Novel microuidic platform accelerates processing of tissue into single cells for molecular analysis and primary culture models

Tissues are composed of highly heterogeneous mixtures of cell subtypes, and this diversity is increasingly being characterized using high-throughput single cell analysis methods. However, these efforts are hindered by the fact that tissues must rst be dissociated into single cell suspensions that are viable and still accurately represent phenotypes from the original tissue. Current methods for breaking down tissues are inecient, labor-intensive, subject to high variability, and potentially biased towards cell subtypes that are easier to release. Here, we present a microuidic platform consisting of three different tissue processing technologies that can perform the complete tissue to single cell workow, including digestion, disaggregation, and ltration. First, we developed a new microuidic digestion device that can be loaded with minced tissue specimens quickly and easily, and then use the combination of proteolytic enzyme activity and uid shear forces to accelerate tissue breakdown. Next, we integrated dissociation and lter technologies into a single device, which enhanced single cell numbers and fully prepared the sample for single cell analysis. The nal multi-device platform was then evaluated using a diverse array of tissue types that exhibited a wide range of properties. For murine kidney and mammary tumor, we found that microuidic processing produced 2.5-fold more single, viable cells. Single cell RNA sequencing (scRNA-seq) further revealed that device processing enriched for endothelial cells, broblasts, and basal epithelium, and did not increase stress responses. For murine liver and heart, which are softer tissues containing fragile cell types, processing time could be reduced to 15 min, and even as short as 1 min. We also demonstrated that periodic recovery at dened time intervals produced substantially more hepatocytes and cardiomyocytes than continuous operation, most likely by preventing damage to fragile cell types. In future work, we will seek to integrate additional operations such as upstream tissue preparation and downstream microuidic cell sorting and detection to create powerful point-of-care single cell diagnostic platforms.


Introduction
Tissues are highly complex ecosystems containing a diverse array of cell subtypes. Signi cant variation can also arise within a given subtype due to differences in activation state, genetic mutations, epigenetic distinctions, stochastic events, and microenvironmental factors. 1,2 This has led to a rapid growth in studies attempting to capture cellular heterogeneity, and thereby gain a better understanding of tissue and organ development, normal function, and disease pathogenesis. [3][4][5][6][7][8][9] For example, in the context of cancer, intratumor heterogeneity is a key indicator of disease progression, metastasis, and the development of drug resistance. [10][11][12][13][14] High-throughput single cell analysis methods such as ow cytometry, mass cytometry, and single cell RNA sequencing (scRNA-seq) are ideal for identifying single cells in a comprehensive manner based on molecular information, 15,16 and these methods have already begun to transform our understanding of complex tissues by enabling identi cation of previously unknown cell types and states. 8, [17][18][19] However, a critical barrier to these efforts is the need to rst process tissues into a suspension of single cells. Current methods involve mincing, digestion, disaggregation, and ltering that are labor intensive, time-consuming, ine cient, and highly variable. 20,21 Thus, new approaches and technologies are critically needed to ensure reliability and wide-spread adoption of single cell analysis methods for tissues. This would be particularly important for translating single cell diagnostics to human specimens in clinical settings. Moreover, improved tissue dissociation would make it faster and easier to extract primary cells for ex vivo drug screening, engineered tissue constructs, and stem/progenitor cell therapies. [22][23][24][25] Patient-derived organ-on-a-chip models, which seek to recapitulate complex native tissues for personalized drug testing, are a particularly exciting future direction that could be enabled by improved tissue dissociation. 22,[26][27][28][29][30] scRNA-seq has recently emerged as a powerful and widely adaptable analysis technique that provides the full transcriptome of individual cells. This has enabled comprehensive cell reference maps, or atlases, to be generated for normal and diseased tissues, as well as identi cation of previously unknown cell subtypes or functional states. 31,32 For example, an atlas recently generated for normal murine kidney uncovered a new collecting duct cell with a transitional phenotype and unexpected level of cellular plasticity. 4 Moreover, an atlas of primary human breast epithelium linked distinct epithelial cell populations to known breast cancer subtypes, suggesting that these subtypes may develop from different cells of origin. 3 For melanoma, scRNA-seq was used to identify three transcriptionally distinct states, one of which was drug sensitive, and further demonstrated that drug resistance could be delayed using computationally optimized therapy schedules. 33 While scRNA-seq is clearly a powerful diagnostic modality, the process of breaking down the tissue into single cells can introduce confounding factors that may negatively in uence data quality and reliability. One factor is the lack of standardization, which can lead to substantial variation across different research groups and tissue types. Another signi cant concern is that incomplete break down could bias results towards cell types that are easier to liberate. A recent study utilizing single nuclei RNA sequencing (snRNA-seq) with murine kidney samples found that endothelial cells and mesangial cells were underrepresented in scRNA-seq data. 34 Finally, lengthy enzymatic digestion times have been shown to alter transcriptomic signatures and generate stress responses that interfere with cell classi cation. [35][36][37][38][39] Addressing these concerns would help propel the exciting eld of scRNA-seq into the future for tissue atlasing and disease diagnostics.
Micro uidic technologies have advanced the elds of biology and medicine by miniaturizing devices to the scale of cellular samples and enabling precise sample manipulation. [40][41][42][43][44] Most of this work has focused on manipulating and analyzing single cells. [44][45][46][47][48] Only a small number of studies have addressed tissue processing, and even fewer have focused on breaking down tissue into smaller constituents. [49][50][51] We developed a micro uidic device that speci cally focused on breaking down cellular aggregates into single cells. 52,53 This dissociation device contained a network of branching channels that progressively decreased in size down to ~ 100 µm, and contained repeated expansions and constrictions to break down aggregates using shear forces. We then developed a device for on-chip tissue digestion using the combination of shear forces and proteolytic enzymes. 54 Finally, we developed a lter device containing nylon mesh membranes that removed large tissue fragments, while also dissociating smaller cell aggregates and clusters. 55 The micro uidic digestion, dissociation, and lter devices each enhanced single cell recovery when operated independently. To date, however, we have not combined these technologies to maximize performance and execute a complete tissue processing work ow on-chip. Moreover, we have not validated micro uidically-processed cell suspensions using scRNA-seq.
