Because of their unique physico-chemical properties, nanoparticles (NPs) are widely used in the agri-food industry as agrochemicals for controlling pests and the delivery of active ingredients, nanosized ingredients, and additives (Chaudhry et al., 2008; Polo et al., 2011; Chen et al., 2016; Grillo et al., 2016). Another interesting and growing nanomaterial application is for functionalizing food processing surfaces and for packaging, mostly to improve mechanical and antimicrobial properties (Dainelli et al., 2008).

Given these widespread applications, NP exposure represents a potential toxicity risk for human health. Of special interest is the effect of NPs on human gut microbiota, considering the range of consumer goods that can be intentionally, or accidentally ingested. However, most of the literature regarding NP effects on human health concern lung cell in vitro cultures (Geary et al., 2016). Only a few animal studies have been made on NP gastrointestinal intake and impact, insufficient to allow a clear safety assessment of ingested NPs (Tran and Chaudhry, 2010; Bergin and Witzmann, 2013; Jones et al., 2015; Pietroiusti et al., 2016) or a better understanding of their impact on the intestinal ecosystem (Fröhlich and Fröhlich, 2016).

A major reason for the scarcity of relevant literature on NP effect on the gut ecosystem is related to the lack of effective and simplified model systems for studying the nature of these complex interactions. Furthermore, most of the studies were conducted with relatively high (bactericidal) NP concentrations, and the effect of sub-lethal exposure (low concentrations) is poorly understood. Thus, there is a critical gap in current knowledge, low NP concentrations (rather than high), following dilution and dispersion, being expected to predominate in both the food chain and the gastrointestinal system (Benn and Westerhoff, 2008; Colman et al., 2013).

In addition, there is growing recognition that intestinal microflora exists as biofilm differing in characteristics from their planktonic counterpart (Donelli et al., 2012). Despite the evidence of human microbiome as a biofilm on gut mucosa (Donelli et al., 2012; Donaldson et al., 2016), the complexity of such intestinal biofilm and its interaction with low NP concentrations (the actual human consumption) and epithelial intestinal cells is still largely unknown.

In order to lay the foundation for understanding the complex interplay among NPs, gut biofilm and its host, the physiological response of a simplified system composed by Escherichia coli mono-species biofilm and Caco-2 intestinal cells exposed to sub-lethal concentrations of silver nanoparticles (AgNPs) was investigated. E. coli was chosen as representative of the gut biofilm, being the best characterized model microorganism and the most frequent commensal aero-anaerobic Gram-negative bacillus of the vertebrate gut (Tenaillon et al., 2010). AgNPs are the most commonly used metal nanoparticles in different applications (Zhang et al., 2016a).

To simulate the anaerobic and aerobic compartments required for bacteria and intestinal cells growth, a semi-batch model based on a transwell permeable support was developed. Interaction between the two compartments was obtained by exposing Caco-2 intestinal cells to the metabolites secreted by E. coli biofilm after its exposure to AgNPs.

To the best of the authors’ knowledge, this study is the first to investigate the effect of AgNPs on Caco-2 cells that takes into consideration previous AgNP-intestinal biofilm interactions, and at concentrations mimicking real human exposure.

In our study, intestinal E. coli biofilm was grown in a transwell device under anaerobic conditions, and the interaction between 1 μg/mL AgNPs, biofilm and Caco-2 cells was assessed by putting into contact the basolateral media containing the biofilm metabolites with Caco-2 cells grown with oxygen. To date, many studies have focused on the direct toxic effects of NPs on exposed cells and organisms, which may not reflect the real situation observed in vivo (Fröhlich and Fröhlich, 2016). In order to mimic the real intestinal epithelium exposure conditions, Caco-2 cells were treated with basolateral media from the microbial barrier, in which metabolites, produced by biofilm after its interaction with AgNPs, are present. In addition, experiments were conducted using a concentration of AgNPs (1 μg/mL) representative of their actual food-related ingestion (Gottschalk et al., 2013; Kuorwel et al., 2015; Fröhlich and Fröhlich, 2016).

