2

URoL: Microbiome

Defining a microbiome as “a collection of different microbes in a specific habitat [that] may include non-host associated microbiomes and host-associated microbiomes,” NSF published three calls for URoL:Microbiome proposals. URoL:Microbiome Theory and Mechanisms (URoL:MTM),¹ published in 2019, sought theoretical predictive frameworks to generate and test hypotheses about the causal relationships within the microbiome and among the microbiome, host, and environment. URoL:Microbiome Interactions and Mechanisms (URoL:MIM)² was published in 2020 and focused on understanding interactions and mechanisms that govern the structure of microbiomes. URoL:MTM offered two submission tracks: Track 1 for a budget up to $500,000 and duration up to 3 years and Track 2 for a budget up to $3,000,000 and duration up to 5 years. URoL:MIM funded projects with a total budget up to $3,000,000 and duration up to 5 years. A third program, URoL:Emergent Networks (URoL:EN),³ published in 2021, is a cross-directorate program of NSF that includes some microbiome research. By December 2022, NSF had funded 29 microbiome projects.

The charge from NSF to URoL:MIM researchers was as follows (with a similar charge to URoL:MTM researchers):

URoL:EN aims to develop a predictive understanding of how key properties of living systems emerge from interactions of factors such as genomes, phenotypes, and environments; and of how emerging networks of systems (organismal, natural, social, and/or human-engineered) respond to or influence evolving environments. Its charge included the following:

___________________

¹ See https://www.nsf.gov/pubs/2020/nsf20513/nsf20513.htm (accessed February 28, 2023).

² See https://www.nsf.gov/funding/pgm_summ.jsp?pims_id=505694 (accessed February 28, 2023).

³ See https://beta.nsf.gov/funding/opportunities/understanding-rules-life-emergent-networks-urolen (accessed February 28, 2023).

The workshop to discuss projects relevant to URoL:Microbiome was held on February 22, 2023. Six PIs (representing four MTM-, one MIM-, and one EN-funded project) participated in the live discussion. These PIs were joined by three moderators—Margaret J. McFall-Ngai, Jennifer Martiny (University of California, Irvine), and Jo Handelsman (University of Wisconsin)—all of whom conduct multidisciplinary research to understand the structure and function of microbiomes. Fifty-five individuals watched the live webcast and 73 have viewed the recording as of May 10, 2023.

Martiny summarized the six key goals of the workshop. Participants were asked to share their scientific advancements; to consider how disciplinary convergence drove or enabled those advances; to describe how the research groups incorporated multidisciplinary, systems-level approaches into their projects; to discuss the broader implications of their scientific advancements and research ecosystem; to explore opportunities for further investigation and future societal needs; and to start highlighting rules of life that might be generalizable across fields and scales, understanding that these microbiome projects are still in their early stages.

Handelsman read selected comments submitted in response to a pre-workshop questionnaire by PIs from 10 funded projects, including 3 projects not represented among the workshop participants.⁴ These comments have been incorporated into the proceedings below, which are organized by topic.

SCIENTIFIC ADVANCEMENTS

Participants described their research projects and briefly outlined their scientific findings to date. A brief overview of these research projects is provided here.

Developing an Experimental Framework to Study Marine Fish Microbiomes

Eric Allen (Scripps Institution of Oceanography) described his group’s URoL:MTM project, which investigates the microbiome dynamics within marine fish species. The group specifically aims to contrast fish with two different feeding ecologies—coastal pelagic herbivorous versus cosmopolitan herbivorous.⁵ Microbiome samples are characterized across multiple body sites via metagenomic analysis and cultivated in vitro. To define the metabolic phenotypes of these microbiota over space and time, the researchers combine high-throughput amplicon data⁶ and metagenomic sequence data^7,8 of gut samples. The major takeaways to date, he said, have been in the area of method development.

___________________

⁴ These three projects consisted of an investigation into the physiological role of the cnidarian microbiome; development of new tools to track and forecast evolutionary change in genetic networks; and a study using metagenome analysis to predict emergent community metabolism in a diverse population of bacteria.

⁵ Coastal pelagic fish live in the water column in coastal areas and away from the sea floor. Cosmopolitan fish have a wide range that may extend across all or most of the world.

⁶ See https://academic.oup.com/nar/article/47/18/e103/5527971 (accessed May 1, 2023).

⁷ See https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7641418 (accessed May 1, 2023).

⁸ See https://www.technologynetworks.com/analysis/articles/lc-ms-what-is-lc-ms-lc-ms-analysis-and-lc-msms-348238 (accessed March 1, 2023).

Given the absence of model systems for this work, Allen and colleagues have been developing synthetic cultivation methods that faithfully replicate the complex microbial communities found in fish guts in a laboratory environment. He described two broadly applicable advances from their work so far, in the form of two new methods: one to estimate microbial biomass from sequence data alone, and another to interrogate the metatranscriptomes of host-associated microbiota by depleting the large host-derived RNA signal. In response to a question from McFall-Ngai regarding the difficulty of distinguishing transient versus resident microbiota in digesta, Allen noted that this is primarily a problem with carnivorous and other “short-gut” fish species. It is easier, he said to distinguish the two types of microbiota in the long guts of herbivorous fish, particularly in the hind chambers, where the environmental signal disappears, much of the primary deconstruction takes place, and there is an expansion of microbes responsible for fermentative metabolism. The researchers combine this information with environmental sampling to distinguish what is native to the gut versus transient, he said. The use of metatranscriptomics also enables live microbes to be distinguished from dead ones.

