Microfluidic Device and Method for Isolation of Nucleic Acid

Abstract

The present invention concerns a microfluidic device for mechanically induced trapping of molecular interactions comprising at least a first unit cell and a second unit cell, each unit cell comprisinga membrane chamber comprising a membrane, a flow channel crossing the membrane chamber and having an inlet and an outlet, and the flow channel crossing the first unit cell being different from the flow channel crossing the second unit cell. Another object of the invention is a method for isolation of specifically bound nucleic acids to target molecules on said microfluidic device followed by its recovery and identification.

Claims

1-24. (canceled)

25. A microfluidic device for mechanically induced trapping of molecular interactions including a first unit cell and a second unit cell, each unit cell comprising: a membrane chamber having a membrane; a flow channel crossing the membrane chamber and having an inlet and an outlet; and the flow channel crossing the first unit cell being other than the flow channel crossing the second unit cell, or being not the same as the flow channel crossing the second unit cell.

26. The microfluidic device according to claim 25, further comprising: a first set of components configured to provide an independent fluidic communication to each unit cell.

27. The microfluidic device according to claim 26, further comprising: a second set of components configured to connect at least the first and the second unit cells together.

28. The microfluidic device according to claim 27, wherein the first set of components or the second set of components include control valves.

29. The microfluidic device according to claim 25, wherein a surface of the microfluidic device is functionalized to capture proteins.

30. The microfluidic device according to claim 29, wherein the proteins include at least one of transcription factors and other biomedically important proteins.

31. A dispenser for parallel loading of multiple samples on a microfluidic device according to claim 25, comprising one inlet and a plurality of outlets to equally distribute the samples, wherein the number of outlets corresponds to the number of unit cells of the microfluidic device.

32. The dispenser according to claim 31, wherein the dispenser is made of polydimethylsiloxane (PDMS).

33. The dispenser according to claim 31, wherein the inlet is directly connected to each outlet.

34. The dispenser according to claim 31, wherein the inlet is connected to the outlets via several channels.

35. A method for isolation of specifically bound nucleic acids to target molecules comprising: providing a microfluidic device according to claim 25; loading a mixture of nucleic acids and target molecules into the microfluidic device; trapping the bound nucleic acid-target molecule complexes; removing the unbound material; collecting the bound nucleic acids; amplifying the bound nucleic acids; and high throughput sequencing of the amplified bound nucleic acids.

36. The method according to claim 35, wherein target molecules include proteins.

37. The method according to claim 36, wherein the proteins are expressed proteins.

38. The method according to claim 35, wherein the proteins include single or multiple proteins.

39. The method according to claim 35, wherein the proteins include transcription factors (TFs) or other biomedically important proteins.

40. The method according to claim 35, wherein the nucleic acids include a random DNA library.

41. The method according to claim 40, wherein the random DNA library include sequencing adapters.

42. The method according to claim 35, wherein the nucleic acids include a random RNA library.

43. The method according to claim 35, further comprising the step of: loading the mixture of nucleic acids and target molecules on the membrane chamber of the microfluidic device.

44. The method according to claim 35, wherein bound nucleic acids are collected at the same time.

45. The method according to claim 35, wherein bound nucleic acids are collected while heating the microfluidic device.

Description

DESCRIPTION OF THE DRAWINGS

[0034] The above object, features and other advantages of the present invention will be best understood from the following detailed description in conjunction with the accompanying drawings, in which:

[0035] FIG. 1 illustrates an exemplary 64-unit micro device according to the present invention, where FIG. 1A shows a 64-unit chip design in which dark grey and light grey colors denote flow and control layers respectively; FIG. 1B shows that each unit of the device can be accessed independently through an individual flow inlet and an outlet but can also be connected with the other units; FIG. 1C shows that switching between individual and common access modes is done through the use of control valves; and FIG. 1D shows that each device consists of or includes a PDMS chip (approximately 25 cm, for example) and an epoxydized glass slide;

[0036] FIG. 2 is a schematic representation of MITOMI-seq procedure, where FIG. 2(1) shows a mixture of nucleic acids and protein introduced into the unit of the microfluidic device, protein is immobilized on the surface of the device with antibody; in FIG. 2(2) the system is incubated for one hour to allow its equilibration and complex assembly; in FIG. 2(3) newly formed complexes are trapped under a flexible PDMS membrane and unbound molecules as well as molecular complexes are washed away; and in FIG. 2(4) proteins are disrupted and selectively bound nucleic acids are collected from the device;

[0037] FIG. 3 shows chip fabrication according to the present invention;

[0038] FIG. 4 shows a mask fabrication process according to the present invention;

[0039] FIG. 5 shows a design of two PDMS passive dispensers (FIGS. 5A and 5B), an external source of compressed pressure is connected to the inlet of the device and is equally distributed between 64 outlets;

[0040] FIG. 6 illustrates extended DNA library design and construction according to the present invention;

