About

Maureen L. Coleman is an Associate Professor at the University of Chicago in the Department of Geophysical Sciences. Her research group studies microbial population diversity, ecology, and evolution, primarily focusing on freshwater and marine systems. Her work employs a combination of laboratory experiments, field observations, and computational analyses to investigate key processes such as the role of marine viruses in biogeochemical cycles, the physiology and genetics of photoheterotrophy, and the phylogenetic and metabolic diversity of microbes in the Laurentian Great Lakes. Dr. Coleman completed her Ph.D. in Environmental Microbiology at the Massachusetts Institute of Technology in 2008 and conducted postdoctoral research in Geobiology at MIT/Caltech until 2011. Her contributions to the field have been recognized with awards such as the 2014 Sloan Research Fellowship in Ocean Sciences. Her research has significantly advanced understanding of microbial interactions, viral impacts on microbial communities, and biogeochemical processes in aquatic ecosystems.

Research topics

• Ecology
• Biology
• Environmental science
• Chemistry
• Geology
• Physical geography
• Genetics
• Computational biology
• Oceanography
• Biochemistry
• Geography

Selected publications

• Dissolved Organic Matter Composition in the Laurentian Great Lakes Ice and Its Contribution to Spring Melt

  Journal of Geophysical Research Biogeosciences · 2026-02-01 · 1 citations

  articleOpen access

  Abstract Dissolved organic matter (DOM) plays a vital role in lakes, but its behavior in winter is poorly understood. This study examined the differences in DOM between lake ice and the upper water column across 18 sites in the Laurentian Great Lakes, integrating in situ sampling and remotely sensed ice data to create a mass budget model to estimate basin‐scale DOM storage and release from ice. We found that the composition of the DOM pool in ice varied based on ice thickness, water DOM composition, nutrients, and dissolved organic carbon concentrations. Calculations of protein‐like, microbial humic‐like, and terrestrial‐like DOM storage in ice under different ice cover scenarios revealed considerable contributions to the upper water column following ice melt, especially for protein‐like DOM which, during extensive ice cover years, contributed an average of 17.7% to the protein‐like DOM pool in spring. This ice‐derived DOM may be an important source of labile carbon for microbial communities, but projected reductions in winter ice cover and duration under climate change may alter DOM dynamics, potentially impacting this important spring carbon subsidy.

  PublisherOA PDFDOI

• The Great Lakes Winter Grab: Limnological data from a multi‐institutional winter sampling campaign on the Laurentian Great Lakes

  Limnology and Oceanography Letters · 2024-11-12 · 5 citations

  articleOpen access

  Abstract Interest in winter limnology is growing rapidly, but progress is hindered by a shortage of standardized multivariate datasets on winter conditions. Addressing the winter data gap will enhance our understanding of winter ecosystem function and of lake response to environmental change. Here, we describe a dataset generated by a multi‐institutional winter sampling campaign across all five Laurentian Great Lakes and some of their connecting waters (the Great Lakes Winter Grab). The objective of Winter Grab was to characterize mid‐winter limnological conditions in the Great Lakes using standard sample collection and analysis methods. Nineteen research groups sampled 49 locations varying widely in depth and trophic status, collecting a range of limnological data. This dataset includes physical, chemical, and biological measurements. These data can be used to examine diverse aspects of Great Lakes ecosystems or integrated with winter observations from other lakes to improve understanding of winter limnology across different aquatic systems.

  PublisherOA PDFDOI

• Tracking nitrogen allocation to proteome biosynthesis in a marine microbial community

  Nature Microbiology · 2023 · 15 citations

  • Biology
  • Biochemistry
  • Computational biology
  PublisherDOI

• Unraveling the functional dark matter through global metagenomics

  Nature · 2023 · 193 citations

  • Computational biology
  • Evolutionary biology
  • Biology

  . Using massively parallel graph-based clustering, we group these proteins into 106,198 novel sequence clusters with more than 100 members, doubling the number of protein families obtained from the reference genomes clustered using the same approach. We annotate these families on the basis of their taxonomic, habitat, geographical and gene neighbourhood distributions and, where sufficient sequence diversity is available, predict protein three-dimensional models, revealing novel structures. Overall, our results uncover an enormously diverse functional space, highlighting the importance of further exploring the microbial functional dark matter.

  PublisherOA PDFDOI

• Diverse and variable community structure of picophytoplankton across the <scp>Laurentian Great Lakes</scp>

  Limnology and Oceanography · 2023-09-04 · 8 citations

  articleOpen access

  Abstract The Laurentian Great Lakes provide economic support to millions of people, drive biogeochemical cycling, and are an important natural laboratory for characterizing the fundamental components of aquatic ecosystems. Small phytoplankton are important contributors to the food web in much of the Laurentian Great Lakes. Here, for the first time, we reveal and quantify eight phenotypically distinct picophytoplankton populations across the Lakes using a multilaser flow cytometry approach, which distinguishes cells based on their pigment phenotype. The distributions and diversity of picophytoplankton flow populations varied across lakes and depths, with Lake Erie standing out with the highest diversity. By sequencing sorted cells, we identified several distinct lineages of Synechococcales spanning Subclusters 5.2 and 5.3. Distinct genotypic clusters mapped to phenotypically similar flow populations, suggesting that there may not be a clear one‐to‐one mapping between genotypes and phenotypes. This suggests genome‐level differentiation between lakes but some degree of phenotypic convergence in pigment characteristics. Our results demonstrate that ecological selection for locally adapted populations may outpace homogenization by physical transport in this interconnected system. Given the reliance of the Lakes on in situ primary production as a source for organic carbon, this work sets the foundation to test how the community structure of small primary producers corresponds to biogeochemical and food web functions of the Great Lakes and other freshwater systems.

  PublisherOA PDFDOI

• eLife Assessment: Molecular mechanisms of microbiome modulation by the eukaryotic secondary metabolite azelaic acid

  2023-07-26

  peer-reviewOpen access1st authorCorresponding

  Photosynthetic eukaryotes, such as microalgae and plants, foster fundamentally important relationships with their microbiome based on the reciprocal exchange of chemical currencies. Among these, the dicarboxylate metabolite azelaic acid (Aze) appears to play an important, but heterogeneous, role in modulating these microbiomes, as it is used as a carbon source for some heterotrophs but is toxic to others. However, the ability of Aze to promote or inhibit growth, as well as its uptake and assimilation mechanisms into bacterial cells are mostly unknown. Here, we use transcriptomics, transcriptional factor coexpression networks, uptake experiments, and metabolomics to unravel the uptake, catabolism and toxicity of Aze on two microalgal-associated bacteria, Phycobacter and Alteromonas, whose growth is promoted or inhibited by Aze, respectively. We identify the first putative Aze transporter in bacteria, a ‘C4- TRAP transporter’, and show that Aze is assimilated through fatty acid degradation, with further catabolism occurring through the glyoxylate and butanoate metabolism pathways when used as a carbon source. Phycobacter took up Aze at an initial uptake rate of 3.8×10-9 nmol cell-1 hr-1 and utilized it as a carbon source in concentrations ranging from 10 μM-1 mM, suggesting a broad range of acclimation to Aze availability. For inhibited bacteria, we infer that Aze inhibits the ribosome and/or protein synthesis and that a suite of efflux pumps is utilized to shuttle Aze outside the cytoplasm. We demonstrate that seawater amended with Aze becomes enriched in bacterial families that can catabolise Aze, which appears to be a different mechanism from that in soil, where modulation by the host plant is required. This study enhances our understanding of carbon cycling in the oceans and how microscale chemical interactions can structure marine microbial populations. In addition, our findings unravel the role of a key chemical currency in the modulation of eukaryote-microbiome interactions across diverse ecosystems.

