PHYSICAL REVIEW E 103, 052410 (2021)

Effectiveness of glass beads for plating cell cultures

Alidivinas Prusokas
Plant and Microbial Sciences, School of Natural and Environmental Sciences, Newcastle University,
Newcastle upon Tyne NE1 7RU, United Kingdom
and Department of Biology, University of York, Heslington, York YO10 5DD, United Kingdom

Michelle Hawkins
Department of Biology, University of York, Heslington, York YO10 5DD, United Kingdom

Conrad A. Nieduszynski
The Earlham Institute, Norwich Research Park, Norwich NR4 7UZ, United Kingdom

*
Renata Retkute
Department of Plant Sciences, University of Cambridge, Cambridge CB2 3EA, United Kingdom

(Received 16 November 2020; revised 9 April 2021; accepted 23 April 2021; published 17 May 2021)

Cell plating, the spreading out of a liquid suspension of cells on a surface followed by colony growth, is a
common laboratory procedure in microbiology. Despite this, the exact impact of its parameters on colony growth
has not been extensively studied. A common protocol involves the shaking of glass beads within a Petri dish
containing solid growth media. We investigated the effects of multiple parameters in this protocol: the number
of beads, the shape of movement, and the number of movements. Standard suspensions of Escherichia coli were
spread while varying these parameters to assess their impact on colony growth. Results were assessed by a
variety of metrics: the number of colonies, the mean distance between closest colonies, and the variability and
uniformity of their spatial distribution. Finally, we devised a mathematical model of shifting billiard to explain
the heterogeneities in the observed spatial patterns. Exploring the parameters that affect the most fundamental
techniques in microbiology allows us to better understand their function, giving us the ability to precisely control
their outputs for our exact needs.

DOI: 10.1103/PhysRevE.103.052410

I. INTRODUCTION of viable cells. The more homogeneously cells are spread on

the surface during plating, the better separated the resulting
A prominent technique in microbiology is cell plating to
cell colonies, leading to a more precise CFU estimation. It is
enable colony growth on nutrient agar. The goal of cell plating
expected that the number of colonies formed is linearly pro-
is to separate cells contained within a small sample volume
portional to the concentration of viable cells in the suspension,
by homogeneously distributing the cell suspension over the
provided that the suspension was mixed and spread well. If the
surface of a plate. This results in the formation of discrete
suspension is not spread well, clusters of cells will be formed
colonies after incubation that can be enumerated or subjected
which visually resemble a single colony and therefore will be
to further analysis. Counting the number of colonies is used
enumerated as a single colony [8]; furthermore, bias may be
in many cell culture protocols: detection of bacteria in food
introduced when analyzing genetic variations and heredity of
samples or clinical specimens [1]; measuring progenitor stem
such clusters of cells [9]. To proceed with further procedures,
cell content [2]; constructing gene-knockout libraries [3];
the cells need to be restreaked to get single colonies, which
cell-based DNA cloning [4]; testing resistance to drugs [5];
takes extra incubation time.
studying evolution of antibiotic resistance [6]; activation of
There are two strategies for cell plating: spread plating with
mucosal-associated invariant T cells [7]. Successful cultiva-
a turntable rod or spread plating with glass beads [10,11].
tion of cells depends critically on the choice of appropriate
Using the first method, a small volume of a cell suspension is
growth media, cell plating method, and incubation conditions.
spread over the plate surface using a sterile bent glass rod as
A colony is defined as a visible cluster of cells growing
the spreading device. The second method, which is the subject
on the surface of a medium, presumably derived from a single
of our study, involves shaking glass beads over the surface
cell. These single progenitor cells are also called colony form-
of the plate. This technique is also known as the Copacabana
ing units (CFUs), which provide an estimate of the number
method [12,13].
The protocol for using sterile glass beads for dispersion of
cells on solid media has the following steps [11,12]: (1) a cell
*
[email protected] suspension is dispensed onto the middle of a round Petri dish

