Sun et al.

Chin Med (2017) 12:16

DOI 10.1186/s13020-017-0137-x Chinese Medicine

RESEARCH Open Access

A simple method for HPLC retention

time prediction: linear calibration using two
reference substances
Lei Sun1,2, Hong‑yu Jin1, Run‑tao Tian3, Ming‑juan Wang1, Li‑na Liu1, Liu‑ping Ye4, Tian‑tian Zuo1
and Shuang‑cheng Ma1*

Abstract 
Background:  Analysis of related substances in pharmaceutical chemicals and multi-components in traditional Chi‑
nese medicines needs bulk of reference substances to identify the chromatographic peaks accurately. But the refer‑
ence substances are costly. Thus, the relative retention (RR) method has been widely adopted in pharmacopoeias and
literatures for characterizing HPLC behaviors of those reference substances unavailable. The problem is it is difficult
to reproduce the RR on different columns due to the error between measured retention time (­ tR) and predicted ­tR in
some cases. Therefore, it is useful to develop an alternative and simple method for prediction of t­ R accurately.
Methods:  In the present study, based on the thermodynamic theory of HPLC, a method named linear calibration
using two reference substances (LCTRS) was proposed. The method includes three steps, procedure of two points
prediction, procedure of validation by multiple points regression and sequential matching. The t­ R of compounds on a
HPLC column can be calculated by standard retention time and linear relationship.
Results:  The method was validated in two medicines on 30 columns.
Conclusion:  It was demonstrated that, LCTRS method is simple, but more accurate and more robust on different
HPLC columns than RR method. Hence quality standards using LCTRS method are easy to reproduce in different labo‑
ratories with lower cost of reference substances.
Keywords:  RP-HPLC, Retention time, Relative retention, Linear calibration using two reference substances, Multi-
component analysis, Traditional Chinese medicines

Background have emerged, and widely used in Chinese pharmaco-

Multi-components analysis is an effective strategy for poeia 2015 edition, the United States Pharmacopoeia
quality control of traditional Chinese medicines (TCMs), (USP39-NF34) and literatures [1–10]. In general, ERS
which have complex chemical profiles. But the classic method provides only one reference chromatogram in
external standard method was severely confined in its the pharmacopoeias, instructions of ERS and literatures.
application due to the high cost of reference substances. But there are hundreds of brands of ­C18 columns in the
As a consequence, substitute reference substance meth- market. It means that the reference chromatogram may
ods such as extractive reference substance (ERS) method be different from the actual chromatogram. Due to the
and single standard to determine multi-components column types and other various factors, the error between
(SSDMC) method for overall quality control of TCMs measured retention time ­(tR) and predicted ­tR by the rela-
tive retention (RR) method cannot be ignored sometimes.
In order to improve the reproducibility of chromato-
*Correspondence: masc@nifdc.org.cn graphic separation and RR, the method of classification
1
National Institutes for Food and Drug Control, No. 2 Tiantan Xili, of ­C18 columns has been proposed [11–15]. The columns
Dongcheng District, Beijing 100050, People’s Republic of China
Full list of author information is available at the end of the article
were divided into three types: A, B and EP. Although the

© The Author(s) 2017. This article is distributed under the terms of the Creative Commons Attribution 4.0 International License
(http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium,
provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license,
and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/
publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Sun et al. Chin Med (2017) 12:16 Page 2 of 12

