Abstract

   Diffuse invasion is the primary cause of treatment failure of
   glioblastoma (GBM). Previous studies on GBM invasion have long been
   forced to use the resected tumor mass cells. Here, a strategy to
   reliably isolate matching pairs of invasive (GBM ^INV ) and tumor core
   (GBM ^TC ) cells from the brains of 6 highly invasive patient‐derived
   orthotopic models is described. Direct comparison of these GBM ^INV and
   GBM ^TC cells reveals a significantly elevated invasion capacity in GBM
   ^INV cells, detects 23/768 miRNAs over‐expressed in the GBM ^INV cells
   (miRNA ^INV ) and 22/768 in the GBM ^TC cells (miRNA ^TC ),
   respectively. Silencing the top 3 miRNAs ^INV (miR‐126, miR‐369‐5p,
   miR‐487b) successfully blocks invasion of GBM ^INV cells in vitro and
   in mouse brains. Integrated analysis with mRNA expression identifies
   miRNA ^INV target genes and discovers KCNA1 as the sole common
   computational target gene of which 3 inhibitors significantly suppress
   invasion in vitro. Furthermore, in vivo treatment with 4‐aminopyridine
   (4‐AP) effectively eliminates GBM invasion and significantly prolongs
   animal survival times (P = 0.035). The results highlight the power of
   spatial dissection of functionally accurate GBM ^INV and GBM ^TC cells
   in identifying novel drivers of GBM invasion and provide strong
   rationale to support the use of biologically accurate starting
   materials in understanding cancer invasion and metastasis.

   Keywords: 4‐aminopyridine, glioblastoma, KCNA1, miRNA, patient derived
   orthotopic xenograft
     __________________________________________________________________

   Using 6 highly invasive glioblastoma (GBM) PDOX models, this study
   isolates matching pairs of invasive (GBM ^INV ) and tumor core (GBM ^TC
   ) cells, demonstrates an elevated invasion capacity in GBM ^INV cells,
   identifies a novel miRNA signature (miRNA ^INV ) of GBM invasion, and
   discovers a commonly shared gene, KCNA1, from miRNA ^INV regulated
   genes that can be pharmacologically inhibited to block GBM invasion in
   vivo.

   graphic file with name ADVS-8-2101923-g007.jpg

1. Introduction

   Glioblastoma multiforme (GBM) is the most malignant brain tumor in
   children and adults. Despite multimodal therapies and significant
   advances in the understanding of tumor biology^[ [70]^1 ^] and
   molecular subgroups,^[ [71]^2 , [72]^3 , [73]^4 ^] the prognosis of
   patients with GBM remains extremely poor,^[ [74]^5 ^] with 5‐year
   survival rates between 5% and 15% in children^[ [75]^6 ^] and a 1‐year
   survival rate of ≈10% in adults.^[ [76]^7 ^] Diffuse infiltration of
   tumor cells into surrounding normal brain tissue, a hallmark of GBM
   growth, is the primary cause of tumor recurrence and treatment
   failure.^[ [77]^8 ^] While major advances have been made in
   understanding the biology of GBM by studying cells from the tumor core,
   little is known about the invasive GBM cells (GBM ^INV ) that migrate
   deep into surrounding normal brain tissues despite the findings of “go
   or grow” mechanism^[ [78]^9 , [79]^10 ^] and epithelial to mesenchymal
   transition.^[ [80]^11 , [81]^12 ^] This is because these GBM ^INV cells
   are not amenable to surgical removal for study, as aggressive surgical
   resection of normal tissue carries the risk of serious and permanent
   neurological deficits.^[ [82]^6 , [83]^13 ^] Additionally, although it
   has been suggested that the blood‐brain barrier (BBB) could be
   compromised and consequently “leaky” in the GBM tumor core (GBM ^TC ),
   GBM ^INV cells are frequently protected by an intact and functional
   BBB, making them even less vulnerable to chemotherapeutic agents than
   GBM ^TC cells.

   Most past and current studies on GBM invasion were and are still forced
   to utilize tumor tissues resected from the primary tumor mass for
   biologic analyses. Although these GBM ^TC cells may be capable of
   invasive growth, they are not actively invading at the time of harvest.
   Since tumor invasion is a complex biological process involving
   functional modifications and dynamic interactions between tumor cells
   and the microenvironment, there remains an urgent need for
   biologically‐ and functionally‐accurate GBM ^INV cells to identify key
   genetic drivers of GBM invasion.

   To overcome this barrier, we developed a panel of patient‐derived
   orthotopic xenograft (PDOX or orthotopic PDX) mouse models through
   direct implantation of pediatric GBM (pGBM) surgical specimens into
   matched locations in the brains of SCID mice. Detailed characterization
   showed that these PDOX tumors replicate the histopathological features
   of pGBM, maintained key genetic abnormalities of the original patient
   tumors, and importantly, remained highly invasive.^[ [84]^14 , [85]^15
   , [86]^16 ^] These transplantable PDOX models thus provide a reliable
   resource for isolating paired and functionally accurate GBM ^INV cells
   from normal mouse brain tissues/parenchyma and GBM ^TC cells from the
   tumor core for biological studies of pGBM invasion. The inclusion of
   different molecular subtypes of pGBMs should further facilitate the
   discovery of commonly shared or subtype‐specific biological changes.

