Chemical analysis using 3D printed glass microfluidics

Additive manufacturing (3D printing) is a disruptive technology that is changing production systems globally.


Introduction
Additive manufacturing (3D printing) technologies and their applications are advancing with unprecedented speed. For example, in medicine 3D bioprinting has allowed printing of tissues such as skin, bone and cartilage. [1][2][3] Custom made implants can also be made by additive manufacturing using Xray or magnetic resonance imaging combined with 3D model design. 4,5 In the pharmaceutical eld, additive manufacturing allows personalized medicine through drug dosage forms with personally tailored drug combinations, doses and drug release proles. [6][7][8] In the chemical sciences, additive manufacturing is important for manufacturing miniaturized and microuidic devices as chemical reactors or [bio]analytical devices. [8][9][10][11][12][13][14] Additive manufacturing of microelectronics is also possible. 15 Furthermore, additive manufacturing has allowed manufacturing of shape changing objects, with the shape transformation obtained, for example, by illumination with light. 16 In these cases, the term 4D printing has been used. Based on these examples, the main advantages of additive manufacturing stem from the fact that, aer converting imaging data or visual observations and measurement data into a 3D computer design, the design can be easily and quickly transferred, with a single apparatus, to a desired complete product in a single stepassembly of multiple parts is typically not needed. Therefore, additive manufacturing enables rapid and affordable prototyping of devices and thus fast cycles of tool development and experiments.
The range of materials which can be used in additive manufacturing (3D printing) is extensive, including biomaterials (e.g. cells, protein, hydrogels), polymers (e.g. thermoplastics and photocurable polymers), ceramics and metals, 9 as well as multiple materials. 14,16 However, the lack of standardized biocompatible and transparent materials has been one of the main factors limiting the adoption of additive manufacturing for microuidics. 10 Glass is one of the best materials for many chemical and biological applications, due to its excellent chemical and heat resistance, transparency, hardness, and high dielectric strength. 17 An additional advantage of glass, especially for chemical and biological measurement applications, is that the surface properties of glass can be easily tuned by well-known surface modication reactions. 18 Such excellent properties have stimulated the development of additive manufacturing methods for glass. However, few examples consider the millimeter scale or below, i.e. going down to microuidic dimensions and devices. One reason for this is the very high working temperature of glass (typically > 1000 K). 17 Glass may also crack due to intrinsic stresses induced during glass fabrication when not properly annealed. The major drawbacks for manufacturing of microuidic glass devices by traditional microfabrication methods include the use of hazardous reagents, high instrumentation costs and challenges in bonding of glass layers. 19 Two recent studies report additive manufacturing (3D printing) methods for transparent glass microuidics based on the use of composite materials containing silica nanopowder: stereolithography 20,21 and direct ink writing. 22 Despite this progress in additive manufacturing of microuidics, a very long heat treatment time (tens of hours) was needed in both studies for removal of solvent and organic binders of the composite material, as well as for sintering to obtain the nal glass structure. Furthermore, since the method requires the outgassing of volatile binders, there is a limit to the wall thickness of the glass structures created with such methods, hampering their applicability. Recently, glass 3D printing using sol-gel technology 23 and a new fused silica glass subtractive machining technique based on a nanocomposite material have been introduced (microchannel structure shown). 24 However, both these methods also include a long heat treatment stage.
Selective laser-induced etching can be used to make 3D microuidic glass structures, but it is not an additive manufacturing method since a laser is initially used to "draw" the structures in a glass specimen followed by wet-chemical etching to remove the exposed glass material. [25][26][27][28] This method is especially suited for quartz glass, and its precision is high ($10 mm). 26 However, production is generally limited to relatively short channels ($20 mm), 26,27,29 but it is possible to increase the channel length by using additional vertical entrances for the etching reagent. 30 Manufacturing of small devices is also possible with femtosecond laser ablation, oen with the glass partly in water. However the resulting resolution and etching quality are worse than with subsequent chemical etching, 31 and microchannel lengths are typically limited to the cm range. 32,33 Femtosecond laser direct writing of porous glass and subsequent annealing allows manufacturing of microchannels with lengths of many centimeters and nanochannels with widths of tens of nanometers. [34][35][36][37] In this study, we report characterization results and chemical applications of microuidic devices (with a possible channel width down to $100 mm) produced by additive manufacturing of molten glass ( Fig. 1 and 2). This is the rst full scientic study to report the devices and their characterization, it is a continuation of our preliminary report showing glass microreactors. 38 The additive manufacturing method used is a further development of the proprietary molten glass 3D printing technology developed at MICRON3DP. 39,40 Additive manufacturing of molten transparent glass has also been reported by Oxman et al., targeted mostly for architectural, artistic and industrial use. [41][42][43] In our study, functional operation of the additively manufactured microuidic glass microreactor is demonstrated with online electrospray ionization (ESI)mass spectrometric analysis of the last reaction step in the synthesis of linezolid, an antibiotic of last resort. In addition, operation of the additively manufactured glass direct infusion device is demonstrated by the analysis of selected drugs using electrospray-mass spectrometry. Moreover, the excellent optical quality of the additively manufactured glass structures is demonstrated by Raman spectroscopic measurements through the reactor channel wall.
Additive manufacturing (3D printing) of the glass devices Additive manufacturing of molten glass is a proprietary technology of MICRON3DP. [44][45][46] The technology is especially designed for 3D printing of materials with a high melting temperature. The technology includes parts such as a heated chamber inside which printing is performed such that thermal stress does not cause cracks, a printing nozzle (lower part inside the heated chamber) and its heating system, a printing base plate mounted inside the heated chamber, a cooling unit for cooling the upper part of the printing nozzle (outside the heated chamber), a system for precise movement of the printing nozzle and/or the printing base plate and a loading system for printing material allowing, for example, printing of different glass colors and continuous printing to avoid cracks. Further details about the manufacturing of the glass devices is discussed in the rst paragraph of the Results and discussion section.

