In bacteria, there are three well-established start codons: AUG, GUG and UUG. In very rare cases, AUA, AUC, AUU and CUG, are also start codons. But among the remaining 57 codons, none are typically considered to be possible start codons, and no one had previously systematically measured the capacity of all 64 codons to initiate translation.
This project, a collaboration between NIST/JIMB and the Endy Lab at Stanford, started with a simple question: what if any codon could be a start codon? Given that the seven known bacterial start codons all initiate with methionine as the N-terminal amino acid, the wide-range of other observed non-canonical translation behaviors, and the large number of molecular interactions involved in initiating translation, it seemed mechanistically possible that translation could initiate from additional codons.
We explored this question by cloning all 63 other codons in place of AUG as the start codons for a GFP gene on a pET20b(+) plasmid. We also cloned a subset of 12 codons onto 3 different vectors, one GFP and two nanoluciferase, at different copy numbers, to explore non-canonical translation initiation under a range of different expression conditions.
We observed statistically significant translation initiation of GFP from 47 of the 64 codons.
To confirm the codon from which translation was initiating, we cloned a 6x-His tag onto the C-terminus of the GFP gene of four of the different start codons: AUC, ACG, GGA and CAU. We purified the protein using the 6x-His tag, and confirmed via MS/MS that translation was initating from our intended start codon, and that in all but one spectrum, the N-terminal amino acid was methionine.
We observed statistically significant translation initiation of nanoluciferase from all 12 of the subset of 12 codons, including UAG, one of the three stop codons, on both low-copy and very-low-copy plasmids. We did not observe significant translation initiation from any non-canonical start codons on the p15A plasmid, because bacterial autofluorescence overwhelmed any fluorescence we might have observed from GFP. Incidientally, this highlights the challenges of measuring small signals in vivo.
We've shown that it is possible for translation to initiate from almost any codon in E. coli. However, our results are only for an engineered, heterologous gene. The bigger question is if, and how frequently, non-canonical translation initiation occurs in wild-type organisms. Given the large number of known wild-type non-canonical translation events, we think it is possible that wild-type non-canonical translation initiation may occur, and that it may be advantageous. We hope that the results that we've shown here may inspire others, especially those with expertise in de novo proteomics, to identify the occurence of wild-type translation initiation from non-canonical start codons.
A common method for estimating the abundance of bacteria in a liquid culture is to measure the optical density (OD) of the culturea at a wavelength of 600 nm (OD600). OD is a measurement of the attenuation of incident light, and any material in the cell that absorbs incident light will contribute to its attenuation. Red fluorescent proteins (RFPs) have an absorbption peak near 600 nm.
The attentuation of incident light by the RFPs will bias estimates of cell abundance made from measurements of OD600. We examined the magnitude of this bias in cultures of RFP-expressing E. coli from the BIOFAB collection, and determined that the magnitude of the bias in OD600 measurements is up to 7%, linearly proportional to the amount of expressed RFP.
Based on these findings, we recommend that when estimating the per-cell fluorescence in a liquid culture, OD should be measured at a wavelength that is not absorbed by any expressed fluorescent protein. Adopting the general practice of measuring OD at 700 nm, in place of the customary 600 nm, would allow unbiased estimates of cell abundance from measurements of OD of cells expressing nearly all visible-spectrum fluorescent proteins.
More information can be found in our publication in ACS Synthetic Biology.
In addition to my current work, here are some of the other projects I have worked on, during grad school and college:
The goal of LAM is to measure the concentration of a protein in solution based on the rotational period of a cluster of magnetic beads. The magnetic beads are coated with affinity molecules against a specific protein. When mixed with the protein, they form a cluster, whose shape depends on the concentration of the protein, and the rotational period of the cluster depends on its shape.
Above is a schematic overview of LAM. In this particular example, the protein being used is thrombin, which we have chosen as a model protein for demonstrating proof-of-principle of this system. The affinity molecules being used are aptamers, specifically the 15-mer and the 29-mer (so called because of their length) anti-thrombin aptamers. The 15-mer and 29-mer aptamers bind to opposite ends of the thrombin molecule, which allows the aptamers to sandwich the molecule. We chose this thrombin-aptamer model because it has been very popular for use in demonstrating proof-of-principle for novel protein deteciton systems.
