Abstract
Simple Summary
Medulloblastoma is the most common malignant pediatric brain tumor. It
can be divided into four molecular subgroups with clear biological and
clinical differences: Group 3, Group 4, SHH, and WNT. The Group 3
subgroup has the lowest overall survival rate, and the WNT subgroup has
the highest. It is known that MYCN and let-7 play a critical role in
medulloblastoma tumorigenesis and progression. By integrating
multi-omics data, including gene expression, methylation, copy number
variation, and miRNA expression, we further divided the SHH subgroup
according to MYCN expression and let-7 activity and found that the
combination of high MYCN expression and high let-7 activity is
associated with worse overall survival.
Abstract
Medulloblastoma (MB) is the most common pediatric embryonal brain
tumor. The current consensus classifies MB into four molecular
subgroups: sonic hedgehog-activated (SHH), wingless-activated (WNT),
Group 3, and Group 4. MYCN and let-7 play a critical role in MB. Thus,
we inferred the activity of miRNAs in MB by using the ActMiR procedure.
SHH-MB has higher MYCN expression than the other subgroups. We showed
that high MYCN expression with high let-7 activity is significantly
associated with worse overall survival, and this association was
validated in an independent MB dataset. Altogether, our results suggest
that let-7 activity and MYCN can further categorize heterogeneous SHH
tumors into more and less-favorable prognostic subtypes, which provide
critical information for personalizing treatment options for SHH-MB.
Comparing the expression differences between the two SHH-MB prognostic
subtypes with compound perturbation profiles, we identified FGFR
inhibitors as one potential treatment option for SHH-MB patients with
the less-favorable prognostic subtype.
Keywords: miRNA, medulloblastoma, prognostic, MYCN, let-7, sonic
hedgehog, ActMiR
1. Introduction
Medulloblastoma (MB) is the most common malignant embryonal brain tumor
in children [[32]1,[33]2]. The 5-year overall survival rate for
patients with medulloblastoma is 65–70% [[34]3]. Treatment for
medulloblastoma usually consists of a combination of surgery,
radiation, and chemotherapy [[35]4]. The risk stratification for
patients is currently based first on age, because patients under 3
years old do not undergo craniospinal radiation therapy, and then on a
combination of tumor size, histology, and metastatic disease
[[36]5,[37]6,[38]7]. The World Health Organization classifies
medulloblastoma, which is very heterogeneous, into four molecular
subgroups with clear biological and clinical differences: sonic
hedgehog-activated (SHH), wingless-activated (WNT), Group 3, and Group
4 [[39]8,[40]9,[41]10,[42]11,[43]12,[44]13,[45]14]. Group 3 has the
worst overall survival rate, and WNT has the best overall survival rate
[[46]3,[47]10]. These subgroups reflect different cells of origin for
the tumor cells [[48]9,[49]15]. Furthermore, integrative genomic
analyses revealed distinct molecular features of the four
medulloblastoma subgroups, such that MB samples can be clustered into
subgroups using features such as transcription, methylation, and copy
number variation (CNV) [[50]10,[51]16,[52]17]. For example, in addition
to each subgroup having unique expression and methylation profiles, WNT
and SHH MB have distinct expression and methylation profiles from Group
3 and Group 4 [[53]8,[54]16,[55]17].
MB is generally treated with surgery followed by chemotherapy and/or
radiation therapy. Given the aforementioned molecular differences among
the subgroups, subgroup-specific targeted therapies have been
investigated, such as SMO inhibitors in SHH-MB, which is known to have
alterations to SUFU, GLI2, and MYCN [[56]18,[57]19]. In addition to
differences among the subgroups, there is also a clear heterogeneity
within the subgroups themselves [[58]16,[59]20], and SHH is the most
heterogenous subgroup in terms of genomic, molecular, and clinical
features [[60]16,[61]21]. Furthermore, SHH-MB tumors with MYCN
amplification are generally resistant to SMO inhibitors [[62]18]. The
TP53 mutation status is a prognostic factor in MB and SHH-MB in
particular [[63]12,[64]22,[65]23,[66]24,[67]25]. TP53 mutations occur
more frequently in the WNT and SHH subgroups (13.8% and 7.6%,
respectively) vs. Group 3 and Group 4 (1.5% and 0%, respectively)
[[68]12]. It is hypothesized that TP53 mutations confer radiation
resistance, so additional therapies are urgently needed to target
high-risk patients in each individual MB subgroup
[[69]12,[70]22,[71]23,[72]24,[73]25].
The MYCN/LIN28/let-7 axis, downstream of SHH, plays a critical role in
SHH-MB [[74]16,[75]17,[76]26,[77]27,[78]28]. A high LIN28B expression
has previously been shown to lead to a poor prognosis in neuroblastoma,
Group 3, and Group 4 MB [[79]29,[80]30]. MYCN is normally highly
expressed in undifferentiated cells and embryonal tumors, which are
thought to be initiated by proliferating progenitor cells unable to
differentiate [[81]31,[82]32,[83]33]. For example, in newborn mice,
high levels of Mycn were found in the brain, and these expression
levels decreased into adulthood [[84]31,[85]34], and human infants are
more likely to have SHH-MB over other MB subgroups [[86]16]. In
addition, the let-7 miRNA family, a group of well-known
tumor-suppressing miRNAs [[87]35,[88]36], is a repressor of MYCN and
can be repressed by LIN28B [[89]37,[90]38]. MicroRNAs (miRNAs)
post-transcriptionally regulate genes and play a critical role in
carcinogenesis and tumor progression [[91]35,[92]39,[93]40,[94]41].
However, the miRNA expression level does not directly reflect miRNA
activity [[95]42,[96]43,[97]44,[98]45], so we developed ActMiR, a
computational tool for inferring the activity of miRNAs in vivo based
on changes in expression levels of target genes [[99]46].
In this study, we investigated let-7 miRNA activity and MYCN
interactions in MB. Since integrative analysis between methylation,
transcription, and CNV can cluster MB into subgroups and because MB
tumors consist of both stromal and tumor cells [[100]47], we
investigated whether the tumor cellular composition can be estimated
based on subgroup-specific genomic features. We further investigated
whether these cellular composition estimates, or together with let-7
miRNA activity, are related to MB prognosis and the drug response in MB
subgroups.
Notably, we identified let-7 miRNA activity as a potential prognostic
biomarker for SHH-MB. We further demonstrated that MYCN expression, in
combination with let-7 activity, stratified patients in SHH-MB into
subtypes associated with different overall survival. SHH-MB patients
with both high let-7 activity and high MYCN expression had
significantly worse survival than other SHH-MB patients. Our results
suggest that the MB patients in each molecular subgroup are still
heterogeneous and that a miRNA-mediated regulatory network can be used
for dissecting heterogeneity and identifying novel subtype-specific
prognostic markers and therapeutic targets.
