Cumulative effects of mutation and selection on susceptibility to bacterial pathogens in Caenorhabditis elegans

Abstract

Understanding the evolutionary and genetic underpinnings of susceptibility to pathogens is of fundamental importance across a wide swathe of biology. Much theoretical and empirical effort has focused on genetic variants of large effect, but pathogen susceptibility often appears to be a polygenic complex trait. Here, we investigate the quantitative genetics of survival over 120 h of exposure (“susceptibility”) of Caenorhabditis elegans to three bacterial species of varying virulence, along with a fourth strain, the OP50 strain of Escherichia coli, the standard laboratory food for C. elegans. We compare the genetic (co)variance input by spontaneous mutations accumulated under minimal selection to the standing genetic (co)variance in a set of 47 wild isolates. Three conclusions emerge. First, mutations increase susceptibility to pathogens. Second, susceptibility to pathogens is uncorrelated with fitness in the absence of pathogens. Third, with the possible exception of Staphylococcus aureus, pathogen susceptibility is clearly under purifying directional selection of magnitude roughly similar to that of competitive fitness in the mutation accumulation conditions. The results provide no evidence for fitness tradeoffs between pathogen susceptibility and fitness in the absence of pathogens.

Introduction

The relationship between pathogens and their hosts holds an exalted place in evolutionary biology (Haldane, 1949; Hamilton, 1980). It is well established that susceptibility of individuals to the harmful effects of pathogens often has a genetic basis. The notion that particular multilocus host genotypes confer resistance or susceptibility to particular pathogen genotypes has been especially important, because “Red Queen” models of the evolution of sex and recombination are predicated on fluctuating epistasis (Barton, 1995), with negative frequency-dependent selection resulting from a host–pathogen arms race providing the most obvious plausible scenario (Hamilton, 1980).

In some cases, the genetic basis of variation in host susceptibility to pathogens can be attributed to variants at one or two loci. For example, variants at a few loci in the human genome (e.g., HBB, ABO, G6PD) are famously implicated in resistance to malaria (Rockett et al., 2014). Often, however, the heritable basis of susceptibility to a particular pathogen appears to be polygenic, with a substantial fraction of unexplained heritability (Cogni et al., 2016; Daub et al., 2013; Hill, 2001; Lefebvre & Palloix, 1996), sometimes even in cases with segregating large-effect loci (e.g., malaria; Band et al., 2019). Pathogens may thus impose significant selection even in the absence of Red Queen dynamics. It seems reasonable, then, to consider “susceptibility to pathogen X” as a generic complex trait; in some cases, there will be loci of large effect that explain much of the variance in susceptibility, but other times there will not.

Two fundamental issues in understanding the biology of complex traits are (i) what is the rate of input of genetic variance (and covariance) by mutation? and (ii) what is the relationship of the trait(s) with fitness? Taken together, these two considerations present a hierarchy of evolutionary hypotheses. The simplest explanation for any pattern of (co)variation is neutral evolution sensu stricto, i.e., the pattern can be sufficiently explained by neutral mutation and random genetic drift. If the pattern cannot be plausibly explained by neutral evolution, the next-simplest explanation is purifying selection against unconditionally deleterious mutations, plus drift. Only if those two explanations prove unsatisfactory need more complicated explanations be considered; for example, scenarios of balancing or positive selection.

For quantitative traits, the analog of the per-nucleotide mutation rate, μ, is the mutational variance V_M = Uα², where is the genomic mutation rate and α² is the average squared effect of a new mutation on the trait (Barton, 1990; Clayton & Robertson, 1955; Lynch & Hill, 1986). For a neutral trait at mutation-drift equilibrium, the genetic variation segregating within a population is V_G = 2N_eV_M, and populations will diverge asymptotically at rate 2V_M (Lynch & Hill, 1986). For traits under purifying selection (directional or stabilizing), at mutation-selection balance (MSB), , where s is the average strength of purifying selection against a mutant allele that affects the trait (Barton, 1990; Charlesworth, 2015; Kondrashov & Turelli, 1992). More precisely, , where is the coefficient of variation of the distribution of fitness effects (DFE) of mutations affecting the trait (Charlesworth, 2015). Because is rarely known with any precision, the relationship V_M/V_G is most often invoked in a comparative framework, where it is common to ignore variation in the DFE and assume that mutations have uniform effects, i.e., V_M/V_G ≈ s (Barton, 1990; Crow, 1993; Houle et al., 1996).

The multivariate extensions of V_G and V_M are the standing genetic and mutational variance–covariance matrices, G and M, of which the diagonal elements are the variances, V_G,i and V_M,i, respectively, and the off-diagonal elements are the genetic/mutational covariance between traits i and j, Cov_G,ij and Cov_M,ij (Lande, 1980; Phillips & McGuigan, 2006). The covariances capture the cumulative effects of pleiotropy and linkage disequilibrium.

