391
becomes progressively more difficult: 10 p.g N of 1-p.m Synechococcus cells would take
hundreds of hours at present sorting rates.
This situation may be changing with improvements in the sensitivity of techniques such as
stable isotope analyses. Labelling studies are now being carried out using sorting followed by
emission spectroscopy (Lipschultz 1990; P. Wheeler, pers. comm.), and the detection limits
for natural abundance mass spectrometry should now be attainable by sorting, at least for
some of the larger phytoplankton cells (B. Fry, M. Altabet, pers. comm.).
Data Analysis: Efficient and appropriate analysis of the copious multiparameter data from flow
cytometry remains, as always, a challenging area for future research. The amount of data
obtained from flow cytometers will continue to increase both in terms of complexity (number
of parameters measured) and number of samples. The latter aspect is already being influenced
greatly by the introduction of preservation of seawater samples in liquid nitrogen for later
analyses of the picoplankton (Vaulot et at. 1989).
There is a wide consensus that manual, iterative techniques such as gating and "brushing"
(Dean 1990) need to be supplemented with automated techniques for "routine" analyses (e.g.,
experiments in which the signatures do not change dramatically). Examples of approaches to
date include curve-fitting methods (Li 1990) and multivariate analyses such as ordination
(Frankel et al. 1989) and clustering (Murphy 1985). These techniques are versatile and avoid
subjective decisions inherent in interactive analyses such as brushing. Curve-fitting techniques
are relatively simple, requiring user-supplied estimates of the number of popUlations present
and the shapes of the distributions in terms of the parameters measured. Ordination techniques
use various mathematical methods to derive new axes which can be linear or non-linear
combinations of the original variables. The data can be projected onto the new axes, reducing
the number of dimensions while preserving the structure of the data. One commonly applied
ordination technique, principle component analysis, seeks to derive new, orthogonal (hence
uncorrelated) axes which account for most of the variance in the data. Frankel et at. (1989)
used principal component analysis to analyze seawater run on a flow cytometer and were able
becomes progressively more difficult: 10 p.g N of 1-p.m Synechococcus cells would take
hundreds of hours at present sorting rates.
This situation may be changing with improvements in the sensitivity of techniques such as
stable isotope analyses. Labelling studies are now being carried out using sorting followed by
emission spectroscopy (Lipschultz 1990; P. Wheeler, pers. comm.), and the detection limits
for natural abundance mass spectrometry should now be attainable by sorting, at least for
some of the larger phytoplankton cells (B. Fry, M. Altabet, pers. comm.).
Data Analysis: Efficient and appropriate analysis of the copious multiparameter data from flow
cytometry remains, as always, a challenging area for future research. The amount of data
obtained from flow cytometers will continue to increase both in terms of complexity (number
of parameters measured) and number of samples. The latter aspect is already being influenced
greatly by the introduction of preservation of seawater samples in liquid nitrogen for later
analyses of the picoplankton (Vaulot et at. 1989).
There is a wide consensus that manual, iterative techniques such as gating and "brushing"
(Dean 1990) need to be supplemented with automated techniques for "routine" analyses (e.g.,
experiments in which the signatures do not change dramatically). Examples of approaches to
date include curve-fitting methods (Li 1990) and multivariate analyses such as ordination
(Frankel et al. 1989) and clustering (Murphy 1985). These techniques are versatile and avoid
subjective decisions inherent in interactive analyses such as brushing. Curve-fitting techniques
are relatively simple, requiring user-supplied estimates of the number of popUlations present
and the shapes of the distributions in terms of the parameters measured. Ordination techniques
use various mathematical methods to derive new axes which can be linear or non-linear
combinations of the original variables. The data can be projected onto the new axes, reducing
the number of dimensions while preserving the structure of the data. One commonly applied
ordination technique, principle component analysis, seeks to derive new, orthogonal (hence
uncorrelated) axes which account for most of the variance in the data. Frankel et at. (1989)
used principal component analysis to analyze seawater run on a flow cytometer and were able