  PublisherOA PDFDOI

• Juan <scp>Luis Ramos</scp> : An exceptional <scp>Editor</scp> of <scp>Environmental Microbiology</scp>

  Environmental Microbiology · 2022-12-19

  editorial

  The authors declare no conflict of interest.

  PublisherDOI

• Genome Streamlining, Proteorhodopsin, and Organic Nitrogen Metabolism in Freshwater Nitrifiers

  mBio · 2022 · 20 citations

  Senior authorCorresponding
  • Ecology
  • Biology
  • Environmental science

  Microorganisms play critical roles in Earth's nitrogen cycle. In lakes, microorganisms called nitrifiers derive energy from reduced nitrogen compounds. In doing so, they transform nitrogen into a form that can ultimately be lost to the atmosphere by a process called denitrification, which helps mitigate nitrogen pollution from fertilizer runoff and sewage. Despite their importance, freshwater nitrifiers are virtually unexplored. To understand their diversity and function, we reconstructed genomes of freshwater nitrifiers across some of Earth's largest freshwater lakes, the Laurentian Great Lakes. We discovered several new species of nitrifiers specialized for clear low-nutrient waters and distinct species in comparatively turbid Lake Erie. Surprisingly, one species may be able to harness light energy by using a protein called proteorhodopsin, despite the fact that nitrifiers typically live in deep dark water. Our work reveals the unique biodiversity of the Great Lakes and fills key gaps in our knowledge of an important microbial group, the nitrifiers.

  PublisherDOI

• Editor's evaluation: Wide-ranging consequences of priority effects governed by an overarching factor

