Diagnostics: Fusion Protein-Based Paper Microfluidics
Since we had to determine baseline conditions of fusion protein-based paper microfluidic devices as a new realm in and of itself, all of our experimentation is foundational and we had to invent the assays to test the devices we created. This trial and error was necessary when exploring protein constructs, paper substrates, channel configurations, and printing methods. After that, we had to test whether our system was viable at all given the conditions we established for it.
Protein Constructs
Our fusion protein constructs revolved around three functional domains: a double cellulose binding domain, a super folder GFP, and a hydrophobic domain. By connecting these domains in various configurations using flexible linkers, we synthesized three novel fusion proteins, as well as a past biobrick from Imperial iGEM 2014, that are able to complement each other in different ways. We expand on the design behind each of the novel parts in the biobrick descriptions linked below.
We were able to synthesize, induce, and extract lysate containing all four of the constructs listed above, and verify their functionality:
*The functionality of BBa_K3260031 is showcased in the “Implementing Protein-Based Paper Microfluidics” section below*
The cell lysate is fluorescing very intensely under UV light, which matches the excitation wavelength of superfolderGFP. This, along with sequencing of the constructs, indicates that we successfully expressed BBa_K1321348 in lysate. We experimented with all of our constructs suspended in their cell lysate, because it saved time and also all of our constructs contained strong affinity domains that we believed out-competed the natural lysate proteins. In BBa_K1321348, BBa_K3260019, and BBa_K3260021, it is the double cellulose binding domain, and in BBa_K3260022, it is one half of a strongly complementary leucine zipper pair.
Above is a test of the functionality of two of our new fusion proteins, BBa_K3260036, and BBa_K3260034. If these two parts assemble correctly on the cellulose surface, there should be fluorescence, which is what we observed here. This means that the leucine zipper components of each protein aligned and held together, allowing the two halves of GFP to stay annealed to the cellulose after washing, thus allowing them complete their chromophore and fluoresce. Because we see fluorescence (highlighted in the color swatch of each sample), the success of the leucine zippers and therefore the functionality of the last two functional domains can be inferred as well, the double cellulose binding domain anchoring the whole complex to the cellulose surface, as well as the hydrophobic domain sticking up from the N-terminal region of BBa_K3260034, and therefore the assembled protein complex. The successful assembly of these two biobricks not only shows the applicability of these constructs for paper-based microfluidics, but it also shows the potential of split-GFP reporting for fusion proteins in functional applications.
Paper Substrates
We tested three main types of “paper" for substrate efficacy: Whatman’s No.4 Filter Paper, Bacterial Cellulose (K. rhaeticus), and archival printer paper. Originally, it was believed that Whatman’s No. 4 chromatography paper would be applicable to the cellulose binding proteins [4]. However, due to some impurities inherent to plant cellulose filter paper (such as the presence of hemicellulose and lignin), and some unpromising results from our preliminary hydrophobicity tests, we decided to try bacterial cellulose from K. rhaeticus. We were able to obtain samples of K. rhaeticus bacterial cellulose from the SoundBio iGEM team in Seattle, Washington (this collaboration is expanded on in our collaborations page). We dried the product out on paper, and on it printed our channel designs (more details on printing below). When we tested our printed proteins on bacterial cellulose, there was no visible capillary action - it seems as if the bacterial cellulose already had some inherent hydrophobic properties that were obscuring the functionality of our hydrophobic proteins.
The top image is of preliminary hydrophobicity tests on Whatman’s No. 4 Filter Paper. The squares were rocked in hydrophobic lysate and dried for varying lengths of time, but when dropping ~2 ul of water on each sample, no increased hydrophobicity or resistance to water was observed between the experimental lysate and control. We then turned to bacterial cellulose of K. rhaeticus which, as mentioned above and displayed in the bottom image, displayed some baseline hydrophobicity that rendered it inapplicable for the functionalization of hydrophobic proteins.