2. Materials and Methods
2.1. Preprocessing of Gene Expression, miRNA Expression, Methylation Data,
and CNV Data
We downloaded 763 MB mRNA expression profiles ([101]GSE85217) and 763
matching MB methylation profiles ([102]GSE85212) from the super series
[103]GSE85218 [[104]16,[105]48]. We also downloaded 285 MB mRNA
expression profiles ([106]GSE37382) and 1097 MB genotype profiles
([107]GSE37384) from the super series [108]GSE37385, which had some
overlap with [109]GSE85218 the dataset that was not listed in the meta
data [[110]17]. For inferring the activity of miRNAs with ActMiR, we
downloaded the training dataset consisting of 73 mRNA expression
profiles and 64 miRNA pediatric brain cancer expression profiles
([111]GSE42658), including MB; of these data, 57 mRNA and miRNA
profiles originated from the same non-control samples [[112]49]. From
this dataset, we used 14 pilocytic astrocytoma samples, 14 ependymoma
samples, 9 MB samples, 5 glioblastoma samples, 5 atypical
teratoid/rhabdoid tumor samples, 4 choroid plexus papilloma samples, 3
diffuse astrocytoma samples, 2 anaplastic astrocytoma samples, and 1
papillary glioneuronal sample. The 9 MB samples consisted of 5 Group 4
samples, 2 SHH samples, and 2 WNT samples.
All mRNA and miRNA expression data were log2 transformed. Copy number
variation (CNV) was called on for the [113]GSE37384 genotype profiles
using version 1.64.0 of the R package DNACopy’s circular binary
segmentation method [[114]50]. For the [115]GSE85212 methylation
profiles, we extracted DNA methylation values (
[MATH: β :MATH]
values) for each probe. In the case of multiple probes mapping to the
same gene, we performed a Spearman’s correlation with the gene
expression and the probe, with the best p-value selected because
methylation near the promoter regions is associated with gene
repression [[116]51,[117]52,[118]53]. To match the genotype data from
[119]GSE37384 to methylation and gene expression data from
[120]GSE85218, we used the tool MODMatcher to determine which samples
were the most correlated and then used clinical data such as age and
sex to confirm their identities [[121]53]. This resulted in 229 samples
with gene expression, methylation data, and CNV data. The results of
this mapping and the original dataset sample names are presented in
[122]Supplementary Table S1.
2.2. Identifying Cis-Regulatory Genes by Integrating Gene Expression,
Methylation Data, and CNV Data
To determine the cis-regulatory genes of MB, we used 229 samples with
gene expression, methylation data, and CNV data by mapping between
[123]GSE85218 and [124]GSE37384. We performed multiple linear
regression on these 229 mapped samples as follows:
[MATH:
Expg~<
/mo>Methyl<
mi>g+CNVg :MATH]
, where
[MATH:
Expg :MATH]
indicates the expression levels of genes
[MATH: g :MATH]
,
[MATH:
Methylg :MATH]
indicates the DNA methylation level in a gene’s
[MATH: g :MATH]
promoter region, and
[MATH:
CNVg :MATH]
indicates CNVs that contained a gene
[MATH: g :MATH]
in cis form. The DNA methylation level was rank-based inverse normal
transformed. Cis-regulatory genes were determined based on the false
discovery rate (FDR) 5% corresponding to p-values < 1 × 10^−7. To
calculate the FDR rates based on p-values, the multipletests function
in the statsmodels package in Python with the Benjamini and Hochberg
method was used [[125]54]. We defined cis-Methyl genes as genes with a
significantly negative coefficient for
[MATH:
Methylg :MATH]
variable in a multiple linear regression, cis-CNV genes as those with a
significantly positive coefficient for the
[MATH: CNV :MATH]
variable, and cis-CNV/Methyl as the unique subset that met both
criteria. At FDR 5% corresponding to p-values < 1 × 10^−7, we
identified 3630 cis-CNV, 589 cis-methyl, and 107 both cis-CNV and
cis-methyl genes. We used a principal component analysis (PCA) to see
how well these cis-regulatory genes can separate MB subgroups.
2.3. Identifying Subgroups in [126]GSE42658
Since [127]GSE42658 did not list the subgroup annotation of the MB
samples, we performed subgroup classification as follows. Using the
expression in [128]GSE85218 of the cis-Methyl genes identified above
(Methods 2.2), we iteratively took the mean difference of each of the
four subgroups (SHH, WNT, Group 3, and Group 4) against a combined set
of the other three. Next, we performed a Spearman’s correlation of each
[129]GSE42658 sample’s cis-Methyl expression against each of the
subgroup’s one vs. all mean difference. The subgroup with the highest
correlation was annotated as the [130]GSE42658 sample’s subgroup.
2.4. Inferring Tumor Purity of SHH-MB Tumors
The expression of the cis-Methyl genes reflects a different cell of
origin for each MB subgroup. To further refine SHH-MB subgroup-specific
up/downregulated genes, we compared single-cell RNA-seq expression
profiles from 25 MB patients ([131]GSE119926) [[132]55], including 23
diagnostic samples and two recurrences. We downloaded TPM values
(transcript per million reads) for each gene and performed log2
transformation. Among 589 cis-Methyl genes, 247 and 248 were
SHH-MB-specific up/downregulated, with genes expressed higher/lower in
SHH-MB than the other subgroups at p < 0.001, respectively. Within the
247 SHH-MB-specific upregulated genes, 13 genes were expressed
specifically in SHH-MB tumor cells (p < 0.001 and expression in
non-SHH-MB tumor cells < 1.5), referred to as SHH-MB tumor
cell-specific expressed genes (i.e., on genes). Similarly, among 248
SHH-MB specific downregulated genes, 6 genes did not express in SHH-MB
tumor cells specifically (p < 0.001, and expression in SHH-MB tumor
cells < 1.5), referred to as SHH-MB tumor cell-specific non-expressed
genes (i.e., off genes). Given the SHH-MB tumor cell-specific on and
off genes, their expression in a bulk tissue profile is
[MATH:
g=gS
HH_onref∗pur<
/mi>ity+gSHH_offref∗(1−purity :MATH]
), with
[MATH:
gSHH
_onref :MATH]
and
[MATH:
gSHH
_offref :MATH]
as the reference expressions of the SHH-MB tumor cell-specific on and
off genes, respectively.
