Peak purity analysis in HPLC and

CE using diode-array technology

Product Note

HPLC3D and 3DCE

ChemStations (DOS Series)
Introduction .......................................................... 2

Techniques for peak purity determination ....... 2

In the elution dimension: signals ................... 2
In the third dimension: spectra ...................... 4
Background correction .............................. 4
Normalization ............................................... 6
Selection of peak spectra .......................... 6
Absorbance threshold ................................ 7
Comparing the peak spectra ...................... 7

More advanced spectral techniques ............... 8

The similarity factor ......................................... 8
Improving sensitivity and reliability .............. 9
Similarity curve ............................................. 9
Threshold curve ........................................... 9
Using specific target spectra ...................... 11
Calculating a similarity factor ...................... 12
Without similarity curve ............................ 12
With similarity curve .................................. 12
Extracting signals automatically ................. 12

Using the HPLC3D and 3DCE ChemStation ........ 13

Examples ......................................................... 13
Pure peak .................................................... 13
Impure peaks .............................................. 13
Automation of peak purity determination .. 15

References ......................................................... 15

Glossary .............................................................. 16

1
Introduction
An essential requisite of a separa-
tion analysis is the ability to verify
the purity of the separated species,
that is, to ensure that no coeluting
or comigrating impurity contrib-
utes to the peaks response. The
confirmation of peak purity should
be performed before quantitative
information from a chromato
graphic or electrophoretic peak is
used for further calculations.
Techniques for peak purity determination
Several techniques are currently
used for peak purity determina-
tion in high performance liquid
chromatography (HPLC)1 and in
capillary electrophoresis (CE).2
With a conventional single wave-
length detector (or a monitor pro-
viding just one single output sig-
nal) such as a refractive index- or
conductivity detector, peak purity
can only be judged from the peak
profile of this signal. Peak profile
however is influenced by a variety
of parameters and depends heavi-
ly on chromatographic or electro-
phoretic resolution. Therefore
peak purity determination based
on the peak profile of a single
signal is a very unreliable and
insensitive method. This is espe-
cially true in CE, where mis-
matched sample and buffer zones
always result in peak skewing.2 As
a consequence, a second approach
involving multiple wavelength
detectors acquiring more than one
signal in parallel has been adop-
ted. Such detectors enable impuri-
ties to be uncovered by methods
2
Neglecting peak purity confirma-
tion means, in quality control, an
impurity hidden under a peak
could falsify the results or, in re-
search analysis, important infor-
mation might be lost or scientific
observations rendered void should
an impurity remain undiscovered.
Validated analytical methods usu-
ally include the peak purity check
as a major item in the list of their
method validation criteria.
that involve overlaying signals to
compare peak profiles and calcu-
late signal ratios. These detectors
have some disadvantages and
these will be discussed in the fol-
lowing section. They can be elimi-
nated easily by a third approach
based on diode-array technology:1
on-line acquisition of UV-Visible
spectra during the peaks elution
several signals at different
wavelengths in parallel
signal extracts from a 3-D data
matrix, containing spectral data
and separation signals, for
peak purity analysis.
Peak purity determination can be
performed in different levels,
tailored to the complexity of the
separation and the users needs.
Detailed descriptions of the
various peak purity routines for
signals and spectra, such as
normalization and overlay, the
mathematical operations and the
different display modes will be
given in the following sections.
Method validation criteria
Selectivity (peak purity determination)
Linearity
Limits of detection and quantification
Quality of data (accuracy and precision)
Ruggedness
Table 1
Peak purity determination a major
criterion in method validation
In the elution dimension:
signals
A valuable tool in peak purity
analysis is the overlay of separa-
tion signals at different wave-
lengths to discover dissimilarities
of peak profiles. The availability
of spectral data in the three-di-
mensional matrix generated by the
diode-array detector enables sig-
nals at any desired wavelength to
be selected and reconstructed for
peak purity determination after
the analysis. A set of signals can
be interpreted by the observer
best when, before being displayed,
it is normalized to maximum ab-
sorbance or to equal areas. A good
overlap, where peak shape and
retention or migration time match,
indicate a pure peak while a poor
overlap indicates an impure peak,
as demonstrated in figure 1.
 pure impure