In this work, we present a micro uidic platform comprised of three different tissue-processing technologies that enhances break down and produces single cell suspensions that are immediately ready for downstream single cell analysis or other use. First, we design a new digestion device that can be loaded with minced tissue and operated with minimal user interaction. Next, we integrate the dissociation and lter technologies into a single unit, and optimize the two-device platform using murine kidney to produce single cells more quickly and in higher numbers than traditional methods. Using the optimized protocol, we evaluate different tissue types using two single cell analysis methods. For murine kidney and breast tumor tissues, micro uidic processing can produce ~ 2.5-fold more epithelial cells and leukocytes, and > 5-fold more endothelial cells, without affecting viability. Using scRNA-seq, we show that device processed samples are highly enriched for endothelial cells, broblasts, and basal epithelium. We also demonstrate that stress responses are not induced in any cell type, and can even be reduced if shorter processing times are employed. For murine liver and heart, signi cant single cell numbers are obtained after only 15 min, and even as short as 1 minute. Interestingly, we nd that substantially more hepatocytes and cardiomyocytes are obtained if sample is recovered at discrete intervals, most likely because these cell types are sensitive to shear forces. Importantly, the micro uidic platform can signi cantly shorten processing time or enhance single cell recovery for all tissue types studies, and in some cases accomplish both, without affecting viability. Furthermore, the entire tissue processing work ow is performed in an automated and reliable fashion. Thus, our micro uidic platform holds exciting potential to advance diverse applications that require the liberation of single cells from tissues.

Device Design And Fabrication
We redesigned our original digestion device so that manual assembly was not required. Instead, minced tissue is loaded through a port at the top of the device, which can then be sealed using a cap or stopcock.
Scalpel mincing of tissue into ~ 1 mm 3 pieces is ubiquitous, and therefore this format will be compatible with a wide array of tissue types and dissociation protocols. The full design layout of the new minced tissue digestion device is shown in Fig. 1A, including the loading port, a chamber that retains the tissue in place, and uidic channels that administer uid shear forces and deliver proteolytic enzymes. These features were arranged across six layers of hard plastic, including two uidic channel layers, two "via" layers, a top end cap with hose barbs and loading port, and a bottom end cap. The tissue chamber is in the uppermost uidic layer, directly beneath the loading port and a 2.5 mm diameter via, and a detailed schematic is shown in Fig. 1B. We employed a square geometry, with 5 mm length and width, to allow tissue to be evenly distributed during loading. Chamber height was 1.5 mm, slightly larger than minced tissue, to prevent clogging during sample loading and device operation. Fluidic channels were placed upstream and downstream of the tissue chamber, and in both cases, we employed four channels that were 250 µm wide. The symmetric channel design was chosen for the minced format because there is a greater emphasis on prevention of clogging. We also extended channel length to 4 mm to prevent larger tissue pieces from squeezing all the way through, but ared the end to make it easier to connect with the underlying via layer.
The dissociation and lter devices will process tissue fragments and cell aggregates that are small enough to leave the tissue chamber of the digestion device. This includes disaggregation via shear forces generated within the branching channel array and via physical interaction with nylon mesh lters. 52,55 For this work, we have integrated the dissociation and lter devices into a single unit to minimize holdup volume and simplify operation. The original designs were both arranged across 7 layers, and thus it was straightforward to combine them, as shown in (Fig. 1C).
The minced digestion and integrated dissociation/ lter devices were fabricated using a commercial laminate process, with channel features laser micro-machined into hard plastic (PMMA or PET). All layers and other components were then aligned and bonded using pressure sensitive adhesive. Photographs of the fabricated devices are shown in Figs. 1D and E.

Platform Optimization Using Murine Kidney
We rst evaluated the new minced digestion device using adult murine kidney samples. The kidney is a complex organ composed of anatomically and functionally distinct structures, and adult kidney tissue has a dense extracellular matrix that is challenging to dissociate into single cells. 52,55,56 Freshly dissected kidneys were minced using a scalpel to ~ 1 mm 3 pieces and loaded into the minced digestion device through the luer port. The device and tubing were then primed with PBS containing 0.25% type I collagenase, the luer input port was sealed using a stopcock, and uid was recirculated through the device using a peristaltic pump. We initially tested ow rates of 10 and 20 mL/min, which were used in previous work. 52,54,55 After 15 or 60 min of recirculation, sample was collected, washed, and genomic DNA (gDNA) was extracted to assess total cell recovery. A control was minced and gDNA was directly extracted to provide an upper recovery limit. At 10 mL/min, gDNA was ~ 15% and 60% of the control after 15 and 60 min, respectively ( Fig. 2A). Increasing ow rate to 20 mL/min improved results to ~ 40% and 85%, respectively. Images of the tissue chamber were captured at the end of each experiment, and representative results are shown in Fig. 2B. We consistently observed tissue remaining in the chamber or adjacent channels at 10 mL/min, corroborating low gDNA recovery results. After 60 min at 20 mL/min, only a small amount of tissue was found within channels/vias, which helps explain why gDNA recovery was slightly lower than the control. Another possibility is that cells were damaged or destroyed during recirculation. To address this concern, we recirculated MCF-7 breast cancer cells through the system and assessed cell number and viability (see Supplementary Information, Fig. S1). We observed that cell recovery decreased by ~ 10% after recirculating through the digestion device, regardless of ow rate or time. Moreover, results were similar after recirculating through the peristaltic pump alone, and cell viability remained high for all conditions tested. This con rms that sample loss was most likely related to hold-up within the system or transfer steps, and not damage. Since 20 mL/min was more effective at clearing the tissue chamber and isolating gDNA, it was used for the remainder of the study.
