1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
Quantifying Community Assembly Processes and Identifying Features that Impose Them
James C. Stegen*, Xueju Lin, Jim K. Fredrickson, Xingyuan Chen, David W. Kennedy,
Christopher J. Murray, Mark L. Rockhold, and Allan E. Konopka
Fundamental and Computational Sciences Directorate, Biological Sciences Division, Pacific
Northwest National Laboratory, Richland, WA, 99352, USA
*Corresponding author, phone: 509-371-6763, Email: James.Stegen@pnnl.gov
Supplementary Methods
Field sampling, DNA processing, and environmental data
We study a bacterial meta-community associated with subsurface sediments within an
unconfined aquifer ~250m (horizontal distance) from the Columbia River in Richland, WA. The
sampled locations are within the Hanford Integrated Field Research Challenge (IFRC) site
(http://ifchanford.pnnl.gov/) (Bjornstad et al 2009). Sediment samples were taken during the
drilling of 26 wells as described in (Bjornstad et al 2009) (see Table S1 for summarized
metadata). DNA was extracted as in (Lin et al 2012a) and sequenced as in (Lin et al 2012b).
Using QIIME (Caporaso et al 2010) sequences were pre-processed by removing any sequences <
200 or > 300 nucleotides long, with a mean quality score < 25 (Huse et al 2007), containing
ambiguous characters, containing a homopolymer longer than 8 nucleotides, missing the primer
sequence, or containing an uncorrectable barcode. Samples were not used if they had < 500
sequences or >5% relative abundance of Propionibacterial (a sign of contamination). Operational
taxonomic units (OTUs) were created using cd-hit (Li & Godzik 2006) with a prefix pre-filter
length of 200 nucleotides, a minimum coverage of 99%, and a minimum similarity of 97%. The
most abundant sequence within each OTU was taken to represent the OTU. For simplicity, below
we refer to OTUs as ‘species.’
Prior to further analyses, each sample was rarefied to 500 sequences (to control betweensample heterogeneity) and only the 1000 most abundant species were retained to reduce the
influences of sequencing errors that may produce singletons. This step is not likely to greatly
influence our results because all community-level analyses account for relative abundances;
patterns are driven primarily by the most abundant species. Representative sequences were
aligned against the SILVA database (Pruesse et al 2007) using PyNAST within QIIME
(Caporaso et al 2010) with a minimum alignment length of 150 and a minimum identity of 75%.
Sequences that failed to align were dropped. FastTree (Price et al 2009) was used to infer a
phylogenetic tree within QIIME (Caporaso et al 2010). All community-level analyses (described
below) were carried out in the R statistical language (http://cran.r-project.org/).
Measured environmental data for each community included its elevation, its horizontal
distance from the Columbia River shoreline, the elevation of the top of the Ringold formation at
its geographic location (see Bjornstad et al 2009), and the composition of its associated
sediments measured as percent mud. These variables provide indirect characterization of
environmental conditions that may influence microbial communities; elevation is related to
vertical gradients in redox and potentially the availability of electron donors and acceptors;
horizontal distance from the Columbia River is related to the magnitude of river intrusion; the
Ringold elevation likely influences how the geochemical environment changes with depth such
1
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
that two communities at the same depth but in locations with different Ringold elevations may
experience different geochemical environments; and percent mud influences numerous physical
and geochemical aspects of the local environment.
To estimate percent mud, grain-size analyses were performed on selected sediment
samples collected during drilling at the IFRC site using a combination of sieve, hydrometer, and
laser diffraction methods (Gee & Or 2002). The mud content (silt- and clay-sized particles) in
each of the samples was estimated from the mass fraction less than 0.063 mm (Folk 1980).
