Declines in large wildlife increase landscape-level
prevalence of rodent-borne disease in Africa
Hillary S. Younga,b,c, Rodolfo Dirzob,1, Kristofer M. Helgenc, Douglas J. McCauleya,b, Sarah A. Billeterd,
Michael Y. Kosoyd, Lynn M. Osikowiczd, Daniel J. Salkelde,f, Truman P. Youngg, and Katharina Dittmarh
aDepartment of Ecology, Evolution, and Marine Biology, University of California, Santa Barbara, CA 93106; bDepartment of Biology and fWoods Institute for
the Environment, Stanford University, Stanford, CA 94305; cDivision of Mammals, National Museum of Natural History, Smithsonian Institution, Washington,
DC 20013; dDivision of Vector-Borne Infectious Diseases, National Center for Infectious Diseases, Centers for Disease Control and Prevention, Fort Collins,
CO 80521; eDepartment of Biology, Colorado Sate University, Fort Collins, CO 80523; gDepartment of Plant Sciences, University of California, Davis, CA 95616;
and hDepartment of Biological Sciences, University at Buffalo, The State University of New York, Buffalo, NY 14260
Contributed by Rodolfo Dirzo, March 26, 2014 (sent for review December 10, 2013)
Populations of large wildlife are declining on local and global
scales. The impacts of this pulse of size-selective defaunation
include cascading changes to smaller animals, particularly rodents,
and alteration of many ecosystem processes and services, poten-
tially involving changes to prevalence and transmission of zoo-
notic disease. Understanding linkages between biodiversity loss
and zoonotic disease is important for both public health and
nature conservation programs, and has been a source of much
recent scientific debate. In the case of rodent-borne zoonoses,
there is strong conceptual support, but limited empirical evidence,
for the hypothesis that defaunation, the loss of large wildlife,
increases zoonotic disease risk by directly or indirectly releasing
controls on rodent density. We tested this hypothesis by experi-
mentally excluding large wildlife from a savanna ecosystem in East
Africa, and examining changes in prevalence and abundance of
Bartonella spp. infection in rodents and their flea vectors. We
found no effect of wildlife removal on per capita prevalence of
Bartonella infection in either rodents or fleas. However, because
rodent and, consequently, flea abundance doubled following ex-
perimental defaunation, the density of infected hosts and infected
fleas was roughly twofold higher in sites where large wildlife was
absent. Thus, defaunation represents an elevated risk in Bartonella
transmission to humans (bartonellosis). Our results (i ) provide
experimental evidence of large wildlife defaunation increasing
landscape-level disease prevalence, (ii) highlight the importance
of susceptible host regulation pathways and host/vector den-
sity responses in biodiversity–disease relationships, and (iii) sug-
gest that rodent-borne disease responses to large wildlife loss
may represent an important context where this relationship is
largely negative.
Kenya
|
dilution effect
We are in the midst of a global extinction crisis (1). Amongmammals, for which population trends are known, more
than 50% of species are currently declining (2). The current
pulse of global defaunation is widely recognized to be size-
selective, with larger species exhibiting greater risks for popula-
tion declines or extinctions (3–5). This bias derives from a variety
of factors, including harvester preference for large-bodied species,
expansive habitat requirements of these large species, and the
fact that large animals typically have life history traits associated
with slow population growth (e.g., low fertility, long generation
time, later age of reproduction) (6, 7). Body size is also strongly
correlated with functional roles (8), and large species thus often
play functionally distinct and impactful roles in ecosystems (9).
The systematic decline of large species, both herbivores and pred-
ators, is thus often associated with pronounced effects on other
aspects of community composition and structure (5, 10, 11), eco-
system function (11), and even evolutionary trajectories (12).
Recently, there has been growing interest in understanding
what the effects of wildlife declines of this type may be for the
emergence and prevalence of infectious zoonotic diseases, with
active debate on both the direction and generality of diversity–
disease relationships (13–18) and on the likely implications for
human health (19, 20). Characterizing the nature of this re-
lationship has broad significance. Zoonotic disease agents pres-
ent a growing threat to global health. At least 60% of all human
disease agents are zoonotic in origin (21), and mammals are the
primary reservoir hosts for most known zoonotic diseases in
humans (22, 23). The suggestion that anthropogenically driven
wildlife declines may lead to increased disease risk in a landscape
has thus generated a great deal of interest because it raises the
possibility that conservation of intact natural landscapes may be
an effective intervention strategy for mediating emerging threats
to public health. Much of the research and debate on this re-
lationship thus far has focused on the mechanism of transmission
interference, or the process by which systematic changes in
community richness (number of species), composition (identity
of species), and host competence (the proportion of individuals
of a species that can maintain and transmit infections) affects
community competence and, ultimately, prevalence of pathogens
(proportion of hosts infected) in a community, without neces-
sarily changing the absolute abundance of susceptible hosts
(individuals that can be infected by a pathogen) (15, 24, 25). The
present study focuses primarily instead on a second, hitherto less
explored pathway, susceptible host regulation, or the process by
which biodiversity loss changes the abundance of susceptible
hosts (26).
Rodents are common reservoir hosts (long-term source hosts
for a pathogen) for many human zoonotic pathogens, such as
Borrelia burgdorferi (Lyme disease) (24), hantaviruses [hantavirus
Significance
Understanding the effects of biodiversity loss on zoonotic
disease is of pressing importance to both conservation science
and public health. This paper provides experimental evidence
of increased landscape-level disease risk following declines in
large wildlife, using the case study of the rodent-borne zoo-
nosis, bartonellosis, in East Africa. This pattern is driven not by
changes in community composition or diversity of hosts, as
frequently proposed in other systems, but by increases in
abundance of susceptible hosts following large mammal declines.
Given that rodent increases following large wildlife declines
appear to be a widespread pattern, we suggest this relationship
is likely to be general.