A variety of in vitro models have been proposed to simulate and study the microbiome of the human gastrointestinal system environment. These can be divided into (i) batch and (ii) dynamic systems (Marzorati et al., 2014). Batch systems usually consist of small reactor vessels or test tubes modeling one single segment of the gastrointestinal tract. Batch experiments offer a very easy and flexible screening tool (Venema and Van den Abbeele, 2013), but they do tend to over-simplify the complexity of the gastrointestinal process (Marzorati et al., 2014). Moreover, they are far from physiological conditions, and suffer, for example, from the accumulation of microbial metabolites, thus inhibiting further microbial activity so that incubation usually cannot be extended to periods over 24 h (Venema and Van den Abbeele, 2013). Instead, continuous dynamic models allow in-depth studies of the biological processes in the gut under more representative physiological conditions (Venema and Van den Abbeele, 2013; Marzorati et al., 2014). Some dynamic systems, e.g., SHIME (Van den Abbeele et al., 2010), EnteroMix (Makivuokko et al., 2005), the Lacroix model (Cinquin et al., 2006), and TIM-2 (Minekus et al., 1999) are composed by two or more chambers connected by vessels or membranes that simulate how the lower or complete digestive tract allows the continuous flux of fluids. These systems thus reproduce the main parameters of human digestion as accurately as possible, and the experimental conditions can be tuned finely, and often, under computer control (Cordonnier et al., 2015). However, such sophisticated systems require not only complex and expensive equipment, not always available on the market, but also long times to set up the experimental conditions, as well and a high a level of expertise and competence not commonly found in all laboratories.

In the present study, the plastic transwell permeable support was used to address the issue of providing an anaerobic atmosphere to the luminal gut microbiota, and an aerobic environment to the intestinal mucosa, using semi-batch working conditions and two independent compartments. Transwell devices are well-known systems, commonly used for in vitro tests to simulate the intestinal absorption of drugs and other substances, included nanoparticles (Hilgers et al., 1990; Williams et al., 2016), and to grow biofilm under different physiological conditions (Standar et al., 2010; Kim et al., 2012; Wu et al., 2015; Wang et al., 2016). Compared to batch static models, transwell systems have the additional advantages of fine control of experimental conditions in different compartments, constant nutrient support and the possibility to collect the metabolites of cultured cells in long term experiments. Moreover, in contrast to dynamic systems, they are cheap, widely available on the market, easy to use, do not require a high expertise, and simplify biofilm formation and quantification, making experiments from among different laboratories more comparable. Thus, the proposed model offers the advantage of a flexible workflow, allowing independent studies in compartements with different atmosphere conditions, i.e., aerobic conditions for Caco-2 cells and anaerobic ones for human gut biofilm. Despite the precise choice to use E. coli as representative of the intestinal microbiota, the present system can be adapted to host different microorganisms, including Bacteroidetes and Firmicutes. We would like to point out that the designed laboratory model system is not intended to be a miniaturized version of human gut ecosystem. Rather, the purpose of the developed laboratory model system was to simplify the complexity of the intestinal ecosystem, so that it can be more easily understood. If we cannot accurately predict the behavior of a simplified laboratory system, it is unlikely we can understand enough to make predictions of natural systems (Jessup et al., 2004).

The influence of anoxic conditions on biofilm response against AgNP exposure was also studied. According to the literature, AgNPs display different properties without and with oxygen (Yang et al., 2013). Thus, it is reasonable to expect that NPs in an anaerobic environment might impact biofilm behavior differently from aerobic systems. It is also worth mentioning that, despite the fact that the fate of AgNPs on microorganisms in aerobic atmosphere has been studied to some degree, bacterial response to AgNPs under anaerobic conditions has been far less investigated, and is less understood (Gitipour et al., 2016). In line with these considerations, we performed experiments simultaneously in anaerobic and aerobic conditions and the effect of the lack/presence of oxygen in AgNP response was assessed.