We then decided to try printing our designs on archival grade printer paper. As will be expanded on below, printing our devices on archival paper yielded successful results and was deemed the most promising way forward for developing protein based paper microfluidics.
Printing Methods
After trying screen printing and manual painting of proteins, we settled on ‘hacking’ the ink cartridges of commercial printers to print out our protein lysate instead of ink. Our team worked with the Canon Pixma MP950 and the Canon Pixma TR4520 to print our proteins. While the MP950 is a higher end Canon printer, for this particular purpose it was not well suited. Error messages that halted function (that are apparently systemic to all units) were inhibitory when we were trying to make the printer “forget” about tampered parts of its hardware. Also the ink cartridges had reader chips that most likely contained information about the color in the cartridge, and whether it had been used before. For a project that is looking to tamper with cartridges, this was not an ideal characteristic. The TR4520, on the other hand, worked extremely well for our needs. It held only two cartridges, color (containing compartments with blue, red, and yellow) ink and black ink, which made it extremely easy to control exactly what was being printed. We left the color cartridge intact, and in our print surrounded where we wanted our protein to be (signified with black ink on our printing template) with a bold blue background. This eased visualization of where our protein should have been printed, and also provided a helpful negative control when our testing liquid ran past its final destination in our microfluidic device. This printer was also extremely economical and easy to attain. We formulated the following general procedure to load and print protein lysate into ink cartridges:
Prepping Cartridge for Lysate Loading:
- Locate the lid, and the seam that connects it to the body of the cartridge.
- Using a large screwdriver, press into this seam until the cartridge pops off, or can be torn off. Try to make this as clean a separation as possible, as the lid needs to be 3. reattached for adequate loading into the printer.
- Remove the white sponge from the cartridge, and clean with DIwater until runoff is clear when sponge is squeezed.
- Rinse inside of cartridge and printhead with DIwater. The printhead is the interface between the sponge and the paper to be printed on. Rinse until runoff from all directions is clear.
- Let both components dry/ dry off cartridge with paper towel.
(used black ink cartridge for protein lysate)
Loading Experimental Lysate:
- Pipette ~1ml of experimental lysate onto printhead pad embedded on the inside of the cartridge. This will aid flow of protein through printhead.
- Pipette ~2ml of experimental lysate onto depressed square shape on white sponge, making sure it’s well saturated.
- Pop sponge back into the cartridge, making sure the depressed square on the sponge itself is aligned with the printhead pad.
- Put lid back on cartridge, using scotch tape to affix it to the body.
- Load cartridge back into the printer.
- Go forth and print.
Designing a Template
The channel designs we chose to implement to test the viability of microfluidics were based off of Liu et. al [1] where there were “sample zones” in which their liquid would be placed, and then “testing zones” where the liquid would hopefully end up after traversing microfluidic channels of various widths. Their use of a “flower”-based mask allowed us to test a range of channel widths at the same time. We also deconstructed this “flower” to separate the variables we were testing and examine them side by side. The print-out template is shown below, with the dimensions illustrated to the side.
The dimensions we chose were meant to show the range of channels normally encountered in both wax channel microfluidics and wider range PDMS devices. We found these designs to cover the full range of our protein functionality. Since we commandeered the black ink cartridge for our experimental lysate, we outlined our desired microfluidic channels in black ink in out file to print out. Then, to delineate where the protein was being printed and where it wasn’t, we surrounded our protein “zone” with blue, which would be printed out as normal printer ink.
Implementing Protein-Based Paper Microfluidics
As mentioned above, the implementation of archival printing paper into our printer system yielded positive results for our protein-based microfluidic devices. To test our microfluidic devices, we used water with green food-coloring.
BBa_K3260031 Single Layer Print:
The video above shows one of our tests with printing with protein lysate. The lysate that we loaded into the black ink cartridge contained BBa_K3260031(our dCBD + our hydrophobic, or Radek) domain. Directly below is a labelling of this video with channel widths for each component in our testing template.