2470-0045/2021/103(5)/052410(7) 052410-1 ©2021 American Physical Society

ALIDIVINAS PRUSOKAS et al. PHYSICAL REVIEW E 103, 052410 (2021)

containing solid media; (2) spherical glass beads are poured

into the middle of the plate; (3) the plate lid is closed; (4) the
plate is agitated with a shaking motion so that the glass beads
roll over the entire surface of the plate; (5) the plate is then
inverted to remove the beads before incubation.
In this study we investigate how the way a plate is agitated
influences the spread of CFUs. We perform experiments with
the bacteria Escherichia coli and use a model from statistical
mechanics, billiard, to explain heterogeneities in the observed
spatial patterns. Billiard is a type of mathematical model
describing a dynamical system where one or more particles
move in a container and collide with its walls [14]. Billiard
systems based on the wave dynamics in cavities, acoustic
resonance in water, atoms bouncing off a beam of light, and
quantum dots have been studied over several decades both
experimentally and theoretically [15–21]. We introduced a
shifting billiard as a conceptual abstraction of the cell plating
with glass beads. Numerical analysis of the dynamics of this
simple yet efficient billiard, indicated a close relationship FIG. 1. Metrics to assess the effectiveness of cell plating:
between plate movement and quality of colony distribution. (a) Counting the number of CFUs, nCFUs . (b) Calculating the mean
(dashed line) of the distribution of distances to nearest CFUs, dNB .
II. CELL PLATING EXPERIMENT (c) Dividing the plate into 5◦ arcs, determining the number of
CFUs in that arc, and calculating the coefficient of variation, CVS .
The experiments were performed on circular plates with (d) Calculating the mean difference between observed and expected
88 mm diameter. Sterile 4-mm glass plating beads (manufac- distances from CFUs and plate center, mS ; black line shows fre-
tured by Sigma-Aldrich) were used for the dispersion of cells quency based on a triangular distribution.
over the surface of a plate. This gives the ratio between glass
bead and plate d = 4/88 = 0.045.
We explored following movement of a plate: number of CFUs in that arc, and estimated the coefficient of
(1) L shape: The plate is moved on a trajectory resem- variation between these numbers, CVS . Next we have calcu-
bling the horizontally reflected letter “L”: up-right-left-down lated distances of CFUs to the center of the plate, binned into
(↑→←↓). 1 mm intervals, and calculated an average difference between
(2) Up-down: The plate is moved up and down along the observed frequencies and expected frequencies, mS . Expected
vertical axes (↑↓). frequencies for the latter metrics were assumed to follow the
(3) Hourglass: The plate is moved on a trajectory re- triangular distribution. Larger values of nCFUs and dNB corre-
sembling the hourglass: diagonal up-left-diagonal down-right spond to a better spread, while smaller values of CVS and mS
(←). indicate a more uniform spread of colonies. Figure 1 shows
We called the full cycle of movement, where a plate returns a graphical representation of these metrics, which gives the
to its initial position, a loop. following values: nCFUs = 335, dNB = 2.8, CVS = 0.46, and
E. coli K-12 MG1655 was incubated in Luria-Bertani (LB) mS = 2.3.
broth overnight (10 g/L Tryptone, 5 g/L yeast extract, 5 g/L
NaCl, pH 7.0) [22]. A single E. coli colony was sampled from IV. THE MODEL
a streaked plate, placed within the LB broth, and incubated
in a 37 ◦ C shaker overnight until saturation. This overnight A. Shifting billiard
stock was serially diluted by factors of 10 in 56/2 salts to the First, we start with a circular billiard which is stationary
required dilution. 100 μl of the cell dilution was plated on in space. Without loss of generality, we set the radius of the
LB agar 1.8% W/V, utilizing the above motions, with a vary- circle to be 1. A single particle is moving without friction
ing number of loops and beads. These plates were incubated between the boundary of the unit circle. Particle movement
overnight at 37 ◦ C and subsequently imaged and shown in is defined by the elastic collision rule: the angle of reflection
Fig. 8 in Appendix. is equal to the angle of incidence [23]. The dynamics of parti-
cle movement can be described in terms of collision maps,
where θk denotes the point of the collision and ϕk denotes
III. ASSESSING EFFECTIVENESS OF PLATING METHOD
the reflection angle for the kth collision, k = 1, . . . , n. We
The colony counts were performed automatically using a adopted a system where θ = 0 corresponds to a point (1,0) in
custom-made code for image analysis; more details are given a Cartesian coordinate system. Figure 2(a) shows a collision
in the Appendix. We used the following metrics to assess of a particle with the boundary at the point θ = π and ϕ = π3 .
plating methods: number of CFUs, nCFUs ; mean distance to The dynamics of a particle in the circle is fully integrable and
nearest CFU, dNB ; and variability and uniformity in CFU is described by the following properties: (i) ϕk = ϕ1 , i.e., the
spatial distribution on a plate. To determine the spatial vari- angle of the reflection does not change; (ii) the coordinates
ability, we have grouped CFUs into 5◦ intervals, calculated the of the point of collision satisfies θk = (θ1 + 2kϕ1 ) mod 2π ;

