A Qualitative Meta-Analysis Reveals Consistent Effects of Atrazine on Freshwater Fish and Amphibians

Objective The biological effects of the herbicide atrazine on freshwater vertebrates are highly controversial. In an effort to resolve the controversy, we conducted a qualitative meta-analysis on the effects of ecologically relevant atrazine concentrations on amphibian and fish survival, behavior, metamorphic traits, infections, and immune, endocrine, and reproductive systems. Data sources We used published, peer-reviewed research and applied strict quality criteria for inclusion of studies in the meta-analysis. Data synthesis We found little evidence that atrazine consistently caused direct mortality of fish or amphibians, but we found evidence that it can have indirect and sublethal effects. The relationship between atrazine concentration and timing of amphibian metamorphosis was regularly nonmonotonic, indicating that atrazine can both accelerate and delay metamorphosis. Atrazine reduced size at or near metamorphosis in 15 of 17 studies and 14 of 14 species. Atrazine elevated amphibian and fish activity in 12 of 13 studies, reduced antipredator behaviors in 6 of 7 studies, and reduced olfactory abilities for fish but not for amphibians. Atrazine was associated with a reduction in 33 of 43 immune function end points and with an increase in 13 of 16 infection end points. Atrazine altered at least one aspect of gonadal morphology in 7 of 10 studies and consistently affected gonadal function, altering spermatogenesis in 2 of 2 studies and sex hormone concentrations in 6 of 7 studies. Atrazine did not affect vitellogenin in 5 studies and increased aromatase in only 1 of 6 studies. Effects of atrazine on fish and amphibian reproductive success, sex ratios, gene frequencies, populations, and communities remain uncertain. Conclusions Although there is much left to learn about the effects of atrazine, we identified several consistent effects of atrazine that must be weighed against any of its benefits and the costs and benefits of alternatives to atrazine use.

The herbicide atrazine (2chloro4ethylamino 6isopropylaminostriazine) is the second most commonly used pesticide in the United States (Kiely et al. 2004) and perhaps the world (Solomon et al. 1996;van Dijk and Guicherit 1999). It is a photosynthesis inhibi tor used to control certain annual broad leaf weeds, predominantly in corn but also in sorghum, sugarcane, and other crops and landscaping. The environmental risk posed by atrazine to aquatic systems is presently being reevaluated by the U.S. Environmental Protection Agency (U.S. EPA 2003(U.S. EPA , 2007. One of the challenges in evaluating the safety of atrazine has been that its biological effects are highly controver sial, and much of the debate in the literature has been targeted at its effects on freshwater vertebrates (Hayes 2004;Renner 2004).
There have been four reviews on the biologi cal effects of atrazine, all of which were funded by the corporation that produced or produces this chemical (Giddings et al. 2005;Huber 1993;Solomon et al. 1996Solomon et al. , 2008. However, none of the past reviews used a metaanalytical approach to identify generalities in responses to atrazine exposure. Metaanalysis, as para phrased from the U.S. EPA, is the systematic analysis of studies examining similar end points to draw general conclusions, develop support for hypotheses, and/or produce an estimate of overall effects (U.S. EPA 2009a). This sort of weightofevidence approach would provide directional hypothe ses for future work on atra zine. Furthermore, it would offer invaluable information to regulatory agencies on general and expected impacts of atrazine on freshwater vertebrates that might help resolve much of the controversy surrounding atrazine. Given the lack of a metaanalytical assessment and the potential importance of any atrazine effects, we set out to conduct an objective, qualita tive metaanalysis on the effects of atrazine on amphibian and fish survival, behavior, meta morphic traits, and immune, endocrine, and reproductive systems.

Atrazine Persistence, Transport, and Exposure
To place the results of this metaanalysis within an ecologic context and to evaluate the relevance of studied atrazine concentrations and exposure regimes, we briefly discuss the fate, transport, and field concentrations of atra zine. Atrazine is persistent relative to most cur rentuse pesticides. CibaGiegy Corporation (1994), the company that previously produced atrazine, reported no detectable change in atra zine concentration after 30 days in hydrolysis studies conducted at pHs between 5 and 7, and an aqueous photolysis halflife of 335 days under natural light and a neutral pH. Halflives from field and mesocosm studies are variable because degradation can depend on various environmental conditions. Nevertheless, sev eral field and mesocosm studies report half lives > 3 months (e.g., de Noyelles et al. 1989;Klaassen and Kadoum 1979).
Atrazine is also relatively mobile-regularly entering water bodies through runoff-and concentrations in surface waters often peak after rains. Several researchers have suggested that atrazine can be transported 1,000 km aeri ally (van Dijk and Guicherit 1999). Indeed, atrazine has been found regularly in surface waters and precipitation great distances from where it is used, such as above the Arctic Circle, albeit at low concentrations (van Dijk and Guicherit 1999).