Overall, the researchers have found that gut microbial diversity is lower in herbivorous than carnivorous fish, noted Allen. The herbivorous gut has three dominant bacterial phyla with a total complexity of roughly 100 species, though the number varies depending on how the metagenomic analysis is performed. There is a high prevalence of mollicutes,⁹ as well as mycoplasma,¹⁰ which are not present in the free-living marine microbiome and are likely contributing to host physiology (producing B vitamins, for example) rather than merely functioning as parasites, Allen noted.

An important focus of Allen’s work is to identify which subset of enzymes are most important for digestion, which could enable development of an inexpensive aquaculture feedstock that has all the necessary nutrients. To do this, the researchers isolate microbiomes from individual fish as they graze on different foods over time and perform metatranscriptomic analysis on the microbes, which allows them to see different enzyme signatures turning on in response to different diets. They have thus been able to identify the most active enzymes from among more than 1,000 carbohydrate-active enzymes present in a single gut compartment. “You can begin to almost reconstruct, based on microbial activity, what these fish have been consuming in the very recent past,” he said, noting that in herbivorous fish, the presence of desulfatases and carbohydrate-active enzymes describe a “very stark demarcation” between free-living microbes and true symbionts native to the host fish.

In response to a question from Handelsman regarding the source of the gut symbionts, most of which are not present in seawater, Allen replied that it is a combination of vertical transmission, sediment (in demersal fish), and coprophagy, which occurs when schooling fish eat the feces of their neighbors, “doing their own little aquatic fecal transplant.” The gut exerts strong selective pressure for the growth of key taxa that contribute to host physiology. To pin down this selective process, the group is seeking to develop a model system that would enable them to manipulate the aquatic microbiome of a coastal pelagic fish in an aquarium and follow it over time.

Assembly, Diversification, and Coevolution of Marine Invertebrate Microbiomes

Robert Thacker (Stony Brook University) described his group’s investigation into how local host diversity and environmental conditions influence the composition of microbiomes

___________________

⁹ Mollicutes describe a class of bacteria lacking a cell wall, which live as parasites of plants or animals.

¹⁰ Mycoplasma is a genus of bacteria within the class Mollicutes.

within marine invertebrate¹¹ hosts. Thacker, who is a co-PI on a URoL:MTM grant (PI: Kent Hatch), studies the vertebrates that settle on sediment tiles distributed around Long Island. One element of this work examines the prevalence of host switching and co-speciation¹² in sponges and their microbiomes. Previously, Thacker’s laboratory had found that Ircinia sponges comprised nine species (not three as had been thought) and that each species hosted a unique microbiome, which appeared to result from vertical transmission. However, recent work from his group and others has uncovered much more evidence of host switching and acquisition from the environment among microbial eukaryotic hosts. Furthermore, although Thacker had previously described co-speciation of large, filamentous cyanobacteria in a tiny clade of Dysideidae sponges, he has since found that broadening this study to a larger clade eliminates any evidence of co-speciation. Instead, the roughly 100 species of cyanobacteria in this sponge microbiome show an independent pattern of host associations. These findings are driving the group’s current investigations, in an effort to better model when and how host switching occurs. Thacker’s observations led McFall-Ngai to wonder whether sponges might offer some insight into the “lore” that intracellular symbionts are more tightly evolved with their hosts than extracellular symbionts, which has been challenged in insects. In sponges, the localization of most symbionts is currently unknown, said Thacker.

Modeling of Microbiomes Across Hosts and Platforms

One difficulty for both experimentalists and modelers is comparing microbiome studies across different hosts and platforms, said URoL:MTM grantee Rebecca Vega Thurber (Oregon State University). To address this, her group is performing the same experiments in three different model aquatic species and seeking to identify system agnostic microbiome measures, which she defined as patterns in microbiome data that are independent of the specific host or microbiome, with the goal of uncovering rules of life and developing new analytical measures for microbiomes. The group is particularly interested in uncovering metrics that enable the prediction of host resiliency against a variety of anthropogenic disturbances, which can inform conservation efforts.

The three aquatic microbiomes studied by Vega Thurber’s group reside in zebrafish gut, coral, and seagrass. These differ considerably in microbiome complexity, ranging from a single non-pathogenic bacterial parasite that dominates the coral microbiome, 40-100 bacterial species in zebrafish gut, and a much more diverse microbiome in the seagrass. The exact same experiment is conducted in each system. The researchers begin with antibiotic administration to disrupt the microbiome, then impose a temperature stress, and then introduce a pathogen (eukaryotic for fish, fungal for seagrass, bacterial for corals). The resulting data are compared across systems, using deep learning to analyze both experimental and pre-existing data sets (including data from a range of host-microbiome systems), with the aim of detecting novel patterns.

So far, the group has found that, following a disturbance, dramatic alterations can be seen in the composition of the microbiome, and also of host disease phenotype, under conditions of ecological dominance (when a microbiome is dominated by just one or two species). In contrast, hosts with more diverse microbiomes are more resilient and their microbiomes are less altered by

___________________

¹¹ Marine invertebrates lack a vertebral column and represent the earliest animals. They are a diverse group that includes jellyfish, corals, shrimp, starfish, tunicates, sponges, and a variety of marine worms, among many others.