In anaerobic conditions E. coli biofilm behavior differed, depending on the type of AgNP treatment. Indeed, when grown in the presence of AgNPs, the biofilm, for the first 72 h, showed a 2.3 ± 0.3 fold increase in viable cells compared to the control, though no reduction in oxidative stress was recorded. Indeed, the promotion of biofilm formation by exposure to sub-lethal levels of AgNPs has already been reported (Königs et al., 2015; Yang and Alvarez, 2015). This has led us, the authors of this manuscript, to speculate that stimulation of biofilm growth is an adaptive response to face the persistent presence of AgNPs, the bacteria responding by adopting a more efficient biofilm lifestyle. The stimulation of biofilm formation as a way to tolerate stress is a well-known adaptive behavior. Indeed, the presence of sub-lethal low concentrations of several antibiotics and chemicals has been shown to induce biofilm formation in a variety of pathogenic bacterial species, enabling such species to continue to exist and resist antibiotic or chemical treatment (Perrin et al., 2009; Kaplan, 2011; Machado et al., 2011).

After the 72 h period of treatment, the oxidative stress level within the biofilm appeared much increased, though the number of viable cells within the treated biofilm was still comparable with that of the control. It is reported that AgNPs can enter cells, inactivating cellular enzymes and generating endogenous ROS by interaction with biomolecules (Yang et al., 2013; Ikuma et al., 2015; Königs et al., 2015; Yang and Alvarez, 2015; Qayyum and Khan, 2016). Furthermore, it is reported that cells use ROS as a signal or cue to adapt to a changing environment (Čáp et al., 2012). Indeed, a connection between biofilm formation and oxidative stress has already been observed (Kaplan, 2011; Čáp et al., 2012; Villa et al., 2012; Cattò et al., 2015, 2017; Jakubowski and Walkowiak, 2015; Gambino and Cappitelli, 2016).

However, when AgNPs were used to treat established biofilm the outcome was different. After adding the AgNPs there was, together with increased ROS levels, the cessation of biofilm development, suggesting the occurrence of a strong cellular oxidative imbalance that the bacteria were unable to overcome efficiently in 24 h, e.g., the activation of drug resistance sessile mechanisms. As reported in recent work published by our research group, 1 μg/mL does not affect E. coli viability or its planktonic growth (Garuglieri et al., 2016). Thus, the data suggest an anti-biofilm mechanism subtler than a simple killing activity. Since a 1 μg/mL AgNP concentration does not affect bacterial life, the best microbial strategy might be to escape from adverse conditions, and not to persist in maintaining a biofilm lifestyle (Villa et al., 2010, 2012; Cattò et al., 2017).

It has been reported that biofilm simultaneously harbors cells in multiple states (e.g., growing, stress-adapted, dormant, inactive) and such a specific physiological status in microbial cells is an important contributing factor to antimicrobial tolerance (Williamson et al., 2012; Stewart, 2015). We found that both treated and untreated biofilm consisted of a predominant spatially distinct subpopulation of GFP-positive cells actively synthesizing new proteins. Indeed, the controls displayed a greater number of active growing cells at the air-biofilm interface than in the deeper biofilm regions where the cells were by far fewer and with very slow growth. Indeed, previous research on Pseudomonas aeruginosa reported the same pattern distribution, and showed that cells at the top of biofilm generally have a high mRNA abundance for genes involved in general metabolic functions, while the mRNA levels for these housekeeping genes are low in cells at the bottom of biofilm (Williamson et al., 2012). In biofilm grown for 96 h with AgNPs, the pattern distribution appeared completely inverted, with the inactive portion of cells on the top, at the interface with air, and the active one at the bottom of the biofilm, exactly where AgNP exposure took place. Indeed, a metabolically active status allows bacteria to sense their environment and actively respond to the presence of a stress factor (Stewart et al., 2015).

When the AgNP treatment involved 72 h-old established biofilm, the cells displayed a distribution pattern similar to the control. These results corroborate the hypothesis that, at this stage, where oxidative stress is far too high, the cells activate alternative defense responses rather than adaptive mechanisms to survive in the sessile life mode.

The results indicate that, in the presence of oxygen, the AgNPs only weakly influenced the biofilm, whereas the combination of oxygen restriction and AgNPs strongly modified the biofilm development. The situation was even worse when the AgNP treatment involved an anaerobic established biofilm. It has been reported that the anaerobic E. coli MG1655 is slow growing and patchy compared to aerobic biofilm, but some features, like the production of extracellular polymeric substances, remain unchanged (Bayramoglu et al., 2017). A closer inspection of the mRNA data reveals that essential cell processes are attenuated in anaerobic biofilm, including protein synthesis, information transfer, cell structure, regulation, and transport, suggesting that the lack of oxygen imposes severe stress on mature biofilm and this could affect some important adaptive mechanisms (Bayramoglu et al., 2017).

