Rapid Development of Cell State Identification Circuits with Poly-Transfection

Summary

Complex genetic circuits are time-consuming to design, test, and optimize. To facilitate this process, mammalian cells are transfected in a way that allows the testing of multiple stoichiometries of circuit components in a single well. This protocol outlines the steps for experimental planning, transfection, and data analysis.

Abstract

Mammalian genetic circuits have demonstrated the potential to sense and treat a wide range of disease states, but optimization of the levels of circuit components remains challenging and labor-intensive. To accelerate this process, our lab developed poly-transfection, a high-throughput extension of traditional mammalian transfection. In poly-transfection, each cell in the transfected population essentially performs a different experiment, testing the behavior of the circuit at different DNA copy numbers and allowing users to analyze a large number of stoichiometries in a single-pot reaction. So far, poly-transfections that optimize ratios of three-component circuits in a single well of cells have been demonstrated; in principle, the same method can be used for the development of even larger circuits. Poly-transfection results can be easily applied to find optimal ratios of DNA to co-transfect for transient circuits or to choose expression levels for circuit components for the generation of stable cell lines.

Here, we demonstrate the use of poly-transfection to optimize a three-component circuit. The protocol begins with experimental design principles and explains how poly-transfection builds upon traditional co-transfection methods. Next, poly-transfection of cells is carried out and followed by flow cytometry a few days later. Finally, the data is analyzed by examining slices of the single-cell flow cytometry data that correspond to subsets of cells with certain component ratios. In the lab, poly-transfection has been used to optimize cell classifiers, feedback and feedforward controllers, bistable motifs, and many more. This simple but powerful method speeds up design cycles for complex genetic circuits in mammalian cells.

Introduction

The field of mammalian synthetic biology has rapidly progressed, from developing simple sense-and-respond parts in cultured cell lines to the optimization of complex networks of genes to address real-world challenges in diagnostics and therapeutics¹. These sophisticated circuits are capable of sensing biological inputs from microRNA profiles to cytokines to small molecule drugs, and implementing logic processing circuits including transistors, band-pass filters, toggle switches, and oscillators. They have also shown promising results in animal models of diseases like cancer, arthritis, diabetes, and many more^1,2,3,4,5. However, as the complexity of a circuit grows, optimizing the levels of each of its components becomes increasingly challenging.

One particularly useful type of genetic circuit is a cell classifier, which can be programmed to sense and respond to cellular states. Selective production of protein or RNA outputs in specific cellular states is a powerful tool to guide and program differentiation of cells and organoids, identify and destroy diseased cells and/or undesirable cell types, and regulate the function of therapeutic cells^1,2,3,4,5. However, creating circuits in mammalian cells that can accurately classify cell states from multiple cellular RNA and/or protein species has been highly challenging.

One of the most time-consuming steps of developing a cell classification circuit is to optimize the relative expression levels of individual component genes, such as sensors and processing factors, within the circuit. To speed up circuit optimization and allow for the construction of more sophisticated circuits, recent work has used mathematical modeling of cell classifier circuits and their components to predict optimal compositions and topologies^6,7. While this has shown powerful results so far, mathematical analysis is limited by the need to systematically characterize the input-output behavior of component genes in the circuit, which is time-consuming. Further, a myriad of context-dependent problems can emerge in complex genetic circuits, causing the behavior of a full circuit to defy predictions based on individual part characterizations^8,9.

To more rapidly develop and test complex mammalian circuits such as cell state classifiers, our lab developed a technique called poly-transfection¹⁰, an evolution of plasmid co-transfection protocols. In co-transfection, multiple plasmid DNA species are complexed together with a positively charged lipid or polymer reagent, then delivered to cells in a correlated manner (Figure 1A). In poly-transfection, plasmids are separately complexed with the reagent, such that the DNA from each transfection complex is delivered to cells in a de-correlated manner (Figure 1B). Using this method, cells within the transfected population are exposed to numerous combinations of ratios of two or more DNA payloads carrying different circuit components.

To measure the ratios of circuit components delivered to each cell, each transfection complex within a poly-transfection contains a constitutively expressed fluorescent reporter that serves as a proxy for cellular uptake of the complex. Filler DNA that does not contain any elements active within a mammalian cell is used to tune the relative amount of the fluorescent reporter and circuit components delivered to a cell in a single transfection complex and is discussed in more detail in the discussion. An example of filler DNA used in the Weiss lab is a plasmid containing a terminator sequence, but no promotor, coding sequence, etc. Cells with different ratios of circuit components can then be compared to find optimal ratios for gene circuit function. This in turn yields useful predictions for choosing promoters and other circuit elements to achieve optimal gene expression levels when combining circuit components into a single vector for genetic integration (e.g., a lentivirus, transposon, or landing pad). Thus, instead of choosing ratios between circuit components based on intuition or via a time-consuming trial and error process, poly-transfection evaluates a wide range of stoichiometries between genetic parts in a single-pot reaction.

In our lab, poly-transfection has enabled the optimization of many genetic circuits, including cell classifiers, feedback and feedforward controllers, and bistable motifs. This simple but powerful method significantly speeds up design cycles for complex genetic circuits in mammalian cells. Poly-transfection has since been used to characterize several genetic circuits to reveal their multi-dimensional input-output transfer functions at high resolution¹⁰, optimize an alternate circuit topology for cell state classification¹¹, and accelerate various published^12,13 and ongoing projects.

Here we describe and depict the workflow for using poly-transfection to rapidly optimize a genetic circuit (Figure 2). The protocol shows how to generate high-quality poly-transfection data and avoid several common errors in the poly-transfection protocol and data analysis (Figure 3). It then demonstrates how to use poly-transfection to characterize simple circuit components and, in the process, benchmark poly-transfection results against co-transfection (Figure 4). Finally, the results of poly-transfection show optimization of the cancer classifier circuit (Figure 5).