Here, we report results of a set of experiments in which we (i) estimate the cumulative effects of spontaneous mutations on the susceptibility of two strains of Caenorhabditis elegans to four bacteria of varying pathogenicity (M) and (ii) estimate the standing genetic (co)variance for susceptibility to those same bacteria from a set of ∼50 C. elegans wild isolates (G). Our first, broad goal is to determine the realms of consistency and of idiosyncrasy in the mutational process in the context of host–pathogen evolution—what features are specific to particular host genotypes, or particular pathogens, and what features are general. Our second, broad goal is to draw inferences about the strength and pattern of natural selection acting on C. elegans’ susceptibility to bacterial pathogens, relative to the inferred strength of selection on competitive fitness.

Materials and methods

Mutation accumulation experiments—basic principles

An mutation accumulation (MA) experiment is simply a pedigree in which replicate populations, derived from a known (ideally) isogenic progenitor, are propagated under conditions in which natural selection is minimized (Halligan & Keightley, 2009; Mukai, 1964). Typically, selection is minimized by minimizing N_e; mutations with fitness effects s < 1/4N_e are effectively neutral (Keightley & Caballero, 1997). MA experiments can be used to illuminate the workings of natural selection in two ways. First, if selection is directional, the change in the trait mean (the mutational bias, ΔM) will point in the opposite direction of selection. On average, fitness declines under MA conditions. Traits that are positively correlated with fitness should, on average, exhibit a negative mutational bias (ΔM < 0); traits that are negatively correlated with fitness should, on average, exhibit a positive mutational bias (ΔM > 0).

Second, they provide the most straightforward and least assumption-laden way to estimate mutational (co)variances, which can be compared to the standing genetic (co)variances as outlined in the Introduction section.

Mutation accumulation lines

Details of the MA experiment are given in Baer et al. (2005). N2 is the standard laboratory strain of C. elegans; PB306 is a wild isolate graciously provided by Scott Baird. The basic protocol follows that of Vassilieva and Lynch (1999) and is depicted in Supplementary Figure S1. Briefly, replicate populations (MA lines) were initiated from a cryopreserved stock of a highly inbred ancestor and propagated by transferring a single immature hermaphrodite at each generation (4-day intervals) under standard laboratory conditions. The lines were propagated for 250 transfers (G₂₅₀) prior to cryopreservation. Data on transfers and additional context are given in Supplementary Table S1. Of the 100 N2 and PB306 MA lines initiated in 2001, we report results from 59 N2 and 65 PB306 lines. Lines not included in this study either were lost or had extremely slow growth rates, likely due to deleterious mutations of large effect. On average, each MA line carries about 65 spontaneous base-substitution and small indel mutations (Rajaei et al., 2021), with a few larger structural variants (Saxena and Baer, unpublished results). The average number of mutations per genome is nearly identical between the N2 and PB306 MA lines; thus, any differences between the two strains in the cumulative effects of mutations are the result of differences in the distribution of mutational (and/or epimutational) effects, rather than in the total number of mutations.

The cryopreserved progenitor (G0) serves as a control in two ways. The difference between the trait mean of the G0 and MA treatments (ΔM) quantifies the cumulative effect of spontaneous mutations on the trait mean. If the G0 progenitor is subdivided into replicate “pseudolines,” which are subsequently treated identically to the MA lines (Supplementary Figure S1), the difference between the among-MA line and the among-pseudoline components of variance represents the cumulative heritable variance resulting from the accumulation of spontaneous mutations (and potentially epimutations, depending on the experimental design; Beltran et al., 2020; Johnson et al., 2020). G0 pseudolines were constructed by thawing a sample of the cryopreserved G0 progenitor and picking individual L4 stage hermaphrodites to new plates, each of which was designated by a unique pseudoline number (PS1, PS2,…PSn). Pseudolines were subsequently treated as if they were MA lines. Unfortunately, due to a management error on the part of the senior author (CFB), not all pathogen treatments include similar numbers of pseudolines (see below).

Competitive fitness assay

Competitive fitness of the MA lines and their G0 progenitors was assayed in May 2005. The details of that assay have been published elsewhere (Yeh et al., 2018), and are reprised in Section 4 of the Supplementary Material. Fitness data are given in Supplementary Table S2.

Wild isolates

A collection of wild isolates of C. elegans was obtained from Erik Andersen (Northwestern University) and cryopreserved in the Baer lab. A list of the wild isolates is given in Supplementary Table S3. The genome sequences of the wild isolates along with collection information are available at https://caendr.org/.

