Statistical Analysis of Data from Limiting Dilution Cloning to Assess

Monoclonality in Generating Manufacturing Cell Lines

Jorge Quiroz
BARDS, Research CMC Statistics, Merck Research Laboratories, Kenilworth, NJ 07033, USA

Yung-Shyeng Tsao
Department of Bioprocess Technology Expression, Biologics and Vaccines, Merck Research Laboratories, Kenilworth, NJ 07033, USA

DOI 10.1002/btpr.2290
Published online May 17, 2016 in Wiley Online Library (wileyonlinelibrary.com)

Assurance of monoclonality of recombinant cell lines is a critical issue to gain regulatory

approval in biological license application (BLA). Some of the requirements of regulatory
agencies are the use of proper documentations and appropriate statistical analysis to dem-
onstrate monoclonality. In some cases, one round may be sufficient to demonstrate mono-
clonality. In this article, we propose the use of confidence intervals for assessing
monoclonality for limiting dilution cloning in the generation of recombinant manufacturing
cell lines based on a single round. The use of confidence intervals instead of point estimates
allow practitioners to account for the uncertainty present in the data when assessing
whether an estimated level of monoclonality is consistent with regulatory requirements. In
other cases, one round may not be sufficient and two consecutive rounds are required to
assess monoclonality. When two consecutive subclonings are required, we improved the
present methodology by reducing the infinite series proposed by Coller and Coller (Hybrid-
oma 1983;2:91–96) to a simpler series. The proposed simpler series provides more accurate
and reliable results. It also reduces the level of computation and can be easily implemented
in any spreadsheet program like Microsoft Excel. V C 2016 American Institute of Chemical

Engineers Biotechnol. Prog., 32:1061–1068, 2016

Keywords: limiting dilution cloning, manufacturing cell lines, monoclonality in one round of
subcloning, monoclonality in two consecutive subclonings, Poisson distribution, recombinant
protein production

Introduction Xpress are three current technologies with the potential to

Development of new production cell lines is essential in
the manufacturing of biologics. Regulatory agencies require
that manufacturing recombinant cell lines be monoclonal.1
This means that transfected cells containing the desired
sequences and its corresponding Master Cell Bank (MCB)
should be derived from a single cell progenitor.2 Thus, assur-
ance of monoclonality of recombinant cell lines is a critical
issue to gain regulatory approval in biological license appli-
cation (BLA).3 In addition, some of the questions about
monoclonality of the manufacturing cell lines that have been
raised in the Investigational New Drug (IND) application
need to be adequately addressed. Some of the requirements
of regulatory agencies are the use of proper documentations
and appropriate statistical analysis to demonstrate the ability
to select single cells.
Due to its simple operation, limiting dilution cloning is
traditionally performed to achieve monoclonality. However,
limiting dilution cloning may be time and labor intensive.
Flourescent-activated cell sorting (FACS), ClonePix, and
Additional Supporting Information may be found in the online ver-
sion of this article.
Correspondence concerning this article should be addressed to
J. Quiroz jorge.quiroz@merck.com
reduce time and effort for clone screening.4 Although these
clone screening technologies may eventually provide a more
definite proof of monoclonality, limiting dilution is currently
used in some ongoing BLA submissions where manufactur-
ing cell lines could not survive after FACS single-cell sort-
ing or the advanced cloning technology was not properly
developed back when the project was initiated. Therefore,
appropriate statistical methodology of data from limiting
dilution cloning continues to be a requirement to assure
monoclonality. In this paper, we focus on the statistical anal-
ysis of data from limiting dilution cloning to assess mono-
clonality in generating manufacturing recombinant cell
lines. We discuss how to assess monoclonality when a single
round is sufficient and when two repetitive rounds are
necessary.
If there is sufficient certainty about the assurance of
monoclonality in one round, a single round of clonality may
be sufficient.3 Traditional statistical analyses to assess mono-
clonality are based on point estimates. However, assessment
of monoclonality in generating manufacturing cell lines is a
confirmatory study conducted to demonstrate that a specified
level of monoclonality meets regulatory requirement. Point
estimates are useful, but they may not allow practitioners to
confirm precisely whether a specified level of monoclonality