2.5. Inferring miRNA Activity with ActMiR
We previously developed a tool for inferring miRNA activity, ActMiR
[[133]46,[134]56], based on the expression levels of miRNA-predicted
target genes in a tissue. Using ActMiR, we trained pediatric
brain-specific miRNA activity models using the miRNA expression and
gene expression data in [135]GSE42658 [[136]46,[137]49]. The method is
regression-based concerning both the miRNA expression and gene
expression. In brief, the ActMiR method first determined the baseline
expression levels of each miRNA’s target genes, i.e., a state where the
miRNA did not regulate gene expression. This baseline expression level
was defined as the average expression of samples with low miRNA
expression. Next, we took the differences between expression levels of
the target genes and the baseline expression level for each sample to
determine how degraded the genes’ expressions were by miRNA. Finally,
we performed a linear regression between these degradation values and
the baseline expression; the resulting coefficient from the fit is the
miRNA activity. However, because not all predicted miRNA target genes
are functionally regulated by miRNAs, we determined the miRNA
functional targets by using an iteratively reweighted least squares
regression method between activity and gene expression [[138]36]. High
anticorrelation between miRNA activity and gene expression indicates
that the gene is a functional target of miRNA. We inferred the activity
and determined the functional target genes for the let-7 miRNA family
using the mean miRNA expression of the following let-7 miRNA family
members: hsa-let-7a, hsa-let-7b, hsa-let-7c, hsa-let-7d, hsa-let-7e,
hsa-let-7f, hsa-let-7g, and hsa-let-7i.
Using the functional target genes determined by ActMiR in
[139]GSE42658, we inferred activity in the [140]GSE85218 MB dataset
[[141]16] following the procedure, as previously described [[142]56].
In brief, based on the expression levels of the negatively associated
functional target genes of each miRNA, we calculated the sum of the
scaled expression levels for each sample. We defined the baseline
samples for each miRNA as the samples with the lowest sum of scaled
expression levels of negatively associated functional target genes. We
defined the bottom 5% of the total samples as the baseline samples.
After defining the baseline samples, the procedure to estimate the
miRNA activity is the same as the standard ActMiR procedure described
above. As MB gene expression is heterogeneous by subgroup, we inferred
the activity on each subgroup separately using the classifications
defined in Reference [[143]16].
2.6. Survival Association of miRNA Activity or Gene Expression
We tested for the overall survival associations with miRNA activity. We
used Cox proportional hazards regression on each MB subgroup with
activity. We also separated samples within each subgroup based on
whether their miRNA activity was higher or lower than the linear
regression line between activity and tumor purity. We fit Kaplan–Meier
survival curves and tested the equivalence of the curves using log-rank
tests for high vs. low activity [[144]57]. We performed the same
procedure for the gene expression levels.
2.7. Interaction between Activity and Gene Expression
We investigated the relationship between the let-7 family’s activity
and MYC, MYCN, and LIN28A/B, because they are in the same signaling
pathway [[145]35]. We calculated correlations between the gene
expression and activity. We also separated samples based on
subgroup-specific MYCN expression and whether the samples were greater
or less than the purity-adjusted subgroup-specific expression (linear
regression line between expression and tumor purity). The resulting
four groups were high activity and high expression (MYCN^high-let-7
activity^high), high activity and low expression (MYCN^low-let-7
activity^high), low activity and high expression (MYCN^high-let-7
activity^low), and low activity and low expression (MYCN^low-let-7
activity^low).
2.8. Identification of Functional Target Genes Enriched for Canonical
Pathways
We annotated the function of miRNAs by comparing their functional
target genes with 1329 canonical pathways from MSigDB databases
[[146]58] and identified biological pathways overrepresented in the
functional target gene set of each miRNA using Fisher’s exact test.
2.9. Detecting Small Molecules That Might Be Effective to SHH Subtypes with
Poor Prognosis
First, we determined the differentially expressed genes (DEGs) between
SHH-MB of high let-7 activity and high MYCN expression vs. the rest of
SHH-MB by using t-tests. We determined DEGs at FDR 1%, which
corresponded to p-values < 1.4 × 10^−4. To calculate the FDR based on
the p-value, the multipletests function in the statsmodels package in
Python with the Benjamini and Hochberg method was used [[147]54]. Next,
based on differentially expressed genes, we determined whether these
genes were upregulated or downregulated for tumors with high let-7
activity and high MYCN expression. Then, we investigated whether these
genes were perturbed by drug treatments using the query tool
([148]https://clue.io/query/, accessed on 3 June 2021) from the
Connectivity Map (CMap) of the Library of Integrated Network-Based
Cellular Signatures (LINCS) gene expression resource [[149]59,[150]60].
A negative enrichment score from CMap indicated that the treatment of
the drug showed opposite expression changes for SHH-MB tumors with high
let-7 activity and high MYCN expression compared to other SHH-MB
tumors.
2.10. Validation
To validate our observations, we applied the analyses on 194 MB mRNA
expression profiles generated by the St. Jude group in the following
datasets [[151]61]: [152]GSE10327 [[153]20], [154]GSE12992 [[155]62],
and [156]GSE30074 [[157]63] and from
[158]http://www.stjuderesearch.org/data/medulloblastoma/, accessed on
26 March 2021 [[159]28], which, overall, included 46 SHH-MB samples. We
normalized these mRNA expression profiles to the [160]GSE85218 SHH
samples using ComBat in the sva R package [[161]64]. We inferred the
miRNA activity based on the models trained on the [162]GSE42658
dataset. To assess the survival as sociations between let-7 activity
and MYCN expression, we separated the samples into four groups based on
purity-adjusted expression and let-7 activity (above or below the
linear regression lines between let-7 activity/MYCN expression and
tumor purity in [163]GSE85218) and calculated the survival differences
between different groups.
2.11. Computational Methods
Analysis was carried out using Python version 3.7.9 and R version 3.6.1
using the packages scipy, lifelines, and survival
[[164]65,[165]66,[166]67]. The figures were generated with seaborn and
ggplot2 [[167]68,[168]69,[169]70].
