Graphical Abstract

   graphic file with name fx1.jpg
   [29]Open in a new tab

Highlights

     * •
       Cell patterning enables quantitative yield and frequency analysis
       of hPSC fate
     * •
       A kinase inhibitor library is used to identify key PSC
       differentiation pathways
     * •
       mTOR inhibition enhances mesendoderm, hemogenic endothelium, and
       blood progenitors
     * •
       siRNA knockdown of RAPTOR, an mTORC1 component, phenocopies mTOR
       inhibition
     __________________________________________________________________

   Zandstra and colleagues perform an HTP screen on micro-patterned hPSCs,
   simultaneously measuring yield and frequency of four early
   differentiation trajectories. mTOR inhibitors such as rapamycin are
   found to enhance mesendoderm induction, and siRNA knockdown of
   RAPTOR/mTORC1 phenocopies this effect. Rapamycin also enhances
   hemogenic endothelium and blood progenitors, identifying mTOR signaling
   as a key node in mesoderm and blood regulation.

Introduction

   Human pluripotent stem cells (hPSCs) and their differentiated
   derivatives offer the exciting opportunity to develop tools to study
   and treat human diseases. However, robust and reproducible control of
   hPSC fate remains challenging. Small molecules offer one approach to
   control hPSC fate, and the discovery and characterization of these
   compounds can be facilitated by cell-based phenotypic high-throughput
   screening (HTS). Emerging data from hPSC assays has revealed variable
   and contradictory observations, even with matched cell lines and
   protocols ([30]Haibe-Kains et al., 2013). Although the factors
   underlying this variability are not completely known, population
   context has been identified as a main contributor to assay
   inconsistency ([31]Snijder et al., 2012). Spatially heterogeneous
   ([32]Peerani et al., 2007) micro-environmental factors such as
   endogenous ligands, extra-cellular matrix proteins (ECMPs), and cell
   subpopulations are strong regulators of hPSC fate. Specifically,
   spatial cell distribution has been shown to affect hPSC self-renewal
   ([33]Maherali and Hochedlinger, 2008), differentiation trajectories in
   both normal and patient-derived cells ([34]Cai et al., 2009,
   [35]Chambers et al., 2009), and disease phenotypes ([36]Sun et al.,
   2012).

   Consequently, robust assays that combine defined culture conditions
   with comprehensive analysis of cell responses to exogenous cues are
   needed. To this end, we developed a chemically defined cell
   patterning-based high-throughput (HTP) assay, engineering colony size,
   local cell density, medium composition, and substrate for rapid and
   robust measurement of hPSC fate responses to exogenous cues
   ([37]Nazareth et al., 2013). We applied the assay to screen a library
   of kinase inhibitors for effects on four early hPSC fates. For each
   compound, the change in yield and purity in the resulting pluripotent,
   neuroectoderm (NE), mesendoderm, and extra-embryonic populations were
   simultaneously tracked, allowing for estimation of selection and
   induction events. Our analysis identified mammalian target of rapamycin
   (mTOR) inhibitors, such as rapamycin, as having a strong
   mesendoderm-inducing effect on hPSCs. Rapamycin was subsequently shown
   to synergize with bone morphogenetic protein 4 (BMP4) and activin A to
   enhance BRACHYURY induction more than 3-fold, an effect that propagated
   to equivalent enhancements of hemogenic endothelium and blood
   progenitor cells. This study demonstrates the advantages of controlling
   micro-environmental parameters and measuring multiple subpopulation
   outputs in parallel on PSC fate screening assays. This strategy should
   enhance discovery in more complex and predictive multi-cell population
   drug-screening assays.

Results

A Kinase Inhibitor Screen of hPSCs Revealed Lineage-Specific Regulators

   We previously developed a 48-hr hPSC screen that employs control of
   spatial cell patterning to configure the hPSC micro-environment for
   rapid and robust response to exogenous cues (96μCP assay) ([38]Nazareth
   et al., 2013) ([39]Figure 1A). Single-cell OCT4 and SOX2 costaining
   enables simultaneous classification of pluripotent (OCT4^+SOX2^+), NE
   (OCT4^−SOX2^+), mesendoderm (OCT4^+SOX2^−), and extra-embryonic/other
   (OCT4^−SOX2^−) cell fates. This platform can be used to screen test
   factor effects on yield (percentage) and purity (absolute number of
   cells) of each subpopulation per colony using in-house software
   ([40]Figure 1B).

Figure 1.

   [41]Figure 1
   [42]Open in a new tab

   HTS of Small-Molecule Regulators of hPSC Pluripotency, Primitive
   Streak, NE, and Extra-embryonic/Other Cell-Fate Decisions

   (A) Assay design. hPSCs are micro-patterned on micro-contact printed
   200-μm diameter extracellular matrix proteins (ECMP) circles in 96-well
   plates. Control or base medium with test compounds is added to each
   well. After 48 hr, imaging is performed and single-cell OCT4 and SOX2
   expression levels are quantified.

   (B) Subpopulation purity and high yield are measured to determine
   inductive effects of compounds. Diagram demonstrates the changes in
   composition and purity of the “pink” cell population.

   (C) Compound distribution ranked by induction efficiency of
   pluripotency, NE, primitive streak, and extra-embryonic/other cell
   fates.

   (D) Sample composite colony images from controls and selected
   responders stained for DNA (DAPI, blue), OCT4 (green), and SOX2 (red).
   BMP4 is the OCT4^−SOX2^− control, CM is the OCT4^+SOX2^+ control,
   FGF2 + SB431542 is the OCT4^−SOX2^+ control, and BMP4 + activin A is
   the OCT4^+SOX2^− control. Colony outlined with dotted white line. Scale
   bar, 100 μm.

   (E) Calcein (viability) staining of unfixed cells at 48 hr treated with
   hit compounds. Non-viable control treated with methanol. Colonies are
   outlined with a dotted white line. Scale bar, 200 μm.

   (F) 67 putative hit compounds ranked by increasing viability. n = 3
   independent biological replicates. Error bars indicate SD.