Protocol

NOTE: Table 1 and Table 2 serve as significant references for this protocol. Table 1 shows reagent scaling for reactions, and Table 2 shows DNA ratio arithmetic for an example poly-transfection described in the protocol (upper half) and for a possible follow-up experiment (lower half).

1. Preparing cells for transfection

1. Ensure that the culture of human embryonic kidney (HEK293) cells is 60%-80% confluent before initiating the protocol. To do this, seed 1 x 10⁶ cells in a 100 mm x 15 mm tissue culture Petri dish 2 days prior, and incubate at 37 °C with 5% CO₂.
   NOTE: Although our protocol focuses on HEK293 cells, other cell types may be substituted.
2. Prepare the media and cells for transfections as described below.
   1. Pre-warm at least 20 mL of a solution of Dulbecco's modified eagle medium (DMEM) with 10% fetal bovine serum (FBS) and 1% non-essential amino acids (NEAA; see Table of Materials) to 37 °C. Pre-warm at least 2.4 mL of Trypsin and 2.4 mL of phosphate-buffered saline (PBS) to 37 °C as well. Pre-warm reduced serum medium to ~16 °C.
      NOTE: All tissue culture work should be performed with care in a biosafety cabinet.
3. Resuspend the cells in the DMEM solution as described below.
   1. Aspirate and dispose of the current media. Dispense 2 mL of PBS onto the HEK293 cell culture to wash the cells. Aspirate and dispose of the PBS.
   2. Dispense 2 mL of Trypsin onto the HEK293 cell culture. Place the Petri dish in an incubator at 37 °C for 3 min or until the cells no longer adhere to the dish. Return the dish to the biosafety cabinet and dilute the cell solution by dispensing 8 mL of DMEM solution into the plate.
   3. Mix the solution by gently pipetting up and down several times. Aspirate all the media and place it in a 15 mL conical tube.
4. Centrifuge the cells at 300 x g for 3 min to pellet them. Aspirate the media (taking care not to aspirate the cells) and discard it. Resuspend the cells in 5 mL of DMEM solution, mixing by gently pipetting up and down.
5. Estimate the current cell concentration using an automated cell counter (listed in the Table of Materials). Seed six wells (in this example) in a 24-well plate with 1 x 10⁵ cells (for a seeding density of 50,000 cells/cm²).
6. Add the DMEM solution up to 500 µL (add DMEM solution to the wells first), and then label one well per treatment as the following: no color control, mKO2 control, TagBFP control, NeonGreen control, all color control, and poly-transfection 1. TagBFP, mKO2, and NeonGreen control wells are single color controls for all fluorescent proteins included in the poly-transfection.

2. Performing transfection

1. Prepare tubes for each DNA aggregate. Set aside 1.5 mL microcentrifuge tubes and label the tubes as: no color control, mKO2 control, TagBFP control, NeonGreen control, all color control, poly-transfection mix 1, and poly-transfection mix 2.
   1. Add 36 µL of reduced serum medium to the no color control, mKO2 control, TagBFP control, NeonGreen control, and all color control tubes. Add 18 µL of reduced serum medium to each of the poly-transfection mix 1 and poly-transfection mix 2 tubes.
      NOTE: The plasmid concentrations are assumed to be 150 ng/µL.
   2. Add 600 ng of filler plasmid to the no color control tube. Add 300 ng of mKO2 and 300 ng of filler plasmid to the mKO2 color control tube. Add 300 ng of TagBFP and 300 ng of filler plasmid to the TagBFP color control tube.
   3. Add 300 ng of constitutive NeonGreen plasmid and 300 ng of filler plasmid to the NeonGreen color control tube. Add 100 ng each of mKO2, TagBFP, and constitutive NeonGreen, as well as 300 ng of filler plasmid, to the all color control tube.
   4. Add 150 ng of mKO2 to the poly-transfection mix 1 tube. Add 75 ng of reporter NeonGreen plasmid and 75 ng of filler plasmid to the poly-transfection mix 1 tube.
   5. Add 150 ng of TagBFP to the poly-transfection mix 2 tube. Add 75 ng of L7ae plasmid and 75 ng of filler plasmid to the poly-transfection mix 2 tube.
2. Create the transfection master mix in a 1.5 mL microcentrifuge tube by combining 216 µL of reduced serum medium with 9.48 µL of transfection reagent (see Table 1 for reagent ratios and reaction scaling). Mix well by pipetting up and down, and set aside.
3. Add 1.58 µL of enhancer reagent to each of the no color control, single color control, and all color control tubes. Add 79 µL of enhancer reagent to each of the poly-transfection mix tubes. Mix each tube individually by pipetting vigorously.
4. Add the transfection master mix to each tube containing DNA.
   1. Add 37.58 µL of transfection master mix to each of the no color control, single color control, and all color control tubes. Mix each tube individually by pipetting vigorously.
   2. Add 18.79 µL of transfection master mix to each of the poly-transfection mix tubes. Mix each tube individually by pipetting vigorously.
5. Dispense the transfection mixes into the wells.
   1. Pipette 65.97 µL of each transfection mix for the no color, single color, and all color controls into the corresponding wells.
   2. Pipette 32.98 µL of the poly-transfection mix 1 into the poly-transfection well and swirl the plate quickly but gently in a tight figure-eight pattern along a flat surface to distribute the complexes effectively. Then, pipette 32.98 µL of the poly-transfection mix 2 into the same poly-transfection well and swirl the plate in the same fashion.
6. Place the plate in an incubator at 37 °C, with 5% CO₂ and without shaking, for a period of 48 h.
   NOTE: To increase cell viability, the cell media can be replaced every 6 h following transfection (though this is not always necessary, and with HEK293 cells and its derivatives, one must be careful not to detach the cells from the plate when changing the media).