   Accumulating data suggest that microRNAs (miRNA), noncoding RNAs of
   ≈20–23 bps,^[ [87]^17 , [88]^18 ^] play a key role in GBM invasion.^[
   [89]^19 ^] miRNAs often bind to target mRNAs through partial
   complementary pairing and either suppress mRNA translation or reduce
   mRNA stability. They are shown to regulate multiple cellular processes
   including cell division, differentiation, and death.^[ [90]^20 ^] While
   the study of miRNAs in pGBM is still in its infancy, over‐expression of
   miRNAs, including miR‐34a, ‐124, ‐128, ‐137, and ‐145, has been
   detected in adult GBM and shown to suppress self‐renewal, inhibit
   tumorigenesis,^[ [91]^21 ^] trigger cell cycle arrest,^[ [92]^22 ^] or
   apoptosis,^[ [93]^23 , [94]^24 ^] and promote invasion in vitro in
   cultured GBM cells and in vivo in subcutaneous xenografts.^[ [95]^19 ,
   [96]^25 ^]

   In this study, we utilized 6 PDOX mouse models of pGBM of different
   subtypes to harvest functionally accurate, paired GBM ^INV cells (that
   have invaded into the surrounding normal mouse brain tissue) and GBM
   ^TC cells (from the primary tumor mass) to examine their functional
   differences in migrating into normal brains, followed by the analysis
   of differentially expressed miRNAs via global microRNA profiling,
   validation of the functional roles of the candidate driver miRNAs both
   in vitro and in vivo in mouse brains using purified GBM ^INV cells. We
   subsequently examined genes that mediate the driver miRNA‐induced pGBM
   invasion, and finally identified a novel candidate driver gene KCNA1
   and demonstrated the therapeutic efficacy of pharmacological targeting
   of this invasion‐driver gene in blocking pGBM invasion and prolonging
   PDOX survival times.

2. Results

2.1. Patient‐Derived Orthotopic Xenograft Tumors Replicate the Highly
Invasive Phenotype of Pediatric Glioblastoma In Vivo

   Diffuse infiltration of tumor cells into surrounding normal brain
   tissue is one of the hallmarks of GBM. To confirm that our pGBM PDOX
   models replicated this important biological feature, we performed a
   systematic analysis of the invasive capacity and mode of tumor cell
   migration of pGBM cells in six PDOX models that had been
   sub‐transplanted in vivo in mouse brains for 6–8 generations as
   described previously (Figure [97] 1a,b).^[ [98]^15 , [99]^26 ^] These
   models were subgrouped as proliferative, proneural, and mesenchymal
   through gene expression analysis (Figure [100]1a),^[ [101]^2 ^] and
   MYCN, MID, G34, and pedRTKIII subtypes with DNA methylation
   profiling(Figure [102]1a).^[ [103]^27 , [104]^28 ^] When the
   tumor‐bearing mice became moribund, whole mouse brains were harvested,
   paraffin embedded, and serially sectioned (>160 sections/mouse brain).
   Standard H&E staining consistently revealed a large tumor mass
   surrounded by a characteristic “ragged edge” (Figure [105]1c and Figure
   [106]S1, Supporting Information). To positively identify single and/or
   small clusters of invasive pGBM cells, we performed immunohistochemical
   (IHC) staining using human‐specific antibodies against vimentin (VIM)^[
   [107]^15 , [108]^29 , [109]^30 ^] and used a straight line reticle
   (eyepiece micrometer) to measure the distances of migration under a
   Nikon 3 head teaching microscope from which 2–3 investigators can
   examine the same fields of the same slides at the same time. In tumors
   with short invasion that can be captured in one image (from 4× to 20×),
   ImageJ was applied to digitally measure the distances. The border line
   between tumor core and invasion (or the edge of tumor core) was defined
   as the front of tumor core where tumor cells were lined up facing
   normal brain tissues; while the leading front of invasion was
   determined by the tumor cells that migrated the farthest (or deepest)
   into the normal brain. A line drawing from the leading front invasive
   cells perpendicular to the tumor core edge was used to measure the
   distances (Figure [110]1c). For each PDOX models, at least three mouse
   brains each with >3 slides of the largest cross sections of tumor mass
   were included. Despite the different molecular subtypes, diffuse
   invasion was detected in all six models (Figure [111]1c and Figure
   [112]S1, Supporting Information). Most of the leading invasive edge was
   composed of single tumor cells followed by increasingly larger
   micro‐tumors closer to the tumor core, covering a distance ranging from
   525 to 2083 µM (mean 1109 ± 375.2 µm). Migration along blood vessels,
   that is, perivascular invasion,^[ [113]^31 ^] was detected in all 6
   models, ranging from 289.2 to 2058.3 µm (914.4 ± 466.7 µm); while
   invasion along neural fibers ranged from 170 to 2666.7 µm (1158 ± 524
   um). Seeding into the cerebral spinal fluid (CSF) was observed in some
   sections; the distances varied from 525 to 2750 µm (1881.3 ± 541.1 µm)
   (Figure [114]1c). Together, these data demonstrated maintenance of the
   invasive pGBM phenotype and all 3 routes of GBM tumor migration in vivo
   in our PDOX models, with single cell migration as the predominating
   mode.