signal A
(offset) signal A
signal B signal B
Figure 1
7.5 8.1 7.5 8.1 Normalized signals
Time [min] Time [min] for pure and impure
peak

mAU
200

150

100

50
signal A signal B

7.6 7.8 8.0 8.2 8.4 8.6

Ratio Time [min]
A-to-B

1.6
1.5
1.4
1.3
1.2 impure pure
1.1 Figure 2
7.6 8.0 8.4 8.6 Signal ratiograms
7.8 8.2
Time [min] for impure and pure
peaks

In addition to overlaying signals, 1 2

their ratios can be calculated and
plotted. The resulting ratiograms
are sensitive indicators of peak
purity (see figure 2). Any signifi-
cant distortion of the ideal rectan-
gular form of the ratiogram
indicates an impurity. However
there are some limitations to be
considered when using signals for
peak purity determination.
The UV-Visible spectra of both the
main compound and the impurity
must be well known in order
to select the most suitable wave-
lengths for the peak profile com-
parison. This fact reduces the
applicability of this type of
impurity detection to a routine-
like search of known impurities
in known main peaks.
The technique is not directly
applicable to research work or
method development. The risk of
overlooking an unknown impurity
remains even when several wave
lengths are selected in parallel.

3
Peak purity determination in
the third dimension:
spectra
Comparing peak spectra is prob-
ably the most popular method to
discover an impurity. If a peak is
pure all UV-Visible spectra ac-
quired during the peaks elution or
migration should be identical,
allowing for amplitude differences
due to concentration. The results
obtained by comparison of these
spectra against each other should
be very close to a perfect 100%
match. Significant deviations can
be considered as an indication of
impurity. Unfortunately the in-
verse is not necessarily true. If the
spectra are not significantly differ-
ent, the peak can still be impure
for one or more of three possible
reasons: (1) the impurity is
present in much lower concentra-
tions than that of the main com-
pound, (2) the spectrum of the
impurity and the spectrum of the
main compound are identical or
very similar, (3) the impurity com-
pletely coelutes or comigrates
with the main compound, with
both having exactly the same peak
profile.
Any peak purity algorithm can
only confirm the presence of
impurities and never prove
absolutely that the peak is pure.
The likelihood of discovering an
impurity rises with increasing
distinction between spectra and
peak profile, higher resolution
between the main compound and
the impurity, and with increasing
absolute and relative concentra-
tion of the impurity.
There are no hard and fast rules
as to which concentrations of
impurities can be detected and
which not. In general, less than
0.51% impurity can be deter-
4
mined if the spectra are distinct
enough. If the spectra resemble
each other very closely, and the
column or capillary system does
not resolve the impurity and main
component well, then only 5%
impurity may be feasible.
Before proceeding, the difference
between the terms spectral im-
purity and peak impurity should
be clearly defined. Spectral impu-
rity indicates a distortion of the
analyte spectrum by the near-con-
stant presence of background ab-
sorbance from solvents, and/or
matrix compounds and/or an im-
purity. Peak impurity, in contrast,
refers to a distortion of the analyte
spectrum by an additional compo-
nent which partially or completely
co-elutes or comigrates with the
major compound.3
Background correction of the
peak spectra
Before the spectra are used for the
peak purity analysis, they should
be corrected for background ab-
sorption caused by the mobile
phase or matrix compounds, by
subtracting the appropriate refer-
ence spectra. Whether such a
background correction needs to
be applied depends on the separa-
tion system employed. For iso-
cratic separations with a properly
balanced diode-array detector, the
solvent's constant spectral contri-
bution will be eliminated by the
automatic subtraction of the sol-
vent spectrum at the beginning of
the run. For gradient separations
where the mobile phases contri-
bution to absorbance may change
with time, background corrections
should be made for each peak
individually. The Hewlett-Packard
HPLC3D and 3DCE ChemStation
softwares offer five different
modes for setting reference
spectra.
Peak apexThe apex spectrum
is subtracted from all other peak
spectra (figure 3). This mode is
recommended if spectra are
collected in the peak controlled
mode. In figure 6 the different
spectra acquisition modes of
Hewlett-Packards diode-array
detectors are illustrated. The
disadvantage of this method is
that the resulting spectra show a
lower absorption (about half of
the expected response) and for
weakly absorbing compounds, the
relative noise level increases.
Baseline spectrathe spectra at
the start and at the end of the
integrated peak are linearly inter-
polated and subtracted from each
peak spectrum (figure 4). This
mode is recommended if spectra
have been collected with the di-
ode-arrays All spectra or All in
peak spectra acquisition modes
(see also figures 6 and 7). A limi-
tation here is that peaks must be
baseline separated, otherwise
impurities from neighboring peaks
are introduced. Selected baseline
spectra should not lie too far
away from the actual peak apex.
Nearest baselineThe baseline
spectrum recorded closest in time
to the right or left of the integra-
tion borders is used here as the
reference spectrum (figure 4).
This mode is recommended if
spectra acquisition has been
performed in peak controlled
mode, and sufficient spectra are
stored in the spectral data file for
there to be a baseline spectrum
close enough for every peak
spectrum.
 without reference spectrum