Half the beads are coated with the 15-mer aptamer, and the other half are coated with the 29-mer aptamer. In this first step, only the 29-mer-coated beads are used. An aliquot of the beads are mixed with a solution containing the protein of interest. Any protein present in the solution will bind to the aptamer located on the surface of the bead.
Now the other beads are added into the solution. At this point, there is minimal free protein floating in solution, so the 15-mer-coated beads will only bind to protein thats bound to the 29-mer aptamer. This starts the process of binding the beads together to form a clump.
The first two steps took place in a microcentrifuge tube. Now, a 1 μL droplet of the bead solution is transferred to a glass slide, which is inverted, forming an inverted droplet (see the next schematic below). A strong permanent magnet is held underneath the inverted droplet, pulling the beads down to the tip of the droplet, which serves to concentrate them in a single point. As the beads fall down, they encounter each other, and free aptamers on the surface of a bead have the opportunity to bind to protein bound to the opposite aptamer on the surface of a bead in a sandwich-type binding event. Once the beads have reached the bottom of the droplet, all the bead-bead binding and cluster formation has been completed.
The shape of the bead cluster depends on the concentration of protein. This is the key principle behind this method. In the presence of a high concentration of protein, as the beads are falling and bumping into each other, they have a high likelihood of binding to each other. This results in the formaton of many, small, branch-like structures. As these branches encounter each other, they bind together, with the resulting bead cluster having a low-density, snowflake-like shape (see images below). As the concentration of protien decreases, the branches that are formed are smaller, and so the density of the cluster increases, and its overall diameter decreases. In the case of no protein being present in solution, the beads will not bind to each other, and the resulting cluster will have a high-density, hexagonally-packed shape (this hexagonal shape isn't really apparent in the cartoon above, but is clearer in the images below). The key principle behind this method is that the shape of the cluster that forms strongly correlates with the concentration of protein present in the solution.
For the final step in the reaction process, the inverted droplet with the bead cluster is placed in a rotating magnetic field. The field is built from two pairs of orthogonally-oriened Helmholtz coils, with one of the pairs driven by a sine wave, and the other pair driven by a cosine wave. The field is generated in LabVIEW, output via a DAQ Board, amplified by a car-stereo amplifier, and then finally fed into the coils to generate the rotating field. The rotational period of the bead cluster is a function of the concentration of protein present in solution. This period can be measured in one of two ways: by recording videos of the beads rotating in the droplet with a microscope, or by shining a laser on the beads as they rotate, which creates a periodic diffraction pattern, which can then be captured by a photodiode, and analyzed in real-time by a LabVIEW program, providing a real-time measurement of the concentration of protein in solution.
Above is a schematic of the inverted droplets in which the key steps of the reaction take place. The droplet has a volume of 1 μL, and is formed by placing the droplet in a circular region surround by Teflon on a glass slide. The Teflon keeps the droplet from spreading, and ensures uniform droplet size.
Steps 1 and 2 from the first part above, the mixing of the beads with the protein, occur in a microcentrifuge tube. This allows for better mixing than would occur if the beads and protein were placed straight into the droplet.
From the bead solution in the microcentrifuge tube, a 1 μL droplet is taken and placed on the Teflon-coated glass slide. The slide has un-coated circular regions where the droplet is placed. The surrounding Teflon keeps the drop from spreading and helps it keep its shape and ensures the uniformity of droplets.
The beads have a density of around 2 g/mL, and so would naturally fall to the tip of the droplet under the forces of gravity if left alone; unfortunately, this process takes over 90 minutes. To speed up the falling of the beads, a permanent magnet is used. This magnet is placed underneath the droplet, and accelerates the rate of falling the beads, while still allowing them to bind to each other as they fall; this process takes 2 minutes.
