Multiple stressors and the cause of amphibian abnormalities

Abstract

The repeated occurrence of abnormal amphibians in nature points to ecological imbalance, yet identifying causes of these abnormalities has proved complex. Multiple studies have linked amphibian abnormalities to chemically contaminated areas, but inference about causal mechanisms is lacking. Here we use a high incidence of abnormalities in Alaskan wood frogs to strengthen inference about the mechanism for these abnormalities. We suggest that limb abnormalities are caused by a combination of multiple stressors. Specifically, toxicants lead to increased predation, resulting in more injuries to developing limbs and subsequent developmental malformations. We evaluated a variety of putative causes of frog abnormalities at 21 wetlands on the Kenai National Wildlife Refuge, south-central Alaska, USA, between 2004 and 2006. Variables investigated were organic and inorganic contaminants, parasite infection, abundance of predatory invertebrates, UVB, and temperature. Logistic regression and model comparison using the Akaike information criterion (AIC) identified dragonflies and both organic and inorganic contaminants as predictors of the frequency of skeletal abnormalities. We suggest that both predators and contaminants alter ecosystem dynamics to increase the frequency of amphibian abnormalities in contaminated habitat. Future experiments should test the causal mechanisms by which toxicants and predators may interact to cause amphibian limb abnormalities.

Full text

Ecological Monographs, 80(3), 2010, pp. 423–440
Ó2010 by the Ecological Society of America
Multiple stressors and the cause of amphibian abnormalities
MARI K. REEVES,
1,2,6
PETER JENSEN,
3
CHRISTINE L. DOLPH,
4
MARCEL HOLYOAK,
2
AND KIMBERLY A. TRUST
5
1
U.S. Fish and Wildlife Service, Anchorage Fisheries and Ecological Services Office,
605 West 4th Avenue, Room G-61, Anchorage, Alaska 99501 USA
2
Ecology Graduate Group and Department of Environmental Science and Policy, University of California,
1 Shields Avenue, Davis, California 95616 USA
3
Integral Consulting Incorporated, 200 Harry S Truman Parkway, Suite 330, Annapolis, Maryland 21401 USA
4
Water Resources Science Program, University of Minnesota, 200 Hodson Hall,
1980 Folwell Avenue, Saint Paul, Minnesota 55108 USA
5
U.S. Fish and Wildlife Service, 1011 East Tudor Road, Anchorage, Alaska 99503 USA
Abstract. The repeated occurrence of abnormal amphibians in nature points to ecological
imbalance, yet identifying causes of these abnormalities has proved complex. Multiple studies
have linked amphibian abnormalities to chemically contaminated areas, but inference about
causal mechanisms is lacking. Here we use a high incidence of abnormalities in Alaskan wood
frogs to strengthen inference about the mechanism for these abnormalities. We suggest that
limb abnormalities are caused by a combination of multiple stressors. Specifically, toxicants
lead to increased predation, resulting in more injuries to developing limbs and subsequent
developmental malformations. We evaluated a variety of putative causes of frog abnormalities
at 21 wetlands on the Kenai National Wildlife Refuge, south-central Alaska, USA, between
2004 and 2006. Variables investigated were organic and inorganic contaminants, parasite
infection, abundance of predatory invertebrates, UVB, and temperature. Logistic regression
and model comparison using the Akaike information criterion (AIC) identified dragonflies
and both organic and inorganic contaminants as predictors of the frequency of skeletal
abnormalities. We suggest that both predators and contaminants alter ecosystem dynamics to
increase the frequency of amphibian abnormalities in contaminated habitat. Future
experiments should test the causal mechanisms by which toxicants and predators may
interact to cause amphibian limb abnormalities.
Key words: abnormality; Alaska, USA; amphibian; invertebrate predators; Kenai National Wildlife
Refuge; malformation; Lithobates sylvaticus; metals; Rana sylvatica; wood frog.
INTRODUCTION
The incidence of limbless amphibians is an alarming
global phenomenon without adequate explanation.
Recent research highlights the role of predators in
causing such missing and shrunken limb abnormalities
(Ballenge
´e and Sessions 2009, Bowerman et al. 2010),
yet predators occur in most amphibian breeding
habitats, and abnormal amphibians are not as consis-
tently detected. Normal frequencies of abnormal am-
phibians have been estimated at 0–2%(Ouellet 2000)
and 0–5%(Johnson et al. 2001). Studies examining large
numbers of frogs across North America support a
baseline frequency of 5%(Johnson et al. 2010). Yet in
some areas, abnormalities occur more frequently, and
such studies have found correlations with chemical
contamination (Ouellet et al. 1997, Hopkins et al. 2000,
Linzey et al. 2003, Taylor et al. 2005, Bacon et al. 2006,
Gurushankara et al. 2007, Reeves et al. 2008). Unfor-
tunately causal mechanisms are lacking. However, a
clear understanding of the mechanisms behind amphib-
ian abnormalities could help managers limit abnormal-
ity occurrence (e.g., Bohl 1997).
Controlled experiments have linked abnormal am-
phibians with parasite infection (Johnson and Suther-
land 2003), chemical contaminants (reviewed in Johnson
et al. 2010), UVB radiation (Ankley et al. 2004), and
amputation injury (Fry 1966, Ballenge
´e and Sessions
2009, Bowerman et al. 2010). The trematode parasite
Ribeiroia ondatrae is the best established cause of
amphibian limb abnormalities. When this parasite
infects the developing tadpole limb bud, it can cause a
wide spectrum of limb malformations, including extra
limbs, missing limbs, bony triangles, and skin webbings
(Johnson and Sutherland 2003). In turn, parasite
abundance has been experimentally linked to poor
water quality from nutrient enrichment (Johnson et al.
2007). Also in controlled experiments, skeletal abnor-
malities have been caused in different taxa directly by
both organic and inorganic chemicals (reviewed in
Johnson et al. 2010). In laboratory and mesocosm
studies, UVB radiation has caused reductions or
deletions of both hind limbs in a symmetrical fashion
that is rare in nature (Ankley et al. 2004). Finally, two
recent field studies suggest that odonate predators cause
Manuscript received 21 May 2009; revised 5 November 2009;
accepted 11 December 2009; final version received 25 January
2010. Corresponding Editor: D. K. Skelly.
6
E-mail: mari_reeves@fws.gov
423

skeletal abnormalities by injuring the developing limb
bud and disrupting normal patterning and growth
(Ballenge
´e and Sessions 2009, Bowerman et al. 2010).
In one of these studies, limb abnormalities were
eliminated by rearing tadpoles in predator-exclusion
cages (Bowerman et al. 2010), and both groups found
that experimental amputation of limbs early in devel-
opment produced shrunken limb malformations at
metamorphosis.
In the Kenai National Wildlife Refuge (KNWR) in
south-central Alaska, USA, abnormality frequency
among Alaskan wood frogs (Rana sylvatica) was higher
in some areas than the expected background prevalence
of 0–5%(Ouellet 2000, Johnson et al. 2001, Reeves et al.
2008). Abnormality frequency was 7.9%for the KNWR
on average, and in some wetlands abnormality preva-
lence exceeded 20%. Moreover, although the parasite R.
ondatrae was not found infecting wood frogs in the
KNWR, abnormalities were more frequent at road-
accessible sites, suggesting a possible interaction be-
tween these malformed frogs and human disturbance
(Reeves et al. 2008).
The KNWR contains a broad array of stressors
known to cause frog abnormalities. Oil drilling and
other road-associated human activities have facilitated
land and water contamination by petroleum products
and polychlorinated biphenyls (Parson 2001). Histori-
cally, persistent organochlorine pesticides and herbicides
have also been used (Parson 2001). Toxic metals occur
naturally at high levels in some areas (Crock et al. 1992),
and metals and hydrocarbons may run off of roads into
adjacent aquatic habitat (Wheeler et al. 2005 ). Global
climate change is evident in south-central Alaska, where
recent warmer and drier conditions have been associated
with shrinking and drying of the KNWR wetlands that
comprise frog breeding habitat (Klein et al. 2005). The
ecological consequences of such rapid, directional
climate change are unknown.
Both skeletal and eye abnormalities have been
documented on the KNWR (Reeves et al. 2008), and
each of these abnormality types has been associated with
different stressors in previous studies. Skeletal abnor-
malities have been associated with parasites (Johnson
and Sutherland 2003), chemicals (reviewed in Johnson et
al. 2010), UVB (Ankley et al. 2004), and predators
(Ballenge
´e and Sessions 2009, Bowerman et al. 2010).
Dominant eye abnormalities in KNWR (unpigmented
irises) have been correlated with recessive genetic
mutations (Nishioka 1977), urban contamination (Ver-
shinin 2002), and early-season temperature extremes
(Vershinin 2002). These abnormalities in Alaskan frogs
were also associated with different stressors by Reeves et
al. (2008): Skeletal abnormalities were correlated with
proximity to road, frog size, and frog developmental
stage, and eye abnormalities were only associated with
year sampled.
The goal of this study was to combine field
observations with manipulative laboratory and field
experiments to provide a greater level of inference
about the causes of amphibian abnormalities on the
KNWR. Specifically, we hypothesized that higher
frequencies of amphibian abnormalities would be
associated with each of the following individual
stressors in KNWR wetlands: (1) low UVB attenua-
tion; (2) higher numbers of predatory invertebrates;
and (3) higher concentrations of inorganic and organic
contaminants. Although water temperature has not
been shown to directly cause amphibian abnormalities,
higher temperatures might cause differential success of
amphibian predators (e.g., Anderson et al. 2001). We
therefore also hypothesized that higher temperatures
would be associated with higher frequencies of am-
phibian abnormalities. Finally, we hypothesized that
these stressors would act in combination (i.e., addi-
tively or synergistically) to increase the frequency of
abnormalities.
We tested these hypotheses by quantifying each of
these stressors at 21 study sites in KNWR during 2004–
2006 and conducting logistic regression and model
comparison using the Akaike information criterion
(AIC) to identify whether any of these stressors
predicted the frequency of amphibian abnormalities
at these sites. Because they may have different causes,
we analyzed skeletal abnormalities and eye abnormal-
ities separately. We also tested whether the presence
andtypeofroadadjacenttoawetlandwouldaffectthe
contaminants in that wetland. Based on our analysis of
field data, we designed three follow-up experiments to
test the role of predation injury and chemical contam-
inants in producing limb abnormalities in Alaskan
wood frogs. First, we used predator-exclusion cages in
high-malformation sites to test the effect of predator
removal on limb and eye abnormalities. Second, we
mimicked predation injury by limb amputation to test
whether amputations caused missing or shrunken limb
malformations in late-stage tadpoles and metamorphs.
Finally, we tested whether exposure to site sediment
and water in the absence of other stressors caused
abnormalities directly or alternatively caused sublethal
toxic effects that increased vulnerability to other
stressors, such as predators. We predicted that (1)
removal of predators would result in a decrease in limb
and eye abnormalities, (2) artificial limb amputation in
late-stage tadpoles would result in the subsequent
development of limb malformations, and (3) amphib-
ians raised in sediment and water from KNWR
wetlands would exhibit more abnormalities than
amphibians raised in control conditions. We show that
while both predation and contaminants can promote
limb abnormalities, and can do so in the same study
system, their effects in combination appear to be
subadditive, a result perhaps explained by greater
(unmeasured) mortality in contaminated wetlands with
abundant predators or by interference with predator
detection.
MARI K. REEVES ET AL.424 Ecological Monographs
Vol. 80, No. 3

