A set of microRNAs coordinately controls

tumorigenesis, invasion, and metastasis

Iacovos P. Michaela,b, Sadegh Saghafiniaa,b,c, and Douglas Hanahana,b,1
a
Swiss Institute for Experimental Cancer Research, School of Life Sciences, Swiss Federal Institute of Technology Lausanne, 1015 Lausanne, Switzerland;
b
Swiss Cancer Center Leman, 1015 Lausanne, Switzerland; and cDepartment of Computational Biology, University of Lausanne, 1015 Lausanne, Switzerland

Contributed by Douglas Hanahan, October 9, 2019 (sent for review August 5, 2019; reviewed by Eduard Batlle and Andrea Ventura)

MicroRNA-mediated gene regulation has been implicated in nucleotides 2 through 7) and a target gene’s complementary sites
various diseases, including cancer. This study examined the role (miRNA-response elements [MREs], usually located in the 3′
of microRNAs (miRNAs) during tumorigenesis and malignant pro- UTR) is mediated by simple Watson–Crick pairing; therefore,
gression of pancreatic neuroendocrine tumors (PanNETs) in a the specificity of the miRNA is determined by the length of the
genetically engineered mouse model. Previously, a set of miRNAs seed region and the accessibility of the MRE. As a result,
was observed to be specifically up-regulated in a highly invasive miRNAs are predicted to be able to bind to many genes, which
and metastatic subtype of mouse and human PanNET. Using potentially allows them to have pleiotropic effects, consequen-
functional assays, we now implicate different miRNAs in distinct tially posing challenges to determining functionally relevant gene
phenotypes: miR-137 stimulates tumor growth and local invasion, targets (12–14).
whereas the miR-23b cluster enables metastasis. An algorithm, MicroRNAs have been implicated in various aspects of cancer
Bio-miRTa, has been developed to facilitate the identification of progression, such as proliferation, apoptosis, and the epithelial–
biologically relevant miRNA target genes and applied to these mesenchymal transition (EMT), and they can either promote or
miRNAs. We show that a top-ranked miR-137 candidate gene, suppress tumorigenesis and metastasis (15–21). Previously we
Sorl1, has a tumor suppressor function in primary PanNETs. reported that distinct miRNAs are differentially expressed dur-
Among the top targets for the miR-23b cluster, Acvr1c/ALK7 has
ing the progression of tumorigenesis in the RIP1-Tag2 (RT2)
recently been described to be a metastasis suppressor, and we
mouse model of pancreatic neuroendocrine tumors (PanNETs)
establish herein that it is down-regulated by the miR-23b cluster,
(9). A subset of these PanNETs, designated as “met-like pri-
which is crucial for its prometastatic activity. Two other miR-23b
mary” tumors (MLPs), are characterized by miRNAs and mRNA
targets, Robo2 and P2ry1, also have demonstrable antimetastatic
transcriptome signatures in common with liver metastases, pos-
effects. Finally, we have used the Bio-miRTa algorithm in reverse
to identify candidate miRNAs that might regulate activin B, the
ing the intriguing hypothesis that PanNET cancer cells acquire the
principal ligand for ALK7, identifying thereby a third family of
necessary capabilities for metastasis during primary tumorigenesis
miRNAs—miRNA-130/301—that is congruently up-regulated con- (9, 22). Our cross-species analysis revealed that 6 miRNAs—miR-
comitant with down-regulation of activin B during tumorigenesis, 132, miR-137, miR-181, miR-23b, miR-27b, and miR-24-1—were
suggestive of functional involvement in evasion of the proapoptotic
barrier. Thus, dynamic up-regulation of miRNAs during multistep Significance
tumorigenesis and malignant progression serves to down-
regulate distinctive suppressor mechanisms of tumor growth, The capabilities for invasion and metastasis underlie the mor-
invasion, and metastasis. tality and morbidity of most forms of human cancer. Currently,
there are no effective therapies specifically targeting these
cancer | metastasis | microRNAs | Acvr1c/ALK7 | PanNETs cancer phenotypes, in part due to the paucity of dominant
mutations that induce them, and indeed losses of suppressors
of invasion and metastasis are increasingly recognized as de-
M any cancer patients will succumb to their disease due to
metastasis, underscoring the imperative to develop anti-
metastatic therapies (1). During this complex process, cancer
terminants, posing challenges for drug development. Our results
implicate epigenetic gene regulation mediated by elevated ex-
cells invade both into adjacent tissue and into blood vessels, from pression of distinct microRNAs in orchestrating invasion and
which they can subsequently extravasate into other organs, and metastasis, evidently by abrogating distinctive suppressor
eventually begin proliferating to form macrometastases (2–4). A mechanisms. Therefore, targeting such microRNAs holds promise
majority of cancer cells in a tumor typically fail to complete this as a strategy to combat malignant cancers with epigenetically
disrupted tumor suppressor mechanisms.
process, emphasizing the significant barriers encountered and
the capabilities necessarily acquired to circumvent them. While Author contributions: I.P.M. and D.H. designed research; I.P.M. and S.S. performed re-
various genetic aberrations of oncogenes and tumor suppressor search; I.P.M. and S.S. contributed new reagents/analytic tools; I.P.M. and S.S. analyzed
genes have been associated with tumorigenesis, there is currently data; and I.P.M. and D.H. wrote the paper.
a dearth of recurrent driver gene mutations specifically impli- Reviewers: E.B., Institut de Recerca en Biomedicina Barcelona; and A.V., Memorial Sloan
cated in metastasis (5). Notably, recent studies have shown that Kettering Cancer Center.
genetic polymorphisms and epigenetic regulators, including chro- The authors declare no competing interest.
matin modifiers and miRNAs, can play important roles during the This open access article is distributed under Creative Commons Attribution-NonCommercial-
“metastatic cascade” (6–9). NoDerivatives License 4.0 (CC BY-NC-ND).

