Evolution of macromolecular crystallography beamlines at the Swiss Light Source and SwissFEL

This article forms part of a virtual issue celebrating the 50th Anniversary of the Stanford SSRL synchrotron radiation and protein crystallography initiative led by Keith Hodgson.

This review highlights the development and evolution of three macromolecular crystallography (MX) beamlines at the Swiss Light Source (SLS) over the past two decades. We discuss key advancements in X-ray optics, detectors, goniometers, sample changers and MX methodology, emphasizing their impact on high-throughput and high-resolution structural biology. Our contributions are presented within the broader context of global efforts in synchrotron-based MX. Looking ahead, we explore the future experiments enabled by SLS 2.0 and new opportunities at SwissFEL to enhance experimental capabilities and drive scientific discoveries.

1. Introduction

Following the success of high-energy third-generation synchrotrons in the 1990s – such as the 6 GeV European Synchrotron Radiation Facility—ESRF (1994), 7 GeV Advanced Photon Source—APS (1995) and 8 GeV Super Photon Ring—SPring-8 (1997) – the Swiss Light Source (SLS) started user operation in 2001 as a medium-energy synchrotron (2.4 GeV) (Nolting et al., 2023). The SLS features a compact design with 288 m circumference, where the booster and storage rings share the same tunnel. The source emittance is 5500 pm rad and 5 pm rad (horizontal versus vertical). From the start, the SLS operated in top-up mode (Böge, 2002), ensuring stable, continuous beamline operation with a 400 mA electron beam current (Ludeke et al., 2006). The success of the SLS marked the beginning of a new era, leading a wave of new national synchrotron facilities around the world (https://lightsources.org/).

The SLS supports various scientific applications across 18 beamlines, with three being dedicated to macromolecular crystallography (MX) (Hendrickson, 2000; Helliwell, 1992). In 2024, the MX beamlines celebrated 10000 structures in the Protein Data Bank (PDB) and 4700 publications (Fig. S1 of the supporting information), which count for half of all publications from the SLS. In addition, numerous structures have been determined for proprietary research in drug discovery (Hennig et al., 2012; Tosstorff et al., 2022; Vulpetti et al., 2023; Käck & Sjögren, 2025). The Swiss Free-Electron Laser (SwissFEL) was established in 2017 to complement the SLS. SwissFEL has two branches to cover both hard X-ray and soft X-ray applications, and two endstations provide capabilities for serial femtosecond crystallography (SFX) with a variety of sample delivery methods (Milne et al., 2017; Nolting et al., 2023).

2. Three MX beamlines at the SLS – X06SA-PXI, X10SA-PXII and X06DA-PXIII

The first MX beamline X06SA-PXI was designed to leverage in-vacuum undulator technology (Hara et al., 1998), extending high-brightness radiation into the hard X-ray regime (3–18 keV) within a medium-energy synchrotron. This achievement was made possible through a collaboration between the Paul Scherrer Institute (PSI) and SPring-8, which facilitated the installation and operation of the first in-vacuum undulator (U24) at X06SA-PXI shortly after the SLS storage ring's commissioning in 2001 (Ingold et al., 2007). The in-vacuum, small-gap, short-period undulator with small phase error operating on higher harmonics proved to be a stable source of high-brightness radiation, leading to the installation of similar undulators at other SLS beamlines, including two U19 undulators at X06SA-PXI (replacing U24) and X10SA-PXII.

The X06SA-PXI source size and divergence are 202 µm × 23 µm and 135 µrad × 25 µrad (horizontal versus vertical, FWHM) at 12.4 keV, respectively. The X-ray optics system was engineered to maximize the flux by collecting the full undulator harmonics while enabling adaptive control of the beam size and divergence at two sample positions. This was achieved by combining a novel double-crystal monochromator featuring a sagittal bender on the second crystal (Schulze-Briese et al., 1998) and a flexural hinge-based mirror bender in the vertical focusing mirror (VFM) (Rossetti et al., 2002). The sagittally bendable Si crystal collected up to 150 µrad of X-rays and focused them in the horizontal direction while the VFM provided achromatic focusing in the vertical direction with two stripes of different surface material, uncoated Si and Rh. The beamline was fully tunable between 5.7 and 17.5 keV and was optimized for energies around the selenium K-edge of 12.66 keV with an energy resolution of 2 × 10⁻⁴ for experimental phasing using multiple- and single-wavelength anomalous diffraction (MAD/SAD). At the starting time of the beamline, about half of the structures needed de novo phasing (Hendrickson, 2014). With the brighter X-ray beam, control of radiation damage became an integral part of the MX data collection and crucial in experimental phasing. To this end, we introduced absolute flux and dose estimation (Owen, Holton et al., 2009) to calculate data collection strategies (Dauter, 1999).

The beamline hosted two experimental endstations for high-resolution diffraction (HRD) and micro-diffraction (MD). The HRD station, optimized for resolving crystals with large unit cells, provided a low-divergence beam (85 µm × 10 µm, 320 µrad × 70 µrad) with a flux of 2 × 10¹² photons s⁻¹ at 12.4 keV. The beam size could be adjusted to match crystal size, enhancing the signal-to-noise ratio. Thanks to the minimal aberration in sagittal focusing and low slope error of the VFM, at the MD station designed for micro-crystallography, tight focusing (25 µm × 5 µm, 1100 µrad × 350 µrad) with the same flux was achieved. Horizontal beam size could be further reduced to 10 or 5 µm using apertures integrated with a micro-diffractometer developed at the European Molecular Biology Laboratory (EMBL, Grenoble) (Perrakis, Cipriani et al., 1999). A secondary source and Kirkpatrick–Baez (KB) mirror system were added in 2012 to enhance flexibility, allowing finer control over the beam size and divergence at the sample position. The beam size range 1–100 µm addressed the user community's demand nicely (Fig. S2 of the supporting information).

The second beamline X10SA-PXII was constructed for proprietary research exclusively with the three founding partners, Max-Planck-Gesellschaft, Novartis and Roche, in 2004 (Diez et al., 2007). The X-ray optics design was built based on the experience of X06SA-PXI, leading to a focused beam size and divergence of 50 µm × 10 µm and 540 µrad × 130 µrad at 12.4 keV, respectively. The energy range was extended to 20 keV with an additional Pt coating on the vertical focusing mirror. A dedicated diffractometer (D3) (Fuchs et al., 2014) was developed with three key features: (1) sub-micrometre sphere of confusion for the horizontal single-axis rotation and fast sample-rastering; (2) beam-shaping apertures for micro-crystallography with a 10 µm beam; (3) on-axis microspectrophotometer for multi-mode optical spectroscopy (Owen, Pearson et al., 2009; Pompidor et al., 2013). Later, a secondary source and kinoform lenses were added to explore refraction-based X-ray focusing for MX beamlines (Lebugle et al., 2018). The beamline served academic research and structure-based drug discovery, and gradually attracted more industry partners. Part of the `in-house research' beam time was also used for the Structural Genomics Consortium (SGC) (Williamson, 2000) before the Diamond Light Source (DLS) was constructed.

The third beamline, X06DA-PXIII, exploited the small electron beam size at a bending magnet source and used a 2.7 T super-bending magnet to push the critical energy to 11 keV for MX applications. The X-ray focusing optics consisted of a vertical collimating mirror in the front end and a toroidal focusing mirror. The focusing aberration of the toroidal mirror was minimized using a 2:1 focusing ratio (MacDowell et al., 2004), which produced a fixed focus of 90 µm × 45 µm with divergence of 2 mrad × 0.5 mrad at the sample position. One unique design was the double channel-cut monochromator for a true fixed-exit of X-rays for the 5–17.5 keV energy range. Energy changes required only two rotations, ensuring both speed and stability. This design was later proven to be beneficial for experimental phasing.