same type of columns was used to repeat the analyti- HPLC (UV–Vis detector) were used. Matlab software
cal methods, the differences in the performance and the was provided by Math Works Inc. USA. 30 C ­ 18 columns
separation effects were still large. And then the methods (shown in Table 1), from 13 manufacturers, included A,
for selecting columns with equivalent selectivity, such as B, and EP types were used. And most columns belong to
the USP approach [16], the PQRI approach [17, 18] and type B according to the previous study [11–15]. Accord-
Katholieke Universiteit Leuven column classification sys- ing to the PQRI approach [17, 18] and using the data from
tem [19–21] were proposed. Take PQRI approach [17, 18] the USP website (http://www.usp.org/USPNF/columns.
as an example, hydrophobicity (H), steric interaction (S), html), the similarity of columns were calculated using
hydrogen-bond acidity (A), hydrogen-bond basicity (B) col1 as the reference column. The similarity (0–13.18)
and ion-exchange capacity (C), were used to describe the showed that the differences among the columns were
performance of the column. And the similarity between a large, which indicated that the selected columns are in a
column and the reference column was calculated by these wide range and have good representative trait.
five parameters. When the similarity was less than three, Reference substances of psoralen, isopsoralen, Chon-
the two columns were regarded to be equivalent. Using glou saponin I, Chonglou saponin II, Chonglou saponin
the equivalent column, the reproducibility of separation VI, Chonglou saponin VII, ethinylestradiol, and herbal
and RR could be improved to some extent. However, in reference substances including Psoraleae Fructus (Pso-
addition to column, many other factors also have great raleae) and Paridis Rhizome (Paridis), were supplied
influences on the chromatograms, such as the dead vol- by the National Institutes for Food and Drug Control,
ume of chromatographic system, the different structure China. Methanol, acetonitrile, and phosphoric acid were
of analytes, the complexity of the chromatographic con- HPLC graded and supplied by the Fischer Company,
ditions, and so on. Therefore, it is necessary to develop USA. Ammonium nitrate (analytical grade) was sup-
a method that takes all aforementioned factors into plied by Beijing Chemical Works. Water was prepared by
account to reduce the prediction error of the ­tR. Milli-Q system, Millipore Company, USA.
According to the thermodynamic theory of liquid chro-
matography, there is a linear relationship between the t­ R of Preparation of sample solution
the compounds on two different HPLC systems (including Psoraleae [26]: weigh 0.5 g of the powder and place it in a
chromatographs and columns) [22]. For better understand- 50-mL stopper conical flask, then add 25 mL of ethanol.
ing, the pdf of reference 22 (in Chinese, Additional file 1) Sonicate the mixture for 30 min and centrifuge for 5 min.
and the English version of reference 22 (only the section Filter the supernatant through a 0.45-μm PTFE filter.
of theory was translated, Additional file  2) are provided. Paridis [27]: weigh 0.5  g of the powder to a stopper
Combined with the above principle and previous studies conical flask, add 25 mL of ethanol, heat under reflux on
[23–25], a novel method using two reference substances for a water bath for 30  min, cool and filter the supernatant
predicting HPLC ­tR has been proposed (linear calibration through a 0.45-μm PTFE filter.
by two reference substances, LCTRS). The ­StR (arithmetic
average of t­R for the same compound on different HPLC Chromatographic conditions
system under the same chromatographic conditions) is Psoraleae [26]: mobile phase A was water and mobile phase
used as the reference value, and the linear regression is used B was methanol. The elution procedure was shown below:
as the basic algorithm for ­tR prediction. In this study, the 0–20 min, 50%B→70%B; 20–45  min, 70%B→85%B;
method was validated in two medicines on 30 ­C18 columns. 45–50  min, 85%B→90%B; 50–60  min, 90%B. Detection
Compared with the RR method, LCTRS method is proved wavelength was 308 nm. Paridis [27]: mobile phase A was
to be more accurate, and more robust on different HPLC water and mobile phase B was acetonitrile. The elution
columns. Hence, it provides a good prospective application procedure was shown below: 0–40  min, 30%B→60%B;
in quantification of multi-components in TCMs as well as 40–50  min, 60%B→30%B; 50–60  min, 30%B. Detection
related substances in pharmaceutical chemicals. wavelength was 203 nm. Column temperatures were both
set at 30 °C and flow rate was 1.0 mL/min.
Methods
The Minimum Standards of Reporting Checklist contains Results
details of the experimental design, and statistics, and HPLC chromatogram of samples
resources used in this study (Additional file 3). The typical chromatograms of Psoraleae and Paridis were
shown in Fig. 1. The peaks were mainly identified by the
Instruments and reagents reference substances. For those peaks without reference
Waters e2695 HPLC (2998PDA detector), Agilent 1260 substances, UV–Vis spectrum and mass spectrum were
HPLC (DAD detector), and Shimadzu LC-2010A HT used for identification.
Sun et al. Chin Med (2017) 12:16 Page 3 of 12

Table 1  tR (min) of four saponins in Paridis on different columns

No. Brand Chonglou saponin VII Chonglou saponin VI Chonglou saponin II Chonglou saponin I

