The Pennsylvania State University

The Graduate School

Department of Chemical Engineering

PURIFICATION OF PLASMID DNA USING ULTRAFILTRATION MEMBRANES

A Dissertation in

Chemical Engineering

by

Ying Li

© 2017 Ying Li

Submitted in Partial Fulfillment

of the Requirements

for the Degree of

Doctor of Philosophy

August 2017

ii

The dissertation of Ying Li was reviewed and approved* by the following:

Andrew L. Zydney

Distinguished Professor of Chemical Engineering

Dissertation Advisor

Chair of Committee

Manish Kumar

Associate Professor of Chemical Engineering

Janna Maranas

Professor of Chemical Engineering Graduate Program Coordinator of the Department of Chemical Engineering

William Hancock

Professor of Bioengineering and Biomedical Engineering

*Signatures are on file in the Graduate School

iii

ABSTRACT

Previous studies have demonstrated that plasmid transmission through ultrafiltration

membranes can be controlled by adjusting the filtrate flux thereby controlling the extent of

plasmid elongation. This phenomenon can be used for the separation of different plasmid

isoforms by exploiting the differences in flexibility of the supercoiled, linear, and open-circular

isoforms. However, there are a number of critical challenges that still need to be overcome in

order to apply these novel membrane-based processes for commercial scale manufacture of DNA

for gene therapy applications and for use as DNA-based vaccines. The overall objectives of this

dissertation are to develop novel strategies to enhance the separation resolution during

ultrafiltration of different plasmid isoforms and control membrane fouling during ultrafiltration

of concentrated DNA solutions.

This work first focused on developing the strategy of pre-conditioning, accomplished by

pre-elongating the DNA by passage through a region with large pore size, to minimize fouling

and enhance DNA separations. Data were obtained using both asymmetric hollow fiber

membranes, with flow in either the normal or reverse orientation, and with composite membrane

structures made by placing a larger pore size flat sheet microfiltration membrane in series with

an ultrafiltration membrane. In all cases, flow through the larger pore size region pre-stretched

the plasmid, leading to an increase in plasmid transmission and a significant reduction in fouling.

This pre-conditioning also provided a significant increase in selectivity for separation of the

linear and supercoiled isoforms. The performance of composite membrane system can be

optimized by controlling the pore size and morphology of the microfiltration membranes.

iv

This work also examined the effects of ionic conditions (including solution ionic strength

and ion type) on separation of the different plasmid isoforms. The transmission of the linear and

open-circular isoforms slightly increased with increasing solution ionic strength (NaCl or MgCl2

concentration) due to shielding of intramolecular electrostatic interactions. The effect of ionic

strength was greatest for the supercoiled plasmid due to changes in its plectonemic structure,

providing opportunities for enhanced purification of this therapeutically active isoform.

Polycation spermine was found to induce DNA condensation at a threshold concentration, above

which transmission of the plasmid DNA dropped rapidly with the membranes becoming nearly

completely retentive to the plasmid. DNA condensation was reduced in the presence of high

concentrations of monovalent salts, potentially providing an opportunity to “tune” the

transmission of the DNA isoforms by proper of addition of spermine and NaCl to the solutions.

Solution conditions also have a significant effect on the fouling characteristics of

supercoiled plasmid DNA isoforms with different numbers of base pairs. Sieving coefficient and

filtrate flux data were analyzed using a model based on the partial blockage of the membrane

pores by trapped plasmids. Fouling increased dramatically at low ionic strength, with the flux

decline parameter for the 3.0 kbp plasmid in a 1 mM NaCl solution being an order of magnitude

greater than that in a 10 mM solution. Fouling was also most pronounced for the larger 16.8 kbp

plasmid, consistent with the greater probability of plasmid trapping at the pore entrance.

Ultrafiltration membranes also have the potential to separate supercoiled plasmids based

on differences in their size (i.e., number of base pairs). An up to 30-fold selectivity between 3.0

and 16.8 kbp plasmids was achieved using commercial ultrafiltration membranes. The reduction

in transmission of the supercoiled plasmids with increasing chain length was a direct result of the

morphology of the supercoiled isoform; no significant affect of plasmid size was seen during

v

ultrafiltration of linear versions of the same plasmids. The supercoiled isoforms adopt a branched

structure due to the under-twisting of the DNA, with the number of branches increasing, and the

DNA transmission decreasing, with increasing chain length.

It is anticipated that the results from this study will provide important information needed

for successfully implementing UF processes into commercial systems for the large-scale

manufacture of therapeutic DNA products.

vi

TABLE OF CONCENTS

LIST OF FIGURES ..................................................................................................................... ix

LIST OF TABLES ..................................................................................................................... xiv

ACKNOWLEDGEMENTS ....................................................................................................... xv

Chapter 1 Introduction................................................................................................................. 1

1.1 DNA therapeutics and plasmid DNA .................................................................................... 1

1.2 Downstream processing of plasmid DNA ............................................................................. 8

1.3 Purification of plasmid DNA using membrane processes .................................................. 11

1.4 Challenges in membrane ultrafiltration of plasmid DNA ................................................... 15

1.5 Dissertation outline ............................................................................................................. 16

Chapter 2 Materials and Methods............................................................................................. 19

2.1 Plasmid DNA ...................................................................................................................... 19

2.1.1 Supercoiled plasmids .................................................................................................... 19

2.1.2 Plasmid isoforms .......................................................................................................... 20

2.2 Buffers ................................................................................................................................. 21

2.3 DNA characterization .......................................................................................................... 23

2.3.1 Agarose gel electrophoresis .......................................................................................... 23

2.3.2 PicoGreen Assay........................................................................................................... 24

2.3.3 NanoDrop Spectrophotometry ...................................................................................... 25

2.4 Membranes .......................................................................................................................... 26

2.5 Membrane characterization ................................................................................................. 32

2.5.1 Membrane hydraulic permeability ................................................................................ 32

2.5.2 Scanning electron microscopy (SEM) .......................................................................... 32

2.6 Sieving experiments ............................................................................................................ 33

2.6.1 Normal flow filtration ................................................................................................... 33

2.6.2 Tangential flow filtration .............................................................................................. 34

2.7 Diafiltration ......................................................................................................................... 35

Chapter 3 Use of Pre-conditioning to Control Membrane Fouling and Enhance Membrane

Performance .......................................................................................................................... 37

vii

3.1 Introduction ......................................................................................................................... 37

3.2 Materials and methods ........................................................................................................ 39

3.3. Results and discussion ........................................................................................................ 40

3.3.1 Membrane Orientation .................................................................................................. 40

3.3.2 Effects of plasmid and membrane pore size ................................................................. 47

3.3.3 Isoform separation ...................................................................................................... 49

3.4 Conclusion ........................................................................................................................... 51

Chapter 4 Preconditioning with Multilayer Composite Membranes..................................... 53

4.1 Introduction ......................................................................................................................... 53

4.2 Materials and methods ........................................................................................................ 56

4.3 Results and Discussions ...................................................................................................... 59

4.3.1 Multilayer composite membrane .................................................................................. 59

4.3.2 Effects of upper layer structure .................................................................................... 61

4.3.3 Physical model .............................................................................................................. 69

4.4 Conclusions and recommendations ..................................................................................... 74

Chapter 5 Enhanced Plasmid DNA Purification by Exploiting Ionic Strength Effects ....... 76

5.1 Introduction ......................................................................................................................... 76

5.2 Materials and Methods ........................................................................................................ 78

5.3 Results and Discussion ........................................................................................................ 78

5.3.1 Linear plasmid .............................................................................................................. 78

5.3.2 Open-circular plasmid ................................................................................................. 81

5.3.3 Isoform separation ....................................................................................................... 83

5.3.4 Physical Interpretation ................................................................................................. 89

5.4 Conclusions ........................................................................................................................ 91

Chapter 6 Effects of Multivalent Salts on Plasmid DNA Ultrafiltration ............................... 92

6.1 Introduction ......................................................................................................................... 92

6.2 Materials and methods ........................................................................................................ 95

6.3 Results and Discussions ...................................................................................................... 95

6.3.1 Supercoiled isoform ...................................................................................................... 95

6.3.2 Effects of monovalent salts ........................................................................................... 98

6.3.3 Isoform separation ...................................................................................................... 101

viii

6.4 Conclusions and recommendations ................................................................................... 104

Chapter 7 Effect of Ionic Strength on Membrane Fouling During Ultrafiltration of Plasmid

DNA ...................................................................................................................................... 107

7.1 Introduction ....................................................................................................................... 107

7.2 Materials and methods ...................................................................................................... 109

7.3 Results and discussions ..................................................................................................... 110

7.3.1 Fouling experiments ................................................................................................... 110

7.3.2 Fouling model ............................................................................................................. 114

7.3.3 Effects of plasmid size ................................................................................................ 117

7.3.4 Effects of membrane pore size ................................................................................... 121

7.4 Conclusions ....................................................................................................................... 125

Chapter 8 Size-based Separation of Supercoiled Plasmid DNA using Ultrafiltration ....... 127

8.1 Introduction ....................................................................................................................... 127

8.2 Materials and methods ...................................................................................................... 130

8.3 Results and discussion ....................................................................................................... 130

8.3.1 Supercoiled plasmids ................................................................................................. 130

8.3.2 Linear plasmids........................................................................................................... 137

8.3.3 Physical interpretation ................................................................................................ 138

8.4 Conclusions ....................................................................................................................... 142

Chapter 9 Conclusions and Recommendations for Future Work ........................................ 144

9.1 Conclusions ....................................................................................................................... 144

9.2 Future recommendations ................................................................................................... 149

Bibiography ............................................................................................................................... 153

ix

LIST OF FIGURES

Figure 1.1: Electron microscopy image of different plasmid DNA isoforms (adopted from

https://commons.wikimedia.org/wiki/File%3APlasmid_emEN.jpg) ............................................. 4

Figure 1.2: Structure of supercoiled plasmid DNAs (reproduced from17) ........................................... 7

Figure 1.3: A) The impurity content in bacteria cell lysate, B) final product purity requirement

specified by guidelines from FDA .................................................................................................. 9

Figure 1.4: Typical industrial scale pDNA purification process ....................................................... 10

Figure 2.1: Molecular structure and physical properties of Tris-HCl and EDTA ............................. 23

Figure 2.2: AGE image for plasmid with different sizes (left panel) and isoforms (right panel) ...... 24

Figure 2.3: SEM of the cross section of A) UltracelTM composite regenerated cellulose and B)

BiomaxTM polyethersulfone membranes (both from MilliporeSigma) and C) modified

polyethersulfone hollow fiber membrane (from Spectrum Labs). Images taken from

manufacturers’ websites................................................................................................................ 28

Figure 2.4: Molecular structure of polysulfone, polyethersulfone, and regenerated cellulose .......... 29

Figure 2.5: Schematic of stirred cell set-up used for ultrafiltration (adopted from Latulippe, 2010)

....................................................................................................................................................... 34

Figure 2.6: Schematic of hollow fiber setup for tangential flow filtration ........................................ 35

Figure 3.1: Use of microfluidic devices to pre-condition DNA.A) Stretching by shear flow

(reproduced from Perkins et al., 2007)73; B) Pre-stretching using obstacle arrays (reproduced

from Chan et al., 2004)74; C) Pre-stretching by conical-shaped microchannel (adopted from

Larson et al., 2006)70 ..................................................................................................................... 38

Figure 3.2: Observed sieving coefficient of the supercoiled 16.8 kbp supercoiled plasmid through

a 500 kDa and 50 kDa hollow fiber membrane in normal orientation. Data were obtained at

a feed flow rate of 100 mL/min using a plasmid concentration of 0.25 µg/mL ........................... 41

Figure 3.3: Observed sieving coefficients (upper panel) and filtrate flux (lower panel) during

constant pressure ultrafiltration of a 3 µg/mL solution of the supercoiled 16.8 kbp plasmid

through a 500 kDa hollow fiber membrane in both the lumen-to-shell (forward) and shell-to-

lumen (reverse) orientations ......................................................................................................... 43

Figure 3.4: Observed sieving coefficient of the supercoiled 16.8 kbp supercoiled plasmid through

a 500 kDa hollow fiber membrane in the reverse and forward orientations. Data were obtained

at a feed flow rate of 100 mL/min using a plasmid concentration of 0.25 µg/mL ....................... 45

x

Figure 3.5: Schematic diagram of plasmid transmission through an asymmetric membrane in the

(A) forward and (B) reverse orientations. ..................................................................................... 47

Figure 3.6: Observed sieving coefficients of the supercoiled 3.0 kbp (p-EMP) and 16.8 kbp (p-

FDY) plasmids through the 500 kDa and 50 kDa PS hollow fiber membranes in the reverse

orientation. .................................................................................................................................... 49

Figure 3.7: Observed sieving coefficients for the linear and supercoiled isoforms of the 3.0 kbp

plasmid through the 50 kDa hollow fiber membrane in the reverse (left panel) and forward

(right panel) orientations ............................................................................................................... 50

Figure 3.8: Agarose gel electrophoresis image of the feed and filtrate samples during

ultrafiltration of a binary mixture of the linear and supercoiled isoforms of the 3.0 kbp plasmid.

Lane 1 - linear 1 kbp DNA ladder; Lane 2 - feed sample; Lane 3 - filtrate sample at Jv = 10

μm/s; Lane 4 - filtrate sample at Jv = 20 μm/s............................................................................... 51

Figure 4.1: Comparison of A) isotropic B) anisotropic and C) composite membrane structure in

sterile filters (reproduced from 79) ................................................................................................ 54

Figure 4.2: Scanning electron microscopy images of the upper surface of the Durapore, Mixed

Cellulose Ester (compiled from https://www.emdmillipore.com/US/en/product/MF-

Millipore), and Nuclepore membranes ......................................................................................... 57

Figure 4.3: Schematic of etching apparatus for preparation of polycarbonate membranes with

conical shape pores ....................................................................................................................... 59

Figure 4.4: Effects of pre-conditioning with different pore size Durapore microfiltration

membranes (upper layers) on transmission of the supercoiled 3.0 kbp plasmid through the

composite membranes. Experiments were performed in TE buffer containing 100 mM NaCl,

with an UltracelTM 100 kDa membrane as the bottom layer ......................................................... 62

Figure 4.5: Sieving coefficient for the supercoiled 3.0 kbp plasmid through composite

membranes as a function of the pore size of the upper layer Durapore microfiltration

membranes at filtrate flux of 40, 60 and 80 µm/s ......................................................................... 63

Figure 4.6: Effects of pre-conditioning with different pore sizes of mixed cellulose ester (MF)

microfiltration membranes on the transmission of the supercoiled 3.0 kbp plasmid through the

composite membranes. .................................................................................................................. 64

Figure 4.7: Effects of pre-conditioning with different pore size Nuclepore microfiltration

membranes on transmission of the supercoiled 3.0 kbp plasmid through the composite

membranes .................................................................................................................................... 65

Figure 4.8: Effects of pre-conditioning with different pore size upper (Durapore) membranes on

transmission of the supercoiled 9.8 kbp plasmid through the composite membranes .................. 68

xi

Figure 4.9: Selectivity of supercoiled 3.0 kbp plasmid over open-circular plasmid through

different composite membranes. Experiments were performed in TE buffer containing 100

mM NaCl, with UltracelTM 100 kDa membrane as bottom layer ................................................. 69

Figure 4.10: Physical model for DNA transport through different pore morphologies A) conical

shape B) gradual transition funnel shape and C) sudden onset funnel shape ............................... 70

Figure 4.11: SEM images of 0.015 µm Nuclepore membranes etched for A) 20 B) 30 and C)

40min. Images show membrane surfaces in the etching solution ............................................... 701

Figure 5.1: Effect of NaCl concentration on the observed sieving coefficients for ultrafiltration

of the linear 3.0 kbp plasmid through the UltracelTM 100 kDa membrane ................................... 79

Figure 5.2: Effect of ion valence and concentration on the observed sieving coefficient for

ultrafiltration of the linear 3.0 kbp plasmid through the UltracelTM 100 kDa membrane ............ 81

Figure 5.3: Transmission of the open circular isoform of the 3.0 kbp plasmid through the

UltracelTM 100 kDa membrane in the presence of 10, 150, or 300 mM NaCl ............................. 82

Figure 5.4: Effect of solution ionic strength on transmission of the open circular 3.0 kbp plasmid

through the UltracelTM 100 kDa membrane at a filtrate flux of 140 ± 5 μm/s. ............................. 83

Figure 5.5: Effect of solution ionic strength on transmission of the linear, open-circular, and

supercoiled 3.0 kbp plasmid through the UltracelTM 100 kDa membrane at a filtrate flux of

140 µm/s ....................................................................................................................................... 84

Figure 5.6: Selectivity between the linear and supercoiled isoforms (top panel) and between the

supercoiled and open-circular isoforms (bottom panel) of the 3.0 kbp plasmid using the

Ultracel™100 kDa membrane ...................................................................................................... 86

Figure 5.7: Agarose gel electrophoresis showing the separation of a binary mixture of the linear

and supercoiled isoforms in TE buffer containing 10 mM NaCl. Lane 1: linear 1 kbp DNA

ladder. Lane 3: feed sample. Lanes 2 and 4: filtrate samples collected using the Ultracel 100

kDa membrane at a filtrate flux of 80 μm/s .................................................................................. 88

Figure 5.8: Agarose gel electrophoresis showing the separation of a binary mixture of the open-

circular and supercoiled isoforms in TE buffer containing 150 mM NaCl. Lane 4: linear 1 kbp

DNA ladder. Lane 1: feed sample. Lanes 2: filtrate samples collected at filtrate flux of 100

μm/s. Lanes 3: filtrate samples collected at filtrate flux of 125 μm/s. .......................................... 89

Figure 6.1: Chemical structure of spermidine and spermine. The basic amino (NH2) groups bind

protons at physiological pH to become positively charged .......................................................... 93

Figure 6.2: Observed sieving coefficients for the supercoiled 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of filtrate flux for experiments performed with 0, 2, 10, 15,

and 30 µM spermine in TE buffer containing 10 mM NaCl. ....................................................... 97

xii

Figure 6.3: Effect of spermine concentration on the observed sieving coefficients for the

supercoiled 9.8 kbp plasmid through 300 kDa Biomax membranes at a filtrate flux of 50 µm/s.

Data obtained in TE buffer containing 10 mM NaCl. .................................................................. 98

Figure 6.4: Observed sieving coefficients for the supercoiled 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of the filtrate flux for experiments performed with various

concentrations of spermine added to TE buffer containing 100 mM (upper panel) a and 1 mM

(lower panel) NaCl ...................................................................................................................... 100

Figure 6.5: Observed sieving coefficients for the supercoiled 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of the filtrate flux for experiments performed with 10 µM

spermine added to TE buffer containing 1, 10 and 100 mM NaCl. ............................................ 101

Figure 6.6: Observed sieving coefficients for the linear 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of the filtrate flux for experiments performed with 0, 2, 6,

10, 15 µM spermine added to TE buffer containing 10 mM NaCl. ............................................ 103

Figure 6.7: Observed sieving coefficients for the open-circular 9.8 kbp plasmid through a 300

kDa Biomax membrane as a function of the filtrate flux for experiments performed with

various concentrations of spermine added to TE buffer. Left - solutions contained 10 mM

NaCl, Right - solutions contained 100 mM NaCl ....................................................................... 104

Figure 6.8: Selectivity for separation of the supercoiled and open-circular 9.8 kbp plasmids

through a 300 kDa Biomax membrane as a function of spermine concentration for experiments

performed at filtrate flux of 50 µm/s in TE buffer containing 10 and 100 mM NaCl. ............... 104

Figure 7.1: Effect of solution ionic conditions on the sieving coefficients (top panel) and filtrate

flux (bottom panel) during ultrafiltration of solutions of the 16.8 kbp supercoiled plasmid

through 300 kDa Biomax membranes. Data were obtained at plasmid concentrations of 3×10-

3 kg/m3 using TE buffer with 1, 10 or 500 mM NaCl. Dashed curves are model calculations

using parameter values given in Table 7.1.................................................................................. 111

Figure 7.2: Ultrafiltration of a 3×10-3 kg/m3 solution of the 16.8 kbp linear plasmid through a 300

kDa Biomax membrane in 1 mM NaCl TE buffer. .................................................................... 114

Figure 7.3: Effect of solution ionic strength on the sieving coefficient (top) and filtrate flux

(bottom) of different size supercoiled plasmids through 300 kDa Biomax membranes. Data

were obtained with 3×10-3 kg/m3 solutions of the 16.8, 9.8 and 3.0 kbp plasmid wi ith 3×10-3

kg/m3 solutions of the 16.8, 9.8 and 3.0 kbp plasmid with 1 and 10 mM NaCl in TE buffer.

Dashed curves are model calculations using parameter values specified in Table 7.1.

118

Figure 7.4: Fouling rate constant k plotted versus radius of gyration of plasmid DNA. Data

include 3.0 and 9.8 kbp supercoiled plasmids in TE buffer containing 1 and 10 mM NaCl, and

16.8 kbp supercoiled plasmids in TE buffer containing 1, 10, and 500 mM NaCl. Fouling

xiii

experiments were performed with Biomax 300 kDa membranes. Error bars are within the size

of the symbols ............................................................................................................................. 120

Figure 7.5: Effect of solution ionic strength on the sieving coefficient (top) and filtrate flux

(bottom) of the supercoiled plasmid through membranes with different MWCO. Data obtained

with 5x10-3 kg/m3 solutions of the 16.8 kbp plasmid with 1 and 10 mM NaCl in TE buffer

using Biomax 300 and 1000 kDa membranes. Dashed curves are model calculations using

parameter values in Table 7.1 ..................................................................................................... 123

Figure 8.1: Observed sieving coefficients of the 3.0, 9.8, and 16.8 kbp supercoiled plasmids

through the 100 kDa Biomax membrane in TE buffer with 300 mM NaCl. .............................. 131

Figure 8.2: Selectivity between the 3.0 and 16.8 kbp supercoiled plasmids as a function of filtrate

flux. Ultrafiltration experiments were performed using 100 kDa Biomax membranes in TE

buffer containing 300 mM NaCl. ................................................................................................ 133

Figure 8.3: Selectivity between the 3.0 and 16.8 kbp supercoiled plasmids as a function of

membrane MWCO. Only the optimal Ψ values were plotted. Ultrafiltration experiments were

conducted in TE buffer containing 300 mM NaCl. The optimal Ψ values were obtained at

filtrate flux of 110, 70, and 5.2 µm/s for the 50, 100, and 300 kDa membranes, respectively .. 134

Figure 8.4: Observed sieving coefficients of the 3.0, 9.8, and 16.8 kbp supercoiled plasmids

through the 100 kDa Ultracel membrane in TE buffer with 500 mM NaCl. .............................. 135

Figure 8.5: Agarose gel electrophoresis (AGE) showing the separation of a binary mixture of the

3.0 and 16.8 kbp supercoiled plasmids using an UltracelTM 100 kDa membrane in TE buffer

containing 500 mM NaCl. Lane 1: linear 1 kbp DNA ladder. Lane 2: filtrate sample collected

at a filtrate flux of 70 µm/s. Lane 3: feed sample. Lane 4: purified 3.0 kbp supercoiled plasmid.

Lane 5: purified 16.8 kbp supercoiled plasmid ........................................................................... 137

Figure 8.6: Observed sieving coefficients of the 3.0, 9.8, and 16.8 kbp linear plasmids through

(a) the Ultracel 100 kDa membrane in TE buffer with 150 mM NaCl and (b) the Biomax 100

kDa membrane in TE buffer with 10 mM NaCl. ........................................................................ 138

xiv

LIST OF TABLES

Table 2.1: Specification of plasmid DNA stock solution............................................................. 20

Table 2.2: Digestion conditions for plasmid DNA ...................................................................... 21

Table 2.3: Detection limit and reproducibility of NanoDrop Spectrophotometry ....................... 26

Table 2.4: Average hydraulic permeability and estimated mean pore radius of various UF

membranes used in this study ....................................................................................................... 31

Table 4.1: Specifications of microfiltration membranes used in this study ................................. 57

Table 4.2: Comparison of observed sieving coefficients of a 3.0 kbp supercoiled plasmid

through an Ultracel 100 kDa membrane with the skin-side up, the skin-side down, and in a

composite structure with a 0.22 µm Durapore membrane ............................................................ 60

Table 4.3: Dimensions of membranes with conical shape pores created by anisotropic chemical

etching of 0.015 µm Nuclepore membranes ................................................................................. 71

Table 7.1: Best fit values of β and k for plasmid ultrafiltration experiments shown in Figs. 7.1–

7.5................................................................................................................................................ 117

xv

ACKNOWLEDGEMENTS

First and foremost, I would like to thank my advisor, Dr. Andrew L. Zydney. I am deeply

grateful for the guidance, encouragement and the time he spent training me as a scientist. His

enthusiasm, insights and dedication to science have made these five years a very rewarding

experience. I cherish the freedom in research he has allowed me, and appreciate all the support

and opportunities he has provided me with during the pursuit of my Ph.D. I would also like to

express my sincere gratitude to the members of my Dissertation committee, Dr. Manish Kumar,

Dr. Janna Maranas and Dr. William Hancock, for their valuable advice and feedback.

My grateful thanks also to the former and current members of the Zydney research group:

Ehsan Espah Borujeni, Achyuta Teella, Elaheh Binabaji, Mahsa Hadidi, Shudipto Dishari,

Youngbin Baek, Ivan Manzano, Zhao Li, Parinaz Emami, Fatemeh Fallahianbijan and Hadi

Nazem-Bokaee, with whom I shared this wonderful experience. Special thanks to Ehsan and

Achyuta for teaching me basic lab skills and Dr. David Latulippe, from whose Ph.D work I

gained a lot of inspiration.

I want to extend my gratitude to a group of talented and diligent undergraduates who

have worked with me for the past five years, including David Currie, Rachel Bolten, Neil Butler,

Kuangzheng Zhu, Weiru Luo and Anirudh Nambiar. I feel happy to have had the chance to

mentor them and I am thankful for their contribution and company in the lab.

I am indebted to the Walter L. Robb Family Chair, the National Science Foundation, and

the Air Products & Chemicals Graduate Fellowship program for providing funding for my Ph.D

study. I wish to thank Dr. Henry C. Foley and Maryam Peer for guiding me to start independent

scientific research; Air Products & Chemicals and the ACS Green Chemistry Institute for giving

xvi

me the opportunities to conduct research projects that are closely related to industrial processes.

Although these works are not included in this thesis, the experiences were certainly influential

for my growth.

I owe sincere appreciation to all of my friends for being part of my life over the past

several years. Special thanks goes to Pengfei Zhan, Xin Lu and Zifeng Li for always being there

to share my happiness and sadness. I would like to specially thank Xiao-Guang Yang, whose

love and support have been an essential part of this journey. I feel lucky to have met you.

Most importantly, I would like to thank my parents for their infinite and unconditional

love. Thanks for providing me with such a warm, supportive and open-minded family. Thanks

for always encouraging me to be positive, reflective and persistent. You have shaped who I am

today.

I hope I am now strong enough to embrace the many possibilities in my future life

journey.

1

Chapter 1

Introduction

1.1 DNA therapeutics and plasmid DNA

The past few decades have witnessed the advent and rapid development of DNA

therapeutics, including gene therapy and DNA vaccines. The concept of gene therapy was first

introduced in the 1960s, with the idea of inserting corrective genetic material to replace defective

genes that are responsible for human disease 1. Instead of tackling only the symptoms, as is the

case with most conventional drugs, DNA-based drugs are designed to directly treat or eliminate

the root cause of the disease. Because of their selective production of a given protein or enzyme,

gene therapy has the potential for enhanced therapeutic effects, lower toxicity and side effects, as

well as prolonged duration of action. Moreover, gene therapy has the potential to bring the

ultimate solution for diseases that are presently untreatable or poorly managed 2.

DNA vaccination first sparked the interests of the scientific community in the early 1990s

when it was discovered that plasmid DNA could induce an immune response to the plasmid-

encoded antigen (Wolff et al., 1990). In contrast to recombinant protein-based vaccines, DNA

vaccines can effectively engage both the cellular and humoral arms of the immune response and

therefore induce more prolonged immunity. At the same time, DNA vaccines can circumvent

many problems and risks associated with recombinant virus vaccines, such as pre-existing

immunity, loss of attenuation, or spread of inadvertent infection3.

A number of studies have demonstrated the potential of DNA therapeutics for treatment

of inherited diseases such as cystic fibrosis and hemophilia, various types of cancer, neurological

diseases such as Alzheimer’s and Parkinson’s, cardiovascular disorders and infectious diseases

2

such as AIDS 4. As of 2016, there were over 2400 DNA therapeutics being studied in clinical

trials worldwide, with over three-quarters in Phase I or I/II (Gene Therapy Clinical Trials

Worldwide). Glybera, the first gene therapy drug, was approved by the European Medicine

Agency and launched in Germany in 2015. It treats lipoprotein lipase deficiency (LPLD), a rare

genetic disorder affecting about 1-2 persons per million people that causes fat to build-up in the

blood leading to cardiovascular disease, diabetes, and life-threatening recurrent bouts of

pancreatitis. Glybera contains engineered adeno-associated viral vectors harboring correct copies

of the lipoprotein lipase gene to restore the LPL enzyme activity, offering a long-term cure for

six years or more 5.

In 2016, the second gene therapy product, Strimvelis, was granted authorization in

Europe. Strimvelis treats an extremely rare, life threatening genetic disorder named adenosine

deaminase severe combined immunodeficiency syndrome (ADA-SCID). Children born with this

single genetic defect are extremely susceptible to everyday infections and must live in plastic,

germ-free chambers. Strimvelis works by infusion of hematopoietic stem cells extracted from the

patient and transduced with gammaretrovirus containing the human adenosine deaminase gene 6.

Although the U.S. Food and Drug Administration (FDA) has not yet approved any human gene

therapy products, it is expected that several DNA therapeutics are on track for approval in the

U.S. in the near future. For example, SPK-RPE 65, a gene therapy agent developed by Sparks

Therapeutics for treating inherited retinal dystrophies, has completed a Phase III clinical trial and

is likely to hit the market by the end of 2017. In addition, the full elucidation of the human

genome and the advancement of CRISPR/Cas9 gene editing technology should provide exciting

new advances in the development of DNA therapeutics.

3

Successful application of gene therapy and DNA based vaccines depends on the

development of a vector that can safely and effectively deliver the gene of interest to the target

cells followed by sustainable expression of the gene product in vivo. There are two main

categories of delivery vehicles, viral and non-viral. Various types of viruses have been developed

as vectors, taking advantage of the natural ability of viruses to deliver genetic materials to cells.

Among the most advanced viral vectors are retrovirus, adenovirus, and adeno-associated virus

(AAV). Viruses generally have high transfection efficiency in a wide variety of cells with

prolonged expression of the desired gene product. The major disadvantages of virus vectors

include insert-size limitations, immunogenicity, and manufacturing challenges7-8. According to

statistics from the Gene Therapy Clinical Trials Worldwide in 2016, 60.3% of clinical trials of

DNA therapeutics are conducted using various virus vectors (percentages of Adenovirus,

Retrovirus and AAV are 21%, 18% and 7%, respectively).

In comparison, non-viral vectors have less efficient transfection and often provide only

transient expression in vivo. However, non-virial delivery systems continue to be a promising

alternative to viral vectors due to safety concerns regarding the use of viruses in humans. Non-

viral delivery is particularly attractive in terms of the lack of immune response, wide range of

insert-gene size, and convenience in large-scale production 9. Non-viral vectors can be delivered

in the form of naked/plasmid DNA or DNA complexed with cationic lipids or polymers 10. There

has been significant interest in using naked plasmid DNA (pDNA) for direct gene transfer due to

its simplicity. Enhanced transfection efficiency of naked pDNA can be obtained using

electroporation and the “gene gun”. These physical approaches allow DNA to directly penetrate

the cell membrane, avoiding enzymatic degradation of DNA molecules by bypassing the

endosome/lysosome 11-12.

4

Plasmid DNA is a double-stranded, circular, extrachromosomal DNA molecule found in

bacteria. Plasmids are able to replicate independently and can be artificially constructed as

recombinant DNA for protein production, as vectors in genetic engineering for cloning, and as

DNA therapeutics. The size of naturally occurring plasmids can vary from very small mini-

plasmids of less than 1 kilobase pairs (1 kbp) to very large megaplasmids that are several

megabase pairs (Mbp) in size 13. The typical plasmid size for DNA therapeutics is 3 to 20 kbp.