  2022-06-26

  peer-reviewOpen access1st authorCorresponding

  Article Figures and data Abstract Editor's evaluation Introduction Results and discussion Materials and methods Appendix 1 Appendix 2 Data availability References Decision letter Author response Article and author information Metrics Abstract Priority effects, where arrival order and initial relative abundance modulate local species interactions, can exert taxonomic, functional, and evolutionary influences on ecological communities by driving them to alternative states. It remains unclear if these wide-ranging consequences of priority effects can be explained systematically by a common underlying factor. Here, we identify such a factor in an empirical system. In a series of field and laboratory studies, we focus on how pH affects nectar-colonizing microbes and their interactions with plants and pollinators. In a field survey, we found that nectar microbial communities in a hummingbird-pollinated shrub, Diplacus (formerly Mimulus) aurantiacus, exhibited abundance patterns indicative of alternative stable states that emerge through domination by either bacteria or yeasts within individual flowers. In addition, nectar pH varied among D. aurantiacus flowers in a manner that is consistent with the existence of these alternative stable states. In laboratory experiments, Acinetobacter nectaris, the bacterium most commonly found in D. aurantiacus nectar, exerted a strongly negative priority effect against Metschnikowia reukaufii, the most common nectar-specialist yeast, by reducing nectar pH. This priority effect likely explains the mutually exclusive pattern of dominance found in the field survey. Furthermore, experimental evolution simulating hummingbird-assisted dispersal between flowers revealed that M. reukaufii could evolve rapidly to improve resistance against the priority effect if constantly exposed to A. nectaris-induced pH reduction. Finally, in a field experiment, we found that low nectar pH could reduce nectar consumption by hummingbirds, suggesting functional consequences of the pH-driven priority effect for plant reproduction. Taken together, these results show that it is possible to identify an overarching factor that governs the eco-evolutionary dynamics of priority effects across multiple levels of biological organization. Editor's evaluation This important study documents the causes and consequences of priority effects in community assembly, using a plant-microbe-pollinator model system. Using an elegant combination of lab and field approaches, the authors provide compelling evidence that early-colonizing microbes alter the nectar pH, which has far-reaching ecological and evolutionary effects. It will likely be of interest to a wide audience and has implications for microbial, plant, and animal ecology and evolution. https://doi.org/10.7554/eLife.79647.sa0 Decision letter Reviews on Sciety eLife's review process Introduction Many ecological communities take one of multiple alternative states even under the same species pool and the same initial environmental conditions (Beisner et al., 2003; Chase, 2003; Scheffer et al., 2001; Schröder et al., 2005). Alternative states can vary not just in species composition, but also in functional, ecosystem-level characteristics such as invasion resistance, total biomass, and nutrient cycling (Bittleston et al., 2020; Delory et al., 2019; Leopold et al., 2017; Pausas and Bond, 2020; Sprockett et al., 2018; Suding et al., 2004). However, it is often hard to predict which of the alternative states an ecological community will take. The main challenge is that alternative states are caused by priority effects, the elusive, historically contingent process in which the order and timing of species arrival dictate the trajectory of community assembly (Debray et al., 2022; Drake, 1991; Fukami, 2015; Palmgren, 1926; Song et al., 2021). To further complicate matters, the strength of priority effects is not always static, but can change rapidly through evolutionary changes in species traits (De Meester et al., 2016; Faillace et al., 2022; Faillace and Morin, 2016; Knope et al., 2011; Urban and De Meester, 2009; Wittmann and Fukami, 2018; Zee and Fukami, 2018). The broad scope of priority effects makes it difficult to fully understand which of the alternative states is realized under what conditions (Fukami, 2015). For any ecological community, this understanding requires simultaneous examination of the compositional, functional, and evolutionary consequences of priority effects. Questions that need to be addressed to understand the causes and consequences of priority effects are indeed wide-ranging. For example, how do priority effects happen (mechanism)? When are priority effects particularly strong (condition)? How rapidly can priority effects change in strength (evolution)? And how do priority effects affect the functional, not just taxonomic, properties of the communities being assembled (functional consequences)? These questions are interrelated, but given that they concern different scales of time and biological organization, it seems reasonable to expect each question to involve a different set of species traits and environmental factors. In this paper, we present empirical evidence that, contrary to this expectation, it is possible for one common factor to underlie many aspects of priority effects, including mechanism, condition, evolution, and functional consequences. Specifically, we show that environmental pH is an overarching factor governing priority effects in a microbial system. Simple microbial systems can help identify general basic principles that organize ecological communities (Cadotte et al., 2005; Drake et al., 1996; Jessup et al., 2004; Vega and Gore, 2018), and the nectar microbiome has recently emerged as a well-characterized simple system for understanding community assembly (Brysch-Herzberg, 2004; Chappell and Fukami, 2018; de Vega et al., 2021; Lachance et al., 2001; Letten et al., 2018; Vannette, 2020). Our study system here consists of the bacteria and yeasts that colonize the floral nectar of the sticky monkeyflower, Diplacus (formerly Mimulus) aurantiacus, a hummingbird-pollinated shrub native to California and Oregon of the USA (Belisle et al., 2012). Initially sterile, floral nectar is colonized by these microbes via pollinator-mediated dispersal (Belisle et al., 2012; de Vega et al., 2022). Microbial communities that develop in nectar through this dispersal are often simple, dominated by one or a few species of nectar-specialist bacteria or yeast (Álvarez-Pérez et al., 2019; Golonka and Vilgalys, 2013; Herrera et al., 2010; Tsuji and Fukami, 2018; Warren et al., 2020). Our previous work has shown that hummingbird-mediated dispersal is highly stochastic because the pollinators differ in the species identity of yeasts and bacteria they carry on their beaks and tongues (Morris et al., 2020; Toju et al., 2018; Vannette et al., 2021; Vannette and Fukami, 2017). Additionally, the outcome of local antagonistic interactions among these microbial species can be sensitive to relative initial abundances (Dhami et al., 2016; Dhami et al., 2018; Grainger et al., 2019; Mittelbach et al., 2016; Peay et al., 2012; Tucker and Fukami, 2014; Vannette and Fukami, 2014). Together, the stochastic dispersal and the history-dependent interactions jointly cause priority effects in nectar microbial communities. Moreover, recent studies indicate that whether flowers are dominated by bacteria or yeasts can affect pollination and seed set, although a mechanism for this effect remains unclear (de Vega et al., 2022; Good et al., 2014; Herrera et al., 2013; Jacquemyn et al., 2021; Junker et al., 2014; Rering et al., 2020; Schaeffer and Irwin, 2014; Vannette et al., 2013; Vannette and Fukami, 2016; Yang et al., 2019). Despite the increasing amount of knowledge about this system, whether or not there is a common factor that explains the various phenomena associated with priority effects has not been investigated in this system, or any other system for that matter, to our knowledge. In this paper, we present and synthesize multiple independent pieces of evidence that all point to the pervasive role of pH across several aspects of priority effects in this nectar microbiome. First, we report the results of a field survey of D. aurantiacus flowers across a 200-km coastline in California, showing that microbial communities in the nectar exhibit distribution patterns that are consistent with the existence of two alternative community states, one dominated by bacteria and the other dominated by yeasts (1—3Figure 1A). Next, we describe findings from laboratory experiments that suggest that the potential alternative states we observed in the field survey are caused by inhibitory priority effects between bacteria and yeasts (Figure 