Author contributions: H.S.Y., R.D., K.M.H., D.J.M., and T.P.Y. designed research; H.S.Y.,
R.D., K.M.H., D.J.M., S.A.B., M.Y.K., L.M.O., D.J.S., and K.D. performed research; H.S.Y. and
R.D. analyzed data; and H.S.Y., R.D., K.M.H., and D.J.M. wrote the paper.
The authors declare no conflict of interest.
Freely available online through the PNAS open access option.
1To whom correspondence should be addressed. E-mail: rdirzo@stanford.edu.
This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
1073/pnas.1404958111/-/DCSupplemental.
7036–7041 | PNAS | May 13, 2014 | vol. 111 | no. 19 www.pnas.org/cgi/doi/10.1073/pnas.1404958111
pulmonary syndrome (HPS)] (27), Yersinia pestis (plague) (28),
and Bartonella (bartonellosis) (29). They are particularly im-
portant hosts for flea-borne diseases, which are absent or low in
prevalence in most larger wild animals, many of which do not
carry fleas (30). A great deal of work in both the Lyme disease
and HPS systems has documented relationships between host
species richness, community composition, and prevalence of
these pathogens (31, 32). Changes in susceptible host abundance
would appear to be a particularly likely pathway by which wildlife
loss and disturbance could affect rodent-borne disease risk. By
contrast to larger-bodied mammals, rodents, particularly small-
sized species (i.e., <500 g), are often relatively robust to human
disturbance and many species live commensally with humans (33,
34). Due to their generally rapid reproductive rates and small
home range sizes, populations can fluctuate dramatically over both
small spatial and temporal scales, and in response to declines or
removals of either rodent predators or rodent competitors, in-
cluding large herbivores (35–37). Because both large predators
and herbivores face a high risk of decline from human distur-
bances (6, 10), susceptible host regulation may be a strong po-
tential pathway by which wildlife loss can affect human disease risk.
There is theoretical support and empirical evidence of a neg-
ative correlation between large wildlife loss and increased ro-
dent-borne disease risk (26, 27, 38). However, we largely lack
experimental data that document if and how the cascading
effects of large wildlife loss have an impact on rodent-borne
zoonoses via susceptible host regulation. In this study, we look at
the overall effects of declines in large wildlife, particularly large
herbivores (>15 kg), on landscape-level risk for the rodent-borne
pathogen Bartonella spp., known to affect human health nega-
tively (bartonellosis) (29). Bartonella is a globally distributed,
facultative intracellular bacterial parasite (Alphaproteobacteria)
that is found in a wide variety of mammals, including humans
and rodents. The identity and pathogenicity of many species
of Bartonella to humans are unknown, particularly in Africa.
However, many species, including several of the species detected
in this study (e.g., Bartonella rochalimae, Bartonella grahamii,
Bartonella elizabethae) are known or suspected to be human
pathogens. Bartonella spp. infection is characterized by long-
lasting intraerythrocytic, relapsing bacteremia that can induce
pathology in multiple organ systems in humans (29). Although
host specificity varies widely, many strains appear to have
rodents as the primary reservoir hosts (39, 40). Bartonella also
appears to be primarily vector-borne [although direct and ver-
tical transmission, from parent to offspring, may also occur (40)],
with fleas being confirmed vectors (41, 42). Bartonella is in-
creasingly being recognized as a zoonotic disease with implica-
tions for human and animal health (29).
We conducted this work in a long-term and well-controlled,
replicated large herbivore removal experiment based in central
Kenya, the Kenya Long-Term Exclosure Experiment (KLEE)
(43). Within this experimental setup, we examined the effects of
large wildlife removal on four different metrics of Bartonella dis-
ease risk: (i) the abundance of infected hosts (rodents), (ii) the
prevalence of infection in these hosts, (iii) the abundance of in-
fected vectors (fleas), and (iv) the infection prevalence in vectors.
As research and debate continue to expand regarding the di-
rection, magnitude, and generality of the relationship between
biodiversity and disease, there is consensus that no single re-
lationship is likely to be universal across pathogens and systems
(15–18, 44, 45). Thus, to make knowledge on the diversity–dis-
ease relationship useful to managers, policy makers, and scien-
tists, research needs to focus on the following: (i) understanding
the context (e.g., types of hosts, vectors, pathogens, disturbance)
in which either positive or negative relationships are most likely
and (ii) understanding the mechanisms by which these rela-
tionships occur (19). Here, based on our results from infection
prevalence of Bartonella spp. in rodents under experimental
defaunation, we argue that the intersection of large wildlife de-
cline and rodent-borne disease may provide a particularly likely
context for negative diversity–disease relationships to be general.
Additionally, we conclude that susceptible host regulation, rather
than transmission interference, likely drives the relationship in
this ecological context.
Results
Rodent and Flea Abundance. A total of 832 rodents, representing
11 species, were captured in five sampling periods in the KLEE
(Fig. 1A). Treatments in which we simulated the loss of large
wildlife demonstrated strong and consistent elevation (compared
with control treatment) in total rodent density per unit area,
despite pronounced seasonal variation in rodent abundance
(Fig. S1; F1,4 = 54.9, P < 0.01). This pattern is quite robust across
years and seasons, having been shown to persist in the KLEE
for more than a decade despite large fluctuations in rainfall
patterns (32, 35, 46). One species, Saccostomus mearnsi (Mearns’s
pouched mouse), dominated capture rates in both exclosures and
open plots, accounting for 75% of all captures. The same pattern
of roughly doubled abundance in exclosure plots across time was
observed when considering just S. mearnsi (Fig. 1B). There was
no significant change observed in rodent diversity between con-
trol and large mammal removal treatments (Fig. 2A; Shannon
diversity index: F1,4 = 0.3, P = 0.6) and no significant change in
overall community composition [analysis of similarity (ANOSIM):
R = 0.30, P = 0.20]. There was also no significant difference in
age or sex structure of populations between exclosure and con-
trol plots (data not shown). Thus, for the rodent community, only
abundance varied significantly among plots.