Wet deposition of atrazine might also be important in some areas. In a review on atmos pheric dispersion of currentuse pesti cides, van Dijk and Guicherit (1999) reported more studies detecting atrazine in rain or air (from European and U.S. sites) than any other currentuse pesticide. The maximum reported wet deposition of atrazine is 154 µg/L from Iowa precipitation (Hatfield et al. 1996). Wet deposition > 1 µg/L was reported regularly in North America and Europe between 1980 and the early 1990s (reviewed by van Dijk and Guicherit 1999). As a reference point, the maximum contaminant level for drinking water set by the U.S. EPA is 3 µg/L atrazine (U.S. EPA 2002).
Surface water is likely the primary source of atrazine exposure for freshwater vertebrates. Data on atrazine concentrations in surface water, however, are more abundant for lotic (streams and rivers) than lentic (lakes, ponds, wetlands, ditches) systems (Solomon et al. 2008), primarily because of the extensive stream monitoring conducted by the U.S. Geological Survey National Water Quality Assessment project and Syngenta Crop Protection, Inc. (U.S. EPA 2007). In lentic systems, water is not replenished as it is in lotic systems, and chemicals can concentrate as lentic systems dry. Maximum reported concentrations in lentic sys tems are often 2.5-10 times higher than maxi mum concentrations in lotic systems (Baker and Laflen 1979;Edwards et al. 1997;Evans and Duseja 1973;Frank et al. 1990; Kadoum and Mock 1978;Kolpin et al. 1997). Additionally, many amphibians develop in ephemeral agri cultural ponds that might receive and concen trate atrazine (Knutson et al. 2004).
Given the limited data on atrazine concen trations in lentic systems, the expected (or esti mated) environmental concentration (EEC) is a reasonable alternative for estimating concen trations to which aquatic organisms are likely to be exposed. GENEEC2 software (U.S. EPA 2009b) calculates standardized EECs used by the U.S. EPA for Tier1 chemical risk screen ing. EECs are important because chemical registration decisions entail comparing lowest observable effect concentrations (LOECs) to EECs to determine whether higherlevel mod eling is warranted. Hence, effects of a chemical near or below the EEC can affect the decision to approve its use.
For present atrazine application rates, EECs based on GENEEC2 software are typically near 100 µg/L but can be higher for some crops. However, the recommended application rates (~ 2 lb active ingredient/acre) are now two to four times less than they were in the early 1990s (~ 8 lb active ingredient/acre). Hence, at the time of atrazine regis tration, LOECs near or below 500 µg/L, a feasible EEC at the time, might have triggered Tier2 testing and might have raised concerns about the safety of atrazine that could have compromised its registration. Given both past and presentday conditions, the lack of thorough data on atrazine concen trations in lentic systems, and the common use of agricultural ponds, ditches, and wet lands by amphibians and fish, we suggest that concentrations near or below historical EECs (≤ 500 µg/L) are ecologically relevant when considering the findings of this metaanalysis. This is arguably conservative given that atra zine concentrations > 500 µg/L have been regu larly recorded in agricultural ponds and ditches (Baker and Laflen 1979;Edwards et al. 1997;Evans and Duseja 1973;Frank et al. 1990;Kadoum and Mock 1978;Kolpin et al. 1997).

Methods
We selected studies for this metaanalysis beginning with those cited by Solomon et al. (2008), the most recent review of atrazine effects on amphibians and fish. We then sup plemented these studies by searching Web of Science (Thomson Reuters, New York, NY) to identify studies that might have been missed by Solomon et al. (2008). The search terms were "atrazine" combined with either "amphibian*" or "fish*".
Selection criteria for inclusion of studies in metaanalyses can affect the conclusions that are drawn (Englund et al. 1999). Hence, we excluded from this metaanalysis studies that had substantial contamination in control treat ments or reference sites (unless a regression approach was taken to analyze the data); no presentation of statistics and withingroup vari ance estimates; considerable inconsistencies that could affect the biological conclusions; spatial confounders associated with atrazine treatments; pseudoreplication; or other consid erable flaws in experimental design. We eval uated whether the exclusion of these studies changed the conclusion of the metaanalysis for each end point (Englund et al. 1999). For the 15 response variables, the inclusion of studies that did not meet our criteria never altered the conclusions of our metaanalyses, and in some cases including these studies actually strength ened the conclusions. Because of this and space limitations, studies that were excluded and why, as well as the directions of effects in these studies, are provided in Supplemental Material available online (doi:10.1289/ehp.0901164.S1 via http://dx.doi.org/).