Plasmid DNA can exist in three different topological conformations (isoforms) as shown

in Figure 1.1:

Supercoiled (covalently closed-circular) DNA is the natural, intact

conformation that results from the coiling of the axis of the DNA double helix.

Nicked open-circular DNA has one strand cut

Linear DNA has free ends and is formed by a breakage in both strands

Figure 1.1: Electron microscopy image of different plasmid DNA isoforms (adopted from

https://commons.wikimedia.org/wiki/File%3APlasmid_emEN.jpg)

5

DNA supercoiling is numerically described by the linking number (Lk), which is the

number of times the two strands of the DNA double helix are intertwined. The linking number

equals the sum of Tw, the number of ‘twists’ or turns of the double helix, and Wr, the number of

coils or ‘writhes’ (Equation 1.1).

𝐿𝑘 = 𝑇𝑤 + 𝑊𝑟 (1.1)

Lk is a constant and can only be changed by breaking the DNA backbone. However, there can

by complementary changes in Tw and Wr at a fixed Lk.. For biological circular DNA, Wr is

typically negative (Lk < Tw) and the DNA is described as being “underwound”. The linking

number difference, ∆Lk, is the difference between the actual number of turns in a plasmid, Lk,

and the number of turns in a relaxed plasmid Lk0 of the same size:

∆𝐿𝑘 = 𝐿𝑘 − 𝐿𝑘0 (1.2)

where Lk0 is determined by dividing the total base pairs of the molecule by the number of bp per

turn as:

𝐿𝑘0 = 𝑏𝑝/10.4 (1.3)

To facilitate comparison between plasmids with different size, specific linking deficits or

superhelical density is usually expressed in terms of the parameter σ, which represents the level

of supercoiling of the molecule independent of its size:

𝜎 = ∆𝐿𝑘/𝐿𝑘0 (1.4)

Linking deficits typically range from -0.05 to -0.07 for natural supercoiled plasmid DNA isolated

from bacteria, and it is dependent on a number of conditions such as solution ionic strength and

temperature.

6

Figure 1.2 illustrates the supercoiling structure of DNA. The supercoiled DNA adopts

interwound or plectonemic conformations as determined by electron microscopy 14-18, atomic

force microscopy 19-20, Monte Carlo simulations 21-23, neutron scattering 24, and analysis from

sedimentation 25 and the products of topoisomerases 26. The supercoiled shape of the DNA

minimizes the unfavorable free energy associated with decreasing Lk from the preferred value of

the relaxed state by changing Wr and Tw. The Waston-Crick twist is referred to as ‘secondary’

winding, with the formation of superhelices referred to as ‘tertiary’ winding. The point at which

two DNA helices cross in projection are called nodes; the superhelix axis is defined as the curve

that passes through the nodes and bisects the area enclosed by the DNA between adjacent nodes.

Electron microscopy studies indicated that the average superhelix winding angle is about 60° and

does not depend on σ 16-17. Most supercoiled DNA displays a branched structure, with the

branching point defined as the intersection of the superhelixes of three or more plectonemic

segments. For example, the molecule shown in Figure 1.2 has 2 branching point, 5 plectonemic

segments and 25 nodes.

7

Figure 1.2: Structure of supercoiled plasmid DNAs (reproduced from 17)

Negative supercoiling is crucial for a number of biological processes. First, the free

energy associated with negative supercoiling facilitates processes which require untwisting or

unwinding of DNA, such as DNA replication and transcription. It also promotes processes

involved in the packaging of DNA within the cells. Finally, the supercoiled DNA conformation

plays a direct role in bringing together and aligning distant DNA sequences and therefore

facilitates the binding of proteins and other ligands to DNA 27-28.

8

1.2 Downstream processing of plasmid DNA

Relatively large amounts of DNA are required for administration (up to several

milligrams per dose in humans) in order to generate a strong therapeutic effect for gene therapy

or a strong immune response for a DNA vaccine. The large dose needed for DNA therapeutics

requires development of large scale processes able to fill the ultimate market demand (on the

order of many kilograms per year for many vaccines). Plasmids are usually produced in a

recombinant Escherichia coli (E. coli) host by fermentation and represent around 3% by mass of

the cell lysate 29. Similar to the case of protein therapeutics, process development for

manufacturing plasmid DNA (pDNA) generally begins with the construction and selection of

appropriate expression vectors and strain selection, followed by optimization of the fermentation

conditions (upstream processing), cell growth, and finally a series of purification steps

(downstream processing). The different stages of pDNA production are integrated and require

concomitant optimization. The downstream purification is significantly affected by the

contaminants generated during the fermentation. For example, the judicious selection of plasmid

vector and host strain, combined with growth-condition optimization, can increase plasmid yield

from 5-40 mg/L to as high as 220 mg/L with a 40% reduction in the RNA content during cell

lysis 30-31.

In general, the purification of biological products is difficult since most products are

labile and therefore require relatively mild operating conditions. In particular, the structure and

physical properties of pDNA (including its size, shape and conformation, and rheological

properties), as well as the diversity of biomolecules present in pDNA containing extracts, impose

additional challenges for establishing efficient purification schemes 32. Figure 1.3 shows the

impurity content in bacterial cell lysates and the final product purity requirement currently

9

specified by guidelines from FDA. Most of the impurities share similar characteristics with

pDNA, such as negative charge (RNA, genomic DNA, endotoxins), similar size, and

hydrophobicity. Furthermore, although the compact supercoiled pDNA isoforms are produced

from fermentation of E. coli, open-circular, linear and denatured pDNA isoforms are also

generated due to conformational changes that occur within the bacterial host or to structural

damage during various downstream processing steps. For instance, random cleavage (enzymatic,

chemical, mechanical, etc) of one or both opposing strands of the intact DNA double helix

structure gives rise to the open-circular and linear isoforms, respectively. Improper alkaline lysis

conditions (e.g., pH above 12.5) have been shown to generate irreversibly denatured covalently

closed pDNA isoforms 33. The resulting isoforms are less efficient in transferring gene

expression. For this reason, the key regulatory agencies specify that the homogeneity of a

therapeutic pDNA product, expressed as percentage of the pDNA in the intact supercoiled

isoform, should be higher than 90%.

Figure 1.3: A) The impurity content in bacteria cell lysate, B) final product purity requirement

specified by guidelines from FDA

*The individual specifications are not fixed values and may be adapted according to further

developments in this field. Guidelines can be obtained from the FDA (http://www.fda.gov/cber/)

or the WHO (http://www.who.int/en/)

10

The purification processes after fermentation are comprised of several unit operations

including cell harvest, cell lysis, cell debris/solid removal, polishing/clarification, buffer

exchange, concentration, and formulation. A typical pDNA purification process is summarized

in Figure 1.4. Chromatography is currently a key step for large scale pDNA purification, and it

has been widely used for removal of key process impurities, including the undesired plasmid

isoforms, genomic DNA, host cell proteins, RNA, and endotoxins. Chromatography is also used

for assessing the purity of clinical grade pDNA for quality and safety purposes. Chromatographic

techniques typically used include reverse-phase HPLC (RP), hydrophobic interaction

chromatography (HIC), anion-exchange chromatography (AEC), size exclusion chromatography

(SEC), and affinity chromatography (AC) (e.g., immobilized metal affinity, triple-helix affinity

and amino acid-DNA affinity chromatography) 34.

Figure 1.4: Typical industrial scale pDNA purification process

There are several critical limitations associated with DNA chromatography. First, the

average pore size in chromatographic resins is smaller than or comparable to the radius of

11

gyration of typical plasmid DNA molecules, therefore the majority of the internal surface area is

inaccessible to the large DNA molecules, leading to very low binding capacities. In contrast to

protein biologics, where loading can range from 10 to 100 g of protein per liter of resin, only 0.2-

2 g of pDNA can typically bind per liter of resin 35. As a result, very large quantities of

chromatographic media are needed for the large-scale purification of pDNA for therapeutic

applications (~ 500-2000 L of resin kg-1 pDNA processed). Second, the separation of different

plasmid isoforms is fairly limited due to their extremely similar chemical and physical

properties. The lack of selectivity has important implications for product purity and yield.

Finally, diffusional mass transfer significantly limits the total throughput of chromatographic

processes, with very long contact times required for satisfactory resolution 36.

1.3 Purification of plasmid DNA using membrane processes

A promising alternative strategy for pDNA purification is the use of ultrafiltration (UF)

membranes with pore sizes on the order of 10 nm, which was first demonstrated by Latulippe

and Zydney 37. Membrane processes are attractive since they are easily scaled and typically

cause little damage to biomolecules. In addition, membranes have unique advantages over

chromatography, as the process is driven by convective flow and does not suffer from diffusional

limitations. At the same time, membranes tend to be more cost effective than chromatographic

separations. Currently, membrane based processes have been widely used in the biotechnology

industry for protein concentration, buffer exchange, virus filtration and depth filtration 38-40.

Latulippe et al. 37 conducted extensive experimental and theoretical studies on the

transmission of plasmid DNA through semi-permeable ultrafiltration membranes. The extent of

DNA transmission was found to be largely dependent on the filtrate flux, with the transmission

12

increasing from essentially zero to nearly 100% as the filtrate flux increased, even though the

pore size of the UF membranes (2-20 nm) was an order of magnitude smaller than the radius of

gyration of pDNA (~100 nm for 3-30 kbp plasmids that are typically used in DNA therapeutics).

The large increase in DNA transmission with increasing filtrate flux was not caused by

concentration polarization effects since the results were found to be independent of the stirring

speed. Instead, Latulippe and Zydney hypothesized that the high transmission was due to

elongation of the plasmid DNA in the converging flow field entering the membrane pores during

ultrafiltration.

The extent of polymer deformation during flow can described in terms of the Deborah

number (De), which is the ratio of the time-scale for polymer relaxation (τ) to the characteristic

time for the fluid flow (γ-1) where γ is the shear rate associated with the flow. Significant

polymer deformation typically occurs when De = τγ ≥ 1. When τ > γ-1, the hydrodynamic force is

greater than the spring restoring force and the polymer stretches out into a string of ‘blobs’ with

a characteristic blob size smaller than the pore size 41. The elongated polymer is then able to

enter the pore and pass through the membrane. Daoudi and Brochard 42 modeled the effects of

chain deformation on the transport of large linear polymers into a single cylindrical pore. The

model predicts a sharp transition in polymer transmission above a critical value of the filtrate

flux, with this critical flux scaling as:

𝐽𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 = 𝑘𝐵𝑇

𝑟𝑝2𝜇

(1.5)

where kB is Boltzmann’s constant, T is the absolute temperature, µ is the fluid viscosity, and rp is

the pore radius. Equation (1.5) predicts that the critical filtrate flux is independent of the polymer

size but varies with the reciprocal of the pore radius squared. It is also proportional to

temperature and the reciprocal of the solution viscosity. These trends agree with experimental

13

results obtained with Ultracel 100, 300 and 1000 kDa membranes 37. However, this elongational

flow model only considers flow through a single cylindrical pore located within an infinite flat

(non-porous) plane. The flow behavior is much more complicated for a real membrane (e.g., the

Ultracel membranes with porosity as high as 50%) since the fluid streamlines from adjacent

pores will interact at a distance above the surface of the membrane that is much less than the Rg

of the plasmids. Latulippe and Zydney 43 modified the elongational flow model by assuming that

plasmid elongation occurred at a critical distance equal to a fraction of the radius of gyration

(β·Rg) with β <1. The resulting model yields a much smaller Jcritical and agrees well with the

critical flux measured experimentally:

𝐽𝑐𝑟𝑖𝑡𝑖𝑐𝑎𝑙 =𝛽3𝜋

6 (

𝑟𝑝2) (

1

𝜆) (

𝑘𝑩𝑇

𝜂) (1.6)

where λ is the ratio of the hydrodynamic radius to the radius of gyration, which is specific to the

type of pDNA and the solution conditions, and β was determined by fitting the model to data for

Jcritical under different experimental conditions. Although Equation (1.6) was in good agreement

with experimental data, the plasmid transmission is found to increase gradually with increasing

filtrate flux, in contrast to the sharp transition predicted by this polymer elongation model. This

discrepancy is likely due to the heterogeneity in the DNA molecules in combination with the

pore size distribution in the membranes. This will be discussed in more detail in the following

chapters.

It is well established that solution conditions, including both the salt concentration and

ionic composition, can significantly affect the conformation and flexibility of DNA due to

intramolecular electrostatic interactions between the negative-charged phosphate groups along

the DNA backbone. As a result, the transmission of pDNA through a UF membrane is also a

strongly affected by solution conditions, as shown by Latulippe and Zydney 44. For example, the

14

sieving coefficient of a 3.0 kbp supercoiled plasmid through Ultracel 300 kDa membranes

increased dramatically (from <0.01 to 0.65) at a constant filtrate flux of 50 µm/s as the NaCl

concentration was increased from 1 to 150 mM. An even more significant increase in

transmission was observed using MgCl2 as the salt concentration was increased from 2 to 40 mM

(similar to the total ionic strength range up to 150 mM NaCl). This could be due to the formation

of intramolecular salt bridges by the divalent magnesium ions, which would further reduce the

effective plasmid size. Dynamic ultrafiltration experiments were also performed with continuous

adjustment of the solution ionic strength. A considerable increase in the observed sieving

coefficient was observed following a 10-min mixing period (after changing the salt

concentration). This could imply that the conformational change in the pDNA with respect to the

change in ionic strength may be on the order of minutes. The dynamic response of the

supercoiled plasmids to changes in ionic environment could have important implications for the

overall performance of diafiltration/buffer exchange processes for DNA purification.

Similar to the case of supercoiled plasmids, the transmission of the linear and open-

circular isoforms is also a strong function of the filtrate flux. However, the critical flux for the

different isoforms were significantly different. Latulippe and Zydney 45 hypothesized that

different pDNA isoforms would display unique Jcritical values due to differences in their ability to

elongate in the converging flow field, thus providing opportunities to separate the isoforms by

simple ultrafiltration. Sieving experiments with the individual pDNA isoforms showed the

lowest values of the critical filtrate flux for the linear isoform followed by the supercoiled and

then the open-circular isoform; these results were confirmed by experiments using binary and

ternary mixtures of the pDNA isoforms. Separations could thus be achieved by operating at a

filtrate flux between the Jcrititcal values for the specific isoforms. For example, the supercoiled

15

DNA could be separated from the linear isoform by selecting a filtrate flux between the critical

flux for the linear and supercoiled isoforms, with the filtrate solution only containing the linear

isoform due to the nearly complete retention of the supercoiled isoform under these conditions. It

is worth mentioning that the separation mechanism in ultrafiltration is dramatically different

from other common methods of DNA separation such as SEC and Agarose gel electrophoresis

(AGE). For example, plasmid retention in SEC is due to equilibrium partitioning between the

mobile phase and the fluid space within the resin particles; therefore, the separation in SEC

depends on differences in the radius of gyration of the different isoforms 46. In this case, the

linear (largest) isoform has the shortest retention time while the supercoiled (smallest) isoform

has the longest retention time. In contrast, the supercoiled isoform migrates fastest

(corresponding to the largest mobility) through the homogeneous gel in AGE while the open-

circular isoform migrates the slowest, which is likely due to either transient impalement or

hindered reptation effects involving the agarose fibers within the gel matrix.

1.4 Challenges in membrane ultrafiltration of plasmid DNA

One of the main challenges in using membranes for biomolecule separations is the

limited selectivity of traditional size-based membrane separations. The general rule of thumb is

that at least a 10-fold difference in size is required for effective separation using conventional

ultrafiltration membranes (although electrically-charged membranes have been shown to have

much greater resolution when separating species with different net charge) 47. This is much less

than the difference in size for the different DNA isoforms, and even the differences in

elongational flexibility are often insufficient to provide sufficiently high resolution separations.

16

In addition to selectivity, another major challenge in membrane UF for pDNA

purification is membrane fouling. Borujeni and Zydney 48 conducted a thorough study on the

fouling behavior of pDNA during ultrafiltration. A significant decline in both filtrate flux and

DNA sieving coefficient was observed during the ultrafiltration of concentrated plasmid

solutions. For example, the filtration of a 20 L/m2 pDNA solution with a concentration of 7.2 x

10-3 kg/m3 caused 3-fold decline in filtrate flux and nearly total loss of plasmid transmission. A

simple mathematical model was developed for the decline in both filtrate flux and sieving

coefficient based on pore blockage by large pDNA molecules. The model accounts for both

partial pore blockage, as described by the parameter β (ratio of the filtrate flux through a blocked

pore to that through an open pore), and the rate at which the pores are blocked, which was

characterized by the rate constant k. Model calculations were in good agreement with the

experimental results, properly capturing the effects of membrane pore size, plasmid

concentration, and plasmid size on the fouling behavior. A variety of strategies have been

applied to control membrane fouling, including modification of membrane surfaces, physical

cleaning (sponges, water jets, or backflushing), chemical cleaning (use acids or bases to remove

foulants and impurities), optimization of operating conditions etc. However, it is critical to find

more efficient approaches to reduce membrane fouling for UF to be applied in large scale

purification of pDNA.

1.5 Dissertation outline

Previous studies have demonstrated that plasmid transmission through UF membranes

can be controlled by adjusting the filtrate flux thereby controlling the extent of plasmid

elongation. This phenomenon can be used for the separation of different plasmid isoforms by

17

exploiting the differences in flexibility of the supercoiled, linear, and open-circular isoforms.

However, there are a number of critical challenges that still need to be overcome in order to

apply these novel membrane-based processes for commercial scale manufacture of DNA for

gene therapy applications and for use as DNA-based vaccines.

The overall objective of this dissertation is to develop novel strategies to enable the

application of UF for pDNA separations and to demonstrate the feasibility of using these

enhanced membrane systems for the purification of plasmid DNA. This includes:

I. Evaluating different approaches that can enhance the separation resolution during

ultrafiltration of different plasmid isoforms

II. Examining different approaches to reduce membrane fouling during ultrafiltation of

concentrated DNA solutions

III. Developing better physical understanding of the factors governing the ultrafiltration

behavior of DNA through small pore size membranes

It is anticipated that the results from this study will provide important information needed for

successfully implementing UF processes into commercial systems for the large-scale

manufacture of therapeutic DNA products.

The details of the general experimental methods and materials used in this research are

summarized in Chapter 2. Chapter 3 investigates the strategy of pre-conditioning during UF of

pDNA, accomplished by exploiting the asymmetric structure of conventional ultrafiltraion

membranes to enhance the separation performance and reduce fouling. Chapter 4 provides a

different approach to pre-conditioning, in this case using composite membrane structures with

layers have different pore size. Chapters 5 and 6 examine the effects of solution ionic strength

and mutilvalent salts on the transmission of different plasmid isoforms through UF membranes,

18

including the identification of strategies to enhance the resolution of membrane systems for

DNA separations. Chapter 7 explores the effects of solution ionic conditions on membrane

fouling phenomena. Chapter 8 discusses the use of UF membranes for the separation of DNA

based on differences in DNA size / length (kbp), including a comparison with other size-based

separation methods such as AGE and SEC. The major conclusions that can be drawn from this

thesis, as well as recommendations for future research in this area, are presented in Chapter 9.

19

Chapter 2

Materials and Methods

This chapter describes the materials, apparatus, and experimental procedures that are

common across the different chapters in this dissertation. Additional details on specific methods

are provided in the appropriate chapters.

2.1 Plasmid DNA

2.1.1 Supercoiled plasmids

Experiments were performed with plasmids from approximately 3 to 17 kilo-base-pair

(kbp) in size. The 2,961 base pair (bp) pBluescript® II KS+ plasmid was obtained from

Strategene. The 9,801 bp plasmid was generated by insertion of a 6,840 bp fragment into the SalI

site of pBluescript® II KS+ plasmid vector. The 16,836 bp plasmid was generated by insertion

of a 13,878 bp fragment into the NotI site of the pBluescript® SK- plasmid (2,958 bp). The three

supercoiled plasmids mentioned above are named as p-EMP, p-MDY, and p-FDY and are

described as having size of 3.0, 9.8 and 16.8 kbp throughout the thesis.

Stock solutions of each supercoiled plasmid were prepared by Aldevron (Fargo, ND) as

follows. A single colony of E. coli DH5α bacteria strain was transfected by the desired plasmid

and used to inoculate a small starter culture that was allowed to grow for approximately 7 hr at

37 °C. The cell culture was then transferred to a shaker flask that contains 2 L of a nutrient-rich

media to grow overnight. The resulting biomass was harvested and lysed, with the lysate purified

20

by anion exchange chromatography. The plasmid stock solution met the stringent lot release

specifications listed in Table 2.1. All solutions were prepared in Tris-EDTA (TE) buffer with

concentrations of 200 – 250 µg/mL. The received stock solution was divided into 110 µL

aliquots to avoid damage of DNA molecules from repeated freeze-thaw cycles. These aliquots

were stored at -20 °C until ready for use in experiments.

Table 2.1: Specification of plasmid DNA stock solution

*Abbreviations: AGE, Agarose gel electrophoresis; EtBr, Ethidium bromide; LAL, Limulus

amebocyte lysate; EU, endotoxin units.

2.1.2 Plasmid isoforms

The linear and open-circular plasmid isoforms were prepared by enzymatic digestion of

the supercoiled isoform using restriction endonucleases that recognize specific nucleotide

Method Specification

Appearance Visual inspection Clear, colorless and free of visible particulates

Concentration UV spectrophotometry =0.20 or 0.25mg/mL

DNA homogeneity EtBr stained AGE Predominantly supercoiled

Endotoxin Kinetic Turbidimetric LAL < 100 EU/mg

Identity EtBr stained AGE Co-migrates with clienet reference DNA or

size confirmed versus a supercoiled marker

260/280 Absorbance Ratio UV spectrophotometry 1.80 – 2.00

Residual host Genomic DNA EtBr stained AGE < 5.00%

Residual host RNA EtBr stained AGE Not visible at 200 ng load

21

sequences in the double-stranded DNA and cleave either both strands (for production of the

linear isoform) or one strand (for the open-circular isoform). The digestion conditions for the

different plasmids / isoforms are summarized in Table 2.2. Restriction endonucleases were

obtained from New England Biolabs (10,000 U/mL) along with the appropriate buffer for the

respective digestion reactions. The digestion mixture was incubated at 37 °C for 3 hr and then

purified using commercially available kits that remove unwanted enzymes, salts, and other

impurities. Purification procedures from the QIAQuick PCR purification kit (Qiagen, CA) were

as follows. The DNA digestion mixture was first loaded onto a silica-gel membrane QIAquick

column in a high salt concentration buffer (buffer PB, pH =7.0). Impurities were washed through

the column by proprietary buffer PE (which contains ethanol at pH 7.7). Purified DNA was then

eluted with a small volume of low salt TE buffer. The temperature of the elution buffer was then

raised to 65°C to recover residual bound plasmid; this was especially important for the larger

size plasmids (9.8 and 16.8 kbp). The exact pDNA concentrations were evaluated using a

NanoDrop UV spectrophotometer (Thermo Scientific). The integrity of the DNA was verified by

AGE as discussed subsequently.

Table 2.2: Digestion conditions for plasmid DNA

Plasmid size Linear Open-circular

3.0 kbp 4 U/µg BamHI 2 U/µg Nt.AlWI

9.8 kbp 1-2 U/µg KpnI 1-2 U/µg Nt.AlWI

16.8 kbp 2 U/µg PaeR7I 1.5-2 U/µg Nt.AlWI

2.2 Buffers

22

Buffer solutions were prepared by diluting a 100× concentrate of 1.0 M Tris-

hydrochloride (Tris-HCl) and 0.1 M ethylenediaminetetraacetic acid disodium salt (EDTA-Na2)

from Sigma-Aldrich. The molecular structure and physical properties of Tris-HCl and EDTA are

shown in Figure 2.1. Tris is widely used as a component of buffer solutions in biochemistry and

molecular biology, especially for solutions of nucleic acids. It has a pKa of 8.07 at 25 °C, and

thus has an effective buffer pH range between about 7.5 and 9.0. EDTA inactivates DNase by

chelating and sequestering cations such as Mg2+ that are required for DNase activity. The TE

buffer solution (i.e., 10 mM Tris-HCl, 1 mM EDTA-Na2) is thus commonly used to solubilize

DNA or RNA while protecting it from degradation. Plasmid DNA can be stored in TE buffer at

4°C for short term use or at -20°C to -80°C for long term storage up to 2 years.

Deionized distilled water with a resistivity greater than 18 MΩ-cm was produced using a

NANOpure Diamond water purification system (Barnstead International, IL). The solution ionic

strength was adjusted with either monovalent salts (e.g., sodium chloride = NaCl) or divalent

salts (e.g., magnesium chloride = MgCl2), both obtained from VWR with certified ACS grade.

The solution pH (7.7 ± 0.1) was measured using a 420APlus pH meter (Thermo Orion), and the

solution conductivity was measured using a 105APlus conductivity meter (Thermo Orion). All

solutions were prefiltered through 0.2 µm pore size Supor 200 disc filters obtained from Pall

Corporation (Port Washington, NY) to remove particulates.

23

Figure 2.1: Molecular structure and physical properties of Tris-HCl and EDTA

2.3 DNA characterization

2.3.1 Agarose gel electrophoresis

Agarose gel electrophoresis (AGE) was used to confirm the integrity and topology of the

different plasmid isoforms. A 0.8% agarose gel solution was prepared by dissolving 0.36 g

agarose powder and 4.5 µL of GelStarTM nucleic acid gel stain (Lonza, NJ) in 45 mL of Tris–

Acetate-EDTA (TAE) buffer. The agarose solution was poured onto a 7×7 cm2 casting tray (Bio-

Rad, CA) with an 8-well comb inserted, and then allowed to set for 30 min at room temperature.

The gel was then loaded into a Mini-Sub Cell GT (Bio-Rad) that had been pre-filled with about

200 mL TAE buffer. The electrophoresis was conducted at a constant electric field of 45–55 V

for 90–120 min. Gel images were obtained using a Fluorchem FC image system. Figure 2.2 is a

typical AGE image showing the separation of plasmid DNA with different sizes and isoforms.

24

For the same voltage, plasmid DNA with the smallest size migrates fastest (from top to bottom),

with the supercoiled isoform migrating faster than either the linear or open-circular isoforms.

Figure 2.2: AGE image for plasmid with different sizes (left panel) and isoforms (right panel)

2.3.2 PicoGreen Assay

DNA concentrations were evaluated by fluorescence detection using the ultrasensitive

nucleic acid stain PicoGreen (Life Technology, CA). All DNA samples were analyzed in

duplicate using Cliniplate 96-well black microplates (Thermo Scientific, PA) and a GENios FL

microplate reader (TECAN). 70 µL of the PicoGreen solution was prepared by diluting the stock

reagent with TE buffer (1:200), with the reagent added to each well along with 70 µL of the

DNA sample. The plates were shaken for 3 min, with the fluorescence intensity evaluated at 530

nm using an excitation wavelength of 485 nm and a temperature of 36 °C. Calibration curves

were constructed using DNA solutions with known concentrations from 0 to 0.5 µg/mL, with

accuracy of 0.25 ng/mL. Since the Picogreen fluorescence is weakly sensitive to the salt

concentration49, calibration standards were included in each plate at the specific ionic conditions.

DNA concentration measurements using fluorescence methods are more sensitive than

25

absorbance, particularly for low-concentration samples, and the use of DNA-binding dyes allows

more specific measurement of DNA than is possible with spectrophotometric methods.

2.3.3 NanoDrop Spectrophotometry

Nucleic acids absorb ultraviolet (UV) light due to the heterocyclic rings of the

nucleotides; the sugar-phosphate backbone does not contribute to absorption. The wavelength of

maximum absorption for both DNA and RNA is 260 nm (λmax = 260nm), with a characteristic

value for each base. The absorption properties of DNA can be used for detection, quantification

and assessment of purity. A260 readings need to be within the instrument’s linear range (generally

0.1–1.0). The actual DNA concentration is then determined by subtracting off the contribution

from the turbidity (measured by absorbance at 320 nm):

Concentration (µg/ml) = (A260 reading – A320 reading) × dilution factor × 50 µg/ml

The DNA purity was estimated by measuring the absorbance from 230 to 320 nm. The most

common purity calculation is based on the ratio of the absorbance at 260 nm to that at 280 nm.

Good-quality DNA will have an A260/A280 ratio of 1.7–2.0. The ratio is best calculated after

correcting for turbidity (absorbance at 320nm).

DNA purity (A260/A280) = (A260 reading – A320 reading) ÷ (A280 reading – A320 reading)

A NanoDrop 1000 Spectrophotometer (Thermo Fisher Scientific, DE) was used to

evaluate the UV absorbance of the different DNA samples. 1-2 µL aqueous nucleic acid samples

were added to the measurement pedestal, which was sufficient to generate accurate and

reproducible results. The detection upper/lower limits and reproducibility of the NanoDrop 1000

Spectrophotometer are summarized in Table 2.3.

26

Table 2.3: Detection limit and reproducibility of NanoDrop Spectrophotometry

Detection Limit (µg/mL) Approx. Upper Limit (µg/mL) Typical Reproducibility

(minimum 5 replicates)

2 3700 (dsDNA); 2400 (ssDNA)

3000 (RNA)

Sample range 2-100 µg/mL: ±2 µg/mL

Sample range > 100 µg/mL: ±2%

2.4 Membranes

Ultrafiltration experiments were performed using both flat sheet membranes (in normal

flow filtration) and hollow fiber membranes (for use with tangential flow filtration or TFF). In

tangential flow filtration, the feed flow is directed parallel to the membrane and thus

perpendicular to the filtrate flow. This allows the retained species to be swept along the

membrane surface and out of the device, which significantly reduces fouling and increases the

filtrate flux50. Membranes are typically characterized by their pore size or nominal molecular

weight cut-off (MWCO), with the latter defined as the molecular wright of a solute that has a

retention coefficient of approximately 90%51. UF membranes with MWCO ranging from 50 to

500 kDa were used in this work.

The key design parameters for membrane processes include the selectivity, volumetric

filtrate flux, and system capacity. For pressure-driven membrane processes, the selectivity is

directly related to the solute sieving coefficient:

𝑆 = 𝐶𝑓

𝐶𝐹 (2.1)

where Cf and CF are the solute concentrations in the filtrate and feed solutions, respectively. The

selectivity (ψ) for the isoform separation is defined as 52:

27

(2.2)

where S1 and S2 are the observed sieving coefficients of the plasmid isoforms that are more and

less highly transmitted through the membrane, respectively. The initial (clean membrane)

volumetric filtrate flux is related to the membrane hydraulic permeability as:

𝐿𝑝 = 𝐽𝑣

∆𝑃 (2.3)

where Jv is the filtrate flux (volumetric filtrate flow rate per unit membrane area) and ∆P is the

transmembrane pressure difference. The filtrate flux during filtration of the actual feed solution

is typically less than that predicted from the fresh membrane permeability due to membrane

fouling and concentration polarization effects. The system capacity is defined as the volume of

feed that can be processed per unit membrane area before the membrane has to be cleaned or

replaced. For constant pressure filtration, this is usually defined as the volume processed up until

the filtrate flux has dropped to less than 10% (or 20%) of its initial value. The change in the

membrane hydraulic permeability can also be used to characterize the extent of membrane

fouling after a filtration run 53.

Most of the membranes used in this study have an asymmetric structure, which is

comprised of a thin skin layer that provides the membrane its functionality and a much thicker

and more porous substrate that provides the membrane its structural integrity. The small

thickness of the skin greatly enhances the permeability compared to symmetric membranes with

similar selectivity. Figure 2.3 is a scanning electron microscope (SEM) image of the cross

section of the BiomaxTM and UltracelTM membranes used in this work showing their asymmetric

structures. The skin thickness is approximately 0.5 – 1 μm while the total membrane thickness is

around 100 μm.

y =S1

S2

28

Figure 2.3: SEM of the cross section of A) UltracelTM composite regenerated cellulose and B)

BiomaxTM polyethersulfone membranes (both from MilliporeSigma) and C)

modified polyethersulfone hollow fiber membrane (from Spectrum Labs). Images

taken from manufacturers’ websites

Polysulfone (PS), polyethersulfone (PES) and regenerated cellulose are among the most

commonly used materials to make ultrafiltration membranes. PS and PES are thermally stable,

easy to fabricate, have wide pH tolerance and are chemically resistant to most acid, base and

chlorine solutions used for membrane cleaning and sterilization. PES is somewhat less

hydrophobic than PS (Figure 2.4) due to the increased proportion of the sulfone groups.

