Fast fragment- and compound-screening pipeline at the Swiss Light Source

The FFCS platform has recently been developed by the Swiss Light Source macromolecular group. It is an integrated next-generation X-ray crystallography fragment-based drug-discovery pipeline for crystal soaking, handling and data collection.


Introduction
The identification of small molecules that modulate protein function and activity is a crucial step in drug development. Fragment-based drug discovery (FBDD) has emerged as an effective and efficient method to identify promising chemical scaffolds for the development of lead compounds (Erlanson et al., 2016). FBDD is an alternative to high-throughput screening (HTS), which uses biochemical or biophysical assays to search for lead compounds within large libraries that have tens of thousands to millions of members (Fox et al., 2006). The size of these libraries presents significant logistical challenges, but they can still sample only a small portion of the possible combinatorial chemical space (Hann & Oprea, 2004) and are frequently expanded to address new targets. HTS is effective in identifying compounds that bind tightly to target proteins, but these molecules are often not drug-like due to high lipophilicity or disadvantageous pharmacokinetic properties. From an imperfect and over-elaborate starting point, optimizing (or retaining) the potency and selectivity of hits while simultaneously improving bioavailability can be extremely challenging. In the past two decades, these inherent difficulties of HTS have triggered the development of FBDD, which has attracted increasing interest from both academia and industry (Erlanson et al., 2016).
In contrast to HTS, FBDD utilizes small libraries (1000-3000 members) of simple molecules (<300 Da in molar mass, clogP <3, <3 rotatable bonds, <3 hydrogen-bond donors or acceptors; Congreve et al., 2003). For simple molecules, a small ISSN 2059-7983 library can effectively sample the respective chemical space. Fragments typically bind weakly to target proteins (K d = 10 À2 -10 À4 M) but with high ligand efficiency. In order to progress towards high-affinity compounds, the binding affinity of hits can be increased by coupling fragments ('linking and merging') or derivatizing a single fragment ('growing'). This process affords a greater understanding and control of the binding mode and pharmacokinetic properties of the resulting lead compounds compared with the development of HTS hits. To date, drug candidates arising from FBDD programmes have achieved considerable success, with six drugs approved by the United States Food and Drug Administration (USFDA) and more than 15 candidates currently in clinical trials (Erlanson et al., 2016;Jahnke et al., 2020;http:// practicalfragments.blogspot.com/2021/11/fragments-in-clinic-2021. The detection of binding during fragment-library screening is confined to assays with appreciable throughput that can detect binding with dissociation constants in the high-micromolar to millimolar range. In these assays, high concentrations of fragments must be used to achieve an appreciable saturation of binding, necessitating the use of co-solvents (usually DMSO) to increase solubility. Commonly used techniques for fragment screening include surface plasmon resonance, nuclear magnetic resonance, thermal denaturation midpointshift assays and mass spectrometry. When evaluating the development potential of fragment hits, other lowerthroughput techniques such as isothermal titration calorimetry, microscale thermophoresis, biochemical assays and X-ray crystallography can be used to provide information about binding affinity and thermodynamics, as well as binding location and orientation.
In addition to its use in validating and characterizing the binding of fragment hits identified by other techniques, X-ray crystallography has achieved great success as a primary screening method. It is one of the most sensitive and reliable techniques for the detection of binding, and gives key structural information on fragment poise. In addition, X-ray crystallographic FBDD (xFBDD) can be 'site-agnostic', identifying and locating fragments bound both at known active sites and at allosteric sites. Thus, it allows druggability to be assessed for the whole accessible protein surface, without assumptions, and in the absence of known high-affinity binders. The use of X-ray crystallography for primary screening in FBDD was first demonstrated in 1997 (Verlinde et al., 1997). Since then, the clear advantages of the method have established it as an integral part of the drug-discovery platforms in several companies (Spurlino, 2011;Hubbard et al., 2007;Davies et al., 2006;Price et al., 2017). In the last halfdecade, the approach has become more widely adopted. Technological advances at MX beamlines in terms of instrumentation, beam intensity and robotics have enabled the development of dedicated screening platforms at synchrotron sources. This approach was spearheaded by XChem at Diamond Light Source (DLS), which offered the first proposal-based full xFBDD (hundreds of fragments) platform to synchrotron users (Douangamath et al., 2021;Krojer et al., 2017). In parallel or subsequently, xFBDD platforms were developed at other synchrotron sources Lima et al., 2021;Cipriani et al., 2012;Lima et al., 2020).
The Swiss Light Source (SLS) macromolecular crystallography (MX) group has recently developed their own xFBDD platform, called the Fast Fragment and Compound Screening (FFCS) platform. In this publication, we present this newly established FFCS pipeline, focusing on hardware and software solutions.

