Higher stress and immunity responses are associated with higher mortality in reef-building coral exposed to a bacterial challenge

Understanding the drivers of intraspecific variation in susceptibility is essential to manage increasingly frequent coral disease outbreaks. We challenged replicate fragments of eight Acropora millepora genotypes with Vibrio spp. to quantify variation in lesion development and to identify host and coral-associated microbial community properties associated with resistance. While Vibrio spp. remained relatively rare in the microbiome of challenged corals, other stress-associated microbial taxa significantly increased in abundance. Contrary to expectations, higher constitutive immunity and more active immune responses did not confer higher resistance to bacterial challenge. Furthermore, more pronounced gene expression responses to bacterial challenge were associated with higher rather than lower mortality. A newly developed gene expression assay based on two genes related to inflammation and immune responses, deleted in malignant brain tumors 1 and a matrix metalloproteinase, predicted mortality under Vibrio treatment both in the initial experiment and in a validation experiment involving another 20 A. millepora genotypes. Instead of mounting more robust responses, resistant corals were largely unaffected by the bacterial challenge and maintained gene expression signatures of healthier condition, including elevated fluorescent proteins and ribosomal biosynthesis along with diminished ubiquitination. Overall, our results support the view that coral disease and mortality is commonly due to opportunistic pathogens exploiting physiologically compromised hosts rather than specific infections, and show, contrary to the prevailing wisdom, that greater immune responses do not necessarily translate into greater disease resistance.


Introduction 56
Global declines in coral cover are compounded by a variety of diseases (1, 2), many of 57 which are ambiguously defined by macroscopic characterizations of lesions (1-3). Several 58 bacterial species from the genus Vibrio have been implicated as etiological agents of some coral 59 diseases (4)(5)(6), but these bacteria may act merely as opportunistic pathogens exploiting 60 compromised hosts (7). Indeed, host immune health is considered to be a major determinant of 61 disease transmission dynamics (8,9). Corals, like all invertebrates, rely entirely on innate 62 immunity for protection from invading pathogens. Features of innate immunity in corals include 63 molecular pattern recognition (10), secreted antimicrobial macromolecules (11), cellular 64 responses (e.g., phagocytosis) (12,13), and physical barriers (e.g., mucus) (14,15). Melanin 65 deposits serve as another physical barrier against invading pathogens (16). The melanin synthesis 66 cascade is activated when pathogen recognition triggers cleavage of prophenoloxidase (PPO) to 67 phenoloxidase (PO). Reactive oxygen species (ROS) produced during melanin synthesis 68 contribute to its cytotoxic effects on pathogens, but also cause self-harm (17) that must be 69 countered by antioxidant enzymes such as catalase (CAT) and peroxidase (POX). 70 Field surveys of naturally occurring coral disease events show variability in disease 71 outcomes among conspecifics, despite the fact that the colonies are exposed to the same 72 environmental stressors and, presumably, the same potential pathogens (8,18). One possible 73 explanation is that some corals resist disease by making greater contributions to constitutive 74 immunity and coral taxa investing more in innate immunity (e.g., production of cytotoxic 75 defenses) are less likely to suffer infectious disease outbreaks (19,20). However, no laboratory 76 experiments have yet confirmed these differences in susceptibility at the intraspecific level or 77 investigated the molecular basis of differential disease outcomes among individuals. 78 In this study, we comprehensively examine coral host immune activity, genome-wide 79 gene expression, Symbiodinium profiles, and coral-associated microbial communities to 80 understand the physiological features underpinning disease resistance in a reef-building coral.  (Fig. S3A). Genes encoding proteins involved in immunity 140 (interferon gamma) and programmed cell death (apoptosis regulator Bcl-W) were upregulated in 141 bacteria-challenged corals, while deleted in malignant brain tumors protein 1 (dmbt1) and a 142 cryptochrome were downregulated. Gene ontology analysis of genes differentially expressed by 143 bacterial treatment found only one significantly enriched term (FDR = 0.1), "small molecule 144 metabolic process", which was downregulated in treated corals. Notably, the gene expression 145 response to bacterial challenge was predominantly driven by more susceptible genotypes, as 146 expression profiles of the most resistant corals remained similar to the control condition (

166
Symbiodinium profiles differ by reef and survival. Symbiodinium communities were profiled 167 using RNA-seq reads mapping uniquely to clade A, B, C, or D Symbiodinium transcriptomes. As 168 expected for A. millepora in this region of the GBR, clade C dominated most genotypes (6/8, 169 Fig. S5) (21). While corals with the highest survival rate also contained the most D-type 170 symbionts (p < 0.001), corals with the second-lowest survival rate were also significantly 171 enriched with clade D symbionts (p < 0.001). 172 173 174

Microbial community profiles differ between individuals and in response to bacterial 175
challenge. Clustering at 97% similarity identified 1238 operational taxonomic units (OTUs). 176 Principal coordinate analysis (PCoA) of weighted UniFrac distances revealed significant 177 differences depending on survival (Fig. 3A). As expected, asymptomatic corals treated with 178 Vibrio spp. had higher abundances of Vibrio than untreated controls (Table S1; FDR = 0.018). 179 Stress-associated taxa (families) were more abundant in treated corals (p = 0.04, Fig. 3B, Table  180 S2). Notably, some corals with high survival already had the highest proportion of stress-181 associated taxa in the control treatment (p = 0.014; Fig. 3B). In addition, A. millepora with the 182 highest survival rate (genotype W30) had significantly more chloroplast-derived OTUs than 183 other genotypes (

