Co-production of contaminated landscapes: anthropogenic loading and food web structure drive mercury bioaccumulation in abandoned gold mines

Artisanal and small-scale mining is a significant and growing livelihood across the global South, 3 which all too often leaves a legacy of contaminated landscapes. Given the increasing reliance of 4 economies on metals and minerals, it is critical to understand what controls contamination 5 outcomes in this rapidly developing extractive practice. Here, we demonstrate that the emerging 6 concept of co-production offers a novel way to elucidate the joint contributions of natural and 7 societal factors in shaping contaminant exposure from artisanal and small-scale mining. 8 Specifically, understanding the co-production of contaminated landscapes requires attention to 9 both the political economy of mining, including how labor and extraction methods differ across 10 mines, as well as the sources and pathways of mercury exposure. In Madre de Dios, Peru, we 11 measured mercury levels in wildlife inhabiting abandoned gold mining sites worked with 12 different extraction technologies. We found that the type of technology used, whether heavy 13 machinery or suction-pump based, influenced mercury loading into mines, and together with 14 differences in food-web structure, mediated mercury biomagnification rates. Mercury 15 concentration increased 2.1 to 3.7-fold per trophic level, and bioaccumulation levels were high in 16 both mined and unmined sites—indicating elevated background levels in the region. We also 17 found evidence of lateral transfer of mercury from abandoned mining pits to terrestrial food webs. 18 This observation indicates that the footprint of mercury contamination extends well beyond 19 individual mines, affecting the larger landscape. Our findings underscore the necessity of 20 understanding the entangled ways in which social and ecological factors contribute to the 21 production of toxic landscapes.


IV
Miners commonly use mercury to extract gold (11,12). Processing gold releases 51 inorganic mercury into the atmosphere, which can be redeposited in nearby waterbodies. In the 52 bottom sediments of these waterbodies, under certain abiotic conditions, inorganic mercury can 53 undergo a microbially-mediated transformation into methylmercury, the most toxic and 54 biologically available form of mercury (13)(14)(15). Many studies have documented bioaccumulation 55 of mercury in the vicinity of active mines (16)(17)(18)(19), in sediments and tailings downstream of 56 mining (20)(21)(22), and even legacy contamination from mines that have long been abandoned (23, 57 24). Bioaccumulation of mercury can cause a wide range of detrimental impacts to wildlife such 58 as a reduction in growth (25)(26)(27), juvenile survivorship (28), reproductive success (29,30), and 59 even mortality (31). Contaminant exposure extends beyond wildlife, as humans are also exposed 60 to mercury through the consumption of fish and other top predators (6). Because ASGM is 61 concentrated in culturally and biologically diverse areas of the globe (7), currently occurring in 62 more than 80 countries in the global South (1,2), understanding risk of contaminant exposure in 63 these unique ecosystems is critical. 64 The emerging concept of co-production (32,33) represents a novel way to conceptualize 65 the joint contributions of natural and societal factors in determining the risk from exposure to 66 contaminants from ASGM. Co-production of ecosystem services (or disservices) is the process by 67 which societies leverage labor, artefacts, and technology to use material and non-material flows 68 of nature to produce goods and services that benefit (or harm) humanity (34). For example, 69 hydropower is an ecosystem service co-produced by using built infrastructure (i.e., dams) to 70 harness river flow regimes -but dam-induced habitat fragmentation and altered flows represent 71 disservices that exact a high environmental cost (35,36). Another example of co-production is 72 gold mining, wherein humans transform the natural wealth of geologic deposits into commodities 73 by using extractive technologies to remove, process, and refine auriferous rock. Just as the 74 V number and spatial arrangement of dams in a river network mediates the amount of harm done to 75 salmonid fisheries due to habitat fragmentation, the intensity and scale of mining impact the 76 conditions of a landscape and its ability to support bacterial production of toxic methylmercury. 77 Such place-specific nuances of ASGM underscore the need to improve our mechanistic 78 understanding of how co-production of contaminated landscapes is mediated by differences in the 79 political economy of gold production and the local ecological context. 80 The case of the bioaccumulation and biomagnification of mercury in wildlife inhabiting 81 abandoned gold mines in Madre de Dios, Peru illustrates the potential of using a co-production 82 approach. In Madre de Dios, a diversity of gold extraction methods co-exist with differing levels 83 of mechanization ranging from non-mechanized artisanal operations to heavily mechanized 84 operations using front loaders and excavators. Mechanization drives variation in the degree of 85 environmental impact (37) and in daily production volume which is often directly correlated with 86 mercury usage. A legacy of the increasing mechanization of ASGM is the creation of networks of 87 abandoned mining pits that cover approximately 28% of the land area deforested due to mining, 88 or an area equivalent to ~20,000 ha (38). Over time, these ponds are colonized by invertebrates, 89 fish, and other wildlife forming new networks of aquatic habitat. Many of these abandoned mines 90 also become sources of wild fish for people inhabiting nearby areas. However, these ponds are 91 also potential hotspots where inorganic mercury can be transformed into methylmercury, and 92 bioaccumulate in wildlife (24,39). In Madre de Dios, there is growing evidence that mercury 93 contamination from ASGM extends beyond the boundaries of mining areas as high levels of total 94 inorganic mercury (THg) have been documented in downstream river sediments (21,22), fish 95 (16), indigenous populations upstream of mining (40), and in wildlife found far from mining areas 96 (19). 97 VI Here we examined how co-production of contaminated landscapes mediated exposure 98 risk to wildlife in three distinct ways. First, we evaluated whether mercury bioaccumulation in 99 wildlife inhabiting abandoned gold mines differed when compared to mercury concentrations in 100 wildlife found in unmined sites. We found that bioaccumulation was high in all sites, even 101 unmined sites, but that bioaccumulation in wildlife was highest in those sites where ASGM 102