1B). We then show that these priority effects are largely driven by bacteria-induced reduction in nectar pH , a finding consistent with the tri-modal distribution of nectar pH among flowers that we observed in the field (Figure 1A). Motivated by these results as well as the high variation we found in the strength of priority effects among yeast strains, we then use experimental evolution in artificial nectar (Figure 1C) to show that the low pH of nectar can cause rapid evolution of yeast traits, resulting in increased resistance to the pH-driven priority effects (Figure 1D). This evolution appears to be constrained by a trade-off between tolerance to a low-pH environment and growth in a neutral-pH environment. Using whole genome resequencing, we explore genomic differences in yeast strains evolved in isolation, in the presence of its primary bacterial competitor, and in a low pH environment (Figure 1E). Finally, we show in a field experiment that the low pH of nectar reduces nectar consumption by flower-visiting hummingbirds (Figure 1F), suggesting that pH reduction is the likely explanation for our earlier observations that nectar bacteria reduced pollination and seed set in D. aurantiacus. We end by weaving these results together in a pH-based story of the wide-ranging consequences of priority effects in this system. Figure 1 Download asset Open asset Schematic of the approaches taken in this study. (A) Field survey to characterize the distribution of yeast and bacteria in flowers as well as nectar pH of Diplacus aurantiacus, (B) initial microcosm experiments to assess strength of priority effects and identify nectar pH as a potential driver, (C) experimental evolution to study adaptation to low-pH and bacteria-conditioned nectar, (D) secondary microcosm experiments to study the effect of adaptation to nectar environments, (E) whole genome resequencing to identify genomic differences between evolved strains, and (F) field experiments to study the effect of low pH on nectar consumption by pollinators. Results and discussion Field observations suggest bacterial vs. yeast dominance as alternative states To study the distribution of nectar-colonizing bacteria and yeasts among D. aurantiacus flowers, we conducted a regional survey in and around the San Francisco Peninsula of California (Figure 1A, Figure 2—source data 1) in June and July of 2015. Floral nectar was sampled from a total of 1152 flowers (96 flowers at each of 12 sites) along an approximately 200km coastline (Dhami et al., 2018; Tsuji et al., 2016; Figure 2A). Bacteria and yeasts were cultured on different media to characterize floral nectar communities, which had been corroborated by molecular sequencing (Vannette and Fukami, 2017) and correlations between cell counts and colony forming units (Peay et al., 2012, Figure 2—figure supplement 1). Figure 2 with 4 supplements see all Download asset Open asset Sites vary in regional dominance of bacteria and yeast. (A) Ninety-six Diplacus aurantiacus flowers were harvested from each of 12 field sites in and around the San Francisco Peninsula in California, USA (Figure 2—source data 1) with (B) variable numbers of flowers classified as bacteria-dominated (blue), fungi-dominated (yellow), co-dominated (green) flowers, or flowers where microbes were too rare to determine (grey) (n=1152). (C) Flowers are often dominated by bacteria or yeast, but rarely both. Each point represents a floral community and inset plot represents zoomed-in version of the plot behind it (n=1152). (D) Co-dominated flowers were observed less frequently than expected. In panel D, each point represents a site, with the numbers indicating the site numbers shown in panels A and B. In panel A, the location of Jasper Ridge Biological Preserve (JR) is also indicated (n=12). Figure 2—source data 1 Field sites in Diplacus aurantiacus field survey. https://cdn.elifesciences.org/articles/79647/elife-79647-fig2-data1-v2.docx Download elife-79647-fig2-data1-v2.docx Figure 2—source data 2 Association between percentage of flowers colonized by yeast or bacteria per plant and the distance between host plants. To determine whether climatic factors and seasonality influence microbial abundance in this system, WorldClim bioclimatic variables (average annual mean temperature, temperature seasonality, and average monthly precipitation) were extracted for each plant and site. These variables, along with sampling date, were modeled as variables predicting bacterial and fungal abundance, respectively, in a linear mixed model with site as a random effect. https://cdn.elifesciences.org/articles/79647/elife-79647-fig2-data2-v2.docx Download elife-79647-fig2-data2-v2.docx Figure 2—source data 3 Association between bioclimate variables, date of sampling, and microbial colonization. Linear mixed model predicting bacterial or yeast abundance by average annual temperature (WorldClim bio1), temperature seasonality (WorldClim bio4), annual precipitation (WorldClim bio12), sampling date, with site location included as a random effect (n=144). https://cdn.elifesciences.org/articles/79647/elife-79647-fig2-data3-v2.docx Download elife-79647-fig2-data3-v2.docx Across all 12 sites, we found that D. aurantiacus flowers were frequently dominated by either bacteria or yeast, but rarely by both (Figure 2C, Figure 2—figure supplement 2). We used a classification method called CLAM (Chazdon et al., 2011) to classify flowers into four groups: bacteria-dominated flowers, yeast-dominated flowers, co-dominated flowers, and flowers with too few microbes to be accurately classified (Figure 2B, Figure 2—figure supplement 3). For each site, we then calculated the proportion of co-dominated flowers that would be expected if bacteria and yeasts were distributed among flowers independently of each other. This analysis showed that the observed proportions of co-dominated flowers were lower than expected by chance alone (Figure 2D; paired t-test: n=12, 95% CI [-5.3,–1.1], p=0.006). This scarcity of co-dominance may have been caused by stochastic factors such as dispersal limitation that creates spatial segregation (Belisle et al., 2012) or deterministic factors such as nectar chemistry that allows for niche partitioning. For example, some flowers may have nectar characteristics that intrinsically favor bacterial growth, while other flowers with different nectar chemistry may preferentially support yeast growth. However, another possibility is that the scarcity of co-dominance was caused jointly by stochastic and deterministic factors. Specifically, if a flower happens to become dominated by bacteria, it may prevent yeast from becoming abundant, and vice versa, through differential modification on nectar chemistry by bacteria vs. yeasts. This mutual suppression would represent inhibitory priority effects, where stochastic dispersal dictates the trajectory of local community assembly because early-arriving colonists deterministically exclude late-arriving immigrants. To examine the possibility of spatial segregation, we regressed the geographic distance between all possible pairs of plants to the difference in bacterial or fungal abundance between the paired plants. If plant location affected bacterial or yeast abundance, we should see a positive relationship between distance and the difference in abundance between a given pair of plants. Contrary to this expectation, we found no strong relationship between distance and the difference in bacterial colonization (Figure 2—figure supplement 4A, Figure 2—source data 2, LM: n=8775, p=0.07, R2=0.0003) and a slightly negative association between distance and the difference in fungal colonization (Figure 2—figure supplement 4B, Figure 2—source data 2, LM: n=8775, p<0.05, R2=0.004), suggesting that spatial segregation is unlikely to explain the observed abundance pattern. It is also possible that climatic variables affected the colonization of bacteria and yeasts. However, in a linear mixed model predicting bacterial or yeast abundance by average annual temperature, temperature seasonality, annual precipitation, sampling date, and site location included as a random effect, none of the predictors were significant (Figure 2—source data 3). This result indicates that the observed abundance pattern is unlikely to have been strongly influenced by spatial proximity, temperature, moisture, or seasonality, reinforcing the hypothesis that the distribution pattern is instead underlain by bacterial and yeast dominance as alternative stable states. To test this hypothesis more directly, we conducted laboratory experiments using Acinetobacter nectaris and Metschnikowia reukaufii, which are the most frequently found species of nectar bacteria and yeasts, respectively. Previously, we reported that M. reukaufii was the most frequently cultured species of fungi in D. aurantiacus nectar in