195
Diagnostic gene expression biomarker identification and validation. Two candidate survival 196 genes were selected based on their high and dynamic expression: deleted in malignant brain 197 tumors protein 1 (dmbt1) and gelatinase and matrix metalloproteinase (mmp). Dmbt1 expression 198 was positively and mmp expression was negatively correlated with survival (p < 0.001 for both) 199 ( Fig. S7). This gene pair was used in a self-normalizing double-gene qPCR assay, sensu (22). The current view of coral immunity is that species with higher constitutive and/or inducible 218 immune activity are more resistant to disease (19,20). However, we found that this hypothesis 219 does not apply to intraspecific differences in disease resistance, as resistance was not correlated 220 with antioxidant and cytotoxic activities, and higher survival was not associated with more 221 robust immune responses to bacterial challenge ( Fig. 1B-D). Rather, resistant corals were largely 222 unaffected by the bacterial challenge according to the selected immune activity assays. This 223 unexpected finding suggests that mechanisms other than host immune activity underlie increased 224 resistance to the bacterial challenge. 225 226 Resistant corals show a "healthier" gene expression profile. Constitutive (baseline) gene 227 expression analysis provided comprehensive insight into the physiology underlying observed 228 differences in predisposition to lesion formation. While we did not observe increased catalase or 229 peroxidase activities in resistant corals, the constitutive upregulation of oxidative stress response 230 genes ( Fig. 2A) suggests that these individuals may "frontload" (sensu (23)) antioxidant activity 231 in some manner that was not detected by our enzymatic assays. Constitutive frontloading of 232 thermal tolerance transcripts is a proposed mechanism by which some corals outperform 233 conspecifics under heat stress (23), and thus offers a tempting explanation for observed 234 differences in immune performance as well. However, surprisingly few genes involved in stress 235 responses or immunity were upregulated in resistant corals prior to bacterial challenge. Instead, 236 these individuals exhibited a more "healthy" gene expression profile: they had elevated 237 ribosomal components typically associated with higher growth rate (24) as well as fluorescent 238 proteins whose abundances have been linked to health status in corals (25, 26). Another signature 239 of higher resistance was diminished abundance of ubiquitination-related transcripts, e.g., genotypes except the most resistant (Fig. 3B). Surprisingly, resistant corals tended to harbor 272 elevated levels of these potentially harmful bacteria even before the challenge (Fig. 3B), which 273 drove the association between bacterial composition and survivorship (Fig. 3A). This finding 274 suggests that resistant corals are not more efficient in excluding harmful bacteria but are 275 generally less sensitive to their presence, which aligns well with their insensitivity to bacterial 276 challenge at the gene expression and physiological levels, as discussed above. 277 The most striking difference in microbial composition is the most resistant coral (W30) 278 harbored more chloroplast-derived OTUs (Fig. S6B), suggesting the presence of endolithic algae 279 or cyanobacteria. It is tempting to speculate that these microbes facilitated host defense by 280 actively secreting antimicrobial compounds, an ability that has been well characterized in 281 cyanobacteria (32, 33). This putative association between chloroplast-derived OTU abundance 282 and lower mortality under bacterial challenge merits further investigation. and PERMANOVA (adonis) were conducted based on weighted UniFrac distances (57) using 376 the R package vegan (58). Significant differences in OTU abundances between reefs of origin, 377 treatments, and survival fractions were assessed using a likelihood ratio test as implemented by 378 the G-test with group_significance.py in QIIME. Individual OTUs were designated as "stress-379 associated", "health-associated", or "unknown" based on previous literature (Table S2). RNA isolation, cDNA preparation, and qPCR were carried out as previously described (22) with 398 the exception that the RNAqueous Total Isolation Kit (Ambion) was used to isolate total RNA. 399 Linear regression implemented in R was used to test for the relationship between survival 400 fraction and the log-difference in expression between the two candidate genes, as in (22)      gene, summed over all isoforms. All subsequent analyses were carried out in R version 3.1.3 (9). Sample outliers were detected using arrayQualityMetrics (10). Low-expressed genes with a mean UTC less than three across all samples were discarded from the analysis.
Primer design. For deleted in malignant brain tumors protein 1, the forward and reverse primers were 5'-TCATGTGACCTGTGTTGGGA-3' and 5'-GGTGACGCTCCGATCAAAC-3', respectively. For gelatinase A and related matrix metalloproteases, the forward and reverse primers were 5'-GTTCCAAAATCGGCCACACC-3' and 5'-CGTTATGCAGGGCTTCCAGA-3', respectively. Primer pair specificity was verified by gel electrophoresis and melt curve analysis of the amplification product obtained with template A. millepora cDNA. Primer efficiencies were determined by amplifying a series of two-fold dilutions of A. millepora cDNA and analyzing the results using function PrimEff of the MCMC.qpcr package in R (11). Briefly, C T (threshold cycle) results were plotted as C T vs. log 2 [cDNA], and amplification efficiencies (amplification factor per cycle) of each primer pair were derived from the slope of the regression using formula: efficiency = 2 -(1/slope) (12).
Microbiome community analysis. DNA was isolated using an RNAqueous kit together with the RNA for gene expression analysis. DNA samples were diluted to 10 ng · µL -1 .
Sequences with six or more consecutive identical bases (12,128 sequences) or incorrect primer sequence (63,719 sequences) were discarded using split_libraries.py in QIIME (Quantitative Insights Into Microbial Ecology (13)). Sequences of 97% similarity were clustered into operational taxonomic units (OTUs). A phylogeny was generated by aligning representative sequences that were filtered to remove gaps and hypervariable regions.