Comparative analysis of helminth infectivity: growth in intermediate hosts increases establishment rates in the next host

Abstract

Parasitic worms (i.e. helminths) commonly infect multiple hosts in succession before reproducing. At each life cycle step, worms may fail to infect the next host, and this risk accumulates as life cycles include more successive hosts. Risk accumulation can be minimized by having high establishment success in the next host, but comparisons of establishment probabilities across parasite life stages are lacking. We compiled recovery rates (i.e. the proportion of parasites recovered from an administered dose) from experimental infections with acanthocephalans, cestodes and nematodes. Our data covered 127 helminth species and 16 913 exposed hosts. Recovery rates increased with life cycle progression (11%, 29% and 46% in first, second and third hosts, respectively), because larger worm larvae had higher recovery, both within and across life stages. Recovery declined in bigger hosts but less than it increased with worm size. Higher doses were used in systems with lower recovery, suggesting that high doses are chosen when few worms are expected to establish infection. Our results indicate that growing in the small and short-lived hosts at the start of a complex life cycle, though dangerous, may substantially improve parasites' chances of completing their life cycles.

1. Introduction

Parasitic worms like acanthocephalans, cestodes, trematodes and nematodes (i.e. helminths) lead perilous lives. Before reproducing, they successively infect multiple hosts from dissimilar taxa and thus face different physiologies and immune systems [1]. They are usually trophically transmitted, with one host being eaten by the next host, and in extreme cases, worms only reproduce after being transmitted through four or five hosts [2]. At every step in these complex life cycles, there is a chance that parasites will not encounter or successfully infect the next host, so longer life cycles (i.e. more consecutive hosts) intuitively seem more hazardous than shorter life cycles. Although longer life cycles can have benefits (e.g. better propagule transmission and larger reproductive sizes [3]), the accumulation of risk with multiple transmission steps is a fundamental cost. To minimize this risk, worms have adaptations to increase the chance of encountering their next host, such as exploiting intermediate hosts at sub-lethal levels [4–7], manipulating host behaviour to increase transmission [8,9] and being a generalist when there are more host species to infect [1,10]. Once in the next host, trophically transmitted worms must avoid being digested, evade host immune responses and migrate to their final infection site—they must establish infection. Comparisons of establishment probabilities across parasite life stages are lacking but they could elucidate how risky complex life cycles truly are.

Higher establishment rates and an increased likelihood of life cycle completion may be tied to worm growth. Transitional body size is linked to fitness in models of complex life cycle evolution for both free-living [11–15] and parasitic animals [3,16–18]. Larger parasite larvae are expected to benefit from shorter developmental times in the next host, as less growth is needed to reach an optimal size for infectivity (in an intermediate host) or sexual maturity (in a definitive host) [19]. Additionally, larger larvae could have higher establishment if they are more resistant to host immune responses, e.g. it likely takes more toxin-releasing leukocytes to expel bigger worms [20]. Larger worms may also be better at avoiding digestion and reaching infection sites, especially if size correlates with energy content [21]. Size can be a proxy for ontogeny too, with larger larvae having developed the structures needed to infect the next host [22]. Experiments have found infection success to increase with larval worm size in some cases [23–25], but not others [26–28]. These studies manipulated size by different means (e.g. by changing infection intensity, host size or host diet), which can affect the relationship between size and establishment [29]. Furthermore, intraspecific variation in larval size is limited, so identifying a correlation between transitional worm size and establishment probability may be easier across species.

Size-dependent establishment could explain some counterintuitive worm growth strategies [16,17]. Helminths commonly grow in intermediate hosts before becoming infective to the next host, but this growth is risky, because intermediate hosts, especially at the start of a life cycle, are almost universally smaller than subsequent hosts. Small hosts can only provide worms with so much space and energy to grow [30–32], and they have shorter lifespans [33]. Instead of growing, an alternative strategy in a small host would be to immediately develop infectivity, await transmission, and then grow larger with lower mortality in the next host. But using small intermediate hosts for just transportation is rare in helminths [34,35], suggesting that larval growth is generally favourable. Growth in short-lived hosts may be worth the risk if it elevates establishment rates.