using apex as reference

peak peak
start end using nearest baseline

apex baseline Figure 3

spectrum spectrum Reference spectra
for peak controlled
spectra

peak 3

Baseline Nearest
Peak segment baseline
peak 5 spectrum
1 b1 to b2 b1
peak 4
peak 1 2+3 b3 to b4 b3
peak 2 4 b3 to b4 b4
peak 6
5 b3 to b4 b3
6 b5 to b6 b5

Figure 4
b1 b2 b3 b4 b 5 b6 Reference spectra
for All spectra or
All in peak spectra

Manual referenceOne or two
reference spectra can be specified
by the user. This mode is used for
interactive spectral evalu- ations
of non-baseline-separated peaks.
Only the spectra in between the
two selected reference points are
used for the purity evaluation.
No referenceSpectral opera-
tions are performed without any
reference (recommended for
exceptional situations only, even
isocratic separations should use a
reference spectrum).
The wavelength range for the
spectra can, and should be,
selected carefully so that only the
significant spectral area is under
observation. This eliminates the
high absorbance of eluants in the
lower UV range that normally
cause high spectral noise. Higher
wavelength ranges should be
omitted if the compounds show
no absorbance to avoid increased
noise and calculation time.

5
 Normalization of the peak
spectra (a) maximumminimum (b) match normalization (c) area normalization
normalization
Before the spectra are compared
they should be normalized. The maximum
ChemStation offers three modes
of normalization (figure 5):
area
maximum absorbance of the minimum
spectrum (or a particular wave- maximum

length range of the spectrum) difference spectrum

area
area of the spectrum (or a wave-
length range of the spectrum) minimum

best possible match of the entire Both spectra have same Area of difference Both spectra have same
spectrum. This type of normaliza- absorbance range spectrum as small area
as possible
tion is recommended to display
spectra for peak purity evaluation
because it tries to make the
difference of both spectra as
small as possible by shifting and
rescaling the spectra. Figure 5
Three normalization modes, (a) maximumminimum normalization, (b) match
normalization and (c) area normalization

Selection of the peak spectra

apex
for comparison
Traditionally, three peak spectra
sufficed for purity determination:
the upslope-, the apex- and the
downslope spectra. This practice
risks overlooking impurities at
the base of the peak. HP diode-
array detectors can acquire all upslope downslope
peak spectra or peak-controlled
spectra with the help of the HPLC3D
and 3DCE ChemStations. Peak
controlled acquisition is subject to
a certain noise absorbance thresh-
old, as described in the following
absorbance baseline
section and shown in figure 6, to threshold
detect peaks. Acquisition either
way provides significantly more Peak controlled
spectral data for more reliable
peak purity evaluations.
All-in-peak