Geophysical logging was also performed in the IFRC wells using a spectral gamma logging
system. The primary natural gamma-emitting radionuclides are 40K, 232Th, and 238U. Percent mud
was found to be positively correlated with both 40K and 232Th, with slightly stronger correlations
for 232Th. Therefore the 232Th log data were interpolated to the sediment sample measurement
locations, and a correlation function was used to estimate percent mud from the interpolated
232
Th values. One sample did not have a 232Th value (well C6207 within the Ringold formation)
from which to calculate percent mud. This sample was approximately a meter within finegrained Ringold material (see Appendix A in Bjornstad et al 2009); to estimate its percent mud
we used the median percent mud across samples taken from deeper than a meter below the top of
the Ringold.
Testing phylogenetic signal
To test for phylogenetic signal we first estimated two dimensions of each OTU’s
ecological niche. Each dimension is estimated as the habitat conditions under which a given
OTU is most abundant (Andersson et al 2010, Pei et al 2011, Stegen et al 2012). Across our
system the most substantial shifts in environmental conditions are likely associated with
subsurface depth and sediment composition; at greater depths the sediment shifts from coarsegrained Hanford material to fine-grained Ringold material and redox conditions shift from
oxidizing to reducing (Bjornstad et al 2009). Our sampled communities are primarily from
oxidizing conditions (Table S1), so we cannot rigorously evaluate redox-state niches. Instead, the
environmental niche of each OTU was characterized as the subsurface elevation where it was
most abundant and sediment composition (the % mud) where it was most abundant (similar to
Pei et al 2011, Stegen et al 2012).
We relate between-OTU niche differences to between-OTU phylogenetic distances using
Mantel correlograms with permutation-based significance tests for each of 50 phylogenetic
distance classes (similar to Diniz-Filho et al 2010). Significance tests were based on 999
permutations using the R function ‘mantel.correlog’ (package ‘vegan’) with a progressive
Bonferroni correction (Legendre & Legendre 1998) and no distance class cutoff.
Turnover in phylogenetic community composition
A given value of βMNTD could be less than, greater than, or equal to the degree of
turnover expected when Selection does not influence turnover in community composition. Less
than expected phylogenetic turnover should result from environmental conditions constraining
community composition by imposing Selection on species’ ecological niches. Greater than
expected phylogenetic turnover should be due to divergent environmental conditions causing
each community to be composed of an ecologically distinct set of species. Note that these
2
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
conditions assume at least a minor amount of exchange of organisms among local communities
through deep evolutionary time (so that individual communities do not evolve evolutionarily
distinct assemblages in situ). This assumption is likely upheld in our system, which is within a
single unconfined aquifer (maximum of 54m separating any two communities) through which
groundwater regularly flows and into which the Columbia River annually intrudes (McKinley et
al in revision). As such, the degree to which βMNTD deviates from a null model expectation
measures the degree to which community composition is limited by Selection on species’
ecological niches.
To quantify the degree to which βMNTD deviates from a null model expectation we used
a randomization procedure that shuffled species names and abundances across the tips of the
phylogeny. After shuffling, βMNTD was recalculated to provide a null value, and repeating the
randomization 999 times provided a null distribution. The difference between observed βMNTD
and the mean of the null distribution is measured in units of standard deviations (of the null
distribution) and is referred to as the β-Nearest Taxon Index (βNTI). βNTI values less than -2 or
greater than +2 indicate that observed βMNTD deviates by more than two standard deviations
from the null model expectation. For a given pairwise community comparison, βNTI < -2 or >
+2 therefore indicates significantly less than or greater than expected phylogenetic turnover,
respectively.
Our randomization procedure was chosen by considering the results of Hardy (Hardy
2008), which show that randomization outcomes are influenced by phylogenetic signal in species
abundances. We find little to no evidence for phylogenetic signal in species abundances, using
Mantel correlograms as above (Fig. S1). In this case Hardy (Hardy 2008) showed that our
randomization procedure provides very close to an exact test for a range of phylogenetic turnover
metrics. We therefore suggest that our specific null model provides robust statistical and
ecological inferences.
Turnover in species composition
Most metrics of turnover in species composition provide no information on whether the observed
degree of turnover is less than, greater than, or similar to the degree of turnover expected if
community assembly was governed primarily by Drift. One exception is Raup-Crick (Chase et al
2011), which generates an expected degree of turnover using a randomization procedure where
species are probabilistically drawn into each local community until empirically-observed local
richness is reached. In the randomization procedure the probability of drawing a given species
from the meta-community species pool is proportional to the number of local sites occupied by
that species (Chase et al 2011).