3. Results
3.1. Inter- and Intra-Tumor Heterogenity of MB Accessed by Cis-Methylation
Regulated Genes
MB is categorized into four molecular subgroups according to gene
expression patterns and clinical features
[[170]8,[171]9,[172]10,[173]11]. The molecular subgroups of MB have
been shown to be associated with distinct clinical features
[[174]8,[175]9,[176]10,[177]11,[178]12,[179]13,[180]14]. Here, we
examined a MB cohort [[181]16] with the multi-omics profiling data
available (Methods). The subgroup and age characteristics of the MB
cohort are described in [182]Table 1. First, we used a PCA on the gene
expression, and as expected, it was able to separate the MB samples
into the four molecular subgroups
[[183]8,[184]9,[185]10,[186]11,[187]12,[188]13,[189]14] ([190]Figure
1A). We further explored whether the expression of genes primarily
regulated by methylation (Methods) can better cluster MB into
subgroups, because DNA methylation shows a highly dynamic pattern
during cellular differentiation, indicating its key function related to
cell fate specification [[191]71,[192]72,[193]73], and patterns of DNA
methylation may provide an indirect assessment of these developmental
origins. Indeed, cis-methyl genes (i.e., expression regulated by the
methylation level in its promoter region; see Materials and Methods for
details) can better separate MB into subgroups ([194]Figure 1B) than
all the genes ([195]Figure 1A). In contrast, cis-CNV genes (i.e.,
expression regulated by its DNA copies; see Materials and Methods for
details) did not improve the subgroup separation ([196]Figure 1C,D).
Table 1.
Number of samples by subgroup and age for the studies, as described by
Northcott et al. [[197]16]. The validation datasets had incomplete age
information [[198]20,[199]28,[200]61,[201]62,[202]63].
Dataset Subgroup Subgroup Number Age Number
Training ([203]GSE85218) SHH 223 0–3 62
4–10 55
10–17 29
18+ 69
WNT 70 0–3 1
4–10 23
10–17 27
18+ 13
Group 3 144 0–3 24
4–10 90
10–17 17
18+ 5
Group 4 326 0–3 11
4–10 181
10–17 108
18+ 14
Validation SHH 46 <3 23
≥3 22
WNT 21 <3 0
≥3 21
Group 3 37 <3 9
≥3 27
Group 4 74 <3 3
≥3 71
[204]Open in a new tab
Figure 1.
[205]Figure 1
[206]Open in a new tab
Subgroup distributions based on the PCA of expression profiles of (A)
all 21,050 genes measured within [207]GSE85212, [208]GSE85217, and
[209]GSE37384; and (B) the 589 cis-methyl genes; (C) 3630 cis-CNV
genes; and (D) 107 genes in the overlap of the cis-CNV and cis-Methyl
genes.
This result indicates that the expression of cis-methyl genes can
capture epigenetic fingerprints linked with the subgroups. Indeed,
different molecular subtypes reflect different cells of origin of tumor
cells [[210]15,[211]74], which have unique epigenetic fingerprints
[[212]29]. Furthermore, the expression of these cis-methyl genes that
associate with cell types can be used to estimate the tumor cell purity
in tumor samples. For example, the expression of cis-methyl genes can
clearly separate SHH-MB (black dots in [213]Figure 1B) from other MB
subgroups. By integrating single-cell RNA sequencing (scRNAseq) data of
the MB samples, we identified a set of cis-methyl genes whose
expression was exclusively on or off in SHH-MB tumor cells (PN
progenitor cells, [[214]74]) (see Materials and Methods for details).
As expected, the expression of this set of genes was able to separate
SHH-MB from the other MB subgroups ([215]Supplementary Figure S1).
Then, the SHH-MB tumor cell fraction in tumor samples was estimated
based on this set of genes (see Materials and Methods).
In addition, the MB subgroups have varying amplifications/deletions in
oncogenic genes [[216]16,[217]17]. As consistent with previous studies
[[218]16,[219]17], both MYC and MYCN were cis-CNVs (p = 1.096 × 10^−8
and 5.234 × 10^−11, respectively; Materials and Methods). Furthermore,
we found that MYC and MYCN are differently expressed by the subgroup
([220]Figure 2A,B). The Group 3 and WNT subgroups showed higher MYC
expression levels, while the SHH and WNT subgroups showed higher MYCN
expression levels than the other two subgroups. The higher expression
of MYCN in SHH compared to the other subgroups is consistent with known
MYCN amplifications within the subgroup [[221]16].
Figure 2.
[222]Figure 2
[223]Open in a new tab
Expression profiles for critical genes in the LIN28 pathway. (A–D) The
expression distributions of MYC, MYCN, LIN28A, and LIN28B,
respectively, in [224]GSE85218. LIN28A has a similar expression for all
four subgroups. SHH and WNT have lower expressions than Group 3 and
Group 4 in LIN28B and higher expression in MYCN [[225]16,[226]20]. MYC
has a unique expression profile: SHH and Group 4 have lower
expressions, while WNT and Group 3 have higher expressions. ◊: Diamonds
indicate outliers.
3.2. MYCN/LIN28/Let-7 Axis
MYCN, LIN28, and let-7 form a feed-forward loop ([227]Figure 3), with
MYCN promoting the transcription of LIN28A/B and LIN28A/B inhibiting
the maturation of let-7 microRNAs and then let-7 inhibiting MYCN
post-transcriptionally
[[228]35,[229]37,[230]38,[231]75,[232]76,[233]77]. The MYCN/LIN28/let-7
axis plays a critical role in SHH-MB
[[234]16,[235]17,[236]26,[237]27,[238]28]. The expression of LIN28B had
subgroup variations, while the expression of LIN28A did not
([239]Figure 2C,D). The SHH and WNT subgroups had lower LIN28B
expression levels than Groups 3 and 4 ([240]Figure 2D), opposite of
what would be expected based on the biology of high MYCN leading to
high LIN28A/B expression ([241]Figure 3), suggesting other molecular
mechanisms regulating the feed-forward loop.
Figure 3.
[242]Figure 3
[243]Open in a new tab
LIN28 signaling pathway. MYC, MYCN, and IL-6 promote transcription for
LIN28A/B. let-7 miRNAs post-transcriptionally repress the expression of
MYC, MYCN, and IL-6.
Next, we examined the overall survival associations of the expression
of LIN28A, LIN28B, MYC, and MYCN using Cox proportional hazards
regression. The expression of MYC, MYCN, and LIN28B was associated with
the overall survival, as expected when examining all the MB samples
together ([244]Table 2), as their expression is also associated with
the subgroups ([245]Figure 2). When examining their associations with
survival in each individual subgroup, the MYCN expression level was not
associated with survival in SHH-MB ([246]Table 2), even though MYCN was
highly expressed in SHH-MB ([247]Figure 2). On the other hand, the
LIN28B expression level was low in SHH-MB ([248]Figure 2), suggesting
that other genes may play an important role in the MYCN/LIN28/let-7
pathways, such as let-7 miRNAs. Therefore, the let-7 miRNA family was
further examined for their impact on the survival of SHH-MB.
Table 2.