[0041] FIG. 7 shows a MITOMI-seq procedure according to the present invention, where FIG. 7A shows a snapshot of three units of the microfluidic device, each unit is isolated from others using the microvalves and has an individual inlet and an outlet as illustrated by the independent passage of color dyes; FIG. 7B shows a MITOMI-seq procedure, Bait TF, target dsDNA and a non-specific competitor poly-dIdC are mixed and loaded in one chamber of the microfluidic device, the mixtures are incubated on-chip for 40 min. then, bound DNA is eluted from all the units of the device simultaneously and collected in one tube, recovered DNA is amplified and sequenced on a HiSeq instrument; on the left of FIG. 7B there is a schematic representation of three individual chambers, and on the right there is corresponding snapshots of an individual chamber taken before and after mechanical trapping;

[0042] FIG. 8 shows motifs identified by MITOMI-seq for mouse TFs;

[0043] FIG. 9 shows motifs identified by MITOMI-seq for fly TFs;

[0044] FIG. 10 shows motifs identified by MITOMI-seq for human TFs;

[0045] FIG. 11 shows JASPAR, HT-SELEX and MITOMI-seq motif occurrence in ChIP-seq peaks of NFKB1 derived from human lymphoblastoid cell line (GEO:GSM935527); and

[0046] FIG. 12 shows pMARE vector maps.

DETAILED DESCRIPTION OF THE INVENTION

MITOMI (Mechanically Induced Trapping of Molecular Interactions) Followed by HT (High Throughput) Sequencing (MITOMI-Seq)

[0047] We first thought of using original MITOMI devices and an established protocol to perform an on-chip selection assay. Initial experiments revealed however that MITOMI devices suffered from a small and uncontrolled carry-over between neighboring units. Such a cross talk between units is typically not a problem for standard MITOMI applications that require a basic fluorescence-based read out. However, if one wants to recover bound DNA material from the device and subsequently analyze it with an extremely sensitive method like HT-sequencing, even small amounts of cross-contamination between samples may possibly skew data interpretation. Therefore there is a need for a device that can perform mechanically induced trapping of interactions but also allows a controlled isolation of individual units to be one of the key components of a successful on-chip selection procedure.

[0048] Thus, we decided to develop a cross talk-devoid device as part of an assay that would allow us to perform a robust on-chip selection of TF binding sequences from a size-unbiased pool of randomized DNA requiring minimal amounts of biological material.

Microfluidic Device for Parallel Isolation of Specifically Bound Nucleic Acids to Protein Targets

[0049] We first designed a micro device that accommodates all the desired features in CleWin software. For convenience we restricted the size of the device to fit a standard 75 by 25 mm glass substrate compatible with an available fluorescent scanner. We designed our device to contain 64 units and set the diameter of the reaction chamber within each unit to 300 m (FIG. 1). We fabricated the device using two-layer soft lithography at the Center of Micro and Nano Technology at EPFL.

[0050] FIG. 1 illustrates an exemplary 64-unit micro device according to the present invention, where FIG. 1A shows a 64-unit chip design in which dark grey and light grey colors denote flow and control layers respectively. FIG. 1B shows that each unit of the device can be accessed independently through an individual flow inlet and an outlet but can also be connected with the other units. FIG. 1C shows that switching between individual and common access modes is done through the use of control valves. FIG. 1D shows that each device consists of or includes a PDMS chip (approximately 25 cm, for example) and an epoxydized glass slide;

[0051] FIG. 1 shows an embodiment of a microfluidic device according to the invention. The device consists of or includes two parts: a PDMS microchip and a substrate (FIG. 1D). A PDMS chip consists of a flow and a control layers which are tightly bonded together. The flow layer is a network of channels and cells, which are tightly controlled by a system of control valves. These valves are essentially thin PDMS membranes that could be deflected by hydraulic forces. Thereby, all manipulations of molecular complexes are performed on the surface of the substrate, in the flow layer of the device. To control the pressure of circulating fluids, the device can be connected to external mechanical or solenoid valves. The mechanical trapping of molecular complexes is done through capping of a circular area with a button valve in each unit cell of the device. FIG. 1B is an enlargement of a part of the microfluidic device. This figure shows four unit cells. Each unit cell comprises: [0052] a membrane chamber, that is crossed by a flow channel that could be operated independently of other unit cells of the device (individual mode, FIG. 1C-1) but could also be connected to other unit cells (common mode, FIG. 1C-2); [0053] a spotting chamber.

[0054] FIGS. 1C-1 and 1C-2 illustrate the use of unit cells in an individual and common mode respectively. In individual mode the direction of the flow is controlled by a first set of components: multiplexer valves, neck valves and common control valves, that allow fluid to enter through an individual inlet and exits through an individual outlet. In common mode unit cells are controlled by a second set of components: individual inlets control valves, neck valves and individual outlet control valves, that connect units together and provide a simultaneous accessed to all the units through flow inlet.