Next, we analyzed single cells using ow cytometry. Cell suspensions were labeled using a panel of antibodies and uorescent probes speci c for EpCAM (epithelial cells), TER119 (red blood cells), CD45 (leukocytes), and 7-AAD (live/dead), as listed in Table 1. We found that single epithelial cell numbers increased with processing time, from 15 to 60 min, producing up to ~ 14,000 cells/mg tissue (Fig. 2C). This represents a 1.5-fold increase relative to the control, which was digested for 60 min under constant agitation, followed by repeated pipetting and vortexing to replicate standard tissue dissociation protocols. We also note that after only 15 min in the digestion device, epithelial cells were statistically similar to the control digested for 4-fold longer time. We also investigated an interval operation format, which involved processing for short time periods, eluting the cell suspension, replacing collagenase in the digestion device, and continuing recirculation. We observed that epithelial cell numbers accumulated through each time point of interval operation in a comparable manner to static operation. This demonstrates that interval collection does not compromise results, and suggests that epithelial cells can withstand longterm recirculation. Epithelial cell viability was ~ 80% for all control and device conditions, further con rming that device processing did not adversely affect cells (Fig. 2D). Results in terms of cell number and viability were similar for leukocytes (see Supplementary Information, Fig. S2A and B). We then investigated whether the integrated dissociation/ lter device can further enhance single cell yield following the digestion device. We performed initial tests using the MCF-7 model, and found that recirculation at 20 mL/min, even for short periods of time, resulted in low viability (see Supplementary  Information, Fig. S1). At 10 mL/min, single cells increased by ~ 20% after 30 s of recirculation, with no change in viability, which is similar to previous work using a syringe pump. 53 Longer recirculation times enhanced single cell numbers but decreased viability. Thus, we selected to evaluate short recirculation times at 10 mL/min using minced kidney that had been processed using the digestion device for 15 min.
As a nal step, sample was passed through the nylon mesh membranes in the lter component at 10 mL/min. Single epithelial cell recovery numbers are presented in Fig. 2E. The digestion device produced 4-fold more single cells than the control that was also digested for 15 min. A single pass through the integrated device increased single epithelial cells by ~ 40% compared to digestion alone, which was ~ 5.5-fold greater than the control. Recirculation through the branching channel array produced fewer cells than the single pass. Epithelial cell viability was ~ 85-90% for all conditions (Fig. 2F). Similar results were observed for leukocytes (See Supplementary Information, Fig. S2). Based on these results, selected a single pass through the integrated dissociation/ ltration device for all work with kidney. We also note that the integrated device obviates the need for a cell straining step prior to ow cytometry.

Single Cell Analysis Of Murine Kidney
We further evaluated kidney under different digestion times using the full micro uidic platform. We also added endothelial cells (via CD31, Table 1) to the ow cytometry panel, since they are di cult to isolate using traditional dissociation methods. 34 Minced tissue was loaded into the digestion device and processed under static (15 or 60 min) or interval (1, 15, and 60 min) formats, and then passed through the integrated dissociation/ lter device one time. Controls were minced, digested (15 or 60 min), disaggregated by vortexing/pipetting, and ltered using a cell strainer. Results for epithelial cells are presented in Fig. 3A, and are generally similar to optimization studies ( Fig. 2C), although epithelial cells increased to ~ 20,000/mg tissue. This was ~ 40% higher than the optimization study due to the integrated dissociation/ lter, and overall more than double the 60 min control. Surprisingly, the 1 min interval produced ~ 1500 epithelial cells/mg, which was similar to the 15 min control. This time point was chosen primarily to eliminate erythrocytes (see Supplementary Information, Fig. S3). Device processing was even more effective for endothelial cells (Fig. 3B), which exceeded the 60 min control by > 5-fold. Findings for leukocytes ( Fig. 3C) were generally similar to epithelial cells. We note a slight decrease in total cell recovery for the interval format relative to the 60 min static condition for all cell types, although this was not statistically signi cant. This modest decrease may have been due to sample loss during transfer and/or priming steps. Alternatively, cell clusters may have eluted in the early intervals, which would have otherwise been broken down if they remained within the digestion device. Population distributions for each cell type and processing condition are shown in Fig. 3D. Relative to the 60 min control, endothelial cells were enriched for all device conditions except the 1 min interval. Leukocytes were present at similar levels except in the 15 min control, where they were under-represented. Viability for all three cell types after device processing were similar to or exceeded controls (see Supplementary  Information, Fig. S4).
