Combinatorial Screening For Protein Crystallization
Advances in Instrumentation for Automated Combinatorial Crystallization Screening
Brent W. Segelke
Timothy Lekin
Dominique Toppani
Lawrence Livermore National Laboratory
Mary Cornett
Jim Johnson
Dave Martin
Innovadyne Technologies Corporation
(This work was performed under the auspices of the U.S. Department of Energy by University of California, Lawrence Livermore National Laboratory under Contract W-7405-Eng-48.)
Abstract
By considering crystal screening as a sampling problem, we have previously demonstrated the inherent efficiency of stochastic combinatorial screening for crystallization. Though stochastic combinatorial screening is efficient, it has been challenging to automate. Robotic liquid handling instruments are generally designed for high throughput mother-daughter transfers or for lower throughput, though versatile, re-arraying.
A newly developed, non-contact liquid handling instrument equipped with 96+8 independently actuated tips (Screenmaker 96+8™) holds the promise of delivering the speed and versatility required to make high-throughput, on the fly, stochastic combinatorial screening practical. The independent actuation enables "any source any destination" liquid handling. The instrument also maintains high precision over a broad range of volumes and viscosities, delivering the necessary versatility for the types, concentrations, and ratios of components used in custom combinatorial screens.
We are currently working to generate performance parameters for the full set of reagents we use and to integrate our CRYSTOOL© design engine with the instrument, to pass instructions and performance parameters for aspirate/dispense operations to the instrument at run-time. This will fully enable stochastic combinatorial screening. Stock reagents arrayed in 96-well deep well blocks are aspirated simultaneously and dispensed in the random order and volume prescribed by worklists generated by the design engine. The same instrument can also be used for rapid setup of crystallization experiments from pre-made screens or for the setup of grid optimization screens.
Key Features for Automated Random Screen Preparation
Aspirate and dispense can be from any source to any destination on a large deck space
The instrument is capable of working with a large range of volumes
The instrument can receive nearly all liquid handling parameters and instructions at runtime
The tips are independently actuated
It takes approximately 1 hour per 96-well screen setup compared to ~10 hours for setup by hand. Despite this significant gain compared to manual setup, screen preparation is our current rate limiting step for total throughput. The MultiProbeII has several other limitations as well.
Limitations of the MultiPROBE®II
The low volume precision determines the minimum screen volume, currently 1mL, which leads to reagent waste and/or screen reuse.
To achieve low volume precision a wet dispense is required, which in turn requires extensive washing and slows the process.
The positional accuracy and low volume precision are inadequate for most 96 well crystallization labware necessitating that a second instrument be used to set up crystallization plates.
Because of the MutliPROBE®II limitations we use a Matrix Technologies Hydra® II Plus-One to setup crystallization plates. The Hydra® II Plus-One transfers our random screen reagent cocktails and protein stock to 96-well sitting drop labware. The Hydra® II Plus-One has two independent dispense heads, one a positive displacement 96-tip head ideally suited to 96-well mother-daughter transfers, and a single non-contact nano-dispenser. The 96-tip head transfers premixed cocktails to the sitting drop reservoir, then the nano dispenser transfers protein to the drop, and finally, the 96-tip head transfers reservoir solution to the drop as well. The 96-tip head is capable of high precision down to <200nL but a wet dispense is required. The Hydra® II Plus-One can set up a sitting drop plate in ~2 minutes but requires nearly 5 minutes of washing between plates, limiting the throughput to 10-12 plates/hour. The Hydra® II Plus-One has a few serious limitations:
Limitations of the Hydra® II Plus-One
Low volume dispense limited to ~500nL due to wet dispense requirement
Single volume for each 96-well mother-daughter transfer
Screenmaker 96+8™
Innovadyne Technologies Inc. has introduced the Screenmaker 96+8™ which has two dispense heads, one with 96 non-contact nano-dispensers and the other with 8 non-contact nano-dispensers.
| Screenmaker 96+8™ |
|
The combination of the capabilities of the core nano-dispense technology and the configuration of the Screenmaker 96+8™ make for a very versatile instrument that promises to overcome the limitation of other instruments like the MultiPROBE®II and the Hydra® II Plus-One to enable random combinatorial screening. The Screenmaker 96+8™ is capable of any destination dispensing over a wide range of volumes and each dispense can be controlled independently.
| 96-Tip and 8-Tip Heads |
|
Screenmaker 96+8™ Key Advantages
Faster:
Plate setup in <2min
Protein dispense <45 seconds using a single tip for protein dispense
As fast as 3 seconds with 8 tips used for protein dispense
Back-filled system liquid (deionized distilled water) eliminates 96 reagent aspirations and enables fast washing
No wet dispense required
Sub-microliter volumes achievable
Random sampling of drop size and drop mixing ratios possible
Core Technology
The core technology of the Screenmaker 96+8™ is the micro-solenoid nano-dispenser. A micro-solenoid dispenser is comprised of a tip, a syringe for aspirating sample, a pressure reservoir and several valves.
| System Schematic - Aspirate |
|
To aspirate sample, the tip is lowered into the sample and valves are switched so the tip and syringe are in line. The syringe draws a column of water which is separated from sample by an air gap. Sample liquid does not travel past the valve.