Our results, obtained with the in vitro model set up in this study, sustain a protective effect of biofilm against the genotoxic effect exerted by a 1 μg/mL AgNP concentration on Caco-2 cells. Nevertheless, at 72 h the results showed an increase in DNA damage regardless of AgNP presence (C72 and NP72): damage that can be attributed to unidentified metabolites produced by the bacterial biofilm and able to exert cell genotoxicity.

Several studies have reported the induction of oxidative stress as a conceivable molecular mechanism of AgNP toxicity (Dubey et al., 2015). However, the in vitro toxicological studies present in the literature suggest that many factors could influence the results, such as AgNP properties (size, concentration, type of surface-coating) and exposure protocols (cell line, cells in adhesion or suspension, state of confluency and differentiation, type of medium, time of exposure) (Zhang et al., 2016b). However, the notable differences found in the in vitro experimental protocols makes it difficult to compare the results obtained in our work. Let us consider only the studies on Caco-2 cells exposed to a comparable range of AgNP concentrations: Sahu et al. (2014b) demonstrated that 20 nm AgNPs decreased the viability of Caco-2 cells, starting from 10 μg/mL, but was not able to exert a genotoxic effect, estimated as presence of micronuclei by flow cytometry, up to 15 μg/mL. However, in a previous study by the same research group, the dsDNA, a measure of cell proliferation and indicative of cellular DNA damage, was significantly reduced starting from 1 μg/mL (Sahu et al., 2014a). Recently, Chen et al. (2016) focused their interests on the toxicological effects of low concentrations of AgNPs (0.7–2 μg/mL) in Caco-2 cells. Their results led them to conclude that, when Caco-2 cells were exposed to 1 μg/mL AgNPs for 24 h, intracellular ROS levels increased significantly, while the mitochondrial membrane potential that indicates mitochondrial damage appeared seriously reduced. Moreover, at 1 μg/mL AgNPs the cell membranes appeared damaged and the NPs began to induce apoptosis (Chen et al., 2016). In accordance with those studies, the present data sustain the lack of a cytotoxic effect of 1 μg/mL AgNPs, as observed by the cell viability of TNP vs. TSB, when determining the significant increase in induced DNA oxidative damage. Using the proposed experimental gut model, the preliminary interaction of 1 μg/mL AgNP concentration with biofilm significantly reduced the genotoxicity of the basolateral media to levels consistent with the basolateral collected from the control biofilm (NP24 vs. C24 and NP48 vs. C48, p > 0.05).

We have come to the conclusion that the results obtained with the in vitro, simplified model set up in this study clearly sustain a protective biofilm effect against the genotoxic effect exerted by a concentration of 1 μg/mL of AgNPs (TNP). Future work will be devoted to deeply investigate the protective role of complex multi-species biofilms on intestinal epithelial cells.

Overall, the obtained data raise potential safety concerns associated with the use of AgNPs in agri-food applications. Despite Caco-2 cells not being significantly affected by AgNPs after bacteria mediation, E. coli biofilm appeared to be seriously impacted when treated with AgNPs in anaerobic conditions. Earlier experiments have revealed that environmentally relevant concentrations of NPs affect short fatty acid production, hydrophobicity, sugar content of the extracellular matrix and electrophoretic mobility of the intestinal microbiota, leading to changes in the microbiota community stability (Taylor et al., 2015). Deviations in the intestinal microbiome have been associated with dozens of diseases, varying from inflammatory bowel disease to diabetes and colorectal cancer (De Vos, 2015). In this research, the important role of biofilm in protecting epithelial cells from AgNP genotoxic effects has been demonstrated. It is likely that changes in the microbial ecosystem could leave intestinal cells more exposed to negative genotoxic stressors ingested with food, e.g., AgNPs, thus increasing the level of cellular DNA oxidative damage and creating a potential for serious AgNP associated diseases.

EG and EM performed the experiments and wrote the manuscript. CC and FV conceived the experiments, participated in discussions, and improved the manuscript. FC and DE coordinated the collaboration of the authors. All authors read and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.