Anthropogenic litter is a novel habitat for aquatic macroinvertebrates in urban rivers

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. Freshwater Biology published by John Wiley & Sons Ltd 1School of Geography, University of Nottingham, Nottingham, U.K. 2Geography and Environment, Loughborough University, Leicestershire, U.K. 3School of Biological, Earth and Environmental Sciences, University College Cork, Cork, Ireland 4Environmental Research Institute, University College Cork, Cork, Ireland


| INTRODUC TI ON
Growing public and political interest around problems of anthropogenic litter (solid manufactured waste items, including but not limited to plastics) has encouraged a recent proliferation of studies into its occurrence and abundance in the environment, and its ecological impacts (e.g. Agamuthu et al., 2019;Beaumont et al., 2019). In marine ecosystems, anthropogenic litter has been shown to reduce organism fitness and cause mortality through entanglement and ingestion (Agamuthu et al., 2019;Kühn et al., 2015), expose organisms to harmful chemicals via leaching of pollutants (Rochman, 2015), modify physical habitat structure (Kiessling et al., 2015), and aid the spread of invasive species (Rech et al., 2016;Tyrrell & Byers, 2007).
Despite urban rivers being some of the most anthropogenically modified landscapes on Earth (see review: Walsh et al., 2005), and acknowledgement of the persistence and prevalence of anthropogenic litter in urban rivers (e.g. McCormick & Hoellein, 2016;Rech et al., 2015), the effects of anthropogenic litter on these ecosystems have yet to be fully explored (Blettler et al., 2018).
Urban rivers are typically limited in habitat diversity and quality due to historical channelisation, dredging and bed/bank stabilisation works.
Coupled with elevated concentrations of pollutants, and changed hydrological and sediment inputs, this has resulted in characteristic low diversity communities in many urban rivers (termed the urban stream syndrome: Walsh et al., 2005). Urban rivers also receive disproportionately large inputs of anthropogenic litter, which, although undesirable, may provide different and more complex shapes and textures of substrate. Anthropogenic litter may also interact with flow patterns, increasing habitat heterogeneity in a similar manner to large rocks, wood, and aquatic macrophytes. These natural habitat structures are largely absent in urban rivers due to regular removal practices and the high frequency of disturbance events (Blauch & Jefferson, 2019).
Anthropogenic litter may therefore act as a proxy for habitats lost through urbanisation, supporting biodiversity that would otherwise be absent (Chapman & Clynick, 2006). For example, flexible anthropogenic litter may perform similar physical functions to aquatic macrophytes.
Macroinvertebrates are known to readily colonise artificial surfaces as long as they are non-toxic (e.g. the use of artificial substrate samplers; Beak et al., 1973). As their community structure is strongly related to habitat, especially the size, diversity, and arrangement of riverbed substrates (Death, 2000;Jowett, 2003), this makes them useful model organisms to assess the effects of anthropogenic litter on urban rivers. The atypical physical structure of anthropogenic litter could provide a novel habitat for some macroinvertebrate taxa, offering opportunities for adaptable species to colonise, and resulting in a community distinct from those living on natural substrates (Czarnecka et al., 2009). Anthropogenic litter with complex physical structure (e.g. items such as bottles which have interiors : Czarnecka et al., 2009; or items with rough surfaces: Boyero, 2003) may also be able to support a greater diversity of organisms through increasing available niche space. This may be especially true where the natural substrate is relatively inhospitable to macroinvertebrates, such as sandy or silty estuarine rivers and harbours with low bed stability (e.g. Francis & Hoggart, 2008;García-Vazquez et al., 2018). The quality and quantity of macroinvertebrate food resources are also affected by microhabitat conditions (Wallace & Webster, 1996) so changes in the abundance or diversity of food resources caused by anthropogenic litter may have cascading effects in macroinvertebrate communities. In addition, anthropogenic litter may preferentially support non-native species that may be better able to take advantage of the novel habitat than native species (Katsanevakis, 2008;Tyrrell & Byers, 2007).
