Aggregation of parasites: description, causes and consequences
In a nutshell: aggregation describes the non-random (non-uniform)
distribution of parasites within hosts; it is a nearly ubiquitous
observation that in a parasite population, some individuals have lots
of parasites while most have very few. Over the next two (three?)
lectures, we'll talk about why this happens (what ecological
mechanisms drive the distribution of parasites among hosts?) and why
it matters.
I'll also go over some of the results from Shaw and Dobson:
these are about variation in mean parasite burden
(Chicken and egg problem: describing characteristics before
showing examples, showing examples before describing characteristics)
Examples
(S&D figure)
Describing aggregation
Characteristics of distributions:
- Mean (sometimes all we know about macroparasite burdens)
- Variance
- Incidence (by definition, the only thing we keep track of for microparasitic infections)
(reminder of microparasite/macroparasite distinction)
All of these are connected: e.g., you can't change the incidence
without
affecting the mean and variance as well.
These summaries themselves are relatively
"nonparametric"; don't assume a specific distribution
or parasite population process.
Expected null values/null distributions:
- Even (variance=0)
- Poisson (variance=mean)
- Normal (variance=whatever)
- Log normal (variance=whatever, but more skewed than a normal)
- Negative binomial distribution (variance>mean; var = m*(1+m/k)) [?
check ED]
(for large values, Poisson approaches evenness ...)
Each of these implies a particular relationship (or relationships)
among mean, variance, incidence.
Importance of null hypotheses but also importance of
getting beyond null hypothesis.
You have to know what you would expect if "nothing interesting were
happening", but you also have to be able to specify what categories of
interesting things you expect to happen, and what the data would look
like if they did.
Description by itself isn't enough, particularly if one set of data
could correspond to several different ideas about what ecological
processes are occurring.
- "Gini" coefficient (even distribution)
- Variance/mean ratio (Poisson)
- Coefficient of variation (standard dev./mean, Poisson or normal)
- Negative binomial k (neg. bin./Poisson)
- Lloyd's index = mean+v/m-1 (consequences/causes: intensifies competition).
Expected number of non-self competitors within a host (can get
relative strength by dividing by mean)
- Taylor's power law (exponents between 1 and 2=between Poisson and Normal-with-constant-CV)
Remember the importance of counting zeros!
Also, note that sample size biases estimates of aggregation downwards.
Data
Shaw and Dobson:
- parasite burden is log-normally distributed (does it make sense to
count numbers? Perhaps in some contexts biomass, or biomass relative
to host body size, would be a better measurement of parasite burden?
Distinction between different ways of thinking about populations
(physiology/ecology, individuals/amounts)
- apparent restriction in mean burden: 90% of mean abundances
between 0.1 and 100. (What might generate these limits?)
- Variance/mean relationship. log var = 1.098 + 1.551 log mean
(or, var=3*mean^1.551) [what does this mean in terms of distributions
and processes?]. High correlation (87% of variation explained);
suggestion of evolutionary/ecological processes acting? "Too
aggregated" vs. "too even"? Connection to virulence (high
virulence=low burdens, evenness); intermediate virulence
hypothesis?
- Prevalence vs mean parasite burden
[Taylor's Power Law fig.]
The rest of Shaw & Dobson mostly talks about variations
in mean burden by ecological type/family/etc. (not classical
phylogenetic analysis, but at least takes phylogenetic associations
into account at some level).
I'm not going to talk about this much, but you should read this
section.
Some of the results are:
- low burdens: 2-host systems
(endoparasitic cestodes, predator-prey mediated transmission)
- high burdens: 1-host systems, nematodes
with passive consumption or active adult transmission stages
The analysis is interesting/suggestive, but suffers from
several problems (as usual: it's easy to nitpick).
- the standard problem of observational data sets: e.g., are 2-host
systems special, or is it just that all of the predator-prey
transmission examples included in the data set are have only 2 hosts?
- how do we disentangle evolutionarily correlated suites of
characters? Which came first, the chickens or the eggs (e.g. endoparasitism,
predator-prey transmission, multiple hosts)? Which characters
"really" drive the observed pattern, or is this a silly question?