| System Schematic - Dispense |
|
To dispense, valves are switched so the tip is in line with the pressure reservoir. The amount of time the micro-solenoid valve is open determines the volume dispensed. Because the sample is dispensed by pressure, the final dispense volume is impacted by sample viscosity. Machine parameters (or performance parameters) need to be tabulated to optimize precision for each "liquid class" over a range of volumes.
| 100 nL Dispense at 1000 Frames/Second |
|
| Enlarge - Video |
Screenmaker 96+8™ Adjustable Parameters for Performance Optimization
Air gap
Aspirate and dispense rate
Predispense (rate, volume, repeats)
Post-aspirate and post-dispense delay
Pulse width
Pressure
Calibration
To relate machine parameter settings to volumes dispensed, Screenmaker 96+8™ is calibrated by simple fluorescence measurements on a fluorescence plate reader. We use a Tecan GENios™ which uses a UV lamp and excitation and emission filters that are well matched to the excitation and emission spectra of fluorescein.
| TECAN Plate Reader |
|
To calibrate the Screenmaker 96+8™, a concentrated fluorescein solution is used. The plate reader has more than sufficient sensitivity to quantify well below 25 µg/mL of fluorescein. This is equivelent to 50nL of maximum concentration fluorescein (50mg/mL in 0.5N NaOH) dispensed into 100 µL of solution.
| Fluorescein Calibration in Wells |
|
The plate reader itself is calibrated by reading from a MATECH™ fluorescence calibration plate. Prior to calibration, our plate reader has a linear response across a range of fluorescence that would be equivalent to 50nL-25µL dispensed with std 2-5%. After calibration, the plate reader has std <1% over this range.
| Calibration with Fluorescent Dye |
|
With the plate reader itself calibrated, the Screenmaker 96+8™ is used to dispense various reagents doped with fluorescein using a range of machine (or performance) parameters. Fine corrections across the range of tips is made in a pulse width correction factor table (see left). Optimum performance parameters for each liquid class for each range of volumes is determined and a pulse width per volume-dispensed table is calculated.
| Calibration Table Data Entry Screen |
|
Validation
A significant portion of the work we've done with Screemaker to date has been focused on optimizing performance parameters for various liquid classes and validating precision dispensing across a range of liquid classes and volumes. The validation experiments are done much the same way as the calibration experiments. We have validated performance parameters over a large range of volumes and viscosities.
| Liquid Class Optimization Curves |
|
|
To date we've been able to obtain machine settings for a range of liquid classes such that approximately 85% of the total crystallization parameter space reachable with the MultiPROBE®II is reachable with the Screenmaker 96+8™. The Screenmaker 96+8™ also enables us to explore lower drop volumes than our current systems.
| Reachable Parameter Space Overlap |
|
Current Status and Effort
Our current effort is focused on developing a complex liquid class and performance parameter library and interfacing our crystallization experiment design engine, CRYSTOOL©, with the Screenmaker 96+8™. For simple, single component reagent transfers, parameter optimization can be done very quickly. The CRYSTOOL© design engine generates complex recipes for >300 million reagent combinations. Even binary cocktails (for example, dilutions of PEG with water) have wide ranging behaviors (see chart below). Performance parameters obviously cannot be individually determined for each combination, so CRYSTOOL© will be modified such that it generates liquid class predictions.
| Binary Cocktail Viscosity Relationships |
|
The Screenmaker 96+8™ software interface is being developed such that a liquid class for each liquid transfer can be assigned at run time. A lookup table will be used to assign performance parameters on the fly.
Below is an example of what may be accomplished with on the fly liquid class corrections. We generated a set of random combinations of reagents that represent a wide range of liquid classes (viscosities). Then, with a simple linear combination model for liquid class we calculated a correction factor and changed the pulse width for each tip. As you can see below, there is a large variance for dispensing 200 nL of random cocktails but the variance is damped significantly after correction (from ~26% std to ~9% std).
| Dispensed Volumes for a Wide Range of Liquids |
|
Appendix: Why Random Screening?
If random screening is such an ordeal to automate, why bother? There are at least three popular sampling strategies used in crystal screening: random screen (Shieh et. al. 1995, Carter et. al. 1979, Jancarick et. al. 1991, Cudney et. al. 1994); Footprint screen (Stura, E.A et. al. 1992) and Grid Screen (McPherson, A. 1982).
If A (figure below) represents the sum total of all possible crystallization experiments (our sample space), the green arrays represent grid screens, the red arrays footprint screens, and the yellow dots random screens.
| Crystallization Experiment Space |
|
Which one is more efficient? Which one should you use? It can be proven by rigorous derivation (not shown) that random sampling is most efficient. Quantification of the efficiency gain from random sampling requires empirical determination from experiment.
From real experiments performed on real proteins, 10 trays (720 experiments) for each sampling protocol (random, footprint, grid) for each of five proteins have been set up and monitored for crystal formation. Sure enough, random sampling is the most efficient on average and in nearly every case, as shown by the number of experiments needed to arrive at the first crystal. Catalase is an interesting contrary example--in this case, every tray has crystals and all methods are nearly equivalent.
| Crystal Formation by Strategy |
|
Other Benefits of Random Screening
Comprehensive screening
Fine screening (optimization screening) requires no additional time or setup
More meaningful database
Below is one example of a very useful analysis that cannot be performed on other crystallization databases--success rate.
| Success Rate by Precipitant |
|
Applications