the 12-site survey in 2015 (Table S2 in Dhami et al., 2018; see also Belisle et al., 2012). As for bacteria, Dhami et al.'s (2018) data are not extensive enough to draw a firm conclusion. However, we here present data from a multi-year survey of bacteria cultured from D. aurantiacus nectar at one site, Jasper Ridge (JR, Figure 3), from 2012 to 2022. The Jasper Ridge data support culture-independent (metabarcoding) data on bacterial species composition of D. aurantiacus nectar at the same site (Vannette and Fukami, 2017) and a nearby site (Toju et al., 2018). Namely, both culture-dependent and culture-independent methods indicate that Acinetobacter spp. were the dominant species of bacteria in D. aurantiacus nectar, followed by Neokomagataea (formerly Gluconobacter) sp. (see further detail in Appendix 1). The most common species of bacteria and yeast observed in our study system, such as species of Acinetobacter, Neokamagataea, Pseudomonas, and Metschnikowia, have been shown to be common nectar specialists in other plants as well (e.g., Fridman et al., 2012; Warren et al., 2020). Taken together, these independent pieces of information collectively indicate that, given their prevalence, A. nectaris and M. reukaufii would be a reasonable pair of species to focus on as a first step toward understanding the possible alternative states in the nectar microbial community of D. aurantiacus in our study landscape. Figure 3 Download asset Open asset Cultured bacteria and yeast from a 6year survey. Cultured nectar bacteria (A) and yeast (B) from a 6-year survey of D. aurantiacus nectar at Jasper Ridge identified by colony PCR. The number placed at the top of each bar indicates the number of colony samples analyzed. Single fungal colonies were isolated on yeast mold agar (YMA) with supplemented 100 mg/L of the antibacterial chloramphenicol. Single bacterial colonies were either isolated on Reasoner’s 2A agar (R2A) supplemented with 20% sucrose and 100 mg/L of the antifungal cycloheximide (2012–2018), or tryptic soy agar (TSA) supplemented with 100 mg/L of the antifungal cycloheximide (2019–2022). Figure 3—source data 1 Primer sequences and PCR cycles for colony PCR. https://cdn.elifesciences.org/articles/79647/elife-79647-fig3-data1-v2.docx Download elife-79647-fig3-data1-v2.docx Evidence for pH-mediated priority effects causing alternative states One way to directly test for priority effects between bacteria and yeasts is to experimentally manipulate initial relative abundance of the two microbial groups. This manipulation would allow us to determine if communities diverge such that bacteria-dominated communities develop if bacteria were initially more abundant than yeast and vice versa. Our prior work that took this approach yielded evidence for strong priority effects in D. aurantiacus nectar (Dhami et al., 2016; Peay et al., 2012; Toju et al., 2018; Vannette and Fukami, 2014). However, these studies either looked at priority effects among yeast species or, in the case of priority effects between yeast and bacteria, focused on the yeast M. reukaufii and the bacterium Neokomagataea (formerly Gluconobacter) sp. (Tucker and Fukami, 2014). The most common bacterial species, A. nectaris, may also engage in strong priority effects against M. reukaufii, but this possibility has not previously been tested. To test for priority effects between M. reukaufii and A. nectaris, we used sterile PCR tubes as artificial flowers. Each tube contained artificial nectar that closely mimicked sugar and amino acid concentrations in field-collected D. aurantiacus nectar (Peay et al., 2012). We altered the arrival order of M. reukaufii and A. nectaris and measured growth after five days, which the of individual D. aurantiacus flowers (Peay et al., 2012; Figure data 1). We found that A. nectaris exerted a strong inhibitory priority effect against M. reukaufii (Figure 4A, Figure data and vice (Figure supplement 1). Figure 4 with 4 supplements see all Download asset Open asset bacteria exert negative priority effects against nectar yeast, to reduction in nectar pH. (A) Metschnikowia reukaufii yeast after five of growth with arrival order with Acinetobacter nectaris bacteria or growth alone with on either the first or or of the experiment (B) pH of nectar after of bacterial of bacteria are associated with lower pH. The of each point represents the in panel A. are on the (C) pH nectar yeast growth in In A and that the same letter placed their were from one Figure data 1 Priority effect experiment used in priority effects experiment with yeast, fully experiment the effect of arrival order on priority effects. Download Figure data 2 Priority effect experiment Results from a linear mixed model the effect of arrival order on yeast growth, where and represents initial arrival by bacteria or yeast, respectively. and represent the growth of yeast at either arrival time or 2). less than or to Download our previous work with Neokomagataea (Tucker and Fukami, we that inhibitory priority effects against yeast were caused by bacteria-induced reduction in nectar pH, yeast growth. with this we found that of bacteria were associated with lower nectar pH Figure of yeast were associated with nectar pH (Figure supplement 2). In addition, yeast in low-pH nectar at low (Figure but not if at high (Figure supplement 3). growth was not affected by nectar pH (Figure supplement Together, these data provide further evidence that nectar pH reduction by bacteria their priority effects on yeast. One likely for this effect of pH on yeast is that, yeast first to nectar, they such as amino and bacterial growth, pH-driven suppression that would happen if bacteria were initially abundant (Tucker and Fukami, 2014; Vannette and Fukami, 2018). possible is that yeast have which the that the will that are to low pH. This mechanism has been in the yeast, and although whether it to M. reukaufii remains to be tested. In to priority effects, or stochastic could to increased variation in the of bacterial dominance among flowers 2016; et al., 2022). In previous work on M. reukaufii that changes to its broad niche et al., However, our experiments initial abundance provide evidence for priority effects between nectar bacteria and yeast. we have to priority effects of early-arriving yeast in our previous (Tucker and Fukami, 2014; Vannette and Fukami, 2018), we focus on the other of the priority effects, where initial dominance of bacteria yeast growth, in this To assess whether the nectar pH reduction by bacteria that we observed in the laboratory experiment (Figure had for variation in nectar pH among flowers in the we conducted a survey of D. aurantiacus nectar at two sites, San and Jasper in June and July of 2022. We found that the distribution of nectar pH from 2 to in a way consistent with the that D. aurantiacus nectar has a mean pH of about and that, yeast and bacteria reduce nectar pH to an average of and (Vannette et al., Specifically, we found that the distribution of nectar pH in the field was Furthermore, a model was a than a model Figure supplement 1). to the model we nectar pH had local at and (Figure which are to from the experiment we reported previously (Vannette et al., where nectar from D. aurantiacus flowers was with no the yeast M. reukaufii, or the bacterium Neokomagataea sp. under laboratory conditions (Figure In addition, we observed that and flowers were more likely to have low nectar pH, the that, in these flowers, microbes would have more chance to and the nectar pH than in and flowers (Figure Figure supplement 2). We also found that the two sites in their of nectar pH (Figure Figure supplement 2). Jasper many of the flowers had low pH consistent with bacterial In at San these flowers had pH that characterize yeast another site within the we measured both nectar pH and bacterial growth and found that D. aurantiacus flowers with bacterial to have lower nectar pH (Figure of these results support for the that microbial abundance and nectar pH are by the priority effects between bacteria and yeasts. Figure with 2 supplements see all Download asset Open asset Field survey of nectar pH in D. aurantiacus. (A) of nectar pH in individual D. aurantiacus flowers at Jasper Ridge and San flowers from which we were to a amount of nectar to of flowers of flowers, had too nectar for pH The tri-modal represent the from a with the indicated by indicate experimental pH from Vannette et al., where bacteria and yeast were in field-collected D. aurantiacus nectar as and pH was measured after four of growth. (B) of nectar pH aurantiacus flowers with and by a in D. and shown for Jasper Ridge and San Jasper and flowers with and respectively, had too nectar to pH. San and flowers with and respectively, had too nectar to pH. (C) Association between bacterial in individual flowers and nectar pH represent flowers

