of 32
Supporting Information for “Single cell activity reveals
direct electron transfer in methanotrophic consortia”
Shawn E. McGlynn, Grayson L. Chadwick,
Christopher P. Kempes, Victoria J. Orphan
To whom correspondence should be addressed; E-mail: vorphan@gps.caltech.edu
Contents
1 Proposed Interaction Mechanisms Between Archaea and Bacteria in-
volved in anaerobic methane oxidation
2
2 Sediment composition, sample preparation, and analytical measurements 3
3 Analysis of NanoSIMS data
8
4 Metrics for Degree of Mixing
11
5 Spatial Effects on Activity of Single Cells
13
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
5.2 Relationship Between Cell Activity and Distance to Surface . . . . . . . . 14
5.3 Relationship Between Cell Activity and Distance to Partner . . . . . . . . 14
5.4 Interfaces in cellular neighborhoods . . . . . . . . . . . . . . . . . . . . . . 14
6 Genomic Evidence for Direct Electron Transfer in ANME-2 Archaea
20
6.1 Distribution of putative multiheme cytochromes in microbial genomes . . . 20
6.2 ANME-2 multiheme cytochrome proteins with putative s-layer domains . . 21
7 Modeling Overview
22
7.1 Diffusive chemical exchange . . . . . . . . . . . . . . . . . . . . . . . . . . 22
7.2 Conceptual Model for Interspecies Direct Electron Transfer . . . . . . . . . 26
WWW.NATURE.COM/NATURE | 1
SU
PPLEMENTAR
Y INFORMATION
doi
:10.10
38/na
ture
1
5512
1 Proposed Interaction Mechanisms Between Archaea
and Bacteria involved in anaerobic methane oxida-
tion
In 1994, Hoehler and co-workers proposed the existence of a microbial consortium medi-
ating methane oxidation coupled to sulfate reduction based on field and laboratory mea-
surements of methane oxidation within methanogenic sediments [23]. In their hypothesis,
methanogens would carry out reverse methanogenesis with concomitant production of
H
2
, and this H
2
would be efficiently scavenged by a sulfate-reducing bacterium, maintain-
ing favorable thermodynamics of the redox couple. Later, using 16S rRNA FISH probes
targeting putative methane-oxidizing ANME archaea, Boetius [10] observed ANME cells
belonging to the Methanosarcinales in consortia with sulfate-reducing members of the
Desulfobacteraceae, supporting the Hoehler proposal of a structured syntrophic rela-
tionship, where the close physical proximity observed between cells would presumably
facilitate intercellular metabolic coupling. Follow up work by Nauhaus and colleagues in-
vestigated molecular hydrogen, formate, acetate and methanol as possible intermediates
between the two organisms, but they were unable to find compelling evidence that these
could function as intercellular electron shuttles during AOM [17]. Instead, these authors
postulated that cells in direct physical contact could possibly utilize redox components
positioned outside the cell as agents of electron transfer. In this scenario, direct electron
transfer occurring through closely packed cells in consortia would be an alternative to the
transfer of a molecular intermediate.
Other possible mechanisms, including methanethiol production and exchange [19] and,
more recently zero valent sulfur transfer [20], have been proposed. In particular, the
interaction mechanism hypothesized by Milucka and co-workers represents a significant
departure from previously proposed interaction scenarios for sulfate-coupled AOM, where
ANME archaea are proposed to independently carry out the full reaction of methane
oxidation coupled to sulfate reduction, with electron transfer to sulfur atoms terminating
at the S(0) oxidation state within ANME cells. This S(0) was then proposed to be
disproportionated in an unprecedented reaction from HS
2
, leading to the formation of
HS
and SO
2
4
in a ratio of 1:7. In this scenario, the ANME archaea were suggested
to be capable of AOM independent of the associated sulfate-reducing bacteria, and the
interaction occurring between organisms would be better described as commensal, rather
than an obligate mutualism.
These above mentioned studies give rise to three possibilities as to the nature of
ANME-SRB interactions which may be occurring during net methane oxidation:
1. An as yet unidentified molecular intermediate other than those tested is involved
in syntrophic the coupling, and this unknown molecule (or mixture of molecules)
obviates the thermodynamic constraints associated with a diffusible intermediate
coupled with the low net energy yield of anaerobic methane oxidation.
WWW.NATURE.COM/NATURE | 2
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
2. SRB are dependent on ANME for the formation of HS
2
, but thermodynamic predic-
tions indicate ANME are not dependent on SRB for the removal of the intermediate
because sulfate reduction to disulfide with methane as the electron donor is exer-
gonic over a wide range of disulfide concentrations [20].
3. ANME-SRB consortia are syntrophic, and syntrophic coupling occurs through direct
passage of electrons to the SRB which are poised at an appropriate potential.
Expected outcomes from microbial interactions within these three scenarios predict
differences in the emergent spatial patterns of cellular activity which can be compared
against our FISH-nanoSIMS data of individual ANME archaea and SRB cells in AOM
consortia:
1.
A molecular intermediate
. In the case of a molecular intermediary of syntrophic
exchange between partnering cells, consortia geometry (that is, size and cellular
arrangement) will be a strong driver on both the magnitude and distribution of
metabolic activity amongst partnering cells, owing to the expected rate of diffusion
compared to cellular growth rates [13].
2.
Sulphate reducing ANME hypothesis
. In this case, ANME activities are not
expected to be related to SRB activities, because the proposed reaction is exergonic
at all reasonable HS
2
product concentrations [20], however, bacterial activities will
be related to spatial proximity to ANME cells, who are proposed to be the source
of zero valent sulphur required for SRB sulfur disproportionation.
