Studies in EcoTrons will be increasing in the near future and will provide unprecedented insights into ecosystem functioning; for example, Roscher et al. found that the functional composition of communities is key in explaining carbon assimilation in grasslands. Mesocosms, which we call EcoPODs, are smaller versions of EcoTrons with higher experimental throughput that bridge laboratory and field studies . Existing EcoPODs have a footprint of approximately 2.1 m2 and can be filled with up to 1.23 m3 of soil of 0.8 m in depth . Using the EcoPOD lysimeter technology, intact soil monoliths can be retrieved from the field and studied under controlled conditions in the laboratory. The above ground portion is approximately 1.5 m tall and, therefore, allows the study of a number of different plants in soil with macro and microorganisms in the context of environmental changes. The contained nature of EcoPODs allows accurate mass balance calculations . EcoPODs allow precise conditioning of above and below ground temperature and moisture and, therefore, can simulate seasonal changes and enable short as well as long term experiments. They are equipped with state of the art sensor technology allowing in situ measurements of key environmental parameters,growing vegetables in vertical pvc pipe activities of organisms, and ecosystems at micrometer to meter scales.
EcoPODs can be equipped with multi and hyperspectral cameras that track plant biomass and physiological states. In conjunction with highly controlled physical and chemical conditions, researchers will be able to track the microbial activity within the system using a variety of genomic tools, including DNA or RNA shotgun metagenomics, proteomics, and metabolomics. This will facilitate tracking of microbial recruitment and activity across all life stages of the plant and can simulate seasonal changes. Broadly, this system can be used for fundamental research questions about biogeochemical cycles and the role of biodiversity in ecosystem processes, as well as applied studies that include biological or chemical components that require increased safety clearance and cannot be easily tested in the field. Because soil ecosystem and phytobiome experiments increasingly rely on in situ sensing over time, EcoPODs can also serve as a test bed for novel and improved sensing capabilities. Complimenting approaches to develop more field relevant laboratory growth systems are composed of one or more single plant chambers such as RootChips , GLO ROOT , EcoFAB , and other systems that enable detailed characterization. For example, RootChips systems provide a high throughput system for rhizosphere imaging, the GLO ROOT systems enable direct imaging of root architecture within soils, and EcoFABs are “fabricated ecosystems” that are aimed at creating model ecosystems on par with the model organisms used for genetic and biological studies. EcoFABs comprise a chamber, biological and abiotic components , and any measurement technologies .
EcoFABs allow real time microscopy for high resolution imaging of plant root architecture and are currently designed to provide sufficient materials for metabolomic, geochemical, and sequence based analyses. They are made using widely accessible 3D printing technologies to fabricate controlled microbiome habitats that can be standardized and easily disseminated between labs . This approach provides flexibility that enables scientists to add or change variables while monitoring microorganisms and their interaction with plants.EcoFABs are also envisioned to facilitate standardization of phytobiome research because construction materials are cheap and construction instructions are available . Analogous to medical drug testing pipelines, which generally begin as high throughput laboratory screens and are gradually scaled up to relevant mammal models and, finally, to human clinical trials, we envision phytobiome research studies to similarly follow a throughput versus relevance gradient from EcoFABs to EcoPODs and, finally, to field studies . Although this suite of fabricated ecosystems is not aimed at simulating the real world, the enhanced control over abiotic and biotic factors in these experimental platforms enables plant root microbiome interaction studies that are not possible in field experiments because fields generally display greater complexity and unpredictability or do not allow for manipulations. Thus, use of fabricated ecosystems can reveal important correlations and causations of individual metabolic reactions as well as biogeochemical cycles. Challenges that have been encountered or are foreseeable include the relatively short experimental durations that can be executed in EcoFABs as well as EcoPODs due to the size limitations of the respective platforms and because of potential increases in parasite pressure as a result of air and water flow limitations.
On the other hand, EcoTrons are not set up for quick turnover experiments and require expensive infrastructure to start and end experiments. Although insights obtained from greenhouse experiments have often not been replicable in the field, we expect that EcoFAB can serve as a reproducible system, in which microscopy and metabolomics can be applied to low complexity microbiomes in the context of plant roots. Data obtained from individual microorganisms can inform microbially based biogeochemical models, as discussed below. We expect EcoPODs and EcoTrons to facilitate in situ sensing,vertical greenhouse climate manipulations, and deep soil monolith access. Links and extrapolations among fabricated ecosystems and the field can be achieved by generating and testing hypotheses across platform scales. For example, field observations may be tested under replicable conditions in EcoTron or EcoPOD and promising microbial candidates could be isolated and further studied in EcoFABs. A reverse workflow is also imaginable, where promising microbial isolates or plants resulting from EcoFAB experiments may be tested in EcoPOD or EcoTron before being potentially released into field experiments. Furthermore, extrapolations could be testable beforehand by taking advantage of archived datasets from sources such as long term observatories, including Neon . Generally, challenges for extrapolating results of these fabricated ecosystems to realistic field conditions could be presented by the limited complexity in these laboratory systems; for example microbial isolates often perform predictably under laboratory conditions but may be inactivated by night temperatures or competitors. There is still a number of unknown unknowns which may significantly affect plant performance, microbial community dynamics, and soil nutrient cycling, and which vary from ecosystem to ecosystem, hence resulting in a disconnect between studies conducted in the laboratory versus in the field. Other challenges are presented by natural climate variability in the field and the uncertainty in climate change predictions, which are significantly affected by socioeconomical drivers . Although laboratory experiments may be conducted based on historic field data or even in tandem with real time field data measurements—for example, using sensor platforms coupled to edge computing —results may have limited applicability under future climate scenarios. However, this is also true for reproducibility of field experiments in general. Studying plant microbiome interactions and soil processes under defined conditions can assist in the identification and evaluation of such unknown unknowns which, in turn, will improve applicability of laboratory results to the field.Microbial communities found on healthy plants are incredibly taxonomically diverse and include bacteria, archaea, fungi, oomycetes, algae, protozoa, nematodes, and viruses.
