Development of One-Step Single-Cell RT-PCR for the Massively Parallel Detection of Gene Expression by Yuan Gong M.S

Development of One-Step Single-Cell RT-PCR for the Massively Parallel Detection of Gene Expression by Yuan Gong M.S. Chemical Engineering Practice Massachusetts Institute of Technology, 2009 B.S. Chemical Engineering and Applied and Computational Mathematics California Institute of Technology, 2007

SUBMITTED TO THE DEPARTMENT OF CHEMICAL ENGINEERING IN PARTIAL FULLFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY IN CHEMICAL ENGINEERING AT THE MASSACHUSETTS INSTITUTE OF TECHNOLOGY JUNE 2014

Signature of Author: ______Yuan Gong Department of Chemical Engineering May 21, 2014

Certified by: ______J. Christopher Love Associate Professor of Chemical Engineering Thesis Supervisor

Accepted by: ______Patrick S. Doyle Professor of Chemical Engineering Chairman, Committee for Graduate Students

Development of One-Step Single-Cell RT-PCR for the Massively Parallel Detection of Gene Expression

Yuan Gong

Submitted to the Department of Chemical Engineering on May 21, 2014 in Partial Fulfillment of the Requirements for the degree of Doctor of Philosophy (Ph.D.) in Chemical Engineering

ABSTRACT

The United Nations estimates that over 35 million people are afflicted with HIV/AIDS in the world. Highly active antiretroviral treatments (HAART) that use a combination of drugs that target the virus at different stages of its life cycle are effective at reducing the HIV plasma levels below levels detectable by the most sensitive assays. However, upon termination of HAART, HIV RNA transcripts are measurable in the blood after 2-3 weeks. This relapse is attributed to the presence of a reservoir of latently infected cells, such as resting CD4+ T-cells. The latent reservoir in resting memory CD4+ T-cells has been estimated to decay with a half-life of as long as 44 months, thus hindering the eradication of HIV. Current knowledge of latent reservoirs came from the isolation of possible reservoir populations by cell surface markers and querying each population for the presence of HIV RNA. These measurements do not have single cell resolution so the exact frequencies of latently infected cells are not known.

In this thesis, we developed and optimized a method to detect cellular transcripts of single cells in an array of nanowells. The limit of detection of the assay was approximately 1.4 copies of DNA in a 125 pL well (18.6 fM) with a false positive rate as low as 4.6x10-5. Combining this assay along with image-based cytometry and microengraving, we generated a multivariate dataset on single cells to understand the relationships between cell phenotype, transcribed genes, and secreted products. We showed that gene expression could not be a surrogate measure for antibody secretion. We were also able to detect rare cells in a population at a frequency as low as 1 in 10,000. We then applied the technology to samples from a patient on HAART for more than 1.5 years. We were able to detect an infection rate of 1:3000 cells that had low levels of HIV RNA in bulk.

Thesis supervisor: J. Christopher Love Title: Associate Professor

Acknowledgments

I would like to thank my advisor, Professor J. Christopher Love, for accepting me in his group. He provided me with enjoyable problems to solve and a lot of guidance and creative ideas on new directions to overcome the many challenges in this work. I also greatly appreciate the advice he gave me during Practice School on how to manage and direct a team of co-workers towards a common goal. I hope I was as good a student to him as he was a mentor to me!

I would like to thank my thesis committee members, Professor K. Dane Wittrup, Professor Darrell Irvine, and Professor Bruce Walker, for all the feedback and advice they gave me to improve this thesis. I also want to thank my collaborators, Dr. Xu Yu and Dr. Maria J. Buzon, for sharing their knowledge and expertise on HIV latency with me. They also patiently waited for me as I optimized the technology to detect HIV in single cells. I am extremely grateful to Aaron Gawlik and Alan Stockdale for designing and building the RT-PCR machine.

It has been a great pleasure to work with all current and past members of the Love Lab: Dr. Adebola Ogunniyi, Dr. Qing Han, Dr. Jonghoon Choi, Dr. Yvonne Yamanaka, Dr. Todd Gierahn, Dr. Rita Lucia Contento, Timothy Politano, Denis Loginov, Dr. Ayca Yalcin Ozkumur, Dr. Qing Song, Dr. Eliseo Papa, Dr. Kerry Love, Vasiliki Panagiotou, Dr. Navin Varadarajan, Dr. Bin Jia, Dr. Sangram Bagh, Dr. Alexis Torres, Viktor Adalsteinsson, Brittany Thomas, Lionel Lam, Abby Hill, Sarah Schrier, Kimberly Ohn, Thomas Douce, Rachel Barry, Dr. Joe Couto, Dr. Konstantinos Tsioris, Dr. Lilun Ho, Dr. Kartik Shah, Dr. John Ballew, Nicholas Mozdzierz, Narmin Tahirova, Ross Zimnisky, John Clark, and Rachel Leeson. Thank you for all of your helpful suggestions and ideas to my project and for creating a fun and enjoyable lab environment!

Finally, I would like to thank my family and friends for supporting me for the past seven years. Your encouragement and love have kept me going throughout my graduate school experience.

Table of Contents

List of Figures ...... 9 List of Tables ...... 11 Introduction ...... 13 1.1. Human immunodeficiency virus ...... 13 1.2. Existing tools to detect HIV-infected cells ...... 14 1.3. Amplification and detection techniques ...... 16 1.4. Objectives and outline of thesis ...... 18 Chapter 2. Materials and Methods ...... 19 2.1. Cell line culture ...... 19 2.2. Fabrication of array of nanowells...... 20 2.3. Cytometry and imaging ...... 21 2.4. One-step reverse transcription polymerase chain reaction (RT-PCR) ...... 22 2.4.1. Primer and TaqMan probe selection ...... 22 2.4.2. Imaging end-point RT-PCR signal ...... 23 2.4.3. Quantitative TaqMan RT-PCR ...... 25 2.4.4. Digital PCR in nanowells ...... 25 2.5. Microengraving ...... 25 2.6. Surface capture of transcripts ...... 26 2.7. Hybridization chain reaction ...... 27 2.8. Data Analysis ...... 28 Chapter 3. Establishing one-step RT-PCR in nanowells ...... 31 3.1. Optimization of cell lysis ...... 31 3.2. Optimization of RT-PCR in nanowells ...... 38 3.3. Optimization of pre-treatment of cells ...... 39 3.4. Optimization of thermocycling ...... 41 3.4. Discussion ...... 46 3.4.1. Limit of detection of transcripts ...... 46 3.4.2. Evaporation ...... 48 3.4.3. Limitations ...... 50 Chapter 4. Characterization of RT-PCR in nanowells ...... 53 4.1. Sensitivity and specificity ...... 53 7

4.2. Integration with microengraving ...... 57 Chapter 5. Identification of target cells in large populations ...... 59 5.1. Cytometry ...... 59 5.2. Activation of cells ...... 62 5.3. Limit of detection of cells ...... 66 5.4. HIV-positive patient sample...... 67 5.5. Discussion ...... 69 Chapter 6. Other methods to detect transcript ...... 73 6.1. Surface capture of transcripts ...... 73 6.2. Hybridization chain reaction ...... 78 6.3. Discussion ...... 81 Chapter 7. Conclusions ...... 83 References ...... 87

List of Figures

Figure 3.1. TaqMan probe mechanism ...... 31

Figure 3.2. Determining the RT-qPCR kit ...... 32

Figure 3.3. Comparison between 1-step Fast qScript with ROX and MGB ...... 35

Figure 3.4. The effect of Tween-20 on the RT-qPCR reaction on 100 cells ...... 35

Figure 3.5. The effect of NP-40 on the RT-qPCR reaction on 100 cells ...... 36

Figure 3.6. Direct comparison between Tween-20 and NP-40 ...... 36

Figure 3.7. Efficiency of RT-qPCR with the addition of 0.05% NP-40 or 0.5% Tween-20 ...... 37

Figure 3.8. Effect of SDS on RT-qPCR on 100 cells ...... 37 Figure 3.9. Temperature model for an increase in temperature from 60 °C to 95 °C after 5 seconds ...... 39

Figure 3.10. Removal of false positives by RNase, DNase-free treatment ...... 40

Figure 3.11. Sample images of digital PCR on a serial dilution of HIVgag DNA ...... 44

Figure 3.12. Limit of detection for PCR ...... 44

Figure 3.13. Various cycle numbers for PCR on HIVgag DNA ...... 45 Figure 3.14. Detection of B2M mRNA transcripts on beads from bulk cellular mRNA extraction ...... 47

Figure 3.15. Detection on B2M mRNA from bulk cellular mRNA extraction ...... 47

Figure 3.16. Effect of cell lysate on RT-PCR ...... 48

Figure 4.1. Schematic of method for parallel single-cell RT-PCR reactions in nanowells ...... 55

Figure 4.2. Detection of mRNA transcripts of constitutively expressed genes in 4D20 cells ...... 56 Figure 4.3. Integrated single-cell analysis of gene expression and secreted antibodies from human B cell hybridomas ...... 57

Figure 5.1. Histograms of surface marker fluorescence on PBMCs ...... 61

Figure 5.2. Comparison between flow cytometry and microscopy for T-cell classification ...... 61

Figure 5.3. Effect of 10x activation conditions on HIVgag and cell viability in ACH2 cells ...... 62

Figure 5.4. Comparison of 1x TNFa and 10x TNFa activations on ACH2 ...... 63

Figure 5.5. Comparison of 1x TNFa and 10x TNFa activations on ACH2 in nanowells ...... 65

Figure 5.6. Effect of activation time on ACH2 in nanowells ...... 65

Figure 5.7. RT-qPCR on bulk cells from HIV-positive sample ...... 69

Figure 6.1. Schematic of mRNA capture on a glass surface ...... 74

Figure 6.2. Sample scans of B2M cDNA (FAM, left) and KanR cDNA (HEX, right) ...... 75

Figure 6.3. Images of swapping dyes in two sets of probes ...... 77

Figure 6.4. Captured cDNA on glass from unlabeled mutPGK1 ...... 77

Figure 6.5. Comparison of HCR with direct detection in 4D20 cells ...... 79

Figure 6.6. Combination of two sets of HCR to detect single nucleotide polymorphism ...... 79

Figure 6.7. Capture of ACTB on PDMS followed by RT-PCR in nanowells ...... 81

List of Tables

Table 2.1 Primer and probe design for RT-PCR...... 22

Table 2.2 Sequences of initiator and hairpin for HCR...... 28

Table 3.1 List of tested RT-PCR lysis conditions...... 41

Table 5.1 Surface markers for PBMC classification ...... 59

Table 5.2 Effect of activation conditions on ACH2 cell detection by RT-PCR in nanowells ...... 64

Table 5.3 Detection of serial dilutions of ACH2 cells in nanowells ...... 66

Table 5.4 RT-PCR on HIV-positive T-cells ...... 69

Introduction

1.1. Human immunodeficiency virus

The United Nations estimates that over 35 million people are afflicted with HIV/AIDS in

the world and almost 1% of the world’s supposedly healthy population (ages 15-49) is infected1.

With the availability of drugs and more awareness on the transmission of the disease, the number of deaths caused by HIV and the number of new infections have dropped over the past decade1.

Highly active antiretroviral treatments (HAART) that use a combination of drugs to target the

virus at different stages of its life cycle are effective at reducing the HIV plasma levels below

levels detectable by the most sensitive clinical assays available (limit of detection of 50

copies/mL) in 3-4 weeks. While HAART is very effective, it is expensive and has known side

effects2, 3. The virus is also known to develop resistance if HAART regimen is not strictly followed4. However, upon termination of HAART, HIV RNA transcripts are measurable in the

blood after 2-3 weeks5-8, demonstrating that HAART is not curative. This relapse is attributed to

the presence of a reservoir of latently infected cells, such as resting CD4+ T-cells9, 10, monocytes

and dendritic cells11, that are not responsive to HAART. Typically, activated CD4+ T-cells that are infected will undergo apoptosis, but in latently infected cells, the life cycle of the virus is interrupted by cellular factors, such as the histone deacetylation and methylation of HIV long terminal repeat (LTR)12. The latent reservoir in resting memory CD4+ T-cells has been estimated to decay with a half-life of as long as 44 months13, thus maintaining a long-lived latently infected

population and hindering the eradication of HIV14. Therefore, research on the identification and

eradication of these latent reservoirs has been pursued as a strategy in HIV treatment. Gene

therapy using Tre recombinase, which is an evolved Cre recombinase, has been shown to excise proviral DNA from the integrated host genome15. Its effect on latently infected cells have not been tested yet16. Recently, the discovery of a population of replication competent proviruses that were not induced by potent reactivation conditions in vitro further complicates the eradication of latently positive cells17.

The gold standard for studying latency is the extraction of resting CD4+ T-cells from a person on HAART. However, the frequency of these latently infected cells are 1 in a million18, so deep mechanistic studies would not be feasible on only these cells. Latently infected primary

CD4+ T-cells can be made in a variety of methods19, 20, but these cells often require a long time to culture19, 20. Latently infected cell lines such as ACH2 do exist and are commonly used to model the phenomenon21. While cell lines are convenient to work with, some biologically relevant limitations to their use exist, such as the cell line is a clonal population with the same integration site, the integration sites are often in transcriptionally inactive regions of the genome22 while the integration sites in resting CD4+ T-cells were in actively transcribed regions23-25, and the cell line grows very quickly whereas latently infected T-cells are resting in vivo18.

1.2. Existing tools to detect HIV-infected cells

The standard method to measure the size of the latent reservoir is a viral outgrowth assay26, 27. Latent infections are identified using a population of highly purified resting T-cells28.

These cells are often taken from patients on HAART since HIV levels in their blood are below the level of detection. The purification process removes, by flow cytometry, cells with markers of various stages of activation such as CD69, CD25, and HLA-DR. To demonstrate that latent 14

infections exist in the population, the purified cells are stimulated with phytohemagglutinin

(PHA)29, gamma-irradiated virus-free PBMCs29, or cross-linking anti-CD3 antibodies and the newly produced virions can be detected5. Since activation of all T-cells is highly toxic30, 31, some

recent reactivation agents that have been investigated are more specific to the reactivation of

latent proviruses. These small molecules include histone deacetylase inhibitors (HDACi) such as

valproic acid, vorinostat, givinostat, belinostat, and panobinostat32-34, disulfiram35, prostratin36-38,

and bryostatin39. Valproic acid has had inconsistent results on the reduction of latent reservoirs in

vivo40, 41.

To verify that the population contains cells with integrated HIV genome, several assays

digest the host genomic DNA with a specific restriction enzyme. Then, the digests are diluted so

that intramolecular ligation is dominant. In Alu polymerase chain reaction (PCR), one primer

binding Alu repeat elements, which are interspersed throughout the genome, and another primer

specific for HIV are used to amplify the integrated DNA18. Common integration sites can be

sequenced by using inverse PCR where the region flanking the HIV genome is amplified42, 43.