The endstation was inspired by the mini-hutch design of beamline 8.3.1 from the Advanced Light Source (ALS) (MacDowell et al., 2004). The experimental hutch had three access windows for different experimental modes. The front window enabled fast sample exchange in manual operation, the side window was used for loading pucks into the sample changer, and the back window, serving as a portal for fully automated in situ screening of crystallization plates (Fig. S3 of the supporting information), was connected to a crystallization facility. The latter was integrated at the beamline to facilitate on-site sample preparation, crystallization and in situ screening experiments (Bingel-Erlenmeyer et al., 2011). The PSI and industry partners – Actelion (now Idorsia), Boehringer Ingelheim, Mitsubishi Chemical, Novartis and Proteros Biostructures – co-financed the beamline. X06DA-PXIII started user operation in 2008.

The three beamlines complemented each other and fostered the development of transformative MX techniques and methods. They are among the most productive MX beamlines worldwide (Zheng et al., 2014; Grabowski et al., 2021). The beamline development timeline and selected highlights are illustrated in Fig. 1. The main beamline characteristics and the X-ray optics design are listed in Table 1 and Table S1 of the supporting information.

Table 1
Main characteristics of three MX beamlines at the SLS and SLS 2.0

	X06SA-PXI		X10SA-PXII		X06DA-PXIII
	SLS	SLS 2.0	SLS	SLS 2.0	SLS	SLS 2.0
Start year	2001	2025	2004	2025	2008	2025
Machine energy (GeV)	2.4	2.7	2.4	2.7	2.4	2.7
Source	U19, 1.9 m	U17, 3 m	U19, 1.9 m	U17, 3 m	SB, 2.7 T	2.1 T
Photon source size, divergence (µm, µrad in FWHM at 12.4 keV)	202 × 23 135 × 25	47 × 10 33 × 28	202 × 23 135 × 25	47 × 10 33 × 28	130 × 20 402 × 308	16 × 16 263 × 251
Energy (keV)	5.7–17.5	5–30	6–20	5–30	5–17.5	4–15
Energy resolution	10^−4	10^−4, 10^−2	10^−4	10^−4	10^−4	10^−4
Flux (photons s^−1 at 12.4 keV)	10^12	10^13, 10^15	10^12	10^13	10^11	10^12
Beam size (µm in FWHM)	5–100	1–100	15–100	1–100	90 × 45 (h × v)	15–100
Goniometer	SmarGon	SmarGon	SmarGon	SmarGon	PRIGo	SmarGon
Detector	EIGER 16M	EIGER 16M, JUNGFRAU 9M	EIGER2 16M	EIGER2 16M	PILATUS 2M	PILATUS4 2M
Sample changer	TELL	TELL2	TELL	TELL2	TELL	TELL2

Figure 1 Annual PDB depositions (SLS – light blue bar, SwissFEL – red bar), development of MX beamlines at SLS and SwissFEL and selected scientific highlights [Lac. Permease (PDB entry 1pv6; Abramson et al., 2003), 70S Ribosome (PDB entry 4v51; Selmer et al., 2006), polyhedra (PDB entry 2oh5; Coulibaly et al., 2007), ion channel (PDB entry 3ehz; Hilf & Dutzler, 2009), GPCR (PDB entry 2x72; Standfuss et al., 2011), CRISPR-Cas (PDB entry 4cmp; Jinek et al., 2014; Yamano et al., 2016), RNA polymerase (PDB entry 5o7x; Engel et al., 2017), SARS protease (PDB entry 6yva; Shin et al., 2020), Chloride-pump (PDB entry 7o8y; Mous et al., 2022), designed GPCR (PDB entry 8oyv; Goverde et al., 2024), Industry drug design (PDB entry 8pfp; Ferretti et al., 2024)].

3. Innovations in MX beamline technology and methods development

Driven by advancements and automation in protein production and crystallization, global structural genomics initiatives, and structure-based drug discovery, continuous innovation in MX beamline technology over the past three decades has transformed synchrotron MX. What was once a specialized technique has now become a widely accessible and indispensable tool for both academic research and industrial applications. The SLS has made a few valuable contributions to this global endeavor. Techniques developed for one specific application were often essential for other unforeseen or yet unknown applications. We highlight some of them here.

3.1. Integrated X-ray optics concept for reliable beamline operation

The X-ray optics system was designed to maintain a stable X-ray beam position across the entire energy range by integrating high-resolution mechanics, real-time beam position monitoring and active feedback. The mechanics of monochromator and mirror benders were engineered for precise beam steering with micrometre-level precision. Such precision is essential for MX experiments, where even minor misalignments can affect the accuracy of measurements and data quality.

However, maintaining a micrometre-sized beam at the sample position over extended periods is challenging due to potential drifts in the electron orbit and thermal fluctuations in the optics system, especially after changing the X-ray energy with the monochromator. To address these issues, we developed a quadrant X-ray position monitor chemical vapor deposition (CVD) diamond, capable of tracking the X-ray beam position with micrometre precision (Schulze-Briese et al., 2001; Pradervand et al., 2004). The 12 µm-thin CVD diamond is nearly transparent over the full energy range of the beamlines and multiple sensors can be installed along the X-ray path to monitor both beam position and angle. This system enables real-time correction of beam drift by steering the X-ray optics and automates energy changes without the need for manual intervention, thereby making beamline operation and MAD/SAD phasing experiments more user-friendly and boosting overall productivity. Additionally, this sensor technology has been commercialized by DECTRIS Ltd and made available to the synchrotron community.

3.2. X-ray detectors from PILATUS, EIGER to JUNGFRAU

For detecting the X-ray diffraction, large format charge-coupled device (CCD) detectors were developed and widely used at MX beamlines at third-generation synchrotrons from the 1990s (Strauss et al., 1990; Gruner et al., 2002). The SLS MX beamlines were initially also equipped with MAR165 and MAR225 CCD detectors (https://www.rayonix.com). To fully harness the increasing brightness at synchrotron beamlines, PSI initiated what was later referred to as PILATUS project to develop a large-format hybrid pixel-array detector for MX applications (Eikenberry et al., 2003). The first generation of PILATUS detectors was commissioned at the SLS, with a PILATUS 6M introduced for user operation at X06SA-PXI in 2007 (Henrich et al., 2009) [Fig. 2(a)]. Building on this innovation, the next-generation detector, EIGER, was launched in 2015, featuring smaller pixel sizes, shorter readout times and higher frame rates (Dinapoli et al., 2011) [Fig. 2(b)].

Figure 2 Detector evolution and selected applications. (a) The PILATUS detector enabled fine-phi slicing data collection (Mueller et al., 2012). (b) The high frame-rate EIGER detector allowed fast 2D diffraction scan for micro-crystallography (Wojdyla et al., 2016). (c) The kilohertz JUNGFRAU detector and millisecond time-resolved crystallography (Leonarski, Nan et al., 2023). (d) Data rate growth and autonomous MX for ligand screening (Huang, Metz et al., 2024).

The unique features of PILATUS/EIGER detectors transformed MX data collection and processing, profoundly impacting synchrotron MX (Förster et al., 2019). Single-photon sensitivity, zero point-spread function and high dynamic range enabled precise recording of both weak and strong diffraction spots. This capability was especially critical for challenging experiments, such as crystallography of ribosomes and large molecular complexes (Fig. S4 of the supporting information) (Neubauer et al., 2009), where the single-photon sensitivity and zero-readout noise allowed thousands of weak intensity reflections at high diffraction angles to be captured, reaching higher resolution from crystals with unit-cell dimensions as large as 1000 Å.