col1 Discovery Ca18 21.234 ± 0.021 23.001 ± 0.006 32.773 ± 0.007 35.118 ± 0.004

col2 Discovery Cb18 16.555 ± 0.010 17.989 ± 0.006 27.161 ± 0.011 29.154 ± 0.002
col3 Xbridge Ca18 21.101 ± 0.004 23.070 ± 0.027 32.483 ± 0.021 34.963 ± 0.006
col4 BDS HypersilCa18 21.014 ± 0.011 22.898 ± 0.018 32.679 ± 0.032 35.170 ± 0.036
col5 Inertsil ODS-2a 22.132 ± 0.009 24.502 ± 0.004 33.176 ± 0.017 35.936 ± 0.003
col6 Kromasil Ca18 21.276 ± 0.012 23.693 ± 0.016 32.618 ± 0.006 35.929 ± 0.007
col7 Luna ­C18(2)a 20.760 ± 0.018 23.362 ± 0.008 30.941 ± 0.011 33.813 ± 0.003
col8 Luna ­C18(2)b 16.551 ± 0.008 18.865 ± 0.023 25.923 ± 0.016 28.310 ± 0.011
col9 Inertsil ODS-3b 17.225 ± 0.006 19.640 ± 0.005 26.754 ± 0.013 29.471 ± 0.005
col10 Alltima Ca18 20.856 ± 0.008 23.752 ± 0.021 31.687 ± 0.011 34.872 ± 0.004
col11 Symmetry Ca18 21.016 ± 0.015 22.076 ± 0.016 32.470 ± 0.023 35.476 ± 0.010
col12 Gemini Ca18 21.300 ± 0.014 23.756 ± 0.007 31.599 ± 0.003 34.337 ± 0.015
col13 CapcellpakC18MGa 21.076 ± 0.012 23.828 ± 0.004 31.627 ± 0.006 34.695 ± 0.001
col14 Zorbax Extend-Ca18 17.201 ± 0.022 19.731 ± 0.003 27.525 ± 0.009 30.504 ± 0.005
col15 Sunfire Ca18 21.652 ± 0.013 24.065 ± 0.024 32.375 ± 0.010 35.197 ± 0.006
col16 Sunfire Cb18 17.501 ± 0.006 19.571 ± 0.018 27.401 ± 0.004 29.826 ± 0.010
col17 Nucleosil ­C18 ­HDa 21.452 ± 0.013 23.735 ± 0.006 32.747 ± 0.019 35.513 ± 0.011
col18 ODS ­Hypersila 19.669 ± 0.010 21.583 ± 0.022 30.361 ± 0.007 32.657 ± 0.005
col19 CapcellpakC18AQa 20.092 ± 0.013 22.560 ± 0.010 29.568 ± 0.006 32.218 ± 0.005
col20 Spherisorb ­ODS2a 18.684 ± 0.006 21.334 ± 0.021 28.580 ± 0.008 31.385 ± 0.004
col21 Zorbax SB-Ca18 18.438 ± 0.015 20.986 ± 0.010 28.085 ± 0.008 30.849 ± 0.007
col22 DiamonsilCa18 22.082 ± 0.021 25.078 ± 0.005 32.475 ± 0.020 35.721 ± 0.023
col23 DiamonsilCb18 16.769 ± 0.007 19.154 ± 0.008 26.240 ± 0.011 28.935 ± 0.004
col24 Diamonsil ­C18(2)a 19.296 ± 0.016 22.819 ± 0.004 31.192 ± 0.019 34.710 ± 0.009
col25 Kromasil Eternity Ca18 19.732 ± 0.017 22.099 ± 0.018 29.535 ± 0.004 32.153 ± 0.021
col26 Shim-pack VP-ODSb 19.015 ± 0.007 21.059 ± 0.014 29.401 ± 0.002 31.845 ± 0.001
col27 Agilent HC-Ca18 23.492 ± 0.247 25.839 ± 0.250 34.884 ± 0.344 37.637 ± 0.330
col28 Agilent TC-Ca18 20.609 ± 0.003 22.574 ± 0.005 30.806 ± 0.003 32.783 ± 0.002
col29 Venusil MP Cb18 17.970 ± 0.007 20.468 ± 0.011 27.287 ± 0.031 29.985 ± 0.007
col30 Nucleosil ­C18 ­ABa 18.348 ± 0.012 20.209 ± 0.017 29.233 ± 0.011 31.893 ± 0.014
StR (Average ­tR) 19.803 22.110 30.319 33.035
a
 4.6 mm × 250 mm × 5 μm
b
 4.6 mm × 150 mm × 5 μm

Standard retention time ­(StR) that the advantages of using ­StR was better than ­tR of any
Under the same chromatographic conditions, measured single column. In this paper, the deviation (ΔtR) of t­Rmea
retention time (­tRmea) of the four saponins in Paridis on and ­tRpre (formula 2) was used to evaluate the merits and
different chromatographic systems (which includes HPLC defects of RR method and LCTRS method.
instruments and columns, hereinafter referred to as col-
umns due to the differences of ­tR mainly caused by col- 
n
StR = tRi /n (n ≥ 1) (1)
umns) were shown in Table 1. The arithmetic average of
i=1
­tR for the same compound on different columns is called
­StR, formula (1). Just like RR, S
­ tR is the reference value for
calculating the predicted retention time ­(tRpre) of analyte
tR = |tR mea − tR pre| (2)
in the samples. Theoretically, under the same chromato- Linear principle of LCTRS
graphic condition, the RR calculated by different columns According to the chromatographic thermodynamic the-
is constants, but ­StR is not. It will be discussed in Sec- ory, Wang et  al. proved that there was a linear relation-
tion  "Minimum number of columns for ­StR calculation" ship between the ­tR of the same compounds on different
Sun et al. Chin Med (2017) 12:16 Page 4 of 12

