Flow Radiocytometry Using Droplet Microuidics

Flow-based cytometry methods are widely used to analyze heterogeneous cell populations. However, their use for small molecule studies remains limited due to bulky fluorescent labels that often interfere with biochemical activity in cells. In contrast, radiotracers require minimal modification of their target molecules and can track biochemical processes with minimal interference and high specificity. Here, we introduce flow radiocytometry (FRCM) that broadens the scope of current cytometry methods to include radiotracers as probes for single cell studies. FRCM uses droplet microfluidics and radiofluorogenesis to translate the radioactivity of single cells into a fluorescent signal that is then read out using a high-throughput optofluidic device. As a proof of concept, we quantitated [ 18 F]fluorodeoxyglucose radiotracer uptake in single human breast cancer cells and successfully assessed the metabolic flux of glucose and its heterogeneity at the cellular level. We believe FRCM has potential applications ranging from analytical assays for cancer and other diseases to development of small-molecule drugs. This paper reports several significant improvements to realize the high-throughput FRCM. First, we present a comprehensive study of a radiofluorogenic probe suitable for use with water-in-oil microemulsions. Optimal working conditions for the radiofluorogenic chemistry were established based on X-ray and FDG exposure. Second, we present microfluidic and optofluidic device designs for efficient drop generation and read. We also demonstrate the use of a 2-photon lithography system for 3D printing of complex microfluidic channel molds. To our knowledge, this is the first reported use of such a system for fabricating the mold of entire microchannels including inlets and outlets. Third, we present the design, fabrication, and operation of an incubation system for radiofluorogenic drops that minimizes physical crosstalk. Finally, as a proof of concept, FRCM was performed to compare FDG uptake by control and GLUT1-knock-down cells, which successfully quantitated the difference in the metabolic flux of glucose in the two biologically different samples and assessed cellular heterogeneity on the level of single cells. We anticipate that, with further integration with downstream cell sorting, this platform will constitute a high-throughput method for studying the distribution of a wide range of radiolabeled molecules in single cells using fluorescence detection. the cells (d) We and added 20 units I. (e) We pipetted 20 times to help cell into single cells and filtered out remaining cell clusters using a 40 μ m-pored mesh strainer (make). (f) We measured cell density using hemocytometer and total radioactivity of the cell sample using a dose calibrator (Atomlab 400, Biodex) to estimate average radioactivity of radiolabeled single cell. (g) We added 1x PBS to achieve a cell number density of ~1 × 10 6 cells/mL and 16% v/v OptiPrep density gradient medium (Sigma-Aldrich) to match the density of the final medium to 1.06 kg/m 3 reported as the density of mammalian epithelial cells 32,33


INTRODUCTION
Single-cell analysis methods such as flow cytometry are widely used to investigate the functional and phenotypic heterogeneity of cell populations 1 . However, in many instances, the use of bulky fluorophores is prone to interfering with the process being studied 2 . In contrast, radiotracers require minimal modification of their target molecules and are capable of tracking biochemical processes with minimal interference and high specificity. In recent years, new approaches such as radioluminescence microscopy 3 have been developed that can perform single cell analysis using clinical radiotracers, i.e. radioactive probes used for imaging scans and cancer therapies.
However, none of the existing methods can analyze more than ~100 cells at a time, preventing statistically robust characterization of highly heterogeneous population of cells. Here, we report a drop-based microfluidic platform named flow radiocytometry (FRCM) which measures radioactivity of single cells with high-throughput and could potentially be integrated with downstream cell sorting.
Radiotracers have found wide usage throughout medicine due to their unique advantages. In addition to their ability to closely target specific molecular pathways, radiotracers can be imaged deep within the tissues of humans and animals. As an example, a fluorine-18-labeled glucose analog, 2-deoxy-2-[ 18 F]-fluoro-D-glucose (FDG), serves as an imaging agent for positron emission tomography (PET) to track glucose consumption in-vivo 4 . FDG-PET enables a direct, real-time measurement of the activities of membrane transporters and downstream cellular enzymes. Owing to the superior specificity and sensitivity of the radiotracer, FDG-PET is routinely used in hospitals to assess organ function or locate disseminated tumors [4][5][6] . Another prominent advantage of radiotracers is that there are about 50 FDA-approved radiotracers available for human use 7 . These have been proven to be reliable and effective from decades of use in hospitals. In addition, PET has served a crucial role for precise pharmacology studies of human subjects in later-stages of drug development 8 . Moreover, novel radiopharmaceuticals are actively developed for cancer theranostics and many are in the pipeline pursuing regulatory approvals [9][10][11][12] .
