L. Ludvigsen et al., CORRELATING PHOSPHOLIPID FATTY-ACIDS (PLFA) IN A LANDFILL LEACHATE POLLUTED AQUIFER WITH BIOGEOCHEMICAL FACTORS BY MULTIVARIATE STATISTICAL-METHODS, FEMS microbiology reviews, 20(3-4), 1997, pp. 447-460
Different multivariate statistical analyses were applied to phospholip
id fatty acids representing the biomass composition and to different b
iogeochemical parameters measured in 37 samples from a landfill contam
inated aquifer at Grindsted Landfill (Denmark). Principal component an
alysis and correspondence analysis were used to identify groups of sam
ples showing similar patterns with respect to biogeochemical variables
and phospholipid fatty acid composition. The principal component anal
ysis revealed that for the biogeochemical parameters the first princip
al component was linked to the pollution effect and to redox processes
and the second principal component described the geological and geoch
emical features of the samples. Dependent on the data transformation o
f the phospholipid fatty acid profiles in either absolute concentratio
ns (Logarithm transformed) or in mol% of total phospholipid fatty acid
s, different groups of samples and outliers were revealed by the princ
ipal component analysis. The principal component analysis on data in a
bsolute concentrations revealed that many phospholipid fatty acids ref
lected the pollution effect on the biomass composition. In contrast, t
he phospholipid fatty acids in mol% divided the samples into one group
of the more polluted samples and another with the nearly unpolluted s
amples. The important phospholipid fatty acids for this grouping were
mainly a few of the normal saturated phospholipid fatty acids (10:0, 1
6:0 and 18:0). Discriminant analysis was used to allocate samples of p
hospholipid fatty acids into predefined classes; A large percentages o
f samples were classified correctly when discriminating samples into g
roups of dissolved organic carbon and specific conductivity, indicatin
g that the biomass is highly influenced by the pollution. In contrast,
the discriminant analysis revealed that on the basis of the profiles
of phospholipid fatty acids no good discrimination between samples sho
wing dominant sulfate reduction and dominant iron reduction could be m
ade, nor between samples showing dominant nitrate reduction and aerobi
c respiration. Partial least square analysis related the phospholipid
fatty acids data to the biogeochemical parameters assuming linear rela
tionships. After selection of the optimal phospholipid fatty acid comb
ination by genetic algorithms, good partial least squares models with
low prediction errors were gained primarily between the biogeochemical
parameters describing total contents of carbon, pH and chloride. The
models predicting specific activity in terms of, e.g,, sulfate reducti
on activity in a sample had relatively higher prediction errors and lo
w correlation coefficients. This indicates that the phospholipid fatty
acid profiles from complex habitats have limited value for identifyin
g more specific microbial populations.