   See also [43]Figure S1.

   We applied this platform to screen a collection of 400 small-molecule
   kinase inhibitors at two concentrations (0.2 μM and 1 μM). To ensure a
   significantly large sample size, we performed further analysis only on
   compounds which yielded >800 imaged cells per well, resulting in 707
   unique conditions including five controls on each plate: base medium
   (blank), PSC supporting mouse embryonic fibroblast (MEF)-conditioned
   (CM), NE inducing transforming growth factor β (TGF-β) inhibitor and
   fibroblast growth factor 2 (FGF2) (TiF), mesendoderm inducing BMP4 with
   activin A (BA), and extra-embryonic inducing BMP4. The overall
   distributions of the compounds with respect to each subpopulation
   indicated that mesendoderm and extra-embryonic/other inducing compounds
   were underrepresented ([44]Figure 1C). Control conditions showed high
   reproducibility across plates ([45]Figure S1A). Calculation of the Z′
   factor, a statistical parameter used to compare HTP assays ([46]Zhang
   et al., 1999), for the pattern-based (Z′ = 0.94)and non-pattern-based
   ([47]Desbordes et al., 2008) (Z′ = 0.29, p < 0.01) assays supported the
   rationale for controlling micro-environments to improve robustness
   ([48]Figures S1B and S1C). Sample images of control conditions and
   select responders are shown in [49]Figure 1D. Viability of endpoint
   cells was confirmed at 48 hr using calcein and Hoechst staining
   ([50]Figures 1E and 1F).

   To visualize factor effects on responding population cell yield and
   purity, we measured the percentage and average number of cells of each
   subpopulation per colony per well. For example, to visualize effects on
   the PSC population we subtracted the percentage and number of
   OCT4^+SOX2^+ cells obtained in the blank to obtain ΔOCT4^+SOX2^+ (%)
   and ΔOCT4^+SOX2^+ (number of cells per colony), respectively
   ([51]Figure 2A). Based on the subpopulation positive controls, we set
   thresholds for yield and purity to obtain a list of “responders.”
   Similar visualizations are shown for NE (OCT4^−SOX2^+) ([52]Figure 2B)
   and mesendoderm (OCT4^+SOX2^−) ([53]Figure 2C). Enlarged plots of the
   threshold regions showing compound names of responders are shown in
   [54]Figure S2. Visual inspection of responding compounds confirmed the
   expected yield and purity effects ([55]Figure 1D). Based on these
   results, we sought to determine whether responding compounds were
   enriched for specific inhibitor targets.

Figure 2.

   [56]Figure 2
   [57]Open in a new tab

   Pathway Enrichment Analysis

   (A) Increase in PSC purity (%OCT4^+SOX2^+) versus increase in yield
   (number of OCT4^+SOX2^+ cells per colony) obtained from 712 unique
   treatments. Values from blank were subtracted (graph origin represents
   base medium alone). Thresholds for purity and yield were used to
   classify “pluripotency responders.” Arrow indicates a sample responding
   compound.

   (B and C) Similar analysis as in (A) to classify (B) NE and (C)
   mesendoderm responders. Dotted lines outline the regions classified as
   responders.

   (D) Enrichment in PSC responders (shown in [58]Figure 1D). p Values
   were obtained for each target, using the hypergeometric distribution,
   and shown in −log base 10. Threshold corresponds to p = 0.001.

   (E and F) Similar analysis as in (D) for (E) NE and (F) mesendoderm
   responders.

   (G) Two-dimensional hierarchical clustering of subpopulation yield and
   purity data across treatments. Left: sample similarity tree, with
   samples clustered. Enhanced primitive streak cluster is in purple, PSC
   cluster in blue, and NE cluster in red. Right: detailed view of
   enhanced primitive streak cluster, indicating compound name,
   concentration, and reported kinase targets indicated in red.

   See also [59]Figures S2 and [60]S3; [61]Table S2.

Specific Kinase Inhibitor Targets for NE and Mesendoderm Induction Identified

   To identify fate-modulating pathways enriched in our library, we
   performed a hypergeometric test to obtain p values for the enrichment
   of each kinase target in the PSC, NE, and mesendoderm responder groups
   (five most enriched targets are shown in [62]Figures 2D–2F; count of
   unique compounds for each kinase target is shown in [63]Figure S3A;
   count of kinase targets in responders and total screen is shown in
   [64]Table S1). For the PSC-enhanced cluster no target family was
   statistically enriched, indicating that the tested kinase inhibitors do
   not robustly rescue hPSCs from differentiation in the blank lacking
   exogenous FGF2 and activin A ([65]Vallier et al., 2005). This suggests
   that, in contrast to differentiation, exogenous pathway agonists, or a
   combination of inhibitors, may be required to maintain pluripotency
   in vitro ([66]Tsutsui et al., 2011). For NE responders we obtained the
   highest enrichment for small molecules targeting the ERK/MEK pathway
   (p = 2.9 × 10^−5) ([67]Figure 2E), as well as a high enrichment (p =
   0.001) of small molecules targeting TGF-β superfamily type I activin
   receptor-like kinase (ALK) receptors ALK4, ALK5, and ALK7. Inhibition
   of TGF-β signaling has been well established to result in hPSC
   differentiation to NE ([68]Smith et al., 2008, [69]Vallier et al.,
   2004). We previously observed low levels of endogenous TGF-β signaling
   in the 96μCP assay, which likely contributes to lower than expected
   significance of ALK4/5/7 inhibitors. Lastly, in the mesendoderm
   responders we found the highest enrichment for mTOR inhibitors (p =
   8.1 × 10^−11) ([70]Figure 2F).