In its current form Raup-Crick does not account for species’ relative abundances (Chase
et al 2011), which can carry information useful for understanding community assembly
processes (Anderson et al 2011). In order to take full advantage of this information we extend
Raup-Crick to consider species’ relative abundances. Practically, this requires a minor addition
to the procedure developed by Chase et al. (Chase et al 2011). In the randomization procedure
we use their method (summarized above) to draw species into a given local community until the
empirical species richness is reached. At that point each species in the randomly assembled
community is represented by one individual. To model stochastic (Drift-based) recruitment,
individuals are drawn into the community, but only into those species assembled in the previous
step. The probability of drawing an individual into a given species is proportional to that species
3
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
abundance in the meta-community. The randomization procedure therefore assumes no influence
of Selection or Dispersal Limitation.
For a pair of communities being compared, both were randomly assembled following the
above procedure. Compositional turnover between the communities was then quantified using
standard Bray-Curtis dissimilarity, which accounts for species relative abundances. For each pair
of communities the randomization was run 999 times, providing a null distribution of expected
Bray-Curtis values. The empirically observed Bray-Curtis value for the pair of communities was
compared to this null distribution following the procedure developed by Chase et al. (Chase et al
2011). Specifically, the number of comparisons between randomly assembled communities that
have a Bray-Curtis value greater than the empirical Bray-Curtis is added to half the number of
ties; ties occur when observed Bray-Curtis is equal to Bray-Curtis based on randomly assembled
communities. The resulting sum gives the probability that stochastic community assembly (Drift)
results in less turnover than empirically observed. As in Chase et al. (Chase et al 2011), we
standardize this probability to vary between -1 and +1 by subtracting 0.5 and then multiplying by
2, and refer to the resulting metric as RCbray.
Similar to Chase et al. (Chase et al 2011), we interpret RCbrayvalues greater than +0.95 or
less than -0.95 as indicating that Selection or Dispersal significantly influence turnover in
community composition. In turn, RCbrayvalues between -0.95 and +0.95 are consistent with a
dominant role of Drift.
160
Combining spatial eigenvectors and measured environmental variables with model-selection
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
To provide the most complete description of spatial and environmental relationships among local
communities we combined spatial eigenvector analysis with measured abiotic variables. The
spatial eigenvectors describe spatial relationships among communities across a range of spatial
scales; the first eigenvector breaks sampling locations into broadly distributed clusters and
subsequent eigenvectors characterize spatial relationships at increasingly fine scales (Borcard &
Legendre 2002, Borcard et al 2011, Heino et al 2011).
To carry out spatial eigenvector analyses we used the R function ‘pcnm’ within the
package ‘vegan’. The ‘pcnm’ function takes a (spatial) distance matrix as input. For analyses
within the Ringold and Hanford formations we used the geographical locations (Eastings and
Northings, Table S1) of each well to build the distance matrix. Within each formation spatial
eigenvectors therefore describe spatial relationships in two-dimensions. The resulting
eigenvectors are referred to as ‘PCNM’ axes in Tables S2-S4. Note that this method was
originally referred to ‘principal coordinates of neighbor matrices’ (Borcard & Legendre 2002),
but is now referred to as ‘Moran’s eigenvector maps’ (Borcard et al 2011).
Eigenvectors were described using two-dimensional spatial relationships within
formations because the horizontal distance separating communities was much larger than the
vertical distance separating communities. Communities retained for analyses within each
formation are therefore distributed across an approximately two-dimensional space. For analyses
across the full system (which includes both formations) we described spatial distances in threedimensions due to the larger vertical distances separating communities. These three-dimensional
Euclidean distances were then used to define spatial eigenvectors. Note that spatial eigenvector
analysis is robust in one, two or three dimensions (Borcard & Legendre 2002).