Fleas comprised more than 95% of the total ectoparasites
sampled from captured rodents, with only occasional lice or
mites (Androlaelaps spp.) and no ticks observed. A total of 1,570
fleas were surveyed in this study. Flea diversity encompassed
four genera. From the genus Xenopsylla, three species were iden-
tified: Xenopsylla brasiliensis, Xenopsylla cheopis (aequisetosa), and
Xenopsylla sarodes sarodes. Other species found were Dinopsyllus
Fig. 1. Working in the KLEE, which examines the effect of large wildlife
(e.g., zebra, giraffe, elephant, gazelle) removal (A), we found that S. mearnsi
abundance was significantly higher (roughly double on average) in plots
where large mammals had been removed compared with open plots (B)
despite strong seasonal variability. (C) There was no significant difference in
the intensity of infestation of fleas per rodent between treatments. (D) As a
result, the density of fleas per hectare is significantly higher (roughly double) in
plots without large wildlife. Error lines represent 1 SE, based on three replicate
blocks. Data for all rodents are shown in Fig. S1. Photography credits: A, KLEE
exclosure, D. Kimuyu (Equus quagga, Nanyuki, Kenya); Inset of flea in C and
D X. sarodes sarodes (male), M. Hastriter and Michael Whiting, Monte L.
Bean Museum, Brigham Young University, Provo, UT.
Young et al. PNAS | May 13, 2014 | vol. 111 | no. 19 | 7037
EC
OL
OG
Y
lypusus, Parapulex echinatus, and Ctenophthalmus calceatus cabirus.
More than 98% of all fleas were Xenopsylla spp. (Fig. S2), and
there was no difference in flea diversity (F1,4 = 0.7, P = 0.5) or
community composition between exclosure and control treat-
ments (ANOSIM: R = 0.15, P = 0.4). Four of the six species of
fleas found were present on S. mearnsi. Three of the species were
found on multiple host species (all host flea associations are
shown in Tables S1 and S2).
S. mearnsi had much higher flea density than any other spe-
cies, and this single species thus accounted for >95% of all fleas
in both exclosure and control plots. Given the very low number
of fleas from other species, and the lack of any differences in
diversity or composition of hosts among treatments, all sub-
sequent analyses were conducted using only S. mearnsi (data with
all rodents are shown in Figs. S1 and S3). There was no signifi-
cant difference in average abundance of fleas per individual
rodent across treatments (Fig. 1C; F1,4 = 4.9, P = 0.09). How-
ever, because of the increases in rodent abundance, there was
more than a doubling (103 ± 23% increase) in total numbers of
fleas per hectare (Fig. 1D; F1,4 = 27.7, P < 0.01). This abundance
pattern was robust and persistent across years (47).
Bartonella Prevalence. We measured Bartonella prevalence (pro-
portion of individuals infected) in both the rodent host
S. mearnsi and its fleas. A total of 157 S. mearnsi individuals were
screened for Bartonella across the three time periods (94 in
exclosures and 63 in open sites). BLAST analyses identified
Bartonella genotypes related to the Sc-tr1 group (from Saccos-
tomus campestris, South Africa) (48), the OY group (including
B. rochalimae, B. grahamii, and B. elizabethae) (49), and several
previously unidentified genotypes.
Bartonella prevalence in hosts did not differ significantly be-
tween treatments with and without wildlife (Fig. S3; F1,4 = 0.4,
P = 0.55), although it did vary across sampling periods (Fig. S3;
F2,3 = 14.6, P = 0.03). There was no significant treatment × time
interaction (F2,3 = 3.7, P = 0.16). Likewise there was no signif-
icant difference in the prevalence of Bartonella in the fleas
sampled from S. mearnsi in control plots compared with wildlife
exclosures (Fig. 3A; F1,4 = 0.6, P = 0.49) or any evidence of
a treatment × time interaction (Fig. 3A; F2,3 = 2.2, P = 0.26).
There was again variation by sampling period (F1,4 = 53.7, P =
0.01). In contrast to these results, the abundance of Bartonella-
infected hosts varied strongly by treatment (Fig. 3B; F1,4 = 19.3,
P = 0.01), with roughly twice as many infected hosts in plots
where wildlife had been removed, due entirely to the higher
density of rodent populations. There was both a sampling period
effect (Fig. 3B; F2,3 = 283.2, P < 0.0001) and a time × treatment
interaction (F2,3 = 55.3, P < 0.01). The abundance of infected
fleas also varied by treatment (Fig. 3C; F1,4 = 10.7, P = 0.03) and
by sampling period (F2,3 = 463.6, P < 0.001). When a time ×
treatment interaction effect occurred, the effects of treatment
were stronger in sampling periods where rodent or vector abun-
dance was higher.
Discussion
As large wildlife continues to decline globally, ecologists, con-
servation scientists, and health practitioners are challenged to
understand and interpret the implications of these changes for
ecological communities and the people who inhabit these spaces.
These results provide strong experimental evidence that the
effects of differential loss of large wildlife can cascade to cause
increases in the abundance of zoonotic pathogens across a land-
scape, via an increase in the abundance of hosts and vectors.
Fig. 2. (A) Rodent diversity (Shannon diversity index ± SE, across plots
within a season) was not significantly different between treatments with
and without large mammalian wildlife. (B) There were also no significant
differences in community similarity between experimental plots (animals
pooled across sampling seasons). Species codes are as follows: ACKE, Acomys
kempi; AEKA, Aethomys kaiseri Arvicanthis niloticus; DEME, Dendromus
melanotis; MANA, Mastomys natalensis; MUAC, Mus cf. acholi; MUMI, Mus
minutoides; MUSO, Mus sorella; MUTE, Mus tenellus; SAME, S. mearnsi; ZEHI,
Zelotomys hildegardeae.