Pathogen infection and survival assay

We used four strains of bacteria: three documented pathogens (the PA14 strain of Pseudomonas aeruginosa, the NCTC8325 strain of Staphylococcus aureus, and the OG1RF strain of Enterococcus faecalis) and the standard laboratory worm food, the OP50 strain of Escherichia coli. We use the term “pathogen” to refer to the three pathogenic species and “bacteria” when OP50 is included, although there is reason to think that OP50 is a weak pathogen of C. elegans, on the grounds that lifespan is extended when worms are fed on killed bacteria compared to live bacteria (Portal-Celhay et al., 2012). The basic protocol is a variant of the “slow killing assay” (Etienne et al., 2015), in which ∼30 L4-stage worms are transferred to a plate spread with bacteria and worms scored as live or dead at 12-hr intervals over the course of 120 hr. For the MA lines, each bacterial species was assayed in three or four experimental blocks; each MA line was assayed in a single block per bacterial species (i.e., line is nested within block). Blocks included three replicates per line from 20 to 24 MA lines and 13 (or fewer) G0 pseudolines. Wild isolates were assayed in one assay block, with three replicates per strain. The assay is described in detail in Section 1 of the Supplementary Material and depicted in Supplementary Figure S2.

Data analysis

Survival probability

Relative survival of a focal type (replicate i of strain x of treatment y on bacteria z) was quantified as the fraction of worms f (f = n_LIVE/n_TOTAL) surviving at time t relative to the mean fraction surviving of the N2 G0 ancestor on OP50, f_REL = f_FOCAL/f_N2,OP50. For all bacteria except PA14, survival was quantified at the culmination of the assay, at t = 120 hr. Survival was much lower on PA14 than on the other strains (Supplementary Figure S3), so survival on PA14 was quantified after 60 hr. Complete survival data from the MA assay are given in Supplementary Table S4. Times were chosen to roughly maximize the total variance in survival (Supplementary Table S5).

Mutational parameters

Two statistics were calculated to quantify the cumulative effects of MA on pathogen susceptibility. First, we estimated the mutational bias (ΔM), the per-generation change in the trait mean; ΔM = (f_REL,MA  − f_REL,G0)/t, where t is the number of generations of MA (t = 250) (Lynch & Walsh, 1998).

Second, we measured the mutational variance, V_M, calculated as (V_{L, MA} − V_L,0)/2t, where V_{L, MA} is the among-line variance of the MA lines, V_L,0 is the among-line variance of the G0 pseudolines, and t is the number of generations of MA. V_M standardized as a proportion of the residual (within-line) variance, V_E, is the “mutational heritability”; , where is the weighted average of the residual variance of the MA lines and the G0 pseudolines. Mutational covariances, Cov_M, were calculated analogously to V_M, from the among-line components of covariance.

Genotypic variances (V_G) and covariances (Cov_G) of the wild isolates were calculated similarly to the mutational (co)variances. There is very little residual heterozygosity in the wild isolates (Rajaei et al., 2021), so wild isolates were treated as inbred lines, in which the genotypic variance V_G = V_L (Zhang et al., 2021). Genetic covariances and correlations were calculated analogously. Broad-sense heritability, , where V_P is the total phenotypic variance. There is some disagreement in the worm quantitative genetics literature whether V_G should be estimated from V_L or V_L/2. The former assumes that the base population is already inbred, and that reproduction is effectively clonal; the latter assumes that the base population is outbred (Falconer, 1989). The traditional calculation of H² from inbred lines assumes that V_G = V_L/2 (e.g., the popular software package H2boot; Arnold & Phillips, 1999), but it is customary in the C. elegans quantitative genetics literature to equate V_G with V_L. For the sake of comparison with other studies, we assume here that V_G = V_L (note that Etienne et al. [2015] used the traditional V_L/2 formulation in their calculation of H²).

Statistical analysis

To account for variation among assay blocks, we first applied the general linear model (GLM) y_ij = μ + b_j + e_ij to each combination of strain (N2 or PB306) and bacterial species independently, where y_ij is the value of f_REL of replicate i in assay block j, μ is the overall mean, b_j is the fixed effect of block j, and e_ij is the residual effect, assumed normally distributed with mean 0. The residuals of this linear model, e_ij, subsume the effects of treatment (MA, G0), line (nested within treatment), and residual error. The residuals of this regression are the dependent variable, y*, in subsequent analyses. Pooled over strains, treatments (MA, G0), and bacteria, the y* are normally distributed (Supplementary Figure S4); we removed outliers with absolute Studentized residuals > 3 (n = 10/2021 samples).

Next, still considering each combination of C. elegans strain (N2 or PB306) and bacterial species independently, we estimated means and variances from the GLM y*_ijk = μ + a_i + L_j|i + e_jk|i, where y*_ijk is the residual from the block mean, μ is the overall mean, a_i is the fixed effect of generation of MA (gen = 0 for G0, gen = 250 for MA), L_j|i is the random effect of line (MA or pseudoline) within each treatment group (MA, G0), and e_jk|i is the residual effect within each treatment group, assumed normally distributed with mean 0. Variances of random effects were estimated independently for each treatment group (banded main diagonal covariance structure). Variances were estimated by restricted maximum likelihood with degrees of freedom determined by the method of Kenward and Roger (1997), as implemented in the MIXED procedure of SAS v.9.4. The fixed effect of treatment was tested by F-test with Type III sums of squares. Significance of among-line variance components (V_L) was assessed by likelihood ratio test (LRT) of the model with the among-line component of variance included against the model with the among-line variance constrained to 0. Since V_M and are composite parameters, confidence intervals were estimated by means of a parametric bootstrap, in which the components (V_{L, MA}, V_L,0, V_E) are normally distributed with means equal to the observed and standard deviations equal to the estimated standard errors. Individual bootstrap estimates of variances were constrained to be non-negative, i.e., bootstrap replicates less than 0 were set equal to 0. By constraining individual variances to be non-negative, we are in fact sampling from a zero-inflated truncated normal distribution.

