Directory UMM :Data Elmu:jurnal:S:Soil Biology And Chemistry:Vol33.Issue1.Jan2001:

Soil Biology & Biochemistry 33 (2001) 25±40
www.elsevier.com/locate/soilbio

Structural and functional analysis of whole-soil microbial communities for
risk and ef®cacy testing following microbial inoculation of wheat roots in
diverse soils
J.V. Gagliardi a,*, J.S. Buyer b, J.S. Angle a, E. Russek-Cohen c
a

Department of Natural Resources Sciences, University of Maryland, College Park, MD 20742, USA
b
Soil Microbial Systems Laboratory, USDA-ARS, Beltsville, MD 20705, USA
c
Department of Animal Sciences, Univeristy of Maryland, College Park, MD 20742, USA
Received 19 April 1999; received in revised form 25 January 2000; accepted 16 May 2000

Abstract
The increasing use of genetically engineered or modi®ed microorganisms (GEMs) has led to regulations regarding the safety of their use.
Intended (target) effects and unintended (non-target) effects of GEMs must currently be evaluated prior to ®eld testing or commercial use. We
present soil and rhizosphere microbial community effects testing of two GEMs, Pseudomonas chlororaphis 3732RN-L11 and Pseudomonas
¯uorescens 2-79RN-L3, parental strains of these organisms and an uninoculated treatment using ®ve diverse soils planted to wheat. An assay

using BIOLOG w GN plates measured microbial community functional responses on wheat roots with adhering soil. Overall differences using
multivariate statistical methods were highest at inoculation, and these effects persisted while the inoculated organisms were detectible on
selective media. Differentiation based on lacZY genes engineered to the chromosome of both GEMs was signi®cant for the 3732 GEM in all
®ve soils tested, but not for the 2-79 GEM in a single soil. Lactose utilization in uninoculated microbial communities varied around a low
baseline value. Direct fatty acid extraction and analysis of soil from around wheat roots was also performed using a novel method. Fatty acid
analysis differentiated the 3732 GEM from all other treatments, but did not distinguish the 3732 parent inoculated from uninoculated
treatments. As with the BIOLOG assay, multivariate statistical differences from fatty acid analysis decreased between GEM inoculated
and uninoculated populations as viable counts of the GEM declined. Neither assay showed measurable community-level effects when
inoculated organisms declined below detection, though three of six soils with surviving GEM populations still had signi®cant effects after
105 days. Published by Elsevier Science Ltd.
Keywords: BIOLOG; FAME; Genetically engineered microorganisms; Pseudomonas; Rhizosphere

1. Introduction
Genetically engineered microorganisms (GEMs) are
coming under increasing scrutiny as their potential uses
and fates are considered. During the risk assessment and
review process prior to approval for release of GEMs,
concerns regarding toxicity and pathogenicity are
addressed, and numerous tests are employed that address
speci®c acute or chronic effects (Monsanto Corporation,

1987, 1988). The potential for exposure, potential environmental interactions, and potential non-target effects are also
considered (Levin and Strauss, 1991). Risk assessments
prior to environmental release speci®cally seek to determine
whether altered or inserted nucleic acids have the potential
* Corresponding author. Tel.: 11-301-504-9214; fax: 11-301-504-8370.
E-mail address: gagliarj@ba.ars.usda.gov (J.V. Gagliardi).
0038-0717/00/$ - see front matter Published by Elsevier Science Ltd.
PII: S 0038-071 7(00)00110-3

to confer production of toxic substances or pathogenic traits
to non-target organisms (Levin and Strauss, 1991). Community level microbial analyses, however, have not been
possible until recently, with the development of new assays
and procedures that show promise for use in complex
substrates. The possibility of simultaneously testing ef®cacy
along with non-target effects on microbial communities
would be an added bene®t.
Assessing the non-target effects of introduced microbes,
including effects of GEMs, is an ongoing area of research.
Several recent studies have assessed the non-target effects of
inoculated strains of Pseudomonas ¯uorescens. Wheat rhizosphere effects on indigenous pseudomonad populations

were more signi®cant than in fallow soil, but were still considered minor (De Leij et al., 1995). Nutrient cycling effects
were found but were deemed negligible in clover-cropped
®elds following a previous crop treatment of sugar beets

26

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

(Moenne-Loccoz et al., 1998). Indigenous pseudomonads
were reduced but there was no change in total microbial
populations, some soil enzyme activities were reduced
after inoculation but others increased, and diacetylphloroglucinol (DAPG) production caused plant stress (Naseby
and Lynch, 1998b). Soil enzyme activities were differentially affected by inoculation plus amendment with kanamycin or lactose compared to uninoculated controls, and
there were soil microbial community effects attributable to
a GMM (GEM) but not the wild-type strain, implying that
the inserted genes have signi®cant effects (Naseby and
Lynch, 1998a). Reduced total microbial activity was
reported after inoculation with a DAPG producing strain,
with increased plant uptake of nitrogen for inoculated peas
attributed to indigenous microbes dying off (Brimecombe et

al., 1998). In a similar study, water-soluble carbon
compounds increased, phosphate decreased, and there
were numerous nutrient transforming enzyme effects after
inoculation of pea rhizospheres, and some effects were
attributable to production of the antibiotic DAPG (Naseby
et al., 1999). The implication from this body of research is
that there are measurable effects from both engineered and
non-engineered strains of P. ¯uorescens in different soil and
rhizosphere types, though these effects may be transient, and
are often dif®cult to interpret in terms of an ecological
perspective.
Whole soil microbial community analyses may also be
employed to assess overall impact, with results on multiple
effects gained in a single assay. One method uses commercially available BIOLOG w GN microplates, originally
developed for identi®cation of Gram-negative bacteria
(Bochner, 1989). Instead of a pure culture, an extract
containing portions of the soil microbial community is
inoculated into the microplate, and the relative utilization
of individual carbon sources is compared using multivariate
statistical methods (Garland and Mills, 1991; Zak et al.,