To conduct a qualitative metaanalysis, we chose to use the votecounting method-in which we tallied the number of studies that did and did not detect effects of atrazinefor several reasons. We quantified the effects of atrazine on 15 response variables from > 125 studies, and vote counting, the simplest approach to metaanalyses, made it feasible to manage this complexity. Vote counting also facilitates identifying response variables that might warrant more sophisticated meta analyses based on effect sizes. Finally, we chose vote counting because it is a conservative approach, biasing results toward detecting no overall effect (Gurevitch and Hedges 1993). Because most atrazine studies conducted analysis of variance to test for dose responses, despite regression analyses providing much greater statistical power (Cottingham et al. 2005), we include studies that had substan tial trends for effects of atrazine (i.e., a non significant increase or decrease) with studies that reported statistically significant effects (α = 0.05). Our criteria for a trend were a clear dose response, a probability value < 0.1, or authors interpreting their nonsignificant result as a trend. Never did including trends change our conclusions of the metaanalysis.

Results and Discussion
Effects of atrazine on fish and amphibian survival. Many researchers have evaluated the effects of atrazine on fish (reviewed by Giddings et al. 2005;Huber 1993;Solomon et al. 1996) and amphibian survival (e.g., Karasov 2000, 2001;Brodeur et al. 2009;Diana et al. 2000;Rohr et al. 2003Rohr et al. , 2004Rohr et al. , 2006b. Our general conclusions from these studies are consistent with the conclusions of authors from previous atrazine reviews (Giddings et al. 2005;Huber 1993;Solomon et al. 1996Solomon et al. , 2008: There is not consistent, published evidence that ecologically relevant concentrations of atrazine are directly toxic to fish or amphibians. There are, however, some important exceptions (e.g., Alvarez and Fuiman 2005;Rohr et al. 2006bRohr et al. , 2008cStorrs and Kiesecker 2004). Because our conclusions are consistent with previous reviews, we did not conduct a metaanalysis on survival.
Effects of atrazine on fish and amphib ian development and growth. Background on metamorphosis. A basic understanding of four concepts about amphibian metamorpho sis is necessary to interpret the effects of any chemical on time to, or size at, metamorpho sis. First, amphibians must reach a minimum size before they can metamorphose (Wilbur and Collins 1973). Second, once they reach this size, they can accelerate development and metamorphose earlier if they are in a stressful environment or metamorphose later if they are in a good environment (Wilbur and Collins 1973). Last, metamorphosis is predominantly controlled by corticosterone and thyroid hor mones (Larson et al. 1998); thus endocrine system disruption can lead to inappropriately timed metamorphosis.
These important facts have profound implications for understanding the effects of pollution on metamorphic traits. For example, imagine that an amphibian shunts energy away from growth to detoxify a chemi cal and, as a result, reaches the minimum size for meta morphosis 5 days later than amphibians not exposed to the chemical. Once this amphibian reaches the minimum size for metamorphosis, it might accelerate its develop mental rate and metamorphose 5 days earlier to get out of the stressful chemi cal environment. In this exam ple, there is no net effect of the chemical on time to metamorphosis despite inarguably hav ing considerable effects on energy use, growth, and development (Larson et al. 1998). A single chemical could delay, accelerate, or have no effect on timing of metamorphosis, depending on chemical type and concentration.
This example highlights four points. First, a lack of an effect of a chemical on timing of metamorphosis does not mean there was no effect on developmental rate or hormones that drive metamorphosis, as concluded by Solomon et al. (2008). Second, non monotonic dose responses in the timing of metamorphosis are expected and are likely common. This is because there are several processes occurring (detoxification, growth, and modulation of developmental timing) that can be temporally volume 118 | number 1 | January 2010 • Environmental Health Perspectives offset and that likely have different (and potentially opposite) functional responses to the same chemical. Third, timing of meta morphosis in response to chemicals should be highly variable. This variation should not be interpreted as inconsistencies across stud ies (e.g., Solomon et al. 2008), because the complexity of meta morphosis is expected to induce extreme variability. Finally, unlike timing of metamorphosis, size at metamor phosis is expected to monotonically decrease with increasing chemi cal concentration across species and studies (controlling for time to metamorphosis) because energy used for detoxification is often taken away from that used for growth and development.
Effects on metamorphic traits. Our quali tative metaanalysis on the effects of atrazine on metamorphic traits is consistent with the predictions described above. Twelve of 21 studies found significant effects of atrazine on metamorphic timing, with 7 showing an increase and 7 showing a decrease in time to metamorphosis; thus, as predicted, the direc tion of the effect was not consistent across studies ( Table 1). Seven of the 21 studies had either clear non monotonic dose responses or were possibly non monotonic (Table 1). These results are consistent with the high variabil ity and high probability of non monotonicity expected for this end point.
Only two studies explicitly quantified the effects of atrazine on both thyroid hormones and timing of metamorphosis, and both showed significant nonmonotonic effects Larson et al. 1998) (Table 1). Further, Larson et al. (1998) revealed delays in growth and develop ment early in life followed by accelerated develop ment and early metamor phosis once a critical size for metamorphosis was reached. Additional studies that quantify the impacts of atrazine on thyroid hormones, cortico steroid hormones, and changes in growth and development through time are needed.