   In addition to target enrichment analysis, two-dimensional hierarchical
   clustering (707 conditions, eight outputs, 5,656 data points) revealed
   several distinct phenotype clusters ([71]Figure 2G, left). A cluster of
   compounds that enhanced PSC (blue dendrogram) also contained the CM
   control ([72]Figure S3B). Similarly, a cluster that enhanced NE (red
   dendrogram) also contained the TiF control ([73]Figure S3C). The few
   compounds that enhanced mesendoderm (purple dendrogram) clustered with
   two mesendoderm induction controls, BA and BA with FGF2 (BAF)
   ([74]Figure 2G, right). In this cluster, 53% of the compounds targeted
   mTOR and no other kinase was targeted by more than one compound.
   Enrichment analysis of the three clusters revealed identically
   significantly enriched pathways as our threshold-based method
   ([75]Figure S3D), further confirming the ERK/MEK and mTOR enrichments
   found in the screen. Based on these results we sought to further
   validate the ERK/MEK and mTOR results.

ERK/MEK Inhibition Induces hPSC Differentiation to NE in a BMP-Dependent
Manner

   Analyzing all the ERK/MEK inhibitors in our library, we observed a
   general increase in the percentage of NE cells at the expense of PSCs,
   with no changes in total cell number ([76]Figures 3A and 3B). To
   further investigate the effect of MEK inhibition, we retested PD0325901
   and found a significant reduction in PSC yield and increase in NE
   frequency (both p < 0.001) ([77]Figure 3C), but found no significant
   increase in NE cells per colony induced relative to the blank control
   (p = 0.43), indicative of selection ([78]Figure 3D). Reports are mixed
   on the effects of MEK inhibition (via PD0325901) on neural induction,
   and there is strong evidence that variability in endogenous factors
   such as BMP has hindered interpretation of these previous results
   ([79]Greber et al., 2011).

Figure 3.

   [80]Figure 3
   [81]Open in a new tab

   PSC and NE Subpopulation Screen Analysis

   (A) Subpopulation percentages from all screened MEK/ERK inhibitors,
   ranked by PSC percentage response.

   (B) Subpopulation cells per colony from all screened MEK/ERK
   inhibitors, ranked by PSC percentage response.

   (C) The MEK inhibitor PD-0325901 increases the percentage of early NE
   and decreases the percentage of PSCs. ^∗p < 0.001, significant
   difference from blank control; n ≥ 3 independent biological replicates.

   (D) MEK inhibitor PD-0325901 does not increase the number of early NE.
   ^∗p < 0.01, significant difference from blank control; n ≥ 3
   independent biological replicates.

   Error bars indicate SD. See also [82]Figure S4.

   To further investigate, we performed a meta-analysis of a previous
   dataset of the response of a panel of cell lines to a set of conditions
   including blank medium, PD0325901, and TiF, using the 96μCP assay
   ([83]Nazareth et al., 2013). This analysis demonstrated that PD0325901
   and TiF variably induce early NE in a cell-line- and passage-dependent
   manner ([84]Figure S4A). Next, to determine the dependencies of
   PD0325901-induced NE on other pathways, we compared the effect of
   PD0325901 on NE induction in the blank, with BA, or with 17 other
   single factors that are agonists or antagonists of major developmental
   pathways. When BMP4 was added with PD0325901, NE induction was reduced
   from 95% to less than 0.5% (p < 0.002), confirming that NE is
   exceptionally sensitive to BMP4, strengthening the possibility that
   endogenous BMP4 abolishes NE induction ([85]Figure S4B). Notably,
   PD0325901 induction of NE was affected by other factors, including
   FGF2, activin A, TGF-β, insulin-like growth factor 1, and heregulin β1,
   congruent with previous proposals that inhibition of MEK, BMP, and
   TGF-β signaling ([86]Greber et al., 2011) or dual BMP and TGF-β
   signaling ([87]Kim et al., 2010) may indeed robustly induce NE
   regardless of variations in endogenous signaling. Together, these data
   confirm that NE induction via MEK inhibition alone is highly variable
   across hPSC lines, support our hypothesis that this induction is
   sensitive to BMP inhibition, and demonstrate that our platform can
   reveal subtle molecular perturbations of PSC fate.

mTOR Inhibitors Dose-Dependently Induce Mesendoderm in hPSCs

   Our analysis indicated a strong effect of mTOR inhibition on
   mesendoderm induction. Conflicting reports show that mTOR inhibition
   inhibits PSCs and enhances serum-mediated differentiation toward
   mesoderm and endoderm ([88]Zhou et al., 2009), or may enhance the
   purity of pluripotent cells ([89]Easley et al., 2010). To validate that
   mTOR inhibitors enhance mesendoderm induction, we screened nine mTOR
   inhibitors at both 1 μM and 0.1 μM, and screened eight inhibitors of
   additional pathways implicated in hPSC regulation. Ranking the
   compounds according to level of effect on mesendoderm induction
   (%OCT4^+SOX2^−), the mTOR inhibitors clearly separated from the others
   and the majority of conditions tested significantly enhanced the
   percentage of OCT4^+SOX2^− cells relative to blank serum-free (SF)
   medium alone (p < 0.05) ([90]Figure 4A). Rapamycin at 1 μM was the
   strongest responder of all the conditions, comparable with the BA
   control. Rapamycin binds to FKBP12 and inhibits the kinase activity of
   mTOR complex 1 (mTORC1), which contains mTOR and RAPTOR ([91]Sabers
   et al., 1995). A separate complex containing mTOR and RICTOR (mTORC2)
   is thought to be rapamycin insensitive, although this is cell-line and
   context dependent ([92]Lamming et al., 2012). In contrast to rapamycin
   and its derivatives such as temsirolimus and everolimus, which only
   target mTORC1, next-generation dual mTORC1/2 inhibitors have been
   developed, such as AZD-8055 and KU-63794. Notably, within the 18 mTOR
   inhibitor treatments (nine compounds at two concentrations), the top
   half of ranked responders are enriched for compounds only targeting
   mTORC1 (p < 0.005).

Figure 4.

   [93]Figure 4
   [94]Open in a new tab

   mTOR Inhibitors in Defined Medium Upregulate Mesendoderm via TGF-β and
   BMP Pathways

   (A) Nine mTOR inhibitors in the OICR library were tested at 1 and
   0.1 μM. Additional inhibitors of kinases implicated in early
   development but not targeting mTOR (“other inhibitors”) were also
   tested. Small molecules were ranked by effect magnitude to separate
   mTOR inhibitors from others. ^∗p < 0.01, ^∗∗p < 0.001 compared with
   blank; n = 3 independent biological replicates.