The limited number of published studies examining fauna living on anthropogenic litter has been focussed on marine (e.g. Chapman & Clynick, 2006;Katsanevakis et al., 2007;García-Vazquez et al., 2018) rather than freshwater environments. These studies report that experimentally introduced anthropogenic litter on a sandy sea bed may locally increase the abundance and diversity of benthic communities, where it provides habitat for hard-substratum dwelling species that are otherwise absent (Katsanevakis et al., 2007). However, where comparable natural habitats are present, such as natural rocky reefs, there may be limited differences in patterns of colonisation of anthropogenic litter and natural substrates (Chapman & Clynick, 2006). So far, only one study has considered pre-existing in situ anthropogenic litter as a component of freshwater habitats (Czarnecka et al., 2009), where it was reported that the macroinvertebrate communities found on anthropogenic litter in a Polish reservoir were more diverse and considerably different in taxonomic composition from those on the surrounding sand bed, but were similar in diversity to those recorded on macrophytes. However, so far there have been no investigations of in situ anthropogenic litter undertaken within non-tidal river systems.
In this study, we compare macroinvertebrate communities inhabiting anthropogenic litter with those on natural rock substrates in three urban rivers to provide the first direct evaluation of the role of anthropogenic litter as riverine habitat. We anticipate that distinct communities would be recorded on the two substrate types, and that faunal diversity would be higher on anthropogenic litter given its heterogeneity relative to natural mineral substrates. If anthropogenic litter provides novel habitats, understanding how it affects macroinvertebrates is important in informing future urban stream management.

| Study sites
Sampling was conducted in three small (first or second order) urban gravel-bed rivers in Leicestershire and Nottinghamshire, U.K.: the River Leen, Black Brook, and Saffron Brook (Figure 1). Each river was sampled over two consecutive days in September and October 2018.
Straightened reaches with homogenous substrate grain-size and morphology were selected to minimise any effect of natural morphological heterogeneity. Sites were similar in dimension, water quality, and discharge, but differed in urbanisation intensity (Table S1).
The predominant natural substrate (substrate is defined here as riverbed material on which an organism lives) at all sites was gravel and cobbles (hereafter rocks), with some interstitial fine sediment (sand and silt). Rocks were comparable in size to anthropogenic litter pieces and could be easily isolated from the riverbed to collect the macroinvertebrates inhabiting them (similar to anthropogenic litter). Hence rocks were chosen for comparison with anthropogenic litter. Both rocks and anthropogenic litter were sampled from the riverbed surface for consistency. There was not any discernible structure to the bed sediments, such as armouring, as subsurface sediments were visually similar to those on the surface.

| Field methods
Anthropogenic litter density was assessed at each site by measuring the area of riverbed (average channel width × river length surveyed) containing 100 pieces of anthropogenic litter. Rock and anthropogenic litter samples were collected from the full width of the channel and the surface layer of the riverbed. Whilst moving upstream in a grid pattern, we collected alternately encountered anthropogenic litter items (providing 50 samples at each site), and a representative sub-sample of 50 rocks by pacing through the sampling area and taking the rock immediately at the sampler's foot (Wolman, 1954). Only items larger than 1 cm in their b-axis length were sampled, as smaller items were difficult to consistently collect and macroinvertebrate numbers would be low on such items.
Items were described in terms of their material composition (fabric, glass, metal, plastic, masonry, rock, or other). Pieces of masonry (e.g. brick, concrete, and roofing tiles) were classified as rocks in comparisons of all anthropogenic litter types against all rocks, as it was thought that they may function like natural mineral substrates.
However, masonry and rock samples were considered as separate materials in analyses of material types to test this assumption.
Macroinvertebrates were collected by transferring items (anthropogenic litter or rocks) from the riverbed into a 1-mm mesh kicknet held directly downstream (following Benke and Wallace's [2003] methodology for sampling macroinvertebrates on large wood). The contents of the net were placed into a sampling bag, along with the item, and preserved with industrial methylated spirit. Large or embedded items were cleaned of macroinvertebrates in the field by scrubbing a set area of 0.03 or 0.06 m 2 depending on their exposed area (0.03 m 2 was roughly equivalent to the median surface area of anthropogenic litter pieces) with a brush to dislodge macroinvertebrates into a kicknet held downstream (Pilotto et al., 2016).