Fig. 1  HPLC chromatograms of samples

HPLC system (mainly considered as columns) under the on 30 columns was used to fit multiple point linear equa-
same chromatographic conditions [22], as expressed in tion. The averages of ΔtR (average ± standard deviation)
formula (3) and Fig. 2a, b. were calculated, as shown in Fig. 3. For both medicines,
the prediction deviation was reduced with increasing
tR coli = a × tR colj + b (3) number of columns. However, the prediction accuracy
will not be significantly improved when the number of
Since formulas (1) and (3) are both linear, thus there columns reaches five, which is considered as a low-cost
is a linear relationship between ­ tR and ­ StR for each and reasonable limit. It is recommended to choose five–
compound, as shown in formula (4) and Fig.  2c, d. It is fifteen columns for ­StR calculation.
noteworthy that the correlation coefficient of the linear Even for the columns with same type of packing mate-
regression is higher than that shown in Fig. 2a, b. rial, there are still some differences among the column
stationary phase, packing techniques and errors in the
tR coli = a × StR + b (4) process of chromatographic analysis. Those differences
Minimum number of columns for ­StR calculation will cause deviation of ­tR. The physical explanation of
Theoretically, ­tR on any column can be used as reference ­StR calculation was to evenly mix and refill the station-
value for linear fitting. But the ΔtR calculated with ran- ary phase of the columns selected. Because of reducing
dom column were instable. Thus, the reasonable number the random and system errors, the prediction result was
of columns for S ­ tR calculation was thoroughly investi- accurate and robust.
gated by random sampling. ­StR was calculated based on
1, 5, 10, 15, 20, 25, and 30 columns combined with non- Procedure of two points prediction
replicate random sampling times of 30, 100, 100, 100, For RR method, only one compound was chosen as refer-
100, 100, and 1, respectively. The value of S
­ tR with ­tRmea ence compound (reference substance required), and RR
Sun et al. Chin Med (2017) 12:16 Page 5 of 12

Fig. 2  Linear fitting results of Psoraleae and Paridis, code No. is the same as that in Fig. 1

of all other compounds were used as reference value for saponin VII and Chonglou saponin I) and sample solu-
calculating ­tRpre. For LCTRS method, two compounds tion were performed on a C­ 18 column (col4: BDS Hypersil
were chosen as reference compounds (reference sub- ­C18). The ­tRmea (21.014 and 35.170 min) of two reference
stances required), and ­StR of all other compounds were compounds in the sample solution were obtained by the
used as reference value for calculating t­Rpre. The refer- reference substances solution (Fig. 4a). Then two points,
ence compounds, the value of RR and ­StR were shown in Chonglou saponin VII (19.803, 21.014) and Chonglou
Tables 2 and 3. saponin I (33.035, 35.170), could be determined in the
Take Paridis as an example. First of all, reference sub- coordinate using ­StR as abscissa and ­tRmea as ordinate.
stances solution of two reference compounds (Chonglou Based on the two points, the following linear equation
Sun et al. Chin Med (2017) 12:16 Page 6 of 12

Fig. 3  Prediction results of different number of columns for ­StR calculation

Table 2  RR and reference compound for RR method

Peak 1 Peak 2 Peak 3 Peak 4 Peak 5 Peak 6 Peak 7 Peak 8 Peak 9 Peak 10 Peak 11

Psoraleae (n = 23) 0.226 0.247 0.615 0.657 0.794 0.808 0.878 1.000a 1.061 1.103 1.152
a
Paridis (n = 30) 0.652 0.729 1.000 1.090
a
  Reference compound

Table 3  StR (min) and reference compound for LCTRS method

Peak 1 Peak 2 Peak 3 Peak 4 Peak 5 Peak 6 Peak 7 Peak 8 Peak 9 Peak 10 Peak 11

Psoraleae (n = 23) 9.271 10.122a 25.143 26.854 32.437 32.951 35.843 40.793a 43.254 44.950 46.937
a a
Paridis (n = 30) 19.803 22.156 30.319 33.035
a
  Reference compound

was given: y  =  1.0698x  −  0.1719 (Fig.  4b). Taking S ­ tR Procedure of multiple points regression
of analytes (Chonglou saponin VI and Chonglou sapo- After assignment of the peaks of analytes in the sample
nin II) into equation, the t­Rpre of Chonglou saponin VI solution by prediction of two points regression, the ­tRmea
(23.481 min) and Chonglou saponin II (32.263 min) were of those peaks should be validated by multiple points
attained, respectively. Finally, in the chromatogram of the regression. In this procedure, ­tRmea and ­StR of refer-
sample solution, the corresponding peaks of Chonglou ence compounds and analytes were used to fit a multiple
saponin VI and Chonglou saponin II can be found within points linear regression: Y = 1.1038x − 1.1075 (Fig. 4d).
the range of t­ Rpre ± tRW ­(tRW is abbreviation of t­ R win- Taking ­StR of analytes (Chonglou saponin VI and Chon-
dow, in this case is 0.6 min), as shown in Fig. 4c. It can be glou saponin II) into new equation, the new t­Rpre of
seen that ΔtR of analytes calculated by prediction of two Chonglou saponin VI (23.297  min) and Chonglou sapo-
points were 0.583 min and 0.416 min (The ­tRmea of ana- nin II (32.358 min) were calculated. If ΔtR of all analytes
lytes were 22.898 min and 32.679 min). were less than the given ­tRL ­(tRL: ­tR limit, in this case
Sun et al. Chin Med (2017) 12:16 Page 7 of 12