Outline of the FFCS pipeline
The FFCS pipeline was developed within the MX group at the SLS. The main aim of the FFCS project was the development of an integrated next-generation pipeline for crystal soaking, handling, and data collection and analysis, which would aid high-throughput drug-discovery processes in the pharmaceutical industry and academic laboratories. The FFCS pipeline includes the following steps ( Fig. 1).
(i) Crystal growth: crystals are grown in SWISSCI MRC-3 plates, which are stored and imaged in a Rock Imager (Formulatrix) at 4 C or room temperature (RT).
(iv) Data collection: at one of the three SLS MX beamlines.
(v) Initial data processing and structure solution: within the beamline computing environment.
The process, which involves handling hundreds of crystals and fragments/compounds, requires the creation of tools to make it faster, more efficient and more automated. The crucial components of the FFCS pipeline are the hardware setup and the dedicated in-house-developed software, the FFCS software suite, which allows easy experiment design, efficient bookkeeping and links all of the steps of the FFCS pipeline. Close collaboration with the group of Gü nter Fritz, University of Hohenheim, enabled the extension of the initial data processing and structure solution with an advanced dataevaluation pipeline developed in his group (Stegmann et al., in preparation).

Hardware setup
In order to accommodate new equipment, the SLS Macromolecular Crystallization Facility (CF), which is localized in close proximity to the MX beamlines, was rearranged substantially. A separate space dedicated solely to the FFCS equipment was created and adapted accordingly. The FFCS hardware setup was to a large extent inspired by the XChem Fragment Screening facility at DLS, which is the largest facility for high-throughput crystallographic fragment screening that is tailored for academic users.
The Crystallization Facility is equipped for robotic crystallization screening using sitting-drop vapour diffusion (or lipidic cubic phase) with a Mosquito robot (SPT Labtech), which is available at both 4 C and RT. Crystallization plates are barcoded, stored and imaged in dedicated 4 C or RT Rock Imager (Formulatrix) plate hotels. The Echo 550 noncontact liquid handler (Labcyte) is mounted on a specialized vibrationfree table to prevent any interference during acoustic droplet injection experiments. To ensure longevity and to maintain high quality, FFCS fragment and compound libraries are stored in barcoded Echo Qualified 384-well or 1536-well COC plates (Labcyte) in a dedicated storage system. The Storage-Pod system (Roylan Developments) consists of storage pods housing the fragment libraries and a controller unit to purge the storage pods with dry nitrogen gas. This enables the storage of fragment and compound libraries in an inert, moisture-free and oxygen-free environment, preventing their dilution, precipitation and damage. The library storage plates are stored at room temperature, since repeated freeze-thaw cycles of DMSO-dissolved substances are associated with compound degradation (Kozikowski et al., 2003). To further extend their durability, the FFCS library storage plates are heat-sealed with a PlateLoc Thermal Microplate Sealer (Agilent) using an aluminium-based material seal.
The harvesting of fragment/compound-soaked crystals is assisted by a Crystal Shifter (CS) robot (Oxford Lab Technologies). The CS device is a motorized xy microscope stage with computer software which enables semi-automated errorreduced crystal picking by hand while preventing drying of the crystallization drops. The harvesting station is also equipped with a puck scanning station (the 3D-printing design of which was kindly shared with us by the DLS staff) equipped with two dedicated video cameras which enable scanning of both the puck and, optionally, the pin barcodes.
An extended effort was put into the FFCS laboratory design to ensure the ergonomic use of space and to guarantee an optimal workflow. Devices are located in close proximity in an arrangement reflecting the steps in the experimental procedure. Solutions such as a liquid-handler vibration-free table and a dedicated library storage system bestow robustness and reliability to the pipeline and improve upon the already existing facilities.