C 2016 American Institute of Chemical Engineers

V 1061
1062
meets regulatory requirement due to the uncertainty present
in the data. Therefore, it is important to report the level of
uncertainty in order to fairly represent the conclusion of the
study. One way to quantify the uncertainty is to compute
confidence intervals on the probability of monoclonality.
Such comprehensive and detailed approaches, to the best of
our knowledge, are not widely used for assessing monoclon-
ality in generating manufacturing cells lines.
If sufficient certainty about assurance of monoclonality in
generating manufacturing cell lines is not achieved in the
first round of subcloning, cell lines from a single well are
subcloned a second time.4 Coller and Coller5 derived a for-
mula to estimate the probability of monoclonality in two
consecutive rounds based on an infinite series. Because the
infinite series may be computationally intensive, a common
practice is to estimate the probability of monoclonality in
two repetitive rounds using the first two terms of the series.
To increase the accuracy, other practitioners include a few
more terms. However, in general, these two approaches
underestimate the probability of monoclonality in two repeti-
tive rounds. Underestimation of the probability of monoclon-
ality may be misleading and may encourage unnecessary
rejection of acceptable dilution levels. In this paper, we
reduce the expression suggested by Coller and Coller to a
simpler series. The simpler series allows one to obtain more
accurate estimation of the probability of monoclonality and
can be easily implemented using any spreadsheet program
like Microsoft Excel.
The rest of the paper is organized as follows. In Section
2, we present an experimental dataset from a limiting dilu-
tion cloning used for assessing monoclonality of generating
a manufacturing cell line. In Section 3, we briefly discuss
the Poisson distribution as used in limiting dilution clon-
ing, and calculate confidence intervals to assess monoclon-
ality when a single round of clonality is sufficient. In
Section 4, we derive a simpler series that provides more
accurate and reliable estimate of the probability of mono-
clonality in two repetitive subclonings. A case study is
used to demonstrate the proposed statistical methodology
in Sections 3 and 4. Section 5 has the conclusion and some
recommendations.
Limiting Dilution Method for Generating
Manufacturing Cell Lines
To demonstrate the issues and solutions for assessing
monoclonality, Figure 1 shows a schematic representation
with first and second rounds of a limiting dilution cloning
experiment as used in generating manufacturing cell lines.
Newly transfected cell clones are not well established and
fragile; therefore, a high cell density is required in the first
step to guarantee maximizing the number of wells with sin-
gle colony. In the first subcloning step, the newly trans-
fected cells were subjected to limiting dilution at an initial
average input cell density of 2500 cells per well. After an
incubation period, surviving cells growth and form colo-
nies. Wells with a single colony are denoted monoclonal. If
probability of monoclonality is not sufficiently high in the
first round, cells from selected monoclonal wells are sub-
cloned a second time. Cells in the second round are better
established and less fragile. Thus, lower cell densities are
used during the second round. To determine the appropriate
density in the second round, several limiting dilution clon-
ings were considered. In this case, the second subcloning
Biotechnol. Prog., 2016, Vol. 32, No. 4
was seeded at four different seeding densities (5, 2.5, 1.25,
and 0.6 cells per well, respectively). The assessment of
monoclonality was performed at each seeding density. The
probability of monoclonality for two consecutive subclon-
ings was calculated by pairing the data from the first sub-
cloning round and one of the dilutions of the second
subcloning round.
Table 1 shows an example dataset from an experiment
with two consecutive rounds where four different dilutions
were considered in the second subcloning round as shown in
Figure 1. The dataset is based on a real experiment and,
without loss of generality, proprietary information is not dis-
closed. The number of wells surveyed is considered large
enough for the purpose of the study.
Data collection is not standardized and varies from lab to
lab. One practice is to record the number of wells showing
no growth, also denoted as empty wells, and the total num-
ber of wells surveyed. A less common practice is to record
the number of wells with single colony and the total number
of wells surveyed. Since colonies may be lumped together,
counting empty wells is preferred and less prone to errors
than counting wells with single or multiple colonies. For the
benefits of ongoing studies with either type of data, the
example dataset shows both types of reporting so that we
can illustrate how to conduct the statistical analyses using
either type of data collection.
Statistical Assessment of Monoclonality
in a Single round of Cloning
In this section, we discuss how to assess monoclonality
when a single round of cloning is sufficient. Since the meth-
odology is based on the Poisson distribution assumption, we
briefly review the Poisson distribution in the limiting dilution
cloning context.
The Poisson distribution
The Poisson distribution has broad application as model
for count data such as the number of wells with specified
number of cells. The Poisson distribution is not only useful
to model limiting dilution data, but can be applied to solve
some of the issues surrounding this type of data.
In limiting dilution method, the Poisson distribution is
commonly expressed in terms of the average number of col-
onies per well, hereafter denoted by l, such that the propor-
tion of wells to contain k number of colonies is
lk 2l
Pl ðkÞ5 e ; (1)
k!
where k50; 1; 2; . . .. More specifically, the proportion of
empty wells, here denoted by s, is expressed as s5Pl ð0Þ5
e2l and the proportion of wells with single colony as
Pl ð1Þ5le2l . Some critical values can also be expressed as
function of s as follows:
1. The average number of colonies per well expressed as a
function of s is
l52ln ðsÞ: (2)
2. The proportion of wells with single colony expressed as a
function of s is
Biotechnol. Prog., 2016, Vol. 32, No. 4 1063

