model, organisms with highly streamlined genomes had reduced energy investment in replication and reduced expression and maintenance
of specialist gene functions. In contrast, organisms with high genome complexity were capable
of responding better to more nutrient- and energy-replete environmental conditions. This assumption was consistent with research illustrating
genome streamlining in some marine bacterio-plankton (15, 16), yet allowed specialist life strategies that enabled more complex organisms to
thrive in oligotrophic environments.
The model constructed genomes for each organism by randomly assigning genes from the
predetermined ensemble (table S3) of known
genes or environmental responses for which an
indicator gene has not been identified. The ensemble accounts for the cycling of nitrogen and
organic matter in the Amazon River plume, for
which we have “omics” observations and biogeochemical data (17–19). The model genetic pathways each mediate a biochemical transformation
or metabolic process. Organismal responses were
adapted using the costs and benefits estimated
for each genetic pathway in the organism’s genome (fig. S1 and table S3). The actual costs and
benefits of maintaining or transcribing genetic
pathways are unknown for most metabolic functions; thus, their relative magnitudes were estimated indirectly. First, the organismal integrated
cost-benefits were scaled to create variance in
the size-structured environmental response functions, consistent with observations (fig. S1). Second, the relative cost-benefits were adjusted to
ensure that all the model genes existed in at least
one organism somewhere within the model domain. Third, for nitrogen fixation and nitrification, for which we have observations of genes,
transcripts, and biochemical rates, cost-benefits
were adjusted until the model simulated reasonable integrated basinwide rates of nitrogen fixation and local rates of nitrification. Last, our
limited direct observations of gene and transcript
abundance provided a constraint on the model.
Importantly, all organisms given a particular genetic function received the same cost-benefits
for that gene, although their integrated responses
varied widely according to the other genes they
carried. Because different microbial communities
emerged in each model run, the basinwide biochemical transformations mediated by the community were not tuned to match observations
and were thus free to vary as environmental conditions changed.
Seven substrates and 68 microbes coexisted
at any given time, which is computationally taxing but still insufficient to explore all genetic
combinations. To incorporate broader genetic
potential, organisms whose biomass never rose
above 1% of the local community anywhere in
the model were replaced with a new organism.
Most organisms were rapidly replaced because
their genomes did not encode a sustainable
organism in the model ocean. Others persisted
for long periods but were ultimately replaced by
new organisms that were better adapted to exploiting the model ocean and community (Fig. 1).
1150 1 DECEMBER 2017 • VOL 358 ISSUE 6367 sciencemag.org SCIENCE
Fig. 2. Normalized transcript concentrations in the Amazon River plume from June observations
in 2010 and modeled June. In (A) to (C), normalized transcript concentrations are shown along the
salinity gradient. (A) Transcription of genes involved in nitrate uptake and storage. (B) Transcription
of the amtB gene for ammonium transport and the glnA gene that indicates nitrogen limitation.
(C) Transcription of the amoA gene for nitrification. (D) Transcription of genes for degrading lignin-related
or aromatic compounds along the normalized colored dissolved organic matter gradient. l, liter.
−80˚ −40˚ 0˚
−80˚ −40˚ 0˚
Fig. 3. Satellite observations and one model estimate of upper ocean particle density in June.
(A) Satellite-derived >5-mm particle number per cubic meter. (B) Model run showing surface >5-mm
particle number per cubic meter assuming conversions of 20 and 25 fg of nitrogen per cell volume (measured
in cubic micrometers) for nano- and microparticles, respectively. (C) Satellite-derived <5-mm particle
number per cubic meter. (D) Model run showing surface <5-mm particle number per cubic meter assuming
a conversion of 1 fg of nitrogen per cell volume (measured in cubic micrometers) for picoparticles. Observations
are 2003–2007 Sea WIFS (Sea-viewing Wide Field-of-view Sensor) climatological data.