Associations of the overall survival (OS) and genes in the LIN28
pathway. Using all MB samples together, LIN28B, MYC, and MYCN were
significantly associated with survival. For SHH-MB, only LIN28B was
significantly associated with the OS. In WNT-MB, none of the four genes
was significant. In Group 3, MYC and MYCN were significant. In Group 4,
MYC’s association was significant.
Subgroup Gene Hazard Ratio Wald p Log-Rank p
Training All Samples MYCN 0.86 1.9 × 10^−2 0.019
MYC 1.1 4.9 × 10^−3 4.6 × 10^−3
LIN28A 1.0 0.98 0.98
LIN28B 1.4 1.1 × 10^−6 8.0 × 10^−7
Training SHH MYCN 1.3 0.25 0.26
MYC 0.97 0.91 0.91
LIN28A 1.1 0.79 0.79
LIN28B 1.5 0.01 9.2 × 10^−3
Training WNT MYCN 2.8 0.45 0.43
MYC 0.40 0.26 0.27
LIN28A 1.8 0.70 0.70
LIN28B 0.82 0.92 0.92
Training Group 3 MYCN 0.74 0.028 0.029
MYC 1.2 0.020 0.019
LIN28A 0.71 0.29 0.29
LIN28B 1.3 0.11 0.11
Training Group 4 MYCN 1.1 0.49 0.49
MYC 1.3 1.01 × 10^−3 8.4 × 10^−4
LIN28A 0.96 0.83 0.83
LIN28B 0.97 0.87 0.87
[249]Open in a new tab
3.3. MB Has Higher Let-7 Activity Than Other Brain Tissues and
Subgroup-Specific Activity
Let-7 miRNAs are tumor suppressors [[250]37,[251]38,[252]78]
post-transcriptionally repressing MYCN expression, as well as feedback
repressing LIN28 expression in the MYCN/LIN28/let-7 pathway
[[253]35,[254]39,[255]40]. Therefore, we explored the prognostic effect
of let-7 activity in MB. There was no MB dataset with samples profiled
for both miRNA and mRNA expression, so we examined miRNA profiles of 57
pediatric brain tumors ([256]GSE42658) [[257]49] that resembled MB.
Within this dataset, the let-7 miRNAs had higher expression than the
other miRNAs ([258]Figure 4A), and the let-7 expression in MB was not
significantly different from its expression level in the other tissues
(MB vs. other tissues, t-test p = 0.65, [259]Figure 4B).
Figure 4.
[260]Figure 4
[261]Open in a new tab
Distributions of the miRNA expression in [262]GSE42658. (A) A histogram
of the mean expression levels of all the miRNAs. Let-7 family miRNAs
are indicated as red vertical lines. (B) Boxplots of the mean let-7
family expression across tissue types. The expression of let-7 members
in MB (red) is not significantly different from its expression in other
tissues or tumor types. (C) The inferred let-7 family activity across
tissues. The Let-7 activity in MB (red) was significantly higher than
in the other tissues in a one vs. others comparison (t-test, p =
0.011). A higher activity of let-7 in MB with similar expression levels
to other tissues indicates that the let-7 family more frequently
regulates the expression of its target genes in MB compared to other
tissues [[263]46,[264]56]. ◊: Diamonds indicate outliers.
Several studies have demonstrated that miRNA functional activity was
not accurately reflected by its expression level
[[265]42,[266]43,[267]44], so we inferred the activity of let-7 using
the previously described ActMiR method [[268]46,[269]56]. All let-7
miRNAs were highly expressed in pediatric brain tumors ([270]Figure
4A), so we inferred let-7 activity using the mean expression of all
let-7 miRNAs rather than individual miRNAs. Although let-7 miRNAs were
not significantly differentially expressed among the different tissues
([271]Figure 4B), let-7 activity in MB was significantly higher than
the activity in most other tissues in a one vs. others comparison
(t-test, p = 0.011), except the activity in choroid plexus papilloma
([272]Figure 4C), indicating the potential regulatory role of let-7
miRNAs in MB.
The functional target genes of the let-7 miRNAs, estimated by
integrating the miRNA expression and gene expression data of
[273]GSE42658, were used to infer the subgroup-specific let-7 activity
for the 763 MB samples in [274]GSE85218 using the ActMiR method
[[275]46] (see Materials and Methods for details). The subgroups SHH
and Group 4 had 136 and 145 let-7 functional targets, respectively,
much larger than the 33 and 38 functional targets in WNT and Group 3,
respectively ([276]Figure 5A). The SHH and Group 4 subgroups were
enriched for a MYCN amplification [[277]16], suggesting a potential
relationship between MYCN and let-7 miRNA activity.
Figure 5.
[278]Figure 5
[279]Open in a new tab
Inferred activity in the let-7 miRNA family and correlation against
MYCN for samples in [280]GSE85218. (A) The bar plot of the number of
inferred functional targets for the let-7 family. p-values indicating
enrichment for functional targets by the hypergeometric test. SHH and
Group 4 have the most functional targets of the let-7 family. (B–E)
Correlations of the let-7 activity and MYCN expression in the subgroups
SHH, WNT, Group 3, and Group 4, respectively. In SHH-MB, let-7 activity
and MYCN expression had a significant Spearman’s correlation, with p =
1.66 × 10^−8. The correlation p-values were 0.411, 0.117, and 0.319 in
WNT, Group 3, and Group 4, respectively.
To annotate the function of miRNAs, we assessed the biological pathways
over-represented in the functional target genes of the let-7 miRNAs
(see Materials and Methods for details) [[281]58] at 5% FDR
corresponding to p-values < 1 × 10^−3. The functional targets of let-7
in SHH-MB were uniquely significantly enriched for the Reactome G1
Phase pathway ([282]Supplementary Figure S2).
Next, because let-7 miRNAs suppress the MYCN expression
post-transcriptionally [[283]35] ([284]Figure 3), we examined the
correlation between let-7 activity and MYCN expression. In SHH-MB,
let-7 activity had a significant negative correlation with MYCN
expression (p = 1.66 × 10^−8; [285]Figure 5B), while let-7 activity was
not significantly correlated with MYCN expression in the other
subgroups ([286]Figure 5C–E). This observation also indicates a
potential regulatory role of let-7 miRNAs in the MYCN/LIN28/let-7
pathways in SHH-MB.