As complex life cycles progress, parasites exploit bigger hosts, and bigger hosts may be more challenging to infect. They have thicker tissues, preventing penetration [36], and they are usually vertebrates with more sophisticated immune systems [37,38]. Big animals also harbour more diverse parasite communities [39–41], so competition among parasites could limit establishment [42], though this is less relevant in laboratory infections with single parasite species. Alternatively, competition might decrease with host size, since bigger hosts provide parasites more space [43] and energy [44]. Large, long-lived hosts also avoid aggressive immune responses that cause collateral tissue damage [45,46], perhaps facilitating parasite establishment. How establishment varies with host size affects how risk accumulates in longer life cycles. For example, if parasites were less able to infect larger hosts, then the likelihood of completing a cycle would decrease more than exponentially with life cycle length, because establishment rates would decline in each successive, bigger host. Lower establishment in larger hosts was observed in a comparative analysis of trematode metacercaria infections [47], but it is not known if this is a general pattern.

To explore how establishment varies across complex parasite life cycles, we collected data from experimental helminth infections with nematodes, cestodes and acanthocephalans. Specifically, we analysed recovery rates: the proportion of recovered parasites from a given dose. We tested whether parasite recovery (i) increases with life cycle progression, (ii) increases with worm size and (iii) decreases with host size. We also assessed whether factors predicting lower recovery were associated with higher doses, which might imply that people use high doses when they expect fewer worms to establish infection. Our results suggest that larval growth mitigates some of the cumulative risk associated with longer life cycles.

2. Methods

(a). Data collection

We searched for recovery rates in 1112 studies, which were part of a helminth life cycle database [48]. The database covered nematodes, cestodes and acanthocephalans, but not trematodes. No comparable database exists for trematodes, although interspecific variation in trematode infectivity has been examined [47,49]. Known life cycles are not equally distributed among helminth taxa, so we explored how taxonomic biases in the life cycle database might carry over into our recovery rate compilation (see supplement ‘Taxonomic biases’). We only included experiments that reported sample size (number of hosts from which a recovery rate was calculated), dose, time until dissection and the number of worms recovered, which narrowed the data to 153 studies. Recovery rate was the ratio of recovered, living parasites to the administered dose.

We recorded whether the first, second or third host in the life cycle was exposed, as well as whether it was an intermediate or definitive host. Parasite size at transmission was extracted from the life cycle database [48]. For example, if we had a recovery rate for parasite x in the second host, then we extracted parasite x's average final size in the first host. Worm lengths and widths were converted to biovolume based on a stage's shape (e.g. volume of a cylinder for thread-like worms, an ellipsoid for round eggs or a ribbon for flatworms). Biovolumes approximate masses when tissue density is constant (e.g. approx. 1.1 g cm⁻³; [50,51]). We also obtained host masses from various databases, such as Pantheria [52], EltonTraits [53], fishbase [54] and the Encyclopedia of Life [55]. Parasite size was missing for approximately 14% of the infections in our data. We imputed reasonable sizes from related species and re-ran the analyses. Imputation did not qualitatively change the results, so we present unimputed results.

Related species may have similar recovery rates. To assess this, we acquired parasite taxonomies from the Open Tree Taxonomy [56], and we built a molecular phylogeny based on ribosomal and mitochondrial sequences (see supplement ‘Phylogenetic inference’). We obtained sequences for 97 species (76% of the total), and we added the rest to the tree according to their taxonomy.

(b). Analyses

We analysed recovery rates with generalized linear mixed models that assumed binomial errors and additive overdispersion [57,58]. All mixed models were fitted with the R package MCMCglmm [59] and we calculated R² values according to Johnson [60].

The study was treated as a random effect. Often, multiple recovery rates were reported per study because a given study might have used several doses, host species and/or infection durations. For a given experimental condition (i.e. dose and time of dissection), studies presented recovery rates either for individual hosts (e.g. host 1 was given y worms with z recovered; host 2 was given y worms…) or for groups of hosts (e.g. n hosts were given y worms each with an average recovery of z worms per host). To make recovery rates comparable across studies, we pooled individual hosts exposed under the same conditions and calculated the total number of administered and recovered worms. Thus, the data are neither at the individual host level (which would inflate the variance within some studies relative to others) nor at the study level (which would ignore important variation within studies due to e.g. different dissection times), but rather at the ‘condition’ level.