In addition, the deadtime-free millisecond readout time made continuous, shutterless data collection feasible, improving data collection precision and reducing data collection time. Moreover, the combination of zero readout noise and fine-phi slicing minimized background noise, enhancing the signal-to-noise ratio and enabling higher diffraction resolution (Mueller et al., 2012; Casanas et al., 2016; Pflugrath, 1999) [Fig. 2(a)]. This technological leap facilitated a paradigm shift in data collection strategies, moving away from high-dose, low-redundancy methods toward low-dose, high-redundancy approaches (Weinert et al., 2015; Winter et al., 2019). By the time PILATUS arrived, several automated data processing pipelines were available (Holton & Alber, 2004; Minor et al., 2006). Still, none were optimized to match data processing time to data collection time, which was reduced to a few minutes with PILATUS. We exploited the parallel data processing of XDS (Kabsch, 2010a; Kabsch, 2010b) using high-performance computing clusters and developed the pipeline go.com (unpublished) with simple decision-making approaches. go.com provided fast feedback on key data processing results (i.e. diffraction resolution, completeness, I/σ, CC_1/2, CC_ano and possible space groups) within a few minutes after data collection, proving essential for the era of high-throughput MX. Similar approaches were used elsewhere thereafter (Winter, 2010; Vonrhein et al., 2011; Monaco et al., 2013).

Furthermore, the high frame rates enabled fast and continuous grid scans for diffraction-based centering and micro-crystallography. In the past, diffraction cartography – used to identify optimal diffraction regions within crystals (Bowler et al., 2010) or to locate microcrystals (Cherezov et al., 2009) – was limited by CCD readout times and noise. Continuous grid scanning was first implemented at DLS using a PILATUS detector (12.5 Hz) (Aishima et al., 2010), and faster protocols (100 Hz) were realized with the EIGER detector at SLS (Wojdyla et al., 2016) [Fig. 2(b)]. These advancements were essential for the development of serial synchrotron crystallography (SSX) (Diederichs & Wang, 2017).

Commercialized and further advanced by DECTRIS Ltd, PILATUS and EIGER detectors are now integral to synchrotron facilities worldwide, significantly improving X-ray data quality and experimental throughput. This technology has played a key role in the exponential growth of protein structure determination over the past two decades, also providing essential datasets for training breakthrough tools like AlphaFold2 (Jumper et al., 2021).

To complement photon-counting detectors, the PSI's detector group developed the JUNGFRAU charge-integrating pixel-array detector for applications in both X-ray free-electron laser (XFEL) and synchrotron environments (Mozzanica et al., 2018) [Fig. 2(c)]. Its adaptive gain technology offers single-photon sensitivity and a high dynamic range without count-rate limitations. The JUNGFRAU detector is particularly well suited for MX applications at low energy and high flux due to its high dynamic range, low noise performance and fast readout speed. These features were key to handling high photon rates and improving data accuracy (Leonarski et al., 2018; Chapman, 2018). While JUNGFRAU performs efficiently at the pulsed sources with moderate repetition rates of some XFELs, e.g. 100 Hz in the case of SwissFEL, achieving `continuous' synchrotron operation requires kilohertz frame rates, presenting challenges in data throughput. To address this, the Jungfraujoch system was developed, capable of handling 38 GB s⁻¹ on a single server using field-programmable gate arrays (FPGAs) and general-purpose GPUs (Leonarski et al., 2020; Leonarski, Brückner et al., 2023). Additionally, Jungfraujoch integrates basic crystallographic data analysis, including background integration, spot finding and indexing (Gasparotto et al., 2024), enabling real-time monitoring, analysis and feedback at kilohertz frame rates.

All in all, continuous advancements in X-ray detectors have played an instrumental role in boosting beamline productivity and data quality, enabling new MX applications, including autonomous experiments and driving an exponential increase in data rates [Fig. 2(d)] (Leonarski, Brückner et al., 2023).

3.3. From single-axis to multi-axis goniometer

Several generations of single-axis goniometers were developed in the early days of SLS MX beamlines. We started with a compact design with a stack of two stepping motors (Pauluhn et al., 2011) [Fig. 3(a)], which was replaced by a flexor device later (Fuchs et al., 2014) [Fig. 3(b)]. When high-precision compact nano-positioning technology became available in 2008, we built a new goniometer using two piezo positioners (SmarAct GmbH) [Fig. 3(c)]. In parallel, we have been following the development of multi-axis goniometers at other synchrotron facilities, noticeably the mini-kappa design at the ESRF (Brockhauser et al., 2013). Challenges were the precision required for collecting data from micrometre-sized crystals with micrometre-sized X-ray beams at synchrotron MX beamlines, avoiding collisions with beamline devices and self-shadowing on the detector. We developed the Parallel Robotics Inspired Goniometer (PRIGo), a new type of multi-axis goniometer with micrometre precision, large collision-free angular range and reduced self-shadowing (Waltersperger et al., 2015) [Fig. 3(d)]. Based on a combination of serial and parallel kinematics, PRIGo utilized linear and rotary piezo positioners to emulate the movements of an arc. A calibration procedure was developed to reach the sphere of confusion <1 µm for Ω and <7 µm for χ, respectively. The PRIGo was installed at X06DA-PXIII in 2012.

Figure 3 Evolution of the goniometer at the SLS. (a)–(c) Single-axis goniometers. (d)–(f) Multi-axis goniometers (Waltersperger et al., 2015).

To make the PRIGo technology accessible to beamlines at other facilities, we teamed up with the company SmarAct GmbH (Oldenburg, Germany) to create the next-generation device, SmarGon, in 2015 [Fig. 3(e)]. With enhanced precision and a simplified design for easier construction and calibration, SmarGon has since been deployed at beamlines at both the DLS and the SOLEIL (Source Optimisée de Lumière d'Énergie Intermédiaire du LURE) synchrotrons. Our collaboration with SmarAct continued to improve the system's mechanical robustness, initialization procedures and calibration. We also developed a new control system (smargopolo) using the Robot Operating System. This latest generation, SmarGon-MCS2 [Fig. 3(f)], was successfully deployed at X06SA-PXI and X10SA-PXII in 2021 (Glettig et al., 2024).

3.4. From MAD/SAD to Native-SAD phasing

The MAD and SAD techniques were revolutionary in MX phasing and became the main experimental phasing methods in the 2000s (Hendrickson, 2000; Hendrickson, 2014). The tunability of synchrotron radiation allowed easy access to the absorption edge of phasing elements to maximize both dispersive and anomalous signals. The success of cryogenic cooling made it possible to measure MAD from a single crystal, greatly improving the data accuracy required for the small amount of the anomalous signal. However, working with heavy elements that needed to be incorporated into protein was laborious and not consistently successful. Seleno­methio­nine derivatization later revolutionized de novo structure determination (Hendrickson et al., 1990), significantly reducing the non-isomorphism problem and taking advantage of the convenience of using the 12.66 keV Se K-edge at synchrotrons. New methods, algorithms and powerful programs for data processing (Otwinowski & Minor, 1997; Leslie, 2006; Kabsch, 2010b,a), phasing (de La Fortelle & Bricogne, 1997; Terwilliger & Berendzen, 1997; Schneider & Sheldrick, 2002; Sheldrick, 2010), density modification (Wang, 1985; Zhang & Main, 1990; Terwilliger, 2000; Sheldrick, 2002; Skubák & Pannu, 2011) and automatic model building (Perrakis, Morris & Lamzin, 1999; Cowtan, 2006; Terwilliger et al., 2008; Pannu et al., 2011; Usón & Sheldrick, 2018) gradually made the SAD phasing the first choice. As data accuracy continued to improve, another approach emerged: exploiting the anomalous signal from sulfur atoms in cysteine and me­thio­nine residues, which are natively present in most proteins, known as native-SAD (Hendrickson & Teeter, 1981; Liu et al., 2012). However, since the anomalous sulfur signal is weak in the typical energy range of MX beamlines, it was necessary to increase the phasing signal-to-noise ratio. This involved reducing experimental systematic errors, using lower energy X-rays to enhance anomalous signals, or employing both strategies.