Fig. 4  Flow chart of LCTRS (Paridis, code No. is the same as that in Fig. 1)
Sun et al. Chin Med (2017) 12:16 Page 8 of 12

is 0.5  min), the prediction was success, otherwise fail- treated as peak series for sequential matching. That is, the
ure (Fig. 4e). In this case, ΔtR were 0.399 and 0.321 min, earlier ­tRpre will be matched to the peak with the earlier
respectively. The step of validation by multiple points ­tRmea. Although ΔtR of peak #6 increased to 1.036 min,
was based on the principle of stepwise linear regression, the match results were correct, as shown in Fig. 5b. This
which can further improve the prediction accuracy. rule can be further applied to multiple-peak series, which
The purpose of setting that the t­RW is larger than has a close ­tR.
­tRL is to increase the amount of suitable columns and
to improve the accuracy of prediction. Generally, the Comparison between LCTRS method and RR method
recommended ranges of ­tRW and ­tRL are 0.8–2.0 and The comparison among unadjusted RR method, adjusted
0.5–1.5 min, respectively. If necessary, the values can be RR method (dead time was measured by ammonium
adjusted in accordance with different samples under dif- nitrate as probe compound), prediction by two points,
ferent chromatographic conditions. If the ΔtR of some and validation by multiple points was summarized in
compounds are large, their ­ tRW and ­ tRL can be set Tables  4 and 5. The results showed that the unadjusted
individually. RR method and adjusted RR method were similar, their
prediction accuracy were bigger and suitable for less pos-
Sequential matching rule itive columns. But the prediction deviation was reduced
If the ­tR of two peaks are too close, e.g. less than 2 min, and the number of positive columns was increased by
there would be a mistake for peak matching by the least LCTRS method. The best was validation by multiple
ΔtR rule. Take Psoraleae for example, as shown in Fig. 5a, points which was based on the prediction by two points.
peak #6 was assigned to peak A in the sample solution on
col6 (Kromasil ­C18) with a small ΔtR of 0.515 min. How- Exclusion of column and compound by linear fitting
ever, peak #5 was not found and peak B in the sample Nonlinear shift of ­tR for a compound on different col-
solution was not matched. When t­ RW was set as 1.2 min, umns could be caused either by different column pack-
­tRmea of peak A was within the window of t­ Rpre of peak ing materials and use of other packing techniques, or by
#5. ­tRmea of the peaks A and B would both fall into the the different compound structure. In order to exclude the
window of ­tRpre of peak #6. Because of the existence of columns and compounds with relatively large nonlinear
one common peak (peak A), peaks #5 and 6 should be shift, linear fitting of ­tRmea and ­StR were performed. The

Fig. 5  Advantage of sequential matching (Psoraleae, code No. is the same as that in Fig. 1)
Sun et al. Chin Med (2017) 12:16 Page 9 of 12

following rules were used to identify the outlier column requirements (the average of correlation coefficient was
and compound. (1) In a regression scatter plot, the com- 0.9989). The outlier columns were col2, 8, 12 (Fig. 6a), 16,
pounds obviously deviated from a regression line (the 19, 20 and 30. For Paridis: no obvious nonlinear devia-
correlation coefficient is usually less than 0.99). (2) ΔtR tion was observed of four saponins. All 30 columns met
was usually larger than 1–2  min. The excluded columns the requirement with average correlation coefficient of
and compounds would not be used for ­StR calculating. 0.9993.
For Psoraleae: no obvious nonlinear deviation was In order to simulate t­R of compound with large
observed of all 11 compounds. 23 columns met the structural difference, reference substances solution of

Table 4  Comparison result by four methods (Psoraleae)

Method Maximum of ΔtR/min Average of ΔtR/min Number of positive ­columnsa

Unadjusted RR 3.494 0.804 2

Adjusted RR 3.001 0.886 3
Prediction by two points 2.194 0.599 5
Validation by multiple points 1.689 0.465 9
a
  The columns which meet the following requirements are called positive column, (1) the resolution of peaks meets the requirements; (2) ΔtR of all pending test
compounds are no more than ­tRL (for Psoraleae is 1.2 min)

Table 5  Comparison of prediction result in four methods (Paridis)

Method Maximum of ΔtR/min Average of ΔtR/min Number of positive ­columnsa

Unadjusted RR 1.811 0.420 12

Adjusted RR 1.562 0.410 9
Prediction by two points 0.836 0.283 25
Validation by multiple points 0.545 0.204 30
a
 tRL = 0.5 min