Mass spectrometric (MS) set-up and conditions for the reaction studies
The MS measurement set-up is shown in Fig. S1. † Deactivated fused silica capillaries (OD 190 mm (based on caliper (0-25 Â 0.01 mm) measurement 180 to < 190 mm), ID 75 mm, length between 320 mm and 370 mm, Trajan Scientic Europe, Milton Keynes, UK) were attached to the additively manufactured glass reactor by inserting them into the inlets and gluing them (capillaries aer gluing $3-11 mm inside the channels) with Duralco 4703 (Cotronics Corp., Brooklyn, USA), a highly thermally and chemically stable epoxy glue. The edge of the microreactor (Fig. 2a) required grinding in some places to reach the channels for silica capillary insertion. The microreactor has in total ve inlet/outlets, two at one end of the base plate, and three at the other end of the base plate (Fig. 2a). In the online experimental results presented ( Fig. 3 and S10 †), the inlet/outlet 1 was plugged with epoxy, and the inlet/outlet 2 was used as the reaction solution outlet (reaction channel length $360 to 380 mm). The reaction solution was conducted via the fused silica capillary to the ion source of an ion trap mass spectrometer. The fused silica capillaries of inlets/outlets 4 and 5 were used (compound 2 from 4 and compound 1 from 5) for reagent infusion into the microreactor. The capillary connected to inlet/ outlet 3 was in this experiment blocked with a column plug and a union (IDEX Europe GmbH, Erlangen, Germany). An alternative measurement arrangement was also used, in which inlets/outlets 1 and 2 were used for reagent infusion, inlets/ outlets 4 and 5 were plugged with column plugs and unions (IDEX Europe GmbH, Erlangen, Germany) and the capillary of inlet/outlet 3 was used to conduct the reaction solution to the ion trap mass spectrometer (reaction channel length $330-350 mm). The two reagent introduction silica capillaries were connected to separate 1 mL syringes (ID 4.6 mm, Hamilton Bonaduz AG, Bonaduz, Switzerland) by ttings and unions (IDEX Europe GmbH, Erlangen, Germany). A PHD 2000 syringe pump (Harvard Apparatus, Holliston, USA) was used for infusing the solvents and/or reagents into the reactor.