Next we performed scRNA-seq, which has been used to catalogue the diverse cell types residing within murine kidney and create atlases. 4, 56-59 Kidney tissue was processed using the device platform and collected at 15 and 60 min intervals, and we also evaluated the 60 min control. Live single cells were isolated from debris and dead cells using uorescence-activated cell sorting (FACS), loaded onto a droplet-enabled 10X Chromium platform, and 34,034 cells were sequenced at an average depth of ~ 60,000 reads/cell. After ltering, we used Seurat to identify (Fig. 4A) and annotate (see Supplementary   Information, Fig. S5) seven cell clusters. 60 This included two clusters of proximal tubules (convoluted, or S1, and straight, or S2-S3), endothelial cells, macrophages, B lymphocytes, and T lymphocytes. The nal cluster was heterogenous, and included cells from the distal convoluted tubule (DCT), Loop of Henle (LOH), and collecting duct (CD), as well as mesangial cells (MC). All seven clusters were represented in control and device conditions (see Supplementary Information, Fig S6A). The relative number of cells in each cluster are shown in Fig. 4B. Proximal tubules were the predominant cell population, representing ~ 53% of the control, which closely matched a recently published mouse kidney atlas. 4 Proximal tubules were further enriched in the 15 min device condition, comprising ~ 86% of the cell suspension. The other cell populations were under-represented relative to the control, most by ~ 2-fold, but reaching as high as 8-fold for macrophages. However, it is unclear whether this was caused by diminished recovery or simply dilution by proximal tubules. The 60 min device interval only contained ~ 29% proximal tubules, but we surmise that most had already been removed in the 15 min interval. Endothelial cells were clearly enriched at 60 min, increasing to ~ 25% of the suspension, while remaining cell types remained close to control values. Similar trends were observed within the DCT, LOH, DC, and MC sub-clusters (see Supplementary Information, Figs. S6B and C). To compare population percentages obtained from scRNAseq ( Fig. 4B) and ow cytometry (Fig. 3D), consideration must be given to which cell populations were likely to express each marker. CD45 and CD31 gene expression was well correlated with the appropriate clusters (see Supplementary Information, Fig. S7). For EpCAM, DCT and CD cells have been shown to express at high levels, while proximal tubules and LOH cells ranged from low to undetectable. 61 Inspection of sequencing results indicated that EpCAM was highly expressed by at least a subset of the DCT, LOH, CD, & MC sub-clusters (see Supplementary Information, Fig. S7). Interestingly, proximal tubules were predominantly EpCAM-negative, but this could be explained by low basal expression and/or a potential secondary factor such as low protein turnover. We used the brightest uorophore to stain EpCAM, phycoerithrin (PE), to help discern low level expression, but it is possible that some cell proximal tubules remained undetectable. Assuming all proximal tubule, DCT, LOH, CD, & MC clusters were EpCAM+, we calculated population percentages of ~ 62, 88, and 40% for the control, 15 m device, and 60 m device conditions, respectively. This is directly in line with ow cytometry results for the 15 m device case, but considerably lower for the others. We note that if ow cytometry missed any of these cell types due to low EpCAM expression, it would only widen the disparity. Instead, we contend that the comprehensive manner in which scRNA-seq identi es cell types is superior to ow cytometry, particularly when a clear positive biomarker for all cell sub-populations is lacking. Flow cytometry is better suited to cell counting, however, and based on those results, device processing consistently produced comparable numbers of cells at 15 min and at least 50% more cells at 60 min, relative to the 60 min control. We used these estimates as weighting factors (1x for 15 min, 1.5x for 60 min), along with percentages in Fig. 4B, to calculate aggregate device platform yields (see Supplementary Information, Table S1). Total endothelial cell recovery was ~ 4-fold greater than the control, while other cell types were ~ 2-to 2.5-fold greater, which all match ow cytometry ( Fig. 3A-C). While the true weighing factors may be slightly different, it does appear that the relative numbers between control and device platform are consistent between ow cytometry and scRNA-sEq. However, the relative numbers across cell types varies considerably, which may have resulted from biasing during FACS collection or droplet loading in the 10X Chromium system, which have been documented previously. 62 Our results suggest a preferential selection of endothelial cells and leukocytes during these steps. Nevertheless, we conclude that our micro uidic device platform can address cellspeci c biasing of kidney tissue during isolation by enriching endothelial cells, which have been shown to be underrepresented using standard tissue dissociation work ows, 34 while maintaining all other cell subtypes at similar numbers.
Lastly, we evaluated stress response genes, which can interfere with cell identi cation using transcriptomic information. Induction of stress responses have been linked to conventional tissue dissociation protocols. 34 Importantly, we found no evidence that exposure to uid shear stresses within the digestion device heightened the stress response for any cell type. This suggests that time was the predominant factor, which can be mitigated using the interval concept in our micro uidic platform.

Processing And Single Cell Analysis Of Murine Breast Tumor Tissue
Solid tumors can exhibit high degrees of intratumoral heterogeneity, which has been directly implicated in cancer progression, metastasis, and the development of drug resistance. 12,13 This heterogeneity has successfully been captured using scRNA-seq and linked to survival for glioblastoma, drug resistance in melanoma, and prognosis for colorectal cancer. 3, 63-67 Moreover, it is expected that expanded application of scRNA-seq in clinical settings will soon emerge to provide molecular and cellular information for guiding personalized therapies. 68 Due to abnormal extracellular matrix composition and density, however, tumor tissues are considered to be amongst the most di cult epithelial tissues to dissociate. 55,69, 70 We evaluated micro uidic processing of mammary tumors that spontaneously arise in MMTV-PyMT transgenic mice. We rst optimized the minced digestion and integrated dissociation/ lter devices separately. The digestion device generated ~ 2-fold more EpCAM + epithelial cells than the controls after 15 and 30 min, and the difference extended to 2.5-fold after 60 min (see Supplementary Information, Fig.  S8A). Viability was higher for device processed samples than controls at all time points (see Supplementary Information, Fig. S8B). We then tested the integrated dissociation/ lter device and again found that a single pass was optimal (see Supplementary Information, Fig. S8C and 9D). In this case, recirculation for 1 and 4 min produced similar cell numbers, but with lower viability.
Results for the full micro uidic device platform are shown in Fig. 5, and were generally similar to kidney, but with 2-to 3-fold lower cell counts/mg tissue. However, the device platform still produced signi cantly more cells than controls. Epithelial cells were ~ 2-fold higher at both time points (Fig. 5A). Endothelial cells were again liberated more effectively by device processing, with 5-fold more cells recovered after 15 min and 10-fold more after 60 min (Fig. 5B). Leukocytes increased by 3-and 5-fold after 15 and 60 min, respectively (Fig. 5C). The interval format produced similar total epithelial cell and leukocyte numbers when compared to the corresponding static time point. However, ~ 30% more endothelial cells were obtained from intervals. We also note that a remarkably large number of epithelial cells (> 15%) were obtained at the 1 min interval. Relative population percentages are shown in Fig. 5D. Device processing enriched for endothelial cells and leukocytes at all but the 1 min time point, which remained similar to controls. Viability for all three cell types were similar to the 15 min control and exceeded the 60 min control (see Supplementary Information, Fig. S9). Thus, the micro uidic platform liberated more single cells from tumor, while also better preserving cell viability.