| Laboratory methods
All anthropogenic litter and rock items were individually washed through a 500-µm mesh sieve, then manually processed to collect macroinvertebrates. Macroinvertebrates were identified to species or genus level where possible. Exceptions were Diptera and Sphaeriidae, which were identified to family, Oligochaeta to subclass, and Acarina to order. Taxonomic levels were consistent between samples and sites. Identification followed Holland (1972), Ellis (1978), Friday (1988), Wallace et al. (1990), Edington and Hildrew (1995), Reynoldson and Young (2000), Killeen et al. (2004), Humpesch (2010), Cham (2012), Dobson et al. (2012), and Elliott and Dobson (2015). Trichoptera (caddisfly) pupae, unlike larvae, could only be identified to family level so were excluded from further analysis. The data analyses outlined in Section 2.4 were repeated with family level data, which included caddisfly pupae, and findings were qualitatively identical (Table S2).
The surface area of each item (anthropogenic litter or rock) was approximated by wrapping the item in tin foil and weighing the resultant foil pieces (1 g: 0.0214 m 2 ; Dudley et al., 2001). The surface area of flexible materials or items with complex shapes (e.g. plastic bags) was determined using equations for the surface area of the approximate geometric shape (Bergey & Getty, 2006). Items that were too large or embedded to be collected from the field were measured in situ.

| Data analysis
All statistical analysis was conducted using R statistical software (version 3.6.3; R Core Team, 2020). Completeness of sampling F I G U R E 1 Map showing the three sampling sites (shown as triangles) on the River Leen, Black Brook, and Saffron Brook relative to the urban areas they flow through and the mainstem rivers was assessed by calculating coverage for anthropogenic litter and rocks at each site. This measure of sample completeness estimates the proportion of total individuals in a community that belong to taxa in the sampled community (Chao & Jost, 2012).
Macroinvertebrate density was calculated by dividing the total macroinvertebrate abundance across taxa by the sampled surface area of an item (0.03 or 0.06 m 2 for partially sampled items).
Macroinvertebrate diversity was assessed by calculating Hill's numbers in vegan (Oksanen et al., 2019). The Hill series are defined to the order q (D q ), which determines the weighting of rare to abundant taxa for each index. D 0 is equivalent to observed taxa richness, which places greater emphasis on rare taxa as it is insensitive to relative frequencies (i.e. evenness); D 1 is equivalent to the exponential of Shannon's entropy, which is weighted towards common taxa; and D 2 to the inverse of Simpson's diversity, which is weighted towards highly abundant taxa (Tuomisto, 2010). Each point in the series therefore provides complementary information on taxa richness and evenness.
The mean surface area of rocks (including masonry) was four times smaller than that of anthropogenic litter items (rocks: 0.03 m 2 ± 0.01 [SE], anthropogenic litter: 0.12 m 2 ± 0.02; two-sample Wilcoxon W = 16,899, p < 0.001). Given that a strong positive relationship exists between item surface area and total macroinvertebrate abundance (Spearman's rank [R s ] = 0.80, p < 0.001), as well as between surface area and observed taxa richness (D 0 ; R s = 0.79, p < 0.001), all subsequent analysis controlled for surface area (by including area in linear mixed effect models and generalised linear models) to account for this difference between substrates.
To test for differences in density and diversity (D 0 , D 1 , and D 2 ) between anthropogenic litter and rocks, linear mixed effects analysis was performed using lme4 (Bates et al., 2015) with significance calculated for parameter estimates using lmerTest (Kuznetsova et al., 2017). To compare diversity, substrate (anthropogenic litter or rock) and sampled surface area were entered as fixed effects, and site (River Leen, Black Brook, or Saffron Brook) included as a random effect. Linear mixed effects models for density excluded surface area, as this factor is already incorporated into the calculation of density for each item, but otherwise model structure was identical. Model validation and checking followed the protocol in Zuur et al. (2009). Significance values for the effect of substrate type were identified by likelihood ratio tests (distributed as χ 2 ) of the full model against a null model without the substrate factor. Linear mixed effects analyses were repeated, substituting the substrate factor for material composition using a single factor with seven levels: fabric, glass, metal, masonry, plastic, rock, and other. Significant differences between material types were examined using parameter estimates and associated p values calculated using Satterthwaite approximation in lmerTest. Thus, we looked for differences between substrates (anthropogenic litter and rock), and between material types (fabric, glass, metal, masonry, plastic, rock, or other) in separate analyses.