052410-2
EFFECTIVENESS OF GLASS BEADS FOR PLATING CELL … PHYSICAL REVIEW E 103, 052410 (2021)

function of v0 for two values α = π2 and α = π4 after the circle

has been shifted up by v = 0.25. For both values of α, the
reflection angle is equal to π2 for v0 = −v and decreases for
other values of v0 .
We assume an arbitrary periodicity of circle shift cor-
responding to a time the particle takes to travel from one
collision to another. Furthermore, change in position is in-
stantaneous. After the first collision, the circle will return
back to its original position with the center located at (0,0).
The trajectory of the particle moving in the shifting circle
is shown in Fig. 2(d). The initial conditions for the particle
are v0 = 0 and α = π4 , and the circle is shifted by v = 0.25.
For this particular set of parameters, the particle is fixed in a
period-two orbit after six collisions with the boundary of the
circle.
Finally, we introduce a more complex movement of the
FIG. 2. Billiard with moving boundaries: (a) configuration of the circle. The circle is shifted according to the following loop: it
system: reflection angle ϕ and collision point θ ; (b) shifted circle: is moved up (the “up” position, U), right (the “right” position,
distance of the shift v, angle α, and coordinate v0 ; (c) reflection angle
R), left (returned to the U position), and down (returned to
after shift v = 0.25; (d) trajectory of the particle for up-down shifting
the O position). Figure 2(e) shows the trajectory for a particle
of the circle; (e) trajectory of the particle for up-right-left-down
with the same initial conditions as in Fig. 2(d), only this time
shifting of the circle. In (d) and (e), O indicates the original position
(blue circle), U the “up” position (red circle), and the left position
it breaks away from the period-two orbit after the circle is
(green circle). shifted to the position R.

