Background
Positron emission tomography (PET) has become a key tool in medical imaging with primary applications being cancer diagnosis, staging, and monitoring of treatment [
1‐
3], with additional applications in many other disciplines including neurology [
4,
5], cardiology [
6,
7], and pharmacology [
8,
9]. As the number of applications and PET probes has grown, so has the need to produce an ever-increasing variety of radiolabeled compounds for both preclinical and clinical uses. Probes are traditionally synthesized by skilled radiochemists using specialized equipment and facilities that reduce their radiation exposure when working with large quantities of short-lived isotopes necessary to produce a final dose sufficient for imaging a human. In recent years, the development of automated radiosynthesizers that can produce a variety of different probes with minimal human intervention or radiation exposure [
10,
11] has aimed to simplify routine synthesis of PET probes, especially for the clinic. As such, these synthesizers can be operated by technicians and do not require a highly trained radiochemist. Additionally, some automated systems can be configured to prepare different PET probes and thus also act as valuable tools for researchers developing new synthesis protocols for novel probes. In order to be useful to chemists, these systems must also provide an intuitive and easy-to-use software interface for the creation and modification of synthesis programs.
There are a variety of radiochemical synthesizers on the market with a range of features and capabilities [
12]. Examples include IBA's Synthera® [
13], GE's FASTlab [
14] and TRACERlab [
15], Eckert & Ziegler's Modular-Lab [
16] and PharmTracer [
17], and Siemens' Explora® GNs [
18]. However, the software that drives these systems tends to be overly complex and requires a deep understanding of the system internals. For example, synthesis programs on the Synthera® are composed of low-level operations such as switching individual valves within the main system or the disposable cassette (integrated fluidic processor). As a result, the program to perform the relatively simple synthesis of 2-[
18F]fluoro-2-deoxy-
d-glucose ([
18F]FDG) requires a program that consists of 227 steps (determined from the protocol supplied by the manufacturer at the time of installation). These programs are written in a scripting language that may not be intuitive to a radiochemist or technician without computer programming experience and hence require the investment of significant time and energy to master. The radiochemist is also required to have a detailed understanding of the mechanisms of fluid transfer and the fluidic architecture of the system. The Modular-Lab software enables programs to be built graphically as flow charts, rather than written as scripts, but the flow chart elements consist of low-level hardware operations or steps to activate visual cues that are presented on an engineering schematic of the system during production runs. The complexity of developing synthesis programs is not a concern for routine production, where fixed programs are run on a regular basis, but becomes a significant hurdle for radiochemists, who frequently develop and optimize automated synthesis protocols for novel probes. Several previous works have attempted to reduce the complexity of developing new synthesis protocols by implementing higher-level unit operations or macros but still required the user to be familiar with the low-level system details and write syntheses in cryptic scripting languages [
19‐
21].
We describe here the development of a software package [
22,
23] to run the ELIXYS (Sofie Biosciences, Culver City, CA, USA), a disposable cassette-based, automated multi-reactor radiosynthesizer [
24‐
28] that is designed for both the development of new synthesis protocols and routine probe production. To facilitate the creation and modification of programs, this software is based on high-level unit operations designed to make intuitive sense to a chemist. A small number of adjustable parameters for each of these operations provide considerable flexibility to implement diverse syntheses and optimize conditions but do not require a detailed understanding of fluidics and low-level hardware architecture. Examples of these unit operations are ‘ADD’ which adds any reagent to any reaction vessel, ‘REACT’ which performs a reaction under sealed conditions, and ‘TRANSFER’ which transfers the contents of one reaction vessel to another with an optional cartridge purification step. The client software uses a drag-and-drop interface to further simplify the programming process and is designed to run on multi-touch tablets and phones. The server software that runs the instrument supports multiple client connections simultaneously, to allow others to watch the current synthesis run for increased transparency and oversight, and is tolerant of failures of client devices without impacting a production run in progress. We describe the software architecture and user interface design to illustrate these differences, evaluate the software in terms of its ease of use, and demonstrate the successful synthesis of several PET probes. It is our hope that this new software will empower radiochemists to focus on chemistry rather than engineering and to develop and produce new probes more quickly.
Results and discussion
Using the client programming interface, we developed a program for the synthesis of [
18F]FDG, the most widely used PET probe, that requires a total of 7 reagents (Table
2) and 18 unit operations (Table
3). The concept of unit operations abstracts the steps of the synthesis from the details of the hardware to such an extent that almost no information about the engineering design of the radiochemical synthesizer is needed to understand the synthesis protocol; indeed, very few details about the underlying hardware can be gleaned from the program alone. The small number of unit operations required to synthesize [
18F]FDG on the ELIXYS hides the complexity of the underlying system from the user; these 18 unit operations translate into 476 low-level hardware operations (e.g., individual motions of reaction vessels and switching of valves). For comparison, similar synthesis protocols for [
18F]FDG on IBA's Synthera® and GE's FASTlab, synthesizers that were programmed with the traditional paradigm of scripting low-level hardware operations, require a total of 227 and 335 program steps to be defined by the user, respectively (Table
4). It is important to note that there are small differences between the ELIXYS and Synthera® [
18F]FDG synthesis protocols (e.g., number of azeotropic drying steps, acid vs. base hydrolysis, cartridges for purification) and that the ELIXYS protocol has not been optimized. [
18F]FDG produced on ELIXYS passed all quality assurance tests (e.g., radiochemical purity, pH, Kryptofix 222, pyrogenicity, sterility, residual solvent levels, TLC retardation factor (
R
f) comparison with cold standard, and radioisotope identity). Decay-corrected radiochemical yield of 65% ± 2% (
n = 3) was obtained for this 45-min synthesis. The uncorrected yield of 52% is slightly lower than the 60% reported in IBA's product literature [
34], but optimization could improve the final yield.
