We report that the efficiency of reprogramming human somatic cells to induced pluripotent stem cells (hiPSCs) can be dramatically improved in a microfluidic environment. Microliter-volume confinement resulted in a 50-fold increase in efficiency over traditional reprogramming by delivery of synthetic mRNAs encoding transcription factors. In these small volumes, extracellular components of the TGF-β and other signaling pathways exhibited temporal regulation that appears critical to acquisition of pluripotency. The high quality and purity of the resulting hiPSCs (μ-hiPSCs) allowed direct differentiation into functional hepatocyte- and cardiomyocyte-like cells in the same platform without additional expansion.


Pluripotent stem cells hold great promise for the modeling of human tissues in vitro, but more efficient methods for reprogramming somatic cells are needed. hiPSCs are characterized by the expression of key genes and signaling pathways. NANOG, OCT4 (POU5F1) and SOX2 constitute a master transcriptional network that supports secondary transcription factor expression (for example, MYC family members and ZFX) in a feed-forward manner and is critical for stabilizing pluripotency1. These transcription factors depend on fibroblast growth factor (FGF) and transforming growth factor-β (TGF-β) signaling pathways to sustain self renewal2,3. Somatic cell reprogramming can be induced by the forced expression of POU5F1, SOX2, KLF4 and C-MYC4. In the initial stage of reprogramming, these factors perturb the endogenous transcriptional network and lead to chromatin remodeling. During maturation and stabilization, endogenous POU5F1, SOX2 and NANOG become expressed and the same signaling networks that govern pluripotency in embryonic stem cells are activated5 (Supplementary Fig. 1).

Using current methodologies, somatic cell reprogramming occurs stochastically with low efficiency6. For instance, modified mRNA (mmRNA) transfection of transcription factors is currently the most efficient technology for reprogramming genetically unmodified human cells, and the maximum yield is only 3 hiPSC colonies per 100 somatic cells seeded6. Increasing yield requires improvements in transcription factor delivery and a better understanding of the epigenetic reorganization and genetic network rewiring that underlie the process. However, the fact that only a few cells undergo full reprogramming in culture makes it difficult to capture critical molecular features and limits opportunities for improvement.

Here, we show that downscaling mmRNA reprogramming to microliter volumes generates a favorable environment for the acquisition of pluripotency. As in our previous work7,8 on stem cell differentiation and pluripotent stem cell culture in a microfluidic context, we found that a confined cell microenvironment has a strong impact on self-regulated autocrine and paracrine signaling. We achieved an average efficiency of over 120 hiPSC colonies per 100 seeded cells. To further facilitate optimal reprogramming, we developed a remotely controlled system for generating hiPSC in feeder- and xeno-free conditions that requires only 200 μL of medium during a 14-d reprogramming. mmRNA technology allows the rapid attainment of transgene-free hiPSCs and, due to the short lifespan of transfected mmRNA, cells can be readily differentiated at the end of the reprogramming protocol within the same system. Together, these features make the process compliant for clinical-grade hiPSC generation and ideal for highly controlled in vitro experiments in basic research.


Technological downscale of reprogramming process

By analyzing published microarray data on the progression of reprogramming9, we observed that defined sets of soluble extracellular signaling molecules are temporally regulated either positively or negatively during the process (Supplementary Figs. 1 and 2 and Supplementary Data 1). We hypothesized that manipulating the concentration of these endogenous extracellular components in the surrounding medium could strongly impact the reprogramming process. To achieve the conditions necessary for testing this hypothesis, we scaled down the volume of culture medium to a few microliters, causing a ten-fold reduction in the ratio of medium volume to cell surface area compared to that of a well plate (Supplementary Fig. 1). At this scale, extrinsic autocrine and paracrine factors released by the cells can concentrate rapidly.

We designed a microfluidic platform for downscaled reprogramming (Fig. 1a, Supplementary Fig. 3 and Supplementary Video 1). The culture chamber contained medium to a height of only 200 μm to ensure significant accumulation of endogenous factors, a model that has proven successful for long-term culture of pluripotent stem cells, germ layer differentiation and delivery of viral particles7,8,10. The platform is extremely versatile, since it can either be run by manual pipetting with minimal expertise (Supplementary Fig. 3) or coupled to advanced liquid handling equipment to increase throughput. We also automated our setup with an integrated system of remotely controlled pneumatic valves that distributed medium with precise timing into the 32 culture chambers of a single device (Fig. 1a).

Figure 1: Reprogramming on a microliter scale.
Figure 1

(a) Microfluidic cell culture proceeds in independent channel-shaped chambers that contain adherent cells. Medium handling is controlled via software that actuates an integrated pneumatic circuit, and a 2-week workflow converts fibroblasts into hiPSCs by transfection of reprogramming factors every 4 h. At right, an automated microfluidic chip. (b) Efficiency of nGFP mmRNA transfection in BJ fibroblasts 24 h after a single transfection. Gray arrows indicate the typical amount of mmRNA used in wells and the optimized amount used in microfluidics with cells at confluence. Error bars, mean ± s.d. (n > 4). (c) Dependence of transfection efficiency on the duration of incubation. Experimental microfluidic data (circles) are compared to simulated data (dots) derived from a stochastic model of the transfection process. Error bars, mean ± s.d. (n > 4). (d) Western blot quantification of exogenous protein in single wells and microfluidic channels 24 h after two daily mmRNA transfections in BJ fibroblasts. Error bars, mean ± s.d. (n = 4). *P < 0.05, **P < 0.01 (see Online Methods for details of statistical tests). (e) Epithelial morphology and transfection progression, monitored by nGFP expression, at day 8 of microfluidic reprogramming. Scale bars, 500 and 150 μm (macro).

We opted for rapid transient expression of reprogramming factors using mmRNA, which has a typical lifespan in cells of 24 h11. However, daily transfections are required for sustained expression over several days11, and the procedure benefits greatly from automation. Using mmRNA encoding a nucleus-targeted green fluorescence protein (nGFP), we obtained 80% nGFP+ cells by a single transfection in microfluidics, with a 60-fold reduction of the overall amount of mmRNA per cell used compared with equally efficient transfection in wells (Fig. 1b and Supplementary Fig. 4). Optimized mmRNA concentration and transfection duration were determined by means of a stochastic mathematical model (Fig. 1c and Supplementary Fig. 5). At optimal conditions of mmRNA delivery, two key players in the reprogramming process, POU5F1 and KLF4 (refs. 12,13), showed higher expression at the protein level in microfluidics (Fig. 1d and Supplementary Fig. 5).

High efficiency and purity of reprogramming at micro-scale

We next examined whether human fibroblasts could be effectively reprogrammed to hiPSCs in the microfluidic platform by sequential delivery of mmRNAs. In addition to the four so-called Yamanaka factors (POU5F1, SOX2, KLF4 and C-MYC), we transfected LIN28 and NANOG daily14, along with nGFP to verify sustained delivery efficiency during the process (Fig. 1e and Supplementary Fig. 5). We initially performed reprogramming in the presence of mitotically inactive feeder cells. μ-hiPSC colonies were first recognized morphologically as early as 7 d after the first mmRNA transfection (Fig. 2a). Colonies can be individually extracted from the system by manual picking or can be collectively pooled, either mechanically or by standard chemical or enzymatic methods (Supplementary Fig. 6). Randomly chosen colonies were thoroughly characterized after clonal extraction (Fig. 2b–e and Supplementary Fig. 7). Immunofluorescence analysis was also performed in situ to verify mesenchymal-to-epithelial transition and pluripotency marker expression at different stages (Fig. 2f,g and Supplementary Fig. 7).