[0055] FIG. 2 is a schematic representation of MITOMI-seq procedure, where FIG. 2(1) shows a mixture of nucleic acids and protein introduced into the unit of the microfluidic device, protein is immobilized on the surface of the device with antibody. In FIG. 2(2) the system is incubated for one hour to allow its equilibration and complex assembly. In FIG. 2(3) newly formed complexes are trapped under a flexible PDMS membrane and unbound molecules as well as molecular complexes are washed away. In FIG. 2(4) proteins are disrupted and selectively bound nucleic acids are collected from the device.

Chip Fabrication

[0056] FIG. 3 shows chip fabrication according to the present invention and FIG. 4 shows a mask fabrication process according to the present invention.

[0057] We first prepared two masters: one for the flow and another for the control layer. Both masters were printed on chrome plates using standard photolithography techniques.

[0058] We then used the chrome masters as masks for fabricating two types of wafers: flow and control. Flow wafers were typically fabricated with AZ9260 positive resist at a thickness of 14 m (Table 1) and control wafers were fabricated with SU-8 negative photo resist at a thickness of 10 m (Table 1). Next, we used the fabricated wafers to mold two layers of a PDMS device. Finally, two PDMS layers were aligned and bonded together; we punched holes using a manual-punching machine (Syneo, USA).

PDMS Dispenser

[0059] The device, which allows individual access to each working unit, has an advantage of manipulating heterogeneous samples simultaneously without any risk of contamination or crosstalk. Parallel loading of multiple samples on the chip in this case requires several external sources of pressure. The commercially available tools, such as pneumatic manifolds, typically can branch an external source of compressed air creating several daughter sources. Most of the suppliers currently provide pneumatic manifolds made of metal or plastic that can bifurcate the pressure source into 8 to 12 parallel outputs. But for manipulating 64 samples simultaneously one would need to use several manifolds connected in a massive control unit. To avoid this complex construction, save space and facilitate the dispensing of samples into individual chambers we designed and fabricated a passive dispenser that is aimed to substitute pneumatic manifold. The aim of this device is to evenly distribute the input compressed air between 64 outlets. Similarly to previously mentioned micro device we fabricated the dispensers using soft lithography. But unlike the two-layer patterned devices, dispenser requires only one molded part that is bonded to non-patterned PDMS. FIGS. 5A and 5B show a design of two PDMS passive dispensers, an external source of compressed pressure is connected to the inlet of the device and is equally distributed between 64 outlets.

Random Nucleotide Library

[0060] We exploited the principle of MITOMI-based affinity selection of bound sites from the randomized DNA library to characterize TF DNA binding specificities. We have designed the target DNA library by randomly introducing all four nucleotides at each position of a DNA sequence. This library of random DNA sequences can then be exposed to a TF of interest after which bound sequences are collected and decoded by sequencing. This approach was adopted by several techniques that aim to identify TF binding preferences including HT-SELEX and B1H. Unlike de Bruijn sequences, the design of a random library is relatively simple and synthesis of target oligos is not cost-prohibitive. At the same time, the length of the random site is not limited and one can easily design a library that would cover all possible 10, 20- or even 30-, 40- or 50-mers. This might be especially useful when identifying the binding preferences of homo- and heterodimers. And finally, randomized DNA libraries can be easily multiplexed and used for the characterization of binding specificities of several factors simultaneously.

MITOMI-Seq: Principle

[0061] Each MITOMI-seq experiment starts with the in vitro expression of TFs of interest and generation of target DNA libraries. We express TFs of interest in 6 l of the TnT SP6 High-Yield Wheat Germ (Promega) protein expression system. To make this expression system compatible with the Gateway cloning format and to allow the fluorescence-based detection of TFs, we shuttle the open-reading frame (ORF) of the TF of interest into one of the custom-made pMARE expression vectors (pMARE-eGFP or pMARE-mCherry (Hens et al., 2011)) (FIG. 12) (for more details, please see Methods section).

[0062] The target DNA libraries are constructed from single stranded synthetic oligos by an enzymatic second strand synthesis. Meanwhile, the surface of the microfluidic device is functionalized to capture tagged TFs (the procedure is similar to the one established for regular MITOMI chips, see Methods).

[0063] The expressed TFs are then mixed with target random DNA libraries and the mixtures are immediately loaded on the micro device. After 40 minutes of incubation, newly formed TF-DNA complexes are trapped under a flexible button membrane, unbound material is removed from the device by washing and bound DNA is collected by continuous elution (for a detailed protocol, please see Methods). Collected DNA is then amplified and sequenced in one lane of HiSeq sequencer (Illumina).

