confirmed that maternally ingested bacteria crossed the midgut, were entrapped in the hemocoel, accumulated in the ovary and then were finally deposited into the egg chorion and yolk. In the tobacco hornworm Manduca sexta, larval ingestion of non-pathogenic and pathogenic bacteria induced both immune responses in the infected host and changes in DNA methylation and histone acetylation (Gegner et al., 2019). In this system, maternal transgenerational immune priming was demonstrated by showing that bacterial structures were translocated from the gut lumen to the egg yolk, resulting in transcriptional reprogramming of immunity-associated genes.

More generally, examples are accumulating that suggest transgenerational immune priming might be common in insects, including in bumblebees (Sadd et al., 2005), moths (Tidbury et al., 2011;Trauer & Hilker, 2013) and beetles (Knorr et al., 2015;Zanchi et al., 2011). The growing recognition of the importance of immune priming in insects has recently led to it being dubbed the 'secret weapon' of the insect world (Sheehan et al., 2020).

The likely importance of bacteria for immune priming in insects is especially interesting because, at least in the Lepidoptera, microbiome composition tends to be highly variable and fluctuates in response to environmental drivers, gut physiology, developmental stage and host diet (Paniagua Voirol et al., 2018). In other insects that consume the egg casing after hatching, this early larval behaviour has been identified as a transmission mechanism for both intracellular and extracellular microbes (Salem et al., 2015). Although vertical transmission is less well understood in Lepidoptera, it is plausible that some portion of the larval microbiome is determined by the extracellular egg microbiota, given the ubiquity of egg casing consumption. For example, the microbiome of gypsy moth larvae is determined by egg mass of origin (which includes both maternal and environmental effects), and also host diet (Mason & Raffa, 2014).

While it is clear that the lepidopteran microbiome can be influenced by many factors, what is less clear is the role of maternally transmitted extracellular bacteria in affecting the lepidopteran immune response, and how these bacteria may interact with ecological factors, such as host plant use (Smilanich et al., 2018).

In this study, we sought to investigate how ecological variation in host plant use may interact with egg-associated microbes to affect the insect immune response in a common North American butterfly, Lycaeides melissa. This species is of interest from an evolutionary ecology standpoint because within the past 200 years it has undergone an expansion of diet to include the exotic legume Medicago sativa. Lycaeides melissa uses this host in many parts of its range, even though larval survival and adult fecundity are reduced when reliant upon M. sativa as compared to development on a common native host plant Astragalus canadensis (Forister et al., 2009). Furthermore, consumption of the novel host plant M. sativa results in a distinct suite of larval microbes, compared to larvae consuming the native host (Yoon et al., 2019). Thus, it is a system well suited for investigating interactions between host plant use and extracellular, egg-associated microbes. Specifically, we used a crossed (plant by population by experimental treatment) design to ask the following questions: (1) Does removing extracellular, egg-associated microbes affect the strength of the caterpillar immune response? (2) Is this effect dependent on the species of host plant consumed during development or (3) the population of origin of the mother?

If the immunological effects of egg-associated microbes are contingent on the identity of the host plant or maternal population of origin, then the outcome of host plant choice by ovipositing L. melissa females potentially affects not only larval performance via nutrition, but could also result in optimal versus non-optimal immune priming associated with other ecological factors. After testing for effects associated with the whole microbial community, we also investigated whether specific microbial taxa were associated with increased or decreased immune function. Finally, we note that our use of the idea of 'extracellular egg-associated microbes' is meant to include all microbes on the external surface of the egg, acknowledging that these microbes may have an origin from within the mother but might also have been acquired from the environment (including the leaf) after oviposition. However, since the female makes the choice of the oviposition site, these other sources are also maternally influenced, albeit not separately assayed in the present study.

Gravid L. melissa females were collected from a population associated with the native host plant A. canadensis at Washoe Lake NV, USA (hereafter referred to as WLA) and from a population utilizing the novel host M. sativa (alfalfa) at Verdi NV, USA (hereafter: VUH) during late June and early July 2016 (approximately 40 females from each location; see Figure S1 for an illustration of our experimental design). Permits were not required for collection at either site. Eggs acquired from these females were randomly assigned to a washing treatment (washed or unwashed, N = 200 eggs per treatment).

For eggs destined to consume the novel host plant M. sativa, these females laid on M. sativa. For eggs destined to consume the native host plant A. canadensis, these females laid on A. canadensis. Females were randomly assigned to either one host treatment or the other.