In contrast to timing of metamorphosis, size at metamorphosis shows a clear dose dependent response to atrazine exposure ( Table 1). Fifteen of 17 studies and 14 of 14 species showed significant reductions, or considerable trends toward reductions, in amphibian size at metamorphosis associated with atrazine exposure, and all of these studies reported effects at ecologically relevant concen trations based on the above criteria (Table 1). Similar growth reductions have been observed in fish (Alvarez and Fuiman 2005;McCarthy and Fuiman 2008). Atrazine consistently reduced amphibian size, which is likely to have adverse effects on amphibian populations because smaller metamorphs generally have lower terrestrial survival, lower lifetime repro duction, and compromised immune function (Carey et al. 1999;Scott 1994;Smith 1987). However, populationlevel effects of atrazine have not been empirically tested for in nature and thus need to be evaluated explicitly.

Effects of atrazine on fish and amphib ian behavior. Effects on locomotor activity.
Twelve of 13 studies reported that atrazine exposure increased amphibian or fish loco motor activity over at least a portion of the concentration gradient tested (Table 2). Interestingly, 4 of 5 studies on fish, but none of the studies on amphibians, reported non monotonic dose responses. For fish, low con centrations of atrazine stimulated hyper activity, but higher concentrations caused reductions in activity. For amphibians, hyperactivity was typically observed at the concentrations tested, but higher concentrations would likely even tually become toxic and reduce activity. All studies conducted on fish detected effects of atrazine on locomotor activity, whereas 88% of the studies on amphibians detected atrazine effects ( Table 2).
The effects of atrazine on amphibian and fish locomotor activity are consistent with atrazineinduced changes in locomotor activity in mammals. Atrazine seems to cause hyper activity in mammals by competing with receptors for the inhibitory neuro transmitter gamma aminobutyric acid, by altering monoamine turn over, and through neuro toxicity of the dopa minergic system (Das et al. 2001;Rodriguez et al. 2005). One study showed that atrazine has similar effects on the nervous system of Ranid frogs (Papaefthimiou et al. 2003), but additional studies are needed that evaluate the mechanisms responsible for atrazineinduced activity changes in fish and amphibians.
Effects on antipredator behaviors. Six of 7 studies reported that atrazine decreased amphibian and fish behaviors associated with predationrelated risk reduction (Table 2). Reduced predation avoidance behaviors can increase predation risk, whereas increased hyperactivity should increase encounter rates with predators (Skelly 1994). Hence, reduced riskreduction behaviors coupled with hyper activity are expected to increase predation. However, there are no published studies on the effects of atrazine on predator-prey rela tionships of which we are aware. Given that atrazine might have effects on both predators and prey, the effects of atrazine on predatorprey interactions are difficult to predict with out additional studies.
Effects on olfaction. Five of 5 studies reported that atrazine exposure reduced olfac tory sensitivity of fish in a dosedependent manner (Table 2). In contrast, 3 of 3 studies on amphibians detected no effects of atrazine on olfaction at much higher concentrations than were tested on fish (Table 2). One study on amphibians stained activated olfactory neurons with agmatine and found no difference in the stimu la tion of olfactory neurons between atra zinetreated and control animals (Lanzel 2008).
Effects on other behaviors. One study showed that atrazine reduced amphibian waterconserving behaviors, which increased their rate of water loss (Rohr and Palmer 2005) (Table 2). Interestingly, both the hyper activity and the reduced waterconserving behaviors occurred hundreds of days after atrazine expo sure had ceased; there was no evidence that these end points recovered from atrazine expo sure, suggesting permanent effects (Rohr and Palmer 2005). Amphibians are extremely sus ceptible to desic ca tion; thus atrazineinduced changes in water conserving behaviors would be expected to increase mortality risk.
Effects of atrazine on fish and amphibian immunity and infections. Effects on immunity. Our qualitative metaanalysis revealed that atrazine exposure consistently reduced immune functioning of fish and amphibians, with 16 of 18 studies finding effects at ecologi cally relevant concentrations. However, many of the end points (16 of 39) were from studies where atrazine was tested as part of a mixture of pesticides, and thus the effects of atrazine were not isolated (Table 3). Nevertheless, atra zine exposure-alone (21 of 27 end points) or in a pesticide mixture (12 of 16 end points)was associated with reduced immune function ing, resulting in an overall reduction in 77% (33 of 43) of the quantified fish and amphib ian immune end points (including trends for a decrease) (Table 3). These results are some what conservative because in one study mul tiple genes associated with immunity were significantly downregulated (Langerveld et al. 2009), but they were counted as a single end point (Table 3).