[0064] To overcome all the difficulties related to the preparation of samples for HT sequencing, we decided to incorporate sequencing adapters directly within the random DNA library design. To do so we designed target DNA libraries that already contain Solexa (Illumina) sequencing-compatible adapters at the 5- and 3-prime ends of the barcoded randomized 30 bp fragment (FIG. 6). The constructs of 121 bp were then synthesized as single-stranded oligos and were later used to generate double-stranded fluorescently labeled target DNA libraries using Klenow 5-3 exo-second strand synthesis (For details see Methods).

MITOMI-Seq: Procedure

[0065] FIG. 7 shows a MITOMI-seq procedure according to the present invention. FIG. 7A shows a snapshot of three arbitrary units of the microfluidic device (unit 1, unit 2 and unit 3), each unit is isolated from others using the microvalves and has an individual inlet and an outlet as illustrated by the independent passage of color dyes.

[0066] FIG. 7B shows a MITOMI-seq procedure, Bait TF, target dsDNA and a non-specific competitor poly-dIdC are mixed and loaded in one chamber of the microfluidic device, the mixtures are incubated on-chip for 40 minutes then, bound DNA is eluted from all the units of the device simultaneously and collected in one tube, recovered DNA is amplified and sequenced on a HiSeq instrument.

[0067] On the left of FIG. 7B there is shown a schematic representation of three individual chambers, and on the right there is corresponding snapshots of an individual chamber taken before and after mechanical trapping.

[0068] For each MITOMI-seq experiment, we mix the DNA libraries with expressed TFs and load these mixtures onto the microfluidic chip (FIG. 7B). After incubating the chip for 40 min at room temperature, we trap the antibody-captured TF-DNA complexes with the button valves and wash the unbound material away. We then also remove non-specifically bound DNA sequences from the immobilized complexes by opening the button for one second under continuous flow of the washing buffer. To elute the remaining (bound) DNA, we resort to heating the microchip to 70 C. for 10 min as explorative experiments revealed that this vastly improved the release of trapped DNA in PBS washing buffer which is applied for 10 min. Recovered DNA is then amplified and sequenced on a HiSeq sequencer (FIG. 7). (For the detailed MITOMI-seq protocol, see Methods).

[0069] The resulting sequencing data is then de-multiplexed using the barcodes and reads are trimmed to the random 30 bp fragment that is located between the two barcodes corresponding to each sample. These 30 bp fragments are subsequently sorted according to their frequency: from high to low, and identical reads are collapsed in one. We then extract the top 1500 reads from each sample and use it to generate representative binding motifs with MEME (Bailey and Elkan, 1994).

Results:

MITOMI-Seq Identifies DNA Binding Motifs of TF Monomers and Dimers

[0070] First we used the extended random library to perform MITOMI-seq on several well-studied TFs for which DNA binding preferences are known. We also included the now well-characterized PPAR:RXR heterodimer to the tested set. Indeed, the binding specificity of PPAR:RXR heterodimer was previously characterized through ChIP-seq data analysis (Nielsen et al., 2008) but importantly, it has never been probed by any of the high- or medium-throughput in vitro techniques such as PBM, HT-SELEX or B1H.

[0071] Motifs obtained by MITOMI-seq for the selected factors closely match those identified previously (Table 2). Interestingly, for PPAR:RXR heterodimer, we recovered a motif similar to the one identified by ChIP-seq (Nielsen et al. 2008).

[0072] Over a span of just a couple of weeks, we have already been able to process 40 various individual TF or TF dimers originating from different species: M. musculus, D. melanogaster and H. sapiens in three separate MITOMI-seq experiments (24 TFs per experiment). Some factors were processed twice. We were able to retrieve 24 high-confidence motifs with E-values ranging from 1.70E-14 to 4.3E-435 (FIGS. 8, 9, 10). Among those 24 were motifs for TFs that were so far deemed off-limits for DNA binding analyses and for which consequently no binding properties were described. Specifically, we were able to identify motifs for five TFs that belong to the family of KRAB zinc finger proteins: ZNF282, ZKSCAN1, ZFP1, ZNF167 and ZNF415 (FIG. 10). We also identified the binding motif of the zinc finger and homeodomain-containing TF ZEB1 (FIG. 8) and this motif was later confirmed by ChIP-seq analysis (data not shown).

Comparison of MITOMI-Seq, PBM and HT-SELEX Platforms: NFKB1, a Case Study

[0073] We next tried to estimate how well MITOMI-seq performs in comparison to the two other commonly used in vitro technologies that enable TF DNA binding measurements: PBM and HT-SELEX. For this purpose, we focused on NFKB1 DNA binding data sets since data are available from all three technologies. Specifically, we compared the HT-SELEX experimental data from selection cycles 2, 3 and 4 (Dolma et al., 2013), normalized PBM probe data (Siggers et al., 2011) and two MITOMI-seq datasets from independent experiments. Note however that we anticipate including additional comparisons involving other factors in the ultimate manuscript. The NFKB1-related analyses have to therefore be considered as a proof of principle.