1994; Winding, 1994; Garland, 1996a,b). Multivariate
methods tend to minimize type I errors while utilizing information regarding correlations among variables to improve
power (Seber, 1984). However, methods that obligately
subtract negative control well reactions or average well
color development (AWCD) values from all plate readings
(Garland and Mills, 1991), or that take multiple readings
over time and select differently incubated readings for coanalysis (Garland, 1996a, 1997) to correct for inoculum
density differences, may mask effects at the community
level that are unrelated to inoculum density. No valid way
exists to standardize inoculum prior to a BIOLOG assay
without spending considerable time and effort, though
there have been recent attempts to rectify this by statistically
adjusting for inoculum density after taking a single plate
reading, including the use of kinetic curves pre-determined
for each sample type analyzed (Lindstrom et al., 1998), the
use of AWCD as a covariate instead of as an obligate adjustment for analysis (Harch et al., 1997), and one that used the
blank well as a covariate (Buyer et al., 1999). The use of

BIOLOG for community analyses currently requires
signi®cant effort to inoculate, read, analyze, and understand

the results, and there may be considerable bias associated
with any of these steps.
Other community level assays employ the many types of
biogenic molecules that differentiate living organisms on
phylogenetic lines. Analysis of phospholipid fatty acids
was seen as a useful tool for characterization of soil microbial communities (Federle, 1986). This method does not
require growth in culture so it is a snapshot of the soil
biological community. Fatty acid analyses speci®c for phospholipids were used in several studies of soil microbial
communities (Zelles et al. 1992; Frostegard et al., 1993,
1996; Zelles et al. 1994; Wander et al. 1995). In a novel
approach, a soil sample is directly saponi®ed and the resulting methylated fatty acids are analyzed (Haack et al., 1994;
Cavigelli et al., 1995; Buyer and Drinkwater, 1997).
Extracting and esterifying fatty acids directly from soil
perhaps ensures a more unbiased expression of soil community structure than separation of constituent parts prior to
analysis. In addition, phospholipid fatty acid analysis is
more dif®cult and expensive than whole soil methods, and
the two methods apparently yield equivalent information
(Haack et al., 1994). A novel method employed for whole
soil fatty acid analysis uses transesteri®cation (R.A. Drijber,
personal communication), which was used for this study and

is described here in detail. Previous comparative studies
have found that BIOLOG soil microbial community multivariate pro®les were similar when compared to phospholipid fatty acids, but both methods showed high variation
between replicates (Baath et al., 1998), and BIOLOG
provided greater differentiation of treatments compared to
directly saponi®ed whole soil community fatty acid pro®les
(Buyer and Drinkwater, 1997).
Several studies have used BIOLOG plates to analyze
microbial communities on the root portion of plants, including equal portions of surface root material extracted by
blending in buffer (Ellis et al., 1995), whole roots grown
in nutrient solution and extracted by mechanical agitation
with glass beads in detergent (Garland, 1996a,b), roots
grown in soil using equal portions of washed roots extracted
by shaking in buffer (Siciliano and Germida, 1998), and
equal portions of roots with adhering soil extracted by shaking in buffer (Siciliano et al., 1998). Fatty acid techniques
for plant root or rhizosphere analysis include direct saponi®cation of oven-dried roots (Graham et al., 1995; Siciliano
and Germida, 1998), direct saponi®cation of root tissue with
adhering soil (Siciliano et al., 1998), and of germinating
seeds in soil using sonication for 5 min followed by vortexing and removal of seed material, followed by saponi®cation of the soil (Buyer et al., 1999). Previous work with
Pseudomonas chlororaphis 3732RN-L11 showed that this
organism invades wheat roots and is recoverable from

surface sterilized roots (Nairn and Chanway, 1999).
In this study, we apply two community level assays: (1)
BIOLOG plates inoculated with unwashed wheat roots, or

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

soil samples, both blended in water and analyzed with a
novel statistical approach; and (2) a novel whole soil
transesteri®cation of fatty acids method, using soil from
around the same wheat roots. Two microorganisms were
chosen for this study based on the availability of wildtype strains, and of GEM strains with similar chromosomal
lacZY inserts for 3732 and 2-79. They were also chosen
because of the existence of previous risk assessments and
®eld release approvals that would allow us to more easily
perform later tests in the ®eld. The major differences
between the two microorganisms tested are different species
identi®cation, and the ability of strain 2-79 to produce the
antibiotic phenazine-1-carboxylate, useful in biocontrol
applications (Thomashow and Weller, 1988). Soils were
inoculated with either a single genetically engineered pseudomonad strain, a non-engineered pseudomonad strain, or

were not inoculated. Treatments within a soil were prepared
and inoculated simultaneously to randomly chosen replicate
intact soil-core microcosms (Angle et al., 1995), so all treatments essentially started with the same indigenous soil
microbial communities and were otherwise similarly incubated and watered for the remainder of the study. Fresh
intact soil-core microcosms were obtained from several
ecologically unique Canadian sites and one US site for
this experiment. Our goal was to use microcosms validated
with ®eld releases (reported elsewhere for this study) for
pre-release testing of GEMs. Field-survival-predictions,
risk-assessment regarding indigenous soil microbial
communities, and ef®cacy testing of the developed organism could potentially be performed simultaneously in these
microcosms.
2. Materials and methods

27

the beginning of each year and are listed (Table 1). Microcosms from all sites were kept in growth chambers with 12h daylight length and 70% relative humidity. Canadian
microcosms were kept at 228C and UM microcosms were
at 288C, which represents the mean growing season daytime
temperature at the sites where the microcosms originated.

