Systems BioEcology The problem: An outdated and inadequate environmental modeling approach Environmental engineers are charged with integrating current scientific knowledge into models of environmental systems so that predictions can be made. Historically, this has been relatively easy. To develop a model of oxygen depletion downstream of wastewater discharges in the Ohio River, Streeter and Phelps (1925) took a rate constant for oxygen consumption by sewage from the lab and simply plugged it into a transport model. This approach was consistent with the scientific knowledge available at that time (i.e. “sewage consumes oxygen”). Since then, there have been tremendous advances in our basic scientific knowledge, especially in the areas of molecular biology, biochemistry and genetics. It is incredible, but we are truly at the brink of understanding life at the molecular level. However, there has been very little change in our environmental modeling approach. Whether they are oxygen-consuming heterotrophs or toxin-producing cyanobacteria, microbes are essentially considered to be chemical molecules – intracellular mechanisms are ignored. They are quantified as a chemical concentration (e.g. mg C/L, ug Chl. a/L). Nutrient uptake, growth, chemical transformation, etc. are considered to be chemical reactions between the microbe and other molecules. We are presently using concepts from chemistry to model biology. There are a number problems with this approach. First, by lumping microbes into concentrations, we lose the ability to resolve population heterogeneity. Also, it prohibits us from simulating intracellular states and mechanisms, which is where most of the basic science progress is being made. This is a huge disconnect between science and models. We have to find a way to integrate current science into our models. When I was developing a model of arsenic transformation by phytoplankton during my doctoral studies at Columbia (Hellweger et al., L&O, 2003; Hellweger and Lall, ES&T, 2004; Hellweger, Appl Organomet Chem, 2005) I was thrilled to find that the gene for arsenate reduction had been identified and its protein structure resolved – yet I could not figure out how to use this information in my chemistry-type model. There was no such thing as a gene or protein in my model (see figure below). After joining the faculty at Northeastern in 2004, I realized that this is a widespread fundamental problem. Our modeling approach is completely outdated and entirely incompatible with current science. The goal of my research program is to make environmental models compatible with current scientific knowledge.
Figure. How can today’s scientific advances be incorporated into yesterday’s chemistry-type models? (A) Structure of ArsA arsenic efflux pump protein (from literature), (B) domains of ArsA (from literature) and (B) equation from the OldLace arsenic transformation model developed during my doctoral research. The solution: “Systems BioEcology” The first step is to realize that microbes are not chemical molecules but complex systems with dynamic internal states and mechanisms. Unlike the traditional chemistry (e.g. Monod) approach, “systems biology” explicitly recognizes this. The second step is to realize that ecosystems themselves are not simple mixed chemical reactors but also complex systems with many populations each made up of individuals with different life histories and states, adaptive behavior, lifecycles and local interactions. Unlike the traditional population-level (e.g. Lotka-Volterra) approach, the “systems ecology” or individual-based modeling (aka agent-based modeling, ABM) approach explicitly recognizes this. Both approaches consider systems to be made up of many lower level entities, and that the system-level behavior emerges from the cumulative behavior and interaction of these low-level entities. The combination of these two approaches – “systems bioecology” – is at the heart of my research program. In Systems BioEcology, ecosystem behavior emerges from the cumulative behavior and interaction of microbes, who’s behavior themselves emerges from the cumulative behavior and interaction of genes and proteins (see figure below). This approach allows us to go from the scale of molecular biology to that of the ecosystem. We can, for example, perform an ecosystem-scale gene knockout experiment in silico to understand how gene X affects the density of population Y and concentration of chemical Z within an ecosystem.
