2.1. Design choices

The mixer provides a dual-inlet section that accepts two capillaries and allows the convergence of two fluid channels via overlapping concentric cones [Fig. 1(a)]. At the start of the mixing channel, the main to side channel diameter ratio is 100:231.7 µm. Due to the centred sample inlet and the 3D hydro­dynamic flow-focusing geometry, a wall contact of the sample is initially (throughout the 500 µm in the downstream direction) prevented. The total length of the mixing channel is 2570 µm (design `J_7'). From this length, the initial 2070 µm have a 231.7 µm channel width, followed by a 300 µm-long tapering section [truncated cone which reduces the inner diameter (ID) down to 75 µm], leading to a 200 µm-long final section with a 75 µm inner diameter. For the geometry at the tip, the ID–OD–D (liquid channel diameter, gas orifice and distance between orifices) are chosen to be 75, 345 and 600 µm; therefore, 75 µm-wide streams are provided for the X-ray beam.

Figure 1 (a) Cross-sectional schematic of the 2PP-3D-printed mix-and-extrude device. (b)–(e) Microscopy images showing the assembly of the mixing-HVE tip with two fluid-feeding fused silica capillaries. (b) Two capillaries, each with 250 µm ID, are inserted into the 3D-printed mixing-HVE access ports and (c) glued with ep­oxy glue. (d) The capillaries (un-glued end) are fiddled through a 0.046′′ ID steel tubing (OD = 1/16′′), which is pulled over the mixing section of the device. (e) A small amount of slightly viscous ep­oxy glue (cured for 1 min) is added onto the gap with a clean capillary or metal wire to secure IP-S to the steel and provide gas tightness.

Inside the first section of the 231.7 µm-wide mixing channel, a modified mixing structure based on the `JKMH#10' Kenics mixer from Knoška et al. (2020) is incorporated. With this, a series of six helical elements are introduced to the mixing channel for repeated flow splitting/stretching. The first blade is positioned 500 µm after the mixing initiation point (overlap of the liquid apertures).

With the 2570 µm-long mixing channel and a combined liquid flow rate of Q_total = 1.46 µl min⁻¹ (i.e. 1.43 µl min⁻¹ for the reactant and 0.03 µl min⁻¹ for the sample), a retention time of 4.1 s before extrusion can be achieved within the 3D-printed part. This corresponds to a stream velocity of v_flow = 5.5 mm s⁻¹ for the 75 µm-wide sample stream. Longer retention times require lower flow rates. For instance, as used during the diffraction data collection described below, a retention time of 18.2 s can be obtained with Q_total = 0.36 µl min⁻¹ (exposed sample has a flow velocity v_flow = 1.3 mm s⁻¹). Another design variation (`J_8') contains a shorter mixing channel (total length: 1685 µm). With the aforementioned flow rates, this shorter mixing-HVE allows retention times between 2.3 and 9.9 s. All 3D CAD files can be found in our online design repository: https://github.com/flmiot/EuXFEL-designs.

At the European XFEL, the X-ray pulses can arrive at 10 Hz, i.e. one pulse per train. From previous studies, it is known that flow velocities as low as v = 0.3 mm s⁻¹ are sufficient for stable viscous sample extrusion while maintaining a pulse displacement of ca 30 µm on the sample and, as a result, effectively avoiding exposure of the same crystal with multiple X-ray pulses (Vakili et al., 2022). Moreover, sample delivery to X-rays that arrive at higher pulse repetition rates, as seen at the FELs SACLA (60 Hz), PAL-XFEL (60 Hz), SwissFEL (100 Hz) and LCLS (120 Hz), can also be accommodated (Shimazu et al., 2019; Lee et al., 2020; Milne et al., 2017; Wells et al., 2022). However, for ≥100 Hz operation, a sample width reduction from 75 to 50 µm is highly recommended. At constant flow rates, this will increase the flow velocity by a factor of 2.2 which, in turn, facilitates viscous extrusion without the need for higher pump pressures.