2.3. Bacterial inocula
In 1996, P. chlororaphis 3.732RN-L11 (3732 GEM)
(Monsanto Corporation, 1987, 1988; Kleupfel et al., 199l)
and an uninoculated treatment were used. In 1997, P.
chlororaphis 3732RN (3732 Parent) was also inoculated
into separate microcosms from each soil. In addition, separate UM soil microcosms were inoculated with P. ¯uorescens
2-79RN-L3 (2-79 GEM) and P. ¯uorescens 2-79RN (2-79
Parent) (Weller, 1983; Monsanto Corp., 1987, 1988). All
GEM and parent organisms utilized were spontaneous rifamycin and nalidixic acid resistant mutants, hence the RN in
the strain names. Both GEMs were engineered to contain the
lacZY genes inserted at a single site on the chromosome
(Barry, 1986, 1988). All inocula were grown to late-log
phase and a density of approximately 10 9 CFU ml 21 in
Pseudomonas F broth made with the same ingredients as
Pseudomonas F agar (Difco, Detroit, MI) but without
agar, and amended with 100 mg ml 21 each of rifamycin
SV and nalidixic acid after autoclaving and cooling. The
cells were immediately placed on ice, pelleted by centrifugation at 5000 £ g, and re-suspended in sterile distilled
water to a density of 0.800 OD640, a density of approximately 5 £ 10 8 CFU ml 21. Each microcosm received 2.5 ml
of prepared inoculum on the soil surface over sown wheat

seeds, followed by 2.5 ml of sterile reverse osmosis grade
(R0) water. Uninoculated microcosms received 5.0 ml of
RO water.

2.1. Intact soil-core microcosms
2.4. Extraction and assay of the rhizosphere
All assays used intact soil core microcosms planted with a
single wheat plant (Triticum aestivum), a Canadian Prairie
Hard Red Spring Wheat, cultivar A. C. Karma. Microcosms
were constructed of 5 cm inner diameter polyvinyl chloride
plumbing pipe 17.5 cm long, bevelled to 458 on the outside
at one end. The core was driven 15 cm into the soil and then
carefully removed intact (Angle et al., 1995).
2.2. Soils for the microcosm experiment
Five ecologically distinct soils from throughout Canada
were obtained in the spring of 1996 and again in 1997. All
sites sent freshly obtained intact soil-core microcosms by
overnight courier to Maryland for this study. Sites AG and
CL were in Ottawa, Ontario; BC was in Vancouver, British
Columbia; SK was in Watrous, Saskatchewan; NR was in
Montreal, Quebec (NR was tested in 1997 only). The UM
soil was from Calverton, MD, the same site used for a 1994
®eld release test (Angle et al., 1995). Soil chemical, physical, and textural characteristics were measured in detail at

Sampling began 4 h after inoculation, on day 0. Day 0
samples consisted of only bulk soil, since seed had not yet
germinated. Subsequent samplings at 14, 28 and 42 days
after inoculation occurred when there was suf®cient root
growth to constitute a sample for analysis. At 56 days the
wheat had matured and begun to senesce. Sampling after
84 days again required the use of bulk soil since the roots
had disintegrated. Soils were extracted from microcosms by
inverting the core and hammering on the bottom edge until
the sample was extruded. The sample was sieved (4 mm)
while carefully separating out wheat roots. The entire root
with adhering soil was used for BIOLOG analysis, while the
sieved soil from around the roots was used for fatty acid
analysis. Since the microcosms had a diameter of only 5 cm
and were 15 cm deep, after 14 days all portions of the
microcosms had contact with wheat roots. Sieved soil was
saved in Te¯on capped glass vials at 2208C then freeze
dried in uncapped vials and re-stored at 2208C prior to
fatty acid analysis, explained in a subsequent section. For

28

13.1
10.2
0.09
0.12
1.2
1.2
701.8
918.6
162.4
206.2
31.3
53.2
56.5
57.0
43
46
MD 96 L
97 L

42
40

15
14

5.8 2.6
5.9 2.4

1.22
1.08

5.33
6.74

10.8
9.2

74.3
93.1

12.9
12.3

12.2
0.13

0.26
0.30
3.3
3.6

1.6
1671.5

4067.4
5146.0
782.0
905.7

196.3

60.8
70.9

312.8
426.7

89.2

95.9
117.3

15.2
21.1

67.1

1.48
1.24

29.97
26.31

5.9 3.7

7.5 5.9
7.4 6.2
22
20

18

34
35
SK 96 L
97 L

22
60
NR 97 SL

44
45

1.15

10.29

14.8

103.9

13.5
12.1
0.15
0.16
2.1
1.9
4547.3
5882.2
482.9
587.2
60.4
52.1
96.7
96.7
59
79
CL 96 SL
97 LS

29
14

12
7

7.5 3.6
7.4 3.4

1.13
0.86

46.90
26.37

27.4
58.9

58.0
131.6

20.2
17.7
0.17
0.13
2.9
2.3
663.0
484.6
120.5
52.0
46.0
63.5
55.2
55.8
86
89
BC 96 S
97 S

9
8

5
3

6.1 5.6
5.8 4.4

0.92
0.82

5.77
5.06

12.8
16.1

119.1
77.6

18.2
15.9
0.09
0.10
1.6
1.6
653.0
620.6
155.8
127.4
177.2
139.9
139.1
163.6
14.3
13.4
57.3
57.9
5.57
5.47
0.88
0.80
6.2 3.6
6.2 2.9
3
2
16
14
81
84
AG 96 LS
97 LS

Yr Text Sand (%) Silt (%) Clay (%) pH OM (%) BD (g cm 321) CEC (meq 100 g 21) SS (mg K 21) N (mg K 21) P (mg K 21) K (mg K 21) Mg (mg K 21) Ca (mg K 21) OC (%) ON (%) C:N ratio

Table 1
Soil characteristics. All values represent the mean of three replicate microcosms, randomly chosen and set aside just prior to inoculation. Values reported are converted from pounds acre 21 to SI units. Soil samples
were analyzed by University of Maryland Soil Testing Laboratory, College Park, MD 20742. Key to terms: Yr ˆ Year; Text ˆ Soil texture; S ˆ Sand; Si ˆ Silt; C ˆ Clay; L ˆ Loam; OM ˆ Organic Matter;
BD ˆ Bulk Density; CEC ˆ Cation Exchange Capacity; SS ˆ Soluble Salts; N ˆ Nitrogenas as NO4; P ˆ Phosphorus as PO4; K ˆ Potassium as K2O; Mg ˆ Magnesium; Ca ˆ Calcium; OC ˆ Total Organic
Carbon; ON ˆ Total Organic Nitrogen, C:N ˆ Carbon to Nitrogen ratio