Females were placed into individual oviposition cages in a lath house the same day they were caught in the field. They were given fresh host plant material for oviposition and fed Gatorade™ every day (Fruit Punch flavour), as well as misted with water several times a day, as we have done in previous experiments. Oviposition took place over 2-3 days, with most females laying between 10 and 20 eggs each. Eggs were removed from host plant material using a sterilized small paintbrush. Half the eggs were left unwashed as controls, and half were immediately washed. Washed eggs were surface sterilized with a series of anti-microbial rinses, beginning with 96% ethanol, and ending with PVP-I Betadine solution (Beard et al., 1992). A subset of eggs from each treatment group was set aside and frozen for sequencing after washing (N = 20 eggs per treatment group). Each rinse was done three times for approximately 30 s per treatment,

13652435, 2022, 12, Downloaded from https://besjournals.onlinelibrary.wiley.com/doi/10.1111/1365-2435.14184 by University Of Nevada Reno, Mat, Wiley Online Library on [01/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License

and eggs were dried before being put back underneath a bright light for hatching. Washed eggs were kept separate from unwashed eggs. About 3-4 days after washing, eggs began hatching. Neonate larvae were transferred to host plant material using a sterilized small paint brush. Larvae were reared individually in petri dishes at ambient temperature and 10 h of light per day, as previously described (Forister et al., 2009). Plants were collected weekly from the same localities where the maternal butterflies were collected and leaves in rearing dishes were replaced every 2-3 days. Immune assays, weight measurements and DNA sequencing occurred once all larvae reached 4th instar (approximately 2 weeks after hatching). Larvae were frozen prior to sequencing.

Larval immune response was measured using three assays: standing phenoloxidase (PO), total PO and melanization via filament injections. Standing PO is a measure of the naturally activated enzyme after the haemolymph is taken from the caterpillars (Gonzalez-Santoyo & Cordoba-Aguilar, 2012). This assay measures the formation of dopachrome, which is assumed to be largely driven by active PO. Total PO is measured by chemically activating all the proenzyme prior to measuring PO activity to measure the basal amount of PO plus any inactive amount (Adamo, 2004). Filament injections serve as a proxy for a parasitism event (attack by a parasitoid wasp or fly that leaves an egg or eggs within the body of the caterpillar) and melanization of filaments is an indicator of immune response (Lavine & Beckage, 1996). Both PO measurement and melanization reflect the strength of the immune response (Hansen et al., 2017;Smilanich et al., 2009).

Standing PO activity was measured by taking 5 μl of haemolymph from the posterior abdominal segment of individual larvae using a sterilized sewing pin. Pins and handling instruments were sterilized with 100% ethanol before each extraction. Individuals were observed after extraction to ensure no larvae bled excessively, which could have confounded our measures of immune response.

Eppendorf tube and chilled on ice while dopamine solution was prepared. Powdered dopamine (Sigma-Aldrich) (0.0257 g) was added to 20 ml of distilled water. Samples were incubated for 20 min at room temperature, then 50 μl of the haemolymph/distilled water solution was added to a 96-well plate with 200 μl of dopamine. The reaction then proceeded in a microplate reader (Bio-Rad iMark) for 45 min (data recorded every 30 s at 490 nm). Data were analysed using Microplate Manager software (Bio-Rad v.6.3). We extracted the kinetic rate for the linear phase of the reaction (0-45 min). In addition, blanks, which consisted of distilled water and dopamine, were included as negative controls for each run. We did not run a positive control with each run; however, samples from all treatment groups (both host treatments) were run together to avoid confounding treatment with instrument variation. For total PO, the same protocol was followed with the addition to the well of 10 μl of the activator CPC (Cetylpyridinium chloride; Sigma-Aldrich) solution at a concentration of 1.0 g/20 ml of distilled water.

After haemolymph extraction, larvae were individually injected with approximately 2 mm of clear, sterilized monofilament. Larvae were returned to their respective petri dishes and given access to plant tissue for 24 h, then frozen and dissected to remove filaments.