The blue regions are plain printer ink, while to the white sections in between the blue and the black outline of the channels are printed protein lysate. We can see on the range of straight, separated channels that the water efficiently wicked up the 250um wide channel and the 200um channel. On the largest “flower” template, the water flowed up the 800um and 900um channels. We also see a successful serendipitous negative control of our device once the water flows past the end of one of the 900um channels and starts penetrating the paper printed with blue ink. This provides a nice contrast with the channel flow and illustrates that while the resolution of the channels isn’t extremely sharp, the hydrophobic protein is certainly performing its function. On the smaller scale flower channels, the water efficiently travels down the 100 - 500 um channels, which is indicative of its functionality on the PDMS-scale resolution.
BBa_K3260031 Double Layer and Negative Control Side by Side:
The video above is another printing lysate test. This test print was part negative control (single layer, but underprinted with normal ink) and part double layer printed BBa_K3260031, which is a dCBD fused to a single hydrophobic domain. Below is a labelling of the video with channel width for each component in our testing template.
On this test we have a firmer grasp on the resolution that is achieved with this printing method, and it is shown that some bleeding into the hydrophobic region of the device design occurs. However, this is sharply contrasted with the complete lack of resolution and function in the negative control, that was single layer printed with our hydrophobic protein but printed on the opposite side with plain blue ink, which undercuts the functionality of the protein. We can see that the water effectively travels up the channels of widths 200 um, 250 um, 800 um, and 900 um.
Conclusion
All in all, we were able to rationally design, synthesize, print, and prototype a microfluidic system whose channel architecture is based around fusion proteins and paper reagents, not wax or PDMS. We were able to create functional microfluidic channels down to 100um in width, which is extremely significant because this falls comfortably within the range of high resolution PDMS device channels (which have widths of 30 nm to 500 um [2, 3]). Even though our channels did experience a significant degree of bleeding, the functionality of their wicking ability even down to 100um widths speaks to their success as a proof of concept. In addition to this, we were able to produce these prototypes using low cost commercial printers, ink cartridges, and paper reagents. Not only did we successfully produce these prototypes with our original fusion proteins idea (BBa_K3260031), but we also experimented with more elaborate protein designs, yielding a fusion system that can be applicable to our paper microfluidic system, but also an extremely wide range of applications involving the functionalization of cellulose with a binding domain, split GFP domains, and complementary leucine zippers (illustrated in BBa_K3260036 and K3260034). Coming back to the central goal of the subproject, this proof of concept for a protein-based paper microfluidic system opens the door for a whole new range of low cost, easily producible, high resolution devices.
- Liu, Ning, et al. “Direct Spraying Method for Fabrication of Paper-Based Microfluidic Devices.” Semantic Scholar , Nanyang Technological University , Sept. 2017.
- Abidin, Ummikalsom, et al. “Replication and Leakage Test of Polydimethylsiloxane (PDMS) Microfluidics Channel.” AIP Publishing, AIP Publishing LLC, 25 Jan. 2019, aip.scitation.org/doi/abs/10.1063/1.5086611.
- Strong, E. Brandon, et al. “Fabrication of Miniaturized Paper-Based Microfluidic Devices (MicroPADs).” Nature News, Nature Publishing Group, 9 Jan. 2019, www.nature.com/articles/s41598-018-37029-0.
- Sher , Mazhar, et al. “Paper-Based Analytical Devices for Clinical Diagnosis: Recent Advances in the Fabrication Techniques and Sensing Mechanisms.” HHS Public Access, Molecular Diagnostics , 26 July 2017.