  PublisherDOI

• Editor's evaluation: Microplankton life histories revealed by holographic microscopy and deep learning

  2022-08-03

  peer-reviewOpen access1st authorCorresponding

  Article Figures and data Abstract Editor's evaluation eLife digest Introduction Results Discussion Methods Appendix 1 Data availability References Decision letter Author response Article and author information Abstract The marine microbial food web plays a central role in the global carbon cycle. However, our mechanistic understanding of the ocean is biased toward its larger constituents, while rates and biomass fluxes in the microbial food web are mainly inferred from indirect measurements and ensemble averages. Yet, resolution at the level of the individual microplankton is required to advance our understanding of the microbial food web. Here, we demonstrate that, by combining holographic microscopy with deep learning, we can follow microplanktons throughout their lifespan, continuously measuring their three-dimensional position and dry mass. The deep-learning algorithms circumvent the computationally intensive processing of holographic data and allow rapid measurements over extended time periods. This permits us to reliably estimate growth rates, both in terms of dry mass increase and cell divisions, as well as to measure trophic interactions between species such as predation events. The individual resolution provides information about selectivity, individual feeding rates, and handling times for individual microplanktons. The method is particularly useful to detail the rates and routes of organic matter transfer in micro-zooplankton, the most important and least known group of primary consumers in the oceans. Studying individual interactions in idealized small systems provides insights that help us understand microbial food webs and ultimately larger-scale processes. We exemplify this by detailed descriptions of micro-zooplankton feeding events, cell divisions, and long-term monitoring of single cells from division to division. Editor's evaluation This paper presents a valuable new method combining holographic microscopy and deep learning to track the behavior and growth of individual plankton. The paper illustrates the method with compelling data from two applications, zooplankton feeding behavior and diatom cell division. This paper will be of interest to plankton ecologists and ocean ecosystem modelers. The results obtained from this method will provide new insights into the trophic strategies of ocean plankton and important constraints for global ocean models. https://doi.org/10.7554/eLife.79760.sa0 Decision letter eLife's review process eLife digest Picture a glass of seawater. It looks clear and empty, but in reality, it contains one hundred million bacteria, about one hundred thousand other single-celled organisms, and a few microscopic animals. In fact, the majority of life in the ocean is microscopic and we know relatively little about it. Nevertheless, these microbes have a major impact on our lives. Microscopic algae known as phytoplankton, for example, produce half of the oxygen we breathe. For animals, birds and other large organisms in the ocean, we have a good understanding of who eats who and where the material ends up. However, for phytoplankton and other microbes, we depend on bulk measurements and averages of large groups. Bachimanchi et al. developed a method to follow individual microbes living in seawater and to observe how they move, grow, consume each other and reproduce. The team combined holographic microscopy with artificial intelligence to follow multiple planktons, diatoms and other microbes throughout their life span and continuously measured their three-dimensional location and mass. This made it possible to estimate how fast the organisms were growing and moving, and to observe what they ate. The experiments revealed new insights into how micro-zooplankton, diatoms and other microbes in the ocean interact with each other. This new method may be useful for researchers who would like to track the movements and whereabouts of microscopic planktons, bacteria or other microbes for extended periods of time. It is also a rapid method for counting, sizing, and weighing cells in suspension. The hardware used in this method is relatively cheap, and Bachimanchi et al. have shared all the computer code with examples and demonstrations in a public database to enable other researchers to use it. Introduction The role of herbivores in structuring plant communities is well established in terrestrial ecology. Already Darwin, in his foundations on evolutionary biology (Darwin, 2004), noted how excluding herbivores from a heath land transformed it into a forest of pine trees with an altogether different species composition. Single-celled micro-zooplankton take on the role of herbivores in the ocean, consuming approximately two thirds (40 Petagrams (Pg) carbon) of the primary production (Calbet and Landry, 2004). In oceanic ecology, the primary production is dominated by unicellular phytoplankton, which produce around 50 Pg of carbon annually, quantitatively slightly exceeding the production of terrestrial plants (Behrenfeld and Falkowski, 1997; Field et al., 1998). Selective grazing shapes the plankton community and drives large-scale processes such as harmful algal bloom formation and carbon export (Irigoien et al., 2005; Selander et al., 2019). Despite its importance, our understanding of the role of micro-zooplankton in shaping oceanic communities is still much less developed than that of macro-organisms, which can more readily be observed at the individual level (Glibert and Mitra, 2022). In fact, rates and fluxes in the oceanic microbial food web are still mainly inferred from indirect measurements or ensemble averages, leaving us with a limited mechanistic understanding. Quantitative estimates of primary production are mostly inferred from satellite images of ocean color (chlorophyll) using moderate resolution spectroradiometers calibrated against in situ isotope incorporation experiments (Hu et al., 2012). Ensemble-level biomass transitions during grazing events by microscopic zooplankton are calculated from dilution experiments (Landry and Hassett, 1982), where the grazer density is manipulated by dilution, and the corresponding net increase in primary production is approximated. While these methods provide good estimates of the magnitude of biomass fluxes, they do not resolve the small-scale individual interactions that drive the large-scale processes. Moreover, indirect measurements of processes such as micro-zooplankton grazing rest on assumptions that are not always fulfilled. For example, feeding rates and growth rates of both predators and prey need to be unaffected by dilution, which is often not true (Dolan et al., 2000). In addition, the dilution technique is based on chlorophyll measurements and does not account for consumption of non-chlorophyll-bearing particles, which leads to underestimation of carbon transfer (Stoecker et al., 2017). Currently, the biomass of individuals is often inferred from volume-to-carbon relationships developed over time for different trophic groups of planktons (Strathmann, 1967; Menden-Deuer and Lessard, 2000), which require cell counting and sizing followed by elemental analysis, but do not allow continuous measurements of the same individual. However, these regression relations are not very precise: the average deviation of individual data points to the regressed expression exceeds 50% (Menden-Deuer and Lessard, 2000). In addition, single cells of the same volume can differ by a factor two in dry mass, which is not possible to detect by volume-to-carbon relationships. To go beyond the current level of detail in marine microbial food webs, we need complementary techniques that can follow individual microplanktons over extended periods, while continuously monitoring their growth rate and predation events. Continuous measurements can be realized using microscopy techniques. For example, holographic microscopy can record holograms of cells under investigation in the form of interference patterns containing phase and amplitude information. The information in the holograms can be used to extract the three-dimensional position of microplanktons as well as their mass (Zangle and Teitell, 2014). Holographic imaging has already found applications in microbial studies, especially for in situ measurements of particle size distributions and their identity (Nayak et al., 2021). However, its full potential has not yet been exploited, namely for the quantitative investigation of the growth and feeding patterns of individual planktons over prolonged times. Arguably, this is because the data acquisition and processing pipelines are very computationally expensive. Here, we solve this problem by employing a technique that combines holography with deep learning. The deep-learning algorithms circumvent the long computational