Macroinvertebrate community composition was compared using the manyglm function in mvabund (Wang et al., 2020). The function fits generalised linear models (GLMs) to the raw counts for each taxa assuming a negative binomial distribution, with substrate type, sampled surface area, and site as explanatory variables without interactions. A sum-of-LR test statistic was obtained with significance assigned using randomisation (999 permutations), where the p value is adjusted for multiple testing using step-down resampling. This approach deliberately specifies a mean-variance relationship, inherent to count data, meaning that it can address the problems of confounded location and dispersion effects and difficulty detecting effects expressed in low-variance taxa, common to distance-based community analysis such as SIMPER and PERMANOVA (Warton et al., 2012). Manyglm tests were also repeated substituting substrate for material composition.
Differences between communities were visualised using boral (Hui, 2020); a model-based approach to unconstrained ordination that fits a latent variable model to raw abundance data and can be interpreted in a similar way to non-metric multidimensional scaling ordination (Hui, 2015). Ordination assumed a negative binomial distribution, and sample identity effects were included so ordination is based on composition rather than relative abundance. Site was included as a fixed effect. Ordination was repeated for individual sites to visualise differences between material types within each site.

| Anthropogenic litter abundance and composition
Anthropogenic litter was abundant at all sites: 4.2 items/m 2 in the River Leen; 1.1 items/m 2 in Saffron Brook; and 0.6 items/m 2 in Black Brook. Anthropogenic litter material types included fabric, glass, metal, plastic, ceramic, and wood. Fewer than five ceramic and wood items were collected across all sites so these have been collated hereafter as other for simplicity. The proportional abundance of anthropogenic litter materials was similar across all sites, with glass, metal, and plastic the dominant materials across sites. These materials each made up approximately one third of the total anthropogenic litter items (Table S3). All fabric items were flexible, as were 5% of metal items, and 69% of plastic items, but no other materials were flexible. Rocks were generally less morphologically complex than anthropogenic litter, having been rounded by fluvial processes.
Whilst most rocks were of natural origin, some appeared to be failed bank protection (based on visual comparison with nearby rip-rap), and 10% were masonry (brick, concrete, or roofing tiles).

| Differences in macroinvertebrate density and diversity
Across all sites, a total of 16,894 individuals from 46 families (61 taxa) were collected (see Table S4 for full list of taxa). The completeness of sampling (checked by calculating coverage) was >0.99 for anthropogenic litter and rock at all sites, indicating that sampling was close to completion. As such, it is reasonable to compare estimates of diversity between anthropogenic litter and rocks, despite their differences in surface area. The density of macroinvertebrates was not significantly different between anthropogenic litter and rocks (χ 2 (1) = 0.81, p = 0.37), or between material types (χ 2 (6) = 7.73, p = 0.26; Table 1). However, macroinvertebrate diversity was significantly higher on anthropogenic litter than on rock, indicating a consistent pattern across all sites and for all diversity measures (χ 2 (1) = 24.54 (D 0 ), 22.63 (D 1 ), 12.28 (D 2 ), p < 0.001; Figure 2).
On average, observed taxa richness (D 0 ) was nearly four taxa per item higher on anthropogenic litter than on rocks, with a mean of 8.3 ± 0.5 (SE) for anthropogenic litter and 4.6 ± 0.4 for rocks. This difference was reduced at a higher order of D, suggesting that the higher diversity on anthropogenic litter reflected greater numbers of low abundance taxa with a small number of dominant taxa.
Material type (fabric, glass, metal, plastic, masonry, rock, or other) also affected diversity measures (χ 2 (6) = 52.18 (D 0 ), 37.20 (D 1 ), 19.26 (D 2 ), p < 0.005; Figure 3). Glass and rock samples were considerably less diverse than other material samples (mean D 0 per item was 5.1 ± 0.5 and 4.2 ± 0.3, respectively), especially fabric (11.0 ± 1.9) and plastic (10.9 ± 0.9). These differences were significant; rock samples were significantly less diverse than masonry, fabric, plastic, and metal samples across all Hill's numbers. Glass samples were less diverse than plastic and metal samples at D 0 and D 1 , but were not different from other materials at D 2 . Plastic, metal, fabric, and masonry samples consistently had the highest diversity across Hill's numbers.