4
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
Spatial eigenvectors only describe spatial relationships among sampling locations. As
such, some eigenvectors may describe the spatial scale(s) at which dispersal operates, while
others may be more related to the spatial structure of environmental variables (Legendre et al
2009). In addition to spatial relationships, we measured four abiotic variables that may influence
community composition. However, these measured variables may also simply describe spatial
relationships among communities. For example, horizontal distance from the Columbia River
may reflect spatial relationships or may reflect different environmental conditions related to
spatially structured river water intrusion (McKinley et al in revision). In addition, measured
abiotic variables may co-vary with each other and/or with spatial eigenvectors. To combine all
variables and minimize co-variation we combined measured abiotic variables with spatial
eigenvectors using principal components analysis (PCA). The resulting PCA axes (Tables S2-S4
provide loadings) were used as independent variables in a model-selection procedure with either
βNTI or RCbray as the dependent variable. Note that three separate sets of PCA axes were
characterized: one for the Hanford formation, one for the Ringold formation, and one for the full
system. As such, labels associated with Hanford formation PCA axes have no relationship to, for
example, labels of Ringold formation axes.
To fit statistical models for βNTI and RCbray we used distance-based redundancy analysis
(Legendre & Anderson 1999) (R function ‘capscale’ within package ‘vegan’) combined with a
model-selection procedure. We used forward model-selection (Blanchet et al 2008) where the
significance of independent variables (α = 0.05) was evaluated step-wise and the order of
variable evaluation was based on improvement in the model’s adjusted R2. Model-selection
proceeded until the next independent variable was non-significant as determined by 1000
permutations (R function ‘ordiR2step’ within package ‘vegan’). Separate model-selection
procedures were carried out for the Hanford, the Ringold and the full system, and βNTI and
RCbray were further evaluated separately. Distance-based redundancy analysis takes positive,
pairwise community distances as input such that βNTI and RCbray were each normalized to vary
between 0 and 1 prior to model-selection; for each, the absolute magnitude of the minimum
(negative) value was added to all values (making all > 0) and the resulting values were then
divided by their maximum (making all > 0 and < 1).
As discussed above, the magnitude of βNTI is governed by the influence of Selection
relative to the influences of Dispersal and Drift. Any PCA axes that explain a significant fraction
of variation in βNTI must therefore reflect one or more environmental variables that impose
Selection. This is true even if a significant PCA axis is unrelated to measured abiotic variables;
the degree to which PCA axes are related to measured abiotic variables was evaluated by
examining PCA axis loadings (Tables S2-S4). If a given PCA axis is significant for βNTI but
measured abiotic variables do not load onto it, we consider this PCA axis to be an unmeasured,
yet influential and spatially structured environmental variable. If measured abiotic variables load
heavily onto a significant PCA axis, we consider the axis to be a measured, influential
environmental variable. Furthermore, all PCA axes non-significant for βNTI were considered to
primarily characterize spatial relationships among communities. This is true even if measured
abiotic variables load heavily; measuring a given abiotic variable does not necessarily indicate
that variable imposes Selection.
Prior to RCbray model-selection we used the βNTI model-selection results to characterize
each PCA axis as an unmeasured environmental, a measured environmental, or a spatial variable.
Following RCbray model-selection, these variable designations were used (in conjunction with
PCA loadings) to interpret the factors imposing Selection or Dispersal Limitation.
5
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
Comparison of inferences to those gleaned from pre-existing approaches
We compared the insights derived from our novel analytical framework to those derived from a
pre-existing approach (similar to e.g., Heino et al 2011, Legendre et al 2009). To achieve a direct
comparison to our approach we used the same PCA axes with the same model-selection
procedure, but with Bray-Curtis dissimilarity as the dependent variable. However, in the ‘preexisting approach’ one must rely on PCA loadings to identify PCA axes as environmental or
spatial; PCA axes with heavy loadings from measured abiotic variables are considered
environmental and all others are considered spatial. The key assumption is that any influence of
measured abiotic variables is through Selection (i.e., measured variables do not impose Dispersal
Limitation). As noted above, this is not necessarily the case. In fact, which abiotic variables
impose Selection and which impose Dispersal Limitation is an empirical question that requires
an answer informed by the ecology of the system rather than PCA axis loadings (which by
themselves carry no ecological information).