Nevertheless, most commercial PS/PES membranes are surface modified to render them more

hydrophilic to reduce protein adsorption54. The regenerated cellulose membranes are extremely

29

hydrophilic due to the large numbers of hydroxyl groups on the surface, which greatly reduces

protein binding and fouling during filtration. The structural integrity of regenerated cellulose

membranes is enhanced by casting the cellulose onto a microporous support matrix (typically

polyethylene). The resulting composite regenerated cellulose (CRC) membranes have a uniform,

robust structure, with high integrity and greater resistance to back pressure. Since regenerated

cellulose membranes are not very stable to extreme acidic or basic conditions, they are most

suitable in applications that do not involve harsh chemical conditions.

Figure 2.4: Molecular structure of polysulfone, polyethersulfone, and regenerated cellulose

30

Two types of flat sheet membranes, UltracelTM (regenerated cellulose) and BiomaxTM

(polyethersulfone), both with MWCO of 100 and 300 kDa, were used extensively throughout

this study. Membrane discs (25 mm diameter) were cut from large membrane sheets, generously

provided by Millipore Corp., using a specially designed cutting device. All membranes were

initially soaked in 90% isopropyl alcohol and then flushed with at least 100 mL of water to

remove residual storage agents and to insure thorough wetting of the pore structure. The

membranes were stored in DI water at approximately 4 °C to prevent collapse of the membrane

pore structure (i.e., irreversible damage) due to drying.

Tangential flow ultrafiltration experiments were performed with hollow fiber PS

membranes obtained from GE Healthcare (Niskayuna, NY) with MWCO of 50 kDa (UFP-50-C-

03M) and 500 kDa (UFP-500-C-03M). Each cartridge contains 30 hollow fibers with 5 × 10-4 m

(0.5 mm) ID and 0.3 m length, giving 140 cm2 total membrane area. New membrane filters were

soaked in 25% ethanol or isopropyl alcohol for 1 hr and rinsed with DI water to ensure complete

removal of glycerol that was used during long term storage / shipping. The modules were

preconditioned by circulating the buffer of interest through the retentate and permeate at a feed

transmembrane pressure of 1.6 to 2.8 bar (25 to 40 psi) for several minutes. Transmembrane

pressure (TMP) was calculated as:

𝑇𝑀𝑃 (∆𝑃) = 𝑃𝑓𝑒𝑒𝑑+ 𝑃𝑟𝑒𝑡𝑒𝑛𝑡𝑎𝑡𝑒

2− 𝑃𝑝𝑒𝑟𝑚𝑒𝑎𝑡𝑒 (2.4)

The shear rate at the lumen surface of the membrane was evaluated assuming Poiseuille flow as:

𝛾 = 4𝑞

𝜋𝑅3 (2.5)

31

where γ is the shear rate in s-1, q is volume flow through the fiber lumen in mL/s per fiber and R

is the fiber radius in cm.

The mean pore size of various ultrafiltration membranes was estimated from

measurements of the hydraulic permeability using the Hagen-Poiseuille equation55:

𝑟𝑝 = (8 𝜇𝛿𝑚𝐿𝑝

)1/2

(2.6)

Equation (2.6) assumes that the membrane is composed of a parallel array of uniform cylindrical

pores with the same length and radius. The membrane thickness, δm, was estimated as 1.0 μm for

the skin layer in these asymmetric membranes, and the membrane porosity, ε, was estimated as

50% 56. Table 2.4 lists the average hydraulic permeability and estimated mean pore radius of the

UF membranes used in this work. Interestingly, the permeability of the 500 kDa PS hollow fiber

was considerably lower than that of the Ultracel 100 kDa flat sheet membrane, which could be

due to inconsistency resulting from different membrane configurations and materials.

Table 2.4: Average hydraulic permeability and estimated mean pore radius of various UF

membranes used in this study

Membrane Permeability (m/s/kPa) Mean pore radius (nm)

Ultracel 100 kDa 2.5 × 10-6 6.4

Biomax 100 kDa 3.6 × 10-6 7.7

Biomax 300 kDa 7.7 × 10-6 11

PS hollow fiber 50 kDa 0.4 × 10-6 2.6

PS hollow fiber 500 kDa 1.1 × 10-6 4.3

32

2.5 Membrane characterization

2.5.1 Membrane hydraulic permeability

The membrane hydraulic permeability is an important indicator of membrane

functionality and is evaluated as the slope of data for the filtrate flux (Jv) as a function of the

transmembrane pressure (∆P) using DI water as shown in Equation (2.2). Measurements were

typically taken at 4 different TMP ranging from 14 to 69 kPa (2 to 10 psi), which were

determined by differential pressure measurements using a digital pressure gauge (Ashcroft, CT).

For TFF cartridges, TMP is calculated based on Equation (2.4) and can be adjusted by either

altering the feed flow rate or restricting the retentate outlet. The filtrate flux is calculated as the

volumetric flow rate through the membrane divided by the effective membrane area. The

volumetric flow rate was evaluated by timed collection of filtrate samples using a digital balance

(Mettler Toledo). Membranes were only used when the permeability was within ±20% of the Lp

suggested by the manufacturer.

2.5.2 Scanning electron microscopy (SEM)

The surface characteristics of the clean membranes and membranes after chemical

modification (Chapter 4) were examined by scanning electron microscopy. A small piece of

membrane (approximately 1 cm2) was cut from larger membrane discs and attached to an

aluminum barrel stub using carbon tape. The membrane surface was imaged using a FEI Nova

NanoSEM 630 Field Emission Scanning Electron Microscope (FESEM) and helix detector under

low vacuum mode (without coating to protect the membrane surface) at magnifications up to

150,000 × using a 3.0 to 5.0 kV electron landing energy.

33

2.6 Sieving experiments

2.6.1 Normal flow filtration

Sieving experiments using flat sheet membranes were conducted using 10 mL Amicon

stirred cells (MilliporeSigma) with effective membrane area of 4.1 cm2. Figure 2.5 shows a

schematic of the stirred cell setup. The plasmid solution was added to the stirred cell, and the

entire apparatus was sealed and connected to an air-pressurized feed reservoir. The stirred cell

was placed on a magnetic stir plate (VWR 205 Autostirrer) to ensure proper mixing of the

solution using a stirring speed of 730 rpm. The pressure of the feed reservoir was adjusted to

obtain the desired filtrate flux. The system was allowed to stabilize (typically after filtration of

approximately 1 mL), with a filtrate sample then collected for subsequent analysis. Data were

reported in terms of the observed sieving coefficient (S0), evaluated as the ratio of the plasmid

concentration in the collected filtrate sample to that in the feed solution (Equation 2.1), with the

feed concentration evaluated from the arithmetic average of samples taken from the stirred cell

immediately before and after collecting the filtrate sample. Ultrafiltration experiments were

performed at room temperature (18–23°C), with filtrate and feed samples stored at 4°C until they

were ready to be analyzed.

34

Figure 2.5: Schematic of stirred cell set-up used for ultrafiltration (adopted from Latulippe, 2010).

2.6.2 Tangential flow filtration

The experimental set up used for the hollow fiber modules is shown schematically in

Figure 2.6. The hollow fiber modules were oriented at a 45°angle to the horizontal to facilitate

sampling and filling of the shell region. The feed was driven through the module using a positive

displacement pump (Masterflex, Gelsenkirchen, Germany). The pressures on the feed and

retentate side were measured using pressure sensors and LabQuest software (Vernier, OR), while

the pressure on the filtrate side was atmospheric. The transmembrane pressure was set by

adjusting a valve on the retentate exit line, with the permeate flow rate evaluated via timed

collection of permeate samples obtained throughout the ultrafiltration. Experiments were

performed in a batch concentration mode at constant transmembrane pressure (by adjusting the

35

pinch valve on the retentate exit line), with the retentate recycled back to the feed reservoir while

permeate was removed. The feed reservoir was constantly mixed using a magnetic stir bar

throughout the experiment. Permeate and feed samples were taken periodically for off-line

analysis.

Figure 2.6: Schematic of hollow fiber setup for tangential flow filtration

2.7 Diafiltration

Diafiltration is a well-established method for obtaining high resolution separations using

membrane processes by effectively “washing” the more permeable species through the

membrane. This is accomplished by continuous addition of buffer to replace the fluid withdrawn

in the filtrate 57. Diafiltration has been widely used for the separation of small molecules from

36

therapeutic proteins 58-59 and also for the removal of host cell proteins from recombinant protein

therapeutics 47. During a constant volume diafiltration process, buffer solution is added

continuously throughout the filtration at the same rate as the permeate is removed so that the

total volume in the feed tank remains constant during the process.

Experimental setup of a diafiltration system is similar to that shown in Figure 2.5. The

stirred cell was initially filled with 12 mL of the plasmid solution in TE buffer while the

polycarbonate feed reservoir was filled with a plasmid free TE buffer solution. The reservoir was

left at atmospheric pressure throughout the diafiltration. The desired filtrate flux was set by

connecting a peristaltic pump to the exit filtrate line. The vacuum generated by the pump caused

buffer in the reservoir to be drawn into the stirred cell at the same volumetric flow rate as the

filtrate is removed. Filtrate and feed samples were collected at designated time intervals to

evaluate the concentration and identity of the pDNA isoforms. The plasmid concentration in the

stirred cell during a diafiltration process can be expressed as a function of the number of

diavolumes (ND) as:

𝐶 = 𝐶0 exp (−𝑁𝐷 ∙ 𝑆𝑜) (2.7)

where C0 is the initial plasmid concentration in the stirred cell, S0 is the observed sieving

coefficient of the particular plasmid isoform, and ND equals to the total volume processed

through the membrane divided by the constant retentate volume in the stirred cell. The filtrate

concentration Cfiltrate can therefore be calculated from:

𝐶𝑓𝑖𝑙𝑡𝑟𝑎𝑡𝑒 = 𝐶0 𝑆0exp (−𝑁𝐷 ∙ 𝑆0) (2.8)

37

Chapter 3

Use of Pre-conditioning to Control Membrane Fouling and Enhance Membrane

Performance

The objective of the work described in this Chapter was to examine the potential of using

pre-conditioning, in this case accomplished by pre-elongating the DNA by passage through a

region with large pore size, to minimize fouling and enhance DNA separations. The majority of

the work presented in this chapter was previously published in the Journal of Membrane Science

(Li et al., 479, 117-122, 2015).

3.1 Introduction

As discussed in Chapter 1, one of the major challenges in plasmid DNA ultrafiltration is

membrane fouling48, 60-61. Borujeni and Zydney48 showed that membrane fouling increased

dramatically at high feed concentrations, with the extent of plasmid transmission decreasing

rapidly during the ultrafiltration process due to partial blockage of the membrane pores by

individual plasmids that became trapped at the pore entrance. The rate of fouling was greater for

the larger plasmids due to the greater probability of a plasmid getting trapped at the pore

entrance. This trapping phenomenon was likely due to the incomplete extension of the plasmid

in the converging flow field into the membrane pores and / or the increased likelihood of “knot”

formation in the larger plasmids62-63.

Plasmid “trapping” has also been observed in micro- / nano-fluidic systems designed for

DNA separations, manipulations, and sequencing 64. A variety of approaches have been used to

38

facilitate DNA extension by “pre-conditioning” the DNA in nanofluidic systems, e.g., by

applying shear forces, passing the DNA through a gel matrix or an array of nano-obstacles 65-69,

or using a conically-shaped constriction to gradually elongate the DNA 70-72. For example, Cao

et al. 65 used a gradient array of nanostructures with decreasing spacing to enhance the extension

of long DNA molecules; DNA that passed through the nanostructured array were able to enter

narrow slit-shaped nanopores under conditions in which the DNA was trapped at the pore

entrance in the absence of “pre-conditioning” by the nanostructured array. Different strategies to

pre-condition DNA using microfluidic systems are schematically shown in Figure 3.1.

Figure 3.1: Use of microfluidic devices to pre-condition DNA.A) Stretching by shear flow

(reproduced from Perkins et al., 2007)73; B) Pre-stretching using obstacle arrays

(reproduced from Chan et al., 2004)74; C) Pre-stretching by conical-shaped

microchannel (adopted from Larson et al., 2006)70

The objective of the work described in this Chapter was to examine the effect of “pre-

conditioning” on the transmission and fouling behavior of plasmid DNA during ultrafiltration.

Data were obtained with asymmetric hollow fiber ultrafiltration membranes, oriented with the

flow through either the skin or the substructure first. The results were analyzed using available

39

models for polymer elongation, with the data used to identify conditions that could provide

enhanced separation of the plasmid DNA isoforms. The results clearly demonstrate the potential

of controlling the membrane pore morphology to pre-condition the DNA, thus, reducing the

extent of fouling and enhancing the performance of membrane systems for DNA purification.

3.2 Materials and methods

Ultrafiltration experiments were performed with hollow fiber polysulfone (PS)

membranes obtained from GE Healthcare (Niskayuna, NY) with nominal molecular weight

cutoffs of 50 kDa (UFP-50-C-03M) and 500 kDa (UFP-500-C-03M). These membranes are

highly asymmetric, with the tight “skin” on the inner surface of the fiber lumen.

200 µg/mL stock solutions of 3.0 and 16.8 kbp supercoiled plasmids were prepared by

Aldevron (Fargo, ND) and stored frozen at -20º C. A small amount of the stock solution was

thawed and diluted with TE buffer containing 150 mM NaCl immediately prior to use in the

ultrafiltration experiment. The linear and open-circular isoforms were prepared in our laboratory

by enzymatic digestion of the supercoiled isoform using restriction and nicking endonucleases

(New England Biolabs, MA), respectively. Plasmid concentrations were determined using the

Quant-iT PicoGreen dsDNA assay kit (Life Technologies, Carlsbad, CA). The quality of the

plasmid stock solutions and the effectiveness of the enzymatic digestions were examined by

agarose gel electrophoresis (AGE) following the procedures described in Chapter 2. AGE was

also used to estimate the relative concentrations of the different isoforms in separation

experiments performed using mixtures of the linear and supercoiled isoforms.

40

Plasmid ultrafiltration experiments were conducted in both the normal (from lumen to

shell) and reverse flow directions. Buffer solution was pumped into the module (through the

fiber lumens) using a Masterflex peristaltic pump at a constant feed flow rate of 100 mL/min,

with the transmembrane pressure adjusted using a ball valve installed on the exit retentate line.

During normal operation, the feed flow was introduced into the fiber lumen with permeate

withdrawn through the shell. The module was oriented at a 45° angle to the horizontal with the

upper permeate port kept closed. During ‘reverse’ operation, the feed flow was introduced into

the shell, with filtration occurring through the membrane substructure and then the skin, with the

filtrate collected from the lumen exit. In both cases, data were obtained in total recycle mode,

with the retentate and permeate lines recycled back to the feed reservoir to maintain a uniform

plasmid concentration. Filtrate samples were collected periodically throughout the ultrafiltration

experiment to evaluate the plasmid concentrations.

3.3. Results and discussion

3.3.1 Membrane Orientation

Figure 3.2 shows typical data for the transmission of plasmid DNA through ultrafiltration

membranes. Experiments were performed using the 16.8 kbp supercoiled plasmid through a 500

kDa and a 50 kDa hollow fiber membrane in normal (lumen-to-shell) operation. The plasmid

transmission increased significantly with increasing filtrate flux for data obtained using the 500

kDa PS membranes. For example, the sieving coefficient increased from So < 0.05 at a filtrate

flux of 6.7 µm/s (24 L/m2/h) to So ≈ 0.9 at a filtrate flux of 45 μm/s (160 L/m2/h). This increase

in plasmid transmission with increasing filtrate flux is due to elongation of the plasmid in the

41

fluid flow entering the membrane pore as discussed by Latulippe and Zydney43. Note that the

radius of gyration of the 16.8 kbp plasmid is approximately 170 nm 75, which is more than 10

times larger than the less than 5 nm mean pore size of the 500 kDa membrane used in these

experiments (Table 2.4). In contrast, the transmission of plasmids through the 50 kDa membrane

was very low (So < 0.1) over the entire range of filtrate flux tested, indicating that the critical flux

for DNA transmission is above 60 µm/s. This is consistent with the analysis provided by

Latulippe and Zydney43, where the critical flux is reversely proportional to the square of the pore

radius.

Figure 3.2: Observed sieving coefficient of the supercoiled 16.8 kbp supercoiled plasmid through

a 500 kDa and 50 kDa hollow fiber membrane in normal orientation. Data were

obtained at a feed flow rate of 100 mL/min using a plasmid concentration of 0.25

µg/mL

0

0.2

0.4

0.6

0.8

1

1 10 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux Jv, (µm/s)

500 kDa

50 kDa

42

Figure 3.3 shows data for the observed sieving coefficient of the 16.8 kbp supercoiled

plasmid during ultrafiltration of a 3 μg/mL (3 × 10-3 kg/m3) solution through a 500 kDa hollow

fiber membrane in both the forward (lumen-to-shell) and reverse (shell-to-lumen) orientations.

Data were obtained at a feed flow rate of 100 mL/min using a constant transmembrane pressure

of 55 kPa (8 psi), yielding an initial filtrate flux of 45 μm/s (160 L/m2/h) in both orientations,

consistent with the very similar values of the membrane permeability determined in the lumen-

to-shell and shell-to-lumen directions. The change of the sieving coefficient throughout the

filtration is plotted as a function of the cumulative filtrate volume normalized by the membrane

area; both experiments were conducted for a total filtration time of more than 840 s. When the

membrane was used in the reverse orientation, the observed sieving coefficient and the filtrate

flux remained essentially constant throughout the ultrafiltration experiment with So values

between 0.92 and 0.87. This high degree of plasmid transmission is again due to the stretching

of the plasmid in the elongational flow field entering the membrane pore.

In contrast to the results in the reverse orientation, the observed sieving coefficient in the

forward orientation (skin-side facing the feed) decreased significantly during the ultrafiltration,

going from So > 0.8 at the start of the experiment to less than 0.4 after filtration of 35 L/m2 due

to membrane fouling. The filtrate flux also decreased slightly during filtration in the forward

direction, from Jv = 47 to 41 μm/s. This small decline in filtrate flux (compared to the large

decline in plasmid transmission) is due to fluid flow around the trapped plasmids in the partially

blocked pores. This is discussed in more detail by Borujeni et al.76.

43

Figure 3.3: Observed sieving coefficients (upper panel) and filtrate flux (lower panel) during

constant pressure ultrafiltration of a 3 µg/mL solution of the supercoiled 16.8 kbp

plasmid through a 500 kDa hollow fiber membrane in both the lumen-to-shell

(forward) and shell-to-lumen (reverse) orientations.

The absence of any fouling in the reverse flow orientation is likely due to the gradual

elongation of the plasmid DNA in the more open pores within the support structure of the hollow

44

fiber membrane. This gradual elongation, or “pre-conditioning”, of the plasmid minimizes the

likelihood of the plasmid getting trapped at the entrance to the narrow (less than 20 nm) pores in

the skin layer of the membrane. This pre-conditioning is absent when the fluid flow is in the

forward orientation, with the plasmids immediately exposed to the very narrow pores. This

behavior is in good agreement with results obtained by Cao et al.65 and Tegenfeldt et al.77 in

nanofluidic devices using an array of nanopores to pre-condition the DNA.

The effect of pre-conditioning on the observed sieving coefficient of the 16.8 kbp

plasmid is examined in more detail in Figure 3.2. Data were obtained with a 0.25 µg/mL

solution of the supercoiled plasmid using a single 500 kDa hollow fiber membrane in two

separate experiments, one with flow in the forward direction (through the skin first) and one with

flow in the reverse direction (through the substructure first). In both cases, the flux was

increased from low to high, with approximately 3 min at each condition to insure complete

washout of the hold-up volume in the hollow fiber system (approximately 2 mL inside the fiber

lumen and 9 mL in the shell region). There was no evidence of any fouling in these experiments

due to the very low concentration of the plasmid; the plasmid sieving coefficient remained

constant at each pressure and the membrane hydraulic permeability after filtration was within 5%

of that of the fresh membrane. There was also no degradation or topological changes in the

plasmid due to either operation of the peristaltic pump or passage of the plasmid through the

membrane as confirmed by agarose gel electrophoresis.

45

Figure 3.4: Observed sieving coefficient of the supercoiled 16.8 kbp supercoiled plasmid through

a 500 kDa hollow fiber membrane in the reverse and forward orientations. Data were

obtained at a feed flow rate of 100 mL/min using a plasmid concentration of 0.25

µg/mL.

Plasmid transmission in both the forward and reverse orientations was a strong function

of filtrate flux. For example, the sieving coefficient in the reverse orientation increased from So

= 0.10 to So > 0.8 as the filtrate flux increased from approximately 3.4 to 8.5 μm/s

(corresponding to 12 to 31 L/m2/h). A qualitatively similar behavior was seen in the forward

direction; however, the sieving curve was strongly shifted to the right (higher filtrate flux). In

both cases, plasmid transmission was negligible at very small values of the filtrate flux; plasmid

elongation (and thus transmission) only became significant above a critical value of the filtrate

flux as discussed by Latulippe et al.37. The critical flux for plasmid transmission was estimated

by extrapolation of the sieving coefficient data to zero on a linear plot (using data with So values

0

0.2

0.4

0.6

0.8

1

1 10 100

Sie

vin

g C

oeff

icie

nt,

S0

Filtrate Flux, Jv (μm/s)

reverse

forward

46

less than 0.5), giving Jcrit = 1.8 ± 0.1 µm/s in the reverse orientation compared to 6.3 ± 0.3 µm/s

in the forward orientation.

The large difference in plasmid transmission in the forward and reverse orientations is

directly related to the different hydrodynamics in the two flow directions as shown schematically

in Figure 3.5. When the flow is in the forward direction, plasmid elongation occurs in the

converging flow field as the fluid approaches the pore. In contrast, the converging flow in the

reverse orientation is determined by the geometry of the “conical-shaped” pore as one moves

through the substructure and towards the skin of the highly asymmetric membrane. Daoudi and

Brochard 42 used scaling arguments to evaluate the critical flux for the transmission of a highly

flexible chain for the two geometries sketched in Figure 3.5. In each case, polymer transmission

was assumed to occur when the Deborah number (the ratio of the time scale for polymer

relaxation to the characteristic time scale for the fluid flow) was greater than or equal to one. For

a single cylindrical pore, analogous to the behavior of the hollow fiber membrane in the forward

orientation, the critical flux was scaled as shown previously in Equation (1.5):

𝐽𝑐𝑟𝑖𝑡 ≈ 𝑘𝐵𝑇

𝜇𝑟𝑝2 (1.5)

where ε is the membrane porosity, kB is the Boltzmann constant, T is the absolute temperature, μ

is the fluid viscosity, and rp is the pore radius. In contrast, the critical flux for a conical pore was

given as 42:

𝐽𝑐𝑟𝑖𝑡′ ≈ α𝐽𝑐𝑟𝑖𝑡 (3.1)

where α = D/x is a measure of the angle of the cone. The parameter α can be estimated for the

hollow fiber membrane based on the pore size in the membrane substructure (D ≈ 1 µm) and the

47

thickness of the region of the hollow fiber where the pore size begins to narrow as one

approaches the skin (x ≈ 10 µm) giving α ≈ 0.1. This is in good order of magnitude agreement

with the data in Figure 3.4, which give a ratio of critical flux in the reverse and forward

directions of 0.3 ± 0.1. Note that although Equations (1.5) is in good qualitative agreement with

the data, the actual values of the critical flux were much larger than those found experimentally.

This behavior was discussed previously by Latulippe and Zydney 43.

Figure 3.5: Schematic diagram of plasmid transmission through an asymmetric membrane in the

(A) forward and (B) reverse orientations.

3.3.2 Effects of plasmid and membrane pore size

The effects of the plasmid and membrane pore size on the observed sieving coefficients

are examined in Figure 3.6. Data were obtained with 0.25 μg/ml solutions of the 3.0 and 16.8

kbp plasmids in TE buffer containing 150 mM NaCl using the 500 kDa and 50 kDa hollow fiber

48

membranes, both with flow in the shell-to-lumen (reverse) orientation. The observed sieving

coefficients for the two plasmids through the 50 kDa membrane were very similar over the entire

range of flux. For example, at Jv ≈ 40 μm/s the sieving coefficient of the 3.0 and 16.8 kbp

plasmids were 0.30 and 0.29, respectively, with Jcrit values of 30 ± 3 and 28 ± 1 μm/s. Similar

results were obtained with the 500 kDa membrane, although the data for the 16.8 kbp plasmid

did appear to be shifted slightly to the right (higher flux or lower sieving coefficients). This

small difference may simply be due to the use of different hollow fiber modules for the two

experiments. The effect of plasmid size on the ultrafiltration of the supercoiled plasmid is

discussed further in Chapter 8.

The plasmid sieving coefficients through the 500 kDa membrane are substantially larger

than the values with the 50 kDa membrane, consistent with the very large difference in critical

flux: 1.7 ± 0.2 µm/s for the 500 kDa membrane and 29 ± 3 µm/s for the 50 kDa membrane. If

we assume that the conical pore angle is approximately the same in both hollow fibers, the ratio

of the critical flux values should be proportional to the square of the difference in pore size (as

given by Equation 1.5). For a membrane with uniform cylindrical pores, the effective pore size

can be estimated assuming Poiseuille flow as:

Lp =εrp

2

8δ (3.2)

where δ is the membrane thickness. The ratio of the permeability of the 50 and 500 kDa

membranes was 0.35 ±0.1 (assuming equal porosity and membrane thickness), which is almost

one order of magnitude greater than the ratio of the critical flux values (0.06). This discrepancy

could be due to the effects of a pore size distribution and hydrodynamic interactions between

adjacent pores, as well as the heterogeneity in the DNA molecules.

49

Figure 3.6: Observed sieving coefficients of the supercoiled 3.0 kbp (p-EMP) and 16.8 kbp (p-

FDY) plasmids through the 500 kDa and 50 kDa PS hollow fiber membranes in the

reverse orientation.

3.3.3 Isoform separation

Figure 3.7 shows results for the transmission of the linear and supercoiled isoforms of the

3.0 kbp plasmid in TE buffer containing 150 mM NaCl through the 50 kDa hollow fiber

membrane in both the shell-to-lumen (left panel) and lumen-to-shell (right panel) orientations.

In both cases, transmission of the linear isoform is greater than that of the supercoiled isoform

due to the greater elongational flexibility of the linear plasmid. The difference in transmission

between the two isoforms is much more pronounced when the filtration is performed in the shell-

to-lumen (reverse) orientation. For example, at a filtrate flux of 20 μm/s (72 L/m2/h) the sieving

0

0.2

0.4

0.6

0.8

1

1 10 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux, Jv (μm/s)

50

coefficient for the linear isoform is greater than 0.98 while that for the supercoiled isoform is less

than 0.02, a selectivity of more than 40-fold for the membrane in the reverse orientation. In

contrast, the highest selectivity that was obtained when using the membrane in the lumen-to-shell

(forward) orientation was only around 2-fold.

Figure 3.7: Observed sieving coefficients for the linear and supercoiled isoforms of the 3.0 kbp

plasmid through the 50 kDa hollow fiber membrane in the reverse (left panel) and

forward (right panel) orientations.

Based on the results in Figure 3.7, ultrafiltration experiments were performed using a

binary mixture of the supercoiled and linear isoforms of the 3.0 kbp plasmid, each at a

concentration of 0.25 µm/mL, in TE buffer containing 150 mM NaCl. Data were obtained using

the 50 kDa membrane in the reverse orientation (flow from shell to lumen) at filtrate flux values

of 10 and 20 µm/s, with the feed and filtrate samples analyzed by agarose gel electrophoresis

(AGE) as shown in Figure 3.8. The first lane shows a DNA ladder used for calibration. The

1 10 100

Filtrate Flux, Jv (µm/s)

supercoiled

linear

0

0.2

0.4

0.6

0.8

1

1 10 100

Sie

vin

g C

oeff

icie

nt,

S0

Filtrate Flux, Jv (µm/s)

51

feed sample is shown in Lane 2; the top and bottom bands correspond to the linear and

supercoiled isoforms, respectively. The filtrate sample, obtained at a flux of 10 μm/s (Lane 3),

did not contain either plasmid (at the sensitivity of the AGE analysis), consistent with the high

degree of plasmid retention seen in Figure 3.7 under these conditions. The filtrate sample

obtained at a flux of 20 μm/s (Lane 4) shows a significant band corresponding to the linear

isoform, with no measurable amount of the supercoiled DNA. Although it is difficult to

quantitatively evaluate the DNA concentrations from the AGE gels, the results in Figure 3.8

clearly demonstrate the potential of using the 50 kDa hollow fiber membrane in the reverse

orientation to obtain highly selective separation between the linear and supercoiled isoforms.

Figure 3.8: Agarose gel electrophoresis image of the feed and filtrate samples during ultrafiltration

of a binary mixture of the linear and supercoiled isoforms of the 3.0 kbp plasmid.

Lane 1 - linear 1 kbp DNA ladder; Lane 2 - feed sample; Lane 3 - filtrate sample at

Jv = 10 μm/s; Lane 4 - filtrate sample at Jv = 20 μm/s.

3.4 Conclusion

52

The experimental data presented in this Chapter clearly demonstrate the potential of using

pre-conditioning to reduce membrane fouling and enhance the separation performance during

ultrafiltration of plasmid DNA through narrow pore size membranes. Pre-conditioning was

accomplished simply by operating the asymmetric membrane in the reverse orientation, i.e., with

the flow directed through the open pores in the substructure before the skin. Pre-stretching of the

plasmid in the more open pores increased the extent of plasmid transmission through the skin

layer of the membrane and it reduced the extent of fouling by decreasing the likelihood of the

plasmid becoming trapped at the pore entrance, similar to results seen in nanofluidic (lab on a

chip) systems for DNA manipulation.

The increase in plasmid transmission was consistent with predictions of a simple scaling

model developed by Daoudi and Brochard 42 to describe the elongation of an idealized polymer

chain through a conical pore. The dependence of the critical flux on the plasmid and pore size

was also consistent with this model. Interestingly, pre-conditioning also caused a significant

increase in the selectivity of the hollow fiber membrane for the separation of the linear and

supercoiled isoforms due to the greater enhancement in transmission of the linear isoform. This

enhanced selectivity provided very high resolution in the separation of these DNA isoforms.

These results clearly demonstrate the potential of enhancing the performance of DNA

ultrafiltration by controlling the pore morphology to properly pre-condition the DNA prior to

transmission through the very narrow pores in the ultrafiltration membranes.

53

Chapter 4

Preconditioning with Multilayer Composite Membranes

The results in Chapter 3 demonstrated that pre-conditioning is a very effective approach

for reducing membrane fouling and enhancing the selectivity for the separation of different

plasmid isoforms. The pre-elongation of DNA molecules can also be accomplished by placing a

large pore size flat sheet microfiltration membrane in series with an ultrafiltration membrane.

This chapter discusses the use of this type of multilayer composite membrane structure,

including the use of membranes with different pore size, for the ultrafiltration of plasmid DNA.

4.1 Introduction

The emergence and rapid development of composite membranes has been a major

milestone in membrane technology since the late 1970s. Primarily developed for reverse

osmosis (RO) and nanofiltration (NF) for water treatment applications, these ‘thin-film

composite’ (TFC) membranes generally contain a thin dense polymer skin formed over a

microporous support 78. Although they are closely related to the Loeb-Sourirajan asymmetric

membrane, the dense skin is formed independently in a separate step. This results in an abrupt

discontinuity between the microporous region and the skin layer. Each layer can have very

different pore size distribution, with isotropic or anisotropic morphology, with appropriate aspect

ratio and thickness. This provides a high degree of flexibility in tailoring the composite

membrane structure / properties for specific application. Figure 4.1 shows a nice comparison of

the isotropic, anisotropic and composite membrane structures.

