Diet modulates the relationship between immune gene expression and functional immune responses

Nutrition is vital to health and the availability of resources has long been acknowledged as a key factor in the ability to fight off parasites, as investing in the immune system is costly. Resources have typically been considered as something of a “black box”, with the quantity of available food being used as a proxy for resource limitation. However, food is a complex mixture of macro- and micronutrients, the precise balance of which determines an animal's fitness. Here we use a state-space modelling approach, the Geometric Framework for Nutrition (GFN), to assess for the first time, how the balance and amount of nutrients affects an animal's ability to mount an immune response to a pathogenic infection. Spodoptera littoralis caterpillars were assigned to one of 20 diets that varied in the ratio of macronutrients (protein and carbohydrate) and their calorie content to cover a large region of nutrient space. Caterpillars were then handled or injected with either live or dead Xenorhabdus nematophila bacterial cells. The expression of nine genes (5 immune, 4 non-immune) was measured 20 h post immune challenge. For two of the immune genes (PPO and Lysozyme) we also measured the relevant functional immune response in the hemolymph. Gene expression and functional immune responses were then mapped against nutritional intake. The expression of all immune genes was up-regulated by injection with dead bacteria, but only those in the IMD pathway (Moricin and Relish) were substantially up-regulated by both dead and live bacterial challenge. Functional immune responses increased with the protein content of the diet but the expression of immune genes was much less predictable. Our results indicate that diet does play an important role in the ability of an animal to mount an adequate immune response, with the availability of protein being the most important predictor of the functional (physiological) immune response. Importantly, however, immune gene expression responds quite differently to functional immunity and we would caution against using gene expression as a proxy for immune investment, as it is unlikely to be reliable indicator of the immune response, except under specific dietary conditions.

used as a proxy for resource limitation. However, food is a complex mixture of macro-and 28 micronutrients, the precise balance of which determines an animal's fitness. Here we use a state-29 space modelling approach, the Geometric Framework for Nutrition (GFN), to assess for the first 30 time, how the balance and amount of nutrients affects an animal's ability to mount an immune 31 response to a pathogenic infection. 32 Spodoptera littoralis caterpillars were assigned to one of 20 diets that varied in the ratio of 33 macronutrients (protein and carbohydrate) and their calorie content to cover a large region of nutrient 34 space. Caterpillars were then handled or injected with either live or dead Xenorhabdus nematophila 35 bacterial cells. The expression of nine genes (5 immune, 4 non-immune) was measured 20 h post 36 immune challenge. For two of the immune genes (PPO and Lysozyme) we also measured the 37 relevant functional immune response in the haemolymph. Gene expression and functional immune 38 responses were then mapped against nutritional intake. 39 The expression of all immune genes was up-regulated by injection with dead bacteria, but only those 40 in the IMD pathway (Moricin and Relish) were substantially up-regulated by both dead and live 41 bacterial challenge. Functional immune responses increased with the protein content of the diet but 42 the expression of immune genes was much less predictable. 43 Our results indicate that diet does play an important role in the ability of an animal to mount an 44 adequate immune response, with the availability of protein being the most important predictor of the 45 functional (physiological) immune response. Importantly, however, immune gene expression 46 responds quite differently to functional immunity and we would caution against using gene 47 expression as a proxy for immune investment, as it is unlikely to be reliable indicator of the immune 48 response, except under specific dietary conditions. 49 relies on the entomopathogenic nematode Steinernema carpocapsae, which vectors X. nematophila, to 208 gain access to an insect host, where it rapidly multiplies, generally causing death within 24-48 hours 209 (Georgis et al., 2006;Herbert and Goodrich-Blair, 2007). However, in the lab we can circumvent the 210 requirement for the nematode by injecting X. nematophila directly into the insect haemocoel (Herbert 211 and Goodrich- Blair, 2007). 