Figure 1.

   Figure 1
   [115]Open in a new tab

   PDOX pGBM tumors were highly invasive in vivo. a) Clinical,
   pathological, and molecular subtype information of the six pGBM tumors.
   b) Orthotopic pGBM tumor implantation strategy and representative gross
   appearance of pGBM PDOX models. Tumor cells (1 × 10^5 in 2 µL) from six
   pGBMs were directly implanted into right cerebra of NOD/SCID mm mice
   (1.5 mm anterior and 3 mm deep) (left panel). The animals were
   monitored daily until they developed neurological deficits or became
   moribund, at which time they were euthanized. Formation of PDOX tumors
   can frequently be observed (arrow, right panel). c) Representative
   images showing the modes of intra‐cerebral invasion in PDOX models
   (left panel) and the quantitative analysis of migration distances of
   all six models (right panel). In additional to H&E staining, human pGBM
   xenograft cells were positively identified through IHC staining using
   human‐specific antibodies against vimentin (arrow). In all six models,
   invasion through single cells, along blood vessels (perivascular
   invasion), along neural fibers, and spread through cerebral spinal
   fluid (CSF) were observed. The distances between the invasive front and
   the border line of tumor core (white dotted line) were measured and
   graphed. Scale bar = 100 µM. n = 3 in each condition. Data are shown as
   mean ± SD.

2.2. Spatial Dissection to Isolate Matching Pairs of Invasive (GBM ^INV ) and
Tumor Core (GBM ^TC ) Cells

   To isolate GBM ^INV and GBM ^TC cells, freshly harvested whole mouse
   brains were sectioned into 1 mm slices to enable gross identification
   of the primary tumor mass (Figure [116] 2a) and to facilitate
   microscopic dissection of the tumor core (from which GBM ^TC cells were
   collected) from “normal” mouse brain tissue (from which GBM ^INV cells
   were collected) (Figure [117]2b).^[ [118]^32 , [119]^33 , [120]^34 ^]
   To purify human GBM ^INV and GBM ^TC cells, we utilized FITC‐conjugated
   human HLA‐ABC antibodies and a cocktail of APC‐conjugated antibodies
   specific to mouse CD24, CD90, CD117, CD133 and performed florescence
   activated cell sorting (FACS). In the invasive front (from “normal”
   mouse brain tissues), human HLA‐ABC^+ cells (GBM ^INV ) ranged from
   10.6% to 54.4% (33.0 ± 0.68%) of the viable cells; in the tumor core,
   human GBM ^TC cell proportions ranged from 85.9% to 97.2% (93.1 ±
   0.58%) (Figure [121]2c and Figure [122]S2, Supporting Information).
   These data provided a quantitative estimate of the diffusive invasion
   of pGBM in mouse brains, using a novel strategy to harvest matched
   pairs of functionally distinct GBM ^INV and GBM ^TC cells for
   biological studies.

Figure 2.

   Figure 2
   [123]Open in a new tab

   GBM ^INV cells possess stronger invasive capacity than their matching
   GBM ^TC cells. a) Slicing of fresh whole mouse brain to facilitate easy
   identification of tumor core. Whole mouse brains were placed on a mouse
   brain matrix and sliced at 1 mm thickness into 10–12 slices. b)
   Separation of tumor core (tumor) from “normal” mouse brains under
   stereotactic microscope. The border between tumor and normal (circle)
   was identified following general guidelines of human brain tumor
   resection during surgery.^[ [124]^32 , [125]^33 , [126]^34 ^] “Normal”
   mouse brain tissues (containing GBM ^INV cells) and tumor mass
   (containing GBM ^TC cells) were placed in cold (4 °C) growth medium in
   separate Petri dishes and dissociated into single cell suspension using
   Gentle Dissociator (Miltenyi). c) Purification of GBM ^INV and GBM ^TC
   cells through FACS. Cell suspensions were incubated with
   FITC‐conjugated monoclonal antibodies against human HLA‐ABC and
   APC‐conjugated monoclonal antibodies against mouse major
   histocompatibility antigen by FACS. The mouse cells (APC‐positive and
   FITC‐negative) were gated out together with the dead cells (propidium
   iodine high). Data are shown as mean ± SD. d) In vitro assay showing
   significantly increased invasive capacity of GBM ^INV cells than the
   matching GBM ^TC cells in two models under two different growth
   conditions. The purified GBM ^INV and GBM ^TC cells from IC‐1406GBM
   (1406) and IC‐3752GBM (3752) were cultured as neurosphere (NS) in
   serum‐free media supplemented with EGF and bFGF and monolayer (Mono)
   cells in traditional FBS‐based medium. The invasive capacity of GBM ^TC
   and GBM ^INV was examined in triplicates by CytoSelect 96‐Well Cell
   invasion assay (*P < 0.05) (Data are shown as mean ± SD). e) In vivo
   validation of higher invasive capacity of GBM ^INV cells than that of
   the matching GBM ^TC cells. Purified GBM ^INV and GBM ^TC cells from
   IC‐1406GBM (1406) and IC‐3752GBM (3752) models were implanted
   separately into the brains of SCID mice. The animals were euthanized
   when they develop signs of neurologic deficits or become moribund.
   Paraffin sections were stained with H&E and the distances from the
   “border” of tumor core (red line) to the far front of the invasive edge
   (black line) were measured (arrow). Note the formation of invasive
   satellite tumors in mouse brains implanted with GBM ^INV cells (arrow
   in the upper panel). Tumor sizes and depths of pGBM invasion were
   quantitated by ImageJ (**P < 0.01, *P < 0.05) (n = 3 per group. Data
   are shown as mean ± SD).