54

Figure 4.1: Comparison of A) isotropic B) anisotropic and C) composite membrane structure in

sterile filters (reproduced from79)

There are a variety of methods for forming composite membranes, including but not

limited to 80:

I. Interfacial polymerization (IP) of reactive monomers on the surface of the support film

II. Casting an ultrathin layer separately, followed by lamination to the support film

III. Dip coating a polymer solution onto the support and drying; or dip-coating a reactive

monomer/prepolymer solution followed by curing with heat or irradiation

IV. Gas phase deposition of the skin layer from a glow-discharge plasma

55

The different layers of the thin film composite membranes can be independently optimized to

achieve the desired selectivity and permeability while offering excellent mechanical strength and

compression resistance.

As demonstrated in Chapter 3, the performance of an ultrafiltration membrane for DNA

separations can be significantly enhanced by “pre-conditioning” the feed by passing the plasmid

solution through a region with large pore size before allowing the plasmids to approach the

narrow pore region. In Chapter 3, this pre-conditioning was achieved by using a highly

asymmetric ultrafiltration membrane in reverse orientation. However, these ultrafiltration

membranes were not designed for this application; thus, the relative size and thickness of the

different regions may not be ideally suited for DNA ultrafiltration. It is also possible to achieve

the pre-conditioning by using composite membrane structures. This is most easily accomplished

by simply physically layer two different membranes together, each of which can be selected to

have the most appropriate pore size and morphology.

The objective of the work described in this Chapter was to examine the performance of

multilayer composite membranes for DNA ultrafiltration, including evaluation of the key factors

that determine the effectiveness of the pre-conditioning layer. In addition, several physical

models are developed to help understand the effects of pre-conditioning on DNA transmission.

Further confirmation of these models was obtained using membranes with conically shaped

pores. These results provide important insights into the role of preconditioning in enhancing the

performance of DNA ultrafiltration, while providing a framework for the design and

optimization of membranes specifically for plasmid DNA separations.

56

4.2 Materials and methods

Ultrafiltration experiments were conducted with Ultracel composite regenerated cellulose

membranes (MilliporeSigma, MA) with MWCO of 100 kDa. These membranes were used in a

composite (sandwich) structure with a variety of microfiltration membranes with different pore

sizes and cast from different polymers as listed in Table 4.1. All membranes are hydrophilic with

low protein binding. Figure 4.1 shows scanning electron microscopy images of the upper surface

of the different microfiltration membranes used in this study. The polycarbonate membranes are

made by a track-etch process yielding very uniform cylindrical pores that are not interconnected.

In contrast, the polyvinylidene fluoride and mixed cellulose ester membranes have a fibrous

network structure resulting in an isotropic (interconnected) pore morphology.

57

Table 4.1: Specifications of microfiltration membranes used in this study

Membrane Material Rating (µm) Thickness (µm) Porosity (%)

Durapore

(Millipore)

Polyvinylidene fluoride

(PVDF)

0.1 125 70

0.22 125 70

0.45 125 70

Nuclepore

(Whatman) Polycarbonate (PC)

0.1 6 2.4

0.2 10 10

0.4 10 13

MF

(Millipore) Mixed cellulose ester

0.1 105 74

0.22 150 75

Figure 4.2: Scanning electron microscopy images of the upper surface of the Durapore, Mixed

Cellulose Ester (compiled from https://www.emdmillipore.com/US/en/product/MF-

Millipore), and Nuclepore membranes

58

Membranes with conical shaped pores were prepared by chemical etching of track-etched

polycarbonate films with pore size of 0.015 µm (Whatman) following the procedures described

in Harrell et al. (2006)81 and Scopece et al. (2006)82. The thickness of the original membrane is 6

µm with a surface pore density of 3 × 108 pores/cm2. Figure 4.3 shows a schematic of the

experimental apparatus. A polycarbonate membrane was placed between two halves of the glass

chamber, and an etching solution (8 M NaOH) was added to one of the half cells. The other half

cell was filled with a stop solution of 1 M KCl and 1 M formic acid. A Pt electrode was placed

into each solution (anode in the etching solution, cathode in the stop solution), and a constant

transmembrane potential difference was applied during etching. A voltage greater than +0.1 V

was used to prevent the etching ion (OH-) from entering the other side of the chamber83. Etching

was performed at room temperature (23°C). The cone angle in the etched membrane can be

controlled by varying the applied transmembrane potential and / or the etching time. After

etching, the membrane was removed from the chamber and immediately immersed in 1 M

formic acid for 1 hr to neutralize any base left in the pores. The membrane was then soaked in DI

water for 1 hr at 40°C, rinsed thoroughly, and stored in air. Field Emission Scanning Electron

Microscopy (FESEM, FEI Nova NanoSEM 630) was used to measure the diameter of the pores

on both sides of the membrane to obtain insights into the conical shape of the pores.

59

Figure 4.3: Schematic of etching apparatus for preparation of polycarbonate membranes with

conical shape pores

4.3 Results and Discussions

4.3.1 Multilayer composite membrane

Initial experiments were performed using composite (or “sandwich”) membranes made

by placing an asymmetric Ultracel 100 kDa ultrafiltration membrane in series with a symmetric

(0.22 µm pore size) Durapore microfiltration membrane. Three sets of experiments were

performed: one with the Ultracel membrane alone in the normal (skin-up) orientation, one with

the Ultracel membrane in the reverse (skin-down) orientation, and one with a Durapore

membrane placed directly on top of the Ultracel membrane (with the Ultracel in the skin-up

orientation). Sieving data were obtained with a 0.25 µg/mL solution of the 3.0 kbp supercoiled

plasmid in TE buffer containing 100 mM NaCl; this low concentration was used to avoid

artifacts associated with membrane fouling. The results are summarized in Table 4.2 at filtrate

flux values of 40 and 60 μm/s (140 and 220 L/m2/h).

60

Plasmid transmission through the Ultracel 100 kDa membrane with the skin-side up was

low (So < 0.15) at both fluxes, consistent with the high value of the critical flux (> 60 µm/s)

determined by Latulippe et al. (2007)37 for this membrane. In contrast, there was significant

transmission of the supercoiled plasmid through the membrane in the skin-side down orientation

(substructure facing the feed), with So = 0.67 at Jv = 60 µm/s. The results with the composite

(sandwich) membrane are similar to those for the Ultracel membrane in the skin-side down

orientation; placing the Durapore membrane on top of the Ultracel membrane caused more than a

3-fold increase in plasmid transmission. This large increase in plasmid transmission is due to the

pre-elongation of the plasmid in the large pores in either the Durapore membrane (for the

composite structure) or the support layer of the Ultracel membrane (when used in the skin-side

down orientation). Note that the Ultracel membrane is itself a composite structure formed by

casting a regenerated cellulose layer on a polyethylene microporous substrate with a pore size

similar to that of the Durapore membrane 47.

Table 4.2: Comparison of observed sieving coefficients of a 3.0 kbp supercoiled plasmid through

an Ultracel 100 kDa membrane with the skin-side up, the skin-side down, and in a

composite structure with a 0.22 µm Durapore membrane

Observed Sieving Coefficient, So

Jv ≈ 40 µm/s Jv ≈ 60 µm/s

Skin Up 0.06 0.13

Skin Down 0.51 0.67

Composite

Membrane 0.20 0.42

61

4.3.2 Effects of upper layer structure

In order to further explore the effect of the upper layer membrane pore size and structure

on the pre-conditioning, a series of experiments were performed using different microfiltration

membranes placed directly on top of an Ultracel 100 kDa ultrafiltration membrane. All

experiments were conducted in TE buffer containing 100 mM NaCl.

Figure 4.4 shows the effects of the pore size of the upper layer membrane on plasmid

transmission through the Ultracel 100 kDa membrane. Experiments were conducted with 0.1,

0.22, and 0.45 µm Durapore membranes as the top layer; these are all homogeneous membranes

with very similar pore morphology. Ultrafiltration experiments performed with a single layer

Ultracel 100 kDa in both the normal (skin up) and flipped (skin down) orientation are also

included for comparison. The sieving coefficients at a filtrate flux of around 20 μm/s were

minimal at all membranes, indicating that the critical flux for plasmid transmission was above

this value for all of the composite membranes examined in Figure 4.4. The greatest plasmid

transmission was obtained using the flipped membrane, with the composite membrane made

used the 0.22 µm Durapore having the greatest sieving coefficients compared to the other

composite membrane structures. The benefits of pre-elongation were slightly reduced using the

0.1 µm pore size Durapore membrane, while DNA transmission through the composite

membrane formed with the 0.45 µm Durapore was nearly indistinguishable from that of the

Ultracel 100 kDa membrane when used in the normal (skin up) orientation.

62

Figure 4.4: Effects of pre-conditioning with different pore size Durapore microfiltration

membranes (upper layers) on transmission of the supercoiled 3.0 kbp plasmid

through the composite membranes. Experiments were performed in TE buffer

containing 100 mM NaCl, with an UltracelTM 100 kDa membrane as the bottom layer.

The data in Figure 4.4 are re-plotted in Figure 4.5 as an explicit function of the pore size

in the upper layer (microfiltration) membrane at several distinct values of the filtrate flux. Where

necessary, the sieving coefficients were determined by linearly interpolating between results

obtained at slightly higher and lower fluxes. Results obtained with a single ‘skin up’ Ultracel 100

kDa membrane were plotted at a pore size in the upper layer equal to 0. It is clear from Figure

4.5 that the effect of pre-conditioning is directly related to the average pore size in the large pore

region. The plasmid transmission was greatest when the pore size in the upper membrane layer

was at an intermediate value, in this case corresponding to the 0.22 µm pore size Durapore

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux, Jv (µm/s)

No Preconditioning

0.1 μm Durapore

0.22 μm Durapore

0.45 μm Durapore

Flipped Membrane

63

membranes. Preconditioning using an upper layer with a larger pore size had a minimal effect on

DNA transmission, suggesting that the plasmid can only be pre-stretched effectively when the

pore size is sufficiently small that passage of the DNA through the membrane causes enough

force to significantly stretch the DNA. The use of an upper layer with a pore size below 0.2 µm

requires the DNA to elongate almost as much as the ultrafiltration membrane itself, causing a

reduction in the sieving coefficient. The very high transmission seen with the Ultracel membrane

in the reverse (flipped) orientation may be due to the graded pore size in moving from the

substructure towards the thin skin layer. This is discussed in more detail subsequently.

Figure 4.5: Sieving coefficient for the supercoiled 3.0 kbp plasmid through composite membranes

as a function of the pore size of the upper layer Durapore microfiltration membranes

at filtrate flux of 40, 60 and 80 µm/s.

Two additional sets of ultrafiltration experiments were conducted using different pore

size mixed cellulose ester (MF) and Nuclepore polycarbonate membranes as the preconditioning

layer, all in composite membrane structures on top of an Ultracel 100 kDa membrane (Figures

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6

Sie

vin

g C

oe

ffic

ien

t, S

0

Upper Layer Pore Size (μm)

40 µm/s

60 µm/s

80 µm/s

64

4.6 and 4.7). In both cases, preconditioning resulted in a significant increase in plasmid

transmission. For example, at a fixed filtrate flux of 60 µm/s, preconditioning using 0.1 and 0.22

µm MF membranes increased the sieving coefficient from 0.13 to 0.60 and 0.31, respectively,

corresponding to 4.6- and 2.4-fold increases in plasmid transmission. Similarly, preconditioning

with 0.1, 0.2 and 0.4 µm Nuclepore membranes yielded sieving coefficients of 0.55, 0.66, and

0.36, respectively. Similar to results obtained with the Durapore membranes, the greatest DNA

transmission was obtained with an intermediate pore size for the upper layer, although in this

case the maximum transmission was obtained with a 0.1 µm pore size for the mixed cellulose

ester membranes and a 0.2 µm pore size for the Nuclepore membranes.

Figure 4.6: Effects of pre-conditioning with different pore sizes of mixed cellulose ester (MF)

microfiltration membranes on the transmission of the supercoiled 3.0 kbp plasmid

through the composite membranes.

0

0.2

0.4

0.6

0.8

1

30 50 70 90

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux, Jv (μm/s)

0.1 μm MF

0.22 μm MF

No preconditioning

Flipped membrane

65

Figure 4.7: Effects of pre-conditioning with different pore size Nuclepore microfiltration

membranes on transmission of the supercoiled 3.0 kbp plasmid through the

composite membranes.

The Durapore and mixed cellulose ester (MF) membranes have similar pore structures

(SEM images in Figure 4.2), bulk porosity, and thickness (values in Table 4.1). Thus, it is

unclear why these membranes didn’t have the same behavior when used as a preconditioning

layer for the ultrafiltration of plasmid DNA. Note that in addition to the difference in optimal

pore size, preconditioning with the mixed cellulose ester membranes gave significantly greater

DNA transmission than preconditioning using the Durapore membranes. The mixed cellulose

ester and Durapore membranes do have very different surface chemistry, with the cellulose

acetate / cellulose nitrate having similar properties to the composite regenerated cellulose in the

Ultracel membranes. This may have improved the adhesion between the mixed cellulose ester

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux, Jv (µm/s)

No Preconditioning

0.1 μm Polycarbonate

0.2 μm Polycarbonate

0.4 μm Polycarbonate

Flipped membrane

66

and Ultracel membranes, enhancing DNA transmission. In addition, SEM images of mixed

cellulose ester membranes show a much ‘lacier’ structure with somewhat higher surface porosity

84-85, which could be more effective in pre-elongating the DNA. It is also possible that there are

differences in the interconnectivity 86 and tortuosity of the pore structures in mixed cellulose

ester and Durapore membranes, although the exact impact of these physical parameters on the

elongation of DNA is unclear.

The structure of the Nuclepore polycarbonate track-etched membranes is very different

from that of the Durapore and mixed cellulose ester membranes. The Nuclepore membranes have

straight-through cylindrical pores formed by bombardment of thin polycarbonate sheets with

fission fragments from radioactive decay of high atomic weight isotopes followed by appropriate

chemical etching of the polymer. As a result, the porosity and thickness of this type of membrane

are much smaller than for other microfiltration membranes (Table 4.1). DNA transmission

through composite membranes formed with an upper layer of a Nuclepore membrane was greater

than that with a Durapore membrane having the same effective pore size, which is likely due to

primarily to the much lower porosity (which leads to higher elongational forces at the same

filtrate flux), possibly in combination with the low tortuosity. However, membrane fouling was

found to be greater in composite membranes formed using the Nuclepore membranes, especially

when the upper layer had a small pore size of 0.1 µm. For example, the hydraulic permeability of

the 0.1 µm Nuclepore membrane dropped by more than 40% after filtration of less than 10 µg

(50 mL of the 0.20 µg/mL feed solution) of the supercoiled 3.0 kbp plasmid at a filtrate flux of

80 µm/s. Note that the data in Figure 4.7 were obtained using 0.1 µg/mL DNA solutions at a

filtrate flux below 80 µm/s to minimize the extent of membrane fouling. The high rate of fouling

67

is directly related to the low porosity of these membranes, leading to a rapid rate of pore

blockage during DNA filtration.

An additional series of experiments were performed with the 9.8 kbp supercoiled plasmid

(Figure 4.8). In this case, DNA transmission was largely unaffected by the preconditioning,

irrespective of the pore size of the upper (Durapore) layer. In contrast, pre-conditioning by

operating the Ultracel membrane in the reverse orientation, i.e., with flow through the membrane

substructure before the skin, yielded sieving coefficients that were much larger than those of the

same membrane oriented with the skin-side up. For example, at a filtrate flux of 40 µm/s, the

sieving coefficient with the flipped Ultracel membrane was more than 8 times larger than that

obtained with any of the composite membranes. Experiments performed with the 16.8 kbp

supercoiled plasmid also showed a dramatic increase in DNA transmission using the flipped

Ultracel 100 kDa membrane (data not shown), with the sieving coefficient increasing from 0.05

to 0.48 at filtrate flux of 70 µm/s. These results suggest that preconditioning of the larger size

DNA might require a gradually converging pore structure, similar to what exists in moving from

the substructure to the skin in the highly asymmetric Ultracel membranes (when used in the

reverse orientation).

68

Figure 4.8: Effects of pre-conditioning with different pore size upper (Durapore) membranes on

transmission of the supercoiled 9.8 kbp plasmid through the composite membranes.

A corresponding set of experiments was performed with open-circular DNA isoform in

order to explore the effect of preconditioning on the separation of plasmid isoforms using

composite membranes. 0.1 µm MF membrane and 0.22 µm Durapore membranes were selected

as they provided the most effective pre-elongation of the supercoiled plasmids. The selectivity

between the supercoiled and open-circular plasmid is defined as the ratio of the sieving

coefficients for the two isoforms. The selectivity obtained using composite membranes having a

0.1 µm mixed cellulose ester MF membrane as the upper layer was comparable to that obtained

without any preconditioning, since the MF membrane caused a similar relative increase in

transmission of both the supercoiled and open-circular isoforms (Figure 4.9). In contrast, the

composite membrane made using a 0.22 µm Durapore membrane as the upper layer had a

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100 120

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux, Jv (μm/s)

No preconditioning

0.1 μm Durapore

0.22 μm Durapore

0.45 μm Durapore

Flipped membrane

69

selectivity of more than 25 at a flux of 60 µm/s due to the large increase in transmission of the

supercoiled plasmid while the sieving coefficient of the open-circular plasmid was still quite low.

Figure 4.9: Selectivity of supercoiled 3.0 kbp plasmid over open-circular plasmid through

different composite membranes. Experiments were performed in TE buffer

containing 100 mM NaCl, with UltracelTM 100 kDa membrane as bottom layer.

4.3.3 Physical model

There are three different physical geometries that might be appropriate to describe the

effects of pre-conditioning on DNA transport through the composite membranes as shown

schematically in Figure 4.10.

0

5

10

15

20

25

30

30 50 70 90

Se

lec

tivit

y

Filtrate Flux, Jv (μm/s)

0.1 μm MF

0.22 μm Durapore

No preconditioning

70

Figure 4.10: Physical model for DNA transport through different pore morphologies A) conical

shape B) gradual transition funnel shape and C) sudden onset funnel shape

The pore structure in integrally skinned asymmetric membranes, when operated in the skin-side

down orientation, can probably be best described by the conically shaped pore structure seen in

Figure 4.10A. The reduction in critical flux in this pore geometry is related to the angle of the

cone, as discussed in Chapter 3. The funnel-shaped pores in Figure 4.10B and C likely provide a

better description of the geometry of the composite membranes made by placing a larger pore

size microfiltration membrane on top of the skin layer for a tight ultrafiltration membrane.

In order to study the effect of the conical pore structure in more detail, a series of

membranes were prepared by anisotropic etching of track-etched polycarbonate membranes 87.

Table 4.3 lists the conditions used for the anisotropic etching along with the pore size measured

on the upper and lower surface of these membranes by FESEM. In each case, the mean pore size

was calculated by averaging the diameter of 20 separate pores. The angle of the anisotropy was

controlled by changing the etching time; the transmembrane voltage was maintained at + 5 V and

the NaOH concentration in the etching solution was kept at 8 M for all experiments. Figure 4.11

shows the SEM of the surfaces of anisotropically etched polycarbonate membranes. The average

permeability of the original 0.015 µm Nuclepore membranes was smaller than 0.1 × 10 -6

m/s/kPa. The permeability of the anisotropically etched membranes was increased due to the

71

large increase in pore size on one side of the membrane. Membranes with this particular conical

shaped pores will be used to study effects of pore angle on ultrafiltration of different isoforms of

plasmid DNA.

Table 4.3: Dimensions of membranes with conical shape pores created by anisotropic chemical

etching of 0.015 µm Nuclepore membranes

Etching Time (min) Pore Size (etch side) Pore Size (stop side) Permeability (µm/s/kPa)

20 35 25 0.1±0.05

30 40 27 0.15±0.05

40 130 25 0.8±0.10

Figure 4.11: SEM images of 0.015 µm Nuclepore membranes etched for A) 20 B) 30 and C)

40min. Images show membrane surfaces in the etching solution.

72

Smith et al.88 and Larson et al.70 used florescence microscopy to directly observe the

extension of different size DNA molecules at different shear rates. The stretching of DNA under

shear flow can be quantified in terms of the Weissenberg number (similar to the Deborah

number):

𝑊𝑖 = 𝜀⊥𝜏 (4.1)

where ε⊥ is the strain rate perpendicular to flow and τ is the longest relaxation time of the

polymer. For example, the average strain rate in a Nuclepore membrane with 0.1 µm cylindrical

pores at a filtrate flux of 60 µm/s is 24 ms-1 (velocity gradient of Jv/ε over a 0.05 µm distance

from pore wall to axis). The longest relaxation time of DNA can be estimated based on its

contour length, where τ scales as L3/2 89-90. The relaxation time for a 48.5 kbp λ-phage DNA was

measured as 150 ms. Therefore the relaxation time of a 3.0 kbp plasmid DNA should be

approximately 2.3 ms, giving Wi = 55. The mean fractional extension of a DNA molecule under

these conditions is around 0.3 according to data obtained by Smith et al. 88 in free shear flow.

Note that the highest attainable fractional extension by shear flow was less than 0.4, which is

much less than that obtained in an elongational flow (mean fractional extension above 0.8 at De

≈ 1 which is where the coil-to-stretch transition occurs)91.

Upstream shear interaction can also increase the stretching efficiency by reducing the

conformational complexity of the pre-stretched DNA at the entrance to the narrow pore 70, 92-94.

Although some initial random coil conformations (e.g., the ‘dumbbell’ or coiled conformations)

can be elongated as fast as the fluid at high strain rates (affine stretching), molecules in folded

states (e.g., ‘hairpin’ conformations) strongly resist stretching in extensional flows since they

have two ends tightly pulled behind the middle section. Larson 92 and Smith and Chu 91 showed

73

that pre-shearing DNA can cause rotation and adjustment of molecular configurations reducing

the tendency of molecules to form folds in subsequent elongational flow.

The shape and size of the funnel structure can also have a large effect on the pre-

conditioning. The stretching efficiency is found to depend strongly on the accumulated fluid

strain 93, 95. In a funnel shaped pore, higher aspect ratio funnels are predicted to achieve higher

stretching efficiencies based on a purely elongational flow model. On the other hand, pre-

shearing would be more significant in a funnel with a smaller opening. Thus, there should exist

an optimal size to maximize the pre-conditioning efficiency, consistent with our experimental

results.

Larson et al.70 studied DNA stretching under different strain rate profiles and proved

experimentally that the shape of the funnel is critical for efficient DNA stretching. The gradual

transition funnel shape shown in Figure 4.10B produces increasing strain rate towards the

entrance to the narrow pore, yielding a highly extended conformation. In comparison, the sudden

onset shape funnel shown in Figure 4.10C caused many of the DNA molecules to be under-

stretched. In particular, it remains questionable whether bursts of elongational flow could stretch

DNA efficiently in the presence of extra shear interactions. Studies have shown that shear-

induced polymer tumbling in the tapered funnel region would diminish stretching efficiency96-97.

Note that the sizes of the 9.8 and 16.8 kbp supercoiled plasmid were measured to be 120 and 170

nm from static light scattering75, close to the mean pore size (0.1 and 0.2 µm) of the upper

microfiltration membranes. Hence these DNA molecules would effectively span the narrow pore

space, experiencing rotational flow in opposing directions simultaneously. The conflicting

influences of opposed rotational flows likely frustrated effective pre-conditioning in the upper

layer of these composite membranes.

74

4.4 Conclusions and recommendations

The data presented in this Chapter demonstrate that pre-conditioning using composite

membranes can enhance the transmission of supercoiled DNA through narrow pore UF

membranes. Data were obtained using a series of composite membranes generated by placing a

larger pore size microfiltration membrane on top of the skin layer of an ultrafiltration membrane.

The increase in plasmid transmission was found to be greatest when using an intermediate pore

size for the upper layer of the composite membranes. The pore size that provided the greatest

transmission through the Ultracel 100 kDa membrane was between 0.1 and 0.2 µm. Membranes

with larger pores provided insufficient pre-elongation, while membranes with smaller pores

required that the plasmids elongate significantly even to enter the upper layer of the membrane.

The greatest increase in transmission was seen when the asymmetric ultrafiltration membrane

was used in the reverse orientation, likely due to the gradual reduction in the pore size as one

moves from the membrane substructure to the tight skin.

Limited experiments performed with the open-circular plasmid suggest that pre-

conditioning using these composite membranes can also significantly increase the selectivity for

the separation between open-circular and supercoiled isoforms. In this case, pre-conditioning

caused a large increase in transmission of the supercoiled isoform while having little effect on

the transmission of the open-circular isoform. The net result was an increase in selectivity from

around 5-fold for the Ultracel 100 kDa membrane alone to more than 25-fold for a composite

membrane formed with a 0.22 µm Durapore membrane as the upper layer.

75

The pre-conditioning in composite membranes was found to be closely related to the pre-

shearing and pre-stretching of DNA molecules before approaching the tapered region. Conical

shape membranes made with anisotropic chemical etching of polycarbonate films containing

straight through cylindrical pores provide an idealized physical model to study the effects of pore

morphology on DNA transport through nanopores. Results presented in this study have

important implications on fabricating composite membranes with desirable pore size and

structures to provide optimal performance for specific needs.

Future studies can be done to further explore the transmission of plasmid DNA through

different shapes of nanopores. Synthesized polycarbonate membranes having conical shaped

pores with different cone angles and pore diameters can be made by controlling the chemical

etching conditions, including the NaOH concentration, etching time, solvent composition, and

transmembrane voltage82, 98. Etching without any applied voltage yields almost cylindrical pores

with a relatively short conical segment at one end 81 (sudden onset model). Ultrafiltration data

obtained with different isoforms and sizes of plasmid DNA can be compared with predictions of

the different physical models developed for flow-induced DNA elongation. Ultrafiltration

experiments could also be performed with more concentrated DNA solutions for longer filtration

times using these composite and conical-shaped membranes to examine the effects of pre-

conditioning on membrane fouling. Knowledge obtained from these studies would be of direct

interest in the design and manufacture of composite membranes with desirable pore size and

shape for separation and purification of plasmid DNA.

76

Chapter 5

Enhanced Plasmid DNA Purification by Exploiting Ionic Strength Effects

The solution structure of plasmid DNA is known to be a strong function of solution

conditions due to intramolecular electrostatic interactions between the charged phosphate groups

along the DNA backbone. The objective of the work described in this Chapter was to determine

whether it was possible to enhance the separation of different plasmid isoforms by proper

selection of the solution ionic strength and ion type during ultrafiltration. Experiments were

performed with three different plasmid isoforms in solutions containing different NaCl or MgCl2

concentrations. The majority of the work presented in this Chapter was previously published in

Biotechnology and Bioengineering (Li et al., 113, 783-789, 2015).

5.1 Introduction

It is well established that the conformation and flexibility of DNA are both functions of

the solution environment due to the strong intramolecular electrostatic interactions between the

charged phosphate groups along the DNA backbone. For example, Borochov et al. 99 reported a

25% reduction in the radius of gyration (Rg) of a linear 6.6 kilo base pair (kbp) DNA molecule

(from 245 to 195 nm) as the sodium chloride concentration was increased from 5 to 100 mM.

This change in NaCl concentration also caused a decrease in the DNA persistence length from 91

to 55 nm. Cherny and Jovin100 observed a gradual transition in the conformation of a 2.96 kbp

supercoiled plasmid from a loose to tightly inter-wound structure as the concentration of NaCl

was increased. Rybenkov et al.101 evaluated the effects of ionic conditions on the effective

77

diameter of the supercoiled helix, with results showing a 3-fold reduction (from approximately

15 to 5 nm) as the Na+ concentration increased from 1 to 100 mM. Kong et al.102 showed that the

transmission of various plasmids through 0.2 µm microfiltration membranes was significantly

improved with the addition of 150 mM NaCl to the formulation buffer, which the authors

attributed to the reduction in repulsive interactions between the charged membrane and the

charged DNA.

There is, however, relatively little data on the effects of solution conditions on the

transmission of plasmid DNA through the small pores in ultrafiltration membranes. DNA

transmission during ultrafiltration occurs by elongation of the DNA in the converging flow field

approaching the membrane pores37. The transmission of the supercoiled plasmid increased with

increasing solution ionic strength (at a given filtrate flux), which was attributed to the reduction

in effective plasmid size due to shielding of the intramolecular electrostatic interactions44.

However, the potential impact of solution ionic strength on the ultrafiltration of the linear and

open-circular plasmids has yet to be determined.

The objective of the work described in this Chapter was to obtain quantitative data for the

effects of salt concentration and ion valence on the transmission of the linear, open-circular, and

super-coiled isoforms of a 3.0 kbp plasmid through small pore size ultrafiltration membranes.

These results were used to identify ionic conditions that significantly enhance the selectivity of

ultrafiltration processes for purification of the therapeutically active supercoiled isoform that is

of interest in both gene therapy applications and for production of DNA-based vaccines.

78

5.2 Materials and Methods

Filtration experiments were conducted using Ultracel™ membranes and the 3.0 kbp

plasmid in the supercoiled, linear, and open-circular conformations. Plasmid solutions with a

final concentration of 0.25 µg/ml were prepared by diluting the stock solution with TE buffer

(solution pH = 7.7 ± 0.1).

The ionic strength of the buffer solution was adjusted by adding appropriate amounts of

sodium chloride (NaCl) or magnesium chloride hexahydrate (MgCl2・6H2O). Plasmid

concentrations were determined using the PicoGreen assay as described in Chapter 2.3. The

PicoGreen® fluorescence intensity signal is mildly sensitive to the salt concentration in the

buffer solution 49. Standard calibration curves were thus prepared over a range of salt

concentrations, with the concentrations evaluated by appropriate interpolation. The results were

used to evaluate the effects of solution conditions on the observed sieving coefficients.

Ultrafiltration using a mixture of two different plasmid isoforms were also performed, with

results analyzed using a combination of the Picogreen assay and AGE as described Chapter 2.3.

5.3 Results and Discussion

5.3.1 Linear plasmid

Figure 5.1 shows typical results for transmission of a 0.25 µg/L solution of the linear 3.0

kbp plasmid through the 100 kDa UltracelTM membrane as a function of filtrate flux. Data were

obtained from four separate experiments, each using TE buffer but with different amounts of

added NaCl. In each case, data were obtained with increasing values of the filtrate flux, with the

value re-checked at low flux to insure that there was no hysteresis. Selected filtrate and feed

79

samples were analyzed by agarose gel electrophoresis to confirm the integrity of the plasmid;

there was no evidence of plasmid degradation or aggregation in any of the salt solutions at either

low or high filtrate flux. There was also no measurable fouling of the membrane. The filtrate flux

remained stable at each applied pressure and the hydraulic permeability of the membrane after

the ultrafiltration experiment (evaluated from data for the filtrate flux as a function of the

transmembrane pressure) was within 10% of that for the clean membrane. The absence of any

fouling is consistent with previous results obtained with similarly dilute plasmid solutions 103.

Figure 5.1: Effect of NaCl concentration on the observed sieving coefficients for ultrafiltration of

the linear 3.0 kbp plasmid through the UltracelTM 100 kDa membrane

At any given value of the filtrate flux, the plasmid sieving coefficient increased with

increasing solution ionic strength. The greatest effect of ionic strength was seen at intermediate

values of the filtrate flux. For example, the sieving coefficient at a filtrate flux of 43 μm/s varied

80

from So = 0.29 in the solution with 1 mM NaCl to So = 0.63 in the 300 mM NaCl solution.

Plasmid transmission in the solutions containing 10 and 150 mM NaCl solution were very

similar for all values of the filtrate flux; the sieving coefficient at a flux of 12 µm/s was So = 0.05

at both ionic strengths while So = 0.69 and 0.73 at a flux of 80 µm/s in the 10 and 150 mM NaCl

solutions, respectively.

Figure 5.2 shows results from a similar set of experiments but with solutions containing

either 10 or 40 mM MgCl2 added to the TE buffer. The data at the two MgCl2 concentrations

were essentially identical over the entire range of filtrate flux with So values differing by less

than 0.05. Results are also shown for the 150 mM NaCl solution (taken from Figure 1) for

comparison. The transmission of the linear plasmid was very similar for all three ionic

conditions. Note that the ionic strength (I) of the 10 and 40 mM MgCl2 solution, calculated as:

I = 1

2 (∑ 𝑐𝑖𝑧𝑖

2𝑖 ) (5.1)

where ci and zi are the concentration and valence of each ionic species, are approximately 43 and

130 mM (using the pKa = 8.1 for Tris at 25 °C), whereas the ionic strength of the solution

containing 150 mM NaCl is 160 mM (accounting for the contribution from the TE buffer). The

data in Figure 5.2 are thus consistent with the results from Figure 5.1, with transmission of the

3.0 kbp linear plasmid nearly independent of solution ionic strength between I = 20 and 160 mM.

