A Parametric Correction Scheme for Anomalous HPLC Resultscultures. The inverse RF was then computed...

13 November 2007 Laboratory for Hydrospheric Processes/Code 614.2 1

A Parametric Correction Scheme for AnomalousHPLC Results

Stanford HookerNASA/Goddard Space Flight CenterGreenbelt, Maryland

Laurie Van HeukelemUMCES/Horn Point LaboratoryCambridge, Maryland


A Summary of the CHORS Results in SeaHARRE-3(For TChl a Spanning 0.02–1.37 mg m-3)

The SeaHARRE-3 results aredivided into methods thatwere properly validated orquality assured (dark bars)and those that were not (lightbars). For the latter, theworst-case average result isshown above the bar.CHORS executed two meth-ods based on a C8 and a C18column, denoted S8 and S18,respectively, and both havesignificant problems: the newS8 method has poor TChl aand nearly adequate PPigresults, while the old S18method has the opposite.Higher-order data productsare not as notably degraded,but the functional form of theS8 uncertainties is corrupted.

.

TChl a PPig Sum Ratio Index0

12

24

36

48

60

Pigment Category

QA (Standards Mix)S8 (Standards Mix)QA (Field Samples)S8 (Field Samples)S18 (Field Samples)

SeaHARRE-3

15%

25%

The uncertaintyof this result is

97%

122%

36%38%

641%

39%

Aver

age

Accu

racy

[%]

The functional form of the S8uncertainties in the higher-

order variables is notreproduced.


An Important Clue from an Unequivocally DamagedSet of Samples

During SeaHARRE-2, one ofthe QA laboratories analyzeda set of unequivocally dam-aged samples (they weredefrosted during shipping).The results showed a QAlaboratory analyzing badsamples was superior to amethod lacking a proper QAscheme and analyzing goodsamples. This was confirmedby the precision data as well.An important aspect of theresults, however, was thatthe functional form of thehigher-order variables wasnot properly reproduced inthe damaged samples. Con-sequently, this is now a testto look for aberrant data oranalyses.


The Parameters Describing the CHORSQuantitation Equation

The quantitation of pigments using the CHORS methodology involves the use ofvolumes, peak areas, and an inverse response factor (the calibration term). Theexperiments and analysis conducted at CHORS from 22–25 June (inclusive)showed the following: a) the volumetric terms not involved with calibration areknown to within 1%; b) the peak areas associated with the natural samples and theinternal standard, appear to be known to within 4%, and c) the calibration process isinadequate for calibration and validation activities and has calibration-to-calibrationvariability exceeding 25%.


The Parametric Correction Scheme

Because of the collection of duplicate seawater samples by A. Mannino and M.Russ (MR) as part of coastal investigations with S. Hooker (MODIS investigator), aswell as the participation of CHORS in the SeaHARRE activity (SH-3), three datasets could be assembled wherein CHORS results could be compared to the corre-sponding analyses by HPL. Using the HPL results as truth (whose accuracies arewithin the current state of the art), the original CHORS calibration can be replacedwith a parametric calibration derived from the HPL results. The HPL calibrationuncertainty is to within 3% (95% confidence) over an 18-month time period.The correction processinvolves matching thequantitated values fromCHORS and HPL on apigment-by-pigment andsample-by-sample basis.In each instance, theoriginal CHORS calibra-tion is removed and anew calibration (inverseresponse) factor is com-puted by assuming theHPL data are correct.


The Final C8 Inverse Response Factors DerivedUsing the Parametric Correction Scheme

The final inverse RFs were derived using the following steps:

If an inverse RF could not be computed from two or three of the duplicate data sets,because of an excessive amount of data at low concentrations, the inverse RF wasalso computed by applying the parametric correction scheme to the data from theDHI Mix distributed during SeaHARRE-3. The pigments in the mix were all atconcentrations ensuring large, well-formed peaks, but the mix did not contain all thepigments found in a natural sample, because it was formulated using phytoplanktoncultures. The inverse RF was then computed by analyzing the potential relative RFsof the pigment from the various available data sets, and selecting the result with thebest overall RRF value. For some pigments (Caro, But, Diato, DVChl a, Neo, Phidea, Pras, and Viola), the final inverse RF was computed using only the parametriccorrection of the DHI Mix results .

• Individual inverse RFs were computed from the three duplicate data sets forall pigments and all samples for which the concentration of the pigment wasat least 0.05 mg m-3.

• An average inverse RF was computed for each pigment from those duplicatedata sets having an average concentration of at least 0.05 mg m-3 for thepigment.