¹² Co-speciation refers to co-evolution of the host and parasite, which may lead to specialization of the parasite to a specific host.

disturbance. Indeed, if host phylogeny is disentangled from the microbiome, then ecological dominance becomes the primary driver of disease phenotype, said Vega Thurber, a finding that is “pretty profound” and consistent with established ecological theory stating that more diverse ecosystems tend to be more robust. Secondly, data show that disturbance leads to an increase in variance of the microbiome, and this increased variance typically remains even after the disturbance has been removed. Thirdly, and perhaps as a result of the first two features, the group has found that network connections within microbiomes, which remain stable across the life history of an organism, fall apart following a disturbance. On the data analysis side, the researchers found that of the multiple types of -omics they have analyzed by deep learning (metagenomics, metatranscriptomics, 16s RNA, and metabolomics), microbial metabolomics showed the best correlation with host phenotype. This last observation was made on data from vertebrates, so its generalizability to other taxa remains to be determined.

Citing findings from David A. Relman and others that the human gut microbiome typically returns to prior levels within several weeks following a short course of antibiotics,¹³ McFall-Ngai asked what the reservoir populations are in each of the systems studied by Vega Thurber’s group and how long it takes them to recover following a disturbance. For the coral system, where 99% of the relative abundance of bacteria consists of a single parasite, the reservoir is unknown, said Vega Thurber. Heating these corals reduces the parasite to undetectable levels, and it takes months to return to normal. Vega Thurber noted that the parasite normally exists intracellularly, but when the parasite population gets large it moves into mucocytes,¹⁴ which release mucus extracellularly to protect the coral. This creates the potential for horizontal transmission of the parasite when one coral passively takes up another coral’s mucus. Less is known about transmission of the seagrass microbiome, but given the plentiful nature of seagrasses, passive transmission of the microbiome is a possibility, said Vega Thurber. The zebrafish gut recovers quickly, likely due to either coprophagy or repopulation by a small reservoir that remains in the gut. One difficulty in comparing across systems is that recovery periods may differ; “the patterns might be the same, but the timeline might be really different,” she added.

Curious about the reason behind the increased dispersion of the microbiome that follows a disturbance, Martiny wondered whether this might result from a breakdown of selective barriers followed by a stochastic process of repopulation, or alternatively, whether it reflects differences among hosts with respect to the preexisting composition of their microbiomes. Two different systems provide similar answers, said Vega Thurber. In zebrafish, infection with a pathogen alters the fish’s immune status (observed as changes in immune-related gene expression), which relieves some barriers to infection, enabling altogether new microbes to populate the gut and cause dysbiosis. Similarly, corals undergo major changes in gene expression in response to temperature stress, as does the native coral parasite. The parasite’s population crashes and there is a concomitant influx of new, fast-growing bacteria, which remain in the coral for weeks or months.

A Search for General Rules Governing Microbiome Dynamics

Jizhong (Joe) Zhou (University of Oklahoma), co-PI on a URoL:MTM grant (PI: Mathew Leibold), discussed his group’s efforts to identify general ecological rules that govern

___________________

¹³ See https://pubmed.ncbi.nlm.nih.gov/22674335 (accessed March 2, 2023).

¹⁴ A mucocyte is a cell that produces and secretes mucus.

microbiome dynamics in different ecosystems, similar to rules that have been observed in nonmicrobial systems, such as those that govern species-area relationships. The group uses microbes grown in enclosed anaerobic digesters (AD)¹⁵ as a model system, which enables frequent sampling and full control of the environment. Microbiome biodiversity, structure, and function are monitored over varying time spans, some exceeding 15 years, and in response to various environmental challenges. This information will be used to develop novel mathematical approaches, statistical tools, and metagenomics-enabled models to generate frameworks for predicting and manipulating the dynamics of AD systems, which are used globally for waste treatment, and to uncover rules that can be tested in a variety of microbiome systems.

Unlike physics or chemistry, ecology has produced few general laws, but one such law governs the species-area relationship, which describes how the number of species increases with area and applies to animal, plant, and microbial communities, said Zhou. In this project, his group aims to identify laws governing the species-time relationship, which they previously found was accelerated in soil bacteria and fungi under warming conditions.¹⁶ Their aim is to determine whether the relationships they detected previously describe a general rule that works across systems. The group conducted a meta-analysis of 6,000 data sets collected until December 2022 from a range of systems including plants, insects, and the human gut, and the results revealed a “very significant relationship between time and turnover (increase in species number),” said Zhou. However, Zhou noted that the researchers observed much greater variation in turnover rates among different communities of microbes than among different plants or animals, for which the change rate is relatively constant across systems.

Zhou described two goals of his project in the realm of high-level ecological theory. One goal is to determine the extent to which microbiome dynamics are stochastic versus deterministic.¹⁷ Although deterministic processes are important (and the focus of most ecology studies), stochastic processes (including genetic drift, diversification, and dispersal) may be more important in microbial communities, he said, while acknowledging that theoretical challenges will make it difficult to prove this. The relative involvement of stochastic versus deterministic processes may itself vary among systems and situations, noted Martiny and McFall-Ngai, who suggested that if a spectrum exists between these two mechanisms in the microbiome biosphere, then there may be rules that govern its application. Another goal, that of identifying a microbial signature that underlies transient microbiome dynamics, faces both theoretical and measurement challenges, said Zhou.

Zhou likened the complexity of the stochastic-versus-deterministic question to an earlier study of his examining the effect of spatial isolation on microbial dynamics, in which spatial limitation, rather than isolation, may have played the larger role due to the scale of the experiment. Scale is critical when analyzing results, he said. This applies both spatially (e.g., one gram of soil versus a 10-meter transect) and temporally (e.g., daily versus yearly sampling) and presents a huge challenge, “not only for microbial ecology but also … for macroecology from any perspective,” he added. To further complicate matters, heterogeneity of the terrestrial environment is likely to be much greater at scale than that of the aquatic environment, he added. Vega Thurber noted an additional difficulty arising from variations in richness size. Because

___________________

¹⁵ In anaerobic digestion, microbes break down biodegradable material (such as manure) in the absence of oxygen.