Fig. 3. There was no significant overall difference in the prevalence of
Bartonella spp. infection either in the dominant rodent S. mearnsi or in the
fleas of S. mearnsi (A) between control and large wildlife removal treat-
ments. However, because of increased abundance of S. mearnsi in sites
without large wildlife, there was a significant increase in the abundance of
infected S. mearnsi (B) and infected vectors (C) in these simulated large
wildlife loss treatments. Error lines represent 1 SE, based on three replicate
blocks. Data is qualitatively similar when considering all rodents (Fig. S3).
7038 | www.pnas.org/cgi/doi/10.1073/pnas.1404958111 Young et al.
Notably, the primary mechanism by which large wildlife loss
appears to affect Bartonella risk in this system is through sus-
ceptible host regulation. Although the effects of susceptible host
regulation have been experimentally documented in plant sys-
tems (50) and, for zoonotic diseases, have been suggested to
result from fragmentation (24), this study is, to our knowledge,
the first experimental demonstration of the phenomenon for
a rodent-borne zoonosis. In this case, host regulation may be
occurring due to both direct (competition) and indirect (vege-
tation structural changes) effects of changes in abundance of
large vertebrates (35, 51) rather than via changes in rodent
predators (47), the interacting species more often considered
(14, 38).
Interestingly, we found no support that transmission inter-
ference, as documented in other systems (52, 53), operates in
this system. There were no significant changes in the diversity or
composition of the rodent host community (or of flea commu-
nities) between treatments. Given the much lower density of
larger mammals relative to small mammals in this landscape,
transmission interference is also not likely relevant when the
whole mammal community is considered. There was also no
change in the prevalence of Bartonella spp. in the dominant host,
S. mearnsi, despite its much higher abundance in exclosures.
Although increased prevalence is often seen when densities of
competent hosts increase, this relationship is expected only
when transmission is density-dependent. For pathogens with fre-
quency-dependent transmission, effects of density of prevalence
are not necessary unless contact rates change (54). Other
studies of Bartonella have observed a similar lack of correlation
between Bartonella prevalence and host population density (55),
although positive and delayed relationships have also been
observed (56, 57).
Although it is theoretically possible that the increased number
of infected fleas on rodents in defaunated landscapes could be
offset by fewer fleas from the (absent) large wildlife, such an
effect is highly unlikely. Most of the animals excluded are
large ungulates, which, for ecological and evolutionary reasons,
almost entirely lack fleas, particularly in the tropics of Africa [the
currently known flea fauna on ungulates has a decidedly Hol-
arctic or South Asian distribution (Vermipsyllidae and Ancistro-
psyllidae, respectively)] (30, 58, 59). Although large carnivores
do carry Pulex and Ctenocephalides fleas (not Xenoypsylla fleas),
these large carnivores are at extremely low densities in the
landscape and are just one of many host groups for these fleas
(58). From a human risk perspective, the wild ungulates and
large carnivores excluded in this study are also unlikely to in-
teract closely with humans, making transmission less likely than
rodent–human transmission. However, we expect that diseases
transmitted by ectoparasites more common on ungulates than on
rodents (e.g., predominantly tick-borne pathogens) will show
very different, even inverted, responses (60, 61).
The conclusion that wildlife decline causes a substantial in-
crease in Bartonella risk assumes that abundance of infected
hosts and vectors is the best metric of disease risk. Using per
capita prevalence responses would have led to a conclusion of no
effect. Clearly, the choice of prevalence metric can produce
qualitatively different outcomes and emphasizes the importance
of including infected host and vector abundance as a response
metric (62). We suggest that per capita prevalence, sometimes
used in the discussion of negative biodiversity relationships
through transmission interference (24), may not be the most
relevant metric to many real world disturbance–disease rela-
tionships. In this case, for example, there would need to be
a nearly 50% decrease in per capita prevalence to counter the
amplifying effects of changes in host and vector abundance
following defaunation.
Although abundance of infected vectors is likely a better re-
sponse metric than per capita prevalence, there are still many
questions that need to be answered to assess if this change in
abundance of infected vectors will translate to a proportionate
increase in human disease in those anthropogenic settings. One
set of questions revolves around patterns of human contact with
these dominant S. mearnsi hosts. Although Bartonella spp. in-
fected S. mearnsi were found in human habitations and in more
disturbed agricultural or pastoral settings (SI Text), the presence
of infected rodents is not itself sufficient for transmission. If
humans do not interact closely with these rodents, disease
transmission may be inconsequential regardless of changes in
abundance. Likewise, if human behavior in defaunated habitats
changes human–S. mearnsi contact patterns, the relationship
between number of infected vectors and human risk will be
nonlinear. However, because defaunation seems most likely to
be associated with increased human contact rates (e.g., human
presence is a cause of defaunation, humans are more willing to
spend time in safer defaunated landscapes), we expect such
changes would only exacerbate the effects of defaunation on
transmission, rendering our conclusions of increased risk con-
servative. There are also outstanding questions about human
vector contact. The most common flea in this study, X. cheopis, is
a known generalist that can feed on humans and effectively
transfer other pathogens (58). Because fleas are also known to
transmit Bartonella spp. to humans, it seems likely that fleas in
this system will be able to transmit Bartonella spp. to humans.
However, the effectiveness and frequency of X. cheopis or
X. sarodes transmission of Bartonella spp. to humans remain un-
determined. As with the transmission of most zoonotic patho-
gens to humans, the relationship is likely to be complex and
multifactorial. The density of infected vectors, although an im-
portant and commonly used indicator of risk, remains an im-
perfect metric for assessing true risk to humans. Explicit links to
human health outcomes will be needed to demonstrate this
connection fully.
More generally, however, given the ubiquity of large wildlife
loss; the pervasive and well-substantiated observations of rodent
increases following such defaunation globally (35, 58, 63, 64);
and the fact that rodents are one of the most frequent hosts of
zoonotic diseases (65), particularly flea-borne diseases (58),
these results suggest that size-selective animal loss may have
a major impact on global risk of rodent-borne diseases. This
study also adds to a growing body of evidence from both the
Lyme and hantavirus systems that pathogen regulation through
susceptible host regulation of rodent-borne diseases may be
equally or more important than transmission interference in
some cases (31, 38). We suggest susceptible host regulation
may be an underappreciated ecological function of the pre-
servation of intact mammalian communities. Further research
on other rodent-borne pathogens, particularly with alternative
transmission pathways (e.g., direct transmission) or with higher
host specificity (66), will be critical to assessing the generality of
our findings.