A large number of studies have combined model-selection with variation partitioning
(Legendre & Legendre 1998) to infer influences of community assembly processes (e.g.,
Legendre et al 2009). Variation partitioning has, however, recently been shown to provide
invalid inferences (Gilbert & Bennett 2010, Smith & Lundholm 2010), especially with respect to
how much stronger one process is than another (Stegen & Hurlbert 2011). As such, we do not
use variation partitioning. Instead, if environmental or spatial variables explain a non-zero
fraction of variation in Bray-Curtis, we consider this evidence for some (non-zero) influence of
Selection or Dispersal Limitation, respectively. How influential each process is, however, cannot
be estimated, nor can the influence of Drift (Anderson et al 2011, Legendre et al 2009). Modelselection results for Bray-Curtis are provided in Table 1 and a comparison of inferences drawn
using our approach versus the standard approach is provided in Figure 4.
6
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
Supplemental Figures
Figure S1
Figure S1. Phylogenetic Mantel correlogram showing generally non-significant phylogenetic
signal for species abundances. Solid and open symbols denote significant and non-significant
correlations, respectively, relating between-species abundance differences to between-species
phylogenetic distances at a given phylogenetic distance lag. Abundance for each species was
found across the entire meta-community (both formations, all communities) after rarefaction (to
500 sequences per community). Significant negative correlations indicate that closely related
species have different abundances, while more distantly related species have more similar
abundances. Importantly, there are no significant positive correlations; abundance is not
phylogenetically ‘conserved’ (sensu Losos 2008). Lack of phylogenetic conservatism in
abundance allows for robust statistical performance of our specific randomization used in
conjunction with phylogenetic turnover (Hardy 2008). Further, there are only two phylogenetic
distances across which there is a significant negative correlation. This rather weak pattern of
‘species abundance phylogenetic overdispersion’ (sensu Hardy 2008) is also consistent with
robust statistical performance of our specific randomization procedure (Hardy 2008).
7
276
277
278
279
280
281
282
283
Supplemental Tables
Table S1. Metadata for sampled communities. Well ID can be associated with well names
provided in Bjornstad et al. (Bjornstad et al 2009). Communities at the Hanford-Ringold contact
are designated as such. Core Elevation is the elevation at which a given community was
sampled. Ringold Elevation is the elevation at the top of the Ringold formation. The water table
has a minimum elevation near 104.3m (Bjornstad et al 2009).
284
285
8
286
287
Table S1 (cont.).
9
288
289
10
290
Table S1 (cont.).
291
292
11
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
Table S2. Loadings of measured environmental variables and spatial eigenvectors (‘PCNM’) on
PCA axes within the Ringold formation. Any loadings weaker than 0.001 are listed as null. A
standard approach to identify PCA axes as either environmental or spatial is based on the
loadings of measured abiotic variables. In this case, PCA axes with heavy loadings from
measured abiotic variables are considered environmental (grey fill across the ‘Loading-Based
Designation’ row). All other PCA axes were considered spatial (black fill). The approach we
employ uses model-selection for βNTI; PCA axes retained during model-selection for βNTI
reflect environmental variables, whether or not measured abiotic variables load on them. If
measured abiotic variables do not load onto a retained PCA axis, it is considered an unmeasured
environmental variable. PCA7 is such a variable within the Ringold; PCA7 was the only axis
retained during the βNTI model-selection procedure (Table 1), yet no measured abiotic variables
load onto PCA7. This indicates that Selection in the Ringold is imposed, in part, by an
unmeasured environmental variable with an estimated spatial structure depicted in Fig. 5.