3.
Direct electron transfer
. Here, ANME-SRB activities should be positively corre-
lated, but the magnitude and distribution of cellular activity within AOM consortia
is less strongly linked to aggregate size or spatial arrangement of ANME-SRB cells
as suggested by the modeling presented here.
2 Sediment composition, sample preparation, and an-
alytical measurements
Sediment sample acquisition
Sediment was obtained from a white mat covered active
methane seep at Hydrate Ridge North (station HR-7; (44
40.02’N, 125
6.00’W; 600m
depth) using a push coring device operated by the ROV Jason II during the AT 18-
10 Hydrate Ridge Aug/Sept 2011 expedition. Push core 47 was processed shipboard
immediately following recovery and the upper 9 cm of sediment stored under N
2
in mylar
at 4
C until dispensing the top 9 cm of the core into a 1L overpressurized CH
4
(30 psi)
large-scale microcosm incubation in anoxic filtered sea water.
WWW.NATURE.COM/NATURE | 3
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
15
N-isotope labeling experimental setup
10ml of the sediment slurry described
above was aliquoted into 72ml serum vials within a Coy anaerobic chamber (2.5% H
2
) and
15
NH
4
was added from an anoxic 500mM solution to achieve a final concentration of 1mM.
Bottles were stoppered with butyl rubber and flushed with methane, then over pressurized
to 20psi methane, covered in aluminum foil, and incubated at 7
C. Geochemical analysis
of the slurry was conducted by ion chromatography revealing 496
μ
M NH
4
, 321
μ
M
thiosulfate, and 24mM sulfate at the start of the incubation.
Microcosm sampling and embedding
Aliquots of the slurry were removed from the
incubations at 6, 20, and 64 days and fixed with 2% paraformaldehyde buffered with
PBS. Fixation was for 1 hour on ice and followed by three PBS washes accomplished by
pelleting the sediment by centrifugation followed by re-suspension (1min, 1000 x g). To
separate microbial consortia from sediment matrix, fixed sediment was first diluted into
PBS to achieve a final volume of 500
μ
l and sonicated in two 15 second bursts at setting
3 ( 6V(rms) output power on a Branson Sonifier W-150 ultrasonic cell disruptor) on ice
with a sterile remote-tapered microtip probe (Branson) inserted into the liquid.
To the resulting suspension was then added 500
μ
l Percoll at the bottom of the tube,
and this mixture was centrifuged at 4
C for 20 min. After centrifugation, the supernatant
containing consortia was removed from the tube and the percoll removed from the solu-
tion by PBS buffer exchange over a 3
μ
M TSTP filter on a 15 ml filter tower. Finally, the
percolled material was concentrated by pipetting approximately 1 ml slowly from a pipet
tip onto the 3
μ
m filter in a small area (approximately 2mm diameter). After this, the ma-
terial was overlaid with molten agar (2% nobel agar in 50mM Hepes pH 7.4 35g/L NaCl).
The agar plug was peeled from the filter after solidifying, sliced into small pieces, and
embedded in technovit 8100 resin (Heraeus Kulzer GmbH) following the manufacturer’s
protocol with the exception that ethanol was used rather than acetone for dehydration,
and the sucrose infiltration step was omitted. Blocks were cut with a Leica microtome
equipped with a glass knife, and 1 micron sections stretched on a water droplet on a
polylysine coated slide with teflon wells (Tekdon Inc). The slides were then air dried
depositing the sections on the polylysine slide and stored at room temperature until use.
Sulfide measurements
The 6, 20 and 64 day time points were assayed for sulfide
production using the method of Cline [57]. Sulfide production in the bottle was estimated
to be 0.0019mM per day over the sampling period.
Average doubling time of microbial consortia in the microcosm experiments,
and rational for using nitrogen isotopes for determining microbial activity
At
the 20 and 64-day time points, the
15
NH
+
4
containing incubation was sampled to estimate
biosynthetic rates within the microcosm. Microbial consortia were separated from the
WWW.NATURE.COM/NATURE | 4
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
sediment matrix, and embedded in Technovit for nanoSIMS analysis as described above.
To estimate the doubling time of consortia in the incubation, nanoSIMS data were ac-
quired on non-phylogenetically identified consortia at the 20 and 64 day time points and
averaged to obtain a specific growth rate, with the assumption that a 50% atom percent
increase in
15
N would represent one doubling period of the organisms. A specific growth
rate of 0.0068/day was calculated, giving a doubling time of approximately 102 days.
There have been numerous estimates of the growth rate of consortia involved in AOM
[16, 61, 62, 24, 63, 64]. Variability in these values may stem from differences in the
incubation set up (for example temperature or methane partial pressure), as well as geo-
graphic, and microbial composition related differences between samples and set ups. In
general though, the relative agreement between methods based on counting biomass and
isotope incorporation amidst significant sample diversity seems to indicate that all these
techniques function as decent indicators of cell growth.
Fluorescence in situ hybridization
The phylogenetic identity of microorganisms
within consortia within the sample was determined using conventional FISH with the
probes described in the table below. FISH hybridization was conducted using standard
protocols [60] with percoll separated aggregates immobilized on 3 micron TSTP filters.
Visualization via epi-fluorescence was accomplished by mounting FISHed material with a
mixture of DAPI-citifluor (5
μ
g DAPI/ml) and imaging with a 60x objective (Olympus).