212 Experiment 1: Within 24 h of moulting to the 6th instar, 400 larvae were divided into 20 groups (n = 213 20 per group), placed individually into 90 mm diameter Petri dishes and provided with ~1.5 g of one 214 of the 20 chemically-defined diets (Table 1). Within each diet, 10 larvae were allocated to the control 215 group and 10 were assigned to the bacteria-challenged group. Following 24 h feeding on the assigned 216 diets (at time, t = 0), 200 larvae were handled then replaced on their diet (control) whilst 200 larvae 217 were injected with 5 µl of a heat killed LD50 dose of X. nematophila (averaging 1272 X. nematophila 218 cells per ml nutrient broth) using a microinjector (Pump 11 Elite Nanomite) fitted with a Hamilton 219 syringe (gauge = 0.5mm). The syringe was sterilised in ethanol prior to use and the challenge was 220 applied to the left anterior proleg. Every 24 h up to 72 h (i.e. 48 h post infection), larvae were 221 transferred individually to clean 90 mm Petri dishes containing 1.5 -1.8 g of their assigned 222 chemically-defined diet. 96 h after moulting into L6, the larvae had either pupated or were placed on 223 semi-artificial diet until death or pupation. The amount of food eaten each day was determined by 224 weighing the wet mass of the chemically-defined diet provided each day to the caterpillars, as well as 225 weighing uneaten control diets each day (3 control diets per diet). The uneaten diet and control diet 226 were then dried to a constant mass (for approx. 72 h), allowing the consumption per larva to be 227 estimated. 228 Experiment 2: The set up for this experiment was identical to Experiment 1, except that each of the 229 400 larvae was injected with 5 µl of either a heat-killed (control) or live LD50 dose of X. nematophila 230 (averaging 1272 X. nematophila cells per ml nutrient broth). 231 Following challenge, hemolymph samples were obtained from all caterpillars at 20 h post infection.  233   Hemolymph samples were obtained by piercing the cuticle next to the first proleg near the head with a  234   sterile needle and allowing released hemolymph to bleed directly into an Eppendorf tube.  235 Immediately following hemolymph sampling, 30 µl of fresh hemolymph was added to a sterile ice-236 cooled Eppendorf containing 350 µl of lysis buffer (RLT + Beta mercaptoethanol -100:1) for later 237 RNA extraction and qPCR analysis (Expts 1 and 2). The remainder of the hemolymph extracted was 238 stored in a separate Eppendorf for further immune assays (Expt 2 only). All hemolymph samples were 239 stored at -80°C prior to processing. 240

Gene expression (Expts 1 and 2) 241
RNA was extracted from hemolymph samples using Qiagen RNeasy mini kit following the 242 manufacturers instructions with a final elution volume of 40 µl. Extracts were quantified using the 243 Nanodrop 2000 and diluted to 0.5 µg/µl for cDNA synthesis. Prior to cDNA synthesis a genomic 244 DNA elimination step was carried out by combining 12 µl RNA (0.5 µg total RNA) plus 2 µl DNA 245 wipeout solution and incubating at 42 °C for 2 min, cDNA synthesis was carried out using Qiagen 246 Quantitect Reverse Transcription kit in a final reaction volume of 20 µl following the manufacturer's 247 instructions, cDNA synthesis was carried out for 30 min at 42 °C followed by 3 min incubation at 95 248 °C and stored at -20 °C. cDNA was diluted 1:5 for use as a qPCR template. 249 Primers and probes were synthesised by Primer Design and qPCR was performed in a reaction 250 volume of 10 µl with 1x Taqman FAST Universal PCR Master mix (Thermo Fisher), 0.25 µM of each 251 primer, 0.3 µM probe and 2 µl of a 1:5 dilution of cDNA. qPCR was carried on the ABI 7500 FAST, 252 cycling parameters included an initial denaturation at 95 °C for 20 sec followed by 40 cycles of 253 denaturation at 95 °C, 3 sec and annealing at 60 °C for 30 sec. All PCRs were run in duplicate. 254 We selected five immune genes, three from the Toll immune pathway: Toll, Prophenoloxidae (PPO), 255 which is the precursor of the phenoloxidase enzyme (PO), responsible for production of melanin 256 during the encapsulation response, and lysozyme, which produces the antimicrobial lysozyme 257 enzyme, active against Gram positive bacteria. We also selected two genes from the IMD immune 258 Moricin, which produces the AMP Moricin, active against Gram positive and negative  259 bacteria, and Relish, which activates transcription of AMP genes (Ligoxygakis, 2013;Wiesner and 260 Vilcinskas, 2010). We also selected three non-immune genes, Tubulin, a component of the 261 cytoskeleton responsible for organelle and chromosomal movement. Armadillo (b-catenin), which 262 facilitates protein-protein interactions and EF1, an elongation factor facilitates protein synthesis. 263 These genes were selected, due to robust amplification, from a pool of potential endogenous controls 264 that were tested in pilot studies. We also selected Arylphorin, which is primarily characterised as a 265 storage protein (Telfer and Kunkel, 1991), however, it is up-regulated in response to bacterial 266 infection and also in response to non-pathogenic bacteria in the diet of Trichoplusia ni caterpillars 267 (Freitak et al., 2007) and so we did not have an a priori expectation as to its behaviour in this species. dopamine is the preferred substrate over L-dopa. It is the natural substrate for insects, it is more 293 soluble than L-dopa and unlike L-dopa, is not subject to spontaneous darkening (Sugumaran, 294 2002). 295 296

Gene expression 298