We included time until dissection (i.e. days post exposure) as a fixed covariate because we expected recovery rates to decrease with time due to parasite mortality. This was the case overall (figure 1a), but within studies, recovery rates could increase, decrease, or be stable over time (figure 1b–d). We, therefore, allowed each study to have a different relationship with time (i.e. a random slopes model [64]). Studies where recovery increased with time were unexpected, though not unprecedented [49], as small parasite larvae can be undercounted shortly after exposure. To assess whether undercounts could bias the results, we re-ran analyses after excluding infections where recovery rates within 4 days post exposure were 50% below the study's mean recovery.

Figure 1.

(a) Recovery rate (worms recovered/dose) decreased with time from exposure until dissection across all infections. However, within studies, recovery rates could decrease (b), be stable (c) or increase (d) with time until dissection. Lines and 95% credible intervals were estimated with mixed models that accounted for the variable time-dependent recovery patterns among studies (i.e. random slopes). Data in (b), (c) and (d) were from [61–63], respectively.

To explore phylogenetic effects, we included the phylogenetic covariance matrix as a random effect [65]. However, phylogeny was confounded with the study because most studies included just one or a few related species. As a result, the phylogenetic variance component was hard to estimate, often small, and did not affect other model parameters (electronic supplementary material, figure S4). Similarly, the model was not improved by adding a nested random effect for worm taxonomy, specifically the higher taxonomic levels above genus that are less confounded with the study (likelihood ratio test, p = 0.86). Thus, recovery rates did not exhibit clear phylogenetic structure after accounting for study effects, and we would not necessarily expect such structure because different life stages of the same parasite species can have quite different recovery rates. Therefore, we did not include phylogeny in our final models.

Recovery rates from more hosts are presumably more accurate than those from fewer hosts. We weighted data points by the log10 sample size, e.g. a recovery rate reported from 10 hosts was given twice as much weight as one from one host. However, weighting had little impact on the results, because recovery rates did not vary with sample sizes. As this weighting scheme was subjective, we present results without weighting by sample size.

We added parasite life cycle characteristics as fixed effects to the model. First, we added which host in the life cycle was exposed (e.g. first, second, third), and we distinguished between infections of intermediate versus definitive hosts. Next, we added parasite size and then host mass, both log transformed. We also tested interactions between life cycle stage and worm and host size to assess whether the relationship between size and recovery is the same in first, second and third hosts. Because the significance of model terms can depend on the sequence in which they are added [66], we assessed each term in series, in the full model, and when added alone to the random slopes model accounting for study effects.

Finally, the studies comprising our dataset were focused on parasite morphology, development, physiology, etc. so presumably researchers chose doses to obtain sufficient parasite material. Therefore, we might expect researchers to use higher doses in systems where the expected recovery rates were low. To test this, we re-fit the models but with log-transformed dose as the response (standard linear mixed models with Gaussian errors and no random slopes), and then we examined whether the correlates of recovery rate were also related to dose.

3. Results

We collected 1659 recovery rates from 153 studies for 127 parasite species (10 acanthocephalans, 29 cestodes and 88 nematodes), representing 54 of 123 families in the life cycle database and encompassing the exposure of 16 913 individual hosts (2720 fish, 67 herptiles, 1224 birds, 2544 mammals and 10 358 invertebrates). Although diverse, we note that the recovery rate data included more mammal parasites, fewer reptile and amphibian parasites, and fewer marine parasites than expected from their biodiversity, largely mirroring gaps in life cycle knowledge (electronic supplementary material, figures S1 and S2). Recovery data were also biased towards more intensely studied parasite families (electronic supplementary material, figure S2). However, trends in recovery rates were robust to excluding overrepresented groups from the data (electronic supplementary material, figure S3).