Hybridization with the ANME-2
538 and DSS member seep1a
1441 probes showed
that approximately 11% of the aggregates in the incubation were ANME-2
538:seep1a
1441
pairs. A similar amount (12%) of ANME-2
538 targeted consortia were found to pair
with a non seep1a
1441 targeted bacterial partner. 3% of the incubation was found to
be seep1a
1441 paired with a non ANME-2
538 targeted archaea or with unidentified
cells (non archaeal). FISH with the ANME-2c-760 probe gave a similar results: 12%
of ANME-2c consortia were found with DSS-658 hybridized cells (likely seep1a given the
above mentioned results), and 9% were with an unidentified partner. 4% of the incubation
was ANME-2c paired with a non-EUB probe identified partner. It was concluded that
approximately half of the ANME-2c-bacterial consortia exists partnerships with the spe-
cific seep1a-DSS group (those hybridized by the ANME-2
538 and ANME-2c-760 probes)
and another half with a non-identified partner.
Use of the newly designed ANME-2b
729 probe showed that approximately 14% of
the consortia in the incubation were ANME-2b paired with a delta
495 targeted bacterial
partner. 3% of the incubation was ANME-2b paired with a non delta
495 targeted part-
ner.
WWW.NATURE.COM/NATURE | 5
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
From the above, it was concluded that the ANME-2c and ANME-2b - deltaproteobac-
terial pairs are prominent aggregate types in the incubation. These were selected for
nanoSIMS analysis. Prior to nanoSIMS, FISH with the seep1a-1441, arc915, and delta-
495a probes was used to identify seep1a-archaeal pairs (very likely seep1a-ANME-2c pairs
given the above results) and archaea paired with deltaproteobacteria (very likely a mix-
ture of ANME-2b and ANME-2c deltaproteobacteria pairs given the above results). Thus,
two major groups of organisms were identified for nanoSIMS analysis: the specific seep1a-
ANME-2c pairs, and a mixture of ANME-2b/2c paired with an unknown deltaproteobac-
terial partner targeted with delta
495.
Fish probes used in this study
1. Arc915 [66]
2. Anme2:538 [67]
3. Anme2c:760 [15]
4. Anme2b:729 (This study)
5. Eub
mix [68]
6. d495a and competitor [69] [70]
7. seep1a
1441 [25]
Preparation of FISH hybridized samples for nanoSIMS analysis
The 20-day
time point sample - where measured sulfide in the incubation was approximately 2.3mM
was the subject of detailed analysis by FISH-nanoSIMS. Mapping of FISH stained consor-
tia for nanoSIMS analysis was carried out on thin sectioned aggregates in Technnovit 8100
resin (1
μ
m thickness) which were mounted onto teflon coated microscope slides. Identifi-
cation of consortia was made using the arc-915(fitc), anme-1-350(cy3), seep1a-1441(cy3),
and delta495 (cy5) probes mixed in a 40% formamide hybridization buffer. After FISH,
the coverslip was removed, and the slide was washed gently in DI water. After air drying,
the slide was carefully broken into the dimensions of the nanoSIMS holder (filing was
necessary for precise fitting) and the sample was gold coated with 30nM gold by sputter
coating for conductivity.
nanoSIMS operation
Prior to analysis, the areas for analysis were pre-sputtered with
a 90pA beam until approximately 15,000cps
12
C
14
N were reached at the analytical settings:
approximately 0.3pA primary Cs
+
ion beam at (D1=4, ES=2) and a 25 micron raster. At
least two image frames were collected for each consortia analyzed. Images were acquired
at either 256x256 or 512x512 pixel resolution, depending on the size of the image captured.
WWW.NATURE.COM/NATURE | 6
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
15
N accumulation by cells was determined from measurement of the
12
C
14
N
and
12
C
15
N
ions.
Transmission Electron Microscopy
A modified protocol from Ghineas and Simionescu
was followed for heme staining [58, 59, 35]. Fixation was accomplished on ice by mixing
one volume sediment slurry with 1/2 volume each of 1) 5 % gluteraldehyde in 25mM
Hepes pH 7.4, 17.5 g NaCl and 2) 8% paraformaldehyde in 37.5mM Hepes pH 7.4, 26.25
g/L NaCl to achieve final aldehyde concentrations of 2% paraformaldehyde and 1.25%
gluteraldehyde. After fixation, washing was completed by 5x 1ml washes with resuspen-
sion and centrifugation (1 minute, 1000 x g) in 50mM Hepes pH 7.4, 35g/L NaCl. The
sediment was then percolled to separate cellular aggregates from inorganic particles and
embedded in molten nobel agar (see above section “Microcosm sampling and embedding”
for more information). A solution of 3,3’-diaminobenzidine tetrahydrochloride (Sigma-
Aldrich, St. Louis, MO (DAB) at a concentration of 0.0543g DAB/ml was made in 1M
HCl by sonication at setting 3 until the powder was dissolved. After sonication, the dis-
solved DAB solution was added into 50mM Tris HCl pH 8 to achieve a final concentration
of 0.0015g DAB/ml buffer. The solution was briefly sonicated again and immediately
filtered through a 0.22
μ
m syringe filter. H
2
O
2
was added from a 30% aqueous stock to
the DAB solution to achieve a final concentration of 0.02%. This H
2
O
2
/DAB solution was
added to agar embedded sediment and incubated for 2.5 hours at room temp on a rocker.
A DAB solution without H
2
O
2
was added to a separate set of samples for comparison.
The DAB solution was removed with 5x 1ml washes with 100mM Hepes pH 7.8.