Methods
Study Site. This work was conducted between November 2009 and November
2011 in the Laikipia County of Kenya, at the Mpala Research Centre. To
determine how the loss of large native wildlife influences disease risk, we
conducted all sampling in the KLEE (0°17′ N, 36°52′ E) (43). Established in
1995, the KLEE uses a block design that includes three replicates of different
types of wildlife exclosures. In this experiment, we used only the total
exclosure sites, which effectively exclude all animals larger than 15 kg, and
the control sites, which allow free access to all wildlife. Each experimental
plot in the KLEE is 4 ha in size, but we sampled only the central hectare to
avoid confounding edge effects.
The KLEE is located on “black cotton” soil (nutrient-rich but poorly drained
vertisols with high clay content) and set within an Acacia drepanolobium
savanna-woodland. Mean annual rainfall at the KLEE site is ∼630 mm and
is weakly trimodal. Resident large wildlife in the area includes elephants
(Loxodonta africana), giraffes (Giraffa camelopardalis), zebras [Equus quagga
Young et al. PNAS | May 13, 2014 | vol. 111 | no. 19 | 7039
EC
OL
OG
Y
(Equus burchellii ) and Equus grevyi], and lions (Panthera leo), among
others (43).
Rodent and Ectoparasite Sampling. Rodent trapping followed methodologies
previously used in long-term monitoring of rodents in these plots. Briefly, in
the inner hectare in each of the total exclosure (n = 3) and control (n = 3)
plots, we set 100 Sherman traps, placed on a 10 × 10-m grid with 10-m
spacing. Traps were baited every evening with peanut butter and oats for
three consecutive nights. Trapping occurred in five sampling periods over
a 2-y period: November 2009, March 2010, November 2010, March 2011, and
November 2011. All captured rodents were identified to species (using
morphological and genetic techniques), sexed, weighed, and marked.
Rodent blood was collected onto Whatman paper. Animals were released at
their capture sites. Due to low capture rates of some species, we assessed
abundance of rodents per site per capture period simply as the minimum
number of animals known to be alive, based on the number of unique
individuals captured per site.
Ectoparasites were sampled using a modification of protocols described by
McCauley et al. (46). Essentially, the animal was held over a tub of ethanol
and then combed for 10 strokes to provide an index of infestation intensity.
All ectoparasites that fell into the tub or could be recovered from the
flea comb were collected and counted. If an animal was subsequently
captured again in the same 3-d trapping session, it was not sampled
for fleas a second time. Ectoparasites were morphologically identified to
species level.
Bartonella Surveillance. Bartonella screening was conducted only for rodents
captured in the first three trapping periods (n = 168 hosts screened). This
screening included all rodents captured in November 2009 and March 2010;
given the very high capture rate in November 2010, only a random sub-
sampling (10%) of those rodents was screened. For each selected rodent
host, we (i) screened the host blood for Bartonella and (ii) screened at least
one flea from each animal if it had fleas. For the subset of hosts with mul-
tiple fleas (n = 131), we also screened all remaining fleas individually from
a randomly selected 30% of individuals (n = 373 fleas screened). Extraction,
PCR screening, and analysis were conducted independently at two sites, the
Centers for Disease Control and Prevention and the University at Buffalo,
The State University of New York. Protocols followed those detailed
by Billeter et al. (67) for fleas and by Kosoy et al. (68) for rodent hosts.
Individual fleas were titrated using a sterile needle, and DNA was extracted
using a Qiagen QIAamp tissue kit. Fleas were screened for the presence of
Bartonella DNA using protocols described by Billeter et al. (67), with primers
targeting 767-bp and 357-bp fragments of the citrate synthase gene, gltA.
PCR products were purified using the QIAquick PCR purification kit (Qiagen).
DNA sequences were analyzed using Geneious R6-1 (Biomatters) and Lasergene
version 8 (DNASTAR) software.
Host prevalence of Bartonella was defined as the number of Bartonella-
positive rodents (all Bartonella strains pooled) over the number of tested
rodents. Vector prevalence of Bartonella was defined as the number of
positive fleas over the total number of fleas screened per host, considering
only the subset of data taken from hosts that both had fleas and where all
fleas were screened. Abundance of infected hosts was defined as the
product of total host abundance (minimum number known alive) and host
prevalence of Bartonella. Abundance of infected vectors was defined as the
product of host abundance, mean vector abundance per host, and vector
prevalence of Bartonella. All metrics were calculated per site (three exclo-
sures, three control sites) per sampling interval.
For all rodent and flea analyses, we used factorial, repeated measures
ANOVA. To compare species similarity across host and flea communities, we
conducted ANOSIMs among treatments pooled across sampling periods.
Reported values are mean ± SE unless otherwise noted. All analyses were
conducted in R 2.12.1 (R Development Core Team, 2010).
ACKNOWLEDGMENTS. We thank Cara Brook, Ralph Eckerlin, Jackson
Ekadeli, Frederick Erii, Lauren Gillespie, Lauren Helgen, Ashley Hintz, Helen
Kafka, Felicia Keesing, John Lochikuya, Margaret Kinnaird, Peter Lokeny,
Darrin Lunde, Scott Miller, Mathew Namoni, Everlyn Ndinda, John Ososky,
John Montenieri, Jack Silange, Michael Hastriter, and Michael Whiting for
help in this project. Financial support for this project came from the James
Smithson Fund of the Smithsonian Institution, the National Geographic
Society, the National Science Foundation (Long Term Research in Environ-
mental Biology Grants BSR-97-07477, 03-16402, 08-16453, 12-56034, DEB-09-
09670, and DEB-1213740), the Natural Sciences and Engineering Council of
Canada, the African Elephant Program of the US Fish and Wildlife Service
(Grant 98210-0-G563), the Woods Institute for the Environment, and the
Smithsonian Institution Women’s Committee. Vector images were provided
courtesy of the Integration and Application Network, University of Maryland
Center for Environmental Science (http://ian.umces.edu/imagelibrary).
