Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

Contents lists available at ScienceDirect

Journal of Pharmaceutical and Biomedical Analysis

journal homepage: [Link]/locate/jpba

Reducing relative response factor variation using a multidetector

approach for extractables and leachables (E&L) analysis to mitigate
the need for uncertainty factors
Mark Anderson Jordi ∗ , Kevin Rowland, Weixi Liu, Xingluan Cao, Jie Zong, Yuan Ren,
Zhu Liang, Xiao Zhou, Michael Louis, Kaitlin Lerner
Jordi Labs, 200 Gilbert St., Mansfield, Ma, 02048, United States

a r t i c l e i n f o a b s t r a c t

Article history: Characterization of Extractables and Leachables (E&Ls) is an important aspect of product quality in impor-
Received 20 February 2020 tant fields such as pharmaceuticals, medical devices and food contact materials. The main goal of an E&L
Received in revised form 17 April 2020 study is identification and quantification of those species which may leach from packaging materials used
Accepted 22 April 2020
to contain pharmaceuticals or which may leach directly out of a medical device or food contact mate-
Available online 1 May 2020
rial and thus may result in patient exposure. It is common practice to perform relative quantitation of
extractables and leachables using surrogate standards due to the large diversity of species observed and
Keywords:
the lack of available reference standards. A key problem in obtaining accurate E&L results arises due to
Extractables
Leachables response factor (RF) variation. Different compounds at the same concentration give different signal inten-
E&L sities and thus have different RF values. Two key aspects of study quality are affected by this problem.
Uncertainty factor First, the evaluation of the number of compounds which are above the toxicologically relevant threshold
Analytical evaluation threshold (analytical evaluation threshold, (AET)) can be affected (RF Problem 1: AET Underreporting). Second, quan-
Response factor database titative accuracy is affected which can reduce the reliability of the margin of safety (MOS) calculations
which serves as the basis of the toxicological evaluation (RF Problem 2: Quantitative Error). RF databases
have been the main solution proposed for solving these problems but do not reduce the underlying RF
variation and lack the scope required to address quantitative error for compounds not contained in the
database. In the absence of other solutions, large uncertainty factors (UF) have been applied in the AET
calculations to account for RF Problem 1: AET Underreporting. These UF factors have been assigned values
of 4 for GCMS and up to 10 for LCMS. Large uncertainty factors have a number of unintended negative
consequences including the need for large amounts of sample concentration (>10X) prior to analysis
resulting in potential compound loss or degradation and increased matrix effects. To overcome these
problems, this publication demonstrates a multidetector approach using an HPLC system coupled with a
Quadrupole Time of Flight Liquid Chromatography Mass Spectrometer (QTOF-LCMS), Charged Aerosol
Detector (CAD) and an Ultraviolet-Visible Detector (UV) and a dual detection Gas Chromatography Mass
Spectrometry (GCMS) system using a Polyarc Reactor system with Flame Ionization Detection (FID).
Herein, it is demonstrated that this combination of methods (the multidetector approach) allowed detec-
tion and accurate surrogate standard quantitation of 217 unique extractables spanning a wide range of
chemical properties (Mw, logP, pKa and boiling point). The combination of optimized detector selection
with appropriate standard selection was verified to provide positive detection for 94% of the compounds
at the AET level and a high level of quantitative accuracy (± 20% for 85% of the compounds and ±40%
for 91% of the compounds) while using only a UF of 2. Unlike the RF database approach, the multidetec-
tor approach is not limited to only those compounds contained in the database but is applicable to the
majority of extractables.
© 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license
([Link]

∗ Corresponding author.
E-mail addresses: mjordi@[Link] (M.A. Jordi), krowland@[Link] (K. Rowland), wliu@[Link] (W. Liu), xcao@[Link] (X. Cao),
jzong@[Link] (J. Zong), dren@[Link] (Y. Ren), zliang@[Link] (Z. Liang), xzhou@[Link] (X. Zhou), mlouis@[Link] (M. Louis),
klerner@[Link] (K. Lerner).

[Link]
0731-7085/© 2020 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license ([Link]
2 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

1. Introduction at a wide range of concentrations for less well-studied polymer

systems. The ideal analytical method would need to provide high
The characterization of extractables and leachables (E&Ls) is an selectivity, universal detectability and high sensitivity for identi-
important aspect of product quality. Leachables from pharmaceu- fication while providing broad dynamic range and high accuracy
tical packaging, medical devices and food contact materials can for quantification (universal response factors). Unfortunately, no
lead to adverse effects on product stability, efficacy and patient single detector technology currently exists which can meet these
safety. Leachables are those compounds or elemental impuri- objectives for all potential E&L analytes. This has been noted in USP
ties which migrate into the drug product from pharmaceutical <1663>, “It is a reality that there is no analytical technique or com-
packaging or into the patient directly or indirectly from medi- bination of analytical techniques that is capable of the discovery,
cal devices under the conditions of use [5]. Extractables are those identification, and quantitation of any and all organic and inorganic
compounds which migrate under conditions which are gener- extractable chemical entities known to science. [4]” It is however
ally harsher than the expected conditions of use and are thus possible to reduce quantitative error and maximize identification
meant to provide an exaggerated list of potential leachables [4]. potential through judicious application of detection methods best
In drug product analyses, leachables studies are typically con- suited to their intended purpose (identification or quantification)
ducted using the drug product. For medical devices, a simulated and through proper surrogate standard selection.
leachables condition (saline or other suitable simulant) is applied To understand the requirements for the analytical methods, it
due to the difficulties in replicating the biological system encoun- is necessary to understand clearly the level of sensitivity needed.
tered by implantable devices. Examples of the deleterious effects Guidance for characterization of extractables and leachables has
of leachables include the reaction of a therapeutic proteins with been issued for pharmaceutical packaging [3–5] and medical
acrylic acid which leached from prefilled syringes or the leaching devices [6–8]. Additional guidelines for control of DNA reactive
of a degradant of the common antioxidant Irgafos 168, (bis(2,4- impurities (mutagenic impurities) is provided in the ICH M7(R1)
di-tertbutylphenyl)phosphate (bDtBBP)) which was found to be guidance [9]. This document defines a threshold of toxicological
highly detrimental to cell growth [15,16]. concern (TTC) (ranging from 1.5 ␮g/day for lifetime exposure to
Key quality aspects in E&L study design include selection of 120 ␮g/day for less than lifetime exposure) which is defined as an
appropriate extraction conditions, proper sample preparation pro- “acceptable intake for any unstudied chemical that poses a negligi-
cedures (prevent loss of E&Ls, enhance detection and remove ble risk of carcinogenicity or other toxic effects.” The TTC and other
matrix interferences) and the suitability of the analytical meth- safety thresholds have been widely used in E&L studies to calculate
ods for identification and quantitation of all extracted E&Ls. an analytical evaluation threshold (AET value) which defines the tox-
Mass spectrometry (MS) detectors are commonly applied for icologically relevant identification threshold (concentration above
both identification and quantification primarily because of their which a compound should be identified and reported) during an
high sensitivity and broad applicability [12–14]. Identification of E&L study [10]. The most current embodiment of the AET relevant
individual leachables and extractables is generally accomplished to medical devices is defined in the ISO 10993-18:2020 guidance
using a combination of gas chromatography–mass spectrome- as follows [8].
try (GCMS) for volatile and semi-volatile compounds and liquid
A 1
chromatography–mass spectrometry (LCMS) for semi-volatile and AET (g/mL) = DBT × ×
B×C×D UF
non-volatile compounds. Secondary detectors for LC and GC have
been applied in some cases but their role is typically not empha- where: DBT is a dose based threshold such as the TTC
sized. Examples include the use of ultraviolet (UV) or evaporative A is the number of devices extracted
light scattering (ELSD) detection for liquid chromatography and B is the extract volume
flame ionization detection (FID) for gas chromatography [14]. C is the maximum number/mass of devices used per patient
The purpose of this article is to demonstrate the suitability D is the dilution or concentration factor
and advantages of the multidetector approach for confirming which UF is an uncertainty factor applied to account for RF variation
extractables are above the analytical evaluation threshold (AET) and has previously been defined as UF = 1/(1-RSD) where RSD is the
and for providing accurate quantification. An analytical method relative standard deviation of a response factor database. A similar
is considered suitable for its intended purpose based on its per- formula was detailed for pharmaceutical packaging as defined by
formance characteristics as determined during method validation PQRI [10]. The current embodiment of the ISO 10993-18 standard
[1,2]. Methods should be selected with their intended purpose defines the UF as mean/[1-(t x std)] where t is degrees of confi-
clearly defined. In an E&L screening study, the goal is to provide dence, mean is the mean of the response factor database and std is
a complete list of all E&Ls present in the extract solution above the standard deviation. The authors would note that this definition
the toxicologically relevant level and to accurately determine their is not equivalent to 1/(1-RSD) and is in our opinion inappropriate
associated concentrations for subsequent toxicological evaluation. as it can result in UF values < 1 in some cases and provides unrea-
The toxicological results from the screening E&L analysis can then sonably large UF values for others cases where there is only modest
be used to identify those leachables which pose significant risk RF variation.
and a determination can be made if additional analyses are war- When performing E&L screening studies, the AET value must be
ranted. For drug products, it is common to use this process to related to the detector response by analyzing a reference standard
identify candidates for inclusion in a leachables assessment of or standards at the AET concentration. The signal from the refer-
stability study samples. For medical devices, the screening study ence standard is then compared to the analytical responses of the
results often serve as a part of a larger biocompatibility assess- E&Ls to determine if they exceed the threshold (reference standard
ment. Based on these objectives, the suitability of E&L screening response at the AET concentration). It is important to keep in mind
methods should be evaluated based on their ability to provide com- that at this stage in the analysis, the identity of the E&L compounds
prehensive identification and quantitation for all E&L’s above the is not yet known. Hence, it is not practical to use the actual E&L com-
relevant toxicological threshold. This is a difficult analytical chal- pounds under study as standards since 1) their identity is not yet
lenge in pharmaceutical packaging studies due to the complexity of known, 2) many of the compounds found to extract are oligomers
many drug product matrices and the high sensitivity required, (typ- or other side products of the polymer system and hence are not
ically low ppb), but can be even more daunting for medical device commercially available and 3) the frequently large number of E&Ls
extracts which often contain hundreds of individual extractables would make this time and cost prohibitive. For these reasons, the
 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334 3