- Model selection: bootstrap confidence limits on the "binary trees"
shown in the data set? How sensitive are the results to small chance events?
- Relationships with parasite size/host size/relative size? (Data
aren't good enough ... but ...)
S&D do offer some explanations for the variation around the
mean-variance line (i.e., ecological/evolutionary correlates of
aggregation), but these are mostly group-based and don't offer
real ecological correlations. It's hard to say why.
More work would, as usual, be necessary to make firm conclusions
about why burdens are higher/lower in these groups.
(Tree-based analysis of variance/mean etc.? Probably not enough
information left in the residuals around the variance-mean line.)
- higher aggregation in some trophically and passively transmitted
nematodes of inverts
- lower aggregation in dipteran parasites of cattle and reindeer
(mobile adults spread around disease)
- in tapeworms and in trophically transmitted parasites with
invert intermediate hosts, aggregation decreases with average
burden ("ceiling" on parasite burden?)
Causes
If we don't have quite enough information to mine aggregate data sets for
hypotheses about causes of aggregation, can we just think about
possible reasons/investigate particular cases instead?
There are some very basic mechanisms that can generate or suppress
aggregation.
- Direct reproduction within individuals: the
wormy get wormier ... increases aggregation.
There are other mechanisms amplifying infection:
parasite pheromones (ectoparasitic arthropods)
[cf bark beetles: parasitoids of pine trees?];
changes in behavior; breakdown in immunity because
of nutrition/pathology; etc..
- Heterogeneity in exposure:
e.g. clumped dispersal (e.g. picking up a whole batch
of eggs/larvae at the same time),
other kinds of
variation in exposure in time and space
(because of parasite prevalence, environmental conditions) ...
increases aggregation.
(Keymer and Anderson 1979 experimental work on spatial
variability, other work on temporal variability.)
This is one of the basic ideas behind the generation of the
negative binomial distributions; it's also why "inappropriate"
lumping (the definition of which depends on your question) leads to
higher estimates of aggregation.
- Heterogeneity between individuals
(phenotypic or genotypic), in
resistance/tolerance etc. (or perhaps in behavior,
which would translate to heterogeneity in exposure).
Evidence:
Heterogeneity in exposure or resistance can occur at different levels:
- inbred or clonal organisms (mice?) in the lab:
variation due to "luck" or exposure.
(How would we expect this "aggregation" to show up
- genetic heterogeneity: the same organisms in a controlled
environment, but allow natural variation (how much is there??
there is some evidence, but much of it is from (i) hybrid zones or
(ii) agricultural populations, neither of which might tell us all that
much about wild populations (although there are differences between
strains in susceptibility to infection). This is a tough question:
once again, the question of maintenance of genetic variability for
apparently favorable traits comes up. Could be RQ, other kinds of
nonequilibrium dynamics (e.g. recently introduced diseases),
drift, mutation-selection balance, ?).
In one simple case (a cestode, Hymenolepis citelli in rodents)
resistance follows simple Mendelian genetics (only homozygous
recessives get the parasite [cf "genetic diseases", where lethal
recessives are maintained in populations]).
- environmental variability between populations
We can go beyond this and think about other
levels: between-individual, between-family,
between-subpopulation, etc. etc.. They can also interact.
(S&D do point out that unlike in other ecological studies,
we at least have a natural "population" to measure, which
prevents some kinds of inappropriate grouping.)
Density-dependent host mortality, or
density-dependent increases in parasite mortality or decreases in
fecundity and growth (indirectly affecting fecundity)
decrease aggregation (think back to Lloyd's index).
An extreme example of this is concomitant immunity, where
the first parasite into a host actually primes the host's immune
system to reject future parasitic attacks (e.g. human filariasis).
An extreme example of evenness (given by Poulin) is the
copepod Leposphilus labrei, parasitic on wrasse:
1922/1924 infected fish had one copepod, the other two had two.
(Note that for this particular example we don't care how many had
zero, but this could be extremely relevant in other contexts.
We might want to treat this disease as "microparasitic".