1. Barnosky AD, et al. (2011) Has the Earth’s sixth mass extinction already arrived?
Nature 471(7336):51–57.
2. Schipper J, et al. (2008) The status of the world’s land and marine mammals: Diversity,
threat, and knowledge. Science 322(5899):225–230.
3. Dirzo R, Mendoza E, Ortíz P (2007) Size-related differential seed predation in a heavily
defaunated neotropical rain forest. Biotropica 39(3):355–362.
4. Fritz SA, Bininda-Emonds OR, Purvis A (2009) Geographical variation in predictors of
mammalian extinction risk: Big is bad, but only in the tropics. Ecol Lett 12(6):538–549.
5. Trebilco R, Baum JK, Salomon AK, Dulvy NK (2013) Ecosystem ecology: Size-based
constraints on the pyramids of life. Trends Ecol Evol 28(7):423–431.
6. Cardillo M, et al. (2005) Multiple causes of high extinction risk in large mammal
species. Science 309(5738):1239–1241.
7. Fritz SA, Purvis A (2010) Phylogenetic diversity does not capture body size variation at
risk in the world’s mammals. Proc Biol Sci 277(1693):2435–2441.
8. Petchey OL, Gaston KJ (2006) Functional diversity: Back to basics and looking forward.
Ecol Lett 9(6):741–758.
9. Bellwood DR, Hoey AS, Choat JH (2003) Limited functional redundancy in high di-
versity systems: Resilience and ecosystem function on coral reefs. Ecol Lett 6(4):
281–285.
10. Estes JA, et al. (2011) Trophic downgrading of planet Earth. Science 333(6040):
301–306.
11. Pringle RM, Young TP, Rubenstein DI, McCauley DJ (2007) Herbivore-initiated in-
teraction cascades and their modulation by productivity in an African savanna. Proc
Natl Acad Sci USA 104(1):193–197.
12. Galetti M, et al. (2013) Functional extinction of birds drives rapid evolutionary
changes in seed size. Science 340(6136):1086–1090.
13. Johnson PT, Preston DL, Hoverman JT, LaFonte BE (2013) Host and parasite diversity
jointly control disease risk in complex communities. Proc Natl Acad Sci USA 110(42):
16916–16921.
14. Ostfeld RS, Holt RD (2004) Are predators good for your health? Evaluating evidence
for top-down regulation of zoonotic disease reservoirs. Front Ecol Environ 2(1):13–20.
15. Ostfeld RS, Keesing F (2012) Effects of host diversity on infectious disease. Annu Rev
Ecol Evol Syst 43(1):157–182.
16. Salkeld DJ, Padgett KA, Jones JH (2013) A meta-analysis suggesting that the re-
lationship between biodiversity and risk of zoonotic pathogen transmission is idio-
syncratic. Ecol Lett 16(5):679–686.
17. Young H, Griffin RH, Wood CL, Nunn CL (2013) Does habitat disturbance increase
infectious disease risk for primates? Ecol Lett 16(5):656–663.
18. Wood CL, Lafferty KD, DeLeo G, Young H, Kuris AM (2014) Does biodiversity protect
against infectious disease? Ecology 95(4):817–832.
19. Myers SS, et al. (2013) Human health impacts of ecosystem alteration. Proc Natl Acad
Sci USA 110(47):18753–18760.
20. Hough RL (2014) Biodiversity and human health: Evidence for causality? Biodivers
Conserv 23(2):267–288.
21. Taylor LH, Latham SM, Woolhouse ME (2001) Risk factors for human disease emer-
gence. Philos Trans R Soc Lond B Biol Sci 356(1411):983–989.
22. Cleaveland S, Laurenson MK, Taylor LH (2001) Diseases of humans and their domestic
mammals: Pathogen characteristics, host range and the risk of emergence. Philos
Trans R Soc Lond B Biol Sci 356(1411):991–999.
23. Wolfe ND, Dunavan CP, Diamond J (2007) Origins of major human infectious diseases.
Nature 447(7142):279–283.
24. LoGiudice K, Ostfeld RS, Schmidt KA, Keesing F (2003) The ecology of infectious
disease: Effects of host diversity and community composition on Lyme disease risk.
Proc Natl Acad Sci USA 100(2):567–571.
25. Lacroix C, et al. (2014) Non-random biodiversity loss underlies predictable increases in
viral disease prevalence. J R Soc Interface 11(92):20130947.
26. Keesing F, Holt RD, Ostfeld RS (2006) Effects of species diversity on disease risk. Ecol
Lett 9(4):485–498.
27. Kallio ER, et al. (2009) Cyclic hantavirus epidemics in humans—Predicted by rodent
host dynamics. Epidemics 1(2):101–107.
28. Gage KL, Kosoy MY (2005) Natural history of plague: Perspectives from more than
a century of research. Annu Rev Entomol 50:505–528.
29. Chomel BB, Kasten RW (2010) Bartonellosis, an increasingly recognized zoonosis.
J Appl Microbiol 109(3):743–750.
30. Whiting MF, Whiting AS, Hastriter MW, Dittmar K (2008) A molecular phylogeny of
fleas (Insecta: Siphonaptera): Origins and host associations. Cladistics 24(5):677–707.
31. Suzán G, et al. (2009) Experimental evidence for reduced rodent diversity causing
increased hantavirus prevalence. PLoS ONE 4(5):e5461.
32. LoGiudice K, et al. (2008) Impact of host community composition on Lyme disease risk.
Ecology 89(10):2841–2849.
