Direct optical mapping of transcription factor binding sites on field-stretched λ-DNA in nanofluidic devices

Mapping transcription factor (TF) binding sites along a DNA backbone is crucial in understanding the regulatory circuits that control cellular processes. Here, we deployed a method adopting bioconjugation, nanofluidic confinement and fluorescence single molecule imaging for direct mapping of TF (RNA polymerase) binding sites on field-stretched single DNA molecules. Using this method, we have mapped out five of the TF binding sites of E. coli RNA polymerase to bacteriophage λ-DNA, where two promoter sites and three pseudo-promoter sites are identified with the corresponding binding frequency of 45% and 30%, respectively. Our method is quick, robust and capable of resolving protein-binding locations with high accuracy (∼ 300 bp), making our system a complementary platform to the methods currently practiced. It is advantageous in parallel analysis and less prone to false positive results over other single molecule mapping techniques such as optical tweezers, atomic force microscopy and molecular combing, and could potentially be extended to general mapping of protein–DNA interaction sites.

2 photoresist developer for 15 seconds) were etched using Inductively Coupled Plasma (ICP, RIE-10ip, Samco) machine. A mixture of CHF 3 /CF 4 /O 2 /Ar gases at flow rates 50/33.3/6.7/30 sccm and bias/RF power 700/300W were used for 2 min and 30 seconds. Surface profiler (Alpha step IQ, KLA Tencor) measurements confirmed features of depth around 1µm. Figure S1. Fabrication process flow of nanoslit devices used in our experiments. Nanoslit features defined by second step UV exposure were etched using a Reactive Ion Etching (RIE Plasmalab 80+, Oxford Instruments) machine. An initial de-scum process using 100 sccm O 2 , 300 mTorr pressure and 150W RF power for 3 minutes to remove any residual photoresist layer in the nanoslit regions is followed by a brief etching using CHF 3 /O 2 mixture at flow rates 85/6 sccm, pressure 100 mTorr and RF power of 70W for 8 minutes. Surface profiler and atomic force microscopy (Nanoscope III, Veeco) measurements showed the nanoslit depths are around 60 nm (Fig. S2a). To be noted, as the slit width (10 µm) and length (200 µm) are big enough to be defined by a simple photomask through direct UV exposure of the photoresist and developing, the nanoslit height (or depth) was merely defined by the RIE etching process without complicated and expensive nanolithography process, such as electron-beam lithography.
Alternatively, a wet-etching process would work due to the ultralow aspect ratio (6x10 -3 ) of our nanoslit.
Above-mentioned fabrication steps were carried out on a 4" fused silica wafer to facilitate batch processing. Each fluidic device is 14x14 mm 2 and thus a 4" wafer can yield nearly 25 working chips. The 4" wafer was subjected to dicing to obtain individual fluidic devices. Then, loading holes were drilled using a sand blaster and the devices are ready for bonding. A small DC voltage is applied to drive the DNA-protein complexes from the reservoirs to the nanoslit region, where they are observed using an inverted epi-fluorescence microscope.

PSQ Bonding Details:
Each diced chip is then bonded using a glass coverslide coated with polysilsesquioxane (PSQ) polymer layer (1

Gel-Shift Assay:
Gel shift assay was done to optimize the conditions for DNA-RNAP complex formation.
A shift assay including the primary antibody and secondary antibody conjugated QD was also done to get some insights on DNA-RNAP-AB-QD complex formation conditions.
A 310 base-pair PCR fragment with P R promoter region (37974 -38032 bases from the 5' were some non-specific complexes too, but the results from the specific complexes were distinct from the non-specific ones (See Fig. 3 in main article).