81

Figure 5.2: Effect of ion valence and concentration on the observed sieving coefficient for

ultrafiltration of the linear 3.0 kbp plasmid through the UltracelTM 100 kDa

membrane

5.3.2 Open-circular plasmid

Figure 5.3 shows data for the transmission of the open-circular isoform of the 3.0 kbp

plasmid through the Ultracel 100 kDa membrane in TE buffer containing either 10, 150, or 300

mM NaCl. The open-circular plasmid was highly retained over the entire range of filtrate flux,

with So < 0.2 even at the highest salt concentration and flux (140 µm/s) and So < 0.01 at a filtrate

flux of 40 µm/s in the solutions containing 10 and 150 mM NaCl. The high retention of the open-

circular isoform is consistent with previous results 45 and reflects the difficulty in fully

elongating the open-circular plasmid.

82

Figure 5.3: Transmission of the open circular isoform of the 3.0 kbp plasmid through the

UltracelTM 100 kDa membrane in the presence of 10, 150, or 300 mM NaCl

The results from Figure 5.3 have been re-plotted in Figure 5.4 as an explicit function of

the solution ionic strength along with data obtained in two separate experiments performed in the

presence of 10 and 40 mM MgCl2. The data in the NaCl and MgCl2 were reasonably consistent

when plotted as a function of the ionic strength, with the transmission of the open-circular

isoform increasing slightly with increasing ionic strength. However, the open-circular isoform

was highly retained over the full range of experimental conditions.

83

Figure 5.4: Effect of solution ionic strength on transmission of the open circular 3.0 kbp plasmid

through the UltracelTM 100 kDa membrane at a filtrate flux of 140 ± 5 μm/s.

5.3.3 Isoform separation

Figure 5.5 summarizes the effects of solution ionic strength on the transmission of the

different plasmid isoforms at a filtrate flux of 140 μm/s (500 L/m2/h); qualitatively similar

behavior was seen at other values of the flux but with lower transmission at lower flux. The data

in Figure 5.5 were all obtained with NaCl solutions, although similar results were seen with

MgCl2. The observed sieving coefficients of the linear 3.0 kbp plasmid were almost independent

of ionic strength, with values ranging from So = 0.82 to 0.93 for ionic strength between 10 and

300 mM. In contrast, transmission of the supercoiled plasmid increased significantly with

increasing solution ionic strength, similar to data reported previously by Latulippe and Zydney

(2008)44, going from So = 0.09 in the 10 mM ionic strength solution to more than 0.71 in the 300

mM solution. The open-circular plasmid also showed a slight increase in transmission with

84

increasing ionic strength (data from Figure 5.4), but this isoform was strongly retained under all

conditions.

Figure 5.5: Effect of solution ionic strength on transmission of the linear, open-circular, and

supercoiled 3.0 kbp plasmid through the UltracelTM 100 kDa membrane at a filtrate

flux of 140 µm/s.

Figure 5.6 shows results for the selectivity between the linear and supercoiled isoforms

(top panel) and between the supercoiled and open-circular isoforms (bottom panel) using the

UltracelTM 100 kDa membrane at both low and high NaCl concentrations. The error bars were

determined from standard propagation of error analysis based on Equation (2.2).

(2.2) y =S1

S2

85

The selectivity between the linear and supercoiled isoforms decreased with increasing filtrate

flux due to the large increase in transmission of the supercoiled isoform. The selectivity was

greatest at low ionic strength due to the large increase in retention of the supercoiled plasmid; the

transmission of the linear isoform was nearly independent of the salt concentration. The

selectivity in the 10 mM NaCl solution was nearly 30 at a filtrate flux around 40 μm/s (140

L/m2/h), with even higher selectivity possible at lower salt concentrations and / or lower filtrate

flux. The maximum selectivity in the 10 mM NaCl solution was more than 3 times higher than

the optimal selectivity achieved in 150 mM NaCl solution. In contrast, the selectivity between

the supercoiled and open-circular plasmids (lower panel) was greatest at high ionic strength due

to the increased transmission of the supercoiled plasmid under these conditions. The selectivity

attains its maximum value at an intermediate filtrate flux (around 80 µm/s) due to the increase in

transmission of both isoforms at very high filtrate flux.

86

Figure 5.6: Selectivity between the linear and supercoiled isoforms (top panel) and between the

supercoiled and open-circular isoforms (bottom panel) of the 3.0 kbp plasmid using

the Ultracel™100 kDa membrane

87

Based on the results in Figure 5.6, an ultrafiltration experiment was performed using a

binary mixture of the supercoiled and linear isoforms of the 3.0 kbp plasmid, each at a

concentration of 0.5 μg/mL, in a TE buffer containing 10 mM NaCl. Data were obtained using an

UltracelTM 100 kDa membrane at a filtrate flux of 80 µm/s (280 L/m2/h). Feed and filtrate

samples were analyzed using agarose gel electrophoresis, with the resulting gel shown in Figure

5.7. The first lane shows a 1 kbp DNA ladder for calibration. The feed sample (Lane 3) has equal

concentrations (corresponding to essentially equal brightness) of the linear and supercoiled

plasmids, with the lower band corresponding to the supercoiled isoform which migrates further

through the gel 104. The filtrate samples (Lanes 2 and 4) clearly show a band corresponding to the

linear isoform, with no measureable amount of the supercoiled plasmid. This behavior is

completely consistent with the high selectivity between the linear and supercoiled isoforms seen

in Figure 5.6 under these conditions. Note that a somewhat higher selectivity could have been

obtained at an even lower filtrate flux (40 µm/s), but the resulting filtrate solution was very dilute

making it harder to see the bands for the linear isoform on the gel.

88

Figure 5.7: Agarose gel electrophoresis showing the separation of a binary mixture of the linear

and supercoiled isoforms in TE buffer containing 10 mM NaCl. Lane 1: linear 1 kbp

DNA ladder. Lane 3: feed sample. Lanes 2 and 4: filtrate samples collected using the

Ultracel 100 kDa membrane at a filtrate flux of 80 μm/s.

Similar experiments were performed with a binary mixture of the supercoiled and open-

circular plasmids in a TE buffer containing 150 mM NaCl at a filtrate flux of 125 µm/s. In this

case, the filtrate samples (Lanes 2 and 3) show a significant band corresponding to the

supercoiled isoform, with no measurable band corresponding to the open-circular DNA (which

lies above that for both the supercoiled and linear plasmids). The high resolution of the

separation was a direct result of the high transmission of the supercoiled isoform in the 150 mM

NaCl solution at high filtrate flux, with the open-circular isoform remaining highly retained

under these conditions.

89

Figure 5.8: Agarose gel electrophoresis showing the separation of a binary mixture of the open-

circular and supercoiled isoforms in TE buffer containing 150 mM NaCl. Lane 4:

linear 1 kbp DNA ladder. Lane 1: feed sample. Lanes 2: filtrate samples collected at

filtrate flux of 100 μm/s. Lanes 3: filtrate samples collected at filtrate flux of 125

μm/s.

5.3.4 Physical Interpretation

Although all three plasmid isoforms showed at least some increase in transmission with

increasing solution ionic strength due to electrostatic shielding effects, the ultrafiltration behavior

of the individual isoforms was very different: the sieving coefficient of the linear plasmid was a

very weak function of the salt concentration while the sieving coefficient of the supercoiled

isoform increased significantly with increasing ionic strength. In contrast, the effect of solution

ionic strength on the radius of gyration (Rg) is very similar for the different plasmid isoforms.

For example, Hammermann et al. 105 observed a 20% reduction in the radius of gyration of a

supercoiled 1.87 kbp plasmid (from Rg = 58 to 47 nm) when the NaCl concentration was

increased from 10 to 100 mM, while Borochov et al. 99 reported approximately the same percent

reduction in Rg of a linear 6.6 kbp plasmid (from Rg = 245 to 186 nm) as the NaCl concentration

90

increased from 5 to 200 mM. In addition, the persistence length of DNA, which provides a

measure of the DNA flexibility, is relatively independent of salt concentration at solution ionic

strength above about 10 mM106. Thus, the large difference in ultrafiltration behavior of the

supercoiled and linear isoforms is probably not due to differences in either the effective size or

persistence length.

Instead, the strong dependence of the transmission of the supercoiled plasmid on the

solution ionic strength is most likely due to changes in the supercoiled structure at different salt

concentrations. Atomic force microscopy (AFM) and scanning electron microscopy (SEM) have

been used to evaluate the plectonemic structure of supercoiled plasmids in different ionic

environments. Images show a significantly tighter supercoiling at high NaCl concentrations107.

Cherny and Jovin100 showed that the number of nodes (distinct crossings of double helical

segments) of a 2.96 kbp plasmid decreased from nearly 15 to approximately 2 as the NaCl

concentration was reduced from 50 to 1 mM. The supercoiled isoform displayed an open ring

structure at very low salt concentrations similar to that of the open-circular plasmid 100. These

changes in structure also led to a reduction in the superhelix diameter with increasing salt

concentration 108-111, where the superhelix diameter is defined as the average distance between

opposing DNA strands in the interwound regions. For example, Rybenkov et al. 108 reported

values of the superhelix diameter ranging from 15 nm in a 10 mM NaCl solution to 3 nm in 1 M

NaCl, with the latter approaching the geometrical diameter of the double helix. The net result is

that the ultrafiltration behavior of the supercoiled isoform is very similar to that of the open-

circular plasmid at low ionic strength but becomes more similar to that of the linear plasmid at

high ionic strength, consistent with the experimental data for the sieving coefficient in Figure

5.5.

91

5.4 Conclusions

Although the effects of ionic environment on DNA conformation are well known, the

experimental data presented in this Chapter provide the first demonstration that it is possible to

enhance the performance of membrane systems for plasmid DNA separations by proper selection

of the ionic conditions. The transmission of the linear isoform was nearly independent of

solution ionic strength whereas the transmission of the supercoiled plasmid increased

significantly with increasing salt concentration. The large effect of solution conditions on the

transmission of the supercoiled plasmid was likely due to a change in conformation from a loose

to tightly inter-wound structure with increasing ionic strength associated with the shielding of the

intramolecular electrostatic interactions.

The changes in isoform transmission caused a significant enhancement of the selectivity

between the linear and supercoiled plasmid isoforms at low ionic strength due to the increased

retention of the supercoiled isoform under these conditions. In contrast, the selectivity between

the supercoiled and open-circular plasmid isoforms was greatest at high ionic strength since the

open-circular isoform was highly retained under all conditions. These results show that it should

be possible to perform a staged diafiltration process for purification of the desired supercoiled

isoform: the linear plasmid would initially be removed in the permeate by a diafiltration process

performed at low ionic strength and low filtrate flux with the supercoiled plasmid then recovered

in the permeate by a second diafiltration at high ionic strength and relatively high filtrate flux.

The ability to control the plasmid transmission by proper selection of the filtrate flux and ionic

conditions could thus provide a new strategy for the high resolution separation of plasmid

isoforms using small pore size ultrafiltration membranes.

92

Chapter 6

Effects of Multivalent Salts on Plasmid DNA Ultrafiltration

It has been well established that the structure of plasmid DNA is a strong function of

solution ionic conditions due to changes in intramolecular electrostatic interactions between the

charged phosphate groups along the DNA backbone. Multivalent cations like spermine play a

critical role in compacting and controlling the structure of supercoiled DNA in living cells. The

objective of the work described in this Chapter was to investigate the effects of the polyamine

spermine on the ultrafiltration of plasmid DNA and to explore the opportunity of using these

polycations to enhance the purification of specific plasmid isoforms. Much of the experimental

data presented in this Chapter were obtained by Rachel Bolten as part of her senior Honors thesis

in Chemical Engineering.

6.1 Introduction

Polyamines are organic compounds having two or more amino groups (NH2). Spermidine

(which contains 3 amine groups) and spermine (which contains 4 amine groups) are two

naturally occurring polyamines (Figure 6.1), both of which are synthesized in living cells via

highly regulated pathways and are critical in regulating a variety of cellular activities112. In

particular, the presence of physiological level polyamines (submillimolar or millimolar) has been

shown to inhibit topoisomerase catalyzed relaxation of negative supercoils in DNA113. Several

studies have suggested that DNA supercoiling, and in turn its biological activity, are controlled

by systematic regulation of the concentrations of these polyamines114.

93

Figure 6.1: Chemical structure of spermidine and spermine. The basic amino (NH2) groups bind

protons at physiological pH to become positively charged.

Polyamines can directly interact with DNA molecules in two ways. First, the positively

charged amine groups can interact electrostatically with the negatively charged phosphates along

the DNA backbone, similar to the interactions that occur with mono- and di-valent cations115.

Cation binding is generally found to stabilize the right handed, B-form double helix structure of

DNA116 and protect the DNA from damage due to radiation or oxidation117. Second, polyamines

can bind in the major and minor grooves of DNA, forming hydrogen bonds with bases from the

opposing strands of the DNA double helix as revealed by recent crystal structures and infrared

spectra of polyamine-DNA complexes118-120. This reduces denaturation of DNA and favors

formation of writhes instead of unwinding of DNA. Shao et al.121 used single molecule

experiments to demonstrate that the presence of spermine or spermidine produced more compact

plectonemes in DNA in the presence of high concentrations of monovalent salts. Theoretical

models confirmed that binding of polyamines reduced the radius and increased the density of

DNA supercoils.

Sato et al. 114 studied the conformational transition of a 12.5 kbp supercoiled plasmid

DNA induced by spermine using fluorescence microscopy. The hydrodynamic radius of the

plasmid decreased from more than 250 nm to less than 150 nm as the spermine concentration

94

was increased from 0 to 10 µM; these concentrations are far too low to cause any significant

increase in electrostatic shielding due to the increase in solution ionic strength. Murphy et al.122

showed that the equilibrium adsorption capacity of a Q-Sepharose anion-exchange resin for a 5.9

kbp supercoiled plasmid DNA at 600 mM NaCl was enhanced by up to 40% in the presence of

2.5 mM spermine. This was primarily due to compaction of the plasmid by spermine, which

allowed the large plasmid to access the surface of small pores within the adsorbents.

Multivalent salts can also induce DNA condensation by neutralizing the high negative

charge of the DNA, thereby reducing inter-helix electrostatic repulsion. Wilson and

Bloomfield123 used the counter-ion condensation theory developed by Manning124 to calculate

that DNA condensation occurs when approximately 90% of the DNA charge is neutralized.

Inter-molecular bridging between DNA helices also helps to promote condensation at low ionic

strength125. DNA condensation primarily occurs via intramolecular interactions at low DNA

concentrations (below 5 µg/mL), with the resulting condensates having a compact spherical or

toroidal shape as observed by both electron and atomic force microscopy126-127. The

hydrodynamic radius of DNA condensates resulting from collapse of λ-phage DNA by spermine

is 41±5 nm as determined by dynamic light scattering128.

Although polyamines are known to have a significant effect on DNA structure, there is

currently no information on the possible role of these multivalent cations on plasmid

ultrafiltration. The objective of the work described in this Chapter is to evaluate the effects of

spermine on plasmid DNA transmission through UF membranes and to explore the opportunity

for enhancing the separation between different plasmid isoforms by proper addition of spermine.

Data were obtained using Biomax 300 kDa membranes with different concentrations of

spermine, both with and without other monovalent salts. The results clearly demonstrate that

95

spermine has a dramatic effect on plasmid ultrafiltration, even at micromolar concentrations,

leading to a large decrease in DNA transmission.

6.2 Materials and methods

Ultrafiltration experiments were performed with 9.8 kbp plasmids in the supercoiled,

open circular, and linear isoforms. Limited experiments were also performed with the 16.8 kbp

supercoiled plasmid. Plasmid DNA solutions were prepared in TE buffer with NaCl

concentrations ranging from 1 to 100 mM. Spermine (MW= 202.34, ≥97%) was obtained from

SigmaAldrich (MO) and stored at -20°C prior to use. TE buffer containing 10 mM spermine and

appropriate NaCl concentrations was prepared and then slowly added to the DNA solution. The

DNA and spermine were allowed to interact for 30 min before use in the ultrafiltration

experiment.

Sieving experiments were performed with Biomax 300 kDa ultrafiltration membranes.

Data were obtained using the standard constant-pressure ultrafiltration procedures described in

Section 2.6.1. In addition, a limited number of constant flux experiments were performed by

connecting a peristaltic pump to the permeate exit line. The entire system was flushed with 10

mL buffer between sieving experiments performed with different salt or spermine

concentrations.

6.3 Results and Discussions

6.3.1 Supercoiled isoform

96

Figure 6.2 shows data for transmission of the supercoiled 9.8 kbp plasmid through the

Biomax 300 kDa membrane in TE buffer containing 10 mM NaCl and various concentrations of

spermine. DNA transmission was largely unaffected by the spermine concentration for spermine

concentrations ranging from 0 to 10 µM. For example, the DNA sieving coefficients at a filtrate

flux of 75 µm/s were So ≈ 0.84 in both the 0 and 2 µM spermine solutions, while the sieving

coefficient for DNA in the 10 µM spermine was 0.74. Results obtained with a different lot of

Biomax 300 kDa membranes yielded very different results, with the addition of 2 or 10 µM

spermine causing a significant increase in DNA transmission (from So = 0.12 in the absence of

spermine to So = 0.93 in the 10 µM solution). At least 3 membranes were tested from each lot,

with the results for the membranes within a given lot being highly consistent. It is unclear what

caused this discrepancy; further study is required to determine the cause of the very different

behavior for these lots of the Biomax membranes.

In contrast to the data at low spermine concentrations, DNA transmission at spermine

concentrations of 15 and 30 µM was negligible over the entire range of filtrate flux, suggesting

the presence of a threshold spermine concentration above which the plasmid is unable to

elongate and pass through the pores of the membrane. Vijayanathan et al. 128 also observed a

critical spermine concentration of 10 µM for DNA condensation in a 10 mM sodium cacodylate

buffer. DNA condensates have been shown to have a toroidal shape with hydrodynamic radius

around 41 nm for 48.5 kbp plasmids 128-129. These large condensates would be completely

retained by the small pores in the Biomax 300 kDa membrane (11.2 nm from Table 2.4), since

the converging flow into the membrane pores is likely to be unable to elongate the condensed

DNA.

97

Additional experiments were performed with the 16.8 kbp supercoiled plasmid under the

same solution conditions examined in Figure 6.2 (data not shown). The results were similar to

those obtained with the 9.8 kbp plasmid, although the sieving coefficients in solutions containing

0 to 10 µM spermine were somewhat lower than that observed in Figure 6.2. Data obtained with

spermine concentrations above 15 µM showed negligible transmission of the 16.8 kbp plasmid,

similar to what was seen with the 9.8 kbp plasmid.

Figure 6.2: Observed sieving coefficients for the supercoiled 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of filtrate flux for experiments performed with 0, 2,

10, 15, and 30 µM spermine in TE buffer containing 10 mM NaCl.

The effect of spermine on DNA transmission is shown more explicitly in Figure 6.3, with

the sieving coefficients at the different spermine concentrations evaluated at a filtrate flux of 50

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux, Jv (µm/s)

0µM

2µM

10µM

15µM

30µM

98

µm/s. The DNA transmission initially increased slightly upon addition of a low concentration of

spermine, but then rapidly decreased to essentially zero for spermine concentrations at 15 µM and

above.

Figure 6.3: Effect of spermine concentration on the observed sieving coefficients for the

supercoiled 9.8 kbp plasmid through 300 kDa Biomax membranes at a filtrate flux

of 50 µm/s. Data obtained in TE buffer containing 10 mM NaCl.

6.3.2 Effects of monovalent salts

Since spermine directly alters the inter- and intra-molecular electrostatic interactions

between the negatively-charged phosphate groups in DNA, a series of sieving experiments were

performed using the supercoiled 9.8 kbp plasmid in TE buffer with different spermine and NaCl

concentrations. Results are shown in Figure 6.4 for data obtained in the presence of 100 and 1

mM NaCl. Increasing monovalent salt concentration significantly increased plasmid

0

0.2

0.4

0.6

0.8

1

0 10 20 30

Sie

vin

g C

oe

ffic

ien

t, S

0

Spermine Concentration (µM)

99

transmission when at low spermine concentrations (upper panel of Figure 6.4), which is

consistent with results obtained by Latulippe and Zydney 44 and also in Chapter 5. For example,

at a filtrate flux of 50 µm/s, the sieving coefficient in the absence of any spermine ranged from

more than So = 0.8 in the TE buffer with 100 mM NaCl to So < 0.2 in the solution with only 1

mM NaCl. Adding low concentrations of spermine had little effect on DNA transmission, with

the sieving coefficients remaining nearly unchanged from the values obtained in the absence of

any spermine over the entire range of filtrate flux. Note that the ionic strengths (I) of the 1, 10

and 100 µM spermine solution, calculated as:

I = 1

2 (∑ 𝑐𝑖𝑧𝑖

2𝑖 ) (5.1)

where ci and zi are the concentration and valence of each ionic species, are approximately 8, 80

and 800 µM, compared to the 11.6 mM contributed by the TE buffer itself.

The effect of spermine on DNA transmission at high spermine concentrations depends

very strongly on the NaCl concentration. For example, the critical spermine concentration,

defined as the spermine concentration at which there is a large reduction in DNA transmission,

was between 2 and 6 µM in the TE buffer with 1 mM NaCl but this increased to 15 – 30 µM for

the buffer with 10 mM NaCl and to between 100 – 200 µM for 100 mM NaCl. This behavior is

consistent with results obtained by Hoopes and McClure130 for DNA precipitation, in which the

addition of 0.1 M salt increased the concentration of spermine needed for DNA precipitation by

more than 10-fold. The critical spermine concentration for precipitation was found to increase

approximately linearly with increasing NaCl concentration for both long (48.5 kbp) and short

(146 bp) DNA131.

100

Figure 6.4: Observed sieving coefficients for the supercoiled 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of the filtrate flux for experiments performed with

various concentrations of spermine added to TE buffer containing 100 mM (upper

panel) and 1 mM (lower panel) NaCl.

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80

Sie

vin

g c

oe

ffic

ien

t, S

0

Filtrate Flux Jv, µm/s

0µM

10µM

30µM

100µM

200µM

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux Jv, (µm/s)

0µM

2µM

6µM

10µM

15µM

101

Figure 6.5: Observed sieving coefficients for the supercoiled 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of the filtrate flux for experiments performed with

10 µM spermine added to TE buffer containing 1, 10 and 100 mM NaCl.

6.3.3 Isoform separation

A corresponding set of ultrafiltration experiments was performed with the linear and

open-circular versions of the 9.8 kbp plasmid using the Biomax 300 kDa membranes. Figure 6.6

shows results for the linear plasmid in TE buffer containing 10 mM NaCl at spermine

concentrations between 0 and 30 µM. The transmission of the linear isoform was greater than

that for the supercoiled isoform over the entire flux range due to the greater elongational

flexibility of the linear isoform45. Similar to the results with the supercoiled plasmid, the sieving

coefficient of the linear plasmid was largely independent of spermine for concentrations at or

below 10 µM. However, DNA transmission dropped to nearly zero when the spermine

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux Jv, µm/s

100mM NaCl

10mM NaCl

1mM NaCl

102

concentration was increased to above 15 µM, reflecting the condensation of the plasmids under

these conditions. The threshold spermine concentration for the linear isoform was similar to that

found for the supercoiled plasmid (data in Figure 6.3). Thus, spermine had relatively little affect

on the separation between the linear and supercoiled plasmids; the selectivity at low spermine

concentrations was essentially the same as that in the complete absence of any spermine while

high spermine concentrations essentially eliminated transmission of both isoforms through the

Biomax 300 kDa membrane.

Figure 6.7 shows results for the sieving coefficients of the open-circular 9.8 kbp plasmid

through the same Biomax 300 kDa membrane in TE buffer containing 10 (left) and 100 mM

(right) NaCl. There was negligible transmission of the open-circular plasmid through the Biomax

300 kDa membrane in the 10 mM NaCl solution at all spermine concentrations, reflecting the

high value of the filtrate flux required to sufficiently elongate these large circular molecules.

The addition of 100 mM NaCl significantly increased the transmission of the open-circular

plasmid, with the observed sieving coefficients essentially independent of the spermine

concentration up to 30 µM spermine. Figure 6.8 shows the selectivity between the supercoiled

and open-circular isoforms for experiments performed with various concentrations of spermine

at a filtrate flux of 50 µm/s. The selectivity of the supercoiled/open-circular separation was

maximum in the absence of or at very low concentrations of spermine. For example, operating in

a 10 mM NaCl solution with 0 or 2 µM spermine resulted in very low transmission of the open-

circular plasmid (So < 0.05) while there was significant passage of the supercoiled DNA (So >

0.7). In addition, operating at a lower NaCl concentration (10 mM) allowed effective separation

at higher filtrate flux, with the open-circular plasmid still strongly retained under these

103

conditions. Increasing NaCl concentration enhances the transmission of both isoforms, with good

separation only possible at very low filtrate flux (<20 µm/s).

Figure 6.6: Observed sieving coefficients for the linear 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of the filtrate flux for experiments performed with

0, 2, 6, 10, 15 µM spermine added to TE buffer containing 10 mM NaCl.

0

0.2

0.4

0.6

0.8

1

0 20 40 60 80 100

Sie

vin

g C

oe

ffic

ien

t, S

0

Filtrate Flux Jv, µm/s

0µM

2µM

10µM

15µM

30µM

104

Figure 6.7: Observed sieving coefficients for the open-circular 9.8 kbp plasmid through a 300 kDa

Biomax membrane as a function of the filtrate flux for experiments performed with

various concentrations of spermine added to TE buffer. Left - solutions contained 10

mM NaCl, Right - solutions contained 100 mM NaCl.

Figure 6.8: Selectivity for separation of the supercoiled and open-circular 9.8 kbp plasmids

through a 300 kDa Biomax membrane as a function of spermine concentration for

experiments performed at filtrate flux of 50 µm/s in TE buffer containing 10 and 100

mM NaCl.

6.4 Conclusions and recommendations

The results presented in this chapter provide the first data for the effects of spermine, a

small polyvalent amine, on the transmission of the supercoiled, linear, and open-circular plasmid

isoforms through ultrafiltration membranes. Adding low concentrations (typically below 10 µM)

of spermine had minimal effect on plasmid transmission through the Biomax 300 kDa

membranes. However, high spermine concentrations caused a dramatic reduction in DNA

0

4

8

12

16

20

0 10 20 30

Se

lec

tivit

y, ψ

Spermine Concentration (µM)

10 mM NaCl

100 mM NaCl

105

transmission, which is likely associated with the condensation of the DNA into a structure that

can no longer be elongated by the flow field into the membrane pores. DNA condensation

appeared to happen at a threshold spermine concentration, above which the plasmids became

almost completely retained by the membrane. The amount of spermine needed to induce DNA

condensation increased with increasing NaCl concentration for both the supercoiled and open-

circular plasmids (data were not obtained at different NaCl concentrations using the linear

plasmid).

Although spermine had a large effect on plasmid transmission, the results obtained in this

Chapter did not find any improvement in the selectivity for the membrane separation between the

different plasmid isoforms upon addition of spermine. At low spermine concentrations, the

plasmid sieving coefficient was essentially independent of the spermine concentration, while at

high spermine concentrations the transmission of all 3 isoforms through the Biomax 300 kDa

membrane was negligible. However, additional experiments might be able to identify conditions

where spermine could enhance the separation behavior. For example, Shao et al. 121 examined

the effect of spermine on DNA structure at high monovalent salt concentrations (>200 mM),

which were sufficient to prevent DNA condensation and which might lead to enhanced DNA

transmission. Experiments under these conditions would likely require the use of ultrafiltration

membranes with much smaller pores than the Biomax 300 kDa membrane to try to identify

conditions where spermine alters DNA transmission by tightening the supercoiled structure,

which could in turn lead to greater selectivity between the open-circular and supercoiled

isoforms.

Although the addition of spermine was not beneficial for DNA isoform separation, it is

very possible that DNA condensation using appropriate concentrations of spermine could be

106

attractive for separation of plasmid DNA from other impurities in the cell lysate. Hoopes and

McClure 130 showed that the intra- and intermolecular precipitation of DNA induced by spermine

was highly selective even in the presence of most proteins or triphosphates. The addition of

spermine could thus be used to significantly increase the retention of a desired DNA product

while allowing host cell proteins to be removed in the permeate. Similarly, unreacted DNA

could be more easily retained, and thus removed, from restriction endonuclease digests by

addition of appropriate quantities of spermine. Note that Hoopes and McClure 130 found that the

concentration of spermine required to precipitate a 100 bp DNA fragment in moderate salt

condition was twice that for a larger (200 bp) DNA, suggesting that the spermine concentration

could be adjusted to enhance the selectivity for the separation between these small nucleotide

fragments.

107

Chapter 7

Effect of Ionic Strength on Membrane Fouling During Ultrafiltration of Plasmid DNA

Results from Chapter 5 demonstrated that the selectivity of membrane systems for DNA

purification can be significantly enhanced by proper selection of the solution ionic strength, but

these data were all obtained in dilute solutions where fouling is negligible. The objective of the

work described in this Chapter was to evaluate the effects of solution conditions on the fouling

characteristics of both supercoiled and linear plasmid DNA isoforms with different length.

Experiments were performed with Biomax membranes with MWCO of 300 and 1000 kDa using

3.0, 9.8, and 16.8 kbp plasmids in solutions containing different NaCl concentrations. The

majority of the work presented in this Chapter was previously published in Separation and

Purification Technology (Li et al., 176, 287-293, 2016).

7.1 Introduction

Membrane processes are highly attractive for the downstream purification of plasmid

DNA, including the separation of the desired supercoiled plasmid from other topological

isoforms 132. However, membrane fouling remains a major factor limiting the performance and

application of these membrane processes. Several studies have demonstrated that even very pure

solutions of plasmid DNA can cause significant fouling, leading to a decline in filtrate flux and

loss of membrane selectivity. For example, Affandy et al.60 showed that the fouling behavior of a

20 kilo base-pair (kbp) and a 56 kbp plasmid DNA during sterile filtration through 0.22 μm

PVDF membranes was due primarily to pore blockage. Borujeni and Zydney 48 observed a rapid

108

decline in both the sieving coefficient and filtrate flux during filtration of a 16.9 kbp supercoiled

plasmid through a 100 kDa polyethersulfone ultrafiltration membrane, with the experimental

data well described by a partial pore blockage model in which the plasmids were trapped at the

pore entrance. The flux and sieving coefficient could be at least partially restored by periodic

backpulsing to remove the trapped plasmids 76.

It is well established that protein fouling is a strong function of solution conditions,

including both pH and ionic strength, due to changes in the intermolecular electrostatic

interactions between the charged protein and the membrane 133-135. DNA is a strongly charged

polyelectrolyte due to the phosphate groups along the DNA backbone. The data presented in

Chapter 5 clearly showed that transmission of the supercoiled plasmid through Ultracel

composite regenerated cellulose membranes increased with increasing ionic strength and that this

could be used to improve the separation of the different plasmid isoforms by proper choice of

solution ionic conditions. However, this work was performed with very dilute DNA solutions

where membrane fouling was negligible.

The objective of the work described in this Chapter was to evaluate the effects of solution

conditions on the rate and extent of membrane fouling during ultrafiltration of more concentrated

plasmid DNA solutions. Data were analyzed using the partial pore blockage model developed by

Borujeni et al.48 to obtain additional insights into the underlying fouling mechanisms and to

develop appropriate strategies for reducing membrane fouling during DNA ultrafiltration.

109

7.2 Materials and methods

Polyethersulfone (Biomax) ultrafiltration membranes (25 mm diameter disks) with

nominal molecular weight cut-off (MWCO) of 300 and 1000 kDa were obtained from EMD

Millipore (Bedford, MA) and used in all ultrafiltration experiments. TE buffer solutions were

prepared by diluting a concentrated stock solution with deionized water. NaCl was added to

achieve the desired ionic strength. All solutions were pre-filtered through 0.2 µm Supor 200

filters (Pall) prior to use.

Data were obtained with three different size supercoiled plasmids (3.0, 6.8 and 16.8 kbp)

obtained from Aldevron, with the linear versions obtained using appropriate restriction

endonucleases as described in Chapter 2. All plasmid samples were stored at -20°C and slowly

thawed at 4°C immediately prior to use.