The procedure for analysing the filaments involved selecting the entire surface of the filament in the photograph, then obtaining a continuous numerical value corresponding to how dark the selected area is. Darker filaments have more melanization and indicate a strong immune reaction. Filaments were photographed at 80X magnification using a dissecting microscope connected to a digital camera (Carl Ziess Discovery V.8, AXIOCAM Software) and their grey value recorded using Adobe Photoshop (CC 2018; Adobe System Inc.). Light for the photographs was provided by a fibre optic dual gooseneck light source with the intensity set at level three (out of six) for all photographs. To obtain the grey value, we used the 'quick selection' tool located in the side menu of the screen. Once the entire filament was selected, we changed the 'mode' to grayscale by selecting 'image' from the top menu ribbon, selecting 'mode', then changing from RGB which is the default setting to grayscale. To get the mean grey value, we chose 'analysis' from the 'image' tab, then selected 'record measurements'. The mean grey value is given along with other measurements of the selected area, with the scale ranging from 0 (black) to 255 (white). Average melanization was calculated by taking the mean grey value measured in Photoshop and then calculating the inverse such that smaller values represent paler filaments and larger values correspond with darker filaments. Darker filaments represent a stronger immune response as darker filaments signify a higher quantity of melanin deposited.

DNA was extracted from whole frozen larvae using a DNeasy blood and tissue kit (Qiagen). An extraction blank was created as a negative control. Library preparation and sequencing were performed by the Genome Sequencing and Analysis Facility (GSAF) at the University of Texas, Austin. A two-step PCR approach was used where template was amplified during the first round of PCR, and barcoded Illumina primers were added to amplicons during the second round. To target the V4 region of the 16S locus, we used the 515-806 primers developed by Wang and Qian (2009). PCR conditions for the first round were 1 cycle of 98°C for 30 s, followed by 12 cycles of 98°C for 30 s, 62°C for 30 s and 72°C for 30 s. Final extension at 72°C was conducted for 5 m. Total reaction volume was 20 μl (1 μl template, 7 μl water, 10 μl NEBNext 2X Master Mix and 1 μl of each primer). AMPure beads were used to clean amplicons (0.8x).

Conditions for the second round of PCR were as above, except seven amplification cycles were conducted instead of 12. AMPure beads were again used to clean amplicons. Positive and negative controls were used by GSAF to verify PCR success. Amplicons were pooled

and sequenced on the Illumina MiSeq platform using paired end 2 × 250 bp sequencing.

Sequences were merged using USEARCH v10.0.240_i86osx32 (Edgar, 2010;Edgar & Flyvbjerg, 2015). Only reads with less than one expected base calling error were retained. Exact sequence variants (ESVs) were defined using the UNOISE3 command (Edgar, 2016a).

ESVs that were less than 64 bp long, were low complexity (e.g. filled with repeats) or that matched PhiX, were discarded. An ESV table was constructed using the 'usearch_global' command of VSEARCH v2.6.2_macos_x86_64 (Rognes et al., 2016). Original, unfiltered reads were queried with the consensus ESV sequences to generate this table, thus recovering information from some low-quality reads that were removed during our stringent filtering process. VSEARCH was used to identify ESVs corresponding with mitochondrial DNA through comparison with the MIDORI (accessed 29 January 2018; Machida et al., 2017), MITOFUN (accessed 7 February 2018; Ntertilis et al., 2013) and RefSeq mitochondria (accessed 7 February 2018; O'Leary et al., 2016) databases. ESVs corresponding with chloroplast DNA were identified through comparison with the RefSeq plastid database (accessed 7 February 2018). Taxonomic hypotheses were generated via SINTAX (Edgar, 2016b) through comparison with the Ribosomal Database Project 16s training set (version 16; Cole et al., 2014) and the Greengenes database ( v13.5, DeSantis et al., 2009). ESVs identified as mtDNA or cpDNA by either database were discarded. When 1% or more of the reads for a given ESV were present in the negative control, then these ESVs were deemed contaminants and removed.

To understand how immune function (standing PO, total PO and melanization) varied by treatment group and was influenced by bacterial diversity and larval mass, we used Bayesian linear models separately for each of the three measures of immune function. Estimates of bacterial diversity were generated for each replicate as described below and used as a predictor variable in models. Response variables were assumed to be normally distributed with a mean that was the output of the linear model and a precision sampled from a gamma distribution with shape and rate parameters of 0.01. Priors for beta coefficients and the intercept term for each sampling group were normally distributed with a mean centred at zero and precision of 0.001. This model structure was also used to estimate the individual effect of abundant bacterial taxa on each metric of immune function. We deemed abundant bacterial taxa those that comprised on average 0.1% or more of the reads in samples (167 taxa met this threshold). Larval mass was included as a covariate in all models.

Beta coefficients for each bacterial taxon were estimated for each treatment group to determine if the influence of a particular taxon was affected by experimental treatments.