occurred. 103
Second, we evaluated whether extraction technologies influenced mercury 104 biomagnification rates by comparing natural lakes in an unmined watershed to areas worked with 105 heavy machinery (HM) and those worked with suction-pump based techniques (SP). We 106 estimated the trophic magnification slope (TMS) by regressing THg concentrations against stable 107 nitrogen isotopes (δ15N) of multiple consumers common to all mining pits. Our estimated TMS 108 of 0.46 ± 0.03 was on the high end of reported values for tropical freshwater, lentic environments. 109 Further, we found that the type of technology used in gold production influenced the degree of 110 mercury loading (as measured by sediment THg concentration), and together with differences in 111 food-web structure across abandoned mines, these factors controlled the rate of mercury 112 biomagnification at each site. 113 Finally, we determined whether mining pits subsidize mercury for terrestrial ecosystems 114 via export of contaminated prey that are then consumed by riparian predators. We collected a 115 riparian predator common to all sites, long-jawed orb-weaving spiders (Family: Tetragnathidae) 116 and compared THg concentration and trophic position of these spiders to those of aquatic 117 consumers. We confirmed that cross-ecosystem subsidies of aquatic prey transfer not only energy 118 but also contaminants to riparian and terrestrial food webs. This result confirms that the footprint 119 of mercury contamination extends beyond boundaries of individual mining pits, further exposing 120 humans and wildlife. 121