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

BIOLOG analysis of root samples and enumeration of
bacteria, extraction and sampling took place immediately.
Sterile RO water was added as a diluent to a ®nal volume of
100 ml for extraction of root samples. Blending with a
Waring blender for 1 min at top speed in a 1-l glass vessel
homogenized the root samples prior to analysis, followed by
1 min of settling to remove large particulates before taking a
sub-sample. Aliquots were taken from the top third of the
solution to avoid pieces of ¯oating organic matter. Samples
were then diluted in an isotonic magnesium±phosphate
buffer (pH 7.0) prior to inoculation onto growth media, or
in 0.85% NaCl for inoculation to BIOLOG plates. Total
culturable bacteria were enumerated using Rhizosphere
Isolation Media (Buyer, 1995) containing 100 mg ml 21
cycloheximide (Fluka Chemicals, St. Louis, MO) and
50 mg ml 21 nystatin (Sigma Chemicals, St. Louis, MO)
after 48-h incubation at 288C. Selective media used to
enumerate all the inoculated organisms was Pseudomonas
F agar (Difco Labs, Detroit, MI) with 10 ml l 21 glycerol
(Fisher Scienti®c, Pittsburgh, PA), supplemented after autoclaving with 100 mg ml 21 each of rifamycin SV, nalidixic
acid, and cycloheximide (Fluka Chemicals, St. Louis, MO),
and 60 mg ml 21 X-GAL (5-Bromo-4-chloro-3-indoyl B-dgalactopyranoside) (Diagnostic Chemicals, Ltd, Oxford,
CT).
2.5. Incubation and reading of BIOLOG plates
Blended root samples were serially diluted 1 to 1000 in
sterile 0.85% NaCl for inoculation to BIOLOG plates. Each
well of a BIOLOG w GN microplate was inoculated with
150 ml using a multi-channel pipettor. BIOLOG plates
were covered and incubated at 228C in the dark for 72 h.
All plates were read once in a microplate reader equipped
with a 590 nm barrier ®lter (Molecular Devices, E-MAX).
The 96 optical density readings from each microplate were
stored as an ASCII ®le then later combined with readings
from other replicates for statistical analysis.
2.6. Whole soil transesteri®cation and extraction of fatty
acids
Transesteri®cation is one of many methods that catalyze
the esteri®cation and methylation of fatty acids from lipids
(Christie, 1995). Transesteri®cation as reported here was
modi®ed for whole soil analysis (R.A. Drijber, personal
communcation) and differs from previous methods that
were based on MIDI microbial fatty acid identi®cation techniques (Microbial ID Inc., 1992) applied to whole soils
(Cavigelli et al., 1995; Buyer and Drinkwater, 1997). Essentially, this method eliminates the need for boiling, which is
necessary when using the MIDI chemistry. All glassware
was soaked in 1 N HCl, rinsed with sterile distilled water,
then inverted in metal racks and autoclaved for 20 min to
prevent sample contamination. All equipment was made
of stainless steel, glass or was Te¯on lined to prevent
hydrocarbon contamination. Each sample consisted of 2 g

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

freeze-dried soil in a 50 ml Pyrex centrifuge tube. Ten ml of
0.2 N KOH in HPLC grade methanol was added and the
tube was vortexed and placed in a 378C water bath for 1 h
with occasional agitation by hand. Following incubation,
the mixture was neutralized with 1 N acetic acid (typically
2 ml). Extraction of fatty acids was accomplished using 5 ml
HPLC grade hexane added to the mixture, which was then
vortexed and centrifuged at 2000 rpm in a ®xed angle rotor
to separate the phases. The hexane layer was then transferred to a 15 ml Pyrex tube using a sterile Pasteur pipette
and the entire hexane extraction repeated, for a total of
10 ml hexane with solubilized fatty acids in the tube.
Hexane was evaporated from each tube under a slow stream
of nitrogen gas in a fume hood. The resulting residue was redissolved in 0.5 ml of a 1:1 mixture of methyl tert-butyl
ether and hexane. Samples were analyzed using a gas chromatograph (Hewlett Packard 5890A, FID detector) using
MIDI EUKARY standards and the MIDI EUKARY (eukaryote) fatty acid database. Individual peaks were identi®ed
based on the relative retention time of known fatty acids in
the standard mixture, which was re-veri®ed every 10
samples. If any calibration failed, the preceding samples
were not used for analysis, and the gas chromatograph automatically shut down until it was re-calibrated. Peak areas
were quanti®ed by integration and the percent of total peak
area for each named peak was calculated. Individual sample
®les were stored, then combined later as a single ASCII ®le
for analysis.

29

used. Fatty acids were analyzed as percent of total peak
area. The ten most signi®cantly different fatty acids for
each soil comparison were used for analysis. To improve
multivariate normality, fatty acid peak values were transformed (% 1 1) 21. Multivariate statistical analysis for both
assays was a canonical discriminant analysis (CDA) using
SAS (Cary, NC) software (see Appendix A for SAS
programming). Multivariate treatment differences are
reported as an overall Wilk's F ratio, a Mahalanobis
distance for paired comparisons, or a MANOVA F ratio
performed on CDA variables for selected time points
(Glimm et al., 1997).
2.8. Survival of inoculated organisms
Selective media plate counts for inoculated microorganisms were measured from 0 to 105 days after inoculation and transformed by log10 …x 1 1†; which improved the
univariate normal distribution and homogeneity of variance
for the data. Linear regression was used since linear effects
compared to quadratic and cubic effects were most signi®cant. The x-intercept value of linear regression represents
the predicted number of days until viable populations fall
below detectible levels, while the slope of the regression
line represents the rate at which inoculated bacteria
declined. Signi®cance between GEM and parent strains is
reported as an F ratio from pair-wise comparison of slopes
(Sokal and Rohlf, 1995) (Table 2).