[0074] First, we asked how well MITOMI-seq-derived PWM models predict HT-SELEX binding models and if there would be a difference in data obtained by the two technologies. We used a representation of an inferred NFKB1 binding model obtained by ChiP-seq in MITOMI-seq or HT-SELEX data as an estimate of the performance of each technique. For five HT-SELEX and MITOMI-seq datasets, we estimated the amount of unique reads obtained after HT sequencing. Next, we used FIMO (Grant et al., 2011) to calculate enrichment of the NFKB1 TRANSFAC motif (ChIP-seq-derived V$NFKB_Q6_01 (Matys et al., 2006)) within unique reads from all five datasets. We found that one MITOMI-seq data set was significantly enriched with sites mapping to the motif (23.70%) as compared to the three datasets obtained by HT-SELEX. A second MITOMI-seq dataset did however not show a similarly high motif enrichment, as the percentage of unique reads that contained the TRANSFAC-derived NFKB1 motif was close to the one found for the 2.sup.nd and the 3.sup.rd cycle data sets obtained using HT-SELEX. Thus, it appears that the motif enrichment in the data obtained by both technologies is comparable, at least for NFKB1, with MITOMI-seq being potentially in greater agreement with ChIP-seq data than HT-SELEX. We speculate that differences in the enrichment found between two MITOMI-seq data sets could be due to variability in the on-chip selection process between two experiments. We already observed that some experiments result in a wider representation of non-specific DNA binders in sequencing data than others. Nevertheless, we would like to argue that this technical bias does not hinder overall motif discovery.

TABLE-US-00001 TABLE 3 NFKB1 motif enrichment in five different datasets generated by MITOMI-seq or HT-SELEX. The number of unique reads in which Total Unique TRANSFAC Motif NFKB1 data set reads reads Sequencer motif occurs enrichment HT-SELEX cycle 2 66061 65807 GAII 1771 2.69% HT-SELEX cycle 3* 440647 296195 HISEQ 21214 7.16% HT-SELEX cycle 4 153828 115256 GAII 11939 10.36% MITOMI-seq 838253 291788 HISEQ 69167 23.70% experiment 1 MITOMI-seq 1033730 746221 HISEQ 35678 4.78% experiment 2 *Cycle 3 was used to construct the HT-SELEX binding model in the original study.

Conclusion:

[0075] A comprehensive understanding of proteinDNA binding properties is of central importance to gene regulation. Several high-throughput technologies have therefore been developed to study TF-DNA binding. The most popular technology allowing in vivo DNA profiling is ChIP-seq (Johnson et al., 2007). However, it is well appreciated that ChIP-seq-derived DNA binding properties might not provide an accurate picture of TF DNA binding specificities. First, the accessible sites in a particular genome might not cover all possible k-mers. Second, in vivo binding is affected by additional factors, such as chromatin structure, nucleosome positioning and co-factors and thus observed DNA binding in vivo may not even be direct. Thus, in contrast to in vivo studies, in vitro DNA binding assays are valuable because they enable the assessment of direct DNA binding properties, allowing the sampling of the full spectrum of DNA k-mers. In the past, in vitro binding models of TFs were defined based on low-throughput techniques and thus had low resolution and limited accuracy. With technological developments, the ability to measure and predict binding sites has improved. A large leap came in the form of PBMs and HT-SELEX. These two high-throughput technologies produced DNA binding specificity data covering hundreds of TFs. But despite these significant technological advances, all available in vitro binding models taken together currently explain the specificities of only about one third of the total number of known TFs. Moreover, the DNA binding properties of most TF homo- and heterodimers, let alone larger complexes still remains vastly unexplored.

[0076] In this study, we aimed to address this problem, presenting a novel platform for the robust characterization of DNA binding specificities of TF monomers and dimers. The platform itself is based on two core technologies: MITOMI and HT-sequencing, which in combination, enable us to examine TF DNA binding from a new perspective and to determine the binding specificities at an unprecedented resolution. MITOMI-seq combines robust selection of sequences bound to a certain TF from a pool of k-mers with subsequent identification of the bound DNA by deep sequencing. As part of these efforts, we also developed an integrative microfluidic device that allows to run several MITOMI-seq assays simultaneously and to process 64 TFs or TF combinations in parallel. The device is based on the MITOMI principle and performs physical trapping of TF-DNA complexes thereby reducing the loss of bound DNA during the washing step to a minimum. This, in turn, illuminates the unique property of the assay, namely, the ability to preserve and analyze interactions over a wide affinity range. Unlike PBM and HT-SELEX, the two most popular in vitro technologies that aim to identify TF binding specificities, MITOMI-seq operates at micro scale and requires minute amounts of biological material. For example, to perform MITOMI-seq on one TF, one needs only few nanograms of protein which can be easily produced through available in vitro expression systems. Thus, tedious bacterial or mammalian protein expression and purification is no longer required which significantly shortens the time needed for one experiment (Table 4) and even allows the analysis of TFs (such as KRAB ZFPs) that are otherwise difficult to study because of in cello expression issues.