The Estimated Accuracy of the C8 Individual PrimaryPigments (PPig) After Final Parametric Correction

The final inverse RF valuesare evaluated by using themto compute new CHORSdata products and comparingthem to the correspondingHPL data (i.e., the HPL dataare the reference data in theuncertainty computations):

Absolute values are used toprevent variance cancellationwhen computing averages(shown in the figure). Thecalibration and validation ob-jective is an average uncer-tainty below 25% (less than15% is desired for algorithmrefinement activities).

. .

25%

15%

0

10

20

30

40

50

Aver

age

Unce

rtain

ty [%

]

MR-1 (N = 104 Mesotrophic–Eutrophic)

SH-3 (N = 24 Oligotrophic–Eutrophic)MR-2 (N = 17 Mesotrophic–Eutrophic)

TChl

a

TChl

b

TChl

cCa

ro But

Hex

Allo

Diad

Diat

o

Fuco

Peri

Zea

Pigment


The Estimated Accuracy of the C8 Corrected DataProducts (For TChl a Spanning 0.02–21.06 mg m-3)

One of the important tests ofthe efficacy of the parametriccorrection is whether or notthe functional form of theuncertainties in the higher-order variables is reproduced(recalling that the CHORSSeaHARRE-3 results had ab-errant relationships). The plotto the right shows the correctfunctional form for the cor-rected data and it also showsthe average uncertainties arevery nearly the same as theuncertainties obtained for thequality-assured laboratoriesfor all three SeaHARRE ac-tivities. Some uncertaintiesare actually lower, which isan artifact of forcing agree-ment with HPL (which usuallyhad the best results).

.

TChl a PPIG Sum Ratio Index0

6

12

18

24

30

Aver

age

Unce

rtain

ty [%

]

Pigment

15%

25%

The functional decrease inuncertainties for the progression

from individual pigments to pigmentsums and pigment ratios, followedby a small increase with pigmentindices is properly reproduced.

MR-1 MR-2 SH-3QA SH-1 QA SH-2 QA SH-3


A Summary of Issues Identified During the FirstInvestigation of CHORS C8 Quantitation Problems

The first investigation of CHORS C8 quantitation problems established a) the para-metric correction scheme and showed it could correct the data to within calibrationand validation requirements (always to within an average uncertainty of 25% andwith few exceptions to within 15%); b) the CHORS calibration working range wasinadequate and contributed significantly to calibration variability (at as much as the50% level); c) some of the other CHORS calibration procedures were less thanoptimal (e.g., substandard pipette choices and the use of unscored vial caps) anddegraded precision; and d) the inverse response factors for new calibration curves(executed using recommended improvements) agreed to within 1% of the responsefactors obtained from the parametric correction scheme.• The original and the new calibration curves were not linear and there was a

difference between the chlorophylls and the carotenoids. The nonlinearitywas also seen in the C18 calibrations.

• The data used in the parametric correction scheme were from July 2005 toJanuary 2006, but the entire CHORS HPLC analysis time period spannedSeptember 2004 to January 2006, so there is a need to demonstrate thesystem and data quality were unchanging over the whole time period.

• The red-to-blue detector ratio during calibrations was not constant and wasevident for both the C8 and C18 methods.


The Temporal Stability of the CHORS C8 HPLCSystem: Inverse RFs at 10 – 15 ng inj-1

The DHI calibrations providean extensive set of actualcalibrations across the entireC8 analysis time period. Byselecting those data with a10 – 15 ng inj-1 concentration,the high variability seen inCHORS calibrations can beavoided. These data show anaverage stability (CV) of ap-proximately 1.5 – 8.0% acrossthe analysis time period,except for the Chl a and Chlb calibrations, which have astability of 16.6 – 28.5%. Thishigher variability was alsoseen in the more frequentSigma calibrations, whichhad a stability of 16.3 – 20.6%.The reasons for this insta-bility are explored later on.

.

J J JJ J

GG G G G

B B

B B

B B

EE E

E

F

FF

A

A

A

0

3x104

6x104

9x104

2005 2005.5 2006

Inve

rse

RF

Time

Parametric CorrectionScheme Time Period

C C CC

CH HH H H HCH

Chl c 2

DiatoZea

a-Caro

Fuco

Peri

SigmaChl a

Sigma Chl b


Convergence and Self Consistency of the C8 InverseResponse Factors from the Parametric Correction

The relative response factor of the CHORS calibrations provides an importantindicator of whether or not the inverse response factors from the parametriccorrection properly converged and are self consistent. For this inquiry, Fuco andMVChl a were chosen as the normalizing pigments. The original CHORS RRFsexhibit a large amount of scatter and an average APD (with respect to the HPL RRFvalues) of 28.9%. The new RRF values from the parametric correction scheme havehave less scatter and an average APD of 13.5%. (A log-log plot is used, so theindividual points are easier to discern with respect to one another.)

