrt-transformed, and LD-pruned.
(XLSX)
The file contains data from the top 20 SNPs associated with cases where Axis.1 ≥ 0 (’resistant group’), from MDS-ordination of zOTUs. Values for Axis.1 corrected for structure, sqrt-transformed, and LD-pruned.
(XLSX)
The file contains the nucleotide sequences of all identified zOTUs in the project.
(FA)
This file contains the technical report for the sequencing of the bacterial metagenome.
(TXT)
This file contains a list of all samples from which the bacterial metagenome was typed and deposited with ENA (accession PRJEB52013). We used universal primers for the V3/V4 region of 16S RNA. A key to sample numbering is given.
(PDF)
File contains several sections. Methods: assessing infection intensity, selecting the 8x8 sub-matrix. Basic data: explanations, lists of colonies and parasite strains. Variation of infection intensity across colonies and strains. Explanations and supporting results for data in Appendices A to E.
(PDF)
Data Availability Statement
All raw data, including the metagenomic data for the microbiota analyses, are deposited in the EMBL-EBI European Nucleotide Archive (ENA) repository (http://www.ebi.ac.uk/ena), under the study accession PRJEB52013.