The proportional relative abundance of all ESVs as a function and is thus appropriate for a situation in which many individual effects (in this case for the many microbial taxa) are being estimated.

Posterior probability distributions (PPDs) for the relative abundance of each ESV within a treatment group were subtracted from PPDs for that microbial taxon in a different treatment group. If the majority (95%) of the PPD of those differences was on either side of zero, then there was high certainty that ESV was influenced by treatment.

To examine differences in diversity among treatment groups, we calculated Shannon's entropy for each PPD sample of Dirichlet parameters for that group (as per Harrison et al., 2020). This resulted in a PPD of diversity, thus propagating our uncertainty in taxon relative abundance through our calculations. Differential diversity among sampling groups was determined through subtraction of these distributions, as described above. For linear models, point estimates of diversity were used and were the means of the PPDs of diversity.

All models were defined in JAGS (Plummer, 2003) and implemented through the rjags package in r (R Core Team, 2017).

Specification of all models is provided in the Supporting Information.

When modelling bacterial relative abundance, the Markov chain Monte Carlo (MCMC) sampler used by rjags was optimized via an adaptation period long enough to approach the theoretical optimum for sample proposition, or for 30,000 steps, whichever came first.

Model adaptation was allowed to run for up to 10,000 steps for linear models. For relative abundance modelling, the first 500,000 MCMC samples chosen were discarded (burn in) and for linear modelling, the first 100,000 samples were discarded. The Geweke statistic was used to determine if burn in was sufficient to arrive at stable distributions for parameters of interest. Subsequent to sampler adaptation and burn in, MCMC sampling continued for 4000 iterations and every fourth sample retained. Two chains were run for each abundance model and convergence of focal parameters determined through the Gelman-Rubin statistic and through visual inspection of chain histories.

Washing treatment reduced bacterial diversity on sequenced eggs as measured by the exponentiated Shannon's diversity (see Figure S2). Overall, the washing treatment had the strongest positive effect on total PO (main effects summarized in Table 1).

This effect of washing was stronger among larvae eating the native host A. canadensis compared to larvae eating the novel, non-native host plant M. sativa (see Table 2 for host plant specific effects). For standing PO, by contrast, our model results do not suggest an effect of egg washing. Furthermore, melanization was affected by host diet. Specifically, there was an overall positive effect of consuming the novel, exotic host plant M.

sativa on melanization (Table 1), although this effect involves a population-specific interaction, as can be seen in Figure 1. While the overall treatment effects are quantified in Table 1 (from a model that estimated those effects while controlling for other factors), Figure 1 shows data (not controlling for larval weight) broken down by host and population of origin, as discussed further below.

Our Bayesian model allows us to compare the relative importance of predictor variables across a scale of continuous probability. Thus, there are several notable main effects that do not cross the 95% probability threshold (for probability of effect in a particular direction) but are still worthy of discussion. For example, egg washing had a 93.7% probability of positively affecting melanization (Table 1). There was also a 93.3% probability of larval mass having a strong positive effect on melanization. This trend was consistent across immune assays, with larval mass having an 88.3% probability of positively affecting total PO and an 87.8%

probability of positively affecting standing PO. There was also an 89.6% probability of a maternal source population effect, with maternal lines originating from WLA (Washoe Lake: the native site)

producing larvae with higher melanization. TABLE 2 Effects of maternal population, egg washing, larval mass and bacterial diversity on caterpillar immune function. Values shown are means of posterior probability distributions (PPDs) for beta coefficients output from Bayesian hierarchical linear models. Separate models were created for caterpillars reared on either host. The proportion of each PPD above zero and highest density intervals are shown in parentheses.

If ~95% or more of a PPD was above or below zero, then there was a high certainty of a non-zero effect of that covariate (bolded). Beta-coefficient estimates for source population describe the average effect of shifting from the reference category to the alternative category (e.g. from VUH to WLA). Shannon's microbial diversity was also tested as a part of this model but had an effect near zero in all cases and thus is not reported here

Washing treatment (unwashed is reference) Weight (mg)

Milkvetch (Astragalus canadensis); native host Total PO 2.8 (-5.4-10.8; 75.0%) 13.3 (4.1-22.5; 99.8%) -1.3 (-5.8-3.2; 28.6%)

Standing PO -2.4 (-15.7-10.9; 36.8%) 0.83 (-13.3-15.9; 53.8%) -2.12 (-9.9-5.5; 28.2%)