| Differences in macroinvertebrate community composition
We checked whether any taxa exclusively inhabited either anthropogenic litter or rocks, excluding taxa that occurred in fewer than five samples. This was necessary to verify that apparent associations were not due to low abundance of a taxon. Under these conditions, no taxa were recorded only on rocks, but five taxa were recorded exclusively on anthropogenic litter. These were: Anisus vortex (Gastropoda: total abundance of 25 across 13 samples), Theromyzon tessulatum Caddisfly). All of these taxa were more common on plastic; especially the three leeches (80% of occurrences on plastic items).
Separate ordinations for each site indicated that metal, fabric, and especially plastic anthropogenic litter items supported the communities which were most dissimilar to those on rocks ( Figure 5).
In contrast, glass samples were similar in composition to those from rocks, as were masonry samples, despite the differences in diversity (D 0 , D 1 , and D 2 ) on these materials. The differences in community composition on different materials were more evident in Black Brook and Saffron Brook than the River Leen. Note: For full details on how statistical tests were performed see Section 2.4 'Data analysis'. linear mixed effect (LME) models tested for differences in density and diversity (D 0 , D 1 and D 2 ) between substrates and between materials by including substrates/material and surface area as fixed effects (surface area was excluded for density tests), and site as a random effect. The test statistic for LME models is the χ 2 test statistic of a likelihood ratio test. The manyglm function tested for differences in community composition between substrates and materials, calculating a sum-of-LR test statistic and associated p value with 999 permutations.

| Differences between types of anthropogenic litter and rock
Previous research on anthropogenic litter in a Polish reservoir (Czarnecka et al., 2009) (Death & Winterbourn, 1995;Nakano et al., 2018). These substrates may help macroinvertebrates avoid dislodgement, provide more reliable food resources, and act as flow refuges (Jowett, 2003). Unstable, highly mobile substrates are likely to be inhabited by less diverse communities (Death, 2008), because substrate movement will inhibit colonisation and continually reset successional trajectories (Czarnecka et al., 2009).
Predicting the relative stability of anthropogenic litter is complicated by its wide variety of shapes and densities (Williams & Simmons, 1997), and therefore, unlike rocks, is it not strongly associated with size. For example, low-density plastic bags are easy to entrain but conversely more likely to become stranded on obstacles such as vegetation or to become partially buried (McCormick & Hoellein, 2016). An added complication is that recruitment of anthropogenic litter in urban rivers can occur independently of flow stage (McCormick & Hoellein, 2016), making it difficult to estimate its exposure time in the river. Further investigation of these dynamics of anthropogenic litter, and the ways it differs to natural substrates, would help build our understanding of the ways anthropogenic litter may affect macroinvertebrate distribution. Nonetheless, the presence of biofilm on all anthropogenic litter and rocks sampled in this study, and the low flows during summer months prior to sampling, means that most items will have been exposed long enough for the colonisation of macroinvertebrates to have occurred.
Small-scale complexity, at a scale similar to the body length of macroinvertebrates, is known to be an important control of faunal distribution (Boyero, 2003;Robson & Barmuta, 1998 Boyero, 2003;Clifford et al., 1989;Clifford et al., 1992;Robson & Barmuta, 1998). Basic life-functions, such as respiration, metabolism, locomotion, and reproduction, are affected by the physical characteristics of the habitat (Lancaster & Downes, 2013), so a more structurally diverse habitat is thought to support greater biodiversity.
Likewise, complex surfaces allow macroinvertebrates to shelter from hydraulic stress, enabling conservation of energy and preventing accidental entrainment into drift (Brooks et al., 2005), as well as providing shelter from predators (Everett & Ruiz, 1993). In this study, we found that macroinvertebrate communities on smooth and flat glass and rocks were less diverse than those on other material types. Similarly, masonry samples (mostly bricks with complex holes or grooves) supported a much higher macroinvertebrate diversity (for D 0 , D 1 , and D 2 ) than rocks, despite there being limited difference in community composition on the two materials. This suggests that although rock and masonry habitats are functionally similar, and so support comparable communities, the greater complexity of masonry samples means that they can support more diverse macroinvertebrate communities.