While these digestion assays only detect integrated HIV proviral DNA, they have varying

efficiencies because the viral DNA integrates at different locations, so the lengths of the

amplified sequences vary.

Although the presence of integrated HIV genome is necessary for identifying latency, it is

not sufficient. Not all integrated HIV genomes produce replication-competent virus after

activation. Deleterious mutations in the reverse transcription and integration into silenced regions

of the host genome may result in the lack of competent virus production42. Identifying integrated

HIV genome and producing competent HIV virions following stimulation cannot, however, be

both applied to the same population. Stimulated cells will produce virus that can infect and

integrate into uninfected cells so it is unclear if the integrated HIV was from previously infected

or newly infected cells, and the process of detecting integrated HIV genome requires killing the

cell, which would prevent their stimulation.

Monie et al. devised a method that can partially bypass the conflicting tests44. Resting T- cells were cultured in the presence of drugs that block the reverse transcription (RT) of HIV mRNA to DNA and the integration of HIV DNA into the host genome. After stimulation, the newly produced virions that bud from latently infected cells can infect other cells, but the drugs prevent the integration of virus genome into the host. Finally, to avoid the varying transcript length of Alu PCR, HIV RNA from the media can be analyzed by RT-PCR primers specific to

HIV mRNA. While this assay can detect latently infected population of cells, no assay that can

detect latency in single cells exists.

1.3. Amplification and detection techniques

The detection of transcribed genes often uses reverse transcription (RT) polymerase chain

reaction (PCR) to convert mRNA into many copies of cDNA. This reaction can amplify many

specific transcripts from single cells—usually sorted into microtiter plates by flow cytometry or

micromanipulation—to recover particular genes of interest or to quantify the amount of mRNA

present45. Traditional assays for studying genetic and proteomic responses to applied external

stimuli typically require more than 1000 cells for each analysis46, 47. The resulting average

measures, however, obscure variations that may exist among individual cells, especially rare

cells, and can lead to misinterpretations of the biology48, 49. Using conventional plates is also

labor-intensive and costly for analyzing a statistically robust numbers of single cells. 16

Miniaturized systems have been developed that use actuated microfluidic systems50,

microdroplets of water-in-oil emulsions51-54, and arrays of microwells55-58 to define individual

PCR reactions requiring only femtoliters to nanoliters of reagents to reduce cost. These

approaches can also increase the efficiency of amplifying limited numbers of templates. On-chip

RT-PCR reactions have been demonstrated for amplifying isolated mRNA59, 60 or small numbers

of individual cells61, 62. Other techniques to amplify and detect weak signals also exist for targets

such as proteins, microRNA, mRNA, and DNA. They include fluorescent in situ hybridization63-

65, hybridization chain reaction66-70, and other isothermal catalytic amplification71-75. These assays have very good limit of detection (as low as 1 fM), but are generally still performed on a bulk sample of cells or tissue.

To establish a single-cell methodology for detecting latent infection, RT-PCR must be efficient in picoliter volumes. It has been demonstrated that 72 parallel PCR reactions in 450 pL volumes on a microfluidic chip was possible50 and this number has expanded to 96 single cell

samples by the Fluidigm Dynamic Array76. Real time RT-PCR has also been done in 1241 oil

droplets with volume 70 pL containing viral RNA77. Using a modified PCR reaction, the 454

sequencing in 75 pL silicon wells has sequenced about one million transcripts on beads55, 78.

Digital PCR reactions have been shown to amplify single copies of DNA in volumes as small as

36 femtoliters using PDMS58. It has also been used as a more sensitive alternative to qPCR for

detecting HIV DNA in a bulk population54. Finally, RT-PCR has been performed directly from

single cells without purifying the mRNA in 20 microliter volumes79. No technology, however,

combines all of these techniques into one-step, high-throughput, single-cell RT-PCR in picoliter

volumes.

1.4. Objectives and outline of thesis

The objectives of this thesis were to establish a RT-PCR technique to detect the presence of target genes in single cells using the array of nanowells as individual containers. By using these nanowells, we would be able to identify which single cells were infected with HIV-1 and interrogate its surface markers to determine its cellular lineage. Since the assay was developed in nanowells, we could also use other processes such as microengraving to link more information on a single cell. With this knowledge, we would be able to identify better cellular targets for

possible eradication of the disease. The specific aims of my doctoral thesis were the following:

1. Develop a new technique for the detection of mRNA transcripts from single cells using

RT-PCR reactions in nanowells for high-throughput screening.

2. Develop and optimize a multiplexed assay for detecting multiple DNA transcripts

produced by single cells.

3. Detect the production of virus in infected cells and determine the frequency and identity

of those cells.

Chapter 2 of the dissertation discusses the materials and methods used to develop and validate

the assay for detecting genetic transcripts in cells. Chapter 3 focuses on the optimization of RT-

PCR in nanowells. Chapter 4 uses the methods developed in Chapter 3 on single cells to

determine sensitivity and specificity and demonstrate the integration with other nanowell assays.

Chapter 5 uses RT-PCR to detect HIV in a cell line and HIV-positive patient. Chapter 6

discusses other methods that were considered for detecting rare transcripts. Finally, Chapter 7

contains a summary of the results and potential future directions for this work.

Chapter 2. Materials and Methods

2.1. Cell line culture

Epstein-Barr virus transformed human hybridoma 4D20 was a generous gift from James

Crowe (Vanderbilt University). The 4D20 cell line produces an IgG1 antibody against the 1918

H1N1 influenza virus. Cells were cultured as a suspension in R15 medium composed of RPMI

1640 (Mediatech) supplemented with 15% fetal bovine serum (PAA Laboratories), 2 mM L- glutamine (Mediatech), and 1x Penicillin-Streptomycin (Mediatech). The cell line was

2 maintained in 25 mm canted-neck flasks (BD Falcon) in 5% CO2 at 37 °C and was split twice a

week to 2.5x105 cells/mL.

The ACH2 cell line, a T-cell clone with one integrated copy of HIV-1, was obtained

through the NIH AIDS Reagent Program, Division of AIDS, NIAID, NIH from Dr. Thomas

Folks. These cells were cultured as a suspension in the same manner as the 4D20 cell line. ACH2 cells were split twice a week to 1x106 cells/mL.

The mutant phosphoglycerate kinase 1 (PGK1) cell line was purchased from the Coriell

Institute (GM14889). These cells were an Epstein-Barr virus transformed B-lymphocyte that contains a nucleoside base change of A491 (normal) to T491 (mutant) at position 491 in the

PGK1 gene and produces the amino acid substitution D163V. These cells were cultured as a

suspension in the same manner as the 4D20 cell line and were split twice a week to a ratio of 1:3.

2.2. Fabrication of array of nanowells

Silicon masters for 50 x 50 x 50 μm3 wells were produced by photolithography (Stanford

Microfluidics Foundry or Georgia Institute of Technology). Each chip fits on a standard glass slide (75 x 25 mm2, Corning) occupying approximately the center 60 x 22 mm2. Several types of

designs were used: the normal field of view (NFOV) array has 72 x 24 blocks of 7 x 7 wells and

the large field of view (LFOV) array has 43 x 14 blocks of 11 x 11 or 12 x 12 wells. A channel

was included to facilitate liquid removal by aspiration from the device and to act as a liquid

reservoir as water is lost into the polydimethylsiloxane (PDMS) during thermocycling. For the

NFOV, every 4 x 4 block was surrounded by channels, and for the LFOV, every block was

surrounded by channels. PDMS (Sylgard 184 Silicone Elastomer Kit, Dow Corning) or RTV615

(Momentive) was vigorously mixed at a mass ratio of 10:1 elastomer base to curing agent and

deaerated for 20 min under vacuum for Sylgard 184 or at least 1 hr for RTV615. Before the first

use, the injection mold and the silicon master were placed under vacuum with a glass vial

containing a few drops of trichloro(1H,1H,2H,2H-perfluorooctyl)silane (Sigma), and baked the

next day at 80 °C for 2 hr. Approximately 5 mL of PDMS was slowly injected into the mold so

that the final device has a thickness of 1 mm and was attached to a pre-cleaned standard glass slide. The injection mold was then cured for 2 hr at 80 °C. The arrays were removed from the mold while hot, and scotch tape (Staples) was applied to seal the wells so that dust did not fall into the wells. The arrays continued to cure in this manner at room temperature until they were used (usually more than a week). The glass backs of the arrays were further cleaned with ethanol, hexane, or acetone before use.

2.3. Cytometry and imaging

For cell surface marker staining, peripheral blood mononucleated cells (PBMCs) were

thawed from -196 °C in 37 °C R15. PBMCs were then washed twice with R15, counted with a hemacytometer, and rested in the 37 °C incubator for at least one hour. After resting, the cells were labeled with a set of surface marker antibodies (1:1000 dilution for each antibody) for either CD3+ or CD3– cells with a cell viability marker (1 nM calcein violet AM, Life

Technologies). The CD3+ panel contained CD4 Alexa Fluor 568 (Biolegend), CCR7 PE/Cy7

(Biolegend), CD45RA Alexa Fluor 647 (Biolegend), CD122 PerCP-eFluor 710 (eBioscience),

and CD95 Alexa Fluor 488 (Biolegend). The CD3– panel contained CD4 Alexa Fluor 568, HLA-

DR PE/Cy7 (Biolegend), CD14 Alexa Fluor 647 (Biolegend), and CD11c Alexa Fluor 488

(Biolegend). For only staining cell viability, the cells (e.g., 4D20 and ACH2) were washed once with PBS and stained with 1 nM calcein violet AM or 1 nM CellTracker Violet (Life

Technologies). After labeling for 30 min at 37 °C, the cells were washed once with PBS, loaded

into nanowells, and imaged on an epifluorescent microscope (Observer.Z1, Carl Zeiss GmbH) at

10x magnification (Objective EC “Plan-Neofluar” 10x/0.3, Carl Zeiss GmbH). A broad spectrum

light source was produced by a xenon lamp in a Lambda DG-4 (Sutter Instrument) and passed

through a “Pinkel” quad-band filter set (Semrock) for specific excitation bandwidths. Emissions

were filtered by a specific emission filters (Semrock) in a filter wheel (Lambda 10-3, Sutter

Instrument) just before image collection by an EM-CCD camera (C9100-13, Hamamatsu

Photonics). The entire system was controlled using the software AxioVision version 4.7 (Carl

Zeiss GmbH). The time settings were 100 ms exposure for each fluorescent channel at 100

EMCCD gain.

2.4. One-step reverse transcription polymerase chain reaction (RT-PCR)

2.4.1. Primer and TaqMan probe selection

Primers and dual-labeled TaqMan probes for housekeeping genes were designed using the online software RealTimeDesign (Biosearch Technologies). Some of the design criteria were intron spanning primers and probes to eliminate genomic DNA amplification, 55 °C melting temperature for primers and 60 °C for probes, probes did not start with 5’ G because of quenching, amplicon length of 90-200 base pairs, and 3’ end with G or C. The primer and probe

Table 2.1 Primer and probe design for RT-PCR.

Name Sequence (5'->3') Tm (°C)

B2M forward TCCAGCGTACTCCAAAGATTCAG 56.7 B2M reverse GAAACCCAGACACATAGCAATTCAG 56.1 B2M probe FAM-CTCACGTCATCCAGCAGAGAATGGA-BHQ1 60.3 GAPDH forward TTGCCCTCAACGACCACTTTG 58.1 GAPDH reverse GAGGTCCACCACCCTGTT 57.0 GAPDH probe FAM-TCCTGGTATGACAACGAATTTGGCTACA-BHQ1 59.8 ACTB forward GATGCAGAAGGAGATCACTGC 55.6 ACTB reverse GCCGATCCACACGGAGTA 56.9 ACTB probe FAM-CAAGATCATTGCTCCTCCTGAGCGC-BHQ1 61.7 4D20 Heavy Chain for. GGTCCTGTGCTGGTGAAAC 56.3 4D20 Heavy Chain rev. GCTCACACCCATTCTATCATTG 53.7 4D20 Heavy Chain probe Q670-CACAGAGACCCTCACGGTGACCT-BHQ2 62.4 HIVgag forward CATGTTTTCAGCATTATCAGAAGGA 53.6 HIVgag reverse TGCTTGATGTCCCCCCACT 59.0 HIVgag Q670 probe Q670-CCACCCCACAAGATTTAAACACCATGCTAA-BHQ2 60.7 HIVgag FAM probe FAM-CCACCCCACAAGATTTAAACACCATGCTAA-BHQ1 60.7 HIV 1LTR forward TTAAGCCTCAATAAAGCTTGCC 53.6 HIV 1LTR reverse GTTCGGGCGCCACTGCTAGA 62.4 HIV 1LTR probe Q670-CCAGAGTCACACAACAGAGGGGCA-BHQ2 62.8 HIV 2LTR forward CTAACTAGGGAACCCACTGCT 56.1 HIV 2LTR reverse GTAGTTCTGCCAATCAGGGAAG 55.4 HIV 2LTR probe Q670-AGCCTCAATAAAGCTTGCCTTGAGTGC-BHQ2 61.4

sequences (Biosearch Technologies) are shown in Table 2.1. All HIV primer and probe sequences were given to us by our collaborators from the Xu lab in the Ragon Institute. We used

6-carboxyfluorescein (FAM) and Quasar® 670 (Q670) with their respective quenchers, Black

Hole Quencher™ 1 (BHQ1) and BHQ2, as the two channels for the probes. The primers and probes were reconstituted in 1x TE buffer (10 mM Tris, pH 8.0, 1 mM EDTA) to a stock concentration of 100 μM. Probes for short-term use were further diluted to 2 μM in water. To minimize the number of freeze-thaw cycles, each aliquot of short-term use probe and 4x master mix were for a total reaction mix of four 80 μL reactions. All primers, probes, and 4x master mix solutions were stored at -20 °C.

2.4.2. Imaging end-point RT-PCR signal

Cells were split the day before their use in experiments. For cell labeling, the cells were first washed once with phosphate buffered saline (PBS, Mediatech), then resuspended in 1 mL

PBS with 1 μL of the labeling dye (CellTracker Violet BMQC or calcein violet AM, Life

Technologies) according to the manufacturer’s recommended concentration. Labeling was carried out at 37 °C for 30 min. Cells with high-viability were isolated with Ficoll-Paque Plus

(GE Healthcare Biosciences) and then treated with 30 μg/mL bovine pancreatic RNase A

(Sigma-Aldrich) at 37 °C for 30 min. The cells were then washed three times with 10% FBS in

RPMI and once with PBS before they were resuspended in 5 mL PBS. After these steps, more than 98% of the cells remained viable as determined by the cellular exclusion of trypan blue

(Life Technologies). Each array of nanowells was cleaned by a 30 s plasma treatment (Plasma

Cleaner PDC-32G, Harrick Plasma) and blocked in 0.5% BSA in PBS for 30 min at room

temperature before loaded with cells by gravity such that about 50% of the wells had cells in

them.