2.3. GBM ^INV Cells Possess Significantly Stronger Invasive Capacity both In
Vitro and In Vivo

   To test our hypothesis that GBM ^INV cells possess stronger migratory
   capacity than those in the GBM ^TC cells, we first compared their
   invasive capability in vitro with a standard invasion assay. Since
   monolayer tumor cells maintained in traditional fetal bovine serum
   (FBS)‐based media do not share biological features with neurospheres
   propagated in serum‐free media (supplemented with EGF and βFGF), which
   favors the growth of cancer stem cells,^[ [127]^15 , [128]^30 ,
   [129]^35 , [130]^36 ^] we incubated pGBM cells in both types of growth
   media to have a better coverage of cell subpopulations and to
   understand the differences between the monolayer and the neurosphere
   cells. GBM ^INV cells from two pGBM models (IC‐1406GBM and IC‐3752GBM)
   exhibited a 29–46% increase in invasion compared to the matching GBM
   ^TC cells (P <0.05), and 3D neurosphere cells were significantly more
   invasive (37–46% higher) than the monolayer cells (Figure [131]2d).
   Since neurospheres grew in suspension, the scratch assay that measures
   the cell motility was not performed.

   To further validate these findings in vivo, we directly implanted
   freshly purified GBM ^INV and GBM ^TC cells from these two pGBM models
   into the brains of SCID mice and examined their invasive capacity on
   paraffin sections via H&E and IHC staining. As anticipated, invasion
   into surrounding normal brain was observed in the xenografts derived
   from GBM ^TC cells, confirming the maintenance of invasive capacity of
   a fraction of GBM ^TC cells even though they were not “invading” at the
   time of harvesting. In xenografts derived from GBM ^INV cells, however,
   the depth of invasion was significantly longer compared to GBM ^TC
   cells (1246 vs 105.1 µm in IC‐1406GBM and 1266.3 vs 114 µm in
   IC‐3752GBM) (Figure [132]2e). While GBM ^TC cells formed large
   intra‐cerebral tumor masses, it was the GBM ^INV cells that developed
   invasive satellite tumors (Figure [133]2e). These in vitro and in vivo
   data demonstrated the functional differences between GBM ^INV and GBM
   ^TC cells, thereby highlighting the importance of using functionally
   accurate GBM ^INV cells in understanding GBM invasion.

2.4. Novel miRNA Drivers of Pediatric Glioblastoma Invasion

   MiRNAs have been implicated in driving biological processes of human
   cancers^[ [134]^37 ^] including tumor invasion and metastasis.^[
   [135]^38 , [136]^39 ^] To test our hypothesis that GBM ^INV cells
   depend on a unique set of miRNAs for invading into normal brain tissue,
   we compared the miRNA expression profiles of matched GBM ^INV and GBM
   ^TC cells derived from the 6 PDOX pGBM models using TaqMan MicroRNA
   array. In addition to four control assays provided by the vendor, we
   included normal human cerebral tissue obtained from warm autopsy (≈4 h
   postmortem) as a reference. Differences in miRNA expression between GBM
   ^INV and GBM ^TC cells were directly compared and the fold changes
   calculated by 2^−ΔCT. Of the 768 miRNAs detected, 23 were significantly
   upregulated (>twofold) in the GBM ^INV cells (hereafter designated as
   miRNA ^INV ), and 22 miRNAs upregulated in GBM ^TC cells (hereafter
   referred as miRNA ^TC ) in at least 4 of the 6 pGBM models
   (Figure [137] 3a and Table [138]S1, Supporting Information).

Figure 3.