Sample Preparation
Fluosphere end-labeling of λ-DNA: This process involves multiple steps like ligation of biotinylated primers to DNA ends, removal of unbound oligonucleotides and binding streptavidin fluospheres to biotinylated DNA molecules. We followed a method similar to that of Perkins et al. Removing Unbound Oligonucleotides: The unbound oligonucleotides from the previous step were removed, as it would interfere with the fluosphere labeling step in the next step. The reason is that these oligonucleotides are much smaller compared to the ligated λ-DNA molecules and can diffuse faster, thereby binding more easily to the streptavidin sites in the fluosphere, thus reducing the DNA end labeling efficiency.
A 100,000 MWCO centrifugal filter (10 ml, Amicon Ultra4, Millipore) was used for this purpose. 1X TE buffer was used in this process. First, 1 ml 1X TE buffer is added to the filter.
Then, another 1ml 1X TE buffer with 1 µl BSA (10 mg/ml, NEB) was added and the column is centrifuged at 2750g for 4 minutes at 25 o C (Z 300K, Hermle). The filter is taken out, replaced with 1X TE buffer up to the 2 ml mark of the filter and the centrifugation process was carried out for two more times with the above mentioned conditions. After this step, the filter was filled with 1ml of 1X TE buffer; ligation solution was added to it slowly using a pipette with wide opened tip. The filter is then filled to the 2 ml mark and centrifuged at 1000g for 16 minutes at 25 o C. Such slow speeds are used to avoid shearing of longer DNA molecules. The solution that got drained to the bottom of the filter was removed and the filter was filled with fresh 1X TE buffer to the 2 ml mark. This process was repeated seven times to ensure maximum removal of unbound oligonucleotides.
The final solution retained at the top of the filter was collected and the DNA concentration was measured using a spectrophotometer (NanoDrop, Thermo Scientific) to ensure that the DNA concentration is adequate for the following steps.
Labeling biotinylated λ-DNA with fluospheres: After the cleanup process, biotinylated λ-DNA was mixed in 1: Labeling DNA molecules without cohesive (cos) sites: λ-DNA has 12-bp overhanging sites at both ends (cos sites), which makes fluosphere end-labeling process easier. Here we used T4 DNA polymerase/Klenow enzyme assisted end labeling for blunt or truncated DNA molecules, showing that our platform can be extended to DNA molecules without cos sites. Here, we will discuss in detail with a simple demonstration that adds support to this claim.
For this, we used a combination of T4 DNA polymerase (with 3'-end to 5'-end exonuclease activity, New England Biolabs) and Klenow enzyme (3'-5' exonuclease deleted, New England Biolabs) to label phage T7 DNA, which, unlike lambda DNA, has blunt ends.
Analyzing T7 DNA sequence, we found that an adenosine is present only as the 19 th base from the 3' end. So, we added a very high concentration of dATP during the T4 DNA polymerase digestion, so that the digestion process cannot proceed beyond this point. Once the digestion is completed, we added Klenow enzyme with 3'-5' exonuclease activity deleted in presence of dNTPs, where dTTP was replaced by biotin-dUTP. After the reaction, free nucleotides were separated from phage T7 DNA molecules using centrifugal filter cleanup method (which we also used for lambda DNA end labeling process, explained in detail earlier in page 8 of this document.) As a simple demonstration, we used 2.8 µm streptavidin-coated magnetic beads (Invitrogen) and added biotinylated T7 DNA molecules. Results showed successful end modification using T4 DNAP/Klenow enzyme process, thus DNA molecules binding to the magnetic beads (Fig. S4a). As a control experiment, we used 2.8 µm magnetic beads without streptavidin modification. Results showed no DNA binding to beads, proving that DNA binds to beads only through biotin-streptavidin coupling (Fig. S4b). showing no binding of T7 DNA molecules.
Next, we also used 40 nm streptavidin transfluospheres to label single DNA molecules.
We used a positively charged polylysine coated coverslide to carry out a simple demonstration.
Although these experiments were not conducted in our nanoslit devices, it proves that blunt DNA like phage T7 DNA can be labeled using this alternate method (Fig. S5a). Alternately, one could also generate PCR products of any sequence along a DNA strand using biotinylated primers. By using a biotinylated primer in a PCR reaction, the corresponding strand of the PCR product can be biotinylated. To demonstrate end labeling of biotinylated PCR products, we generated a 7 Kb long PCR product using M13 phage DNA as a template, where one end of the PCR product was biotinylated. Later streptavidin-fluospheres and biotinylated 7 Kb PCR products were mixed together for effective end labeling (Fig. S5b). Observation Solution: Above prepared DNA-RNAP-AB-QD complex solution was mixed with equal amounts of OBS buffer solution containing glucose oxidase (50 µg/ml, Sigma), catalase (10 µg/ml, Roche), β-mercaptoethanol (0.5% v/v, Sigma), and YOYO-I dye in DMSO (10 µM, 1:5 dye: base-pair ratio) and allowed to sit at room temperature for 10-15 minutes. This is to ensure proper mixing of dye molecules with the DNA molecules thereby ensuring uniform labeling.