Fig. 1. Effect of RF variation on the AET threshold. Note the difference in peak magnitude for two standards at equal concentration (black arrows) resulting in different
threshold levels and additional compounds which must be identified (green arrows). (For interpretation of the references to colour in this figure legend, the reader is referred
to the web version of this article).

common practice is to use a surrogate standard to determine the dard then in the absence of any adjustment for RF differences, the
response associated with the AET level. This practice introduces the concentration of the target compound will be underestimated. The
potential for significant error into the E&L study due to response concentration of each target compound is used to calculate a mar-
factor variation [11,17]. The response factor (RF) for a given com- gin of safety (MOS) value. MOS values are used to assess the level of
pound is the amount of signal per unit concentration. The RF can risk to the patient from exposure to that particular extractable or
be further defined as the slope of the calibration curve for the ana- leachable at the reported concentration. An underestimated con-
lyte under a given set of analytical conditions and using a specific centration could result in the calculation of an inflated MOS value
detector system [18]. Additionally, it is often analytically expedi- potentially adversely affecting the accuracy of the biocompatibil-
ent to define the relative response factor (RRF) as the slope of the ity assessment and underestimating the toxicological risk. On the
calibration curve for a target compound divided by the slope of the other hand, if the RF value is greater than that of the surrogate
calibration curve for a surrogate compound [18,21]. RRF values are standard, the compounds concentration will be overestimated. This
frequently preferred over RF values as they account for variability could result in an underestimated MOS value and thus a potentially
over time including detector drift and sensitivity changes. Finally, safe device could be considered to not be biocompatible.
it should be noted that linearity of the detector is assumed when To mitigate RF Problem 1: AET Underreporting, a UF was intro-
using the slope as a measure of the RF as non-linearity would consti- duced into the AET equation. Some authors have suggested a UF of
tute variation in the response factor as a function of concentration. 4 for GCMS and a UF of 10 has been proposed for LCMS [17]. For
When comparing a target compound with a surrogate standard, studies utilizing the threshold of toxicological concern (TTC), the
RRF values which approach a value of one indicate equal responses associated AET values typically require a detection limit in the mid
for equal concentrations. If the RRF is below one then the concen- to low ppb (ng/mL) range. As an example, consider a theoretical
tration of the target compound will be underestimated. This could long-term contacting device for which the surface area is 30 cm2 ,
result in the exclusion of compounds from the E&L study that are one device is used and the concentration factor is 1. Using the most
actually above the AET level (RF Problem 1: AET Underreporting). conservative TTC value of 1.5 ␮g/day and a surface area to extrac-
Fig. 1 shows an example of the determination of the AET thresh- tion volume ratio of 3 cm2 /mL (derived from ISO-10993-12), an AET
old for a Gas Chromatography Mass Spectrometry (GCMS) analysis value of 150 ng/mL is calculated prior to applying a UF factor. If a
and demonstrates how surrogate standard selection can negatively UF of 10 is applied to account for RF variation then this drops to
affect the determination of which compounds exceed the thresh- just 15 ng/mL.
old. From this demonstration, it is clear that the peak magnitude  g  A 1
associated with the AET is not a single value but is instead a distri- AET = DBT × ×
bution based on the RF values for all E&L compounds. The degree mL B×C×D UF
of RF variation observed is a function of both the method utilized 1 1
(analytical conditions) and the detector system applied (choice of = 1.5g/day × × = 15 ng/mL
30/3 × 1 × 1 10
instrumentation).
A second, significant problem also arises due to RF variation.
For analyses requiring large extraction volumes, (100 mL-5000
After it is determined that a compound is at or above the tox-
mL for large volume parenterals, or devices using high solvent
icologically relevant level, it is then necessary to determine the
volumes such as dialyzers) this value can be sub part per billion.
compound’s identity and quantity for subsequent toxicological
This level of sensitivity approaches or exceeds the limits of what
evaluation. Differences in the response factor for the E&L compound
current analytical technology can achieve for screening analysis
as compared to the surrogate standard can adversely affect quan-
(non-targeted). It should also be noted that including a UF in the
titative accuracy (RF Problem 2: Quantitative Error) [18,21]. If the
AET calculation does nothing to resolve RF Problem 2: Quantitative
RF value for a compound is less than that for the surrogate stan-
error. It is therefore strongly desirable to mitigate response factor
4 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