308
309
12
310
13
311
312
313
314
315
316
317
318
319
320
321
Table S3. As for Table S2, but within the Hanford formation. Within the Hanford PCA1 and PCA3 were retained during modelselection for βNTI (Table 1). Those axes are therefore considered environmental variables that impose Selection, while all other PCA
axes were considered to primarily reflect spatial relationships among communities (as opposed to environmental differences among
communities). Distance-to-the-river and elevation (of communities) are the measured variables loading most heavily on PCA1 and
PCA3, respectively. RCbray model-selection retained PCA axes 1, 3 and 7 (Table 1). No measured variables load onto PCA7. Note that
PCA axes are specific to the formation being studied; PCA7 for the Hanford is unrelated to PCA7 for the Ringold. The loading
patterns suggest that Selection results from vertically and horizontally structure environmental variables, while Dispersal Limitation
results from isolation of communities due to spatially structured, but unmeasured aspects of the system, such as complex hydrologic
flow paths.
322
323
14
324
325
326
327
328
329
330
331
332
333
334
335
336
Table S4. As for Table S2, but across the Hanford and Ringold formations (the ‘full’ system). Model-selection for βNTI across the
full system retained PCA axes 1 and 32 (Table 1), and measured abiotic variables load heavily on both. Elevation (of communities)
was the measured variable loading most heavily on both PCA1 and PCA32, although the percent-mud loading on PCA1 was nearly as
strong (0.576 vs. -0.554). PCA1 and PCA32 were therefore considered measured environmental variables. PCA19 was also selected,
but no measured abiotic variables loaded on this axis, suggesting that PCA19 reflects an unmeasured environmental variable that
imposes Selection. All other PCA axes were considered to primarily reflect spatial relationship. These patterns suggest that
unmeasured factors and factors associated with elevation, such as percent-mud, impose Selection across the full system. Modelselection for RCbray also retained PCA1 and PCA32, but retained 11 other axes as well (Table 1). These additional 11 axes were
identified as spatial variables by the βNTI model-selection, and are therefore considered to primarily reflect factors that impose
Dispersal Limitation. Given the large number of selected variables and their spatial complexity, we hypothesize that the degree of
organismal exchange among local communities is governed by spatially complex hydrologic flow paths.
15
337
338
Table S4 (cont.).
16
339
340
Table S4 (cont.).
17
341
18
342
Supplemental References
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
Anderson MJ, Crist TO, Chase JM, Vellend M, Inouye BD, Freestone AL et al (2011). Navigating the
multiple meanings of β diversity: a roadmap for the practicing ecologist. Ecol Lett 14: 19-28.
Andersson AF, Riemann L, Bertilsson S (2010). Pyrosequencing reveals contrasting seasonal dynamics of
taxa within Baltic Sea bacterioplankton communities. ISME 4: 171-181.
Bjornstad BN, Horner JA, Vermeul VR, Lanigan DC, Thorne PD (2009). Borehole completion and
conceptual hydrogeologic model for the IFRC Well Field, 300 Area, Hanford Site. PNNL-18340,
Pacific Northwest National Laboratory, Richland, WA.
Blanchet FG, Legendre P, Borcard D (2008). Forward selection of explanatory variables. Ecology 89:
2623-2632.
Borcard D, Legendre P (2002). All-scale spatial analysis of ecological data by means of principal
coordinates of neighbour matrices. Ecol Model 153: 51-68.
Borcard D, Gillet F, Legendre L (2011). Numerical Ecology with R. Springer: New York, NY.
Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Bushman FD, Costello EK et al (2010). QIIME
allows analysis of high-throughput community sequencing data. Nat Methods 7: 335-336.
Chase JM, Kraft NJB, Smith KG, Vellend M, Inouye BD (2011). Using null models to disentangle
variation in community dissimilarity from variation in α-diversity. Ecosphere 2: art24.
Diniz-Filho JAF, Terribile LC, da Cruz MJR, Vieira LCG (2010). Hidden patterns of phylogenetic nonstationarity overwhelm comparative analyses of niche conservatism and divergence. Global Ecol
Biogeogr 19: 916-926.