B. Model for plating with glass beads

(iii) if ϕ < π and is a rational multiple of π , then the particle
trajectory is periodic with the particle following the sides Let  denote the unit disk, and ∂ denote its boundary.
of some regular polygon; (iv) otherwise, if πϕ is irrational, Particle movement within a billiard is defined by the elastic
then the particle trajectory densely fills the ring between the collision rule: the angle of reflection is equal to the angle of
boundary of the unit circle and the boundary of a smaller circle incidence [14,23]. Denote by qt = (xt , yt ) the coordinates of
with radius cos2 (ϕ) [14]. For case (iii), the number of polygon the moving particle at time t, and by νt = (ut , wt ) its velocity
sides depends on the reflection angle. For example, if ϕ1 = π2 , vector. Then its position at time t + t can be computed by
the polygon is a square and the particle moves in period-four xt+t = xt + tut , (1)
orbit along the sides of the square; and for ϕ1 = π the particle
runs back and forth along a diameter of the circle and is locked yt+t = yt + twt . (2)
in period-two orbit.
Next we introduce a class of billiard—shifting billiard. The When the particle collides with the boundary ∂, its ve-
circle is shifted in such a way that its center moves distance locity vector ν gets reflected across the tangent line to ∂ at
v along the vertical axis, i.e., the coordinates of its center are the point of collision and the new postcollision velocity vector
(0, v). As the circle boundary is moved, the collision point can be computed as
also changes. For a particle moving along the x axis and νnew = νold − 2 νold , n n, (3)
perpendicular to the vertical axis, i.e., with θ = π and ϕ = π ,
the collision point becomes a for v > 0 (the circle shifted where n is the unit normal vector to the boundary and ν, n
up) or a for v < 0 (the circle shifted down), as shown in denotes the scalar product. For an arbitrary collision between
Fig. 2(b). We denote by α the angle between the trajectory of two particles, the postcollisional velocities are given by
the particle and the direction of the circle shift, (α ∈ [0, π ]).  old 
νnew
1 = νold
1 + 0.5(1 + r) ν2 − ν1
old
· kk, (4)
We also denote by v0 the coordinate where the line going
through the trajectory of the moving particle intercepts the  old 
νnew = νold
2 + 0.5(1 + r) ν2 − ν1
old
· kk, (5)
circle shift directory. Graphical representation of α and v0 is 2
shown in Fig. 2(b). If the angle α < π2 , this indicates that the where 0  r < 1 is the normal coefficient of restitution,
particle moves in the same direction as the circle has been which is typically 0.8 for glass beads [24]; ν · k denotes the
shifted. Otherwise, if the angle α > π2 , then the particle moves vector product, and k is the collision vector, directed from the
in the direction which is opposite to the direction that the center of the second particle to that of the first particle:
circle has been shifted. Potentially, v0 ∈ (−∞, ∞), but not all
α-v0 values will give trajectories contained within the moving k = (q1 − q2 )/q1 − q2 . (6)
circle boundaries. Allowed values of v0 are within the interval We have the following setup for the model: there are N
[v01 , v02 ], where v01 = v − sin(α)−1 if α  π2 , v01 = v − 1, if glass beads which at a time t are randomly distributed within
α > π2 ; and v02 = 1 if α  π2 , v02 = sin(α)−1 if α > π2 . a unit circle in such a way that they do not overlap, i.e., the
As the point of the collision changes, so does the angle Euclidean distance between beads centers is larger than 2d.
of the reflection. Figure 2(c) shows the reflection angle as a We assumed that the plating dish is shifted a unit distance (i.e.,

052410-3
ALIDIVINAS PRUSOKAS et al. PHYSICAL REVIEW E 103, 052410 (2021)

corresponding to a diameter of the plate) within each step of TABLE I. Results of the dilution experiment.
a loop. All particles are stationary at the start of a simulation,
and start moving after collision with the moving boundary ∂. Five loops Ten loops
For an arbitrary time unit T = 1, we further assume that the Metrics 10−5 10−6 10−7 10−5 10−6 10−7
boundary is moved with a velocity V . Simulations are run at
discrete time intervals with a fixed time step t = 0.01 × T . nCFUs 964 286 59 1050 382 71
At each time step, we calculate if any particle is close enough dNB 1.3 2.1 8.5 1.5 2.0 8.3
to the boundary, or any two particles are close to each other, CVS 0.59 0.75 1.2 0.45 0.62 0.97
and create a list of collision events. After each collision, the mS 7.76 3.39 1.09 8.77 3.59 1.34
direction and velocity of the involved particle are updated.
We assume that velocities of particles are reduced with each
time state by a constant, cv , due to the presence of growth Parameter inference was performed using an adaptive mul-
media. After running pilot simulations, we found that values tiple importance sampling framework [25]. The likelihood
V = 3 and cv = 0.99 produced trajectories consistent with the was obtained by assuming that the observed number of
experiment data. During simulations we track the position of colonies is Poisson distributed:
each particle with respect to the center of the plating dish.