Next, a 1% OsO
4
solution was made by dilution of a 4% aqueous stock into 100mM
Hepes pH 7.8. 1/2 ml of the 1% OsO
4
solution was added to each of the tubes containing
the agar embedded sediment samples and this was incubated 90 min on ice. The samples
were then washed with 5x 1ml changes of 100mM Hepes pH 8 solution and the samples
embedded into LR white by a graded ethanol series (15 minutes each of 25%, 50%, 75%,
100% x 3 times, follwed by 50% LR White Resin, 50% ethanol on a rocker for 30min. The
samples were then moved to 100% LR White Resin for 1 hour on a rocker followed by a
LR white replacement and placement at 56
C for 2 days for polymermization. The blocks
were then sectioned at 200nm and floating sections were mounted on copper grids which
had been briefly flamed and rinsed in water. Thin sections were examined and imaged
by a FEI Tecnai Spirit TEM operated at 120 kV. Conventional transmission electron
microscopy 2K by 2K images were acquired using TVIPS F224 CCD camera.
WWW.NATURE.COM/NATURE | 7
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
3 Analysis of NanoSIMS data
Introduction
Previous SIMS measurements have not indicated that growth in these
consortia was limited to syntrophic partner interfaces [16], however that work was not
conducted at sufficient resolution to quantitatively determine activity relationships as was
done here. In this work, our goal was to specifically analyze the activity of individual cells
as it relates to their surroundings - below we describe how this analysis was performed.
This section covers our workflow for taking a FISH image and a corresponding nanoSIMS
isotope map and producing a finalized dataset consisting of cell locations, phylogenetic
identities and isotope ratios. This data is then used as a starting point for all down-
stream spatial analyses. Briefly, this process involves a transformation of the FISH image
onto the nanoSIMS image; manual drawing of regions of interest (ROIs) around individ-
ual cells on the nanoSIMS image; phylogenetic classification of those cells based on the
FISH image; extraction of nanoSIMS isotope information for each ROI; and finally the
inverse-transformation of the ROI centroids. Each process is explained in detail below,
and
Extended Figure 1
illustrates the process.
ROI isotope data generation
The initial processing of all FISH and nanoSIMS data
was done in the Matlab program Look@nanoSIMS (LANS), which is designed to read
nanoSIMS .im data files, and has a range of analysis tools [48]. In this study, the LANS
interface was used for pre-processing the data, and ROI drawing, while the rest of the
analysis was left to custom Matlab scripts. Below are the steps conducted in LANS.
First, a nanoSIMS .im file is loaded into LANS, and the planes of data are aligned and
combined following the program manual [48]. Next, the “Align External Template Image”
option is used to warp the FISH image of the aggregate onto the nanoSIMS image. This
process is essential for correct identification of cells on the nanoSIMS image because in the
process of acquiring data it is common for the isotope map to appear slightly warped when
compared to the corresponding light microscopy image. Warping was accomplished by
selecting well-resolved fiducial marker points in side-by-side FISH and nanoSIMS images,
and then constructing a transform function from these points.
Extended Figure 1a
and
1b
shows example markers on the FISH and the nanoSIMS image respectively.
Ex-
tended Figure 1c
illustrates the resulting warp of the FISH image onto the nanoSIMS
image.
To illustrate the necessity for using a transform function instead of manually overlay-
ing the FISH and the nanoSIMS images,
Extended Figure 1d
shows a manual overlay
of the original FISH image in yellow, and the transformed FISH image in blue. The effect
is slight, but it is clear that the overlay is quite accurate at the top of the image, and off
by approximately one cell length at the bottom of the aggregate. This precludes the use
WWW.NATURE.COM/NATURE | 8
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
of simple image manipulations such as resizing or rotating to attain an accurate overlay.
Also, such manual transformations cannot be accurately inverted, which is an integral
part of this analysis (as described later in this section).
With the FISH image transformed onto the nanoSIMS image, regions of interest (ROIs)
corresponding to single cells are selected. Each cell in an aggregate is outlined by hand,
and this process is exclusive, i.e. no pixel can then be assigned to two different ROIs.
After ROIs were defined the following data was exported for each aggregate. First, a
“.mat” file was saved that contained the counts for each isotope of interest (in our case,
14
N
12
C,
15
N
12
C,
12
C,
13
C,
31
P,
32
S, and
34
S). Next, the transformed FISH image, as well
as the points that were used to define the transform function were saved. Finally, a file
was saved that contains the index number and spatial information for each ROI.
Each ROI needs to be phylogenetically classified based on the information from the
FISH image. This is completed in a semi-automated fashion by a custom Matlab script.
This script takes a warped FISH image and the ROI data, and for each ROI the average
intensity of each channel in the FISH image is calculated, and the phylogenetic identity
is assigned automatically based on which channel is most intense over the entirety of the
ROI. Occasionally the background fluorescence is such that this automated selection fails
for some of the ROIs, and these were then re-assigned phylogenies by hand based on
manual inspection of the FISH image.
Extraction of nanoSIMS Data
After ROIs have been classified, each ROI is used as
a mask to extract the isotope counts contained within it. For each ROI the raw counts
for each isotope is stored for each pixel, as well as the average counts across the whole
ROI. This data is combined with the ROI classification data and exported.
Inverse Transformation of ROI Spatial Data
Much of the subsequent analysis for
this study involves the spatial distance between cell centroids. This is a trivial calculation,
however the ROIs and their centroids must again be transformed, this time using the
inverse of the original transform applied to the FISH image. This is because, as noted
above, the nanoSIMS introduces slight, non-linear spatial warping. As a result, for the
accurate measurement of distance between centroids in microns, the x,y coordinates of
centroids were transformed back into the “FISH space” with the inverse of the original
transform
Extended Figure 1f
. Reliable distance calculations can be made in units of
microns using our pixel to micron conversion for microscope images.