The reaction mix used the qScript One-Step Fast qRT-PCR kit with ROX (Quanta

Biosciences). It contained 1x One-Step Fast Master Mix with ROX, 1 μM of each primer, 200 nM of each probe, 1x qScript One-Step Fast RT, 80 U of SUPERase-In RNase Inhibitor (Life

Technologies), and 0.05% NP-40 (Tergitol, Sigma) in a total volume of 40-80 μL per array. The

final NP-40 concentration was later increased to 0.5% to lyse the cells more effectively. Before

adding the reaction mix, the array was washed with 1x Tris-buffered solution (TBS, 20 mM Tris, pH 7.5, 150 mM NaCl) and quickly rinsed with 0.5x TBS or water. The reaction mix was applied to the nanowells and spread using a pipet tip before the device was sealed onto another glass slide. Excess reaction mixture was removed along the sides and the entire device was placed on an Eppendorf Mastercycler Gradient (Eppendorf) with a glass slide adaptor (in situ Adapter,

Eppendorf). Mineral oil (Sigma) was added to improve the heat conductivity between the adaptor and the device. The thermocycle profile was initially 40 min at 50 °C, 2 min at 95 °C, 12 cycles of 40 s at 95 °C and 1 min at 65 °C, and 38 cycles of 40 s at 95 °C and 1 min at 60 °C, with the lid maintained at 50 °C. It was common to observe dried wells and warped wells (pincushion distortion) around the perimeter of the array. The array also became cloudy from the penetration of water into the PDMS. The use of RTV615 instead of Sylgard 184 as the silicone reduced the number of dried wells almost completely (>98% usable array) and the thermocycles were later reduced to 15 min at 50 °C, 2 min at 95 °C, 35 cycles of 40 s at 95 °C and 1 min at 60 °C.

2.4.3. Quantitative TaqMan RT-PCR

Quantitative TaqMan RT-PCR (RT-qPCR) was run on LightCycler 480 (Roche). The

same thermocycles were used in RT-qPCR as in the nanowell RT-PCR. Each reaction had a final

volume of 20 μL in a clear, LightCycler-specific, 96-well plate (Roche). The cycle numbers that exceeded the threshold intensity (Ct) were calculated with the built-in analysis module. The final

PCR products were also imaged on 2% agarose gel with ethidium bromide to verify amplicon

length. The desired DNA band was cut from the gel and purified with a QIAquick gel

purification kit (Qiagen) for digital PCR in nanowells. The purified DNA concentration was

measured by forming a column with 1.5 μL of DNA on a NanoDrop 1000 spectrometer

(NanoDrop).

2.4.4. Digital PCR in nanowells

Serial dilutions of purified DNA (e.g., HIVgag PCR product) were made and 1 μL of

each dilution was added to 80 μL of reaction mix. The reaction mix contained the primer and

probe set for the desired DNA template as well as the primer and probe set for a negative gene

(e.g., ACTB) in a separate channel as a negative control. The nanowell arrays were treated as

normal (e.g., plasma cleaning and blocking) and rinsed in water for the final step before adding

the reaction mix.

2.5. Microengraving

Detailed procedures for microengraving can be found in Ogunniyi et al. Nature Protocols

(2009) vol. 4 (5) pp. 767-82. Briefly, cells were labeled for cell viability (CellTracker violet),

loaded into the nanowells, and imaged. The nanowells were then sealed with a glass slide that

was functionalized with anti-IgG1 antibodies at 37 °C. After 2 hr, the glass slide was separated

from the nanowells and the captured IgG1 was detected following the application of a secondary,

goat anti-human IgG1 antibody conjugated with Alexa Fluor 647 (Life Technologies).

Data from the microscopy, microengraving, and RT-PCR were collected and filtered.

Only wells that contained a single live cell initially, and had a single cell after RT-PCR (detected

by non-specific staining with the reference dye, ROX) were tabulated. Spots on the microarray

generated by microengraving that had a signal-to-noise ratio greater than 2 for more than 55% of

its pixels and a coefficient of variation less than 80 were considered positive for IgG1 secretion.

2.6. Surface capture of transcripts

One method to capture transcripts on glass slides was to use amine-epoxy linkage. Glass slides were cleaned with 2.5 M sodium hydroxide (NaOH) in 60% ethanol (EtOH) for 2 hours and reacted in a 0.1% (3-glycidoxypropyl)trimethoxysilane in 100% EtOH supplemented with traces of glacial acetic acid as an acid catalyst for 2 hours at 40 °C. The slides were washed twice with 100% EtOH and baked at 120 °C overnight to remove residual water. One surface of the epoxy slides was then reacted with the primer mix conjugated with a 5’ amine-C9 group in 0.15

M NaOH at 80 °C for 2 hours. The free epoxy groups were washed and blocked with 0.2 M Tris

and 0.1% sodium dodecyl sulfate (SDS) at 50 °C for 4 hours before the primer-conjugated glass were ready.

Another method to attach transcripts to surfaces (e.g., PDMS or glass slide) was to use oligonucleotides with a 5’ amine group linked to amine groups on the surface by p-phenylene diisothiocyanate (PDITC)80. To functionalize the PDMS with amine groups, the array of

nanowells was plasma treated for 5 minutes and then placed in a 10% (3- 26

aminopropyl)triethoxysilane (APTES) in water and rocked for 1 hr at room temperature in a 4- well polystyrene dish (Nunc). After two 30 s, manual water washes, the array was dried at 80 °C,

overnight. The next day, the dried array was soaked in acetone for 3 min to rewet the well

surfaces, and washed in dimethylformamide (DMF) before reacting in 0.2% (w/v) PDITC in

10% (v/v) pyridine/DMF for 3 hr at room temperature. Excess PDITC was removed by two

DMF washes, one methanol wash, and one 100 mM sodium bicarbonate, pH 9.0 wash. The

desired amine-conjugated oligonucleotide that was reconstituted in water was diluted to 25 μM

in 50 mM sodium borate, pH 8.5. This solution was reacted with the PDITC-conjugated array

under a lifterslip in a humidified box. The next day, the reaction was quenched with 1x TBS for

10 min, and blocked with 0.5% BSA in PBS for 1 hr at 80 °C. After three PBS washes, the array

was loaded with cells for downstream processing. Cells were loaded and lysed in MES lysis

buffer (20 mM MES pH 6.0, 500 mM NaCl, 10 mM EDTA, 0.01% NP-40, 10 mM DTT).

2.7. Hybridization chain reaction

The hairpin pairs and initiator sequences for hybridization chain reaction (HCR) were

modified from the sequences A, H1 and H2 from literature67. The modifications include moving

blocks of bases, using the complementary sequences, and adding 9 adenosines at the end of the

initiators where the oligonucleotide was attached to either an antibody (through a 3’ thio-C3

linker on the initiator) to detect a target protein or another oligonucleotide to detect a target gene.

The four sets of initiator ( ) and hairpins ( and ) sequences (5’ to 3’) were ordered from An H2n-1 H2n Integrated DNA Technologies and purified by high performance liquid chromatography (Table

2.2).

All HCR oligonucleotides were reconstituted to 100 μM in 1x SPSC (0.1 M sodium phosphate, 1 M sodium chloride). Each initiator and hairpin were denatured to 95 °C for 2 min, immediately placed on ice for 1 min, and kept at room temperature until used. To test the specificity of the hairpins to their initiators, each of the four initiators were diluted to 2 μM in 1x

TBE buffer (90 mM Tris-borate, 2 mM EDTA, pH 8.3) and 1 μL was spotted onto a poly-L- lysine coated glass slide at different locations. A mixture of the four sets of hairpins (20 μM each) was dispensed onto the slide under a lifter slip (Electron Microscopy Sciences) for 2 hr at room temperature. The slide was washed with PBS/0.05% Tween-20 and PBS, and scanned using a

Genepix 4200AL (Molecular Devices). The commercial software package Genepix Pro 6.1 was used to extract the fluorescence for each spot in each of the four channels.

Table 2.2 Sequences of initiator and hairpin for HCR.

ID Sequence (5'->3')

A1 GCA CGT CCA CGG TGT CGC TTG AAT AAA AAA AAA

H1 FAM-ATT CAA GCG ACA CCG TGG ACG TGC ACC CAC GCA CGT CCA CGG TGT CGC ACC

H2 FAM-GTT GCA CGT CCA CGG TGT CGC TTG AAT GCG ACA CCG TGG ACG TGC GTG GGT

A2 GCA GCC GTA GAC TAG TGC GCG AAT AAA AAA AAA

H3 TYE563-ATT CGC GCA CTA GTC TAC GGC TGC ACG ACC GCA GCC GTA GAC TAG TGC CAC

H4 TYE563-GTT GCA GCC GTA GAC TAG TGC GCG AAT GCA CTA GTC TAC GGC TGC GGT CGT

A3 CGT CGG CAT CTG ATC ACG CGC TTA AAA AAA AAA

H5 TYE665-TAA GCG CGT GAT CAG ATG CCG ACG TGC TGG CGT CGG CAT CTG ATC ACG GTG

H6 TYE665-CAA CGT CGG CAT CTG ATC ACG CGC TTA CGT GAT CAG ATG CCG ACG CCA GCA

A4 CGT GCA GGT GCC ACA GCG AAC TTA AAA AAA AAA

H7 TEX615-TAA GTT CGC TGT GGC ACC TGC ACG TGG GTG CGT GCA GGT GCC ACA GCG CTG

H8 TEX615-CAA CGT GCA GGT GCC ACA GCG AAC TTA CGC TGT GGC ACC TGC ACG CAC CCA

2.8. Data Analysis

Images generated by automated microscopy were analyzed using custom software

(Enumerator, mabanalyze, and CellProfiler). The location, the number of cells, and the 28

fluorescence intensity of each channel were tabulated in a text file. This information was filtered and plotted using MATLAB (MathWorks). The data were filtered to remove wells with more than four cells because too many cells gave inaccurate measures of the well intensity. Wells with large variation in the reference channel (greater than two standard deviations from the mean reference signal) were also removed to eliminate wells with no liquid and wells with a high degree of covariance in fluorescence. This filter was important because dividing by a low reference signal would give a relative intensity that was too high and artificially positive. This artificial positive signal was especially problematic for the FAM channel. For each block of wells, the mean gene-specific fluorescence intensity of empty wells (Iempty) was calculated and used to determine the relative fluorescence of every well (Iwell/Iempty). A histogram was plotted to bin the relative fluorescence intensities. The histogram peak for Iwell/Iempty of empty wells was fit to a Gaussian curve to compute estimated values for the mean and standard deviation of negative reactions. The threshold value on the relative fluorescence for positive reactions was set to be three standard deviations above the mean. From this value (e.g., Iwell/Iempty = 1.4), the sensitivity, specificity, and positive predictive value were determined for each gene. For the analysis of

Q670 fluorescence data from HIV-infected cells, a threshold of 1.5 times the mean empty fluorescence, which corresponded to approximately 8-12 standard deviations from the mean, was used. Such a high cutoff was possible because the signal intensity from the Q670 dye was much brighter than that of the FAM dye.

Chapter 3. Establishing one-step RT-PCR in nanowells

3.1. Optimization of cell lysis

To test the RT-PCR efficiency in the nanowells, we used beta-2-microglobulin (B2M) as the target gene because B2M is constitutively expressed in the 4D20 cell line. B2M primers and

TaqMan probes were designed to reverse transcribe bases 122 to 211 from the mature mRNA

(GENBANK NM_004048). No additional steps or reagents were used to remove genomic DNA from the reaction. The signal we used to determine a positive reaction came from the digestion of a quenched TaqMan probe. If the target gene were present, the intact probe would bind to the

Figure 3.1. TaqMan probe mechanism. When the probe is intact, the emitted light by FAM is quenched by the BHQ-1. As the Taq polymerase extends the primer, its exonuclease will cut the probe, thus freeing FAM. FAM is no longer within the proper distance from BHQ-1 for quenching, so it can be detected.

desired gene by complementary base pairing. As the PCR progressed, the Taq enzyme would cleave the probe because it has 5’ to 3’ exonuclease capabilities (Figure 3.1). Thus, the fluorophore (e.g., FAM or Quasar 670) would no longer be at a fixed distance from the quencher and the detection of its fluorescence would be possible by epifluorescence microscopy.

Several commercial kits were tested in tubes to identify the kit with the best efficiency for use in subsequent RT-PCR reactions. Two of kits were the qScript 1-Step Fast RT-PCR with

ROX (Quanta Biosciences) and QuantiFast SYBR Green RT-PCR (Qiagen). We found that using a template HeLa mRNA, the qScript 1-Step Fast RT-PCR with ROX had the lower Ct

0.6

0.5

0.4 QuantiFast SYBR 10 ng QuantiFast SYBR 0.1 ng 0.3 QuantiFast SYBR 10 pg QuantiFast SYBR 1 pg

Relative Fluorescence Relative 0.2 qScript ROX 10 ng qScript ROX 0.1 ng qScript ROX 10 pg 0.1 qScript ROX 1 pg

0 0 5 10 15 20 25 30 35 40 45 50 Cycle number

Figure 3.2. Determining the RT-qPCR kit. The 1-step Fast qScript with ROX kit was significantly better than the QuantiFast SYBR Green kit. The Ct value for qScript at 10 pg mRNA was approximately the same as the Ct value for the 0.1 ng mRNA for QuantiFast.

value (a measurement of reaction effectiveness), so it was the best at the reverse transcription of mRNA from single cells (Figure 3.2). In fact, the fluorescence profile of the 10 pg of mRNA 32

with qScript was similar to that of the 100 pg of mRNA with QuantiFast, which suggested that

qScript was about 10 times more sensitive to mRNA than QuantiFast. We also tested two types of kits from Quanta Biosciences (qScript with ROX and qScript MGB). From serial dilutions of mRNA, we observed that both kits were similarly sensitive, but the qScript kit with ROX had a higher relative fluorescence than the MGB kit for the 10 pg mRNA sample (Figure 3.3). Since our target genes will be rare cellular mRNA, we chose the more sensitive detection and higher relative fluorescence kit, 1-step Fast qScript with ROX, for all future studies.

To adapt a typical multistep RT-qPCR in a PCR tube to a one-step reaction in nanowells, we added a detergent to help lyse the cells. We chose to test two non-ionic detergents, Tergitol type NP-40 (NP-40) and Tween-20, and one ionic detergent, sodium dodecyl sulfate (SDS), for their ability to help lyse cells during RT-PCR55. These detergents were commonly added in the

RT-PCR kits to help stabilize the enzymes in the reaction mix. It was also important to ensure

that the addition of even more detergents did not significantly hinder the reaction itself.