¹⁶ See https://pubmed.ncbi.nlm.nih.gov/30911147 (accessed March 3, 2023).

¹⁷ The outcome of a stochastic process is based on a random probability distribution, so the outcome of any individual event cannot be predicted with certainty. In contrast, the outcome of a deterministic process can be predicted with certainty based on the inputs.

systems with low diversity are very sensitive to dispersion, it is “statistically a little bit risky” to make conclusions about stochasticity based on comparisons of low-diversity and high-diversity systems, she said.

The Role of Fungal–Bacterial Interactions in Metal Transformation and Detoxification

Unlike many organic pollutants, metal species cannot be degraded, but they can be transformed to less toxic forms. Jose Cerrato (University of New Mexico) and colleagues are using their URoL:MIM grant to study the roles played by soil microbiomes, specifically fungal hosts and their symbiotic bacteria, in transforming and detoxifying the various toxic metals that were deposited in New Mexico’s soil by uranium mining. Although fungi are important contributors to the rhizosphere¹⁸ and to nutrient transport in plants, the field of biogeochemistry has focused almost exclusively on bacteria, leaving a gap in the literature, said Cerrato. His group is therefore trying to understand on a mechanistic level how fungal-bacterial relationships affect metal transformations. They are integrating field observations with laboratory experiments, including advanced transmission electron microscopy, X-ray spectroscopy, and microbiome analysis.

The researchers are currently studying a copper- and iron-rich smelting site in Michigan. They are focusing on nearby plants, which represent a rhizosphere community “that would love to host both bacteria and fungi to facilitate uptake of metals,” said Cerrato. In New Mexico, the group isolated microbes from waters with pH 2-4 due to acid mine drainage. They have also begun to study a river system that was contaminated by a spill in Silverton, Colorado, in 2015. Cerrato previously characterized metal speciation at the site, and his URoL collaborator, Gregory Bonito, isolated fungi and is performing a community analysis.

Ultimately, Cerrato’s work is designed to elucidate general rules governing how the structure and function of the fungal microbiome is altered in response to environmental stressors. Natural systems experience gradients in pH, dissolved oxygen, and redox gradient that change seasonally or even daily, said Cerrato, and microbial physiology is very sensitive to these conditions. The researchers will explore how these gradients affect behavior and function of the microbes, which has been explored for many bacteria but not for fungi.

Toxic Cyanobacterial Blooms in California’s Rivers

While toxic cyanobacterial blooms are a common and well-studied phenomenon in lakes, little is known about these blooms when they occur in streams. This is an emerging environmental issue affecting streams across the globe, said Rosalina Christova (California State University, San Marcos), co-PI on a URoL:EN project designed to explore the conditions responsible for proliferation and toxin production by benthic cyanobacteria (PI: Ramesh Goel). Her group will combine intensive field observations spanning multiple seasons with laboratory experiments, -omics investigations, and bioinformatics to develop a predictive model of the timing of benthic anatoxin-producing mats in streams in northern California. This project began in January 2023, just 1 month before the workshop.

The researchers will isolate different strains of anatoxin-producing cyanobacteria in the genera Microcoleus and Anabena and characterize them on a molecular level, to identify which

___________________

¹⁸ The rhizosphere comprises the roughly 1 millimeter-wide zone of soil surrounding the root of a plant and includes all the organisms in that zone. It nurtures organisms that support plant growth.

strains are producing anatoxins and to study the interactions between organisms in the microbial mats. Because this group of cyanobacteria is unable to flourish in low-nutrient lakes, these interactions among microbes may prove important for their ability to flourish in low-nutrient streams. In a preliminary experiment, Christova’s laboratory cultured anatoxin-producing strains in media completely lacking in nitrogen and found that the cyanobacteria continued to thrive and produce anatoxin for 45 days. This result indicates the depth of the management issue, she said, as not only are these blooms not confined to agricultural areas but also they impact low-nutrient reference streams, streams in protected natural areas (including Zion National Park), and many streams in northern California.

LARGER THEMES AND RULES OF LIFE

“Maybe the ultimate goal is to make predictive models that will take one microbiome and be able to predict behaviors of others, as we can do in many other areas of biology, but have not been able to do very broadly in microbiome work,” said Handelsman. “This is a holy grail of community biology. Can you take elements of the community, whether it’s organisms, genes, metabolites, or something else, and predict the properties that those individuals will express when they’re together, an emergent property from the individual pieces?” Participants reflected on how their work might uncover more generalizable rules to further this aim.

Rethinking Co-phylogeny and Host Switching

Two participants shared observations that made them reconsider the likely role of co-phylogeny (i.e., a long-term process of co-evolution) in establishing the relationship between a host and its microbial symbionts. Vega Thurber described a bacterial parasite of corals that had a reduced genome, on the basis of which she had assumed it to be an intracellular obligate parasite. However, microscopy showed that it moves from an intracellular to an extracellular state, and other work indicates that it has a free-living or semi-free-living stage, which researchers had not expected. The bacterium is horizontally transmitted, which was also unexpected, and shows no evidence for co-speciation with the host.