Reagent	Amount			Scaling
Reduced serum medium for DNA mixture	36 μL per control tube, 18 μL per poly-transfection tube			0.05 μL Reduced serum medium/ng DNA per tube, with 10-20% extra volume to account for pipetting
DNA	300-600 ng per tube
P3000	1.58 μL per control tube, 0.79 μL per poly-transfection tube			0.0022 μL P3000/ng DNA per tube, with 10-20% extra
Reduced serum medium for Lipo master mix	36 μL per control tube, 18 μL per poly-transfection tube			0.05 μL reduced serum medium/ng total DNA, with 10-20% extra volume to account for pipetting
Transfection and enhancer reagent	1.58 μL per control tube, 0.79 μL per poly-transfection tube			0.0022 μL Lipofectamine 3000/ng DNA, with 10-20% extra

Table 1: Reagent scaling for transfections. The table indicates the correct ratio of reagent to include for the DNA quantity included in a single well. This can be used to scale reactions effectively and form master mixes. Quantities of reagent have been scaled to include a 20% excess.

3. Preparing cells for flow cytometry

1. Pre-warm at least 4.2 mL of the DMEM solution to 37 °C. Pre-warm at least 4.2 mL of Trypsin and 4.2 mL of PBS to 37 °C. Keep fluorescence-activated cell sorting (FACS) buffer solution at 4 °C until ready to use.
2. Resuspend the cells in the FACS buffer solution (PBS supplemented with 1% BSA, 5 mM ethylenediaminetetraacetic acid [EDTA], and 0.1% sodium azide [NaN₃], to reduce clumping; see Table of Materials) as described below.
   1. Aspirate and dispose of the current media in each well. Dispense 5 mL of PBS into each well to wash the cells. Aspirate and dispose of the PBS.
   2. Dispense 5 mL of Trypsin into each well. Place the plate in an incubator at 37 °C for 3 min, or until the cells no longer adhere to the dish. Return the plate to the biosafety cabinet and dilute the cell solutions by dispensing 5 mL of DMEM solution into each well.
   3. For each well, mix the solution by gently pipetting up and down several times. For each well, aspirate all the media and place it in a 15 mL conical tube.
3. Centrifuge the cells at 300 x g for 3 min to pellet them. Aspirate the media, taking care not to aspirate the cells and discard it. In each tube, resuspend the cells in 5 mL of FACS buffer solution, mixing by gently pipetting up and down.
4. Pass each cell suspension solution through a strainer (to remove clumps) into separate flow cytometry conical tubes. Keep these tubes on ice for no more than 1 h, and perform flow cytometry as soon as possible.

4. Performing flow cytometry

NOTE: Operating a flow cytometer requires proper training and knowledge of the necessary tasks. As software and equipment may vary and users should be trained generally, this section refers to specific operations that are useful to perform.

1. First, examine the cells transfected with the filler plasmid control (no color control) to select for cell characteristics and avoid anomalies (including aggregates, debris, etc.). While there are many combinations of parameters to distinguish cells, use the following three general options as a good way to visualize distinguishing features.
   1. Look at the side scatter area (log or linear scale per preference/cell type) versus the forward scatter area (linear scale).
   2. Look at the side scatter height (log scale) versus the side scatter width (linear scale).
   3. Look at the forward scatter width (linear scale) versus the forward scatter height (linear scale).
2. Next, look at the single color controls. Use the all color control to tune the instrument voltages, such that the signals from each fluorescent protein are normalized to equivalent arbitrary units (a.u.) of fluorescence. Next, run the single color controls for each fluorescent protein, which are used to set the compensation matrix, allowing for bleed-through correction.
   NOTE: Ideally, the full dynamic range of the fluorescence values should be visible. Further normalization of fluorescent protein signals can be done via conversion to standardized units (e.g., molecules of equivalent fluorescein [MEFLs; see Beal et al.¹⁵]). To enable MEFL conversion during analysis, run rainbow calibration beads. Such calibration is also useful for reducing instrument-to-instrument and day-to-day signal variation¹⁶.
3. Run the poly-transfection sample tube.
   NOTE: Where possible, it is recommended to run 1,000 x 10^ (^ = mixes) cells, as higher-dimensional poly-transfections need to be subdivided into more bins during analysis, and each bin needs enough cells (ideally >10) to make statistically significant comparisons.

5. Performing post-experiment analysis

1. Initially, use data from the controls (and, if applicable, the beads) to ensure accurate results. Use one of the available software tools to perform live cell gating (using gates outlined above), compensation, and autofluorescence correction.
   NOTE: We typically use either custom MATLAB code (e.g., https://github.com/jonesr18/MATLAB_Flow_Analysis or Cytoflow¹⁷), which has both a graphical user interface and a python library suitable for the pre-processing phase and for poly-transfection analysis.

Method	Complex	ng Fluorescent Marker	ng L7ae (fraction)	ng Reporter (fraction)	ng Filler DNA (fraction)	Total (ng)
Poly-transfection 1						600
→	1	150	75 (½)		75 (½)	300
→	2	150		75 (½)	75 (½)	300
Poly-transfection 2						600
→	1	150	25 (1/6)		125 (5/6)	300
→	2	150		125 (5/6)	25(1/6)	300

Table 2: DNA amounts for poly-transfection demonstrated in the protocol, and an example follow-up experiment with tuned plasmid ratios. The upper half of the table shows the composition of plasmids used in a simple poly-transfection experiment. The lower half shows the composition of an updated experiment that adjusts the plasmid ratios to better subsample a hypothetical concentration space, where the gene expression modulator is at a more optimal 1:5 ratio relative to its reporter, yielding more transfected cells to sample around this ratio.