variation and with it the need for large UF values. This would pro- sitivity and specificity of the mass spectrometry detectors coupled
vide enhanced reliability for the toxicological evaluation and would with the more consistent responses of the CAD, UV and GCMS-FID
increase the viability of the AET value. detectors was shown to significantly improve quantitative accuracy
The amount of error introduced into an E&L study due to the and provide comprehensive detection for all species analyzed.
RF problems is directly proportional to the amount of RF variation
observed for the detectors used for quantitation. As noted above, 2. Materials and methods
mass spectrometry detection has been widely used in E&L analysis
for both identification and quantitation. It is well known in the lit- 2.1. Chemical reagents and materials
erature that the response factors for different compounds can differ
substantially when using MS detection [19–21,23]. This is partic- The standards used for relative quantitation listed in Table 2 and
ularly significant for liquid chromatography mass spectrometry extractables standards of the 217 compounds used for the deter-
(LCMS) but is also observed to a lesser extent for gas chromatog- mination of the response factors were obtained from commercial
raphy mass spectrometry (GCMS) due to the different means used sources in high purity. Purity of each standard was further verified
to ionize the samples. In a study reported by Blanz et al., which during the analysis by confirming the absence of significant impuri-
included the LCMS analysis of 132 structurally diverse drug candi- ties peaks during QTOF-LCMS-UV-CAD and GCMS-FID analysis. The
dates and 233 corresponding metabolites, 27 metabolites showed majority of the standards were purchased from Sigma-Aldrich Co.
a response factor less than 0.2 and 7 had a response factor greater (St. Louis, MO, USA), TCI America (Portland, OR, USA) or Alfa Aesar
than 2 [21]. In LCMS, the response can further be affected by sample (Ward Hill, MA, USA). LC/MS grade methanol, isopropanol, acetoni-
matrix effects (ion suppression or ion enhancement) which result trile (ACN), ammonium hydroxide and formic acid were purchased
in different response factors for a single compound as a function of from Honeywell (Houston, TX, USA). Dichloromethane (DCM) and
the sample matrix or mobile phase composition [19,29]. Tetrahydrofuran (THF) were analytical grade from Pharmco-Aaper
Ultraviolet (UV) detection at low wavelength is currently the and were used without further purification. Ultrapure water (18
overwhelming method of choice for relative quantitation when M cm, EMD Millipore Billerica, MA) was used. Chemical proper-
screening for drug impurities [20,22,23]. UV response factors can ties for the extractables (Mw, boiling point, log P values and pKa)
change for different compounds due to changes in the molar were obtained from reputable websites such as PubChem, Chem-
extinction coefficient (type of chromophore) but this approach is Spider, ChemSRC and ChemAxon.
generally preferred to MS relative quantitation for compounds with
a consistent chromophore as it has the advantage of not being sus-
2.2. Standard preparation
ceptible to sample matrix effects. Other detection methods have
been applied to provide “varying degrees of orthogonality” and
Stock solutions of the standards for RF determination were pre-
increase the “likelihood of separating and detecting all” impurities
pared by individually dissolving them at 1 mg/mL. Portions of the
including Chemiluminescent nitrogen detectors (CLND), flame ion-
stock solutions were then combined into working solutions con-
ization detection (FID), refractive index (RI) and 1 H NMR [23,27,28].
taining groups of standards which were subsequently diluted to
Mass based detectors including charged aerosol detection (CAD)
concentrations of 5, 10 and 25 ␮g/mL for analysis. Methanol was
and evaporative light scattering (ELSD) were applied to supple-
used to dissolve standards for QTOF-LCMS-UV-CAD while DCM was
ment UV because they do not require a chromophore and they
applied for GCMS-FID analysis. In rare instances, standards which
provide signals which are less dependent on the “properties of
were insoluble in methanol or DCM were initially dissolved into
the individual analytes” [26,28]. While both mass-based detectors
another suitable solvent (acetonitrile or tetrahydrofuran) and then
have limitations, including moderately less sensitivity than other
subsequently diluted into methanol or DCM.
methods, the CAD detector showed particular promise as reduced
variation in response was observed across a wide range of chemi-
cals. 2.3. GCMS-FID analysis
The purpose of this publication is to demonstrate the suitability
of the multidetector approach to obtain comprehensive and accu- The slope of the calibration curves (RF values) of each
rate quantification for an extensive collection of extractables (217 extractable were determined using the mass selective detector
extractables) without the use of large uncertainty factors (UF) by (MSD) and flame ionization detector (FID). All GCMS-FID analy-
surrogate standard quantitation. The extractables analyzed were ses were performed on an Agilent 7890B gas chromatograph with
selected to represent the universe of potential extractables and dual detection combining both a 5975 MSD and an FID (Agilent
covered a very wide range of properties (Mw, log P, pKa and Technologies, Santa Clara CA). The two detectors were connected
boiling point) and contained a significant number of compounds post column using a flow splitter for simultaneous detection. A pol-
reported to be difficult to analyze (specifically 42 organic acids yarc catalytic reactor system (Activated Research Company, Eden
and 38 amines). The multidetector approach consists of an HPLC Prairie, MN) was connected after the column splitter and prior to
system using a Quadrupole Time of Flight Liquid Chromatography the FID to convert all organic molecules to methane for FID detec-
Mass Spectrometer (QTOF-LCMS), Charged Aerosol Detector (CAD) tion. MS analyses were performed in electron ionization (EI) mode.
and an Ultraviolet-Visible Detector (UV) and a dual detection Gas The mass spectrometer was set to monitor a mass range of 29–500
Chromatography Mass Spectrometry (GCMS) system coupled with amu. Separations were performed on an HP-5MS 30 m x 0.25 mm
a Polyarc Reactor system with Flame Ionization Detection (FID). In x0.25 ␮m film thickness (Agilent Technologies, Santa Clara CA). The
this approach, the QTOF-LCMS detector is used solely for identifi- GC oven was ramped from 28−270 ◦ C at a ramp rate of 15 ◦ C per
cation of compounds above the AET threshold while the CAD and minute and held for 10 min at 270 ◦ C.
UV are applied for quantitation in HPLC. GCMS and FID are applied
for identification and quantification with the primary advantage of 2.4. QTOF-LCMS-UV-CAD analysis
the FID being its ruggedness for quantitation and the GCMS its iden-
tification capabilities. The multidetector approach described herein The slopes of the calibration curves (RF values) for each
relies on the combination of HPLC (QTOF-LCMS-UV-CAD) and GC extractable were determined using a Agilent 6545 quadrupole time
(GCMS-FID) to provide a high rate of coverage for the full range of of flight mass spectrometer (QTOF) using a Dual Agilent Jet Stream
extractables and leachables studied. The combination of high sen- Electrospray Ionization source in both positive and negative ion
 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334 5

modes, an Agilent 1260 Diode Array Ultraviolet Visible (UV) detec- weights, logP values, boiling point, pKa and pKb) for the extracta-
tor and a Corona Veo RS charged aerosol detector (CAD) (Thermo bles used to probe the RF variation of the method. The multidetector
Fisher Scientific, Waltham MA) combined with an Agilent 1290 approach utilizes a triple detection system combining QTOF-LCMS-
HPLC system (Agilent Technologies, Santa Clara CA). The QTOF anal- UV-CAD detectors and a dual detection system with GCMS-FID.
ysis parameters included: Gas temperature 350 ◦ C; VCap 3500 V; QTOF-LCMS was operated in both positive and negative ion modes.
Nebulizer (N2) 15 psi; Drying gas (N2) 12 L/min; Sheath gas 8 L/min; Thus, a total of 6 different signals were acquired for each extractable
Sheath Gas Temperature 400 ◦ C.; mass range m/z 80−3200. The compound (LCMS Positive ion, LCMS Negative ion, UV, CAD, GCMS
UV detector was set to monitor at a wavelength of 230 nm using and FID). The RRF values were normalized using the average RRF
bandwidth of 4 nm. The CAD detector used a gas pressure of 58 value for the complete distribution for each detector. This has
PSI, evaporation temperature of 35 ◦ C and a charge voltage of 2.37 several advantages. The primary advantage being that it greatly
kV. Separations were performed on an Agilent Zorbax Eclipse Plus simplifies data interpretation such that a value of 1 is equal to the
C-8, 1.8 ␮m, 2.1 × 50 mm (Agilent Technologies, Santa Clara CA) average RRF in each distribution. An RRF value greater than 1 then
maintained at 45 ◦ C. The mobile phase consisted of 0.05% formic indicates a strongly responding extractable and a value less than
acid/0.03% ammonium hydroxide/5% methanol in water (solvent 1 indicates a poorly responding extractable in comparison to the
A) and 0.05% formic acid in methanol (solvent B) and Isopropanol average RRF for the distribution. Table 1 summarizes the RRF dis-
(solvent C) delivered at 0.65 mL/min. Gradient elution had an initial tribution results for each detector. Fig. 3 shows the resulting RRF
condition of solvent A (100%) for .3 min, at 5.3 min (B:C 80%/20%), distribution plots for all 6 detectors. Only the compounds which
at 8.3 min (B:C 50%/50%), at 10.54 min (B:C 50%/50%). Run time was showed a response on a given detector system are included in that
10.54 min with a 1.5-minute post time to allow for equilibration. detector’s RRF distribution. This is appropriate because only com-
The HPLC effluent was introduced directly into the MS system. pounds which show a response would be used for quantitative
assessment (non-responding compounds are not measured at all
and must be detected by another approach). All 217 extractables
2.5. Data processing
analyzed in this study showed a response by one or more detectors
demonstrating the wide applicability of this strategy.
Data reduction was performed in Mass Hunter Qualitative Anal-
The distribution plots for the 6 detectors were found to differ
ysis Version B.10.0.10305.0 (Agilent Technologies, Santa Clara CA).
substantially in terms of RRF variability and in terms of the breadth
Integration of the total ion chromatogram (TIC) was used for deter-
and type of compounds covered. This is a reflection of the different
mination of the peak areas for GCMS. Integration of the compound
principles upon which each detector is based. LCMS positive and
chromatogram (summation of all compound related ions) was used
negative mode detected 72% and 34% of the overall extractables,
for determination of the peak area for QTOF-LCMS. The UV signal
respectively, making LCMS positive mode the single most compre-
at 230 nm and the CAD and FID chromatograms were integrated
hensive detector in terms of compound coverage. Unfortunately,
without further data processing. A linear regression analysis was
the LCMS positive and negative mode distributions showed a very
used to determine the slope of the calibration curve (RF value) for
wide range of RRF values with a maximum number of compounds
each extractable using duplicate injections at 5, 10 and 25 ␮g/mL.
between 0 and 0.2 and the second most abundant region at > 1.8.
Irganox 245 was used as the surrogate standard for determination
This is reflected in the very large% RSD values for the distributions
of the RRF values for QTOF-LCMS-UV-CAD. Decane was used as the
(LCMS Positive% RSD, 108%, Negative Mode LCMS%RSD, 119%). This
surrogate standard for determining RRF values in GCMS-FID. Fol-
indicates that a disproportionate number of compounds observed
lowing the determination of all RRF values for each detector, the
by LCMS give either very weak or very strong responses in compar-
RRF values were further normalized using the average RRF value
ison with the average (RRF = 1). This has very significant negative
for each distribution thus making the choice of surrogate standard
implications for both RF Problem 1: AET Underreporting and RF
immaterial to the final values.
Problem 2: Quantitative Error when LCMS is applied for relative
quantitation placing great importance on appropriate surrogate
3. Results and discussion standard selection. Only 33% of the compounds detected by positive
mode LCMS and 22% of the compounds detected by negative mode
3.1. Relative response factor distribution results LCMS had an RRF value between 0.6–1.4 (±40% of the actual con-
centration) without surrogate standard optimization. This further
In order to understand the magnitude of RF Problem 1: AET demonstrates that the approach of using a single internal standard
Underreporting and RF Problem 2: Quantitative Error, it is nec- for relative LCMS quantitation will frequently result in very sub-
essary to understand the RF distribution for the universe of stantial error in the quantitative values. LCMS detects compounds
extractables for each detector system for a particular E&L screening based on their ability to ionize. The ionization process is well known
method. To estimate the universe of extractables, 217 extractable to vary widely based on sample chemistry with ionizability being
compounds were analyzed containing a wide range of chemical favored for those compounds containing heteroatoms with acidic
properties including volatiles, semi-volatiles and non-volatile com- or basic character and disfavored for those without these groups
pounds. Highly volatile organic compounds (HVOCs) (compounds [21]. Compounds which do not have heteroatoms or other ion-
that exert a vapor pressure greater than 80 mm Hg when measured izable groups show very weak response. This is the fundamental
at 20 ◦ C) have been excluded as these compounds are appropri- reason for the increased number of compounds at the extremes of
ate only for headspace GCMS and other specialized HVOC methods the LCMS RRF distributions. Studies in our laboratories have shown
and should be analyzed separately. The properties examined for that this problem is not unique to QTOF instruments but is also
inclusivity included molecular weight (93–1177 amu), logP val- observed for triple quadrupole systems due to the fact that they
ues (−2.38−25.156), boiling point (102->600 ◦ C) and pKa values rely on the same ionization processes.
(−0.83−18.9). The extractables standards had a wide range of func- In contrast, UV and CAD showed RRF distributions which are
tional groups including amines, acids, alcohols, phenols as well more symmetrical. While there was still a range of RRF values,
as sulfur, silicon and phosphorous containing species and many the distributions tended to be more even and to have a maximum
others. Table 2 shows a partial listing of some of the extractables closer to the center of the distribution. This has significant pos-
analyzed to provide a sense of the diversity of compounds analyzed. itive implications for reducing RF Problem 2: Quantitative Error
Fig. 2 shows the distribution of chemical properties (molecular as error is minimized when the majority of compounds have an
6 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