Despite the unique strength and utility of radiotracers, their use for single-cell analysis has been limited due to the poor resolution of conventional radionuclide imaging methods. PET has spatial resolution of millimeters at best due to the fundamental resolution limit imposed by positron range 13 . Recently, researchers have worked to develop on-chip radiometric imaging platforms to overcome the fundamental challenge and achieve single-cell resolution. This research field emerging over the last decade has recently been reviewed by Liu and Lan 14 . A notable work in the field is radioluminescence microscopy (RLM) developed in 2012 at Stanford 3,15 . The proposed technique combines scintillator-mediated radioluminescence and epi-fluorescence microscopy to visualize radiotracer uptake of living single cells [15][16][17] . RLM offers the best resolution (i.e., tens of microns) for radionuclide imaging to date 14 , and several RLM studies have successfully assessed heterogeneity of cellular uptake of various clinical radiotracers 3,[18][19][20] .
Although RLM offers high spatial resolution, its use remains limited because it can analyze only 40 -200 cells at a time, and each acquisition requires at least 5 minutes, and can last as long as an hour 21 . In fact, the limited throughput is fundamentally related to the stochastic nature of radioactive decay. Radioactive decay events are spontaneous and random, in contrast to fluorescence which can be measured on demand upon stimulation. Therefore, long exposures are required to detect a statistically robust number of radioactive events and achieve sufficient signal-to-noise ratio. RLM is also challenging to integrate with cell sorting for post analysis, and all these shortcomings preclude any large-scale application of RLM.
We developed FRCM using radiofluorogenesis and microfluidic technology to achieve highthroughput single cell radiometry and enable future on-chip integration with downstream cell sorting. Radiofluorogenesis converts dim and discontinuous radioactive signals from single cells into integrated bright and continuous fluorescence. The radiofluorogenic translation is realized using molecular probes which irreversibly convert to fluorophores in response to ionizing radiation. Using a microfluidic platform, we encapsulated radiolabeled single cells together with radiofluorogenic probes in water-in-oil drops (Fig. 1a). Within the drop, the ionizing particles (positrons, electrons, or alpha particles) emitted from an encapsulated single cell produce reactive oxygen species (ROS) via water radiolysis (Fig. 1b). These ROS can instantaneously react with radiofluorogenic probes, generating a fluorescent signal proportional to the level of radioactivity. The translation of radioactivity into fluorescence enabled a 2-order-of-magnitude reduction in acquisition time per cell, which made possible high-throughput single-cell radiometry by measuring the fluorescence of individual drops using a continuous-flow optofluidic platform. The strategy of co-encapsulating single cells and ROS probes in microdrops was first proposed and attempted by Gallina et al. in our group 21 . Although Gallina et al. demonstrated the potential feasibility of the approach, their assay suffered from significant crosscontamination between drops, polydispersity in drop size, and low readout throughput. This paper reports several significant improvements to realize the high-throughput FRCM. First, we present a comprehensive study of a radiofluorogenic probe suitable for use with water-in-oil microemulsions. Optimal working conditions for the radiofluorogenic chemistry were established based on X-ray and FDG exposure. Second, we present microfluidic and optofluidic device designs for efficient drop generation and read. We also demonstrate the use of a 2-photon lithography system for 3D printing of complex microfluidic channel molds. To our knowledge, this is the first reported use of such a system for fabricating the mold of entire microchannels including inlets and outlets. Third, we present the design, fabrication, and operation of an incubation system for radiofluorogenic drops that minimizes physical crosstalk. Finally, as a proof of concept, FRCM was performed to compare FDG uptake by control and GLUT1-knockdown cells, which successfully quantitated the difference in the metabolic flux of glucose in the two biologically different samples and assessed cellular heterogeneity on the level of single cells.