Figure 2: hiPSCs are obtained by reprogramming at the micro-scale.
Figure 2

(a) Time course (days 16–19) of the generation and expansion of a representative hiPSC colony inside a microfluidic channel. Scale bar (same for all images), 750 μm. (b,c) Fibroblast-derived hiPSCs express the pluripotency markers listed and, after defined monolayer differentiation, specific germ-layer markers as revealed by immunofluorescence. Scale bars (same for all figures), 100 μm. (d) hiPSCs can generate embryoid bodies (EBs) that randomly express markers of the three germ layers (meso, mesoderm; ecto, ectoderm; endo, endoderm) (Supplementary Fig. 7) and teratomas with structures typical of the three germ layers, such as neural rosettes, cartilage and gut-like structures. Scale bars, 100 μm; except EB, 200 μm. (e) μ-hiPSCs obtained as in b and expanded for 12 passages are karyotypically normal. (f,g) In situ characterization of early reprogramming (day 7) and hiPSC formation within a microfluidic channel by immunofluorescence. Antibody to E-cadherin (E-cad) was used to detect mesenchymal–epithelial transition. At day 7 most cells in the channel were E-cad+ (white arrows indicate the lower pneumatic layer of automated chip). A representative μ-hiPSC shows expression of pluripotency markers by immunofluorescence within an automated chip (day 21). Scale bars, 250 μm.

Reprogramming yield was quantified by an efficiency index, defined as the number of hiPSC colonies expressing pluripotency markers (NANOG and TRA-1-60) at day 15, from the first mmRNA transfection, per 100 cells initially seeded (Fig. 3a). This efficiency index is numerically equal to the percentage of colonies per cell seeded commonly reported in the literature. After establishing an incremental mmRNA dosage strategy to balance exogenous delivery, cell proliferation and cell death, we obtained a high reprogramming yield in microfluidics, up to 16 (6 ± 4) using supporting feeder cells (Fig. 3a,b, Supplementary Fig. 8 and Supplementary Video 2).

Figure 3: Micro-scale reprogramming occurs with high efficiency and purity.
Figure 3

(a) Improvement of reprogramming yield by protocol optimization. Reprogramming efficiency index is the number of colonies expressing both NANOG and TRA-1-60 per 100 cells seeded. Well, conventional reprogramming in 6-well plate; μF feeder, microfluidic non-optimized reprogramming with feeder layer; μF feeder incremental, similar to the previous one but adapting mmRNA dose to cell density during the first reprogramming days; μF feeder-free and μF feeder-free incremental, similar to the previous two but without feeder layer. (b) TRA-1-60+ colonies at the end of the reprogramming protocol with feeder layer. Left, bright field image of a microfluidic culture channel with superimposed TRA-1-60-labeled cells; right, 10-channel microfluidic chip area showing the spatial distribution of TRA-1-60+ colonies in black. (c) Immunofluorescence of NANOG+/TRA-1-60+ μ-hiPSCs derived in feeder-free conditions in microfluidics. Scale bars, 100 μm. Dot plot of hiPSC purity, measured as the ratio of NANOG+ cells to the total number of cells in culture (day 15, 72 h after the last mmRNA transfection). (d) Morphology of μ-hiPSCs derived from renal epithelial (RE) cells isolated from urine samples at day 15 (top) and 19 (bottom). Scale bars, 200 μm. Dot plot represents the fold change in reprogramming yield of RE cells in microfluidics compared to wells of a female (XX) and two male patients (XY). (e) Reprogramming efficiencies of various primary cell samples. RE cells were isolated from individuals with anti-trypsin deficiency (ATD), and skeletal muscle fibroblasts (SkMFs) from muscles of individuals with Duchenne muscular dystrophy (DMD). (ae) Each dot represents an independent microfluidic channel. Black bar, mean of respective data points shown.

To uniquely identify the correlation between reprogramming and endogenous factors, we performed reprogramming in chemically defined and xeno-free conditions (Supplementary Fig. 9), avoiding cell feeder layers, which would uncontrollably perturb medium composition. Under feeder-free optimal conditions, we obtained an efficiency index of up to 150 (121 ± 38) (Fig. 3a–c and Supplementary Fig. 9), representing an approximately 50-fold greater reprogramming efficiency in micro-scale compared to a multi-well plate (2.1 ± 0.7) and the highest efficiency reported in the literature6,11. We also showed that the balance between initial cell density and the transfection-induced toxicity has to be optimized; for instance, even when using a higher mmRNA dose, a ten-fold increase in cell seeding density decreases reprogramming efficiency (Supplementary Fig. 10). We explored how different sources of variability could affect reprogramming efficiency in microfluidics using nested ANOVA for different random factors, cell batches (n = 3) and microfluidic devices (n > 2), each with multiple independent culture channels (n > 2) for a total of 32 experimental data points (Fig. 3a). The analysis showed no significant effect of cell batches and microfluidic devices on the reprogramming efficiency index and it confirmed system robustness (Supplementary Fig. 10). Moreover, μ-hiPSC purity at day 15 from the first reprogramming transfection, measured as NANOG+/TRA-1-60+ cells over total number of cells in culture, was 85% (84.8 ± 5.7%, Fig. 3c). Thus, microfluidic reprogramming demonstrated high homogeneity, as showed by imaging cells undergoing reprogramming every 12 h for 17 d (Supplementary Videos 3 and 4).

Although usually only a few hiPSC colonies are expanded for further applications downstream of the reprogramming process, having a system that boosts efficiency is very important for generating hiPSCs from recalcitrant cell types such as slowly proliferating and early senescent primary cells. A particularly convenient cell source from which to obtain hiPSCs from a patient is urine15. The reprogramming efficiency of urine-derived epithelial cells by a nonintegrating method has been reported to be variable from patient to patient, and fewer than 0.02 colonies per 100 cells seeded16. Using mmRNA within the microfluidic system, we successfully reprogrammed urine-derived epithelial cells from three different donors (Fig. 3d and Supplementary Fig. 11), confirming a 50-fold increase in reprogramming efficiency compared with the efficiency of a well plate. Next, we reprogrammed cells, either freshly isolated or stocked in cell banks, from patients affected by genetic diseases (Fig. 3e). One application of our system was the reprogramming of skeletal muscle fibroblasts isolated from patients affected by Duchenne muscular dystrophy, which have a limited lifespan before senescence onset. To our knowledge, this was the first time hiPSCs were obtained from the target tissue of a specific genetic disease, and this process could have important implications in in vitro studies of various diseases on account of the possibility of tissue-specific genetic mutations occurring during disease progression.

Amplified endogenous signaling enhances reprogramming

To further investigate the effect of a confined environment on reprogramming, we profiled gene expression in colonies freshly obtained in microfluidics (p0 μF) and a multi-well plate at day 15 (Fig. 4a,b) using microarrays. We designed the experiment to also assess whether potential differences in gene expression of hiPSC clones derived in microfluidics and in a well could be eliminated after expansion in standard conditions. p0 μF colonies evidenced a distinctive expression profile, but the differences did not involve the expression of the main pluripotency associated transfection factors (Fig. 4c). After a three-passage conventional expansion, μ-hiPSCs converged to the same expression profile of hiPSCs derived in multi-well plates (Fig. 4b and Supplementary Data 2). The expression profile of colonies obtained in all four conditions was within the variability of pluripotent stem cells reported in the literature (Supplementary Fig. 12). The most significantly enriched gene ontology categories characterizing genes differentially expressed between wells and microfluidics are related to the extracellular space and receptor-mediated signaling (Fig. 4d and Supplementary Data 3).