Fig. 2. Histograms showing the distribution of properties for the 217 extractables standards used to determine RF variation.

Table 1
Normalized relative response factors distribution statistics.

Detector LCMS Positive LCMS Negative UV 230 nm CAD GCMS FID

Average 1.0 1.0 1.0 1.0 1.0 1.0

RSD 108% 119% 60% 65% 52% 54%
Min 0.0013 0.0065 0.0364 0.0336 0.0318 0.0098
Max 6.1140 7.9816 2.8493 2.6467 2.8401 3.1218
Percent detected 72% 34% 52% 56% 59% 59%
% of compounds where RRF > .6 and < 1.4 33% 22% 44% 41% 61% 55%

RRF value closer to the center of the distribution (RRF = 1). In detector for many oligomeric and polymeric species. UV detection
comparison to LCMS, UV and CAD showed %RSD values of 60% is based on the absorption of UV light (in this case 230 nm) and
and 65% respectively, indicating that a much greater number of was found to be dependent upon the type and presence of chro-
compounds fall within a more reasonable range of quantitative mophores. Examples of strongly absorbing compounds included
accuracy. Thus, a significant improvement in quantitative accu- those containing aromatic structures such as hindered phenolics
racy would be expected simply by replacing LCMS with UV and (an important class of polymer antioxidants) while poorly respond-
CAD for relative quantitation even without optimized standard or ing compounds typically had a single weak chromophore such as
detector selection. UV and CAD detectors successfully detected 52% acrylic monomers which are better suited to GC detection.
and 56% of the extractable compounds respectively, and a total of GCMS and FID response showed very analogous RRF distribu-
74% of all compounds combined. This included 80% of the com- tions. While there was still a range of RRF values, the distributions
pounds detected by LCMS indicating a high degree of overlap. 44% were found to be the most Gaussian of all of the detectors with the
of the compounds detected by UV at 230 nm and 41% of the com- majority of compounds showing values closer to the center of the
pounds detected by CAD had an RRF value between 0.6–1.4 (±40% distribution (RRF = 1). This indicates that the majority of the com-
of the actual concentration) without standard optimization. As pre- pounds would be expected to show more reasonable quantitative
viously reported, CAD detection was found to correlate strongly accuracy reducing RF Problem 2: Quantitative Error. This data sup-
with compound boiling point, showing more consistent detection ports that GCMS or FID relative quantitation would be expected
for compounds with boiling points above 400 ◦ C and discrimina- to generally be more reliable than that for LCMS. GCMS and FID
tion against low boiling compounds. This makes CAD an excellent showed %RSD values of 51% and 53% respectively indicating that
 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334 7

Table 2
Partial list of the 217 extractables analyzed showing the diversity of compounds suitable for the method and the list of Surrogate Standard compounds. Normalized RRF
values for each detector are also listed.

Name Molecular formula CAS RRF LCMS Pos RRF LCMS Neg RRF UV 230 nm RRF CAD RRF GCMS RRF FID

Example extractable compounds

Bis(2-ethylhexyl) phthalate C24H38O4 117-81-7 1.921 ND 0.622 0.802 1.706 1.855
4-Chlorobenzene sulfonic acid ClC6H4SO3H 98-66-8 ND ND ND ND 0.081 0.034
Dibenzyl phosphate C14H15O4P 1623-08-1 0.134 2.266 ND 1.411 ND ND
Caprolactam C6H11NO 105-60-2 1.045 ND ND ND 0.658 1.200
1-naphthylamine C10H9N 134-32-7 0.580 ND 1.400 ND 0.918 0.248
Hexadecylamine C16H35N 143-27-1 1.514 ND ND 2.009 0.543 0.387
Tributylamine C12H27N 102-82-9 1.872 ND ND ND 0.708 1.614
5-Amino-1-pentanol C5H13NO 2508-29-4 0.126 ND ND 2.597 0.095 0.663
Dodecamethylcyclohexasiloxane C12H36O6Si6 540-97-6 ND ND ND ND 0.630 0.182
Butyl methacrylate C8H14O2 97-88-1 ND ND ND ND 0.692 1.088
Irganox 245 C34H50O8 36443-68-2 1.705 1.744 0.550 1.570 ND ND
Irganox 3114 C48H69N3O6 27676-62-6 0.513 ND 0.814 0.894 ND ND
Irganox 1076 C35H62O3 2082-79-3 1.210 ND 0.334 0.634 ND ND
Irganox 1035 C38H58O6S 41484-35-9 4.529 1.263 0.508 1.354 ND ND
Irganox 1024 C34H52N2O4 32687-78-8 0.157 2.225 0.750 1.647 ND ND
Tinuvin 320 C20H25N3O 3846-71-7 1.263 0.060 1.005 ND 2.126 1.287
Diphenyl phthalate C20H14O4 84-62-8 0.120 0.034 1.177 0.938 1.667 1.243
Bisphenol A C15H16O2 80-05-7 ND ND 1.655 1.822 2.579 1.449
Bisphenol A diglycidyl ether C21H24O4 1675-54-3 1.259 ND 1.371 1.451 ND ND
Octaethylene glycol C16H34O9 5117-19-1 1.763 ND ND 2.139 ND ND
Citric acid C6H8O7 77-92-9 0.058 0.212 ND 1.463 ND ND
1,3,5-Triphenylbenzene C24H18 612-71-5 0.001 ND 1.976 0.819 ND ND
LCMS surrogate standards
1-octane sulfonic acid C8H18O3S 5324-84-5 ND 3.164 ND 1.639 ND ND
Irganox 245 C34H50O8 36443-68-2 1.695 1.744 0.551 1.570 ND ND
Phthalic acid C8H6O4 88-99-3 0.047 1.061 1.421 1.005 0.532 0.726
Glyceryl trioleate C57H104O6 122-32-7 0.438 ND ND 0.185 ND ND
Oleamide C18H35NO 301-02-0 1.414 ND ND 2.680 1.504 0.779
Diisobutyl phthalate C16H22O4 84-69-5 0.908 ND 0.853 ND 0.837 1.423
Diphenyl phthalate C20H14O4 84-62-8 0.119 0.033 1.172 0.931 1.667 1.243
4,4 -Dichlorodiphenyl sulfone (DCDPS) C12H8Cl2O2S 80-07-9 0.003 ND 1.402 0.149 1.398 1.158
Irgacure 184 C13H16O2 947-19-3 0.019 ND 1.163 ND 1.045 1.920
Diethylene glycol cyclic phthalate (Howflex Gbp) C12H12O5 13988-26-6 0.458 ND 0.947 ND 1.173 1.311
GCMS surrogate standards
Benzaldehyde C7H6O 100-52-7 ND ND 0.986 ND 0.815 1.758
Decane C10H22 124-18-5 ND ND ND ND 0.741 2.502
Decamethylcyclopentasiloxane (D5) C10H30O5Si5 541-02-6 ND ND ND ND 0.770 0.456
4-Methoxyphenol C7H8O2 150-76-5 ND ND 0.815 ND 1.092 1.802
Butylated hydroxytoluene (BHT) C15H24O 128-37-0 ND ND 0.584 ND 1.222 1.606
Hexadecane C16H32 544-76-3 ND ND ND ND 0.875 2.299
Irgacure 184 C13H16O2 947-19-3 0.019 ND 1.163 ND 1.045 1.920
Diisobutyl phthalate C16H22O4 84-69-5 0.908 ND 0.853 ND 0.837 1.423
Eicosane C20H42 112-95-8 ND ND ND ND 1.227 3.154
Tetracosane C24H50 646-31-1 ND ND ND ND 1.249 1.812