We anticipate that, with further integration with downstream cell sorting, this platform will constitute a high-throughput method for studying the distribution of a wide range of radiolabeled molecules in single cells using fluorescence detection. Figure 1. Illustrations showing production of reactive oxygen species (ROS) from radioactive beta decay and radiofluorogenic conversion from the ROS. (a) A single radioactive cell is encapsulated into a water-in-oil drop. As the radiotracer decays, the emitted beta radiation creates ROS such as hydroxyl radicals, which in turn mediate the radiofluorogenic conversion.

A. Experimental overview and design
The goal of FRCM is to measure the heterogeneous behavior of single cells on the basis of their transport and incorporation of radiolabeled substrates. For instance, the metabolic flux of glucose into single cells can be assessed by measuring their radioactivity after incubation with a radiolabeled glucose analog. In this work, we quantitated [ 18 F]FDG radiotracer uptake in thousands of single human breast cancer cells (MDA-MB-231). It is important to assess cellular heterogeneity of cancer considering that higher uptake of FDG is interpreted as a distinctive property of cancer tissues. Reports have shown that increased glycolysis in cancer is associated with higher cell proliferation, metastatic potential and therapeutic resistance [22][23][24] . This implies that FRCM may provide crucial information for cancer therapeutics and identification of rare cell subclones. The assessment of heterogeneity in FDG uptake from ~100 cancer cells has been done via RLM, but we here present statistically robust high-throughput data for thousands of cells.

a. Drop design and generation
In the FCRM assay, cells are incubated with a radiolabeled substrate of interest, then washed and analyzed following a three-stage process comprised of drop encapsulation, incubation, and read ( Fig. 2a). Using a flow focusing microfluidic chip, called hereafter DropGen microchip, single radioactive cells were mixed with a radiofluorogenic probe (Fig. 2b) and encapsulated in water-in-oil drops (Fig. 2c). The device generated 800 pL monodisperse drops at a throughput of 200 per second. In addition, since drops may be empty or contain multiple cells, we used fluorescent quantum dots (QDs) to label cells and allow them to be counted in each drop.
The assay is designed such that the fluorescence of a radiofluorogenic drop accurately represent the amount of radiotracer in the encapsulated cell. To achieve this goal, unwanted crosscontamination and nonspecific fluorescence activation of probe should all be minimized.
Specifically, radiofluorogenic probe, radiotracers, and ROS should all be hermetically encapsulated to avoid cross-contamination between drops. In addition, the radiofluorogenic fluorescence activation should be driven predominantly by extracellular radiation-induced ROS, not by intracellular ROS either inherently present or generated from oxidative stress. To address these requirements, we selected a hydrophilic radiofluorogenic probe, dichlorodihydrofluorescein (DCFH), as the ROS sensor. This helped create a strict compartmentalization of the three reactants (i.e., DCFH, FDG, and ROS) in the water drop and cell (Fig. 1b). The hydrophilic probe DCFH was observed to be strictly confined within the water phase, preventing cross-contamination between adjacent drops. In addition, the hydrophilicity of DCFH limited its transport into cells, thus mitigating the nonspecific fluorescence activation by cell-intrinsic ROS. As for ROS, the major radiation-induced radical, • OH, is short-lived (c.f., nanoseconds order of lifetime) and remains confined to the drops in which it is created. In contrast metabolism-driven intracellular ROS such as • O2 − and H2O2 are long lived, and therefore cell-containing drops were kept ice-cold to suppress ROS production and diffusion.
Additionally, we thoroughly optimized drop size and chemical composition of the water phase to achieve high signal-to-background ratio (SBR). These considerations are discussed in detail in Supplementary Material. We designed the incubation system such that radiofluorogenic conversion of a drop is proportional to the amount of radiotracer taken up by the encapsulated cell, with no unwanted cross-contamination between adjacent drops. Considering that 18 F has a positron range of Rmean = 0.6 mm 26 in water, it was important to separate the drops to avoid physical crosstalk. First, we kept the cell concentration in the sample as low as 13,000 -25,000 cells per mL, such that only 5-9% of drops contained a cell. Second, we collected the drops into a 2.8 m-long microbore tubing for incubation. The narrow lumen (inner diameter of 500 µm) forced the drops to line up along a few rows (Fig. 2d). Third, the flow rate ratio of the continuous phase to the dispersed phase in drop generation was 3 to 1, and the larger volume of the continuous phase further spread out the drops in the incubation tubing. Together, these features of the assay provided significant spatial distancing between cell-containing drops, minimizing physical crosstalk. Other approaches were implemented for optimal incubation of radioactive cells within droplets, which are detailed as Supplementary Material including Fig. S1. After incubation, the box was placed upright to concentrate the drops by driving them upward through buoyancy along the inclining serpentine path (Fig. 2e). This drop packing procedure was crucial to achieve high throughput in drop reading by removing undesirable extra spacing between the drops.