Fig. 6  Outlier column (a) and Outlier compounds (b), code No. is the same as that in Fig. 1
Sun et al. Chin Med (2017) 12:16 Page 10 of 12

ethinylestradiol mixing with four Chonglou saponins ends, otherwise they are in the middle or near the same
were used to measure ­tR of those five compounds on 30 end]. The coverage corresponding to the smallest ΔtR
columns. Nonlinear shift of ethinylestradiol was observed was 80–100%.The results of Psoraleae (Fig.  7a) showed
on col1 (Fig.  6b), 2–6, 8, 11, 15–18, 26–28 and 30. It the advantage of choosing reference compounds with
appears that the HPLC retention behaviors of ethinyle- smaller linear deviation, when the coverages of t­R were
stradiol and four Chonglou saponins were significantly similar. Therefore, the optimized reference compounds
different on this chromatographic condition. It further for Psoraleae were peak #2 and peak #8, rather than
indicated that the classification and similarity evaluation peak #1 and peak #11 which had a maximum coverage
of columns should be based on the characteristics of col- of ­tR but with more deviation from linearity. The selec-
umns as well as analytes. tion rule decreases the randomness of choosing reference
If the outlier compounds cannot be excluded. The fol- compounds and the amount of calculation (or the ΔtR of
lowing approaches could be used: (1) specify one or more all possible reference compound pairs will be calculated
suitable columns; (2) provide reference substances for every time). The accurate and simple selection proce-
those compounds; (3) use UV–Vis spectrum and/or mass dures were as follows. Firstly, Select two reference com-
spectrum for assistant peak identification. pound pairs with large ­tR coverage (80–100%). Secondly,
exclude compounds with large linear deviation based on
Selection of two reference compounds the linear fitting results. Thirdly, calculate the ΔtR of the
Ideally there should be no difference in selecting any of rest reference compound pairs and select reference com-
the two compounds as reference compounds. However, pound pairs with the smallest ΔtR.
because of the difference of HPLC instruments, columns, tR2 − tR1
compounds structure, complexity of elution condition, Coverage of tR = (5)
tRlast − tRfirst
and accidental error of analysis, different selection of ref-
erence compound pairs will make differences. In order to tR2 is ­tR (or ­StR) of second reference compound; ­tR1 is ­tR
find out the rule for reference compounds selection, each (or ­StR) of first reference compound; t­ Rlast is t­ R (or S­ tR) of
combination of possible reference compound pairs for last compound; t­ Rfirst is ­tR (or ­StR) of first compound.
the two medicines was studied. The average of ΔtR cor- In summary, the establishment procedures of LCTRS
responding to each reference compound pair were cal- were as follows. Firstly, select five–fifteen different
culated and shown in Fig.  7. It can be seen that, for the brands of ­C18 columns, and record the HPLC chroma-
two medicines, the ΔtR of prediction by two points step tograms of reference substances and sample on all col-
would be decreased with increasing coverage of t­R [as umns. Secondly, calculate initial S ­ tR by using all columns,
shown in formula (5). The coverage of t­ R is a reflex of the and perform linear fitting of ­tR on each column with ­StR.
relative position of the two reference compounds. The Exclude outlier columns and compounds, and recalculate
first compound is at one end (with smaller ­tR), the last final ­StR using remaining columns. Finally, select two ref-
compound is at the other end (with bigger ­tR). If the cov- erence compounds with large ­tR coverage and low linear
erage is high, the two reference compounds are near both deviation.

Fig. 7  Selection of two reference compound (abbreviated as RC)

Sun et al. Chin Med (2017) 12:16 Page 11 of 12

Discussion need lots of reference substances for peak identification.

Advantages of LCTRS method But it may be not affordable for routine analysis and
According to the study of Wang et al. [22], ­tR of the com- research using all reference substances. LCTRS is a simple
pounds on different HPLC system follows the linearity and low-cost alternative method for peak identification.
principle. The RR method can be regarded as external Compared with RR method, it need one more reference
standard one point method, which means the regres- substance but is more accurate and suitable for more
sion line is forced to pass origin. However, most of the HPLC columns. LCTRS method provides a good prospec-
linear equations have intercepts, which is why the devia- tive application for overall quality evaluation of TCMs
tion of unadjusted RR method was large. For considering and impurities analysis in pharmaceutical chemicals.
the dead time, adjusted RR method should be better than
unadjusted RR method in theory. But the probe com- Additional file
pound for dead time measurement would be interacted
with mobile phases and stationary phases of the columns. Additional file 1. PDF of reference 22. The authors have got the permis‑
The interaction would increase the error in dead time sion from the copyright holder to use the article.
measurement. So the prediction accuracy of this method Additional file 2. English translation of reference 22. The authors have
was not improved in practice. For prediction by two got the permission from the copyright holder to use the article.
points and validation by multiple points, dead time, gra- Additional file 3. The minimum standards of reporting checklist.
dient delay, volume exclusion effect of stationary phase,
retention behavior of homologous compounds and so
forth, were fully considered. Thus, the prediction accu- Abbreviations
TCMs: traditional Chinese medicines; RR: relative retention; LCTRS: linear
racy was significantly improved. Stepwise linear regres- calibration using two reference substances; tR: retention time; SSDMC: single
sion was used in the validation by multiple points step, standard to determine multi-components; StR: arithmetic average of ­tR; ΔtR:
which further improved the prediction effect. deviation of ­tR; tRpre: predicted retention time; tRmea: measured retention
time; CR: calibrated retention.