   Figure 3
   [139]Open in a new tab

   Differentially expressed miRNAs between GBM ^INV and GBM ^TC cells. a)
   Hierarchical clustering of over‐expressed and downregulated miRNA in
   GBM ^INV cells. miR‐126, miR‐487b, and miR‐369‐5p were selected for
   functional study from the 23 upregulated miRNAs, and miR‐185 and
   miR‐589 were selected from the 22 downregulated miRNAs in least 4 of
   the 6 pGBM models (P < 0.05). b) Representative images showing
   successful infection of pGBM cells by Lentivirus‐miR‐GFP. The purified
   GBM ^INV cells from IC‐1406GBM were grown as neurospheres and infected
   with lentivirus‐miRNA‐off (miR487b‐off, miR‐126‐off, miR‐369‐5p‐off,
   and mix‐off) for 72 h (MOI 1:1) and examined for the expression of GFP.
   Non‐infected cells were included control. c) Confirmation of
   lentivirus‐mediated miRNA knock‐down of miRNA ^INV using RT‐qPCR. Both
   neurosphere (NS) and monolayer (Mono) cells derived from GBM ^INV cells
   of IC‐1406GBM (1406) and IC‐3752GBM (3752) were tested. (**P < 0.05)
   (Data are shown as mean ± SD).

2.5. Silencing miRNA ^INV in GBM ^INV Cells Suppresses Invasion In Vitro

   To examine the functional role of the upregulated miRNAs in the GBM
   ^INV cells (i.e., miRNA ^INV ) we selected two newly discovered miRNAs
   that exhibited high fold changes and high frequency, that is,
   miR‐369‐5p with >3.11‐folds in 5/6 models and miR‐487‐5p with >
   3.3‐fold in 4/6 models, as well as, one miRNA (miR‐126, >2.7‐fold in
   5/6 models) that was reported to be involved in tumor invasion.^[
   [140]^40 , [141]^41 ^] We examined the effects of loss‐of‐function and
   gain‐of‐function of these 3 microRNAs through lentivirus‐mediated
   transduction assays. Successful transduction was confirmed with
   Lenti‐GFP (Figure [142]3b) and the efficient knock‐down of target miRNA
   ^INV (>70%) with RT‐qPCR (Figure [143]3c).

   For the loss‐of‐function analysis of miRNA ^INV , the invasive capacity
   of puromycin‐selected and GFP^+ GBM ^INV cells from two highly invasive
   PDOX models, IC‐1406GBM and IC‐3752GBM, was examined using CytoSelect
   96‐Well Cell Invasion Assay in quadruplicates. As shown in Figure [144]
   4a, silencing miR‐126, ‐369‐5p, and ‐487b with Lenti‐miRNA‐126‐off,
   ‐369‐5p‐off, and ‐487b‐off, (MOI = 1:1 for 72 h) did not affect cell
   proliferation in either the GBM ^INV monolayer or neurosphere cultures
   but induced significant suppression of invasion in GBM ^INV cells grown
   as neurospheres (P < 0.05 compared to the untreated and the GBM ^INV
   cells transduced with Lenti‐non‐target‐off, n = 3) from both pGBM
   models. In the monolayer cells, only invasion of IC‐1406GBM ^INV cells
   was inhibited (Figure [145]4a and Figure [146]S3a, Supporting
   Information). These data indicated the selectivity of the miRNA ^INV
   function in cell invasion, particularly in the 3D neurospheres that
   exhibited increased invasive capacity (in Section [147]2.3), and
   support their role as candidate miRNA drivers of pGBM invasion.

Figure 4.

   Figure 4
   [148]Open in a new tab

   Functional validation of miRNA ^INV in GBM ^INV cell invasion both in
   vitro and vivo. a) In vitro loss‐of‐function assay showing the
   suppression of GBM ^INV cell invasion by lentivirus mediated silencing
   of miRNAs ^INV . The puromycin‐selected and FACS‐purified GFP^+ GBM
   ^INV cells from IC‐1406GBM (1406GBM ^IN ^V ) and IC‐3752GBM (3752GBM
   ^INV ) were examined both as neurospheres (NS) and monolayer cells
   (Mono). Data were normalized to the cell only group of neurospheres and
   presented as % of control. While cell proliferation was not affected (P
   > 0.05) (top panel), silencing of miR‐126, ‐369‐5p, ‐487b alone, and in
   combination (miR‐mix‐off) with Lentivirus‐miRNA‐off led to significant
   suppression of cell invasion (lower panel, P < 0.05) as examined by
   CytoSelect 96‐Well Cell Invasion Assay. (**P < 0.01, *P < 0.05 compared
   to the control group, n = 3. Data are shown as mean ± SD). b) In vitro
   gain‐of‐function assay showing the activation of GBM ^TC cells
   following lentivirus mediated transduction of miRNA ^INV . GBM ^TC
   cells (that were not actively invading) were transduced with
   Lentivirus‐miRNA to increase the expression of the miRNA ^INV . Cell
   proliferation was not affected (upper panel), but cell invasion was
   increased, particularly in neurosphere (NS) cells from IC‐1406GBM
   (1406GBM ^TC ) and IC‐3752GBM (3752GBM ^TC ) (**P < 0.01, *P < 0.05
   compared to the control group, n = 3. Data are shown as mean ± SD). c)
   In vivo confirmation of suppressed GBM ^INV cell invasion following the
   silencing of miRNA ^INV s. GBM ^INV cells from IC‐1406GBM (1406GBM ^IN
   ^V ) were transduced with Lentivirus‐miRNA‐off followed by
   puromycin‐selection. The FACS‐purified GFP^+ GBM ^INV cells were
   implanted into the brains of NOD/SCID mice (1 × 10^5 cells per mouse
   brain) and monitored for signs of neurological deficits or moribund
   when the animals were euthanized. 1406GBM ^INV cells were identified
   through H&E staining of whole mouse brains and IHC staining of
   paraffin‐embedded sections using human‐specific antibodies against VIM.
   d) The slides with the largest cross section of intra‐cerebral
   xenografts were examined to quantitatively evaluate the tumor size and
   the distances (arrow)between the invasive front (red line) and tumor
   core “board line” (black dotted line) were measured by ImageJ (*P <
   0.05). Scale bars represent 100 µM. Data are shown as mean ± SD.