c. Drop read
After incubation, we injected the drops into the second optofluidic microchip, called hereafter DropRead microchip, to measure the DCF fluorescence (Fig. 2f). The DropRead microchip was mounted on a microscope stage to monitor microfluidic operation and obtain brightfield and fluorescence images. We also used the microscope illumination for fluorescence excitation. A single excitation light source (480 nm wavelength) was sufficient to excite both DCF and QD at the same time. The emission peaks of DCF and QD were respectively 525 nm and 800 nm, and we did not observe any overlap between the two emission measurements.
The drops were arranged in a single file and spaced to be investigated one by one with no overlap. This spacing was achieved by adding a flow of continuous oil phase. The drops passed in front of two optical fibers, which were inserted into the DropRead microchip to collect fluorescence from QD and DCF, respectively. The blue excitation light was focused by a 40x objective lens onto the location shown in Fig. 2f, covering the two fiber ends. As the drops flowed in front of the optical fibers, QD fluorescence was collected via optical fiber 1 and measured by photomultiplier tube (PMT) 1, then, shortly afterwards, DCF fluorescence was collected via optical fiber 2 and measured by PMT 2 (Fig. 3a). We synchronized the two PMT measurements in time using a common trigger signal. The plots in Fig. 3b show raw PMT signals as a function of elapsed time. Peaks and valleys were identified, and their heights were measured using custom MATLAB scripts (See Supporting Material including Fig. S2 for more details).
Peak height was used as the signal (S) representing the fluorescence intensity of individual drops. Excluding outliers, valley heights were averaged to estimate background (B; i.e., intrinsic fluorescence of the continuous oil phase) and the normalized drop fluorescence was computed as The fluorescence peaks from both PMTs were analyzed. Typically, the DCF fluorescence peak (PMT 2) had a width of 12 ms and appeared 4 ms after the corresponding QD peak (PMT 1). The combination of the two PMT signals enabled us to differentiate cell-containing drops from empty ones. Additionally, if two or more peaks were detected within 6 ms in the PMT 1 signal, we assumed the drop contained multiple cells and excluded the measurement from the dataset. The measured cell encapsulation probability in these experiments followed Poisson statistics, as expected for random encapsulation (see Supplementary Material including Tables S1 and S2 for more details).
The proposed optofluidic configuration offers crucial advantages in radiofluorogenic drop read over conventional epifluorescence microscopy. First, the proposed system is capable of detecting the dim fluorescence of empty drops and drops containing non-radioactive cells with high sensitivity. The measurement of these drops is necessary to accurately capture the full range of cellular heterogeneity, including cells taking up minimal amount of radiotracer. Minimal exposure of drop to excitation light is also important to minimize unwanted photo-activation of DCFH, which contributes to higher background. The resident time of each drop under excitation window was approximately 12 ms, which is much shorter than the time required to image a field of view using fluorescence microscopy. In addition, high sensitivity was achieved in our design by employing sensitive PMT photodetectors, which not only detected weakly radioactive single cells but also captured single cells labeled with minimal amount of QDs. Finally, thanks to precise microfabrication techniques, our system measures each drop at the same location, thus improving reproducibility compared to microscope set-up that are subject to variations in excitation and light collection efficiency over the field of view. We here describe why we chose DCFH for FRCM among many ROS probes available. Our foremost consideration was that DCFH is hydrophilic and, therefore, remained confined to the water drops and did not leak out to continuous phase, at least over 12 hours. In contrast, many other ROS probes (e.g., Amplex Red) are amphiphilic and thus prone to leaking out of aqueous drops. For example, Gallina et al. observed that DHRh123 partitioned to the oil phase and adsorbed onto the PDMS microchannel wall, reducing sensor concentration within the drop while increasing background fluorescence. Despite the notable advantage of using DCFH for a droplet assay, Gallina et al. ruled it out for use in their radiofluorogenic assay because the authors believed the sensitivity of DCFH to ionizing radiation drastically deteriorated over time 21 . However, repeating the same assay we achieved far more stable reagent than reported by Gallina et al. (Table 1). The data plots and linear regression analysis for the measured sensitivity and stability