[0077] We demonstrated that MITOMI-seq-derived specificity models generally agree with the TF binding models identified by ChIP-seq or by available in vitro methods (Table 2). We also showed that MITOMI-seq could identify DNA binding preferences of not only monomers or homodimers but alsoof TF heterodimers. Particularly, using MITOMI-seq data, we were able to generate relevant binding models for PPAR:RXR and Clk:Cyc heterodimers through de novo motif discovery. In addition, we identified for several factors binding motifs that were never reported before. Two identified motifs (for ZEB1 and ZNF282) were later confirmed by ChIP-seq.

[0078] To understand the potential of MITOMI-seq data and how similar its performance is to PBM or HT-SELEX data, we also compared the output of MITOMI-seq to that of the other two in vitro technologies. A close comparison of NFKB1 datasets generated from HT-SELEX and MITOMI-seq revealed that unlike SELEX, MITOMI does not result in an over-selection of the same DNA sequences but instead provides a larger fraction of informative sites that contribute to a specificity model. Thus, binding models generated from MITOMI-seq data may potentially be more accurate and comprehensive compared to HT-SELEX models. At the same time, we showed that, similar to observations with the PBM approach, the DNA binding data from MITOMI-seq can guide DNA binding affinity estimates. We also showed that the sequences that were enriched in the NFKB1 MITOMI-seq data set were also ranked as highest NFKB1 affinity sites in the PBM data. This suggests that, in principle, MITOMI-seq and PBM can produce comparable data. But unlike PBM, MITOMI-seq allows the probing of a much larger sequence space making it an ideal platform for the identification of sites bound by TF dimers or by factors that recognize long DNA sequences.

[0079] One common concern about the DNA binding data obtained by various in vitro technologies, including MITOMI-seq, is how relevant in vitro models are to in vivo binding and if there are certain advantages or disadvantages of each in vitro technology with respect to one another. Currently we are investigating this problem and trying to assess on how well can MITOMI-seq data predict in vivo TF DNA binding in general and if there any advantage of it compared to PBM and HT-SELEX. As a first attempt to evaluate the goodness of the motifs derived by MITOMI-seq, HT-SELEX and the motif retrieved from JASPAR database (ChIP-seq based) we quantified the occurrence of each of the motifs within the ChIP-seq peaks corresponding to NFKB1 binding in lymphoblastoid cells. We used sensitivity at 1% false-positive and area under the receiver operating characteristic curve to gauge the binding prediction (see Orenstein and Shamir, 2014 for details). We found the area under curve (AUC) value to be higher for MITOMI-seq-derived motif compared to HT-SELEX and JASPAR motifs (FIG. 11. JASPAR, HT-SELEX and MITOMI-seq motif occurrence in ChIP-seq peaks of NFKB1 derived from human lymphoblastoid cell line (GEO:GSM935527). This proves that in the case of NFKB1 binding the motif derived by MITOMI-seq is more accurate in predicting binding sites of NFKB1 genome-wide compared to previously available binding models.

TABLE-US-00002 TABLE 4 Comparison of PBM, SELEX-seq and MITOMI-seq technologies. PBM (Protein binding Property microarrays) SELEX-seq MITOMI-seq Sequence space limited large large Throughput low high high Nature of the medium-to-high high low-to-high interactions affinity affinity affinity Protein yes yes no purification needed Ability to monitor yes no yes heterodimers Amount of protein g ng ng required Hands-on ~2 days ~2 days 4-5 hours procedure Need for the yes no no preprinted microarray

Methods:

Protein Cloning and Expression

[0080] To enable the expression of TFs and their immobilization and fluorescence-based detection, we explored different strategies. We found that the wheat germ (WG) in vitro transcription translation expression system containing translation enhancer (TE) sequences from the barley yellow dwarf virus (BYDV) (Promega) yielded the most robust and reproducible protein expression (data not shown). To make this expression system compatible with Gateway TF ORF clone format and to allow the fluorescence-based detection of TFs, we generated several novel vectors, pMARE, that differ by fluoro C-terminal fusions (FIG. 12). To do so we cut the pF3A WG (BYDV) Flexi vector (Promega) with NcoI and DraI, and removed the barnase cassette. The NcoI site was blunted and the Gateway reading frame A cassette (Invitrogen) was ligated in. Subsequently, the eGFP, mCherry, YFP of eBFP coding sequences containing a stop codon at the 3-end was incorporated between the KpnI and Sac/restriction sites using standard cloning techniques. TFs were then subcloned from the Entry clones into the pMARE vector by standard Gateway cloning.