Compatibility of LCTRS method and RR method Authors’ contributions

Both LCTRS method and RR method are equivalent in LS and SCM conceived the study. LS and HYJ designed the study. LS and LPY
conducted the experiments. LS, RTT, MJW, LNL and TTZ analyzed the data. LS
mathematics. Formulas can be expressed in the same wrote the manuscript. RTT and SCM revised the manuscript. All authors read
form. In the LCTRS, calibrated retention (CR) is defined and approved the final manuscript.
as the ratio of ­StR of analytes to reference compounds, as
Author details
shown in formula (6). Different from RR, CR is based on 1
 National Institutes for Food and Drug Control, No. 2 Tiantan Xili, Dongcheng
statistics of S
­ tR. Thus, its prediction accuracy was equal District, Beijing 100050, People’s Republic of China. 2 Xinjiang Institute for Food
to LCTRS (only equal to prediction by two points). and Drug Control, Urumqi 830004, China. 3 Chemmind Technologies Co.,
Ltd., Beijing 100085, China. 4 Huainan Food and Drug Inspection Center,
tRi − tR1 Huainan 232007, China.
CR = RR = (6)
tR2 − tR1 Acknowledgements
We would like to thank Ms. Yu Pang from Center for Drug Reevaluation and Mr.
where ­tRi is S­ tR of analytes in CR, or t­ R of analytes in RR; Chao Zhang from Suzhou Institute for Drug Control for the useful suggestions
t­ R1 is ­StR of the first reference compound in CR, or dead on this paper.
time in adjusted RR, or zero in unadjusted RR; ­tR2 is ­StR
Competing interests
of the second reference compound in CR, or ­tR of refer- The authors declare that they have no competing interests.
ence compound in RR.
Availability of data and materials
All data generated or analysed during this study are included in this published
Conclusion article [and its supplementary information files].
A new method for t­R prediction of HPLC chromato-
graphic peaks was proposed. 16 compounds in two medi- Funding
This project was financially supported by Youth Development Research Foun‑
cines under isocratic or gradient elution conditions were dation of NIFDC (2013WA8), National Natural Foundation of China (81303214)
tested through three brands of HPLC instruments with and National Major Scientific and Technological Special Project for “Significant
30 different brands of ­C18 columns. It is demonstrated New Drugs Development” (2014ZX09304307-002).

that the method is simple, accurate, and robust for more

HPLC columns. Furthermore, the calculation approach Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub‑
of ­StR and the selection rule of the two reference com- lished maps and institutional affiliations.
pounds were discussed.
Both multi-components analysis in TCMs and determi- Received: 20 February 2017 Accepted: 30 May 2017