METHODS
Study area and selection of study sites
The KNWR comprises 797 200 ha in south-central
Alaska, including four designated wilderness areas:
Mystery Hills, Swanson River, Skilak Lake, and
Tustemena Lake (Wilderness Act of 1964: 16 U.S.C.
1131–1136). The refuge also contains 345 km of roads,
most of which were developed to support two oil and gas
fields that began production on the refuge in the 1950s.
In this study, we sought to evaluate the relationship
between stressors and amphibian abnormalities across a
gradient of human disturbance within the refuge. We
therefore classified areas of the refuge as either
developed (adjacent to roads) or as remote (.1km
from a road and within the four designated wilderness
areas). The effects of road contamination on the
surrounding environment are thought to be negligible
beyond a 1-km distance (Trombulak and Frissell 2000);
we therefore considered sites .1kmfromroads
essentially undisturbed by local human activities.
We used 1:24 000 United States Geological Survey
topographic quadrangle maps and digital aerial photo-
graphs to identify potential amphibian breeding sites in
both developed and remote study areas. Sites were
identified as potential amphibian breeding areas if they
appeared wet or marshy in maps or photographs and if
they were 1 km in length. We identified all potential
breeding sites in developed areas, that is, all potential
sites located adjacent to gravel roads, small paved roads,
and paved highways that transected or bordered the
refuge. In wilderness areas of the refuge, we placed more
stringent geographic limits on our search for breeding
sites because air travel was prohibited. We therefore
selected study sites that could be accessed within 1–2
days of travel via foot, canoe, or motorized boat. In the
Mystery Hills wilderness, we limited our search for
potential wetlands to within 1 km of the Fuller Lakes
Trail, a footpath maintained by the KNWR. In the
Swanson River wilderness, we limited our search to
within 1 km of the Swanson River Canoe Route, a series
of lakes connected by portages maintained by the
KNWR. In the Skilak Lake wilderness, we limited our
search to within 1 km of Skilak Lake itself, which is 15
km in length and traversable by motorized boat.
Tustemena Lake is also traversable by boat. However,
Tustemena Lake is .30 km long, and boating condi-
tions can become hazardous quickly in bad weather. We
therefore limited our studies within this area to within 1
km of the 15 km of shoreline nearest to the Kasilof River
boat launch.
We identified a total of 138 potential new study sites
using these methods: 102 sites in developed areas and 36
sites across all four wilderness areas. We visited each site
in May or early June of 2004 or 2005 (depending on the
site) and evaluated whether safe site access was possible
and whether the site contained water. Sites identified in
maps or photographs that did not contain standing
water were excluded from further analysis. If a breeding
area was very shallow and it appeared dry on aerial
photographs taken in July, we also excluded it from
further analysis. One site visit and one aerial photograph
are clearly not sufficient to fully characterize the active
hydroperiod of a wetland, and this strategy undoubtedly
biased our site selection towards deeper and larger
wetlands. However, this approach enabled us to screen
out sites that were susceptible to drying, which was
necessary given our need to limit our pool of possible
study sites to those wetlands where collection of
sufficient metamorph data was likely. A total of 53 sites
were excluded because of insufficient water (44 in
developed areas, nine in remote areas). An additional
two sites in remote areas were excluded because they
could not be accessed due to extremely dense vegetation
around the wetland perimeter.
If .30 cm of water was present or if 0–30 cm of water
was present but the site appeared wet in aerial
photographs taken during July, we visually searched
and dip-netted the entire perimeter of the wetland for
evidence of wood frog egg masses or tadpoles. This
method led us to identify wood frogs at 38 developed
sites and 21 wilderness sites.
We revisited all 21 wilderness sites in late June/July of
2004 or 2005 and found an insufficient number of
metamorphs (i.e., 50) to evaluate abnormality preva-
lence at 12 of them. We were, however, able to obtain an
adequate number of metamorphs (50) to evaluate
abnormality prevalence at nine sites: four in Mystery
Hills, two in Swanson River, two in Skilak Lake, and
one in Tustemena Lake. We selected all nine of these
sites for inclusion in this study (Table 1).
We also obtained adequate numbers of metamorphs
from 19 of the 38 developed sites in late June/July of
2004 or 2005. Metamorph collections were not obtained
from the remaining 19 developed sites either because
metamorphs were found to be insufficiently abundant or
because sampling crews could not revisit the sites before
metamorphosis occurred. We then selected 12 of these
19 sites for inclusion in this study to represent (1)
different types of developed areas (oil and gas field
roads, other gravel roads, paved highways), (2) variabil-
ity in wetland depth, and (3) a number of different
geographic areas within the refuge (Table 1; Appendix
A: Fig. A1).
Study species
Rana sylvatica (Hillis 2007) is the only amphibian
common in most of Alaska and the only amphibian in
KNWR. Wood frogs breed explosively just after
snowmelt, lay eggs in late April or early May, and
metamorphose in late June through early August
depending on temperature, timing of snowmelt, and
wetland hydroperiod (Herreid and Kinney 1967).
Synchronous breeding and development cause larvae
to metamorphose within a 5–7 day window at most
breeding ponds (Herreid and Kinney 1967; M. K.
August 2010 425MULTIPLE STRESSORS AND ABNORMAL FROGS

Reeves, personal observation). Synchronous breeding
and one cohort per season aid malformation studies by
reducing variation in age at sampling. As the only
tadpole species in KNWR, wood frogs do not encounter
interspecific predation by other tadpoles, nor are they
known to cannibalize live conspecifics (M. K. Reeves,
personal observation; R. Relyea, unpublished data).
Abnormality assessment and classification
Amphibian abnormality assessment and classification
methods are described more fully in Reeves et al. (2008).
Briefly, 50–100 metamorphic frogs, stages 42–46 (Gos-
ner 1960), were assessed for abnormalities at each site.
This method controlled tadpole stage by limiting
sampled animals to recent metamorphs between fore-
limb emergence and complete tail resorption. Snout-to-
vent length (SVL) and tail length were measured, and
developmental stage was recorded. After field examina-
tion, abnormal frogs were euthanized with MS-222,
photographed, preserved in 70%ethanol, and sent to the
U.S. Geological Survey, National Wildlife Health
Center, or Ball State University for radiographs to aid
abnormality classification. All abnormalities were clas-
sified by a single researcher using field notes, photo-
graphs, and radiographs. Normal frogs were released at
the capture site (with the exception of frogs collected for
parasitology). We followed guidelines for the use of live
amphibians outlined by the American Society of
Ichthyologists and Herpetologists (ASIH 2004) when
handling and sampling specimens. All animal collections
were authorized under a sampling permit approved by
the State of Alaska, Department of Fish and Game.
Between site visits, equipment was disinfected with 5%
bleach solution.
Parasitology
We selected both normal and abnormal frogs from
each site for parasitological examination. We collected
2–37 frogs (as available) from each site and sent them
live to the University of Wisconsin, Lacrosse (UWL).
Not all frogs survived each trip, which sometimes led to
sample loss. All animals were kept in a walk-in cooler
(48C) until necropsied. Metamorphs were euthanized
with 0.1%MS-222 solution (Argent Chemical Labora-
tories, Redmond, Washington, USA). With the aid of a
stereo-dissecting microscope, the number, location, and
life stage of all parasites were recorded. Representative
metacercariae were manually excysted, prepared as wet
mounts, and photographed using a combination of
bright-field, phase contrast, and Nomarski differential
interference microscopy.
Measurement of explanatory environmental variables
Amphibian abnormality data were collected from
2000 to 2006, but contaminant samples were collected in
2004 and 2005, predator sweeps were done in 2005 and
2006, and UVB and temperature were only quantified
adequately for statistical analysis in 2006. Our statistical
analysis ignores interannual variability in all the
predictors, except in frog size and developmental stage,
because we did not have data for each predictor variable
across all sampling years. These predictors were
therefore aggregated (see Appendix A) and a single
value for each predictor was assigned to each site. Each
TABLE 1. Breeding site information for this study of abnormalities in wood frogs, Rana sylvatica, at 21 wetlands on the Kenai
National Wildlife Refuge, south-central Alaska, USA.
Site Years sampled
Total no.
frogs sampled
Overall skeletal
and eye abnormality
prevalence (%)
Skeletal abnormalities (%)à
Mean Median Minimum Maximum
1 2000–2006 481 11.2 9.8 9.4 3.7 15.9
2 2000, 2004–2005 122 4.9 0.0 0.0 0.0 0.0
3 2000–2006 270 7.0 6.7 6.5 3.9 10.0
8 2000–2006 436 10.8 9.9 10.0 3.3 15.2
12 2000–2002, 2004–2006 328 6.4 4.1 3.6 1.7 6.3
14 2000–2006 516 3.1 2.6 1.5 0.0 7.3
21 NS§ 0     
31 2004–2006 253 6.3 1.7 1.0 0.0 4.0
46 2004–2006 202 9.4 8.5 9.6 4.0 12.0
47 2004–2006 243 5.3 2.0 1.9 1.0 3.2
51 2004 50 10.0 4.0 4.0 4.0 4.0
54 2004–2005 59 13.6 10.7 10.7 10.7 10.7
55 2004–2005 104 4.8 1.9 1.9 0.0 3.8
56 2004–2005 100 4.0 3.0 3.0 2.0 4.0
60 2004–2006 203 8.9 4.6 5.7 2.0 6.0
62 2005 50 12.0 12.0 12.0 12.0 12.0
90 2004–2006 192 13.5 9.7 9.5 3.7 16.0
95 2004–2005 103 2.9 1.9 1.9 0.0 3.8
97 2004–2006 134 2.2 0.8 0.8 0.0 1.6
111 2004–2005 109 11.0 5.5 5.5 1.8 9.3
141 2004 54 1.9 0.0 0.0 0.0 0.0
Number of abnormal frogs at site divided by total number of frogs examined at each site over all years.
àStatistics compiled only from sampling events at which 50 or more frogs were examined.
§ Disease caused mass tadpole mortality each year at this site, and metamorphs were never sampled (NS).
MARI K. REEVES ET AL.426 Ecological Monographs
Vol. 80, No. 3