Figure 4: Enhanced endogenous signaling in microfluidics promotes reprogramming.
Figure 4

(a) Experimental design for microarray analysis. Freshly derived colonies (passage 0, p0) obtained in microfluidics or in wells were dissected into two halves (4,000 cells each). One was immediately processed for total RNA extraction, and the other was seeded in a new well and expanded for three passages in conventional well plates (p3). Colonies of 8,000 cells from each condition were dissected again at p3 to extract total RNA. RNA samples were analyzed by microarray. (b) Principal component analysis of freshly derived and expanded hiPSCs from microfluidics (n = 4) and wells (n = 4), grouped according to the results of hierarchical clustering. Each clone was isolated from reprogrammed samples in different microfluidic chips or well plates. (c) Expression level of selected pluripotency markers in the four conditions (p0 and p3 in the two culture systems). Color bar, relative median-centered gene expression (log2). (d) Microarray-derived expression profiles of genes differentially expressed between the four conditions (p0 and p3 in the two culture systems) belonging to the intersection category shown in Figure 1b. Color bar, same as in c. (e) Effect of medium change frequency. Additional changes every 4 h were performed compared with standard daily medium management to promote the washout of endogenous factors. Doubling the number of medium changes per day (6 d−1) compared with the standard experiment in microfluidics (3 d−1) significantly reduced reprogramming efficiency (**P < 0.01; see Online Methods for details of statistical tests). (f) TGF-β pathway activation during reprogramming measured as SMAD2/3 activity of a reporter line. Medium in microfluidics is exposed to significantly higher conditioning of cell-released endogenous signals. Error bars, mean ± s.d. (n = 3). n.d., not detectable. (g) TGF-β dynamics during reprogramming in microfluidics under nontreated (−) and exogenously supplemented TGF-β1 conditions for the whole duration of reprogramming (n = 2). (h) Feeder-free reprogramming efficiency in microfluidics under conditions described in g and in the presence of the TGF-β inhibitor SB431542. SB2-2, 2 μM inhibitor concentration for the whole duration of reprogramming; SB2-10, 2 μM inhibitor concentration for the first 6 d, followed by 10 μM during the second half of the process (n = 10). a.u., arbitrary units.

To verify the contribution of the extracellular accumulation of cell-secreted factors during reprogramming, we increased the frequency of medium change, expecting to increase the flushing of cell-secreted factors7,8. Consistent with this hypothesis, the efficiency of reprogramming was significantly decreased by increasing the frequency of medium change from f = 3 d−1 to f = 6 d−1, despite the higher delivery of the propluripotency exogenous factors from the reprogramming medium (Fig. 4e). We also obtained consistent results when we compared the efficiency of reprogramming within microfluidic culture chambers of different heights (Supplementary Fig. 13 and Supplementary Table 1).

Overall, these results support the hypothesis that, in our standard microfluidic system (200 μm height and low frequency of medium change), there is an optimal balance between endogenous and exogenous factors for cell-reprogramming progression.

Microarray results showed that multiple extracellular components of the TGF-β pathway were upregulated in freshly generated colonies (p0 μF, Fig. 4d). Since exogenous TGF-β is not present in the medium, relevant differences between the two culture systems could be due to differential accumulation of cell-secreted TGF-β-family ligands. Consistent with this, medium conditioned by cells undergoing reprogramming produced an exceptionally strong activation of the downstream effector, SMAD2/3, in luciferase-reporter cells, with a specific temporal profile (Fig. 4f,g). When we exogenously promoted or inhibited TGF-β signaling throughout the reprogramming process, no μ-hiPSCs were obtained (Fig. 4h), although some phenotypic differences between promotion and inhibition were observed (Supplementary Fig. 14). We conclude that, in the microfluidic confined environment, cells are able to self-regulate signaling toward maximal reprogramming efficiency and TGF-β pathways play an important but complex role during reprogramming.

Differentiation of freshly derived μ-hiPSCs into functional cells

We next questioned whether the freshly generated μ-hiPSCs (p0 μF) could differentiate into specific phenotypes without an intermediate stage of expansion. The high purity of freshly generated hiPSCs (Fig. 3c) should allow defined differentiation, limiting undesired extrinsic signals from nonpluripotent cells. Signaling pathways regulating the exit from pluripotency and the induction of specific germ layer differentiation should have been in place (Fig. 5a). Indeed, our results show that specific markers of the three germ layers are selectively expressed after directed differentiation of freshly generated μ-hiPSCs (Fig. 5b).

Figure 5: μ-hiPSCs can produce patient-specific differentiated functional cell types.
Figure 5

(a) Extracellular soluble components of signaling pathways play a major role during cell reprogramming and specification into the three germ layers. (b) μ-hiPSC colonies, freshly generated in microfluidic channels and specifically differentiated toward the three germ layers, selectively express early germ-layer commitment markers, as detected by qPCR. Mean expression ± s.d. (n = 10). (c) Cardiomyocytes specified from primary μ-hiPSCs display cTNT expression and sarcomeric organization (see also Supplementary Fig. 15 and Supplementary Video 5). Scale bar, 50 μm. Top left, magnification showing sarcomeric structure details. (d) Differentiation of primary μ-hiPSC into the hepatic lineage. Colonies differentiated into large areas of mature cells, displaying functional markers. CK18 and actin in green, nuclei in blue, albumin (Alb) in red, PAS. Scale bars, 250 μm, 50 μm (albumin). (e) Schematic summary describing the integrated process of reprogramming and differentiation in microfluidics without intermediate passaging. (f) Vision of high-throughput derivation of human tissues for biological assays on a population scale.

Moreover, we verified that freshly generated μ-hiPSCs (p0 μF) could undergo functional differentiation, which requires the activation or repression of specific signaling pathways at defined temporal windows. Contracting cardiomyocyte-like cells showing remarkable troponin-T expression and sarcomeric organization were obtained in microfluidics by sequential exogenous promotion and inhibition of WNT signaling (Fig. 5c, Supplementary Fig. 15 and Supplementary Video 5). Polygonal-shaped CK18+ hepatocyte-like cells, occasionally polynucleated, were also obtained by promoting Activin and WNT pathways of freshly generated μ-hiPSCs (Fig. 5d). Hepatocyte-like cell functional maturation of extensive morphologically differentiated areas was confirmed by albumin secretion and glycogen storage.


We have presented the first reprogramming process completely performed at micro-scale, a process that provides a number of advantages. Previous studies only used microfluidics as a tool to mechanically improve initial transcription factor transduction or nuclear fusion17,18. Our system requires only a few microliters of medium per day for each independent culture channel, with an overall 100-fold reagent savings compared with a well of a six-well plate. Cost reductions, coupled with system automation, make it feasible to increase throughput to hundreds of parallel reprogramming experiments from different cell sources. In our lab prototype, we can already perform 32 parallel independent experiments in a single chip, and multiple chips can be run simultaneously.

For basic research applications, this system offers the possibility of manipulating the cell culture microenvironment with high precision, and can easily be implemented through automation. Moreover, microfluidics is well suited to comply with good manufacturing practice requirements because the system can be closed and remotely controlled, maximizing robustness and strongly decreasing the probability of contamination.

A number of groups have worked on increasing reprogramming efficiency using either integrating reprogramming methods19,20 or doxycycline-inducible systems21,22,23,24. A recent study showed that reprogramming is a deterministic process using secondary fibroblasts differentiated in vitro21; this result still awaits confirmation for fibroblasts derived from human samples22. Patient-derived somatic cell reprogramming necessarily demands non-integrating transcription factor induction systems that, though less efficient, reduce the risk of genomic aberrations.

Our method showed a 50-fold improvement over the most efficient reprogramming reported using human cells without genetic modifications. Moreover, the process achieved high homogeneity, with 85% of hiPSCs in culture at day 15 after the first reprogramming transfection. High efficiency seems attributable to temporally defined and self-regulated cues that cells undergoing reprogramming provide within the microfluidic environment, cues that bear further study. This system can also be successfully used for dissection of reprogramming mechanisms, given that the majority of cells in culture synchronously reacquire a pluripotent phenotype. Moreover, this system makes it possible to differentiate patient-specific μ-hiPSCs without cell expansion and to produce functional cardiomyocyte- and hepatocyte-like cells in approximately a month (Fig. 5e). Besides lowering costs, this time reduction has important implications on cell quality, reducing the probability of genetic aberrations.