2.2. Beam time injection setup

The SPB/SFX instrument of the European XFEL comprises two interaction regions: a high-vacuum upstream sample environment (interaction region upstream, IRU) and an in-helium downstream interaction region (interaction region downstream, IRD) with respect to the X-ray beam. The instrument setup at IRD provides high flexibility on sample deliveries, including the HVE injection setup. The injection rod funnel is attached to a helium atmosphere sample chamber, located between the vacuum out-coupling acoustic delay line (ADL) and detector (JUNGFRAU 4M), and the inserted nozzle rod is positioned using the XYZ motors of the goniometer support tower (Round et al., in preparation).

The assembled nozzle, connected to capillaries and 1/16′′ OD steel tubing (IDEX, part No. U-145), was connected to the nozzle holder using an F333N fitting (IDEX) (Weierstall, 2014). A slit on the insertion rod [Fig. S1(b)] was created 10 cm above the o-ring to shorten the sample capillary length. The nozzle holder was screwed onto the metal rod and the rod was inserted into the IRD sample chamber, and then the capillaries were carefully taken through the slit and linked to the sample reservoirs of the HVEs using 0.015–0.0625′′ ID polyether ether ketone tubing and a custom No. 10–32 UNF steel fitting with a 1/16′′ through-hole (similar to part No. F-354, IDEX). ThorLab 0.5′′ mounting rods were used to mount the two HVE systems on the insertion rod to bridge the short distance to the nozzle [Fig. 2(a)]. The final capillary length was 30 cm, with an interior volume of 14.7 µl.

Figure 2 (a) Beamline injection setup surrounding the mixing-HVE device inside the helium-purged sample chamber at the SPB/SFX instrument and (b) in the laboratory test station during in-air operation.

2.3. Sample preparation

The preparation of the iq-mEmerald protein and crystal is described in the supporting information, and the following procedures were included to prepare samples suitable for HVE injection. LCP was prepared with a 7:3 ratio of monoolein to water using two gas-tight glass syringes and a coupler. To embed iq-mEmerald crystals (ca 5 × 15 µm) and CuCl₂ in LCP, a 10%(v/v) crystal pellet and 20 mM CuCl₂ solution were mixed with prepared LCP separately. To investigate the mixing inside the nozzle, iq-mEmerald protein embedded in LCP was used instead of protein crystals to provide a continuous fluorescence signal; a 65:35 ratio of monoolein and 10 mg ml⁻¹ of iq-mEmerald protein in 50 mM Tris (pH 8.0) or 10 mM CuCl₂ was mixed and resulted in 130 µM protein and 3.5 mM CuCl₂ in LCP, respectively (Fig. 3). The prepared samples were loaded into the HVE sample reservoirs and prepared for injection (Han et al., 2021).

Figure 3 Stereomicroscopy image of iq-mEmerald crystals (ca 5 × 15 µm) embedded in LCP, used for the mix-and-extrude device characterization.

The samples were extruded using two HPLC pumps (Shimadzu, LC-20AD XR) connected to the two HVE injector systems (Weierstall et al., 2014). The pumps were first run rapidly to fill the capillary with samples (up to 2.14 µl min⁻¹). When the extruded sample volume was determined to be around 13 µl, shortly before the samples reached the nozzle, the flow rates were reduced to the desired values.

3.1. Device characterization

We investigated the fluorescence quenching of iq-mEmerald protein in LCP mixed with copper ions for characterization of the mix-and-extrude nozzle on account of its facile visualization by means of optical microscopy. Moreover, the reaction time is so fast that it can be ignored in the mixing time scale of our nozzle, and the diffusion time of copper ions is expected to be fast compared with that of other ligands. The mixing ratio of crystal and copper ions was fixed to 1:3 so that the concentration of copper does not affect the quenching time and fluorescence intensity. The retention time for the characterization was triggered between 2.3 and 7.6 s (Fig. S4).

The same injection test was performed under a fluorescence microscope for quenching observation in iq-mEmerald crystals in LCP at high resolution (Movie S1 of the supporting information). Independently of the retention time, we could observe fluorescence quenching of mixed species within the jet, meaning that the diffusion of copper ions in LCP and the quenching reaction in iq-mEmerald crystals happened in less than 2.3 s. The diffusion and the reaction of the test system was very fast, and thus the fluorescence signal disappeared even before the sample was extruded from the nozzle. To measure how fast this diffusion and reaction happens within the mixing device, we reduced the intrinsic fluorescence of the IP-S photoresist by curing the device in a UV (λ = 385 nm) chamber (XYZ printing, 3UD10XEU01K) for 1 h followed by hard-baking at 80°C for 3 h.