Finally, the beads are placed in a rotating magnetic field, which is composed of two orthogonally-oriented pairs of Helmholtz coils, as shown in the schematic at below. The field is powered by a LabVIEW program run through the computer. The laser-diode detection setup, also discussed earlier, is also shown in the schematic below. The image shows the laser, the focusing lenses, the magnetic field build from the two Helmholtz coils, and the photodiode, all of which are connected to a computer offscreen.
Below are a series of images showing the dependence of the shape of the bead cluster on the concentration of the protein, in this case thrombin. Shown below are three sets of images, each one with a different concentration of beads used for making the cluster (220 ug/mL, 70 ug/mL, and 22 ug/mL). The concentration of thrombin used in each image is shown to the left of the image. Notice that the as the concentration of beads used in the droplet decreases, the concentration of protein needed to affect the shape of the cluster also decreases. The control clusters have no thrombin. The clusters are shown at two different magnifications.
These images were taken with a 5 MP camera connected to a brightfield microscope, with 20x and 40x water-immersion objectives. The scale bars are 20 microns.
At the low protein concentrations, the beads form a densely-packed cluster. These clusters show hexagonal packing, which is the maximally efficient packing arrangement for spheres. This hexagonal packing is shown bekiw in the zoomed-in portion of the 220 ug/mL control image shown above. The scale bar above is 5 microns.
One of the primary advantages of LAM is that it can be performed without the use of a microscope, using a simple laser and photodiode setup. This setup consists of a single low-power laser (20 mW, 650 nm), focused on a droplet containing the bead clusters. The droplet also serves as a lens, focusing the beam onto the cluster of beads at the bottom of the droplet. Shining the beam on the bead clusters produces a projection of the bead clusters, which is captured by an array of photodiodes located underneath the beads. When the beads are rotating in the magnetic field, their projection produces a periodic pattern that the diodes pick up, and through a LabVIEW program that I have written, perform a Fourier transform on the raw signal to calculate the rotational period of the beads. A schematic and a photograph of this setup are shown below.
Below is a screenshot from the LabVIEW program that converts the raw input from the 9-diode array into a rotational period for the cluster.
One of the advantages of LAM is that the dynamic range and limit of deteciton (LOD) can be tuned based on the concentration of beads used in the droplet. The LOD and dynamic range occur when a certain percentage of available binding sites are occupied (roughly 0.1% for the LOD, and 0.5-10% for the dynamic range). Fewer beads used in the droplet means that the LOD can be pushed further down. Below is a series of dose-response curves at different bead concentrations. They have been normalized to facilitate comparisions between the different bead concentrations. Each point on the graph represents the average of 10 measurements, with the error bars representing the standard deviations.
The LOD for each bead concentration can be calculated based on the mean + 3*SD of the control.
The LOD of 80 fM shown here is one of the lowest LODs ever reported for a thrombin-based biosensor. We chose to use thrombin as our analyte, because it has been a very popular target for showing proof-of-principle work for new biosensing methods. Considering the relative simplicity and portability of this method compared to other sub-picomolar sensing techniques, we believe that LAM has the potential to be attractive for further devevlopment towards clinical applications.
The above data was collected using the microscope and image tracking software. However, below is data collected using the laser and photodiode setup, validating its use for performing LAM.
Finally, the images of the bead clusters can be analyzed via Image Analysis software (FracLac ImageJ plug-in) to detemrine their fractal dimension and lacunarity. This could provide an alternative way of quantifying the effect of protein concentration on the bead clusters, and for measuring the protein concentration.
More information about this work can be found in our publication in Biosensors and Bioelectronics.
LAM initially started with a slightly different approach. Instead of forming clusters with magnetic beads all of the same size, LAM initially started by adding magnetic beads to larger, non-magnetic spheres, with the number of beads added to the sphere a function of protein concentration. As more beads were added to the sphere, the sphere would rotate at a higher frequency. The concept is illustrated below.
Below is some data showing results generated using this approach:
And some images:
While we thought this method was pretty neat, unfortunately, this method presented us with several challenges. The primary issue was that the spheres and beads are quite heavy relative to the force of the aptamer-protein bond that was holding them together. This made the resulting complex very fragile, and it had to be handled very delicately to avoid rupturing the bonds are destroying the structure.