predictor variable must be considered a proxy for the
true values and variables, because it is probable that
each predictor variable changes with time. Nevertheless,
it would have been surprising if such temporal variation
created false positive results.
Sediment and water sampling for organic
and inorganic contaminants
Sediment and water samples were collected during
amphibian abnormality assessments in 2004 and 2005.
Water and sediment samples were taken during meta-
morph collections from eight ponds in 2004 and 13 in
2005 and tested for organic and inorganic contaminants
including total metals, polychlorinated biphenyls (PCBs)
and other organochlorines (OCs), polycyclic aromatic
hydrocarbons (PAHs), and dioxins and furans (Appen-
dix A). Contaminant sampling enumerated 89 organic
and 28 inorganic analytes in sediment and 22 inorganic
analytes in water for each site (Appendix B: Tables B1–
B6). To reduce these data for statistical analysis, we first
compared contaminants detected at study sites to
established toxicity thresholds (Buchman 2008). Three
sediment benchmarks were used for comparison: the
lowest effects level (LEL), a level of sediment contam-
ination that can be tolerated by most benthic organisms;
the threshold effects level (TEL), the concentration
below which significant adverse effects on biota are not
expected; and the probable effects level (PEL), the
concentration at which significant adverse effects
become likely (Buchman 2008). The threshold concen-
tration used for water was the criterion continuous
concentration (CCC), the chronic limit for the priority
pollutant in fresh water (Buchman 2008). To be retained
for statistical analysis, we required a contaminant to be
over at least one toxic threshold at one or more sites and
detected at 20%of study sites or more (see Appendix B:
Tables B1 and B2 for comparisons to thresholds).
Contaminants exceeding thresholds were then grouped
according to whether they were organic (pesticides,
PCBs, dioxins/furans, and PAHs) or inorganic (metals
and other elements) and then reduced with principal
components analysis (PCA). The PCA (Appendix A)
produced four vectors (two representing organic and
two inorganic) that were then used as predictor variables
in statistical analysis of abnormality data from the field
sites.
Temperature
We measured temperature every 30 min with data
loggers at all sites (Hobo, Tidbit Loggers, Onset Data
Corporation, Pocasset, Massachusetts, USA). Loggers
were floated at a depth of 5 cm adjacent to egg masses
and deployed before or during frog breeding at most
sites (late April to early May). They were retrieved when
frogs were assessed for abnormalities or when the site
dried. Temperature measurements were made at all sites
in 2005 and 2006 and a subset of sites in 2004. For
statistical analyses in this study, we used mean
temperature from 2006 only, due to concerns about
unsystematic deployment of temperature loggers in 2004
and 2005.
UVB
All UVB measurements were taken with a broadband
UV meter (Macam Photometrics, Livingston, UK; peak
spectral response ¼311 nm and bandwidth range ¼292–
330 nm) in 2006. The instrument was factory calibrated
using standards traceable to the British Standard
TABLE 1. Extended.
Eye abnormalities (%)à
Latitude (N) Longitude (W)
Distance
to road (km)
Road
typeMean Median Minimum Maximum
2.2 3.2 0.0 4.3 6084400 17015085200 19 00.0 gravel
4.2 4.2 0.0 8.3 6084300 5015085300 400.0 gravel
0.0 0.0 0.0 0.0 6084300 33015085300 23 00.0 gravel
1.0 0.7 0.0 3.3 6083700 35015084800 27 00.0 gravel
3.4 1.7 0.0 9.4 6084200 51015084800 55 00.0 gravel
0.4 0.0 0.0 1.8 6084500 5015083000 1400.1 gravel
    6083900 2301518100 4800.0 gravel
6.0 5.9 2.0 10.0 6081200 801508200 3309.2 gravel
1.3 1.9 0.0 2.0 6084600 36015083200 21 02.6 gravel
3.7 3.8 1.0 6.3 6084600 37015083200 50 02.9 gravel
6.0 6.0 6.0 6.0 6083000 4401508300 3003.0 paved
1.8 1.8 1.8 1.8 6083000 5801508300 4003.4 paved
2.9 2.9 1.9 3.9 6083100 3701508400 1004.7 paved
1.0 1.0 0.0 2.0 6083100 2501508300 5104.3 paved
5.0 6.0 1.9 7.0 6082600 11015083000 28 04.4 gravel
0.0 0.0 0.0 0.0 6082600 4015083100 1004.9 gravel
3.2 3.2 2.7 3.7 6082700 4301518500 3700.0 paved
1.0 1.0 0.0 1.9 6083100 52015082700 52 00.1 paved
0.9 0.9 0.0 1.8 6083100 54015082500 40 00.0 paved
5.5 5.5 5.5 5.6 6082800 59015082700 21 00.0 gravel
1.9 1.9 1.9 1.9 6083600 26015081700 46 00.0 gravel
August 2010 427MULTIPLE STRESSORS AND ABNORMAL FROGS

Institute and field calibrated daily. We calculated
percentage of UVB attenuation (i.e., the percentage of
surface UVB that reaches a 10-cm depth) for each
wetland from UVB measurements at 0 and 10 cm depth.
The UVB measurements were not taken at five sites for
logistical reasons or because the sites dried before
measurement was possible. For each of the remaining
sites, multiple measurements were recorded on different
dates in variable weather conditions; the number of
measurements per site ranged from 0 to 13 (mean ¼4.1).
For statistical analyses, we used the median UVB
percentage of attenuation at 10-cm depth.
Invertebrate predator assessments
We identified potential invertebrate predators and
quantified their densities during early and late summer
at each site in 2005 and 2006. Invertebrates were
collected by sweeping a 0.3 30.3 m D-frame net (350-
mm mesh net) at a depth of ,1 m on a 10 m long
transect through vegetative perimeters of the ponds,
where tadpoles were captured. Three 10-m transects
were swept, to sample ;2120 L of water. Samples were
removed from the net, separated from vegetation and
debris in the field, placed in pre-labeled Whirl-pak bags
(Nasco, Fort Atkinson, Wisconsin, USA), and preserved
with 70%ethanol. All individuals were identified to the
lowest practical taxonomic level, either genus or species,
under 103magnification using appropriate keys (Clif-
ford 1991, Merritt et al. 1995, Needham et al. 2000),
enumerated, and preserved in labeled glass vials with
70%ethanol.
We calculated the mean abundance of the following
predatory taxa in both early and late seasons using both
years of data: all dragonflies, all damselflies, three beetle
taxa (Dytiscus spp., Rhantus spp., and Ilybius spp.), and
the total abundance of predaceous invertebrates. Al-
though we quantified predatory fishing spiders and
leeches, their abundances were too low to retain them
for statistical analyses and still meet assumptions of
linearity. We only retained three species of beetles for
statistical analysis because it is unlikely that the other
beetle species present would be capable of injuring or
removing the limb bud of a developing wood frog
tadpole (D. A. Yee, personal communication). We
retained all damselflies for statistical analysis, because
we believed they could plausibly injure the developing
wood frog tadpole limb (P. J. Crowley, personal
communication). Raw abundance data are enumerated
in Appendix B.
Statistical analysis and hypothesis evaluation
for field data
We used the Akaike information criterion (AIC) to
determine which measured environmental variables best
predicted amphibian abnormalities in the KNWR
(Burnham and Anderson 2002). We used generalized
linear models (GLIM; PROC GENMOD in SAS
version 9.1.3, SAS Institute, Cary, North Carolina,
USA) fitted with a binomial error distribution to obtain
maximum likelihood and parameter estimates. This
method is the equivalent of a logistic regression that
also allowed us to treat frogs at the same site in the same
year as repeated measures (which are potentially
correlated). As additional indicators of model fit, we
calculated quasi-R
2
and performed Hosmer-Lemeshow
goodness-of-fit tests (SAS version 9.1.3).
We built our models by first testing all predictors
representing univariate hypotheses separately. Specific
hypotheses tested during this step involved predators,
contaminants, temperature, and UVB as independent
causes of limb and eye abnormalities. In addition to the
variables detailed above, metamorph size and develop-
mental stage (Gosner 1960) were used as covariates in
every model for skeletal abnormalities because previous
work showed they were correlated with this abnormality
type (Reeves et al. 2008). Size and developmental stage
were treated like all other predictor variables in the eye
abnormality analysis. One site was excluded from
statistical analysis because it was sampled during a
forest fire and concentrations of some organic contam-
inants were approximately an order of magnitude higher
than other sites (see Appendix A for more information).
We retained variables from the top three univariate
models with the lowest AIC and highest quasi-R
2
and
then created two-factor models by testing each of these
three best variables in combination with every other
univariate predictor. We then (arbitrarily) retained the
top three models from this step based on lowest AIC and
highest quasi-R
2
and tested each of these with every
other univariate predictor. We kept adding variables to
the three best models from each step until we saw no
further improvement in AIC and quasi-R
2
. If there were
more than three models with likelihood values within
two points of the ‘‘best’’ (lowest AIC) model in that step,
then all models within two points of the best model were
taken to the next step in the analysis. Once we had
obtained the best-fitting model for each abnormality
type, we tested interaction terms by running each
possible interaction among predictor variables singly
and comparing models with AIC and quasi-R
2
for the
models without interactions. Interactions were only
retained if they improved model fit over the best-fitting
model without interactions (see Tables 2–5 for model
selection process and results).
Because we knew roads were associated with skeletal
abnormalities (Reeves et al. 2008), we tested whether
road type significantly influenced the composition of
inorganic contaminants in site sediment and water. We
used only inorganic contaminants in the multivariate
response because we detected few organic contaminants
that should be associated with road runoff (e.g., PAHs;
Wheeler et al. 2005). We performed an analysis of
similarity (ANOSIM; Primer version 6.1.6, Primer-E,
Lutton, Ivybridge, UK) on a dissimilarity matrix of the
Euclidian distance between sites, with road type (paved,
MARI K. REEVES ET AL.428 Ecological Monographs
Vol. 80, No. 3