Recovery varied among studies; the study random effect alone accounted for 32% of the variation in recovery rates (table 1). This increased to 44% when allowing recovery rate to vary with dissection time differently in each study (i.e. random slopes; figure 1). Although variable, recovery rates usually decreased with time post infection; of the 84 studies with at least four dissection time points, recovery rate decreased significantly over time in 37 and increased significantly in 25. Recovery increased over time in fewer studies (22) after we excluded 30 infections where recovery seemed underestimated. Excluding these presumed undercounts did not qualitatively change the results, though it did decrease the variance explained by the random slopes from about 12 to 9%.

Table 1.

Generalized linear mixed models (binomial errors) for recovery rate. The initial model included study as a random effect and time of dissection (log transformed days post exposure) as a fixed effect. Then, we sequentially added random slopes (study by time interaction), life cycle stage characteristics, and, finally, parasite and host sizes. Model improvement was assessed with likelihood ratio tests and R². Marginal R² (R²_m, with 95% credible intervals) represents the variation explained by fixed effects, while conditional R² (R²_m) represents that explained by random and fixed effects combined. We calculated R² values for each term alone and in sequence to assess their individual and combined effects on recovery rate.

step	d.f.	R2m		R2c
		alone	sequence	alone	sequence
variation within and among studies
study random effect and dissection time fixed effect (random intercepts)*a	2	0.011 [0.004–0.021]	0.011 [0.004–0.021]	0.322 [0.268–0.387]	0.322 [0.268–0.387]
+ study × dissection time (random slopes)*a	2	0.005 [0–0.022]	0.005 [0–0.022]	0.437 [0.374–0.499]	0.437 [0.374–0.499]
life cycle stage
+ next host in cycle (1st host, 2nd host, 3rd host)*a	2	0.059 [0.025–0.107]	0.059 [0.025–0.107]	0.431 [0.368–0.495]	0.431 [0.368–0.495]
+ intermediate or definitive host?*	1	0.035 [0.013–0.064]	0.081 [0.045–0.127]	0.49 [0.425–0.557]	0.475 [0.411–0.541]
+ next host × intermediate versus definitive	2	0.096 [0.054–0.144]	0.096 [0.054–0.144]	0.488 [0.426–0.551]	0.488 [0.426–0.551]
worm and host size
+ worm biovolume*a	1	0.075 [0.039–0.117]	0.113 [0.076–0.159]	0.41 [0.353–0.473]	0.468 [0.408–0.529]
+ worm biovolume × stage	5	0.124 [0.082–0.17]	0.124 [0.082–0.17]	0.478 [0.416–0.542]	0.478 [0.416–0.542]
+ host mass*a	1	0.040 [0.016–0.073]	0.161 [0.114–0.214]	0.493 [0.422–0.566]	0.514 [0.45–0.576]
+ host mass × stage	5	0.134 [0.083–0.194]	0.176 [0.127–0.231]	0.526 [0.458–0.59]	0.526 [0.461–0.591]

*p < 0.01, likelihood ratio test for term when added in sequence.

^aModel parameter is significant (p < 0.01) in final model.

The average recovery rate was 11% [95% CI = 7–16%], 29% [20–39%] and 46% [25–70%] in first, second and third hosts, respectively (figure 2). The model also suggested that recovery rates were higher in first hosts that were intermediate hosts than first hosts that were definitive hosts (figure 2). However, this trend appears to be explained by other predictors like host mass (i.e. first intermediate hosts are smaller than first definitive hosts), because parameters contrasting intermediate and definitive hosts were not significant in the full model (table 1).

Figure 2.

Recovery rates by parasite life stage, i.e. whether the exposed host was the first, second or third host in the life cycle and whether it was an intermediate or definitive host. Observed recovery rates are overlaid on boxplots. Below each box is the number of species. Larger points are means and 95% credible intervals estimated with mixed models accounting for variation among studies. Means were estimated at the median dissection time (18 days post exposure).

