^{1}

^{*}

^{2}

^{3}

^{1}

^{a}To whom correspondence should be addressed. E-mail:

All four coauthors—KRL, JD, JRM, and MWS—participated in the conception and design of the research, data collection and analyses, and writing of the paper.

The authors have declared that no competing interests exist.

We review the evidence for the role of climate change in triggering disease outbreaks of chytridiomycosis, an emerging infectious disease of amphibians. Both climatic anomalies and disease-related extirpations are recent phenomena, and effects of both are especially noticeable at high elevations in tropical areas, making it difficult to determine whether they are operating separately or synergistically. We compiled reports of amphibian declines from Lower Central America and Andean South America to create maps and statistical models to test our hypothesis of spatiotemporal spread of the pathogen

Amphibian populations are declining across the globe at an alarming rate, with over 43% of species in a state of decline [

One cause of amphibian declines is chytridiomycosis, hypothesized to be an invasive disease [

Predicting the impacts of an invasive disease will require an understanding of the biotic and abiotic factors that influence the interactions between the host and pathogen. Temperature and moisture influence most aspects of the biology of amphibians [

Just as temperature, rainfall, and humidity influence the biology of amphibians, so too might these factors affect the growth, persistence, and ecology of a potential pathogen. Changes in regional or local climate may directly or indirectly alter pathogen development and survival rates, disease transmission, and host susceptibility [

Recent global climate change is well documented [

In this paper, we evaluate the data regarding the declines of

Because few die-offs and population declines of amphibians have ever been observed directly, especially prior to the 1990s, attribution of the causes and the timing of these events is often only a rough estimate, and in many cases, no additional data will ever be available. Furthermore, in the case of

Findings [

“Undoubtedly, the LYO does not accurately represent the timing of a disappearance in some cases, especially in tier two. Thus uncertainty is high for any particular species, and the strength of our conclusions lies in the broad patterns. Errors in the data could generate these patterns [of correlation of timing of disappearance with unusually warm years] only if sampling were biased so that the LYO tends to follow a relatively warm year irrespective of the timing of disappearances. A decline in a cool year might be misclassified as having occurred in a warm year,

Ultimately, testing hypotheses such as the CLEH or

LYO, as the year of last record [

Pounds et al. [

Elementary statistical theory indicates that increases in sampling error, independent of bias in the parameter estimator, will reduce model fit simply by increasing the variance in the variables. In the dataset, DOD differed from LYO for 24–27 of 54 species, ten of these were Stable despite a reported LYO, and for those species with different dates for LYO and DOD, the mean difference was 11.2 y ± 8.2 (standard deviation [SD]) (

List of Species and Approximate Date of Decline (DOD) Included in Our Analyses

DOD sites are indicated by open circles. Black bars indicate the hypothesized leading edge of the wave of

We tested the robustness of the relationship between AT and LYO data reported in [

(1986) indicates

We gathered records of approximate DOD and

Rates (kilometers/year) of spread were calculated by dividing the distance between pairs of locations by the interval between the DODs. Rates of spread in Central America (our Wave 1) have been described previously [

We propose

Further, for each wave, we ran additional regression analyses [

Similar to our approach for LYO, we developed a Monte Carlo procedure that simulated error in DOD to determine how robust the regression analyses of each wave were to sampling error in DOD. For each wave, we added error sampled from a uniform distribution to each datum backward in time for up to 20 y. For the error simulations on each wave, we ran 10,000 trials for each year of error that was added to the time since earliest DOD. From these trials, we calculated the mean and 95% confidence interval on the slope of the relationship (i.e., β) between the number of years since the earliest DOD against distance from the earliest DOD using linear regressions. We considered the amount of error necessary to falsify the hypothesis of an epidemic wave as that when the 95% confidence interval of the slope overlapped zero.