   (B) Ten-point dose curves of mTOR inhibitors rapamycin and AZD-8055
   show a dose-dependent increase of % mesendoderm. ^∗p < 0.05 compared
   with blank control; n = 3 independent biological replicates. Blank and
   BA response shown as reference lines.

   (C) Ten-point dose curves of rapamycin show a bimodal increase in %
   PSCs. ^∗p < 0.05 compared with blank (0 rapamycin); n = 3 independent
   biological replicates.

   (D) Ten-point dose curves of rapamycin and AZD-8055 effect on number of
   PSCs per colony at subnanomolar concentrations of rapamycin but not
   AZD-8055. ^∗p < 0.05 compared with blank (0 rapamycin); n = 3
   independent biological replicates.

   (E) 14 factors (ligands and small molecules) targeting early
   development pathways tested alone and with rapamycin to screen for
   interactions. n = 3 independent biological replicates. ^∗p < 0.05,
   compared with blank + rapamycin condition.

   (F) Dose curves of rapamycin performed alone and with BMP4 (10 ng
   ml^−1) and activin A (100 ng ml^−1) showing average percentage
   of BRACHYURY^+ cells per well was quantified. ^∗p < 0.05 and ^∗∗p <
   0.005 both compared with the equivalent condition without rapamycin;
   n = 3 independent biological replicates. Response to blank and BMP4 and
   activin A alone (“BA”) are shown as reference lines.

   (G) Same experiment as in (F) showing the number of BRACHYURY cells per
   colony. ^∗p < 0.05, ^∗∗p < 0.005 compared with the equivalent condition
   without rapamycin; n = 3 independent biological replicates.

   (H) Images of cells stained for DNA (DAPI, blue) and BRACHYURY
   expression (red), with treatments as indicated. Colony outlined with
   dotted white line. Scale bar, 200 μm.

   Error bars indicate SD. See also [95]Figure S5.

   To further validate the effect of mTOR inhibitors on hPSCs, we
   performed ten-point dose curves of two mTOR inhibitors, rapamycin and
   AZD-8055, and again measured the four early OCT4/SOX2 subpopulations.
   As expected, both compounds increased the percentage of mesendoderm
   cells in a dose-dependent manner ([96]Figure 4B). Surprisingly, a
   bimodal response to rapamycin was observed in the PSC response, where
   at 0.001 and 0.01 nM there was a moderate but significant enhancement
   of the percentage and number of PSCs. This enhancement was not seen at
   concentrations above 0.1 nM, where mesendoderm induction peaks at
   0.1 μM ([97]Figure 4C). The yield, and not just percentage, of PSCs are
   enhanced by low concentrations of rapamycin ([98]Figure 4D). This
   bimodal response was seen with two additional replicates of rapamycin
   (not shown) but was not seen with AZD-8055, which may indicate that the
   mechanism is via mTORC1 inhibition ([99]Figure 4D). Additional studies
   are required to understand this effect more completely.

mTOR Inhibitor-Mediated Induction of Mesendoderm Is BMP-Receptor and
TGF-β-Receptor Dependent

   To gain insight into the mechanism behind mTOR inhibition inducing
   mesendoderm in hPSCs, we screened 14 agonists and antagonists of early
   development signaling pathways, alone and with rapamycin ([100]Figures
   4E and [101]S5). TGF-β1 and FGF2 moderately enhanced rapamycin-induced
   mesendoderm. Inhibition of either phosphatidylinositol-3-kinase (using
   LY-294002), a downstream effector of mTOR implicated in mesendoderm
   differentiation ([102]Singh et al., 2012), or Wnt (using IWP4) did not
   significantly alter the effect of rapamycin. The effects of rapamycin
   were abolished by the BMP type I receptor (ALK2/3) inhibitor
   LDN-193189, MEK inhibitor PD-032591, and TGF-β receptor (ALK4/5/7)
   inhibitor SB-431542. Consistent with our results, MEK inhibition has
   been shown to divert differentiation from mesendoderm ([103]Yu et al.,
   2011), indicating that mTOR inhibition is insufficient to overcome this
   effect. The LDN-193189 and SB-431542 results imply that rapamycin's
   effect on mesendoderm is dependent on BMP and TGF-β receptor-mediated
   signaling, which are both required for hPSC differentiation toward
   mesendoderm ([104]Vallier et al., 2009). Addition of either (or both)
   of these inhibitors to rapamycin, relative to rapamycin alone, results
   in loss of the OCT4^+SOX2^− population, no difference in the
   OCT4^+SOX2^+ population, and a moderately enhanced SOX2^+OCT4^−
   population ([105]Figure S5A), with no change in the total number of
   cells per colony ([106]Figure S5B).

   Based on the blocking effects of SB-431542 and LDN-193189, we reasoned
   that rapamycin, which induces mesendoderm alone in SF-defined medium,
   may act additively or synergistically with endogenous TGF-β/BMP
   signaling. We therefore generated a rapamycin dose curve with (rap+BA)
   and without (rap+BA) exogenous BMP4 (10 ng ml^−1) and activin A (100 ng
   ml^−1), and analyzed mesendoderm induction by BRACHYURY ([107]Zhang
   et al., 2008) expression. The percentage and number of BRACHYURY^+
   cells in both rap+BA and rap+BA increased in a rapamycin dose-dependent
   manner. The maximum percentage of BRACHYURY^+ cells achieved with
   rapamycin alone (19%) was comparable with the percentage achieved with
   BA alone (18%); however, rap+BA resulted in a more than 3-fold
   increased yield of BRACHYURY^+ (57%) compared with BA alone (p < 0.005)
   ([108]Figure 4F). The number of BRACHYURY^+ cells per colony was also
   enhanced in rap+BA relative to BA alone by 3.8-fold (p < 0.005)
   ([109]Figure 4G). Sample images are shown in [110]Figure 4H. Similar
   trends were observed in dose curves generated from non-patterned hPSCs
   with rapamycin, or an alternative mTOR inhibitor (temsirolimus) and a
   BA background ([111]Figures S5C–S5F).