In a similar vein, although Thacker’s previous observations highlighted sponges with unique, species-associated microbiomes that were vertically transmitted to the larvae, suggesting co-speciation, more recent work in other sponges has uncovered considerable evidence of host switching and acquisition from the environment. This has caused him to rethink the role of co-phylogeny in the sponge-microbiome relationship. Vega Thurber noted that these examples point to the difficulty of untangling co-evolutionary relationships. “It’s an interesting and exciting time to try to understand the patterns, she said. “Can we make global assumptions about where these organisms live with their hosts and what their actual transmission state is, based on aspects of their genome? Maybe we can’t.”

“There’s clearly an environment symbiotic relationship interaction going both ways where the environment influences the interaction, [and] the interaction influences the microhabitat that it’s existing in,” said Handelsman. A URoL PI who was unable to attend the meeting described the rules of life being uncovered by their work as follows:

Predictive Potential of Metabolomics with Respect to Host Disease and Behavior

In what way is a host’s disease phenotype related to its microbiome? Vega Thurber’s group performed deep learning analysis on vertebrate microbiomes using multiple types of - omics, including metagenomics, metatranscriptomics, 16s RNA, and metabolomics, and compared each against metadata from the host phenotype. Using language modeling, they found that metabolomics (i.e., a quantitative measure of different metabolites across the microbiome) provides the best match to the host phenotype, enabling a deep learning algorithm to distinguish among different disease states of mice based on microbiome metabolomics alone. “Metatranscriptomics and metagenomics did not improve the ability to determine the phenotype. The metabolomics alone—it was a very, very high confidence level … it could identify which of the microbiomes were associated with the disease phenotype.” Vega Thurber noted that this could be helpful for microbiome research in the future, raising the question of “can we skip the metagenomes and metatranscriptomes and just look at metabolomics … or do we actually need all of them?”

In his comparisons of microflora from the guts of carnivorous versus herbivorous fish, Allen has made a similar observation. Using LC-MS/MS to analyze the metabolomes, he noted that tens of thousands of spectra are generated, though only 3% have a spectral match to a database, and most of these are lipids. But “as soon as you go into the suite of overall spectral diversity, then all of a sudden, you begin to see really different signatures, which are almost predictive and reproducible across species that have different feeding ecologies” he noted, suggesting that metabolomics may prove a powerful tool for phenotype prediction based on unique patterns of peaks, even when the identities of 97% of the peaks are unknown.

In Handelsman’s system, the metabolite profile looks very different when comparing a single organism to a pair or a three-member community; in at least one case, changes in a peak correspond to the same molecule being handed off from one organism to the next and undergoing a different modification each time. “The unknown molecules can be the most exciting ones—let’s not throw those away,” she added. Zhou cautioned that metabolite profiles may be relevant to short-term conditions such as host disease phenotype, but other questions (e.g., host preference) are likely to require a more stable indicator, such as genomics.

A URoL PI who was unable to attend the meeting referenced work they had recently published “showing that genomes can be used to predict emergent community metabolism” and illustrating an emergent rule that “gene presence and absence can predict metabolite dynamics.”

Impacts of Microbiome on Resilience to Disturbance

It is a tenet of ecological theory that diversity within an ecosystem increases its resistance to disturbance. In a large modeling experiment of coral microbiomes across coral phylogeny, Vega Thurber’s group found that, once host phylogeny was disentangled from the microbiome, ecological dominance was the primary driver of disease phenotype. Hosts with more diverse microbiomes had greater resilience. This observation may indicate a rule of life, suggested McFall-Ngai.

MULTIDISCIPLINARY RESEARCH/MULTI-TEAM RESEARCH: EXAMPLES OF CHALLENGES AND STRATEGIES FOR SUCCESS

“Macroecology was completely generated outside of thinking about microbes,” said McFall-Ngai, noting that the landmark ecology textbook does not address microbes.¹⁹ Although microbiology and symbiosis cross disciplinary lines by their very nature, researchers still have to contend with the different cultures that have developed in the various scientific disciplines, “and overcoming those different cultures is a major milestone that we have to work on…. It makes one think that maybe the field of ecology ought to be completely reconstructed to take into account the fact that the microbes were foundational and that the ecology that we see in the macro world arose from that. We continue to do this anthropocentric looking down, instead of looking up.”

Despite recent advances in microbiome research that constitute “the biggest change in our view of the biosphere since Darwin,” she said, “we continue to be in a situation in which biology has been siloed since the 1950s … reflected in departments of universities as well as in the funding agencies.” As an illustration of the gulf that persists between macro- and microbiology, McFall-Ngai described her experience on the American Society for Microbiology’s Climate Change Task Force: “I kept saying, ‘you know, microbes do live with animals and plants.’” On the flip side, one of the microbiologists on the task force attended a 3-day National Academies’ meeting on climate change, where microbes were not mentioned once. This is a “major problem” that “the fields of symbiosis and microbiomes can begin to address and begin to integrate … in ways that they haven’t been [addressed] before,” she added.

PIs described examples of the challenges posed by multidisciplinary research and shared their strategies for overcoming them.

Challenge of Obtaining and Effectively Using High-Quality Data

Participants expressed concerns regarding the quality of data that are generated from microbiomes, often in very high volumes. “This is a big challenge and opportunity,” said Handelsman; “we’re collecting all these data all the time, and we have trouble figuring out how to use it all … it’s so easy to collect way too much.” “When we study community dynamics with various omics tools, the key issue is reproducibility,” said Zhou, recalling an experiment in which the results of three successive sequencing reactions from the same sample overlapped by less than 10%. Quantification and sensitivity also present technical issues, “so we should be very, very careful when we address research questions using omics tools. One critical question is whether the data quality is good enough.” He said that his group has frequent discussions about the extent to which experimentally-derived data can address theoretical questions, noting that they occasionally modify the questions to conform to the limitations of the methods, while at the same time ensuring that the models remain relevant to microbial systems, another important challenge.