MicroRNAs are small noncoding RNAs ∼22 nt in length that Data deposition: The RNA sequencing data reported in this paper have been deposited in
the Gene Expression Omnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo
are able to posttranscriptionally repress the expression of (accession no. GSE131887). The Bio-miRTa R package source code is deposited at
mRNAs, and therefore regulate diverse signaling pathways and Hanahan Lab website and accessible via https://www.epfl.ch/labs/hanahan-lab/data-and-
biological processes (10, 11). After their biogenesis, the mature tools/.
strand of a miRNA, along with an Argonaute protein, form the 1
To whom correspondence may be addressed. Email: [email protected].
miRNA-induced silencing complex (miRISC), which is respon- This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.
sible for the translational silencing and mRNA decay of its target 1073/pnas.1913307116/-/DCSupplemental.
genes. The pairing between the miRNA seed region (miRNA First published November 8, 2019.

24184–24195 | PNAS | November 26, 2019 | vol. 116 | no. 48 www.pnas.org/cgi/doi/10.1073/pnas.1913307116

similarly up-regulated in human and mouse metastasis-associated whether the enhanced tumor growth was consequent to the initial
MLP tumors (22). In the current study, we have investigated these seeding of the cancer cells or subsequent tumor growth, miRNA
and other miRNAs and developed an algorithm to identify and expression was induced 1 wk after injection of the cells into the
associate candidate target genes with the miRNAs that might pancreas, by which point solid tumors had become established.
regulate them. As a result, we revealed “tumor suppressors” that Once again, up-regulation of miR-137, but not of the miR-23b
are down-modulated by distinct miRNAs and collectively impli- cluster, reduced survival, reflecting enhanced tumor growth (Fig.
cated in circumventing barriers to invasion and metastasis. 1B), substantiating its ability to promote primary tumorigenesis.
Results MicroRNA-137 Elicits an Invasive Capability. We next immuno-
MicroRNA-137 Enhances Primary Tumor Progression. The MLP- stained a series of tissue sections from similarly sized orthotopic
associated and metastasis-specific miRNA signatures contains 33 transplant tumors, collected from end-point mice, for the driving
miRNAs—16 up-regulated and 17 down-regulated—compared to SV40 T antigen oncoprotein so as to reveal the cancer cells, and
noninvasive insulinoma tumors (ITs) (9), of which a subset is observed that up-regulation of miR-137 increased the proportion
similarly altered both in mouse and human MLP PanNETs (SI of invasive carcinoma (IC; putatively the MLP subtype) versus
Appendix, Fig. S1 A and B) (22). In order to prioritize their in- noninvasive IT tumors, when compared to the control group
vestigation, we further examined expression of each in both sub- (Fig. 1 C–E). In contrast, expression of the miR-23b cluster had
types of primary tumor—IT and MLP—and in liver metastases in no impact: the proportions of IT vs. IC/MLP were the same as
the RT2 PanNET model. The results led us to focus on miR-137 with the control group (Fig. 1 D and E). A flow-stimulated in-
and the miR-23b cluster, due to their distinctive patterns of ex- vasion assay (23) further established the capability of miR-137 to
pression in invasive MLP tumors and liver metastases (SI Ap- enhance invasiveness of βTC3 cells (Fig. 1F). We also evaluated
pendix, Fig. S1 C and D). Notably, the paralog cluster targeting the a second cell line, STC-1, derived from an intestinal carcinoid
same MREs—the miR-23a cluster—is also up-regulated in mouse tumor that developed in the same RT2 mouse model (24), which
PanNET metastases (SI Appendix, Fig. S1E). was similarly engineered to overexpress these miRNAs (SI Ap-
To probe potential functional effects, miR-137 and the miR- pendix, Fig. S1 I and J); miR-137 (but not the miR-23b cluster)
23b cluster were overexpressed in the βTC3 IT cell line, using a again enhanced flow-stimulated invasiveness (Fig. 1G).
piggyBac transposon system enabling doxycycline (DOX)-inducible These functional analyses collectively reveal that miR-137 is