Genetic covariance matrices (M and G for MA lines and wild isolates, respectively) were estimated analogously; the details are given in Section 2 of the Supplementary Material.

Comparison of M and G

To maximize power, we pooled samples over the two sets of MA lines and recalculated among-line (co)variances as outlined in the methods. These (co)variances include the heritable non-genetic component of variance, so we designate the “(epi)mutational” covariance matrix M* to highlight the distinction. Since the assay of the wild isolates was not designed to partition the among-line (co)variance into genetic and non-genetic components (i.e., we estimated G* rather than G), M* is the appropriate covariance matrix to compare to G*.

Results and discussion

Our overarching purpose is comparative evolutionary genetics in the context of host–pathogen relationships—what features are specific to particular host genotypes, or particular pathogens, and what features are general. Survivorship of the N2 G0 progenitor on OP50 (food) sets the benchmark by which the effects of the pathogenic bacteria are assessed. On average, the N2 progenitor suffered approximately 11% mortality over the course of the 120-hr assay (note that the assay was conducted at 25°, which is stressfully warm for C. elegans). Relative to that benchmark, survivorship was further reduced by as little as 1.5% in the PB306 G0 progenitor on OP50 and by as much as 84% in the N2 G0 progenitor on PA14. Averaged over the two strains (N2 and PB306) and treatment groups (MA and G0), the relative pathogenicity of the three pathogenic bacteria is OP50 < E. faecalis < S. aureus << PA14 (Supplementary Figure S4; Supplementary Table S6). Interestingly, the relative pathogenicity was different for the wild isolates, for which the rank order of pathogenicity was E. faecalis < S. aureus < OP50 << PA14 (Supplementary Figure S4). The discrepancy appears to be due to N2 and PB306 having substantially higher than average survivorship on Escherichia coli OP50. PB306 was included in the wild isolate assay, and its relative survival was similar to the MA assay for all bacterial species except E. faecalis (albeit n = 3 replicates in the WI assay vs. >100 replicates in the MA assay). N2 has been selected to eat OP50 for thousands of generations, so Escherichia coli (or at least OP50) should probably be considered a pathogen (or at least as “unhealthy food”) for all strains except N2. It is well known that worms live longer on killed OP50 than on live OP50 (Portal-Celhay et al., 2012), so its status as a pathogen has some precedence.

MA lines

Means

We expect that survival on OP50 (food, at least for N2) should be unambiguously beneficial, and both N2 and PB306 exhibit a significant negative mutational bias on OP50 (Table 1; Figure 1). Predictions about the direction of selection for survivorship on pathogenic bacteria are less clear because fitness tradeoffs between pathogen susceptibility and traits such as growth rate and yield in the absence of pathogens are well documented (Biere & Antonovics, 1996; Duncan et al., 2011; Kraaijeveld & Godfray, 2008; Tian et al., 2003).

Table 1.

Means (SEM).

Strain	Treatment	Survival	OP50	E. faecalis	PA14	S. aureus
N2	G0	Surv60	0.914 (0.009)	0.869 (0.010)	0.338 (0.016)	0.740 (0.015)
		Surv120	0.888 (0.011)	0.756 (0.012)	0.044 (0.006)	0.607 (0.015)
		f REL	1 (0.012)	0.852 (0.014)	0.370 (0.017)	0.683 (0.017)
	MA	Surv60	0.869 (0.013)	0.831 (0.017)	0.340 (0.023)	0.706 (0.014)
		Surv120	0.837 (0.014)	0.678 (0.026)	0.056 (0.010)	0.556 (0.015)
		f REL	0.943 (0.016)	0.764 (0.029)	0.372 (0.025)	0.626 (0.017)
		ΔM (104)	−2.3 (0.76)	−3.6 (0.13)	−1.1 (1.10)	−2.3 (0.97)
PB306	G0	Surv60	0.905 (0.014)	0.759 (0.011)	0.408 (0.053)	0.739 (0.014)
		Surv120	0.875 (0.020)	0.663 (0.014)	0.235 (0.043)	0.632 (0.015)
		f REL	0.985 (0.022)	0.712 (0.016)	0.447 (0.058)	0.711 (0.017)
	MA	Surv60	0.857 (0.016)	0.775 (0.013)	0.353 (0.021)	0.679 (0.015)
		Surv120	0.818 (0.019)	0.616 (0.015)	0.155 (0.015)	0.535 (0.018)
		f REL	0.920 (0.021)	0.693 (0.017)	0.386 (0.022)	0.603 (0.020)
		ΔM (104)	−2.6 (1.1)	−0.6 (0.96)	−2.5 (1.24)	−3.8 (1.02)
Average		ΔM (104)	−2.5	−2.1	−1.8	−3.1
Wild isolates		Surv60	0.788 (0.019)	0.871 (0.019)	0.633 (0.019)	0.819 (0.019)
	Surv120	0.697 (0.017)	0.773 (0.017)	0.051 (0.017)	0.736 (0.017)
	f REL	0.746 (0.023)	0.828 (0.023)	0.665 (0.023)	0.785 (0.023)