It was found that the same set of unit operations was sufficient to synthesize several simple probes in 12 to 15 steps including
N-succinimidyl-4-[
18F]fluorobenzoate ([
18F]SFB), [
18F]fluorothymidine ([
18F]FLT), and (
S)-
N-((1-allyl-2-pyrrrolidinyl)methyl)-5-(3-[
18F]fluoropropyl)-2,3-dimethoxybenzamide ([18F]Fallypride). The performance of these syntheses was comparable to literature reports and will be published separately. Consistent with the results for [
18F]FDG, the ELIXYS program for [
18F]SFB consisted of only 15 unit operations compared to the 206-step program to produce [
18F]SFB with Synthera®. Similarly, synthesis programs for [
18F]FLT on ELIXYS and Siemen's Explora RN required 15 and 87 steps, respectively. Complex probes that utilize all three reaction vessels, such as
d-[
18F]FAC and
l-[
18F]FMAU required more steps than one-pot syntheses (i.e., 42 steps).
d-[
18F]FAC and
l-[
18F]FMAU were produced with acceptable quality and with yield and synthesis time comparable to literature reports [
29]. Since no other commercial system is capable of synthesizing these probes with comparable yields, program lengths cannot be compared to other systems but would likely be substantially shorter. Of particular importance is the observation that no tweaking beyond the adjustment of parameters in the high-level unit operations was required to produce the variety of probes above with yields comparable to those obtained on more traditional radiosynthesizers. A manuscript detailing several different chemical syntheses on the ELIXYS system will be published separately.
The consistent finding that program length is considerably shorter on the ELIXYS system suggests that it will be easier and faster to create and edit synthesis programs compared to conventional software approaches. It is also likely that these programs will have fewer bugs and be easier to debug.
As another measure of the ease of programming, six students with no prior experience with automated radiosynthesizers took part in an exercise to estimate the learning curve of the software. The students were able to complete programming the system within 1 h and subsequently set up and execute the program on the ELIXYS synthesizer with very little assistance. While not a controlled trial, this exercise was informative both because the students found the system easy to use overall and because it highlighted several areas of the user interface that were less intuitive than originally thought. The observations from this exercise will be used to enhance the client software.
When running a synthesis program, the ability to shorten or extend the duration of timed unit operations such as reaction or evaporation steps was found to be advantageous for synthesis development and optimization. In combination with the live video from the active reactor, we found this feature useful in dealing with situations such as determining the optimum length of time to evaporate to a specific level, measuring the time required to reach the endpoint of a reaction where a visible change (e.g., color) is associated with reaction completion, determining the required time to transfer a new type or volume of liquid from a reagent vial to the reaction vessel, and measuring the length of time required for the contents of a reaction vessel to be purified through a new type of cartridge. The actual duration of each adjustable step is recorded to the synthesis run history to enable the user to review how long each step actually took as well as to replay a previous program run.
In terms of reliability, in over 100 production runs, we have not observed any failures due to software error. Furthermore, as designed, we have noted that the production run has continued without interruption in several cases where a client device failed, e.g., due to running out of battery power, or when network connectivity was temporarily disrupted.
Although ELIXYS is currently supporting only preclinical PET probe production, we are in the process of implementing several software features to enable compliance with cGMP guidelines and plan to place the upgraded system into a laboratory that supports clinical production of PET probes under both 21 CFR 212 and USP 823. Additionally, Sofie Biosciences is working on amending the CMC section of an ongoing phase 1 clinical trial for d-[18F]FAC to include ELIXYS for the production of this probe.
Competing interests
RMvD and MM own equity in Sofie Biosciences, and RMvD is a consultant for Sofie Biosciences. SBC and KMQ have worked for Sofie Biosciences as paid consultants. ML declares that he has no competing interests.
Authors’ contributions
All authors - SBC, KMQ, ML, MDM, and RMvD - were involved in the design of the software architecture and user interface. KMQ performed low-level programming of the PLC; SBC and KMQ programmed the core server applications, and SBC programmed the client application. All authors contributed to the testing and refinement of the complete software package. ML developed and performed the [18F]FDG synthesis. SBC, ML, and RMvD wrote the manuscript. All authors read and approved the final manuscript.