Larger parasite larvae had higher recovery rates (figure 3), with a doubling of worm size estimated to increase the odds of infection 19% (CI = 13–25%). This trend cannot be fully explained by difficulty detecting small worms; we still found recovery to increase significantly with worm size (p = 0.006) after restricting the data to larger, hard-to-overlook stages (i.e. the top third in larval size). Worm size alone explained 7.5% of the variation in recovery rates, but it only explained approximately 2% of the variation beyond that accounted for by life stage (table 1), suggesting that parasite growth may cause the increased recovery rates in later life stages (figure 2). The parasite size by stage interaction was borderline significant (p = 0.06) but weak, explaining an additional 1% of the variation in recovery (table 1 and figure 3).

Figure 3.

Recovery rate as a function of worm size. Colours indicate whether the first, second or third host in the life cycle was exposed (number of parasite species, n = 54, 60 and 14, respectively). Lines and 95% credible intervals were estimated with mixed models accounting for variation among studies. (Online version in colour.)

Recovery rates decreased with host mass within life cycle stages but not across them (figure 4). Controlling for life stage, a doubling of host mass decreased the odds of infection 9% (CI = 5–12%). The rate of decrease weakened from first to second to third hosts (estimated slopes: −0.17, −0.09 and 0.01, respectively; figure 4), but this trend was not statistically clear, as the host mass by stage interaction was not significant and explained just 1% of the variation in recovery (table 1).

Figure 4.

Recovery rate as a function of host mass. Colours indicate whether the first, second or third host in the life cycle was exposed (number of parasite species, n = 54, 60 and 14, respectively). Lines and 95% credible intervals were estimated with mixed models accounting for variation among studies. (Online version in colour.)

Higher doses were used with earlier life stages, smaller worms and larger hosts (figure 5a,b). Furthermore, dose was negatively correlated with the recovery rates predicted by the full model in table 1 (figure 5c).

Figure 5.

Dose as a function of (a) worm size, (b) host mass and (c) predicted recovery rates from parasite life stage, parasite size and host mass (i.e. predictions from the final model in table 1). Colours indicate which host in the life cycle was exposed. Lines and 95% credible intervals were estimated with mixed models accounting for variation among studies. (Online version in colour.)

4. Discussion

Helminth establishment rates increased with life cycle progression, which reduces mortality risk in longer life cycles. For example, the average recovery rate of helminth propagules given to their first host was 11%. If this establishment probability were constant (and assuming the next host was always encountered), then roughly 1 in 750 worms (0.11³ = 0.13%) would complete a three-host life cycle. But recovery rates increased with life cycle progression, such that about 1 in 68 worms (0.11 × 0.29 × 0.46 = 1.5%) would complete a three-host life cycle under the same assumptions. Better establishment rates, thus, substantially raise the likelihood of completing complex life cycles. Several experiments have reported that parasite larvae taken from a facultative intermediate host have higher establishment than larvae that did not infect the facultative host (e.g. [67,68], review in [69]), further hinting that exploiting intermediate hosts can increase worm establishment in subsequent hosts. This trend may exacerbate parasite aggregation at later life stages [70] though perhaps not universally [71,72].

The increase in establishment seemed to be caused by parasite growth, as larger infectious stages had higher recovery rates both within and across stages. Detection bias may play a part in this trend. Smaller parasites can be harder to find and count, leading to lower recovery. However, recovery still increased with worm size after we excluded apparent undercounts as well as when we only analysed larger worms that are difficult to overlook. Larger parasites may be harder for hosts to kill [73] and/or they may be in better condition and thus able to avoid digestion and migrate to infection sites. However, body size need not reflect energy content in larval worms [74]. Smaller worms grown under resource limitations, such as in smaller hosts or in higher-intensity infections (i.e. crowding), had lower establishment in some studies [22–24] but not others [26–28]. Such mixed results partly stem from recovery rates being inherently variable. Life cycle stage, worm size and host size together only accounted for 17% of the variation in recovery rates; the remaining 83% was between (35%) and within studies (48%) and presumably attributable to a slew of unmeasured factors (e.g. parasite and host genetics, age, rearing, etc.). Furthermore, in experiments with a single species, larval sizes can only be manipulated so much and perhaps not enough to notably alter establishment. In our analysis, worm size spanned nine orders of magnitude, and we identified a clear trend for larger worms to have higher recovery rates.