3.4. Stratifying SHH-MB by MYCN Expression and Let-7 Activity
The SHH-MB subgroup is the most heterogeneous among the MB subgroups
[[287]16,[288]55]. As MYCN is highly expressed in SHH-MB ([289]Figure
2) and let-7 miRNA is potentially active ([290]Figure 5) in SHH-MB, we
investigated the potential of MYCN expression and/or let-7 activity as
a prognostic biomarker ([291]Table 3). MYCN is amplified in some SHH-MB
tumor cells [[292]16,[293]55], such that MYCN expression was
significantly correlated with SHH-MB tumor purity, as expected
(Spearman’s correlation coefficient = 0.3265, p = 6.193 × 10^−7;
[294]Supplementary Figure S3A). On the other hand, let-7 activity was
significantly anticorrelated with SHH-MB tumor purity (Spearman’s
correlation coefficient = −0.3891, p = 1.788 × 10^−9;
[295]Supplementary Figure S3B). Note that the LIN28A/B expression
levels were very low ([296]Figure 2 and [297]Supplementary Figure S4A)
and did not correlate with the tumor purity (Spearman’s correlation
coefficient r = 0.02 and −0.01, p = 0.7 and 0.85 for LIN28A and LIN28B,
respectively).
Table 3.
Associations of the overall survival rate and genes in the LIN28
pathway with let-7 activity based on the SHH-MB samples in
[298]GSE85218. LIN28A/B KM curves for SHH-MB are shown in
[299]Supplementary Figure S11A,B.
Test Variable Hazard Ratio Wald p Log-Rank p
Surv~MYCN MYCN 1.3 0.25 0.26
Surv~LIN28A LIN28A 1.1 0.79 0.79
Surv~LIN28B LIN28B 1.5 0.011 0.0092
Surv~let-7 let-7 169.5 0.32 0.32
Surv~purity purity 3.3 0.11 0.11
Surv~age age 0.99 0.55 0.55
Surv~MYCN + let-7 MYCN 1.7 0.066 0.069
let-7 2.9 × 10^5
Surv~purity + MYCN + let-7 purity 1.4 × 10^−3 0.025 0.024
MYCN 1.7
let-7 7.7 × 10^3
[300]Open in a new tab
Next, the SHH-MB samples were partitioned into high/low MYCN expression
groups based on the purity-adjusted mean MYCN expression level
([301]Supplementary Figure S3A) or into high/low let-7 activity groups
based on the purity-adjusted mean let-7 activity ([302]Supplementary
Figure S3B). Neither MYCN or let-7 were significantly associated with
the survival rate (MYCN: log-rank test p = 0.182, [303]Figure 6A and
let-7: log-rank test p = 0.362, [304]Figure 6B). When combining MYCN
and let-7 activity information together, the SHH-MB patients with both
high MYCN expression and high let-7 activity (MYCN^high-let-7
activity^high) had a much worse overall survival rate than the other
three groups (MYCN^high-let-7 activity^low, MYCN^low-let-7
activity^high, and MYCN^low-let-7 activity^low) (log-rank test p =
0.0028, [305]Figure 6C).
Figure 6.
[306]Figure 6
[307]Open in a new tab
Kaplan–Meier overall survival curves for SHH-MB based on the MYCN
expression, let-7 family activity, and MYCN + let-7 family activity for
the samples in [308]GSE85218. High activity/expression (blue) was
defined as the group of samples with activity/expression greater than
or equal to the linear regression line between the MYCN/let-7 family
and purity, and low activity/expression (orange) was defined as the
group of samples with activity/expression less than the linear
regression line. (A) The Kaplan–Meier survival curves for SHH with high
vs. low MYCN expression (log-rank test p = 0.182), (B) the Kaplan–Meier
survival curves for SHH with let-7 activity (p = 0.362), and (C) the
Kaplan–Meier survival curves for let-7 activity and MYCN expression.
MYCN^high-let-7 activity^high showed a significantly worse outcome
(log-rank tests with p = 0.0028) than the combined group (Others) of
MYCN^high-let-7 activity^low, MYCN^low-let-7 activity^high, and
MYCN^low-let-7 activity^low. The Others group in 7C is represented as
individual groups in [309]Supplementary Figure S12.
3.5. Validation in an Independent MB Cohort
To validate our observations, we combined 194 MB mRNA expression
profiles generated by the St. Jude’s Children’s Research Hospital group
[[310]61], including [311]GSE10327 [[312]20], [313]GSE12992 [[314]62],
[315]GSE30074 [[316]63], and from
[317]http://www.stjuderesearch.org/data/medulloblastoma/, accessed on
26 March 2021 [[318]28] (see Materials and Methods), referred to as the
validation cohort collectively. The characteristics of the validation
cohort are summarized in [319]Table 1. Like the [320]GSE85218 dataset,
the MYCN expression in SHH-MB was significantly higher than in the
other groups (p = 1.420 × 10^−9, [321]Supplementary Figure S5B). The
same analyses above were applied to the validation cohort (the results
are listed in [322]Table 4).
Table 4.
Associations of the survival rate and genes in the LIN28 pathway based
on samples in the validation dataset.
Subgroup Gene Hazard Ratio Wald p Log-Rank p
Validation all MYCN 1.3 0.019 0.019
MYC 1.0 0.58 0.58
LIN28A 1.7 0.041 0.044
Validation SHH MYCN 1.4 0.14 0.14
MYC 0.77 0.49 0.49
LIN28A 5.3 0.074 3.5 × 10^−5
Validation WNT MYCN 5.2 0.15 0.076
MYC 0.20 0.23 0.17
LIN28A 65.8 0.27 0.25
Validation Group3 MYCN 0.90 0.82 0.82
MYC 1.2 0.30 0.29
LIN28A 2.0 0.28 0.28
Validation Group4 MYCN 1.7 1.1 × 10^−3 4.3 × 10^−4
MYC 1.1 0.64 0.64
LIN28A 0.66 0.39 0.39
[323]Open in a new tab
For SHH-MB in the validation cohort, the inferred let-7 activity and
MYCN expression were significantly anticorrelated (Spearman’s
correlation r = −0.393, p = 0.00689) ([324]Supplementary Figure S6A),
consistent with the observations in the [325]GSE85218 data. Their
associations with survival are listed in [326]Table 5. Similarly, the
inferred SHH-MB tumor purity was correlated with MYCN expression (r =
0.2961, p = 0.04572) and anticorrelated with let-7 activity (r =
−0.3261, p = 0.0270). Then, the SHH-MB samples were partitioned into
high/low MYCN expression or high/low let-7 activity groups based on the
purity-adjusted mean values ([327]Supplementary Figure S3C,D). MYCN
expression or let-7 activity alone were not significantly associated
with survival (p-values = 0.27 and 0.63, respectively; [328]Figure
7A,B). When combining MYCN expression and let-7 activity information
together, the SHH-MB patients with both high MYCN expression and high
let-7 activity had a significantly worse survival rate than
MYCN^high-let-7 activity^low, MYCN^low-let-7 activity^high, and
MYCN^low-let-7 activity^low (log-rank test p-value = 0.037; [329]Figure
7C). All these findings were consistent with our observations in the
[330]GSE85218 dataset.