GENETICS
miRNA expression that includes the fluorescent protein mKate able to enhance progression to invasive carcinomas, in particular
along with luciferase (SI Appendix, Fig. S1 F–H). We first exam- by stimulating invasiveness, whereas the miR-23b cluster did not
ined the effect of miRNA overexpression in experimental primary have a discernible effect, consistent with its modest up-regulation
tumors arising in an orthotopic transplant model of PanNET, in- in MLP tumors compared to miR-137.
volving the injection of βTC3 cells into the pancreas of immuno-
compromised mice. Initially, we induced expression of miRNAs MicroRNA-23b Cluster Enhances Liver Metastasis. The intriguing
in vitro by adding DOX to the cell culture media and then kept observation that the miR-23b cluster was markedly up-regulated in
administering DOX in vivo via DOX-containing chow. Expression metastasis compared to MLP primary tumors, and its lack of effect
of miR-137, but not the miR-23b cluster, accelerated tumor on the latter, led us to assess its effects on metastasis. Since
growth, which led to shorter overall survival (Fig. 1A). To examine analysis of metastasis in mice bearing βTC3 cell-derived orthotopic

Fig. 1. MicroRNA-137 reduces survival and increases invasion of PanNET. (A and B) Survival of mice with orthotopic PanNETs elicited by inoculation of control
βTC3 cells, or miRNA overexpressing cells. MicroRNA expression was induced either concomitantly (A) or 7 d after injection of the cells into the pancreas
parenchyma (B). n = 3 to 7. Log-rank test. *P < 0.05, **P < 0.01. (C) Quantification of tumor volume at the endpoint of the survival studies. (D) Quantification
of the distribution between IT and IC tumors. n = 28 to 37. Fisher’s exact test. ns, not significant. ***P < 0.001. (E) H&E staining and immunostaining for the
SV40 T antigen (Tag) oncoprotein to reveal cancer cells in tumor tissue sections. (F and G) Effect of miRNA expression on the invasive capabilities of βTC3 (F)
and STC-1 (G) in a flow stimulated invasion assay. n = 3 to 6. Student’s t test. ***P < 0.001.