Folk RL (1980). Petrology of Sedimentary Rocks. Hemphill Publishing Co.: Austin, Texas.
Gee GW, Or D (2002). Particle-size analysis. In: Dane JH, Topp GC (eds). Methods of Soil Analysis, Part
4. Physical Methods. Soil Science Society of America: Madison, Wisconsin. pp 255-293.
Gilbert B, Bennett JR (2010). Partitioning variation in ecological communities: do the numbers add up?
Journal of Applied Ecology 47: 1071-1082.
Hardy OJ (2008). Testing the spatial phylogenetic structure of local communities: statistical performances
of different null models and test statistics on a locally neutral community. J Ecol 96: 914-926.
Heino J, Grönroos M, Soininen J, Virtanen R, Muotka T (2011). Context dependency and
metacommunity structuring in boreal headwater streams. Oikos: no-no.
Huse SM, Huber JA, Morrison HG, Sogin ML, Mark Welch D (2007). Accuracy and quality of massively
parallel DNA pyrosequencing. Genome Biol 8.
Legendre P, Legendre L (1998). Numerical Ecology. Elsevier Science: Amsterdam, The Netherlands.
Legendre P, Anderson MJ (1999). Distance-based redundancy analysis: Testing multispecies responses in
multifactorial ecological experiments. Ecol Monogr 69: 1-24.
Legendre P, Mi XC, Ren HB, Ma KP, Yu MJ, Sun IF et al (2009). Partitioning beta diversity in a
subtropical broad-leaved forest of China. Ecology 90: 663-674.
Li WZ, Godzik A (2006). Cd-hit: a fast program for clustering and comparing large sets of protein or
nucleotide sequences. Bioinformatics 22: 1658-1659.
Lin X, Kennedy D, Peacock A, McKinley J, Resch CT, Fredrickson J et al (2012a). Distribution of
Microbial Biomass and Potential for Anaerobic Respiration in Hanford Site 300 Area Subsurface
Sediment. Appl Environ Microb 78: 759-767.
Lin X, Mckinley J, Resch CT, Lauber C, Fredrickson J, Konopka AE (2012b). Spatial and temporal
dynamics of microbial community in the Hanford unconfined aquifer. ISME 6: 1665-1676.
Losos JB (2008). Phylogenetic niche conservatism, phylogenetic signal and the relationship between
phylogenetic relatedness and ecological similarity among species. Ecol Lett 11: 995-1003.
McKinley JP, Zachara JM, Resch CT, Kaluzny RM, Miller MD, Vermeul VR et al (in revision). River
water intrusion and contaminant uranium contributions from the vadose zone to groundwater
during the annual Spring rise in Columbia River stage at the Hanford Site, Washington.
19
391
392
393
394
395
396
397
398
399
400
401
402
403
404
Pei NC, Lian JY, Erickson DL, Swenson NG, Kress WJ, Ye WH et al (2011). Exploring Tree-Habitat
Associations in a Chinese Subtropical Forest Plot Using a Molecular Phylogeny Generated from
DNA Barcode Loci. Plos One 6.
Price MN, Dehal PS, Arkin AP (2009). FastTree: Computing Large Minimum Evolution Trees with
Profiles instead of a Distance Matrix. Mol Biol Evol 26: 1641-1650.
Pruesse E, Quast C, Knittel K, Fuchs BM, Ludwig WG, Peplies J et al (2007). SILVA: a comprehensive
online resource for quality checked and aligned ribosomal RNA sequence data compatible with
ARB. Nucleic Acids Res 35: 7188-7196.
Smith TW, Lundholm JT (2010). Variation partitioning as a tool to distinguish between niche and neutral
processes. Ecography 33: 648-655.
Stegen JC, Hurlbert AH (2011). Inferring Ecological Processes from Taxonomic, Phylogenetic and
Functional Trait β-Diversity. Plos One 6: e20906.
Stegen JC, Lin X, Konopka AE, Fredrickson JK (2012). Stochastic and deterministic assembly processes
in subsurface microbial communities. ISME 6: 1653-1664.
405
406
20