n
 i  i 
Next, we formulate a stochastic spatially explicit model of L( |nobs ) = Poisson nobs n , (9)
cell dispersal via the movement of glass beads. At time t = 0 sim
i=1
there are n viable cells in suspension which is dispensed onto
the middle of the plate. The positions of cells within the plate where = {n, ron , roff }; nobs is the observed number of
are modeled using a Gaussian distribution with mean at the colonies; nsim is the number of colonies obtained after sim-
center of the plate and variance equal to half of the plate’s ulating the model with parameter set .
radius. Next, we considered two stochastic events: (i) a cell
is attached to a bead which is passing within d/2 distance
V. RESULTS
from the cell; and (ii) a cell is detached from the bead and
is deposited on the surface of the plate. We assume that the First, the plating was carried out with different concentra-
probabilities of these events at time t are tions of E. coli cells. We used the L-shape movement with
ten glass beads, and either five or ten loops. The results of
experiment are shown in Table I. As cell concentration de-
pon (t ) = exp(−ront ), (7) creased, the mean distance between the closest CFUs, nCFUs ,
and spatial variability of colonies, CVS , increased. The num-
poff (t ) = exp(−rofft ). (8) ber of colonies was 950–1050 for 10−5 dilution, 280–380 for
10−6 dilution, and 50–70 for 10−7 dilution. It is suggested that
After simulating the dispersal of cells, we calculated which for manual counting, the best range of a number of colonies
cells were within 1.5 μm of other cells, forming clusters. We should be between 30 and 300 [26]. Therefore, for further
further assumed that detectable individual CFUs are formed analysis we have chosen 10−6 dilution as this gives a good
by either separated cells or these clusters. distribution of colonies.

FIG. 3. Experimental results. Methods analyzed are as follows: hourglass (HG: ←), L shape (L: ↑→←↓), and up-down (UD: ↑↓).
Each row corresponds to a particular experimental setup. The first two columns show the number of beads used and the number of loops in
each experiment. Red arrows indicate direction from less uniform to more uniform distribution of CFUs.

052410-4
EFFECTIVENESS OF GLASS BEADS FOR PLATING CELL … PHYSICAL REVIEW E 103, 052410 (2021)

FIG. 4. Experimental images of E. coli CFUs and density his-

tograms of simulated particle trajectories for the up-down movement.
Columns correspond to a different number of glass beads. Blue
circles indicate the initial positions of particles.

Figure 3 displays a summary from the experiments: each

row corresponds to a different method of cell plating, number
of glass beads, and number of loops. We have analyzed 40 FIG. 6. Density histograms of simulated particle trajectories for
configurations in total. We observed that the number of CFUs the L-shape movement. Columns correspond to a different number
increased when the number of loops increased from 5 to 25. of glass beads, and rows correspond to a different number of loops.
Most of the experiments produced a consistent number of Blue circles indicate the initial positions of particles.
colonies. The exceptions were for cells spread using either
100 loops, or 25 beads and 25 loops, which gave much lower spatial variability, CVS ; however, the number of CFUs was not
colony numbers. Methods producing the smallest amount of as high as for L-shape or up-down trajectories. Interestingly,
CFUs also had the largest values of dNB and CVS . the most natural way to move plates, i.e., wrist movement,
We could identify three groups of configurations, which was not performing as well as the other methods of plate
produced nCFUs > 300: (i) L-shape movement with 2–10 movement.
beads and 25 loops; (i) L-shape movement with 10–25 beads Next, we used the model of shifting billiard to investi-
and 5 loops; and (iii) up-down movement with 3–25 beads gate the observed spatial patterns in the up-down movement.
and 50 loops. However, the latter configuration performed Figure 4 shows experimental images of E. coli CFUs and
quite poorly with respect to other metrics, especially dNB and density histograms of simulated particle trajectories for the
mS . In terms of the number of steps the plate is moved, 50 up-down movement. For model simulations with N beads, we
up-down moves are equivalent to 25 L-shape moves. The have added an extra bead N, but kept the initial positions of
mean distance to the nearest CFUs for the up-down method the 1 . . . (N − 1) beads the same as in simulations with N − 1
had the smallest values overall. The hourglass movement has beads. It can be seen then even with 50 loops, the trajectories
performed well for 10 loops and 3–25 glass beads in terms
of the values of nCFUs , and had one of the lowest values of