Knockdown of RAPTOR Phenocopies the Mesendoderm-Inducing Effects of Rapamycin

   Functionally, mTOR operates as part of two protein complexes, mTORC1
   and mTORC2 (reviewed in [112]Sabatini, 2006), which are differentially
   regulated by upstream signals and have different downstream effectors.
   Although rapamycin was long thought to inhibit mTORC1, recent studies
   revealed that rapamycin can also inhibit mTORC2 in a
   cell-type-dependent manner ([113]Lamming et al., 2012). To examine the
   role of each mTOR complex in rapamycin-induced mesendoderm
   differentiation, we targeted RAPTOR (mTORC1 specific) and RICTOR
   (mTORC2 specific) with small interfering RNA (siRNA). PSCs were seeded
   in pluripotency medium with siRNA against either RAPTOR (siRAPTOR) or
   RICTOR (siRICTOR). Two days after seeding, the medium was exchanged
   with fresh medium under three conditions (+blank, +rapamycin, or +BA),
   in the presence of siRNA. Analysis of OCT4 and SOX2 expression was
   conducted at day 4 ([114]Figure 5A). In blank medium, siRAPTOR enhanced
   the percentage of mesendoderm (OCT4^+SOX2^−) cells relative to control
   scrambled siRNA (siCTRL) (p < 0.0005), but siRICTOR had no significant
   effect. Addition of both siRAPTOR and siRICTOR gave a phenotype similar
   to siRAPTOR alone, and gave rise to a significantly higher percentage
   of mesendoderm cells than siRICTOR alone (p < 0.05). In the presence of
   rapamycin, siRAPTOR again had a significant effect (p < 0.0005), and
   siRICTOR was not significantly different from siCTRL. This result
   indicates that rapamycin (at 0.1 μM) is not completely inhibiting
   mTORC1. In the presence of BA, siRAPTOR also enhances the percentage of
   OCT4^+SOX2^− cells relative to siCTRL (p < 0.0005). Interestingly, in
   this medium condition siRICTOR also enhanced the percentage of
   OCT4^+SOX2^− (p < 0.0005), indicating that the effect of mTORC2
   inhibition is dependent on TGF-β/BMP signaling. Focusing on the BA
   conditions, we repeated the siRNA assay and stained for BRACHYURY. As
   expected, siRAPTOR significantly enhanced both percentage
   ([115]Figure 5B) and number ([116]Figure 5C) of BRACHYURY^+ cells,
   similarly to rapamycin treatment. Sample images are shown in
   [117]Figure 5D. No effect was observed from siRICTOR on BRACHYURY
   induction, indicating that the effect of rapamycin on BRACHYURY
   induction is primarily through mTORC1 and not mTORC2. These results
   also indicate a divergence between the OCT4^+SOX2^− readout and
   BRACHYURY expression and further corroborate that rapamycin is acting
   via inhibition of mTOR signaling, not via non-specific effects.

Figure 5.

   [118]Figure 5
   [119]Open in a new tab

   mTOR Inhibition Synergizes with BMP4 and Activin A in Inducing
   Mesendoderm

   (A) The effect of siRNA on OCT4^+SOX2^− mesendoderm. siRNA was applied
   to non-patterned hPSCs in different medium conditions as noted. ^∗p <
   0.05, ^∗∗∗p < 0.0005; n = 3 independent biological replicates.

   (B) The effect of siRNA on BRACHYURY expression. siRNA was applied in
   different medium conditions as noted to non-patterned hPSCs. ^∗p <
   0.05; n = 4 independent biological replicates.

   (C) Same experiment as in (B), showing number of BRACHYURY^+ cells per
   field and total cells. ^∗∗p < 0.005, ^∗∗∗p < 0.0005; n = 4 independent
   biological replicates.

   (D) Images of cells stained for DNA (DAPI, blue) and BRACHYURY
   expression (red), with treatments as indicated. Scale bar, 500 μm.

   Error bars indicate SD.

mTOR Inhibition Impedes Endoderm and Enhances Mesoderm Differentiation

   Based on the observation that mTOR inhibition enhances the purity and
   number of mesendoderm cells, we investigated the effect of rapamycin on
   specific downstream mesoderm and endoderm progenitors. We first
   differentiated hPSCs, in either BA or rap+BA (0.1 μM) for 42 hr, and
   assessed various markers of NE (PAX3, PAX6), mesendoderm (MIXL1,
   BRACHYURY), mesoderm (MEIS1, MYB), cardiac precursor (MESP1, NKX2.5,
   PDGFRa), hematopoietic (RUNX1), definitive endoderm (EOMES), primitive
   endoderm (GATA6, AFP), and trophectoderm (HAND1) lineages via qRT-PCR
   ([120]Figure S6A). In general, rapamycin supplementation enhanced all
   differentiated lineage markers, with the exception of HAND1, a marker
   of trophectoderm, and EOMES, associated with definitive endoderm (DE).

   We next specifically tested the effect of rapamycin on DE induction.
   Differentiation to DE is a multi-step process including a 1-day
   induction phase (D0) where cells are exposed to activin A (100 ng
   ml^−1) and the glycogen synthase kinase 3β (GSK3β) inhibitor CHIR99021
   (CHIR, 2 μM) ([121]Rezania et al., 2012). We tested various
   concentrations of rapamycin at D0 and measured the resulting
   FOXA2^+SOX17^+ DE at day 5. Although DE differentiation protocols can
   be extremely efficient in specific cell lines ([122]Nostro et al.,
   2011), differentiation efficacy is known to vary greatly between
   different cell lines ([123]Nazareth et al., 2013). Thus we tested four
   cell lines with varying DE induction efficiencies and demonstrated that
   rapamycin dose-dependently reduced DE induction (%FOXA2^+SOX17^+ cells)
   in PDX1, HES2, and HES3-RUNX1-GFP cell lines, and had no effect on H9
   DE induction ([124]Figure 6A). These results show that rapamycin does
   not enhance endoderm differentiation and may be specific to mesoderm
   induction.