Compared to other non-integrating reprogramming methods such as Sendai virus, mmRNA technology shows tunable balance between transfection efficiency and cytotoxicity, which is a fundamental prerequisite for feasible micro-scale reprogramming. Ultimately, the proposed process makes it feasible to rapidly and reliably derive hiPSCs from a large cohort of patients. Improved methods for cell functional maturation will further accelerate the use of this system for high-throughput studies25,26,27 (Fig. 5f), like the systematic investigation of disease predisposition, drug cytotoxicity and efficacy, and response to environmental factors in relation to the genetic background of the cell donor.


Cell culture.

Human foreskin BJ fibroblasts (Miltenyi Biotec) and skeletal muscle fibroblasts (SkMF, provided by M. Mora and S. Zanotti, Telethon Network of Genetic Biobanks, Italy) from patients with Duchenne muscular dystrophy were cultured in Dulbecco's modified Eagle Medium (DMEM, Life Technologies) supplemented with 10% FBS (FBS, Life Technologies). HFF-1 fibroblasts (ATCC) were cultured in DMEM with 15% FBS. Renal epithelial cells (RE) were isolated and expanded from three healthy donors, male and female, in RE medium (Lonza) as previously described28. Inactivated human newborn foreskin fibroblasts NuFF-RQs (AMS Biotechnology) were seeded on 0.2% type-A gelatin (Sigma-Aldrich) at 260 cells mm−2 in DMEM with 10% FBS, in case of use as a feeder layer for reprogramming and for Pluriton medium (Stemgent) conditioning. RE cells from patients with anti-trypsin-deficiency (ATD) were freshly isolated and provided by I. Ferrarotti (Fondazione IRCCS Policlinico San Matteo).

hiPSCs were mechanically passaged on mitomycin-treated mouse embryonic fibroblasts (MEF, Millipore) with daily changes of hiPSC medium (DMEM/F12, 20% knockout serum replacement, 1% NEAA, 1% glutamine, 1% β-mercaptoethanol (all Life Technologies), 20 ng ml−1 b-FGF (Peprotech). Alternatively, freshly derived hiPSCs were directly cultured in feeder-free medium StemMACSTM iPS-Brew XF (Miltenyi Biotec) or transferred from MEF without any adaptation. hiPSCs were passaged every 3–4 d on vitronectin-XF-treated plates using gentle cell dissociation reagent (both Stemcell Technologies). All cell lines were cultured at 37 °C and 5% CO2 atmosphere. All commercially available cell lines are certified mycoplasma free. All cells used in this study have been verified for absence of mycoplasma contamination before reprogramming and after hiPSC generation and expansion.

Microfluidic platform.

Microfluidic platforms were fabricated according to standard soft-lithographic techniques and molded in polydimethylsiloxane (PDMS) as previously described8,10. Briefly, Sylgard 184 (Dow Corning) was cured on a 200-μm-thick patterned SU-2100 photoresist (MicroChem) in order to obtain a single PDMS mold with multiple independent channels. The PDMS mold was punched and sealed on a 75 × 25 mm microscope glass slide (Menzel-Glaser) by plasma treatment. Channels were rinsed with isopropanol and distilled water to check proper flow before autoclaving. When required, a syringe step-motor pump (Cavro, Tecan) was used to periodically control medium flow rate into the microfluidic channels at 120 μL min−1 for 5 s. 0.5 ID Tygon tubings (Cole-Parmer) and 21G stainless steel needles were used to connect the microfluidic channels to the pump head.

Microfluidic platform with integrated medium distribution system.

Multilayer soft lithography was used to fabricate this type of microfluidic platform, composed of a 3-layer PDMS chip29,30. Transparent photomasks were printed at 8,000 dpi. A 4-inch silicon wafer (Siegert) was used for molding. Control mold was made by SU8 2050 (Microchem), obtaining 45 μm square channels. Flow mold had 45-μm round channels made by SPR 220-7 (Dow Corning); to avoid unintentional cross-section valves, 90-μm square channels have been made over the previous by a second SU8 2100 layer. Culture channel mold has 220-μm square section made by SU8 2100. Every mold was previously coated for 1 h at room temperature with chlorotrimethylsilane vapors (Sigma-Aldrich). Sylgard 184 (Dow Corning) was mixed at 5:1 ratio (base:cure agent) for the flow layer (FL), 20:1 ratio for the control layer (CL) and 10:1 ratio for the cell culture layer (CCL). All layers were partially cured at 333 K before peeling, cutting and punching: 45 min for FL, 60 min for CL and 70 min for CCL. After alignment, another 2 h at 353 K completed the curing. The PDMS chip was finally bonded by plasma activation on a large glass slip (75 × 50 mm, Ted Pella) covered with a thin (<0.5 mm) 20:1 PDMS layer. The microfluidic platform was fully assisted by an automated medium delivery and distribution system into the culture channels. A periodic 120 μL min−1 perfusion for 5 s was controlled twice a day by Cavro pumps (Tecan) and a lab-made software interface written in Labview (National Instruments). The experimental setup is shown in Figure 1a and Supplementary Figure 3.

Code availability.

Labview executables and MATLAB scripts produced in lab for this work are available as Supplementary Software.

Surface coating in microfluidic setup.

After autoclaving and before cell seeding, microfluidic culture channels were treated for surface functionalization. Extracellular matrix proteins were either adsorbed or chemically bound to the silanized glass bottom of each channel to provide a durable coating for cell culture (Supplementary Fig. 2). For the adsorbed substrates either fibronectin (Fn, 10 μg mL−1, Sigma-Aldrich) or type-A pork gelatin (Gel, 0.1% and 0.6%, Sigma-Aldrich) were incubated within the microfluidic channels for 1 h at 37 °C before rinsing with DPBS. For glass silanization, surface was treated either with a 5% water solution of (3-aminopropyl)-triethoxysilane (APTES, Sigma-Aldrich) for 20 min, or a 0.3% ethanol solution of 3-(trimethoxysilyl)-propyl methacrylate (TMSPMA, Acros Organics) for 5 min. Covalent bonding of adsorbed Gel on APTES was obtained by treating Gel with 0.5% v/v glutaraldehyde for 10 min. Covalent bonding of Gel on TMSPMA was obtained by treating Gel with methacrylic anhydride (Sigma-Aldrich) in PBS buffer for 1 h at 60 °C to produce an acrylate-reactive variant (GelMA). A TMSPMA–GelMA bonding was performed for 15 min by adding 0.1% ammonium persulphate and N,N,N′,N′-tetramethylenethylenediamine (Sigma-Aldrich) before GelMA (0.1% or 0.6%) injection into each channel. After surface biofunctionalization, channels were extensively rinsed with DPBS before cell seeding. In multilayered platforms, the distribution layer was functionalized by directly incorporating 0.3% TMSPMA in the PDMS before curing.