Note. Survival measures are Surv60 = % surviving at 60 hr; Surv120 = % surviving at 120 hr; see the Materials and methods section for details of calculations of f_REL and ΔM. Values of ΔM in bold font are significantly different from 0 (p < .05); see the Materials and methods section for details of statistical tests.

Figure 1.

Distribution of line mean relative survival (f_REL; see the Materials and methods section for explanation) on the four bacterial species (left to right: Escherichia coli OP50, Enterococcus faecalis, Pseudomonas aeruginosa PA14, and Staphylococcus aureus). G0 ancestral pseudolines in orange, MA lines in green, wild isolates in purple. Dashed line (f_REL = 1) is the mean survival of the N2 G0 pseudolines on Escherichia coli OP50. Asterisks identify samples with significant V_{L, MA}. (A) Top panel; N2; (B) middle panel, PB306; (C) bottom panel, wild isolates.

Whether the deleterious effects of mutations are exacerbated under stressful conditions is a topic of active investigation (e.g., Katju et al., 2018; Matsuba et al., 2013; Yun & Agrawal, 2014), but there is no theoretical reason to expect them to be unless selection is soft (Agrawal & Whitlock, 2010), which it is not in these assays. Infection by a pathogen is prima facie stressful. If deleterious effects are greater under pathogen stress, we predict that ΔM will be greater (more negative) in the pathogen treatments than in the OP50 treatment, and that ΔM will be greatest (most negative) in the most virulent pathogen, PA14.

As a whole, support for those predictions is weak. Point estimates of ΔM are negative in all eight strain/bacteria combinations (Table 1), although not significantly different from zero in three of the eight, and less on PA14 than on OP50. In only two cases—N2 on E. faecalis and PB306 on S. aureus—is ΔM substantially greater (∼1.5X) on the pathogen than on OP50.

Variances

To begin, we estimated the among-line component of variance (V_L) for each combination of bacteria, strain, and treatment (Table 2). For the MA lines, V_{L, MA} includes the combined contributions of new mutations and heritable, non-genetic effects; for the G0 pseudolines, V_L,0 includes only the heritable, non-genetic effects. V_{L, MA} is significantly greater than zero in five of the eight strain/pathogen combinations in the MA lines, and nearly significant in a sixth (PB306 in E. faecalis; Supplementary Table S7). There was no significant V_{L, MA} on S. aureus in either strain.

Table 2.

Variances in relative survival, f_REL (SEM).

Strain	Statistic	OP50	E. faecalis	PA14	S. aureus
N2	V L (MA) 102	0.64 (0.25)	2.24 (0.87)	1.07 (0.49)	0.28 (0.34)
	V L (G0)102	0.042 (0.056)	0	0.17 (0.16)	0.36 (0.30)
	V E (MA) 102	1.48 (0.22)	2.74 (0.56)	2.61 (0.43)	3.29 (0.48)
	V E (G0) 102	0.85 (0.12)	2.06 (2.81)	1.73 (2.39)	2.33 (3.90)
	V E (wt. ave) 102	1.17 (0.17)	2.40 (0.42)	2.17 (0.33)	2.81 (0.44)
	V M  105	1.20	4.47	1.81	0
	103 (95% CI)	1.03 (0.18, 2.07)	1.87 (0.46, 3.71)	0.83 (0, 1.93)	0
PB306	V L (MA) 102	1.40 (0.46)	0.62 (0.38)	1.20 (0.39)	0.44 (0.42)
	V L (G0)102	0.059 (0.160)	0.13 (0.29)	0.11 (0.21)	0
	V E (MA) 102	2.56 (0.36)	2.86 (0.42)	2.42 (0.32)	3.66 (0.56)
	V E (G0) 102	2.97 (0.43)	2.56 (0.44)	2.44 (0.32)	2.70 (0.41)
	V E (wt. ave) 102	2.77 0.39)	2.71 (0.43)	2.43 (0.32)	3.18 (0.48)
	V M  105	2.69	0.99	2.18	0.88
	103 (95% CI)	0.97 (0.27, 1.78)	0.37 (0, 0.99)	0.90 (0.20, 1.63)	0.28 (0, 0.88)
Ave	V M  105	1.95	2.73	1.99	0.44
	103	1.00	1.11	0.86	0.14
	rm,w	0.05 (0.18)	−0.08 (0.23)	−0.08 (0.19)	0.25 (0.07)
Wild isolates	V G  102	1.68 (0.58)	0.57 (0.37)	4.01 (1.22)	2.19 (0.63)
	V E 102	2.13 (0.37)	2.77 (0.45)	3.75 (0.63)	1.63 (0.27)
	H 2 (95% CI)	0.44 (0.21, 0.61)	0.17 (0, 0.34)	0.52 (0.31, 0.66)	0.57 (0.36, 0.70)