Table 5.
Associations of the survival rate and genes in the LIN28 pathway, let-7
activity, and their combined effects for the SHH-MB samples in the
validation dataset. LIN28A KM curves for SHH-MB are shown in
[331]Supplementary Figure S11C.
Test Variable Hazard Ratio Wald p Log-Rank p
Surv~MYCN MYCN 1.4 0.14 0.14
Surv~LIN28A LIN28A 5.3 0.074 3.5 × 10^−5
Surv~LIN28B LIN28B N/A N/A N/A
Surv~let-7 let-7 11.4 0.55 0.55
Surv~purity purity 0.20 0.81 0.81
Surv~age age <3 1.2 0.79 0.79
age ≥3 N/A
Surv~MYCN + let-7 MYCN 1.8 0.039 0.025
let-7 7.7 × 10^3
Surv~purity + MYCN + let-7 purity 0.98 0.090 0.057
MYCN 1.8
let-7 7.7 × 10^3
[332]Open in a new tab
Figure 7.
[333]Figure 7
[334]Open in a new tab
Associations of the survival rate and MYCN expression and let-7
activity based on SHH-MB samples in the validation datasets. High
activity/expression (blue) was defined as the group of samples with
activity/expression greater than or equal to the linear regression line
between MYCN/let-7 family and purity, and low activity/expression
(orange) was defined as the group of samples with activity/expression
under the linear regression line. (A) The Kaplan–Meier overall survival
curves for MYCN expression (log-rank test p = 0.267), (B) the
Kaplan–Meier overall survival curves for let-7 family activity
(log-rank test p = 0.627), and (C) the Kaplan–Meier survival curves for
let-7 activity with MYCN expression. MYCN^high-let-7 activity^high
(blue) showed a significantly worse outcome (log-rank tests p = 0.037)
than MYCN^high-let-7 activity^low, MYCN^low-let-7 activity^high, and
MYCN^low-let-7 activity^low.
3.6. Drug Repurposing to Identify Potential Therapeutic Treatment for SHH
Subgroups
MYCN-amplified SHH-MB is generally resistant to SMO inhibitors, a
targeted therapy for SHH-MB [[335]18,[336]19,[337]79]. It has been
suggested that DFMO would have a greater effect on MYCN-amplified
tumors [[338]80]. ODC inhibition by DFMO decreases LIN28 and increases
let-7 [[339]35,[340]76,[341]81], so the low expression of LIN28A/B in
SHH-MB and the high let-7 activity suggest that DFMO is unlikely to be
effective for SHH-MB patients with high MYCN expression and high let-7
activity who have a less-favorable survival rate. Thus, there is an
urgent need to develop therapeutics for SHH-MB patients, especially for
patients with high let-7 activity and high MYCN expression.
To understand the molecular mechanisms driving the survival differences
([342]Figure 6), we compared SHH-MB with both high MYCN expression and
high let-7 activity against the other SHH-MB samples and identified 172
differentially expressed genes at FDR < 1%, including 125 upregulated
and 47 downregulated genes in the MYCN^high-let-7 activity^high SHH-MB
group. Using these genes, we performed a pathway enrichment analysis
and found that the upregulated genes were enriched for KEGG neuroactive
ligand receptor interaction (p = 3.448 × 10^−5) and other
ligand–receptor pathways, which may indicate cell-to-cell interactions
as potential drug targets ([343]Supplementary Figure S7).
Next, we compared these differentially expressed genes with drug
signatures in the Connectivity Map (CMap) of the Library of Integrated
Network-Based Cellular Signatures (LINCS) gene expression resource
[[344]59,[345]60] using the query tool ([346]https://clue.io/query/,
accessed on 3 June 2021). We identified 21 drug candidates whose
effects may reverse the DEG signature (negative scores lower than −90,
[347]Table 6). Some drugs with the same mechanisms of actions showed
consistent enrichment scores < −90. For example, FGFR inhibitors,
including orantinib, PD-173074, and brivanib, were in the candidate
drug list ([348]Table 6). Previous studies have shown that FGFR
promotes MB tumor cell invasion in vitro [[349]82], the blockade of
FGFR represses brain tissue infiltration in vivo [[350]82], and FGFR
inhibitor decreases the viability and proliferation in MB cell lines
[[351]83]. These results suggest that the FGFR signaling pathway may be
a potential target for treating SHH-MB patients with both high MYCN
expression and high let-7 activity.
Table 6.
CMap results for the 21 drugs with negative CMap connectivity scores.
The target genes of these drugs are shown as a full table in
[352]Supplementary Table S3. ^1 CMap connectivity score. A score of 95
indicates that only 5% of the reference gene sets showed stronger
connectivity than the current query to the perturbagen in question. ^2
Number of drugs that have the same mechanism of action and of which
scores are smaller than −90. The number in parentheses indicates the
total number of drugs with the same mechanisms of action. ^3 Enrichment
statistics of drugs with the same mechanisms of action. Only shown
scores are smaller than −90.