For technological validation, an mmRNA encoding for nuclear GFP (StemMACS Nuclear eGFP, Miltenyi Biotec) was used alone, carried into the cells by different transfection reagents: RNAiMAX (RiM, Life Technology), Stemfect (SF, Stemgent), and StemMACS Transfection Reagent (SM, Miltenyi Biotec), according to the transfection protocols below. Comparisons of transfection efficiencies between microfluidics and wells and between different transfection reagents were carried out with the same cell density. SF was used if not specified. SM was used for feeder-free protocols.

hiPSCs were generated via mmRNA-mediated reprogramming, adapting the protocols reported in literature for feeder11 and feeder-free31 hiPSC derivation, from multiple cell sources: human foreskin BJ and HFF-1 fibroblasts, skeletal muscle fibroblasts (SkMF) and urine-derived renal epithelial cells. In particular, reprogramming was optimized with incremental dosage of mmRNAs during the first four daily transfections, with 25%, 50%, 75%, 100% mmRNA amount of subsequent transfections at 0.5 ng mm−2. As reported by others32, the incremental mmRNA dosage for high reprogramming performances is important because a proper balance between efficient transcription factors delivery, cell proliferation rate and transfection-induced cell mortality has to be achieved.

For reprogramming with a feeder layer, NuFF-RQs were seeded at day −2. At day −1, the reprogramming target fibroblasts (either BJ or HFF) were seeded at different densities (5, 10 and 26 cell mm−2) in DMEM with 10% FBS. At day 0, 2 h before the first mmRNA transfection, medium was switched to defined TGF-β-free Pluriton reprogramming medium (Stemgent) supplemented with 200 ng mL−1 B18R (eBioscience) to suppress single-strand RNA-induced immune response mediated by type I interferons.

The transfection mix was prepared according to the manual of the StemMACS mRNA Reprogramming Kit (Miltenyi Biotec) pooling two solutions: the first obtained diluting 5X 100 ng μL−1 mmRNA of POU5F1, SOX2, KLF4, c-MYC, NANOG, LIN28 (provided by S. Wild and M. Jurk at Miltenyi Biotec) and nGFP, with stoichiometry 3:1:1:1:1:1:1 (ref. 11), in transfection buffer solution, and the second diluting 10 × transfection reagent in transfection buffer solution. The two solutions were mixed in 2:1 volume ratio (RiM and SF) or 3:1 (SM), and the final solution was incubated for 15 min (RiM and SF) or 20 min (SM). Where not specified, SF transfection reagent was used.

Transfections using the final mmRNA solution were started at day 0 and repeated daily for at least 12 d. In well, the final mmRNA solution was added dropwise while gently rocking the plate, 4 h before daily medium changes with B18R-supplemented Pluriton medium. In microfluidics, the final mmRNA solution was added to different percentages of B18R-supplemented Pluriton medium, pipetted inside a reservoir and manually or automatically perfused inside each channel. Fresh B18R-supplemented Pluriton was added to the reservoir and perfused after a transfection period of 4 h and 12 h thereafter. Experiments performed at doubled medium change frequency had the same mmRNA transfection incubation time (4 h) and medium changes every 4 h. To compensate for progressive NuFF death during reprogramming, NuFF-conditioned B18R-supplemented Pluriton medium was used from day 6. Pluriton medium was conditioned daily with 4 ng mL−1 b-FGF on a separate NuFF culture in well.

At the end of the transfection series, hiPSCs were cultured for two days in Pluriton medium without B18R. Few experiments were stopped at this point for performing immunofluorescence analysis and determining a reprogramming efficiency index, defined as the ratio of the number of double stained NANOG+/TRA-1-60+ colonies per 100 target cells seeded. Otherwise, hiPSC colonies were picked and passaged as previously described. Microfluidic hiPSC colonies were collected either by coring the rubber of the microfluidic chip with a biopsy punch or by a preferential detachment using a high flow rate corresponding to a shear-stress of 25 Pa32.

Feeder-free reprogramming was performed solely seeding BJ, HFF, SkMF or RE cells at day −1–, before the first transfection, and using B18R-supplemented Pluriton medium for the whole duration of reprogramming. RE cells were also kept in RE medium during the first 5 d of reprogramming transfections. Double-stained colonies for NANOG and TRA-1-60 obtained in microfluidics were catalogued as hiPSCs with progressive numbering and labeled thereafter. If not specified, hiPSCs were generated within the manually controlled version of the microfluidic chip (Supplementary Fig. 3).

Immunofluorescence and colorimetric assays.

Immunofluo-rescence analysis was performed in either conventional wells or microfluidic channels with the same protocols. Cells were fixed in 4% (w/v) paraformaldehyde (Sigma-Aldrich) for 10 min and stained with primary antibodies in 5% goat serum with 0.1% (v/v) Triton-X-100 (Sigma-Aldrich). Membrane markers were stained without cell permeabilization. Primary antibodies: POU5F1 1:200 and SSEA-4 1:250 (sc-5279; sc-21704, Santa Cruz), NANOG 1:100 (RCAB0004P-F, Reprocell), TRA-1-60 1:250 and TRA-1-81 1:250 (MAB4360; MAB4381, Millipore), SOX2 1:200 (NB110-37235, Novus Biologicals); AFP 1:200 and β-III-TUBULIN 1:200 (A8452; T3952, Sigma-Aldrich), BRACHYURY-T 1:100 (ab20680, Abcam), CK18 1:100 (GTX105624S, GeneTex), ALBUMIN 1:25 (MAB1455, R&D). Alexa488 or Alexa594 conjugated rabbit or mouse secondary antibodies (1:200) were used (A11005; A11001; A11008; A11012, Life Technologies). Nuclei were stained with Hoechst 33342 (Life Technologies). Images were acquired with a DMI6000B fluorescence microscope with motorized stage (Leica Microsystems). The alkaline phosphatase (AP) assay was performed either by a live AP kit in KnockOut DMEM (both Life Technologies) with a 45-min incubation, or after cell fixation by AP-staining kit II (Stemgent) with a 10-min incubation of the staining solution. Glycogen storage analysis was performed by periodic acid–Schiff (PAS) staining (Sigma-Aldrich) following manufacturer's specifications.

Image analysis.

Image analysis was performed as previously described10. Briefly, pairs of images of Hoechst 33342-stained nuclei and nGFP+ cells were analyzed using the software MATLAB R2012b (The MathWorks). Hoechst images were binary transformed for nuclei localization after contrast adjustment and morphological filtering. In the detected nuclei positions, mean nGFP fluorescence intensity was evaluated. A background correction was performed to account for differences in cell substrates and in image acquisition conditions. A threshold of fluorescence intensity was chosen to discriminate between nGFP positive and negative cells. A minimum of three images per channel or well (>100 cells per image) in a least five channels or wells were acquired at 10 × magnification for quantification of nGFP+ cells. At least five images per channel or well and five channels or wells were acquired at 20 × for quantification of single nucleus fluorescence intensity (>600 cells).

Stochastic model.

A mathematical model was developed to describe the transfection process of nGFP-encoding mmRNA at different concentrations (from 0.48 to 20 pg (100 cell)−1 in microfluidics) and transfection durations (1, 2, 4 and 12 h). The model is in the form of a stochastic simulation algorithm in order to capture the discrete nature of mmRNA-containing vesicles and cells (Supplementary Fig. 5).

Cells are described as a two-dimensional partially adsorbing boundary layer at the bottom of the culture volume. In each simulation, they are randomly positioned within a regular square grid, covering a 1-mm2 surface on the cartesian plane xz. The grid spacing, Δx and Δz, was fixed at 57.7 μm, corresponding to the side of a square having the same area as the average cell, determined experimentally. In the experiments we observed that BJ cells, seeded at a density of 250 cell mm−2, reached a density of 300 cell mm−2 after 24 h, when transfection occurred, and of 600 cell mm−2 after another 24 h, when nGFP images were taken. Thus, we simulated the transfection process at 300 cell mm−2 and assumed that proliferation rate was not affected by cell transfection.