Surface Passivation:
Non-specific binding of biomolecules to micro-and nano-channel surfaces happens to be  (9). Here, we tried to see whether it also helps to reduce non-specific interaction of proteins to channel surface. Also, our channel surface is very hydrophilic, which helps to reduce the non-specific binding to a greater extent. Earlier experiments have shown that proteins tend to show adsorb more on hydrophobic surfaces than hydrophilic surfaces (10). Also, studies have shown that increase in protein concentration shows increase in adsorption (11).
To demonstrate that our channels are resistant enough to protein adsorption, the channels were flooded with a very high concentration (50 nM) of streptavidin quantum dots, many folds higher than the concentration used during our DNA-protein complex experiments. QDs were driven from microchannels into the nanoslit region using a small electric field. We did not see Andor) was used to acquire images with an equivalent pixel resolution of 100nm.

Image Processing and Analysis:
DNA length measurements were carried out manually using ImageJ (NIH) software.
Distance between end labeled fluosphere and quantum dots were carried out using centroid localization method (12). First, the point spread function (PSF) for our optical setup was determined and images were iteratively deconvolved using a custom macro written in ImageJ.
The deconvolved images were then cropped appropriately before further analysis. High precision localization of QDs was carried out by deconvolving the collected distribution of photons (or counts) to the PSF of the system. Using this method, we could achieve localization precision around 2.5 nm for a typical quantum dot point spread function. The position co-ordinate values obtained from this localization were used in finding the distance between two quantum dots.
DNA contour length changes due to YOYO-I dye labeling (for 1:5 dye:base-pair ratio, the contour length increases from 16.5 µm to around 22 µm) (13) and DNA stretching in nanoslits (In our experiments, we achieve around 87% DNA stretching) were taken into account in calculating the final RNAP binding position values. Distance measurement results obtained from the above steps were converted to values in kilobases. Although each slit has 20-30 DNA molecules parallely stretched (See Fig. 2 in main article), only those DNA molecules that has the end labeled fluospheres and at least one RNAP-QD complex bound to stretched DNA were considered for construction of position histograms.
Additionally, only DNA molecules 9 µm or longer were considered, such that all expected binding sites are included. This is to avoid any discrepancy in the obtained position histogram.
Results from many molecules (~200) were collected and a histogram was plotted between DNA length (in µm) along X-axis and RNAP binding frequency (counts) along Y-axis using OriginPro 9.0. A multi-peak Gaussian fitting of the obtained histogram was carried out using the PeakAnalyzer function in OriginPro 9.0 ( Fig. 4 in main article).
The concept of high precision QD localization is shown in Fig. S6. In this way, we could obtain localization precision around 2.5 nm for a typical QD PSF. The value 2.5 nm refers to the standard error in determining the center or mean value of the distribution. Centroid localization with very high precision has been reported and now a common practice for single molecule experiments. The standard error in determining the center/mean value can be pushed down to 1