All spectra

Figure 6
Acquisition of spectra at different peak sections

6
The number of spectra per peak 3 peak spectra
to be processed can be three or apex
more. If the value is set to three - 23 width 2
+ 3 width
then three spectra are taken at
roughly equidistant points during
the peaks elution or migration. If
5 peak spectra apex
spectra have been acquired in - 38 width + 38 width
peak-controlled mode during the
- 45 width + 45 width
run, you should select All spectra
for purity determination because
in most practical cases this corre- apex
7 peak spectra
sponds to the three to five spectra - 38 width + 38 width
that will have been recorded. If all - 45 width + 45 width
spectra were acquired during the - width + width
run, a setting of five to seven
spectra to be processed is wise 9 peak spectra apex
(see figure7). Too many spectra - 38 width + 38 width
will only increase calculation time - 5
8 width + 58 width
7
- 98 width- 8 width + 78 width + 9 width
and display time, without provid- 8

ing any more significant informa-

tion. All peak spectra
(all spectra recorded during
elution of the peak)
Absorbance threshold
Setting an absorbance threshold
puts limit to the number of spectra
Figure 7
considered for peak purity. This Position of the selected peak spectra for the different peak spectra selection
reduces the contribution of modes (width is the peak width at half peak height)
spectral noise to normalization
and overlay (figure6).

Comparing the peak spectra ties is the display of difference files result from spectral noise
After selection, correction for spectra, generated by subtracting which may be caused by the in-
background influences and norma- these normalized spectra from the strument (figure8, upper section),
lizing, the spectra can be overlaid other peak spectra selected. The whereas systematic trends will be
to check for possible spectral im- profiles of the difference enable a observed if real spectral differ-
purities. Figure 8 (lower section) further conclusion to be drawn: ences caused by a spectral impu-
shows the normalized and overlaid randomly distributed residual pro- rity occur (figure 9, upper sec-
spectra of a pure peak and figure tion).
9 an impure peak. Any significant
dissimilarity encountered in the
comparison of the spectra re-
corded across the peak indicates
the presence of an impurity. How-
ever no conclusions can be drawn
concerning the kind, number and
level of impurity. An additional aid
in interpreting spectral dissimilari-

Figure 8
Normalized spectra and randomly Figure 9
distributed residual spectra resulting Systematic trends of difference spectra
from spectral noise indicating spectral impurity

7
More advanced spectral techniques

The similarity factor where r is defined as

The ChemStations special peak i=n

purity software routine does not { (A i Aav) ( Bi Bav ) } 16 mAU

only allow the display of spectra
r= i=1

i=n i=n
and their differences, it is also
i=1
(Ai Aav)
2
i=1
2
( Bi Bav )
Absorbance
260 nm
able to calculate a numerical spectrum 2 Similarity 999.968
value to characterize the degree Slope 1.06818
50
of dissimilarity of the peak spec- and where Ai and Bi are measured Intercept 0.04693
tra, a so-called similarity factor, absorbances in the first and sec- 40
based on the match of the peak ond spectrum respectively at the 30
spectra to one another. same wavelength; n is the number 20
of data points and Aav and Bav the 10
Several statistical techniques are average absorbance of the first and 0
0 10 20 30 40 50 Absorbance
available for comparison of second spectrum respectively (see spectrum 1
spectra. Since UV-Visible spectra also figure 10). At the extremes, a
contain only a small amount of similarity factor of 0 indicates no
fine structure, the least-square-fit match and 1000 indicates identical
coefficient of all the absorbances spectra. Generally, values very 18 mAU
at the same wavelength gives the close to the ideal similarity factor
best result. The similarity factor (greater than 995) indicate that 260 nm
used in the ChemStation is defined the spectra are very similar, values
Figure 10
as: lower than 990 but higher than 900 Similarity of absorbance at the indivi-
Similarity = 1000 x r2 indicate some similarity and un- dual wavelengths plotted for a pair of
derlying data should be observed spectra gives the similarity factor
more carefully. Figure 11 shows

(a) very similar spectra (b) different spectra (c) spectra with impurity (d) spectra with noise

Similarity 999.95 Similarity 45.065 Similarity 983.101 Similarity 992.214

Slope 1.12 Slope 0.44 Slope 1.61 Slope 1.61
Intercept 0.18 Intercept 12.4 Intercept -6.45 Intercept -0.016

Spectral difference 0.6% Spectral difference 55% Spectral difference 4.8% Spectral difference 0.5%

Figure 11
Graphical display of similarity factor for different pairs of normalized spectra

8
 examples of similarity factors for
(a) peak without impurity
identical, similar, different and and noise
(b) peak without impurity
but with noise
noisy spectra. The slope of the
regression lines represents the
ratio of the concentration of the
two spectra.