   To determine if gain‐of‐function of miRNA ^INV promotes invasion in GBM
   ^TC cells, which have lower levels of these 3 miRNAs, we transduced the
   non‐invading tumor core cells IC‐1406GBM ^TC and IC‐3752GBM ^TC cells
   with Lenti‐miRNA‐126, ‐487b, and ‐369‐5p in quadruplicates. The
   increased expression of these miRNA ^INV (alone or in combination) did
   not alter cell proliferation, similar to GBM ^INV cells (Figure [149]4b
   and Figure [150]S3b, Supporting Information) but resulted in
   significantly elevated invasion in 3D neurospheres of IC‐1406GBM ^TC
   and in monolayers of IC‐3752GBM ^TC . In monolayers of IC‐1406GBM ^TC
   and neurospheres of IC‐3752GBM ^TC , overexpression of a single miRNA
   ^INV failed to promote invasion; however, simultaneous overexpression
   of all 3 miRNA ^INV caused a significant increase in invasion
   (Figure [151]4b and Figure [152]S3b, Supporting Information). These
   data indicated collective/cooperative activities of these miRNA ^INV in
   promoting pGBM invasion and suggested a complex nature of the
   underlying biology of GBM invasion.

2.6. Silencing miRNA ^INV Significantly Suppresses Pediatric Glioblastoma
Invasion in Mouse Brains

   GBM invasion is an active process involving dynamic interactions
   between tumor cells and their microenvironment. To validate the
   functional roles of miRNA ^INV in vivo in a microenvironment similar to
   human brain tissue, FACS‐purified IC‐1406GBM ^INV cells were transduced
   with Lenti‐mir‐126‐off, ‐369‐5p‐off, and ‐487b‐off (MOI 1:1) to silence
   the 3 miRNA ^INV and subsequently implanted into the brains of NOD/SCID
   mice (1 × 10^5 cells/mouse brain, n = 5 per group). Compared with the
   100% (5/5) tumor uptake rate seen with untreated IC‐1406GBM ^INV cells
   and 80% (4/5) in the non‐target lentivirus‐transduced group, the tumor
   take rates were reduced to 60% (3/5) after implantation of cells with
   Lenti‐miRNA‐126‐off, 50% (3/6) with Lenti‐miRNA‐369‐5p‐off, 33% (2 of
   6) with Lenti‐miRNA‐487b‐off, and 60% (3/5) with a combination of all 3
   Lenti‐miRNA‐off (Figure [153]4c). Animal survival times were not
   significantly different among the tumor‐bearing mice (Figure [154]S5,
   Supporting Information).

   We next examined whether silencing the 3 miRNA ^INV blocked GBM ^INV
   invasion in vivo in mouse brains. Except for the miR‐487b‐off group, in
   which 2 mouse brains were analyzed, there were 3 mouse brains in all
   the remaining groups (control, non‐target, miR126‐off, miR‐369‐5p‐off,
   and miR‐combination‐off). To quantitatively evaluate the invasive
   potential, slides with the largest cross‐section of intra‐cerebral
   xenografts were examined (Figure [155]4c). The non‐target control
   (miR‐non‐target‐off) and untreated IC‐1406GBM ^INV exhibited similar
   invasive capacity. Lentiviral‐mediated silencing of miR‐126, ‐369‐5p,
   and ‐487b caused significant reduction of invasion depth, ranging from
   >75% by miR‐487b‐off to ≈90% by miR‐369‐5p‐off and >95% by miR‐126‐off
   (Figure [156]4d). When the tumor sizes were compared, the differences
   among the six groups were not significantly different (Figure [157]4d)
   and IHC examination of stem cell (Nestin), neural (MAP2), glial (GFAP),
   cell proliferation (Ki67), and mitochondrial markers failed to reveal
   major differences as well (Figure [158]S6, Supporting Information).
   Altogether, silencing the miRNA ^INV (miRNA‐487b, ‐369‐5p, and ‐126) in
   GBM ^INV cells blocked pGBM invasion and reduced tumorigenicity in
   vivo, supporting a critical role of these miRNA ^INV in maintaining the
   invasive phenotype of pGBM cells.