times and, once trained, allow rapid determination of three-dimensional position and dry mass of individual microplanktons over extended time periods. We evaluate this method on nine plankton species belonging to different trophic levels and representing the major classes of microplankton. We highlight that unlike other methods, our approach makes it possible to follow and weigh single cells throughout their lifetime, being especially useful to detail micro-zooplankton and mixotrophic life histories as feeding events can be quantitatively measured. Furthermore, the estimated dry mass can be tagged to single planktons detected in the experiments. We can track and identify both prey and predator cells and closely follow the transfer of mass from cell to cell. Finally, we observe the growth and cell divisions in diatoms by their long-term monitoring over more than one cell cycle. Results Experimental setup and deep-learning data analysis Figure 1 shows an overview of the holographic microscopy experimental setup and the deep-learning data analysis pipeline to estimate the position and dry mass of the planktons. We use an inline holographic microscope in a lens-less configuration (see details in Methods, 'Holographic imaging'). A monochromatic LED light source illuminates the sample suspension that contains the planktons under investigation. As the light passes through the sample, it acquires a complex amplitude that depends on the optical properties of the materials it traverses, generating inline holograms (Figure 1—figure supplement 1), which encode the three-dimensional position of the planktons as well as their size and refractive index. A CMOS camera located on the opposite side of the sample acquires the holograms for further analysis with a frame rate of 10⁢fps, and an exposure time of 8⁢ms. Figure 1 with 3 supplements see all Download asset Open asset Experimental setup and deep-learning data analysis. (a) Holographic microscope: Planktons suspended in a miniature sample well are imaged with an inline holographic microscope. The (cropped) example holographic image features two different plankton species: Oxyrrhis marina and Dunaliella tertiolecta (full image in Figure 1—figure supplement 1). (b) Deep-learning network 1: A regression U-Net (RU-Net, see details in Figure 1—figure supplement 2), trained on simulated holograms, uses individual holograms to predict output maps containing the segmentation of the planktons, their z-position, their dry mass m, and the distances Δ⁢x and Δ⁢y from the closest plankton for each pixel (to be used for the accurate localization of planktons). (c) Plankton 3D position and dry mass: The information obtained by the RU-Net permits us to reconstruct the 3D position of the planktons along with their dry mass (color bar). (d) Plankton sequences: Using the plankton positions obtained by the RU-Net, we extract sequences of 64×64-pixel holograms centered on an individual plankton. (e) Deep-learning network 2: The sequences are then used by a weighted-average convolutional neural network (WAC-Net, see details in Figure 1—figure supplement 3), trained on simulated data, to refine the estimations of m and z. (f) Dry mass time series: Example of a refined dry mass prediction in picograms (pg) for a micro-zooplankton (Oxyrrhis marina, orange line) and a phytoplankton (Dunaliella tertiolecta, blue line) obtained by the WAC-Net. In order to measure the position and dry mass of the planktons, the recorded holograms are analyzed by a regression U-Net (RU-Net, Figure 1b and Figure 1—figure supplement 2, see details in Methods, 'RU-Net architecture and training'). The RU-Net is a deep-learning architecture based on a modified U-Net, with two parallel arms in the upsampling path. The output of the RU-Net is a five-channel image where each channel corresponds to a heat map containing: a segmentation of the planktons from the background used to obtain a rough estimate of their x⁢y (in-plane) position; their estimated z (axial) position; the plankton estimated dry mass m; and the distances Δ⁢x and Δ⁢y from the closest plankton for each pixel (used to improve the in-plane localization). This RU-Net is implemented and trained on simulated input–output image pairs (4000 samples) using the Python software package DeepTrack 2.0 (Midtvedt et al., 2021a). The output heat maps are finally processed to obtain a prediction of the plankton three-dimensional position and their dry mass, as shown in Figure 1c. In order to increase the accuracy of the dry mass estimations, we extract time sequences of holographic images cropped around an individual plankton (Figure 1d and Figure 1—figure supplement 3) and further analyze them with a second deep-learning network. This is a weighted-average convolutional neural network (WAC-Net, Midtvedt et al., 2021b), Figure 1e and Figure 1—figure supplement 3, see details in Methods, 'WAC-Net architecture and training'. The WAC-Net determines a single estimated value of the equivalent spherical radius, as well as a more accurate value of the dry mass of the plankton in the sequence, through a weighted average of the latent representation of various holograms with learnable weights. The number of frames in the sequence is limited to 15 frames for training the WAC-Net. For inference, the length of the sequence is dependent on the application. For example, when analyzing feeding events we aim to capture dry mass dynamics on short time scales, and the sequence length is therefore restricted to a single frame. For the division events, the sequence length is 15 frames, as they occur over longer times ranging from hours to days with more recorded frames. Also the WAC-Net is implemented and trained with simulated data (4000 15-frame sequences of 64⁢px×64⁢px images) using DeepTrack 2.0 (Midtvedt et al., 2021a). Figure 1f shows an example of the dry mass output of the WAC-Net in picograms (pg)when applied on a sliding window over a sequence of holograms corresponding to a micro-zooplankton (Oxyrrhis marina) and a phytoplankton (Dunaliella tertiolecta). Dry mass estimates The combination of RU-Net and WAC-Net permits us to measure the dry mass of each plankton at any point in time. For example, Figure 2a shows a portion of an inline hologram of the micro-zooplankton species, O. marina, tracked by the RU-Net (circles). Individual O. marina cells are then traced for 30 frames and their holograms are further processed with WAC-Net to obtain an estimation of the dry mass for each cell. The orange histogram in Figure 2b shows the dry mass distribution estimated by WAC-Net. Figure 2 with 1 supplement see all Download asset Open asset Dry mass estimates. (a) Phytoplankton species Oxyrrhis marina as detected by RU-Net on a portion of experimental hologram (see Figure 1—figure supplement 1 for the complete hologram). (b) Dry mass distributions for O. marina (illustrated in the inset) obtained by applying weighted-average convolutional neural network (WAC-Net) to the experimental holograms (orange) and by volume-to-carbon relationships (gray, Menden-Deuer and Lessard, 2000). The red line is the value of the average mass estimate obtained from elemental analysis. (c) Comparison of the dry mass estimations obtained by WAC-Net and by the volume-to-carbon method for nine different species of diatoms (Thalassiosira pseudonana, Thalassiosira weissflogii), phytoplantons (Isochrysis galbana, Rhodomonas salina, Dunaliella tertiolecta), and micro-zooplanktons (Oxyrrhis marina, Kryptoperidinium triquetrum, Alexandrium minutum, Scrippsiella acuminata). The two measurements have a correlation coefficient of ρ=0.988. The dashed line represents the best fit and the error bars show the standard deviations of the distributions. The insets illustrate each species. To benchmark the dry mass measurements, we used the volume-to-carbon relationships from Menden-Deuer and Lessard, 2000 followed by an extrapolation of elemental composition using extended Redfield ratios (Anderson, 1995 see Methods, 'Dry mass estimation by volume-to-carbon relationships'). The gray histogram in Figure 2b shows the results for the case of O. marina. The dry mass predicted by the volume-to-carbon relationships (394 ± 123 pg, the uncertainty represents the standard deviations of the distribution) agrees well with the dry mass estimated by our technique (338 ± 126 pg, orange histogram). Importantly, in contrast to the volume-to-carbon relation method, the dry mass estimated by our approach can be tagged to individual cells in the image. This additional feature can be used to study the dry mass evolution of single cells (e.g., in the following sections, we will exploit this possibility in two exemplary studies of feeding and cell division events). We repeated this analysis for nine species of planktons belonging to different taxonomic groups and trophic levels in the marine ecosystem (see Methods, 'Plankton cultures'): phytoplankton species (Isochrysis galbana, Rhodomonas