We tested an initial range of final concentrations (0.75%, 0.5%, 0.25%, 0.1%, 0.05%,

0.005% and 0%) for Tween-20 and NP-40 on reactions with 100 4D20 cells and observed that

the optimal concentration for Tween-20 was 0.5% (Figure 3.4) while the optimal NP-40

concentration was approximately 0.75% (Figure 3.5). The NP-40 concentration was less clear because the two no detergent controls were not consistent. A direct comparison of 0.5%, 0.25%, and 0.05% Tween-20 to 0.75%, 0.5% and 0.05% NP-40 showed that 0.5% and 0.05% NP-40 and

0.5% Tween-20 were the best conditions (Figure 3.6). At cycle 30, 0.5% NP-40 was slightly better than the other conditions, but 0.05% NP-40 was chosen for initial tests to minimize the interference with the kit. To determine the efficiency of the RT-qPCR reaction, the logarithm

(base 10) of the cell number was plotted against the Ct value at that cell number and regressed to

a linear line to determine the slope. The efficiency of the reaction, which is (10-1/slope – 1), was approximately 1.00 for 0.05% NP-40 and 0.39 for 0.5% Tween-20 (Figure 3.7). This result suggested that Tween-20 was interfering with the RT-qPCR reaction, so subsequent reactions only used NP-40 as the detergent.

These serial dilutions of detergent also revealed that amplification of genomic DNA did occur in bulk reactions (no RT controls) despite using intron-spanning primers and probes, but it required 10 more PCR cycles than the amplification of cDNA (Figure 3.6). This result further suggested that the removal of genomic DNA was unnecessary given an appropriate number of cycles.

RT-qPCR experiments with the ionic detergent SDS showed strong inhibition of the reaction at SDS concentrations greater than 0.01% (Figure 3.8). At concentrations of 0.01% and

0.005%, the SDS had a Ct value that was lower than 0.05% NP-40. These results were expected since SDS is often used to denature proteins in SDS-PAGE, so in addition to breaking up the cell more effectively than NP-40, SDS could also be denaturing the RT and Taq enzymes, thus inhibiting the reaction completely.

Figure 3.3. Comparison between 1-step Fast qScript with ROX and MGB. Both kits performed equally for sensitivity to mRNA. The ROX kit had a higher relative fluorescence than the MGB kit.

0.6

0.75% Tween20 0.5 0.5% Tween20

0.4 0.25% Tween20

0.1% Tween20 0.3 0.05% Tween20 0.2 0.005% Tween20 Relative Fluorescence Relative

0.1 0% Tween20 0% Tween20 0 0 10 20 30 40 50 60 Cycle Number

Figure 3.4. The effect of Tween-20 on the RT-qPCR reaction on 100 cells. The optimal concentration of Tween-20 was 0.5%.

Figure 3.5. The effect of NP-40 on the RT-qPCR reaction on 100 cells. The optimal concentration for NP-40 was approximately 0.75%, but it was not clearly determined since the no detergent controls were not consistent.

0.45 0.4 0.75% NP40 0.35 0.5% NP40 0.3 0.05% NP40 0.25 0.2 0.5% Tween20 0.15 0.25% Tween20

Relative Fluorescence Relative 0.1 0.05% Tween20 0.05 No Detergent 0 No RT, No Detergent 0 10 20 30 40 50 60 Cycle Number

Figure 3.6. Direct comparison between Tween-20 and NP-40. Both 0.5% and 0.05% NP-40 were similar to 0.5% Tween-20.

40 y = -6.9353x + 39.031 35 R² = 0.9212 30

25 y = -3.3175x + 30.576 R² = 0.9934 20

Ct number Ct 0.05% NP-40 15 0.5% Tween-20 Linear (0.05% NP-40) 10 Linear (0.5% Tween-20)

0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 Log(Cell number)

Figure 3.7. Efficiency of RT-qPCR with the addition of 0.05% NP-40 or 0.5% Tween-20. RT-qPCR efficiency was 1.00 for NP-40 and 0.39 for Tween-20.

0.9

0.8

0.7

0.6 100, 0.005% SDS 0.5 100, 0.01% SDS 100, 0.05% SDS 0.4 100, 0.1% SDS 100, 0.5% SDS 0.3 100, 1% SDS Ct Threshold Relative FluorescenceRelative 0.2

0.1

0 0 10 20 30 40 50 60 Cycle Number

Figure 3.8. Effect of SDS on RT-qPCR on 100 cells. SDS strongly inhibited the reaction at concentrations greater than 0.01%.

3.2. Optimization of RT-PCR in nanowells

When translating the reaction from a typical 20 µL tube to an array of 125 pL reactors, several additional design considerations were addressed. One extra consideration was heat conduction. The PCR machine was designed for PCR tubes with 0.2 mm thickness that fit perfectly in each well. This fit allowed direct contact of the wall of the tube with the metal heat conductor. For RT-PCR in nanowells, a metal adaptor for glass slides was fitted to the 96-well format. Also, the glass slide was approximately 1 mm thick, compared to the 0.19 mm thickness of the PCR tube. To ensure good thermal contact, 20-30 µL of mineral oil was added between the glass and the metal adaptor. The thermal conductivity of water, glass, mineral oil, and PDMS are on the order of 0.1 to 1 W/mK. A COMSOL Multiphysics model for heat conduction through the glass slide from 60 °C to 95 °C showed that by 5 s, the glass slide and liquid in the nanowells would reach the desired temperature (Figure 3.9). To be more conservative, an additional 10 seconds were added to every step of the thermocycle to allow the temperature to equilibrate.

Although the kit could be run at “fast” cycling profile with shorter denaturing and extension steps, we opted for the normal length.

Another concern for the RT-PCR in nanowells was that the size of the wells would increase the concentration of cellular RNase as well as the total surface of the array. Due to the small reactor volume, RNase concentrations increased dramatically compared to large tube reactions, so SUPERase In, an RNase inhibitor, was added to the reaction mix. To reduce the non-specific adhesion of enzymes to the walls of the nanowells, the PDMS and the glass slide were blocked with bovine serum albumin (BSA). Initial tests also included 0.5% BSA in the reaction mixture to further reduce non-specific binding, but this additive was not necessary.

Figure 3.9. Temperature model for an increase in temperature from 60 °C to 95 °C after 5 seconds. This model showed that in 5 seconds, the temperature profile of the glass slide and thin water (e.g., liquid in the nanowells) had raised to 94 °C.

3.3. Optimization of pre-treatment of cells

Methanol, water and thermal treatments were tested for the most efficient method to lyse

cells in nanowells. Methanol was chosen as a protocol to fix the cells and help reduce the activity

of RNase and DNase. Water was effective at lysing cells since it is hypotonic compared to the

cellular cytoplasm. Initial data showed that treating cells with pre-cooled methanol (-20 °C) for

10 min after the cells were loaded, quickly using water as a hypotonic lysis before adding the reaction mixture, and adding a heat lysis step at 50 °C for 30 min with reverse transcription

resulted in 80% positive reactions (Table 3.1). However, 40% of empty wells were also bright.

These false positives were postulated to be from mRNA contamination from the supernatant of

the cell suspension. To test this hypothesis, the cell suspension was treated with RNase A for 30 min at 37 °C and washed 3 times to remove the RNase A before the cells were loaded onto the array. Using RNase A that was DNase-free, we saw near complete elimination of the false positive signal (Figure 3.10). Therefore, all future reactions had a pretreatment step for the cell suspension to digest any free mRNA in the supernatant. Interestingly, using RNase that was not

Figure 3.10. Removal of false positives by RNase, DNase-free treatment. The histogram on the right show that most wells with no cells had a FAM/ROX less than 2 (cutoff for positive signal), while wells with cells did have specific positive signal.

labeled as DNase-free had a FAM/ROX ratio of approximately 3 in every well (data not shown).

This minor difference showed that DNase was not inhibited during the reaction and that the maximum signal from FAM was approximately 3. Adding DNase to the master mix also verified these results. Other pretreatment steps that were required were the quick wash with water before applying the reaction mixture to help burst the cells open. The methanol treatment of the cells, however, did not improve the percent of true positives in wells with cells.

Table 3.1 List of tested RT-PCR lysis conditions. Lysis conditions Fraction of bright wells Parameter Methanol Detergent Water lysis RT # of cycles Empty Occupied Yes Yes Yes Yes 50 0.3853 0.8131 Yes No Yes Yes 50 0.3832 0.8098 Base case Yes Yes Yes No 50 0.0123 0.0492 Yes No Yes No 50 0.0102 0.0375 Positive control Yes Yes Yes Yes 50 0.4536 0.8194 No detergent Yes No Yes Yes 50 0.4062 0.6758 No probe Yes Yes Yes Yes 50 0.0001 0 No primer Yes Yes Yes Yes 50 0.0003 0.0008 Yes Yes Yes Yes 40 0.1509 0.421 Yes No Yes Yes 40 0.1188 0.3899 40 Cycle Yes Yes Yes No 40 0.0077 0.0335 Yes No Yes No 40 0.0103 0.037 No Yes Yes Yes 40 0.4355 0.7955 No No Yes Yes 40 0.5229 0.8088 No methanol No Yes Yes No 40 0.0043 0.0745 No No Yes No 40 0.0253 0.1876 Yes Yes Yes Yes 50 0.4724 0.7876 Yes No Yes Yes 50 0.2811 0.5679 Quick water rinse Yes Yes No Yes 50 0.0001 0 Yes No No Yes 50 0.0019 0.0004

3.4. Optimization of thermocycling

One of the concerns with running too many thermocycles was that the bulk RT-qPCR on cells showed positive fluorescence at 10 cycles later in the sample with no reverse transcriptase added than in samples with the RT enzyme. This result indicated that genomic DNA would be amplified given enough cycles. To determine how many cycles were needed, we first ran digital

PCR with HIVgag PCR product as the template on the array and started with 50 or 70 cycles.

These experiments used a serial dilution of the HIVgag DNA ranging from an average input

number of DNA of 8 copies/well to 0.125 copies/well (Figure 3.11). Even though 70 cycles were

run, the fluorescence intensity did not increase significantly compared to the 50 cycles. This lack

of increase in the signal further showed that we could attain a maximum signal in the nanowells

at an optimal cycle number. The variability in the positive signal did not disappear with more

cycles, so it suggested that the variability was innate to the system. One cause for the fluctuation

in positive signal could be the distribution of probe in each well.

Several assumptions were made to estimate the fraction of wells that should be positive.

Considering that the nanowells were initially filled with water and assuming that the limit of

detection for DNA was 1 copy per well, the estimated fraction of bright wells in the array was

diluted by a factor of approximately 0.14 or a positive fraction of 1.12 (i.e., 1) to 0.0175. Plotting

this estimated fraction with the actual fraction of bright wells, the limit of detection was

determined to be approximately 1.4 copies instead of 1 copy per well (Figure 3.12). This number

of transcripts in a 125 pL volume corresponded to a concentration of 18.6 fM.

To determine the optimal cycle number, we added enough HIVgag DNA (100 fold more

than 0.4 estimated fraction) to the array so that every well would be bright. We also included the 42

ACTB or GAPDH primers and probe as a negative control. Plotting the histogram of the relative

intensities showed that at 15 cycles, the fluorescence was just barely above the negative peak. By

cycle 20, the fluorescence was twice the negative peak, and by 30 cycles, the fluorescence was

approximately 3 times the negative peak (Figure 3.13). Interestingly, the no DNA control at 35 cycles had a small positive peak, but the mean of this peak was not higher than that of the 30 cycles. This false positive showed that there was some HIVgag contamination in the negative control and that 30 cycles were enough for a maximum fluorescence signal of 3 times the negative peak. Note that theoretical calculations on the perfectly efficient amplification of 1

molecule of DNA showed that 24 cycles were needed to exceed the concentration of 200 nM

probe. Since we added enough DNA for approximately 40-80 copies per well, the 30 cycles would correspond to about 35-37 cycles for the detection of single copy of DNA. Therefore, all future experiments used 35 cycles.

Figure 3.11. Sample images of digital PCR on a serial dilution of HIVgag DNA. From the top to bottom, then left to right, the average input DNA number per well was 8, 4, 2, 1, 0.5, and 0 copies. These input values corresponded to a fraction of bright wells of 1.12 (i.e., 1), 0.56, 0.28, 0.14, 0.07, and 0.

0.6 y = 0.7164x + 0.0009 R² = 0.9667 0.5

0.4

0.3

0.2 Actual fraction Actual

0.1

0 0 0.1 0.2 0.3 0.4 0.5 0.6 Estimate fraction

Figure 3.12. Limit of detection for PCR. The limit of detection was 1/slope or approximately 1.4 copies of DNA.

RTPCR HIVgag 45e-4ng RTPCR ACTB (neg) 0.4 0.25 Preimage Preimage 0.35 20 cycles 20 cycles 10 cycles 10 cycles 0.2 15 cycles 15 cycles 0.3

0.25 0.15

0.2

0.1 0.15 Normalized frequency Normalized frequency

0.1 0.05 0.05

0 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 ROX normalized intensity ROX normalized intensity

PCR HIVgag 18e-4ng PCR ACTB (neg) 0.45 0.35 Preimage, no DNA Preimage, no DNA 0.4 5 cycles 5 cycles 0.3 30 cycles 30 cycles 0.35 20 cycles 20 cycles 35 cycles, no DNA 0.25 35 cycles, no DNA 0.3

0.25 0.2

0.2 0.15 Normalized frequency 0.15 Normalized frequency 0.1 0.1

0.05 0.05

0 0 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 ROX normalized intensity ROX normalized intensity Figure 3.13. Various cycle numbers for PCR on HIVgag DNA. The progression of signal can be seen over the number of cycles. After 20 cycles, the fluorescence intensity was approximately twice that of the preimage (0 cycle), and after 30 cycles, it was approximately 3 times that of the preimage. Note that 35 cycles, no DNA had some contaminating DNA in it, but its fluorescence was not greater than that of the 30 cycles.

3.4. Discussion

3.4.1. Limit of detection of transcripts

Although we had shown that the assay could detect a few copies of DNA, it was not clear what the limit of detection of mRNA was. In a control experiment with diluted standards of cell-free mRNA bound to oligo-dT beads, we were able to detect positive reactions from the beads (Figure 3.14). However, the number of copies of the mRNA on the beads and in bulk could not be measured accurately. Similar to digital PCR, mRNA could be detected in nanowells

(Figure 3.15), but the exact number of B2M mRNA could not be determined. Although this result cannot be compared directly to conditions in which residual components of the lysed cell remain, it would suggest that nanowell-based RT-PCR allows for detection of small quantities of mRNA.