in Table 1 are respectively shown in Fig. S3. Detailed protocol is also discussed in Supplementary Material. In Fig. 4a, the background increases over time despite the absence of ionizing radiation, which we ascribe to nonspecific activation induced by excitation light and oxygen. In order to assess the photo-activation of DCFH induced by the excitation light, we performed the same kinetic spectroscopy and compared fluorescence increment rate for varying the time intervals between measurements and found that DCFH fluorescence increased 2.4 times faster when excitation light was given every 1 hour versus 20 hours (Fig. S4a). In addition, to assess the influence of oxygen on the nonspecific activation of DCFH, we compared two DCFH samples, one covered with paraffin oil on top, and one with no oil cap so that oxygen could freely diffuse into the test solution. DCFH fluorescence intensities increased faster and reached larger values when the samples were not capped with oil (Fig. S4b). These two experimental results confirm that both light and oxygen contribute to the nonspecific fluorescence activation of DCFH, as also observed previously by others 25,27 . We also note that the higher the DCFH concentrations, the larger the slopes (Fig. 4a). Linear regression shows that the slope is linearly proportional to the DCFH concentration (Fig. S4c).
When FDG is added to the solution, the fluorescence of the mixture increases more quickly than for the control DCFH samples (Fig. 4b). This is because radioactive decay of 18 (Fig. 5a). No significant increase of background fluorescence was observed in the waterphase, suggesting minimal leakage of cellular ROS. We observed that some encapsulated cells exhibited visible fluorescence, and we attribute this to DCFH which was partitioned into the cell and converted to DCF in reaction to cell-intrinsic ROS. In contrast, when cells were incubated with FDG, we observed a large difference in fluorescence intensity between empty drops and cell-containing drops (Fig. 5b). PMT measurements acquired using FRCM show a similar trend: for non-radioactive cells, we found little difference between drops containing cells and empty drops (Fig. 5c), whereas for cells incubated with FDG, there is a clear difference in terms of peak heights between empty drops and cell-containing drops. (Fig. 5d). The results clearly demonstrate that the increase in drop fluorescence measured by FRCM is directly caused by the radiotracers taken up by the cells.

Initial bulk radioactivity measurements confirmed that GLUT1 knockdown cells took up 53%
less FDG compared to the control shSCR cells (Fig. 6b). However, unlike FRCM, bulk gamma counting does not provide information about heterogeneity at the cellular level. The radioactivity of the single cells from both the groups were therefore measured using FRCM. By doing so, we aimed to confirm our earlier results showing that the FRCM signal was specific to FDG uptake.
Second, we aimed to demonstrate the radiometric accuracy of the proposed technique and show that it could differentiate between two different levels of radiotracer uptake by single cells.
The half-violin box plot in Fig. 6c summarizes the drop read results for all the experiments. In the plot, each dot corresponds to the pulse height measured by the PMT for a single fluorescent drop. The shaded area on the right-hand side shows the frequency distribution of the drop fluorescence measurements. The two grey-colored data sets are measured from the control experiment, which compared empty drops and drops containing non-radioactive cells. As previously mentioned, we attribute the small increase in fluorescence in cell-containing drops to DCFH partitioned into the cell and activated. We believe the any leakage of long-lived ROS out of an encapsulated cell was significantly suppressed as cells were kept ice-cold throughout the experiment. The red-colored data represent the FDG uptake of control cells treated with scrambled shRNA, and the blue-colored that of GLUT1 knockdown cells. As expected, on average, the drops containing radioactive single cells (i.e., radiolabeled shSCR and shGLUT1) exhibited much brighter fluorescence than the drops containing non-radioactive cells. In addition, the FRCM signal was significantly lower when GLUT1 was knocked down in cells (P value < 10 -4 ). On average, FRCM pulse height signal was 52% lower for the shGLUT1 cells compared to the control shSCR cells. This decrease in comparable to the 53% decrease in uptake measured using bulk gamma counting, suggesting that FRCM is a quantitative approach. This finding confirms that the increase in drop fluorescence was specifically due to FDG taken up by cells via GLUT1. This also demonstrated that the proposed technique has excellent radiometric accuracy, being capable of successfully differentiating the weak and strong levels of radiolabeling.