Figure 6.

   [125]Figure 6
   [126]Open in a new tab

   mTOR Inhibition Enhances HE and Blood Induction

   (A) Four hPSC cell lines were differentiated to DE using a 5-day
   protocol and analyzed for FOXA2^+SOX17^+ expression. Rapamycin was
   added as indicated during the D0 phase. n = 3 independent biological
   replicates. Error bars indicate SD.

   (B) Representative flow cytometry scatterplots of VECAD and CD34
   expression in day 8 hPSC-derived HE with (top) and without (bottom)
   rapamycin.

   (C) Time course showing VECAD^+CD34^+ cells from days 1–8 of
   differentiation showing percentage (top) and number per well (bottom)
   with and without rapamycin. N indicates number of independent
   biological replicates. Error bars indicate SD.

   (D) Comparison of the maximum VECAD^+CD34^+ percentage obtained with or
   without rapamycin. ^∗p < 0.05; n = 3 independent biological replicates.
   Error bars indicate SD.

   (E) Comparison of the fold increase in number of CD34^+VECAD^+ achieved
   with rapamycin relative to without rapamycin, on days 6 and 8. n = 3
   independent biological replicates. Error bars indicate 99% confidence
   interval.

   (F) CFC comparison of definitive blood obtained with and without
   rapamycin. n = 3 independent runs of the HES2 (one replicate) and
   RUNX1-GFP (two replicates) cell lines. Error bars indicate SD. ^∗p <
   0.05 (paired Student's t test).

   (G) Summary of the role of rapamycin in hPSCs.

   See also [127]Figure S6.

   Next, to test the effect of mTOR inhibition on mesoderm-derived
   lineages, we evaluated hematopoietic differentiation of hPSCs. During
   embryogenesis, hematopoietic cells are derived from specialized
   endothelial cells called hemogenic endothelium (HE), isolated using
   vascular endothelial cadherin (VECAD) and CD34 expression ([128]Eilken
   et al., 2009). To track HE induction in our studies, hPSC
   (HES3-RUNX1-GFP cell line) aggregates were cultured in chemically
   defined medium ([129]Pick et al., 2007) in the presence of rapamycin
   (0.1 μM) from day 0 to 2 and VECAD^+CD34^+ expression was assessed
   throughout an 8-day time course. Controls were cultured in the absence
   of rapamycin. The CD34^+VECAD^+ expression profile revealed that
   rapamycin addition from day 0 to day 2 enhances both purity and yield
   of HE ([130]Figures 6B and 6C). Note that day-6 rapamycin-generated
   CD34^+VECAD^+ cells were also CD73^− and CD184^− ([131]Figures S6B and
   S6C), further confirming their HE phenotype ([132]Ditadi and Sturgeon,
   2015). The maximum CD34^+VECAD^+ percentage reached in controls was 10%
   ± 4%, falling within the range of previously reported results in the
   literature ([133]Ditadi and Sturgeon, 2015), whereas in rapamycin 30% ±
   11% purity was attained (p < 0.05) ([134]Figure 6D). Rapamycin addition
   also enhanced HE purity in the HES2 cell line ([135]Figure S6D),
   demonstrating that our observations are not limited to the
   HES3-RUNX1-GFP line. Moreover, rapamycin conditions exhibited a 4-fold
   increase in CD34^+VECAD^+ cells on day 6 and a 2.5-fold increase on day
   8 (99% confidence intervals shown in [136]Figure 6E). Furthermore,
   kinase insert domain receptor (KDR, also known as vascular endothelial
   growth factor receptor 2) expression, an early marker of all blood
   cells ([137]Shalaby et al., 1997), nearly doubled on day 4 in rapamycin
   ([138]Figure S6E). To functionally assess whether early rapamycin
   treatment enhances blood progenitor induction, we seeded cells from
   cultures initiated either in the presence or absence of rapamycin in
   methylcellulose colony-forming cell (CFC) assays ([139]Csaszar et al.,
   2012). A 1.7-fold increase (p < 0.05) in the number of CFCs per 10^5
   cells was obtained in rapamycin-treated conditions (202 ± 42 CFCs [mean
   ± SEM]) compared with control conditions (117 ± 26 CFCs) ([140]Figures
   6F and [141]S6F). Additionally, gene expression analysis at days 0, 1,
   and 2 confirmed that rapamycin addition significantly increased the
   expression of BRACHYURY (p < 0.002) and MIXL1 (p < 0.0002)
   ([142]Figure S6G). These results demonstrate that rapamycin treatment
   during days 0 to 2 of hematopoietic differentiation of hPSCs enhances
   the blood progenitor phenotype. A summary of our findings on the role
   of mTORC1 in hPSC lineage-fate decisions is presented in
   [143]Figure 6G.

Discussion

   We have applied an engineered micro-environment based HTS platform to
   screen a kinase inhibitor library simultaneously for regulators of
   PSCs, NE, and mesendoderm. Small-molecule control of cell fate is
   attractive for scale-up purposes, and a focused library could be
   instrumental in elucidating endogenous regulators of hPSC fate
   decisions. In line with a previous screen ([144]Desbordes et al.,
   2008), we found no single compound able to maintain long-term
   pluripotency, indicating that exogenous pathway agonists or possibly
   combinations of small-molecule inhibitors may be required for hPSC
   maintenance. Our multi-lineage readout enabled the discovery of
   specific pathways endogenously activated in hPSCs amenable to
   small-molecule control of lineage-specific differentiation. Patterning
   enabled a more rapid (48 hr versus 7 days) assay and robust response
   than non-pattern-based screens ([145]Desbordes et al., 2008).
   Additionally, the short assay duration limits the number of emergent
   subpopulations, allowing simultaneous detection of these subpopulations
   to be tracked by OCT4 and SOX2 costaining. In conjunction with
   cell-number readouts, selection and induction events can be reasonably
   discriminated.