2.7. miRNAs ^INV Targeted a Set of Shared Genes and Signaling Pathways

   miRNA‐mediated gene regulation is very complex.^[ [159]^42 ^] The
   activity of a given miRNA on a transcript may result in target mRNA
   degradation, blockage of translation, or increased mRNA expression.^[
   [160]^18 ^] Further, a single miRNA can target multiple mRNAs to
   coordinately regulate their expression; in contrast, multiple miRNAs
   can target a single mRNA. To identify the mRNA targets of the 3 miRNA
   ^INV , we 1) searched TargetScan for an updated list of target genes of
   the 3 miRNA ^INV (miR‐126, ‐487b, and ‐369‐5p), 2) performed global
   gene expression profiling in the same six pairs of GBM ^INV and GBM ^TC
   cells using normal childhood cerebral RNA as control, and 3) generated
   a list of differentially expressed genes of the 3 miRNA ^INV between
   GBM ^INV and GBM ^TC , (fold difference >1.5 or <0.5, P[INV/TC] <
   0.05). The P‐values of differentially expressed genes between GBM ^INV
   and normal human cerebral tissues (P[INV/Normal] ) and between GBM ^TC
   and normal tissues (P[TC/Normal] ) were also calculated.

   For miR‐126, a total of 231 target genes were differentially expressed,
   including 126 downregulated (<0.5‐fold) and 105 upregulated genes
   (>1.5‐fold) in GBM ^INV compared with GBM ^TC cells (Figure [161] 5a
   and Table [162]S2, Supporting Information). For miRNA‐487b, there were
   37 target genes (23 downregulated and 14 upregulated) (Figure [163]5a
   and Table [164]S3, Supporting Information), 22 (62.8%) of which were
   shared targets of miR‐126 (Figure [165]5A). For miR‐369‐5p, which has
   not been associated with any human disease, only seven target genes
   were found (Figure [166]5a and Table [167]S4, Supporting Information).
   The levels of most of the genes identified in the tumor core (GBM ^TC )
   were not significantly different from those in normal human cerebral
   tissues, including 203/247 (81.8%) genes targeted by miR‐126, 33/35
   (94.2%) by miR‐487b, and 6/7 (85.7%) by miR‐369‐5p (Tables
   [168]S2–[169]S4, Supporting Information). Therefore, these genes could
   have been missed if only GBM ^TC cells were utilized to compare with
   normal tissues.

Figure 5.

   Figure 5
   [170]Open in a new tab

   KCNA1 is a common target gene of miRNA ^INV . a) Integrated analysis of
   mRNA profiling with the 3 miRNA ^INV via TargetScan identified the
   private and shared target genes of miR‐126, ‐487b, and ‐369‐5p. KCNA1
   is the only gene commonly targeted by all 3 miRNA ^INV .b)
   Protein‐protein interaction network analysis with STRING identified
   major binding partners of KCNA1. c) List of down‐ or upregulated miRNAs
   in the GBM ^INV cells that also target KCNA1. d) Over‐expression of
   KCNA1 mRNA in the GBM ^INV cells compared with the matching GBM ^TC
   cells in 4/6 pGBM models examined. Data are shown as mean ± SD. e,f)
   Elevated expression of KCNA1 protein in vivo in the invasive front (GBM
   ^INV ) as compared with the tumor core (GBM ^TC ) cells as analyzed
   with IHC in the six pGBM PDOX models. Slides incubated without the
   primary antibodies were uses as control.

   Pathway enrichment analysis of the differentially expressed target
   genes was performed through Ingenuity with Fisher's exact test. When
   examined individually, each of the 3 miRNA ^INV affected a distinct
   collection of pathways. For example, the top canonical pathways
   affected by miR‐126 were dTMP novo biosynthesis, cardiomyocyte
   differentiation, cardiac‐adrenergic, cAMP‐mediated signaling, and
   ERK/MAPK signaling. The top miR‐487b‐affected pathways were protein
   citrullination, neutral pathway, pyrimidine ribonucleotide
   interconversion, and the Wnt/Ca+ pathway. Although miR‐369‐5p only
   targeted nine genes, the top affected pathways were B‐cell development,
   antigen presentation, and the autoimmune thyroid disease pathway
   (Figure [171]S7a, Supporting Information). When analyzed for the shared
   pathways, all 3 miRNA ^INV regulated cell‐to‐cell signaling and cell
   interaction (Figure [172]S7b, Supporting Information), while organismal
   injury and abnormalities, and nervous system development and function
   were affected by 2 of the 3 miRNA ^INV (Figure [173]S7b, Supporting
   Information). Collectively, the 3 miRNA ^INV modulated ERK/MAPK and
   cAMP‐mediated signaling (in canonical pathways) (Figures [174]S7a,
   [175]S8b–[176]S11b, Supporting Information), cancer, organismal injury
   (in disease) (Figure [177]S7b, Supporting Information), protein
   synthesis, cellular development, cell death and survival, cell‐to‐cell
   signaling, and interaction (in molecular functions) (Figures [178]S7b,
   [179]S8b–[180]S11b, Supporting Information), and embryonic, organ,
   organismal, reproductive system, and tissue development (in
   physiological systems involved in cancer invasion and metastasis)
   (Figures [181]S8b–[182]S11b, Supporting Information).