The FRCM results also highlight notable heterogeneity in cellular metabolism, as shown by the wide distribution of values in Fig. 6c. Using RLM, we previously found large heterogeneity in FDG uptake, even among cells from the same population 29 . This heterogeneity reflects the tremendous metabolic plasticity of cancer cells, which are able to use a variety of different metabolic substrates such as glucose and glutamine to satisfy their energetic requirements. To quantify the level of heterogeneity in these cell lines, we computed the coefficient of variation for both cell population and found it to be 50% for the control cells and 46% for the GLUT1 knockdown cells. Another remarkable feature of the data is the biphasic distribution of FDG uptake for shGLUT1 cells. This unusual distribution suggests that this sample is likely composed of two different sub-populations of cells. The first subpopulation of cells shows almost no detectable FDG uptake based on the pulse-height signal. This may be because GLUT1 expression was efficiently suppressed. These cells likely exist in a state of metabolic quiescence or may rely on other metabolic fuels such as glutamine. The second subpopulation contains cells that have higher FDG uptake, which may be due to the variable efficiency of GLUT1 silencing.
These cells may also transport FDG through other glucose transporters such as GLUT3.

CONCLUSION
We present a method of quantifying radiolabeled small molecules in single cells through the permanent activation of radiofluorogenic droplets. By converting stochastic radioactive decays into a continuous fluorescence signal, this approach allows rapid, high-throughput radiometry of single cells. The use of radiofluorogenesis significantly reduced exposure time per cell, from 30 s/cell for RLM down to 100 ms/cell of the proposed FRCM. This 100-fold reduction in exposure time per cell allowed rapid, high-throughput radiometry. FRCM successfully measured over 500 single cells in less than a minute. Importantly, the continuous-flow platform also allows integration with downstream cell sorting for post-analysis assays such as RNA sequencing.
Another notable advantage is that the proposed platform is versatile and compatible with any PET or SPECT radiotracer. In this proof of concept, we used FDG and measured the metabolic flux of glucose in single cells. But other radiotracers can be used to measure, for instance, radiolabeled drug incorporation, substrate transport, or therapeutic isotope uptake in single cells.
We can also use additional fluorescent probes for multiplexing. Accordingly, the extent of applications of this technique is not limited to radiopharmaceutical studies but can also include additional fluorescence-based single cell assay. The platform would be useful for pharmaceutical study assessing cellular heterogeneity. For example, in-vitro tumor models are used to test the efficacy of pharmaceuticals prior to clinical tests. These in-vitro models can be either 2D culture or 3D culture including patient-derived organoids or spheroids. In our work, as a proof of concept, we used 2D in-vitro tumor model to assess cellular heterogeneity in metabolism. For future work, primary single cells derived from human tumors or complex 3D tumor models can be investigated using our proposed technique. We believe that the approach has potential applications ranging from analytical assays for cancer and other diseases to the development of small-molecule drugs.

A. Microchip fabrication
Microfluidic platforms capable of generating and reading droplets were designed and built.
Polydimethylsiloxane (PDMS) microfluidic chips with channel depth modulations were fabricated using replica molding protocols. The reusable molds were fabricated using Nanoscribe (Nanoscribe Photonics Professional GT, Nanoscribe GmbH, Germany), a submicron resolution 3D printer based on two-photon polymerization of negative photoresist. We find a number of advantages in using Nanoscribe for microchannel mold fabrication over conventional photolithography methods. First, accurate channel depth modulation was easily achieved whereas the conventional way of spin coating liquid-type negative photoresist is influenced by slight differences in temperature, humidity and device control and thus prone to inaccurate and non-uniform depth modulation. Second, sharp rectangle cross-sectional shape of channels was easily achieved whereas a mask-based photolithography is prone to creating undesirable trapezoidal shapes.