gravel, wilderness; Table 1) as a predictor of inorganic
contaminants signature in site sediment and water.
Predator exclusion experiment
Predator exclusion cages (diameter 3depth ¼76.2 3
76.2 cm) were obtained from Team NuMark (Victoria,
Texas, USA). Fiberglass window screen was sewn with
fishing line to the sides and bottom of the cage to reduce
cage mesh size and prohibit entry by invertebrates; a
hole with a cinch cord was left in the cage tops so
tadpole development could be monitored by opening the
cage. Each cage was augmented with four fishing net
buoys to keep the cage top above the water surface.
Three cages were deployed at each of three ponds in
2005 and 2006. In 2005 ponds included two developed
sites and one remote site, and in 2006 all sites were road-
accessible because of time limitations. In the first year,
egg masses were collected from study sites, hatched
indoors in clean water (Alaska’s Best Water, Anchorage,
Alaska, USA; hereafter referred to as ‘‘spring water’’) in
site-specific aquaria, reared to Gosner stage 25 (begin-
ning of limb development), and deployed in cages. In
TABLE 2. Model results for skeletal abnormalities: model evaluation metrics.
Model Likelihood Parameters Trials AIC DAIC Quasi-R
2
(%)
Intercept only 780.4789 4 3744 1568.958 NA NA
Saturated model 737.0573 13 3744 1500.115 NA NA
Univariate models step 1
Larval dragonfly abundance 754.4664 4 3744 1516.933 0 60
Organic PCA 2 762.6345 4 3744 1533.269 16.336 41
Sum predators 763.1809 4 3744 1534.362 17.4288 40
Temperature 763.2784 4 3744 1534.557 17.6238 40
Organic PCA 1 764.5033 4 3744 1537.007 20.0736 37
UVB 764.8249 4 3744 1537.65 20.7168 36
Metal PCA 2 765.4382 4 3744 1538.876 21.9434 35
Larval damselfly abundance 765.5634 4 3744 1539.127 22.1938 34
Larval beetle abundance 765.9215 4 3744 1539.843 22.91 34
Metal PCA 1 766.0831 4 3744 1540.166 23.2332 33
Best models step 2
Organic PCA 2 þlarval dragonfly
abundance
746.037 5 3744 1502.074 0 79
Metal PCA 2 þlarval dragonfly
abundance
750.9626 5 3744 1511.925 9.8512 68
Best models step 3
Metal PCA 2 þlarval dragonfly
abundance þorganic PCA 2
741.9137 6 3744 1495.827 0 89
UVB þlarval dragonfly abundance þ
organic PCA 2
744.9576 6 3744 1501.915 6.0882 82
Metal PCA 1 þlarval dragonfly
abundance þorganic PCA 2
745.6634 6 3744 1503.327 7.4998 80
Larval damselfly abundance þlarval
dragonfly abundance þorganic PCA 2
745.7939 6 3744 1503.588 7.7608 80
Sum of predaceous invertebrates þlarval
dragonfly abundance þorganic PCA 2
745.818 6 3744 1503.636 7.809 80
Larval beetle abundance þlarval
dragonfly abundance þorganic PCA 2
745.8973 6 3744 1503.795 7.9676 80
Organic PCA 1 þlarval dragonfly
abundance þorganic PCA 2
745.9437 6 3744 1503.887 8.0604 80
Average temperature þlarval dragonfly
abundance þorganic PCA 2
746.027 6 3744 1504.054 8.227 79
Interaction terms
Best model with dragonfly 3metal
interaction
739.8483 7 3744 1493.697 0 94
Best model with dragonfly 3organic
interaction
740.447 7 3744 1494.894 1.1974 92
Best model (metal PCA 2 þlarval
dragonfly abundance þorganic PCA 2)
741.9137 6 3744 1495.827 2.1308 89
Best model with organic 3size interaction 741.7028 7 3744 1497.406 3.709 89
Best model with metal 3size interaction 741.8577 7 3744 1497.715 4.0188 89
Best model with inorganic 3organic
interaction
741.8993 7 3744 1497.799 4.102 89
Notes: For model step 1, all models are presented. For model steps 2 and 3, only models within 10 AIC points of the best-fitting
model are presented. Change in AIC is calculated relative to the best model in that modeling step. Quasi-R
2
is calculated by
dividing the difference in log likelihood between the present model and the saturated model by the difference in log likelihood
between the saturated model and the intercept-only model. The saturated model included intercept and all parameters tested. The
abbreviation ‘‘NA’’ means ‘‘not analyzed.’’
August 2010 429MULTIPLE STRESSORS AND ABNORMAL FROGS

2006, eggs were counted at the wetland and placed
directly into cages to hatch. The cages were deployed
close to vegetation in water 75 cm deep. In 2005, 100
tadpoles were placed in each cage and were checked
twice monthly to monitor tadpole development and
effectiveness of predator exclusion. In 2006, only 50 eggs
were added to each cage because survival in cages was
poor in 2005 and the metamorphs were small (10–13
mm SVL). At the end of each summer, metamorphs
were measured, examined for abnormalities, and re-
leased into the pond when they reached Gosner stage 42
or later. We conducted a Gtest (Sokal and Rohlf 1994)
to determine whether the frequency of abnormalities in
the exclosures was significantly different from the
frequency in the wild frogs at the same sites during the
years in which the cages were deployed.
Amputation experiment
Experimental animals were harvested from eight
recently laid egg masses, Gosner stages 9–12, on 13
May 2008. Eggs were reared to Gosner stage 28, at
which point limb amputations began. Tadpoles (n¼32,
TABLE 3. Model results for skeletal abnormalities: parameter estimates for each factor.
Model Intercept
Developmental
stage Frog size
Intercept only 2.8747
Saturated model 10.2512 0.1469 0.1358
Univariate models step 1
Larval dragonfly abundance 10.8955 0.2222 0.1133
Organic PCA 2 12.0494 0.2616 0.1247
Sum predators 11.9744 0.2577 0.1331
Temperature 12.8122 0.24 0.119
Organic PCA 1 12.7336 0.2732 0.1146
UVB 13.6824 0.2774 0.0997
Metal PCA 2 12.7329 0.2622 0.0876
Larval damselfly abundance 12.461 0.2657 0.1142
Larval beetle abundance 12.6961 0.2682 0.1016
Metal PCA 1 12.9016 0.2728 0.1032
Best models step 2
Organic PCA 2 þlarval dragonfly abundance 9.4175 0.2017 0.1529
Metal PCA 2 þlarval dragonfly abundance 10.2286 0.1907 0.0796
Best models step 3
Metal PCA 2 þlarval dragonfly abundance þorganic PCA 2 8.6393 0.1667 0.1175
UVB þlarval dragonfly abundance þorganic PCA 2 10.2102 0.2078 0.1539
Metal PCA 1 þlarval dragonfly abundance þorganic PCA 2 9.4536 0.2008 0.1502
Larval damselfly abundance þlarval dragonfly abundance þorganic PCA 2 9.679 0.2052 0.1452
Sum of predaceous invertebrates þlarval dragonfly abundance þorganic PCA 2 9.1948 0.1985 0.161
Larval beetle abundance þlarval dragonfly abundance þorganic PCA 2 9.5663 0.2044 0.1537
Organic PCA 1 þlarval dragonfly abundance þorganic PCA 2 9.4607 0.2006 0.1487
Mean temperature þlarval dragonfly abundance þorganic PCA 2 9.3872 0.203 0.1523
Interaction terms
Best model with dragonfly 3metal interaction 8.5359 0.1711 0.1376
Best model with dragonfly 3organic interaction 7.9233 0.151 0.1232
Best model (metal PCA 2 þlarval dragonfly abundance þorganic PCA 2) 8.6393 0.1667 0.1175
Best model with organic 3size interaction 8.8085 0.1695 0.1145
Best model with metal 3size interaction 8.6169 0.1641 0.1131
Best model with inorganic 3organic interaction 8.6843 0.1673 0.1167
Notes: For model step 1, all models are presented. For model steps 2 and 3, only models within 10 AIC points of the best-fitting
model are presented. Developmental stage is stage of each individual metamorph between 42 and 46 according to Gosner (1960).
Frog size is snout to vent length (mm) of each individual metamorph. Larval abundance values are the average of two abundance
measures of all dragonflies (Aeshna spp., Leuchorrinia spp., and Libellula spp.), damselflies (Coenagrionid spp. and Lestes spp.), and
beetle (Rhantus spp., Ilybius spp., and Dytiscus spp.) larvae captured in early-season sweeps in 2005 and 2006. Metal PCA vector 1
explained 33%of the variance and was positively correlated (r0.5) with the following elements: iron, manganese, and nickel in
sediment; aluminum, barium, iron, and manganese in water. This vector was also negatively correlated (r0.5) with sediment
arsenic and cadmium. Metal PCA vector 2 explained 25%of the variance and was positively correlated (r0.5) with copper, iron,
nickel, and zinc in sediment. Organic PCA vector 1 explained 38%of the variance and was positively correlated (r0.5) with the
following compounds: total PCBs, aldrin, heptachlor-epoxide, mirex, and chlordane. Organic PCA vector 2 explained 25%of the
variance and was positively correlated (r0.5) with lindane and DDT and negatively correlated (r0.5) with total PCBs. UVB
values are the percentages of UVB attenuation from the surface to a 10-cm depth in site water. Mean temperature values used in
modeling are the average water temperatures at 5-cm depth measured at sites in 2006. The sums of predaceous invertebrates values
used in modeling are the means of two abundance measures of all dragonfly, damselfly, and larval beetle predators. The saturated
model included intercept and all parameters tested. The abbreviation ‘‘NA’’ means ‘‘not analyzed.’’
MARI K. REEVES ET AL.430 Ecological Monographs
Vol. 80, No. 3

Gosner stages 28–34) were amputated from 16 to 24
June 2008. These stages of development range from the
limb bud slightly longer than its diameter to the foot
paddle having four of five finger indentations differen-
tiated. Tadpoles were anesthetized with MS-222 (1:2000)
for 3 min prior to surgery. Once tadpoles were
nonresponsive, the right hind limb bud was amputated
and the left limb kept intact for comparison. Tadpoles
were examined for other gross abnormalities, staged,
and placed in individual 3-L glass bowls of spring water
to recover. Between each amputation, a control tadpole
was anesthetized and examined, but not amputated, so
that siblings (n¼25) of each amputated animal were
allowed to develop as controls. A scalpel was used in the
first few amputations, but was too large to make exact
cuts in tiny limb buds; we performed the remaining
amputations with small, pointed tweezers.
After surgery, glass bowls with one tadpole each were
placed outdoors in three water baths containing both
amputated and control animals in a randomized block
design. Animals were kept outdoors until the end of the
experiment. Water was changed weekly until 1 August
2008 and twice per week thereafter to maintain water
quality. Tadpoles were fed NASCO frog brittle for
Xenopus ad libitum after each water change. Tadpoles
were reared until 29 August 2008, at which time each
was euthanized by overdose of MS-222, photographed,
assessed for malformations, and preserved in 10%
neutral buffered formalin. We tested for significant
differences in the probability of survival between injured
and control animals with logistic regression.
Site sediment and water exposure experiment
We exposed developing wood frogs (from eggs to
metamorphosis) to sediment and water from six
wetlands at which abnormalities have consistently been
found (see Appendix A for additional details). Two
controls were also used: one with only spring water and
the other with water and quartz sand to simulate the
potential negative effects of sediment in the bowls (e.g.,
incomplete water changes). We tested parentage by
rearing tadpoles from six parent pairs from two different
sites to control for heritable traits or differences in
tadpole fitness. A randomized block design was used to
test effects due to site and parentage.An experimental
unit was one chemically clean, 3-L glass bowl. Twelve
TABLE 3. Extended.
Larval abundance
Metal
PCA 1
Metal
PCA 2
Organic
PCA 1
Organic
PCA 2 UVB
Mean
temperature
Sum of
predaceous
invertebrates
Interaction
termDragonfly Damselfly Beetle
0.0304 0.009 0.0026 0.1649 0.2139 0.0209 0.2464 0.4859 0.1286 0.004
0.0287
0.2045
0.0021
0.1045
0.1113
0.2619
0.0502
0.0023
0.0247
0.0145
0.0357 0.3281
0.0333 0.1271
0.0411 0.1494 0.34
0.0331 0.3704 0.2929
0.036 0.0504 0.3581
0.0364 0.0016 0.3394
0.0348 0.3185 0.0006
0.0362 0.0211 0.3369
0.0371 0.0327 0.3233
0.036 0.3328 0.0067
0.0394 0.2847 0.3785 0.0145
0.0434 0.1632 0.4986 0.0097
0.0411 0.1494 0.34 NA
0.0409 0.1576 0.7581 0.0224
0.041 0.2691 0.3453 0.0068
0.041 0.1441 0.3432 0.0133
August 2010 431MULTIPLE STRESSORS AND ABNORMAL FROGS

