On the other hand, the lipid-like component was dominant in aerosol collected from the Ross Sea (RS) in the Southern Ocean, where sea ice occurs. The relative abundance at each station is shown in Table 3.3. a) Molecular DOM compositions in seawater, sea ice and aerosol with air mass origins based on their return trajectories. The number of OTUs indicated in the figures means the dominant bacterial OTUs in seawater or in aerosol samples.
Introduction
Motivation & Background
It threatens human health by preventing the control of infectious diseases (Qiao et al., 2018); therefore, the study of ARG in the marine environment is imperative. Although deep learning models are black box models, they can improve performance by training from observational data (Andrychowicz et al., 2016) and simulating non-linear environmental phenomena. In particular, deep learning models have been widely used to improve prediction of hydrological models (Parmar et al., 2017; Sumi et al., 2012).
Contents of Each Chapter
Research Objectives
Literature Review
Dissolved Organic Matters (DOM) in Ocean-Atmosphere Interactions
Submicrometer sizes of organic matter play an important role in marine aerosol, although inorganic sea salt is dominant in the mass fraction with supermicrometer sizes in marine aerosol (Cavalli et al., 2004; O'Dowd and De Leeuw, 2007; O (Dowd et al., 2004). Especially, studies have increasingly reported that primary marine aerosol is highly related to biological activity and aerosolization from the sea can be influenced by the hydrophobicity of microorganisms (Michaud et al., 2017; O'Dowd et al., 2004; Sharoni et al. al., 2015). Furthermore, using the chamber system, studies have suggested that organic matter including microorganisms accelerate the production of primary aerosols in the ocean (Hoffman and Duce, 1976; Sharoni et al., 2015; Tseng et al., 1992).
Bacteria and their antibiotic-resistant genes (ARGs) in Ocean-Atmosphere Interactions
In seawater close to sewage treatment plants (sewage treatment plants), different types of gentamicin resistance genes were found in treated water from treatment plants (Heuer et al., 2002). Recently, studies have been published showing that ARG spreads with microplastics in the ocean and affects the marine ecosystem (Imran et al., 2019; Moore et al., 2020; Yang et al., 2019b). Meanwhile, bacterial communities in surface seawater vary in latitude, which is closely related to temperature (Ghiglione et al., 2012; Sul et al., 2013).
Deep Learning Model
Monitoring Dissolved Organic Matter (DOM) and Interaction between Ocean, Sea Ice, and
Introduction
It was assumed that using various techniques, including optical analysis and Orbitrap, the DOM molecular composition would exhibit unique physical, chemical and biological properties according to latitude-specific environmental characteristics. Furthermore, as ice is characterized in the Southern Ocean, it was compared with previous studies related to ice in the Southern region. In this study, molecular characterization of seawater, sea ice, and aerosol particles was observed to provide changes in ocean DOM by latitude, and more clear evidence of atmospheric aerosol DOM sources in the Southern Ocean through coupling with sea ice.
Materials and Methods
- Seawater and ice samples
- Aerosol samples
- Orbitrap analysis and data processing
- Hybrid Single Particle Lagrangian Integrated Trajectory (HySPLIT) and Statistical
In particular, the CDOM value at narrow wavelength intervals (254 nm) was selected in this study because it has been widely used as a proxy for terrestrial DOMs in previous studies ( Asmala et al., 2012 ; Zhao et al., 2016 ). The negative electrospray ion mode using a heated electrospray ionization source was operated under the following conditions: the flow rate of ultrapure nitrogen sheath gas was maintained at 7 a.u., the spray needle voltage was maintained at 4.5 kV, and the capillary temperature was 320 °C. The molecular formulas in the van Krevelen diagrams are divided into seven categories based on previous studies (Hockaday et al., 2009;.
Results and Discussion
- OM composition of ocean
- OM composition of sea ice
- OM composition of ambient aerosol
- Linkage between ocean, sea ice, and aerosol
In particular, Pusceddu et al. 1999) measured a lipid content of up to 68% in the biopolymer carbon flux collected in sediment cores beneath the Terra Nova Bay ice pack in late summer. The Ross Sea is considered the most biologically active region in Antarctica due to its remarkably high biological primary production in summer, when sea ice is declining (Arrigo et al., 2008). Previous literature has found a phenomenon in which a large amount of lipid substances (e.g., carboxylic acid) was released into the atmosphere during the sea ice-related process (Feltracco et al., 2021).