Figure 1. A schematic representation of first and second rounds of a limiting dilution cloning as used in generating manufacturing
cell lines. Transfected cells were subcloned by two consecutive limiting dilution steps. The first subcloning was seeded at an
average of 2500 cells per well. The second subclonings were seeded at four different seeding densities (5, 2.5, 1.25, and 0.6
cells per well, respectively). The assessment of monoclonality was done at each subcloning round. The probability of mono-
clonality for the two consecutive subclonings was calculated by combining the information from the first subcloning step
and one of the dilutions of the second subcloning round.

Pl ð1Þ5le2l 52sln ðsÞ: (3)

Table 1. Limiting Dilution Data from Case Study
Number of wells
3. The proportion of wells showing some growth as a func-
showing
tion of s is
Subcloning Input cell No-growth Single
round density (empty wells) colony Total 12Pl ð0Þ512s: (4)
First 2500 6826 799 7680
Second 5 749 166 960
2.5 745 93 864 The probability of monoclonality is expressed as a function
1.25 886 69 960 of s to facilitate constructions of confidence intervals. The
0.6 926 31 960 confidence intervals will be useful later on when statistical
1064
tests are used for assessing whether the probability of mono-
clonality is consistent with some specified value.
Assessment of monoclonality based on the number of
empty wells
Assurance of monoclonality is expressed as a probability,
which is defined as the proportion of wells with single col-
ony out of all the wells showing cell growth. More specifi-
cally, the probability that a cell line is monoclonal in a
single round is defined as
proportion of wells with single colony
PM1 5 : (5)
proportion of wells with one or more colonies
Combining Eqs. 3 and 4, the probability of monoclonality
can be expressed as a function of s as follows:
Pl ð1Þ 2sln ðsÞ
PM1 5 5 : (6)
12Pl ð0Þ 12s
We use confidence intervals to assess monoclonality in gen-
erating manufacturing cell lines. Since there is no direct
way, confidence intervals for probability of monoclonality,
PM1, may be derived by taking advantage of the fact that
PM1 is expressed as a strictly monotone function of s as
shown in Eq. 6.6 Using confidence intervals on s with well-
known statistical properties will ensure that confidence inter-
vals on PM1 also have good statistical properties. This is
critical when using statistical intervals to conduct statistical
tests with managed risks of producing unreliable
conclusions.
Since subcloning does no warrant the same result each
time, the proportion of empty wells, s, is estimated from the
data and confidence intervals are required to account for the
uncertainty associated with the estimation process. Confi-
dence intervals on s can be accomplished using a binomial
approximation. There are several methods for constructing
confidence intervals on a binomial proportion. Staszewski7
used the popular Wald confidence interval to construct
bounds for the proportion of empty wells in the presence of
two cultures, and used it to compute a conservative working
bounds for probability of monoclonality of an interested cul-
ture. The Wald confidence interval is based on the traditional
normal approximation to the binomial proportion estimated
as the ratio of number of empty wells to the total number of
surveyed wells. However, Brown et al.8 demonstrated that
the Wald confidence interval for binomial proportion does
not perform well even for large sample size and recom-
mended the use of either the Wilson’s or the Agresti-Coull
method instead. On another application, Evans et al.3 used
the Wilson’s method to calculate confidence intervals for the
probability of single cells per droplet during FACS single-
cell sorting. The Wilson’s method, however, may perform
poorly near the values of 0 and 1.