tadpoles were tested per site and 16 per parent pair, for a
total of 96 experimental units.
We first evaluated hatching success by placing 20 eggs
from a single parent pair (Gosner stages 3–8) in each
experimental unit and counting hatchlings 19 d later. To
maximize survival, the most vigorous tadpole was left in
each container. This may have biased our experimental
results for time to metamorphosis and size, but it would
have biased them towards not detecting deleterious
effects because tadpoles tested were among the healthiest
of their sibling group. At metamorphosis (Gosner stage
42), each frog was weighed, measured, and assessed for
the presence of abnormalities by standard protocols
(U.S. Fish and Wildlife Service 1999, 2007). The date of
metamorphosis was recorded. Tadpoles were euthanized
by overdose of MS-222. We used general linear models
in SAS to test whether site and parent significantly
affected mass, time to metamorphosis, and size at
metamorphosis (SVL). We used logistic regression to
test parent and site effects on hatching success and
survival.
RESULTS
Between 2000 and 2006, across 21 breeding sites, 299
(7.5%) of 4009 metamorphs examined were abnormal
(Appendix C: Table C1 and Fig. C1). Individuals often
had more than one abnormal body part, resulting in 330
separate abnormalities. Prevalence of skeletal abnor-
malities at individual breeding sites ranged from 0 to
16%(overall prevalence 6%), with the highest prevalence
at a site adjacent to the Sterling Highway in 2006 (Table
1). Eye abnormality frequency varied from 0 to 10%
(overall prevalence 2%), with the highest frequency at a
remote site in 2004 (Table 1). Skeletal abnormalities
were found at most sites in most years, whereas half of
the sites had at least one year with no eye abnormalities.
TABLE 4. Model results for eye abnormalities: model evaluation metrics.
Model Likelihood
Para-
meters Trials AIC DAIC
Quasi-R
2
(%)
Intercept only 351.8538 1 3744 705.7076 NA NA
Saturated model 336.7327 13 3744 699.4654 NA NA
Best models step 1
Larval beetle abundance 346.0375 2 3744 696.075 0 38
Developmental stage 347.0449 2 3744 698.0898 2.0148 32
Larval dragonfly abundance 348.6181 2 3744 701.2362 5.1612 21
Frog size 349.1052 2 3744 702.2104 6.1354 18
Mean temperature 349.7564 2 3744 703.5128 7.4378 14
Larval damselfly abundance 349.8507 2 3744 703.7014 7.6264 13
Metal PCA 1 350.0667 2 3744 704.1334 8.0584 12
Organic PCA 2 351.4335 2 3744 706.867 10.792 3
Organic PCA 1 351.6591 2 3744 707.3182 11.2432 1
Sum of predaceous invertebrates 351.8185 2 3744 707.637 11.562 0
Metal PCA 2 351.8212 2 3744 707.6424 11.5674 0
UVB 351.8414 2 3744 707.6828 11.6078 0
Best models step 2
Larval beetle abundance þdevelopmental stage 342.3522 3 3744 690.7044 0 63
Larval beetle abundance þfrog size 342.7243 3 3744 691.4486 0.7442 60
Larval dragonfly abundance þlarval beetle abundance 344.3766 3 3744 694.7532 4.0488 49
Developmental stage þlarval dragonfly abundance 344.7845 3 3744 695.569 4.8646 47
Best models step 3
Developmental stage þfrog size þlarval beetle abundance 339.9631 4 3744 687.9262 0 79
Larval dragonfly abundance þlarval beetle abundance
þfrog size
341.2563 4 3744 690.5126 2.5864 70
Developmental stage þlarval beetle abundance
þlarval dragonfly abundance
341.2595 4 3744 690.519 2.5928 70
Developmental stage þlarval beetle abundance þUVB 341.5328 4 3744 691.0656 3.1394 68
UVB þfrog size þlarval beetle abundance 341.6715 4 3744 691.343 3.4168 67
Developmental stage þlarval beetle abundance
þmetal PCA 2
341.8641 4 3744 691.7282 3.802 66
Developmental stage þlarval beetle abundance
þsum of predaceous invertebrates
341.9315 4 3744 691.863 3.9368 66
Organic PCA 2 þdevelopmental stage
þlarval beetle abundance
341.936 4 3744 691.872 3.9458 66
Developmental stage þlarval beetle abundance
þlarval damselfly abundance
341.983 4 3744 691.966 4.0398 65
Notes: For model step 1, all models are presented. For model steps 2 and 3, only models within 5 AIC points of the best-fitting
model are presented. Change in AIC is calculated relative to the best model in that modeling step. Quasi-R
2
is calculated by
dividing the difference between the present model and the saturated model by the difference between the saturated model and the
intercept-only model. The abbreviation ‘‘NA’’ means ‘‘not analyzed.’’
MARI K. REEVES ET AL.432 Ecological Monographs
Vol. 80, No. 3

More than 20 types of abnormalities were seen
(Appendix C: Table C1), of which 72%were skeletal
and 28%were eye. Ectromelia (partial limb), micromelia
(shrunken limb or limb element), and amelia (limb
totally missing) collectively accounted for half of all
skeletal abnormalities (Appendix C: Fig. C1 and Table
C1), and unpigmented iris (eye totally black) was the
dominant eye abnormality (85%of eye abnormalities
were of this type). The more unusual abnormalities
included microcephaly (shrunken head), scoliosis
(curved spine), cutaneous fusion (skin webbing), and
kinked tails (Appendix C: Table C1). The rarest
abnormality type was polymelia (extra limb); only one
specimen had an extra limb.
Contaminants
Organic and inorganic contaminants were frequently
detected above toxic thresholds in study site sediment
and water (Appendix B: Tables B1–B6 ). At every site,
total PCBs exceeded the threshold effects level (TEL;
Buchman 2008), and at 11 of 21 sites (52%)they
exceeded the probable effects level (PEL; Buchman
2008). Other organochlorines also surpassed toxic
thresholds in multiple study sites: DDT and its
metabolites (DDD and DDE), chlordane (alpha and
gamma), and heptachlor epoxide were all greater than
TELs in sediment from at least one site (Appendix B:
Table B1), and aldrin, lindane metabolites (alpha-
BHC, gamma-BHC), and mirex were over the lowest
effects level (LEL) in one or more locations. For
inorganics, sediment As, Cd, Cu, Fe, Mn, Ni, and Zn
all exceeded at least one sediment screening level at one
or more sites (Appendix B: Table B1), and water Al,
Ba, Cd, Cu, Fe, and Mn all surpassed chronic aquatic
toxicity thresholds (Appendix B: Table B2). Other
contaminants were found in sediment and water from
multiple sites and these are presented in Appendix B
(Tables B1–B6).
Predators
Predatory invertebrates identified in 2005 and 2006
included nine genera of diving beetles, three genera of
dragonflies, and two genera of damselflies (Appendix B:
Tables B7 and B8 contain raw abundance data).
Parasites
Of the 404 specimens examined for parasites, none
were infected with Ribeiroia ondatrae,norwere
planorbid snail hosts seen at any sampling site. There
have been no records of R. ondatrae in Alaska or north
of approximately 508N latitude in North America
(P. T. J. Johnson, unpublished data). Other parasites
were found, including four types of trematode meta-
cercariae and four protist taxa, but none of these are
known to cause limb abnormalities in frogs (prevalence
and intensity of parasites are presented in Appendix B:
Table B9). Overall, parasite abundance and diversity
were low.
Statistical assessment of skeletal abnormalities
Skeletal abnormalities were best predicted by a model
including dragonflies, metals, and organic contaminants,
along with frog size and developmental stage (Fig. 1,
Tables 2 and 3; Appendix D: Fig. D2). This model had a
quasi-R
2
of 0.89, and the Hosmer-Lemeshow test failed
to detect a significantly poor model fit (P¼0.84). The
metals parameter in this best-fitting model was inorganic
PCA vector 2 (correlated with sediment Cu, Fe, Ni, and
Zn). For every one-unit increase in the metals vector, the
odds of a frog having a skeletal abnormality increased
by 16%(odds ratio [OR] ¼1.16, 95%CI ¼1.02–1.32).
An increase in organic contaminants had a similar effect.
For each one-unit increase in the organic vector
positively correlated with DDT, lindane, and phenan-
threne, the odds of having a skeletal abnormality
increased by 40%(OR ¼1.40, 95%CI ¼1.17–1.68).
The early-season abundance of dragonfly larvae was
also positively related to the risk of skeletal abnormal-
ities, and each additional dragonfly nymph increased the
odds of a frog having a skeletal abnormality by 4%(OR
¼1.04, 95%CI ¼1.03–1.06). Although this number may
seem low, this is because the OR reports the additional
risk for each unit increase in dragonflies, which were
abundant in some study sites (mean ¼24, range ¼0–71).
Although water temperature was not among the best
predictors for skeletal abnormalities, dragonflies and
damselflies were more abundant in warmer sites (Fig. 2).
Frog size and developmental stage were also correlated
with skeletal abnormalities in this model. Smaller
metamorphs (OR ¼0.89, 95%CI ¼0.83–0.96) and later
stage metamorphs were more likely to have skeletal
abnormalities (OR ¼1.18, 95%CI ¼1.04–1.34 ).
The best skeletal abnormality model AIC was 21 units
lower than the model with only dragonfly larvae and the
(size and stage) covariates (Tables 2 and 3), 43 units
lower than the model with only metals and covariates,
and 38 units lower than the model with only the organic
contaminants and covariates. This comparative analysis
indicates that all three factors (metals, organic contam-
inants, and dragonflies) produced a substantially better
model than any of these factors alone. The only
interaction term that improved model fit by .2 AIC
points was a dragonfly 3metal interaction (Tables 2 and
3). This interaction term was small and negative, which
implies a sub-additive interaction between these two
factors.
Statistical assessment of eye abnormalities
The best model for eye abnormalities associated them
with more predatory beetles, smaller frogs, and earlier
stage frogs (Tables 4 and 5). The best model included
the early-season abundance of Dytiscus spp., Rhantus
spp., and Ilybius spp. (square-root-transformed, OR ¼
1.14, 95%CI ¼1.07–1.22), developmental stage (OR ¼
0.79, 95%CI ¼0.65–0.96 ), and size (OR ¼0.89, 95%CI
¼0.81–0.97). This model had a quasi-R
2
of 0.79, and a
Hosmer-Lemeshow test for this model also failed to
August 2010 433MULTIPLE STRESSORS AND ABNORMAL FROGS