2.7. Statistical analysis for the BIOLOG and FAME assays
3. Results
A completely randomized design was used, with microcosms of each soil randomly assigned to treatments. Four
replicate samples from each treatment £ soil combination
were analyzed at each time point over the course of
105 days. Data generated from each soil for both assays
were analyzed separately due to concerns that ecological,
physical, and chemical differences would confound results.
A total of 784 samples were analyzed using BIOLOG plates
and 736 using fatty acid analysis. We used Q±Q plots of
multivariate data (Khattree and Naik, 1995) to assess multivariate normality, a requisite prior to multivariate analysis.
With multivariate methods one also needs to be sensitive to
situations where large numbers of variables are used with
comparatively poor replication. For BIOLOG analysis, the
number of variables was reduced to the nine carbon sources
most signi®cantly different for each soil by utilizing an Fratio test of signi®cance (proc GLM; SAS 1988). Covariates
were used for univariate and multivariate functional analysis to adjust for variations of inoculum density, sample size
and plate variations (see Appendix A for SAS programming). The covariates used were the negative control well
O.D., the log(10) transformed number of inoculated culturable bacteria, and the fresh root weight for each sample.
BIOLOG optical density data was multivariate normally
distributed using the covariates, so no transformation was

The introduced microorganisms on wheat roots were
detectable on selective media in all treatments for the length
of this experiment, and predicted survival using linear
regression ranged from 133 to 391 days in the MD and
SK soils, respectively. Survival was similar for the 3732
GEM and parent in all cases (Table 2). R-square values
for the linear regressions range from 0.52 for the 2-79
GEM in the UM soil, to 0.84 for the 3732 parent in the
CL and SK soils (Table 2).
Microbial community responses using BIOLOG plates
were not visible in most reaction wells until after 48 h of
incubation. At 72 h, there were visible reactions in the
majority of wells and all plates had several wells with no
visible reaction. Least square O.D. means were analyzed in
a linear covariance regression model (SAS 1988; Appendix
A) for all carbon sources and graphed, but only lactose
utilization showed a consistent response. This analysis
showed that lactose response values for the 3732 GEM
inoculated to the wheat rhizosphere ®t a regression model
better using covariables …R2 ˆ 0:82†; in comparison to
unadjusted well readings …R2 ˆ 0:65† or when negative
control well values were subtracted from unadjusted well
readings …R2 ˆ 0:64† (Fig. 1). O.D. response values
analyzed with covariables and expressed as least square

30

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

Table 2
Regression parameters for GEM and parent strains of Pseduomonas chlororaphis 3732 and Pseudomonas ¯uorescenes 2-79 survival and recovery
over time in six soils are reported as selective plate counts from 14 to 105
days after inoculation transformed log10(x 1 1). Linear regression was used
since linear effects compared to quadratic and cubic effects were most
signi®cant. Signi®cance between GEM and parent survival for each strain
is reported as an F ratio from pair-wise comparison of slopes (Sokal and
Rohlf, 1995)
Soil

Inoculum

Y-int

Slope

R2

X-int a

Pr . F b

AG

3732

Parent
GEM

4.87
5.77

2 0.031
2 0.034

0.53
0.76

157
170

NS

BC

3732

Parent
GEM

5.13
6.21

2 0.028
2 0.034

0.54
0.70

184
182

NS

CL

3732

Parent
GEM

5.87
6.07

2 0.023
2 0.020

0.75
0.76

261
303

NS

NR

3732

Parent
GEM

6.11
6.76

2 0.028
2 0.023

0.82
0.80

218
291

NS

SK

3732

Parent
GEM

6.14
6.74

2 0.027
2 0.017

0.84
0.54

256
391

NS

UM

3732

Parent
GEM
Parent
GEM

5.61
6.16
5.41
4.78

2 0.038
2 0.036
2 0.041
2 0.026

0.70
0.81
0.61
0.52

146
171
133
181

NS

2-79

NS

a
The x-intercept represents the predicted number of days the GEM will
survive.
b
The probability of a signi®cant difference between slopes using an Fratio test; NS ˆ Not signi®cantly different at a ˆ 0:05:

means also tended not to vary with inoculum density as
much as other methods (Fig. 1).
Lactose utilization values were the only BIOLOG assay
variable that showed a consistent response when comparing
inoculated treatments to uninoculated controls. Lactose
utilization patterns are shown for the 3732 GEM, 3732
parent, and uninoculated treatments in the SK soil,
compared to 3732 GEM survival and total inoculum levels
(Fig. 1). Uninoculated and 3732 parent inoculated lactose
utilization values were indistinguishable and varied below
the visible range (OD , 0.5) throughout the experiment
(Fig. 1). Wheat root microbial populations inoculated with
the 3732 GEM exhibited high lactose utilization readings,
with levels declining to background as viable 3732 GEM
declined (Fig. 1). These results were similar in the UM soil
(Fig. 2) and in the AG, BC, CL and NR soils (not shown).
When the 2-79 GEM and parent were assayed over time
in the UM soil, lactose utilization by wheat root populations
was not seen for any treatment and the pattern of inoculated
treatments was not differentiable from uninoculated controls
(Fig. 3). Survival rates (slopes) for the 2-79 and 3732 GEMs
in the UM soil were similar (Table 2) as were inoculum
CFU, though recoverable organisms were lower for 2-79
in the UM soil than for 3732 at most time points (Figs. 2
and 3). Pure cultures of the 3732 and 2-79 GEM and parent
inoculated to BIOLOG w GN plates indicated that the
inserted lacZY genes in both the 3732 and 2-79 GEM strains
conferred the ability to utilize lactose and lactulose.

Fig. 1. SK soil 3732 GEM survival (dark hexagons) and total viable bacteria (light hexagons) are shown with lactose utilization patterns over time. BIOLOG
least square mean optical density readings adjusted for covariates are shown for 3732 GEM inoculated (dark circles), 3732 parent inoculated (dark squares)
and uninoculated (dark diamonds). Alternate methods for calculation of well readings shown are unadjusted well optical density (light circles) and optical
density with blank well values subtracted (gray circles). BIOLOG well optical density readings adjusted with covariables are more independent of inoculum
density. Lines represent linear regressions extending to both axes.