Fig. 4(d) depicts the spatial/temporal evolution of the fluorescence signal at four downstream positions: inlet, mixer, nozzle and jet. Each plot shows the extracted grey value intensity from the 2D fluorescence microscopy image as a function of the channel width [indicated with yellow dotted lines in Figs. 4(a)–4(c)]. At the `inlet', where the two liquid channels overlap, the mixing time is equal to 0, and hence the mixture (blue) still has the same signal as pure protein (red). The fluorescence signal at the `mixer' position corresponding to a mixing time of 1.8 s (based on the distance travelled of 1.02 mm) decreases to ca 10% of the initial 130 µM concentration at the t₀ position and further downstream; as more time passes, there is no significant fluorescence signal inside the nozzle tip or in the free-flowing jet, indicating full quenching by mixing.

Figure 4 Observation of fluorescence quenching inside the 3D-printed mixing-HVE device. (a) Fluorescence microscopy image of the empty device which serves as the background image. (b) Background-corrected image of the sample, i.e. iq-mEmerald protein (130 µM) in LCP flowing at 0.36 µl min−1 inside the device. (c) Background-corrected image showing the mixing of iq-mEmerald (130 µM) and CuCl2 (3.5 mM) entering with a 1:3 flow rate ratio (0.36 and 1.07 µl min−1, respectively). The resulting flow velocity of the extruded mixture is 5.4 mm s−1 and the total retention time amounts to ca 4.4 s. (d) Pixel intensity profiles from the 16-bit fluorescence images (relative grey values extracted via ImageJ 1.53k) for four downstream positions: inside the 100 µm-wide sample orifice (inlet), inside the 231.7 µm-wide mixing channel within the Kenics section, 1 mm downstream from the inlet (mixer), within the final 75 µm-wide section of the device material before extrusion (nozzle), and in air, 300 µm away from the nozzle tip (jet). The line positions are denoted with the yellow dotted lines in (a). The green boxes in (d) indicate the geometric restrictions of the device/jet, which are the channel widths and jet diameter, respectively.

3.2. Diffusion in LCP

One of the biggest challenges in time-resolved serial crystallography, which determines distinct intermediate structures throughout the reaction, remains the rapid mixing of microcrystals with ligands (Brändén & Neutze, 2021; Schmidt, 2013). Even in investigations using aqueous solutions, the diffusion time of the ligand to the crystal centre varies according to the crystal size and packing, i.e. crystals that are smaller in size or with a higher water content per unit cell are unquestionably better candidates for time-resolved crystallography, as the ligand diffusion time would be decreased for such crystals. For membrane protein crystals in LCP, diffusion is even more challenging. LCP has a similar packing structure to crystals, with the exception of a wider water channel than protein crystals. In addition, the inner diameter of our device mixing channel is 231.7 µm, which increases the diffusion time by 3.1 s for oxygen and 13.2 s for glucose by calculation if the ligands are assumed to diffuse directly into the centre of the LCP stream (Atkins & Paula, 2006). To reduce this diffusion time caused by the width of the stream, a Kenics architecture was employed in our mixing channel to support the mixing of crystal and ligand in LCP. With the series of helical elements for repeated flow splitting/stretching, we exponentially increase the diffusive interface with each additional element. Mixing time point uniformity/dispersion becomes more dependent on spatial location along the mixer and highly insensitive to flow rate fluctuations. This feature is in contrast to flow-focusing designs that tune diffusion distances through flow rate differentials. Even with these efforts to obtain discrete intermediates by reducing diffusion time, we are still likely to observe a mixture of multiple intermediates by mixing with this device. With current advancements in data analysis (e.g. Xtrapol8; De Zitter et al., 2022), it is projected that mixed diffraction data of multiple states can be separated.