FIG. 7. Stochastic model of cell dispersal. Log-likelihood values

of sampled parameters: gray shows values less than −300; red di-
FIG. 5. Experimental images of E. coli CFUs for the L-shape amond shows MLE. (b)–(d) Comparison between simulations and
movement. Columns correspond to a different number of glass beads, experimental data for L-shape trajectory with (b) two beads, (c) five
and rows correspond to a different number of loops. beads, and (d) ten beads.

052410-5
ALIDIVINAS PRUSOKAS et al. PHYSICAL REVIEW E 103, 052410 (2021)

of beads follow the movement of the plate. This mimics ex- cells getting stuck on beads and hence removed during the
actly how the CFUs are positioned in experiments, i.e., along debeading process [27].
and parallel to a vertical axis. As the number of beads is
increased, there is a possibility of a glass bead escaping the
VI. CONCLUSIONS
expected trajectory due to a random collision with other glass
beads. In this work we characterize the effectiveness of glass
Furthermore, we used the model to explain heterogeneities beads for plating cell cultures. We examined the number of
in the observed spatial patterns when using L-shape move- colonies, the mean distance between closest colonies, and the
ment. Figure 5 shows representative images of E. coli CFUs variability and uniformity of spatial distribution of colonies.
and Figure 6 shows corresponding density histograms of sim- Our results indicate that the Copacabana method is highly
ulated particle trajectories for the L-shape movement. It can efficient, although care needs to be taken when choosing the
be seen from the simulations, that when the number of glass trajectory of plate movement, number of beads, and number
beads or a number of loops is low, the surface of the plate of loops.
cannot be fully explored by the moving beads, and particular An exploratory look at the experimental data revealed some
areas (top left quarter) are hardly visited by any of the glass interesting attributes that required further investigation using
beads at all. In order to improve surface coverage, it is neces- a mathematical modeling approach. We introduced shifting
sary to increase either the number of beads or the number of billiard as a conceptual abstraction of cell plating with glass
loops. For example, 5 glass beads with 25 loops, or 10 glass beads. Numerical analysis of the dynamics of this simple yet
beads and 10 loops, produced a reasonable coverage of the efficient billiard indicated a close relationship between plate
plate surface. movement and quality of colony distribution. Simulations
Finally, we have fitted the stochastic model of cell dispersal show that when the number of glass beads or the number of
to experimental data. After a pilot investigation, we have fixed loops is low, the surface of the plate cannot be fully explored
n = 500. Sampled parameter sets {1/ron , 1/roff } and corre- by the moving beads. This could be improved by increasing
sponding log-likelihood values are shown in Fig. 7(a). We either the number of beads or the number of loops. However,
have calculated the maximum likelihood estimate (MLE) as a very high number of loops (i.e., 100) should be avoided,
ron = 0.010 416 667 and roff = 0.035 714 29 (shown as red as experimental data and simulations showed that the number
diamond). Using this parametrized model we simulated cell of colonies produced was much lower compared to other
plating for an L-shape trajectory with two, five, and ten configurations.
beads and the number of loops in the range from 5 to 100 We have analyzed the properties of cell spread using dig-
[Figs. 7(b)–7(d)]. We have found that 20–30 loops give a itized images and mathematical modeling. Optical density
maximum number of nCFUs . (OD) readings have been used to evaluate the number of CFUs
Our full model agreed with experimental observations of based on the time it takes to reach the predetermined OD
the decrease in the number of colonies when using a very of 0.1 when compared with a reference standard curve [28].
high number of loops. Similar phenomena have been observed However, this technique would not provide the necessary de-
when plating central memory cells, which was attributed to tails for the spatial analysis we performed.