Target DNA Library Preparation

[0081] Randomized extended DNA libraries were ordered as single stranded oligos from IDT. The adapter sequences and barcodes used for each library are listed in the Table 5. The oligo containing a Cy5 5-fusion: /5Cy5/CAA GCA GAA GAC GGC ATA CG (SEQ ID NO 9) was used as a primer of the complementary strand synthesis by means of Klenow exo-extension reaction (NEB Cat No M0212). Detailed reaction conditions are described in the MITOMI-seq procedure. The libraries were then purified using MinElute PCR purification kit (Qiagen) and diluted with ddH20 in a ration 1:10. 50 ng of poly-dIdC (Sigma) were added to each 10 l of the diluted library.

MITOMI-Seq Procedure (Detailed Protocol):

[0082] 1. Sample Preparation:

[0083] 1.1. Set Up the Expression Mix for the TFs as Follows: [0084] 4 WITT mix (TnT SP6 High-Yield Wheat Germ master mix, Promega) [0085] 200 ng plasmid DNA [0086] Nuclease-free dH2O till 6 l total volume [0087] Incubate at 25 C. for 3 hours

[0088] 1.2. Synthesis of the dsDNA Libraries: [0089] 5 l Buffer 2 (NEB) [0090] 5 l dNTPs [0091] 0.5 l Cy5 labeling primer (500 M) [0092] 1.5 l Library oligoes (200 M) [0093] 37 l dH2O [0094] 94 C.-5 min [0095] 50 C.-60 sec [0096] place tubes on ice [0097] add 1 l of Klenow 3-5 exo-(NEB Cat No M0212) [0098] 37 C. 60 min [0099] keep at 0 C. [0100] Use MinElute to purify the double-stranded libraries, elute in 12 l of EB. [0101] Dilute the libraries 1:10 in ddH2O and add 50 ng of dIdC to each 10 l of diluted libraries.

[0102] 2. MITOMI:

[0103] 2.1. Surface Chemistry: [0104] Load all control lines with dH2O, at 3 psi, check that all valves work: switch all ON that the air pushes water through the pipes [0105] Usually Button valves are loading earlier, switch it off once it is filled. [0106] Increase the pressure in valves till 10 psi and make sure that that valves work properly. [0107] Run BSA-biotin for 10 min 2 mg/ml BSA-bio @ 5 psi (keep chambers isolated) [0108] Wash with PBS for 2 minutes [0109] Run Neutravidin (500 ug/ml in PBS) for 10 minutes [0110] PBS for 2 minutes [0111] Close button [0112] PBS for 1-2 min [0113] BSA-biotin for 10 min 2 mg/ml BSA-bio @ 5 psi (keep chambers isolated) [0114] PBS for 3-4 min [0115] Anti-GFP-biotin 3 min (1:100) (keep chambers isolated) [0116] Open button [0117] Continue Anti-GFP-biotin for 15 min (1:100) (keep chambers isolated) [0118] PBS for 5 min

[0119] 2.2. Sample Loading and MITOMI: [0120] Mix: [0121] 5 l expressed non-purified TF [0122] 5 l diluted ds DNA library [0123] (optional) 5 l of a partner TF [0124] Load the mixtures through individual inlets [0125] Incubate at RT for 40 min [0126] Scan with a fluorescent scanner [0127] Close the button [0128] Wash with PBS for 20 min [0129] Scan with a fluorescent scanner

[0130] 3. Elution and Library Preparation for HT-Sequencing [0131] Open the button [0132] Incubate 70 C. for 10 min [0133] Elute with PBS for 10 min @ 70 C. and amplify as follows:

TABLE-US-00003 5xHF KAPA buffer 10 L dNTPs (supplied with the kit) 1.5 L primer GA2seq FW (10 M) 0.5 L primer GA2seq RV(10 M) 0.5 L KAPA HiFi polymerase 0.5 L Eluted DNA 37 L Total volume 50 L [0134] Cycle: [0135] 95 C. 2 min [0136] 98 C. 20 sec| [0137] 65 C. 15 sec|15 [0138] 72 C. 90 sec| [0139] 72 C. 2 min [0140] 4 C. forever [0141] PCR purify using Minelute kit from QIAGEN. Elute in 10 L of EB.

[0142] The detailed information about primers, barcodes and libraries used in this study could be found in a supplementary TableS1.

Data Analysis

[0143] Raw Illumina reads were processed using custom perl scripts, FASTX-tools. Read statistics and HMM were implemented using custom scripts. De novo motif discovery was done with MEME.