WWW.NATURE.COM/NATURE | 9
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
On the relationship between two-dimensional slices and whole aggregate be-
havior
It has not escaped our attention that the data presented in this paper represents
two-dimensional slices of aggregates that are three-dimensional and spheroidal in shape.
A potential concern within our empirical observations is that the inferences made from
two-dimensional slices could be missing the effects of the layers above and below the slice.
However, if strong gradients, consistent with previous diffusive models, existed in the
three-dimensional aggregate these should be apparent in the two-dimensional slice for the
correlations between various distances and activity. If the three-dimensional structure
represented a shell of the two types with strong gradients then any two-dimensional slice
would be the same given symmetry. If the geometry were completely random then the sur-
rounding cells should on average match what surrounds a cell in two-dimensions. The basic
assumption for any geometry in between these two extremes is that the two-dimensional
slice gives representative statistics for the three-dimensional spatial arrangement.
WWW.NATURE.COM/NATURE | 10
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
4 Metrics for Degree of Mixing
Introduction
ANME-SRB consortia display spatial arrangements that vary substan-
tially even within narrow sediment horizons. The canonical aggregate is the so-called
“shell-type”, in which an archaeal core is surrounded by a layer of bacteria. Aggre-
gates are also found in the “mixed-type” however, in which the archaea and bacteria are
more evenly distributed throughout, with the perfectly mixed end-member resembling a
checkerboard of the two species. Consortia fall somewhere along a continuous spectrum
of mixing between the perfect shell and the perfect checkerboard morphologies.
It has been proposed in the literature that syntrophic communities optimize their
activity by achieving the highest degree of mixing between the partners (references and
discussion in the main text). In order to test whether we could detect such a pattern in
our unique dataset, we set out to develop a metric that captures the degree of mixing of
an aggregate in a single analytical value. An extensive literature search did not reveal
anything specifically designed for microbial communities that fit this description. The
closest measure we could find is available in the Daime software package [49], which has
a spatial arrangement analysis function designed for FISH images. Unfortunately this
function outputs a plot that describes how clustering varies with distance. While this is a
very useful tool in some instances, it does not satisfy our requirement for a single metric
value for the degree of spatial mixing within each consortium.
After our literature search failed to yield an acceptable metric, we designed our own
based on statistics originally developed for measuring spatial autocorrelation. For this we
assigned an identity value of 1 or 0 to each ROI, then examined how spatially autocor-
related the identity values were over the entire aggregate. Two metrics were developed
which approached the problem in slightly different ways, but behaved similarly on both
computer-simulated mock aggregates and observed aggregate data (see
Extended Fig-
ure 4
). One of the metrics, Moran’s I [50], is usually applied to continuous data, while the
second, join counting [51], is used for categorical data. Both metrics compare the value
of a measurement at a specific location with other nearby measurements. In our case, the
measurement is the phylogenetic affiliation of the ROIs in question, and for Moran’s I the
value of 1 for archaea and 0 for bacteria was arbitrarily chosen.
The formulations of the metrics are shown below. In both cases a weight function must
be applied which describes how the neighboring ROIs are weighted in the calculation of
the spatial autocorrelation. This weight function can either be a continuous function of
distance (decreasing weight with increasing distance), or a function which gives equal
WWW.NATURE.COM/NATURE | 11
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
weight to measurements occurring in the predefined neighborhood of the measurement
in question, a zero weight to all measurements outside that neighborhood. For the re-
sults presented in the main text the common weight function of inverse square of the
distance between the measurements was used, although other powers of inverse distance
were tested, and they had no effect on the general trends presented in the text (data not
shown).
In the following equations, the weight function
w
ij
is always equal to 1
/r
2
ij
, where
r
ij
is
the distance in microns between ROIs i and j. The functions of the form
f
aa
are piecewise
functions, which return 1 if the subscript condition is met. For example, if ROIs i and j
are an archaea and bacteria,
f
aa
=
f
bb
= 0 and
f
ab
= 1. The functions
J
aa
,
J
bb
and
J
ab
simply add up all the weights associated with the specific joins, archaea-archaea, bacteria-
bacteria, and archaea-bacteria, respectively. Since we need a single value to capture the
spatial mixing of partners within an aggregate we combined
J
aa
,
J
bb
and
J
ab
for the
calculation of “J”. As shown below these sums are normalized to the number of joins of
that type (
n
a
choose 2, etc), and ratio of the average within-species join to the average
between-species join is calculated. If this ratio is large, it means that on average the joins
within a species are closer, and overall the consortia consists of segregated populations of
bacteria and archaea.