Score ^1 Name Description Number of Drugs ^2 Enrichment Score ^3
−99.26 cobalt(II)-chloride HSP inducer 1 (3) -
−98.8 amonafide Topoisomerase inhibitor 1 (16) -
−98.64 embelin HCV inhibitor 1 (3) -
−97.22 hyperforin Cyclooxygenase inhibitor 1 (57) -
−97.02 parthenolide NF-kB pathway inhibitor 1 (12) -
−95.75 dapsone Bacterial antifolate 1 (3) -
−95.6 brivanib FGFR inhibitor 3 (3) −97.12
−94.79 ketoconazole Sterol demethylase inhibitor 1 (6) -
−94.68 piperacillin Bacterial cell wall synthesis inhibitor 1 (29) -
−94.61 ziprasidone Dopamine receptor antagonist 1 (65) -
−92.54 tienilic-acid Sodium/potassium/chloride transporter inhibitor 1
(4) -
−92.07 sitagliptin Dipeptidyl peptidase inhibitor 1 (3) -
−91.88 orantinib FGFR inhibitor 3 (3) −97.12
−91.59 FCCP Mitochondrial oxidative phosphorylation uncoupler 1 (2) -
−91.52 XMD-885 Leucine rich repeat kinase inhibitor 1 (3) -
−91.3 geldanamycin HSP inhibitor 1 (13) -
−90.67 harpagoside Acetylcholinesterase inhibitor 1 (17) -
−90.59 tyrphostin-AG-1478 EGFR inhibitor 1 (42) -
−90.43 AG-370 PDGFR receptor inhibitor 1 (9) -
−90.06 PD-173074 FGFR inhibitor 3 (3) −97.12
−90.05 kinetin-riboside Apoptosis stimulant 1 (10) -
[353]Open in a new tab
4. Discussion
Our integrative analysis of gene expression, methylation, and CNV
identified cis-methylation genes that better separated MB patients into
molecular subgroups than all genes, indicating subgroups are probably
linked to an epigenetic memory of development lineages
[[354]15,[355]29,[356]55,[357]74]. By leveraging inferred miRNA
activity, we identified a subset of SHH-MB patients with a worse
survival rate than other SHH-MB patients ([358]Figure 6C), and the
association was validated in an independent SHH-MB cohort ([359]Figure
7C). The MYCN^high-let-7 activity^high SHH-MB subset did not overlap
with the SHH-MB subtypes based on a similarity network fusion (SNF)
analysis [[360]16]. SHH-α and SHH-β, subtypes of SHH-MB by SNF, were
enriched for MYCN amplification and were associated with worse survival
outcomes [[361]16] (five-year survival 69.8% and 67.3% for SHH-α and
SHH-β, respectively vs. 88–88.5% for other SHH-MBs). The
MYCN^high-let-7 activity^high SHH-MB patient group consisted of 32
samples, where 10 of the samples fell into the SNF subtypes SHH-α or
SHH-β (five and five samples, respectively) (Fisher’s exact test p =
0.05 and p = 0.6).
High MYCN expression being associated with inferior survival in SHH-MB
is expected, as MYCN is an oncogene [[362]16]. However, let-7, a known
tumor suppressor, is downregulated in multiple tumor types
[[363]37,[364]78,[365]84]. It is counterintuitive that MYCN^high-let-7
activity^high SHH-MB patients had a worse overall survival rate than
the other SHH-MB patients, especially MYCN^high-let-7 activity^low
SHH-MB patients. It has been shown that high levels of MYCN protein
persist in MYCN-amplified neuroblastoma after transfecting high levels
of let-7 miRNA and that MYCN, which is the most abundant target for
let-7 miRNA, actually acts as a sponge for let-7 miRNAs in these cells
[[366]85]. Our observation that MYCN^high-let-7 activity^high SHH-MB
was associated with worse survival might be due to the MYCN sponge
effect on let-7 miRNAs dominating the let-7-repressing effect on MYCN
transcripts. Indeed, MYCN expression and let-7 activity were not
correlated in MYCN^high-let-7 activity^high SHH-MB, while they were
significantly anticorrelated in other SHH-MB ([367]Supplementary Figure
S8).
High LIN28 expression is known to be significantly correlated with a
shorter survival time in MB [[368]86]. LIN28B regulates multiple
oncogenic processes, in part by downregulating the let-7 miRNA family
in MB, and LIN28B expression is associated with survival in Group 3 and
Group 4 MB [[369]29]. However, whether LIN28/let-7 play an important
role in SHH-MB is not previously known, indicating that our finding
that let-7 plays a potential critical regulatory role in SHH-MB is
novel. Beyond validation in independent datasets ([370]Figure 7),
experimental validations, such as RT-qPCR after let-7 knockdown or
overexpression in SHH-MB cells, are needed to confirm the regulatory
relationships between let-7 and its target genes.
About 8% of SHH-MB patients carry TP53 mutations based on a
whole-genome sequencing study [[371]12]. A separate study showed that
SHH-MB with TP53 mutations have worse survival than SHH-MB without TP53
mutations [[372]23]. The functional impact of TP53 mutations is
heterogeneous, resulting in a spectrum of functional changes from loss
of function to gain of function [[373]87]. In our training dataset
([374]GSE85218), the TP53 mutation status of five SHH-MB tumor samples
was characterized by whole-genome sequencing [[375]12,[376]88]. One of
the five SHH-MB samples carried a TP53 mutation. To assess the
relationship between the TP53 mutation status and potential p53 protein
function, we estimated the p53 pathway activity for each SHH-MB sample
in the training dataset based on ssGSEA [[377]89,[378]90]. The SHH-MB
sample with TP53 mutations was within the lowest quartile of the p53
pathway activity ([379]Supplementary Figure S9A); however, SHH-MB
patients with p53 pathway activity lower than the activity in this
TP53-mutated patient had better survival than the other SHH-MB patients
([380]Supplementary Figure S9B). These results together indicate a high
complexity in including the TP53 mutation status or p53 functional
status in prognostic biomarker models.
The overall survival rates of the four MB subgroups were not
significantly different ([381]Supplementary Figure S10), and SHH-MB was
the most heterogeneous subgroup among the four MB subgroups [[382]12].
SMO inhibition is a targeted therapy for SHH-MB, but SHH-MB patients
with high MYCN expression were resistant to SMO inhibitors in general
[[383]18,[384]19,[385]79]. Furthermore, p53 mutations are associated
with MYCN amplification and are implicated in conferring resistance to
both radiation therapy and SMO inhibitors in SHH-MB
[[386]22,[387]23,[388]24,[389]25]. A subset of SHH-MB patients with
high MYCN expression (MYCN^high-let-7 activity^high SHH-MB) had a
significantly worse survival rate than the other SHH-MB ([390]Figure 6
and [391]Figure 7). Thus, it is urgent to identify effective treatments
for these MYCN^high-let-7 activity^high SHH-MB patients. DFMO, an
inhibitor of LIN28A/B, has been identified as a potential treatment
option for embryonal tumors [[392]80]. It is worth noting that the
LIN28A/B expression levels were low in SHH-MB ([393]Figure 2) and
similar to other tissues ([394]Supplementary Figure S13), so it is
unlikely that DFMO is effective in SHH-MB. Our analysis suggests that
FGFR inhibitors are potential drug candidates for MYCN^high-let-7
activity^high SHH-MB ([395]Table 6). We note that the effects of these
predicted candidates have not been validated in SHH-MB cancers. Further
in vitro and in vivo experiments are needed to validate and strengthen
our findings and demonstrate their therapeutic values. Our results
suggest that the MB patients in each molecular subtype are still
heterogeneous, and integrated genomic analyses can be used for
dissecting their heterogeneity and identifying novel subtype-specific
prognostic biomarkers and therapeutic targets.
5. Conclusions
We applied an integrated genomic analysis on MB datasets and inferred
the let-7 miRNA activity in MB. We identified a SHH-MB subset with high
MYCN expression and high let-7 activity associated with a worse
survival rate than the other SHH-MB and validated the association in an
independent MB cohort.
Acknowledgments