123
We found high levels of mercury bioaccumulation and biomagnification in taxa sampled 124 across all mined and unmined sites (Fig. 2b, and SI Appendix, Table S1). However, the highest 125 mercury concentrations were found in wildlife collected in abandoned gold mines as compared to 126 unmined sites (Table S1). Taxonomic groups showed relatively consistent trophic positions 127 across sites (as measured by δ15N), but spanned >2 trophic levels from snails (caenogastropoda), 128 lowest in the food chain, to piranha (Serrasalmus spp.) and wolf fish (Hoplias malabaricus), 129 apical predators (Fig. 2a). Accordingly, predatory fishes contained the highest average 130 concentrations of THg (Fig. 2b). We also found strong evidence of biomagnification. Total Beyond differences in bioaccumulation and biomagnification between unmined and 141 mined sites, we found evidence of co-production of contamination: THg loading was influenced 142 by both social and natural factors (Table 1). To fully parse the relative importance of trophic 143 position, THg loading in sediments, and the effect of extraction technology on bioaccumulation 144 of mercury in biota, we compared fits of different explanatory model structures. Specifically, we 145 compared support across four linear mixed-effect models using an information-theoretic approach 146 via BIC (Table 1). We found that including an interaction effect between organism trophic 147 VIII position and extraction technology (i.e., whether the site was worked with HM, SP, or was an 148 unmined site) greatly improved model fit over a restricted model that included the effects of 149 technology and trophic position separately. We also evaluated the two best fit models using log-150 likelihood tests (Table 1, models 3 & 4). We found that model 4, which included the interaction 151 effect as well as the fixed effect of mercury loading (THgSed), was significantly different from 152 model 3 (c 2 (1) = 5.806, p = 0.016). This finding suggests that mercury loading, influenced by the 153 type of extraction technology used, drives patterns in mercury bioaccumulation across a diversity 154 of taxonomic groups. 155

156
In addition, we found that sediment total mercury concentration (THgSed), a proxy for 157 mercury loading to each mining pit or reference lake system, varied across sites worked with 158 different technologies (Fig. 3, SI Appendix, Fig. S1) with a significant difference in mean total 159 mercury concentration across site types (F2,36 = 3.438, p = 0.043). The highest levels of loadings 160 were recorded in pits worked by suction pumps, followed by pits worked by HM, and then 161 unmined oxbow lakes. A post-hoc Tukey HSD analysis indicates that sites worked with suction 162 pump machinery were significantly different in THgSed concentrations than those in unmined 163 sites. There was no difference in THgSed between unmined sites and heavy machinery sites. 164 However, we found that one of the oxbow lakes in our control sites was elevated in THgSed 165 possibly due to the fact that this lake is a palm swamp with low concentrations of dissolved 166 Table S2). If this lake were excluded from the analysis, then the mean 167 THgSed would be significantly different between all paired comparisons of sites. 168 169 Finally, we found evidence of lateral transfer of mercury from abandoned mining pits to 170 terrestrial, riparian consumers (Fig. 4). Stable isotopes of δ15N as well as field observations of 171 feeding behavior, confirm that a riparian predator, long-jawed orb weaving spiders, consume riparian consumers bioaccumulated mercury at concentrations that fell within those predicted by 174 the relationship between THg concentrations and trophic position of our aquatic taxa. 175 176 177