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

31

Fig. 2. UM soil 3732 GEM survival (dark hexagons) and total viable bacteria (light hexagons) are shown with lactose utilization patterns over time. BIOLOG
least square mean optical density readings adjusted for covariates are shown for 3732 GEM inoculated (dark circles), 3732 parent inoculated (dark squares)
and uninoculated (dark diamonds) treatments. Lines represent linear regressions extending to both axes.

Colonies of both the 2-79 and 3732 GEMs were also lactose
positive on differential media containing X-GAL.
The most signi®cantly different carbon sources between
treatments in the majority of soils are shown (Table 3).
Lactose and lactulose utilization were signi®cantly different

when comparing 3732 GEM inoculated and uninoculated
treatments (Table 3), though not between uninoculated
and 3732 parent inoculated wheat root communities.
Lactose and lactulose utilization, representing function of
the inserted lacZY genes, did not differentiate 2-79 GEM

Fig. 3. UM soil 2-79 GEM survival (dark hexagons) and total viable bacteria (light hexagons) are shown with lactose utilization patterns over time. BIOLOG
least square mean optical density reading adjusted for covariates are shown for 2-79 GEM inoculated (dark circles), 2-79 parent inoculated (dark squares) and
uninoculated (dark diamonds) treatments. Lines represent linear regressions extending to both axes.

32

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

Table 3
The most signi®cantly different carbon sources were determined using an F-ratio test within each soil (SAS Ð proc GLM) and then results were compared to
look for common effects. If these signi®cant carbon sources appeared in more than half of these tested soils, they are reported here. Percent signi®cance
indicates the number of soils where the listed effect is seen
Treatment comparison

3732 GEM vs. uninoculated

1996

1997

Carbon source

%Sig.

a -d-Lactose b
Lactulose b
p-hydroxy Phenylacetic acid

80
80
60

a

Carbon source

%Sig. a

a -d-Lactose b
Lactulose b
p-hydroxy Phenylacetic acid
l-Aspartic acid

100
67
50
50

3732 Parent vs unionculated

d-Psicose
Mono-methyl Succiante

3732 GEM vs. 3732 Parent

a -d-Lactose b
Lactulose b
b -methyl-d-glucoside

2-79 GEM vs. uninoculated

d-Psicose
a -keto-Butyric acid
d-Saccharic acid

2-79 GEM vs. 2-79 Parent

Methyl Pyruvate d,l-a Glycerol phosphate
Citric acid

2-79 Parent vs. uninoculated

l-Leucine
a -keto-Butyric acid
Lactulose

a
b
c

50
50
100
67
50
c

c

c

The percent of soils where this carbon source utilization is signi®cantly different for the treatment comparison.
Lactose and lactulose utilization differentiated 3732 GEM inoculated, from 3732 parent inoculated and uninoculated treatments in most comparisons.
Carbon sources listed for 2-79 are three most signi®cantly different since 2-79 was inoculated to the UM soil only.

inoculated, from 2-79 parent inoculated, or uninoculated
wheat root communities in the UM soil (Table 3).
Functional diversity, de®ned as the total number of
visible BIOLOG well responses (O.D. . 0.5), showed a
small signi®cant difference in 2 out of 90 time point
comparisons. Total community response for uninoculated
treatments, de®ned as the least square mean of all covariate
adjusted well O.D.s in a BIOLOG plate, gave the highest
values for NR, followed by CL, SK, BC, UM and AG,
respectively. The mean of x-intercepts from survival predictions for all organisms inoculated to each soil (Table 2), a
measure of survival longevity, showed the highest values for
SK followed by CL, NR, BC, AG and UM, respectively.
CDA revealed that approximately 65% of the variance
was represented by graphing the ®rst variable; adding a
second variable represented up to 85% of the variance.
Figures shown use only the ®rst two canonical variables,
though MANOVA utilized the ®rst three CDA variables.
The CDA variables for the SK soil analysis are graphed
for all time points together (Fig. 4), and the same variables
were culled for select time points (Figs. 5 and 6). No significant difference between treatments was seen with the
BIOLOG analysis after day 42 in the SK soil; effects are
shown for day 84 (Fig. 6). The SK soil results are similar for
the other soils, although the CL and NR soils in 1997 still
showed signi®cant multivariate differences at 105 days.
Signi®cant effects (MANOVA F ratios) in all soils

decreased over time in all soils as levels of the inoculated
organisms also decreased.
Signi®cantly different fatty acids between treatments were
compared to signature fatty acids (Vestal and White, 1989;
Frostegard et al., 1993; Cavigelli et al., 1995) then sorted by
type for each soil and treatment combination (Table 4) and
the patterns compared. When comparing signi®cant effects
(fatty acids) between treatments using previously reported
signature fatty acids (Table 4), inoculation changed both
the eubacterial and eukaryotic structure when comparing
GEM by parent, GEM by uninoculated or parent by uninoculated effects in most of the soils. There were no reported
signature fatty acids that consistently differentiated treatments. The inoculated pseudomonads mostly contained
fatty acids common to all living organisms (10:0, 12:0,
14:0, 16:0) (data not shown). Some Gram-negative eubacterial signature fatty acids (10.0 30H, 12:0 20H, 12:1 30H,
17:0 cyclo) were constituent to the inoculated organisms,
but were not signi®cantly different between the uninoculated and inoculated wheat root communities (data not
shown). Using a method for measuring biomass with fatty
acids (Cavigelli et al., 1995), treatments were analyzed for
signi®cant levels of the fatty acids 14:0, 16:0 and 18:0. Few
signi®cant differences were seen, and the highest biomass at
day 0 was not always for inoculated treatments. Fatty acids
that were signi®cantly different between inoculated and
uninoculated treatments were 25:0 (for 3732) and 24:0

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

33

Fig. 4. Functional assay (BIOLOG) canonical discriminant analysis using covariates and the nine most signi®cantly different carbon source utilizations
between treatments for each soil, shown for the SK soil. G ˆ 3732 GEM inoculated (gray circles); P ˆ 372 parent inoculated (black circles); U ˆ unionculated
(light squares). The probability of a great Mahalnobis distance indicates an overall signi®cant difference between 3732 GEM inoculated and uninoculated
treatments, and between 3732 GEM inoculated and 3732 parent inoculated treatments. There is no signi®cant difference between 3732 parent inoculated and
uninoculated treatments.