In order to construct confidence intervals on the probabil-
ity of monoclonality, PM1, we construct confidence intervals
for the proportion of empty wells using the Agresti-Coull
method.9 The Agresti-Coull method is simple to implement,
performs well for small and large sample sizes, and does not
have the disadvantage of the Wilson’s method of poor per-
formance near 0 and 1.10 In addition, the Agresti-Coull inter-
val has coverage probability generally much closer to the
stated level than the traditional Wald intervals.11 Therefore,
Biotechnol. Prog., 2016, Vol. 32, No. 4
the Agresti-Coull interval is preferred because it provides
more reliable conclusions
In monoclonal studies, it is usually desired to demonstrate
that the level of monoclonality exceeds some specified level.
In this context, the data must support that the assurance of
monoclonality is greater than 90%. This can be achieved by
computing the lower bound of a two-sided 90% confidence
interval on the probability of monoclonality. This is equiva-
lent to a one-sided 95% confidence interval. This will ensure
a test of size a50:05 with 100ð122aÞ%590%. If the
lower bound is greater than the specified level, we conclude
that the data is consistent with the specified level of
monoclonality.
The lower bound of a two-sided ð122aÞ% Agresti-Coull
confidence interval on a binomial proportion is given by
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b
s ð12b sÞ
CIAC 5b
s 2j (7)
T
where
T  5T1j2 ;
X10:5j2 (8)
b
s5 ;
T
b
s represents an estimate for s, X represents the observed
number of empty wells (or the number of wells with single
colonies, depending on the type of data), T is the number of
surveyed wells, and j5z12a denotes the 12a quantile of the
standard normal distribution. If anffi upper bound is required,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Eq. 7 becomes bs 1j ðb s ð12b s ÞÞ=T  .
Confidence intervals on PM1 are obtained by exchanging s
with the calculated bounds of the confidence interval in
Eq. 6.
Assessment of monoclonality based on the number of wells
with single colony
When using wells with single colony, confidence intervals
for the probability of monoclonality can be determined using
a slight modification of the approach discussed in Section
3.2. First, Agresti-Coull confidence intervals can be calcu-
lated for the proportion of wells with single colony, Pl ð1Þ.
Using the calculated confidence interval on the proportion of
wells with single colony, estimated values for l can be
obtained by solving for l in the expression Pl ð1Þ5le2l .
The corresponding values for l are then used to estimate the
proportion of empty wells using the expression s5e2l .
Finally, the estimated values for the proportion of empty
wells are used for constructing bounds for the probability of
monoclonality as discussed in Section 3.2.
A case study
We now demonstrate how to perform the statistical analy-
sis to assure monoclonality using the data in Table 1. For
simplicity, the method based on empty wells is denoted
Approach 1 and the one based on wells with single colony is
denoted Approach 2.
Approach 1: Based on the Number of Empty Wells. To
illustrate how to assess monoclonality using Approach 1,
consider the first subcloning round where the average input
density is 2,500 cells per well. For demonstration purposes,
we will also calculate the probability of monoclonality for
Biotechnol. Prog., 2016, Vol. 32, No. 4 1065

Table 2. Estimated Proportion of Empty Wells with One-Sided 95% Confidence Intervals Using Both Approaches
Approach 1 Approach 2
Proportion of Proportion of
Input cell density empty wells Lower bound empty wells Lower bound l Upper bound
2500 88.9% 88:3% 88.9% 88.3% 0.1172 0.1243
5 77.9% 75:7% 80.6% 78.0% 0.2156 0.2486
2.5 86.1% 84:1% 88.4% 86.4% 0.1232 0.1462
1.25 92.1% 90:7% 92.4% 90.9% 0.0791 0.0956
0.6 96.3% 95:3% 96.6% 95.6% 0.0348 0.0452