ND = None Detected.

a much greater number of compounds fall within a more reason- 3.2. Multidetection as a solution to RF problem 1: AET
able range of quantitative accuracy as compared to LCMS (%RSD of underreporting
107% and 119%). This is in line with the %RSD values determined in
other studies (50.9% for GCMS and 44% for FID) [17]. Thus, when a One of the most important aspects of E&L study quality is the
compound is detected by both LCMS and GCMS, the GCMS values decision as to which peaks should be identified and reported in
would generally be more reliable in terms of quantitative accuracy the study because they are at or above the toxicologically relevant
even without optimized standard selection. 61% of the compounds level. The analytical evaluation threshold (AET) approach is com-
detected by GCMS and 56% of the compounds detected by FID monly applied to relate the toxicologically relevant threshold to
had an RRF value between .6–1.4 (±40% of the actual concentra- the observed analytical responses. For a compound to be identified
tion) without standard optimization. The GCMS-FID distributions and reported in the study, two criteria must be met. First, the com-
are modestly narrower than the UV and CAD distributions (%RSD pound must be detected (give a measurable response on at least
51–53% as compared to 61–66% for UV/CAD). GCMS and FID both one detector) and second, the compounds must have an RRF factor
successfully detected 60% of the extractable compounds. It was fur- sufficiently strong that it equals or exceeds the threshold set using
ther observed that 50% of the extractables were detected by both the surrogate standard. In our previous publication, we noted that
QTOF-LCMS-UV-CAD and GCMS-FID indicating a high degree of the number of compounds determined to be at the toxicologically
overlap between methods. This allows useful opportunities for ver- relevant level could be dramatically affected by the choice of surro-
ification of compound identity and further protection against under gate standard [11]. It is important to understand that the surrogate
reporting of compound concentration. Finally, an indication of the standard response defines the location of the AET threshold (see
sufficiency of the data contained in this publication to estimate the Fig. 1). If a surrogate standard with an RRF of 1 is selected, then any
universe of extractables (HVOCs excluded) was noted in the fact compound with an RRF equal or greater than 1 would need to be
that the RSD of the distributions changed by less than 15% for the identified and reported. In setting the AET threshold value, it is pre-
addition of the last 67 compounds. ferred that the surrogate standard selected would be at the center
8 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

Fig. 3. Histograms showing the distribution of RRF Values for 6 different detectors for 217 extractables.

of the distribution (RRF = 1 for a normalized RRF distribution). This dard with an RRF of 0.5 is selected this is equivalent to using a UF
allows the application of any UF factors to be applied starting at the of 2 with a surrogate standard with an RRF of 1 (RRF/UF = ½ = .5).
center of the distribution. If a standard was selected which had an Even if an appropriate surrogate standard or standards are
RRF value less than 1, then an additional measure of conservatism selected, it is an unfortunate fact that no single detector system
would be added to the study (more inclusive) and conversely a can detect all compounds. However, it is not necessary that every
value greater than one builds into the study design less conser- compound is detected by every detector in order to obtain complete
vatism (less protective of safety). Surrogate standard selection and coverage of all extractables. When using the multidetector approach,
the UF factor work together in tandem such that if a surrogate stan- it is only necessary that a compound show sufficient signal on at
 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334 9

least one detector. For instance, in a study conducted using only then depends on how many detectors can be applied at the AET
GCMS and LCMS, it is only necessary that the compound be detected level. In our laboratories, it has been observed that CAD detection
by either GCMS or LCMS to be adequately accounted for in the is not as sensitive as LCMS or UV. The results of this study indi-
toxicological evaluation. It is important to understand that for the cate that application of a QTOF-LCMS-UV and GCMS-FID utilizing
multidetector approach described herein, the signal intensity on any a UF of 2 results in 91% coverage. This remains a high level of cer-
one detector is not correlated to the signal observed by any other tainty especially when one considers the difficulty of detecting all
detector. This is because each detector operates based on differ- unknown species at low or sub ppb levels and the fact that CAD
ent fundamental principles. As described above, the combination of detection at a limit of detection (LOD) greater than the AET may
QTOF-LCMS-UV-CAD with GCMS-FID (multidetector approach) pro- still catch some remaining compounds. In considering the negative
vides five distinct mechanisms of detection (Ionizability for LCMS, potential consequences of extensive sample concentration (>10X),
light absorption for UV, charged particle detection for CAD, elec- it seems advisable that 91% coverage may be the most protective
tron ionization (EI) for GCMS and charged ion detection for Polyarc option from a safety perspective even when the CAD LOD cannot
FID). To be included in the study, a compound only needs to 1) be reach the AET.
detected by any one of these five mechanisms (one of the five detec- One potential draw back of the multidetector approach is the
tors) and 2) give a signal which is greater than the AET threshold potential for overreporting. It is important to note that inclusion
value on any one detector. of a compound in the study at this stage does not mean that the
To confirm the sufficiency of the multidetector approach to compound is actually at the toxicologically relevant level. It is pos-
mitigate RF Problem 1: AET Underreporting, Fig. 4 shows a plot sible that a compound has an RRF greater than 1 on a particular
comparing the maximum RRF value for each compound using detector (compound is overestimated). If this occurs, the actual
either just GCMS and LCMS or using the multidetector strategy concentration may be below the toxicologically relevant level but
(QTOF-LCMS-UV-CAD and GCMS-FID). Each dot on the figure repre- the compound would be considered to be above the threshold (false
sents the highest RRF value for a given compound after comparing positive). While this consequence may be less severe than that for
all detectors. All compounds represented by a value greater than underreporting, the increased cost and time to manufacturers can
1 would be included in the study using a UF of 1. If a UF of 2 was also have negative implications for patients. For this reason, we
applied, then all compounds with an RRF of greater than 0.5 would would recommend that a 2nd AET evaluation be performed fol-
be included. Additional UF values can be considered by dividing the lowing compound quantitation. Inclusion in the study at this early
RRF of 1 by the desired UF value. An examination of the data indi- stage only means that a compound should be subjected to addi-
cates that 54% of the compounds would be determined to be above tional scrutiny as to its actual identity and concentration. Following
the toxicologically relevant threshold using a UF of 1 with only quantitation as described below, a final decision will need to be
LCMS and GCMS while 76% would be included using the multidetec- made regarding if the compound should be included in the study
tor approach. If a UF of 2 is applied then this improves significantly based on the resulting quantitative values. This 2nd AET evaluation
to 85% for the combination of LCMS and GCMS and 94% for the is made at a stage in the analysis when more information is avail-
multidetector approach. Applying a UF of 10 results in 95% cover- able which allows for optimum surrogate standard and detector
age for LCMS and GCMS and 99% for the multidetector approach. selection and thus is based on better information.
Thus, using the multidetector approach with a UF of 2 was shown Regardless of the strategy applied for AET verification, it is
to be equally protective (1% difference) as a UF of 10 when using important to stress the necessity of verifying the limit of detec-
LCMS and GCMS only. While both strategies may be equally inclu- tion (LOD) at the AET threshold at the time of sample analysis. The
sive, the multidetector strategy has several advantages over the use authors have frequently seen studies for which method sensitivity
of a larger UF factor (UF > 4). First, the multidetector approach can was not demonstrated but rather assumed. This is a crucial aspect
reduce underreporting due to signal overlap. This is due to the fact of study quality without which there are no assurances that the tox-
that it is much less likely that an interfering signal would cover the icologically relevant compounds are being detected. Those skilled
peak for a given extractable on multiple detectors simultaneously. in the art know that method LOD values change daily based on tran-
Second, instrument sensitivity is not infinite. Thus, if the AET calcu- sient conditions such as instrument cleanliness (LCMS ion sources
lation uses a larger UF value, then it frequently becomes necessary being particularly vulnerable) and thus verification as a part of sys-
to use extensive sample concentration (>10X). Extensive sample tem suitability is essential. Ideally, this should be accomplished by
concentration has many potential negative consequences including analyzing a group of standards at the AET concentration with a
potential compound loss or degradation, increased matrix inter- range of RRF values whose normalized average approaches one and
ferences which could cause underreporting (the very thing we which are representative of the RRF distribution obtained using a
are attempting to avoid), the requirement for a larger amount of method specific E&L database. This proves that the method has the
original extract (i.e. more devices or packaging must be extracted needed level of sensitivity in order to detect those compounds at
increasing cost and study time) and the potential for increased false or above AET and to cover the use of UF factors. This brings us to
positives due to contamination or concentration of background one final advantage of the multidetector approach which is that UV,
impurities. In some instances, it is also impossible to sufficiently CAD and FID detectors are generally more rugged than MS and are
concentrate the sample to reach the AET when using a large UF less susceptible to loss of sensitivity throughout the analysis due to
value (UF > 4). This is especially common for salt containing samples the high matrix backgrounds often observed in E&L analyses.
where concentration factors are limited due to salt solubility. While
small amounts of sample concentration (<10X) are often necessary 3.3. Multidetection and standard selection as solutions to RF
in order to increase method sensitivity, excessive concentration problem 2: Quantitative error
should be avoided where possible. The multidetector approach pro-
vides a means by which to obtain sufficient assurance of detection A second key aspect of E&L study quality is quantitative accu-
without resorting to large UF values and the corresponding large racy. The accuracy of surrogate standard quantitation is directly
concentration factors. related to the difference in the RRF values for the surrogate stan-
It is important to note that for some analyses, the AET thresh- dards and the extractable compounds. Thus, if an extractable
old may be so low (low ng/mL levels) that method sensitivity on compound has an RRF of 0.2 and it is quantitated using a compound
an individual detector may not be sufficient and a full multidetec- with an RRF of 1, the compounds reported concentration will only
tor approach may not be feasible. The inclusiveness of the strategy be 20% of the actual value. Similarly, if a compound with an RRF of
10 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