This microbial complexity makes it impossible to definitively establish causal relationships between plant and microbial genotypes and phenotypes as well as environmental factors. Instead, representative synthetic communities of defined complexity enable systematic bottom up approaches in gnotobiotic systems under controlled and reproducible conditions to determine causal relationships . In order to systematically test plant microbial community dynamics and functions in relation to the chemical composition of the surrounding environment, comprehensive strain collections representing the phylogenetic and functional diversity of the plant microbiota have been established thanks to the cultivability of an unexpectedly large fraction of the members of the plant microbiota . This high cultivability of plant associated bacteria is likely based on low complexity food webs, continuous substrate supply by the plant, and an essentially aerobic environment . In addition to cultivation and subsequent whole genome analysis, screening SynComs of various complexity for interactions and metabolic activity in correlation with environmental parameters has been a bottleneck. Microfluidics tools such as massively parallel on chip coalescence of microemulsions enable screening of 100,000 communities per day . For example, bacterial isolates can be screened individually and in combinatory sets as SynComs for various useful properties, including plant growth promoting functions such as suppression of pathogens or degradation of harmful substrates, for their potential in bio fuel production, or as environmental remediation agents. Such tools coupled with high throughput DNA or RNA sequencing and long read sequencing platforms including PacBio and Oxford Nanopore , as well as metabolomics and various activity assays , now allow rapid profiling composition, function, and activity of SynComs as well as complex native microbial communities residing in soils and on plants.The quantity of data generated by the new technologies described above surpasses the capabilities of traditional analysis methods. Nevertheless, to gain insight, we need to integrate and fuse different data streams. To accomplish this, we must overcome the heterogeneous data types and lack of standards for data exchange. Ultimately, we need systems that can dynamically pull in diverse data from different devices and experimental modalities and intelligently interpret it using background knowledge in order to derive new hypotheses or make predictions such as being able to predict the consequences of specific environmental changes on plant health mediated by the microbiome. Machine learning methods and, in particular, deep learning have proven particularly useful for classification problems involving large datasets such as environmental data generated from technologies, including thermal sensing and LiDAR. Supervised ML techniques will learn to classify entities based on vectors of data characteristics, trained from prelabeled data. DL techniques involve the use of multilayer architecture neural networks . Different DL architectures can be applied to different problems. Convolution networks can be applied to image detection and recognition problems , whereas recurrent architectures such as long short term memory can be applied to time series data. One of the challenges of phytobiome data are the paucity of sample data or lack of resolution in imaging and instrumentation. One DL architecture designed to address this is the generative adversarial network . A GAN can generate plausible synthetic data by utilizing two NNs that are trained together in an adversarial scenario—one network attempts to distinguish real examples from fake ones, and the other creates plausible example data to fool the first. Over time, both models improve, and the generated examples become more plausible, reflecting real world characteristics of the domain without the need for explicit encoding of priors. In the context of phytobiome data, GAN could, for instance, help to synthesize and denoise imaging data . Although DL has seen tremendous gains and achieved much over the last decade, there are still a number of challenges. The input data must be in vector form, which is straightforward for sensor data; however, complex biological information must be embedded in a suitable fashion. NNs are famously inscrutable—they do not provide any explanation as to why they produce a particular result. This is particularly problematic in the face of adversarial attacks, in which the NN is deliberately fooled by fake data designed to elicit a misclassification. The burgeoning field of explainable artificial intelligence attempts to use a variety of techniques to make DL decision making less of a black box process. The field of DL and ML has seen a rapid advance in recent years but, in many cases, DL methods may not yield improvements over traditional methods. DL methods are best applied for complex multidimensional data such as imaging data or for predictions involving complex latent nonlinear mechanisms; for example, as found in ecosystem models. Some have successfully applied DL methods to modeling distinct ecosystem parameters such as soil temperature over a soil depth profile , and processes such as ice shelf melting as part of the Energy Exascale Earth System Model . DL methods will also gain importance in microbe enabled soil biogeochemical models that aim to predict links between climate change, elevated CO2 concentrations, plant–microbe interactions, and soil nutrient cycling . For example, the ecosys model allows for the incorporation of microbially based models using traits such as growth rate , optimal temperature , and resulting enzyme activity , as well as genome size. Microbial traits, which can be obtained from genomic data, help to identify and quantify trade offs .