Note. V _L = among-line variance; V_E = environmental variance; V_M = mutational variance; = mutational heritability; 95% CI in parentheses; r_m,_w = mutational correlation of relative survival with competitive fitness; V_G = genetic variance; H² = broad-sense heritability; 95% CI in parentheses. Values of and H² in bold font are statistically significant (p < .05).

The same analysis for the G0 pseudolines has less power than for the MA lines because there are fewer pseudolines than MA lines (N2, n = 9, 9, 39, 39 for PA14, OP50, E. faecalis, and S. aureus, respectively; PB306 n = 6, 9, 39, 39). For E. faecalis and S. aureus, there are sufficiently many pseudolines (13/assay block) to credibly test the significance of a variance component; in neither strain is V_L,0 significantly different from zero for those two bacteria. For OP50 and PA14, we report point estimates of V_L,0 and its standard error, and use those estimates in calculations of V_M, but make no claims about statistical significance of V_L,0.

There is significant mutational heritability on two of the four bacterial species in each of the two strains (Table 2). Averaged over all bacterial species except S. aureus, for which is very low or absent, ≈ 0.001, the canonical mutational heritability for a wide variety of traits in many organisms (Conradsen et al., 2022; Houle et al., 1996). We previously estimated mutational parameters for survival on PA14 of (almost) the same set of PB306 MA lines (Etienne et al., 2015). The survival assay (spotted bacterial lawn) and measure of survivorship in that study (LT50) both differed from this study (spread lawn, relative survival at t₆₀), but the estimated of LT50 was ≈ 0.001, very close to the estimate from this study. The point estimate of in N2 in this study is nearly the same as that of PB306, although it does not quite meet the 5% threshold of significance. ΔM of PB306 on PA14 was near zero in both studies.

Mutational correlations with competitive fitness

Under the MA conditions, the PB306 MA lines decline in competitive fitness at about 0.1%/generation (ΔM _w = −1.13 ± 0.25 × 10⁻³/gen), very close to the estimate of ΔM _w for N2 in the same assay (−1.09 × 10⁻³/gen; Yeh et al., 2018). Pooled over the two sets of MA lines, mutational correlations (r_M) between f_REL on bacterial species i and relative competitive fitness are shown in Table 2. For all four bacterial species, r_M between f_REL and competitive fitness in the MA environment is ∼0. Evidently, mutations (and/or epimutations) that contribute to relative survival have no consistent cumulative effect on competitive fitness under the MA conditions.

Wild isolates

There is significant broad-sense heritability (H²) for relative survival on three of the four bacterial species, the exception being on E. faecalis (Table 2). The wild isolate assay did not include pseudoline controls for heritable non-genetic variance, so the estimates of H² are upper bounds on the genetic heritability. Surprisingly, H² is greatest on S. aureus, on which there is little if any mutational variance. By way of comparison, we recalculated H² for survival on PA14 from the set of 114 wild isolates reported by Etienne et al. (2015) with LT50 as the summary statistic. Mean LT50 was 63.5 hr. The recalculated estimate, H² = 0.41, is slightly less than the estimate from relative survival at t₆₀ in this study (H² = 0.52).

Comparison of M and G

For M* (Table 3), a model with banded main diagonal covariance structure (off-diagonal elements constrained to 0) fit slightly better than a model with unstructured covariance (ΔAICc = 1.2; chi-square = 11.0, df = 6, LRT p > .08). For G* (Table 4), the unstructured covariance fit slightly better than the model with banded main diagonal (ΔAICc = 1.5; chi-square = 14.2, df = 6, LRT p < .03). Given the weak support for non-zero off-diagonal elements, it is unsurprising that no consistent pattern emerges from either the mutational or standing genetic correlations, beyond the observation that none of the correlations are large and negative. For that reason, we do not attempt formal matrix comparisons of M* vs. G*.

Table 3.

(Epi)mutational (co)variance matrix M*.

Bacteria	OP50	E. faecalis	PA14	S. aureus
OP50	0.28 (0.08)	0.04 (0.06)	−0.01 (0.05)	0.10 (0.05)
E. faecalis	0.15 (0.23)	0.23 (0.09)	0.10 (0.06)	0.08 (0.05)
PA14	−0.05 (0.18)	0.42 (0.22)	0.25 (0.07)	0.05 (0.05)
S. aureus	0.63 (0.35)	0.52 (0.41)	0.36 (0.33)	0.09 (0.07)

Note. Covariances shown above the diagonal, correlations below the diagonal. Values in bold font are significantly different from 0 (p < .05). See the Materials and methods section for details of estimation of M*.