Fig. 4. Multidetector solution for RF Problem 1: AET Underreporting.

2 is quantitated using a surrogate standard with an RRF of 1 then its any optimization of surrogate standard selection and excluding the
reported concentration will be 200% of the actual value. The mag- LCMS detector.
nitude of this error for the estimated universe of extractables can In the LCMS/GCMS strategy, 6 compounds were found to have no
be seen by considering the data in Fig. 3 and in Table 1 where it is signal from which to perform the quantitation. In the multidetector
shown that errors exceeding 40% are the most common outcome approach 5 compounds showed no signal on UV-CAD or GCMS-FID.
on all but the GCMS detector. Even on the GCMS detector, 39% of However, there is one important distinction, in the case of the mul-
the compounds are poorly estimated (>40% error). There is cur- tidetector approach; the compounds which were missed on UV-CAD
rently no mechanism in the toxicological risk assessment process or GCMS-FID did respond in LCMS. Thus, the data at least exists from
(margin of safety (MOS) calculation) to account for this error and which an analyst could choose to perform LCMS formal quantita-
some discussion of the addition of an additional uncertainty fac- tion for those compounds. In the case of the LCMS/GCMS strategy,
tor designed to account for analytical uncertainty has occurred at no data exists to indicate that 6 compounds were missed. In exam-
national conferences. Those who are frequent practitioners in E&L ining the nature of these compounds, it was observed that for the
analysis know that this would have serious negative consequences LCMS/GCMS strategy these compounds were primarily higher Mw
for the toxicological process as many studies have compounds compounds (average Mw 527 amu) with few or poorly ionizing
with MOS values close to one and these products would poten- functional groups. It has frequently been observed in our labora-
tially need to be unnecessarily rejected if an additional UF factor tories that many E&L extracts contain substantial oligomer peaks
was added to account for this analytical uncertainty. It is therefore (500–2000 amu) which show no signal at all in LCMS or GCMS
highly desirable that a method for improved quantitative accuracy but are detectable by CAD. In many cases, these oligomers are one
be identified. of the most abundant species in the extract but they are missed
Based on the data in Fig. 3 and Table 1 it is clear that no sin- using the LCMS/GCMS approach. In the case of the multidetector
gle detector has a sufficiently universal response to provide highly approach, 2 of the 5 compounds which showed no quantitative
accurate quantitation for the universe of extractables. Fortunately, signal were highly volatile (Hexamethylcyclotrisiloxane and 3-(3-
two mechanisms are available to improve quantitative accuracy acetoxypropyl)heptamethyltrisiloxane) and may be better suited
using the multidetector approach: 1) proper detector selection and to HGCMS. In the remaining 3 of 5 cases, they were generally highly
2) proper surrogate standard selection. Since the response on each polar volatile acids (leucic acid and heptafluorobutyric acid). This
detector is independent of all other detectors, proper selection of is not to imply that the method is not good for acidic species (pKa
the detector (detector which produces an RRF closest to 1) can be < 5) as 42 of 44 acidic species tested were quantifiable.
used as a means to significantly improve quantitative accuracy. Another strategy which can be used to improve quantitative
Fig. 5 shows a comparison of the quantitative accuracy which can accuracy is appropriate surrogate standard selection. If a surrogate
be obtained using optimized detector selection for the combina- standard can be selected which has an RRF value close or equal to
tion of GCMS and LCMS as compared to the multidetector approach that of the extractable compound, then quantitative error can be
using (UV-CAD and GCMS-FID). We intentionally omit the LCMS minimized or eliminated. In the current study, 10 LC and GC surro-
detector in the multidetector approach as the RF variability in this gate standards were analyzed as shown in the bottom of Table 2.
detector is so large and the difficulty in determining proper surro- These standards were selected to provide a diversity of chemistry
gate standard selection so great (see discussion below) that we do and because they elute at a range of retention times allowing the
not feel this detector can be applied confidently for relative quan- creation of calibration curves for each surrogate standard. The rea-
titation. In the LCMS/GCMS approach, only 33% of the compounds son that 10 surrogate standards were selected was to provide a
are within ± 20% of the actual value and only 55% are within ± minimum of four compounds which respond by each detector. The
40% even using the optimum detector. In contrast, using the mul- dataset was then examined by choosing the standard with the RRF
tidetector approach, 40% of compounds are within ± 20% of the value closest to that of the extractable to serve as its surrogate stan-
accurate value and 72% are within ± 40%. In the LCMS/GCMS strat- dard (RRF of surrogate standard closest to RRF of extractable) and
egy, 12% of the compounds are significantly underreported (<40% then further optimized using the detector which resulted in the RRF
of the true value) while only 7% are underreported using the multi- value closest to 1.0. The results of this analysis are shown in Fig. 6.
detector approach. These improvements are accomplished without It was found that this approach resulted in 85% of the compounds
 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334 11

Fig. 5. Comparison of quantitative accuracy for GCMS/LCMS versus the multidetector approach using only optimized detector selection.

Fig. 6. Quantitative accuracy using the multidetector approach with optimized surrogate standard and detector selection.

being within ±20% of the accurate value and 91% being within 40%. than the LCMS detector. Based on this one consideration alone, the
This is a clear demonstration that the multidetector approach con- use of any detector other than LCMS is strongly preferred for rela-
tains the data necessary to produce accurate quantitation with the tive quantitation. Also, a comparison of different detector responses
addition of a reasonable number of surrogate standards. Accuracy (UV vs CAD vs GCMS vs FID) can be used to assess if one detector
which is generally within ±40% is in keeping with published regu- might be strongly underestimating. Even more importantly, detec-
latory expectations for food migration studies (a related field) using tor and surrogate standard selection can be reasonably guided by an
formal quantitation [30]. Given the inherent difficulties associated assessment of the chemistry of the extractable following identifica-
with surrogate standard quantitation and considering the RSD for tion. Selection of an appropriate surrogate standard for UV and CAD
even less variable methods such as GCMS, an accuracy result within can be guided by relatively simple principles which can be readily
50–200% would be a reasonable expectation. The ability of the mul- determined from the elemental formula of the extractable obtained
tidetector approach to exceed this performance standard supports from accurate mass data. For instance, the number of double bond
its adoption as an effective analytical practice. In this analysis, the equivalents (DBE) is generally an excellent predictor of the reli-
percentage of extractables for which each detector provided an ability of UV relative quantitation while molecular weight can be
optimum response (listed in the order UV, CAD, GCMS, FID) was used as a reasonable predictor of CAD response. GCMS or FID values
22.6, 28.3, 33.5 and 15.6% respectively demonstrating the need for are usually calculated and reported separately from UV and CAD,
all four detectors to obtain optimum quantitative accuracy in the providing a separate concentration value increasing the certainty
multidetector approach. that any safety assessment is not underestimating the concentra-
One potential critique of this approach might be to question the tion of an individual extractable. Thus, it is not necessary to achieve
ability of an analyst to choose the optimum detector or surrogate the optimum detector or standard selection in every case to sub-
standard. While this critique is not entirely unmerited, it is impor- stantially improve the quantitative accuracy of the extractables
tant to remember that the RRF distributions for UV, CAD, GCMS and analysis.
FID are narrower than that for LCMS. This reduces the burden on Finally, it may be asked if LCMS can be applied effectively so long
the analyst to make a perfect detector selection. The UV, CAD, GCMS as sufficient surrogate standards are available. Unfortunately, it is
and FID RRF distributions all showed sufficient quantitative accu- our experience that LCMS produces widely different RRF values for
racy (±40% accuracy) for a higher percentage of the compounds seemingly very similar compound chemistry. For instance, consider
12 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