Michael et al. PNAS | November 26, 2019 | vol. 116 | no. 48 | 24185
tumors is not feasible due to severe hypoglycemia consequent to MicroRNA-23b Cluster Has a Prosurvival Effect during the Initial
excessive insulin secretion by the cancer cells, we used experimental Stages of Metastasis. To further investigate the prometastatic
metastasis models. effect of the miR-23b cluster, we varied the timing of DOX-
Up-regulation of the miR-23b cluster proved to selectively mediated induction of miR-23b cluster expression in βTC3 cells
enhance liver metastasis following i.v. inoculation of cancer cells, (Fig. 3A). In the first 2 conditions, DOX was added to the media
whereas overexpression of miR-137 had a very modest effect, as during cell culture before tail vein injection, and we either con-
scored by the increased number of metastatic foci (Fig. 2A), as tinued using DOX-containing chow following inoculation into
well as by higher levels of luciferase (Fig. 2B). Ex vivo fluorescent mice (ON/ON), or not, instead using regular chow (ON/OFF). In
imaging of the livers immediately after dissection established that the remaining 2 conditions, DOX was not added to the media,
the metastatic foci were positive for the mKate reporter expressed and it was either supplied immediately after tail vein injec-
along with luciferase in the transposon vector (Fig. 2C), further tion (OFF/ON) or not supplied at all (OFF/OFF). Whereas the
confirming their identity as transduced cancer cells, as did ON/OFF condition was prometastatic, phenocopying the ON/ON
immunostaining for the SV40 T antigen oncoprotein (Fig. 2D). condition, the OFF/ON was not (Fig. 3 B and C). Quantification
To further evaluate the prometastatic effect of the miR-23b of bioluminescence in the liver over time further established that
cluster, we similarly evaluated miRNA-transduced versions of
the miR-23b cluster did not affect the expansion phase of me-
the aforementioned STC-1 cell line. Up-regulation of the miR-
tastasis (Fig. 3D and SI Appendix, Fig. S2 A and B), signifying
23b cluster significantly enhanced the metastatic ability of the
that the miR-23b cluster was important during the initial stages
STC-1 cells (Fig. 2 E–H). In addition, we assessed the endoge-
nous levels of miR-23b cluster expression in a series of cancer of colonization.
cell lines derived from PanNET tumors arising in the RT2 model To dissect the prometastatic mechanism, we examined pro-
that displayed a spectrum of metastatic potential (25) and found liferation and apoptosis of control and miR-23b cluster-
that miR-23b and miR-27b were expressed at higher levels in the overexpressing βTC3 cells in the liver 24 h after intrahepatic
more highly metastatic cell lines (SI Appendix, Fig. S1 K and L). injection, which efficiently seeds cells into the liver, facilitating
Notably, both the miR-23b cluster and the miR-23a cluster are characterization of their fate during the initial stages of coloni-
similarly up-regulated in high-grade human PanNET tumors zation. We observed that miR-23b cluster overexpression led to
(“NF-MLP,” i.e., “nonfunctional,” namely with low-insulin ex- decreased apoptosis and increased proliferation compared to
pression) (SI Appendix, Fig. S1 M and N). control cells (Fig. 3 E and F and SI Appendix, Fig. S2 B and C),
Collectively, the results reveal that up-regulation of the miR-23b whereas there was no difference in the total number of cancer
cluster selectively enhances the metastatic capability of PanNET cells lodged in the liver (SI Appendix, Fig. S2C). Thus, we conclude
and intestinal carcinoids without significantly affecting tumor that the miR-23b cluster facilitates cancer cell growth following
growth and invasion. initial metastatic seeding of the liver.

Fig. 2. Expression of the miR-23b cluster enhances liver metastasis. (A–G) PanNET experimental liver metastasis assays via tail vein injections using control
βTC3 cells or miRNA overexpressing cells. (E–H) Intestinal carcinoid liver metastasis assays via tail vein injections using control STC-1 cells, or miRNA over-
expressing cells. (A and E) Quantification of metastatic foci in the liver derived from PanNETs (A; n = 5) or intestinal carcinoids (E; n = 7 to 11). (A) One-way
ANOVA followed by Holm–Sidak’s multiple comparisons test. ns, not significant. *P < 0.05, **P < 0.01. (E) Student’s t test. ***P < 0.001. (B and F) Bio-
luminescence imaging before killing (radiance; p/sec/cm2/sr). (C and G) mKate fluorescence imaging of excised livers. (D and H) H&E staining and Tag
oncoprotein immunostaining of liver tissue sections.

24186 | www.pnas.org/cgi/doi/10.1073/pnas.1913307116 Michael et al.

 GENETICS
Fig. 3. Characterization of the prometastatic effects of the miR-23b cluster. (A–D) PanNET experimental liver metastasis assays via tail vein injections using
miR-23b cluster overexpressing βTC3 cells with distinctive regimens of DOX-induced expression. (A) DOX administration schedules. (B) Survival of mice variably
induced to express the miR-23b cluster. n = 5. Log-rank test. ***P < 0.001. (C) Bioluminescence imaging of representative mice at day 30 for the 4 induction
schedules (radiance; p/sec/cm2/sr). (D) Time course of the bioluminescence signal in the liver during the expansion phase. (E and F) PanNET experimental liver
metastasis assays via intrahepatic injection using control βTC3 cells or miR-23b cluster overexpressing cells. Quantification of apoptosis (E; Tag+/Cleaved
Caspase 3) and proliferation (F; Tag+/EdU) of βTC3 control and miR-23b cluster overexpressing cells, 24 h after intrahepatic injection. n = 3 to 8. Student’s
t test. *P < 0.05, ***P < 0.001.