Various approaches have been developed to reduce measurement systematic errors. This included crystal alignment with a multi-axis goniometer to measure the Friedel pairs of reflections on the same diffraction image, `inverse beam' strategy to collect Friedel pairs with similar X-ray dose, multi-orientation data collection with a three-circle goniometer for high true redundancy (Pal et al., 2008) and high-redundancy data collection strategy to improve data precision (Liu, Chen et al., 2011). In addition, Hendrickson demonstrated that multi-crystal averaging effectively reduces systematic errors and enhances signal-to-noise ratio (Liu, Zhang & Hendrickson, 2011) and applied it for de novo native-SAD structure determination (Liu et al., 2012, 2013, 2014). We integrated these ideas by combining the use of the newly developed single-photon-counting detector (DECTRIS PILATUS) and the multi-axis goniometer (PRIGo). We proposed, using 6 keV energy, accessible at most tunable beamlines, a multi-orientation, low-dose, high-redundancy data collection strategy to effectively average out systematic errors by sampling crystal orientations, diffraction geometry and pixel-response variation of X-ray detector in one experiment (Weinert et al., 2015) [Fig. 4(a)]. The method was used to solve the largest native-SAD structure and was used routinely at X06DA-PXIII (Basu, Finke et al., 2019) [Fig. 4(d)].

Figure 4 Instrumentation and methods development for native-SAD phasing. (a) Multi-axis goniometer (PRIGo) and PILATUS 2M detector at beamline X06DA-PXIII. (b) Deep-UV laser shaped crystal (Basu, Olieric et al., 2019). (c) Helium sample environment at beamline BL-1A, PF, Japan. (d) The largest native-SAD structure [reproduced from Weinert et al. (2015)]. (e) Advantage of integrating detector for low-energy native-SAD [reproduced from Leonarski et al. (2018)]. (f) Harvesting anomalous signal at low-enegy with JUNGFRAU detector (unpublished data).

Using X-rays close to the sulfur and phosphor absorption edges for anomalous scattering applications, including native-SAD phasing, has been pioneered by Stuhrmann and coworkers in the 1990s (Lehmann et al., 1993; Stuhrmann et al., 1995; Stuhrmann et al., 1997). However, experimental complications from working at such low energies were not met at the time at MX beamlines. Dedicated beamlines were later constructed to reduce both air and sample absorption and to improve diffraction geometry and detector efficiency. Beamlines I23 at DLS (Wagner et al., 2016) and BL-1A at the Photon Factory (PF) (Liebschner et al., 2016) are two examples. They used either a vacuum or a helium sample environment to reduce air absorption, a kappa-goniometer to improve data completeness at low energy, and specialized detectors (one `cylindrical' PILATUS 12M at I23 and two EIGER 4M in a V-shape configuration at BL-1A) with extra-low-energy calibrations. Both beamlines have been used successfully for native-SAD phasing. Still, at energies below 5 keV, sample absorption becomes significant, making it problematic to perform phasing routinely. Crystal shaping technology with a deep-UV laser developed at SPring-8 came to the rescue and was made available for routine use at BL-1A (Kitano et al., 2005). We used spherically shaped crystals to demonstrate the advantages of low-energy native-SAD using 4.6 keV (2.7 Å) at BL-1A (Basu, Olieric et al., 2019) [Figs. 4(b) and 4(c)].

Even with special calibration of single-photon-counting detectors, the `corner-effect' at low energies remains problematic and limits the data accuracy required for native-SAD. We demonstrated that JUNGFRAU charge-integration eliminated the corner-effect and can produce more accurate data for low-energy phasing experiments (Leonarski et al., 2018) [Fig. 4(e)]. In collaboration with BL-1A, we showed that enhanced anomalous signal at 3.75 keV (3.3 Å) could be harnessed effectively [Fig. 4(f)] by combining crystal-shaping, multi-orientation data collection and a JUNGFRAU 4M in a helium chamber [Fig. 4(c)]. Thanks to the kilohertz frame rate of the JUNGFRAU, 360° data sets could be collected with fine-phi slicing protocol at 100 ° s⁻¹ fast rotation, which made the multi-orientation strategy fast and efficient. We applied this method to solve dozens de novo structures in 2020 (unpublished).

The significance of synchrotron experimental phasing can not be overstated (Hendrickson, 2023), but the success of experimental phasing challenged its own existence. With the advent of accurate protein structure prediction tools like AlphaFold2 (Jumper et al., 2021) and RoseTTAFold (Baek et al., 2021), nearly all structures can now be solved by molecular replacement (Keegan et al., 2024). Fortunately, instruments and methods developed at synchrotron beamlines go beyond experimental phasing. For example, multi-orientation allows improved coverage of reciprocal space for more complete and accurate data for MX structure refinement (Bricogne, 2020) and sampling of real space for X-ray tomography. The higher quality data with fewer systematic errors could help with studies on functional binding and drug design. In addition, it comes in handy when aligning chips and fixed targets for serial crystallography. The kilohertz data acquisition enables millisecond time-resolution in serial time-resolved crystallography (Leonarski, Nan et al., 2023).

3.5. From in situ crystallography to multi-temperature MX

Since the beginning of the 21st century, crystallography at cryogenic temperature has become the standard method at the synchrotron, which increased the X-ray dose limit to two orders of magnitude and greatly facilitated the logistics of transportation of fragile crystals (Garman & Schneider, 1997). Nevertheless, conducting initial diffraction screening directly in the crystallization container proved beneficial for further crystallization optimization and the evaluation of post-crystallization treatments (Martiel et al., 2018). Jean-Luc Ferrer at the FIP beamline at the ESRF pioneered in situ diffraction from 96-well crystallization plates using a six-axis robot (Jacquamet et al., 2004). We followed this idea and made an automated pipeline enabling the robotic transfer of crystallization plate from the storage hotel to the beamline for in situ diffraction screening at X06DA-PXIII in 2010 (Bingel-Erlenmeyer et al., 2011). Users could change the beamline configuration to in situ diffraction from the GUI within 2 min and select crystallization plates from the Formulatrix Rock Imager 1000 or from a supplemental plate holder. A SCARA robot picked up the plate, transferred it to a Stäubli six-axis robot through the back window of the mini-hutch at X06DA-PXIII and the Stäubli presented the selected drop at the sample position for X-ray data collection [Fig. 5(a)]. This approach was later elaborated into a dedicated in situ diffraction screening beamline at the DLS (VMXi) reaching a much larger scale and higher automation (Sanchez-Weatherby et al., 2019; Mikolajek et al., 2023). To facilitate in situ data collection from smaller crystals, various miniaturized devices were developed with thin materials, including silicon films (Zarrine-Afsar et al., 2012; Mueller et al., 2015; Roedig et al., 2016; Dunge et al., 2024), silicon nitride windows (Coquelle et al., 2015), polymers (Axford et al., 2016; Baxter et al., 2016; Schubert et al., 2016; Guo et al., 2018; Doak et al., 2018; Cipriani et al., 2012) and graphene (Sui et al., 2016).

Figure 5 In situ crystallography: from RT to cryo and back. (a) In situ screening of a 96-well plate [reprinted with permission from Bingel-Erlenmeyer et al. (2011). Copyright (2011) American Chemical Society]. (b) In meso in situ serial crystallography (Huang et al., 2016). (c) Apply IMISX chip for multi-temperature crystallography (Huang, Aumonier et al., 2024).