Table 3. Estimated Probability of Monoclonality Assuming One interval), that the probability of monoclonality for the first
Round with Corresponding One-Sided 95% Confidence Intervals round is at least 93.8%. This result may be used by practi-
Using Both Approaches tioners to assess whether the level of monoclonality is con-
Approach 1 Approach 2 sistent with some specified level. For example, suppose that
Input cell Probability of Lower Probability of Lower to assure the monoclonality of a specified cell line, it is
density monoclonality bound monoclonality bound required that the monoclonality be at least 90%. Since the
2,500 94.2% 93.9% 94.2% 93.9% calculated lower bound is greater than 90%, the data provide
5 88.0% 86:7% 89.6% 88:1% sufficient evidence to support that the level of monoclonality
2.5 92.7% 91:6% 94.0% 92:9% exceeds the required probability level of 90%.
1.25 96.0% 95:2% 96.1% 95:3%
0.6 98.1% 97:6% 98.3% 97:8%
Approach 2: Based on the Number of Wells with Single
Colony. To demonstrate how to obtain the results using
each dilution in the second round as individual single Approach 2, first we calculate a lower bound for a two-sided
rounds. 90% Agresti-Coull confidence interval on the proportion of
Before we calculate confidence intervals for the probabil- wells with single colony, Pl ð1Þ.
ity of monoclonality, a confidence interval for s is required. In this example, the value of X 5 799 and T 5 7680. The
For a lower bound of a two-sided 100ð122aÞ%590%; a5 values of a50:05 and j5z0:95 51:645. From Eq. 8, an esti-
0:05 and j5z120:05 5z0:95 51:645. From the data, X 5 6826, mate for the proportion of wells with single colony is
T 5 7680. Thus, T  5768011:6452 57683. From Eq. 8, the
estimated proportion of empty wells, b
s , is given by X10:531:6452 79910:531:6452
Pld
ð1Þ5 5
T 7683 (12)
X10:5j2 682610:531:6452 6827 800
b
s5 5 5 588:9%: (9) 5 50:1041510:41%:
T 7683 7683 7683
The calculated lower bound for the proportion of empty where T  was calculated in Section 3.4.1.
wells, s, is
Since the proportion of empty wells, s, is estimated from
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
b
s ð12b sÞ s5e2l , upper bounds on l are required for constructing one-
Ls 5b
s 2j sided confidence intervals on s using this approach. the cal-
T
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi culated upper bound for Pl ð1Þ is
0:889ð120:889Þ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
50:88921:6453 588:3%;
7683 0:10413ð120:1041Þ
0:104111:6453 50:1098510:98%: (13)
Treating each dilution in the second round as individual sin- 7683
gle round, the calculated proportion of empty wells for each Based on these results, we can use the Microsoft Excel func-
dilution in the second round with corresponding lower bound tion SOLVER to find corresponding values for l that satisfy
is shown under the columns with heading “Approach 1” in the expression Pl ð1Þ5le2l and subsequently calculate confi-
Table 2. dence intervals for the proportion of empty wells. In particu-
The estimated probability of monoclonality is lar, the value l50:1172 is obtained when 0:10415le2l , and
the corresponding estimated proportion of empty wells is
s ln ðb
d1 5 2b s Þ 20:8893ln ð0:889Þ
PM 5 594:2%: (10) s~5e20:1172 50:889588:9%. Similarly, the value l50:1243
12b s 120:889 satisfies 0:10985le2l . Note that the values 0.1041 and
The calculated lower bound for the probability of monoclon- 0.1098 were obtained from calculations in Eqs. 12 and 13.
ality is Thus, a calculated lower bound for the proportion of empty
wells, s, is e20:1243 588:3%.
2Ls ln ðLs Þ 20:8833ln ð0:883Þ
LPM 5 5 593:9%; (11) The probability of monoclonality with corresponding
12Ls 120:883
lower bound can be calculated using Eqs. 10 and 11 using
where Ls is the lower bound for the proportion of empty the results obtained in the previous paragraph. The results
wells. The calculated probability of monoclonality based on are the same as those obtained in Approach 1.
the number of empty wells with corresponding lower bounds The calculated proportion of empty wells using Approach
are shown under the columns with heading “Approach 1” in 2 with corresponding lower bounds for the second round are