We then performed scRNA-seq, again using the 15 and 60 min device intervals and the 60 min control. A total 24,527 cells were sequenced at an average of ~ 45,000 reads per cell. We identi ed 6 clusters corresponding to epithelial cells, macrophages, endothelial cells, T lymphocytes, broblasts, and granulocytes (Fig. 6A). Epithelial cells were the predominant cell population, representing 62.0% of control cells (Fig. 6B). Epithelial percentages increased slightly in the 15 min interval and decreased in the 60 min interval. We identi ed three sub-clusters within the epithelial population corresponding to luminal, basal, and proliferating luminal cells based on expression of Krt14, Krt18, and Mki67, respectively (see Supplementary Information, Fig. S10). The luminal sub-type dominated, as expected for MMTV-PyMT tumors. The basal subpopulation was enriched with device processing, while the proliferating luminal subpopulation was under-represented. These results suggest that basal epithelium is more di cult to dissociate. Comparing cell populations between scRNA-seq and ow cytometry was more straightforward since EpCAM, CD45, and CD31 were all correlated well with the expected cell types (see Supplementary  Information, Fig. S11). However, broblasts were not detected by ow cytometry, and account for a signi cant portion of the 60 min device condition. As with kidney, tumor epithelial percentages were signi cantly higher in ow cytometry data, which would further suggest biasing during sorting and/or droplet encapsulation. If we combine the population percentages in Fig. 6B with the same weighting factors used for kidney (1x for 15 min, 1.5x for 60 min), we can again calculate aggregate device platform yields (see Supplementary Information, Table S2). Differences for the device aggregate relative to the control were ~ 2-fold for epithelial cells and 2.5-to 3-fold for T lymphocytes and macrophages, which are all similar to ow cytometry results (Figs. 5A and C). Endothelial cells were ~ 4-fold greater for the device platform, which is lower than the 10-fold difference from ow symmetry (Fig. 5B). Notably, broblasts were enriched by 10-fold using the device platform. Our results con rm that tissue processing with the micro uidic device platform can improve isolation of all cell types by at least 2.5-fold, as well as di cult to liberate cell types such as endothelial cells, broblasts, and basal epithelium by 4-to 10-fold.
Finally, we determined stress response scores as described for kidney. The importance of stress responses can be heightened for tumor since some response genes, such as members of the Jun and Fos families, have been associated with metastatic progression and drug resistance. 38, 71-73 Stress response scores were similar across all cell types and conditions for tumor (Fig. 6C). It is possible that tumor cells are more sensitive to dissociation-induced transcriptional changes, and that even shorter intervals would be necessary to lower these responses.

Isolation Of Hepatocytes From Murine Liver
The liver plays a major role in drug metabolism and is frequently assessed for drug-induced toxicity. In fact, liver damage is one of the leading causes of post-approval drug withdrawal. 22,74,75 Thus, in vitro screening of drugs against primary liver tissue is a critical component of preclinical testing. Increasingly, organ-on-a-chip systems are being employed to better maintain hepatocyte functionality and activity in culture settings and to enable personalized testing on patient-derived primary cells. 27,76 While liver is softer and generally easier to dissociate, hepatocytes are well known to be fragile, and thus liver presents a unique dissociation challenge. 77 As such, we hypothesized that shorter device processing times would be effective for liver. For these experiments, murine liver was minced into 1 mm 3 pieces and hepatocytes were detected based on ASGPR1 expression. We rst processed liver using the minced digestion device for either 15 or 60 min. After 15 min, hepatocyte recovery was ~ 4-fold higher for the device than the comparable control (Fig. 7A). Continued digestion of the control increased hepatocyte numbers further. Counterintuitively, continued processing in the digestion device diminished hepatocyte yield by approximately half. We believe this nding was caused by the combination of two factors: softer liver tissue is effectively broken down at earlier time points and fragile hepatocytes are more sensitive to damage from recirculation. We also tested a single pass through the integrated dissociation/ ltration device, and found that hepatocyte recovery decreased. This was likely due to the large size of hepatocytes (~ 30 µm), which caused them to be retained or damaged by the 15 µm membrane. It also appears that damage may have been additive, as viability dropped to 45% after 60 min digestion device treatment and passing through the integrated device, while all other conditions were ~ 80% (Fig. 7B). Removing the 15 µm lter from the integrated dissociation/ lter device increased hepatocytes by 30% relative to the digestion device alone, and by nearly 3-fold relative to the control, while maintaining viability (see Supplementary Information, Fig. S12).
Based on the initial optimization studies, we concluded that the micro uidic device platform should utilize short processing times, and use the modi ed dissociation/ lter device with only the 50 µm lter. After 5 min digestion device processing, ~ 700 hepatocytes were recovered/mg liver tissue (Fig. 7C). This was 4-fold higher than the 15 min control and just slightly less than the 60 min control (~ 1000 hepatocytes/mg). Increasing digestion device processing time to 15 min enhanced hepatocyte recovery by 40%, to the same level as the 60 min control. The most striking results were observed under the interval format. After only 1 min, ~ 700 hepatocytes/mg tissue were recovered. Adding the 5 and 15 min intervals resulted in ~ 2400 hepatocytes/mg, for a ~ 2.5-fold enhancement relative to both the 60 min control and 15 min static conditions. Hepatocyte viability remained at 90% for controls and most device conditions (see Supplementary Information, Fig. S13A). Similar trends were observed for endothelial cells (Fig. 7D) and leukocytes (Fig. 7E), including signi cant recovery from the 1 min interval and enhanced overall cell numbers using the interval format. For endothelial cells, interval operation was ~ 1.5-fold higher than the 60 min control and 15 min static device cases. For leukocytes, static device operation produced > 2.5-fold more cells than the 60 min control, and interval operation further enhanced recovery to ~ 3.5-fold. Given the strong performance of the device platform with leukocytes and their relative abundance in liver compared to kidney and tumor, cell suspensions were enriched for leukocytes in comparison to the 60 min control (Fig. 7F). This was particularly true for the static time points and the 1 min interval. Interestingly, the three interval conditions contained very different representations of hepatocytes and leukocytes, suggesting that the choice of elution time could serve as a means to crudely select for one population over the other, if that was so desired. Viability for endothelial cells and leukocytes remained similar to or greater than controls (see Supplementary Information, Figs. S13B and C).