All statistical analyses were conducted using the R statistical package version 3.2.2 (R Core Team, 299 2018). Gene expression data were normalised using NORMA-Gene (Heckmann et al., 2011), a data 300 driven approach that normalises gene expression relative to other genes in the dataset rather than to 301 specifically identified reference genes. It is particularly suited to data sets with limited numbers of 302 assayed genes. Normalised gene expression data were then standardized using the mean (µ) and 303 standard deviation (σ) of each trait (Z = (X-µ) ⁄ σ) prior to analysis. The two experiments, run at 304 different times, had only one treatment in common, (1 -handled vs heat-killed bacteria, 2-heat-killed 305 vs live bacteria). For ease of interpretation, we wanted to analyse both experiments in a single model. 306 To test the validity of this approach, we first compared the gene expression, physiological immune 307 response data and the data for the total amount of food consumed across both experiments for the 308 heat-killed treatment only. There was no significant difference between any of the measures across 309 experiments, with the exception of the total amount of food eaten, and expression of the Moricin gene.

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
Therefore, all data were analysed in a single model, with the exception of those two response 311 variables, where data from the two experiments were analysed separately. 312 Data were analysed for each gene separately using linear mixed-effects models in the packages lme4 313 (Bates et al., 2015) and lmerTest (Kuznetsova et al., 2017). For each gene, the plate that the samples 314 were run on was included as a random effect. A comparison was made of 90 candidate models for 315 each gene, which comprised 30 models covering different combinations of dietary attributes (Table  316 3 The same approach was taken for the physiological immune measurements, lysozyme, PPO and PO 332 activity, except for these variables, standard linear fixed effects models were run as data were 333 collected in a single experiment. The same set of 90 models as described above were fitted, with the 334 addition of 180 extra models that included the additive and interactive effects of the expression of the 335 relevant gene, after correction for the plate to plate variation (residuals from the null model containing 336 the random effect of plate only) -the lysozyme gene for lysozyme activity and the PPO gene for PPO 337 and PO activity. 338 Time to death was analysed for experiment 2, where larvae were injected with dead or live bacteria 340 only. Data were analysed using Cox's proportional hazard models in the package (Therneau, 2015). 341 The same sets of models as described above were fitted (Table 3)

How does consumption vary across diets and bacterial challenge treatments? 352
The total amount of food consumed varied across the diets. For experiment 1, comparing handled 353 caterpillars versus those injected with heat-killed bacteria, the best model predicting consumption was 354 model 30 (Pe*Ce+Pe2+Ce2), but this was indistinguishable from the same model that included the 355 additive effects of treatment (Treatment+ Pe*Ce+Pe2+Ce2, delta=1.34). 356 For experiment 2, comparing dead and live bacterial injections, the best model predicting 357 consumption was model 20 (Co*R+Co 2 +R 2 ), but as for the handled versus dead treatments in 358 experiment 1, this model was indistinguishable from the same model that included the additive effects 359 of treatment (Treatment+Co*R+Co 2 +R 2 , delta=0.51). 360 For all treatment groups, it can be seen that consumption tended to increase as the calorie density of 361 the diet decreased (Figure 1a,b,d,e), suggesting that food dilution constrained caterpillars from being 362 able to take in sufficient nutrients, as expected, and that on the more calorie-dense diets caterpillars 363 over-consumed nutrients. However, this increase in total consumption was more extreme on the high- protein than on the low-protein diets, suggesting that caterpillars were willing to overeat protein to 365 gain limiting carbohydrates. 366 Overall consumption tended to decrease with treatment -dead-bacteria treated caterpillars ate less 367 than handled, and live-bacteria treated caterpillars ate less than dead-bacteria treated (Figure 1a  immune genes (Arylophorin, EF1, Armadillo and Tubulin), the variation in expression levels was 377 lower; for Arylophorin, EF1 and Armadillo, live bacteria triggered the down-regulation of gene 378 expression relative to handled caterpillars, whilst there was no effect for Tubulin ( Figure 2). For 379 Arylophorin, Armadillo and Tubulin, injection with dead bacteria up-regulated gene expression 380 relative to handled caterpillars but there was no effect for EF1 ( Figure 2). The best supported model 381 for every gene tested was model 30, with the bacterial treatment interacting with the amount of 382 protein and carbohydrate eaten (Treatment*(Pe*Ce+Pe 2 +Ce 2 )). However, although the fit of these 383 models was generally good (r 2 > 0.26-0.86), with the exception of Moricin, the amount of variation 384 explained by the fixed part of the model was very low (r 2 < 0.12; Table 4; Figures 3-5). This means 385 that the majority of the variation in gene expression was caused by variation