nation of related substances in pharmaceutical chemicals

Sun et al. Chin Med (2017) 12:16 Page 12 of 12

References 15. Dehouck P, Visky D, Vander Heyden Y, Adams E, Kovács Z, Noszál B,

1. Gao XY, Jiang Y, Lu JQ, Tu PF. One single standard substance for the Massart DL, Hoogmartens J. Characterisation of reversed-phase liquid-
determination of multiple anthraquinone derivatives in rhubarb using chromatographic columns by chromatographic tests: comparing column
high-performance liquid chromatography-diode array detection. J Chro‑ classification based on chromatographic parameters and column perfor‑
matogr A. 2009;1216(11):2118–23. mance for the separation of acetylsalicylic acid and related compounds. J
2. Hou JJ, Wu WY, Da J, Yao S, Long HL, Yang Z, Cai LY, Yang M, Liu X, Jiang Chromatogr A. 2004;1025(2):189–200.
BH, Guo DA. Ruggedness and robustness of conversion factors in method 16. Bidlingmeyer B, Chan CC, Fastino P, Henry R, Koerner P, Maule AT, Marques
of simultaneous determination of multi-components with single refer‑ MRC, Neue U, Ng L, Pappa H, Sander L, Santasania C, Snyder L, Wozniak T.
ence standard. J Chromatogr A. 2011;1218(33):5618–27. HPLC column classification. Pharm Forum. 2005;31(2):637–45.
3. Hou JJ, Wu WY, Liang J, Yang Z, Long HL, Cai LY, Fang L, Wang DD, Yao S, 17. Dolan JW, Maule A, Bingley D, Wrisley L, Chan CC, Angod M, Lunte C,
Liu X, Jiang BH, Guo DA. A single, multi-faceted, enhanced strategy to Krisko R, Winston JM, Homeier BA, McCalley DV, Snyder LR. Choosing an
quantify the chromatographically diverse constituents in the roots of equivalent replacement column for a reversed-phase liquid chromato‑
Euphorbia kansui. J Pharm Biomed Anal. 2014;88(88C):321–30. graphic assay procedure. J Chromatogr A. 2004;1057(1–2):59–74.
4. Wu X, He J, Xu H, Bi K, Li Q. Quality assessment of Cinnamomi Ramulus 18. Snyder LR, Dolan JW, Carr PW. The hydrophobic-subtraction
by the simultaneous analysis of multiple active components using high- model of reversed-phase column selectivity. J Chromatogr A.
performance thin-layer chromatography and high-performance liquid 2004;1060(1–2):77–116.
chromatography. J Sep Sci. 2014;37(18):2490–8. 19. Visky D, Haghedooren E, Dehouck P, Kovács Z, Kóczián K, Noszál B,
5. Liao H, Li Q, Liu R, Liu J, Bi K. Fingerprint analysis and multi-ingredient Hoogmartens J, Adams E. Facilitated column selection in pharmaceutical
determination using a single reference standard for Saposhnikoviae analyses using a simple column classification system. J Chromatogr A.
Radix. Anal Sci. 2014;30(12):1157–63. 2006;1101(1–2):103–14.
6. Zhong JS, Wan JZ, Liu YP, Ding WJ, Wu XF, Xie ZY. Simultaneous HPLC 20. Kóczián K, Haghedooren E, Dragovic S, Noszál B, Hoogmartens J,
quantification of seven chromones in Aloe barbadensis Miller using a Adams E. Column selection for pharmaceutical analyses based on a
single reference standard. Anal Methods. 2014;6(12):4388–95. column classification using four test parameters. J Pharm Biomed Anal.
7. Du YY, Li Q, Liu JJ, Yin YD, Bi KS. Combinative method using multi-compo‑ 2007;44(4):894–905.
nents quantitation by single reference standard and HPLC fingerprint for 21. Dragovic S, Haghedooren E, Németh T, Palabiyik IM, Hoogmartens J,
comprehensive evaluation of Rhodiola crenulatav H. Ohba. Anal Methods. Adams E. Evaluation of two approaches to characterise liquid chroma‑
2014;6(15):5891–8. tographic columns using pharmaceutical separations. J Chromatogr A.
8. Da J, Cheng CR, Yao S, Long HL, Wang YH, Khan LA, Li YF, Wang QR, Cai 2009;1216(15):3210–6.
LY, Jiang BH, Liu X, Wu WY, Guo DA. A reproducible analytical system 22. Wang LX, Xiao HB, Liang XM. A new method to improve the reproduc‑
based on the multi-component analysis of triterpene acids in Ganoderma ibility of retention time on reversed phase ­C18 columns in different
lucidum. Phytochemistry. 2015;114:146–54. laboratories. Chin J Anal Chem. 2003;31(10):1232–6.
9. Lu WY, Liu YG, Yang HS, Sheng Y, Shi HM, Yu LL. Simultaneous HPLC quan‑ 23. Sun L, Jin HY, Pang Y, Ma SC. Two reference substances for determina‑
tification of five major triterpene alcohol and sterol ferulates in rice bran tion of multiple components (1): linear calibration using two reference
oil using a single reference standard. Food Chem. 2014;148(4):329–34. substances for identification of chromatographic peaks. Chin J Pharm
10. Zhang YW, Li Q, Lv CX, Liu XJ, Chen XH, Bi KS. Simultaneous deter‑ Anal. 2013;33(8):1424–30.
mination of four active components in Alisma orientale (Sam.) Juz. 24. Sun L, Pang Y, Jin HY, Liu LN, Ma SC. Two reference substances for deter‑
by HPLC–DAD using a single reference standard. J Pharmaceut Anal. mination of multiple components II influence of detection wavelength
2015;5(2):85–92. selection on quantitative analysis. Chin J Pharm Anal. 2013;33(9):1578–86.
11. Gilroy JJ, Dolan JW, Snyder LR. Column selectivity in reversed-phase 25. He B, Yang SY, Zhang Y. A new method of calibration and positioning
liquid chromatography: IV type-B alkyl-silica columns. J Chromatogr A. in quantitative analysis of multi components by single marker[J]. Acta
2003;1000:757–78. Pharm Sin. 2012;12:1653–9.
12. Gilroy JJ, Dolan JW, Carr PW, Snyder LR. Column selectivity in reversed- 26. Department of Health Hong Kong Special Administrative Region The
phase liquid chromatography: V. Higher metal content (type-A) alkyl- People’s Republic of China. Hong Kong Chinese Materia Medica Stand‑
silica columns. J Chromatogr A. 1026;2004(1026):77–89. ards, vol. 3, Government Logistics Department. Beijing; 2012. p. 101.
13. Wilson NS, Gilroy J, Dolan JW, Snyder LR. Column selectivity in reversed- 27. National Commission of Chinese Pharmacopoeia. Pharmacopoeia of
phase liquid chromatography: VI. Columns with embedded or end- People’s Republic of China, vol. I, China Medical Science Press. Beijing;
capping polar groups. J Chromatogr A. 2004;1026(1–2):91–100. 2015. p. 260.
14. Visky D, Vander Heyden Y, Iványi T, Baten P, De Beer J, Kovács Z, Noszál
B, Dehouck P, Roets E, Massart DL, Hoogmartens J. Characterisation of
reversed-phase liquid chromatographic columns by chromatographic
tests: rational column classification by a minimal number of column test
parameters. J Chromatogr A. 2003;1012(1):11–29.