Improving sensitivity and

reliability
980 980
The similarity curve and thresh- similarity similarity
old curve functions improve the curve . curve
sensitivity and reliability of the
1000 1000
peak purity evaluation by using
all the spectra acquired during
the elution or migration of a peak
rather than just three or four.

Similarity curve Peak spectrum, 20 mAU Noise, 0.1 mAU

The mathematical fundamentals Figure 12

used in the similarity curve calcu- Similarity curves for a pure peak with and without noise,
lations are those used for the pu- plotted in relation to the ideal similarity factor (1000) and the
user-defined threshold (980)
rity factor, however they are dis-
played in another format. All spec-
tra from a peak are compared with
one or more spectra selected by Absorbance range [mAU]
25 mAU
the operator, an apex spectrum for 25
example. The degree of match or
spectral similarity is plotted over
20
time during elution. An ideal pro-
file of a pure peak is a flat line at
1000 as demonstrated in figure
Spectral impurity 5 mAU
12(a).
exceeds noise level
At the beginning and end of each
peak, where the signal-to-noise 10
ratio decreases, the contribution
of spectral background noise to
the peaks spectra becomes impor- 5
Spectral impurity
1 mAU
tant. The contribution of noise to within noise level
the similarity curve as shown in 1

figure12(b). 1000 990 980 970 960 950 940

Decreasing similarity
Threshold curve
Figure 13
The influence of noise on weakly Similarity factor as function of the noise level
absorbing spectra can be seen in
figure 13. The similarity factor
decreases with increasing signal-
to-noise ratio or constant noise
level with decreasing absorbance
range.

9
A threshold curve shows the ef-
(a) peak without impurity but with noise (b) peak with impurity and noise
fect of noise on a given similarity
curve. The effect increases rap-
idly towards both ends of the
peak. In essence, a threshold 5% impurity
curve is a similarity curve with
background noise contribution. 980 980 similarity
threshold curve
Figure 14(a) shows both the simi- curve
larity curve and the threshold similarity threshold
curve
curve for a pure peak with noise, curve
1000
figure 14(b) for an impure peak. 1000

The determination of noise thre-

shold is performed automatically Noise, 0.1 mAU Impurity spectrum, 1 mAU
based on the standard deviation of Figure 14
pure noise spectra at a specified Effect of impurity and noise on similarity and threshold curves
time with a user selectable num-
ber of spectra, usually at 0 min-
utes with 14 spectra. In subse-
quent analyses you can define the (a) curves as calculated (b) as natural logarithm (c) as ratio
noise threshold as a fixed value
based on your experience. threshold l (threshold)
n
curve

The threshold curve, represented ln (similarity)

by the broken line, gives the range
for which spectral impurity lies
within the noise limit. Above this
threshold, spectral impurity ex-
1
ceeds the spectral background similarity 995
curve ratio
noise and the similarity curve
intersects the threshold curve
indicating an impurity providing 995
the reference and noise param- 1000 1000 0
eters have been sensibly chosen. Figure 15
The software offers three modes Threshold and similarity curves, (a) as calculated, (b) ln(threshold) and
ln(similarity) and (c) as a ratio
to display the similarity and
threshold curves: (1) without any
transformation, see figure 15(a);
(2) as the natural logarithm, ln ,
see figure 15(b) with the advan-
tage of more detail for the peak
For a spectral pure peak the ratio
apex in the lower part of the
values are below unity and for
graphic; and (3) as a ratio
spectral impure peaks the values
1000 - similarity are above unity. The advantage of
ratio = this mode is that only one line is
1000 - threshold
displayed, instead of two which
see figure 15(c). makes the interpretation more
simple.