Melanization 20.2 (4.7-35.9; 99.4%) 23.12 (6.4-40.8; 99.8%) -3.43 (-12.4-5.2; 23.1%) Weight 1.1 (-5.7-7.9; 62.2%) 11.22 (4.4-18.1; 100%) -Alfalfa (Medicago sativa); introduced host Total PO -4.88 ((-11.7-2.2; 8.5%) 3.03 (-3.6-9.6; 81.4%) 4.09 (0.5-7.6; 98.8%) Standing PO 3.0 (-9.3-15.5; 68.2%) -1.6 (-14.0-11.1; 40.4%) 8.1 (1.8-14.6; 99.1%) Melanization -0.13 (48.2%) 8.72 (87.8%) 7.45 (97. 2%) Weight -2.4 (-8.1-3.4; 20.8%) 3.6 (-2.0-9.4; 87.7%) -Abbreviation: PO, phenoloxidase.

We detected a number of interesting interactions apparent when the results are subset by host plant treatment (Table 2). For example, larval mass appears to have a strong positive effect on total PO (98.8% probability), standing PO (99.1% probability) and melanization (97.2% probability) for larvae consuming the novel host M.

sativa, but these patterns do not occur with larvae consuming the native host A. canadensis (Table 2). Thus, the main effect discussed above of larval mass positively impacting immune response is most likely driven by larvae consuming the novel host plant, M. sativa.

Furthermore, while washing treatment had positive effects broadly for immune response, this treatment appears to affect larvae consuming the native host plant more strongly (Table 2, Figure 1). Another insight obtained by partitioning the model by host plant and population is that there appear to be maternal source population effects for larvae consuming the native host, but less so on the novel host. For example, originating from WLA appears to give larvae consuming the native host a positive boost for melanization (Table 2), which appears to be driven by a strong difference between WLA and VUH origins in the unwashed treatment (Figure 1). The same effect was not seen in larvae consuming the novel host plant M. sativa. Lastly, larvae grew larger when eggs were washed, and this effect was much stronger in larvae consuming the native host plant (Table 2).

We analysed 2,522,363 reads that merged successfully, passed stringent quality filtering and aligned to one of the 3738 ESVs identified. ESVs with fewer than 30 reads across samples were removed, which left 2240 ESVs. Taxa observed were primarily Proteobacteria, with the Enterobacteriaceae particularly well represented. In general, washing reduced overall bacterial diversity (Figure 2). Taxonspecific analyses demonstrated that the influence of any given bacterial taxon shifted depending upon treatment (Figure 3). For instance, several Enterobacteriaceae had particularly strong negative 1 and 2 for main effects by host plant and population of origin). The native host A. canadensis is abbreviated as AsCa, and the exotic host M. sativa is abbreviated as MeSa. The native host population (Washoe Lake) is represented as WLA, and the exotic host population (Verdi) is represented as VUH. Boxplots show median values and interquartile ranges for raw data; whiskers extend from the 10th to 90th percentiles. Numbers above each treatment group denote treatment groups with high certainty of differing from the focal group (marked with (1)), with a probability equal to or greater than 95% associations with total PO for all treatments, except when caterpillars were fed the native host and were reared from unwashed eggs.

We observed 216 ESVs that were hypothesized to be Wolbachia.

These ESVs shifted in relative abundance depending on treatment, but the total proportion of reads ascribed to members of the genus was always between 11% and 12%. Most Wolbachia taxa were negatively associated with immune function, however there were some Wolbachia taxa that had relatively weak positive effects (Figure 3).

The importance of insect-plant-microbe interactions has recently gained substantial attention as a driver of both evolutionary and ecological processes (Mogren & Shikano, 2021). However, knowledge gaps remain in how vertically transmitted, extracellular microbes might mediate novel host plant use and plant-herbivore interactions.

Our results demonstrate that extracellular, egg-associated microbes can have long-lasting impacts on larval immune function as measured by total PO, and, to a lesser extent, melanization (see Figure 4). We also found long lasting-impacts as reflected in greater larval weight from washed as compared to unwashed egg treatments. These impacts have the potential to cascade up to higher trophic-level interactions, as the PO pathway is one of the main mechanisms larval lepidopterans use to defend against parasitoids (Hansen et al., 2017;Slinn et al., 2018).

Removal of egg-associated microbes had a positive effect on several metrics of immune function, including both total PO and melanization. This result was contrary to what we expected, as previous studies have shown that maternal bacterial priming can increase the strength of the immune response (Freitak et al., 2009).