The stochastic model is discrete in both time and space. A vesicle in position (x(t), y(t), z(t)) in a time interval Δt will jump to the neighbor position on the lattice with probability Pjump = DΔt/δ2 (ref. 33), where δ, equal to 10 μm, is the lattice spacing and the capture boundary layer34. Δt, the time interval simulated, is arbitrary as long as the inequality 2DΔt<<δ2 holds, where D is the diffusion coefficient of vesicles in medium. D was estimated to be 10−12 m2 s−1 by the Stokes–Einstein equation:

where kB is the Boltzmann constant, T absolute temperature (310.15 K), μ medium viscosity (691.6 10−6 Pa˙s35, approximated by water property), and R the vesicle radius, assuming a spherical shape (approximately 300 nm36).

With y named as the direction perpendicular to the plane of cell culture, the boundary condition at y = 0 (the cell plane) was defined as partially adsorbing; a vesicle is adsorbed with probability Pads = cRNA and reflected otherwise, where cRNA represents the vesicle concentration in the bulk and k is a positive constant that describes the reactivity of the boundary and was fitted using transfection experimental data (k = 1.4 × 103 μm2pg−1 for RiM, and k = 8 × 103 μm2pg−1 for SF). Thus, the overall probability a vesicle enters a cell at the boundary is given by two contributions, the probability of reaching the boundary and the probability of adsorption:

provided a vesicle is at distance δ from the boundary at the previous time point simulated.

The characteristic time of diffusion, τ, along the microfluidic culture channel height (H = 200 μm) was calculated by Einstein's relation:

and is approximately 5.5 h. Thus, we simulated the process within the capture boundary layer, assuming that the medium bulk plays as an infinite source of vesicles for transfections lasting up to 4 h. While, for a transfection time, ttransf, of 12 h, we included a multiplicative corrective factor, ɛ, in (2), defined by the following expression:

The model was solved using MATLAB.

RT-PCR and qPCR analyses.

Microfluidic channels were first perfused with D-PBS, then with iScript (Bio-Rad) for total RNA extraction. The solution was left in the channels for 2 min before collection. Total RNA was isolated with the RNeasymini kit (Qiagen), treated with DNase (Life Technologies), and quantified using NanoDrop spectrophotometer. RNA (0.1 μg) was reverse transcribed into cDNA (Life Technologies). The list of primers used is available in Supplementary Table 2. PCR was performed with Platinum Taq polymerase (Life Technologies). Electrophoresis was performed in a 2% (w/v) agarose gel with SYBR Green (Life Technologies). qRT–PCR was performed with TaqMan gene expression assay probes (Life Technologies) according to manufacturer's instructions. Reactions were performed on ABI Prism 7000 machine and results were analyzed with ABI Prism 7000 SDS software. GAPDH expression was used to normalize Ct values of gene expression, and data were shown as relative fold change to control cells using the delta–delta Ct method.

Western blotting analysis.

Whole cell lysate was obtained by solubilization of cells in 10 μL of 5% deoxycholic acid (DOC, Sigma-Aldrich) and cOmplete protease inhibitor (Roche). PAGE was performed with 4–12% NuPAGE polyacrylamide gels and MOPS buffer (Life Technologies). Proteins were blotted on PVDF membranes (Life Technologies) and detected with Carestream films (Kodak). A 1:1,000 dilution of primary antibodies (POU5F1, sc-5279 Santa Cruz; KLF4, sc-20691 Santa Cruz) and HRP-conjugated secondary antibodies (mouse, Bio-Rad; rabbit, Life Technologies) were used. Whole cell lysate from HES2 human embryonic stem cells was used as POU5F1 reference.

TGF-β assay.

Cell-conditioned media were collected every 24 h during the reprogramming process. In the microfluidic system this corresponds to mixing the medium collected three times per day. To activate latent TGF-β, conditioned media were heat treated for 5 min at 368 K before use. CAGA12 SMAD2/3 reporter HaCaT cell line was kindly provided by S. Piccolo's Lab (Department of Molecular Medicine, University of Padova) and cultured in DMEM supplemented with 10% FBS. For luciferase assay, cells were plated in 24-well plates at 90% confluence and incubated overnight in DMEM without serum. Cells were then treated with conditioned media for 10 h and supplemented with 1 μM TGF-β receptor inhibitor SB431542 (SB, Peprotech) where indicated. Luciferase expression was detected as previously described37. Data were normalized on total protein content, which was determined through Bradford assay.


3 × 106 cells were harvested, resuspended in 20 μL of Matrigel and injected subcutaneously into Rag−/− γc−/− mice. Teratomas were evident 8 weeks after cell implantation. Teratomas were harvested and fixed in 4% PFA, followed by cryosectioning. Hematoxylin and eosin staining were performed according to standard protocols.


Q-banded karyotype was performed by the Cytogenetic and Molecular Genetics Laboratory at the University of Brescia (Italy).

Microarray and bioinformatics.

A microarray computational analysis was performed on data of temporal progression of human fibroblast reprogramming (GSE50206) using MATLAB. Gene expression values were normalized by the 75th percentile shifts. Genes were selected that belonged to the intersection of GO:Extracellular region with TGF-β, WNT, MAPK, HIF and AKT–PI3K pathways (according to the KEGG database). Hierarchical clustering was performed with average linkage and Euclidean distance. Normalized data were clustered into three clusters using the k-means algorithm.

Microarray analyses performed in this study were executed 48 h after the last reprogramming mmRNA transfection; four freshly derived colonies of comparable size (8,000 cells) were selected both from the microfluidic system and from a parallel reprogramming experiment in well plates. After a 24-h conditioning in StemMACS iPS-Brew XF medium, each colony was split into two halves: one half (passage p0) was used for total RNA extraction and the other half was further expanded in feeder-free conditions for three passages (p3) in well plates before total RNA extraction from another sectioned colony of comparable size (Fig. 4a). Each colony was lysed using SuperAmp Lysis Buffer (Miltenyi Biotec) and stored appropriately according to the instructions of the SuperAmp Preparation Kit. The samples were sent on dry ice to Miltenyi Biotec, where amplification, cDNA quantification using ND-1000 Spectrophotometer (NanoDrop Technologies), evaluation of cDNA integrity using a 2100 Bioanalyzer (Agilent Technologies) and microarray analysis were performed. 250 ng of each of the cDNAs were labeled with Cyanine 3 and hybridized (17 h, 318.15 K) to an Agilent Whole Human Genome Oligo Microarrays 8x60K v2. Fluorescence signals of the hybridized Agilent Microarrays were detected using Agilent's Microarray Scanner System (Agilent Technologies). Agilent Feature Extraction Software (FES) was used to read out the microarray image files. Extracted signals were analyzed using GeneSpring v12.6 software (Agilent Technologies). Gene expression values were normalized by the 75th percentile shifts, and baseline corrected to the median of all samples. Differentially expressed genes between pairs of conditions were found using ANOVA with Tukey post-test (significance set at P < 0.05), combined with a two-fold change expression threshold. Differentially expressed genes were checked for functional enrichment using the DAVID bioinformatics database (http://david.abcc.ncifcrf.gov/). Principal component analysis (PCA) was performed with variance-based weights, and hierarchical clustering was performed with Euclidean distance and the nearest-neighbor linkage clustering method using MATLAB. Microarray data are available at the National Center for Biotechnology Information Gene Expression Omnibus database under the series accession number GSE59534.

Comparison with microarray data sets deposited in public databases (GEO and Synapse Commons Repository) was performed after selection of relevant samples obtained with similar Whole Human Genome Agilent arrays (GSE50206, GSE42445, syn1447097, syn1449098). Single-batch data normalization was performed as above. Only probes common to all the data sets were taken into consideration, and genes whose detection was compromised in at least one sample were excluded from the analysis. After this filtering procedure, 18,479 genes were further processed. Data sets merging was performed after batch-effect removal using an empirical Bayes method implemented in ComBat R-code38. PCA was then performed in MATLAB as above.

Aspecific differentiation.