   To further deduce the protein networks, we used STRING,^[ [183]^43 ^] a
   database tool for predicting protein‐protein interactions directly and
   indirectly for analyzing miRNA ^INV targeted networks. In miR‐126
   targeted genes, 2 major networks were observed centering on DNAJC10
   (DnaJ heatshock protein family member C10)^[ [184]^44 ^] and RAB33B
   (member RAS oncogene family) (Figure [185]S8a, Supporting
   Information).^[ [186]^45 ^] In miR‐487 target genes, MED28 (mediator
   complex subunit 28) was linked to the Mediator complex, a coactivator
   involved in the regulated transcription of nearly all RNA polymerase
   II‐dependent genes (Figure [187]S9a, Supporting Information).^[
   [188]^46 ^] In miR‐369‐5p, a network of potassium voltage‐gated channel
   family surrounding KCNA1 was identified (Figure [189]S10a, Supporting
   Information). For each of the miRNA ^INV target gene groups, there were
   genes that remain isolated. By combining all the target genes of the 3
   miRNA ^INV s, we were able to expand the protein networks to link the
   nodes of DNAJC10, RAB33B, FLT1 (Fms Related Tyrosine Kinase 1)^[
   [190]^47 ^] and EXOC5 (Exocyst Complex Component 5) (Figure [191]S11a,
   Supporting Information).^[ [192]^48 ^] These sets of STRING identified
   novel protein networks critical to GBM invasion.

2.8. KCNA1 is a Common Computational Target Gene of the miRNAs ^INV

   Since all 3 miRNA ^INV actively suppressed pGBM invasion, we examined
   if they shared common target gene(s) by comparing the target genes from
   each of the three miRNA ^INV ^s that were identified (detailed in the
   previous section). KCNA1 was found to be the sole computational target
   gene shared by miR‐369‐5p, miR‐126, and miR‐487b
   (Figure [193]5a–[194]c). KCNA1 is potassium voltage‐gated channel
   subfamily A member 1, known to be involved in diverse physiological
   processes from repolarization of neuronal or cardiac action potentials
   to regulating calcium signaling and cell volume.^[ [195]^49 , [196]^50
   ^] To functionally validate KCNA1 as a molecular target of miRNA ^INV ,
   we examined its expression after the loss‐of‐function and
   gain‐of‐function of the 3 miRNA ^INV . In GBM ^INV cells, which express
   high levels of KCNA1 compared to GBM ^TC cells and normal brain tissue,
   silencing miR‐126, ‐369‐5p, and ‐487b significantly downregulated KCNA1
   (25–45% compared with non‐target control) in both neurosphere and
   monolayer cultures of IC‐1406GBM ^INV and IC‐3752GBM ^INV cells
   (Figure [197] 6a, top panel). Conversely, enhancing miRNA ^INV
   expression in GBM ^TC cells (in which KCNA1 expression was originally
   low) of IC‐1406GBM ^TC and IC‐3752GBM ^TC increased KCNA1 mRNA
   expression in neurospheres by miR‐369‐5p and ‐487b; in monolayer cells
   by miR‐487b and mild elevation by miR‐126 (Figure [198]6a, lower
   panel).

Figure 6.

   Figure 6
   [199]Open in a new tab

   KCNA1 is a molecular target of miRNA ^INV . a) Silencing miRNA ^INV led
   to decreased KCNA1 mRNA expression in purified GBM ^INV cells from
   IC‐1406GBM (1406GBM ^INV ) and IC‐3752GBM (3752GBM ^INV ) grown as
   neurosphere (NS) and monolayer (Mono) (upper panel). Conversely,
   over‐expressing miRNA ^INV in GBM ^TC cells of these two models,
   resulted in elevated KCNA1 mRNA expression in both culture conditions
   (lower panel). KCNA1 expression was quantitated via qRT‐PCR (*P <
   0.05). Data are shown as mean ± SD. b) KCNA1 protein expression was
   suppressed in vivo in the tumor core (GBM ^TC ) and invasive front (GBM
   ^INV ). GBM ^INV cells from IC‐1406GBM (1406GBM ^INV ) were transduced
   with Lentivirus‐miRNA off to silence miR‐126, ‐369‐5p, ‐487b alone
   (miR‐126‐off, miR‐369‐5p‐off, miR‐487‐off) and in combination
   (miR‐mix‐off), followed by implantation into the brains of SCID mice.
   KCNA1 protein expression was examined through IHC. Tumor cells
   receiving no (Control) or lentivirus‐non‐target‐off
   (miR‐Non‐target‐off) were included as references.