179
Co-production drives mercury accumulation and biomagnification in AGSM sites 180 We found strong novel evidence that mercury contamination in Madre de Dios, Peru, is Water is pumped out of the ponds to constructed sluice boxes sometimes more than 200 meters 189 away to wash auriferous material that is brought from elsewhere on-site. Therefore, the method of 190 gold extraction, whether using HM or SP technologies plays an important role in directly 191 mediating the quantity and location of mercury discharges on the landscape. Our finding adds 192 evidence to the argument that local or regional-level differences in ASGM practices may affect 193 mercury loading into aquatic ecosystems (41). which is known to affect biomagnification rates, varied by sites worked with different extraction 198 technologies. In pits worked with heavy machinery, the average FCL was 2.70, for SP pits it was X 2.30, and for reference lakes it was 2.05. While we do not know the mechanism driving 200 differences in FCL across sites worked with different technologies, we postulate that differences 201 in feeding behavior of the same organism across sites (i.e., omnivory in predatory fish) or food 202 web complexity (i.e., addition or insertion of top consumers lengthening the food chain) may be 203 responsible for this variation (42). Understanding the exact mechanism is particularly important 204 given that wildlife in these landlocked ponds will readily bioaccumulate contaminants from 205 dietary exposure and differences in FCL will mediate the accumulation of contaminants at the top 206 of the food chain. However, trophic position alone did not explain variation in bioaccumulation of 207 THg in wildlife across sites, as taxa with the same trophic position (e.g., Hypostomus spp., 208 Serrasalmus spp.) were consistently higher in THg concentration in SP pits relative to HM pits. 209 Only when we took the full suite of co-production factors -trophic position, extraction 210 technology, and mercury loading -into account did our model (Table 1)  We found extremely high mercury biomagnification rates in our study sites. Our 215 estimated TMS (0.46 ± 0.03) is more than three times that of the mean global value for freshwater 216 tropical sites, 0.12 ± 0.12 (43), and at the high end for studies restricted to Amazonia, which 217 ranged from 0.21 -0.43 (44 -49). These magnification rates have led to levels of top-predator 218 mercury bioaccumulation that represent a significant public and environmental health risk. sediments downstream of mining sites and soils taken from areas proximate to active mining, but 253 this study is the first to demonstrate that abandoned mining pits are an important source of 254 mercury to the terrestrial biota inhabiting these highly impacted systems. More work is needed to 255 fully understand the risk of contaminant exposure to higher trophic levels organisms. Previous 256 work suggests that the degree to which these mining pits export mercury via emergent aquatic 257 insects depends on the life history of these insects, as metamorphosis of aquatic larvae into adults 258 reduces heavy metal burden in aquatic predators such as dragonflies and consequent 259 accumulation in cross-system predators (52). In addition, body size of emergent aquatic insects, 260 mediated by top-down interactions, can also alter contaminant flux into terrestrial ecosystems 261 (53). In addition, these mosaics of abandoned mining pits form new habitat for migratory birds 262 and other higher trophic levels organisms whose mobility allows them to further extend the 263 footprint of mercury contamination from ASGM. 264 265 While these pits can act as sources of methylmercury to wildlife, it should be noted that 266 these results also indicate that there is high heterogeneity in mercury bioaccumulation and 267 biomagnification potential within abandoned gold mining landscapes driven largely by the 268 differences in the political economy of ASGM. The diversity in mining operations drive variation 269 in quantity and location of mercury discharges such that not everywhere in a mined landscape 270 will there be evidence of high levels of mercury in biota. In turn, areas far removed from gold 271 mining could have high levels of mercury in wildlife due to cross-ecosystem subsidies and 272 mobility of consumers. Thus, sampling only one compartment such as the sediment of mining pits 273 or the soils around abandoned mines can lead to inconclusive results, as differences in mercury XIII bioaccumulation higher up in the food chain are not apparent from the relatively small differences 275 in mercury loading in sediment (on the order of 10 ng). Further, proposals to utilize these 276 contaminated landscapes for production of fish through aquaculture or for farming should take 277 into account the type of technology that was used in gold production (38). Sites worked with HM 278 may be safer candidates for remediation for aquaculture than sites worked SP. Additionally, top 279 consumers like predatory fish, fish-eating birds, and mammals should be closely monitored due to 280 their susceptibility to accumulate mercury especially from sites where SP technologies were used. 281 282

Conclusion 283
Environmental contamination is not an isolated or unique phenomenon, but instead it is 284 part of a ubiquitous pattern that implicates human's continued consumption of resources in the 285 degradation of the natural world (54). Our globalized economy's reliance on non-renewable 286 resources is unlikely to diminish in the near future, especially with the use of precious metals in 287 climate-change mitigating technologies like solar panels and electric car batteries (55)(56)(57). To 288 have any hope of mitigating the worst effects of contaminant exposure or remediating already 289 contaminated landscapes, we must understand the entangled ways in which social and ecological 290 factors contribute to the production of toxic landscapes. A co-production approach, that explicitly 291 incorporates the social and natural, can help guide the kind of studies that are urgently needed to 292 ensure that our planet continues to provide the life supporting services that we depend on while 293 also ensuring just outcomes for those people who depend on resource extraction for their 294 livelihoods. 295 296