I
=
n
ij
w
ij
ij
w
ij
(
x
i
x
)(
x
j
x
)
i
(
x
i
x
)
2
(S1)
J
aa
=
ij
w
ij
f
aa
(
x
i
,x
j
)
(S2)
J
bb
=
ij
w
ij
f
bb
(
x
i
,x
j
)
(S3)
J
ab
=
ij
w
ij
f
ab
(
x
i
,x
j
)
(S4)
(S5)
J
=
J
aa
(
n
a
2
)
+
J
bb
(
n
b
2
)
2
J
ab
n
a
n
b
(S6)
Testing spatial mixing metrics on simulated aggregates
The spatial mixing met-
rics shown above have different ranges of values. Moran’s I and the J value approach
-1 and 0, respectively, for cases of perfect mixing. In the other extreme, Moran’s I and
J approach 1 and
, respectively, for cases of perfect segregation. For random distri-
butions of cells, Moran’s I and J equal 0 and 1, respectively. To make sure the spatial
WWW.NATURE.COM/NATURE | 12
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
mixing metrics we developed performed as predicted, we constructed a number of mock
aggregates comprised of 8x8 grids where each point was assigned a 1 for archaea or 0 for
bacteria. The mock aggregates were made to span the full spectrum from full segregation
to perfect mixing. Examples of mock aggregates and their corresponding mixing metrics
are shown in
Extended Figure 4a
. The metrics behaved as expected.
Permutation tests for significance of mixing in the ANME-SRB consortia
Methods exist for Moran’s I for calculating the statistical significance of spatial autocor-
relation, however, since this metric has an expectation of normally distributed continuous
data, we were not able to use these significance calculations to determine which aggre-
gates were significantly mixed or segregated beyond what would be expected from random
variations.
To address this question we turned to non-parametric statistics and performed 300 per-
mutation tests per AS and AD consortia. In these permutation tests the phylogenetic
identifications of the ROIs were randomly redistributed to the (x,y) indices of the ROIs,
and Moran’s I and J were calculated for all permuted aggregates. If the calculated value
of a metric for an aggregate was greater or less than the permuted values more than 95%
of the time it was considered to be significantly more or less segregated than random,
respectively.
Extended Figure 4b
and
4c
display the results of this test for the AS and
AD datasets. The small black dots show the values for the various permutation tests,
and the larger colored dots show the actual value for that aggregate. If an aggregate was
more segregated than random (p
<
0.05) it is colored green, if it was more mixed than
random (p
<
0.05) it is colored purple, and if it was not significantly different than the
random permutations it is colored red. It is worth noting that the spatial arrangements
of the majority of consortia was more segregated than random, whereas there was only a
single aggregate in our dataset which was more mixed than random.
5 Spatial Effects on Activity of Single Cells
5.1 Introduction
Our data on the location, relative activity, and identity of syntrophic partners allowed
us to address questions pertaining to the controls on single cell anabolic activities. The
spatial analyses performed are described in the following sections. To compare patterns
of activity across the entire dataset, the z-score of the activities within populations within
aggregates are often used instead of the raw activity values. This normalization allowed
us to compare the controls on ROI activities between aggregates, and between types,
without the confounding effects of overall aggregate activity differences.
WWW.NATURE.COM/NATURE | 13
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
5.2 Relationship Between Cell Activity and Distance to Surface
One simple question that we were able to address with our dataset was: how do archaea or
bacterial activities vary as a function of distance to the exterior surface of the aggregate?
To answer this question each cell that was on the surface of the aggregate was marked by
hand, these were assigned 0 microns as their distance to the surface. The rest of the cells
were assigned distances to the surface by finding the shortest distance from their centroid
to a centroid of a surficial cell.
Extended Figure 8
shows archaea and bacteria ROI
z-score activities as a function of distance to the surface of the aggregate for both AS and
AD consortia. It is apparent from our data is that there is no significant concentration of
above average or below average activity cells of either type near or far from the surface
of the aggregate. Note: as with all analyses conducted in this study, we only have a two
dimensional slice through a three dimensional body, so all measurements of distance are
best approximations.
5.3 Relationship Between Cell Activity and Distance to Partner
In previous modeling studies it was found that the proximity to syntrophic partner was
a strong determinant of cellular activity within AOM consortia [13]. We were quite
interested to see to what extent this spatial effect was present in our empirical dataset.
We plotted ROI activity z-scores vs. the distance to nearest partner for archaeal and
bacterial ROIs for AS
Figure 2
and AD
Extended Figure 6
. We also plotted the
activity z-scores vs. the average distance to three nearest partners, with similar results
(data not shown). Unlike the models previously developed based on diffusible substrates,
there appears to be no significant trends in activity with distance to syntrophic partner
for either cell type, in either the AS or AD datasets.
5.4 Interfaces in cellular neighborhoods
It is often informative to examine patterns in spatially indexed data by constructing neigh-
borhoods and asking questions about how the values of observations are dependent on the
characteristics of their neighborhoods. To this end we applied numerous neighborhood
construction algorithms including Delaunay Triangulation; Spheres of Influence; Gabriel
Neighbors; Relative Neighbors; and 1, 2, 3 and 4 nearest neighbors. All neighborhoods
were constructed by importing x,y coordinates of the ROI centroids into the R statistical
package [54]. Neighborhoods were made with the spdep package [52][53]. The choice of
neighborhood method is largely arbitrary, and seemed to have little effect on the outcomes
of the analyses presented in this study. For the presented data we used the Spheres of
Influence neighborhoods, as they seems the most reasonable and free from artifacts by
visual inspection (see
Extended Figure 9a
).
WWW.NATURE.COM/NATURE | 14
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
As described in the main text, we would predict that there was an enhancement in
15
N
enrichment for those cellular ROIs which had syntrophic partners in their neighborhoods.