10
Using specific target spectra

The ChemStation permits calcula-

tions of the purity factor and simi-
larity curves relative to different
target spectra, as shown in table 2
and figure 16. As a general rule,
the option Compare the average
spectrum provides the most valu-
able information for unknown
impurities. The flexibility of being
able to select a specific target
spectra is valuable in those cases
where the analyst has to assume
where the impurity is, or needs to
improve the sensitivity of the pu-
rity evaluation. An example may
help to show how this principle
can be applied: if the impurity is
assumed to be located on the elut-
ing or migrating tail of the peak,
selecting the tail- or apex spec-
trum to be compared with all
other spectra will provide the
most significant information.
Figure 16 gives the ratio curve for
the front, apex, tail and average
spectrum of a peak which con-
tains an impurity after the re-
sponse maximum (apex). The
front spectrum gives a small spec-
tral impurity at the end of the
peak. The deviation in this first
ratio curve is small since the front
spectrum absorbed little (giving a
rather high threshold curve). The
apex spectrum gives a low impu-
rity in the front of the peak (the
apex spectrum contains only a
very small amount of the impu-
rity) and high impurity at the tail.
The tail spectrum (with a high
amount of impurity) gives a spec-
tral impurity at the front of the
peak. The average spectrum (a
mean of the peak spectra selected,
in this case the upslope-, apex-,
and downslope spectra) indicates
spectral impurity in the total peak.
Peak with
2% impurity
1
0
ratio of ratio of ratio of ratio of
front spectrum apex spectrum tail spectrum average spectrum
to all other spectra to all other spectra to all other spectra to all other spectra
Figure 16
The ratio curves for different target spectra from the same peak
This average spectrum contains,
Compare each individual spectrum
of course, the spectral contribu- to all others
tion of the impurity. In this par- to the apex spectrum
ticular case the average contains to the average spectrum (of all peak
spectra)
more contribution from the impu-
to the front spectrum
rity than the apex spectrum, show- to the tail spectrum
ing a higher spectral impurity at to the front and the tail spectra
the elution or migration front, and
lower impurity at the tail, com- Table 2
Different reference spectra for spectra
pared with the ratio curve of the comparison
apex spectrum.
The profile of the similarity-,
threshold- and ratio curves de-
pends on the position, level and
spectral differences of the impu-
rity and, as such, no general state-
ments can be made on shape. Pro-
file will differ from situation to
situation.