30H (for 3732 and 2-79). Levels of these fatty acids were
highest on day 0, and in some soils levels remained above
background through day 14, then were indistinguishable
from levels in uninoculated treatments (data not shown).
MANOVA using CDA variables from the fatty acid
analysis showed decreasing signi®cance with time and
with decreasing numbers of inoculated organisms in all
soils. The greatest multivariate signi®cant differences
were seen just after inoculation, when there were high levels
of the inoculated organisms. In most soils there was no
signi®cant difference 42 days after inoculation, though in
the SK soil (Figs. 7±9) and NR soil (not shown), treatments
were still signi®cantly different after 84 days, though signi®cance steadily decreased over time. For the 2-79 GEM
and parent in the UM soil (not shown), effects were similar
to multivariate separation patterns and signi®cant differences seen for the 3732 GEM and parent.
Comparing the two multivariate methods, BIOLOG proved
more sensitive using MANOVA analysis for the BC, CL, and
UM (with 3732) soils. Fatty acid MANOVA analysis was a
more sensitive indicator of overall inoculated organism effects
for the AG, SK and UM (with 2-79) soils. Both methods were
equally sensitive with MANOVA analysis for the NR soil.

4. Discussion
We processed and inoculated dilutions from entire wheat
root samples and chose a single plate incubation time of

72 h for the BIOLOG assay to simplify previously reported
methods (Garland and Mills, 1991; Garland, 1996a) and to
prevent bias due to sub-sampling of root tissue. It was not
possible to standardize inoculum density for each BIOLOG
plate and equally impossible to read the plates multiple
times. We also used a different method for calculating
BIOLOG well readings than those previously reported
(Garland and Mills, 1991; Garland, 1996a, 1997). Subtracting the negative control well from other plate values caused
deterioration of multivariate normality, an assumption that
must be met prior to parametric multivariate analysis
(Johnson and Wichern, 1992). Adjusting each data point
of a plate, regardless of the signi®cance of the adjusting
variable, produces a bias that alters measured effects,
which mutes otherwise obvious trends. For these reasons,
we found that using covariables with BIOLOG readings, in
both univariate and multivariate models, gave the most
repeatable results. This technique proved repeatable in
two separate years using six different soils, and we observed
the same univariate and multivariate effects following
inoculation of P. chlororaphis 3732RN-L11 (3732 GEM)
compared to uninoculated controls.
The BIOLOG assay was sensitive to small changes, as
shown by the lactose utilization response by a small group
of organisms in a large total viable population (Figs. 1 and
2). The lacZ trait is rare in soil bacteria (Barry, 1986, 1988)
so this was a good indicator of the inoculated GEMS.
Lactose utilization was the most signi®cant difference seen
between 3732 GEM inoculated and all other treatments

34

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

Fig. 5. Functional assay (BIOLOG) canonical discriminant analysis results on the day of inoculation (day 0) for the SK soil. G ˆ 372 GEM inoculated (gray
circles); P ˆ 3732 parent inoculated (black circles); U ˆ unionculated (light squares). MANOVA analysis of CDA variables indicates a signi®cant difference
between treatments.

(Table 3). Linear regression of lactose utilization compared
to viable 3732 GEM regression showed that BIOLOG
analysis accurately portrayed potential activity of the inoculated 3732 GEM in a whole community analysis, since the

regressions were parallel (Fig. 1), though in other soils function was reduced compared to survival (Fig. 2). Linear
regression for lactose utilization (Fig. 1) showed that
when viable 3732 GEM dropped below 10 4 CFU g 21 root,

Fig. 6. Functional assay (BIOLOG) canonical discriminant analysis results 84 days after inoculation for the SK soil. G ˆ 3732 GEM inoculated (gray circles);
P ˆ 3732 parent inoculated (black circles); U ˆ unionculated (light squares). MANOVA analysis of CDA variables indicates a signi®cant difference between
treatments.

35

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

Table 4
The ten most signi®cantly different fatty acids for each treatment comparison were determined using a F-ratio test within each soil (SAS Ð pro GLM). These
fatty acids were then compared to previously reported signature fatty acids (Vestal and White 1989; Frostegard et al., 1993; Cavigelli et al., 1995). Each row
contains 10 `*' symbols (one for each of the fatty acids analyzed) divided into columns corresponding to their reported signature fatty acid grouping, if any.
G ˆ 3732 GEM inoculated, P ˆ 3732 parent inoculated, L ˆ 2 2 79 GEM inoculated, N ˆ 2-79 parent inoculated, U ˆ Uninoculated)
Soil

Inoculum

Year

Comparison

AG

3732

1996
1997

G-U
G-U
P-U
G-P
G-U

3732
BC

1996
1997

3732
CL

1996
1997

3732
NR

1996
1997

3732

1996

SK

UM

3732

2-79

a
b
c
d
e
f

1996

1997

G-U
P-U
G-P
G-U
G-U
P-U
G-P
ND f
G-U
P-U
G-P
G-U

G-Neg a

*
*
*
*
***
*

*
*
**

G-U
P-U
G-P

*

G-U
G-U
P-U
G-P

*
*
**
*

L-U
N-U
L-N

**
*
*

G-Pos b

Eub c

Euk d

None e

**
***
*
****
***

***
**
*

**

***
*****
****
*****
*******

***
***
**
**

*
***

*

**
**
***

*
***
**

*
**

****
***
*******
**

*
***

*****
****
****

**
*
*

****
****
*****
*****

*

**
*
***

*******
*****
***

***

*
*
*
*

**
*
***
*

*****
****
****
*****

**

*
*
*

**
**
*

****
*****
*****

***
**
**
*
*
****
*
***

**
*
**
*

**

Characteristic of Gram negative eubacteria
Characteristic of Gram positive eubacteria
Characteristic of eubacteria in general
Characteristic of eukaryotes in general
Characteristic of all organisms or not reported as a signature fatty acid.
Not performed in 1996.