Taken together, the performance of the micro uidic processing platform with liver was quite unique relative to kidney and tumor. We believe that this caused by the fact that uid shear forces are needed to break down tissue, but can also damage some cell types that have already been liberated. All tissues require proper balancing of these effects. For softer tissues like liver, the balance must be shifted away from breakdown and towards preservation, particularly for sensitive hepatocytes, which can be accomplished using interval recovery. Endothelial cells and leukocytes also exhibited some sensitivity to over-processing, although to a lesser degree. It is unclear whether this nding can be generalized to other tissues, including kidney and tumor. Liver sinusoidal endothelial cells are highly specialized, with abundant fenestrae and no underlying basement membrane. 78 These properties could also make sinusoidal endothelial cells particularly sensitive to damage. For leukocytes, we did not distinguish between those that originated within the liver, which would predominantly be Kupffer cells, from those that came from blood, which may be less sensitive to shear. Future studies directly assessing Kupffer cells, as well as hepatic stellate cells, would be of high interest, particularly to make progress towards complex liver models that utilize multiple cell types. 74,79−81

Isolation Of Cardiomyocytes From Murine Heart
Heart failure is another leading cause of drug withdrawal from the market, combining with liver failure to account for ~ 70% of withdrawals. 74,75 Thus, there is robust interest in developing heart-on-chip technologies using primary cardiomyocytes for preclinical drug screening. 30,74,75,82,83 Cardiomyocytes have been shown to be highly sensitive to mechanical and enzymatic dissociation techniques. 84,85 In addition, they are disproportionately long in one direction, on the order of 100 µm and more. 86 For these experiments, murine heart was minced into ~ 1 mm 3 pieces and cardiomyocytes were detected based on Troponin T expression. Since Troponin T is an intracellular marker, we used a xable viability dye, Zombie Violet, in place of 7-AAD. Given potential concerns about cardiomyocyte size and shape, we rst tested the effect of lter pore size in the integrated dissociation/ ltration device. After 15 min processing with the minced digestion device, sample was passed through the original integrated dissociation/ lter device with both 50 and 15 µm pore size membranes or the modi ed version with only the 50 µm membrane. Cell numbers and viability were similar for all conditions (see Supplementary Information, Fig. S14), and thus we selected to use the original version with both membranes for heart tissue.
Next we evaluated the full micro uidic platform at different digestion times. We again focused on shorter processing times due to the potential sensitivity of cardiomyocytes. After 5 min treatment with the digestion device, ~ 2000 cardiomyocytes were recovered per mg heart tissue (Fig. 8A). This was lower than both the 15 and 60 min controls, by ~ half and one-third, respectively. Increasing digestion device processing to 15 min increased recovery to ~ 12,000 cells/mg, which was ~ 2-fold higher than the 60 min control. As with kidney, the interval format further increased cardiomyocyte recovery to ~ 18,000 cells/mg. Endothelial cell (Fig. 8B) and leukocyte (Fig. 8C) yields from the micro uidic device platform were signi cantly lower than the 60 min control. The interval format did improve recovery for both cases, but the 60 min control remained higher by ~ 2-fold for endothelial cells and ~ 1.5-fold for leukocytes. Based on this differential recovery, device platform processing resulted in signi cant enrichment of cardiomyocytes (Fig. 8D). Viabilities for all three cells types were similar to controls (see Supplementary   Information, Fig. S15). Considering results for all tissues in a comprehensive manner, heart likely lies in between the kidney/tumor and liver extremes. The tissue is still challenging to break down, which is why recovery was low at the early time points. Digestion was likely to be particularly ineffective on its own for cardiomyocytes due to strong intracellular connections formed by desmosomes and adherins junctions, while the micro uidic platform provided the shear stresses necessary to break these connections and separate cardiomyocytes. However, the sensitivity of cardiomyocytes to mechanical damage is a confounding factor, which makes longer digestion times unlikely to improve results. Endothelial cells can arise from both vessels and the endocardium that lines the walls of the atrial and ventricular chambers.
We contend that endocardium was liberated effectively by digestion alone since the chambers can be readily accessed by collagenase. As seen for kidney and tumor, however, blood vessels require longer time for effective release of endothelium, even with the device platform. This suggests that our results were dominated by endocardium, and that damage was the predominant reason for reduced recovery.
The fact that interval recovery improved results for all cell types assessed in both heart and liver indicates that this mode is critical for optimal performance. In fact, temporal resolution should likely be increased, or ideally, be continuous, to prevent cell damage. Nevertheless, the micro uidic platform as currently con gured and operated in this study consistently improved the recovery of single cells from diverse tissue types based on increased total cell yield, decreased processing time, and in some cases, both.