In situ crystallography for membrane protein crystals grown in lipidic cubic phases (LCP) in glass plates was another challenge due to the size of microcrystals and high diffraction background. The glass plate and viscosity of the LCP made crystal harvesting challenging. In collaboration with Caffrey, we developed a sandwich crystallization setup with two thin COC films, which allowed harvesting the whole LCP bolus containing microcrystals for in situ serial data collection (Huang et al., 2015). The in meso in situ serial crystallography (IMISX) chips can be prepared with standard LCP crystallization robots. One advantage of the compact format of IMISX chip is that the whole chip can be cryo-cooled, allowing preparation of samples at users' laboratories and sending them in a dryshipper (Dewar) for serial X-ray data collection at synchrotron beamlines (Huang et al., 2016). The IMISX chip is compatible with most sample changers, enabling integration into automation workflows at MX beamlines [Fig. 5(b)], and the IMISX kit is commercially available via MiTeGen. Similar ideas have been pursued to improve throughput (Broecker et al., 2018; Huang, Meier et al., 2020) and automation (Felisaz et al., 2019; Healey et al., 2021).

Although the IMISX method was primarily developed for the determination of membrane protein structures (El Ghachi et al., 2018; Apel et al., 2019; Jaeger et al., 2019; Olatunji et al., 2021; Li et al., 2021), it provided a general and adaptable platform for studying structures from cryogenic temperature to room temperature (RT) at standard MX beamlines (Huang, Olieric et al., 2020). Easy access to multiple temperatures holds great potential for the study of dynamic processes (Douzou et al., 1970; Horrell et al., 2018; Yao et al., 2021; Tsai et al., 2022; Huang et al., 2022; Greisman et al., 2024; McLeod et al., 2025). Recently, we have used it to reveal the changes in ligand binding of endothia­pepsin at multiple temperatures (Huang, Aumonier et al., 2024) [Fig. 5(c)].

Another innovative approach to RT MX uses ultrasonic acoustic levitation to suspend liquid droplets containing protein crystals. The rapid spinning of the crystal within the levitating droplet enables efficient sampling of reciprocal space, while a fast frame-rate X-ray detector captures diffraction images in a manner similar to serial crystallography (Tsujino & Tomizaki, 2016). The method was later extended to levitate thin films as sample holders (Kepa et al., 2022) and holds promise for studying dynamics through droplet mixing.

3.6. From micro-crystallography to serial crystallography

Pioneered at the ESRF in the 1990s (Cusack et al., 1998), protein micro-crystallography became instrumental in the structure determination of crystals <20 µm in size, such as those of G-protein coupled receptors (GPCRs) (Smith et al., 2012). The micro-focusing capability was one unique feature at X06SA-PXI. The 5 µm focused beam allowed de novo structure determination of polyhedra from a few micrometre-sized crystals in 2007 (Coulibaly et al., 2007) [Fig. 6(a)] and of microcrystalline insulin (Wagner et al., 2009). The high flux density reduced the crystal lifetime to a few seconds due to radiation damage (Owen et al., 2006; Holton, 2009). Therefore, multi-crystal merging was routinely used to obtain a complete data set (Coulibaly et al., 2009). Soon after, the micro-beam feature was introduced at X10SA-PXII to meet the industry's demand for membrane protein drug discovery projects.

Figure 6 From micro-crystallography to serial crystallography. (a) The first polyhedra structure was solved by experimental phasing with multiple 5–10 micrometre-sized crystals (Coulibaly et al., 2007). (b) High-quality data from serial rotation crystallography for native-SAD phasing of PepT (Huang et al., 2018). (c) The first demonstration of RT structure determination with SFX-like serial still crystallography at a synchrotron [reproduced from Botha et al. (2015)]. (d) Demonstration of synchrotron serial crystallography in native-SAD phasing of A2A (Weinert et al., 2017).

It quickly became critical that automation was necessary to identify and center micro-crystals. This led to the development of fast grid scans, which became indispensable in locating micro-crystals grown in LCP, as LCP turns opaque upon cooling. The diffraction-based grid scan was first reported at the Stanford Synchrotron Radiation Lightsource (SSRL) in 2007 (Song et al., 2007). Similar implementations with small beams were realized at DLS (Aishima et al., 2010), ESRF (Bowler et al., 2010) and APS (Cherezov et al., 2009). Using an EIGER detector, we achieved 100 Hz grid scan with real-time diffraction hit analysis using a combination of continuous 2D scan, DISTL spot finding (Zhang et al., 2006) and our DA+ software suite that uses messaging and streaming technologies (Wojdyla et al., 2016). We also explored X-ray imaging-based crystal identification methods, namely scanning transmission X-ray microscope and full-field X-ray imaging (Martiel, Huang et al., 2020). Alternative methods based on UV fluorescence (Stepanov et al., 2011) and SONNIC were developed (Calero et al., 2014; Madden et al., 2013). These methods could locate micrometre-sized crystals with zero or near-zero X-ray doses, but they did not provide information on X-ray diffraction quality and required additional instruments. Therefore, we focused on the X-ray diffraction-based method and automated serial rotation crystallography by collecting a small wedge of data (typically 10°) from each crystal (CY+) (Basu, Kaminski et al., 2019), similar to MeshAndCollect at ESRF (Zander et al., 2015) or the ZOO method at SPring-8 (Hirata et al., 2019). We further developed an automated data processing and merging pipeline to process each data wedge separately, select isomorphous data sets, and merge them until the desired data quality was reached (Basu, Kaminski et al., 2019). The CY+ GUI and automation in serial data processing made serial rotation crystallography more accessible to our user community. A simple and deterministic data-scaling and selection method was later developed with Diederichs, particularly effective for experimental phasing by anomalous diffraction (Assmann et al., 2020). Unlike conventional crystallography, which uses one single crystal, serial crystallography consumes more samples. Still, averaging can minimize systematic experimental errors effectively and produce highly accurate data for the most challenging experimental phasing experiment, native-SAD (Huang et al., 2018) [Fig. 6(b)]. The same is true for detecting weak binding ligands (Pearce et al., 2017) and extracting excited states from time-resolved crystallography data (Ursby & Bourgeois, 1997; Genick, 2007).

Following the success of SFX at XFEL facilities in the 2010s (Boutet et al., 2019), synchrotron facilities embraced the new technology (Stellato et al., 2014; Henkel & Oberthür, 2024; Gati et al., 2014). They incorporated innovations in serial sample delivery (Sierra et al., 2018), the measurement of still diffraction images, and novel data processing and merging techniques (White et al., 2012; Sauter et al., 2014). In collaboration with our beamline partner, the Max Planck Institute, we demonstrated SSX using serial sample delivery with the high-viscosity extrusion injector, micro-focused X-ray beam and high frame-rate detector PILATUS 6M at beamline X10SA-PXII (Botha et al., 2015). We showed that high-quality RT data can be obtained and alluded to the possibility of studying protein structural dynamics using SSX [Fig. 6(c)]. A similar demonstration was conducted with a CCD detector at ESRF (Nogly et al., 2015). Later, we applied the method to several proteins and made the method routine with a faster EIGER detector at beamline X06SA-PXI (Weinert et al., 2017) [Fig. 6(d)]. The method was further enhanced by utilizing wide-bandpass or polychromatic (`pink') beams to improve data quality and reduce sample consumption (Meents et al., 2017; Martin-Garcia et al., 2019; Tolstikova et al., 2019).