Vega Thurber concurred that “it’s too easy and too cheap to make a lot of data that in many cases might muddle the waters instead of making it more optically clear what the patterns are.” This problem may be exacerbated by grants that encourage using the most advanced technology available. Vega Thurber said that her group is “very technologically focused … we

___________________

¹⁹ See https://www.academia.edu/15092278/Ricklefs_The_Economy_of_Nature_6th_txtbk (accessed March 10, 2023).

want to increase innovation, but at the same time we need to reflect on the methods … and what are they telling us.” Furthermore, as was previously noted by Zhou, “[the utility of the data] is also context-specific … the question really matters,” she said, adding, “I do think that we’re getting better at understanding what these data are telling us.”

Each type of data gives different insights, and choosing which method to use depends on the context, continued Vega Thurber. “Some are less precise, some are more accurate, but they give us a tool.” For example, time series data are selected to maximize temporal resolution, whereas spatial studies maximize spatial resolution. Amplicon sequencing has flaws “but can provide very cheap and easy-to-analyze data sets over very, very refined timescales [and] large spatial scales; that is going to be something that will give you an answer.” But different tools may be needed to address questions of function. “Each tool that we develop is always going to have a limitation, both at a detection level but also at a question level, and then trying to figure out how we can combine some of these tools and theory … trying to pick the right one at the right time for the right question is really the crux.”

“The biggest strength of this project is that we have different methods,” said Christova. URoL integrative research teams address a problem from multiple perspectives using diverse methods, combining microscopy, molecular analysis, and field data, for example. “Combining different methods to answer the same question helps us to have more reliable data” and validates the methods, she said.

Zhou cautioned that it is misleading to equate the abundance of data that can be derived from a given sample with the actual quantity of samples. “We don’t have enough data, far away from enough, to ask ecological questions,” he said, giving the example of time series dynamics, which requires a minimum of 40 to 100 time points, “but it’s experimental, we don’t have this kind of flexibility; this is also a very big issue.”

Even given an abundance of data, Vega Thurber noted that it is often difficult to parse the directionality of the response, “Is the microbiome directly influencing the host? Is the environment directly influencing the host and then the microbiome? Without some additional statistical methods to define directionality, I think we often are grasping at straws for what the mechanism of these responses really is.” Another significant challenge, she said, is combining data sets across biological scales of complexity. “The timescale [in] which a host response occurs compared to the microbiome can blow your mind … I think that’s a general issue in our communities, trying to understand how to compare data across biological scales.”

Microbiome science is a new field, and its methods are still developing, said Handelsman. “The advances are coming quickly in terms of the methods, but we’re still working to sort out the quality of the data [and] what is the appropriate data … that’s likely going to depend on the system and the question.” The essential role of data scientists in microbiome research underscores this point, she noted.

Approaches to Evaluating Big Data

One potential outcome from the URoL program could be to establish standards for evaluating the quality of big data, said Dennis Discher (University of Pennsylvania). He gave an example of single cell RNA-seq data sets. “I’m looking for gene A scaling with gene B. If I don’t see that, I call it a pretty noisy, weak data set. If I see too many zeros in a single cell RNA-seq I say that’s not worth looking at, because I know RNA is being expressed more globally, more generally, for a specific set of … housekeeping genes.” Some aspects of gene expression might be universally applicable to all microbes, in which case, Discher asked, “can we get at rules of

life that say this data set is on the very low-quality side” based on a “signature of life” like the cell cycle? “I’ve gone public and said ‘these data sets aren’t worth looking at from the point of view of, for example, stoichiometry’…. I’m not seeing enough of this sort of thing being done.”

Zhou noted that assessments of the relative abundance of various microbes in a growth study are more difficult to validate than Discher’s gene expression analyses, because “when we do community studies we have no idea which of them should be there or should not be there, so we cannot judge whether the data quality is really good or not … especially when you do all the meta-analysis, data quality is a big issue.” To make matters worse, different bioinformatic protocols can yield dramatically different results from the same data. “So it’s very hard to judge whether the data set is really good or not. It really depends on our questions…. I say the pattern should be the same no matter how you process the data, but the exact results may be different,” he added.

Deep learning methods raise particular concerns owing to their reliance on pattern recognition, which may lead to results that do not make physical sense, said Discher. “In some fields, people are using physical [and] chemical laws to constrain, inform, and advance deep learning…. As we look for rules of life … don’t let the machine learning just work on pattern recognition and come up with things that violate rules of life. Constrain it in some way.”

Discher and McFall-Ngai suggested that the cohort of URoL microbiome researchers consider establishing “Rules of Life standards” to help validate data sets. McFall-Ngai described one way such an effort could push the science forward. She said that, while it may be difficult to see signatures of symbiosis in transcriptome data, there are clear differences between symbiotic and non-symbiotic transcriptomes when hypothetical genes are taken into account. However, this has not been pursued because the platforms are so different that it is hard to make comparisons among them. Therefore, working “to try to figure how we might be able to come up with rules … would be an exciting goal,” she said.