The effective size of the different plasmids was determined by size exclusion

chromatography (SEC) using a PL Aquagel-OH 60 size exclusion column (designed for

separating polymers from 200 to 105 kDa in size). Data were obtained with an Agilent 1200

series HPLC system (Agilent Technologies) using a sample injection volume of 80 µL with

sample detection by an Agilent 1100 series refractive index detector. The temperature of the

column was kept at 35 °C. The mobile phase was a Tris-EDTA buffer (pH = 7.7 ± 0.3) with the

NaCl concentration chosen to match the ionic strength of the sample. The measured retention

volume of each sample was compared with that of dextran standards with molecular weight

between 2000 and 5000 kDa (American Polymer Standards Corp), which are comparable in size

to the plasmids used in this study. The effective radii of the dextran standards were calculated

from the empirical correlation:

110

𝑅𝑒𝑓𝑓 = 3.1 × 10−11 × (𝑀𝑤)0.47752 (7.1)

as described elsewhere136.

7.3 Results and discussions

7.3.1 Fouling experiments

Figure 7.1 shows typical data for the normalized filtrate flux (bottom panel) and observed

sieving coefficients (top panel) during ultrafiltration of 3 µg/mL (= 3 × 10-3 kg/m3) solutions of

the supercoiled 16.8 kbp plasmid through the Biomax 300 kDa membranes. Data were obtained

with the plasmids suspended in TE buffer containing 1, 10, or 500 mM NaCl. Each experiment

used a fresh membrane; the permeabilities of the 3 membranes were all within ±20% (6.2 ± 1.1 ×

10-12 m). The transmembrane pressure for each experiment was chosen to provide similar initial

values of the sieving coefficient (So = Cf/Cb where Cf is the filtrate concentration and Cb is the

bulk concentration), giving pressures of 3.2, 1.1, and 0.8 psi (corresponding to 22, 7.6, and 5.5

kPa) for the 1, 10, and 500 mM NaCl solutions, respectively. The higher pressure (and thus

higher filtrate flux) required for the solutions with the lower NaCl concentrations has been

discussed previously by Latulippe and Zydney (2008) 44 and in Chapter 5 and is related to the

change in plasmid conformational flexibility. The data are plotted against the total plasmid mass

throughput, evaluated from the measured filtrate volume and concentration as:

𝑚

𝐴= ∫ 𝐽𝐶𝑓𝑑𝑡

𝑡

0 (7.2)

At least two experiments were performed at each condition to confirm the reproducibility of the

data.

111

Figure 7.1: Effect of solution ionic conditions on the sieving coefficients (top panel) and filtrate

flux (bottom panel) during ultrafiltration of solutions of the 16.8 kbp supercoiled

plasmid through 300 kDa Biomax membranes. Data were obtained at plasmid

concentrations of 3x10-3 kg/m3 using TE buffer with 1, 10 or 500 mM NaCl. Dashed

curves are model calculations using parameter values given in Table 7.1.

The flux decline and loss of transmission was most pronounced for the plasmid in the 1

mM NaCl solution. In this case, the sieving coefficient decreased by nearly a factor of 50 (from

112

So = 0.98 to 0.02) while the normalized filtrate flux decreased by more than 60% (to a value of

J/Jo = 0.39) after filtration of less than 50 mg/m2 of the supercoiled plasmid. A similar degree of

fouling (decline in filtrate flux and sieving coefficient) occurred with the 10 mM NaCl solution

but after filtration of 100 mg/m2 of the supercoiled plasmid. In contrast, there was minimal

fouling during ultrafiltration of the supercoiled plasmid in the 500 mM NaCl concentration

solution, with the filtrate flux and sieving coefficient both decreasing by less than 20% after

ultrafiltration of more than 125 mg/m2. These differences in the extent of fouling were also seen

in the values of the membrane permeability determined after the plasmid ultrafiltration

experiment (obtained after gently rinsing the membrane with TE buffer). The ratio of the fouled

to clean membrane permeability was Lp/Lp0 = 0.53 ± 0.05 for the membrane used with the 1 mM

NaCl solution compared to values of Lp/Lp0 = 0.61 ± 0.05 and 0.80 ± 0.03 for the 10 and 500

mM NaCl solutions, respectively.

As discussed by Borujeni and Zydney 48, the decline in filtrate flux and sieving

coefficient seen in Figure 7.1 is due to blockage of the membrane pores caused by trapping of the

plasmids at the pore entrance. Pore blockage causes a much smaller reduction in the filtrate flux

than in plasmid transmission since the “blocked” pores allow significant fluid flow even though

they are impermeable to transmission of subsequent plasmids. This model is discussed in more

detail in the next section.

In addition to the “dilution effect” associated with flow through the “blocked” pores, the

decline in filtrate flux will itself cause a corresponding decline in plasmid transmission due to the

reduction in the extent of plasmid elongation at lower filtrate flux. In order to determine the

impact of the change in filtrate flux on the observed plasmid sieving coefficients, a series of

filtration experiments were performed using fresh Biomax 300 kDa membranes but with the

113

initial filtrate flux chosen to match the measured flux obtained at the end of each fouling run.

This corresponds to J = 40, 32, and 21 µm/s for the 1, 10, and 500 mM NaCl conditions,

respectively. The flux was set by adjusting the transmembrane pressure based on the previously

measured permeability of the specific membrane; the measured filtrate flux during the

experiment was in good agreement with the expected value. Data were obtained using very

dilute plasmid solutions to avoid any fouling during this set of experiments. The measured values

of the plasmid sieving coefficients were similar in the 3 salt solutions: So = 0.81, 0.89, and 0.90.

The slightly lower value of the sieving coefficient in the 1 mM NaCl solution is due to the larger

effective size of the plasmid at low ionic strength; this more than compensates for the higher

filtrate flux. Thus, the small values of the sieving coefficients seen in Figure 7.1 are a direct

result of the dilution of the filtrate caused by fluid flow through the “blocked” pores at the end of

the ultrafiltration experiment, with very little contribution from the change in filtrate flux.

The large degree of fouling seen in Figure 7.1 is also directly linked to the topology of

the supercoiled plasmids. For comparison, a corresponding fouling experiment was performed

with a 4.0 µg/mL solution of the linear p-FDY plasmid in a 1 mM NaCl solution, with results for

the filtrate flux and sieving coefficient shown in Figure 7.2. Both the sieving coefficient and

filtrate flux remained essentially constant over the entire course of the experiment, up to a

throughput of more than 220 mg/m2, indicating that there was negligible fouling/pore blockage

by the linear isoform even in the 1 mM salt solution. The linear plasmids are much easier to

elongate than the supercoiled DNA, and are thus much less likely to become trapped at the pore

entrance.

114

Figure 7.2: Ultrafiltration of a 3x10-3 kg/m3 solution of the 16.8 kbp linear plasmid through a 300

kDa Biomax membrane in 1 mM NaCl TE buffer.

7.3.2 Fouling model

The sieving coefficient and filtrate flux data in the different NaCl solutions were

analyzed using the fouling model developed previously by Borujeni and Zydney 48. As discussed

in the previous section, this model assumes that the membrane pores are occasionally blocked by

plasmids that become trapped at the pore entrance, with these blocked pores assumed to be

completely impermeable to the transport of additional plasmids while allowing some filtrate flow

due to the very open conformation of the trapped plasmids.

The filtrate flux and sieving coefficient are both expressed as functions of the fraction of

open pores (f = Aopen/εAm where ε is the membrane porosity) as:

115

𝐽

𝐽0= 𝛽 + 𝑓(1 − 𝛽) (7.3)

𝑆0

𝑆𝑜𝑝𝑒𝑛=

𝑓

𝛽+𝑓(1−𝛽) (7.4)

where β is the ratio of the flow rate through a partially blocked pore to that through an open pore.

The rate of pore blockage is assumed to be directly proportional to the mass of DNA that passes

through the open pores:

𝑑𝑓

𝑑𝑡= −𝑘(𝐽𝑜𝑝𝑒𝑛𝐶𝑓𝑆𝑜𝑝𝑒𝑛𝑓) (7.5)

where k is a constant describing the rate of fouling. The filtrate flux and sieving coefficient

through an open pore are assumed to be constant since the transmembrane pressure difference

across the pore remains constant. The concentration of the feed solution will gradually increase

with time due to plasmid retention; this was determined numerically using a simple differential

mass balance:

Ct+∆ t =Ct +q

VCFeed -C féë ùû∆ t

(7.6)

Mass balance closure was verified based on the measured plasmid concentration in the stirred

cell at the end of the filtration experiment.

The dashed curves in Figure 7.1 show the calculated values of the sieving coefficient and

filtrate flux based on Equations (7.3) to (7.6) using the best fit values of β and k determined by

minimizing the sum of the squared residuals between the experimental data and model

calculations for both So and J/Jo using a membrane porosity of ε = 0.5. The value of β is

unaffected by the uncertainty in ε as it is defined as the filtrate flow rate through the blocked

pore divided by that in the open pore, which does not depend on the porosity. The fouling rate

116

constant k is proportional to the change in ε according to Equation (7.5) as Jopen = J/ε. However

the absolute values of k should be comparable when experiments were performed with the same

type of membrane (Biomax 300 kDa) since the porosities of the different membranes are very

similar. The values of Sopen and Jopen were taken from the values of the sieving coefficient and

filtrate flux measured at the start of the given ultrafiltration experiment. The model calculations

are in very good agreement with the data, properly describing the more gradual decline in the

filtrate flux (compared to the sieving coefficient) due to the fluid flow through the blocked pores.

The best fit values of β are essentially independent of the NaCl concentration, varying

between β = 0.41 ± 0.01 and 0.48 ± 0.02, where the plus / minus limits on the fitted parameters

were determined based on a 5% increase in the sum of the squared residuals. This behavior is

consistent with the assumed physical picture of the pore being blocked by individual (trapped)

plasmid molecules, with the flow through these “blocked” pores occurring in the “gaps” within

the plasmid and between the plasmid and the pore walls. In contrast, the rate constant for fouling

was significantly affected by the solution ionic strength. There is a nearly 50% decline in the

fouling constant from k = 0.94 × 10-4 to 0.50 × 10-4 m2/kg when the salt concentration was

increased from 1 mM to 10 mM, with k further decreasing to 0.13 × 10-4 m2/kg in the 500 mM

solution.

117

Table 7.1: Best fit values of β and k for plasmid ultrafiltration experiments shown in Figs. 7.1–

7.5.

7.3.3 Effects of plasmid size

Figure 7.3 shows the effect of NaCl concentration on membrane fouling for plasmids

with different size. Experiments were performed using 3.0 µg/mL solutions of the 9.8 and 16.8

kbp supercoiled plasmids in TE buffer containing 1 and 10 mM NaCl. The transmembrane

pressures were maintained at 1.6 and 1.3 psi (11 and 9.0 kPa), giving initial filtrate flux of Jo =

80 and 55 µm/s for the 1 and 10 mM NaCl solutions, respectively. These flux values correspond

to initial sieving coefficients of So = 0.93 ± 0.03. The extent of fouling was much less

pronounced for the smaller 9.8 kbp plasmid, particularly in the 10 mM NaCl solution. For

example, the sieving coefficient decreased from So ≈ 0.9 to 0.05 after filtration of less than 100

mg/m2 for the 16.8 kbp (p-FDY) plasmid in the 10 mM solution, while the sieving coefficient

only decreased to So ≈ 0.4 after more than 145 mg/m2 for the 9.8 kbp (p-MDY) plasmid in the

same NaCl solution. Data obtained with the 3.0 kbp (p-EMP) plasmid in a 10 mM NaCl solution

118

yielded an essentially constant sieving coefficient and filtrate flux after filtration of more than

250 mg/m2, indicating that fouling was completely absent under these conditions.

Figure 7.3: Effect of solution ionic strength on the sieving coefficient (top) and filtrate flux

(bottom) of different size supercoiled plasmids through 300 kDa Biomax membranes.

Data were obtained with 3x10-3 kg/m3 solutions of the 16.8, 9.8 and 3.0 kbp plasmid

wi ith 3×10-3 kg/m3 solutions of the 16.8, 9.8 and 3.0 kbp plasmid with 1 and 10 mM

NaCl in TE buffer. Dashed curves are model calculations using parameter values

specified in Table 7.1.

119

The dashed curves in Figure 7.3 are again the model calculations using the best fit values

of the fouling parameters, k and β. The β value decreases slightly with increasing plasmid size,

varying from β = 0.70 ± 0.07 for p-EMP to β = 0.44 ± 0.03 for p-FDY. This behavior is

consistent with the physical picture of the larger plasmids providing more extensive blocking of

the pore area.

The best fit values of the fouling rate constant k for the 3 plasmids at the different NaCl

concentrations are summarized in Table 7.1. The fouling rate constant increases with increasing

plasmid size at a fixed set of solution conditions. For example, the best fit value of the rate

constant increases from k = 0.12 × 10-4 to 0.94 × 10-4 m2/kg in the 1 mM NaCl solution as the

plasmid size increases from 3.0 to 16.8 kbp. A similar increase is seen in the 10 mM NaCl

solution, consistent with the greater probability of the larger plasmids becoming trapped in the

membrane pores. The greater rate of pore blockage for the larger plasmids is mostly likely

related to their more branched structure as discussed by Li et al. 137. The supercoiled plasmids

adopt a branched structure due to the under-twisting of the DNA, with the number of branches

increasing with increasing chain length. Therefore the chance of blocking a membrane pore is

much higher for the 16.8 kbp plasmid compared with the 3.0 kbp plasmid. This is further

illustrated in Chapter 8. Interestingly, the fouling rate constant for the 16.8 kbp plasmid in a 1

mM NaCl solution was very similar to that of the 9.8 kbp plasmid in a 10 mM NaCl solution,

indicating that an increase in the NaCl concentration is able to compensate for the greater fouling

tendency of the larger plasmid.

120

The relationship between the plasmid size / ionic conditions and membrane fouling was

examined in more detail by plotting the fouling rate constants as a function of the radius of

gyration (Rg) of the plasmid DNA (Figure 7.4). The plasmid radii of gyration were estimated

from the retention volume determined by SEC using mobile phases of differing ionic strength

(TE buffer with 1, 10 and 500 mM NaCl). The radius of a given plasmid decreases with

increasing ionic strength due to the reduction in the magnitude of the intramolecular electrostatic

repulsion between the charged phosphate groups along the DNA backbone. For example, the

radius of the 16.8 kbp plasmid in the 1 mM NaCl solution is 240 nm compared to only 150 nm in

the 500 mM NaCl solution.

Figure 7.4: Fouling rate constant k plotted versus radius of gyration of plasmid DNA. Data include

3.0 and 9.8 kbp supercoiled plasmids in TE buffer containing 1 and 10 mM NaCl,

and 16.8 kbp supercoiled plasmids in TE buffer containing 1, 10, and 500 mM NaCl.

Fouling experiments were performed with Biomax 300 kDa membranes. Error bars

are within the size of the symbols.

121

The fouling rate constant, k, is strongly correlated with the effective plasmid size. The

values of k range from a low of k = 0.01 × 10-4 m2/kg for the plasmid with the smallest effective

size (Reff = 120 nm for the 3.0 kbp plasmid in 10 mM NaCl TE buffer) to a high of k = 0.94 ×

10-4 m2/kg for the plasmid with the largest size (Reff = 240 nm for the 16.8 kbp plasmid in 1 mM

NaCl TE buffer). In addition, plasmids with similar effective size (but different numbers of base

pairs), have very similar fouling rate constants. For example, the 3.0 kbp plasmid in 1 mM NaCl

TE buffer has Reff = 140 nm while the 16.8 kbp plasmid in 500 mM NaCl TE buffer has Reff =

150 nm, with these two plasmids having k = 0.12 and 0.13 × 10-4 m2/kg, respectively.

7.3.4 Effects of membrane pore size

An additional series of experiments was performed with the larger pore size Biomax

1000 kDa membranes, with the results in the 1 and 10 mM NaCl solutions shown in Figure 7.5.

Also shown for comparison are data for the 300 kDa membrane taken from Figure 7.1.

Experiments with the 1000 kDa membrane were done using 5.0 µg/mL solutions of the

supercoiled p-FDY plasmids in TE buffer at constant pressures of ∆P = 2.5 and 1.2 kPa for the 1

and 10 mM NaCl solutions, respectively, with these pressures chosen to give similar initial

values for the plasmid sieving coefficients. The hydraulic permeability of the Biomax 1000 kDa

membranes was 30 × 10-12 m, which is approximately 5 times the value obtained with the

Biomax 300 kDa membranes.

The rate of fouling was much less pronounced with the larger pore size membrane at both

NaCl concentrations. For example, the sieving coefficient data for the Biomax 1000 kDa

122

membrane using the 10 mM NaCl solution showed only a very small decline, from So = 0.97 to

0.87, after a mass throughput of as much as 170 mg/m2. A much greater decline in sieving

coefficient and filtrate flux was seen with the 1 mM NaCl solution, although in both cases the

fouling was less pronounced than that seen with the 300 kDa membrane even at the higher salt

concentration. In contrast to the data with the 300 kDa membrane, the decline in filtrate flux with

the 1000 kDa membrane is comparable to the decline in sieving coefficient, suggesting that there

is proportionally less filtrate flow through the blocked pores in the 1000 kDa membrane.

123

Figure 7.5: Effect of solution ionic strength on the sieving coefficient (top) and filtrate flux

(bottom) of the supercoiled plasmid through membranes with different MWCO. Data

obtained with 5×10-3 kg/m3 solutions of the 16.8 kbp plasmid with 1 and 10 mM

NaCl in TE buffer using Biomax 300 and 1000 kDa membranes. Dashed curves are

model calculations using parameter values in Table 7.1.

In order to understand these differences in fouling behavior in more detail, the data in

Figure 7.5 were analyzed using the fouling model to determine the best fit values of k and β, with

124

results summarized in Table 7.1. The mean pore size (rp) of the two membranes was estimated

from the measured hydraulic permeability using the Hagen-Poiseuille equation assuming

uniform cylindrical pores with length δm = 1 µm (equal to the thickness of the selective skin

layer) and porosity ε = 0.5:

𝑟𝑝 = (8𝛿𝑚𝐿𝑝

)1/2

(7.7)

Equation (7.7) gives rp = 22 nm and 9 nm for the 1000 kDa and 300 kDa membranes,

respectively. The values of the fouling rate constant for the Biomax 1000 kDa membrane are

about a factor of 3 smaller than those for the Biomax 300 kDa membrane at both NaCl

concentrations, which is consistent with the lower probability of plasmid trapping in the larger

pore size membranes. Interestingly, the best fit values of β are also much smaller for the Biomax

1000 kDa membranes (β = 0.14 ± 0.05) compared to the results obtained with the Biomax 300

kDa membranes (β = 0.44 ± 0.03), even though one would expect to have greater fluid flow

through a larger pore that is blocked by a single plasmid. The small values of β for the Biomax

1000 kDa membrane could be due to the presence of multiple DNA molecules in the blocked

pores, which would significantly reduce the available open space for fluid flow. This is also

reflected in the somewhat greater reduction in filtrate flux compared to that seen for the sieving

coefficient for the fouling experiments conducted with the Biomax 1000 kDa membranes (Fig.

7.5). Additional studies will be needed to fully clarify the origin of this behavior.

Table 7.1 summarizes the best fit values of the fouling parameters for all of the

experimental conditions. The fouling rate constant k and the pore blockage parameter β are both

strong functions of the solution ionic strength, the membrane pore size, and the plasmid size.

These changes are likely related to the conformational properties of the supercoiled plasmid. In

125

particular, electrostatic shielding at high salt concentrations reduces the intersegmental repulsion

in the DNA molecules, which causes the plasmids to adopt a more tightly interwound

conformation 101, 107. The more highly condensed and tightly interwound structure not only

facilitates transmission of the supercoiled plasmid through the small pore size ultrafiltration

membranes, it also reduces the probability of pore blockage (and membrane fouling) due to

trapping of plasmid at the pore entrance.

7.4 Conclusions

The data presented in this Chapter provide the first quantitative analysis of the effects of

solution ionic strength on membrane fouling during ultrafiltration of plasmid DNA. Membrane

fouling was negligible with the linear plasmid isoform. In contrast, the supercoiled plasmid

showed a significant decline in both filtrate flux and transmission, with the rate of fouling

decreasing significantly at high ionic strength and for the smaller size plasmids. Membrane

fouling in this system appears to be due to trapping of the large supercoiled plasmids at the pore

entrance 48; previous studies by Borujeni et al.76 showed that the flux and transmission could be

largely restored by periodic backpulsing to push these plasmids out of the pores. The fouling rate

constant was highly correlated with the effective plasmid size (as determined by size exclusion

chromatography), potentially allowing one to predict the rate of fouling for different size

plasmids under different ionic conditions based on limited fouling data in combination with SEC

measurements of the effective plasmid size.

The results from these fouling experiments could have significant implications in the

design of membrane processes for the purification of plasmid DNA. The dramatic reduction in

126

fouling in the higher ionic strength solution could be exploited by operating the membrane

process at relatively high salt concentrations, particularly for processes with higher DNA

concentrations. However, the data presented in Chapter 5 showed that the greatest selectivity for

the separation of the linear and supercoiled isoforms was achieved at low salt concentrations due

to the increase in retention of the supercoiled isoform under these conditions. The design and

optimization of membrane processes for purification of the desired supercoiled isoform thus

requires a detailed understanding of the tradeoffs between the selectivity and fouling behavior in

the different ionic strength solutions.

127

Chapter 8

Size-based Separation of Supercoiled Plasmid DNA using Ultrafiltration

Most previous studies of membrane-based separations have shown no effect of DNA size

on plasmid transmission through small pore size ultrafiltration membranes, consistent with the

predicted behavior for flexible polymer chains. However, supercoiled plasmids are known to

have a highly “branched” structure with the number of branches dependent on the DNA length.

This difference in branching could lead to a significant dependence of the transmission on the

plasmid size, providing opportunities for size-based separations using ultrafiltration.

The objective of the work described in this Chapter was to evaluate the transmission of

different size supercoiled plasmids during ultrafiltration. Data were obtained with 3.0, 9.8, and

16.8 kbp plasmids using both cellulosic and polyethersulfone ultrafiltration membranes with

different molecular weight cutoffs. Initial experiments were performed with purified samples of

the supercoiled or linear isoforms, with the results used to identify appropriate conditions for

plasmid separation. The majority of the work presented in this Chapter was previously published

in the Journal of Colloid and Interface Science (Li et al., 472, 195-201, 2016).

8.1 Introduction

DNA purification is a critical step in many microbiological processes, forensic analyses,

and in the large scale production of gene therapy agents and DNA based vaccines 34. This

includes the removal of other nucleic acids such as genomic DNA, RNA, and DNA dimers (e.g.,

linked plasmids), as well as plasmids with incorrect constructs 33. These nucleic acid separations

128

are particularly difficult since the DNA has similar surface charge and affinity, although there

can be large differences in the size of these species.

Agarose gel electrophoresis (AGE) is the standard method for size-based DNA

separations, with differences in DNA mobility through the gel arising from hydrodynamic

interactions with the agarose matrix 138-139. However, AGE is limited to laboratory-scale

separations, it is very time-consuming, and it can be difficult to recover DNA from the gel and

remove the stain used for visualization. Density gradient centrifugation using CsCl can also be

used for size-based DNA separations, although this usually requires more than 16 h of

ultracentrifugation 140.

Several size exclusion chromatography (SEC) resins have been specifically developed for

DNA separations, including the Sephacryl S-1000 and the Superose 6B 141. These large pore size

resins can provide reasonable resolution for DNA separations, particularly between very large

genomic DNA and smaller plasmids, although the throughput tends to be very low due to the

significant diffusional resistance arising from the large size of the DNA 142. In addition, baseline

resolution can be difficult to achieve due to the broad peaks and the physical and chemical

similarity between the impurities and the supercoiled plasmid 143. For example, McClung and

Gonzales used the Superose 6 resin for purification of plasmids from E Coli extract containing

DNA fragments with good resolution, but all plasmids from 4 to 150 kbp eluted at the same

retention volume, with no fractionation of these plasmids on the basis of size or length 144.

Raymond et al. 145 used the Sephacryl S-1000 resin for purification of supercoiled DNA, with

good (but incomplete) removal of RNA and genomic DNA for both 4.4 and 12 kbp plasmids.

Membrane separations have replaced SEC in many size-based separations due to the

large increase in throughput and the significant reduction in processing time. For example, buffer

129

exchange in the formulation of therapeutic proteins is now done almost entirely by ultrafiltration

/ diafiltration 79, 146. Membrane systems can also be used for much higher resolution size-based

separations, e.g., between protein monomers and dimers 147. However, previous studies of

membrane systems for DNA separations have generally shown little if any dependence of

plasmid retention on the DNA size 37, 43. Latulippe and Zydney 43 hypothesized that this was due

to the elongation of the plasmid in the converging flow field approaching the membrane pores,

with the larger plasmids having more time to elongate as they approach the pore. This behavior

is in good agreement with predictions of scaling models developed to describe the elongation of

single polymer chains during passage through isolated small pores 42. However, it is well known

that supercoiled plasmids adopt a more complex 3-dimensional morphology, which could lead to

very different behavior during ultrafiltration.

The objective of this Chapter was to investigate whether ultrafiltration could be used for

the separation of supercoiled plasmids based on differences in their size, i.e., number of base

pairs. Initial experiments were conducted with purified plasmids of different size to determine

the effect of the plasmid size on transmission through different pore size ultrafiltration

membranes. Corresponding experiments were performed with linear plasmids to confirm the role

of the supercoiled structure on the ultrafiltration behavior. Appropriate conditions were identified

and then applied for the separation of a binary mixture of supercoiled plasmids with different

size. The results were in good agreement with a simple physical model for the transmission of

“branched” polymers, providing further confirmation of the potential for using ultrafiltration for

size-based separation of supercoiled plasmid DNA.

130

8.2 Materials and methods

200 × 10-6 kg/m3 (200 μg/mL) stock solutions of supercoiled plasmids were obtained

from Aldevron (Fargo, ND) and stored frozen at -20 °C. Three different size plasmids: 3.0, 9.8,

and 16.8 kilo base pair (kbp), were used in the experiments. The linear plasmid isoforms were

prepared from the supercoiled isoforms using restriction endonucleases that recognize and cleave

a specific nucleotide sequence in the double-stranded DNA following procedures described in

Chapter 2.

25 mm diameter polyethersulfone (Biomax) ultrafiltration membranes with nominal

molecular weight cut-offs of 50 kDa (PBQK02510), 100 kDa (PBHK02510), and 300 kDa

(PBMK02510) were provided by MilliporeSigma. Limited experiments were also performed

using 100 kDa Ultracel composite regenerated cellulose membranes (PLHK02510,

MilliporeSigma). DNA concentrations were analyzed using the PicoGreen Assay and the size,

integrity, and topology of the different plasmid isoforms were confirmed using Agarose gel

electrophoresis (AGE).

A 10 mL stirred cell (MilliporeSigma) was used in the ultrafiltration experiments. The

stirring speed in the stirred cell was adjusted to 730 rpm for all experiments. The pressure in the

system was controlled by air pressurization of the polycarbonate feed reservoir that was

connected to the stirred cell using pressures from 0 to 60 kPa (approximately 0-8 psi). Additional

details on the plasmid filtration experiments are provided in Chapter 2.

8.3 Results and discussion

8.3.1 Supercoiled plasmids

131

Figure 8.1 shows the transmission of 0.20 - 0.25 μg/mL solutions of the individual 3.0,

9.8, and 16.8 kbp supercoiled plasmids through a Biomax 100 kDa membrane as a function of

filtrate flux. Two samples were obtained at each filtrate flux, with the concentrations measured in

duplicate and reported as the mean values. The data were highly reproducible; the error bars on

the sieving coefficients lie within the size of the symbols and are not shown. There was no

evidence of membrane fouling during the experiments with these dilute plasmid solutions – the

membrane hydraulic permeabilities before and after each ultrafiltration experiment were

statistically indistinguishable. The integrity of the plasmids in both the feed and filtrate samples

was confirmed by AGE; there were no visible structural changes of any DNA sample due to

either filtration through the membrane or prolonged stirring during the ultrafiltration.

Figure 8.1: Observed sieving coefficients of the 3.0, 9.8, and 16.8 kbp supercoiled plasmids

through the 100 kDa Biomax membrane in TE buffer with 300 mM NaCl.

132

The sieving coefficients for all three plasmids were essentially zero (So < 0.02) at filtrate

flux below 40 μm/s (corresponding to 140 L/m2/h). The sieving coefficients increased

significantly with increasing filtrate flux due to the elongation of the plasmid in the converging

flow field entering the membrane pores as discussed previously. Note that the 16.8 kbp plasmid

has a radius of gyration of 169 nm (determined by static light scattering75), while the mean pore

radius of the Biomax 100 kDa membrane is less than 10 nm. The largest sieving coefficients

were obtained with the 3.0 kbp plasmid. For example, at a filtrate flux around 140 μm/s (500

L/m2/h), the sieving coefficient of the 3.0 kbp plasmid was above 0.9 while that for the 16.8 kbp

plasmid was below 0.25. Similar behavior was observed at both lower and higher salt

concentrations (150 and 500 mM), although the actual values of the sieving coefficient tended to

increase slightly with increasing ionic strength.

The selectivity between two different sizes of plasmids can be evaluated using Equation

(2.2), where in this case the subscripts “1” and “2” would refer to the 3.0 and 16.8 kbp plasmids,

respectively. The data in Figure 8.1 have been re-plotted in Figure 8.2 in terms of the selectivity,

with the sieving coefficients for the 3.0 and 16.8 kbp plasmids at the specific flux values

determined by interpolation of the raw data using a smoothed polynomial fit. The selectivity

goes through a maximum value of approximately ψ = 9 at a filtrate flux around 70 µm/s due to

the more rapid initial increase in transmission of the 3.0 kbp plasmid compared to that of the

16.8 kbp plasmid at low filtrate flux. The reduction in selectivity at high flux is due to the

increase in transmission of the larger plasmid.

133

Figure 8.2: Selectivity between the 3.0 and 16.8 kbp supercoiled plasmids as a function of filtrate

flux. Ultrafiltration experiments were performed using 100 kDa Biomax membranes

in TE buffer containing 300 mM NaCl.

The effect of membrane pore size on the selectivity between the 3.0 and 16.8 kbp

plasmids is examined in Figure 8.3 based on data obtained with the Biomax 50, 100, and 300

kDa membranes in TE buffer containing 300 mM NaCl. In each case, the maximum value of the

selectivity is shown; this occurred at a filtrate flux of 110 µm/s for the 50 kDa membrane and at

Jv = 5 µm/s for the 300 kDa membrane. The selectivity was very low with the large pore size

Biomax 300 kDa membrane, withψ < 2.5 for all conditions, due to the similar (and relatively

large) values of the sieving coefficients for both plasmids. The maximum selectivity for the

Biomax 50 kDa membrane was only about ψ ≈ 4.5 due to the high degree of retention for both

plasmids through this small pore size membrane. The greatest selectivity was seen with the

Biomax 100 kDa membrane (data from Figure 8.2). It is possible that the small pore membrane

134

could separate plasmids of different sizes, but that would require a very high filtrate flux

(transmembrane pressure) in order to get significant plasmid transmission, which is beyond the

operation range of the stirred cell apparatus used in this work.

Figure 8.3: Selectivity between the 3.0 and 16.8 kbp supercoiled plasmids as a function of

membrane MWCO. Only the optimal Ψ values were plotted. Ultrafiltration

experiments were conducted in TE buffer containing 300 mM NaCl. The optimal Ψ

values were obtained at filtrate flux of 110, 70, and 5.2 µm/s for the 50, 100, and 300

kDa membranes, respectively.

To confirm that the size-dependent transmission of the supercoiled plasmids was not

unique to the Biomax membranes, additional sieving experiments were performed using the

Ultracel 100 kDa regenerated cellulose membrane. As seen in Figure 8.4, the behavior of the

Ultracel 100 kDa membrane is similar to that seen with the Biomax 100 kDa membrane, with

much larger transmission of the 3.0 kbp supercoiled plasmid compared to that of the 9.8 and 16.8

135

kbp plasmids. The maximum selectivity between the 3.0 and 16.8 kbp plasmids was ψ ≈ 30 at a

filtrate flux around 40 μm/s (L/m2/h). The Ultracel membrane also showed a large selectivity

between the 3.0 and 9.8 kbp plasmids, with ψ ≈ 12 under these conditions.