To more accurately determine the limit of detection, cells were added to the digital

PCR. This addition simulated the effect of cellular lysate on the detection of DNA in nanowells. RT-PCR mixture was spiked with HIVgag DNA to have approximately 20% positive wells, and 4D20 cells (HIV negative) were loaded in the nanowells at high density so that the cell occupancy was from 0 to 4 or more cells (Figure 3.16). Wells with no cells had the expected 20% of bright wells, but if a well had cells, the fraction of bright wells dropped. From 0 to 1 cell, the drop was 4-fold, with another 4-fold decrease from 1 cell to 4 or more cells. Since the DNA was randomly distributed onto the array, the expectation was that the fraction of bright nanowells would be equal no matter how many cells were in them.

The drop in the fraction of bright nanowells showed that the cell lysate was interfering with the PCR, making the reaction less sensitive to small number of DNA copies. The limit of 46

detection of DNA with cell lysate in the nanowell would be 6 to 23 copies of DNA.

Figure 3.14. Detection of B2M mRNA transcripts on beads from bulk cellular mRNA extraction. 4D20 mRNA was extracted with oligo-dT beads. The beads were settled into nanowells and RT-PCR was run. A positive fluorescence signal for reactions with RT enzyme was present around 3 and was absent in the no RT enzyme control.

Figure 3.15. Detection on B2M mRNA from bulk cellular mRNA extraction. 10 ng or 100 pg of bulk cellular mRNA was added to the reaction mix and applied to the array. The exact number of mRNA copies in each well was unknown.

RTPCR 18.4e-6ng gag 0.25

0.2

0.15

0.1 Fraction positive for HIV gag

0.05

0 0 1 2 3 4 Number of cells

Figure 3.16. Effect of cell lysate on RT-PCR. HIVgag DNA was spiked into an RT-PCR reaction with 4D20 cells. Having just one cell in a nanowell reduced the fraction of positive fluorescence by 4-fold from that of wells with no cells.

3.4.2. Evaporation

Evaporation caused by detachment of the array from the sealing glass slide and

permeation of water from the nanowells into the PDMS were observed after the thermocycles81.

The loss of water also caused the nanowells to pucker and shrink into a pincushion distortion. In

case of extreme loss of water, the entire nanowells would detach from the glass slide and had no

fluorescence signal in them. This evaporation was especially prevalent in the few rows and

columns of entire blocks bordering the edges of the array. Also, wells that bordered evaporated

blocks had brighter signal in all fluorescence channels. The increased fluorescence often required additional wells to be removed, thus reducing the percentage of usable wells in the array.

Changing the PDMS from Sylgard 184 to RTV615 significantly decreased the number of

evaporation events to less than 10% of the array and often to less than 5%. One explanation is

that the RTV615 was softer than Sylgard 184, so it could more easily mold to a flat glass slide if there were surface unevenness on the array. The RTV615 was also more stuck to the glass slide than Sylgard 184 as more force was needed to pry the glass slide off the array after thermocycling.

Another important feature of the array was the presence of microfluidic channels. To maximize the loading of cells in wells, arrays without channels were often used. However, for

RT-PCR, arrays with no channels had evaporation in greater than 20% of the array. This result suggested that the channels acted as a liquid reservoir in the array. The importance of a reservoir had been observed to prevent evaporation in femtoliter digital PCR in PDMS58. To increase the

reservoir volume further, we added channels between every block of nanowells, up from the original design of every four blocks. In the original four block format, wells bordering a channel were removed because they had brighter fluorescence than the wells that did not border channels.

Interestingly, if there were channels between every block, the outer wells did not have a significantly brighter fluorescence. This phenomenon was observed in the array of 12 x 12 nanowells as well as the 11 x 11 design, where the wells were closer to the channels, so it was not proximity to the channel that caused the brighter fluorescence. One possible explanation

could be that while the overall loss of liquid could be the same, the fraction lost in all the

channels dropped since the size of the reservoir increased by 4-fold.

Finally, dropping the total number of thermocycles from 50 to 35 further alleviated the

loss of water issues to just the bordering few rows and columns of the outer blocks. Not only did

the reduction in thermocycles increase the percentage of usable wells in the array, it also 49

separated the positive and negative fluorescence signal. While cycles more than 35 did not

increase the mean positive fluorescence signal, they did increase the mean negative fluorescence

signal.

3.4.3. Limitations

One of the limitations of this method is that it was not quantitative. Unlike RT-qPCR in

tubes, RT-PCR in nanowells was an endpoint measurement of the fluorescence from TaqMan

probes, so only a presence or absence of a target gene could be obtained. The fluorescence was

not measured after every cycle. Questions such as how much transcript was in the cell could not

be answered by this platform. While we could detect the fluorescence increase in 5 cycle

increments, as shown in the thermocycle optimization experiments, this required a different array

for each cycle that was measured. When the same array was reimaged multiple times, the entire

array was photo-bleached and lost all fluorescence signals after 3 to 4 reimages. In fact, a drop of

ROX signal was observed in the last quarter of the array even before the entire array was imaged

for the first time. These observations could be attributed to the high transmittance of light of

PDMS82. As one block of nanowells was imaged, some of the light also hit other blocks, so when

the last block of the array was imaged, some light could have already hit the last block more than

600 times.

A second limitation to this assay was that it could not consistently detect nuclear

transcripts such as 1LTR circles in HIV-infected cells. In bulk RT-qPCR experiments, the no RT control still had a positive fluorescence profile, but at a later cycle number than the sample with

RT enzyme. Bulk qPCR of ACH2 cell lysates showed detectable levels of 1LTR DNA circles and these circles were exclusively located in the nucleus83. However, when RT-PCR was run on

ACH2 cells, the 1LTR signal was not consistent: most wells with ACH2 cells were not positive for 1LTR. Since the limit of detection for PCR was approximately 10 copies of DNA, this lack of signal showed that the nucleus was not fully lysed by the reaction mixture.

Chapter 4. Characterization of RT-PCR in nanowells

4.1. Sensitivity and specificity

To establish the feasibility for in situ lysis and detection of an expressed gene of interest in wells containing cells, we used a human B cell hybridoma (4D20) that produces an antibody

(IgG1) against the 1918 influenza virus84. Lysis of the cells and subsequent reverse transcription of a constitutively expressed gene (beta-2-microglobulin, B2M) was achieved in the closed reactors at 50 °C for 40 min. Then, the array was subjected to 50 rounds of thermocycling to amplify the transcribed cDNA and hydrolyze the quenched fluorophore from the labeled probes

(Figure 4.1a). The array was imaged to detect the fluorescent signals evolved from the digested probes (Figure 4.1b). The images were analyzed using a custom program to determine the location of each well, the number of cells per well, and the fluorescence intensities of both the released probe and reference dye. These data were then filtered to discard wells with more than four cells and wells with a large coefficient of variation in the soluble reference signal (ROX).

To normalize for regional variations of the measured intensities, we calculated the relative fluorescence as the ratio of the gene-specific signal (Iwell) to the mean of the gene-specific signal of nearby empty wells (Iempty) (Figure 4.1c, top).

To determine the threshold value for a positive RT-PCR reaction, we fit the relative fluorescence of the wells containing no cells to a single Gaussian distribution to obtain estimates for the mean and standard deviation of the peak representing negative reactions (0.96±0.12). We defined positive reactions as those wells containing cells with a ratio greater than three standard deviations above the mean ratio determined for empty wells. The percentage of positive events

scored in control experiments in which either the primers, probe, or reverse transcriptase were

excluded was less than 0.01% (Figure 4.1c). The lack of positive events scored upon omission of

reverse transcriptase from the reaction indicates that the genomic DNA was not amplified, and

implies that it is not necessary to remove residual genomic DNA from the reaction when using

intron-spanning primers. Digestion of the gene-specific probe with DNase I in the reaction mixture prior to application to an array without cells yielded a measured ratio of 2.65±0.08 (data not shown). This experiment, in combination with the cell-based experiments, suggested that the maximum relative fluorescence for a positive reaction is about 2.7, and that 50 rounds of thermocycling were sufficient to obtain this endpoint.

Figure 4.1 (a) Schematic of method for parallel single-cell RT-PCR reactions in nanowells. Cells are deposited in nanowells, filled with a solution of components for RT-PCR, and then sealed to a glass slide. The thermal lysis, first strand synthesis, and amplification of cDNA are conducted on a thermocycler. The fluorescence intensity of cleaved probes is detected by epifluorescent microscopy. (b) Fluorescent micrographs of gene-specific (B2M) and a reference signal (ROX) confined in individual, sealed nanowells. (c) Histogram of the relative fluorescence of wells that contain cells. Positive reactions have a relative fluorescence greater than 1.4.

Next, we determined the sensitivity and specificity of the method using three genes that are commonly employed as standards for RT-qPCR (B2M, glyceraldehyde 3-phosphate dehydrogenase (GAPDH), beta-actin (ACTB)), as well as the heavy chain of the antibody produced by the 4D20 hybridomas (HC) (Figure 4.2). The threshold values for positive reactions were determined for all four genes. Based on the maximum threshold of 1.4, the sensitivity and specificity of the assay were greater than 84% and 98%, respectively. The positive predictive

value, which indicates the confidence in the assignments, was greater than 95%. It is expected

that the sensitivity will be lower for one-step, single-cell RT-PCR than for RT-PCR using bulk purified mRNA since the single reaction does not individually optimize the release of mRNA

Figure 4.2. Detection of mRNA transcripts of constitutively expressed genes in 4D20 cells. Boxplot of Iwell/Iempty for four genes (ACTB, GAPDH, B2M, and HC). The boxplot follows Tukey’s convention. The median is marked with a red line, and the upper and lower edges of the box indicate the values of the upper and lower quartiles. Notches on the box adjacent to the median value represent its 5% significance level. Whiskers extending from each end of the box represent extreme values within 1.5 times the interquartile range. The numbers of wells included in each box are indicated below each one. The red dashed line indicates the minimum value for positive reactions used for all four genes. from cells, or the subsequent RT and PCR steps. We also note that the apparent sensitivity could be lower than determined: it is possible that a small fraction of the cells were not expressing the target gene at the time of the assay.

4.2. Integration with microengraving

For integrated single-cell analysis of both gene expression and secretory phenotypes, the method described here can also be combined with other nanowell-based techniques such as imaging cytometry and microengraving—a technique for quantifying the frequencies and rates of secretion of proteins for populations of single cells85, 86. We sought to determine if the detection of transcripts for HC in the hybridomas correlated with antibody secretion in the period of time immediately beforehand. To examine this relationship between transcribed genes and secreted proteins, 4D20 cells were labeled with a live cell marker, loaded into nanowells, and imaged to quantify the number of cells in each well. The array containing cells was then sealed with a functionalized glass slide to capture secreted antibodies by

Figure 4.3. Integrated single-cell analysis of gene expression and secreted antibodies from human B cell hybridomas. 4D20 cells were labeled with a live cell stain (Celltracker Violet) and interrogated for IgG1 secretion and heavy chain mRNA. Sample images of correlated data for representative phenotypes are shown (left). The relative fluorescence of the RT-PCR is false colored from red-orange (no reaction) to green (positive reaction). Positive IgG1 secretion is false colored red. The graphic profile (right) shows the distribution of phenotypes measured. The area of each circle is proportional to the number of each phenotype enumerated.

microengraving85. After two hours, the glass slide was removed and probed for captured antibodies, while the cells in the nanowells were then subjected to on-chip RT-PCR to detect

HC mRNA (Figure 4.3). Out of 6,086 wells with single cells, 5,392 cells (88.6%) expressed the heavy chain mRNA, but only 1,795 cells (29.5%) secreted IgG1 during the preceding period of time. Most of the cells secreting IgG1 also had detectable transcripts (90%).

These data provide direct evidence that analyzing transcribed genes alone does not necessarily provide a suitable surrogate for complex functional activities such as secretion.

Chapter 5. Identification of target cells in large populations

5.1. Cytometry

To investigate what types of cells were infected by HIV, we used imaging cytometry to

determine the surface markers on PBMCs. The markers that we were interested in identifying

were CD4, CCR7, CD45RA, CD122, CD95, CD14, HLA-DR, and CD11c (Table 5.1). The markers were split into two panels, one for CD3+ cells and one for CD3– cells, to reduce

crosstalk between the fluorescence channels when imaging on the microscope. After imaging,

the normalized fluorescence intensities from the surface markers were plotted on histograms,

similar to the RT-PCR data (Figure 5.1). The histograms from the microscopy data showed

similar results as a flow cytometer. We observed that some markers were more bimodal (e.g.,

CD4) and other markers had a more continuous spectrum for the fluorescence signal (CCR7).

Using the threshold of the mean plus three standard deviations of the negative peak as the

Table 5.1 Surface markers for PBMC classification

Alexa Alexa Alexa PerCP- Fluorescent Fluor Fluor Fluor eFluor dye: 488 568 647 PE/Cy7 710 CD3+ panel: CD95 CD4 CD45RA CCR7 CD122 Cell type CD3– panel: CD11c CD4 CD14 HLA-DR Effector memory T CD3+ + – – Central memory T CD3+ + – + Naïve T CD3+ – + + + – Stem cell like memory T CD3+ + + + + + Cytotoxic T CD3+ – Monocytes CD3– + Dendritic cells CD3– + + Other APC CD3– +

threshold for a positive signal, we obtained very similar results for classifying the CD3+

populations of T-cells as the flow cytometer (Figure 5.2). The CD95 histogram was unusually broad with a mean of 0.03, so this marker was not considered for identifying the stem cell like memory T-cells; only CD122 was used. The thresholds for the other markers were bunched closely together around 0.03 to 0.04, and if we used these cutoffs for CD95, the percentage of

CD122+CD95+ cells were slightly lower (~10%) than just CD122+. To match the flow

percentages with the microscopy percentages, our threshold for CCR7 should be slightly higher

and CD45RA slightly lower.

6000 4000 CD11c CD95 CD4 3500 CD4 5000 CD14 CD45RA HLA-DR CCR7 3000 CD122 4000 2500

3000 2000

1500 Frequency 2000

1000

1000 500

0 0 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3

Relative fluorescence intensity

Figure 5.1. Histograms of surface marker fluorescence on PBMCs. The mean plus 3 standard deviations from the negative peak were used to determine the threshold between positive and negative in each marker. This cutoff corresponds to a statistical false positive rate of 0.13%.

Figure 5.2. Comparison between flow cytometry and microscopy for T-cell classification. The flow cytometry plots were plotted for CCR7 vs. CD45RA and CD95 vs. CD122. The purple lines indicated the thresholds and the purple numbers represented the fraction of cells in the quadrant. Black numbers near the purple numbers were the results from microscopy.