F I G U R E 5
Output of latent variable model ordination of macroinvertebrate data for all sites separately to more easily show differences between material types (rock, masonry, fabric, glass, metal, plastic, other), which are shown using different shaped symbols As well as differences in diversity on anthropogenic litter and rock, differences in the macroinvertebrate community composition recorded on different materials suggests that these materials function as distinct habitat types because of their variation in physical structure. For instance, the similar communities found on rocks, masonry, and glass could relate to their rigid and hard structure. In contrast, plastic and fabric sample communities were clearly distinct from rocks. These distinctions between materials were much less clear in the River Leen, where anthropogenic litter and rock communities were more homogenous. The reasons for this are not known. A symptom of urban rivers with degraded habitat is community homogenisation (i.e. urban stream syndrome: Walsh et al., 2005); however, the River Leen had the highest mean D 0 per sample (7.5,compared to 7.0 in Black Brook,and 4.8 in Saffron Brook), suggesting that it is not urbanised to the extent of only being able to support disturbance-tolerant taxa. A possible explanation is that the higher density of anthropogenic litter in the River Leen (4.2 items/m 2 ), and therefore greater proximity between materials, has enabled migration of macroinvertebrates between items, and increased community similarity. In Black Brook and Saffron Brook, the anthropogenic litter was more isolated within the riverbed, and so communities were more variable, as reported by Czarnecka et al. (2009).
Plastic and fabric macroinvertebrate communities were the most diverse and most dissimilar to those on rocks. The most obvious difference between fabric, plastic, and rocks, is that all fabric and most plastic items were flexible, suggesting that substrate flexibility may influence and structure macroinvertebrate distributions. The closest natural analogue for this type of habitat is macrophytes or organic detritus. It is possible that anthropogenic litter could provide a substitute for this type of habitat, which is commonly removed and thus absent from many urban rivers (Old et al., 2014;Walsh et al., 2005).
For instance, flexible anthropogenic litter could replicate the geomorphic role of macrophyte stands in lowland rivers (e.g. Folkard, 2011), where macrophytes locally reduce flow velocity and encourage the deposition of the fine sediments and organic detritus.
There were no macrophytes to sample in the urban river reaches studied, but taxa that are typically associated with aquatic vegetation were abundant on flexible materials. Of those taxa found only on anthropogenic litter, three tend to live on macrophytes: C. splendens, A. vortex, and B. contortus. C. splendens larvae have strong preferences for complex vegetation, which provides cover and plentiful prey species (Goodyear, 2000). The gastropods A. vortex and B. contortus, which were primarily found on plastic, generally inhabit plants which provide shelter, oxygenate flows, and provide a surface for biofilm development which they feed on by scraping (Boycott, 1936).
Other taxa also showed associations with flexible materials, although they were also on rocks and other types of anthropogenic litter. Asellus spp. and Gammarus spp. are omnivorous detritivores that live amongst and feed on decomposing plant material (Gledhill et al., 1993). These taxa are unable to consume anthropogenic litter but were still strongly associated with it in this study (especially plastic; 53 and 48%, and fabric: 31 and 30% of occurrences respectively), suggesting that the accumulation of fine organic matter around flexible materials could attract shredders and collectors. The fine sediment collected around flexible materials may also provide habitat for organisms that prefer soft sediments (e.g. Sphaeriidae, Oligochaeta, and some Chironomidae, all of which were recorded in greater numbers on plastic and fabric). However, if macrophytes were present in these rivers, it is possible that many of these taxa would preferentially inhabit this natural substrate, especially those taxa that directly feed on macrophytes. Past studies have recorded different macroinvertebrate communities on plastic and natural leaves, with lower macroinvertebrate abundance and diversity on plastic leaves (Hofer & Richardson, 2007;Quinn et al., 2000).
As well as interacting with fine sediments and organic matter, anthropogenic litter may also affect other macroinvertebrate food resources. We observed that biofilm, an important food for scrapers, had developed on the exposed surface of anthropogenic litter, as well as on rocks. If this biofilm is of a different quality or quantity to that which develops on rocks, this could influence macroinvertebrate distribution. For instance, it has been shown that more complex surfaces are more quickly colonised by biofilm and that this in turn will attract macroinvertebrates (Clifford et al., 1992). Distinct biofilm communities have been found to colonise different materials in marine environments (studies tested a range of plastic polymers and glass; Oberbeckmann et al., 2014;Pinto et al., 2019), meaning that biofilm quality could be affected by material composition.