Representative Results

In Figure 1, we compare co-transfection to poly-transfection. In a co-transfection, all plasmids are delivered in the same transfection mix, resulting in high correlation between the amount of each plasmid any single cell receives (Figure 1A). While the number of total plasmids delivered to each cell varies significantly, the fluorescence of the two reporter proteins in the individual cells across the population is well-correlated, indicating that the two plasmids are being co-delivered at a fairly constant ratio. Transfected cells may uptake few or many complexes, but since each complex has correlated amounts of each plasmid, the co-transfection only explores a small diagonal region of the concentration space between the two plasmids. In contrast, in a poly-transfection, plasmids are delivered in multiple transfection complexes, resulting in de-correlated delivery of plasmids in different transfection mixes (Figure 1B). Transfected cells uptake different combinations of the complexes, resulting in cells that contain varying dosages of plasmids from neither, one, or both complexes.

Figure 2 shows a representative poly-transfection workflow. In general, a poly-transfection experiment consists of the following steps: define the problem, create the transfection mixes, incubate the cells, flow cytometry, and analysis. There is also an optional step to repeat the poly-transfection with optimized part ratios if the initial poly-transfection results are not able to precisely narrow down the optimal part ratios. These steps are outlined in the protocol.

In Figure 3, some examples of well-performed co- and poly-transfections and common errors are provided. Figure 3A shows a well-performed co-transfection with a tight correlation between TagBFP and eYFP-expressing plasmids that were co-delivered. In contrast, the co-transfection in Figure 3B shows poor correlation between these two plasmids. In the shown figure, the poor correlation is due to adding enhancer reagent to the reduced serum medium before the plasmids were added. Such poor correlation in the experiments could also be due to different promoters driving expression of the fluorescent proteins, poor mixing during transfection mix creation, or allowing the complex to incubate for too short or too long of an interval.

Figure 3C shows a well-performed poly-transfection, with good coverage of the two-dimensional space and good compensation of any spectral bleed-through between fluorescent proteins. Figure 3D shows poly-transfection data with a low number of live cells, which is difficult to subdivide into enough bins with a sufficient number of cells in each bin for analysis. To ameliorate this issue, the starting cell population should be in good health and not overgrown, and the experimental toxicity reduced by using a different transfection reagent and/or transfecting with less total DNA. Figure 3E shows a poly-transfection in which the transfection efficiency was poor, resulting again in sparse coverage of the two-dimensional space for analysis. In this case, the number of cells passing morphology gating is high, but the cells do not express fluorescent proteins. To improve efficiency, transfection reagent choice should be optimized and the manufacturer's protocol followed closely. Each transfection mix must contain an equivalent amount of DNA mass, ensuring the cells have an approximately equal chance of receiving each type of complex, and each transfection mix should made in accordance with the kit instructions. Different transfection kits are optimal for different cell types; the kit that gives the best efficiency in the cell type of interest should be used. Finally, some cell types are difficult to transfect, regardless of the kit used. In these cases, a larger number of cells should be transfected, and as many as possible collected during flow cytometry to ensure a sufficient number of cells to analyze. Figure 3F shows poly-transfection data where one of the fluorescent protein markers is clearly showing spectral bleed-through into another fluorescent protein. Single color controls should always be run for all fluorescent proteins in the system and used to generate a compensation matrix that should be applied to all poly-transfections prior to analysis.

When performing poly-transfection experiments for the first time (overall or for a new experimental system), one should perform benchmarking against standard co-transfection. One approach is to individually measure the input-output transfer function for key system parts using both co-transfection (via tuning DNA dosages) and poly-transfection.

In Figure 4, we demonstrate benchmarking of the translational repressor L7Ae, adapted here from the original poly-transfection publication¹⁰. L7Ae is an RNA binding protein that recognizes RNA kink-turn (KT) motifs²⁰; when two KTs (2xKT) are placed in the 5' untranslated region (UTR) of an mRNA, downstream open reading frame (ORF) translation can be effectively suppressed by L7Ae²¹. L7Ae has been previously used to build RNA-based cell type classifiers²¹ and other RNA-based circuits²². To test L7Ae by standard co-transfection, one can tune the ratios of the plasmids encoding L7Ae and its target reporter within different transfection mixes (Figure 4A and Table 3). For poly-transfection, one should independently deliver constitutive L7Ae and corresponding 2xKT reporter in separate transfection mixes, such that a wide range of plasmid ratios are delivered to the cells (Figure 4B). After performing transfection and flow cytometry, data analysis can be performed. To make the data comparable, one can collate the individual co-transfection results into a single dataset, then perform a similar multi-dimensional binning to the poly-transfection data (Figure 4C,D). Comparing the dose-response curves of L7Ae repressing the output reporter, it can be seen that the median output level per bin is very similar between the co-transfection and poly-transfection data (Figure 4E-G).

In general, we have seen that poly-transfection data correlates well with co-transfection data across broad DNA ratios, with less precision at highly skewed DNA dosage ratios (in part due to lower cell coverage in bins for poly-transfection experiments, especially for parts that are very strongly active at low plasmid dosages). To improve the measurement accuracy for such sensitive parts, their expression level can be reduced with a weaker promoter, upstream open reading frames¹⁸, or microRNA silencing-mediated fine-tuners (miSFITs)¹⁹.