FFCS software suite
The in-house-developed FFCS software suite facilitates experiment design and seamless step-by-step progression within the FFCS pipeline while efficiently record-keeping on behalf of the user. The FFCS suite consists of the FFCS database (FFCS DB) and four graphical interfaces: the FFCS GUI, Xi GUI, soakedMe GUI and shiftMe GUI ( Schematic of the FFCS pipeline. Initial crystals are optimized in SWISSCI MRC-3 plates and tested for DMSO resilience. Plates are stored in Rock Imager systems at 4 C or 20 C and are imaged at regular intervals. Selected fragments are contactlessly transferred from small-volume stock solutions to preselected crystal drops with the Echo liquid-handling device. Subsequent semi-automated crystal harvesting, supported by the Crystal Shifter robot, allows the efficient freezing of soaked crystals. (remote access is also possible in the form of a full graphical user session via NoMachine; https://www.nomachine.com/). The FFCS GUI sends input files for Echo and Crystal Shifter experiments, while the soakedMe (Echo PC) and shiftMe (Crystal Shifter PC) GUIs report the results of soaking and fishing experiments back to the FFCS GUI.

FFCS database and FFCS GUI
The FFCS DB system receives and stores all of the metadata produced by users throughout FFCS campaigns. The FFCS DB is deployed on a dedicated virtual machine (Linux OS) as a docker container (https://www.docker.com/). The FFCS DB benefits from a MongoDB (https://www.mongodb.com/) schema-less design, which does not enforce a strict format of metadata. Currently implemented FFCS DB collections are 'Plates', 'Wells' and 'Libraries'. Each collection includes an implemented validator to check for required keywords for each insert operation. Moreover, operation is not permitted if one of the keywords does not exist or the keywords are of the wrong type. All Python software that needs access to data in the FFCS DB utilizes a dedicated client, which is tailored to retrieve specific data as required by the database.
The FFCS GUI is a frontend PyQt5 (Python 3.6) application developed to interact with documents stored in the FFCS DB. It can be accessed via a dedicated FFCS user console that utilizes the same group accounts (nonpersonal e-accounts) to authenticate users at the beamline consoles. This allows the storage of all FFCS project-related documents in the same working space as is used for X-ray diffraction data collection. The e-account FFCS folder includes (i) copies of the crystallization drop images, (ii) Echo and Crystal Shifter input files (in .csv format), (iii) a data collection spreadsheet compatible with the TELL GUI, user software which controls the TELL automatic sample changer (Martiel et al., 2020), (iv) the results of the automated data-processing pipelines (adp and dimmer), (v) a spreadsheet with FFCS experiment reports and (vi) fragment-library file(s) (in .sdf format).
Upon startup of the FFCS GUI, the user is presented with a menu listing all ongoing FFCS campaigns associated with a given e-account (the campaign name functions as a unique identifier); alternatively, users can start a new campaign (Fig. 3a). The FFCS GUI tabs represent consecutive stages of a given campaign, namely 'Plates', 'Cryo', 'Soaking', 'Redissolve' and 'Fishing'. Crystallization plates can be added to the ongoing FFCS campaign in the Plates tab. The plate ID can be typed in manually or it can be scanned with the barcode reader attached to the user console. The list of crystallization plates associated with the given FFCS campaign is retrieved by the Xi GUI and updated after the dispensing locations have been selected. The Cryo tab is an optional step of the FFCS pipeline and must be performed before the final soaking experiment. The Soaking tab allows the selected fragments list to be combined with the targeted crystallization drops (Fig. 3b). The 'Export to Soak' button triggers ZeroMQ (https://zeromq.org/) transfer of the soaking-definition file to the Echo PC and of the Fishing definition file to the Crystal Shifter PC. It also starts a soaking duration timer, which is automatically refreshed every minute in the 'Soak Duration' column. After soaking has been completed, the Echo experiment report is sent to the FFCS DB, the Soaking tab timer is stopped, and the soaking status displayed in the FFCS GUI is updated accordingly. The Redissolve tab is an optional step of the FFCS pipeline which was created on demand for a challenging project that required DMSO-free soaking (Bedi et al., 2020). The final Fishing tab displays the results of crystal harvesting with the Crystal Shifter robot and includes an 'Export Data Collection Excel Sheet' button, which enables the export of a datacollection sample spreadsheet that is compatible with the SLS data-acquisition software.

Xi GUI
The Xi GUI is a PyQt5 interface (Python 3.6) which allows the easy identification of drops with suitable crystals and the precise definition of the fragment-dispensing location for targeted soaking experiments. Upon  startup the GUI (i) connects to the FFCS DB and identifies FFCS campaigns for a given user (based on the e-account name), (ii) extracts the barcodes of the crystallization plates assigned to a selected campaign and (iii) copies images from the Rock Maker storage to the dedicated FFCS campaign folder. Images of crystallization drops are displayed in the central Xi GUI window and the dispensing location can be assigned with a single mouse click (Fig. 4). Dispensing locations can be added/removed with left/right mouse clicks, while moving to the next/previous image is controlled with the left/right keyboard arrows. The total number of wells for a given plate, as well as the number of targeted wells, are displayed in the bottom left corner of the GUI. Selected well locations are exported to the FFCS DB as a list of x, y coordinates, which are displayed in the FFCS GUI Soaking tab.