To test this effect across the entire dataset, we split each partner into two groups, those at
syntrophic interfaces (with a syntrophic partner in their neighborhood) and those with-
out (see
Extended Figure 7c and d
for depiction of a neighborhood and the resulting
interfacial cells). For each archaea or bacteria in each aggregate we then conducted a
2-sample t-test to determine whether there was a significant difference between the mean
activities of the interfacial or non-interfacial cells. The results of these test are displayed
in supplementary tables 2-5. Significance was determined at a P
<
0.05 level, with a Bon-
ferroni correction applied for the multiple comparisons within aggregate types (41 AS and
21 AD). Consortia with significant differences between the interfacial and non-interfacial
cells are bolded and underlined. Two observations from this analysis are clear: 1) in very
few consortia is there a significant difference in the activity of cells with and without part-
ners in their immediate neighborhood, and 2) the ratio of activities between the interface
and non-interfacial cells is nearly 1 in all cases, even those with significant differences in
partner activities. Both of these observations contradict the classic assumptions of steep
gradients of cellular activity caused by diffusible intermediates.
WWW.NATURE.COM/NATURE | 15
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
Aggregate
Total
ROIs
Total
Interface
ROIs
Total Non-­‐
interface
ROIs
Interface
mean
activity
Non-­‐Interface
mean activity
Interface:Non-­‐
interface
activitiy ratio
P-­‐value
Pass with
Bonferroni
Correction
1
218
179
39
0.049
0.046
1.063
0.0443
0
2
11
9
2
0.098
0.090
1.082
0.1353
0
3
17
5
12
0.058
0.057
1.021
0.8509
0
4
91
56
35
0.082
0.081
1.020
0.3328
0
5
93
82
11
0.071
0.069
1.026
0.7219
0
6
68
58
10
0.064
0.054
1.183
0.2266
0
7
61
45
16
0.043
0.038
1.145
0.1402
0
8
6
6
0
0.008
NaN
NaN
NaN
NaN
9
50
28
22
0.073
0.070
1.042
0.0720
0
10
42
29
13
0.072
0.061
1.188
0.0658
0
11
30
26
4
0.078
0.079
0.986
0.7812
0
12
11
11
0
0.051
NaN
NaN
NaN
NaN
13
14
14
0
0.091
NaN
NaN
NaN
NaN
14
12
11
1
0.088
0.078
1.131
NaN
NaN
15
21
21
0
0.087
NaN
NaN
NaN
NaN
16
72
72
0
0.057
NaN
NaN
NaN
NaN
17
37
34
3
0.107
0.101
1.056
0.3044
0
18
13
4
9
0.069
0.057
1.206
0.4364
0
19
14
12
2
0.082
0.097
0.845
0.0725
0
20
18
17
1
0.099
0.105
0.942
NaN
NaN
21
10
5
5
0.085
0.081
1.047
0.6286
0
22
38
35
3
0.081
0.078
1.040
0.3087
0
23
70
63
7
0.113
0.108
1.045
0.2809
0
24
5
4
1
0.084
0.089
0.941
NaN
NaN
25
34
32
2
0.048
0.033
1.441
0.5533
0
26
69
67
2
0.020
0.015
1.381
0.6040
0
27
154
153
1
0.079
0.082
0.966
NaN
NaN
28
13
11
2
0.040
0.036
1.099
0.6055
0
29
31
21
10
0.090
0.075
1.200
0.2236
0
30
120
114
6
0.037
0.036
1.015
0.8316
0
31
18
18
0
0.106
NaN
NaN
NaN
NaN
32
43
36
7
0.095
0.095
1.003
0.9038
0
33
37
36
1
0.068
0.052
1.302
NaN
NaN
34
90
89
1
0.114
0.120
0.950
NaN
NaN
35
44
27
17
0.079
0.079
1.000
0.9868
0
36
58
43
15
0.112
0.103
1.095
0.0006
1
37
38
35
3
0.053
0.055
0.968
0.5090
0
38
30
24
6
0.045
0.052
0.868
0.4419
0
39
103
90
13
0.058
0.055
1.054
0.5913
0
40
44
39
5
0.084
0.086
0.983
0.3668
0
41
19
14
5
0.101
0.098
1.032
0.3972
0
Table S1
:
AS Consortia: Archaea Interface vs. Non-Interface Activity Com-
parison
WWW.NATURE.COM/NATURE | 16
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
Aggregate
Total
ROIs
Total
Interface
ROIs
Total Non-­‐
interface
ROIs
Interface
mean
activity
Non-­‐Interface
mean activity
Interface:Non-­‐
interface
activitiy ratio
P-­‐value
Pass with
Bonferroni
Correction
1
203
168
35
0.064
0.053
1.209
0.0001
1
2
36
21
15
0.113
0.122
0.926
0.0297
0
3
11
4
7
0.148
0.137
1.079
0.2532
0
4
71
54
17
0.115
0.114
1.009
0.8041
0
5
88
77
11
0.070
0.055
1.278
0.0053
0
6
60
52
8
0.071
0.069
1.028
0.7875
0
7
76
46
30
0.085
0.086