Conclusion
We have presented a novel micro uidic tissue processing platform comprised of a newly introduced digestion device that facilitates loading and processing of minced specimens, as well as a newly integrated dissociation/ lter device that completes the dissociation work ow so that the single cell suspension is immediately ready for downstream analysis or alternative application. The new minced digestion device accelerated tissue break down and produced signi cantly more single cells than traditional methods, while the integrated dissociation/ lter device increased yield further, all without affecting viability. This was determined for a diverse array of tissue types that exhibited a wide range of properties, as well as two different single cell analysis methods, ow cytometry and scRNA-sEq. We also introduced a novel processing scheme, interval operation, which allowed us to extract single cells at different time points during micro uidic digestion. We found that for tissues that were physically tougher and more robust, such as kidney and tumor, micro uidic processing produced similar cell numbers in dramatically less time (15 vs 60 min), and approximately 2.5-fold more single cells in total. scRNA-seq further con rmed that endothelial cells, broblasts, and basal epithelial cells were highly enriched by the micro uidic platform, with each increasing by 4-to 10-fold. Additionally, we found that shorter digestion times were associated with lower stress responses for some cell types, but otherwise micro uidic processing did not add to the stress response in any case. These results clearly con rm that the micro uidic tissue processing platform holds exciting potential to advance scRNA-seq studies by reducing cell subtype-biasing, processing time, and/or stress response. For tissues that were softer and may contain sensitive cell types, like liver and heart, we found that processing times could be dramatically reduced and that interval operation was critical to avoid cell damage and maximize recovery. These results will advance goals in tissue engineering and regenerative medicine, and could be particularly exciting for patient-derived organ-on-a-chip models for pharmacological studies. By focusing on minced specimens, our micro uidic tissue processing platform can readily be incorporated into the dissociation work ows for most, if not all, organs and tissues. Minimizing tissue pre-processing would be advantageous, and will be pursued in future work, along the lines of our original core digestion device. Another future goal will be to decrease interval recovery time points to further explore protection of fragile cells, intentional enrichment of certain cell subtypes, and lowering of stress responses. Ideally, we would integrate a cell separation strategy that would make it possible to elute single cells from the platform as soon as they are generated. We will also evaluate other tissues, with a focus on optimizing performance for diverse tissue properties and cell subtypes, as well as explore alternative proteolytic enzymes such as cold-active proteases. 36,38 Finally, we envision incorporating micro uidic cell sorting and detection capabilities into the platform to create fully integrated and point-of-care technologies for cell-based diagnostics and drug testing, with a focus on human tissues for clinical applications.

Materials And Methods
Device Fabrication. Micro uidic minced digestion and integrated dissociation/ lter devices were fabricated by ALine, Inc. (Rancho Dominguez, CA). Brie y, uidic channels, vias, and openings for membranes, luer ports, and hose barbs were etched into PMMA polyethylene terephthalate (PET) layers using a CO 2 laser. Nylon mesh membranes were purchased from Amazon Small Parts (15, and 50 µm pore sizes; Seattle, WA) as large sheets and were cut to size using the CO2 laser. Device layers and other components (hose barbs, nylon mesh membranes) were then assembled, bonded using adhesive, and pressure laminated to form monolithic devices.
Murine Tissue Models. Kidney, liver, and heart were harvested from freshly sacri ced BALB/c or C57B/6 mice (Jackson Laboratory, Bar Harbor, ME) that were determined to be waste from a research study approved by the University of California, Irvine's Institutional Animal Care and Use Committee (courtesy of Dr. Angela G. Fleischman). Mammary tumors were harvested from freshly sacri ced MMTV-PyMT mice (Jackson Laboratory, Bar Harbor, ME). For kidneys, a scalpel was used to prepare ~ 1 cm long x ~ 1 mm diameter strips of tissue, each containing histologically similar portions of the medulla and cortex. Tissue strips were then further minced with a scalpel to ~ 1 mm 3 pieces. Liver, mammary tumor, and heart were uniformly minced with a scalpel to ~ 1 mm 3 pieces. Minced tissue samples were then weighed and either processed with the devices as described below. Controls were placed within microcentrifuge tubes, collagenase type I solution (StemCell Technologies, Vancouver, BC), and the luer port was closed off using a stopcock. The experimental setup consisting of the device, tubing, and peristaltic pump were then placed inside a 37 °C incubator to maintain optimal enzymatic activity. The collagenase solution was recirculated through the device and tubing using the peristaltic pump at a ow rate of 10 or 20 mL/min for a speci ed time.
Quanti cation of DNA Recovered from Cell Suspensions. Puri ed genomic DNA (gDNA) content of digested kidney tissue suspensions were assessed using a Nanodrop ND-1000 (Thermo Fisher, Waltham, MA) following isolation using a QIAamp DNA Mini Kit (Qiagen, Germantown, MD) according to manufacturer instructions. gDNA for device processed samples represents the cellular contents eluted from the device after operation, while gDNA for control samples represent the total amount of gDNA present in these samples.
Integrated Dissociation/Filter Device Operation. Following processing with the minced digestion device, tubing was connected from the outlet of the minced digestion device to the inlet of the integrated dissociation and ltration device. If recirculation was intended, tubing was connected from the cross-ow outlet to the peristaltic pump, while the outlet of the integrated device was closed off with a stopcock. Fluid was then pumped through the dissociation component at 10 mL/min ow rate. For nal collection of the sample, or if only 1 pass through the dissociation component was utilized, the cross-ow outlet was closed off with a stopcock, and sample was pumped through at 10 mL/min and collected from the e uent outlet. Following all experiments, devices were washed with 2 mL PBS + to ush out and collect any remaining cells. For time interval recovery, each PBS + wash was followed by repriming of the device and tubing with collagenase solution, and the outlet of the minced digestion device was reconnected to the peristaltic pump for continued recirculation until the next collection period.
Analysis of Cell Suspensions using Flow Cytometry. Cell suspensions were analyzed using tissue speci c ow cytometry panels shown in Table 1 instructions. An Illumina NovaSeq 6000 platform (Illumina, San Diego, CA) was used to sequence the samples at a depth of ~ 60,000 reads/cell for kidney and ~ 45,000 reads/cell for mammary tumor.