FIG. 8. Flowchart representing image processing. First, the image was analyzed to detect where the plate is. Second, positions of CFUs
were detected by fitting centroids.

052410-6
EFFECTIVENESS OF GLASS BEADS FOR PLATING CELL … PHYSICAL REVIEW E 103, 052410 (2021)

In conclusion, we recommend using an L-shape trajectory APPENDIX: IMAGE ANALYSIS

with two to ten beads and 25 loops which should produce good
Each image has been binarized with a low threshold in
colony separation in terms of the number and spatial spread
order to find the boundaries of a plate; this gives scaling
of colonies. Recently, automated systems have been utilized
for transforming pixels into millimeters (Fig. 8). Next, we
for high-throughput cell handling and analysis [29]. Manual
extracted single-channel images corresponding to each of the
transformation and spreading on agar plates are still used
color channels in the image. We have found that working with
in otherwise fully automated liquid handling platforms [30].
the image in the red channel improved the detection of CFUs.
Therefore, we believe that our results should be of interest
Finally, we have fitted centroids representing CFUs by varying
when using automated platforms and could guide program-
a threshold of binarization.
ming of automatic cell plating protocols.

[1] F. Goetz, T. Bannerman, and K. Schleifer, The genera Staphy- [16] M. Fließer, G. J. O. Schmidt, and H. Spohn, Magnetotransport
lococcus and Macrococcus, in The Prokaryotes, edited by M. of the Sinai billiard, Phys. Rev. E 53, 5690 (1996).
Dworkin, S. Falkow, E. Rosenberg, K. H. Schleifer, and E. [17] V. Lopac, I. Mrkonjić, and D. Radić, Chaotic dynamics and
Stackebrandt (Springer, New York, 2006), Vol. 4, pp. 5–75. orbit stability in the parabolic oval billiard, Phys. Rev. E 66,
[2] B. Wognum, N. Yuan, B. Lai, and C. Miller, Colony forming 036202 (2002).
cell assays for human hematopoietic progenitor cells, Methods [18] J. Nagler, M. Krieger, M. Linke, J. Schonke, and J.
in Molecular Biology (Springer, New York, 2011), Vol. 946, Wiersig, Leaking billiard, Phys. Rev. E 75, 046204
pp. 267–283. (2007).
[3] K. Horie, C. Kokubu, J. Yoshida, K. Akagi, A. Isotani, A. [19] C. P. Dettmann and O. Georgiou, Transmission and reflection
Oshitani, K. Yusa, R. Ikeda, Y. Huang, A. Bradley, and J. in the stadium billiard: Time-dependent asymmetric transport,
Takeda, A homozygous mutant embryonic stem cell bank ap- Phys. Rev. E 83, 036212 (2011).
plicable for phenotype-driven genetic screening, Nat. Methods [20] D. M. Naplekov and V. V. Yanovsky, Billiard with a handle,
8, 1071 (2011). Phys. Rev. E 94, 042225 (2016).
[4] T. Strachan and A. Read, Amplifying DNA: Cell-based DNA [21] L. Zarfaty, A. Peletskyi, E. Barkai, and S. Denisov, Infi-
cloning and PCR, Human Molecular Genetics, 4th ed. (Garland nite horizon billiards: Transport at the border between Gauss
Science, New York, 2011), pp. 163–190. and Levy universality classes, Phys. Rev. E 100, 042140
[5] C. Gianyecchio, N. Lozano, C. Henderson, P. K. A. Bullivant, (2019).
A. Valencia, L. Su, G. Bello, M. Wong, E. Cook, L. Fuller, J. [22] S. Luria and J. Burrous, Hybridization between E. coli and