Collaboration on Methods Papers at the Start of the Project

Participants discussed several strategies that they have found to be helpful for building cohesive teams and bridging across disciplines. Prior to starting their research projects, NSF tasked the various working groups to write methods papers describing their planned studies. “It was so helpful because we had to understand each other’s figures, we had to understand each other’s language … we really understood the limitations of each system, the limitations of the computational analysis, the limitations of the databases that people were working on … and now we’re much more cohesive and understand each other better,” said Vega Thurber. Cerrato described how the collaborative research enterprise itself has been a great learning opportunity for him as a PI.

Freestyle Meetings to Address Experimental Design

Communication among PIs and students across different disciplines can be difficult, said Zhou. His group addresses this by holding monthly meetings without a fixed agenda, in which researchers hold freestyle discussions to address experimental design and data collection from multiple perspectives. This structure has led to intense discussions “to ensure our data collected is appropriate to the theoretic questions,” he said. Thacker said his group has benefited by adhering to a biweekly meeting schedule. Both groups are dispersed around the country, making regular meetings all the more important.

Importance of Physical Proximity

Despite the availability of virtual communication platforms, several participants emphasized that the ability to gather in-person still matters. Vega Thurber’s students benefited from their educational exchanges, which all took place within a single institution. In contrast, collaborations that spanned a significant distance faced particular challenges as a result of the COVID-19 pandemic. Thacker’s plans to exchange students and PIs between Long Island field sites and Alabama computer laboratories were delayed by 1 year due to the pandemic, which also required him to exchange a planned collaboration with high school students for a course involving college students instead. The pandemic also interfered with Vega Thurber’s ability to travel to field stations where some experimental sites are located, delaying her data collection.

Forging Collaborations Across URoL Groups and Beyond

Several participants are actively engaged in developing communities of researchers to expand the impact of URoL-funded projects beyond their group or outside the URoL program altogether. Cerrato noted that a group is currently being formed to connect fungal researchers from across the community of URoL grantees.

Zhou’s group has plans to disseminate their work among the broader community of ecologists by organizing sessions at society meetings and, in the final year of the grant, hosting a national or international workshop that brings together people from different fields. Thacker’s group is working on methods to improve the efficiency of mining GenBank and other data sets. Once they finalize the user interface, the researchers plan to organize a workshop to share their modeling efforts with the broader scientific community. “It’s a good way to build data sets from the microbial perspective instead of the host perspective,” Thacker said.

SUGGESTIONS FOR FUTURE ITERATIONS OF THE UROL PROGRAM

Participants did not offer suggestions for the URoL program as a separate topic of discussion. However, Zhou suggested gathering the URoL PIs to teach microecology over his web platform, which has received more than 200,000 visits from 19 countries. The platform’s audience consists of mostly undergraduate and graduate students and postdoctoral researchers who are interested in ecology, environmental engineering, and microbiology. Vega Thurber suggested holding more frequent meetings for URoL researchers, which “would be really utilitarian and helpful for everyone.”

SOCIETAL IMPACT AND SIGNIFICANCE OF WORK

Handelsman noted “so many connections here … to important questions and advances that we could make for society.” Some of these connections were mentioned during the meeting.

Addressing the Impacts of Climate Change on the Environment and Human Health

Benthic toxic cyanobacterial blooms constitute an emerging and widespread environmental problem, said Christova. Though first reported in New Zealand, these blooms

affect waterways in the United States, Canada, Europe, and around the globe, and their prevalence has increased over the past decade. “Anatoxin producers [are] developing long seasons in different conditions,” she added, noting that cyanobacteria “like it hot, they really flourish under high temperatures and sun activity.” Both plankton in lakes and benthic stream algae have extended their seasons from the summer months through the fall, even producing some winter blooms, and these blooms generate large amounts of toxins. The extended duration of blooms is likely due to climate change, said Christova, though more study is needed to test that theory. The ability of stream blooms to thrive under low-nutrient conditions extends their range beyond agricultural areas to many areas of northern California, including protected areas.

Toxic blooms have both economic and health consequences, said Christova. The knowledge gained through her project should help with studies of toxic blooms in lakes, microbial community dynamics in engineered bioreactors, and microbial biofilms in water distribution systems. Other participants noted that blooms in recent years caused the municipal water systems to be shut down in Toledo, Ohio, and Portland, Oregon. Every year, Portland sees deaths of cattle and dogs that drink the water from contaminated reservoirs. “It’s a big problem,” said Vega Thurber.

A PI who was unable to attend the workshop wrote that “bacterial communities produce significant N₂O and CO₂,” and that their URoL-funded work “suggests ways to engineer communities to control these emissions” and thereby ameliorate their contribution to climate change.

Increasing the Food Supply

Understanding how the fish gut microbiome contributes to digestion will allow for its manipulation in aquaculture settings to optimize fish growth, fish health, and food quality while reducing costs, said Allen. “Why feed a fish a fish, when you could feed a fish partially digested seaweed that has all the protein and nutrients you need, from basically cheap feedstocks?” This can have broad societal impacts, he added, such as improving fish farming in rural areas where malnourishment and dietary protein deficits are systemic.

Improving Strategies for Wastewater Treatment and Toxic Metal Decontamination

Most wastewater treatment is based on activated sludge, said Zhou, which is very energy intensive. For this reason, environmental engineering is trending toward the use of anaerobic digesters, which are much more energy efficient. By providing fundamental knowledge critical for predicting microbiome behavior in this type of engineered system, Zhou’s URoL project has the potential to improve wastewater management.

Toxic spills from legacy mining sites contaminate waterways that are essential to life in the Southwest and have been especially concerning for members of the Navajo Nation who rely on the Colorado River for agriculture and other purposes, said Cerrato. His URoL research is aimed at bioremediation and metal transformation to improve the environment, health, and safety. In addition, some of the nanomaterials produced through these processes can be used in diverse fields, such as medicine, catalysis, and environmental engineering.