salina, Dunaliella tertiolecta); micro-zooplankton species (Kryptoperidinium triquetrum, Alexandrium minutum, Scrippsiella acuminata, along with Oxyrrhis marina which is used in the above discussion); and diatomic species (Thalassiosira weissflogii, Thalassiosira pseudonana). These results are summarized in Figure 2c. The data points and error bars represent the means and standard deviations of the dry mass distributions estimated by our method and the volume-to-carbon method. The two estimates correlate very well (correlation coefficient ρ=0.988). A detailed dry mass distribution comparison (along with equivalent spherical radius distribution comparison) for different species can be seen in Figure 2—figure supplement 1. As a further independent test, we also estimated the dry mass from the elemental analysis of carbon and nitrogen content in O. marina (extrapolated to the other fundamental elements hydrogen, oxygen, and phosphorous through Redfield ratios, Anderson, 1995, see Methods, 'Dry mass estimation by elemental analysis'). The resulting dry mass (453 pg, indicated with a red line in Figure 2) also confirms that our method arrives at realistic numbers. The average value indicated by the red line in Figure 2b lies within the distributions predicted by holographic estimate. Feeding events We use the phytoplankton species D. tertiolecta and the micro-zooplankton species O. marina as the prey and predator, respectively. Figure 3a–c shows the 3D traces of prey (blue) and predator (orange) during a feeding event (see 3D movie of the feeding event in Video 1). In the pre-feeding phase Figure 3a, corresponding to about 10⁢s or 100⁢frames (see also Figure 3d), the predator explores the sample volume in a random fashion. It passes the prey cell closely on a couple of occasions before it makes contact (see Videos 1 and 2 and , Figure 3—figure supplement 2, and Figure 3—figure supplement In the feeding phase (Figure for about or the predator makes contact with the prey and a about a location for while handling the In the phase (Figure 10⁢s or see also Figure 3d), the predator to its behavior and on its for new Figure 3 with 3 supplements see all Download asset Open asset Feeding events. 3D of a feeding event where (a) a predator micro-zooplankton (Oxyrrhis marina, orange a prey phytoplankton (Dunaliella tertiolecta, blue (b) on and (c) finally (see Video 1 and Figure 3—figure supplement The of traces is on the holographic images in the (see also Figure 3—figure supplement 1). (d) Dry mass time of predator and prey estimated by weighted-average convolutional neural network (WAC-Net) in the different (e) The pre-feeding dry mass distributions of the predator Oxyrrhis marina and the prey Dunaliella tertiolecta and the dry mass distribution of predator are in the The dry mass increase between and of the predator is indicated in the The dry mass of the predator the dry mass of the prey (f) is a correlation between dry mass of predators and dry mass of prey for feeding events. The dashed line represents the best Video 1 Download asset This be in because does may still the for Download as Download as Download as Feeding event 1. Video 2 Download asset This be in because does may still the for Download as Download as Download as Feeding event Figure shows the dry mass time of prey and predator during the feeding As the feeding events on a short time to the frame rate of the we use WAC-Net with a sliding window of one the resolution of the dry mass The dry mass distributions of the prey and predator in and are shown by the (Figure to the In the pre-feeding the prey dry mass is measured to be ± 1 and the predator ± The represent the standard error of the The dry mass distribution of the predator is ± The in predator dry mass and pre-feeding closely the prey dry mass (Figure This that the predator has its a of the of the dry mass during each individual feeding In Figure we the results of the dry mass increase in feeding events. The increase in the predator dry mass in the phase well with the pre-feeding dry mass of the prey (correlation coefficient The of the best fit line also that on average of prey is by the predator in a feeding it is possible to individual feeding rates and, predator cells are followed over also growth that how much of the biomass is into predator of a plankton The technique we have developed can follow the life histories of planktons, over time from hours to To demonstrate we use a diatom species, weissflogii, which is and a of (Figure we image a and two of its continuously the in their dry mass using the which we already used to estimate the dry mass of in Figure (see Methods, 'Holographic imaging'). We a of diatoms in the sample which we with a light source light source not to the holographic imaging to the cell Figure with 1 supplement see all Download asset Open asset and cell division of a life of a diatom and its (a) the cell (b) into two cells (c) the cells to grow, cell division (f) Dry mass time through estimated by weighted-average convolutional neural network (WAC-Net) (see also Video 3 and Figure supplement 1). cell dry mass is estimated when it has at least of around it to of the the corresponding times are indicated by the gray dashed A in the dry mass can be with the cells in the relation between the of the dry of the cells and the dry mass of the cell for different division events The dashed line represents the best Video 3 Download asset This be in because does may still the for Download as Download as Download as Figure shows the growth and division of a diatom imaged over a small portion of the The cell in Figure into two approximately into the (Figure that the biomass does not between the division in terms of cell size already been shown in both bacteria and our experiments show that the cells of the biomass from the cell. the two cells slightly (Figure and the cell with the biomass of the two at (Figure Figure shows the dry mass of the and cells as the We that, while the dry mass of these cells is continuously the WAC-Net estimates the most when the cells are we the dry mass measurements as when the cells have at least of around them before or each these times are indicated by the gray dashed in Figure The cell dry mass at is estimated at ± 2 The dry mass of its two cells at as as the two cells is ± 1 and ± 1 pg, is to the dry mass of the cell. As the one of the cells a second of cells (Figure dry are ± 3 and ± 1 Figure their is to the mass of their cell. The uncertainty in the dry mass value represents the standard error of the for frames around the point dashed in Figure We have repeated this with various cell with multiple division events. Figure shows the correlation between the cell dry mass and the of the dry for cell It is to that the division events of occur when the cell between and pg, with a value of This of a value prediction in dry mass for a division event is for the time to this method and is example of the of information that can be by employing a technique that can continuously measure single cells throughout the cell cycle. Discussion The of combining holographic microscopy with deep-learning algorithms lies in the to position and dry mass of individual plankton cells over extended time periods. The method is and and quantitative of trophic interactions such as feeding and biomass increase throughout the cell detail to the life histories of marine The standard methods to the biomass of cells elemental analysis on cells from single species or the biomass from volume-to-carbon relationships from multiple elemental of plankton organisms of different (Strathmann, 1967; Menden-Deuer and Lessard, 2000). analysis has the of detailed measurements of individual and it is and provide individual cell relationships can provide biomass estimates of individual living cells as long as the volume of the cells can be measured (Menden-Deuer and Lessard, the around the is (e.g., the estimated value for O. marina used in Menden-Deuer and Lessard, 2000 is than that measured by elemental Moreover, the volume-to-carbon relationships do not account for the of the cell (e.g., as in 'Dry mass cells of the same species can differ by more

  PublisherDOI

Recent grants

• Collaborative Research: Teasing apart coexisting cyanobacteria in the Laurentian Great Lakes

  NSF · $747k · 2018–2023

Frequent coauthors

• Sallie W. Chisholm

  Massachusetts Institute of Technology

  28 shared

• Matthew B. Sullivan

  The Ohio State University

  17 shared

• Dianne K. Newman

  California Institute of Technology

  16 shared

• Jack A. Gilbert

  University of California, San Diego

  14 shared

• Claudia Steglich

  University of Freiburg

  13 shared

• Sean M. Gibbons

  Institute for Systems Biology

  12 shared

• Adam C. Martiny

  University of California, Irvine

  12 shared

• Edward F. DeLong

  University of Hawaiʻi at Mānoa

  8 shared

Labs

• Maureen L. ColemanPI

Awards & honors

• 2014 Sloan Research Fellow in Ocean Sciences