Owing to the complexity of the LCP structure and membrane protein, it is difficult to predict the diffusion and reaction time of ligand binding in LCP; for instance, if the ligand is diffused in an aqueous channel and the binding site of the membrane protein is located in an aqueous channel, the diffusion time would be comparable to that in aqueous solution. Alternatively, if binding occurs in the lipid, it is probably too time consuming to be studied using this type of device. Therefore, we assume that this mixing-HVE is suited for the examination of reactions taking place on the cell/membrane surface, which is what most current drugs targeting membrane proteins bind to (Yin & Flynn, 2016). However, due to the overall unpredictability of reaction in LCP, it is highly recommended to characterize the reactions in LCP prior to X-ray diffraction using different biochemical and biophysical techniques in order to confirm that the reaction takes place on a timescale compatible with the operation of these nozzles.

3.3. Time-resolved X-ray diffraction

To demonstrate the performance of the device under real life serial membrane protein crystallography conditions/requirements, we collected pre/post-mixing diffraction patterns of GPCR (G protein-coupled receptor) membrane protein crystals in LCP on the SPB/SFX instrument of the European XFEL (experiment No. 2826) using a photon energy of 12.4 keV, an X-ray pulse repetition rate of 10 Hz and a beam size of ca 2 × 2 µm. The pulse energy was 2.2 mJ and the pulse duration was ca 100 fs. The detector (JUNGFRAU 4M) was positioned 120 mm from the sample stream.

For injection, a co-flowing sheath of helium gas was used and the supply pressure was adjusted to 140 psi (corresponding to 24 mg min⁻¹) to maintain a stable sample flow. The distance from the nozzle tip to the X-ray interaction was ca 200 µm. For in situ mixing, a combined flow rate of 0.36 µl min⁻¹, i.e. 0.11 µl min⁻¹ (sample in LCP) + 0.25 µl min⁻¹ (ligand in LCP), with associated pressures of ca 450 and 420 psi at the HPLCs, was applied. This led to a flow velocity of 1.35 mm s⁻¹ for the extruded mixture (sample width = 75 µm) and a probed time point of ca 18.5 s. The un-mixed sample was pumped with a flow rate of 0.36 µl min⁻¹ (i.e. 1.35 mm s⁻¹) with an associated pressure of ca 480 psi at the HPLC. Fig. 5 shows exemplary diffraction patterns exhibiting strong reflections from the crystals in both states. Details on the sample and data statistics are currently being prepared for a forthcoming article.

Figure 5 Diffraction patterns of membrane protein crystals (a) without mixing and (b) after mixing with ligand (t ≃ 18.5 s) collected with the JUNGFRAU 4M detector (pixel size is 75 × 75 µm) at room temperature (experiment No. 2826). The detector consists of eight front-end modules, each being 80 × 40 mm (1024 × 512 pixels) large. The prominent ring arises from monoolein scattering at 4.5 Å resolution. In the insets, the beamline side-view microscopy images show the respective viscous extrusion from the nozzle tip (the scale bar denotes 100 µm).

3.4. Outlook

The current setup, which utilizes externally positioned fluid-feeding capillaries, can be susceptible to capillary rupture. Moreover, sample-conserving inner diameters below the 250 µm used here can be prone to generating high pressures during injection. Note that the capillary length of 30 cm used generated a void volume of 14.7 µl. Especially when considering the difficulties of membrane protein sample preparation and the available sample reservoir capacities of 40 and 120 µl, this sample loss might not be negligible. To reduce the length of the capillaries, a more compact mixing-HVE composed of two hydraulic systems within a single, protective injector rod is desired. In addition, the installation of injection infrastructure with a large footprint should be avoided to prevent interference with other significant instrumentation (detector, laser setup, electronics, optics etc.) already present at the beamline.

The design of another type of HVE described in the recent publication by Shimazu et al. (2019) allows a capillary connection to the sample reservoir without the need for direct screwing. Thus, we assume that an adaptation of their design would enable the advancement of a compact mixing-HVE utilizing very short capillaries.

As described above, the diffusion time of diluted species in LCP media is difficult to predict. To cover a broader range of diffusion and reaction time, modification of the device design is under consideration. For shorter retention time, the length of the mixer can be reduced. To access longer retention times, a modular assembly approach can be pursued in which the 3D-printed mixer and nozzle are connected by a capillary extension of customized length. In addition, once the compact mixing-HVE is commissioned and shorter capillaries can be used, the ID of capillaries and mixer/nozzle can be reduced because pressure build-up due to long capillaries is no longer a concern. Then, even higher sample flow rates can be applied to the system: with faster sample movement in the mixer, the retention time can be reduced further.