across plates. For 386 Moricin, when comparing the handled and dead treatments, 74% of the variation explained by the 387 model was explained by the fixed terms due to the massive up-regulation of Moricin in the dead-388 bacteria injected larvae (Figures 2, 3a,b). The difference between the dead and live treatment groups 389 was much smaller and comparable to the other immune genes ( Variation in the expression of all of the genes was explained by main and interactive effects of the 391 amount of protein and carbohydrate eaten, and in interaction with the bacterial treatment, suggesting 392 that the response to diet for each gene differed across treatments. A visualisation of these response 393 surfaces (Figures 3-5) shows that, for the immune genes, whilst there is general up-regulation between 394 handled and dead bacterial challenges, the response surfaces are fairly flat, i.e. diet does not explain 395 much variation in gene expression. However, for the live challenge, expression tends to peak at 396 moderate protein but high carbohydrate intake, which corresponds to the highest intakes on the 33% 397 protein diet for Toll, PPO, Lysozyme and Relish, and on the 17% protein diet for Moricin For PPO activity, AICc could not discriminate between several of the diet models, with seven being 408 equally well supported (delta < 2; Table 5). Of these models, the top six contained protein and protein 409 squared with additive or interactive effects of bacterial treatment or gene expression (Table 5). For the 410 models that included treatment, the estimates show that PPO activity was increased with live bacterial 411 infection. For PO activity, AICc could not discriminate between 11 different models (delta < 2; Table  412 6). However, the top three models were the same as for PPO, with protein plus protein squared with 413 additive or interactive effects of PPO gene expression. Only two of the models contained treatment 414 effects and both in interaction with diet components. For lytic activity in the hemolymph, three 415 models were equally well supported, all of which contained Lysozyme gene expression interacting M A N U S C R I P T

A C C E P T E D ACCEPTED MANUSCRIPT
with dietary components, which were either protein and protein squared, as for PO and PPO, or the 417 P:C ratio (Table 7); none of the models contained treatment, suggesting that lysozyme activity is up-418 regulated in response to the presence of bacteria and not whether they are alive or dead. As for gene 419 expression, the overall explanatory power of the models was quite low, (r 2 < 0.12; Tables 5-7). 420 For ease of comparison, all 3 physiological immune traits were plotted against the protein content of 421 the diet, as this model was common to all three traits, and the expression of the relevant gene, which 422 featured in the majority of the selected models (Tables 5-7). The effect of treatment was excluded as it 423 did not feature in the majority of the models. For each trait, activity in the hemolymph tended to 424 increase with gene expression, as we might expect, but this was strongly moderated by the protein 425 content of the diet (Figure 6). For PO and PPO activity, on low protein diets enzyme activity was low 426 and there was little correspondence between gene expression and the physiological response, but as 427 the protein content of the diet increased, this relationship became more linear (Figure 6a,b). For lytic 428 activity the pattern was different in that enzyme activity increased strongly with the protein and less 429 strongly with lysozyme gene expression up to about 45% protein, thereafter there was consistently 430 high lytic activity across all levels of gene expression (Figure 6c). 431 432

Does immune gene expression predict survival? 433
Survival was reduced in the live bacterial treatment group relative to those injected with dead bacteria 434 (Hazard ratios 1.25-1.31 for models without treatment interactions, Table 3), however, this effect was 435 moderated by Moricin expression (Figure 7 a,b). In the dead-bacteria treatment group, Moricin did not 436 explain time to death, but in the live-bacteria treatment group, larvae with high levels of Moricin 437 expression had an increased risk of death relative to those with low expression (Figure 7 a,b; Hazard 438 ratios 1.20-1.24 for models without GE interactions, Table 3). Of the top 5 models, 4 included the 439 additive and interactive effects of protein and carbohydrate eaten as well as their squared terms (Table  440 3). To visualise the effects of diet on survival we plotted thin-plate splines for time to death against 441 the amount of protein and carbohydrate consumed. The patterns differ between dead and live bacterial Our dietary manipulation was successful in inducing caterpillars to consume over a large region of 472 nutrient space, allowing us to independently assess the effects of macronutrient composition and the 473 calorie content of the diet on immunity. There was evidence for compensatory feeding; caterpillars 474 did not consume the same amount of every diet. As expected, caterpillars ate more as the calorie 475 density of the food decreased (Figure 1), but this varied across diets, such that consumption was 476 highest on the high protein diets, suggesting that caterpillars were willing to over-eat protein to gain 477 limiting carbohydrates. However, as has been found in previous studies (Adamo, 1998 there is evidence that X. nematophila can inhibit Cecropin, Attacin and Lysozyme gene expression (Ji 499 and Kim, 2004;Park et al., 2007). It may be that, rather than specifically inhibiting AMP gene 500 expression, X. nematophila inhibits the expression of all genes. 