Table 3. shown under the columns with heading “Approach 2” in
From Eq. 11, we can state, with 95% confidence (this is Table 2. Table 2 also shows the estimated values for l with
based on the lower bound of the two-sided 90% confidence corresponding upper bounds.
1066
The probabilities of monoclonality with corresponding
lower bounds using Approach 2 are shown in Table 3. The
calculated probabilities of monoclonality estimated using
Approaches 1 and 2 are expected to differ slightly due to the
sampling variability in the counting process. The results indi-
cate that there is a high level of agreement suggesting that
there are no practical differences between these results.
Assessing Monoclonality in Two Consecutive
Subclonings
If there is no sufficient evidence of monoclonality in the
first subcloning, cells from a monoclonal well are subcloned
a second time. The statistical analysis to assess monoclonal-
ity in multiple rounds of serial subcloning is not as straight-
forward as in the one round case. The analysis is very
complex and there is no simple formula to estimate the prob-
ability of monoclonality. Coller and Coller5 derived an
expression to estimate the probability of monoclonality in
two repetitive subclonings under the assumption that every
viable clone is equally likely to be monoclonal. The expres-
sion derived by Coller and Coller is
1 1
PM2 5w2 ð1Þ1w1 ð1Þð12w2 ð1ÞÞ1 w2 ð2Þw1 ð2Þ1 w2 ð2Þw1 ð3Þ
2 3
1 1 1
1 w2 ð2Þw1 ð4Þ1 . . . 1 2 w2 ð3Þw1 ð2Þ1 2 w2 ð3Þw1 ð3Þ
4 2 3 where
1 1 1
1 2 w2 ð3Þw1 ð4Þ1 . . . 1 3 w2 ð4Þw1 ð2Þ1 3 w2 ð4Þw1 ð3Þ
4 2 3 w2 ð1Þ5
1
1 3 w2 ð4Þw1 ð4Þ1 . . .
4 w1 ðcÞ5
⯗ c!ð12s1 Þ
1 1 1 ðc21Þ=c
1 p21 w2 ðpÞw1 ð2Þ1 p21 w2 ðpÞw1 ð3Þ1 p21 w2 ðpÞw1 ð4Þ1 . . .
2 3 4 WðcÞ 5
⯗ 12s2
(14)
where
Pl ðcÞ lc e2li
wi ðcÞ5 5 i ; (15)
12Pl ð0Þ c!ð12si Þ
Pl ðcÞ is defined in Eq. 1, 12Pl ð0Þ in Eq. 4, i 5 1 corre-
sponds to the first subcloning round and i 5 2 to the second
round, c represents the number of different clones in a well
with c51; 2; . . ., and si represents the proportion of empty
wells in the ith subcloning round. Alternatively, from Eq. 2,
the expression in Eq. 15 can be expressed as function of pro-
portion of empty wells, s,
si ð2ln si Þc
wi ðcÞ5 (16)
c!ð12si Þ
where si represents the proportion of empty wells in the ith
subcloning round. One advantage of the expression in Eq. 16
is that the calculations only require the estimated proportion
of empty wells.
The expression to calculate the probability of monoclonal-
ity in two consecutive rounds derived by Coller and Coller5
assumes that clones in a well with multiple colonies growth
at the same rate. That is, the input cells for the second round
contain equal number of cells for each clone. In reality,
some clones may grow faster than other.12 Therefore, as
observed by Coller and Coller, should one clone growths
Biotechnol. Prog., 2016, Vol. 32, No. 4
faster than another, the probability of monoclonoality of
wells with multiple colonies in the second round would
increase. As a result, the series derived in Eq. 14 provides a
conservative working bound for the probability of monoclon-
ality. Since the probability of monoclonality in the more
realistic cases is very complex, practitioners continue to use
Eq. 14 as a worst-case scenario analysis.
Due to the number of terms in Eq. 14, calculations of the
probability of monoclonality in two repetitive rounds may be
computationally intensive. Thus, the probability of mono-
clonality is sometimes estimated using only the first two
terms of the series. To increase accuracy, some practitioners
use three or more terms. For simplicity of notation, the
approximation based on the first two terms of Eq. 14 is
denoted Approximation 1 and the one based on three or
more terms is denoted Approximation 2.
In general, results from Approximation 1 and 2 underesti-
mate the conservative working bound for the probability of
monoclonality encouraging unnecessary rejection of dilution
with good monoclonal level. An alternative expression to the
infinite series in Eq. 14 that consists of a simpler series that
leads to higher level of accuracy is
X1
PM2 5w2 ð1Þ1 cw1 ðcÞWðcÞ (17)
c51
2s2 ln s2
; (18)
12s2
s1 ð2ln s1 Þc
; (19)
 