Production: In-Lab Drug Production
For the cell-free expression system, the pET-23b(+) plasmid was successfully linearized (Figure 1) and the overhangs for Gibson assembly were successfully added for the hG-CSF and teriparatide inserts (Figure 2). The Gibson assembly of the hG-CSF insert into the linearized plasmid was verified by sequence analysis. For the B. subtilis cellular system, pBS2E was successfully linearized (Figure 3) and Gibson assembly was verified to be successfully through sequence analysis. For a control, the chromogenic protein Prancer Purple from ATUM Paintbox was successfully linearized for Gibson into pET-23b(+) (Figure 4).
After we got our sequencing results and confirmed that we had constructs for pET23-teriparatide-EK, pET23-teriparatide-TEV, pET16-teriparatide-TEV, and pET16-hGCSF-TEV, we moved on to expressing and purifying them in E. coli and cell-free reactions. E. coli expression and subsequent His-tag purification gave inconclusive results due to large amounts of native protein making the drug bands unclear. The cell-free process is discussed below.
Establishing a Cell-free System
We made all the mixtures required for the cell-free reactions, and extracted lysate from E. coli BL21 (DE3) cells using the protocols here. To test for contamination, we streaked 50uL aliquots of each solution onto LB plates and incubated them at 37° °C overnight. The amino acid mix, salt mix, energy mix, did not have any contamination; an abundance of growth from the lysate aliquot indicated contamination, which was undesirable (Fig. 1). To remove the unlysed cells from our solution, we modified the protocol by adding an extra centrifugation step in smaller volumes, and at higher speeds. The result of our modification lead to decreased in-tact bacteria presence from an uncountable amount of colonies per 50uL to 62 ± 5 colonies per 50uL Figure 2.
With our mixtures and solutions in place, we ran preliminary cell-free experiment using an Red Fluorescent Protein-encoding plasmid to test if our platform worked at all. After leaving the cell-free reaction to incubate overnight, we qualitatively observed production of RFP (Fig. 3). At this point, we did not collect quantitative fluorescence values. The red arrows pointing at experimental groups that produced a small amount of fluorescence and the blue arrows pointing to the negative control (no plasmid inserted). In Figure 4 we see our positive control, where we expressed the RFP in E. coli. The cellular-expression positive control is significantly brighter than the cell-free experimental group. This may be because of the greater concentration of RFP present in the positive control. From this data, we knew that RFP production was observable.
Given the initial encouraging results we set to establish our first set of cell-free experiments to see if we could enhance the reaction by manipulating concentrations of T7 Polymerase and Murine Inhibitor on cell-free, and see if anything significantly helps protein yield, or if anything is not as essential, so we can avoid lyophilizing it. The reactions were run for 5hrs. Afterwards, the fluorescence of the samples was measured using the spectrophotometer. Figure 5 shows qualitative results of the experiment.
There is a clear indication that the experiments did not work. The fluorescent data for the experimental groups was well below the controls (data not shown), indicating there was no fluorescence, hence no protein product at all. Under UV light, there was also no observed red color that was evident in the initial experiment.
We began our lengthy troubleshooting process. There were many points to begin from: an ineffective lysate batch, DNA plasmid quality and concentration, environmental contamination by RNases, low protein activity, decomposition/precipitation of cell-free reaction mixes, degradation of the solutions at room temperature while preparing the reactions.
First we tested if changes in T7 Polymerase quantities would influence the reaction, so we doubled the concentration from 10U to 20U, but observed no RFP production, and because we observed no fluorescence under UV light, we did not attempt to quantify our results using the spectrophotometer.
Then we suspected there might be problems with the DNA, so we began by altering the DNA concentration by testing adding 100ng of plasmid to the reaction instead of the previous 30ng, and testing one of the paper-microfluidic protein constructs that contained a green fluorescent protein (GFP). Both experiments did not yield any observable red or green fluorescence.
New cell lysates were extracted and cell-free mixes were remade in case of any chemical degradation or contamination that may have occurred to render the solutions dysfunctional, and reactions were run again. Again, there was no obvious production of RFP.