MS measurement conditions in the reaction studies
The outlet of the glass microreactor was coupled to a commercial electrospray ionization source of an Agilent 6330 ion trap (Agilent Technologies, Santa Clara, USA). The ion trap was operated in positive mode, with the capillary voltage À3 kV. Nitrogen was used as drying gas and introduced with a ow rate of 4 L min À1 . The drying gas temperature was 300 C and the m/ z scan range was 50-600 or 50-700 for the MS and MS n experiments. Mass spectra were recorded continuously during the online experiments, and the presented mass spectra are the average over 1 min at time points when the ion currents for the used ow rate had stabilized. The integer value of the precursor ion was isolated for the MS n experiments with the isolation width 1.0 m/z. MS n data was recorded for 30 s for each precursor ion and the product ion mass spectra were averaged over this time. The mass spectrometric data was processed with the soware DataAnalysis for 6300 series ion trap LC/MS version 3.4, Build 192 (Agilent Technologies, Santa Clara, USA).

Reaction conditions
Initially, the additively manufactured microreactor was lled with Chromasolv®-grade acetonitrile by infusing it through two of the inlets with 1 mL syringes using typically the rst or the only infusion ow rate in the online reaction experiment to record a background mass spectrum (a mass spectrum of the solvent used for preparing the samples without reagents). Aer this, the online reaction studies were conducted. A 1 mL syringe was lled with a 100 mM solution of (5S)-5-(aminomethyl)-3-[3-uoro-4-(4-morpholinyl)phenyl]-1,3-oxazolidin-2-one in acetonitrile and another one with a 100 mM solution of acetic anhydride in acetonitrile. This procedure gives the concentration of 50 mM per reagent at the beginning of the glass reactor channel. Solutions producing reagent concentrations of 170 and 10 mM at the beginning of the glass reactor were also tested (MS spectra not reported). The syringes were coupled to the separate inlets of the glass reactor as described above in the MS measurement set-up section and the mass spectrometric data acquisition started. Infusion ow rates of 4.0 mL min À1 , 1.6 mL min À1 and 750 nL min À1 per syringe were used, giving total ow rates through the reactor of 8.0, 3.2 and 1.5 mL min À1 , respectively. Reaction experiments using only one total ow rate and experiments in which the total ow rate was decreased (8.0, 3.2 and 1.5 mL min À1 ) during the same measurement time were performed. All reaction experiments were conducted at room temperature. Mass spectra were recorded continuously during the reaction experiments and averaged at time points when the signal represented the stabilized infusion ow rate. Of note, in the experiments in which the ow rates decreased during the experiment, there are delay times before the observed signals stabilized to the characteristic level of each ow rate. In addition to the online experiments, pure (5S)-5-(aminomethyl)-3-[3-uoro-4-(4-morpholinyl)phenyl]-1,3oxazolidin-2-one (1) and pure linezolid (3) (typically each 50 mM in acetonitrile) were analyzed by direct infusion electrospray ionization mass spectrometry using the same MS operating settings as for the online reaction study. This was done to collect reference MS and MS n data for these compounds.

Mass spectrometric set-up and conditions for the direct infusion experiments
Stainless steel nano-ESI needles (length 30 mm, OD 150 mm, ID 30 mm Thermo-Fisher Scientic Oy, Vantaa, Finland) were glued in the channels of the direct infusion device with Duralco 4703 (Cotronics Corp., Brooklyn, USA) (Fig. 2b). The direct infusion device was positioned in front of the capillary extension of an Agilent 6330 ion trap using an xyz-stage (Märzhauser Wetzlar GmbH & Co. KG, Wetzlar, Germany) attached to an ion source frame. One end of a copper wire was wrapped around the nano-ESI needle, while its other end was wrapped around a metal rod of the ion source frame to ground the nano-ESI needle.

MS settings for the direct infusion experiments
The Agilent 6330 ion trap (Agilent Technologies, Santa Clara, USA) was operated in positive ESI mode, using a capillary voltage of À3 kV. The temperature of the drying gas (nitrogen) was 100 C, and the ow rate of the drying gas was 2 L min À1 . The used m/z scan range was 50-600. The presented mass spectra are the average over 1 min.