   Our analysis revealed that mTOR inhibitors added alone to SF-defined
   medium can induce mesendoderm, and our follow-up studies confirmed that
   rapamycin enhances the formation of HE and blood progenitors. Ten-point
   dose curves revealed a bimodal effect where rapamycin at low
   concentrations mildly enhances the purity and number of OCT4^+SOX2^+
   PSCs, and higher concentrations reduce the OCT4^+SOX2^+ population and
   enhance mesendoderm differentiation. These results reconcile two
   previous studies: in line with [146]Easley et al. (2010) we observe no
   negative effect on pluripotency of rapamycin at low concentrations, and
   in line with [147]Zhou et al. (2009) we observe loss of pluripotency
   and induction of mesendoderm at higher concentrations. In contrast to
   Zhou et al., we found that rapamycin strongly inhibits DE induction.
   This difference may be serum mediated or cell-line dependent.
   Additionally, emerging evidence suggests that separate induction
   mechanisms may confer distinct mesoderm/endoderm subtype potentials
   ([148]Mendjan et al., 2014).

   Although both GSK3β and mTOR inhibition appear to enhance mesendoderm,
   rapamycin-treated hPSCs appear to generate significantly higher
   frequencies of CD34^+VECAD^+CD73^− HE phenotype cells compared with
   CHIR99021 (GSK3β inhibitor)-treated conditions. GSK3β inhibition has
   been reported to direct hPSC differentiation to blood progenitor
   ([149]Sturgeon et al., 2014) and cardiac cells ([150]Lian et al.,
   2012), supporting its role as a driver of lateral plate-derived
   tissues. Our results indicate that mTOR inhibition is an efficient
   driver of hPSC differentiation to blood progenitor competent mesoderm.
   Comparing the mesendoderm subtypes that emerge in GSK3β- and
   mTOR-inhibited differentiation conditions is an important future
   direction of our study.

   The observation that subnanomolar concentrations of rapamycin have a
   moderate enhancing effect on pluripotency corresponds with findings of
   [151]He et al. (2012) showing that rapamycin at 0.01 and 0.03 nM
   enhances mouse somatic cell reprogramming efficiency nearly 2-fold
   while concentrations of 0.05 nM or greater have no effect or reduce
   reprogramming efficiency. Autophagy regulates homeostasis of
   pluripotency-associated transcription factors including OCT4 in hPSCs,
   with autophagy inhibition increasing transcription factor levels yet
   leading to differentiation ([152]Cho et al., 2014). Accordingly, low
   levels of mTOR inhibition may inhibit autophagy and enhance OCT4 levels
   in hPSCs sufficiently to enhance pluripotency, with higher levels of
   mTOR inhibition increasing OCT4 levels beyond a threshold to initiate
   differentiation. Alternatively, given that differentiated cell types
   such as hPSC-derived extra-embryonic endoderm are known to inhibit
   pluripotency ([153]Peerani et al., 2007), it may be that rapamycin
   selectively inhibits these cell types, leading to increased hPSC
   numbers. Another possibility is that the cell-fate effects of discrete
   mTOR inhibition levels may be directly mediated by differential SMAD
   activation, as mTORC1 inhibition activates SMAD1 and SMAD5 in human
   prostate cancer cells ([154]Wahdan-Alaswad et al., 2012), and mTORC2
   inhibition can enhance SMAD2 and SMAD3 activity via regulation of the
   SMAD2/3-T220/T179 linker residue ([155]Yu et al., 2015). Understanding
   the complex interplay between discrete levels of mTOR signaling,
   autophagy, pluripotency transcription factor homeostasis, signaling
   pathway activation, and effects on cell-fate choice and subpopulation
   dynamics remains an important direction of future studies.

   The methods we present here are applicable to systems with
   heterogeneous subpopulations and complex micro-environmental
   regulation, such as in vitro stem cell and cancer models, and can be
   applied iteratively. Changes in micro-environmental cues have been
   attributed to cell-fate transitions ([156]Bonfanti et al., 2010), and
   heterogeneity in micro-environmental context (such as population size,
   local cell density, and relative cell position) can deterministically
   explain phenotypic variance ([157]Snijder et al., 2009). Improved
   techniques to control the cellular micro-environment in conjunction
   with improved single-cell analytics will further develop our
   understanding of molecular cell-fate regulation.

Experimental Procedures

Cell Culture

   We obtained hESC lines H9 (WA09) and H7 (WA07) from the WiCell Research
   Institute, and HES3-RUNX1-GFP (HES3-derived RUNX1-GFP) from A. Elefanty
   (Monash University). All experiments were performed with H9 unless
   stated otherwise. These cell lines were cultured according to
   previously published methods ([158]Nazareth et al., 2013). All cell
   stocks tested negative for mycoplasma contamination. See
   [159]Supplemental Experimental Procedures for additional cell line
   information.

Micro-Patterning hPSCs onto 96-Well Plates and Application of Small Molecules
or Controls

   We patterned hPSCs on Matrigel into standard 96-well plates as
   previously described ([160]Nazareth et al., 2013) (see
   [161]Supplemental Experimental Procedures). Dissociation of hPSCs was
   performed using TrypLE for 3 min. TrypLE was inactivated with medium
   containing 20% KO-serum replacement (Invitrogen). Cells were
   centrifuged and resuspended in NutriStem (NS; StemGent 001-0005) and
   10 μM ROCK inhibitor Y-27632. Cells were seeded at 10^5 cells per well
   (or as described in the text) into 96-well plates, coated with either
   patterned or non-patterned Matrigel, and incubated for 6 hr. Cells were
   then washed twice with PBS and incubated for 42 hr in fresh test medium
   (SF supplemented with growth factors or small molecules, or CM). SF
   medium consists of DMEM/F12, 1× non-essential amino acids, 50 U ml^−1
   penicillin, 50 μg ml^−1 streptomycin, 10 μg ml^−1 bovine transferrin,
   0.1 mM β-mercaptoethanol (all Invitrogen), 2% fatty acid-free Cohn's
   fraction V BSA (Serologicals), 1× trace elements A, B, and C
   (Mediatech), 50 μg ml^−1 ascorbic acid (Sigma), and 7 μg ml^−1
   recombinant human insulin. The 400-compound library was obtained from
   the Ontario Institute of Cancer Research (OICR) ([162]Tables S2 and
   [163]S3). The following control conditions were included on each
   96-well plate: blank (SF medium with no added factors), CM, B (100 ng
   ml^−1 BMP4), BA (10 ng ml^−1 BMP4 and 100 ng ml^−1 activin A), and TiF
   (40 ng ml^−1 FGF2 and 10 μM SB431542).