detect a significantly poor model fit (P¼0.72). No other
model was within 2 AIC points of this model, and AIC
values increased (indicating worse model fit) when we
added interaction terms.
Statistical assessment of roads
Road type was a significant predictor of inorganic
contaminants in site sediment and water (Fig. 3). All
road types (paved, gravel, and wilderness) differed
significantly in metals based on ANOSIM (global R¼
0.38, P¼0.001; gravel vs. paved, R¼0.73, P¼0.005;
gravel vs. wilderness, R¼0.16, P¼0.017; paved vs.
wilderness, R¼0.76, P¼0.005).
Predator exclusion experiment
A total of 213 caged tadpoles reached metamorphosis
in 2005 and 2006, and none had limb abnormalities. At
the study sites where cages were deployed, 6.3%(28/443)
of wild tadpoles had limb abnormalities during the same
two-year period. The prevalence of skeletal abnormal-
ities was significantly different (Gtest, P,0.0001)
inside cages vs. in wild frogs. No significant difference
was found in unpigmented irises inside vs. outside of
cages (7/443 outside cages vs. 4/213 inside cages; P¼
0.7829); however, power was only 0.05, and we therefore
do not further evaluate this finding. Additional details
are included in Appendix D.
TABLE 5. Model results for eye abnormalities: parameter estimates for each factor.
Model Intercept
Developmental
stage Frog size
Intercept only 3.9461
Saturated model 10.7463 0.2005 0.0927
Best models step 1
Larval beetle abundance 4.222
Developmental stage 9.4837 0.3095
Larval dragonfly abundance 3.6314
Frog size 1.6196 0.126
Mean temperature 2.352
Larval damselfly abundance 3.7404
Metal PCA 1 4.013
Organic PCA 2 3.9398
Organic PCA 1 3.9386
Sum of predaceous invertebrates 3.898
Metal PCA 2 3.9464
UVB 4.0201
Best models step 2
Larval beetle abundance þdevelopmental stage 7.775 0.2759
Larval beetle abundance þfrog size 1.721 0.1368
Larval dragonfly abundance þlarval beetle abundance 3.9414
Developmental stage þlarval dragonfly abundance 8.5488 0.2818
Best models step 3
Developmental stage þfrog size þlarval beetle abundance 8.4716 0.241 0.1211
Larval dragonfly abundance þlarval beetle abundance þfrog size 1.5399 0.1319
Developmental stage þlarval beetle abundance þlarval dragonfly abundance 7.2172 0.2577
Developmental stage þlarval beetle abundance þUVB 8.2778 0.275
UVB þfrog size þlarval beetle abundance 0.9985 0.142
Developmental stage þlarval beetle abundance þmetal PCA 2 8.1084 0.2838
Developmental stage þlarval beetle abundance þsum of predaceous invertebrates 7.0256 0.2553
Organic PCA 2 þdevelopmental stage þlarval beetle abundance 7.3843 0.2671
Developmental stage þlarval beetle abundance þlarval damselfly abundance 7.3241 0.2628
Notes: For model step 1, all models are presented. For model steps 2 and 3, only models within 5 AIC points of the best-fitting
model are presented. Developmental stage is stage of each individual metamorph between 42 and 46 according to Gosner (1960).
Frog size is snout to vent length (mm) of each individual metamorph. Values used in modeling are the average of two abundance
measures of all larval dragonflies (Aeshna spp., Leuchorrinia spp., and Libellula spp.), damselflies (Coenagrionid spp. and Lestes
spp.), and beetles (Rhantus spp., Ilybius spp., and Dytiscus spp.) captured in early-season sweeps in 2005 and 2006. Metal PCA
vector 1 explained 33%of the variance and was positively correlated (r0.5) with the following elements: iron, manganese, and
nickel in sediment; aluminum, barium, iron, and manganese in water. This vector was also negatively correlated (r0.5) with
sediment arsenic and cadmium. Metal PCA vector 2 explained 25%of the variance and was positively correlated (r0.5) with
copper, iron, nickel, and zinc in sediment. Organic PCA vector 1 explained 38%of the variance and was positively correlated (r
0.5) with the following compounds: total PCBs, aldrin, heptachlor-epoxide, mirex, and chlordane. Organic PCA vector 2 explained
25%of the variance and was positively correlated (r0.5) with lindane and DDT and negatively correlated (r0.5) with total
PCBs. UVB is the percentage of UVB attenuation from the surface to a 10-cm depth in site water. Temperature values used in
modeling are the average water temperatures at 5-cm depth measured at sites in 2006. Predaceous invertebrate values used in
modeling are the average of two abundance measures of all dragonfly, damselfly, and larval beetle predators. The saturated model
included intercept and all parameters tested.
MARI K. REEVES ET AL.434 Ecological Monographs
Vol. 80, No. 3

Limb amputation experiment
Amputation resulted in limb malformations (Appen-
dix D: Fig. D1), with 29 of 32 amputated animals
developing micromelia or otherwise malformed hind
limbs, whereas only 1 of 25 control animals had a
shrunken hind limb. There was no significant difference
in survival between amputated and control animals (P¼
0.86), and no scarring was apparent at the end of the
study.
Sediment and water toxicity experiment
Of the 67 animals that survived to metamorphosis
when raised in site sediment and water, none had limb
abnormalities, which is significantly different from the
6.5%(26/397) of wild frogs with limb abnormalities
found in 2006 at the same six study sites from which
sediment and water were taken (Gtest, P¼0.004).
Several experimental frogs did have unpigmented irises
(3/67, 4.5%), and this was not significantly different
from the field frequency of 2.3%for this abnormality
type (9/397 animals, Gtest, P¼0.3195; but power was
only 0.08). Water and sediment from different sites
produced highly significant differences in hatching
success (v
2
¼737.03, 7 df, P,0.0001), survival (v
2
¼
23.92, 7 df, P¼0.001), time to metamorphosis (ANOVA
F
66,7
¼5.85, P,0.0001), and length at metamorphosis
(F
63,7
¼3.36, P¼0.005; Fig. 4), but not mass at
metamorphosis (F
61,7
¼1.83, P¼0.1030). Parentage also
caused significant differences in hatching success (v
2
¼
95.25, 5 df, P,0.0001), length (F
63,5
¼2.61, P¼0.04),
and time to metamorphosis (F
66,5
¼3.20, P¼0.01), but
not mass at metamorphosis (F
61,5
¼1.57, P¼0.19) or
survival (v2¼5.82, 7 df, P¼0.32) to metamorphosis.
The nature of sublethal toxic effects of sediment and
water were site-specific. Some sites had significantly
reduced hatching success, yet developmental period and
size at metamorphosis that were comparable to controls
(sites 12 and 90; Fig. 4). Other sites had reduced
hatching success and a longer development period (sites
1 and 8; Fig. 4) or frogs were impaired in all three
measured responses: significantly reduced hatch, longer
development time, and smaller size at metamorphosis
(site 2; Fig. 4).
DISCUSSION
We found that skeletal abnormalities were related to
multiple stressors: toxic metals, organic contaminants,
and dragonfly predators, whose abundance was influ-
enced by temperature. Our field results are challenging
to interpret, but the results of the follow-up experiments
suggest a possible mechanism. The sediment and water
TABLE 5. Extended.
Larval abundance Metal Organic
UVB
Mean
temperature
Sum of
predaceous
invertebratesDragonfly Damselfly Beetle PCA 1 PCA 2 PCA 1 PCA 2
0.0135 0.0076 0.1013 0.0348 0.0172 0.1156 0.0435 0.5078 0.055 0.0011
0.1345
0.0308
0.1044
0.008
0.1434
351.4335
0.0665
0.0004
0.0175
0.0382
0.1219
0.146
0.0229 0.1131
0.0262
0.133
0.0222 0.1254
0.0189 0.1049
0.1477 0.3075
0.1765 0.3577
0.1269 0.0705
0.1422 0.0015
0.1286 0.1281
0.0038 0.1126
August 2010 435MULTIPLE STRESSORS AND ABNORMAL FROGS

toxicity experiment produced no skeletal abnormalities,
and in situ predator exclusion likewise prevented skeletal
abnormalities. Interpreted together, these results indi-
cate that predators were required to produce the
frequent skeletal abnormalities in field sites, and toxic
compounds were not capable of doing so in the absence
of predators in the subset of sites we tested. We propose
a hypothesis to explain these results: Contaminants
(metals and pesticides) interfere with predator detection
and avoidance, leading to either inadequate assessment
of predation risk from invertebrates or reduced ability to
escape predators due to smaller size, longer development
period, or behavioral impairment. The results are more
complex for eye abnormalities. We observed this
abnormality type in both the predator exclusion
experiment and the sediment and water toxicity exper-
iment, at frequencies that were not statistically different
from those found in the field. Our statistical models
showed correlations with the abundance of predatory
beetles, smaller size, and earlier developmental stage.
Combining these results leads us to suggest that eye
abnormalities may result indirectly from predators, but
it is also likely that factors we did not measure, such as
food quality (e.g., see Toomey and McGraw 2009) or
genetic predisposition (Nishioka 1977), also influence
this abnormality type.
The best predictors of skeletal abnormalities in the
KNWR were early-season abundance of dragonfly
larvae, metals, and organic contaminants. Perhaps not
surprisingly, frog size and frog developmental stage were
significant covariates. Upon first consideration, it seems
reasonable to postulate that contaminants are directly
interfering with normal amphibian development, there-
by increasing skeletal-abnormality prevalence. Indeed,
our field results from one site appear to lend support to
the idea that abnormalities are chemically induced.
Frogs at this study site (KNA54; Table 1) had an
unusual skeletal malformation at a high frequency
(10.7%), in which all four limbs were bulbous and
symmetrically short. This malformation type is consis-
tent with others reported for amphibians exposed to
environmental contaminants (Johnson et al. 2010). The
following contaminants exceeded toxic thresholds at this
FIG. 2. Relationship between early-season larval dragonfly
abundance and mean site temperature. Data points represent
the mean early-season abundance in 2005 and 2006 and the
mean site temperature in 2006.
FIG. 1. Univariate relationships between skeletal abnor-
malities in wood frogs (Rana sylvatica) and (A) metals (quasi-
R
2
¼0.35), (B) organic contaminants (quasi-R
2
¼0.41), and (C)
dragonfly larvae (quasi-R
2
¼0.60) from 21 wetlands on the
Kenai National Wildlife Refuge, south-central Alaska, USA.
Invertebrates were collected by sweeping a 0.3 30.3 m D-frame
net (350 mm mesh) at a depth of ,1 m on a 10 m long transect
through vegetative perimeters of the ponds, where tadpoles
were captured. Three 10-m transects were swept, to sample
;2120 L of water. See Methods: Sediment and water sampling...
and Appendix A for explanations of the metals and organics
principal components analysis vectors.
MARI K. REEVES ET AL.436 Ecological Monographs
Vol. 80, No. 3