Measurement conditions for the direct infusion experiments
Solutions of 10 mg mL À1 verapamil hydrochloride (20 mM) and testosterone (35 mM) in a mixture of Chromasolv®-grade methanol : water 80 : 20 + 1 vol% formic acid were prepared, respectively. The channel (45 mm channel in Fig. 1d and 2b) of the direct infusion device was lled with the sample mixture by manual direct infusion using a 1 mL syringe (ID 4.6 mm, Hamilton Bonaduz AG, Bonaduz, Switzerland). Complete lling of the channel was conrmed by lling the channel until it was ooded, visually ensuring that there were no air bubbles in the channel (and in case of air bubbles, the device was tilted vertically so that the air bubbles could exit at the infusion inlet, aer which the device was again lled with direct manual infusion), and nally depositing a droplet of the sample mixture at the sample reservoir. When the lled device was positioned in front of the mass spectrometer inlet, acquisition of mass spectra was initiated, the capillary voltage of the mass spectrometer was switched on and optimization of the positioning of the device in front of the mass spectrometer was done until a stable electrospray current was obtained. In all the experiments of sufficient quality with this device, the capillary current of the mass spectrometer was below 100 nA.

Raman spectroscopy
A time-gated Raman spectrometer was used to measure the Raman spectra (TimeGated® 532, TimeGate Instruments, Oulu, Finland). The system consisted of a 532 nm picosecond pulsed laser (pulse width < 100 ps, repetition rate 40-50 kHz, average output power approximately 60 mW), sampling optics (ber), spectrometer with a complementary metal-oxidesemiconductor (CMOS) single-photon avalanche diode (SPAD) detector (8 cm À1 spectral resolution) and a computer with MATLAB-based measurement soware (MathWorks, Massachusetts, USA). A sampling probe (BWTek Raman Probe, B&W Tek, Newark, USA) with a focal distance of 5 mm and spot size of approximately 85 mm was used.
Reference spectra of cyclohexane and ethyl acetate were measured in a 4 mL glass vial (through the glass). For the microreactor device (Fig. 2a) measurements, the sampling probe was placed above the channel in the glass microreactor (Fig. S2 †). Spectra without solvent and then with cyclohexane or ethyl acetate were measured from 700 to 1500 cm À1 and 2600 to 3200 cm À1 from the same spot. The solvents were infused into the glass reactor using syringe pumps (Harvard Apparatus, Pump 11 Elite and PHD 2000, Massachusetts, USA, in Fig. S2 † only one pump shown) and 1 mL syringes with an infusion rate of 5 mL min À1 . The time-resolved spectral datasets, including Raman and photoluminescence signals, were collected with the time gate sequentially moved in 50 ps steps using the electronic delay generator from 0 to 2 ns. The processed sum Raman spectra consisted of six such cycles and the time window of 0.2 to 0.7 ns was used when uorescence subtraction was performed. Min-max normalization was also applied. The total measurement time for each sample was approximately 6 min. The measurements were carried out at ambient temperature, lighting and humidity.

Microscopy
To obtain the channel cross-section images, either intact glass devices, or glass channel cross-section samples, were used. For optical microscopy, an object of known size, included in the image was used to obtain the scale bar. For scanning electron microscopy (TM-1000, Hitachi, Tokyo, Japan), the samples were coated with gold by sputtering (Bal-Tec SCD 050, Schalksmühle, Germany). The sputtering voltage was 60 mA and the time was 60 s.