5.2. Activation of cells

To detect copies of mRNA in a cell, an activation protocol was used to boost the

production of virus so that the mRNA would be more likely detected by RT-PCR. The goal of

the activation was to increase mRNA copies within the cells without killing the cells since dead

cells often released their mRNAs into the supernatant. We also wanted the protocol for activation

to be as short as possible. We stimulated ACH2 cells with phorbol 12-myristate 13-acetate with

ionomycin (PMA/iono), tumor necrosis factor alpha (TNFa), and suberoyl anilide hydroxamic

acid (SAHA, vorinostat) at two concentrations (1x and 10x) for a series of times (6 hr, 24 hr and

48 hr). The 1x concentrations of the activators were 1 ng/mL TNFa, 1 μM SAHA, and 1 ng/mL

PMA with 1 μM iono, and the 10x concentrations were used as a potential way to decrease the

time needed for activation. For the 10x activation condition, TNFa and PMA/ionomycin

treatments at 24 hr were the best activation conditions to induce HIVgag mRNA production in

20 1 18 0.9 16 0.8 14 0.7 12 0.6 10 0.5 8 0.4

6 Cell viability 0.3 4 0.2 0.1 HIVgag (fold difference) 2 0 0

Activation conditions (10x concentration) Activation conditions (10x concentration)

Figure 5.3. Effect of 10x activation conditions on HIVgag and cell viability in ACH2 cells. RT- qPCR was used to determine the relative fold differences in HIVgag levels (left). All results were relative to the ACH2 cells that were not activated. Cell viability was measured by trypan blue staining (right). TNFa was the best stimulus for maintaining cell viability over 48 hr.

cells (Figure 5.3). While PMA/iono was similar to TNFa in activating HIV production,

12 1 10 0.9 8 0.8 6 0.7 0.6 4 0.5 2

HIVgag (fold difference) 0.4

0 Cell viability 0.3 0.2 0.1 0 Unactivated TNFa 6hr, TNFa 12 hr, TNFa 6 hr, TNFa 12 hr, ACH2 10 ng/mL 10 ng/mL 1 ng/mL 1 ng/mL Figure 5.4. Comparison of 1x TNFa and 10x TNFa activations on ACH2. RT-qPCR was used to determine the relative fold differences in HIVgag levels (left). All results were relative to the resting ACH2 cells. No significant difference was detected between the two concentrations. The 1x may even be better than the 10x. Cell viability was measured by trypan blue staining (right). Cell viability did not drop significantly over 12 hr.

PMA/iono activated cells were dying more than TNFa activated cells over time. Further

experiments with TNFa showed that the HIVgag signal activated with 1x TNFa was similar to

that of the 10x TNFa and cell viability did not decrease much from the cells that were not

activated (Figure 5.4). These experiments also showed that after three washes, the virus in the

supernatant of the cell suspension was at a low enough concentration that it was not detectable

by RT-qPCR.

After determining that the best activation condition by bulk RT-qPCR measurements was

1 ng/mL TNFa, we wanted to verify that it was still the best condition for nanowells by single-

cell RT-PCR. For a direct comparison, we activated ACH2 cells in the nanowells for 6 hr with

10x TNFa or 24 hr with 1x TNFa (Figure 5.5). HIV-negative PBMCs were added to simulate a

patient sample. We observed significantly more positive reactions in wells with cells that were

activated with 1x TNFa for 24 hr than with 10x TNFa for 6 hr. While both cases had positive

reactions in wells with no cells, the 10x TNFa activation had a large portion of wells with no

Table 5.2 Effect of activation conditions on ACH2 cell detection by RT-PCR in nanowells

Activation condition Positive wells Positive wells Chi-squared of ACH2 with ACH2 (%) with no ACH2 (%) p-value 1 ng/mL, 24 hr 38.3 2.36 <0.0001 10 ng/mL, 6 hr 14.3 6.45 <0.0001 Negative (PBMC only) 0.0474 1 ng/mL, 24 hr 21.9 1.42 <0.0001 1 ng/mL, 18 hr 18.7 0.418 <0.0001 0 ng/mL, 6 hr 2.80 0.423 <0.0001 Negative (PBMC only) 0.0375

cells that were positive. This result suggested that the mRNA that was induced by the activation

was not contained within the cell. The exact source of the mRNA was not known, but possible

sources include budded virus particles or dead cells. The negative control (PBMC) had a false

positive rate of 0.047% (40 positive wells on the entire array).

The 1 ng/mL TNFa activation for 24 hr still had a considerable number of positive

signals in wells with no cells. We experimented with shorter activation times to see if we could

reduce the number of positive signals in wells with no cells (Table 5.2). We observed that there was an approximately 3-fold drop in the fraction of wells with no cells that had a positive signal in the 18 hr and resting ACH2 cells. The percent of cells that were activated also dropped in the

18 hr activation compared to the 24 hr activation, and this drop was 3 percentage points (15% decrease). We also noticed that there was a drop between the experiments on different days for the 24 hr activation (38.3% to 21.9%), and this 43% decrease could be from the degradation of

TNFa by the extra freeze thaw in the later experiment or day-to-day variability. In all cases, the detected positive signal was from mRNA in a cell since the Chi-squared contingency tests all had p-values less than 0.0001. 64

0.35 1 ng/mL TNFa, 24 hr, No cell 1 ng/mL TNFa, 24 hr, Cell 0.3 10 ng/mL TNFa, 6 hr, No cell 10 ng/mL TNFa, 6 hr, Cell 0.25 PBMC, No cell PBMC, Cell

0.2

0.15 Normalized frequency 0.1

0.05

0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 ROX normalized intensity

Figure 5.5. Comparison of 1x TNFa and 10x TNFa activations on ACH2 in nanowells. ACH2 cells were activated for either 24 hr at 1 ng/mL TNFa or 6 hr at 10 ng/mL in nanowells. PBMCs were added after the activation to act as a negative control.

Figure 5.6. Effect of activation time on ACH2 in nanowells. ACH2 cells were activated with 1 ng/mL TNFa for 24 hr, 18 hr, and 0 hr (no activation). Only the histograms of the wells with cells were plotted.

Table 5.3 Detection of serial dilutions of ACH2 cells in nanowells

Serial dilution of Wells with Positive wells Positive wells Positive wells activated ACH2 cells ACH2 with ACH2 with ACH2 (%) with no ACH2 (%) Dilution 1:100, 24 hr 2662 1149 43.16 8.93 Dilution 1:500, 24 hr 427 113 26.46 1.12 Dilution 1:5000, 24 hr 45 17 37.78 0.0901 Negative (PBMC only) 0.0291 Dilution 1:200, 18 hr 1666 751 45.1 10.85 Dilution 1:2000, 18 hr 208 78 37.5 0.82 Dilution 1:10000, 18 hr 25 5 20.0 0.47 Negative (no PBMC) 0.0677 Neg. PBMC HIVGag FAM 0.37 Neg. PBMC 1LTR Q670 0.62 Neg. PBMC FAM+Q670 0.0046

5.3. Limit of detection of cells

To determine the limit of detection of HIV infected cells, we activated a serial dilution of

ACH2 cells (1:100, 1:200, 1:500, 1:2000, 1:5000 and 1:10000) in nanowells. After the 18 hr or

24 hr activation, HIV-negative PBMCs were added to the array before running RT-PCR. The negative controls (PBMCs only) had 23 and 47 bright wells in the arrays and these numbers correspond to a false positive rate of 2.91x10-4 and 6.77x10-4, respectively (Table 5.3). Since the false positive rates were so low, we would need at least 1400 positive wells with a cell to expect that one of those wells was a false positive well. Therefore, if the number of events in the array were less than 1400, we would only count the wells with cells in them. Again, the p-values from

Chi-squared contingency tests were all less than 0.0001, and showed that the positive signals in wells with cells were from cells that had internal HIV rather than a well with cell and contaminating HIV.

As a method to further reduce the number of false positives, we tested the use of two

probes in different fluorescence channels, specifically 1LTR Q670 and gag FAM as the two

probes to detect HIV. We found that the false positive rate for gag FAM and 1LTR Q670

individually were approximately 0.37% and 0.62%, respectively, but the false positive rate for wells that were positive in both channel was 0.0046%. The false positive rate for wells that were double positive was approximately 100 fold improvement over the individual rates.

The number of cells that were in wells followed the expected serial dilutions, but the number of cells that were detected by RT-PCR varied. The percentage of wells with ACH2 cells that were positive ranged from 20 to 45% of all wells with ACH2 cells. Since values were all greater than the percentage of positive wells in the entire chip, the positive wells with ACH2 cells were not from random distribution of positive signal on the array. Interestingly, the overall positive percentages fit the serial dilutions better than the positive wells with ACH2 cells. This observation could be because RT-PCR detected the virus particles produced by the ACH2 cells during activation that went into a random well. The serial dilution experiment also showed that we could detect very few positive wells on the array with RT-PCR (1 in 10000).

5.4. HIV-positive patient sample

Since we showed that we could detect less than 10 positive wells with cells, we

proceeded with a preliminary study on T-cells from an HIV-positive patient. The patient had

been treated with HAART for more than 1.5 years. A side-by-side comparison between RT- qPCR on a bulk sample of 1 million cells and RT-PCR in nanowells was performed. Comparing the relative number of viral RNA in the bulk sample, we observed that the infected T-cells after

activation with PMA/ionomycin for 18 hr had more RNA than infected T-cells that were not activated (Figure 5.7).

For RT-PCR in nanowells, the T-cells were run in a staggered manner so that the unused

cells were resting in normal R15 media. One set of arrays were run after 18 hr activation, 18 hr

activation plus 4 hr rest, and 18 hr activation plus 7 hour rest (Table 5.4). To further reduce the

probability of a false positive result, we used both FAM and Q670 probes for HIVgag and

required that a well had to be double positive in both channels to be considered a well with a

positive signal. With this extra criterion, we determined that the HIV-infected population of T-

cells had a frequency of 1:2000 to 1:13000, depending on the condition. The additional 4 to 7 hr

rest times appeared to increase the frequency of positive events. While the exact reason for this

increased frequency was unknown, it could be explained by the sticking of virus to the T-cell or

endocytosis of the virus by the T-cell. This basal level of virus had been observed after

immediate washing of the T-cells87. There were also fewer events in the PMA/ionomycin

activated cells, even though this activation was much more potent than panobinostat. This

discrepancy could be a sampling issue since only 1 array was used for PMA/ionomycin in each

set whereas panobinostat had 3 arrays.

Figure 5.7. RT-qPCR on bulk cells from HIV-positive sample. RT-qPCR was performed on 1x106 cells to determine the relative number of viral RNA in each sample.

Table 5.4 RT-PCR on HIV-positive T-cells Corrected Wells Double Double ratio of HIV Total with Total positive, positive, infected Condition wells cells cells cell no cell cells 18 hr act, pano, HIV T-cell 86578 13103 13637 4 0 1:3505 18 hr act, pano, HIV T-cell 86600 11873 12318 7 5 1:6143 18 hr act, pano, HIV T-cell 86528 12638 13295 3 11 1:13024 18 hr act, PMA/iono, HIV T-cell 86309 16430 17862 0 4 18 hr act + 4 hr rest, pano, HIV T-cell 85332 15650 16348 14 35 1:2824 18 hr act + 4 hr rest, pano, HIV T-cell 86499 15259 15995 0 3 18 hr act + 4 hr rest, pano, HIV T-cell 82912 15512 16304 12 17 1:2067 18 hr act + 4 hr rest, PMA/iono, HIV T-cell 84483 22165 24454 3 0 1:8151 18 hr act + 7 hr rest, pano, HIV T-cell 77631 16498 17596 12 15 1:2290 18 hr act + 7 hr rest, PMA/iono, HIV T-cell 79249 37128 46374 14 1 1:3595 25 hr rest, no act, HIV T-cell 57895 39087 53097 26 3 1:3028 23 hr act, TNFa, ACH2 51303 1967 1983 81 1214 1:24

5.5. Discussion

We had shown that imaging cytometry by microscopy was possible and comparable to flow cytometry. Each of the two panels of surface markers gave reasonable percentages for the classifications of PBMCs. In flow cytometry, we drew gates depending on the background

fluorescence of a negative sample before the actual sample was run, and in microscopy, the thresholds were derived from the mean intensity of the negative sample plus three standard deviations after measuring the values of the sample. The three standard deviations correspond to a false positive rate of 0.13%. In spite of not having a powerful laser for the microscope, this threshold worked well even for markers (e.g., CCR7) that had a continuous level of display on the cell surface.

In the patient sample, the expected frequency of HIV-infected T-cells was 1:1000088. The

observed frequencies by RT-PCR in nanowells were as much as 5 times higher than expectation.

This observation could be explained by the production of new virus particles. HIV life cycle

models have shown that the time from integration of HIV into the genome to new production of

virus particles was 7-17 hr89, 90. Therefore, during the 18 hr activation period, new virus particles

could have been produced and released into the supernatant to attach to uninfected cells. More

viral production would occur during the 4 and 7 hr resting period between the sets of RT-PCR.

Detection of these virus particles by RT-PCR in nanowells with cells would result in a higher

frequency of infection. The production of virus could also explain the correlation of overall

positive signal in the serial dilution of ACH2 cells to the dilution amount.

The resting HIV-infected T-cells were our negative control since it had such a low

number of RNA in the bulk RT-qPCR. But with RT-PCR in nanowells, we could detect a similar

frequency in this resting sample as in the activated samples. This result was not entirely

surprising since all of the cells were from one blood draw, so the frequencies of all the conditions

should be similar with a sensitive assay. While we did not have a true negative control (i.e., HIV-

negative PBMCs) run with the patient sample, our previous negative controls showed that the false positive rate was less than 4.6x10-5 using two probes.

These preliminary results also showed that we could detect some cells that were infected

in the resting sample. These resting cells could have been producing virus at low levels (e.g., not

latently infected) or could have spontaneously reactivated (may have been latently infected) and

reflect the residual viremia in patients even on HAART6, 91. The precise case could not be

determined by the assay. After activation, a portion of the latently infected cells would reactivate

so that the HIV mRNA would be detectable. Our one experiment showed that there was an

increase in the frequency of infection when the cells were activated by panobinostat, but further

experiments would be needed to draw any clear conclusions.

Chapter 6. Other methods to detect transcript

6.1. Surface capture of transcripts

To overcome the limitations of RT-PCR in nanowells, we wanted to capture the information from single cells onto a glass slide in a stable manner. The format would be cDNA since it is more stable than mRNA and the attachment of cDNA to the glass would be through a covalent bond so that potential downstream applications such as PCR would not disrupt the bond.

One covalent bond that was easily introduced and well-studied for microarrays was the epoxy- amine bond. Our goal was to detect the transcripts expressed in the cells of each well by probing with fluorescently labelled DNA.

First, we established the theoretical feasibility of this approach with some estimates. A typical mammalian cell contains 10-30 pg of total RNA of which 1-5% is mRNA, so one cell has approximately 0.6 pg of mRNA (20 pg RNA, 3% mRNA) (Invitrogen oligo-dT bead product insert). The average number of nucleotides in mRNA is approximately 2,200 bases plus 250 adenosines in the poly-A tail, so the average molecular weight of mRNA is 788,000 g/mole.