11
 Calculating a similarity factor
Two approaches are available,
depending on whether you elect
to use similarity curves or not.
Without similarity curves
The target spectrum, as specified
by the user, is compared with all
the other selected peak spectra
(3, 5, 7, 9 or all). The mean of all
similarity values below the
specified threshold gives the
similarity factor. If no value falls
below the threshold, the similarity
factor is calculated as the mean of
all values.
With
similarity curves
The mean of all similarity values
exceeding the specified or calcu-
lated threshold gives the similarity
factor, on the condition that at
least three consecutive values lie
over the threshold. If you choose
to use the threshold curve, the
threshold is calculated as the
mean of the same data points used
to calculate the similarity factor.
The spectra used to construct the
similarity curve are indicated by
tick marks in the graphical display
(as annotated in figure 17). The
black marks denote spectra under
the threshold limit. Dark gray
marks denote spectra exceeding
the threshold limit but excluded
from the similarity factor calcula-
tion since insufficient neighboring
spectra lay above the threshold.
Light gray marks denote spectra
included in the similarity factor
calculation since at least two
more, neighboring, spectra also
exceed the threshold.
1
0
similarity value within the threshold limit
similarity value exceeds threshold, but not used in purity calculation
similarity value exceeds threshold three consecutive times,
used in purity calculation
Figure 17
Annotation of the spectra used for the calculation of the similarity curve and the
similarity factor
Extracting signals
automatically from spectra
data
The ChemStation software lets
you automatically select peak
signals for display. These peak
signals are taken from the user
specified signals acquired during
the analysis or are extracted from
the spectral data recorded with all
spectra set during acquisition. The
extraction is optimized to find as
much difference as possible in the
curvature of the signals. This
gives an extra conformation for
peak impurity and in many cases
an indication of the location of the
impurity (in some cases even
when more than one impurity is
present).
 Calculating a similarity factor
Two approaches are available,
depending on whether you elect
to use similarity curves or not.
Without similarity curves
The target spectrum, as specified
by the user, is compared with all
the other selected peak spectra
(3, 5, 7, 9 or all). The mean of all
similarity values below the
specified threshold gives the
similarity factor. If no value falls
below the threshold, the similarity
factor is calculated as the mean of
all values.
With
similarity curves
The mean of all similarity values
exceeding the specified or calcu-
lated threshold gives the similarity
factor, on the condition that at
least three consecutive values lie
over the threshold. If you choose
to use the threshold curve, the
threshold is calculated as the
mean of the same data points used
to calculate the similarity factor.
The spectra used to construct the
similarity curve are indicated by
tick marks in the graphical display
(as annotated in figure 17). The
black marks denote spectra under
the threshold limit. Dark gray
marks denote spectra exceeding
the threshold limit but excluded
from the similarity factor calcula-
tion since insufficient neighboring
spectra lay above the threshold.
Light gray marks denote spectra
included in the similarity factor
calculation since at least two
more, neighboring, spectra also
exceed the threshold.
1
0
similarity value within the threshold limit
similarity value exceeds threshold, but not used in purity calculation
similarity value exceeds threshold three consecutive times,
used in purity calculation
Figure 17
Annotation of the spectra used for the calculation of the similarity curve and the
similarity factor
Extracting signals
automatically from spectra
data
The ChemStation software lets
you automatically select peak
signals for display. These peak
signals are taken from the user
specified signals acquired during
the analysis or are extracted from
the spectral data recorded with all
spectra set during acquisition. The
extraction is optimized to find as
much difference as possible in the
curvature of the signals. This
gives an extra conformation for
peak impurity and in many cases
an indication of the location of the
impurity (in some cases even
when more than one impurity is
present).
13
containing 5% of an impurity with
a very similar spectrum to that of
the main compound. The overlay
of the normalized spectra and the
extracted signals clearly indicate
the presence of an impurity. In
this case the overlay of spectra
and their obvious differences
would be sufficient to recognize
the impurity. Residual spectra
confirm the impurity: a systematic
difference not originating from
random noise. The ratiograms
possess the skewed profile of an
impure peak and the similarity
curve exceeds the threshold
curve, leading to the warning mes-
sage. From the extracted signals
we can conclude that the impurity
Figure 20
is at the end of the peak. Peak purity determination of a 1% impurity

Figures 20 and 21 illustrate the

ChemStations capabilities to
discover very small impurities of
1% and less (exactly 0.5%), with
almost identical spectra to the
main analyte. The spectra overlay,
the residual spectra and even the
ratiogram do not provide sensitive
enough information to uncover the
impurity. However the similarity
and threshold curve are more
sensitive and are able to reveal the
hidden impurity (see zoomed
detail).

Threshold curve
Figure 21
Discovering of 0.5% impurity by
similarity and threshold curve
Similarity curve

14
Automation of peak purity
determination

All the routines described here

can be performed both interac-
tively and fully automatically. All
parameter preferences can be
stored in a single method, and
printed as documentation of the
analysis to aid in the execution of
Good Laboratory Practices (GLP).
Starting such a method initiates
injection, separation, data analy-
sis and reporting of the samples in
one step, unattended. Printed re-
ports contain all the graphical
information from the screen plus
additional numerical results.

Despite the ChemStations capac-

ity to run sample analyses auto-
matically and unattended, we rec-
ommend that you examine the
results yourself first visually be-
fore passing any purity judgement.
Incorrect results most often arise
when a column or capillary sys-
tem fails to resolve peaks or the
ChemStation integrates some
baseline event erroneously lead-
ing to a false reference spectrum
being selected.
References
1
L. Huber, Applications of diode-
array detection in HPLC, Hewlett-
Packard Primer, Publication
number 12-5953-2330, 1989.
2
D. N. Heiger, High performance
capillary electrophoresis,
Hewlett-Packard Primer, Publica-
tion number 12-5091-6199E, 1992.