ZeroMQ communication
We created a suite of ZeroMQ servers and clients and two dedicated GUIs to enable the easy transfer of files between computers within the FFCS pipeline. As mentioned in the previous section, upon user request the FFCS GUI sends input files to the Echo and Crystal Shifter PCs. These input files contain sufficient information for the FFCS Soaking and Fishing pipeline steps without disclosing any confidential or user-specific information. After completion of the soaking/fishing steps, dedicated GUIs (soakedMe and shiftMe) assist users in sending results back to the FFCS DB. The soakedMe GUI (Echo PC) sends an Echo dispensing status report, which includes error messages. The shiftMe GUI (Crystal Shifter PC) combines results from Crystal Shifter fishing and the barcode reading station before sending the report back. Both soaking and fishing experiment results are displayed for user inspection in the FFCS GUI.

Data collection and initial analysis
The X-ray diffraction data are collected manually or automatically on one of the   three SLS MX beamlines (PXI X06SA, PXII X10SA and PXIII X06DA). Manual data collection is performed using the SLS MX DA+ data-acquisition (daq) software GUIs, namely the DA+ GUI (Wojdyla et al., 2018) and the TELL GUI, which controls the TELL automatic sample changer (Martiel et al., 2020). Alternatively, unattended data collection with the Smart Digital User (SDU) automation software package is available at all three MX beamlines (Wojdyla et al., in preparation). SDU is a software alternative to the user controlling the experiment with the DA+ GUI and is an integral part of the SLS MX distributed daq software stack, communicating with many other software instances such as the TELL GUI and the mxdb database. SDU data collection and processing options can be defined by users in a sample spreadsheet (the template can be downloaded from the SLS MX webpage); alternatively, beamline-specific default parameters are applied. The automatic data processing (adp) provides instantaneous feedback on a subset of collected data (the first 'fast_xds' step) as well as fully processed data for the total angular range. In the second 'complete' step users can select one of the following processing pipelines: the in-housedeveloped gopy, autoPROC (Vonrhein et al., 2011) or xia2dials (Winter et al., 2018). The sample spreadsheet includes a column to indicate the name of the molecularreplacement model (located in a dedicated folder) which is utilized in the dimple pipeline (Wojdyr et al., 2013)   dedicated dimmer daemon wrappers. In the case of crystals soaked using the FFCS pipeline, the dimmer daemon extracts SMILES (simplified molecular-input line-entry system) codes for the fragments from the FFCS DB and generates ligand coordinates and restraints using eLBOW from the Phenix suite (Moriarty et al., 2009). The resulting dimmer output is saved in a dedicated FFCS folder (organized on a percampaign and per-sample basis), easing further downstream data analysis.