location: As, Cu, Ba, heptachlor-epoxide, DDT, and
PCBs. Although we did not measure it directly, survival
here may also have been poor during the study period.
Eggs were documented each year (2004–2006), yet we
only obtained 50 metamorphs for abnormality assess-
ments in 2004. In 2006, we also observed early-stage
tadpoles with kinked tails, another common symptom of
toxicant exposure (Johnson et al. 2010). Aeshnid
dragonflies were the only dragonfly species found here
and were measured at low abundance (mean early-
season abundance ¼3), suggesting dragonflies were
unlikely to have caused such a high frequency of skeletal
abnormalities at this site. This site therefore implies a
role of toxicants independent of predators in inducing
skeletal malformations. Because toxicants and predators
tended to co-occur in most sites, our data do not allow
us to rule out an effect of toxicants independent of
predators.
Nevertheless, our controlled experiments showed that
although sediment and water from other KNWR study
sites were toxic to developing wood frogs, exposure of
tadpoles to these media during development did not
directly produce asymmetrical limb abnormalities in the
laboratory or in the predator exclusion experiment.
Thus contamination alone is probably an insufficient
explanation for the abnormalities observed in our study
system.
Moreover, in most KNWR study sites, a strong
positive association between larval dragonfly abundance
and skeletal abnormalities suggests that invertebrate
predators played an important role in whether wood
frogs developed normally. The association between
odonate predators and amphibian limb malformations
has recently been documented in two other areas
(Ballenge
´e and Sessions 2009, Bowerman et al. 2010),
and our field associations and experimental work
validate these results. A predator-inflicted injury to a
tadpole limb can result in either a malformation, in
which a partially amputated limb bud regenerates
incompletely, or in a missing limb segment, if the injury
occurs after a tadpole has lost its regenerative ability
(Fry 1966; see also Appendix D: Fig. D1). Of the 237
skeletal abnormalities we observed in the field, 92%were
FIG. 4. Effects of site sediment and water exposure on (A)
eggs (hatching success), (B) larvae (days from hatch to
metamorphosis), and (C) frogs (size measured as snout to vent
length at metamorphosis). Both the sand and water columns are
control treatments. Sand was included as a positive control to
simulate the effects of incomplete water changes in the
treatments with sediment. Values are given as mean þSE.
Different lowercase letters above bars indicate statistically
significant differences (at P,0.05). Site numbers correspond to
those in Table 1 and Appendix B.
FIG. 3. Nonmetric multidimensional scaling (NMDS) plot
of metals in site sediment and water. Key to symbols: circles,
sites adjacent to gravel roads; triangles, sites in wilderness areas;
crosses, sites adjacent to paved roads. An NMDS plot
maximizes the rank-order correlation between the calculated
Euclidian distance between samples and the physical distance in
two-dimensional ordination space. Stress is a measure of the
correlation between these two distances, with values ranging
from ,0.05 (excellent correlation, easily interpretable) to .0.2
(poor correlation, ordination challenging to interpret). Values
toward the edges of the plot represent higher or more extreme
elemental concentrations, but the specific elements differ across
the ordination space. Values closer to the center of the plot
represent more moderate or lower elemental concentrations.
August 2010 437MULTIPLE STRESSORS AND ABNORMAL FROGS

missing or shrunken limbs or digits and were therefore
consistent with an injury-based mode of origin (Appen-
dix C: Table C1). Moreover, in 21%of skeletal
abnormalities, a recent (still bleeding) injury to a limb
or digit was evident (Appendix C: Table C1). Thus, it
seems likely that injury from failed predation attempts
may be a primary cause of the asymmetrical limb
abnormalities in the KNWR.
If dragonfly predation causes most wood frog skeletal
abnormalities in the KNWR, how can we explain the
correlation between contaminants and these abnormal-
ities in our results? The correlations between skeletal
abnormalities and toxicants are particularly interesting
because both inorganic and organic contaminants occur
at high concentrations relative to toxic thresholds
(Buchman 2008) in some KNWR wetlands. One
possible effect of these contaminants is that frogs
become more vulnerable to predation in their presence.
For example, some metals may inhibit the ability of
tadpoles to detect and avoid dragonfly predators, as has
been convincingly demonstrated in fish. In two compel-
ling studies on Cu neurotoxicity to salmon, as little as 2
ppb Cu in water inhibited fish olfaction (Sandahl et al.
2007), and 5 ppb Cu caused smolts to alter their normal
predator avoidance behavior, thereby suffering a signif-
icantly greater number of predator attacks than
untreated controls (J. K. McIntyre, unpublished manu-
script). Given that amphibians are also known to emit
and respond to olfactory chemical cues (Schoeppner and
Relyea 2005), further study of metal effects on wood
frog olfaction and predator avoidance behavior is
warranted. Organic contaminants can also cause behav-
ioral impairment (Cooke 1972), which could reduce
predator avoidance ability in contaminated sites. One of
the pesticides correlated with skeletal abnormalities in
the KNWR is DDT, a persistent organochlorine
pesticide used historically for mosquito control. At high
concentrations, DDT has been shown to alter tadpole
swimming behavior (Cooke 1972). DDT has also been
found to bioaccumulate in frogs in natural settings
(Hofer et al. 2005), especially at colder temperatures
(Licht 1976), so even though DDT concentrations in
KNWR site sediments were not as high as those
resulting in mortality and ‘‘hyperactivity’’ (Cooke
1972), other sublethal toxic effects are possible.
Contaminants have also been shown to reduce
tadpole survival or size at metamorphosis, which might
increase vulnerability to predation (Johnson et al. 2010).
In our field assessment, smaller frogs were more likely to
have skeletal abnormalities, making it plausible that
toxicants render frogs small and weak, thereby hinder-
ing their ability to escape predators. In our experimental
work, site sediment and water reduced frog size, showing
that a size-mediated mechanism of increased predation
may indeed be driven by contaminants. Alternatively,
toxicants may slow frog development, increasing the
time span during which tadpoles are vulnerable to
predators (see also Fig. 4). Although amputation injury
did not affect survival in our amputation experiment,
this may not be the case in natural settings when other
stressors co-occur, such as predators and contaminants.
It should be noted that whereas frog size may be either a
cause or an effect of wood frog abnormalities, our
finding that skeletal abnormalities are more prevalent at
later developmental stages is probably due to sampling
bias created by different capture techniques (as discussed
in Reeves et al. 2008). In summary, the correlation that
we have shown between toxicant presence and reduced
tadpole viability (i.e., poor hatching success and
survival, slower growth, and smaller size at metamor-
phosis) suggests that poor environmental conditions in
KNWR wetlands may be having population-level effects
for which the physical abnormalities serve as indicators.
This question warrants additional research.
There was a negative interaction in our statistical
models between metals and dragonflies, indicating sub-
additivity in the effects of these variables. This is
somewhat dissatisfying because it might indicate mild
collinearity between metals and dragonflies, and we do
not have independent evidence to confirm or refute the
occurrence of an interaction. One possible interpretation
is that the effects of dragonflies were greater at low metal
concentrations, perhaps because metals interfere with
olfaction for prey detection in dragonflies. Another
possibility is that mortality (which we did not measure in
the field) is differentially higher at sites with high metals
concentrations and abundant dragonflies, leading us to
measure fewer abnormal frogs because injured tadpoles
did not escape predators in these contaminated sites.
The change in AIC from adding the interaction was also
small (2 AIC points; Tables 2 and 3). We therefore
refrain from interpreting this further while noting that
both metals and dragonflies still increased skeletal
abnormalities independently (Tables 2 and 3), despite
the presence of the interaction.
Consistent with Reeves et al. (2008), who reported
that proximity to roads increased the risk of skeletal
abnormalities in wood frogs throughout Alaska, we
found that road presence was associated with chemical
contamination. Roads may release contaminants by
changing the geochemistry of disturbed areas and
mobilizing metals from the roadbed, especially during
storm events (Wheeler et al. 2005 ). Road traffic also
releases hydrocarbons and metals through exhaust and
mechanical wear (Wheeler et al. 2005). Conversely,
some elements, such as As, Ba, and Se, appear to be
naturally elevated in some areas of KNWR (Crock et al.
1992). Further study is needed to determine the
contribution of roads to the abnormalities in KNWR
and thus whether management actions are needed.
The majority (85%)ofeyeabnormalitieswere
unpigmented irises. Unlike skeletal abnormalities, we
observed unpigmented irises in both the predator
exclusion experiment and our sediment and water
toxicity experiment at frequencies not significantly
different from those observed in the field, which suggests
MARI K. REEVES ET AL.438 Ecological Monographs
Vol. 80, No. 3