3
H.-J. P. Sievert and
A. C. J. H. Drouen, Spectral
matching and peak purity in
Diode-Array Detection in High-
Performance Liquid Chromatog-
raphy, 1993, 51125, Marcel
Dekker, New York.

15
Glossary

Absorbance threshold Puts a limit on the signal response considered to represent background or baseline response. Analytes
(signal) showing an absorbance stronger than the limit trigger the integrator's peak recognition function.
Absorbance threshold Puts a limit on the spectra used for peak purity analysis, so that only those showing an absorbance stronger
(spectrum) than a particular level are considered in the calculations, thereby reducing the contribution of noise.
Apex The point of maximum response of an analyte's peak at a particular wavelength.
Background absorption Absorption caused by the mobile phase, electrolyte or matrix compounds. Signals and spectra can be
corrected by subtracting the appropriate reference signal or reference spectrum recorded at a point in the
separation where no analyte absorbance was recorded.
Difference spectra The residual absorbance resulting from the subtraction of one spectrum from another.
Downslope The portion of the peak after the peak apex but before the peak end recognised by the
integrator.
Impurity Anything co-eluting or co-migrating with the analyte under study.
Method A method for a separation should, ideally, contain all the conditions needed to reproduce the analytical
results. For post-run data evaluation, such as peak purity analysis, the evaluation conditions can be stored
together with the separation conditions in the HPLC3D or CE3D ChemStation's method.
Peak-controlled acquisition Spectra can be recorded selectively during the separation, on the peaks only, with an absorbance threshold
for the signal. Any response above this level triggers the acquisition of spectra at regular intervals.
Peak purity analysis By observing an analyte's absorbance characteristics during elution, an analyst can study a peak's homo-
geneity, and make conclusions about the absence of co-eluants or co-migrants. This is what is meant by peak
purity. Non-homogeneous characteristics indicate peak impurity.
Ratiogram A plot of the ratio of the absorbance at two distinct wavelengths over time. A flat trace indicates a homo-
geneous peak, while any distortions indicate an impurity at the wavelengths chosen.
Reference signal Control signal recorded at a wavelength unaffected by analyte absorbance.
Reference spectrum Control spectrum recorded at a point on the baseline near to the analyte peak.
Signal Absorbance measured at a specific wavelength over time. When plotted, this trace gives a profile of the
separation, a chromatogram in HPLC, an electropherogram in CE.

Similarity curve The degree of similarity of one particular analyte spectrum, that at the apex of the peak for example, to all the
other spectra recorded during the separation. The similarity curve can also be plotted as a natural logarithm
to enhance the detail at the peak's apex, or as a ratio to the threshold curve.
Similarity factor Mathematical operations can reduce an analyte's absorbance characteristics to a numerical expression of
(also known as purity factor) peak purity, such that for example, an ideal pure peak would have a purity factor of 1000. A purity factor less
than 990 indicates a degree of uncertainty or probable impurity.
Spectrum Absorbance measured at several discrete wavelengths over a wavelength range. Diode-array technology
can record this data at 100 ms intervals, fast enough for on-line acquisition during the separation.
Spectral similarity The degree of agreement between the absorbance values of a set of spectra.
Target spectrum The spectrum where the most difference in peak homogeneity will be apparent, and therefore where the peak
purity analysis should be targeted at. If an impurity is suspected in the tail, then the target should be the tail.
Threshold curve Shows the effect of noise on a given similarity curve. The threshold curve can also be plotted as a natural
logarithm to enhance the detail at the peak's apex, or as a ratio to the similarity.
Upslope The peak front, the portion of the peak after the peak start recognised by the integrator but before the
peak apex.

The information contained in this note

is subject to change without notice.

Copyright Hewlett-Packard
Company, 1993. All Rights Reserved.
Reproduction, adaption or translation
without prior written permission is
prohibited, except as allowed under the
copyright laws.

Printed in Germany 09/93 on paper

stock bleached without chlorine.

Publication Number
12-5091-7502E

16