Effects on infections. Similar to the effects of atrazine on amphibian and fish immunity, atrazine exposure was consistently associated with an increase in infection end points in fish and amphibians at ecologically relevant con centrations (Table 4). Atrazine elevated trema tode, nematode, viral, and bacterial infections (Table 4). Of the studies with sufficient statis tical power and without obvious confounders, 12 of 14 of the infection end points increased or showed a strong trend toward increasing, indicating either more infected individuals, more infections per individual, faster matura tion, or greater reproduction of the parasite within the host, or greater parasiteinduced host mortality (Table 4). As with immunity, these patterns should be considered with cau tion because many of these end points (6 of 16) came from studies where atrazine was part of a mixture of pesticides tested. Nevertheless, atrazine exposure, alone (4 of 7 end points) or in a pesticide mixture or field study (9 of 9 end points), was associated with an increase in infection end points (Table 4). In general, high concentrations of atrazine seem to be directly toxic to trematodes and viruses, pos sibly reducing infection risk for amphibians (Forson and Storfer 2006a;Koprivnikar et al. 2006;Rohr et al. 2008b), whereas more eco logically common concentrations seem to increase amphibian susceptibility, elevating infection risk (Forson and Storfer 2006b;Gendron et al. 2003;Kiesecker 2002;Rohr et al. 2008c).
Several atrazine studies collected immuno logic data only from animals that were also exposed to parasites, thus confounding immune parameters with parasite exposure and loads (Christin et al. 2003;Forson and Storfer 2006b;Gendron et al. 2003;Hayes et al. 2006;Kiesecker 2002;Rohr et al. 2008c). However, in each of these studies, atrazine was associated ). a Aatrex is 59.2% inactive ingredients. b Community-level study. c Authors show that atrazine modifies the thyroid axis for both X. laevis and B. americanus. d All five atrazine concentrations tested reduced frog size relative to controls, but no within-group variance estimates were provided. e 200 ppb developed faster than 2,000 ppb. f Only a single egg mass; might not reflect general response. g Use only 50% of the metamorphs in the time to metamorphosis analysis without describing how they selected this subset of metamorphs or why they used only 50% for time to metamorphosis but 100% of the metamorphs for size at metamorphosis. h Authors report an interaction between atrazine and time for frog length, indicating that control animals were larger than those exposed to atrazine by the end of the experiment. i Tested as a mixture of 5 µ/L atrazine and 5 µ/L carbaryl. j Compared ponds with and without atrazine; effects might be due to other factors. k Frogs lose weight at metamorphosis, thus mass measurements were confounded by grouping tadpole and metamorph weights. l Provide no within-group variance estimate. m No statistics provided but conclude that there was no effect of atrazine. n Graphs for developmental rate through time are indiscernible. o Detected effects in only one of two experiments and for females only. p p = 0.080 for regression analysis, one-tailed test. q Results depended on developmental stage; authors showed that atrazine modifies thyroxine and corticosterone hormones. r Results depended on drying conditions. volume 118 | number 1 | January 2010 • Environmental Health Perspectives with both reduced immune parameters and elevated parasite loads. The elevated infections associated with atrazine cannot be explained by parasites reducing immune responses. Hence, the parsimonious explanation for both of these findings is that atrazine reduced immune responses, which elevated infections, especially given that it is often bene ficial for vertebrates to upregulate immunity upon infection (Raffel et al. 2006). Despite the apparent consistency in the effects of atrazine on immunity and infections . a Community-level study. b Larval red drum are often found in freshwater, so they were included in this meta-analysis. c Mixture of 0.5:0.5 and 1.0:1.0 atrazine and simazine; thus, total concentration of triazine was 1 and 2 ppb, respectively. d Increased salamander water loss and thus desiccation risk.  (Table 3), much remains to be learned about the effects of atrazine and other chemicals on parasite-host inter actions (Raffel et al. 2008;Rohr et al. 2006a). For instance, we know lit tle about how atrazineinduced changes affect population or community dynamics or most human diseases. Effects of atrazine on fish and amphibian gonadal morphology. General morphologic end points. Sex differentiation is the process by which gonads develop into either testes or ovaries from an undifferentiated or bipoten tial gonad (Hayes 1998). This process is dis tinct from reproductive maturation where the differentiated gonad becomes reproductively functional (e.g., undergoes spermato genesis in males). Determining if atrazine induces changes in gonadal morphology is an impor tant step in evaluating whether it can influence sexual differentiation.
Atrazine consistently affected male gonadal morphology in fish and amphibians ( Table 5). Seven of the 10 studies including results on males and females reported strong trends or statistically significant altera tions (6 stud ies) in at least one aspect of general gonadal morphology associated with atrazine expo sure. Alterations included discontinuous and multiple testes, sexually ambiguous gonadal tis sue, testicular ovarian follicles (TOFs), altered gonadal somatic index (GSI; ratio of gonad weight to body weight), expanded testicular lobules, and spermatogenic tubule diameter (Table 5).
Effects on ovarian morphology are gener ally less obvious than those on testicular mor phology and are typically dismissed without quantification. None of the three studies on fish or amphibians included in our meta analysis found significant effects of atrazine on ovarian morphology, suggesting that atrazine induces fewer gonadal abnormalities in females than males. However, additional studies are necessary to fully evaluate the effects of atrazine on female gonadal morphology.