Embryoid bodies (EB). hiPSC colonies were treated with CTK (0.1 mg mL−1 collagenase IV, 0.25% trypsin, 0.01 M CaCl2 and 0.2% KSR in dH2O) for 30 s, mechanically scratched with a serological pipette and resuspended in EB medium (DMEM/F12, 20% knockout serum replacement, 1% L-glutamine, 1% NEAA, β-mercaptoethanol, all Life Technologies). EB were cultured in ultra-low adhesive plates (Corning) for 20 d and then transferred on custom-made PDMS microwells with a Matrigel (BD)-coated glass bottom. Characterization was performed after 6 d.

Aspecific differentiation in monolayer. EB medium (without β-mercaptoethanol) was used to aspecifically differentiate hiPSC colonies for 6 d. In microfluidics, differentiation was performed by perfusing medium every 12 h. In wells, medium was changed every 48 h.

Early germ layer differentiation.

hiPSCs were differentiated in the following germ-layer-specifying media. Ectoderm: DMEM/F12, 1% NEAA, 1 mM L-glutamine, 0.1 mM beta-mercaptoethanol, 20% KnockOUT serum replacement (all Life technologies), 1 μM dorsomorphin. Mesoderm: Supplemented StemPro-34 (Life Technologies), 5 ng mL−1 b-FGF, 2 mM L-glutamine, 50 μg mL−1 ascorbic acid, 150 μg mL−1 transferrin, 0.3 ng mL−1 Activin-A, 10 ng mL−1 BMP-4, 46 μg mL−1 methyl-thio-glycerol. Endoderm: Days 1–2, RPMI, 2% B27 supplement, 100 ng mL−1 Activin-A, 50 ng mL−1 Wnt3a. Days 3–5, RPMI, 2% B27 supplement, 100 ng mL−1 Activin-A.

Cardiac differentiation.

Small molecules were used to promote cardiac differentiation of hiPSC colonies. RPMI with B27 without insulin (cardiac basal medium, CBM, Life Technologies) and with 10 μM CHIR99021 was perfused in microfluidics every 12 h for the first 24 h. Thereafter, the medium was changed every 24 h. CBM was used in the following 36 h and CBM with 4 μM IWP-4 in the next 24 h. CBM was then used until cardiac maturation at day 14 from the beginning of differentiation protocol. Medium change in wells was performed only at changes of medium composition as above.

Hepatic differentiation.

hiPSC colonies obtained in microfluidics were maintained in StemMACS iPS-Brew XF medium to grow over the channel surface. RPMI–B27 was supplemented with 100 ng ml−1 Activin-A and 0.5 mM sodium butyrate for 3 d. Medium was changed to KO-DMEM, 20% Serum Replacement (both from Invitrogen), 1 mM L-glutamine, 1% NEAA, 0.1 mM β-mercaptoethanol, 1% DMSO (Sigma-Aldrich) for 6 d. Hepatic-like cells were maturated with L15 medium (Sigma-Aldrich) supplemented with 8.3% FBS, 8.3% tryptose phosphate broth, 10 μM hydrocortisone 21-hemisuccinate, 1 μM insulin (all from Sigma-Aldrich) and 2 mM L-glutamine containing 10 ng ml−1 hepatocyte growth factor and 20 ng ml−1 oncostatin M (both from R&D) for 6 d. Medium was changed every 12 h in the microfluidic system and every 24 h in wells.

Statistical analysis.

For statistical analyses, single pairwise comparisons were analyzed using Student's t-test with P < 0.05 (*), P < 0.01 (**) or P < 0.001 (***) indicating significance. Multiple comparisons were performed by one-way ANOVA with Tukey post-test, with P < 0.05 (*), P < 0.01 (**) or P < 0.001 (***) indicating significance. Values are expressed by means and s.d. The variability of reprogramming efficiency with different cell batches within different microfluidic devices and channels was verified by nested ANOVA, using Minitab 17 statistical software. Throughout the text, n indicates the number of replicates, referring to a combination of independent experiments performed in at least two different chips or in two different batches of cells. A list of single-experiment conditions and analyses is available in Supplementary Table 3.

Accession codes.

Microarray data are deposited in the Gene Expression Omnibus (GEO) database with accession number GSE59534.


Primary accessions

Gene Expression Omnibus

Referenced accessions

Gene Expression Omnibus


  1. 1.

    et al. Core transcriptional regulatory circuitry in human embryonic stem cells. Cell 122, 947–956 (2005).

  2. 2.

    , & Activin/Nodal and FGF pathways cooperate to maintain pluripotency of human embryonic stem cells. J. Cell Sci. 118, 4495–4509 (2005).

  3. 3.

    et al. Activin/Nodal signaling maintains pluripotency by controlling Nanog expression. Development 136, 1339–1349 (2009).

  4. 4.

    et al. Induction of pluripotent stem cells from adult human fibroblasts by defined factors. Cell 131, 861–872 (2007).

  5. 5.

    et al. A late transition in somatic cell reprogramming requires regulators distinct from the pluripotency network. Cell Stem Cell 11, 769–782 (2012).

  6. 6.

    et al. A comparison of non-integrating reprogramming methods. Nat. Biotechnol. 33, 58–63 (2015).

  7. 7.

    et al. Functional differentiation of human pluripotent stem cells on a chip. Nat. Methods 12, 637–640 (2015).

  8. 8.

    , , & Optimal periodic perfusion strategy for robust long-term microfluidic cell culture. Lab Chip 13, 4430–4441 (2013).

  9. 9.

    et al. Induction of pluripotency in human somatic cells via a transient state resembling primitive streak-like mesendoderm. Nat. Commun. 5, 3678 (2014).

  10. 10.

    , , , & Stochastic model-assisted development of efficient low-dose viral transduction in microfluidics. Biophys. J. 104, 934–942 (2013).

  11. 11.

    et al. Highly efficient reprogramming to pluripotency and directed differentiation of human cells with synthetic modified mRNA. Cell Stem Cell 7, 618–630 (2010).

  12. 12.

    et al. A mesenchymal-to-epithelial transition initiates and is required for the nuclear reprogramming of mouse fibroblasts. Cell Stem Cell 7, 51–63 (2010).

  13. 13.

    et al. Sequential introduction of reprogramming factors reveals a time-sensitive requirement for individual factors and a sequential EMT–MET mechanism for optimal reprogramming. Nat. Cell Biol. 15, 829–838 (2013).

  14. 14.

    et al. Induced pluripotent stem cell lines derived from human somatic cells. Science 318, 1917–1920 (2007).

  15. 15.

    et al. Generation of human induced pluripotent stem cells from urine samples. Nat. Protoc. 7, 2080–2089 (2012).

  16. 16.

    et al. Generating a non-integrating human induced pluripotent stem cell bank from urine-derived cells. PLoS ONE 8, e70573 (2013).

  17. 17.

    et al. A vector-free microfluidic platform for intracellular delivery. Proc. Natl. Acad. Sci. USA 110, 2082–2087 (2013).

  18. 18.

    , , , & Microfluidic control of cell pairing and fusion. Nat. Methods 6, 147–152 (2009).

  19. 19.

    , & How microRNAs facilitate reprogramming to pluripotency. J. Cell Sci. 125, 4179–4787 (2012).

  20. 20.

    et al. Two miRNA clusters reveal alternative paths in late-stage reprogramming. Cell Stem Cell 14, 617–631 (2014).

  21. 21.

    et al. Deterministic direct reprogramming of somatic cells to pluripotency. Nature 502, 65–70 (2013).

  22. 22.

    et al. Derivation of novel human ground state naive pluripotent stem cells. Nature 504, 282–286 (2013).

  23. 23.