3.7. Time-resolved serial crystallography at SwissFEL and SLS

SFX revived time-resolved MX and pushed time-resolution to femtoseconds (Moffat & Lattman, 2023). The first SwissFEL experimental station Alvra offered SFX with injector-based sample delivery methods and a JUNGFRAU 16M detector (Milne et al., 2017) [Fig. 7(a)]. Automated experiment logging and data processing improved the efficiency and feedback of SFX. Alvra attracted internal and external expert user groups and established itself as a productive SFX facility. Highlights include the dynamics and mechanism of a light-driven sodium pump (Skopintsev et al., 2020) and a chloride pump (Mous et al., 2022), the drug release from tubulin (Wranik et al., 2023), DNA repair process (Maestre-Reyna et al., 2023; Christou et al., 2023), and the first molecular events of vision (Gruhl et al., 2023). At the Bernina experimental station, a dedicated instrument (SwissMX), including the robotic sample changer TELL (Martiel, Buntschu et al., 2020), was developed to provide SFX with fixed-target sample delivery (Ingold et al., 2019). The SwissMX was further developed at the Cristallina experimental station soon after, aiming to increase user experiment capacity with automated fixed-target approaches [Fig. 7(b)]. The first fixed-target pump–probe experiment has been published recently (Gotthard, Flores-Ibarra et al., 2024).

Figure 7 Time-resolved crystallography at (a) SwissFEL Alvra [reproduced from Mous et al. (2022), reprinted with permission from AAAS], (b) SwissFEL Cristallina (Gotthard, Flores-Ibarra et al., 2024), (c) SLS X06SA-PXI [reproduced from Weinert et al. (2019), reprinted with permission from AAAS] and (d) MAX-IV MicroMAX (Leonarski, Nan et al., 2023).

In collaboration with Standfuss's group at PSI, the synergy between SwissFEL and SLS was effectively leveraged to develop time-resolved serial synchrotron crystallography (TR-SSX) at X06SA-PXI. The millisecond time resolution enabled us to capture large conformational changes during the pumping cycle of bacteriorhodopsin (Weinert et al., 2019) [Fig. 7(c)] and the reaction of a blue light photoreceptor domain (Gotthard, Mous et al., 2024). The TR-SSX data also complemented the SFX data in the studies of a light-driven chloride pump (Mous et al., 2022) and drug release mechanism from tubulin (Wranik et al., 2023).

The development of the Jungfraujoch kilohertz data-acquisition system (Leonarski, Brückner et al., 2023) with the JUNGFRAU detector also enabled probing of multiple time points from microseconds to seconds in one experiment sequence at synchrotron sources. In collaboration with the MicroMAX team at the first fourth-generation synchrotron, MAX-IV, we showed that a static serial data set could be obtained within minutes and that a 1 ms resolution was achieved (Leonarski, Nan et al., 2023) [Fig. 7(d)]. In addition, JUNGFRAU can be operated in burst mode to reach microsecond-level time resolution (Sikorski et al., 2023).

3.8. MX beamline automation, high-throughput screening and unattended beamline operation

Beamline automation as a global effort had a profound impact on modern synchrotron crystallography (Arzt et al., 2005; Soltis et al., 2008). MX beamline operations have evolved from manual sample mounting to robotic sample exchange, from on-site data collection to remote operation, from human oversight to fully unattended operation.

The cryogenic crystallography workflow made robotic sample exchange routinely attainable, leading to the development of sample changers at most synchrotrons over the last two decades (Muchmore et al., 2000; Cohen et al., 2002; Ohana et al., 2004; Cork et al., 2006; Ueno et al., 2004; Arzt et al., 2005; Cipriani et al., 2006; Papp et al., 2017; O'Hea et al., 2018). We started with the Cryogenic Automated Transfer System (CATS) system (Ohana et al., 2004) for its versatility. Indeed, the system offered wet-mounting, dry-mounting and in situ plate screening capabilities (Jacquamet et al., 2004). Later, inspired by the DLS BART system (O'Hea et al., 2018), we developed high-throughput enabling large-capacity sample loader (TELL) for SwissFEL and SLS with a large capacity dewar holding 480 samples in UniPuck format (Martiel, Buntschu et al., 2020). The SUNA gripper developed at Deutsches Elektronen-Synchrotron (DESY) was later replaced with our gripper based on the original ALS design (Cork et al., 2006). This in-house development improved reliability, speed and compatibility for both cryogenic and RT sample exchange.

Thanks to automation, the remote operation of our MX beamlines steadily increased until the start of the COVID-19 pandemic in 2020. The SLS MX beamlines never stopped operation throughout the pandemic and offered beam time to academics and industry with dedicated fast access for drug discovery programs against COVID-19 (Shin et al., 2020; Gao, Qin et al., 2021; Gao, Zhu et al., 2021; Qin et al., 2023; Sutanto et al., 2021; Huang, Metz et al., 2024).

The sophistication of automation and continuous improvement of fast X-ray detectors gradually increased the throughput so that hundreds of data sets could be collected daily. This throughput gain was transformative and made crystallographic fragment-based screening a reality (Davies & Tickle, 2012). The XChem team at the DLS pioneered this work with streamlined processes, from sample preparation to data management and analysis. They built a full X-ray screening facility at beamline I04-1 in 2015 (Fearon et al., 2024). Similar facilities were constructed at other synchrotrons (Lima et al., 2020; Wollenhaupt et al., 2021; Cornaciu et al., 2021; Barthel et al., 2024; Huang et al., 2025). The SLS fast fragment and compound screening pipeline (FFCS) was built in 2020, focusing on industrial applications (Kaminski et al., 2022; Stegmann et al., 2023). Our service includes the complete pipeline from crystallization, soaking, crystal harvesting to X-ray data collection and analysis.

With the increasing throughput, unattended beamline operation became a highly valued mode of operation by users. A fully automated pipeline from crystal mounting to structure refinement was pioneered for Eli Lilly and Company at LRL-CAT, APS (Wasserman et al., 2012). MASSIF-1 was developed to automate characterization and data collection at the ESRF, taking crystal size, flux and X-ray dose into account (Bowler et al., 2015). At SPring-8, the automatic data collection was optimized for micro-crystallography (Hirata et al., 2019). Our implementation focused on higher throughput for industrial applications. The combination of the fast TELL sample changer, rapid grid scan and short data collection allowed us to process 25 samples per hour before the SLS 2.0 upgrade (Smith et al., 2023). Current development focuses on further improving throughput, making it available for RT data collection and extending the automation to SSX applications.

3.9. Impact beyond MX

Advancements in beamline instrumentation and automation have the potential to influence fields beyond MX. For example, small-angle X-ray scattering tensor tomography (SAS-TT) is a powerful technique for studying the multiscale architecture of hierarchical structures (Liebi et al., 2015). SAS-TT measurements typically require a few hundred 2D SAS scans at various sample orientations, which is often a manual and lengthy process. To make SAS-TT accessible to a broader user base, we automated the complete SAS-TT tomogram sampling and reduced data acquisition time by leveraging the precision of the SmarGon multi-axis goniometer and the speed of 2D scanning at the X06SA-PXI beamline (Appel et al., 2024). An experiment that used to take up to four days was realized in 1.2 h. In addition, the cryogenic MX setup allows automatic sample exchange and reduces radiation damage for SAS-TT applications in life science. The automation expertise and productivity enhancements developed in MX could be extended to other synchrotron beamlines, including tomography (Albers et al., 2024), coherent diffraction imaging and spectroscopy.

4. Future development and SLS 2.0 upgrade

In the last decade, we were excited to see revolutionary extensions of the structural biology toolbox, from MX, SAXS and NMR to cryoEM (Kühlbrandt, 2014), cryoET (Turk & Baumeister, 2020), microED (Nannenga et al., 2014), X-ray bio-imaging (Albers et al., 2024) and accurate structure prediction (Jumper et al., 2021), allowing comprehensive multi-scale investigation of a wide range of biological samples. Except for NMR, other experimental techniques primarily capture static structures at cryogenic temperature. As for the atomic resolution technique, 190000 X-ray static structures (https://biosync.rcsb.org) have been determined since the first protein diffraction at SSRL 50 years ago (Phillips et al., 1976) thanks to bright synchrotron lights, routine experimental phasing, fast X-ray detectors, and automation in laboratories and beamlines. The cryogenic MX will continue to routinely provide high-resolution structures with even higher throughput.