TOFs as a natu ral phenomenon. Jooste et al. (2005) and Solomon et al. (2008) argued that experiments with high numbers of TOFs in control Xenopus laevis support the hypoth esis that TOFs are normal in some X. laevis populations. Although it was argued long ago that some anurans in some environments tran sition through a hermaphroditic phase during development (Witschi 1929), the literature we reviewed does not argue that adult amphib ians commonly have oocytes within testicular tissue or are naturally hermaphroditic (Eggert 2004;Hayes 1998). Indeed, X. laevis sexually differentiates (without a transitional/hermaph roditic stage) during the larval period prior to sexual maturation (Iwasawa and Yamaguchi 1984). Thus, cases of gonadal abnormalities in healthy adult X. laevis populations should be rare. Given that simultaneous hermaphroditism has not been previously reported in X. laevis despite decades of research on their reproduc tive biology, an equally or more plausible expla nation for high numbers of TOFs in control animals (e.g., Jooste et al. 2005;Orton et al. 2006) is exposure to some type of unmeas ured endocrinedisrupting contaminant.
Effects of atrazine on fish and amphibian sex ratios. Given that atrazine exposure has been proposed to feminize gonadal develop ment (Hayes et al. 2002(Hayes et al. , 2003, it might lead to femalebiased sex ratios. Many studies, how ever, have severe methodo logic errors, such as contaminated controls or inadequate data reporting [see Supplemental Material, Table S1 (doi:10.1289/ehp.0901164.S1)], preventing a conclusive synthesis of the effects of atrazine on sex ratios. None of the sexratio studies used the most accepted and powerful approaches for testing for changes in sex ratios (e.g., h Effects could be due to inactive ingredients. i Effects could be due to chemicals other than atrazine that might be in the pond water used to make the stock solutions. j All LC 50 s were calculated incorrectly. Wilson and Hardy 2002). Only four studies, all on X. laevis, were of sufficient quality to be included in our metaanalysis, and only one found that atrazine induced a femalebiased sex ratio (see Supplemental Material, Table S2 (doi:10.1289/ehp.0901164.S1)].

Effects of atrazine on fish and amphibian gonadal function.
Chemicals that alter gonadal development can affect gonadal function, such as germ cell (e.g., spermatogenesis in males) and steroid hormone production (McCoy et al. 2008;McCoy and Guillette, in press), and thus can lead to altered reproductive success.
Effects on testicular cell types. Spermato genesis is the process through which mature male gametes (spermatozoa) are produced from precursor cells (spermatogenic cells). The relative ratios of different spermatogenic cell types, rather than abundance of spermatozoa alone, is the most sensitive metric of altered spermato genesis. Unfortunately, few studies on effects of atrazine on spermatogenesis met our inclusion criteria. Two of two studies demon strated that atrazine was associated with altered spermatogenesis and that several cell types were affected (Table 6). Thus, atrazine appears capa ble of altering spermatogenesis, but the contexts and generality of these effects cannot be firmly established. Our analysis once again highlights a need for more rigorous investigations.
Effects on sex hormone concentrations. Sex hormone production is an important function of gonads that can be altered by gonadal abnormalities (McCoy et al. 2008).
Indeed, altered hormone concentrations are the defining characteristic, in many cases, of endocrine disruption. Six of seven studies on fish and amphibians document strong trends or significantly (five studies) altered sex hor mone concentrations associated with atrazine exposure (Table 6). Although many of these studies were conducted in the field and are therefore correlative, the consistency of these results across studies suggests that atrazine alters sex hormone production and should be considered an endocrinedisrupting chemical. A more thorough understanding of the effects of atrazine on hormone concentrations will require more detailed studies that account for the inherent variability of endocrine system processes.  Table S1 (doi:10.1289/ehp.0901164.S1). a No test statistics or degrees of freedom are presented; however, means and variances were presented either in the text or in a figure of the article. b Xenopus are typically sexually differentiated at the gross morphologic level at metamorphosis; individuals in this study exposed to 25 µg/L were so sexually ambiguous they were initially considered intersex (having both testicular and ovarian issues). c Atrazine concentration for the nonagricultural reference site during 2003 was reported incorrectly; repeated attempts to contact the authors for clarification have not been forthcoming. d When atrazine concentrations were highest (2003), TOFs per individual occurred in higher numbers; TOFs were positively associated with atrazine, nitrate, and quantity of pesticides in a multivariate comparison, suggesting that atrazine is contributing to TOFs. e Concentrations were between ND and 2 except on two occasions at one site, when levels were 65 and 250 µg/L. f Authors argued that differences in GSI between agricultural and nonagricultural sites cannot be due to atrazine because GSI does not correlate with atrazine concentration; however, they presented no statistics to support this claim. g The relationship between detection of atrazine and the presence of one or more intersex cricket frogs approached significance (p = 0.07). h The actual concentration of the 30-µg/L treatment was 125 µg/L. volume 118 | number 1 | January 2010 • Environmental Health Perspectives Effects on reproductive success. Repro ductive success is strongly linked to population persistence and is likely one of the most impor tant end points in toxicologic studies. Five studies that evaluated the effects of atrazine on measures of reproductive success met our meta analysis requirements (Table 6). Two studies on adult fish, Pimephales promelas, found no signifi cant effect of atrazine on number of eggs produced, fertilization success, proportion of hatchlings, or larval development. However, one of these studies (Bringolf et al. 2004) found several non significant, adverse trends ( Table 6). Two of three studies on amphibians found no effects of atrazine on hatching success, whereas one showed reduced hatching success and delayed hatching (Table 6). Given the mixed results, the effect of atrazine on reproductive success needs to be studied more thoroughly.