Monitoring Bacteria with their Antibiotic-Resistant Genes (ARGs) and Interaction between
Introduction
Antibiotic resistance occurs when bacteria change their genes in response to antibiotics and become resistant (Baquero et al., 1998). The untreated ARBs from WWTPs are typically discharged into the marine environment via combined sewage overflows (Zhang et al., 2020). Meanwhile, microbial communities harboring ARGs are diverse, and the abundance and diversity vary, due to differences in seawater temperatures that depend on latitude (Ibarbalz et al., 2019; Salazar et al., 2019; Sunagawa et al., 2015).
Materials and methods
- Sampling
- Chamber and airborne bacteria sampling
- DNA extraction and quantitative polymerase chain reaction (qPCR) for quantifying
- DNA extraction and next-generation sequencing for bacterial community analysis
- Statistical analysis
The volume of the chamber was 100 L, with each nozzle at the top for inlet of fresh air and circulating seawater, and three outlets for collection of aerosol samples. Meanwhile, Shannon's Index is a measure of the uniform distribution of species within a group, that is, diversity. In both networks, OTUs with a maximum relative abundance of more than 0.1% were selected and 254 of 322 OTUs in particle-attached bacteria and 166 of 262 OTUs in free-living bacteria were used for analysis.
Results & discussion
- Abundance and diversity of ARGs along the transect
- Environmental parameters affecting distribution of total ARGs
- Impact of intI1 on the abundance and diversity of ARGs
- Relationships between ARGs and bacterial communities
- Bacteria linkage between ocean and aerosol
For example, bla cluster genes were also found in the pristine Indian Ocean (Calero-Cáceres and Balcázar, 2019), Alaskan soil (Allen et al., 2009) and Arctic permafrost. In the Southern Ocean, PC2 together with distance from land and salinity contributed to the total ARGs (Fig. 4.5b). Therefore, this suggests that distance from land is a significant factor determining the abundance of total ARGs in the Southern Ocean.
ARGs were also found in pristine Arctic soil and may have spread by horizontal gene transfer (McCann et al., 2019). In the free-living bacteria network, 32 nodes and 31 edges were represented by 166 genus-level OTUs (Fig. 4.7a). The number of bacterial OTUs of particle-associated bacteria was approximately 1.2 times greater than that of free-living bacteria in the networks.
Both Bacteroidetes and Proteobacteria were the dominant group of bacteria in both particle-attached and free-living bacterial groups, which have a wide distribution in the ocean regardless of latitude (Sul et al., 2013). The abundance of OTUs belonging to Bacteroidetes in particle-attached bacteria, representing 46%, was 1.8 times higher than in free-living bacteria. At the genus level, the diversity of aerosolized bacteria was greater in the Southern Ocean than in the Pacific Ocean.
This was in contrast to the abundance results, which were higher in the Pacific Ocean than in the Southern Ocean.
Development of Deep Learning Models for Predicting Antibiotic-Resistant Genes (ARGs)
Introduction
A global estimate is that 50 million severe respiratory diseases can be attributed to polluted seawater every year (Byeon et al., 2011; Shuval, 2003). However, the validity of this approach has been questioned (Badgley et al., 2010; Office, 2015), as studies have reported a poor relationship between faecal indicator bacteria and other bacterial species/pathogens (Ishii et al., 2014; Zhang et al. ., 2016). Conventional methods of analysis are time-consuming; it takes an average of 5.2 days to verify incubation results (McAdam et al., 2012).
Current molecular biology techniques, such as quantitative polymerase chain reaction (qPCR), have been used to identify and quantify some ARGs (de Castro et al., 2014; Schmieder and Edwards, 2012). Although qPCR is simpler and faster compared to conventional techniques such as the culture method or traditional PCR (Kralik and Ricchi, 2017; Smith and Osborn, 2009), routine monitoring is limited by the high cost of qPCR analysis (Sakthivel et al., 2012). . Although multiplex PCR was developed to save time and effort by reacting several individual PCRs simultaneously, it is less accurate because it responds to non-specific amplification products (Jansen et al., 2011; Sakthivel et al., 2012).