Figure 8.4: Observed sieving coefficients of the 3.0, 9.8, and 16.8 kbp supercoiled plasmids

through the 100 kDa Ultracel membrane in TE buffer with 500 mM NaCl.

Actual separation of the 3.0 and 16.8 kbp plasmids was done by performing an

ultrafiltration experiment with a binary mixture of the two supercoiled plasmids, each at a

concentration of 0.25 μg/mL, using the same solution conditions as in Figure 8.4. Data were

obtained at a filtrate flux of 70 μm/s using an Ultracel 100 kDa membrane; this corresponds to a

selectivity about 8-fold. This higher flux was chosen to increase the transmission of the 3.0 kbp

plasmid and enhance the accuracy of the AGE used to evaluate the performance of the

136

ultrafiltration process. Figure 8.5 shows the AGE image for filtrate and feed samples obtained 2

min after the start of the ultrafiltration. The first lane shows a 1.0 kbp DNA ladder; Lanes 4 and

5 are the purified 3.0 and 16.8 kbp supercoiled plasmids for reference. The 16.8 kbp supercoiled

plasmid sample contains some higher molecular weight species (faint band near the top of the gel

in Lane 5), which could be plasmid dimers or low levels of the open circular isoform of this

plasmid. The feed sample for the ultrafiltration experiment (Lane 3) consists of equal amounts of

the 3.0 and 16.8 kbp plasmids, with the lower band corresponding to the smaller 3.0 kbp plasmid

since it migrates faster through the gel during the electrophoresis; again, some high molecular

weight species is seen near the top of the gel. The filtrate sample (Lane 2) shows only a single

band corresponding to the 3.0 kbp plasmid with no detectable levels of the 16.8 kbp plasmid or

the higher molecular weight species, consistent with the sieving coefficient data for the

individual plasmids shown previously in Figure 8.4. These data clearly demonstrate that

ultrafiltration can be used for the separation of different supercoiled plasmids on the basis of

size.

137

Figure 8.5: Agarose gel electrophoresis (AGE) showing the separation of a binary mixture of the

3.0 and 16.8 kbp supercoiled plasmids using an UltracelTM 100 kDa membrane in

TE buffer containing 500 mM NaCl. Lane 1: linear 1 kbp DNA ladder. Lane 2: filtrate

sample collected at a filtrate flux of 70 µm/s. Lane 3: feed sample. Lane 4: purified

3.0 kbp supercoiled plasmid. Lane 5: purified 16.8 kbp supercoiled plasmid.

8.3.2 Linear plasmids

A corresponding series of experiments was performed with linear versions of the three

plasmids, each generated by enzymatic digestion of the corresponding supercoiled isoform.

Figure 8.6 presents results for the linear 3.0 and 16.8 kbp plasmids through the Ultracel 100 kDa

(left panel) and Biomax 100 kDa (right panel) membranes in TE buffer containing 150 mM and

10 mM NaCl, respectively. Similar results were obtained at other solution ionic strength and with

both the smaller and larger molecular weight cutoff membranes. The sieving coefficients of the

linear plasmid are considerably larger than the values seen with the supercoiled plasmids

(Figures 8.1 and 8.4) at the same filtrate flux due to the greater elongational flexibility of the

linear isoform; this is discussed in detail by Latulippe and Zydney 45 and has been exploited for

138

the separation of the different DNA isoforms by ultrafiltration. The sieving coefficients for the

two plasmids are very similar for all values of the filtrate flux, with differences of less than 8%

(except for a single data point at 40 μm/s for the Biomax 100 kDa membrane). Thus, the good

selectivity for the supercoiled plasmids seen in Figures 8.1 to 8.5 is completely absent with the

linear isoforms; the selectivity between the 3.0 and 16.8 kbp linear isoforms was less than 1.2

under all experimental conditions.

Figure 8.6: Observed sieving coefficients of the 3.0, 9.8, and 16.8 kbp linear plasmids through (a)

the Ultracel 100 kDa membrane in TE buffer with 150 mM NaCl and (b) the Biomax

100 kDa membrane in TE buffer with 10 mM NaCl.

8.3.3 Physical interpretation

A number of previous experimental and theoretical studies have examined the

transmission of linear polymers through narrow pores. Daoudi and Brochard 42 used scaling

139

arguments to show that the critical volumetric flow rate required for passage of the chain through

a pore scales as:

𝑞𝑐 ≈ 𝑘𝐵𝑇

𝜂 (8.1)

independent of the polymer chain length (where kB is Boltzmann’s constant, T is the absolute

temperature, and η is the solution viscosity). Sakaue et al.148 used a force balance analysis to

show that passage of a linear polymer through a small pore is controlled by the injection of the

first polymer “blob”, which is again independent of the polymer chain length. Both molecular

dynamic simulations 149 and experimental studies 150 have confirmed the universal transmission

behavior of linear polymers with different chain length.

These results strongly suggest that the size-dependent transmission seen in Figures 8.1 to

8.5 is due to the unique molecular structure of the supercoiled plasmid isoform. Supercoiled

DNA adopts a plectonemic (interwound) conformation with numerous branch points due to the

helical “twists” in the circular plasmid (leading to local contortions or “writhe”). This branching

is driven by the increase in entropy associated with the expanded conformation, but is

enthalpically unfavorable due to the additional bending energy required to form the branches 22-

23, 151. Previous studies have shown that the extent of branching is proportional to the size of the

supercoiled plasmid 23. The degree of branching is often quantified by the number of superhelix

“ends” 152. For example, Hammermann et al. 105 showed that a 1.868 kbp supercoiled DNA was

essentially unbranched (N = 2), whereas a 5.243 kbp molecule adopted a conformation with

N ≈ 3. Fathizadeh et al. 153 used molecular dynamics simulations to model the structure of

supercoiled plasmids with lengths between 1.2 and 6 kbp and showed that the average number of

superhelix ends increased from 2 (no branching) for the smallest plasmid to 4.5 ± 0.5 for the 6

140

kbp plasmid. Boles et al. 17 used scanning electron microscopy to count the average number of

branch points per DNA as 1.6 and 2.9 for a 3.5 and 7 kbp supercoiled plasmid, respectively.

Vologodskii and Cozzarelli 22 evaluated the branching frequency of supercoiled DNA as a

function of DNA length using Monte Carlo simulations, with the results showing N = 2, 7, and

12 for plasmids with lengths of 3, 10 and 17 kbp (similar to the size of the plasmids examined in

this work).

Several investigators have developed simple scaling models for the transport of polymer

chains with different topologies (e.g., branching) through small cylindrical pores154-156. The

passage of a polymer chain through a cylindrical pore happens when the confinement and

hydrodynamic forces on each individual blob are balanced (i.e. fc = fh). Here 𝝃, u (=q/D2, where

D refers to the pore diameter), and le are the “blob” diameter, flow velocity, and the blob’s

effective length along the flow direction, respectively; and each “blob” is defined as the

maximum portion of the confined chain whose confinement free energy becomes of order

thermal energy (kBT). The critical flow rate for polymer chains to pass through the pore was

found to be a function of the polymer topology (number of branches):

𝑞𝑐

𝑞𝑐,𝑙𝑖𝑛𝑒𝑎𝑟= (

𝐷

𝜉)2 (8.2)

where 𝑞𝑐,𝑙𝑖𝑛𝑒𝑎𝑟 = 𝑘𝐵𝑇

3𝜋𝜂 (

ξ

𝑙𝑒). In the limit of small pore size (i.e., membrane pore diameter <<

polymer length), the critical flow rate for injection of a branched polymer into a small pore was

found to scale as

𝑞𝑐,𝑏𝑟𝑎𝑛𝑐ℎ

𝑞𝑐,𝑙𝑖𝑛𝑒𝑎𝑟= 𝑛𝑏𝑟𝑎𝑛𝑐ℎ

1/4 (8.3)

141

where nbranch is defined as the number of branching points of the hyperbranched chain (equal to

the number of ends minus two). Thus, Equation (8.3) predicts that the critical flow rate increases

by a factor of approximately 1.5 in going from the 3.0 to 9.8 kbp plasmids, with a 1.8-fold

increase in the critical flow rate for the 16.8 kbp plasmid.

The experimental data for plasmid transmission as a function of filtrate flux can be used

to estimate the critical flow rate by defining qc as the flux at which So ≈ 0.1. The data in Figure

8.1 give values of the critical flux for the 3.0, 9.8, and 16.8 kbp plasmids of 47, 52, and 81 μm/s

based on linear interpolation of the sieving coefficients. Thus, the ratio of the critical flow rate

for the 16.8 kbp plasmid to that for the 3.0 kbp plasmid is 1.7, in excellent agreement with the

factor of 1.8 given by Equation (8.3). Similar results were seen with the other membranes and

with the 9.8 kbp plasmid, providing further evidence that the observed differences in

transmission of the different size supercoiled plasmids is due to differences in the underlying

topologies associated with the writhe / branching of the longer plasmids.

Although there have been no prior experimental studies showing the effects of DNA

branching on plasmid ultrafiltration, Ge and Wu 157 examined the transmission of linear and star-

shaped polystyrene through ultrafiltration membranes with well-defined 20 nm pores. The

polystyrene chains were synthesized by coupling “living” polystyl chains of different lengths

using divinylbenzene. The linear polystyrenes showed much greater transmission than the

corresponding star-shaped polymers (with the same total chain length), consistent with the

behavior predicted by Equation (8.3) and in good qualitative agreement with the data obtained in

this study for the ultrafiltration of supercoiled versus linear plasmids.

142

8.4 Conclusions

The results presented in this Chapter clearly demonstrate the potential of using

ultrafiltration membranes to separate supercoiled plasmids based on differences in their size (i.e.,

number of base pairs). Plasmid transmission through small pore ultrafiltration membranes is due

to the elongation of the DNA chain in the converging flow field approaching the membrane

pores, with minimal transmission below a critical value of the filtrate flux. However, the

different size supercoiled plasmids have different critical flux, leading to greater transmission of

the smaller plasmids. The Ultracel 100 kDa membrane showed a selectivity between the 3.0 and

16.8 kbp plasmids as high as 30, with similar behavior seen with the Biomax polyethersulfone

membrane. Note that a selectivity of 30 could provide a 100-fold purification with 90% yield

using a diafiltration process based on the expected trade-off between the product yield and

purification factor in membrane separation systems 52. For example, the yield of a product that is

collected in feed (retentate) tank during diafiltration is given as:

𝑌 = exp(−𝑁𝐷𝑆) (8.4)

where ND is the number of diavolumes and S is the sieving coefficient of the species of interest.

The purification factor can be calculated as:

P =Y 1-y (8.5)

where ψ is the selectivity.. Thus, a membrane system with selectivity of 30 would give P > 100

with Y = 90%.

The potential for using ultrafiltration for separation of different size supercoiled plasmids

was confirmed by agarose gels of filtrate samples obtained in an experiment using a binary

143

mixture of the 3.0 and 16.8 kbp plasmids. To our knowledge, this is the first quantitative

demonstration of a size-based separation of large plasmid DNA using ultrafiltration.

The reduction in transmission of the supercoiled plasmids with increasing chain length is

a direct result of the morphology of the supercoiled isoform; no significant affect of plasmid size

was seen during ultrafiltration of linear versions of the same plasmids. The supercoiled isoforms

adopt a branched structure due to the under-twisting of the DNA, with the number of branches

increasing with increasing chain length. The experimental results obtained in this Chapter are

consistent with the critical flux determined by a scaling analysis for branched polymers,

providing further confirmation of the origin of this size-based ultrafiltration behavior. Note that

previous studies of plasmid ultrafiltration by Latulippe et al.37, Borujeni and Zydney158 and

Arkhangelsky et al.61 did not observe any significant dependence of plasmid transmission on the

size of the supercoiled DNA, although these experiments were done with considerably larger

pore size membranes (1000 kDa molecular weight cutoff and 20 nm pores, respectively) which

would be expected to have minimal selectivity based on the results in Figure 8.3.

Membrane systems could provide an attractive alternative for the purification of

supercoiled plasmid DNA, both for laboratory analysis and in the preparation of gene therapy

agents or DNA-based vaccines. Membranes are relatively inexpensive, and they provide much

faster separations with greater throughput and scalability than size exclusion chromatography

(SEC). Although additional studies will be needed to quantify the selectivity for different DNA

separations, the data obtained in this Chapter suggest that properly designed ultrafiltration

processes could potentially provide higher resolution separations of supercoiled plasmids than is

possible using currently available SEC resins, particularly for very large size plasmids.

144

Chapter 9

Conclusions and Recommendations for Future Work

9.1 Conclusions

There is growing interest in the use of plasmids for applications in DNA therapeutics,

including both gene therapy and DNA vaccination. The large dosage required for DNA

therapeutics has created a critical need for the development of cost-effective processes for large-

scale plasmid purification. Conventional separation methods, such as different modes of

chromatography, suffer from low binding capacity and significant mass transfer limitations due

to the large size of the plasmid DNA. In addition, the separation of open-circular and linear

plasmid isoforms from the desirable supercoiled isoform is particularly challenging since these

molecules have the same molecular weight and base sequence, with very similar size and surface

charge.

Membrane processes could provide an attractive alternative for industrial scale plasmid

DNA purification since they are easily scalable and only weakly affected by diffusional

limitations associated with the large size of the plasmids. Membranes have been widely used in

the biotechnology and pharmaceutical industries for primary clarification, concentration, and

formulation. Latulippe et al. 37 and Latulippe and Zydney 43 successfully demonstrated that

narrow pore size ultrafiltration membranes can also be used for high resolution purification of

different plasmid isoforms based on differences in their ability to elongate in the converging flow

field entering the membrane pores. The linear plasmid requires the smallest flow rate to be

sufficiently stretched to pass through narrow pores, followed by the supercoiled and then the

145

open-circular isoforms. Separation therefore can be achieved by operating at a filtrate flux

between the critical flux values for the specific isoforms.

There are two major bottlenecks limiting the performance of membrane systems in the

purification of plasmid DNA: (A) the limited selectivity that can be achieved between the

different isoforms, and (B) fouling of the ultrafiltration membranes due to the blockage of the

membrane pores (which reduces both he flux and the selectivity). The overall objectives of this

dissertation were:

I. Designing novel strategies to enhance the resolution of plasmid isoform

separation and / or control membrane fouling

II. Developing a more detailed fundamental understanding of the key factors

governing both the transmission of DNA through nanopores and membrane

fouling phenomena during plasmid ultrafiltration, including the effects of the

membrane structure and pore morphology, solution ionic conditions, size of

plasmid DNA etc.

The key findings from this dissertation are summarized as below.

Chapter 3 introduced and examined the idea of pre-conditioning, accomplished by pre-

elongating the DNA by passage through a region with large pore size, to minimize fouling and

enhance DNA separations. Data were obtained using asymmetric hollow fiber membranes, with

flow in either the normal or reverse orientation. Flow through the larger pore size region pre-

stretched the plasmid, leading to an increase in plasmid transmission and a significant reduction

in fouling. This pre-conditioning also provided a significant increase in selectivity for separation

of the linear and supercoiled isoforms. These results clearly demonstrate the potential for

146

dramatically increasing the performance of membrane systems for plasmid DNA separations by

controlling the pore morphology to pre-stretch the DNA before passing through the narrow pores

of an ultrafiltration membrane.

The data presented in Chapter 4 demonstrated that pre-conditioning using composite

membranes can also enhance the transmission of supercoiled DNA during ultrafiltration. Data

were obtained using a series of composite membranes generated by placing a larger pore size

microfiltration membrane on top of the skin layer of an ultrafiltration membrane. The increase in

plasmid transmission was found to be greatest when using an intermediate pore size for the upper

layer of the composite membranes. Upper layers with too large a pore size provided insufficient

stretching of the DNA while layers with too small a pore size gave minimal improvement in

performance compared to that of the ultrafiltration membrane alone. This also provided

opportunities to enhance the separation between the open-circular and supercoiled plasmid

isoforms. Several physical models were developed to help understand the effects of pre-

conditioning on DNA transmission. Further confirmation of these models is going to be obtained

using membranes with conically shaped pores. These results provide important insights into the

role of preconditioning in enhancing the performance of DNA ultrafiltration, while providing a

framework for the design and optimization of membranes specifically for plasmid DNA

separations.

Chapter 5 examined the enhancement in separation of the different plasmid isoforms that

could be achieved by proper selection of the solution ionic strength and ion type. Experiments

were performed with a 3.0 kbp plasmid using composite regenerated cellulose ultrafiltration

membranes. The transmission of the linear isoform was nearly independent of solution ionic

strength. In contrast, the transmission of the open-circular and supercoiled plasmids both

147

increased with increasing NaCl or MgCl2 concentration due to the change in plasmid size and

conformational flexibility. The effect of ionic strength was greatest for the supercoiled plasmid,

providing opportunities for enhanced purification of this therapeutically active isoform. This

behavior was confirmed using experiments performed with binary mixtures of the different

isoforms. These results clearly demonstrate the potential for enhancing the performance of

membrane systems for plasmid DNA separations by proper selection of the ionic conditions.

Chapter 6 discussed the effects of different levels of the polycation spermine on the

transmission of supercoiled, linear, and open-circular plasmid isoforms through small pore size

ultrafiltration membranes. Adding low concentrations (below 10 µM) of spermine to TE buffer

was found to have little effect on the transmission of all plasmid isoforms through the Biomax

300 kDa membranes. For supercoiled and linear plasmids, DNA condensation happened at a

threshold spermine concentration, above which transmission of the plasmid DNA dropped

rapidly with the membranes becoming nearly completely retentive to the plasmid. The amount of

spermine needed to induce DNA condensation increased with increasing NaCl concentration.

Once the plasmids were condensed into densely packed structures, they were unable to pass

through the narrow pore ultrafiltration membranes even at high filtrate flux, conditions that

would have readily elongated the DNA in the absence of any spermine. DNA condensation was

reduced in the presence of high concentrations of monovalent salts, potentially providing an

opportunity to “tune” the transmission of the DNA isoforms by proper of addition of spermine

and NaCl to the solutions.

Chapter 7 evaluated the effects of solution conditions on the fouling characteristics of

both supercoiled and linear plasmid DNA isoforms with different numbers of base pairs. Sieving

coefficient and filtrate flux data were analyzed using a model based on the partial blockage of the

148

membrane pores by trapped plasmids. Fouling increased dramatically at low ionic strength, with

the flux decline parameter for the 3.0 kbp plasmid in a 1 mM NaCl solution being an order of

magnitude greater than that in a 10 mM solution. Fouling was also most pronounced for the

larger 16.8 kbp plasmid, consistent with the greater probability of plasmid trapping at the pore

entrance. These results provide important insights into the development of membrane systems

for plasmid DNA filtration in different solution conditions.

The results presented in Chapter 8 demonstrated the potential of using ultrafiltration

membranes to separate supercoiled plasmids based on differences in their size (i.e., number of

base pairs). The Ultracel 100 kDa membrane showed a selectivity between the 3.0 and 16.8 kbp

plasmids as high as 30-fold, with similar behavior seen with the Biomax polyethersulfone

membrane. The reduction in transmission of the supercoiled plasmids with increasing chain

length was a direct result of the morphology of the supercoiled isoform; no significant affect of

plasmid size was seen during ultrafiltration of linear versions of the same plasmids. The

supercoiled isoforms adopt a branched structure due to the under-twisting of the DNA, with the

number of branches increasing, and the DNA transmission decreasing, with increasing chain

length. To our knowledge, this is the first quantitative demonstration of a size-based separation

of large plasmid DNA using ultrafiltration membranes.

Overall, the results from this dissertation provide a variety of strategies to effectively

enhance the performance of membrane systems for the ultrafiltration / separation of plasmid

DNA. The insights obtained from these studies provide a better understanding of the key

physical factors that determine the transmission of plasmid DNA through ultrafiltration

membranes and a guide to the systematic design of large scale plasmid purification processes.

149

9.2 Future recommendations

Our current studies on membrane ultrafiltration of plasmid DNA were mostly performed

in small scale systems (15 mL stirred cell) with a single plasmid isoform. Actual separations of

isoform mixtures can be accomplished in linearly-scalable tangential flow filtration modules

using a diafiltration mode of operation (with the more permeable plasmids effectively washed

through the membrane by the continuous addition of diafiltration buffer). This could potentially

allow much greater throughput than was seen in the normal flow filtration experiments

performed in stirred cells. For example, van Reis et al. 147 developed the use of high performance

tangential flow filtration (HPTFF) for large-scale separations of recombinant protein products,

which can be scaled to membrane areas of at least 100 m2.159 Future work on using TFF for the

separation of plasmid DNA isoforms would be focused on demonstrating that the approaches

developed previously for protein purification could be extended to the large scale production of

plasmid DNA.

Appropriate conditions for DNA separations could first be identified based on small scale

studies of the individual plasmid isoforms. A combination of strategies identified in this thesis

could then be applied to enhance the separation efficiency as well as minimize membrane

fouling, including the use of membranes with pore morphologies that effectively pre-stretch the

DNA. The diafiltration would begin at a relatively low filtrate flux (above the critical flux for the

linear isoform but below the critical flux for the supercoiled plasmid) to wash the linear plasmid

through the membrane. The flux would then be increased to allow the supercoiled plasmid to be

recovered in the permeate solution, while the least flexible open-circular plasmid is retained by

the membrane. The separation process could also be done using a multi-stage ultrafiltration

150

system with different modules operated using different pore size and / or filtrate flux to obtain

the desired separation.

The experimental studies described in this dissertation used pure plasmid DNA solutions

prepared by Aldevron following stringent specifications. Future studies with simulated (or

actual) cell lysate solutions are critical to determine the effect of additional impurities (such as

genomic DNA and host cell proteins) on the performance of membrane system for plasmid

purification. Initial experiments would likely be conducted with model systems using specific

proteins having a range of size (i.e., molecular weight) and surface charge (i.e., isoelectric point).

Subsequent work would examine the purification of plasmids directly from E. Coli cell lysate.

Depth filtration would likely be used to remove large cell debris. Filtration experiments would

then be performed over a range of filtration conditions, with the smaller host cell proteins

removed in the permeate while the large plasmids are retained by the membrane. A variety of

filtration conditions would likely need to be examined, including a range of membrane pore size

and materials, solution pH, and salt concentrations, with the goal of minimizing membrane

fouling while enhancing the separation performance.

In addition to DNA therapeutics, numerous types of RNA-based therapeutics have

emerged and are under extensive clinical development. RNA therapeutics exploit various

oligonucleotides that bind to RNA by base pairing in a sequence-specific manner yet have

different mechanisms of actions and effects 160. RNA-therapeutics include inhibitors of mRNA

translation (antisense), agents of RNA interference (RNAi), catalytically active RNA molecules

(ribozymes), and RNAs that bind proteins and other molecular ligands (aptamers) 161. Hundreds

of RNA-based therapeutics are currently under clinical investigation for diseases ranging from

genetic disorders to HIV infection to various cancers 162. RNAs are less stable than DNA, so

151

there are additional difficulties associated with their production and purification. RNA is a single

stranded nucleotide chain that shares many similar properties as other large biopolymers such as

DNA and polysaccharides. Unlike DNA ultrafiltration, transmission of polysaccharides through

ultrafiltration membrane is governed by concentration polarization effects 136. Future work

examining the transport mechanisms of RNA through narrow pore UF membranes would be of

considerable interest, both for developing a more fundamental understanding of nucleic acid

ultrafiltration and for developing new downstream processes for purification of novel RNA-

based therapeutics. This would likely include studies examining the effects of membrane

properties (e.g., pore size and pore morphology) and solution conditions (e.g., ionic strength and

ionic composition) on RNA ultrafiltration.

An interesting alternative to pressure-driven ultrafiltration would be to use an applied

electric field to drive the transport of the charged DNA molecules through the pores in UF

membranes. There have been extensive theoretical and experimental studies on the electrical

field driven translocation of single/double stranded DNA through nanopores. Kasianowicz et

al.163, Mathe et al.164, and Meller et al.165-166 examined the transport of single stranded DNA

through a nanometer size channel within a naturally-occurring membrane protein, with the rate

of transport determined by monitoring transient changes in ionic current. Li et al.167, Storm et

al.168, and Wanunu et al.169 studied the transport of double stranded DNA through solid state

nanopores, including the molecular dynamics involved in DNA translocation. A detailed

experimental analysis of electrically-driven UF processes for plasmid purification would provide

important insights into the underlying physics governing DNA transport and a basis for the

design of effective separations using this approach, both for small scale analytical separations as

well as the possibility of developing larger-scale electrically-driven separation processes. It

152

might even be possible to use a combination of convective flow and electric fields to provide

further improvements in membrane separation. In addition, membranes with different surface

charges can be produced by chemical modification, providing an additional control of DNA

transport by exploiting both intra- and inter-molecular electrostatic interactions.

Further studies should also be performed to better understand the fundamental

mechanisms controlling transport of plasmid DNA through the narrow pores in ultrafiltration

membranes. This would include the effects of pore size distribution and hydrodynamic

interactions between adjacent pores, both of which could be examined theoretically using

computational fluid dynamics simulations to evaluate the velocity profiles and extensional flow

field during filtration through model membranes consisting of an array of cylindrical pores with

different packing geometries and pore size distribution. The transport of DNA molecules in

membranes with well-defined pore dimensions and geometries can also be examined

experimentally by constructing various micro/nano-fluidic devices composed of slits, channels,

and arrays of posts 170. These micro/nano-fluidic devices would also provide opportunities to

study single DNA molecules in idealized environments rather than the bulk behavior averaged

over large numbers of DNA molecules as in membrane systems.

153

Bibliography

1. Mountain, A., Gene therapy: the first decade. Trends in Biotechnology 2000, 18 (3), 119-

128.

2. Verma, I. M.; Somia, N., Gene therapy-promises, problems and prospects. Nature 1997,

389 (6648), 239-242.

3. Ulmer, J. B.; Wahren, B.; Liu, M. A., Gene-based vaccines: Recent technical and clinical

advances. Trends Mol Med 2006, 12 (5), 216-22.

4. Ferraro, B.; Morrow, M. P.; Hutnick, N. A.; Shin, T. H.; Lucke, C. E.; Weiner, D. B.,

Clinical applications of DNA vaccines: Current progress. Clin Infect Dis 2011, 53 (3), 296-302.

5. Ylä-Herttuala, S., Endgame: Glybera finally recommended for approval as the first gene

therapy drug in the European union. Molecular Therapy 2012, 20 (10), 1831.

6. Schimmer, J.; Breazzano, S., Investor outlook: Rising from the ashes; GSK's European

approval of Strimvelis for ADA-SCID. Human Gene Therapy Clinical Development 2016, 27

(2), 57-61.

7. Bessis, N.; GarciaCozar, F. J.; Boissier, M. C., Immune responses to gene therapy

vectors: Influence on vector function and effector mechanisms. Gene Ther 2004, 11 Suppl 1,

S10-7.

8. Patil, S. D.; Rhodes, D. G.; Burgess, D. J., DNA-based therapeutics and DNA delivery

systems: A comprehensive review. The AAPS Journal 2005, 7 (1), E61-E77.

9. Niidome, T.; Huang, L., Gene therapy progress and prospects: Nonviral vectors. Gene

Ther 2002, 9 (24), 1647-52.

10. Schaffert, D.; Wagner, E., Gene therapy progress and prospects: Synthetic polymer-based

systems. Gene Therapy 2008, 15 (16), 1131-8.

154

11. Bodles-Brakhop, A. M.; Heller, R.; Draghia-Akli, R., Electroporation for the delivery of

DNA-based vaccines and immunotherapeutics: Current clinical developments. Mol Ther 2009,

17 (4), 585-92.

12. Herweijer, H.; Wolff, J. A., Progress and prospects: Naked DNA gene transfer and

therapy. Gene Ther 2003, 10 (6), 453-8.

13. Li, S. D.; Huang, L., Gene therapy progress and prospects: Non-viral gene therapy by

systemic delivery. Gene Ther 2006, 13 (18), 1313-9.

14. Rhoades, M.; Thomas, C., The P22 bacteriophage DNA molecule: II. Circular

intracellular forms. Journal of molecular biology 1968, 37 (1), 41IN357-56IN661.

15. Laundon, C. H.; Griffith, J. D., Curved helix segments can uniquely orient the topology

of supertwisted DNA. Cell 1988, 52 (4), 545-549.

16. Adrian, M.; ten Heggeler-Bordier, B.; Wahli, W.; Stasiak, A. Z.; Stasiak, A.; Dubochet,

J., Direct visualization of supercoiled DNA molecules in solution. The EMBO Journal 1990, 9

(13), 4551.

17. Boles, T. C.; White, J. H.; Cozzarelli, N. R., Structure of plectonemically supercoiled

DNA. Journal of molecular biology 1990, 213 (4), 931-951.

18. Bednar, J.; Furrer, P.; Stasiak, A.; Dubochet, J.; Egelman, E. H.; Bates, A. D., The twist,

writhe and overall shape of supercoiled DNA change during counterion-induced transition from

a loosely to a tightly interwound superhelix: Possible implications for DNA structure in vivo.

Journal of Molecular Biology 1994, 235 (3), 825-847.

19. Lyubchenko, Y. L.; Shlyakhtenko, L. S., Visualization of supercoiled DNA with atomic

force microscopy in situ. Proceedings of the National Academy of Sciences 1997, 94 (2), 496-

501.

155

20. Lyubchenko, Y. L.; Shlyakhtenko, L. S.; Gall, A. A., Atomic force microscopy imaging

and probing of DNA, proteins, and protein-DNA complexes: silatrane surface chemistry. DNA-

Protein Interactions: Principles and Protocols, Third Edition 2009, 337-351.

21. Klenin, K. V.; Vologodskii, A. V.; Anshelevich, V. V.; Dykhne, A. M.; Frank-

Kamenetskii, M. D., Computer simulation of DNA supercoiling. Journal of MolecularBiology

1991, 217 (3), 413-419.

22. Vologodskii, A. V.; Cozzarelli, N. R., Conformational and thermodynamic properties of

supercoiled DNA. Annual Review of Biophysics and Biomolecular Structure 1994, 23 (1), 609-

643.

23. Vologodskii, A. V.; Levene, S. D.; Klenin, K. V.; Frank-Kamenetskii, M.; Cozzarelli, N.

R., Conformational and thermodynamic properties of supercoiled DNA. Journal of Molecular

Biology 1992, 227 (4), 1224-1243.

24. Hammermann, M.; Brun, N.; Klenin, K. V.; May, R.; Tóth, K.; Langowski, J., Salt-

dependent DNA superhelix diameter studied by small angle neutron scattering measurements

and Monte Carlo simulations. Biophysical Journal 1998, 75 (6), 3057-3063.

25. Rybenkov, V. V.; Vologodskii, A. V.; Cozzarelli, N. R., The effect of ionic conditions on

the conformations of supercoiled DNA. I. sedimentation analysis1. Journal of Molecular Biology

1997, 267 (2), 299-311.

26. Wasserman, S. A.; Cozzarelli, N. R., Supercoiled DNA-directed knotting by T4

topoisomerase. Journal of Biological Chemistry 1991, 266 (30), 20567-20573.

27. Higgins, C. F.; Dorman, C. J.; Stirling, D. A.; Waddell, L.; Booth, I. R.; May, G.;

Bremer, E., A physiological role for DNA supercoiling in the osmotic regulation of gene

expression in S. typhimurium and E. coli. Cell 1988, 52 (4), 569-584.

156

28. Pruss, G. J.; Drlica, K., DNA supercoiling and prokaryotic transcription. Cell 1989, 56

(4), 521-523.

29. Sousa, F.; Prazeres, D. M.; Queiroz, J. A., Improvement of transfection efficiency by

using supercoiled plasmid DNA purified with arginine affinity chromatography. The Journal of

Gene Medicine 2009, 11 (1), 79-88.

30. Lahijani, R.; Hulley, G.; Soriano, G.; Horn, N. A.; Marquet, M., High-yield production of

pBR322-derived plasmids intended for human gene therapy by employing a temperature-

controllable point mutation. Human Gene Therapy 1996, 7 (16), 1971-1980.

31. Gonçalves, G. A.; Bower, D. M.; Prazeres, D. M.; Monteiro, G. A.; Prather, K. L.,

Rational engineering of Escherichia coli strains for plasmid biopharmaceutical manufacturing.