.

J JJ

JJJ

J

J

J

J

J

J

J

JJ

J

J

JJ

J

J

J

0.4

1

6

0.4 1 6

Orig

inal

CHO

RS R

RF

HPL RRF

Original CHORSAPD Results

Average = 28.9%

J

JJ

J

J

J

J

J

JJ

JJ

J

JJJ

J

JJ

J

J

J

0.4

1

6

0.4 1 6

New

CHO

RS R

RF

HPL RRF

J CarotenoidJ Chlorophyll

Parametrically-Corrected

APD ResultsAverage = 13.5%


The Effects of Using HPL Calibration Practiceson Accuracy

HPL dilution procedures wereused at CHORS to prepareeight Chl a solutions whichwere injected on the CHORSHPLC using cap septa thatfacilitate injection volume ac-curacy. The correlation coef-ficient (0.9989) is very goodand the slope agrees with theparametric value (1,180.4). Abetter indicator of accuracyare the percent residuals ofthe data with respect to thefinal parametric calibration,which show a nonlinear re-sponse of the CHORS HPLCsystem (also seen in thelarge nonzero y-intercept). Ifthe CHORS working rangeis used, the slope is 811.8.

Correlation coefficients exceeding 0.9999 areachievable, as are average percent residuals to within 2%.

.

RPD wrt Final Parametric Calibration [%

]

JJ

J

J

J

J

JJ

-45

-30

-15

0

E

E

E

EEEEE0

3x106

6x106

9x106

0 2x103 4x103 6x103

Peak

Are

a

Purity-Corrected Concentration

E CalibrationJ RPDs

f (x) = 1,162.3 x – 77,147.2R2 = 0.9989


Individual Examples of the Uncertainties in the C8Parametric Correction Scheme

Although the averageuncertainties for theprimary pigments andthe higher-order dataproducts are to withincalibration and vali-dation requirements,some individual pig-ments do not satisfythis criteria for a largenumber of samples,particularly at lowerconcentrations. Manyof the discrepanciesare a direct result ofanalyst-to-analyst dif-ferences in how smallpeaks are quantitatedand are present in alldatabases (larger un-certainties are normalhere).

.

JJ

JJJ

J

JJJJJJ JJ

JJJ JJJJ

J

J J JJJ

J

J

J

J

J

J JJJJ

JJ

JJJ

JJ

JJJ

J

JJJ

JJJ J JJ

J

JJ

JJ

J JJJJJ J

JJJJJJJ

JJJJJ

JJ

J

J

JJJJJJ

JJJJ

J

JJ

J

J

J

J

J

JJJ

JJJJ

J

JJ

J

JJ

J

JJ

J

JJ

J

J

J

JJJJ

JJ

J

J

J J

J

JJJJJ

J

J

J0

25

50

75

100

0.001 0.01 0.1 1 10

APD

[%]

Concentration [mg m-3]

JJ

J

JJ

JJJ

J

JJ

JJJJ JJ

J

JJJ

J

JJ

JJ

J

J JJ

J

J

JJJJ

J

JJJ

JJJ

JJJ

JJ

J

JJJJ

J

J

J J

J

J

JJ

JJ

JJJ

J

JJJ

J

JJ

J

J J

J

J

JJ

JJ

JJ

J

J

J

J

JJJ

J

J

JJJ

J

J

JJJ

J

J

J J

JJ

J

JJ

J

J

J

JJ

J

J

JJJJJ

J

J

J

JJJ

J

J

JJ JJJ

JJJJ J

J

J0

25

50

75

100

0.001 0.01 0.1 1Concentration [mg m-3]

JJJ

J

J

J

JJJ

J

J

J

J

J

J

J

JJJJJJ JJ

J

JJJ

J

J

JJ

J

J

JJ J

J

J

J

JJ J JJJ

J

J

J

J

J

J

J

JJJJ JJ JJJJJ

J

J

JJJ JJJJ

J

J

JJ

J

JJ

J

J

J

J

J

J

J

J

J

J

J

J

J

J

J

J

J

JJ

J

J

J

J

J

J

J

JJJJJJ J

J

J

JJJJJJJJJJJJJJJJJJ J

JJ

JJ JJ0

25

50

75

100

0.001 0.01 0.1 1Concentration [mg m-3]