We note that cleaning of silicon wafer substrate with 500W oxygen plasma treatment for 10 minutes (Drytek2 DRIE100 plasma etcher) prior to Nanoscribe fabrication significantly enhanced adhesion of photoresist onto the substrate surface and thus durability of the mold.
The mold surface was rendered hydrophobic with trichloro(1H, 1H, 2H, 2Hperfluorooctyl)silane (Sigma Aldrich, MO, USA). Then, the PDMS base (Sylgard 184, Dow corning) and curing agent were mixed in a 10:1 ratio, poured over the mold, degassed in vacuum chamber, and cured at ~ 60°C for over 4 hr. We peeled off the cured PDMS from the mold and punctured holes for inlet and outlet ports using a 1 mm diameter disposable biopsy punch

C. Droplet incubation and incubator fabrication
The droplets were collected into a 2.8-meter long polytetrafluoroethylene (PTFE) tubing and incubated in the dark and 4°C for 4 hours. The inside of the tubing was rendered hydrophobic with Novec TM 1720 Electronic grade Coating prior to use. The tubing holder and the box were 3D printed using Ultimaker S5 (Ultimaker B.V., Utrecht, Netherlands).

D. Droplet read and image acquisition
After incubation, we injected the droplets into another optofluidic microchip to read their fluorescence. The drops were arranged and spaced in a single file to be investigated one by one with no crosstalk. Red and green fluorescence emission from each droplet was collected through two optical fibers and measured using two photomultiplier tubes (PMTs). The red fluorescence emitted from quantum dots was delivered first through the 1 st fiber, which was used to confirm single cell encapsulation. Within 15 ms, the green fluorescence emitted from radiofluorogenic probe was delivered through the 2 nd fiber, which informed on the amount of FDG uptake of the corresponding single cell. Immersion oil was placed in between the fiber end and PDMS wall to reduce scattering and reflection and achieve a good optical coupling. The microscope light was used as a fluorescence excitation source and focused on a droplet investigation area by 40x objective lens (as shown in Figure 2f). The red emission was detected using a Hamamatsu PMT concentration of DCFH-DA was 10 μM. We caution that, as opposed to the protocol described by Sleiman et al. 30  were seeded into 6-well plate and incubated for 24 hours. Then cell media were removed, and 1 mL of virus particles were added with 1μg/mL of polybrene and 1 mL of regular DMEM with 10% FBS and 1% antibiotic-antimycotic. After incubating for 48 hours, knockdown cells were selected with 2 μg/mL puromycin for 2 weeks. GLUT1 knockdown was confirmed by Western blot after removing N-glycans using PNGase F.

N. Western blot of shGLUT1 and shSCR
To determine SLC2A1 knockdown in MDA-MB-231 cells, GLUT1 protein expression was evaluated using Western blot. Since GLUT1 is highly glycosylated, N-glycans were removed using PNGase F. As the first procedure, whole cell protein was extracted with RIPA buffer supplemented with protease inhibitor cocktail. Then, 10 μg of protein was mixed with 2 μL of PNGase F (NEB #P0707S) and incubated at 37°C overnight. After incubation, we added sample buffer and ran BOLT 4−12% Bis-Tris gel for western blot (Thermo Fisher Scientific #NW04120). After a transfer to nitrocellulose membrane using Trans-Blot Turbo transfer system (BioRad), the membrane was probed with anti-GLUT1 antibody (Abcam #ab115730). Figure 1 Illustrations showing production of reactive oxygen species (ROS) from radioactive beta decay and radio uorogenic conversion from the ROS. (a) A single radioactive cell is encapsulated into a water-in-oil drop. As the radiotracer decays, the emitted beta radiation creates ROS such as hydroxyl radicals, which in turn mediate the radio uorogenic conversion. (b) Proton (P) decays into neutron (N), emitting positron (e+) and neutrino (v). The emitted positron travels a nite distance, interacting with water molecules along its path and creating reactive oxygen species (ROS), until it annihilates with an electron (e-), resulting in two antiparallel 511 keV photons. (c) Non uorescent dichlorodihydro uorescein (DCFH) is converted into uorescent dichloro uorescein (DCF) after losing two protons from interacting with radiation-induced ROS.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. SIFlowRadioCytometryForSubmission.pdf