33. Mills JN (2006) Biodiversity loss and emerging infectious disease: An example from the
rodent-borne hemorrhagic fevers. Biodiversity 7(1):9–17.
34. Stenseth NC, et al. (2003) Mice, rats, and people: the bio-economics of agricultural
rodent pests. Front Ecol Environ 1(7):367–375.
35. Keesing F (2000) Cryptic consumers and the ecology of an African savanna. Bioscience
50(3):205–215.
7040 | www.pnas.org/cgi/doi/10.1073/pnas.1404958111 Young et al.
36. Leirs H, et al. (1997) Stochastic seasonality and nonlinear density-dependent factors
regulate population size in an African rodent. Nature 389(6647):176–180.
37. Previtali MA, Lima M, Meserve PL, Kelt DA, Gutiérrez JR (2009) Population dynamics
of two sympatric rodents in a variable environment: Rainfall, resource availability,
and predation. Ecology 90(7):1996–2006.
38. Levi T, Kilpatrick AM, Mangel M, Wilmers CC (2012) Deer, predators, and the emer-
gence of Lyme disease. Proc Natl Acad Sci USA 109(27):10942–10947.
39. Hayman DT, McDonald KD, Kosoy MY (2013) Evolutionary history of rat-borne Bar-
tonella: The importance of commensal rats in the dissemination of bacterial in-
fections globally. Ecol Evol 3(10):3195–3203.
40. Kosoy MY, et al. (1998) Isolation of Bartonella spp. from embryos and neonates of
naturally infected rodents. J Wildl Dis 34(2):305–309.
41. Stevenson HL, et al. (2003) Detection of novel Bartonella strains and Yersinia pestis in
prairie dogs and their fleas (Siphonaptera: Ceratophyllidae and Pulicidae) using
multiplex polymerase chain reaction. J Med Entomol 40(3):329–337.
42. Billeter SA, Levy MG, Chomel BB, Breitschwerdt EB (2008) Vector transmission of
Bartonella species with emphasis on the potential for tick transmission. Med Vet
Entomol 22(1):1–15.
43. Young TP, Okello B, Kinyua D, Palmer TM (1997) KLEE: A long-term multispecies
herbivore exclusion experiment in Laikipia, Kenya. African Journal of Range and
Forage Science 14(3):94–102.
44. Randolph SE, Dobson AD (2012) Pangloss revisited: A critique of the dilution effect
and the biodiversity-buffers-disease paradigm. Parasitology 139(7):847–863.
45. Venesky MD, Liu X, Sauer EL, Rohr JR (2013) Linking manipulative experiments to
field data to test the dilution effect. J Anim Ecol, 10.1111/1365-2656.1259.
46. McCauley DJ, Keesing F, Young T, Dittmar K (2008) Effects of the removal of large
herbivores on fleas of small mammals. J Vector Ecol 33(2):263–268.
47. McCauley DJ, Keesing F, Young TP, Allan BF, Pringle RM (2006) Indirect effects of
large herbivores on snakes in an African savanna. Ecology 87(10):2657–2663.
48. Pretorius AM, Beati L, Birtles RJ (2004) Diversity of Bartonellae associated with small
mammals inhabiting Free State province, South Africa. Int J Syst Evol Microbiol 54:
1959–1967.
49. Inoue K, et al. (2009) Exotic small mammals as potential reservoirs of zoonotic Bar-
tonella spp. Emerg Infect Dis 15(4):526–532.
50. Mitchell CE, Tilman D, Groth JV (2002) Effects of grassland plant species diversity,
abundance, and composition on foliar fungal disease. Ecology 83(6):1713–1726.
51. Young HS, et al. (2013) Effects of mammalian herbivore declines on plant commu-
nities: Observations and experiments in an African savanna. J Ecol 101(4):1030–1041.
52. Clay CA, Lehmer EM, Jeor SS, Dearing MD (2009) Sin nombre virus and rodent species
diversity: A test of the dilution and amplification hypotheses. PLoS ONE 4(7):e6467.
53. Ezenwa VO, Godsey MS, King RJ, Guptill SC (2006) Avian diversity and West Nile virus:
Testing associations between biodiversity and infectious disease risk. Proc Biol Sci
273(1582):109–117.
54. Weber N, et al. (2013) Badger social networks correlate with tuberculosis infection.
Curr Biol 23(20):R915–R916.
55. Bai Y, Kosoy MY, Ray C, Brinkerhoff RJ, Collinge SK (2008) Temporal and spatial
patterns of Bartonella infection in black-tailed prairie dogs (Cynomys ludovicianus).
Microb Ecol 56(2):373–382.
56. Telfer S, et al. (2007) Contrasting dynamics of Bartonella spp. in cyclic field vole
populations: The impact of vector and host dynamics. Parasitology 134(Pt 3):413–425.
57. Telfer S, et al. (2007) Ecological differences and coexistence in a guild of micro-
parasites: Bartonella in wild rodents. Ecology 88(7):1841–1849.
58. Bitam I, Dittmar K, Parola P, Whiting MF, Raoult D (2010) Fleas and flea-borne dis-
eases. Int J Infect Dis 14(8):e667–e676.
59. Wall R, Shearer D (1997) Veterinary Entomology: Arthropod Ectoparasites of Veteri-
nary Importance (Chapman and Hall, London).
60. Keesing F, Allan BF, Young TP, Ostfeld RS (2013) Effects of wildlife and cattle on tick
abundance in central Kenya. Ecol Appl 23(6):1410–1418.
61. Perkins SE, Cattadori IM, Tagliapietra V, Rizzoli AP, Hudson PJ (2006) Localized deer
absence leads to tick amplification. Ecology 87(8):1981–1986.
62. Salkeld DJ, Lane RS (2010) Community ecology and disease risk: Lizards, squirrels, and
the Lyme disease spirochete in California, USA. Ecology 91(1):293–298.
63. Caro T (2001) Species richness and abundance of small mammals inside and outside an
African national park. Biol Conserv 98(3):251–257.