0.993
0.8771
0
8
21
10
11
0.013
0.012
1.047
0.7600
0
9
24
22
2
0.093
0.084
1.102
0.0002
1
10
55
36
19
0.102
0.106
0.961
0.3770
0
11
29
27
2
0.068
0.054
1.255
0.7180
0
12
13
11
2
0.058
0.044
1.297
0.1469
0
13
21
14
7
0.107
0.114
0.943
0.0233
0
14
22
14
8
0.097
0.075
1.293
0.1447
0
15
45
25
20
0.091
0.087
1.040
0.2170
0
16
106
100
6
0.044
0.041
1.080
0.3318
0
17
25
24
1
0.115
0.062
1.860
NaN
NaN
18
11
5
6
0.134
0.135
0.990
0.8599
0
19
12
12
0
0.081
NaN
NaN
NaN
NaN
20
15
15
0
0.093
NaN
NaN
NaN
NaN
21
11
7
4
0.147
0.168
0.871
0.1118
0
22
34
32
2
0.097
0.084
1.152
0.0283
0
23
65
60
5
0.116
0.093
1.245
0.1118
0
24
10
6
4
0.064
0.063
1.017
0.9069
0
25
39
30
9
0.064
0.063
1.002
0.9841
0
26
78
75
3
0.033
0.027
1.247
0.4740
0
27
212
194
18
0.082
0.052
1.578
0.0025
0
28
33
14
19
0.060
0.064
0.933
0.6379
0
29
11
11
0
0.130
NaN
NaN
NaN
NaN
30
101
98
3
0.033
0.036
0.907
0.4406
0
31
11
11
0
0.131
NaN
NaN
NaN
NaN
32
42
36
6
0.086
0.070
1.225
0.0112
0
33
54
40
14
0.060
0.053
1.136
0.0008
1
34
100
97
3
0.107
0.112
0.957
0.0895
0
35
25
20
5
0.086
0.091
0.950
0.1769
0
36
61
50
11
0.122
0.105
1.167
0.1025
0
37
50
39
11
0.084
0.084
1.000
0.9897
0
38
49
35
14
0.076
0.063
1.204
0.0987
0
39
99
86
13
0.065
0.059
1.098
0.4257
0
40
28
28
0
0.080
NaN
NaN
NaN
NaN
41
6
6
0
0.130
NaN
NaN
NaN
NaN
Table S2
:
AS Consortia: Bacteria (Seep SRB 1a) Interface vs. Non-Interface
Activity Comparison
WWW.NATURE.COM/NATURE | 17
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
Aggregate
Total
ROIs
Total
Interface
ROIs
Total Non-­‐
interface
ROIs
Interface
mean
activity
Non-­‐Interface
mean activity
Interface:Non-­‐
interface
activitiy ratio
P-­‐value
Pass with
Bonferroni
Correction
1
66
25
41
0.041
0.042
0.974
0.7713
0
2
41
34
7
0.052
0.037
1.411
0.1061
0
3
45
15
30
0.060
0.078
0.776
0.0144
0
4
24
20
4
0.050
0.033
1.540
0.0467
0
5
56
50
6
0.070
0.062
1.130
0.3636
0
6
27
19
8
0.067
0.069
0.970
0.5758
0
7
93
74
19
0.051
0.029
1.749
0.0002
1
8
4
2
2
0.092
0.085
1.080
0.646
0
9
28
28
0
0.026
NaN
NaN
NaN
NaN
10
4
4
0
0.117
NaN
NaN
NaN
NaN
11
7
7
0
0.073
NaN
NaN
NaN
NaN
12
21
16
5
0.007
0.005
1.326
0.0718
0
13
35
31
4
0.040
0.052
0.756
0.3512
0
14
21
19
2
0.027
0.037
0.742
0.1254
0
15
19
19
0
0.031
NaN
NaN
NaN
NaN
16
27
8
19
0.027
0.043
0.629
0.0726
0
17
10
7
3
0.120
0.122
0.984
0.6861
0
18
12
7
5
0.004
0.004
1.034
0.7902
0
19
15
14
1
0.073
0.056
1.301
NaN
NaN
20
94
28
66
0.055
0.055
1.001
0.9877
0
21
9
4
5
0.121
0.101
1.203
0.7292
0
Table S3
:
AD Consortia: Archaea Interface vs. Non-Interface Activity Com-
parison
WWW.NATURE.COM/NATURE | 18
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512
Aggregate
Total
ROIs
Total
Interface
ROIs
Total Non-­‐
interface
ROIs
Interface
mean
activity
Non-­‐Interface
mean activity
Interface:Non-­‐
interface
activitiy ratio
P-­‐value
Pass with
Bonferroni
Correction
1
65
23
42
0.048
0.049
0.987
0.8216
0
2
40
39
1
0.056
0.084
0.672
NaN
NaN
3
28
12
16
0.047
0.048
0.965
0.4372
0
4
11
11
0
0.062
NaN
NaN
NaN
NaN
5
47
44
3
0.052
0.034
1.510
0.2217
0
6
54
17
37
0.061
0.062
0.989
0.6446
0
7
70
59
11
0.086
0.077
1.116
0.1846
0
8
7
4
3
0.062
0.073
0.841
0.532
0
9
53
45
8
0.036
0.034
1.054
0.5196
0
10
20
8
12
0.097
0.101
0.964
0.7594
0
11
16
6
10
0.075
0.074
1.009
0.7576
0
12
15
15
0
0.006
NaN
NaN
NaN
NaN
13
41
30
11
0.063
0.065
0.967
0.771
0
14
36
28
8
0.011
0.015
0.760
0.3623
0
15
26
26
0
0.039
NaN
NaN
NaN
NaN
16
35
8
27
0.051
0.053
0.971
0.2181
0
17
5
4
1
0.119
0.118
1.007
NaN
NaN
18
48
9
39
0.004
0.004
0.956
0.2593
0
19
58
28
30
0.073
0.069
1.069
0.0237
0
20
58
21
37
0.054
0.044
1.243
0.0246
0
21
32
6
26
0.056
0.060
0.932
0.5416
0
Table S4
:
AD Consortia: Bacteria (deltaproteobacteria) Interface vs. Non-
Interface Activity Comparison
WWW.NATURE.COM/NATURE | 19
SUPPLEMENTARY INFORMATION
RESEARCH
doi:10.10
3
8/na ture
1
5512