Development of Bio-miRTa: An Algorithm for Identification of Notably, the first step of Bio-miRTa is generic to the miRNA,
Biologically Relevant miRNA Targets. MicroRNAs are known to and does not factor in the biological context; therefore, it could
play context-dependent roles, and as such their functionally im- be used independently of step 2 to provide a ranked list of all
portant gene targets can in principle vary in distinct pathological possible targets in principle of miRNA(s) based on their pre-
conditions. Therefore, we developed an algorithm, Bio-miRTa, diction scores (Fig. 4 and Materials and Methods).
for the identification of biologically relevant miRNA targets (Fig. In the second step, the algorithm factors in the biological
4), which enables the ranking of the miRNA candidate target context, by incorporating various datasets provided by the user,
genes according to the biological process of interest. As outlined which can be grouped into 2 general categories: 1) experimental
below, Bio-miRTa is structured into 2 steps: In the first step, all datasets from functional miRNA perturbations (e.g., miRNA
potential targets of the miRNA(s) of interest are extracted, and in gain of function [GOF] and miRNA loss of function [LOF]), and
the second, the list is filtered by experimental or descriptive 2) descriptive datasets comparing relevant states or conditions in
datasets that are related to the biological process of interest, in experimental mouse models and human transcriptomic data
order to score and rank potential target genes accordingly. where the candidate miRNA is differentially expressed, and
In the first step of Bio-miRTa, the algorithm extracts all po- hence its targets could also be differentially expressed. For the
tential targets of the miRNA(s), regardless of the biological assigned biological score (BS) in the aforementioned categories
context. During this step, it uses 4 independent target prediction of datasets, we rationalized that the BS of the experimental
algorithms [TargetScan (26), miRanda (27), DIANA (28), and datasets (i.e., BSED) should be greater than the BS of the de-
probability of interaction by target accessibility [PITA] (29)] in scriptive datasets (i.e., BSDD), since in the latter case there could
which the target genes are primarily predicted based on the be other complex modulations of target genes, and as such, al-
presence of an MRE in the target gene, and other algorithm- though relevant, descriptive datasets might provide weaker evi-
specific parameters. Since the number of false positive pre- dence compared to the functional perturbations of a miRNA(s).
diction is high in computational algorithms (30, 31), Bio-miRTa Therefore, we assigned BSED = 2 to the first experimental
also incorporates 2 datasets of experimentally validated miRNA dataset, and BSDD = 1 to the descriptive dataset. In the case of
targets [TarBase (32) and starBase (33)] into the prediction additional datasets, the BS score is increased by 1 for each ad-
pipeline. Each gene predicted as a target of a miRNA by any of ditional dataset in either category, and then a normalized BS is
the 6 datasets is assigned a prediction score (PS; with values of applied to all datasets in the corresponding category. Finally,
0.15 for each prediction algorithm and 0.2 for each experi- Bio-miRTa calculates the final score (FS) by summing the PS
mentally validated dataset). The final PS for a gene is calcu- from the first step and the BS from the second step, which leads
lated by the sum of PS for each dataset (i.e., PSmax = 1; Fig. 4). to the final ranking of the candidate target genes (Fig. 4).
Bio-miRTa reports a gene as a potential target of a miRNA if it To validate the algorithm, we used the transcriptome data
is presented in at least 2 prediction algorithms, or 1 prediction from a recent study in which the miR-200 family was inactivated
algorithm and 1 experimentally validated dataset (i.e., PS ≥ 0.3). in the RT2 PanNET model (15). Notably, the gene encoding for

Michael et al. PNAS | November 26, 2019 | vol. 116 | no. 48 | 24187
Fig. 4. Flowchart of the Bio-miRTa algorithm. In the first step the algorithm extracts the miRNA gene targets using information from target prediction
algorithms (TargetScan, miRanda, DIANA, and PITA) and databases with experimentally validated targets (TarBase and starBase). A PS is assigned to the
potential target genes that appear in the respected database, and those with a cumulative PS < 0.3 are filtered out. In the second step, the algorithm in-
corporates biologically relevant information provided by the user and assigns a BS accordingly. Finally, it scores and ranks the target genes according to the
FS. The details of and rationale for each step are described in the Results and Materials and Methods sections.