Interestingly, there is some evidence that lepidopteran larvae fed bacteria-free diets gain an immune boost when compared to larvae fed a diet supplemented with non-pathogenic bacteria. For example, in the lepidopteran pest species Trichoplusia ni, larvae fed a bacteria free diet had higher levels of PO than larvae fed diets supplemented with non-pathogenic Escherichia coli and Micrococcus luteu (Freitak et al., 2007). However, the same larvae that experienced a reduction in their PO levels also experienced an increase in their general antibacterial activity levels, suggesting some immunological compensation was occurring. Thus, it is possible our larvae also experienced some level of compensation in an unmeasured metric of immune function, such as general anti-bacterial activity or the production of anti-microbial peptides. Interestingly, the aforementioned study with T. ni also showed that bacteria-free larvae had increased pupal weight, a result that is consistent with our own study, where egg washing positively affected 4th instar larval weight. Future ecoimmunology studies should attempt to measure as many different types of immune effectors as possible in order to best capture potential compensatory patterns.

Given the positive immune response we observed and the increased size of caterpillars from washed eggs, it seems unlikely that our washing treatment negatively affected larval development. Betadine is known to have low toxicity and previous studies using Betadine washes on amputated flatworms have shown that the washing treatment had no effect on either growth or regeneration when compared to a control spring water treatment (Shaffer, 2010).

Previous literature reviews have found complex relationships between host plant diet and immune response in lepidopterans (Lampert, 2012;Singer et al., 2014). In some cases, larval consumption of the novel, low-quality host plant resulted in a weakened immune response compared to consumption of the typical, high-quality host plant (Diamond & Kingsolver, 2011;Resnik & Smilanich, 2020).

In other cases, consumption of the alternative, lower quality host plant actually resulted in a stronger immune response (Klemola et al., 2007;Yang et al., 2008). Thus, although we might predict that higher quality hosts (in terms of performance, e.g. larval weight) will result in a more robust immune response due to higher resource availability, this relationship is increasingly proving to be highly context dependent. An added source of complexity is the role of plant chemistry (phytochemistry) in mediating plant-insect interactions.

Previous work in this system has shown that variation in phytochemical diversity can significantly affect immune response, with higher phytochemical diversity resulting in lower PO activity in fourth instar larvae (Yoon et al., 2019). Numbers above each treatment denote treatment groups with a high certainty (at 95% probability) of differing from the focal group.

Treatment groups are numbered from left to right. For instance, treatment group 3 (AsCa-WLA-Unwashed) was estimated to differ from all other treatment groups. We observed high certainty of pairwise differences among treatment groups in most cases (see Supplemental Tables) Furthermore, the role of egg-associated microbes in mediating the larval immune response also appears to be context dependent: microbes were broadly important for predicting immune strength for all larvae, but more so for larvae consuming the native host plant.

Previous studies with L. melissa have demonstrated that consumption of the novel host plant M. sativa results in a significantly different larval microbiome, when compared to the native host plant A.

canadensis (Yoon et al., 2019). Interestingly, this study also demonstrated that larval microbial diversity had direct negative effects on immune response (as measured by standing PO) when larvae consumed the novel host M. sativa, but had no effect when larvae consumed the native host A. canadensis. Since it is thought that the larval lepidopteran microbiome is mostly shaped by transient diet-derived microbes (Hammer et al., 2017), these results suggest that changes Descriptions are based on the means of posterior probability distributions (PPDs) for beta coefficients output from Bayesian hierarchical linear models. If ~95% or more of a PPD was above or below zero, then there was a high certainty of a nonzero effect of that covariate (bolded). See Tables 1 and 2 for exact values and density intervals FIGURE 3 Associations between the relative abundance of bacterial taxa and caterpillar immune function. Barplots show standardized effect sizes for each bacterial taxon associated with specific immune response variables. Effect sizes are beta coefficients from a hierarchical Bayesian linear model that included larval mass as a covariate. Response variables are shown in the left margin. Bars are coloured by taxonomy as shown in key. The orange rectangle in the top row highlights several putative Enterobacteriaceae with particularly strong associations with total PO 13652435, 2022, 12, Downloaded from https://besjournals.onlinelibrary.wiley.com/doi/10.1111/1365-2435.14184 by University Of Nevada Reno, Mat, Wiley Online Library on [01/12/2022]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License