    , , , & Combinatorial modulation of signaling pathways reveals cell-type-specific requirements for highly efficient and synchronous iPSC reprogramming. Stem Cell Rep. 3, 574–584 (2014).

  24. 24.

    et al. Small molecules facilitate rapid and synchronous iPSC generation. Nat. Methods 11, 1170–1176 (2014).

  25. 25.

    , , & Translational strategies and challenges in regenerative medicine. Nat. Med. 20, 814–821 (2014).

  26. 26.

    , , , & Revisiting lab-on-a-chip technology for drug discovery. Nat. Rev. Drug Discov. 11, 620–632 (2012).

  27. 27.

    , & Human-on-chip for therapy development and fundamental science. Curr. Opin. Biotechnol. 25, 45–50 (2014).

  28. 28.

    et al. Generation of human induced pluripotent stem cells from urine samples. Nat. Protoc. 7, 2080–2089 (2012).

  29. 29.

    , , , & Monolithic microfabricated valves and pumps by multilayer soft lithography. Science 288, 113–116 (2000).

  30. 30.

    , , , & Versatile, fully automated, microfluidic cell culture system. Anal. Chem. 79, 8557–8563 (2007).

  31. 31.

    , , & Feeder-free derivation of human induced pluripotent stem cells with messenger RNA. Sci. Rep. 2, 657 (2012).

  32. 32.

    et al. Adhesion strength-based, label-free isolation of human pluripotent stem cells. Nat. Methods 10, 438–444 (2013).

  33. 33.

    & Reactive boundary conditions for stochastic simulations of reaction–diffusion processes. Phys. Biol. 4, 16–28 (2007).

  34. 34.

    et al. Predictive microfluidic control of regulatory ligand trajectories in individual pluripotent cells. Proc. Natl. Acad. Sci. USA 109, 3264–3269 (2012).

  35. 35.

    , & Viscosity of liquid water in the range −8 °C to 150 °C. J. Phys. Chem. Ref. Data 7, 941 (1978).

  36. 36.

    , , , & Delivery of RNA and its intracellular translation into protein mediated by SDS-CTAB vesicles: potential use in nanobiotechnology. BioMed Res. Int. 2013, 734596 (2013).

  37. 37.

    et al. USP15 is a deubiquitylating enzyme for receptor-activated SMADs. Nat. Cell Biol. 13, 1368–1375 (2011).

  38. 38.

    , & Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 8, 118–127 (2007).

Download references


This research was supported by Progetti di Eccellenza Ca.Ri.Pa.Ro. and Progetto Strategico TRANSAC of University of Padova. E.S. was supported by Progetti Giovani Studiosi 2010 (DIRPRGR10) of University of Padova. We would like to thank M. Piccoli (Fondazione Istituto di Ricerca Pediatrica Città della Speranza) for teratoma assay, A. Zoso (University of Padova, Italy) for western blot optimization, D. Bellotti and J. Skripac for karyotype analyses at the Cytogenetic and Molecular Genetics Laboratory at the University of Brescia (Italy), M. Mora, S. Zanotti and Telethon Network of Genetic Biobanks for providing SkMF, and I. Ferrarotti for ATD samples (IRCCS, Pavia, Italy). Furthermore, we would like to thank S. Rüberg, S. Tomiuk and M. Knauel for microarray analysis, and S. Wild and M. Jurk, all at Miltenyi Biotec, for providing material and support for reprogramming technologies. We thank S. Piccolo and G. Martello (University of Padova, Italy) for critical revision of the manuscript.

Author information

Author notes

    • Camilla Luni
    •  & Stefano Giulitti

    These authors contributed equally to this work.


  1. Department of Industrial Engineering, University of Padova, Padova, Italy.

    • Camilla Luni
    • , Stefano Giulitti
    • , Elena Serena
    • , Luca Ferrari
    • , Alessandro Zambon
    • , Onelia Gagliano
    • , Giovanni G Giobbe
    • , Federica Michielin
    •  & Nicola Elvassore
  2. Venetian Institute of Molecular Medicine (VIMM), Padova, Italy.

    • Camilla Luni
    • , Stefano Giulitti
    • , Elena Serena
    • , Luca Ferrari
    • , Alessandro Zambon
    • , Onelia Gagliano
    • , Giovanni G Giobbe
    • , Federica Michielin
    •  & Nicola Elvassore
  3. Miltenyi Biotec GmbH, Bergisch Gladbach, Germany.

    • Sebastian Knöbel
    •  & Andreas Bosio


  1. Search for Camilla Luni in:

  2. Search for Stefano Giulitti in:

  3. Search for Elena Serena in:

  4. Search for Luca Ferrari in:

  5. Search for Alessandro Zambon in:

  6. Search for Onelia Gagliano in:

  7. Search for Giovanni G Giobbe in:

  8. Search for Federica Michielin in:

  9. Search for Sebastian Knöbel in:

  10. Search for Andreas Bosio in:

  11. Search for Nicola Elvassore in:


S.G., C.L. and N.E. designed and performed the experiments; C.L. performed computational analyses. S.G., E.S. and L.F. characterized hiPSCs. S.G. and F.M. developed microfluidic hPSC culture. G.G.G. designed hepatic differentiation. A.Z. and O.G. designed and realized the automated microfluidic platform. S.K. and A.B. contributed to microarray analysis and reprogramming design. C.L., S.G. and N.E. designed the study and wrote the paper.

Competing interests

C.L., S.G., and N.E. are coinventors on a patent application describing the reprogramming and differentiation processes in microfluidics, application number PD2013A000220. S.K. and A.B. are employees of Miltenyi Biotec.

Corresponding author

Correspondence to Nicola Elvassore.

Integrated supplementary information

Supplementary information

PDF files

  1. 1.

    Supplementary Text and Figures

    Supplementary Figures 1–15 and Supplementary Tables 1 and 2

Excel files

  1. 1.

    Supplementary Table 3

    List of single-experiment conditions and reprogramming efficiency results. List of characterized clones and performed analyses.

  2. 2.

    Supplementary Data 1

    Bioinformatics analysis of microarray expression data (GSE50206). List of genes belonging to the intersection of GO: Extracellular region with TGF-beta, WNT, MAPK, HIF and AKT-PI3K pathways along human fibroblast reprogramming.

  3. 3.

    Supplementary Data 2

    Bioinformatics analysis of microarray data (GSE59534). Expression profiles of differentially expressed genes between the four conditions: p0 μF, p0 well, p3 μF, and p3 well.

  4. 4.

    Supplementary Data 3

    Bioinformatics analysis of microarray data (GSE59534). Functional annotation analysis of differentially expressed genes between the four conditions: p0 μF, p0 well, p3 μF, and p3 well.

Zip files

  1. 1.

    Supplementary Software

    MATLAB code for the stochastic mathematical model of mmRNA transfections. LabView code for controlling the automated microfluidic platform.


  1. 1.

    Automated reprogramming chip

    Validation of the automated microfluidic platform with integrated medium distribution system by food dyes.

  2. 2.

    Cells reprogrammed in microfluidics

    Morphological appearance of freshly obtained hiPSC colonies within 10 microfluidic channels at day 18 from beginning of reprogramming.

  3. 3.

    Reprogramming progression in microfluidics

    Time lapse of a feeder-free reprogramming process of human fibroblasts in microfluidics. Frames were captured every 12 h for 17 days. Day 0 represents the day of the first mmRNA transfection. At the end of the process, cells were fixed and stained for NANOG immunofluorescence detection.

  4. 4.

    Reprogramming progression in microfluidics

    Enlargement of a selected area from Supplementary Video 4.

  5. 5.

    Differentiation of freshly-derived hiPSC

    Contracting culture of cardiomyocytes. Freshly obtained microfluidic hiPSC colonies were directly differentiated in well towards the cardiac lineage.

About this article

Publication history






Further reading