“Sometimes [microbiomes are] the problem, as in the case of toxin production, and other times they’re going to be part of the solution,” noted Handelsman.

Environmental Management

In their written responses, several participants anticipated developing theoretical frameworks that could improve efforts at environmental management. Examples included “predicting how host-microbiome interactions respond to environmental variation, which holds implications for how we optimally manage wildlife, livestock, agricultural, and human health;” and understanding “how ecological restoration (reintroduction of an extirpated native species) alters biological networks in ways that may modify success of the restoration process.” Another project “will provide key insights into cnidarian biology, which is urgently needed for the design and implementation of global preservation efforts.” Stating this more generally, this last respondent wrote that “understanding the function of microbiota in complex systems might aid in the rescue of threatened ecosystems like coral reefs.”

EDUCATION AND TRAINING

Participants discussed their strategies for training a future generation of researchers who can tackle complex scientific problems from a multidisciplinary perspective. Several participants noted the importance of training students in teams led by experts from multiple fields.

Training Students to Bridge Across Silos

“One of the ways we’ve tried to solve these siloed language and culture barriers is through our students,” said Vega Thurber. In her group, two of the PIs are computer engineers and three are experimental microbiologists. All the computer engineering students take Vega Thurber’s microbial systems ecology class, and all of her students take machine learning. “They struggle, but … they came out of it so knowledgeable, and … they are that collective voice back to us,” she said. However, this approach is not necessarily feasible for every group. Whereas all the PIs on Vega Thurber’s project are located at the same university, Zhou’s collaborators span four different institutions, making this type of cross-disciplinary training untenable, he said.

In the summer of 2023, the students in Thacker’s group will cross disciplines by spending time in one another’s laboratories collecting data. “I think that’s actually going to help the computer science students a lot who are saying, ‘what are these types of data that we have’ … that they haven’t actually held in their hands yet and seen in person,” he said. Meanwhile, Thacker’s biology students will learn how to construct a database.

The field of environmental engineering has undergone an evolution in this respect, said Cerrato. “In spirit, environmental engineering does try to encompass a holistic view of having biology, chemistry, engineering, and modeling at the core to get a process approach,” though “it’s easier said than done,” he added.

McFall-Ngai addressed the broader divide among the sciences. She noted that, whereas biology majors all take chemistry and physics, chemistry and physics majors are not required to take biology. However, some traditionally STEM-focused universities are now requiring students from all majors to take biology. Although this “would go a long way to helping with some of these problems in cultural differences,” none of the participants’ universities were moving in that direction. “What’s been most productive for us is hiring faculty who are from those disciplines into the biology departments, and … they become essentially half-biologists, and you get really good collaboration,” said Handelsman. “Many of the solutions to climate change … have

biology at their base, but chemistry and physics and engineering at their solution, and they’re not very well positioned, at this point anyway,” mused McFall-Ngai.

Maximizing Inclusion by Addressing Questions That Matter to Students

“It’s incredible how the curiosity, the energy of students is so driving, so exciting,” said Cerrato, who recounted a student challenging him on what question a particular experiment was intended to answer. “I was very happy to see the students actually guide us on what are the right questions to ask,” he said. “We have to pay attention to giving students a voice.” Encouraging students’ agency is particularly important for improving inclusion, said Cerrato. “Environmental engineering and environmental science is such an incredible field to attract underrepresented minorities … especially because we have a high Hispanic population, we have Native American people … and I am always looking for ideas on how to minimize the barriers that we don’t know we have.” One way to lower barriers is to get the students excited in the research, which he accomplishes by focusing on the applications of the research. “As much as we want the science to be the fundamental driver,” Cerrato said, students will be engaged when the research addresses issues they care about. “The proposal ends up being transformed by the students, which I’m glad [about], because then we get it better.”

Preparing a Generation of Scientists to Address Emerging Environmental Problems

Each URoL project aims to contribute to training the next generation of microbiome researchers. For example, Christova’s group will train at least three PhD students and several additional graduate students, contributing to the overall ability to address the problem of toxic cyanobacterial blooms in the future. Cerrato’s project provides training that bridges the fields of environmental engineering, microbiology, geochemistry, bioinformatics, and art.

“We basically have a class of students doing all this work,” said Thacker, referring to his research into the community structure of sponge microbiomes. When the COVID-19 pandemic disrupted his plans to work with high-school students, he switched to a course-based undergraduate research experience. “It’s a ton of work, but it’s worth it. The students love it,” he said. “Basically, it’s like having 30 students in your research lab.” Supporting and integrating STEM education from high school through post-graduate training is a major focus of his project.

Incorporating Traditional Ecological Knowledge into Microbiome Ecology

In response to a question about whether the panelists were seeing interactions of traditional ecological knowledge (TEK) and microbiome ecology, Vega Thurber remarked that the NSF long-term ecological research (LTER) program contained some “beautiful” work along these lines, particularly in forests and soil systems, and that a lot of research is being co-developed to engage tribal communities and elders to incorporate TEK in microbiome research. In Vega Thurber’s LTER project at the Moorea Coral Reef, she said, “we’re working with some grassroots restoration groups of the Tahitian people to look at microbiomes and rivers, particularly in regard to things that are user-inspired, like ciguatera presence, using molecular data and traditional ecological knowledge to figure out where ciguatera is primarily located, [to] prevent ciguatera poisoning in the fishing communities.”