Long short-term memory (LSTM), a type of recurrent neural network (RNN), is widely used as an effective environmental tool for water quality simulation and prediction due to its ability to extract features from time series data (lin Hsu et al., 1997). For example, Barzegar et al. 2020) recently used LSTM and hybrid LSTM models to predict lake water quality variables. Accordingly, it is considered a suitable neural network (NN) for predicting pollutant distributions and water quality over time (Wang et al., 2019; Wang et al., 2017).
We previously investigated the occurrence of ARGs at a combined sewer overflow (CSO) site in Gwangalli Beach in relation to rainfall and tides (Jang et al., 2021).
Material and methods
- Sampling location and period
- Data acquisition
- Data-driven modeling
The LSTM uses three types of gates named according to the function they perform (i.e. output gates (Γ), forgotten gates (Γ) and input gates (Γ)). We applied a one-dimensional (1D) CNN to the output sequence of the LSTM model to improve the prediction performance by further extracting features from the LSTM output (Figure 5.4). The size of the weight matrix determines the total number of parameters in the CNN, while the size of the convolution operation is based on the size of the kernel.
-CNN hybrid model structure for time step 1 to t. a) and (b) indicate the input and LSTM layers of the conventional LSTM, respectively. Row inputs (?@=) calculated from the IA layer are used instead of the original LSTM inputs for prediction. To account for this, many researchers have proposed an “attention” mechanism that forces the NN to focus on a certain part of the input ( Luong et al., 2015 ).
The size of the weight matrix is directly proportional to the LSTM or CNN learning capacity. In the Bayesian optimization method, this surrogate function is optimized instead of the actual objective function. These figures illustrate how the validation MSE of the models decreased during the optimization process.
From the 218 ARG samples, we used 150 to train the NN, and the remaining 68 were used to evaluate the model performance.
Results and discussion
- Spatial distribution of bacterial contamination
- Factors affecting temporal variations in ARGs
- Effect of rainfall on diversities and diversity of ARGs
- Hyperparameter optimization
- ARG prediction results
- Important variables for ARG prediction
In the LSTM-CNN model, the ReLU activation function was the most efficient for both LSTM and the CNN layer. Training and test losses of LSTM-CNN model in (a–d) single ARG and (e) multi-ARG predictions. Our LSTM-CNN model differs in that the LSTM output was applied to the CNN.
Scatter plots of observations and model predictions with trend lines for individual ARG LSTM-CNN predictions. Distribute plots of observations and model predictions with trendlines for multi-ARG LSTM-CNN predictions. All R2 values, except aac(6')-Ib-cr in training, were lower than those of LSTM-CNN for single ARG predictions (Table 5.5e).
In the case of predicting multiple ARGs, IA-LSTM could predict multiple ARGs simultaneously with higher accuracy than LSTM-CNN. Notably, the performance of NN for blaTEM with IA-LSTM was better than that of LSTM-CNN during training and testing. However, due to the black-box effect, conventional hybrid LSTM and LSTM-CNN models are unintelligible regarding the importance of input variables.
As illustrated in the time series results, both LSTM-CNN and IA-LSTM respond to the delayed ARG release after the rainfall (Fig.
Concluding Remarks
Metagenomic profiles of antibiotic resistance genes (ARGs) between human-impacted estuarine and deep ocean sediments. Occurrence of antibiotic resistance genes and bacterial pathogens in water and sediments in urban recreational waters. Metagenomic characterization of antibiotic resistance genes in full-scale reclaimed water distribution systems and corresponding drinking water systems.
Fate of antibiotic resistance genes in mesophilic and thermophilic anaerobic digestion of chemically enhanced primary treatment sludge (CEPT). Occurrence and spatial distribution of antibiotic resistance genes in the Bohai Sea and Yellow Sea, China. High-throughput profiling of antibiotic resistance genes in urban park soil with reclaimed water irrigation.
Emergence of antibiotic resistance super genes in the lower reaches of the Yangtze River in China: prevalence and antibiotic resistance profiles. The role of the natural environment in the emergence of antibiotic resistance in gram-negative bacteria. Evidence for co-selection of antibiotic resistance genes and mobile genetic elements in metal-contaminated urban soils.
Variation in the abundance and diversity of antibiotic resistance genes and bacterial communities in the western and southern Pacific Ocean.