Figure. Systems BioEcology is the combination of Systems Biology (A) and Systems Ecology (B)
Resonating circadian clocks enhance fitness in cyanobacteria in silico
Many different organisms, including the unicellular cyanobacterium Synechococcus elongatus PCC 7942, have a circadian clock. Advances in molecular biology and biochemistry are rapidly identifying and characterizing the components of the clock. However, its function inside the cell is not well-understood, and the mechanism by which it affects fitness is unknown. This paper presents a simulation modeling study that combines existing systems biology models of basic intracellular mechanisms (e.g. DNA replication), photosynthesis, post-translational oscillator and transcriptional/translational feedback loop with an agent-based systems ecology model of individual cells (a new approach called “systems bioecology”). The model is calibrated against and can reproduce the major observed features of the clock, including entrainment to light/dark (LD) cycles, free-running rhythms, phase shifts by dark pulses and KaiC protein dynamics. The observed decrease of the free-running period (FRP) with light intensity (Aschoff’s Rule) emerges unexpectedly in the model. This behavior of the model can be eliminated by changing how the ribosome concentration varies with the growth rate, consistent with the observed behavior of an ldpA mutant, which suggests a role for that gene. The model predicts the observed fitness advantage of wild-type and period mutants when the FRP and LD periods match. The mechanism in the model is higher amplitude (resonance) of the clock, which leads to stronger expression of photosynthesis genes and growth rate. This may be the mechanism actually responsible for the fitness effect in this case, but it does not explain the existence of the circadian clock from an evolutionary perspective. Agent-Based Modeling of Cyanobacteria
The cyanobacterium Anabaena flos-aquae and many other phytoplankton species have a complex life cycle that includes a resting stage (akinete). We present a new agent-based (also known as individual-based) model of Anabaena that includes the formation and behavior of akinetes. The model is part of a coupled Eulerian–Lagrangian model and can reproduce the main features of the observed seasonal and interannual population dynamics in Bugach Reservoir (Siberia), including an unexpectedly large bloom in a year with low nutrient concentrations. Model analysis shows that the internal loading of phosphorus (P) due to germination from the sediment bed is ,10% of the total input. However, most of the long-term nutrient uptake for Anabaena occurs in the sediment bed, which suggests that the sediment bed is not just a convenient overwintering location but may also be the primary source of nutrients. An in silico tracing experiment showed that most water column cells (~90%) originated from cells located in the sediment bed during the preceding winter. An in silico gene knockout experiment (akinete formation is prohibited) showed that the formation of resting stages is of critical importance to the survival of the population on an annual basis. A nutrient-reduction management scenario indicates that Anabaena densities increase because they are less sensitive to water column nutrient levels (because of the sediment bed source) than other species. Photosynthesis genes in cyanophages
Several viruses infecting marine cyanobacteria carry photosynthesis genes (e.g. psbA, hli) that are expressed, yield proteins (D1, HLIP) and help maintain the cell’s photosynthesis apparatus during the latent period. This increases energy and speeds up virus production, allowing for a reduced latent period (a fitness benefit), but it also increases the DNA size, which slows down new virus production and reduces burst size (a fitness cost). How do these genes affect the net ecological fitness of the virus? Here, this question is explored using a combined systems biology and systems ecology (“systems bioecology”) approach. A novel agent-based model simulates individual cyanobacteria cells and virus particles, each with their own genes, transcripts, proteins and other properties. The effect of D1 and HLIP proteins is explicitly considered using a mechanistic photosynthesis component. The model is calibrated to the available database for Prochlorococcus ecotype MED4 and podovirus P-SSP7. Lab- and field-scale in silico survival, competition and evolution (gene packaging error) experiments with wildtype and genetically engineered viruses are performed to develop vertical survival and fitness profiles, and to determine the optimal gene content. The results suggest that photosynthesis genes are non-essential, increase fitness in a manner correlated with irradiance, and that the wildtype has an optimal gene content.
Agent-Based Modeling for Biogeochemistry
Cultural eutrophication is an important environmental problem in the US and other countries. The effective management of a water body’s trophic state requires an accurate biogeochemical (water quality) model. Present models use a lumped-system modeling (LSM) or population-level modeling (PLM) approach that assumes average properties of a population within a control volume. For modern biogeochemical models that formulate phytoplankton growth as a nonlinear function of the internal nutrient concentration (e.g. Droop kinetics), this averaging assumption can introduce a significant error. Agent-based modeling (ABM) or individual-based modeling (IBM) does not make the assumption of average properties and therefore constitutes a promising alternative for biogeochemical modeling. However, to date, ABM for biogeochemical modeling remains largely unexplored. The goal of this research project is to advance the science of agent-based modeling for biogeochemistry. The animation compares the ABM (IBM) and LSM (PLM) approaches for a simplified scenario, consisting of a time-variable point source discharge into a river. The model-predicted nutrient and phytoplankton concentrations are significantly different for the IBM and PLM approaches. The difference is a result of intra-population variability in nutrient status. In the IBM, individual cells have different life-histories and therefore different nutrient contents. Some individuals have accumulated nutrients in excess of their immediate growth requirement (luxury uptake), and some have less nutrients. If the cells with lots of nutrients would “give” some of their nutrients to the cells with few nutrients, the population-average growth rate would increase (as in the PLM results). One way of looking at the situation is as capitalism (IBM) vs. communism (PLM).
Artificial Sewer Networks for Hydrologic Simulations
Simulating urban hydrology using actual sewer networks can be tedious and even practically impossible for large and/or old areas, especially when considering high spatial resolution requirements of physically-based models. Is there an alternative to using actual sewer networks? What if the general presence of watershed features at a certain scale and configuration is important (e.g. the model needs to have gutters at a certain density connected to catch basins), but their exact location and exact configuration within the watershed is not important. If that is true, then we could generate an artificial network based on some user specified characteristics (e.g. land use, slope, …) and use that as input to the model. The idea is that, although the artificial network may look different, it will produce the same results when applied in a model. In this research project, we are working on developing and testing algorithms for generating artificial sewer networks for hydrologic simulations in urban hydrology. The animation shows a sequence of artificial sewer networks generated for the highly-urbanized Faneuil Brook sub-basin in Boston. The artificial networks were created using the Artificial Network Generator (ANGel), which uses the dendritic and space-filling ‘Tokunaga’ fractal tree geometry. We are presently using actual and artificial networks as input to the SWMM model and comparing the model results based on the flood hydrograph and various other parameters (e.g peak flow). |