1=c 1=c 1=c
s2 12s2 1s2 ln s2
; (20)
c51; 2; . . ., and s1 is the proportion of empty wells in the
first subcloning round and s2 the proportion of empty wells
in the second subcloning round. The approximation based on
the simpler series in Eq. 17 is denoted Approximation 3.
The idea behind the simpler series consists of reducing the
series in Eq. 14 by collapsing the subseries column-wise into
one term. Details for the derivation of the simpler series is
provided in supplementary materials available online.
For comparison purposes, Figure 2 shows the probability
of monoclonality in two repetitive subclonings for various
scenarios using the three approximations described above.
For accuracy purposes, Approximations 2 and 3 were based
on ten terms. Later on, in our case study four terms may be
sufficient. In particular, this is true when the probability of
monoclonolity is large. For illustration, we considered the
same proportion of wells showing growth in each of the two
subcloning steps. The horizontal axis shows the proportion
of wells showing growth (instead of the actual probability of
monoclonality at each stage) to facilitate the visual inspec-
tion of the graph.
Figure 2 indicates that, in general, Approximation 2 pro-
vides better estimation of the probability of monoclonality
than Approximation 1. The graph also indicates that Approx-
imation 3 significantly improves estimations of probability of
monoclonality relative to Approximation 1 and 2.
The underestimations of Approximation 1 and 2 might
incorrectly suggest that the level of monoclonality is below
Biotechnol. Prog., 2016, Vol. 32, No. 4
Figure 2. Estimations of conservative working bound for prob-
ability of monoclonality in two rounds based on
Approximation 1 (two-terms), Approximation 2
(multiple terms), and Approximation 3 (simpler
series).
some specified target. For example, suppose that the speci-
fied level of monoclonality is 90% and the proportion of
wells showing growth in the first and second subclonings are
both 65%, this is indicated by the horizontal and vertical
dashed gray lines in Figure 2. From the curves representing
Approximations 1 and 2, the conservative working bound for
the probability of monoclonality is 81.1% using Approxima-
tion 1 and 86.8% using Approximation 2. These results sug-
gest that the level of level of monoclonality did not meet the
specified level. On the other hand, the conservative working
bound for the probability of monoclonality is 89.8% from
the curve representing Approximation 3. The result from
Approximation 3 suggests that, for practical purposes, the
level of monoclonality met the specified level of 90%.
Case study
Once again consider the data in Table 1. Suppose that the
goal is to assure monoclonality with two repetitive subclon-
ings. For illustration purposes, we consider the first subclon-
ing and the second subcloning where the initial average cell
density is 5 colonies per well. In this case, s1 50:889½56862
=7680588:9% and s2 50:780½5749=960578:0%.
From Eq. 16, w1 ð1Þ50:9423, and w2 ð1Þ50:8809. The
probability of monoclonality with two repetitive subclonings
using the first two terms (Approximation 1) w2 ð1Þ50:8809
and w1 ð1Þð12w2 ð1ÞÞ50:9423ð120:8809Þ50:1122 is PM2 5
0:880910:112250:9931  99:3%. Based on the first four
terms in the first row of Eq. 14 (Approximation 2), the proba-
bility of monoclonality was 0:880910:11221
0:003010:000150:9962  99:6%. From Eq. 17 (Approxima-
tion 3), using the first four terms was 0:88091
0:112210:003410:000150:9966  99:7%. In Approximations
2 and 3, more than four terms did not add significantly to the
final probability and were not considered in the calculations.
Since the proportion of empty wells is large, all three approxi-
mations provide very similar results as suggested by Figure 2.
One interpretation of these results is that a conservative
working bound for the probability of monoclonality in two
repetitive subclonings is 99.3% using Approximation 1, 99.6%
using Approximations 2 and 99.7% using Approximation 3.
The results based on the three approximations for the
other limiting dilutions are shown in Table 4. The results
1067
Table 4. Estimated Probability of Monoclonality with Two Repeti-
tive Subclonings
Approximation
Input cell
density 1 2 3
5 99.3% 99.6% 99.7%
2.5 99.6% 99.8% 99.8%
1.25 99.8% 99.9% 99.9%
0.6 99.9% 99.9% 99.9%
from Approximations 2 and 3 for the other dilution levels
were based on four terms since the other terms did not add