the case of Bisphenol A diglycidyl ether quantified against bisphe- dors of E&L services have reported having databases greater than
nol A as a surrogate standard. Bisphenol A is observed to ionize only 2500 compounds [25]. While it is currently unknown how many
in negative mode while bisphenol A diglycidyl ether ionizes only compounds constitute the universe of all extractables, discussions
in positive mode making relative quantitation impossible. As a sec- at national meetings have frequently proposed numbers ranging
ond example, consider the relative quantitation of 5 very similarly from 10,000–100,000 or potentially more. In our own laborato-
structured hindered phenolic antioxidants (Irganox 245, Irganox ries, we have identified and added more than 4000 extractables
3114, Irganox 1076, Irganox 1035 and Irganox 1024) whose RRF into a proprietary database but are daily detecting many new com-
data is shown in Table 2. All of these compounds are hindered phe- pounds which were as yet previously unencountered. This problem
nols of relatively high monoisotopic mass (586, 783, 530, 642 and is further exacerbated by a lack of commercially available reference
552 amu), which elute at similar retention times (5.06, 5.96, 6.26, standards. Many of the most frequently encountered extracta-
5.78 and 5.31 min) and which have reasonably similar elemental bles are degradation products of antioxidants or oligomeric side
formulas (C34H50O8, C48H69N3O6, C35H62O3, C38H58O6S and products of the polymerization process. There are generally no
C34H52N2O4). Based on their structure, it would seem logical to commercial sources from which to purchase these standards. In
propose that LCMS negative mode should be used for relative quan- addition, as new polymer systems and additives are being created
titation and that any one of these compounds would be a reasonable the database approach would lag behind needing to catch up to
surrogate standard for all the others. However, two of these com- current technology. For these reasons, the RRF database approach
pounds gave no response at all in LCMS negative mode. The RSD of seems to leave very significant gaps given that the most extensive
the RRF values in LCMS positive and negative mode was 106% and databases (including our own) likely contain far less than 25% of the
96% respectively. This is in spite of the fact that they have very sim- extractables universe and do not include many frequently encoun-
ilar chemistry and the same ionizable functional group (phenolic tered extractables due to their not being commercially available.
compounds). These seemingly random RRF values make intelligent For these reasons, it is our opinion that the chief utility of an E&L
and rational surrogate standard selection in LCMS very difficult and RRF database is to define the RF distribution of the method to aid
in our opinion, impractical. Comparing that to CAD and UV, the RSD in standard selection, facilitate identification of compounds and to
of the RRF values was only 36% and 32%. This level of variability provide accurate quantitation for a subset of the most commonly
allows for reasonable levels of confidence when selecting surrogate observed commercially available extractables. Qualification of each
standards. E&L method should demonstrate that quantitative accuracy can be
obtained not only for compounds contained in the database but
3.4. RF Databases and the multidetector strategy also for compounds not in the database using relative quantita-
tion with surrogate standards. This should be done using surrogate
As a final point of discussion, a number of recent publications standard quantitation for a significant number of compounds not
have proposed to use an RF database as a solution to both RF in the database with varied chemical properties with reasonable
Problem 1: Underreporting and RF Problem 2: Quantitative Error quantitative accuracy (the authors would suggest ±40% for >90% of
[24,25]. In regards to RF Problem 1: Underreporting, it is acknowl- compounds) and reasonably comprehensive compound coverage
edged that having a comprehensive database would allow for (the authors would suggest >90%). This proves that the method can
determination of a UF value which would be sufficiently protec- successfully handle compounds which are not in the database. This
tive to avoid under reporting. However, this approach does nothing is an essential aspect of insuring method quality since the majority
to reduce RF variation and hence the underlying need for a UF of E&L compounds are not commercially available as standards.
value or the uncertainty which it is designed to mitigate. UF val- The multidetector approach can be applied broadly and without
ues provide an added measure of conservatism by reducing the the requirement for a reference standard of each compound as it
corresponding AET. For large UF values, the resulting AET may reduces RF variation.
extend below the instrumental detection limit. In order to reach
the reduced AET, it is frequently necessary to perform additional
sample concentration (>10X) which increases the risk of compound 4. Conclusions
loss and increases matrix effects which can potentially conceal
extractables and leachables and complicates the process of iden- The accurate quantification of extractables and leachables
tification. A larger UF also increases the risk of over reporting presents a significant analytical challenge due to the large diver-
of compounds which are actually below the threshold but which sity of extractable compounds and the lack of a universal detector.
have higher RF values. Some authors have suggested that the cre- The generally applied strategy of using LCMS and GCMS for rela-
ation of an associative database could be used to “provide alerts tive quantitation has significant, difficult to resolve issues due to
to potential omissions [25]. “Presumably this would then allow the wide RRF variation inherent primarily in the LCMS results. The
for additional targeted screening to be performed specifically for magnitude of this error was estimated for the universe of extracta-
the associated extractables. While the potential of this approach bles (HVOCs excluded) using 217 extractables and it was found
seems valuable, the authors also acknowledge that “the power that only 33% of the compounds were within ± 20% of the actual
of the database is derived from the number of substances in the value and only 55% were within ± 40%. Correction of this error
database” and thus this approach is not helpful for catching those for LCMS analysis using optimized surrogate standard selection
compounds which are not present in the database or which are also did not show readily predictable, logical trends based on the
not correctly associated. It further requires that additional targeted chemistry of the extractables. In contrast, the use of the multidetec-
work be performed. In contrast, the multidetector approach requires tor approach (proper detector selection) with optimized surrogate
no such prior knowledge and actually serves to reduce the need standard selection (using 10 surrogate standards) was shown to
for a UF by reducing RF variation. This prevents under reporting produce accurate values within ±20% for 85% of the compounds
without the potential for negative consequences associated with and 91% of compounds were within ±40% substantially reducing RF
increased sample concentration. Similarly, the ability of an RRF Problem 2: Quantitative Error. It was further demonstrated that the
database to improve Problem 2: Quantitative Error is based on the multidetector approach significantly enhances the assurance that
premise that the extractable under study is in the database. The compounds are not missed during the initial discovery phase of
largest commercial databases currently available include on the the analysis thus reducing RF Problem 1: Underreporting. This is
order of 1500 compounds (Agilent E&L Database) and some ven- accomplished not through the use of UF values but through reduc-
 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334 13