Immunocytochemistry and High-Content Image Analysis

   Immunocytochemistry was performed as previously described
   ([164]Nazareth et al., 2013; see also [165]Supplemental Experimental
   Procedures). Single-cell x-y-coordinate and protein-expression data
   were exported in tab-delimited text files and imported into in-house
   software, ContextExplorer (Joel E.E. Ostblom, E.J.P.N., and P.W.Z.,
   unpublished data) for exploration of colony-level details. For figures,
   16-bit TIFF images were obtained for each channel, identical contrast
   adjustment was performed on all controls, and channels were combined
   into pseudo-colored composite images.

96-Well siRNA Transfection

   Human PSCs were first seeded into non-patterned MG-treated 96-well
   plates at 25,000 cells per well in NS, and incubated for 48 hr. For
   siRNA transfection, Lipofectamine RNAiMAX transfection reagent
   (Invitrogen) was used as per manufacturer's instructions, adapted for
   hPSCs and 96-well plates (see [166]Supplemental Experimental
   Procedures).

Viability Analysis

   To assess viability, we modified the patterned hPSC assay as follows.
   Instead of fixing at 48 hr, calcein AM (Invitrogen) (1:1,000) and
   Hoechst 33342 (1:1,000) were added directly to the medium, incubated
   for 30 min, and then imaged.

Differentiation of hPSCs toward Blood Progenitors and Definitive Endoderm

   For ultra-high-throughput production of size-specified cell aggregates,
   400-μm diameter micro-well plates (commercially available as
   Aggrewells, STEMCELL Technologies) were manufactured in-house in
   24-well plates ([167]Ungrin et al., 2008). For depletion of MEFs, hPSCs
   were dissociated with TrypLE and plated onto Geltrex (diluted 1:50) for
   48 hr in NS. After 48 hr, TrypLE was added for 5 min and then quenched
   with 50% fetal bovine serum in DMEM/F12. The resulting single-cell
   suspension was seeded at 6 × 10^5 cells per well of a micro-well plate
   (500 cells per aggregate) in the presence of Y-27632 (for 24 hr) at
   1,500 rpm for 5 min. Aggregates were cultured in hypoxia for 8 days in
   base medium supplemented with cytokines BMP4 (40 ng ml^−1, R&D
   Systems), vascular endothelial growth factor (VEGF) (50 ng ml^−1, R&D),
   SCF (40 ng ml^−1, R&D), basic FGF (5 ng ml^−1, Peprotech), and
   rapamycin (0.1 μM, Sigma). Note that rapamycin is highly labile, and we
   recommend minimizing freeze-thaw cycles and titrating for maximum
   effect. The base medium consists of StemPro34 (Invitrogen), ascorbic
   acid (50 μg ml^−1, Sigma), L-glutamine (1% v/v, Invitrogen),
   penicillin/streptomycin (1% v/v, Invitrogen), 1-monothioglycerol (4 ×
   10^−4 M, Sigma), and transferrin (150 μg ml^−1, Roche). On day 2,
   hPSC-derived cells were transferred to low-cluster six-well plates, in
   medium without rapamycin, and cultured for an additional 6 days shaken
   at 85 rpm (Orbi-Shaker XL, Benchmark Scientific). Endoderm induction to
   FOXA2^+SOX17^+ DE was performed as previously described ([168]Rezania
   et al., 2012).

qRT-PCR

   qRT-PCR was performed as described previously ([169]Onishi et al.,
   2012). In brief, RNA was isolated using the PureLink RNA Mini Kit
   (Ambion) and reverse transcribed with Superscript III (Life
   Technologies). PCR was performed on the QuantStudio 6 Flex (Life
   Technologies) using SYBR Master Mix (Roche) and primers provided by the
   Center for Commercialization of Regenerative Medicine. Gene NCBI-RefSeq
   accession numbers are provided in [170]Table S4.

CFC Assay

   A minimum of 1.5 × 10^5 cells were seeded on day 8 in 35-mm Greiner
   dishes in methylcellulose-based MethoCult H4435 Enriched medium
   (STEMCELL). Samples were scored based on morphology 14 days after
   plating as described by technical manuals provided by STEMCELL.

Statistical Analysis

   Statistics were computed using one-way ANOVA, two-factor ANOVA, or
   linear regression as indicated. All statistics were computed in MATLAB
   using p values as indicated. N values are independent replicates, not
   technical replicates, except where noted. Enrichment analysis was
   performed using the hypergeometric distribution in Microsoft Excel. Z′
   factor was used to assess assay quality ([171]Zhang et al., 1999). Z′
   is defined as Z′ = 1 − (3σ[positive] + 3σ[negative])/|μ[positive] −
   μ[negative]|, where μ and σ are the mean and SD of the positive
   (CM-treated) and negative (BMP4-treated) control as labeled.
   Hierarchical clustering was performed with MeV (MultiExperiment Viewer,
   [172]http://www.tm4.org/) using Euclidan distance as the similarity
   metric (centered) and centroid linkage as the clustering method.
   Exploratory analysis was performed with Microsoft Excel and Tableau
   (Tableau Software, [173]www.tableau.com).

Author Contributions

   E.N. designed, performed, and analyzed most experiments. N.R. performed
   all blood-induction experiments. T.Y. provided cell-culture support.
   M.P. and R.A. provided the kinase inhibitor library and related
   support. E.N. and P.W.Z. designed the project and wrote the manuscript.

Acknowledgments