the transcription factor Zeb1 was predicted by Bio-miRTa as the For the second step, we applied 1 functional perturbation
top target of the miR-200 family (SI Appendix, Fig. S3 and dataset for each miRNA, and 2 descriptive datasets. The func-
Dataset S1). Zeb1 was shown to be the main gene target in the tional miRNA perturbation datasets were composed of signifi-
aforementioned study (15), and moreover, previous studies also cantly down-regulated genes upon miR-137 overexpression (GOF)
implicated the role of the miR-200 family in the regulation of Zeb1 for the miR-137 inquiry, and significantly down-regulated genes
and epithelial-to-mesenchymal transition (34). Other known targets upon miR-23b cluster overexpression (GOF) for the miR-23b
of the miR-200 family, such as Dlc1, Reck, and Ogt, were also cluster inquiry (35). In each analysis the BSED was therefore
identified by the algorithm, as well as a potential gene target, Mbnl1, equal to 2 (Fig. 4). The 2 descriptive datasets contained down-
regulated genes in mouse and human MLP tumors and meta-
which plays an important role in pre-mRNA alternative splicing
static lesions compared to the lower grade insulinoma/islet tu-
(SI Appendix, Fig. S3). mors (as aforementioned, MLP/metastasis have higher miR-137/
Using Bio-miRTa to Identify Candidate Target Genes for miR-137 and
miR-23b cluster expression compared to insulinoma/islet tumors).
We incorporated these 2 descriptive datasets into both miR-137
the miR-23b Cluster. To assess its utility in identifying relevant
and miR-23b cluster target predictions, assigning a BS = 1 for each
target genes, the first step of Bio-miRTa (Fig. 4) was applied to dataset, resulting in a BSDD = 2, and a maximum combined BS
miR-137 and the miR-23b cluster, whose functional importance of 4 (Fig. 4).
had been established above (Figs. 1 and 2). In the case of the In step 1, we identified 2,340 and 7,563 candidate target genes
miR-23b cluster, which is composed of 3 miRNAs, in the first with PS ≥ 0.3 for miR-137 and the miR-23b cluster, respectively.
step of Bio-miRTa, the algorithm separately subscored each of the Notably, this number was substantially reduced by the applica-
3 miRNAs that targeted a candidate target gene, and then com- tion of step 2 and the combined scoring of both steps. Particu-
bined the scores in order to calculate the corresponding PS for the larly, after filtering in only those candidate target genes that
cluster (SI Appendix, SI Materials and Methods and Fig. S4). appear in the GOF dataset and at least 1 of the 2 descriptive

24188 | www.pnas.org/cgi/doi/10.1073/pnas.1913307116 Michael et al.

 GENETICS
Fig. 5. Identification of miRNA gene targets using Bio-miRTa. (A and B) Highly ranked candidate target genes for miR-137 (A) and the miR-23b cluster (B)
identified using the Bio-miRTa algorithm. The first column labeled as “gene” shows the candidate target genes sorted according to the FS shown in the last
column on the Right. The columns with the headers “target prediction algorithms” and “experimentally validated targets” report the presence of an MRE for
the corresponding miRNA in the candidate target gene. The columns with the header “differential gene expression” show the differential gene expression in
βTC3 cells after miRNA overexpression in βTC3 cells (miRNA GOF), in mouse IT versus MLP tumors in RT2 mice (mouse PanNETs) and in the human IT versus MLP
PanNETs (human PanNETs). The intercalated columns show the PS (based on the presence of MRE according to the first step of the algorithm), BS (based on
the differential gene expression), and FS (the sum of the PS and BS scores). The PS scored in the case of the miR-23b cluster is normalized as shown in SI
Appendix, Fig. S4. The top 5 genes of each miRNA are shown; Datasets S2 and S3 contain the complete list of ranked genes.