The earliest records of

We calculated rates of spread assuming the Venezuela record represents a second, independent introduction of

According to the CLEH, extinction risk for

The use of LYO instead of DOD, in combination with the arbitrary 1998 cut-off date [

Although it is widely assumed that the decline of amphibians in 1987 at Monteverde Cloud Forest Reserve, Costa Rica, was the result of an outbreak of

Merino-Viteri [

We found evidence of directional spread of

Although the oldest record of

Overall, there is support for directional spread of ^{2} = 0.97), Wave 3a (^{2} = 0.47), Wave 3b (^{2} = 0.40), and Wave 4 (^{2} = 0.49). The relationship for Wave 2 was not supported (^{2} values than the other waves, suggesting that sampling error in DOD likely reduced our ability to measure spatiotemporal patterns in

The relationship between time since the earliest DOD within a wave and distance of spread of ^{2} = 0.97), (C) Wave 3a (^{2} = 0.47), (D) Wave 3b (^{2} = 0.40), and for (E) Wave 4 (^{2} = 0.49), but the relationship for (B) Wave 2 was not supported (

The previously published [

Plotted are the mean correlation coefficients (± 95% CI) for the relationship between AT and LYO. Numbers above the points represent the number of simulated runs that found a relationship equal to or greater in magnitude than that found in [

(A) Error is added symmetrically around LYO from a normal distribution of mean zero and increasing standard deviation.

(B) Error is added forward in time from a uniform distribution.

(C) Error is added forward in time with respect to LYO and is sampled from a Poisson distribution.

(D) Error is added forward in time from an exponential distribution.

Unlike the relationship between AT and LYO, the relationship between time since the earliest DOD and distance of spread of

(A) When adding error backward in time from a uniform distribution, the relationship in Wave 1 remained statistically significant when random error up to 16 y was applied, in (B) Wave 3a up to 20 (+) y, in (C) Wave 3b up to 18 y, and in (D) Wave 4 up to 20 (+) y was applied.

An increasing percentage of

Bars show the number of species at each elevation category while gray depicts the number of species in decline and white depicts stable species. The percentage of species in decline is written on each bar. Total number of species included in the analysis was 51.

None of the 64 individual frogs (

Our statistical analyses of the histological results of Ecuadorian frogs [

Our results make two main points. First, we present analyses supporting a classical pattern of disease spread across naive populations, at odds with the CLEH proposed by [

Our analyses of data for declines of amphibians in the Andean region of South America identified spatiotemporal patterns that were robust to sampling error in DOD and consistent with the spread of an introduced, invasive pathogen [

Results of histology support our hypothesis that

Of particular importance when evaluating our results in light of the known errors in DOD are the results of the Monte Carlo simulations. The statistically significant spatiotemporal regressions of

Interestingly, the pattern of wave-like spread was more robust to sampling error than the CLEH. This result is likely due to resolution of LYO and DOD. Both LYO and DOD have inherent error because neither was collected with foreknowledge of pending declines—unlike data in Central America [^{2} = 0.92). In this case, Wave 2 would be nearly as well supported as Wave 1, but note that the slope of the relationship is rather low, so outliers will greatly affect this wave. Such a finding underscores the need for quality data.

In addition to identifying four regions with wave-like spread, results indicated that declines, extinctions, and epidemics occurred years after the leading edge of the wave had passed. This has likely prevented others from recognizing the invasion pattern, such as the logistic regressions using the entire dataset based on geographic coordinates performed by Pounds et al. [

Geography may influence gene flow in plants and animals [

Our estimates of

Our analyses and re-analyses of data related to the CLEH all fail to support that hypothesis. First, our simulations indicate that the correlations reported in [

Second, we show the elevational pattern in

Third, it was suggested [

Although we propose at least three independent introductions of

Our analyses support a hypothesis that

That

Disease dynamics are the result of a complex process involving multiple factors related to the hosts, the pathogen, and the environment that may affect disease patterns on many spatial scales. Disease dynamics are affected by micro- and macro-climatic variables [

A straightforward example of how climate change might drive amphibian extinctions from invasion of

Future studies need to first determine whether or not

(32 KB DOC)

This research was stimulated by an invitation by S. Collinge and C. Ray to participate in the Symposium on Climate Change and Disease at the 2007 Ecological Society of America annual meeting. Thanks to S. Stuart, J. Chanson, V. Katariya at GAA; B. Young, A. Pounds, S. Lotters, L. Coloma, S. Ron (RANA); C. Garrard (Utah State University) for assistance with Geographic Information System (GIS) mapping; and to F. Brem, J. Savage, and the Los Angeles County Museum for histological examination of Monteverde amphibians. Thanks to R. Speare, H. McCallum, L. Coloma, S. Ron, M. Bustamante, J. Longcore, F. Brem, A. Kelly, K. Zippel, R. Gagliardo, and D. Fenolio for discussions of the study or reviews of earlier drafts.

air temperature

climate-linked epidemic hypothesis

date of decline

Global Amphibian Assessment

last year observed