Table 4.

(Epi)genetic (co)variance matrix G*.

Bacteria	OP50	E. faecalis	PA14	S. aureus
OP50	0.31 (0.11)	0.10 (0.06)	0.26 (0.12)	−0.05 (0.08)
E. faecalis	0.53 (0.31)	0.11 (0.07)	0.17(0.10)	0.07 (0.06)
PA14	0.54 (0.20)	0.60 (0.32)	0.73 (0.23)	−0.05 (0.12)
S. aureus	−0.13 (0.24)	0.31 (0.29)	−0.10 (0.22)	0.40 (0.12)

Note. Covariances shown above the diagonal, correlations below the diagonal. Values in bold font are significantly different from 0 (p < .05). See the Materials and methods section for details of estimation of G*.

The strength and form of natural selection

The finding that ΔM < 0 in every case implies that mutations that reduce survival upon exposure to pathogenic bacteria are deleterious, on average. That is not surprising in hindsight, although it is certainly plausible that fitness tradeoffs could result in stabilizing selection on pathogen susceptibility. The lack of correlation between survival on pathogens and competitive fitness in the MA environment indicates that pleiotropy between mutational effects in the two contexts is not strong, but reveals little about selection in the natural environment.

To get a rough idea of how strong selection on these traits is, the neutral expectation provides a benchmark. For a neutral trait in a predominantly selfing organism such as C. elegans, at mutation-drift equilibrium, V_G ≈ 4N_e V_M (Lynch & Hill, 1986). N_e in C. elegans has been estimated to be on the order of 10⁴–10⁵ (Gilbert et al., 2022), so if V_M ≈ 2 × 10⁻⁵ (the average for relative survival on OP50), we expect V_G ≈ 0.8–8, about 40–400 times greater than the observed value of ∼0.02. The discrepancy is similar for PA14, and slightly greater for E. faecalis. Although there is considerable uncertainty associated with all of these estimates, the ability to survive exposure to those three species of bacteria is clearly not a neutral trait; mutations affecting survival on those bacteria are subject to effective purifying selection. S. aureus may present a different scenario; we return to it below.

If selection is sufficiently strong to be approximately deterministic (i.e., ignoring drift), at MSB, , where s represents the average strength of selection against a mutation affecting the trait. If we make the (questionable) assumption of uniform mutational effects (see Charlesworth 2015 for caveats), and s ≈ 0.001 for mutations affecting relative survival on OP50, roughly the same on the virulent pathogen PA14 and a bit stronger on E. faecalis. By way of comparison, a similar calculation for competitive fitness (measured in the MA conditions) revealed s ≈ 0.005 (Yeh et al., 2018; the value is halved if V_G = V_L rather than V_L/2), which is in close agreement with a direct estimate of the average mutational effect  ≈ 0.0035, calculated by dividing the cumulative decline in competitive fitness of the MA lines by the average number of mutations carried by an MA line. These point estimates are obviously rough, but it is evident that mutations that affect susceptibility to pathogens experience significant selection in nature. Whether the selection is direct, imposed by exposure to pathogens in nature, or indirect, due to pleiotropic effects on fitness in the absence of pathogens, cannot be ascertained from these experiments. However, the absence of a mutational correlation between relative survival and relative fitness (in the MA context) implies that it might be the former. Recent studies have revealed a complex transgenerational epigenetically inherited (TEI) behavioral response specific to the PA14 strain of P. aeruginosa that varies among wild isolates (Moore et al., 2021), which reinforces the inference that PA14 imposes direct selection in nature.

MSB is not the only mechanism by which genetic variation may be maintained; balancing selection (BS) is another. There are several well-documented balanced polymorphisms in C. elegans (summarized in Lee et al. 2021), and there are numerous examples from diverse organisms in which pathogens do contribute to BS (Fumagalli et al., 2009; Hawley & Fleischer, 2012; Horger et al., 2012; Quéméré et al., 2021). If BS is the predominant mechanism responsible for the maintenance of genetic variation, there is no reason to expect V_G and V_M to be correlated. In fact, V_M and V_G are nearly always highly positively correlated (Farhadifar et al., 2016; Houle, 1998; McGuigan et al., 2015). BS and MSB are often set up as alternative hypotheses for the maintenance of genetic variation, but it is realistic to expect that uniformly deleterious alleles will always contribute to trait variation, whereas balanced polymorphisms may or may not. If that scenario is generally true, the effect of BS would be to weaken the relationship between V_M and V_G, and that does not seem to be the case. If, in contrast, BS is the exception rather than the rule, the occasional trait experiencing BS would appear as a high outlier, with more V_G than predicted by its V_M, and selection as inferred from the ratio of V_M to V_G would appear anomalously weak.