Neal, and P. Yeh, Variation in mutant prevention concentrations, shigella, J. Bacteriol. 74, 461 (1957).
Front. Microbiol. 10, 42 (2019). [23] M. Berry, Regularity and chaos in classical mechanics, illus-
[6] M. Ragheb, M. Thomason, C. Hsu, P. Nugent, J. Gage, A. trated by three deformations of a circular billiard, Eur. J. Phys.
Samadpour, A. Kariisa, C. Merrikh, S. Miller, and D. S. H. 2, 91 (1981).
Merrikh, Inhibiting the evolution of antibiotic resistance, Mol. [24] A. Lorenz, C. Tuozzolo, and M. Louge, Measurements of im-
Cell 73, 157 (2019). pact properties of small, nearly spherical particles, Exp. Mech.
[7] J. Hagel, L. Garner, M. Bilton, H. Mehta, T. Leng, C. Hackstein, 37, 292 (1997).
P. Phalora, A. Amini, H. Akther, N. Provine, M. Edmans, C. [25] R. Retkute, P. Touloupou, M. Basanez, D. Hollingsworth,
Willberg, and P. Klenerman, Human MAIT cell activation in and S. E. Spencer, Integrating geostatistical maps and trans-
vitro, MAIT Cells (Springer, New York, 2020), pp. 97–123. mission models using adaptive multiple importance sampling
[8] M. Jennison, The relations between plate counts and direct medRxiv:2020.08.03.20146241 (2020), doi: https://doi.org/10.
microscopic counts of Escherichia coli during the logarithmic 1101/2020.08.03.20146241.
growth period, J. Bacteriol. 33, 461 (1937). [26] D. Tomasiewicz, D. Hotchkiss, G. Reinbold, R. Read, and P.
[9] J. Cepl, A. Blahuskova, Z. Neubauer, and A. Markos, Variations Hartman, The suitable number of colonies on plates for count-
and heredity in bacterial colonies, Commun. Integr. Biol. 9, ing, J. Food Prot. 43, 282 (1980).
e1261228 (2016). [27] B. Aksoy, E. Czech, C. Paulos, and J. Hammerbacher, Com-
[10] D. Hartman, Perfecting your spread plate technique, putational and experimental optimization of T cell activation
J. Microbiol. Biol. Educ. 12, 204 (2011). bioRxiv:629857 (2019), doi: https://doi.org/10.1101/629857.
[11] E. Sanders, Aseptic laboratory techniques: Plating methods, [28] T. Ben-Yehezkel, S. Nagar, D. Mackrants, Z. Marx, G. Linshiz,
J. Visualized Exp. 63, e3064 (2012). U. Shabi, and E. Shapiro, Computer-aided high-throughput
[12] M. Worthington, R. Luo, and J. Pelo, Copacabana method for cloning of bacteria in liquid medium, Biotechniques 50, 124
spreading E. Coli and yeast colonies, Biotechniques 30, 738 (2011).
(2001). [29] E. Whitehead, F. Rudolf, H. Kaltenbach, and J. Stelling, Auto-
[13] K. Mills, J. Gareau, and A. Garcia, Low-cost modification to mated planning enables complex protocols on liquid-handling
the Copacabana method for spreading transformation mixtures, robots, ACS Synth. Biol. 7, 922 (2018).
Biotechniques 39, 188 (2005). [30] S. Hans, M. Gimpel, F. Glauche, P. Neubauer, and M.
[14] N. Chernov and R. Markarian, Chaotic Billiards (American Cruz-Bournazou, Automated cell treatment for competence
Mathematical Society, Providence, RI, 2006). and transformation of Escherichiacoli in a high-throughput
[15] W. Bauer and G. F. Bertsch, Decay of Ordered and Chaotic quasi-turbidostat using microtiter plates, Microorganisms 6, 60
Systems, Phys. Rev. Lett. 65, 2213 (1990). (2018).

052410-7