Increased establishment at larger sizes may explain why worms normally grow in small intermediate hosts. Intermediate hosts are almost always smaller than definitive hosts, so worms cannot grow as large in them. Furthermore, small hosts have short life expectancies that can be cut even shorter by excessive worm growth [3,5,75]. So why should worms grow in small hosts at the start of their life cycle when they can grow larger and safer in subsequent hosts? Besides shortening maturation time in the next host [16,19], larval growth increases establishment and higher establishment offsets some, though perhaps not all, of the mortality risks. For example, we found that a doubling of worm size increased recovery odds 19%. To double in size, helminths require approximately 3 days on average [48], and in 3 days, the survival odds for the smallest intermediate hosts, like copepods, decrease 26% (approx. 0.1 deaths per day; [33]). Thus, time spent growing in small hosts remains risky, even if the associated increase in establishment greatly mitigates this risk.

If worms refrained from growing until their final host, establishment rates might decrease with life cycle progression, because bigger hosts were less susceptible to infection. Lower worm establishment in larger hosts was also found in a comparative analysis of trematode metacercaria [47], as well as in experimental infections with various helminths [76–79], though the trend is not universal [80,81]. Larger hosts may simply have greater physical barriers to infection, like thicker tissues [36], or they may invest more in specific immune defenses [45,46,82], given that they live longer, harbour more parasite biomass [44] and commonly have more diverse parasite communities [39–41]. Whatever the mechanism, low establishment in large hosts hints at barriers to evolving simple, direct life cycles. For example, many parasites first grow in invertebrate intermediate hosts before infecting large grazing mammals, even though these grazers obviously consume tiny parasite propagules, as some helminths infect them directly [83]. In some of these cases, skipping the intermediate host and instead directly infecting big hosts as small eggs could result in unsustainably low establishment rates. Moreover, the benefits of larval growth in an intermediate host seem to outpace the costs of infecting larger hosts directly, since recovery increased faster with parasite size than it decreased with host size.

In experimental infections, doses are chosen by researchers, and we found that higher doses were used in experiments with lower recovery. High doses can cause lower recovery through competition among worms and/or stronger host immune responses. Several studies in our dataset reported negative dose-dependent recovery rates (e.g. [84–86]) as have comparative analyses of trematode infectivity [47,49]. However, given the breadth of host–parasite systems included in our analysis (doses varied from 1 to over 10 000), we think causation mostly goes the other way: lower recovery incentivizes researchers to use higher doses. The goal of experimental infections is often obtaining sufficient material to study parasite morphology, development, physiology, etc., and presumably, people draw on their experience with a given host–parasite system to choose appropriate doses. If so, using higher doses with larger hosts and smaller infective stages indirectly supports our finding that recovery is lower in these systems. Then again, using smaller doses with larger worms is also certainly influenced by how easy it is to obtain infective stages; collecting thousands of parasite eggs is often simpler than generating thousands of larvae in intermediate hosts. In any case, we hope these trends in recovery rates inform dose choices in future experiments, though we acknowledge that picking doses is complicated by the high variability in recovery among studies.

To conclude, we compared infectivity across diverse helminth species, although well-studied mammal parasites were overrepresented. We found that helminths have a better chance of establishing infection at each successive life stage, even as they infect bigger, less susceptible hosts. The increase in establishment with life cycle progression seems driven by growth, with larger parasites having higher establishment rates. Thus, growing in the small and short-lived hosts at the start of a complex life cycle, though dangerous, may substantially improve parasites' chances of completing their life cycle.

Data accessibility

Data are available from the Dryad Digital Repository: https://dx.doi.org/10.5061/dryad.2rbnzs7n0 [87]. Code to reproduce our analyses can be found here: https://github.com/dbenesh82/comparative_helminth_infectivity.

Authors' contributions

S.F. collected the data. All authors participated in analyses and drafting the manuscript. D.P.B. conceived, designed and coordinated the study and wrote the final version of the manuscript.

Competing interests

We declare we have no competing interests.

Funding

This study was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation)—project number BE 5336/3-1.