Effects of atrazine on fish and amphibian vitellogenin.
Vitellogenin is an egg yolk precursor protein produced in the livers of female fish and amphibians. Estrogens induce vitello genin synthesis in both males and Hence, these data do not support the hypothe sis that atrazine is strongly estrogenic to fish. Effects of atrazine on fish and amphibian aromatase. Cytochrome p450 aromatase cata lyzes the conversion of androgens to estrogens in gonads and is critical for maintaining a bal ance between these sex hormone classes. Hayes et al. (2002) hypothesized that decreases in testosterone associated with atrazine exposure in their study could be driven by an atrazine induced increase in aromatase and a concomi tant increase in the conversion of testosterone and other androgens to estrogens. This hypoth esis seemed reasonable because atrazine was known to increase aromatase in human can cer cell lines and in alligator gonadal-adrenal mesonephros (Crain et al. 1997;Sanderson et al. 2000). However, since 2002, several studies have explicitly tested whether atrazine increases aromatase in fish and amphibians, and only one of six studies included in our metaanalysis found that atrazine was associ ated with increased aromatase gene expres sion [see Supplemental Material, Table S2 (doi:10.1289/ehp.0901164.S1)].
Effects of atrazine on fish and amphibian populations and communities. Although there are too few studies examining the effects of atrazine on freshwater vertebrate populations to warrant metaanalysis, and virtually all communitylevel studies infer-rather than test for-indirect effects (Rohr and Crumrine 2005), the effects of atrazine on populations and communities warrants a brief discussion. Any chemical that affects physiology, growth, develop ment, reproduction, survival, or species inter actions can affect population and commu nity dynamics (Clements and Rohr 2009;Rohr et al. 2006a). However, the effects of contami nants might not result in immediate popula tion declines because the survivors of chemical exposure frequently have less competition for resources, thus providing densitymediated compensation for adverse effects of the chemi cal (Rohr et al. 2006b). Demonstrating that a factor is the cause of any population decline is, indeed, incredibly difficult (Rohr et al. 2008a). Rohr et al. (2006b) revealed significant and delayed declines in Ambystoma barbouri sala mander populations at 4, 40, and 400 µg/L atrazine, above and beyond the counter acting effects of densitymediated compensation. Although this study provided greater ecologic realism than many studies on atrazine, cau tion should be taken extrapolating these effects to populations in nature because this study was conducted in laboratory terraria. There is certainly a need for controlled studies on the effects of pesticides on wildlife populations.
Several studies have examined the effects of atrazine on amphibian and fish commu nities (Boone and James 2003;de Noyelles et al. 1989;Kettle 1982;Rohr and Crumrine 2005;Rohr et al. 2008c). Many of these stud ies reported alterations in fish or amphibian growth and abundance that seem to be caused by atrazineinduced changes in photo synthetic organisms (reviewed by Giddings et al. 2005;Solomon et al. 2008). At ecologically relevant concentrations, atrazine is expected to have a bevy of indirect effects by altering the abun dance of periphyton, phytoplankton, and macrophytes (Huber 1993;Solomon et al. 1996). However, none of these studies dis tinguish between direct and indirect effects of atrazine on fish or amphibians.
There are several field studies comparing amphibian populations or species richness between atrazineexposed and unexposed hab itats (Bonin et al. 1997;Du Preez et al. 2005;Knutson et al. 2004). All of these studies are correlational, and none thoroughly consid ered or ruled out alternative hypotheses for the observed patterns.