Surface roughness measurements
Surface roughness of the additively manufactured glass microreactors was characterized with a Dectak XT (Bruker, Germany) stylus prolometer. Fig. 1 illustrates the concept of additively manufacturing (3D printing) microuidics from molten glass. Using a custom STL to G-code technique, MICRON3DP can manufacture glass channels as narrow as 100 mm (Fig. 2). The soware-based designs can be modied during manufacturing to reduce the cross-section and increase the yield of glass devices. When the nal printing model does not contain holes at the end of the channels, the rims of the manufactured microuidic parts are cut away to open the channels. The glass additive manufacturing process is carried out directly on a glass plate inside a heated chamber, thus preventing thermal shock (Fig. 1b). The manufacturing temperature for the soda-lime glass was 1000 C in the present study. Manufacturing (printing) times are typically on a time-scale of minutes. For example, the object presented in the movie 38 was additively manufactured in 15 min. In addition to the time needed for additive manufacturing, there is an annealing time of approximately 20 to 30 min, which is calculated based on the glass volume manufactured. Soon aer the annealing, the product can be used. Multiple microuidic tools can be printed on the same glass base plate in one print job. A movie showing additive manufacturing of glass is presented in the ref. 40.

Results and discussion
Two prototype devices were designed to demonstrate both the printing capabilities and the performance in synthetic and analytical chemistry microuidic applications. An additively manufactured microreactor device is presented in Fig. 2a (glass base plate size 100 Â 50 Â 2 mm 3 ). The microreactor device has two reagent inlets/outlets at one end and three inlets/outlets at the other end. The three-inlet/outlet topology can be used for addition of a dilution solvent, another reagent or a quenching reagent. The total length of the channel is approximately 400 mm. This geometry is useful, for example, for the analysis of reactions that need higher reagent concentrations than the online analytical equipment can tolerate. Microreactor devices with the same design, but with smaller and larger channel cross-sections are shown in Fig. S3. † However, the uidic functionality and leak tightness of these devices was not tested. The second design was a direct infusion device ( Fig. 2b and S8, † glass base plate size 100 Â 50 Â 1 mm 3 ), which has a straight channel with a sample reservoir at the other end. The channels with a glued electrospray needle have lengths ranging from 35 to 60 mm, from which the 45 mm channel was used for mass spectrometric experiments. The direct infusion devices show that small glass channels can be additively manufactured for direct infusion electrospray ionization mass spectrometry. In addition, a third design manufactured was a spiral device ( Fig. S9 †), this contains an extremely long single channel ($2.5 m) with an inlet/outlet at each end. The integrity of the spiral device was tested by pumping air through it when the device was submerged under water. The ESI movies (Movies S1 and S2 †) reveal air bubbles emerging from the exit as well as from the inlet side. Altogether, these devices demonstrate the versatility of additive manufacturing with molten glass. Neither the channel length nor the sidewall thickness is limited, unlike with techniques requiring either development or outgassing steps. Furthermore, 3D shapes including 90 vertical turns can be manufactured as demonstrated by the devices in Fig. 2b and S9. † The geometries and sizes of microchannel cross-sections were examined with optical and electron microscopy. Fig. 2ch shows example cross-sections for the manufactured glass devices. The smallest channel ( Fig. 2c and f) has a triangle crosssection with a width of $140 mm and a height of $100 mm. The mid-size channel ( Fig. 2d and g), has a "tilted rectangle" crosssection with a width and height of roughly 400 and 600 mm, respectively. The largest channel (Fig. 2e and h) has a triangular cross-section and is about 2 mm wide and high. The crosssection shape and size variation is demonstrated for the small channel reactor (Fig. S3) in Fig. S4, † the mid-size channel microreactor (Fig. 2a) in Fig. S5 and S6, † and the large channel structure (Fig. S3) in Fig. S7. † In addition, the size and shape variation of the channel ends of the direct infusion device (Fig. 2b) are presented in Fig. S8. † As can be seen from Fig. S4-S7, † the channel size and shape within a glass device are similar in the entire device and it is possible to manufacture different channel shapes (see Fig. S4-S8 † for comparison).