Figure 5 demonstrates a successful application of poly-transfection for optimization of a cell type classifier, again adapted from Gam et al¹⁰. The classifier is a relatively simple design that produces an output in response to expression of miR-21-5p, a miRNA over-expressed in many tumor cells²³. In the absence of miR-21, classifier output is repressed by a bacterially derived transcriptional repressor, BM3R1²⁴, whose adaptation was previously shown to work in mammalian cells²⁵. When miR-21 is present, it binds four target sites placed in both the 3' and 5' UTRs of BM3R1, knocking down its expression and thereby allowing output transcription (Figure 5A). To optimize this system, the three circuit components were delivered in separate transfection mixes: (1) BM3R1 (with miR-21 target sites), (2) output reporter (mKO2), and (3) Gal4-VP16, which activates transcription of the output (Table 4). Note that the output promoter operates on the logic (Gal4-VP16) and not BM3R1, which suppresses output even in the presence of Gal4-VP16²⁵. Each part encodes a transfection marker, TagBFP, mNeonGreen, and iRFP720, respectively, on the same plasmid to indicate the relative DNA dosage of each complex. In general, increased Gal4-VP16 expression should increase reporter mKO2 output, while increased BM3R1 expression should result in lower mKO2 output. Because BM3R1 is knocked down by miR-21-5p, output expression should be higher in HeLa cells, which have higher levels of miR-21-5p and thus express less BM3R1 than in HEK cells.

After poly-transfection into both HEK293 and HeLa cells, we obtained a 3D distribution of different plasmid ratios (Figure 5B-HEK cells). Gam et al.¹⁰ subsampled this distribution at various ratios by considering cells within a particular Euclidean distance from a ratio of interest; this distance should be wide enough to include enough cells to allow statistically significant results from the analysis, but narrow enough to avoid unnecessary noise from including too broad a set of component ratios of the three parts. Gam et al.¹⁰ then used an optimization algorithm to identify the ratio of parts that maximized classification accuracy for co-transfection (i.e., transfected HeLa cells are positive for output expression while transfected HEK cells are not). The optimal ratio of the parts was found to be 10.9:1.5:1: Gal4-VP16:output:BM3R1; cells subsampled around the optimal ratio within the whole poly-transfection space are shown in Figure 5C. This subsampling predicted that, at this ratio, the co-transfected circuit has a 91% specificity, 62% sensitivity, and 77% accuracy when classifying HEK293 versus HeLa cells (Figure 5D). Co-transfection with plasmid ratios set to this optimum yielded even better results: 99% specificity, 68% sensitivity, and 84% accuracy¹⁰. Further, the ratios guided the implementation of a single-plasmid version of the circuit, with relative expression tuned by using different truncated promoters and upstream ORFs (uORFs), yielding a circuit with 91% specificity, 90% sensitivity, and 90% accuracy¹⁰. Thus, poly-transfection is a powerful tool to guide the design of cell classifiers and gene circuits more broadly.

Figure 1: Comparing co-transfection and poly-transfection. (A,B) Overview and comparison of plasmid delivery with co-transfection of two plasmids and poly-transfection of two plasmids. For each transfection method, the leftmost diagram shows the formation of transfection complexes between negatively charged DNA and positively charged lipids. In these examples, each colored plasmid (blue and red) encodes the expression of a different fluorescent protein. The center diagram shows examples of plasmid delivery to cells and also a schematic for the expected distributions in a histogram or scatter plot. Color intensity on the histogram corresponds to fluorescence from the corresponding plasmid color. The rightmost diagram shows real data from cells transfected using each given method. (A) In a co-transfection with two different plasmids, both plasmids are mixed together before adding transfection reagent, resulting in highly correlated packaging of the two plasmid species. In actual co-transfection data, cells exhibit correlated delivery of both plasmids (right). (B) In a poly-transfection, each set of co-delivered plasmids corresponding to a circuit part and a transfection marker is mixed with the transfection reagent separately, resulting in complexes that contain only those plasmids (left). In actual poly-transfection data, cells explore a wide range of concentration space with many different plasmid stoichiometries explored simultaneously (right). This figure has been modified from¹⁰. Please click here to view a larger version of this figure.

Figure 2: Poly-transfection workflow. Step 1: Define the problem. Start with a system with multiple component parts for which the ideal ratio between parts is unknown, and choose a starting ratio between component parts. Step 2: Create the transfection mixes. Each transfection mix contains a circuit component and a fluorescent transfection marker. The amount of circuit part delivered to a cell correlates with the fluorescence of the marker. Step 3: Incubate the cells.In a typical workflow, we incubate cells for 48 h between transfection and flow cytometry. Step 4: Flow cytometry.Run the appropriate controls, as discussed in the protocol, and then run the samples. Step 5: Analysis.Use the transfection markers to bin the cells according to the amount or ratios of parts that the cell received. Use the output protein(s) to measure circuit performance. Find the bins/ratios of parts that optimize circuit performance. Step 6: Repeat with optimized part ratios (optional).If the optimal bins/ratios are at highly skewed ratios, the pilot poly-transfection may not narrow down the optimal part ratios precisely. Repeat the poly-transfection with tuned part ratios, such that a cell that received an equal amount of each transfection mix now receives a close-to-optimal ratio of parts, as determined by the prior round. This figure has been modified from¹⁰. The image of an incubator was used from Servier Medical Art and the image of a cytometer from BioRender. Please click here to view a larger version of this figure.

Figure 3: Representative positive and negative poly-transfection results. (A) Well-performed co-transfection of two fluorescent reporters, showing a tight correlation. A total of 20,000 cells are plotted both here and in (B) for comparison. It's a good idea to perform a similar small trial co-transfection of two fluorescent reporters before starting a poly-transfection experiment, to ensure that plasmids are well correlated within transfection mixes. (B) Noisier co-transfection: plasmids are not well correlated within each transfection mix, which can cause fluorescent markers in a poly-transfection to be a poor marker of plasmid ratios. (C) Well-performed poly-transfection results showing good cell count, transfection efficiency, and compensation. (D) Low live cell count, which does not allow for sub-sampling into bins with statistically significant numbers of cells. (E) Poor transfection efficiency, which does not allow for sub-sampling into bins with statistically significant numbers of cells. (F) Lack of appropriate compensation, which causes fluorescence data to be a poor proxy for the amount of fluorescent protein in the system. Use single color controls to determine an ideal linear compensation matrix, and apply it to the data before further processing. All color controls are also recommended depending on the choice of software. Please click here to view a larger version of this figure.