TABLE-US-00004 TABLE5 Listofbarcodesandadaptersequencesused forextendedlibrary. Sample barcodeFW barcodeRV R30_B1 CATGCTC GAGCATG R30_B2 ACGCAAC GTTGCGT R30_B3 TCGCAGG CCTGCGA R30_B4 CTCTGCA TGCAGAG R30_B5 CCTAGGT ACCTAGG R30_B6 GGATCAA TTGATCC R30_B7 GCAAGAT ATCTTGC R30_B8 ATGGAGA TCTCCAT R30_B9 CTCGATG CATCGAG R30_B10 GCTCGAA TTCGAGC R30_B11 ACCAACT AGTTGGT R30_B12 CCGGTAC GTACCGG R30_B13 AACTCCG CGGAGTT R30_B14 TTGAAGT ACTTCAA R30_B15 ACTATCA TGATAGT R30_B16 TTGGATC GATCCAA R30_B17 CGACCTG CAGGTCG R30_B18 TAATGCG CGCATTA R30_B19 AGGTACC GGTACCT R30_B20 TGCGTCC GGACGCA R30_B21 GAATCTC GAGATTC R30_B22 GCATTGG CCAATGC R30_B23 TGACGTC GACGTCA R30_B24 GATGCCA TGGCATC Cy5primer CAAGCAGAAGACGGCATACG SEQIDNO9 AdaptorFW ACACTCTTTCCCTACACGACGCTCTTCCGATCT (1.b) SEQIDNO10 AdaptorRV GATCGGAAGAGCTCGTATGCCGTCTTCTGCTTG (1.a) SEQIDNO11

BIBLIOGRAPHY

[0144] Bailey, T. L., and Elkan, C. (1994). Fitting a mixture model by expectation maximization to discover motifs in biopolymers. Proc. Int. Conf. Intell. Syst. Mol. Biol. ISMB Int. Conf. Intell. Syst. Mol. Biol. 2, 28-36. [0145] Garcia-Cordero, J. L., and Maerkl, S. J. (2014). A 1024-sample serum analyzer chip for cancer diagnostics. Lab. Chip. [0146] Geertz, M., Shore, D., and Maerkl, S. J. (2012). Massively parallel measurements of molecular interaction kinetics on a microfluidic platform. Proc. Natl. Acad. Sci. U.S.A. 109, 16540-16545. [0147] Grant, C. E., Bailey, T. L., and Noble, W. S. (2011). FIMO: scanning for occurrences of a given motif. Bioinformatics 27, 1017-1018. [0148] Hens, K., Feuz, J.-D., Isakova, A., Iagovitina, A., Massouras, A., Bryois, J., Callaerts, P., Celniker, S. E., and Deplancke, B. (2011). Automated protein-DNA interaction screening of Drosophila regulatory elements. Nat. Methods 8, 1065-1070. [0149] Hong, J. W., and Quake, S. R. (2003). Integrated nanoliter systems. Nat. Biotechnol. 21, 1179-1183. [0150] Johnson, D. S., Mortazavi, A., Myers, R. M., and Wold, B. (2007). Genome-Wide Mapping of in Vivo Protein-DNA Interactions. Science 316, 1497-1502. [0151] Jolma, A., Yan, J., Whitington, T., Toivonen, J., Nitta, K. R., Rastas, P., Morgunova, E., Enge, M., Taipale, M., Wei, G., et al. (2013). DNA-Binding Specificities of Human Transcription Factors. Cell 152, 327-339. [0152] Laser, D. J., and Santiago, J. G. (2004). A review of micropumps. J. Micromechanics Microengineering 14, R35-R64. [0153] Maerkl, S. J., and Quake, S. R. (2007). A systems approach to measuring the binding energy landscapes of transcription factors. Science 315, 233-237. [0154] Matys, V., Kel-Margoulis, O. V., Fricke, E., Liebich, I., Land, S., Barre-Dirrie, A., Reuter, I., Chekmenev, D., Krull, M., Hornischer, K., et al. (2006). TRANSFAC and its module TRANSCompel: transcriptional gene regulation in eukaryotes. Nucleic Acids Res. 34, D108-110. [0155] Nielsen, R., Pedersen, T. A., Hagenbeek, D., Moulos, P., Siersbaek, R., Megens, E., Denissov, S., Brgesen, M., Francoijs, K.-J., Mandrup, S., et al. (2008). Genome-wide profiling of PPARgamma:RXR and RNA polymerase II occupancy reveals temporal activation of distinct metabolic pathways and changes in RXR dimer composition during adipogenesis. Genes Dev. 22, 2953-2967. [0156] Orenstein, Y., and Shamir, R. (2014). A comparative analysis of transcription factor binding models learned from PBM, HT-SELEX and ChIP data. Nucleic Acids Res. [0157] Siggers, T., Chang, A. B., Teixeira, A., Wong, D., Williams, K. J., Ahmed, B., Ragoussis, J., Udalova, I. A., Smale, S. T., and Bulyk, M. L. (2011). Principles of dimer-specific gene regulation revealed by a comprehensive characterization of NF-B family DNA binding. Nat. Immunol. 13, 95-102. [0158] Unger, M. A. (2000). Monolithic Microfabricated Valves and Pumps by Multilayer Soft Lithography. Science 288, 113-116. [0159] Weibel, D. B., Kruithof, M., Potenta, S., Sia, S. K., Lee, A., and Whitesides, G. M. (2005). Torque-actuated valves for microfluidics. Anal. Chem. 77, 4726-4733.