298
Study Site & Field Sampling. We conducted our study in the Department of Madre de Dios, 299 Peru, a region located at the western edge of the Amazon basin (Fig. 1). We sampled across four sites, with each site characterized by the use of either suction-pump (SP) or heavy machinery 301 (HM) extraction technologies as well as one unmined site. On the surface, the mining pits created 302 by these two different mining technologies appear similar, however they differ in key aspects. 303 Suction pump-based technologies create mining pits that are more heterogeneous in their 304 bathymetric profile and on average much deeper than their counterparts created with heavy 305 machinery. The volume of gold-bearing alluvium processed daily by these two types of 306 production also differs. Daily production volume is approximately proportional to the level of 307 mechanization and subsequently influences the amount of mercury used in amalgamation of gold 308 at the end of a day's shift. 309

310
To test whether production practices mediate mercury biomagnification, we sampled multiple 311 abandoned gold mining pits (³2) at each site. A total of 10 mining pits were sampled including 312 seven mining pits where SP technologies were used and three mining pits where HM was used. 313 We were unable to access more sites worked with HM due to the rapidly changing security To estimate the rate of biomagnification across sites, we sampled for the same taxa across all pits 322 and oxbow lakes including benthic macroinvertebrates and algivorous and piscivorous fish. 323 Benthic macroinvertebrates common across all pits and oxbow lakes included predatory 324 dragonflies (Families: Gomphidae, Libellulidae), giant water bugs (Family: Belostomatidae), and water scorpions (Family: Nepidae). Primary consumer invertebrates (herbivorous/detritivorous 326 strategies) included (Family: Caenogastropoda) and burrowing mayflies (Family: 327 Polymitarcyidae). Benthic macroinvertebrates were live sorted to family level and then kept in 328 water for 12-24 hours to clear gut contents. In the lab, the invertebrates were counted, measured 329 and pooled by family to produce a composite sample for each pit. The invertebrates were kept 330 frozen until they were freeze-dried at -56°C for 72 hours. Gastropods were used for estimation of 331 baseline d 15 N following Post (2002, 58). In addition, a maximum of three individuals of the 332 following fish species were collected at each mining pit or oxbow lake; piranha (Serrasalmus 333 spp.), wolf fish (Hoplias malabaricus), and armored catfish (Hypostomus spp.). Fish were 334 collected using a gill net deployed on the same day as invertebrate collection; length and weight 335 were measured, and dorsal muscle tissues were removed, placed on ice until they could be frozen, 336 and later freeze-dried. independent variable) on total mercury concentration in biota (THg; continuous dependent 396 variable). We tested for an interaction effect between technology and trophic position, which 397 would indicate that the slope estimate for the relationship between trophic position and mercury 398 concentration was dependent on the type of technology used. If the interaction effect was not 399 significant, we removed the term and assessed the main effects of trophic position and technology 400 independently. We also tested for differences in mercury loading across pits worked with the 401 same technology using ANOVA and a post-hoc Tukey HSD test to evaluate whether there was a 402 difference in sediment mercury loading across sites. Finally, we used a generalized linear mixed 403 effects model, lme4 (lmer package, 71) to evaluate the proportion of variance in total mercury 404 concentration of biota that was explained by a model that accounted for the random effect of pits 405 nested within sites worked with different extraction technologies, as well as the fixed effects of 406 mercury loading as estimated by the average pit or lake sediment total mercury concentration 407 (THgSed ), trophic position, and the interaction between trophic position and extraction 408 technology. We compared our models using Bayesian information criterion (BIC) scores and chi-409 square log-likelihood ratio tests.