After we had developed this version, we started tinkering around with different ideas on how to improve on LAM, and we eventually ended up with the LAM method discussed on the first part of this page (which we'll call Version 2.0). We ended up discovering that Version 2.0 had several advantages over Version 1.0:
Version 2.0 had a much lower limit of detection. Version 2.0 had a LOD about 3 orders of magnitude lower than Version 1.0. The aptamer-thrombin bond is relatively weak compared to the force of gravity on the beads. Allowing the beads to mix in a droplet and then fall to the bottom of the droplet, where the air-water interface provides a cushioned landing, is a relatively gentle mixing procedure. Mixing the beads in a centrifuge tube, and then transferring them by pipette into a fluidic cell to be imaged, is a much more strenuous mixing process, which likely ruptured many of the sphere-aptamer-thombin-aptamer-bead bonds. Therefore, a much higher number of thrombin molecules were needed to secure the bond to indicate attachment.
Version 2.0 has narrower error bars. The errors bars in the graph for Version 1.0 were much bigger than those for Version 2.0, in part because bonds are more likely to rupture, as mentioned in the previous item, but also because there are fewer beads involved in the binding process for Version 1.0. A sphere in Version 1.0 can hold a maximum of around 100 beads, and typically holds much fewer beads, especially at the lower concentrations, where only a few beads are needed to indicate a binding event. Version 2.0 has hundreds of beads in each bead cluster, regardless of the thrombin concentration. By increasing the number of beads used to transduce a signal, the effects of small fluctations are reduced.
Version 2.0 is better suited toward automation. In Version 2.0, the beads all collect at the bottom of a droplet, whose location is predetermined by the Teflon patterned onto the slide. Therefore, the user will always know where the beads will be located on the chip, simplifying the process for automating the detection process. In Version 1.0, the spheres and beads are injected into a fluidic cell, but the exact location of the beads is not confined to a particular location, requiring the cell to be scanned to find the spheres and beads. Attempts to confine them to specific locations on chip via microfluidic traps were unsuccessful because the confinement significantly impeded rotation.
My initial exposure to aptamers came through this project, which I conducted at Sandia National Labs as an intern during my first summer in grad school. The project was a standard capillary electrophoresis project, where the fluoresecently labeled aptamers are mixed with the target-containing solution. Aptamer-target complexes have a higher mass-to-charge ratio than free aptamers, and so when the mixed sample is introduced into the electric field, the aptamer-target complexes will elute at a different rate than free aptamers, which can be detected by exciting the fluorescently-tagged aptamers.
This project demonstrated that aptamers can be advantageous over antibodies for capillary electrophoresis applications, because aptamers typically have significantly smaller moleclar weights than antibodies, and when targeting molecules that have molecular weights smaller than antibodies, the ratio of complexed to free affinity molecules is higher, providing better data resolution and a more accurate diagnostic system. This project also demonstrated that when working in serum, undesirable aptamer-serum non-specific interactions can be mititgated by the addition of a 20x excess of an competitive arbitrary aptamer. Finally, we demonstrated that the use of a preconcentration membrane helps improve the sensitivity of the system. Below is a sample of data from our publication in Analytical Chemistry, which can be found here.
My very first research experience began in the summer after I graduated high school. I was accepted into a summer research program at the Los Angeles Biomedical Research Institute at Harbor-UCLA Medical Center, and I was assigned to the Pulmonmary Rehabilitation and Clinical Trials Center run by Dr. Richard Casaburi. It was a phenomenal experience, where I learned how much fun it could be to conduct research. I continued to work there during summers and occassionally during the year throughout college.
I participated in a research study examining the effects of different ambulatory oxygen therapies on patients' abilities to maintain an active lifestyle, which improves their health. One of the assessment tools in the study was an accelerometer, which the patients wore at all times. My job was to design an algorithm that would analyze the data from the accelerometer, remove invalid data that was generated from not wearing the device or from riding in a car, and produce an activity profile for the patient. Activity profiles from different patients were then compared to determine the effectiveness of the different therapies. Below I've included a sample activity profile from one of the patients, showing their activity over the course of a typical day. More information about the project can be found here.