the eye abnormalities do not result from a direct
interaction with predators. Nevertheless, our statistical
models implicate predatory beetles in this abnormality
type. We therefore infer that indirect interactions with
these predators may be important. It has been estab-
lished in birds that the carotenoid pigments that color
the eyes have dietary precursors and therefore indicate
diet quality (Toomey and McGraw 2009). It is possible
that a lack of colored pigment in the frog iris is also a
result of poor diet, which could explain the associations
between this abnormality type and both predatory
beetles and smaller size. Perhaps in areas of high
predator abundance, tadpoles forage less and hide more,
thereby consuming a lower-quality diet that leads to less
colorful pigmentation. On the other hand, while
experimental results related to eye abnormalities are
scant, available evidence suggests unpigmented irises
may also be related to recessive genetic mutations
(Nishioka 1977), and it is probable that some factor
we did not measure (e.g., individual fitness or population
health) influences the expression of this abnormality
type. Further investigation is needed to determine the
effect of these eye abnormalities on the vision of
individual frogs and the broader viability of wood frog
populations.
We conclude that skeletal abnormalities are caused by
multiple stressors: organic and inorganic contaminants
and dragonfly predators. We propose that eye abnor-
malities may result indirectly from reduced foraging in
areas with abundant predators, leading to poor nutrition
and smaller size in frogs that express this abnormality
type. For both skeletal and eye abnormalities, the nature
of interaction between abiotic and biotic factors is a
critical area for future research, as is the relationship
between environmental stressors, amphibian abnormal-
ities, and population health. It is important to examine
such multiple stressors in systems that include other
amphibian species and also to explore possible interac-
tions with parasites where they occur.
ACKNOWLEDGMENTS
We thank the following: staff at KNWR, C. Caldes, J. Hall,
J. Morton, and R. West; Unocal, Chevron, and Marathon Oil
Co. employees, especially G. Merle; at UC-Davis, M. Johnson,
A. K. Miles, and N. Willits; for field assistance, E. Brown, M.
Fan, S. Jensen, S. Keys, N. Maxon, E. Moreno, M. Nemeth, M.
Perdue, J. Ramos, C. Schudel, J. Stout, H. Tangermann, C.
Wall, and H. Zimmer; for parasitology, D. Larson (University
of Alaska, Fairbanks) and P. Johnson (University of Colorado,
Boulder); for radiography, D. Green (USGS), L. Guderyahn
(Ball State University), and M. Lannoo (Indiana University);
and for invertebrate sampling and assessment, W. Walton
(UC–Riverside). We honor the memory of D. Sutherland,
formerly with the University of Wisconsin, La Crosse, who
performed most parasitology for this project. The USFWS
Division of Environmental Quality provided financial support
(FFS Number 7N23; DEC ID 200470001).
LITERATURE CITED
Anderson, M. T., J. M. Kiesecker, D. P. Chivers, and A. R.
Blaustein. 2001. The direct and indirect effects of temperature
on a predator–prey relationship. Canadian Journal of
Zoology 79:1834–1841.
Ankley, G. T., S. J. Degitz, S. A. Diamond, and J. E. Tietge.
2004. Assessment of environmental stressors potentially
responsible for malformations in North American anuran
amphibians. Ecotoxicology and Environmental Safety 58:7–
16.
ASIH [American Society of Ichthyologists and Herpetologists].
2004. Guidelines for use of live amphibians and reptiles in
field and laboratory research. hhttp://www.asih.org/files/
hacc-final.pdfi
Bacon, J. P., J. A. Gray, and L. Kitson. 2006. Status and
conservation of the reptiles and amphibians of the Bermuda
islands. Applied Herpetology 3:323–344.
Ballenge
´e, B., and S. K. Sessions. 2009. Explanation for missing
limbs in deformed amphibians. Journal of Experimental
Zoology (Molecular and Developmental Evolution) 312B:1–
10.
Bohl, E. 1997. Limb deformities of amphibian larvae in Aufsess
(Upper Franconia): attempt to determine causes. Mu
¨nich
Contributions to Wastewater Fishery and River Biology 50:
160–189.
Bowerman, J., P. T. J. Johnson, and T. Bowerman. 2010.
Sublethal predators and their injured prey: linking aquatic
predators and severe limb abnormalities in amphibians.
Ecology 91:242–251.
Buchman, M. F. 2008. NOAA screening quick reference tables.
Report 08-1. Office of Response and Restoration Division,
National Oceanic and Atmospheric Administration, Seattle,
Washington, USA.
Burnham, K. P., and D. R. Anderson. 2002. Model selection
and multimodel inference: a practical information-theoretic
approach. Springer-Verlag, New York, New York, USA.
Clifford, H. F. 1991. Aquatic invertebrates of Alberta.
University of Alberta Press, Edmonton, Alberta, Canada.
Cooke, A. S. 1972. The effects of DDT, Dieldren and 2,4-D on
amphibian spawn and tadpoles. Environmental Pollution 3:
51–68.
Crock, J. G., R. C. Severson, and L. P. Gough. 1992.
Determining baselines and variability of elements in plants
and soils near the Kenai National Wildlife Refuge, Alaska.
Water, Air, and Soil Pollution 63:253–271.
Fry, A. E. 1966. Hind limb regeneration in Rana pipiens larvae.
Copeia 1966:530–534.
Gosner, K. L. 1960. A simplified table for staging anuran
embryos and larvae with notes on identification. Herpeto-
logica 16:183–190.
Gurushankara, H. P., S. V. Krishnamurthy, and V. Vasudev.
2007. Morphological abnormalities in natural populations of
common frogs inhabiting agroecosystems of the central
Western Ghats. Applied Herpetology 4:39– 45.
Herreid, C. F., and S. Kinney. 1967. Temperature and
development of the wood frog, Rana sylvatica, in Alaska.
Ecology 48:579–590.
Hillis, D. M. 2007. Constraints in naming parts of the tree of
life. Molecular Phylogenetics and Evolution 42:331–338.
Hofer, R., R. Lackner, and G. Lorbeer. 2005. Accumulation of
toxicants in tadpoles of the common frog (Rana temporaria)
in high mountains. Archives of Environmental Contamina-
tion and Toxicology 49:192–199.
Hopkins, W. A., J. Congdon, and J. K. Ray. 2000. Incidence
and impact of axial malformations in larval bullfrogs (Rana
catesbeiana) developing in sites polluted by a coal-burning
power plant. Environmental Toxicology and Chemistry 19:
862–868.
Johnson, P. T. J., J. M. Chase, K. L. Dosch, R. B. Hartson,
J. A. Gross, D. J. Larson, D. R. Sutherland, and S. R.
Carpenter. 2007. Aquatic eutrophication promotes patho-
genic infection in amphibians. Proceedings of the National
Academy of Sciences USA 104:15781–15786.
August 2010 439MULTIPLE STRESSORS AND ABNORMAL FROGS

Johnson, P. T. J., K. B. Lunde, E. G. Ritchie, J. K. Reaser, and
A. E. Launer. 2001. Morphological abnormality patterns in a
California amphibian community. Herpetologica 57:336–
352.
Johnson, P. T. J., M. K. Reeves, S. Krest, and A. E. Pinkney.
2010. A decade of deformities: advances in our understanding
of amphibian malformations and their implications. Pages
515–540 in D. W. Sparling, G. Linder, C. A. Bishop, and S.
Krest, editors. Ecotoxicology of amphibians and reptiles.
Second edition. Society of Environmental Toxicology and
Chemistry (SETAC) Press, Pensacola, Florida, USA.
Johnson, P. T. J., and D. R. Sutherland. 2003. Amphibian
deformities and Ribeiroia infection: an emerging helminthi-
asis. Trends in Parasitology 19:322–335.
Klein, E., E. E. Berg, and R. Dial. 2005. Wetland drying and
succession across the Kenai peninsula lowlands, south-
central Alaska. Canadian Journal of Forestry Research 35:
1931–1941.
Licht, L. E. 1976. Time course of uptake, elimination, and
tissue levels of
14
C-DDT in wood frog tadpoles. Canadian
Journal of Zoology 54:355–360.
Linzey, D. W., J. Burroughs, L. Hudson, M. Marini, J.
Robertson, J. Bacon, M. Nagarkatti, and P. Nagarkatti.
2003. Role of environmental pollutants on immune func-
tions, parasitic infections, and limb malformations in marine
toads and whistling frogs from Bermuda. International
Journal of Environmental Health Research 13:125–148.
Merritt, R. W., K. W. Cummins, and M. B. Berg. 1995. An
introduction to the aquatic insects of North America. Third
edition. Kendall Hunt, Dubuque, Iowa, USA.
Needham, J. G., M. J. Westfall, and M. L. May. 2000.
Dragonflies of North America. Revised edition. Scientific
Publishers, Gainesville, Florida, USA.
Nishioka, M. 1977. Color variants induced by radiation and
their inheritance in Rana nigromaculata. Science Report of
the Laboratory for Amphibian Biology, Hiroshima Univer-
sity 2:25–89.
Ouellet, M. 2000. Amphibian deformities: current state of
knowledge. Pages 617–646 in D. W. Sparling, G. Linder, and
C. A. Bishop, editors. Ecotoxicology of amphibians and
reptiles. Society of Environmental Toxicology and Chemistry
(SETAC), Pensacola, Florida, USA.
Ouellet, M., J. Bonin, J. Rodrigue, J. DesGranges, and S. Lair.
1997. Hindlimb deformities (ectromelia, ectrodactyly) in free-
living anurans from agricultural habitats. Journal of Wildlife
Diseases 33:95–104.
Parson, T. A. S. 2001. Kenai National Wildlife Refuge
Contaminant Assessment. U.S. Fish and Wildlife Service,
Anchorage, Alaska, USA.
Reeves, M. K., C. L. Dolph, H. Zimmer, R. S. Tjeerdema, and
K. A. Trust. 2008. Road proximity increases risk of skeletal
abnormalities in wood frogs from National Wildlife Refuges
in Alaska. Environmental Health Perspectives 116:1009–
1015.
Sandahl, J. F., D. H. Baldwin, J. J. Jenkins, and N. L. Scholz.
2007. A sensory system at the interface between urban
stormwater runoff and salmon survival. Environmental
Science and Technology 41:2998–3004.
Schoeppner, N. M., and R. A. Relyea. 2005. Damage, digestion
and defence: the roles of alarm cues and kairomones for
inducing prey defences. Ecology Letters 8:505–512.
Sokal, R. R., and F. J. Rohlf. 1994. Biometry: the principles
and practice of statistics in biological research. Third edition.
Freeman, New York, New York, USA.
Taylor, B., D. Skelly, L. K. Demarchis, M. D. Slade, D.
Galusha, and P. M. Rabinowitz. 2005. Proximity to pollution
sources and risk of amphibian limb malformation. Environ-
mental Health Perspectives 113:1497–1501.
Toomey, M. B., and K. J. McGraw. 2009. Seasonal, sexual, and
quality related variation in retinal carotenoid accumulation
in the house finch (Carpodacus mexicanus). Functional
Ecology 23:321–329.
Trombulak, S. C., and C. A. Frissell. 2000. Review of ecological
effects of roads on terrestrial and aquatic communities.
Conservation Biology 14:18–30.
U.S. Fish and Wildlife Service. 1999. Standard operating
procedures for abnormal amphibian surveys. hhttp://www.
fws.gov/contaminants/amphibian/SOPs.htmli
U.S. Fish and Wildlife Service. 2007. Standard operating
procedures for abnormal amphibian classification. hhttp://
www.fws.gov/contaminants/amphibian/AbnormalSOPs.
htmli
Vershinin, V. L. 2002. Ecological specificity and microevolution
in amphibian populations in urbanized areas. Advances in
Amphibian Research in the Former Soviet Union 7:1–161.
Wheeler, A. P., P. L. Angermeier, and A. E. Rosenberger. 2005.
Impacts of new highways and subsequent landscape urban-
ization on stream habitat and biota. Review of Fish Science
13:141–164.
APPENDIX A
Supplemental methodological information for contaminants sampling, data reduction, and the toxicity experiment (Ecological
Archives M080-014-A1).
APPENDIX B
Tables summarizing contaminants, invertebrate abundance, and parasite presence (Ecological Archives M080-014-A2).
APPENDIX C
Summary and photographs of abnormalities in wood frog populations at the Kenai National Wildlife Refuge, Alaska
(Ecological Archives M080-014-A3).
APPENDIX D
Supplemental results for predator exclusion and limb amputation experiments (Ecological Archives M080-014-A4 ).
MARI K. REEVES ET AL.440 Ecological Monographs
Vol. 80, No. 3