Asymmetric Versus Symmetric Filter Wheels and Associated Processing Algorithms: Results from Asynchronous Fluorescence Imaging Photometer Measurements of Phytoplankton.

The use of rotating filter wheels is common in photometric applications. Traditional filter wheel designs typically exhibit a number of filter openings spaced evenly about the circumference of the wheel. In this work we examine a number of shortcomings of this traditional filter design in measurements of phytoplankton fluorescence made with our fluorescence imaging photometer (FIP). We present an alternative asymmetric wheel design that offers a number of advantages over the traditional design as well as a new processing algorithm designed to accommodate convolution of signals from adjacent channels inherent in measurements collected with the asymmetric design. This approach eliminates the need for a separate signal to establish timing and wheel position, unambiguously establishes filter order even when the direction of rotation is unknown, allows for better estimates of signal baseline, and is more resilient to effects of vibration and other dynamic processes that could occur on the time scale of wheel rotation. We demonstrate performance improvements for phytoplankton fluorescence measurements associated with the new wheel design and algorithm compared with previously published methods using the FIP. Both the improved image processing algorithm and filter wheel design were found to reduce noise in our measurements significantly.

Single-Cell and Bulk Fluorescence Excitation Signatures of Seven Phytoplankton Species During Nitrogen Depletion and Resupply.

Phytoplankton play a vital role as primary producers in aquatic ecosystems. One common approach to classifying phytoplankton is fluorescence excitation spectroscopy, which leverages the variation in types and concentrations of pigments among different phytoplankton taxonomic groups. Here, we used a fluorescence imaging photometer to measure excitation ratios ("signatures") of single cells and bulk cultures of seven differently pigmented phytoplankton species as they progressed from nitrogen N-replete to N-depleted conditions. Our objective was to determine whether N depletion alters the fluorescence excitation signature of each species and, if so, how quickly they recover when N (as nitrate) was resupplied, because these factors affect our ability to classify the species correctly. Of the seven species studied, only Proteomonas sulcata, a marine cryptophyte, showed measurable changes in single-cell fluorescence excitation ratios and bulk fluorescence excitation spectra. These changes were likely due to decreases in the cellular concentration of phycoerythrin, a N-rich pigment, as N became scarce. Within 3 h of resupply of N, fluorescence signatures began returning to pre-depletion values and were indistinguishable from N-replete cells by 80 h after resupply. These data suggest that our classification approach is robust for non-PE containing phytoplankton. PE-containing phytoplankton might exhibit systematic changes in their signatures depending on their level of N depletion, but this could be detected and the phytoplankton re-classified following a few hours of incubation in N replete conditions.

Fluorescence Excitation Spectroscopy for Phytoplankton Species Classification Using an All-Pairs Method: Characterization of a System with Unexpectedly Low Rank.

An all-pairs method is used to analyze phytoplankton fluorescence excitation spectra. An initial set of nine phytoplankton species is analyzed in pairwise fashion to select two optical filter sets, and then the two filter sets are used to explore variations among a total of 31 species in a single-cell fluorescence imaging photometer. Results are presented in terms of pair analyses; we report that 411 of the 465 possible pairings of the larger group of 31 species can be distinguished using the initial nine-species-based selection of optical filters. A bootstrap analysis based on the larger data set shows that the distribution of possible pair separation results based on a randomly selected nine-species initial calibration set is strongly peaked in the 410-415 pair separation range, consistent with our experimental result. Further, the result for filter selection using all 31 species is also 411 pair separations; The set of phytoplankton fluorescence excitation spectra is intuitively high in rank due to the number and variety of pigments that contribute to the spectrum. However, the results in this report are consistent with an effective rank as determined by a variety of heuristic and statistical methods in the range of 2-3. These results are reviewed in consideration of how consistent the filter selections are from model to model for the data presented here. We discuss the common observation that rank is generally found to be relatively low even in many seemingly complex circumstances, so that it may be productive to assume a low rank from the beginning. If a low-rank hypothesis is valid, then relatively few samples are needed to explore an experimental space. Under very restricted circumstances for uniformly distributed samples, the minimum number for an initial analysis might be as low as 8-11 random samples for 1-3 factors.

Focus-independent particle size measurement from streak images: a comparison of multivariate methods.

Our laboratories have recently developed a flow-through imaging photometer to characterize and classify fluorescent particles between 3 and 47 µm in size. The wide aperture of the objective lens (0.7 NA) required for measuring spectral fluorescence of single particles restricts the depth of field, such that a large sample volume results in many particles that are out of focus. Here, we describe numerical methods for determining the size of these objects, regardless of their distance from the focal plane, using image processing and multivariate calibration. An intensity profile is extracted from the images and is used as the input for a variety of calibration methods, including partial least squares, neural networks, and support vector machines. The capabilities of these methods are examined to establish the best method for particle sizing that is independent of focus. We found that support vector machines provided the best results, with size estimation error of ±3.1 µm.