FFCS implemented
In order to evaluate the effectiveness and reliability of the FFCS pipeline under real-life conditions, we performed a set of in-house campaigns. Starting from initial tests with lysozyme and thermolysin, we progressed to endothiapepsin, the classic fragment-screening target (Kö ster et al., 2011;Huschmann et al., 2016;Schiebel et al., 2016;Wollenhaupt et al., 2020). One of the first academic collaboration FFCS projects was performed on protein crystals of two homologous bacterial enzymes. These systems functioned as challenging 'real-life' test cases for our pipeline. The two crystal systems relied on different precipitants, PEG and high salt, and both crystal systems had to be adapted by a distinct post-crystallization treatment to cope with the addition of large amounts of DMSO in the soaking procedure (Stegmann, Sharpe et al., in preparation). For both projects, the 500 most diverse fragments (selected based on the structure descriptor FragFP) and 50 project-specific fragments were selected from one of the FFCS pipeline fragment libraries (Maybridge Library Ro3, Thermo Fisher). Once the soaking conditions had successfully been optimized, for each of the studied crystal systems two to three days of intermittent laboratory work were sufficient to perform soaking and crystal harvesting ($250 crystals per day). The assisted harvesting with the Crystal Shifter robot resulted in an average of 50-60 crystals being fished per hour for both tested systems. These numbers may be higher for a trained user considering that both projects exhibited some skin formation on top of the crystallization droplets. The X-ray diffraction data sets were collected on beamlines PXI X06SA and PXIII X06DA either manually or with the SDU software. Both manual data collection and SDU resulted in 12-14 data sets with a full 360 rotation per hour. The resolution for diffraction data collected within the reported projects after optimization was 1.7 and 1.5 Å on average (resolution cutoff as defined by the gopy protocol in the adp software; Wojdyla et al., 2018) for projects 1 and 2, respectively, exhibiting a rather narrow distribution across all data sets (Figs. 5a and 5b). Data sets processed with adp (with the in-house XDS-based gopy pipeline) were fed into dimmer workers which calculate F o À F c difference density maps and generate fragment coordinates. Subsequent steps of data analysis were performed using the data-evaluation pipeline developed in the group of Gü nter Fritz. In brief, the pipeline consists of the following steps: (i) the creation of chemical restraints for each fragment with AceDRG (Long et al., 2017), (ii) automatic fitting of fragments into the difference density with rhofit (Smart et al., 2014), (iii) post-refinement including the proper placement of side chains and water molecules with REFMAC5 (Murshudov et al., 2011), Coot (Emsley et al., 2010) and phenix.refine (Liebschner et al., 2019), and (iv) estimation of the binding affinities of bound fragments by evaluation through a scoring function with smina (Koes et al., 2013) and SmOG (Noel et al., 2016). A bash script executing analysispipeline steps relies on a spreadsheet that contains the names of the data sets and the corresponding SMILES codes of the soaked fragments (input can be derived from the campaign spreadsheet exported from the FFCS GUI). An interactive helper script defines data source and destination paths and enables parallelization of the calculations. The entire fitting and evaluation method will be described and exemplified in detail elsewhere (Stegmann, Vering, Steuber & Fritz, in preparation). The hit rates for the two example projects were very similar, at 6.2% and 6.4%, respectively, and showed well defined, tightly bound fragments (Table 1).

Summary
With recent advances in MX automation and the consequent increase in popularity of xFBDD, we have implemented the FFCS pipeline at the SLS to offer users an additional avenue of using X-ray crystallography to generate and evaluate starting points for lead development in a rapid manner. Whereas the hardware setup for the FFCS pipeline is based on existing XChem solutions, the FFCS software suite was developed in-house. The SLS MX FFCS pipeline has been in operation since 2019, with a number of successful academic collaborations and industry customer projects. At the onset of the COVID-19 pandemic the project portfolio was expanded by a Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) protein target (Sutanto et al., 2021). Experience from the first FFCS campaigns clearly emphasized the importance of two crucial aspects of the pipeline: (i) DMSO optimization and (ii) software flexibility.
Before carrying out a large-scale soaking step, which can be performed with hundreds of fragments or compounds in a matter of hours, we first optimize the soaking conditions in order to obtain optimal data quality. Here, we vary the soaking time and concentration of the DMSO added to ascertain the appropriate soaking parameters. This often requires two to three rounds of iteration (a laboratory part combined with X-ray data collection). For crystal systems that require cryoprotection, additional rounds of testing are completed. Understanding the significance of the DMSO optimization step is necessary in order to perform a successful xFBDD  campaign Collins et al., 2018), especially in the case of challenging crystallization systems such as the collaborative project described in the section above.
The advantage of the flexible in-house-developed FFCS software suite was clearly demonstrated in a collaborative project on the human YTHDC1 domain. DMSO optimization revealed that the active site is occupied by a tightly bound solvent molecule that prevents the binding of a positive control. Based on results from initial optimization we instead performed DMSO-free soaking, in which preselected fragments dissolved in DMSO were deposited onto a crystallization plate with the Echo device, air-dried to remove the solvent and redissolved in the crystallization buffer. Subsequently, 70 YTHDC1 domain crystals were manually transferred to crystallization drops for soaking and harvested (both steps were assisted by the Crystal Shifter), resulting in 30 fragment-bound structures (Bedi et al., 2020). The dedicated Redissolve tab was created in the FFCS GUI on demand for this particular project, enabling the tracking of a DMSO-free soaking experiment in the FFCS DB.
In recent years great progress has been made in beamline automation, data collection and automatic data processing. This has enabled the development of dedicated xFBDD screening pipelines, which are now offered by many synchrotron facilities. The high-throughput laboratory setup combined with easy-to-use software makes more advanced early drug-discovery experiments routine for both expert and inexperienced users. At the same time, synchrotron facilities have widely expanded their services well beyond standard X-ray data collection, providing pre-and post-experiment steps such as sample preparation and data analysis, all within the same synchrotron environment.