Another problem with the DNA was contamination. We only received a very small quantity of RFP plasmid from iGEM, so we had miniprepped it. Miniprep DNA tends to contain a lot of nucleases, and these nucleases, RNases in particular, inhibit cell-free reactions. To diminish RNase activity, we employed Murine RNase inhibitor, and tried to keep our lab environment sterile and RNase free. The experiments we ran that manipulated RNase quantity also yielded no production of RFP. However, it is possible we underestimated the degree of contamination and did not add enough RNase inhibitor. Our SDS-PAGE highlights the samples that contained RNase Murine inhibitor in blue versus the samples that did not contain RNase inhibitor in red. There is no observed production of any protein besides those latent in the solution. At this point, we had also received a commercial cell-free kit, and began to test reactions on it. One of the commercial cell-free reactions we ran is R1 (Fig. 6) - no production of RFP was observed either despite the edition of Murine RNase inhibitor.
Realizing that we were underestimating the amount of Murine RNase inhibitor we needed for the reaction, we increased the quantity in a commercial cell-free trial from 20U of Murine to 100U of Murine, and observed a band in the 27-30kDa range, which corresponded to the size of RFP (Fig. 7). This is also reflected in the positive control sample (expression of RFP in E. coli).
While this result was promising, we could not replicate it consistently, so we had to continue reiterating through the trouble-shooting process. We tested different control plasmids and our peptide-drug plasmids too. During one reaction, we were able to successfully produce teriparatide in a commercial cell-free system (Fig. 8).
In tandem, we also tried to establish a yeast cell-free system that could perform the post translational modifications required for the insulin. However, we came across the same problems as the E. coli system, and weren’t able to observe anything qualitatively or quantitatively.
Despite promising initial results with our lab-characterized cell-free expression system, we were unable to get it to preform consistently most likely due to excess RNase contamination, and degradation of unstable compounds in the mixes. Likewise, we were unable to get the commercial cell-free system to preform consistently for the same reasons. If our experimentation process had gone more smoothly, we would have scaled down our cell-free experiments to fit inside the microfluidic expression chip, and manipulated concentrations of T7 Polymerase, RNase inhibitor, DNA concentration, and incubation time to enhance peptide-drug production to our desired concentrations.
Expression and Purification Microfluidics
For the drug purification system, a full set of chips to produce and purify medicine were designed, fabricated and tested. This included a 3D printed cell free expression chip, a column for Immobilized Metal Affinity Chromatography (IMAC), and a column for Size Exclusion Chromatography (SEC). All of our chips were fully fluid functional.
In order to tackle existing constraints on the minimum feature capabilities of our 3D printer, we used several different approaches to manufacture our expression chip, falling into two main catagories. The first group of iterations was a design that required us to print two halves of our expression chip, and use photo-activated resin to seal them together. After these failed due to leaking and misalignment, we iterated again and printed the chip as a single piece. With some minor adjustments to our design and protocol we were successful: its interfaces were water-tight and we were able to successfully facilitate protein-expression by flowing through the expression chip.
These chips were then flow tested to see if there were any significant issues with bubbling, leakage or delamination of the substrate. We found we had manufactured several fluid functional chips for each of our 3 designs. Videos of our chips being flow tested are shown below.
While some difficulties were encountered loading chromatography resin in our initial prototypes, we were able to confirm the concept behind the purification modules we created by running a large scale size exclusion chromatography to separate yellow food coloring from an orange chromoprotein.
After testing SEC purification on a large scale, we returned to our microfluidic designs and created a new prototype with an enlarged media loading port and multiple layers of smaller bead blockers. With this new design we were able to load SEC media into our prototype.