Biotechnology Journal 2012, 7 (2), 251-261.

32. Ghanem, A.; Healey, R.; Adly, F. G., Current trends in separation of plasmid DNA

vaccines: a review. Anal Chim Acta 2013, 760, 1-15.

33. Prather, K. J.; Sagar, S.; Murphy, J.; Chartrain, M., Industrial scale production of plasmid

DNA for vaccine and gene therapy: plasmid design, production, and purification. Enzyme and

Microbial Technology 2003, 33 (7), 865-883.

34. Ferreira, G. N.; Monteiro, G. A.; Prazeres, D. M.; Cabral, J. M., Downstream processing

of plasmid DNA for gene therapy and DNA vaccine applications. Trends in Biotechnology 2000,

18 (9), 380-388.

35. Stadler, J.; Lemmens, R.; Nyhammar, T., Plasmid DNA purification. J Gene Med 2004, 6

Suppl 1, S54-66.

36. Carnes, A.; Williams, J., Plasmid DNA manufacturing technology. Recent Patents on

Biotechnology 2007, 1 (2), 151-166.

157

37. Latulippe, D. R.; Ager, K.; Zydney, A. L., Flux-dependent transmission of supercoiled

plasmid DNA through ultrafiltration membranes. Journal of Membrane Science 2007, 294 (1-2),

169-177.

38. Van Reis, R.; Zydney, A., Membrane separations in biotechnology. Current Opinion in

Biotechnology 2001, 12 (2), 208-211.

39. Eon-Duval, A.; MacDuff, R. H.; Fisher, C. A.; Harris, M. J.; Brook, C., Removal of RNA

impurities by tangential flow filtration in an RNase-free plasmid DNA purification process.

Analytical Biochemistry 2003, 316 (1), 66-73.

40. Kong, S.; Aucamp, J.; Titchener-Hooker, N. J., Studies on membrane sterile filtration of

plasmid DNA using an automated multiwell technique. Journal of Membrane Science 2010, 353

(1-2), 144-150.

41. Brochard, F.; De Gennes, P., Dynamics of confined polymer chains. The Journal of

Chemical Physics 1977, 67 (1), 52-56.

42. Daoudi, S.; Brochard, F., Flows of flexible polymer solutions in pores. Macromolecules

1978, 11 (4), 751-758.

43. Latulippe, D. R.; Zydney, A. L., Elongational flow model for transmission of supercoiled

plasmid DNA during membrane ultrafiltration. Journal of Membrane Science 2009, 329 (1-2),

201-208.

44. Latulippe, D. R.; Zydney, A. L., Salt-induced changes in plasmid DNA transmission

through ultrafiltration membranes. Biotechnol Bioeng 2008, 99 (2), 390-8.

45. Latulippe, D. R.; Zydney, A. L., Separation of plasmid DNA isoforms by highly

converging flow through small membrane pores. J Colloid Interface Sci 2011, 357 (2), 548-53.

158

46. Latulippe, D. R.; Zydney, A. L., Size exclusion chromatography of plasmid DNA

isoforms. J Chromatogr A 2009, 1216 (35), 6295-302.

47. van Reis, R.; Zydney, A., Bioprocess membrane technology. Journal of Membrane

Science 2007, 297 (1), 16-50.

48. Borujeni, E. E.; Zydney, A. L., Membrane fouling during ultrafiltration of plasmid DNA

through semipermeable membranes. Journal of Membrane Science 2014, 450, 189-196.

49. Singer, V. L.; Jones, L. J.; Yue, S. T.; Haugland, R. P., Characterization of PicoGreen

reagent and development of a fluorescence-based solution assay for double-stranded DNA

quantitation. Analytical Biochemistry 1997, 249 (2), 228-238.

50. Van Reis, R. D., Tangential flow filtration process and apparatus. U.S. Patent 5,256,294

1993.

51. Mochizuki, S.; Zydney, A. L., Theoretical analysis of pore size distribution effects on

membrane transport. Journal of Membrane Science 1993, 82 (3), 211-227.

52. van Reis, R.; Saksena, S., Optimization diagram for membrane separations. Journal of

Membrane Science 1997, 129 (1), 19-29.

53. Zydney, A. L.; Ho, C. C., Effect of membrane morphology on system capacity during

normal flow microfiltration. Biotechnology and Bioengineering 2003, 83 (5), 537-543.

54. Mochizuki, S.; Zydney, A. L., Effect of protein adsorption on the transport characteristics

of asymmetric ultrafiltration membranes. Biotechnology Progress 1992, 8 (6), 553-561.

55. Deen, W., Hindered transport of large molecules in liquid‐filled pores. AIChE Journal

1987, 33 (9), 1409-1425.

56. Rao, S.; Zydney, A. L., Controlling protein transport in ultrafiltration using small charged

ligands. Biotechnology and Bioengineering 2005, 91 (6), 733-742.

159

57. Zeman, L. J.; Zydney, A. L., Microfiltration and Ultrafiltration: Principles and

Applications. M. Dekker: 1996.

58. Rao, S.; Zydney, A. L., High resolution protein separations using affinity ultrafiltration

with small charged ligands. Journal of Membrane Science 2006, 280 (1), 781-789.

59. Molek, J. R.; Zydney, A. L., Separation of PEGylated α-lactalbumin from unreacted

precursors and byproducts using ultrafiltration. Biotechnology Progress 2007, 23 (6), 1417-1424.

60. Affandy, A.; Keshavarz-Moore, E.; Versteeg, H. K., Application of filtration blocking

models to describe fouling and transmission of large plasmids DNA in sterile filtration. Journal

of Membrane Science 2013, 437, 150-159.

61. Arkhangelsky, E.; Sefi, Y.; Hajaj, B.; Rothenberg, G.; Gitis, V., Kinetics and mechanism

of plasmid DNA penetration through nanopores. Journal of Membrane Science 2011, 371 (1),

45-51.

62. Dai, L.; Renner, C. B.; Doyle, P. S., Metastable tight knots in semiflexible chains.

Macromolecules 2014, 47 (17), 6135-6140.

63. Dai, L.; van der Maarel, J.; Doyle, P. S., Extended de Gennes regime of DNA confined in

a nanochannel. Macromolecules 2014, 47 (7), 2445-2450.

64. Dorfman, K. D.; King, S. B.; Olson, D. W.; Thomas, J. D.; Tree, D. R., Beyond gel

electrophoresis: microfluidic separations, fluorescence burst analysis, and DNA stretching. Chem

Rev 2013, 113 (4), 2584-667.

65. Cao, H.; Tegenfeldt, J. O.; Austin, R. H.; Chou, S. Y., Gradient nanostructures for

interfacing microfluidics and nanofluidics. Applied Physics Letters 2002, 81 (16), 3058-3060.

160

66. Cao, H.; Yu, Z.; Wang, J.; Tegenfeldt, J. O.; Austin, R. H.; Chen, E.; Wu, W.; Chou, S.

Y., Fabrication of 10 nm enclosed nanofluidic channels. Applied Physics Letters 2002, 81 (1),

174-176.

67. Teclemariam, N. P.; Beck, V. A.; Shaqfeh, E. S.; Muller, S. J., Dynamics of DNA

polymers in post arrays: Comparison of single molecule experiments and simulations.

Macromolecules 2007, 40 (10), 3848-3859.

68. Balducci, A.; Doyle, P., Conformational preconditioning by electrophoresis of DNA

through a finite obstacle array. Macromolecules 2008, 41 (14), 5485-5492.

69. Viero, Y.; He, Q.; Bancaud, A., Hydrodynamic manipulation of DNA in nanopost arrays:

Unhooking dynamics and size separation. Small 2011, 7 (24), 3508-3518.

70. Larson, J. W.; Yantz, G. R.; Zhong, Q.; Charnas, R.; D'Antoni, C. M.; Gallo, M. V.;

Gillis, K. A.; Neely, L. A.; Phillips, K. M.; Wong, G. G.; Gullans, S. R.; Gilmanshin, R., Single

DNA molecule stretching in sudden mixed shear and elongational microflows. Lab on a Chip

2006, 6 (9), 1187-99.

71. Wong, P. K.; Lee, Y.-K.; Ho, C.-M., Deformation of DNA molecules by hydrodynamic

focusing. Journal of Fluid Mechanics 2003, 497, 55-65.

72. Chen, Q.; Diao, S.; Wu, C., How does a supercoiled DNA chain pass through a small

conical glass pore? Soft Matter 2012, 8 (20), 5451-5458.

73. Perkins, T. T.; Quake, S. R.; Smith, D. E.; Chu, S., Relaxation of a single DNA molecule

observed by optical microscopy. Science-AAAS-Weekly Paper Edition-including Guide to

Scientific Information 1994, 264 (5160), 822-825.

74. Chan, E. Y.; Goncalves, N. M.; Haeusler, R. A.; Hatch, A. J.; Larson, J. W.; Maletta, A.

M.; Yantz, G. R.; Carstea, E. D.; Fuchs, M.; Wong, G. G., DNA mapping using microfluidic

161

stretching and single-molecule detection of fluorescent site-specific tags. Genome Research

2004, 14 (6), 1137-1146.

75. Latulippe, D. R.; Zydney, A. L., Radius of gyration of plasmid DNA isoforms from static

light scattering. Biotechnol Bioeng 2010, 107 (1), 134-42.

76. Borujeni, E. E.; Li, Y.; Zydney, A. L., Application of periodic backpulsing to reduce

membrane fouling during ultrafiltration of plasmid DNA. Journal of Membrane Science 2015,

473, 102-108.

77. Tegenfeldt, J. O.; Bakajin, O.; Chou, C.-F.; Chan, S. S.; Austin, R.; Fann, W.; Liou, L.;

Chan, E.; Duke, T.; Cox, E. C., Near-field scanner for moving molecules. Physical Review

Letters 2001, 86 (7), 1378.

78. Lau, W. J.; Ismail, A. F.; Misdan, N.; Kassim, M. A., A recent progress in thin film

composite membrane: A review. Desalination 2012, 287, 190-199.

79. van Reis, R.; Zydney, A., Bioprocess membrane technology. Journal of Membrane

Science 2007, 297 (1-2), 16-50.

80. Cheryan, M., Ultrafiltration and Microfiltration Handbook. CRC press: 1998.

81. Harrell, C. C.; Siwy, Z. S.; Martin, C. R., Conical nanopore membranes: Controlling the

nanopore shape. Small 2006, 2 (2), 194-8.

82. Scopece, P.; Baker, L. A.; Ugo, P.; Martin, C. R., Conical nanopore membranes: Solvent

shaping of nanopores. Nanotechnology 2006, 17 (15), 3951-3956.

83. Apel, P. Y.; Korchev, Y. E.; Siwy, Z.; Spohr, R.; Yoshida, M., Diode-like single-ion

track membrane prepared by electro-stopping. Nuclear Instruments and Methods in Physics

Research Section B: Beam Interactions with Materials and Atoms 2001, 184 (3), 337-346.

162

84. Ho, C.-C.; Zydney, A. L., Effect of membrane morphology on the initial rate of protein

fouling during microfiltration. Journal of Membrane Science 1999, 155 (2), 261-275.

85. Ho, C.-C.; Zydney, A. L., Protein fouling of asymmetric and composite microfiltration

membranes. Industrial & Engineering ChemistryRresearch 2001, 40 (5), 1412-1421.

86. Ho, C.-C.; Zydney, A. L., Measurement of membrane pore interconnectivity. Journal of

Membrane Science 2000, 170 (1), 101-112.

87. Spohr, R.; Apel, Y. P.; Korchev, Y.; Siwy, Z.; Yoshida, M., Method for etching at least

one ion track to a pore in a membrane and electrolyte cell for preparing the membrane. U.S.

Patent 7,001,501, 2006.

88. Smith, D. E.; Babcock, H. P.; Chu, S., Single-polymer dynamics in steady shear flow.

Science 1999, 283 (5408), 1724-1727.

89. Randall, G. C.; Doyle, P. S., DNA deformation in electric fields: DNA driven past a

cylindrical obstruction. Macromolecules 2005, 38 (6), 2410-2418.

90. Zimm, B. H., Dynamics of polymer molecules in dilute solution: viscoelasticity, flow

birefringence and dielectric loss. The Journal of Chemical Physics 1956, 24 (2), 269-278.

91. Smith, D. E.; Chu, S., Response of flexible polymers to a sudden elongational flow.

Science 1998, 281 (5381), 1335-1340.

92. Larson, R., The role of molecular folds and ‘pre-conditioning’in the unraveling of

polymer molecules during extensional flow. Journal of Non-Newtonian Fluid Mechanics 2000,

94 (1), 37-45.

93. Perkins, T. T.; Smith, D. E.; Chu, S., Single polymer dynamics in an elongational flow.

Science 1997, 276 (5321), 2016-2021.

163

94. Hur, J. S.; Shaqfeh, E. S.; Larson, R. G., Brownian dynamics simulations of single DNA

molecules in shear flow. Journal of Rheology 2000, 44 (4), 713-742.

95. Larson, R.; Perkins, T.; Smith, D.; Chu, S., Hydrodynamics of a DNA molecule in a flow

field. Physical Review E 1997, 55 (2), 1794.

96. Babcock, H. P.; Teixeira, R. E.; Hur, J. S.; Shaqfeh, E. S.; Chu, S., Visualization of

molecular fluctuations near the critical point of the coil− stretch transition in polymer elongation.

Macromolecules 2003, 36 (12), 4544-4548.

97. Teixeira, R. E.; Babcock, H. P.; Shaqfeh, E. S.; Chu, S., Shear thinning and tumbling

dynamics of single polymers in the flow-gradient plane. Macromolecules 2005, 38 (2), 581-592.

98. Harrell, C. C.; Lee, S. B.; Martin, C. R., Synthetic single-nanopore and nanotube

membranes. Analytical Chemistry 2003, 75 (24), 6861-6867.

99. Borochov, N.; Eisenberg, H.; Kam, Z., Dependence of DNA conformation on the

concentration of salt. Biopolymers 1981, 20 (1), 231-235.

100. Cherny, D. I.; Jovin, T. M., Electron and scanning force microscopy studies of alterations

in supercoiled DNA tertiary structure. Journal of Molecular Biology 2001, 313 (2), 295-307.

101. Rybenkov, V. V.; Vologodskii, A. V.; Cozzarelli, N. R., The effect of ionic conditions on

DNA helical repeat, effective diameter and free energy of supercoiling. Nucleic Acids Research

1997, 25 (7), 1412-1418.

102. Kong, S.; Titchener-Hooker, N.; Levy, M. S., Plasmid DNA processing for gene therapy

and vaccination: Studies on the membrane sterilisation filtration step. Journal of Membrane

Science 2006, 280 (1-2), 824-831.

164

103. Li, Y.; Borujeni, E. E.; Zydney, A. L., Use of preconditioning to control membrane

fouling and enhance performance during ultrafiltration of plasmid DNA. Journal of Membrane

Science 2015, 479, 117-122.

104. Tweedie, J. W.; Stowell, K. M., Quantification of DNA by agarose gel electrophoresis

and analysis of the topoisomers of plasmid and M13 DNA following treatment with a restriction

endonuclease or DNA topoisomerase I. Biochemistry and Molecular Biology Education 2005, 33

(1), 28-33.

105. Hammermann, M.; Steinmaier, C.; Merlitz, H.; Kapp, U.; Waldeck, W.; Chirico, G.;

Langowski, J., Salt effects on the structure and internal dynamics of superhelical DNAs studied

by light scattering and Brownian dynamics. Biophysical Journal 1997, 73 (5), 2674-2687.

106. Hagerman, P. J., Flexibility of DNA. Annual Review of Biophysics and Biophysical

Chemistry 1988, 17 (1), 265-286.

107. Shlyakhtenko, L. S.; Miloseska, L.; Potaman, V. N.; Sinden, R. R.; Lyubchenko, Y. L.,

Intersegmental interactions in supercoiled DNA: atomic force microscope study.

Ultramicroscopy 2003, 97 (1), 263-270.

108. Rybenkov, V. V.; Cozzarelli, N. R.; Vologodskii, A. V., Probability of DNA knotting and

the effective diameter of the DNA double helix. Proceedings of the National Academy of

Sciences 1993, 90 (11), 5307-5311.

109. Rybenkov, V. V.; Vologodskii, A. V.; Cozzarelli, N. R., The effect of ionic conditions on

the conformations of supercoiled DNA. I. Sedimentation analysis. Journal of Molecular Biology

1997, 267 (2), 299-311.

165

110. Rybenkov, V. V.; Vologodskii, A. V.; Cozzarelli, N. R., The effect of ionic conditions on

the conformations of supercoiled DNA. II. Equilibrium catenation. Journal of Molecular Biology

1997, 267 (2), 312-323.

111. Stigter, D., Interactions of highly charged colloidal cylinders with applications to double‐

stranded DNA. Biopolymers 1977, 16 (7), 1435-1448.

112. Kusano, T.; Berberich, T.; Tateda, C.; Takahashi, Y., Polyamines: Essential factors for

growth and survival. Planta 2008, 228 (3), 367-381.

113. Peng, H. F.; Jackson, V., In vitro studies on the maintenance of transcription-induced

stress by histones and polyamines. Journal of Biological Chemistry 2000, 275 (1), 657-668.

114. Sato, Y. T.; Hamada, T.; Kubo, K.; Yamada, A.; Kishida, T.; Mazda, O.; Yoshikawa, K.,

Folding transition into a loosely collapsed state in plasmid DNA as revealed by single-molecule

observation. FEBS Lett 2005, 579 (14), 3095-9.

115. Ma, C.; Bloomfield, V. A., Condensation of supercoiled DNA induced by MnCl2.

Biophysical Journal 1994, 67 (4), 1678-1681.

116. Tabor, H., The protective effect of spermine and other polyamines against heat

denaturation of deoxyribonucleic acid. Biochemistry 1962, 1 (3), 496-501.

117. Ha, H. C.; Sirisoma, N. S.; Kuppusamy, P.; Zweier, J. L.; Woster, P. M.; Casero, R. A.,

The natural polyamine spermine functions directly as a free radical scavenger. Proceedings of

the National Academy of Sciences 1998, 95 (19), 11140-11145.

118. Ouameur, A. A.; Tajmir-Riahi, H.-A., Structural analysis of DNA interactions with

biogenic polyamines and cobalt (III) hexamine studied by Fourier transform infrared and

capillary electrophoresis. Journal of Biological Chemistry 2004, 279 (40), 42041-42054.

166

119. Jain, S.; Zon, G.; Sundaralingam, M., Base only binding of spermine in the deep groove

of the A-DNA octamer d (GTGTACAC). Biochemistry 1989, 28 (6), 2360-2364.

120. Tari, L. W.; Secco, A. S., Base-pair opening and spermine binding—B-DNA features

displayed in the crystal structure of a gal operon fragment: implications for protein-DNA

recognition. Nucleic Acids Research 1995, 23 (11), 2065-2073.

121. Shao, Q.; Goyal, S.; Finzi, L.; Dunlap, D., Physiological levels of salt and polyamines

favor writhe and limit twist in DNA. Macromolecules 2012, 45 (7), 3188-3196.

122. Murphy, J. C.; Fox, G. E.; Willson, R. C., Enhancement of anion-exchange

chromatography of DNA using compaction agents. Journal of Chromatography A 2003, 984 (2),

215-221.

123. Wilson, R. W.; Bloomfield, V. A., Counterion-induced condensation of deoxyribonucleic

acid. A light-scattering study. Biochemistry 1979, 18 (11), 2192-2196.

124. Manning, G. S., The molecular theory of polyelectrolyte solutions with applications to

the electrostatic properties of polynucleotides. Quarterly Reviews of Biophysics 1978, 11 (02),

179-246.

125. Bancroft, D.; Williams, L. D.; Rich, A.; Egli, M., The low-temperature crystal structure

of the pure-spermine form of Z-DNA reveals binding of a spermine molecule in the minor

groove. Biochemistry 1994, 33 (5), 1073-1086.

126. Chattoraj, D. K.; Gosule, L. C.; Schellman, J. A., DNA condensation with polyamines: II.

Electron microscopic studies. Journal of Molecular Biology 1978, 121 (3), 327-337.

127. Lin, Z.; Wang, C.; Feng, X.; Liu, M.; Li, J.; Bai, C., The observation of the local ordering

characteristics of spermidine-condensed DNA: atomic force microscopy and polarizing

microscopy studies. Nucleic Acids Research 1998, 26 (13), 3228-3234.

167

128. Vijayanathan, V.; Thomas, T.; Shirahata, A.; Thomas, T., DNA condensation by

polyamines: A laser light scattering study of structural effects. Biochemistry 2001, 40 (45),

13644-13651.

129. Gosule, L. C.; Schellman, J. A., Compact form of DNA induced by spermidine. Nature

1976, 259 (5541), 333-335.

130. Hoopes, B. C.; McClure, W. R., Studies on the selectivity of DNA precipitation by

spermine. Nucleic Acids Research 1981, 9 (20), 5493-5504.

131. Toma, A. C.; de Frutos, M.; Livolant, F.; Raspaud, E., DNA condensed by protamine: a

"short" or "long" polycation behavior. Biomacromolecules 2009, 10 (8), 2129-34.

132. Kahn, D. W.; Butler, M. D.; Cohen, D. L.; Gordon, M.; Kahn, J. W.; Winkler, M. E.,

Purification of plasmid DNA by tangential flow filtration. Biotechnology and Bioengineering

2000, 69 (1), 101-106.

133. Chan, R.; Chen, V., Characterization of protein fouling on membranes: opportunities and

challenges. Journal of Membrane Science 2004, 242 (1), 169-188.

134. Hadidi, M.; Zydney, A. L., Fouling behavior of zwitterionic membranes: Impact of

electrostatic and hydrophobic interactions. Journal of Membrane Science 2014, 452, 97-103.

135. Palecek, S. P.; Zydney, A. L., Intermolecular electrostatic interactions and their effect on

flux and protein deposition during protein filtration. Biotechnology Progress 1994, 10 (2), 207-

213.

136. Hadidi, M.; Buckley, J. J.; Zydney, A. L., Ultrafiltration behavior of bacterial

polysaccharides used in vaccines. Journal of Membrane Science 2015, 490, 294-300.

137. Li, Y.; Butler, N.; Zydney, A. L., Size-based separation of supercoiled plasmid DNA

using ultrafiltration. J Colloid Interface Sci 2016, 472, 195-201.

168

138. Meyers, J. A.; Sanchez, D.; Elwell, L. P.; Falkow, S., Simple agarose gel electrophoretic

method for the identification and characterization of plasmid deoxyribonucleic acid. Journal of

Bacteriology 1976, 127 (3), 1529-1537.

139. Viovy, J.-L., Electrophoresis of DNA and other polyelectrolytes: Physical mechanisms.

Reviews of Modern Physics 2000, 72 (3), 813.

140. Clewell, D. B.; Helinski, D. R., Supercoiled circular DNA-protein complex in

Escherichia coli: purification and induced conversion to an open circular DNA form.

Proceedings of the National Academy of Sciences 1969, 62 (4), 1159-1166.

141. Diogo, M.; Queiroz, J.; Prazeres, D., Chromatography of plasmid DNA. Journal of

Chromatography A 2005, 1069 (1), 3-22.

142. Rathore, A.; Velayudhan, A., Scale-up and optimization in preparative chromatography:

principles and biopharmaceutical applications. CRC Press: 2002; Vol. 88.

143. Prazeres, D. M.; Ferreira, G. N.; Monteiro, G. A.; Cooney, C. L.; Cabral, J. M., Large-

scale production of pharmaceutical-grade plasmid DNA for gene therapy: problems and

bottlenecks. Trends in Biotechnology 1999, 17 (4), 169-174.

144. McClung, J. K.; Gonzales, R. A., Purification of plasmid DNA by fast protein liquid

chromatography on Superose 6 preparative grade. Analytical Biochemistry 1989, 177 (2), 378-

382.

145. Raymond, G. J.; Bryant, P. K.; Nelson, A.; Johnson, J. D., Large-scale isolation of

covalently closed circular DNA using gel filtration chromatography. Analytical Biochemistry

1988, 173 (1), 125-133.

146. Kurnik, R. T.; Yu, A. W.; Blank, G. S.; Burton, A. R.; Smith, D.; Athalye, A. M.; van

Reis, R., Buffer exchange using size exclusion chromatography, countercurrent dialysis, and

169

tangential flow filtration: Models, development, and industrial application. Biotechnology and

Bioengineering 1995, 45 (2), 149-157.

147. van Reis, R.; Gadam, S.; Frautschy, L. N.; Orlando, S.; Goodrich, E. M.; Saksena, S.;

Kuriyel, R.; Simpson, C. M.; Pearl, S.; Zydney, A. L., High performance tangential flow

filtration. Biotechnology and Bioengineering 1997, 56 (1), 71-82.

148. Sakaue, T.; Raphaël, E.; De Gennes, P.-G.; Brochard-Wyart, F., Flow injection of

branched polymers inside nanopores. EPL (Europhysics Letters) 2005, 72 (1), 83.

149. Ding, M.; Duan, X.; Lu, Y.; Shi, T., Flow-induced ring polymer ranslocation through

nanopores. Macromolecules 2015, 48 (16), 6002-6007.

150. Freed, K. F.; Wu, C., Comparison of calculated and measured critical flow rates for

dragging linear polymer chains through a small cylindrical ube. Macromolecules 2011, 44 (24),

9863-9866.

151. Lu, M.; Guo, Q.; Marky, L. A.; Seeman, N. C.; Kallenbach, N. R., Thermodynamics of

DNA branching. Journal of Molecular Biology 1992, 223 (3), 781-789.

152. Marko, J. F.; Siggia, E. D., Statistical mechanics of supercoiled DNA. Physical Review E

1995, 52 (3), 2912-2938.

153. Fathizadeh, A.; Schiessel, H.; Ejtehadi, M. R., Molecular dynamics simulation of

supercoiled DNA rings. Macromolecules 2015, 48 (1), 164-172.

154. Wu, C.; Li, L., Unified description of transportation of polymer chains with different

topologies through a small cylindrical pore. Polymer 2013, 54 (5), 1463-1465.

155. Li, L.; He, C.; He, W.; Wu, C., How does a hyperbranched chain pass through a

nanopore? Macromolecules 2012, 45 (18), 7583-7589.

170

156. Ge, H.; Pispas, S.; Wu, C., How does a star chain (nanooctopus) crawl through a

nanopore? Polymer Chemistry 2011, 2 (5), 1071-1076.

157. Ge, H.; Wu, C., Separation of linear and star chains by a nanopore. Macromolecules

2010, 43 (21), 8711-8713.

158. Borujeni, E. E.; Zydney, A. L., Separation of plasmid DNA isoforms using centrifugal

ultrafiltration. Biotechniques 2012, 53 (1), 49-56.

159. van Reis, R.; Goodrich, E. M.; Yson, C. L.; Frautschy, L. N.; Dzengeleski, S.; Lutz, H.,

Linear scale ultrafiltration. Biotechnology and Bioengineering 1997, 55 (5), 737-746.

160. Kole, R.; Krainer, A. R.; Altman, S., RNA therapeutics: beyond RNA interference and

antisense oligonucleotides. Nat Rev Drug Discov 2012, 11 (2), 125-40.

161. Burnett, J. C.; Rossi, J. J., RNA-based therapeutics: Current progress and future

prospects. Chem Biol 2012, 19 (1), 60-71.

162. Chery, J.; Näär, A., RNA therapeutics: RNAi and antisense mechanisms and clinical

applications. Postdoc Journal: A Journal of Postdoctoral Research andPpostdoctoral Affairs

2016, 4 (7), 35.

163. Kasianowicz, J. J.; Brandin, E.; Branton, D.; Deamer, D. W., Characterization of

individual polynucleotide molecules using a membrane channel. Proceedings of the National

Academy of Sciences 1996, 93 (24), 13770-13773.

164. Mathé, J.; Aksimentiev, A.; Nelson, D. R.; Schulten, K.; Meller, A., Orientation

discrimination of single-stranded DNA inside the α-hemolysin membrane channel. Proceedings

of the National Academy of Sciences of the United States of America 2005, 102 (35), 12377-

12382.

171

165. Meller, A.; Nivon, L.; Branton, D., Voltage-driven DNA translocations through a

nanopore. Phys Rev Lett 2001, 86 (15), 3435-8.

166. Meller, A., Dynamics of polynucleotide transport through nanometre-scale pores. Journal

of Physics: Condensed Matter 2003, 15 (17), R581.

167. Li, J.; Gershow, M.; Stein, D.; Brandin, E.; Golovchenko, J. A., DNA molecules and

configurations in a solid-state nanopore microscope. Nature Materials 2003, 2 (9), 611-615.

168. Storm, A. J.; Storm, C.; Chen, J.; Zandbergen, H.; Joanny, J.-F.; Dekker, C., Fast DNA

translocation through a solid-state nanopore. Nano Letters 2005, 5 (7), 1193-1197.

169. Wanunu, M.; Sutin, J.; McNally, B.; Chow, A.; Meller, A., DNA translocation governed

by interactions with solid-state nanopores. Biophysical Journal 2008, 95 (10), 4716-4725.

170. Levy, S. L.; Craighead, H. G., DNA manipulation, sorting, and mapping in nanofluidic

systems. Chem Soc Rev 2010, 39 (3), 1133-52.

VITA Ying Li

EDUCATION

The Pennsylvania State University, Chemical Engineering, PhD Candidate Aug. 2012-present

The National University of Singapore, Chemical Engineering, B.S (1st Class Honor) May 2012

AWARDS AND HONORS

MilliporeSigma Life Science Award in Bioseparations (2016)

Air Products Fellowship (2016)

American Institute of Chemical Engineers (AIChE) Separations Division Graduate Student Research

Award (2015)

American Chemical Society (ACS) Biochemical Technology (BIOT) Travel Award (2016)

North America Membrane Society (NAMS) Student Travel Award (2014)

Dean’s List, Department of Chemical Engineering, National University of Singapore

Singapore MIT Alliance Research Fellowship (2010)

PUBLICATIONS

Li, Ying, Ehsan Espah Borujeni, and Andrew L. Zydney. "Use of preconditioning to control membrane

fouling and enhance performance during ultrafiltration of plasmid DNA." Journal of Membrane Science

479 (2015): 117-122.

Borujeni Ehsan Espah, Ying Li, and Andrew L. Zydney. "Application of periodic backpulsing to reduce

membrane fouling during ultrafiltration of plasmid DNA." Journal of Membrane Science 473 (2015): 102-

108.

Li, Ying, David Currie, and Andrew L. Zydney. "Enhanced purification of plasmid DNA isoforms by

exploiting ionic strength effects during ultrafiltration." Biotechnology and Bioengineering 113.4 (2016):

783-789. – featured as Spotlight.

Li, Ying, Neil Butler, and Andrew L. Zydney. “Size based separation of supercoiled plasmid DNA using

ultrafiltration.” Journal of Colloid and Interface Science (2016): 195-201.

Li, Ying, Kuangzheng Zhu, and Andrew L. Zydney. “Effects of ionic conditions on membrane fouling

during ultrafitlration of plasmid DNA”. Separation and Purification Technology (2017):176, 287-293

Anirudh Nambiar, Ying Li and Andrew L. Zydney. “Countercurrent Staged Diafiltration for Formulation

of High Value Proteins. ” Submitted to Biotechnology and Bioengineering (2017)

CONFERENCE PRESENTATIONS

Ying Li, Ehsan Espah Borujeni, and Andrew L. Zydney. 24th Annual North American Membrane Society

Meeting, Houston, TX, June 2014 (presentation and poster)

Ying Li, Ehsan Espah Borujeni, and Andrew L. Zydney. Membranes: Materials & Processes, Gordon

Research Seminar, New London, NH, July 2014 (poster)

Ying Li and Andrew L. Zydney. 2015 AIChE annual meeting, Salt Lake City, UT, November 2015 (2

presentations) – featured as Best Presentation in the session.

Ying Li and Andrew L. Zydney. 251st ACS annual meeting, San Diego, CA, March 2016 (presentation).

Ying Li and Andrew L. Zydney. 2016 AIChE annual meeting, San Francisco, November 2016

(presentation).

Ying Li, Anirudh Nambiar, and Andrew L. Zydney. 252nd ACS annual meeting, San Francisco, CA, April

2017 (presentation).