Using these values, one cell has on the order of 500,000 strands of mRNA (maximum of ~1 million strands). The footprint of the nanowell on the glass slide is 50 μm by 50 μm, or 2500 μm2.

The maximum density of amine groups on a glass surface is about 4 amine/nm2, which is approximately 109 amine per well92, and a more typical oligomers surface count is 2.5x108 per well93. Therefore, more than 200 fold excess of capture probes are on the 2500 μm2 area for every strand of mRNA in a single cell.

Figure 6.1. Schematic of mRNA capture on a glass surface. Cleaned glass slides were first conjugated with epoxy groups that react with amine-oligo-dT primers before they were sealed to an array of nanowells containing cells with a reaction mixture for RT. The RT reaction was run at 50 °C for 4 hours to lyse the cell and synthesize the first-strand cDNA. These cDNA can be detected by sequential probing with fluorescence probes.

With feasibility shown, the experiment was designed as follows. First, a batch of epoxy-

modified slides was manufactured (Figure 6.1). The primer mix was composed of 100 µM oligo- dT30 to capture mRNA and 1 µM kanamycin resistance (KanR) primer as a uniformity control.

Similar to RT-PCR in nanowells, 4D20 cells were dispersed onto the array, Superscript III RT

solution containing NP-40, dNTPs, and RNase inhibitor were applied onto the array, and the 74

Figure 6.2. Sample scans of B2M cDNA (FAM, left) and KanR cDNA (HEX, right). The numbers at the upper left corner of each spot indicates the number of cells in the well. Every bright spot had at least one cell in the corresponding well.

primer-conjugated glass was pressed onto the array. The device was sealed shut in a hybridization clamp and incubated at 50 °C for 4 hours. After the RT reaction, the glass was probed for B2M and poly-T or KanR.

We observed that every spot with a brighter B2M signal had a cell, but interestingly, not every well with a cell was bright (Figure 6.2, left). These results suggested that the signal we detected was not entirely from nonspecific binding of probes to cellular debris and that reverse transcription was occurring at the surface of the glass. While wells with more cells generally had stronger signal, this rule was not universally applicable. It may be that the surface capture efficiency was not uniform throughout the slide, the cells actually had less mRNA, or the RT efficiency was just lower in that well. The uniformity control of the fluorescence signal (KanR) showed that across the array, the primers were distributed evenly (Figure 6.2, right). Wells with cells, however, still had a brighter signal than wells without cells. Curiously, we observed that

wells with stronger B2M signal also had stronger KanR signal, suggesting that bleed-through from the FAM channels may account for the variation. Similar results were observed when poly-

T was the target.

Figure 6.3. Images of swapping dyes in two sets of probes. MutPGK1 cells were labelled with Celltracker Red and deposited onto a chip and sealed with a capture slide for cell lysis and RT. The captured cDNA on the glass was probed with either HEX labelled T491 (left) or 665 labelled T491 (right) and imaged in the HEX channel.

Figure 6.4. Captured cDNA on glass from unlabeled mutPGK1. The glass was probed with either HEX-T491 and 665-A491 (left) or HEX-A491 and 665-T491 (right). The images show the ratio of the HEX (green) and 665 (red) signals. The signal from the captured cDNA is much weaker when the cells are not labelled with Celltracker Red, but the signal is specific for the correct labelled probe.

To further characterize the specificity of the assay, we targeted the PGK1 gene in a normal cell line (4D20) and a mutated cell line (mutPGK1). The mutation is homozygous in mutPGK1 where the 491st nucleotide is substituted with T (T491) instead of the normal A491. 77

To identify the cells in the wells and determine well occupancy, mutPGK1 cells were labelled with Celltracker Red, and imaged before the lysis and RT steps. After the RT step, the slide was probed with either HEX-labelled T491 or 665-labelled T491 (Figure 6.3). We observed that the

HEX channel was always brighter, independent of the fluorophore used. One reason could be that the Celltracker Red in the cell debris was also stuck on the glass and this signal was stronger than the fluorophores on the probes. To test this hypothesis, cells not labelled with Celltracker were used and indeed the signal dropped (Figure 6.4). This result confirmed that cell debris that was labelled by the Celltracker dyes were sticking to the glass surface and masking the specific signal from the probes. This could also explain why the spots with cells also had higher KanR and oligo-dT signal since these probes all had the same fluorescent dye. We did observe that the weak signal was correct for identifying the single nucleotide polymorphism. Since the highly transcribed gene PGK1 had such a low signal, this method would not be feasible to detect low copies of transcript that may be in latently infected cells without some amplification step.

6.2. Hybridization chain reaction

One possible amplification step was to use hybridization chain reaction (HCR). In this method, an initiator sequence was attached to the probe instead of a fluorescent dye. This initiator sequence would trigger the opening and attachment of a pair of hairpins that have been fluorescently labeled. Thus, a large number of fluorescent molecules could be attached to a single binding event, allowing for more fluorescence signal to be detected. In addition, multiple hairpin and initiator sets have been designed to have specific amplification of signal to different targets with minimal signal bleedthrough68.

Figure 6.5. Comparison of HCR with direct detection in 4D20 cells. We observed strong signal over background when using HCR to amplify the signal (left). Detection of B2M mRNA with 10 fluorescent probes had high background and less than half of the array was useable (right).

Figure 6.6. Combination of two sets of HCR to detect single nucleotide polymorphism. In both figures, the normal phenotype A491 was conjugated with a TYE563 dye (false colored green) and the mutant phenotype T491 was conjugated with TYE665 dye (false colored red). (Left) Normal 4D20 cells had more T491 signal than A491, and (right) mutant PGK1 cells had more A491 signal than T491. Both sets had a lot of background noise.

To test the feasibility of HCR as an amplification strategy, we compared the background and signal quality over a direct detection of 10 probes. This experiment showed that HCR generated a strong signal above the background noise (Figure 6.5). The image quality was

observed throughout the array in HCR whereas less than half the array was useable when detecting with a set of 10 probes.

We proceeded to use HCR to detect the single nucleotide polymorphism in PGK1 between the normal 4D20 cells and mutant PGK1 cells. Here, each array was loaded with either

4D20 cells or mutant PGK1 cells and both sets of hairpins and initiator conjugated probes were used. We found, oddly, that the normal 4D20 cells were brighter in the mutant signal and the mutant PGK1 cells were bright in both (Figure 6.6). We also observed that both arrays had noisier background than just the single set of hairpins and probe. These higher background intensities and swapped signal suggested that there was a lot of random sticking to the surface of the glass by the HCR hairpins and/or probes. Since HCR produced such a strong signal, any random sticking or insufficient washing would give false positive signal. We were also uncertain if the epoxy on the surface was completely pacified to prevent nonspecific sticking.

Figure 6.7. Capture of ACTB on PDMS followed by RT-PCR in nanowells. The pre-image (left) showing the location of the cells was matched to the RT-PCR results for ACTB (right). The wells that matched properly are boxed in green. Wells with cells but no signal are boxed in orange (left) and wells with no cells but with signal are boxed in orange (right).

Since the capture of transcripts was possible, we tried a second approach that covalently attached oligo-dT to the surface of the PDMS so that cells could be lysed with a more powerful buffer before running RT-PCR. We found that after adding DNase and proteinase K to the MES lysis buffer, a lot of the nonspecific signal could be removed. However, there were still unmatched cells and wells with positive signal in the RT-PCR (Figure 6.7).

6.3. Discussion

The goal of surface capture was to allow for a two-step procedure to improve the sensitivity and specificity of RT-PCR. By capturing the transcripts onto a surface, we would be able to remove cellular debris and other contaminations from the reaction and recover the 1.4 transcript sensitivity of the reaction in nanowells. This method could also allow us to capture

transcripts in the nucleus that we were previously unable to access because of weaker lysing

conditions.

We have shown that surface capture of mRNA onto a glass slide is possible as well as the

reverse transcription of the transcript to cDNA. These bound transcripts could then be detected

with fluorescent probes and the fluorescent signal amplified with HCR. However, the signal was

also convoluted by the presence of cellular debris such as proteins and genomic DNA. If the

debris were also fluorescently labeled by Celltracker, the debris signal was overwhelming the probe signal on bound transcripts. Removing these labels did allow the detection of specific single nucleotide polymorphisms on the glass, but the signal was very weak. It was not clear if the low signal was from insufficient capture of the target transcript. We modified the surface chemistry to link the oligo-dT to the PDMS walls so that a larger surface area (i.e., 5-fold) was

available to capture the mRNA. The results from the modified surface were not better than

running RT-PCR in the nanowells directly on the cells, even for the detection of highly

transcribed genes. One possibility for the additional false positive and false negative signals was

that the transcripts were not completely bound to the surface after the cells were lysed and the

cover glass was removed. This would allow the mRNA to freely diffuse to other wells from a

well previously containing cells. More work would be needed to understand the cause of these

issues.

Chapter 7. Conclusions

In this thesis, we have developed and optimized a simple RT-PCR assay to detect the gene expression in tens of thousands of individual cells in parallel with high sensitivity and specificity in an array of nanowells. By confining these reactions to small 125 pL wells, we showed that the limit of detection was approximately 1.4 copies of DNA in nanowells and that the false positive rate was as low as 0.000046 or 1 in 22000 events. By combining with microengraving and image-based cytometry on a cell line that produces antibodies, we showed that most cells had the requisite transcripts for their antibody, but only a subset of those cells secreted the antibody.

We have also applied the technology to detect HIV-infected cells in peripheral blood mononuclear cells. By using our technology, we are better suited to attain the infection rate in an

HIV-infected sample. Previous studies on infection rate used bulk measurements with either RT- qPCR or digital PCR to attain a measure on how much RNA or DNA was in the sample. These measurement, however, do not accurately represent the number of infected cells in the sample because the measured number of transcripts lacked single cell resolution. Our technology is capable of detecting single cell events by isolating the cells in individual nanowells. A second advantage of our technology is that it can be easily integrated with other processes that can be performed in nanowells such as image-based cytometry and microengraving. The cells can be first imaged by cytometry to determine the identity of the cells in each well and then measured for interesting functional phenotypes such as secretion by microengraving before performing

RT-PCR on them. This combination provides a multivariate and direct measure of the

relationships between the presence of transcribed genes, the cell surface markers, and functional

cellular activities for many individual cells. To acquire this variety of data without our technology, we would need to first isolate the cells by the desired surface markers, assess the secretions of each population, and then determine the RNA content. All of these measurements lack single cell resolution and it would not be possible to link secretion with cellular transcripts in one cell.

Despite the advantages of RT-PCR in nanowells, there are still areas for improvements.

Some disadvantages of the current approach are that the measures of gene expression are not

quantitative, the cell nucleus cannot be efficiently accessed, and the number of fluorescent labels

that can be distinguished distinctly (~4-6 for most fluorescent microscopes) will limit the number

of transcripts detected per cell. Capture of DNA and RNA transcripts on the walls or a glass slide

could also potentially solutions to the first two challenges. These captured transcripts could be

attained with more stringent lysis conditions that would obliterate the cell nuclei and be detected

by fluorescent in situ hybridization for counting of positive spots. We have begun examining

these possibilities and observed that significant blocking of nonspecific binding was needed to

use surface capture and amplification of signal. By using advanced detection methods such as

time-of-flight (TOF) mass spectrometry, the limitations of fluorescent labels could be addressed.

Combinations of TOF mass spectrometry and amplification techniques could potentially be used

to detect small amounts of transcripts in a cell.

In summary, the one-step RT-PCR assay described in this thesis provides a powerful method to determine the presence and absence of transcripts. The technology can be used to identify HIV-infected cells and their surface markers when linked with cell cytometry. This

combination of multiple assays on one array of nanowells allows for a simple and efficient method to better understand the cells associated with HIV latency. When combined with microengraving, our approach can evaluate relationships between the transcription of genes and the secretion of the translated products—a useful intersection to evaluate the suitability of surrogate markers for monitoring clonal production in biomanufacturing or clinical factors in diagnostics.

References

1. UNAIDS. Global report: UNAIDS report on the global AIDS epidemic 2013. URL: http://www.unaids.org/en/media/unaids/contentassets/documents/epidemiology/2013/gr2 013/UNAIDS_Global_Report_2013_en.pdf. 2. Carr, A. Toxicity of antiretroviral therapy and implications for drug development. Nat Rev Drug Discov 2, 624-34 (2003). 3. May, M. et al. Impact of late diagnosis and treatment on life expectancy in people with HIV-1: UK Collaborative HIV Cohort (UK CHIC) Study. BMJ 343, d6016 (2011). 4. Mills, E.J. et al. Adherence to HAART: a systematic review of developed and developing nation patient-reported barriers and facilitators. PLoS Med 3, e438 (2006). 5. Wong, J.K. et al. Recovery of replication-competent HIV despite prolonged suppression of plasma viremia. Science 278, 1291-5 (1997). 6. Palmer, S. et al. Low-level viremia persists for at least 7 years in patients on suppressive antiretroviral therapy. Proc Natl Acad Sci U S A 105, 3879-84 (2008). 7. Chun, T.W. et al. Relationship between pre-existing viral reservoirs and the re-emergence of plasma viremia after discontinuation of highly active anti-retroviral therapy. Nat Med 6, 757-61 (2000). 8. Davey, R.T., Jr. et al. HIV-1 and T cell dynamics after interruption of highly active antiretroviral therapy (HAART) in patients with a history of sustained viral suppression. Proc Natl Acad Sci U S A 96, 15109-14 (1999). 9. Chun, T.W. et al. In vivo fate of HIV-1-infected T cells: quantitative analysis of the transition to stable latency. Nat Med 1, 1284-90 (1995). 10. Chomont, N. et al. HIV reservoir size and persistence are driven by T cell survival and homeostatic proliferation. Nat Med 15, 893-900 (2009). 11. Coleman, C.M. & Wu, L. HIV interactions with monocytes and dendritic cells: viral latency and reservoirs. Retrovirology 6, 51 (2009). 12. Williams, S.A. et al. NF-kappaB p50 promotes HIV latency through HDAC recruitment and repression of transcriptional initiation. EMBO J 25, 139-49 (2006). 13. Siliciano, J.D. et al. Long-term follow-up studies confirm the stability of the latent reservoir for HIV-1 in resting CD4+ T cells. Nat Med 9, 727-8 (2003). 14. Finzi, D. et al. Latent infection of CD4+ T cells provides a mechanism for lifelong persistence of HIV-1, even in patients on effective combination therapy. Nat Med 5, 512- 7 (1999). 15. Sarkar, I., Hauber, I., Hauber, J. & Buchholz, F. HIV-1 proviral DNA excision using an evolved recombinase. Science 316, 1912-5 (2007). 16. Hauber, I. et al. Highly significant antiviral activity of HIV-1 LTR-specific tre- recombinase in humanized mice. PLoS Pathog 9, e1003587 (2013). 17. Ho, Y.C. et al. Replication-competent noninduced proviruses in the latent reservoir increase barrier to HIV-1 cure. Cell 155, 540-51 (2013). 18. Chun, T.W. et al. Presence of an inducible HIV-1 latent reservoir during highly active antiretroviral therapy. Proc Natl Acad Sci U S A 94, 13193-7 (1997). 19. Bosque, A. & Planelles, V. Induction of HIV-1 latency and reactivation in primary memory CD4+ T cells. Blood 113, 58-65 (2009). 87