The main difficulties encountered with our second set of prototypes were that the IMAC channel height of 100um was too low for the beads in the media, which caused them to not be able to enter the chip. While the SEC could be loaded, its input and output channels were small enough that some beads that passed through our bead blockers could still clog channels. Future prototypes for this will need to have both taller channels and wider inlet and outlet channels
Our results with microfluidics showed very promising results. Overall we were able to mill and test 2 fluid functional PDMS microfluidics, and 1 3D printed chip. All of our chips were fully fluid functional, and some significant strides were made in adding chromatography media to our chips. Our hardware reached a similar level of development to many other excellent iGEM projects which focused on microfluidics. BostonU 2017 created a repository of microfluidic devices, which were milled, tested and fluid functional, but not tested on biologics. Our team has fluid tested our chips to the same extent, although we also encountered difficulties incorporating biologics on our chips. UCopenhagen 2018 also sought to create a drug purification system that would work in space, and was able to 3D print several versions of their hardware. We were also able to 3D print hardware for drug production, although both of our teams encountered difficulties incorporating hardware into a fully realized drug production and purification system. Overall our team made significant strides in developing microfluidic hardware that could someday represent a major breakthrough in the way drugs are produced, both in space and on Earth.
Lyophilization
We preformed three lyophilization experiments. In the first experiment, we freeze-dried B. subtilis in growth media on plastic, E. coli in growth media on plastic surface, E. coli in growth media on a filter paper surface, and E. coli in growth media on printer paper. After overnight lyophilization, we rehydrated the cells at different time points (t=0d, 1d, 2d, 8d, 10d, 15d, 30d) and preformed serial dilutions to calculate cell viability. Some of the data is omitted due to contamination of growth plates and procedural errors.
Our results show that B. subtilis has consistent growth even at the one month time point. This is expected because B. subtilis are hardy spore-forming bacteria known to have survive for up to six years in space [1]. However, E. coli viability dramatically decreases one day after lyophilization on paper and plastic substrates when no lycoprotectant is present (fig. 1.). Hence, we turned began testing lycoprotectants to see if them can increase cell viability when freeze-dried.
Next, we tested different concentrations of trehalose as a lycoprotectant for E. coli. There were four experimental groups: 0% Trehalose, 1% Trehalose, 5% Trehalose, and 10% Trehalose. The samples were lyophilized overnight and rehydrated at time points t = 0d, 1d, 7d, 15d, after which serial dilution experiments were preformed to calculate cell viability. We observed that E. coli in the 0% Trehalose condition exhibited the best cell viability overtime, which is contrary to the previous results and to the literature. However, on rehydration at t = 0d, the 10% Trehlose condition had the greatest CFU count compared to all the other groups (fig. 2), which suggest that Trehalose had the best inital viability, but procedural or lyophilzation errors may have results in its quick deterioration.
In our third lyophilization experiment, we tested the preservation ability of skim milk, sucrose, saline, and growth media on VmaxTM. Not much growth was observed at the two rehydration time points (t = 0d,1d), except for a few colonies under the sucrose condition at t = 0d (Fig. 3). These results were also unexpected, but were attributed to errors in the experimental procedures; during this time, we were encountering problems with VmaxTM growth from the manufacturers end, so it is possible that the cells themselves were not viable in the first place.
It is interesting that the cell viability for many of the experimental groups decreases a few days after lyophilization. While this dramatic decrease could be attributed to keeping our freeze-dried samples at room temperature, which encourages degradation, it is possible that there were errors in our freeze-drying procedure that inhibited true exploration of lyophilization capabilities. We left our experiments lyophilize overnight (around 10-14hrs), whereas other procedures lyophilize for 24 to 48hrs; there is the possibility that not all the water was removed from our samples. Additionally, not all freeze-drying experiments simply let the samples sit in the lyophilizer – there are more intricate cycles of freezing and drying, and seperationg of unstable compounds that we were unable to conduct due to laboratory and time constraints [4].