However, a more comprehensive study of different materials in an urban river only found differences in biofilm composition and gross primary production between solid surfaces (plastic, glass, aluminium, and tiles) and soft organic materials (leaf litter and cardboard), rather than between all material types (Hoellein et al., 2014). Further research into a wider range of food resources and materials in rivers is necessary to expand our understanding of this topic.
The differences in community composition between substrates were driven by the taxa that differed most in abundance between anthropogenic and rock samples. These tended to be generalist taxa which are tolerant of poor habitat conditions. Anthropogenic litter samples were dominated by Chironomidae, Oligochaeta, and A. aquaticus, which dominate communities where there is organic enrichment and low dissolved oxygen concentrations (Armitage et al., 1995;Pennak, 1978). The corresponding high abundances of macroinvertebrates that feed on these taxa may be an indirect response to the increase in prey availability. This includes Limnophora spp. (which was only found on anthropogenic litter), the four leech species (especially E. octoculata and H. stagnalis; Elliott & Dobson, 2015), and P. flavomaculatus (Edington & Hildrew, 1995). All of these taxa were recorded more frequently on fabric or plastic than on rocks in this study. During sample collection it was noted that some plastic bags were associated with organic-rich fine sediments and accompanying signs of anoxia. In marine and estuarine environments, plastic bags have been linked to localised anoxia through preventing gas and nutrient exchange process at the sediment-water interface (Green et al., 2015), although this effect is moderated when they are regularly in motion (Clemente et al., 2018), which possibly explains why we still found diverse communities on such materials. The reduced difference between anthropogenic litter and rock diversity at higher Hill's numbers suggests that although the complexity of anthropogenic litter can support a diverse fauna, the taxa that are best able to exploit anthropogenic litter are those of lower conservation value, which can tolerate reduced habitat quality, and so numerically dominate the communities.

| Implications for river management
The rivers studied here were limited in habitat heterogeneity, as is typical of urban rivers globally (Walsh et al., 2005). In such cases, the habitat provided by anthropogenic litter may support biodiversity, both by providing complex and stable habitat for a wide range of organisms, and by representing a unique habitat distinct from rock substrates. Although rocks were the more abundant substrate in the rivers studied here, anthropogenic litter supported novel and more diverse communities, including five unique taxa. In urban rivers where it is not possible to restore instream large wood or macrophytes, anthropogenic litter may accidentally offset a lack of habitat diversity by providing an analogue for these natural substrates which are typically absent. In particular, these results suggest that flexible materials may replicate in-channel macrophyte habitat. It is possible that where anthropogenic litter provides a novel habitat structure, such as in sandy habitats (García-Vazquez et al., 2018), it may enable colonisation by non-native taxa that would not inhabit these environments under natural conditions (Tyrrell & Byers, 2007). Although this was not observed in the rivers studied here, it should be considered in future investigations, especially as urban areas are key sites for the establishment and spread of non-native and invasive species (Francis et al., 2019).
Anthropogenic litter removal and reduction of inputs should be the aim of management strategies, given that rivers are a key source of marine anthropogenic litter (Rech et al., 2014), anthropogenic litter is environmentally damaging (Rochman, 2015), and because of the negative societal and social impacts of littered waterways (Williams & Deakin, 2007). However, the results presented herein suggest that the removal of anthropogenic litter from urban rivers may not lead to biodiversity improvements in the immediate area and may even reduce biodiversity at the local scale. Anthropogenic litter removal efforts should therefore be carefully managed to reduce disturbance to the wider environment (Chapman & Clynick, 2006). This could mean preferentially removing some materials rather than others or improving habitat complexity following the clearance of anthropogenic litter to replace the removed habitat.
Importantly, these responses to anthropogenic litter suggest that even small-scale enhancement of substrate complexity could have positive effects on the local-scale biodiversity and ecological health of urban rivers (Francis & Hoggart, 2008). Understanding the types of habitat provided by anthropogenic litter and comparison to a wider range of natural substrates such as macrophytes and large wood, could help inform mechanisms to provide these functions using alternative and less damaging materials. Higgins, Vanessa Cutts, Philip Jagessar, and Harry Sanders. Thanks also to two anonymous reviewers and the associate editor for comments that substantially improved the clarity of the results presented in this paper.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are openly available in Open Science Framework at https://doi.org/10.17605/ OSF.IO/ K24T7 (Wilson et al., 2020)