Figure 4: Benchmarking poly-transfection against co-transfection. (A) In a co-transfection, one ratio of translational repressor L7Ae to fluorescent reporter plasmids can be tested in each experimental well. (B) In poly-transfection, many ratios of L7Ae to reporters can be tested in the same well. See Table 3 for details of the poly-transfection plasmid mixes. (C,D) Binning workflow for benchmarking poly-transfection against multiple co-transfections. A total of 10 biexponentially-spaced bins were assigned for each transfection marker dimension, which approximate the levels of each of the two plasmids. Here, bins that denote different levels of plasmid #2 are shown. Binning was performed on both a collated set of 11 co-transfection samples spanning various plasmid ratios (C) and data from a single poly-transfection, comprising about 500,000 cells each (D). Colors correspond to sets of bins defined by gene 2 (TagBFP) levels. (E-G) Benchmarking poly-transfection against co-transfection for a representative circuit. Median output fluorescence was evaluated for the cells in each bin and compared between methods for the representative system, L7Ae translational repression. (E) Constructs for measuring L7Ae activity. mKO2 fluorescence serves as an estimate for the delivery of L7Ae (gene #1), while TagBFP fluorescence serves as an estimate for delivery of the regulated mNeonGreen output (gene #2). (F) Multi-dimensional titration curves for L7Ae. Each line represents the set of bins at one level of TagBFP, as denoted in (C) and (D). Solid lines denote poly-transfection data, while dashed lines denote co-transfection data. At each binned level of reporter plasmid, reporter output decreases as L7Ae increases. (G) Visual comparison between co-transfection and poly-transfection for the L7Ae system. Each point represents the measured output in corresponding bins for poly-transfection and co-transfection measurements, where more equivalent values are closer to the red 1:1 line. Overall, the differences observed between poly-transfection- and co-transfection-derived are low, which provide confidence in the reliability of the poly-transfection method. This figure has been modified from¹⁰. Please click here to view a larger version of this figure.

Figure 5: Data analysis for poly-transfection. In poly-transfection, cells can be grouped into bins or slices for analysis. Figure 4 showcases a form of analysis where the cells are binned into slices that correspond to levels of a system component, such as reporter plasmid level, which can be useful for model fitting and understanding dosage responses. Another useful strategy is to analyze circuit performance at different ratios of circuit components. (A) Diagram of a classifier circuit for optimization. Levels of TagBFP, NeonGreen, and iRFP720 correspond to levels of BM3R1, output mKO2, and Gal4-VP16, respectively. See Table 4 for details of the poly-transfection plasmid mixes. (B) Experimental setup of poly-transfection mixes. (C) Flow cytometry data collected from poly-transfection with the circuit in (A). The levels of each of the three reporter fluorescent proteins as a result of circuit poly-transfection into HEK cells, showcasing the wide range of circuit component ratios present in the data. Similar results were obtained from circuit transfection in both HEK and HeLa cells. (D) Subsampling poly-transfection at a particular ratio. To analyze the data, we scanned a large number of ratios between circuit components and determined classifier performance at each ratio. As an example, we show here one particular ratio that demonstrated good performance (Gal4-VP16 = 435 ng of DNA, reporter = 60 ng, and BM3R1 = 40 ng). Plotted in blue is the corresponding ratio of the fluorescent markers for the three different circuit components. Next, we subsampled the data by only considering points within a particular Euclidean distance from the fluorescence trajectory. We subsampled cells at the same trajectory from both HEK and HeLa transfections. (E) Comparison of circuit performance at the same trajectory in HEK and HeLa cells. By comparing statistics such as sensitivity, specificity, and accuracy of classification across many ratios, component ratios can be optimized to generate an ideal genetic circuit. This figure has been modified from¹⁰. Please click here to view a larger version of this figure.

Complex	ng L7Ae plasmid	ng Reporter plasmid	OptiMEM (µL)	P3000 (µL)	Lipo 3000 (µL)
1	250		75	1.5	1.5
2		250	75	1.5	1.5

Table 3: Transfection mixes corresponding to Figure 4. HEK293 cells were poly-transfected with these transfection mixes in one well of a 24-well plate.

Complex	ng BM3R1 plasmid	ng Gal4-VP16 plasmid	ng Reporter plasmid	OptiMEM (µL)	P3000 (µL)	Lipo 3000 (µL)
1	900			75	1.5	1.5
2		900		75	1.5	1.5
3			900	75	1.5	1.5

Table 4: Transfection mixes corresponding to Figure 5. HEK293 and HeLa Cells were each poly-transfected with these transfection mixes in one well each of a 6-well plate.

Discussion

Rapid prototyping methods such as computer-aided design (CAD), breadboarding, and 3D printing have revolutionized mechanical, electrical, and civil engineering disciplines. The ability to quickly search through many possible solutions to a given challenge greatly accelerates progress in a field. We believe that poly-transfection is an analogous technology for biological engineering, enabling rapid prototyping of genetic circuits. Additionally, other rapid prototyping technologies require hands-on sequential iteration of multiple possible solutions, whereas poly-transfection is able to explore many solutions simultaneously. Poly-transfection enables a large number of combinations of genetic circuit component ratios to be tested in a single well and has been used to optimize several published genetic circuits^10,11,13. The experimental protocol is a simple extension of standard co-transfection, allowing for straightforward adoption by many mammalian cell researchers. Generally, a very close agreement between poly-transfection and co-transfection results is seen¹⁰ (an example is Figure 4). Thus, poly-transfection can be used in most cases where plasmid ratios are titrated within co-transfection mixes and outputs are measured at the single-cell level.