While we did attempt to freeze-dry cell-free solutions, none of the results are reliable because the solutions prior to lyophilization may not have been functional given the difficulties we had in trying to develop our cell-free system. If our cell-free system was operating smoothly, we would
Overall, our lyophilization confirm that B. subtilis would be the most storable chassis for cellular expression of peptide drugs for the Astropharmacy because their cells were consistently viable over an extended period of time at room temperature. However, we were unsuccessful in replicating long-term viability in E. Coli and VmaxTM despite our use of lycoprotectants suggested by literature. This may be because our experiments were rather sporadic and not as systematic as they could have been. The experiments should have focused on a single species surveying various conditions to establish a more defined protocol that could then be applied and modified to different species. Future experiments would have investigated other modes of drying (like desiccation), and further tested lyophilization on paper to enhance the manufacturing and transportability of a lyophilized Astropharmacy on a paper-based system.
- Microbes and Molecules Get a Space-Stress Test | Science Mission Directorate. https://science.nasa.gov/science-news/science-at-nasa/2010/22oct_ooreos. Accessed 20 Oct. 2019.
- ATCC - Methods for Freezing and Freeze-Drying Bacteria-140. https://www.atcc.org/support/faqs/840a5/Methods+for+freezing+and+freeze-drying+bacteria-140.aspx. Accessed 16 Oct. 2019.
- Zhang, Miao, et al. “Freeze-Drying of Mammalian Cells Using Trehalose: Preservation of DNA Integrity.” Scientific Reports, vol. 7, no. 1, July 2017, pp. 1–10. www-nature-com.revproxy.brown.edu, doi:10.1038/s41598-017-06542-z.
- Smith, Mark Thomas, and Scott D. Berkheimer. “Lyophilized Escherichia Coli -Based Cell-Free Systems for Robust, High-Density, Long-Term Storage.” BioTechniques, vol. 56, no. 4, Apr. 2014. DOI.org (Crossref), doi:10.2144/000114158.
- Pardee, Keith, Alexander A. Green, et al. “Paper-Based Synthetic Gene Networks.” Cell, vol. 159, no. 4, Nov. 2014, pp. 940–54. DOI.org (Crossref), doi:10.1016/j.cell.2014.10.004.
- Labconco FreeZone 2.5L -84C Benchtop Freeze Dryers, 115V US Models Includes. p. 2.
- Leslie, Samuel B., et al. “Trehalose and Sucrose Protect Both Membranes and Proteins in Intact Bacteria during Drying.” APPL. ENVIRON. MICROBIOL., vol. 61, 1995, p. 6.
- Lopatkin, Allison J., and Lingchong You. “Synthetic Biology Looks Good on Paper.” Cell, vol. 159, no. 4, Nov. 2014, pp. 718–20. DOI.org (Crossref), doi:10.1016/j.cell.2014.10.003.
- Lu, Yuan. “Cell-Free Synthetic Biology: Engineering in an Open World.” Synthetic and Systems Biotechnology, vol. 2, no. 1, Mar. 2017, pp. 23–27. DOI.org (Crossref), doi:10.1016/j.synbio.2017.02.003.
- Ogonah, Olotu W., et al. “Cell Free Protein Synthesis: A Viable Option for Stratified Medicines Manufacturing?” Current Opinion in Chemical Engineering, vol. 18, Nov. 2017, pp. 77–83. DOI.org (Crossref), doi:10.1016/j.coche.2017.10.003.
- Pardee, Keith. “Perspective: Solidifying the Impact of Cell-Free Synthetic Biology through Lyophilization.” Biochemical Engineering Journal, vol. 138, Oct. 2018, pp. 91–97. DOI.org (Crossref), doi:10.1016/j.bej.2018.07.008.
- Pardee, Keith, Shimyn Slomovic, et al. “Portable, On-Demand Biomolecular Manufacturing.” Cell, vol. 167, no. 1, Sept. 2016, pp. 248-259.e12. DOI.org (Crossref), doi:10.1016/j.cell.2016.09.013.