High-resolution cryogenic structures are the cornerstone for investigating the functions of biomolecules. However, this static perspective can overlook the subtle yet critical conformational changes and dynamic movements that proteins undergo during catalysis, ligand binding and allosteric regulation. That also, to some extent, limited modeling and molecular dynamics simulation due to the lack of high-resolution structural data at physiological temperatures and along energy landscapes. Because AlphaFold2/3 was trained predominantly on cryogenic temperatures, it also faces limitations in capturing the dynamic molecular interactions vital for many biological processes and identifying potential therapeutic compounds (Agarwal & McShan, 2024). Therefore, it is essential to determine 3D structures at biologically relevant temperatures and time scales to gain a deeper understanding of protein function and dynamics (Henzler-Wildman & Kern, 2007; van den Bedem & Fraser, 2015; Nam & Wolf-Watz, 2023). Integration of `dynamic' experimental structures, AI-based structure prediction, modeling and molecular dynamics simulation will make protein design and structure-based drug discovery more accurate, efficient and accessible (Reardon, 2024).

After mastering cryogenic MX at third-generation synchrotrons, one next grand challenge is whether fourth-generation synchrotrons and XFEL facilities can offer cryocrystallography-like user-friendliness and throughput to emerging dynamic MX applications. Impressive development has already been accomplished (Henkel & Oberthür, 2024); dedicated on-site sample preparation laboratories were built at and near synchrotron and XFEL facilities (Han et al., 2021), and specialized crystal transportation and crystallization setups were developed for RT samples (Baxter et al., 2016). Such RT sample preparation facilities and sample transportation methods are certainly needed. Still, we would reason that the true potential of synchrotron and XFEL RT MX will be severely limited if the sample preparation and X-ray data collection are coupled in time and space. Then, the success of beam time will be heavily dependent on the timing of delicate and often not very reproducible crystal growth and the capacity of sample laboratories at synchrotron facilities. We foresee active developments to improve the productivity of dynamic MX applications at synchrotron and XFEL facilities in the coming decade. This calls for corresponding development in data analysis tools as well. Finally, we encourage the user community to explore MX beyond static cryogenic structures. Our recent developments in the context of the SLS 2.0 upgrade are briefly described below.

4.1. From cryogenic temperature back to RT

Despite the recent revival of RT MX and its appealing applications in protein dynamics and drug discovery (Fischer, 2021; Thorne, 2023), sample logistics severely limit its broad adoption. Indeed, transporting delicate crystals at RT from the research laboratories to the synchrotron facilities is challenging, and the synchronization between sample preparation and beam time presents another obstacle. In contrast, the sample logistics problem has been solved in cryo-crystallography, which enabled users worldwide to optimize their samples, harvest them and preserve them in optimal conditions ahead of MX measurements. These samples are stored and shipped to the facility when beam time becomes available. This cryogenic workflow has provided a continuous supply of samples to the facilities, enabling the exponential growth of experimental structures determined at synchrotron facilities (https://biosync.rcsb.org/). To address the logistics issues at RT, we recently proposed a Cryo2RT workflow to determine high-quality RT structures from previously cryocooled crystals (Huang, Aumonier et al., 2024). This seemingly unconventional method leverages the advantages of conventional cryogenic MX, allowing us to circumvent the RT sample logistics problems by incorporating the proven cryogenic sample preparation, transportation and beamline automation workflow. Since nearly all crystals can be cryo-cooled, the reverse process could also be applicable in general (Kriminski et al., 2002; Juers & Matthews, 2001; Juers et al., 2007). The method can be readily adapted to standard MX beamlines with a humidity control device, e.g. HC-Lab (https://www.arinax.com) and Watershed (https://www.mitegen.com).

Conventional cryogenic crystallography has distinct advantages, including high resolution and sensitivity to weak binders, but cryo-cooling often drives the system to one low-temperature state. Recent studies show that catalytic efficiency is related to sampling multiple conformational states, challenging the traditional static view of enzyme catalysis (Yabukarski et al., 2022). The temperature-dependent binding modes and conformations could be accessible with RT crystallography (Fraser et al., 2009; Weik & Colletier, 2010; Fraser et al., 2011; Fischer et al., 2015; Keedy et al., 2018; Greisman et al., 2024), but the RT experiment has other challenges. The foremost is reproducibility – RT structures are affected by the crystallization conditions, the crystal harvesting conditions and the potential structure changes during X-ray data collection, such as dehydration, rehydration, X-ray beam heating, radiation damage etc. The experimental conditions must be carefully controlled and varied systematically to capture relevant structure ensembles.

Radiation damage is another challenge. The femtosecond pulses provided by XFEL sources enabled high-resolution structure determination without radiation damage by the `diffraction before destruction' principle in SFX. The SFX damage-free structures serve as the `ground truth' (Hirata et al., 2014; Williams et al., 2025) and are particularly important in studying metalloproteins (Kern et al., 2015; Bowman et al., 2016; Hirata et al., 2014; Suga et al., 2015). At synchrotrons, however, a balance between reaching high resolution and minimizing X-ray dose needs to be found. For example, multi-crystal and serial crystallography could increase attainable diffraction resolution from small crystals by distributing X-ray dose (de la Mora et al., 2020). Unlike X-ray damage at 100 K, radiation damage at RT is both dose and dose-rate dependent (Southworth-Davies et al., 2007; Rajendran et al., 2011; Owen et al., 2012; Warkentin et al., 2012, 2011). The radiation damage is also temperature and time dependent, and the damage mechanism is less understood (Warkentin et al., 2013). We have been using fast frame-rate X-ray detectors to track radiation damage in millisecond timescales at RT (Rajendran et al., 2011; Huang et al., 2015). The recent advances in kilohertz MX with JUNGFRAU detector (Tolstikova et al., 2019; Leonarski, Nan et al., 2023; Leonarski, Brückner et al., 2023) could even `outrun' slow radiation damage processes (Warkentin et al., 2013; Thorne, 2023). With increased flux at higher energy in fourth-generation synchrotrons and X-ray detectors with high-Z sensors, high-energy MX has the potential to further reduce radiation damage (Dickerson & Garman, 2019) and improve diffraction resolution (Jaho et al., 2024). Note that the `low reproducibility' of RT MX experiments could complicate the situation, calling for more systematic and comprehensive studies on dynamic radiation damage processes and their impact on RT structures.

Finally, analysis and interpretation of RT structures are not part of the standard MX practices (Woldeyes et al., 2014). The corresponding data processing, refinement (Burnley et al., 2012; Du et al., 2023), modeling (Riley et al., 2021) and visualization pipelines need to be developed and deployed at future MX beamlines.

4.2. New dimensions in macromolecular crystallography

After 100 years of structural biology research, static 3D structures of proteins can be determined to high resolution and predicted with high accuracy. The next forefront is dynamic structural biology to understand the intricate behaviors of proteins beyond static views, uncover how molecular conformations shift in response to environmental factors (e.g. temperature, pH), binding events or catalytic cycles in a time-resolved manner. Temperature, binding pose and time are readily accessible in MX. A temperature range from glass transition to physiological temperatures (∼200–300 K) can be controlled with standard cryojets, which were previously used primarily to cool down crystals to 100 K. Various small format fixed-targets developed for SSX can be used directly for multi-temperature MX data collection. This large temperature window enables us to probe the thermodynamics and kinetics of protein dynamics, changes in water structures and ligand interactions at atomic resolution (Rasmussen et al., 1992; Tilton et al., 1992; Horrell et al., 2018; Ringe & Petsko, 2003; Greisman et al., 2024). For example, a 10–20°C temperature change can alter ligand binding (Huang et al., 2022; Du et al., 2023). This small temperature change is within the reach of X-ray induced beam heating (Kriminski et al., 2003; Warren et al., 2019; Baxter et al., 2024), raising concerns and calling for careful experimental control from X-ray beam characteristics (size, profile, flux) to sample preparations.