The surface roughness of the outer surface of the microreactors was characterized by placing the tip of a stylus prolometer on the top of a channel and scanning longitudinally. The surface was scanned over 400 mm at 10 different spots to obtain the average roughness, R a, and root mean square (RMS) roughness, R q . This resulted in R a and R q averages of 47 nm and 58 nm respectively, with their corresponding standard deviations being 17 nm and 22 nm, respectively. We assume that these numbers represent the surface roughness of one glass layer, which should be on the same order of magnitude on the outside and inside of the channels, since molten glass is printed. Undulation in the channel wall structure (see Fig. 2c-e, and S4-S8 †) is believed to be caused by the glass layer thickness.
The total volume of the mid-size microreactor device (Fig. 2a) was calculated from the overall channel length of $400 mm and the cross-sectional dimensions (the cross-section area estimated as the average of four channel (Fig. S6 †) cross-sections is 0.124 mm 2 with STD of 0.015 mm 2 ) obtained from the electron micrographs. The volume calculated in this way was 50 mL. A larger volume, approximately 65-70 mL, was obtained for the total measurement system (reaction channel length $360-380 mm and connections to syringe pump and to MS; see microreactor in Fig. 2a and set-up in S1 †) used for online mass spectrometric reaction experiments. This estimation is based on the reaction solution ow rate and the delay aer introducing the reagents until detection of reagent (1) and nal product (3) by mass spectrometry.
All in all, there are ve key advantages of the glass additive manufacturing method developed and used: (1) fabrication of microchannel dimension as low as 100 mm (Fig. 2f), (2) the ability to print large complete devices (sizes up to 200 Â 200 Â 350 mm 3 ) on top of a glass base plate, (3) the ability to print very long microchannels (Fig. 2a (400 mm) and S9 † (2.5 m)), (4) only the layer dimensions are limiting the sidewall thickness, and (5) rapid manufacture of microuidic devices (within tens of minutes instead of tens of hours). Additive manufacturing of different glass qualities (soda-lime, borosilicate) and colored glass is also possible, even though in this study the microuidic devices were limited to clear soda-lime glass. These unique additive manufacturing characteristics are based on recently developed printer nozzle and heating system structures.
The microreactor (Fig. 2a) was used for online analysis of the last step in the synthesis of linezolid (Fig. 3a) 47 by infusing (5S)-5-(aminomethyl)-3-[3-uoro-4-(4-morpholinyl)phenyl]-1,3oxazolidin-2-one, reagent 1, through one of the microreactor inlets and acetic anhydride, 2, (both dissolved in acetonitrile) through another inlet and conducting the reaction solution via an outlet directly to an ion trap for continuous mass spectrometric analysis. The mass spectrum presented in Fig. 3 is from an experiment in which the initial reagent ow rate of 4.0 mL min À1 per reagent (total ow 8.0 mL min À1 ) was decreased in two steps, rst to 1.6 mL min À1 per reagent (total ow 3.2 mL min À1 ) and then down to 750 nL min À1 per reagent (total ow rate 1.5 mL min À1 ). A mass spectrum with the reagent infusion ow rate of 750 nL min À1 is presented in Fig. 3b and mass spectra with reagent infusion ow rates of 4.0 mL min À1 and 750 nL min À1 are shown in Fig. S10. † Protonated molecules of both reagent 1 and the nal product 3 can been seen in these mass spectra at m/z ratios of 296 and 338, respectively. Another observation was that the intensity of the protonated linezolid reaction product [3 + H] + increases in relation to the protonated reagent [1 + H] + at m/z 296, as the reaction time increases due the decreased reagent infusion ow rate (Fig. S10 †). The offline experiment (reaction in a vial) showed the same trend, i.e. the signal of protonated linezolid, [3 + H] + increases in relation to [1 + H] + as the reaction time increases (Fig. S12 †). MS n (multiplestage mass spectrometry) mass spectra and the proposed fragmentation routes for the m/z 296 and 338 are presented in the ESI † (Fig. S13-S15, Schemes S2 and S3 †). Other reaction related ions present and tentatively identied in the mass spectra ( Fig. 3b and S10 †) are: m/z 241, a fragment of m/z 296 formed in the ionization process; m/z 360, [3 + Na] + ; m/z 633, [1 + 3 + H] + ; and m/z 675, [3 + 3 + H] + .
The direct infusion device (Fig. 2b) was used for measuring mass spectra of verapamil and testosterone using electrospray ionization mass spectrometry. The electrospray was achieved with a help of a stainless steel nanospray needle glued into the direct infusion device. The length of the channel used in this experiment was 45 mm. Protonated verapamil and testosterone at m/z 455 and 289, respectively, were clearly seen in the mass spectra (Fig. S16 †).