The complexity and scale of poly-transfection increase with the complexity of genetic circuits to optimize. As the number of dimensions to analyze increases, the combinatorial ratios of different parts and the number of cells needed for their analysis increases exponentially. For example, to optimize the classifier in Figure 5, cells were transfected in a 6-well plate format, and data was collected on at least 1.5 million live cells. We have also optimized a classifier with an additional circuit component¹⁰, necessitating transfection at a 10 cm plate scale and the collection of millions of cells. At that scale and above, DNA production is time-consuming, and transfection reagents are expensive. That said, poly-transfection uses orders of magnitude less cells, DNA, and transfection reagents than testing a comparable number of circuit component ratio combinations in individual wells. Additionally, a considerable amount of time is saved by making orders of magnitude fewer transfection mixes.

Poly-transfection enables advanced analysis methods for flow cytometry data. The two main approaches are (1) binning cells according to the expression of each transfection marker and (2) extracting cells at defined ratios of transfection markers, thereby simulating co-transfection. The former selects cells with a specific combination of DNA dosage for each complex (and thus each part), yielding input-output transfer functions. The latter selects cells at a specific ratio of DNA dosages for each part, simulating co-transfection at such a ratio of plasmid DNA masses. Plotting the expression level of circuit output reporter(s) in the subsampled selections gives a measure of output at the defined levels or ratios of inputs. For circuit optimization, one can identify the best performing bins/ratios as defined by the purpose of the circuit-in the case of cell classifiers, these are the bins/ratios where the circuit is ON in the target cell type and OFF in non-target cell types. When choosing a bin size, one should take into consideration the level of precision required, as well as the number of cells per bin. Smaller bins focus on the ideal combination of circuit components more precisely, but if a bin has too few cells, it can introduce undesirable noise to the measurements.

Future improvements in computational analysis of poly-transfection data may facilitate optimization even with low cell coverage per bin/ratio. Currently, we exclude data from bins with fewer than a set threshold of cells (e.g., 10) to avoid overly noisy measurements. In combination with repeated measurements, this enables more accurate and precise calculations of dose-response curves, classifier accuracies, and other metrics defined on a per-bin basis. However, each cell in poly-transfection can be considered to be an independent experiment, measuring at a fine resolution the circuit output(s) at a precise level of inputs. With this in mind, mechanistic and phenotypic models of dose responses and circuit optimizations can directly fit to the distribution of gene expression from each marker and reporter, rather than their binned summary statistics¹⁰. This enables more robust measurements and takes advantage of the many individual data points collected per experiment. Such modeling approaches could thereby yield predictive circuit characterization and optimization even with sparsely sampled high-dimensional poly-transfections. Further, machine learning methods such as response surface methodology and random forest regression may be used to analyze relatively sparse, high-dimensional data¹⁰.

An alternate approach to increasingly large poly-transfections is to sequentially optimize circuits using hierarchies of poly-transfections. In this approach, first, poly-transfection is used to optimize a module comprising a subset of genetic components within a larger circuit. Then, these optimized modules are delivered as separate poly-transfection mixes, enabling one to find the optimal ratio/dosage of each circuit module. This approach uses a significantly lower number of cells, DNA, and transfection reagents. However, groups of components are not necessarily modular with respect to other components, and thus an optimal ratio of component parts in a module measured in isolation could be suboptimal within the larger circuit context⁸.

Poly-transfection methods are also limited by the laser/filter configuration of the flow cytometer used to collect data. In addition to measured fluorescent outputs, each transfection mix contains a fluorescent protein marker. Thus, it is critical to select fluorescent proteins that can be well compensated on the cytometer. We previously systematically analyzed the bleed-through of a panel of 22 fluorescent proteins on a five-laser flow cytometer (the BD LSRFortessa), and found that the set of Sirius, TagBFP, mNeonGreen, mKO2, and iRFP720 could be used together and compensated well without significant issues¹⁰. However, the optimization of larger circuits may require advanced cytometry methods, such as spectral cytometry to separate fluorescent proteins with more overlapping spectra, which our lab is currently optimizing with up to eight fluorescent proteins. Databases such as the fluorescent protein database (https://www.fpbase.org) are useful for selecting fluorescent proteins with particular spectral overlap. However, this process can be complicated by various factors, some of which are specific to particular cytometers. For instance, the usage of certain red proteins like tdTomato may result in undesirable bleed-through into blue channels only in certain laser/filter configurations¹⁰. Additionally, several far-red proteins have shown non-linear relationships between DNA dosage and fluorescence output¹⁰, reducing their use as effective markers for DNA dosage.

For consistency across experiments that are performed with varying numbers of complexes (one, two, three, etc.), it is useful to use the same fractional amount of plasmid per transfection mix. For example, if testing a three-gene system with each gene encoded in one of three plasmids (A, B, and C), one could co-transfect the circuit plasmids at a 1:1:1 ratio along with an equal ratio of a plasmid encoding a transfection marker. This would make each plasmid one-quarter of the total mass of the mix, and thus approximately one-quarter of the mass of each transfection complex that is formed. When poly-transfecting A, B, and C in separate complexes with their own reporters, instead of mixing the plasmids 1:1 with reporters or maintaining their total DNA mass, it is more consistent to keep each plasmid at one-quarter of the mass of each complex, with filler DNA taking up the remaining mass (see Table 5 for examples). This is because the distribution of gene expression among transfected cells depends less on the total mass of DNA delivered in the complexes, and more on the fractional amount of DNA in each complex¹⁰. If ratios other than 1:1:1 are desired, calculate the corresponding fractions of the total DNA mass in the transfection mix for each part. For example, Table 5 [bottom] shows a 1:1:4 ratio.