Temperature-dependent and time-resolved MX will enable us to study dynamic structural enzymology (Douzou et al., 1970; Tilton et al., 1992; Fraser et al., 2009; Bhabha et al., 2015; Beyerlein et al., 2017; Kupitz et al., 2017; Keedy et al., 2018; Bradford et al., 2021; Yao et al., 2021; Stachowski et al., 2022; Greisman et al., 2024; McLeod et al., 2025; Banari et al., 2025). By controlling the diffusion process and altering the reaction rate at different temperatures, we could capture reaction mechanisms and dynamics of conformational changes (Tsai et al., 2022). Advanced temperature-jump techniques using infrared lasers could probe relaxation processes from micro- to millisecond resolution utilizing fast-frame-rate X-ray detectors (Wolff et al., 2023). Alternatively, fast laser heating combined with re-vitrification can trap dynamics in microsecond resolutions, as shown in time-resolved cryo-electron microscopy (Lorenz, 2024). This innovative microsecond cryo-trapping techniques could be adapted to MX at synchrotron beamlines.

Novel instrumentations have been explored to control binding events by mixing substrate/ligand and protein crystals. Mix-and-extrude with microchannels and micronozzles (Vakili et al., 2023), liquid application method with fixed-target (Mehrabi et al., 2023), acoustic-based drop-on-drop (Fuller et al., 2017) and levitation methods (Tsujino & Tomizaki, 2016; Kepa et al., 2022) have been developed to control the initiation of the reaction within the crystals and capture intermediate states that occur as the enzyme catalyzes its reaction in millisecond to second time scales. However, most methods require specialized instruments and significant expertise. To advance this area of research, these techniques must be democratized and made available to a broader range of researchers. Achieving this involves developing standardized protocols, user-friendly software and automated systems that can handle the complexity of these experiments on standard MX beamlines.

4.3. MX opportunities at the SLS 2.0 upgrade and SwissFEL

The SLS 2.0 upgrade is currently underway (Willmott & Braun, 2024). We have redesigned our three MX beamlines to exploit the brighter source and address MX's future needs (Fig. 8). We strive to increase throughput for conventional cryogenic MX further, make RT MX routine and explore new dimensions in MX. Fully unattended beamline operation will be offered to academic and industry users, while hands-on, on-site support will be provided to explore new opportunities. There will be a paradigm shift in the usage of beam time. In the past, a significant part of beam time was dedicated to solving new structures of large molecular complexes and membrane proteins by screening weakly diffracting crystals and their derivatives. These demands have been moderated by the recent breakthroughs in cryoEM (Kühlbrandt, 2014) and AlphaFold2 (Jumper et al., 2021). In the future, more beam time will be available to study structures and their dynamics.

Figure 8 SLS 2.0 upgrade and MX applications. (a) Synergies among three MX beamlines at SLS 2.0 and two end-stations at SwissFEL. (b) MX applications.

The higher machine energy (2.7 GeV), low source emittance of 157 × 10 pm rad, and advances in undulator technology, monochromator, and X-ray mirrors and refractive optics make it easier to expand X-ray characteristics for MX beamlines. The X-ray optics design has evolved accordingly (Fig. 9). At the SLS, previous X06SA-PXI and X10SA-PXII started with direct focusing, and the beam-defining apertures near the sample position were used to make micro-beams at the cost of flux. Later, both beamlines were changed to two-stage focusing for more flexible beam size and divergence control [Fig. 9(a)]. With the low emittance source of the SLS 2.0, two-stage focusing is no longer necessary for undulator beamlines X06SA-PXI and X10SA-PXII. A secondary optics element could serve to tune the primary focusing. For a bending magnet source, where the large acceptance angle in the horizontal plane and beam collimating in the vertical plane are required, two-stage optics is advantageous. Thanks to the much-reduced source size at the SLS 2.0, moderate focusing is sufficient to achieve a micro-beam with less beam aberration and maintain an acceptable beam divergence for the new X06DA-PXIII [Fig. 9(b)].

Figure 9 Evolution of X-ray optics design of MX beamlines from SLS to SLS 2.0. (a) The low emittance source favors direct focusing at undulator sources. (b) Small source size enables variable focusing with acceptable beam divergence at the bending magnet source. The illustrations are from top views.

The new X06SA-PXI and X10SA-PXII beamlines will deliver micro-focused low-divergence X-rays with a monochromatic flux >10¹³ photons s⁻¹ at 12.4 keV. The beam size can be varied from 1 to 100 µm with KB mirrors to match the crystal size. In addition, we will use two sets of 2D beryllium compound refractive lenses for quick beam resizing for industry applications at X10SA-PXII. The new bending magnet X06DA-PXIII beamline combines a toroidal prefocusing/collimating mirror and a KB-focusing mirror system via a horizontal secondary source to achieve a variable beam size from 15 to 100 µm with flux of 10¹² photons s⁻¹ at 12.4 keV. The X-ray bandwidth and energy range will be extended to 1% at PXI, 30 keV at PXI and PXII, and 4 keV at PXIII. The main beamline characteristics are listed in Table 1. All three new beamlines can perform conventional MX, each with its own specific focus: PXI for dynamics and X-ray imaging, PXII for industry applications, PXIII for autonomous operation [Fig. 8(a)].

We are developing a new MX software suite to accelerate and advance MX research at the new beamlines at the SLS 2.0. The suite integrates beamline device control; orchestrates MX experiments; accelerates data acquisition; assists structure determination and interpretation; facilitates information flow for experimental feedback; and interfaces to beamline developers, operations and users. We aim to achieve `intelligent' beamline operation with real-time data analysis, machine-learning-based experimental feedback and steering to enable autonomous experiments for routine and advanced MX applications.

Increases in source brightness, advances in detector technology and sample delivery methods, and expansion in MX experimental techniques also lead to a formidable data challenge. For example, kilohertz data acquisition will be routine for rotation and serial crystallography to enhance throughput and capture dynamic processes. Innovative and sustainable IT solutions are needed to release the full potential of future light sources. We will address such challenges by exploiting edge computing, utilizing high-performance computing (HPC) infrastructure at PSI, and exploring cloud-based services. Recently, we have enabled kilohertz data analysis in Jungfraujoch so that initial image analysis, compression and reduction could be carried out for each image before data are written to storage. With ever-powerful GPU and FPGA technologies and available AI/ML data analysis methods, data-driven experimental feedback and steering could be implemented in the future. The pre-processed data will feed local HPC or cloud service for complete analysis and optimization. We expect disruptive innovative data analysis methods will be developed for future experiments.

Three MX beamlines at the SLS 2.0 and two endstations at SwissFEL will offer our user community a comprehensive set of high-resolution structure techniques, from high-throughput applications at cryogenic and around RT to damage-free and time-resolved structures [Fig. 8(b)]. The enriched experimental structures could advance physics-based molecular dynamics simulations and AI-driven structure prediction methods in structural biology. This will expand our knowledge of protein structure and function and ultimately accelerate de novo protein design, discoveries in drug development and beyond. After the 10000 static structures at the SLS, we are at a new beginning with SLS 2.0 and SwissFEL.