501 As Moricin was most strongly up-regulated in response to infection, we tested how its expression 502 X. nematophila is a Gram-negative bacterium, and is clearly triggering Moricin and Relish expression, 515 but as Toll is only marginally up-regulated in response, it is probably the IMD pathway that is 516 controlling this response. Another possible explanation for why live bacteria appear to trigger a down-517 regulation of gene expression is that our sampling protocol (20 h post-challenge) did not allow us to 518 catch peak expression levels (note that bacterial loads tend to peak in S. littoralis at around 24h). 519 Expression of lysozyme and PPO in the Glanville fritillary butterfly was not up-regulated 24 h after 520 injection with M. luteus cells (Woestmann et al., 2017) , whilst in the silkworm, up-regulation of 521 lysozyme in response to fungal infection occurred in two peaks, from 9-18 h, and then between 30 and 522 48 h (Hou et al., 2014) . This may be a fungal-specific response, or it might mean that we would have M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT seen higher gene expression had we assayed over an extended time period. It is also possible that the 524 timing of gene expression peaks earlier after live, rather than dead bacterial injection, further studies 525 would be required to elucidate the time course of gene expression for the different genes to be certain 526 of this. However, as non-immune genes also appear to follow the same pattern, reduced expression in 527 response to live vs dead bacteria, the hypothesis that infection results in down-regulation of gene 528 expression in general is a reasonable assumption. 529 Arylphorin is primarily characterised as a storage protein (Telfer and Kunkel, 1991) For two of the immune genes, PPO and Lysozyme, we were able to simultaneously measure the 545 activity of the relevant protein in the hemolymph as a measure of the functional immune response. 546 Thus, we were able to assess how well gene expression predicts functional immune activity and 547 whether this relationship changes with the diet. Here, we found that for each functional immune 548 response, PPO activity, PO activity and lysozyme activity, expression of the relevant gene does 549 predict the response, but only at certain intakes of protein ( Figure 6). For example, PPO and PO M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT activity increase linearly with the expression of the PPO gene, but only above ~30% dietary protein 551 ( Figure 6). This suggests that the availability of dietary protein limits the translation of PPO mRNA 552 into PPO protein, and the activation of PPO into PO. In contrast, the expression of the gene is not 553 limited by protein availability, and so gene expression can be high when dietary protein is low, but it 554 is ineffective as it does not result in a comparable functional immune response. The lytic response is 555 also affected by dietary protein, however, in this case, the relationship between gene expression and 556 lytic activity is consistently weak and above 45% protein maximal lytic activity is achieved at low 557 gene expression, and increased expression does not improve the response. As for PPO, this suggests 558 that protein limits the translation of lysozyme up to about 45% protein. 559 These results are not surprising when you consider the costs associated with the production of protein. 560 It is estimated that only 10% of the energetic costs of protein production are spent on transcription; 561 translation is much more energetically expensive and relies on the availability of amino acids to build 562 the relevant protein (Warner, 1999). It is likely, therefore, that whilst transcription of immune genes 563 might still be up-regulated in response to infection under low protein conditions, the translation of the 564 protein might be reduced, impairing the correlation between mRNA and protein abundance. It is also 565 possible that gene expression would be a better predictor of the functional response at different time 566 points, if there is a lag between gene expression and protein translation. Again, this would require 567 further investigation. However, given the much stronger relationship between the physiological 568 immune responses and protein availability, it still seems likely that the relationship between the two 569 will differ across diets. Our results suggest that caution should be used when interpreting gene 570 expression as a measure of "investment" into a particular trait, or as a measure of the strength of a 571 particular immune response. It is surprisingly common in ecological studies for gene expression to be 572