Relative survival on S. aureus seems to fit that pattern, at least at first glance. Averaged over both sets of MA lines, V_M for relative survival on S. aureus is only about 20% of V_M on the other bacteria, whereas V_G is not atypically low, and H² is the highest of the four bacteria. If N_e is closer to 10⁴ than 10⁵, the selection coefficient inferred from V_M/V_G (S ≈ 2 × 10⁻⁴) is approaching effective neutrality (1/N_e, the “drift barrier”). However, averaged over the two sets of MA lines, ΔM is the greatest (most negative) on S. aureus, which implies that accumulated (epi)mutations significantly reduce relative survival on S. aureus.

It is not unheard of in the MA literature for trait means to change significantly even in the absence of significant mutational variance (Shabalina et al., 1997 is a well-cited example in Drosophila). It is a recurring phenomenon in C. elegans MA experiments, and it is most common for traits that have inherently high variability (e.g., male competitive mating success, Yeh et al., 2018; metabolite concentration, Johnson et al., 2020). The phenomenon may be more detectable in C. elegans than in other systems for two related reasons. The ability to easily maintain cryopreserved stocks permits the contemporaneous assay of the unevolved G0 ancestor and the MA lines, and transgenerational epigenetic inheritance (TEI) is extremely well characterized in C. elegans (Beltran et al., 2020; Rechavi & Lev, 2017). If TEI contributes to the heritable variance in an unbiased way and of similar magnitude to mutational variance (Johnson et al., 2020), it could overwhelm the signal of mutational variance even in the presence of a directional effect of mutations. Alternatively, if a trait is especially sensitive to microenvironmental noise, V_E may overwhelm V_M such that among-line variance is undetectable at the given sample sizes. That may be the case for S. aureus in the MA lines, for which V_E is greatest on S. aureus in both the N2 and PB306 lines. However, to make things even less clear, V_E is smallest on S. aureus in the wild isolates.

Given the time- and labor-intensive properties of both MA experiments and pathogen survival assays, it is perhaps unsurprising that there are very few studies that have quantified the effects of spontaneous mutations on pathogen susceptibility. Recently, Mendel et al. (2024) investigated the effects of spontaneous mutations on the susceptibility of Drosophila serrata to Drosophila C virus (DCV). They reported a mutational heritability for LT50 of ≈ 0.004–0.005, significantly greater than the estimates of in this study and toward the high end of estimates of . The mutational properties (rate and spectrum) at the genomic level in D. serrata remain unpublished, but recent estimates from parent-offspring sequencing of D. melanogaster and D. simulans revealed a base substitution/small indel rate 1–2X that of C. elegans (Wang et al., 2023). The genomes of those two fly species are somewhere between 36% (D. simulans) and 67% (D. melanogaster) larger than that of C. elegans, so the mutational target is probably somewhat larger. However, the per-genome, per-generation rate of transposable element insertions is about 50% greater than that of new base substitutions/small indels. Based on limited data, we infer that spontaneous transpositions constitute <1% of all spontaneous mutations (Saxena and Baer, unpublished results), so if D. serrata has similar TE activity to the other Drosophila species, that could explain the greater mutational heritability compared to C. elegans. Of course, there are many other differences between flies and worms, and between bacteria and viruses. Somewhat surprisingly, C. elegans is only known to have one viral pathogen (the Orsay virus, Felix et al., 2011), which has been established as a laboratory model system (Batachari et al., 2024). The MA lines and wild isolates are freely available and await the intrepid PhD student to do the investigation.

Conclusions

1. The cumulative effect of mutations is to increase susceptibility to pathogens. Moreover, survival upon exposure to pathogens is uncorrelated with competitive fitness under the MA conditions. Thus, there is no evidence of tradeoffs between susceptibility to pathogens and fitness in the absence of pathogens.

2. To the extent that M* and G* are reliable estimators of M and G, respectively, it is evident that survival on these bacteria is subject to effective purifying selection.

3. The number of MA lines included (124) and the number of generations of MA (250, ∼1000 days) are quite large. The number of wild isolates (47) is not so large, but is not small. Nevertheless, the results are (depending on one’s perspective) either intriguingly or distressingly variable between strains (N2, PB306) and bacteria. Nevertheless, the one point of direct comparison—PB306 on PA14—yielded results that are very consistent with a previous study. In our opinion, the way forward lies in development and adoption of high-throughput phenotyping methods to increase the practical sample sizes in these kinds of assays.

Data availability

All data are included as supplementary tables and are archived in Dryad at DOI: 10.5061/dryad.5qfttdzjs. All code is available in Figshare at 10.6084/m9.figshare.28046366.

Author contributions

S.S. performed experiments and analyzed data; L.M.J. performed experiments, analyzed data, and wrote the initial draft manuscript; M.M.I.R. analyzed data; S.R. performed experiments; and C.F.B. conceived and designed the experiments, performed experiments, analyzed data, wrote the final version of the manuscript, and obtained funding.

Funding

This work was supported by the National Institutes of Health awards GM107227 and GM127433.

Conflict of interest

The authors declare no conflicts of interest. Editorial processing of the manuscript was conducted independently of C.F.B., an Associate Editor of Evolution.