ing RF variation and through the synergies between detectors. As References

a step towards improving quality assurance in E&L analyses, it is
recommended that the sensitivity of E&L methods be verified at [1] International Conference on Harmonisation, Q2(R1) Validation of Analytical
Procedures: Text and Methodology, Retrieved from
the time of the analysis utilizing a group of surrogate standards at [Link]
the AET concentration whose average RRF approaches the center of validation-analytical-procedures-text-methodology-step-5 [Link] (last
the method’s RRF distribution (RRF ≈ 1 for normalized values) and Accessed 1 January, 2020), 2020.
[2] Food and Drug Administration, Guidance for Industry: Q2B Validation of
which also contains a standard with an RRF value approximately Analytical Procedures: Methodology, Retrieved from
equal to one divided by the UF (1/UF). This proves adequate method [Link] (last Accessed 1 January, 2020),
sensitivity. It is further recommended that the distribution of RRF 1996.
[3] United States Pharmacopeia, USP <661.2> Plastic Packaging Systems for
values for the surrogate standards should be sufficiently broad to Pharmaceutical Use, Retrieved from [Link]
allow for correction of RF Problem 2: Quantitative Error. To identify default/files/usp pdf/EN/USPNF/revisions/661.2 rb [Link] (last Accessed 1
appropriate standards, each laboratory should develop a method January, 2020), 2020.
[4] USP 40 General Chapter Assessment of Extractables Associated With
specific RF database, the primary purpose of which is to aid in iden-
Pharmaceutical Packaging/Delivery Systems <1663>, United States
tification, improve quantitative accuracy for frequently observed Pharmacopeial Convention, Rockville, MD, 2017, pp. 2020–2035.
extractables and leachables and confirm that the RRF values for the [5] USP 40 General Chapter Assessment of Drug Product Leachables Associated
standards are at appropriate locations in the distribution. The use With Pharmaceutical Packaging/Delivery Systems. <1664>, United States
Pharmacopeial Convention, Rockville, MD, 2017, pp. 2035–2047.
of an RF database does not reduce the underlying problem of RF [6] International Organization for Standardization, Biological Evaluation of
variation and as a solution for RF Problem 1; AET Underreport- Medical Devices, Part 12: Sample Preparation and Reference Materials,
ing and RF Problem 2; Quantitative Error has limited scope. RF 10993-12: Retrieved from [Link] (last
Accessed 1 January, 2020), 2012.
databases allow calculation of a method specific UF but do noth- [7] International Organization for Standardization, Biological Evaluation of
ing to reduce the underlying method uncertainty and the resulting Medical Devices, Part 17: Establishment of Allowable Limits for Leachable
lowered AET values can require additional sample concentration Substances, 10993-17: Retrieved from
[Link] (last Accessed 1 January, 2020),
which can lead to compound loss and increased matrix effects. 2002.
Similarly, quantitative error is reduced only for those compounds [8] International Organization for Standardization, Biological Evaluation of
in the database. The RF database approach is in essence a targeted Medical Devices, Part 18: Chemical Characterization of Medical Device
Materials Within a Risk Management Process, 10993-18: Retrieved from
quantitative approach applied to non-targeted analysis. In contrast, [Link] (last Accessed 1 January, 2020),
the multidetector approach is at its core a non-targeted analytical 2020.
approach for a non-targeted analytical problem (extractables and [9] International Conference on Harmonisation, M7(R1) Assessment and Control
of DNA Reactive (Mutagenic) Impurities in Pharmaceuticals To Limit Potential
leachables screening). This study has demonstrated that the mul-
Carcinogenic Risk: Guidance for Industry, Retrieved from
tidetector approach with a UF of 2 produced 94% coverage for all [Link] (last Accessed 1 January, 2020),
tested extractables and was equally protective (1% difference) as a 2020.
UF of 10 when using LCMS and GCMS only. [10] D. Norwood, L. Nagao, The Analytical Evaluation Threshold (AET), Product
Quality Research Institute, 2015, Retrieved from
[Link] AET [Link] (last
Funding Accessed 9 June, 2017).
[11] M. Jordi, S. Khera, K. Rowland, L. Jiang, P. Solomon, J. Nelson, S. Lateef, J.
Woods, S. Martin, F. Aiello, N. Chen, Qualitative assessment of extractables
This research did not receive any specific grant from funding from single-use components and the impact of reference standard selection, J.
agencies in the public, commercial, or not-for-profit sectors. Pharm. Biomed. Anal. 150 (2018) 368–376.
[12] D. Jenke, Moving forward towards standardized analytical methods for
extractables and leachables profiling studies, PDA J. of Pharm. Sci. Technol. 69
Authors statement (2015) 471–476.
[13] Y. Gao, N. Allison, Extractables and leachables issues with the application of
single use technology in the biopharmaceutical industry, J. Chem. Technol.
The Relative response factor data used to generate this publi- Biotechnol. 91 (2016) 289–295.
cation is part of a proprietary database developed and maintained [14] K. Li, G. Rogers, Y. Nashed-Samuel, H. Lee, A. Mire-Sluis, B. Cherney, R. Forster,
by Jordi Labs. The authors are happy to answer questions related to P. Yeh, I. Markovic, Creating a holistic extractables and leachables (E&L)
program for biotechnology products, PDA J. Pharm. Sci. Technol. 69 (2015)
this data but must maintain the proprietary nature of the database 590–619.
due to business interests. [15] L. Marghitoiu, J. Liu, H. Lee, L. Perez, K. Fujimori, M. Ronk, M. Hammond, H.
Nunn, A. Lower, G. Rogers, Extractables analysis of single-use flexible plastic
biocontainers, PDA J. Pharm. Sci. Technol. 69 (2015) 49–58.
Declaration of Competing Interest [16] M. Hammond, H. Nunn, G. Rogers, H. Lee, A. Marghitoiu, L. Perez, Y.
Nashed-Samuel, C. Anderson, M. Vandiver, S. Kline, Identification of a
The authors declare that they have no known competing finan- leachable compound detrimental to cell growth in single-use bioprocess
containers, PDA J. Pharm. Sci. and Technol. 67 (2013) 123–134.
cial interests or personal relationships that could have appeared to [17] D. Jenke, A. Odufu, Utilization of internal standard response factors to
influence the work reported in this paper. estimate the concentration of organic compounds leached from
pharmaceutical packaging systems and application of such estimated
concentrations to safety assessment, J. Chromatogr. Sci. 50 (2012) 206–212.
Acknowledgment [18] M. Iqbal, K. Rao, G. Sridhar, P. Raju, G. Deshpande, J. Babu, Characterization
and relative response factor determination of process related impurity in
naproxen by nuclear magnetic resonance spectroscopy, J. Pharm. Biomed.
I would like to acknowledge the dedicated staff at Jordi Labs
Anal. 56 (2011) 484–490.
who tirelessly supported this study. We wish to recognize the con- [19] Guide to achieving reliable quantitative LC-MS measurements, in: M. Sargent,
tributions of the following people; Xiaoqing Kong, Jordan Tocher, A. Contributors Sage, C. Wolff, C. Mussell, D. Neville, G. Lord, M. Saeed, R. Lad,
Nisansala Hewage, Hongda Cao, Pravin Pathak, Xunzhe Cui and R. Godfrey, S. Hird, V. Barwick (Eds.), RSC Analytical Methods Committee, first
edition, 2013, ISBN 978-0-948926-27-3.
Lindsay Ross [20] E. Adams, M. Rafiee, E. Roets, J. Hoogmartens, Liquid chromatographic
analysis of streptomycin sulfate, J. Pharm. Biomed. Anal. 24 (2000) 219–226.
[21] J. Blanz, G. Williams, J. Dayer, T. Délémonté, W. Gertsch, P. Ramstein, R.
Appendix A. Supplementary data Aichholz, M. Trunzer, D. Pearson, Evaluation of relative MS response factors of
drug metabolites for semi-quantitative assessment of chemical liabilities in
Supplementary material related to this article can be found, drug discovery, J. Mass Spectrom. 52 (2017) 210–217.
[22] P. Sun, X. Wang, L. Alquier, C. Martanoff, Determination of relative response
in the online version, at doi:[Link]
factors of impurities in paclitaxel with high performance liquid
113334.
14 M.A. Jordi, K. Rowland, W. Liu et al. / Journal of Pharmaceutical and Biomedical Analysis 186 (2020) 113334

chromatography equipped with ultraviolet and charged aerosol detectors, J. [28] P. Hong, A. Phoebe, M. Jones, Study of relative response factors and mass
Chromatogr. A 1177 (2008) 87–91. balance in forced degradation studies with liquid
[23] S. Baertschi, Analytical methodologies for discovering and profiling chromatography/photo-diode array detector/evaporative light scattering
degradation-related impurities, Trends Anal. Chem. 25 (2006) 758–767. detector/mass spectrometry system, J. Chromatogr. A 1512 (2017) 61–70.
[24] M. Rozio, A. Rosato, L. Iadarola, S. Carrara, Compounds Relative Response [29] S. Gustavsson, J. Samskog, B. Langstrong, K. Markides, B. Långström, Study of
Factor, a Reliable Quantification Within Extractables Testing, White Paper, signal suppression in liquid chromatography-electrospray ionization mass
Eurofins Biopharma Product Testing Italy, 2019, December. spectrometry using volatile ion-pairing reagents, J. Chromatogr. A. 937 (2001)
[25] P. Christiaens, J. Beusen, P. Verlinde, J. Baeten, D. Jenke, Identifying and 41–47.
mitigating errors in screening for organic extractables and leachables: part 1; [30] Guidance for Industry, Preparation of Premarket Submissions for Food
Introduction to errors in chromatographic screening for organic extractables Contact Substances: Chemistry Recommendations, Retrieved from
and leachables and discussion of the error of omission, PDA J. Pharm. Sci. [Link]
Technol. 74 (2020) 90–107. documents/guidance-industry-preparation-premarket-submissions-food-
[26] O. Wahl, U. Holzgrabe, Impurity profiling of ibandronate sodium by contact-substances-chemistry (last Accessed 9 March, 2020),
HPLC–CAD, J. Pharm. Biomed. Anal. 114 (2015) 254–264. 2020.
[27] M. Nussbaum, S. Baertschi, P. Jansen, Determination of relative UV response
factors for HPLC by use of a chemiluminescent nitrogen-specific detector, J.
Pharm. Biomed. Anal. 27 (2002) 983–993.