Sequencing fastq les were aligned using 10x Genomics Cell Ranger software (version 3.1.0) to an indexed mm10 reference genome. Cell Ranger Aggr was used to normalize the mapped reads for cells across the libraries for each data set. Genes that were not detected in at least 3 cells were discarded from further analysis. Cells with low (< 200) or high (> 3000 for kidney; >4000 for mammary tumor) unique genes expressed were also discarded, as these potentially represent low quality or doublet cells, respectively. 4 Cells with high mitochondrial gene percentages were also discarded (> 50% for kidney and > 25% for mammary tumor), as these can also represent low quality or dying cells. 87 The Seurat pipeline was used for cluster identi cation, 60 principle component analysis (PCA) was performed using genes that are highly variable, density clustering was performed to identify groups, and Uniform Manifold Approximation and Projection (UMAP) plots were used to visualize the groupings. For kidney, cell clusters were annotated using two approaches. First, top differential genes in each cluster were examined to determine the cell type of the cluster based on expression of known marker genes (e.g. Kap, Napsa, and Slc27a2 for S2-S3 proximal tubules, 36 Gpx3 for S1 proximal tubules, 36 Emcn for endothelial cells, 34 Slc12a1 for loop of Henle, 4 Slc12a3 for distal convoluted tubule, 4 etc. Second, since a well-established atlas of murine kidney was available, we used a cell scoring method 63 to compare marker gene signatures from each of our clusters to published datasets 4,88 to con rm cluster annotations (see Supplementary Information, Fig. S5). For tumor, cell clusters were annotated by examining top differential genes in each cluster to determine cell type based on expression of known marker genes (e.g. EpCAM for epithelial cells). Cellular stress responses were assessed using a previously developed scoring method to compare stress response gene expression from each cluster to a previously published dataset of known stress response genes. 39,63 Statistics. Data are represented as the mean ± standard error. Error bars represent the standard error from at least three independent experiments. P-values were calculated from at least three independent experiments using students t-test.  Tissue fragments and cell aggregates from the digestion device will be further broken down by hydrodynamic shear forces and nylon mesh lters. (D) Pictures of the fabricated minced digestion device.
(E) Picture of the fabricated dissociation/ lter device.

Figure 2
Device optimization using murine kidney. (A) Kidneys were harvested, minced, and processed using the minced digestion device at 10 or 20 mL/min ow rate for 15 or 60 min, and total genomic DNA (gDNA) was quanti ed. The gDNA was extracted directly from the control, and thus this represents maximum recovery. Results at 20 mL/min ow rate were superior, particularly at the shorter time point. (B) Pictures of tissue within the minced digestion device chamber before and after 15 or 60 min of processing at 10 (i) or 20 (ii) mL/min ow rate. Signi cant tissue remained at 10 mL/min, while tissue was larger absent at 20 mL/min. (C) Single EpCAM+ epithelial cells were quanti ed by ow cytometry after samples were processed with the minced digestion device for 15, 30, or 60 min. We also evaluated recovery of sample at different time intervals, with more collagenase added to continue processing of remaining tissue. (D) Epithelial cell viability was ~80% for all control and device conditions. (E) Samples were processed with the integrated dissociation/ lter device following 15 min of digestion device treatment. A single pass through the integrated device produced optimal results. (F) Epithelial cell viability was at ~85-90% for all conditions. Error bars represent standard errors from at least three independent experiments. * indicates p < 0.05 and ** indicates p < 0.01 relative to the control at the same digestion time. # indicates p < 0.05 relative to the static condition at the same digestion time. Scale bar represents 5 mm.

Figure 3
Micro uidic platform results for murine kidney. Kidneys were harvested, minced, processed with the digestion device for different 15 or 60 min, passed through the integrated dissociation/ ltration device one time, and resulting cell suspensions were analyzed using ow cytometry. We also evaluated interval recovery at 1, 15, and 60 min time points from the same tissue sample. Controls were minced, digested for either 15 or 60 min, pipetted/vortexed, and passed through a cell strainer. (A) Single EpCAM+ epithelial cells increased by 2.5-fold with micro uidic processing. Interval results were comparable to static, and the 1 min interval produced comparable cell numbers to the 15 min control. Trends were similar for (B) endothelial cells and (C) leukocytes. Micro uidic processing was particularly effective for endothelial cells, yielding >5-fold more cells than the control at 60 min. (D) Population distributions obtained for each processing condition. Endothelial cells were enriched for all device conditions except the 1 min interval relative to controls. Error bars represent standard errors from at least three independent experiments. * indicates p < 0.05 and ** indicates p < 0.01 relative to the control at the same digestion time.    Micro uidic platform results for murine liver. (A,B) Liver was harvested, minced, and evaluated with the minced digestion device alone and in combination with the integrated dissociation/ lter device.
Hepatocytes were identi ed and quanti ed by ow cytometry. (A) The digestion device increased hepatocyte recovery by ~4-fold at 15 min, but continued digestion and passing through the integrated dissociation/ lter device one time decreased hepatocyte yield, likely due to the large size and fragile nature of hepatocytes. (B) Hepatocyte viability was ~75-80% for all conditions, except the 60 min integrated condition. (C-F) Results using shorter digestion times and a single pass with a dissociation/ ltration device containing only the 50 µm lter. (C) After only 5 min of micro uidic processing, 4-fold more cells were obtained than the 15 min control and only slightly less than the 60 min control. Interval recovery enhanced hepatocyte yield by ~2.5-fold relative to the 60 min control and 15 min static conditions. The 1 minute interval contributed substantially, producing ~70% as many hepatocytes as the 60 min control. Similar results were observed for (D) endothelial cells and (E) leukocytes, although the bene t of intervals was less pronounced. (F) Population distributions obtained for each processing condition. Micro uidic processing generally enriched for leukocytes, although there was a shift to hepatocytes for the later intervals. Error bars represent standard errors from at least three independent experiments. * indicates p < 0.05 and ** indicates p < 0.01 relative to the 60 min control at the same digestion time. # indicates p < 0.05 and ## indicates p < 0.01 relative to the static condition at the same digestion time. Micro uidic platform results for murine heart. Hearts were resected, minced, processed with the micro uidic platform (both 50 and 15 µm membranes), and analyzed by ow cytometry. We employed shorter digestion device time points due to the sensitivity of cardiomyocytes. (C) Micro uidic processing produced ~12,000 cardiomyocytes per mg after 15 min, which was ~2-fold higher than the 60 min