JJJ

JJ

J

J

J

J

J

J

J JJ

J

J

J

J

JJ

J

J

J

JJ

J

J

J

JJ

J

J

J

JJ

JJ

J

J J J

J

JJ

J

J

J J

J

J

J

J

J

JJ

JJ

J

JJ

J

J

JJ

JJJ

JJ

J

J

J

JJ

JJJJJJJ

JJ

J

J

JJJJJJJ

JJ

J

J

JJ

J

J

J

J

J

J

J

J

J

JJJ

J

JJJ

JJJJJJ

JJ

J

JJ

J

J

JJJ

J

J

J

J

J

J

J J

JJ

JJJ

J

0

25

50

75

100

0.001 0.01 0.1 1 3JJJJJ

J JJJJJJ JJJJ

J

J

J

JJ

J

J JJ

J

JJ

JJ

J

JJJ

J

J

JJJ

JJ

J

J

JJ

J

J

J

JJJJJ

J

JJ

J

JJJJJJJ

JJJJJJ

J

JJ

JJ

J

J J

J

JJ

JJ

J

J

J JJJ

J

JJ

JJJJJJ

J

J

J

J

J

JJJ

JJ

JJ

JJJ

JJJJ

J

J

J

J

JJ J

JJ

J

JJJ

J

J

JJ

JJ

J

JJ

J

JJJJ

0

25

50

75

100

0.001 0.01 0.1 1

JJ

JJ

JJJJ

JJJJ

JJ

JJJJ

JJJ

JJ

JJ

J

J

JJ

J

JJJJJ

J

J

JJJJJ

JJ

J

J

J

JJ

JJ

JJ J

JJ

J JJ

JJ

J JJ

J

JJJJ JJJJ

J

JJ

JJJJ

JJJ

J

JJJJ

J

JJ

JJ JJ

J

J

J

J

J

J

JJJJJ

J

JJ

JJJ

J

JJ

J JJ

J

J

JJ

JJ

J

J

JJJ

J

J J

JJJ

JJ

J

JJ

J0

25

50

75

100

0.02 0.1 1 10 30

APD

[%]

TChl a TChl b TChl c

Fuco DiatoPeri

Low (<0.1) and very low (<0.01) concentrations are shaded yellow.


The Performance of the C8 Parametric CorrectionScheme in Terms of Detector Nonlinearity

.

JJJJ

JJ

J

J J

J

J

J

JJ

EEEE

EE

E

E

E

E

E

E

EE

JJ

J

J

JJ

J

E

E

EE

E E

E

J JJ

J

JJJÉ É É É ÉÉ É

-50

-30

-10

10

1 10 100 1000

Resid

ual t

o Ca

libra

tion

Curv

e [%

]

Chl a Amount on Column [ng inj-1]

J 664 nmE 436 nm

Buffer

J 664 nmE 436 nm

Water

J 664 nmÉ 436 nm

No Premix

0%

LargestChange

The nonlinearity in CHORS calibrations are reduced if the buffer is replaced bywater, and is completely removed at 436 nm if there is no premixing (left plot). Thereason the parametric correction is not so negatively influenced by the nonlinearityis the majority of the data—the average response of which almost exclusivelyinfluence the parametric correction factors—correspond to the area of largestchange in the expression of the nonlinearity (right plot).


Limitations of the Parametric Correction of theCHORS C8 HPLC Data

The parametric correction scheme cannot correct all of the CHORS pigments withequal efficacy, because the HPL pigment set is different, and HPL pigmentintegration procedures were not the same as those used by CHORS—particularlyfor pigments at low concentrations or with recurring coelution problems—whichinfluences both the duplicate data sets used to derive the corrections and the in situsamples to be corrected. The combination of method-to-method and natural differ-ences within the overall data set partitions the final data set into five groups, with theuncertainties ranging from lowest to highest, respectively (the marginal pigments areshown in green and the unacceptable pigments shown in red):• The higher order pigments: pigment sums, ratios, and indices (which are all

derived from the primary pigments).• The primary pigments: TChl a, TChl b, TChl c, Caro, But, Hex, Allo, Diadino,

Diato, Fuco, Peri, and Zea (some of which are derived from the secondarypigments).

• The secondary pigments: MVChl a, DVChl a, Chlide a, MVChl b, Chl c1, Chlc2, and Chl c3.

• The tertiary pigments: Lut, Neo, Phytin a, Phide a, Pras, and Viola.• The uncorrectable pigments: Gyro-diester, α-Caro, β-Caro, and DVChl b.

A Parametric Correction Scheme for Anomalous HPLC Resultscultures. The inverse RF was then computed...

Documents

Transcript of A Parametric Correction Scheme for Anomalous HPLC Resultscultures. The inverse RF was then computed...