Method	Complex	ng A (fraction)	ng B (fraction)	ng C (fraction)	ng Reporter (fraction)	ng Filler DNA (fraction)	Total (ng)
Co-transfection	1	150 (¼)	150 (¼)	150 (¼)	150 (¼)	0 (0)	600
Poly-transfection							600
→	1	50 (¼)			50 (¼)	100 (½)	200
→	2		50 (¼)		50 (¼)	100 (½)	200
→	3			50 (¼)	50 (¼)	100 (½)	200
Co-transfection	1	75 (⅛)	75 (⅛)	300 (½)	150 (¼)	0 (0)	600
Poly-transfection							600
→	1	25 (⅛)			50 (¼)	125 (⅝)	200
→	2		25 (⅛)		50 (¼)	125 (⅝)	200
→	3			100 (½)	50 (¼)	50 (¼)	200

Table 5: Examples of using filler DNA for consistency across co- and poly-transfection experiments. Here, two generic examples of co-transfecting three plasmids with the same amount of filler DNA is shown. In the top example, the plasmids are all at equal masses. In the bottom example, one plasmid is delivered at a higher mass compared to the others. The fraction of total DNA that would be used in a co-transfection complex is transmuted to poly-transfection complexes, with the remaining amount of DNA (up to total amount for each complex, which is fixed and equal across complexes) constituted with filler DNA.

The overall objective of filler DNA is to maintain similar dosages of plasmids per cell, regardless of the number of unique transfection mixes. As a crude approximation of this principle, separating three plasmids into three mixes while maintaining the same DNA mass of each (and the appropriate ratio of transfection reagent) reduces the percent of cells receiving a given complex by one third compared to a single mix containing all three plasmids. However, each transfected cell subsequently receives a ~3x higher gene dosage. Reducing the relative amount of the plasmids alone without filler DNA is thus insufficient to maintain consistent plasmid dosages, as this simply decreases transfection efficiency without altering the underlying distribution of expression among the transfected cells¹⁰. On the other hand, filler DNA allows for adjusting the DNA dosage without affecting the overall transfection efficiency (though detectable transfected cells may decrease due to lower signal)¹⁰. Following the crude approximation above, reducing the fraction of DNA in the poly-transfection mixes to one-third and adding filler DNA maintains the relative gene dosages compared to the original co-transfection. Filler DNA therefore ensures efficient and accurate transfection complex formation.

To alter the range of expression of a gene of interest covered by its reporter, the fraction of each test plasmid and/or promoters used to drive the gene there can be tuned relative to the reporter. If a part is strong/highly active at low DNA dosages, using a lower DNA fraction can help center the dynamic range of the part in the middle of the fluorescence distribution of the reporter in transfected cells. Likewise, for parts that are weak/only active at high DNA dosages, using a higher fraction can be useful. However, care must be taken with such tuning, as reducing DNA fractions too much increases the stochasticity of delivery to the cells compared to the reporter¹⁰, and high gene expression levels can overload cellular gene expression machinery^12,14. To avoid stochasticity, and in cases where each gene is delivered to cells at a fixed ratio (e.g., if the circuit is to be generated as a single lenti, PiggyBac, or Landing Pad vector), the relative expression of each gene can be tuned using stronger/weaker promoters, small uORFs¹⁸, and/or miSFITs¹⁹.

Though the protocol describes a reverse transfection technique, poly-transfection is also possible with forward transfection. In reverse transfection, the cells are simultaneously seeded and transfected; in forward transfection, the cells are transfected ~24 h after first plating at approximately one-half the density that would be used in reverse transfection (to allow for cell division by the time of transfection). In general, reverse transfection is more efficient but also more toxic, and we have noticed some differences in the shape of transfection distributions based on the reagent, transfection time, and cell line. Thus, the transfection method of choice should be optimized for each cell line to maximize the number of transfected cells and coverage of the multi-dimensional concentration space of plasmid dosages per cell.

Overall, poly-transfection enables the rapid optimization of mammalian genetic circuits. Many possible ratiometric combinations of circuit components can be easily tested in a single well. Additionally, since poly-transfection contains more information about the levels of each part in a system than a conventional transfection, it has been found to be highly valuable for characterizing the behavior of various genetic parts^10,11,12,13. The adoption of poly-transfection is anticipated to accelerate the pace of developing new and improved gene circuits for use in mammalian cells.

Materials

Name	Company	Catalog Number	Comments
15mL Corning Falcon conical tubes	ThermoFisher Scientific	14-959-53A
24-well petri dish	Any company of choice		(Non-pyrogenic, Sterile, RNase, DNase, DNA and Pyrogen Free)
Bovine serum albumin	NEB	B9000S
Centrifuge	Any company of choice		Capable of exposing 15mL Falcon tubes to 300 rcf
Countess 3 Automated Cell Counter	ThermoFisher Scientific	AMQAX2000
Countess Cell Counting Chamber Slides	ThermoFisher Scientific	C10228
Cytoflow	Non-commercial software package		https://cytoflow.readthedocs.io/en/stable/#
DMEM	VWR	10-013-CV	Use the correct media for your cell type
EDTA	ThermoFisher Scientific	03690-100ML
Fetal bovine serum	Sigma Aldrich	F4135
HEK cells	ATCC	CRL-1573	Use the relevant cell type for your experiments. HEK cells tend to transfect very efficiently.
HeLa cells	ATCC	CRL-12401	Use the relevant cell type for your experiments.
Lipofectamine 3000 and P3000 enhancer	ThermoFisher Scientific	L3000001	Use the correct reagent for your cell type; transfection and enhancer reagent
LSRFortessa flow cytometer	BD Biosciences	N/A
MEM Non-Essential Amino Acids Solution	Gibco	11140050
Microcentrifuge Tubes, 1.5 mL	Any company of choice
Opti-MEM	ThermoFisher Scientific	31985070	reduced serum medium
Phosphate buffered saline	ThermoFisher Scientific	70011044
Rainbow calibration beads	Spherotech	URCP-100-2H
Sodium azide	Sigma Aldrich	S2002
Trypsin	VWR	25-053-CI