The potential of optical detection for in-line analysis with additively manufactured glass devices was studied by measuring Raman spectra of cyclohexane and ethyl acetate through the channel of the microreactor device (Fig. 2a, 4 and S2 †). The Raman spectrum of the empty glass reactor contained a broad peak centered at 1112 cm À1 which is associated with stretching of the Si-O-non-bridging and Si-O-Si bridging groups (Fig. 4a). 48,49 No bands were present between 2600-3200 cm À1 (Fig. 4a). With the exception of the mode centered around 1112 cm À1 , the additively manufactured glass otherwise exhibits minimal Raman scattering and consequent interference in the spectral regions of interest for organic system analyses. 50 Cyclohexane exhibited a strong symmetric CH 2 stretching mode at 2855 cm À1 and two antisymmetric CH 2 stretching modes at 2928 cm À1 and 2942 cm À1 (Fig. 4b (vial) and 4c (microreactor)). 51 Ethyl acetate exhibited a strong CH 3 stretching signal at 2948 cm À1 (Fig. 4b (vial) and 4c (microreactor)). The small solvent sampling volume affected the observed Raman signal intensities of the solvents. Nevertheless, the signals showed that such additively manufactured reactors have sufficient optical transparency for non-invasive in-line Raman measurements.
Time-gated Raman spectroscopy (Fig. 4d-i) has the further advantage that Raman spectra can be efficiently obtained in the presence of uorescence and other photoluminescent phenomena. 52,53 This is achieved through temporal rejection of the uorescence signal, which, unlike the near instantaneous Raman signal, typically occurs on the order of nanoseconds (or even longer). 54 This expands the range of possible glass materials (e.g. soda-lime glass) for printing, without concern for their photoluminescent properties. Efficient uorescence rejection also makes the non-contact sampling optics more exiblehere a simple non-contact ber-optic Raman probe with a theoretical spot size of approximately 85 mm has been used and sampling could be performed directly through the printed glass. To date, non-contact sampling in glass microuidic systems has generally involved microscope objectives and optimized selection of the bottom glass plate material through which the non-contact sampling has been performed. 55 Furthermore, time-gated instrumentation allows measurements to be performed in ambient lighting without interference. The present results demonstrate that Raman (and other optical) analysis is possible with additively manufactured glass microuidic set-ups. The sensitivity of the Raman analysis (e.g. for complex mixtures and low concentration components) could be further optimized by, for example, modifying the microuidic device geometry and increasing the Raman scattering signal intensity with, for example, surface enhanced Raman scattering (SERS) approaches. 50

Conclusions
True additive manufacturing, 3D printing, of microuidic (channels < 500 mm) devices from molten glass has been demonstrated. The main advantages of the devices stem from the excellent properties of glass for chemical and biological applications and the fact that large complete devices can be manufactured in tens of minutes. The latter property is especially important when one considers the advantages of 3D printing such as custom manufacturing and rapid prototyping. The applicability of one of the devices was shown with online mass spectrometric analysis of the last reaction step in the synthesis of linezolid, while another was used for direct infusion MS measurements of drugs. The excellent optical quality of the additively manufactured glass was demonstrated by measuring the Raman spectra of solvents. Relevant optical technologies include uorescence, Raman, and UV spectroscopies as well as colorimetry. The optical sampling possibilities are diverse: depending on the optical technique, measurements could be performed in the backwards (including through the base-plate) or transmission (forwards) directions anywhere along the microchannel path. There is also potential for simultaneous measurements at different positions along the channels as well as imaging. We see a bright future for these micrometer or millimeter scale glass devices, for example, in ow chemistry, an area of chemistry with large commercial success. In addition, applications in analytical and medicinal chemistry, biology and biochemistry are likely to result from this study.

Conflicts of interest
MICRON3DP has a commercial interest in the results.