20. Marini, A., Harper, J.M. & Romerio, F. An in vitro system to model the establishment and reactivation of HIV-1 latency. J Immunol 181, 7713-20 (2008). 21. Duh, E.J., Maury, W.J., Folks, T.M., Fauci, A.S. & Rabson, A.B. Tumor necrosis factor alpha activates human immunodeficiency virus type 1 through induction of nuclear factor binding to the NF-kappa B sites in the long terminal repeat. Proc Natl Acad Sci U S A 86, 5974-8 (1989). 22. Jordan, A., Bisgrove, D. & Verdin, E. HIV reproducibly establishes a latent infection after acute infection of T cells in vitro. EMBO J 22, 1868-77 (2003). 23. Schroder, A.R. et al. HIV-1 integration in the human genome favors active genes and local hotspots. Cell 110, 521-9 (2002). 24. Brady, T. et al. HIV integration site distributions in resting and activated CD4+ T cells infected in culture. AIDS 23, 1461-71 (2009). 25. Wang, G.P., Ciuffi, A., Leipzig, J., Berry, C.C. & Bushman, F.D. HIV integration site selection: analysis by massively parallel pyrosequencing reveals association with epigenetic modifications. Genome Res 17, 1186-94 (2007). 26. Finzi, D. et al. Identification of a reservoir for HIV-1 in patients on highly active antiretroviral therapy. Science 278, 1295-300 (1997). 27. Siliciano, J.D. & Siliciano, R.F. Enhanced culture assay for detection and quantitation of latently infected, resting CD4+ T-cells carrying replication-competent virus in HIV-1- infected individuals. Methods Mol Biol 304, 3-15 (2005). 28. Han, Y., Wind-Rotolo, M., Yang, H.C., Siliciano, J.D. & Siliciano, R.F. Experimental approaches to the study of HIV-1 latency. Nat Rev Microbiol 5, 95-106 (2007). 29. Hermankova, M. et al. Analysis of human immunodeficiency virus type 1 gene expression in latently infected resting CD4+ T lymphocytes in vivo. J Virol 77, 7383-92 (2003). 30. Prins, J.M. et al. Immuno-activation with anti-CD3 and recombinant human IL-2 in HIV- 1-infected patients on potent antiretroviral therapy. AIDS 13, 2405-10 (1999). 31. van Praag, R.M. et al. OKT3 and IL-2 treatment for purging of the latent HIV-1 reservoir in vivo results in selective long-lasting CD4+ T cell depletion. J Clin Immunol 21, 218- 26 (2001). 32. Contreras, X. et al. Suberoylanilide hydroxamic acid reactivates HIV from latently infected cells. J Biol Chem 284, 6782-9 (2009). 33. Rasmussen, T.A. et al. Comparison of HDAC inhibitors in clinical development: effect on HIV production in latently infected cells and T-cell activation. Hum Vaccin Immunother 9, 993-1001 (2013). 34. Archin, N.M. et al. Administration of vorinostat disrupts HIV-1 latency in patients on antiretroviral therapy. Nature 487, 482-5 (2012). 35. Xing, S. et al. Disulfiram reactivates latent HIV-1 in a Bcl-2-transduced primary CD4+ T cell model without inducing global T cell activation. J Virol 85, 6060-4 (2011). 36. Williams, S.A. et al. Prostratin antagonizes HIV latency by activating NF-kappaB. J Biol Chem 279, 42008-17 (2004). 37. Biancotto, A. et al. Dual role of prostratin in inhibition of infection and reactivation of human immunodeficiency virus from latency in primary blood lymphocytes and lymphoid tissue. J Virol 78, 10507-15 (2004).

38. Spina, C.A. et al. An in-depth comparison of latent HIV-1 reactivation in multiple cell model systems and resting CD4+ T cells from aviremic patients. PLoS Pathog 9, e1003834 (2013). 39. Mehla, R. et al. Bryostatin modulates latent HIV-1 infection via PKC and AMPK signaling but inhibits acute infection in a receptor independent manner. PLoS One 5, e11160 (2010). 40. Lehrman, G. et al. Depletion of latent HIV-1 infection in vivo: a proof-of-concept study. Lancet 366, 549-55 (2005). 41. Siliciano, J.D. et al. Stability of the latent reservoir for HIV-1 in patients receiving valproic acid. J Infect Dis 195, 833-6 (2007). 42. Chun, T.W. et al. Quantification of latent tissue reservoirs and total body viral load in HIV-1 infection. Nature 387, 183-8 (1997). 43. Han, Y. et al. Resting CD4+ T cells from human immunodeficiency virus type 1 (HIV- 1)-infected individuals carry integrated HIV-1 genomes within actively transcribed host genes. J Virol 78, 6122-33 (2004). 44. Monie, D. et al. A novel assay allows genotyping of the latent reservoir for human immunodeficiency virus type 1 in the resting CD4+ T cells of viremic patients. J Virol 79, 5185-202 (2005). 45. Yamamura, S. et al. Single-cell microarray for analyzing cellular response. Anal Chem 77, 8050-6 (2005). 46. Engvall, E. & Perlmann, P. Enzyme-linked immunosorbent assay, Elisa. 3. Quantitation of specific antibodies by enzyme-labeled anti-immunoglobulin in antigen-coated tubes. J Immunol 109, 129-35 (1972). 47. Chen, C. et al. Real-time quantification of microRNAs by stem-loop RT-PCR. Nucleic Acids Res 33, e179 (2005). 48. Di Carlo, D. & Lee, L.P. Dynamic single-cell analysis for quantitative biology. Anal Chem 78, 7918-25 (2006). 49. Bengtsson, M., Hemberg, M., Rorsman, P. & Stahlberg, A. Quantification of mRNA in single cells and modelling of RT-qPCR induced noise. BMC Mol Biol 9, 63 (2008). 50. Warren, L., Bryder, D., Weissman, I.L. & Quake, S.R. Transcription factor profiling in individual hematopoietic progenitors by digital RT-PCR. Proc Natl Acad Sci U S A 103, 17807-12 (2006). 51. Beer, N.R. et al. On-chip, real-time, single-copy polymerase chain reaction in picoliter droplets. Anal Chem 79, 8471-5 (2007). 52. Kiss, M.M. et al. High-throughput quantitative polymerase chain reaction in picoliter droplets. Anal Chem 80, 8975-81 (2008). 53. Mazutis, L. et al. Droplet-based microfluidic systems for high-throughput single DNA molecule isothermal amplification and analysis. Anal Chem 81, 4813-21 (2009). 54. Strain, M.C. et al. Highly precise measurement of HIV DNA by droplet digital PCR. PLoS One 8, e55943 (2013). 55. Leamon, J.H. et al. A massively parallel PicoTiterPlate based platform for discrete picoliter-scale polymerase chain reactions. Electrophoresis 24, 3769-77 (2003). 56. Lindstrom, S., Hammond, M., Brismar, H., Andersson-Svahn, H. & Ahmadian, A. PCR amplification and genetic analysis in a microwell cell culturing chip. Lab Chip 9, 3465- 71 (2009). 89

57. Nagai, H., Murakami, Y., Morita, Y., Yokoyama, K. & Tamiya, E. Development of a microchamber array for picoliter PCR. Anal Chem 73, 1043-7 (2001). 58. Men, Y. et al. Digital polymerase chain reaction in an array of femtoliter polydimethylsiloxane microreactors. Anal Chem 84, 4262-6 (2012). 59. Marcus, J.S., Anderson, W.F. & Quake, S.R. Microfluidic single-cell mRNA isolation and analysis. Anal Chem 78, 3084-9 (2006). 60. Marcus, J.S., Anderson, W.F. & Quake, S.R. Parallel picoliter rt-PCR assays using microfluidics. Anal Chem 78, 956-8 (2006). 61. Toriello, N.M. et al. Integrated microfluidic bioprocessor for single-cell gene expression analysis. Proc Natl Acad Sci U S A 105, 20173-8 (2008). 62. Kumaresan, P., Yang, C.J., Cronier, S.A., Blazej, R.G. & Mathies, R.A. High-throughput single copy DNA amplification and cell analysis in engineered nanoliter droplets. Anal Chem 80, 3522-9 (2008). 63. Patterson, B.K. et al. Detection of HIV-1 DNA and messenger RNA in individual cells by PCR-driven in situ hybridization and flow cytometry. Science 260, 976-9 (1993). 64. Patterson, B.K. et al. Detection of HIV-1 DNA in cells and tissue by fluorescent in situ 5'-nuclease assay (FISNA). Nucleic Acids Res 24, 3656-8 (1996). 65. Bagasra, O. Protocols for the in situ PCR-amplification and detection of mRNA and DNA sequences. Nat Protoc 2, 2782-95 (2007). 66. Dirks, R.M. & Pierce, N.A. Triggered amplification by hybridization chain reaction. Proc Natl Acad Sci U S A 101, 15275-8 (2004). 67. Venkataraman, S., Dirks, R.M., Rothemund, P.W., Winfree, E. & Pierce, N.A. An autonomous polymerization motor powered by DNA hybridization. Nat Nanotechnol 2, 490-4 (2007). 68. Choi, J., Love, K.R., Gong, Y., Gierahn, T.M. & Love, J.C. Immuno-hybridization chain reaction for enhancing detection of individual cytokine-secreting human peripheral mononuclear cells. Anal Chem 83, 6890-5 (2011). 69. Choi, H.M., Beck, V.A. & Pierce, N.A. Next-Generation in Situ Hybridization Chain Reaction: Higher Gain, Lower Cost, Greater Durability. ACS Nano (2014). 70. Choi, H.M. et al. Programmable in situ amplification for multiplexed imaging of mRNA expression. Nat Biotechnol 28, 1208-12 (2010). 71. Cairns, M.J., Hopkins, T.M., Witherington, C., Wang, L. & Sun, L.Q. Target site selection for an RNA-cleaving catalytic DNA. Nat Biotechnol 17, 480-6 (1999). 72. Yang, C.J. et al. Linear molecular beacons for highly sensitive bioanalysis based on cyclic Exo III enzymatic amplification. Biosens Bioelectron 27, 119-24 (2011). 73. Duan, R. et al. Lab in a tube: ultrasensitive detection of microRNAs at the single-cell level and in breast cancer patients using quadratic isothermal amplification. J Am Chem Soc 135, 4604-7 (2013). 74. Chapin, S.C. & Doyle, P.S. Ultrasensitive multiplexed microRNA quantification on encoded gel microparticles using rolling circle amplification. Anal Chem 83, 7179-85 (2011). 75. Wang, F., Elbaz, J., Orbach, R., Magen, N. & Willner, I. Amplified analysis of DNA by the autonomous assembly of polymers consisting of DNAzyme wires. J Am Chem Soc 133, 17149-51 (2011).

76. Dominguez, M.H. et al. Highly multiplexed quantitation of gene expression on single cells. J Immunol Methods 391, 133-45 (2013). 77. Beer, N.R. et al. On-chip single-copy real-time reverse-transcription PCR in isolated picoliter droplets. Anal Chem 80, 1854-8 (2008). 78. Margulies, M. et al. Genome sequencing in microfabricated high-density picolitre reactors. Nature 437, 376-80 (2005). 79. Wang, X. & Stollar, B.D. Human immunoglobulin variable region gene analysis by single cell RT-PCR. J Immunol Methods 244, 217-25 (2000). 80. Hoffmann, J., Hin, S., von Stetten, F., Zengerle, R. & Roth, G. Universal protocol for grafting PCR primers onto various lab-on-a-chip substrates for solid-phase PCR. Rsc Advances 2, 3885-3889 (2012). 81. Randall, G.C. & Doyle, P.S. Permeation-driven flow in poly(dimethylsiloxane) microfluidic devices. Proc Natl Acad Sci U S A 102, 10813-8 (2005). 82. Chang-Yen, D.A., Eich, R.K. & Gale, B.K. A monolithic PDMS waveguide system fabricated using soft-lithography techniques. Journal of Lightwave Technology 23, 2088- 2093 (2005). 83. Sloan, R.D. & Wainberg, M.A. The role of unintegrated DNA in HIV infection. Retrovirology 8, 52 (2011). 84. Yu, X. et al. Neutralizing antibodies derived from the B cells of 1918 influenza pandemic survivors. Nature 455, 532-6 (2008). 85. Ogunniyi, A.O., Story, C.M., Papa, E., Guillen, E. & Love, J.C. Screening individual hybridomas by microengraving to discover monoclonal antibodies. Nat Protoc 4, 767-82 (2009). 86. Han, Q., Bradshaw, E.M., Nilsson, B., Hafler, D.A. & Love, J.C. Multidimensional analysis of the frequencies and rates of cytokine secretion from single cells by quantitative microengraving. Lab on a Chip, DOI:10.1039/B926849A (2010). 87. Saleh, S. et al. Expression and reactivation of HIV in a chemokine induced model of HIV latency in primary resting CD4+ T cells. Retrovirology 8, 80. 88. Eriksson, S. et al. Comparative analysis of measures of viral reservoirs in HIV-1 eradication studies. PLoS Pathog 9, e1003174 (2013). 89. Murray, J.M., Kelleher, A.D. & Cooper, D.A. Timing of the components of the HIV life cycle in productively infected CD4+ T cells in a population of HIV-infected individuals. J Virol 85, 10798-805. 90. Kim, S.Y., Byrn, R., Groopman, J. & Baltimore, D. Temporal aspects of DNA and RNA synthesis during human immunodeficiency virus infection: evidence for differential gene expression. J Virol 63, 3708-13 (1989). 91. Dinoso, J.B. et al. Treatment intensification does not reduce residual HIV-1 viremia in patients on highly active antiretroviral therapy. Proc Natl Acad Sci U S A 106, 9403-8 (2009). 92. Taylor, S., Smith, S., Windle, B. & Guiseppi-Elie, A. Impact of surface chemistry and blocking strategies on DNA microarrays. Nucleic Acids Res 31, e87 (2003). 93. Beattie, W.G. et al. Hybridization of DNA targets to glass-tethered oligonucleotide probes. Mol Biotechnol 4, 213-25 (1995).