utilization by microbial communities in 3732 GEM inoculated microcosms was indistinguishable from uninoculated
or 3732 parent inoculated treatments. This puts BIOLOG
assay sensitivity for activity of the 3732 GEM at a theoretical 10 CFU g 21 root, since a 10 23 dilution was inoculated
to the plates. We can infer that when the 2-79 and 3732
GEMs were at similar levels, that lacZY activity should
have been similar. We cannot fully explain why the inserted
genes in the 2-79 GEM functioned in pure culture but not as
part of the wheat rhizosphere microbial population here,
though 2-79 was only tested in one soil, and recoverable
2-79 was lower than for 3732 at most sampling times.
Also, the UM soil was among the poorest in terms of survival of inoculated organisms, which may have played a role.
Using other reporter genes or traits should yield equally
sensitive responses, and functional assays may therefore be
applicable for testing introduced organisms containing other

traits. Judging non-target effects to an indigenous microbial
community is another matter. The signi®cant multivariate
functional assay effects seen were mostly from responses of
the inoculated microorganisms, and uninoculated treatments
were not signi®cantly different from inoculated treatments
after the introduced microorganisms had died off. Most
effects other than lactose utilization varied randomly with
no de®nable pattern. We assume that if there were effects in
these already diverse metabolic communities, we could
have quanti®ed them due to the sensitivity of this assay.
Since 2-79 was intended as the microorganism to produce
the most signi®cant community level effects, and it did not
function in the UM soil microbial community as we had
expected, we cannot de®nitively say if this functional
assay will effectively show non-target effects at the microbial community level though if they occur, they may be
masked by activity of the introduced microbes.

36

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

Fig. 7. Structural assay (FAME) canonical discriminant analysis results for the SK soil. Canonical discriminant analysis used the 10 most signi®cantly different
fatty acid measurements between treatments within each soil. G ˆ 3732 GEM inoculated (gray circles); P ˆ 3732 parent inoculated (black circles); U ˆ
unionculated (light squares). The probability of a greater Mahalanobis distance indicates an overall signi®cant difference between 3732 GEM inoculated and
uninoculated treatments, and between 3732 GEM inoculated and 3732 parent inoculated treatments. There is no signi®cant difference between 3732 parent
inoculated and uninoculated treatments.

Risk analysis of genetic constructs, in terms of transfer to
indigenous microorganisms, may employ the lacZY genes
prior to constructing a similar GEM with other inserted
traits. In other words, test the overall safety of the genetic
construct with lacZY only, and infer whether this trait is
transferable to non-target microbes. Also, for ef®cacy testing of the genetic construct, assess whether this inserted
`model' gene functions in the target environment.
The fatty acid assay results complemented the functional
assay in terms of total microbial community response when
quanti®ed with multivariate statistical methods. Multivariate effects decreased as viable inoculated organisms
declined. However, we lacked a signature fatty acid to
complement lactose utilization as an indicator of sensitivity
for the inoculated organisms. Signi®cant effects between
treatments using previously reported signature fatty acids
(Table 4) for eubacteria and eukaryotes indicate effects by
the inoculated organisms, but not in populations that we
quanti®ed. No de®nable trends for individual fatty acids
constituent to the inoculated microorganisms were seen,
though others (Tunlid et al., 1989) have reported increases
in a microorganism constituent fatty acid following inoculation with Flavobacterium balustinum 299 to cucumber
roots, along with an increase in a community fatty acid.
The lack of signature fatty acids to track the microbes inoculated here was not a drawback, since multivariate statistical
methods quantify changes to multiple variables, and represent effects not visible in any single variable.
The fatty acids that were signi®cantly different in all

soils between inoculated and uninoculated treatments
(25:0 and 24:0 30H) may be components of eukaryotes.
Higher molecular weight fatty acids typically are reported
as constituents of eukaryotic organisms (Vestal and
White, 1989; Frostegard et al., 1993; Cavigelli et al.,
1995), although these particular fatty acids have not yet
been attributed to a speci®c group of organisms. These
fatty acids could be constituents of grazing protozoa or
nematodes that bene®ted from the inoculated organisms
as a food source. Protozoa and nematodes have been
distinguished in soils previously using phospholipid
fatty acids (Grif®ths et al., 1999). Protozoa proved to be
rare when we attempted to enumerate them (data not
shown) so these comparisons were inconclusive. The
inoculated pseudomonads were originally proposed for
use as biocontrol agents (Monsanto Corp., 1987, 1988),
so these fatty acids could be constituent to fungi that were
inhibited. We could not measure speci®c pathogenic
fungal levels directly during this study since selective
media is currently not available (D.M. Weller, personal
communication). These fatty acids also may have been
from germinating seeds. Increased soil levels of these
fatty acids seen just after inoculation and until 14 days
may indicate that the inoculated pseudomonads stimulated the growing plants.
Since the fatty acid assay assessed the entire soil organism population and not just viable and culturable bacteria, it
was potentially a better method for determining non-target
effects. The fatty acid assay contains more variables so there

J.V. Gagliardi et al. / Soil Biology & Biochemistry 33 (2001) 25±40

37

Fig. 8. Structural assay (FAME) canonical discriminant analysis results at inoculation (day 0), shown here for the SK soil. G ˆ 3732 GEM inoculated (gray
circles); P ˆ 3732 parent inoculated (black circles); U ˆ unionculated (light squares). MANOVA analysis of CDA variables indicates a signi®cant difference
between treatments.

is a greater potential to assess rare, slow gro