significantly to the results.
Conclusion
Assurance of monoclonality of recombinant cell lines is a
critical issue to gain regulatory approval in biological license
application. Some of the requirements of regulatory agencies
are the use of proper documentations and appropriate statisti-
cal analysis to demonstrate monoclonality. In this article, we
use confidence intervals for assessing monoclonality for lim-
iting dilution cloning in the generation of recombinant manu-
facturing cell lines. Since there is no direct way, confidence
intervals on the probability of monoclonality are constructed
expressing it as a function of proportion of empty wells. We
use the Agresti-Coull method for constructing confidence
intervals on the proportion of empty wells because they have
very well-known statistical properties. Consequently, confi-
dence intervals for probability of monoclonality will also
exhibit good statistical properties since they are based on the
confidence intervals for proportion of empty wells.
We recommend that whenever possible, data to assess
monoclonality should consist of number of empty wells and
total number of wells surveyed. Here, we demonstrate how to
assess monoclonality in one single round for the recom-
mended data collection. However, since data collection is not
standardized and varies from lab to lab, one less common
practice consists of collecting the number of wells with single
colony and total number of wells surveyed. For the benefit of
ongoing study with this type of data, we demonstrate how to
assess monoclonality based on wells with single colony.
When assessing monoclonality in two repetitive rounds, an
infinite series expression derived by Coller and Coller is used.
Because the series is computationally intensive, the first two,
or more than three terms, are used to estimate the probability
of monoclonality. However, these approaches tend to underes-
timate the conservative working bound for the probability of
monoclonality. Here, we reduced to a simpler series expres-
sion the complex series frequently used to assess monoclonal-
ity derived by Coller and Coller. In general, the simpler series
expression is recommended because it improves the estima-
tion of the conservative working bound for the probability of
monoclonality and can be easily implemented in any spread-
sheet program like Microsoft Excel.
Literature Cited
1. European Medicines Agency, Committee for Medicinal Products
for Human Use. Guideline on production and quality control of
monoclonal antibodies and related substances. Draft, 2007.
Accessed: August 2015.
2. International Conference on Harmonisation of Technical
Requirements for Registration of Pharmaceutical or Human Use.
1068
Derivation and Characterization of Cell Substrates Used for
Production of Biotechnological/biological Products. Q5D, 1997.
Accessed: August 2015.
3. Evans K, Albanetti T, Savery J, Miro-Quesada G, Rajan B,
Groves C. Assurance of monoclonality in one round of cloning
through cell sorting for single cell deposition coupled with
high resolution cell imaging. Biotechnology Prog. 2015;31:
1172–1178.
4. Lai T, Yang Y, Ng SK. Advances in mammalian cell line devel-
opment technologies for recombinant protein production. Phar-
maceuticals 2013;6:579–603.
5. Coller HA, Coller BS. Statistical analysis of repetitive subclon-
ing by the limiting dilution technique with a view torward
ensuring hybridoma monoclonality. Hybridoma 1983;2:91–96.
6. Mood AM, Graybill FA, Boes DC. Introduction to the Theory
of Statistics. Probability and Mathematical Statistics, 5th ed.
McGraw-Hill Book Company: New York, 1974.
Biotechnol. Prog., 2016, Vol. 32, No. 4
7. Staszewski R. Cloning by limiting dilution: An improved esti-
mate that an interesting culture is monoclonal. Yale J Biol Med.
1984;57:865–868.
8. Brown LD, Cai TT, DasGupta A. Interval estimation for a bino-
mial proportion. Stat Sci. 2001;16:101–133.
9. Agresti A, Coull BA. Approximate is better than” exact” for inter-
val estimation of binomial proportion. Am Stat. 1998;52:119–126.
10. Agresti A. Categorical Data Analysis. Wiley Series in Probabil-
ity and Statistics, 3rd ed. New Jersey: Wiley, 2013.
11. Suess EA, Trumbo BE. Introduction to Probability Simulation and
Gibbs Sampling with R. New York: Springer, Use R! Series, 2010.
12. Underwood PA, Bean PA. Hazards of the limiting-dilution
method for cloning hybridomas. J Immunol Methods. 1998;107:
119–128.
Manuscript received Jan. 7, 2016, and revision received Mar. 24,
2016.