Caveats. We would be remiss not to men tion some caveats regarding this metaanalysis. First, a problem with many metaanalyses is the "filedrawer" effect. This refers to the fact that researchers tend to place the results of experiments showing no effects in their file drawer, and many journals tend to publish fewer studies showing no effects than those with effects (Gurevitch and Hedges 1993;Osenberg et al. 1999). This might be less of a problem in studies on pesticides because these chemicals are designed to kill biota; thus in many cases, the null hypothesis might be an effect rather than the absence of one. Additionally, a substantial industry contingent works to ensure that both significant and non significant effects of chemicals get published. Indeed, in the review of atrazine by Solomon et al. (2008), there were approximately No test statistics or degrees of freedom were presented; however, means and variances were presented either in the text or in a figure of the article. c Authors reported no significant correlation between atrazine and sex hormones in their abstract when, in fact, these end points were negatively correlated; contrary to the authors' conclusion, the negative correlations across sexes and age groups reported in their study are unlikely to occur because of a low sample size or sampling error. d Authors argued that differences in hormone levels between agricultural and nonagricultural sites cannot be due to atrazine because hormone concentrations do not correlate with atrazine concentration; however, they presented no statistics to support this claim. e Low samples sizes (7-8 fish) likely precluded detecting these considerable effects.
volume 118 | number 1 | January 2010 • Environmental Health Perspectives 63 cases where atrazine had significant adverse effects and 70 cases where atrazine had no sig nificant effects (Rohr JR, McCoy KA, unpub lished data), suggesting that the filedrawer effect is unlikely to be strongly biasing submis sion and publication of nonsignificant atrazine results. However, we cannot completely dis count the possibility that the filedrawer effect generated a bias toward greater publication of significant effects of atrazine. Another admonishment is that some of the end points in this metaanalysis were not independent of one another. For example, we tallied multiple end points from a single study despite the possibility that they might not be entirely independent.
Finally, we must consider the findings of this metaanalysis on atrazine relative to alter native strategies for weed control. If the alter native to atrazine is another chemical, then we should ideally compare the effects of atrazine to the replacement chemical. In fact, atrazine might be less detrimental to fresh water ver tebrates than a replacement herbicide. If the alternative to atrazine does not entail a chemi cal replacement, then the effects revealed here might indeed be disconcerting. However, we also cannot ignore the benefit, if any, that atrazine provides. Interestingly, several studies estimate that atrazine increases corn yields by only 1-3% (reviewed by Ackerman 2007). To adequately evaluate any chemical, we should ideally conduct a thorough costbenefit analysis that considers the focal chemi cal and alternatives to its use and is based on comprehensive and accurate knowledge [see Ackerman (2007) for a review and critique of atrazine cost-benefit analyses].

Conclusions
As in past reviews, we found little evidence that atrazine consistently causes direct mortality of freshwater vertebrates at ecologically relevant concentrations, but there is evidence that atra zine might have adverse indirect ecologic effects. However, in contrast to a previous review on atrazine (Solomon et al. 2008), we unveiled consistent effects of atrazine at ecologically rel evant concentrations for many other response variables in our metaanalysis. The discrepancy between our findings and the conclusions of previous reviews could be partly a function of differences in criteria for including studies in the group used to draw general conclusions about atrazine effects. Past reviews (e.g., Solomon et al. 2008) did not clearly define their inclu sion criteria, did not make it clear which studies affected their conclusions (or how they came to their conclusions), and regularly dismissed significant effects of atrazine.
Here we reveal that, for fresh water verte brates, atrazine consistently reduced growth rates, had variable effects on timing of meta morphosis that were often nonmonotonic, elevated locomotor activity, and reduced anti predator behaviors. Amphibian and fish immunity was reliably reduced by ecologi cally relevant concentrations of atrazine, and this was regularly accompanied by elevated infections. Atrazine exposure induced diverse morphologic gonadal abnormalities in fish and amphibians and was associated with altered gonadal function, such as modified sex hor mone production. This suggests that atrazine should be considered an endocrinedisrupting chemical. Finally, we do not have a thorough appreciation of the reproductive repercussions of atrazine.
Several end points had enough well conducted studies to warrant more sophisti cated metaanalyses based on effect sizes (e.g., growth, timing of metamorphosis, activity, immunity, infections, gonadal abnormalities). Metaanalyses based on effect sizes can pro vide parameter and standard errors estimates and thus can be useful for probabilistic risk assessment and for predicting atrazine effects.
Although we found consistent effects of atrazine on freshwater vertebrates, the con sequences of these effects remain uncertain. We know little about how atrazineinduced changes in vertebrate growth, somatic develop ment, behavior, immunity, gonadal develop ment, or physiology affect reproduction, populations, gene frequencies, or communities. However, it was Sir Austin Bradford Hill who wisely stated in his address to the Royal Society of Medicine in 1965 that All scientific work is incomplete [and] . . . liable to be upset or modified by advancing knowledge. That does not confer upon us freedom to ignore the knowledge we already have, or to postpone action that it appears to demand at a given time. (Hill 1965) Whatever action is taken in the reevaluation of atrazine by the U.S EPA, we strongly encour age regulators to consider the consistent effects of atrazine on various taxa and to weigh these effects against any benefits atrazine provides and the alternatives to atrazine use.

correction
Corrections have been made from the origi nal manuscript published online: Criteria for identifying results showing "substantial trends" has been clarified; the number of studies has been corrected in the text; and the "effect direction" for relevant studies has been corrected in Tables 1, 3, and 5.