64. Goheen JR, et al. (2013) Piecewise disassembly of a large-herbivore community across
a rainfall gradient: The UHURU experiment. PLoS ONE 8(2):e55192.
65. Luis AD, et al. (2013) A comparison of bats and rodents as reservoirs of zoonotic vi-
ruses: Are bats special? Proc Biol Sci 280(1756):20122753.
66. Kedem H, et al. (2013) Multiple effects of host species diversity on co-existing host-
specific and host-opportunistic microbes. Ecology, 10.1890/13-0678.1.
67. Billeter SA, Gundi VA, Rood MP, Kosoy MY (2011) Molecular detection and identifi-
cation of Bartonella species in Xenopsylla cheopis fleas (Siphonaptera: Pulicidae)
collected from Rattus norvegicus rats in Los Angeles, California. Appl Environ Mi-
crobiol 77(21):7850–7852.
68. Kosoy MY, et al. (1997) Distribution, diversity, and host specificity of Bartonella in
rodents from the Southeastern United States. Am J Trop Med Hyg 57(5):578–588.
Young et al. PNAS | May 13, 2014 | vol. 111 | no. 19 | 7041
EC
OL
OG
Y
Supporting Information
Young et al. 10.1073/pnas.1404958111
SI Text
The applications of this study for human health depend on the
assumptions that Saccostomus mearnsi individuals that interact
closely with humans will also carry infected fleas capable of
biting humans. Additional trapping conducted in human-domi-
nated landscapes (homes, agricultural fields, and pastoral land-
scapes) in nearby areas of Laikipia district during the time period
of this study supports this conclusion. Of 25 S. mearnsi sampled
in these anthropogenic landscapes, the mean flea density was
12.0 ± 10.4 fleas, which is well above the levels observed in the
Kenya Long-Term Exclosure Experiment (KLEE). As in the
KLEE sites, these fleas were primarily composed of Xenopsylla
cheopis and Xenopsylla sarodes (94% of fleas); X. cheopis in
particular are known to be generalists that bite humans and
transmit pathogens. The fleas of S. mearnsi in human-dominated
areas were also dominated by Xenopsylla spp. Of the 21
S. mearnsi from anthropogenic landscapes that were screened for
Bartonella spp., 16 (76%) were positive for Bartonella spp., pre-
dominantly of Bartonella elizabethae, a suspected human patho-
gen. Of the 97 fleas from S. mearnsi from anthropogenic
landscapes that were screened for Bartonella, 56% were positive
for at least one strain of Bartonella. For the 18 fleas from
S. mearnsi found specifically in human dwellings, 79% were pos-
itive for at least one strain of Bartonella. Although detailed sur-
veillance of Bartonella prevalence in human-dominated landscapes
will be needed to assess direct human health outcomes, all initial
surveillance suggests that the relevant host–vector pathogen re-
lationships exist to make human transmission likely.
Young et al. www.pnas.org/cgi/content/short/1404958111 1 of 3
Fig. S1. Examining all rodents and their fleas instead of only S. mearnsi does not qualitatively change the interpretations of effects of defaunation on rodent
or flea abundance. (A) Abundance of all rodents was approximately twice as high in large wildlife exclosures compared with open plots, which is a significant
difference. (B) Density of fleas per rodent did not vary among treatments. (C) Thus, the density of fleas per hectare is also significantly higher (approximately
doubled) in plots where large wildlife has been removed. Error lines represent 1 SE, based on three replicate blocks.
Fig. S2. Flea communities of S. mearnsi in both large wildlife exclosures and control sites (pooled) are dominated by two Xenopsylla species. Fleas in the
Xenopsylla genus are competent vectors of Bartonella as well as other pathogens of relevance to human health (e.g., Yersinia pestis).
Young et al. www.pnas.org/cgi/content/short/1404958111 2 of 3
Fig. S3. Prevalence of infection in S. mearnsi was not significantly different among treatments across seasons; however, there were significant differences
in prevalence over time (as also observed in prevalence among fleas). The interaction effect was not significant. Error lines represent 1 SE based on three
replicate blocks.
Table S1. Average (±SD) number of hosts and fleas/host per site, pooled across all five
sampling periods and three experimental blocks
Individuals per hectare Fleas/host
Species Exclosure Open Exclosure Open
Acomys kempi 0.3 ± 0.6 1.3 ± 2.3 0 0
Aethomys kaiseri 0.3 ± 0.6 0 1.0 0
Arvicanthis niloticus 4.7 ± 3.5 0 1.9 ± 1.9 0
Dendromus melanotis 1.0 ± 1.7 1.3 ± 1.5 0 0
Mastomys natalensis 1.3 ± 0.6 0 0.3 ± 0.6 0
Mus minutoides 28.3 ± 16.3 17.7 ± 8.1 0 0
Mus cf. acholi 1.7 ± 2.9 1.0 ± 1.0 0 0
Mus sorella 3.3 ± 0.6 4.3 ± 3.8 0 0
Mus tenellus 0.3 ± 0.6 1.7 ± 0.6 0 0
Saccostomus mearnsi 143 ± 15.5 64.3 ± 17.4 2.2 ± 0.3 2.8 ± 0.9
Zelotomys hildegardeae 1.0 ± 0 0.33 ± 0.58 8 0
Table S2. Matrix of the presence of flea species (columns) found on each rodent host species (rows)
Species
Ctenophthalmus
cabirus
Parapulex
echinatus
Xenopsylla
brasiliensis
Xenopsylla
cheopis
Xenopsylla
sarodes
Dinopsyllus
lypusus
Acomys kempi X
Aethomys kaiseri X
Arvicanthis niloticus X X
Mastomys natalensis X
Saccostomus mearnsi X X X X
Zelotomys hildegardeae X
None of the species of Mus (Mus minutoides, Mus cf. acholi, Mus sorella, and Mus tenellus ) or Dendromus melanotis had fleas; therefore, they are not listed.
Young et al. www.pnas.org/cgi/content/short/1404958111 3 of 3