datasets (FS > 3), there were only 20 and 27 candidate target involving perturbations in miRNA expression, and that Bio-
genes for miR-137 and the miR-23b cluster, respectively (Fig. 5 miRTa can be reversed, so as to identify candidate miRNAs via
A and B and Datasets S2 and S3). the MREs present in a particular gene of interest.
Initially, we focused on candidate target genes of miR-137. Then
we turned to investigate the top-ranked miR-23b candidate genes MicroRNA-137 Candidate Target Genes Are Associated with Poor
identified by this algorithm, in particular the Tgfβ superfamily Prognosis in Human Cancer Types. To begin assessing the associa-
receptor Acvr1c (Activin receptor type-1C; commonly known as tion of miRNA-137 and its target genes with human cancers, we
ALK7), which we have previously reported to be a metastasis first audited available public transcriptomic datasets of human
suppressor (25). breast cancer in The Cancer Genome Atlas (TCGA) and found a
clear association of high miR-137 expression with comparatively
Using Bio-miRTa in Reverse to Identify Candidate miRNAs Targeting worse overall survival (OS) in breast cancer (Fig. 6A). Given this
the ALK7 Ligand Activin B. We previously reported (25) that ex- association, we further queried available transcriptomic data-
pression of the principal ALK7 ligand, activin B, encoded by bases for a possible association of putative miR-137 target genes
the Inhßb gene, is down-regulated during the early stages of with breast cancer prognosis. The analysis revealed that com-
PanNET tumorigenesis; thus we hypothesized that it too might be paratively low expression of a miR-137 gene signature, composed
repressed by miRNAs. Although we lacked a functional pertur- of the 5 top-ranked miR-137 candidate genes (Fig. 5A), was
bation dataset, Bio-miRTa identified a family of 4 similar miRNAs— strongly associated with poorer relapse-free survival (RFS) of
miR-130a/b and miR-301a/b—predicted to bind a conserved MRE breast cancer patients (Fig. 6B), as was the individual expression
within the 3′ UTR of human and mouse Inhßb; concordantly, of each gene (Fig. 6C and SI Appendix, Fig. S6 A–D). Thus, in
these miRNAs proved to be up-regulated during the sequential agreement with a previous report (37), it is plausible that miR-
stages of tumorigenesis in the RT2 PanNET model (SI Ap- 137 enhances invasive tumor growth in human breast cancer, and
pendix, Fig. S5 A–D and Dataset S4). Using luciferase reporter thereby contributes to progression of the disease. In addition to
assays, we verified the ability of these miRNAs to bind to the breast cancer, we assessed the association of miRNA-137 with 20
identified MRE within the 3′ UTR of Inhßb mRNA and thereby additional cancer types using TCGA data and found that its
repress expression of the reporter (SI Appendix, Fig. S5 E and F). expression was associated with poorer OS of patients with head
Furthermore, the biochemical interaction of miR-301a with Inhβb and neck, ovarian, and uterine corpus endometrial cancer, while
mRNA has been previously established using the HITS-CLIP it was associated with better OS of patients with kidney renal cell
assay (36). As such, we infer that activin B expression may also and liver hepatocellular carcinoma (SI Appendix, Fig. S7). Future
be regulated by dynamic expression of miRNAs during multistep studies will be required to delineate the evidently context-
PanNET tumorigenesis, motivating future functional studies. dependent roles of miRNA-137 in these different cancer types.
Notably, this analysis demonstrates that the algorithm can be Among the candidate miR-137 target genes (Fig. 5A), we
informative in the absence of a functional (GOF/LOF) dataset began by assessing Sorl1 (also known as SORLA and LR11),

Michael et al. PNAS | November 26, 2019 | vol. 116 | no. 48 | 24189
Fig. 6. Association of miR-137 and its top-ranked candidate target genes with breast cancer prognosis, and characterization of the Sorl1 target gene in
primary PanNET tumors. (A) Association of miR-137 expression with overall survival in a cohort of breast cancer patients including all subtypes. (B and C)
Association of an miR-137 gene signature (comprising the top 5 predicted target genes) and 1 of its components, SORL1, with RFS in human breast cancer. (D)
The MRE for miR-137 in the 3′ UTR of SORL1. The miRNA seed, the MREs for human and mouse SORL1, as well as the mutated MRE used for the reporter
assays, are shown. (E) Luciferase reporter assays. n = 6. Student’s t test. ***P < 0.001. (F–H) Orthotopic PanNET tumor assays using control, miR-137 over-
expressing, and Sorl1 knockdown βTC3 cells. The miR-137 overexpression and Sorl1 knockdown were induced by DOX food concomitant with injection of the
engineered cells into the pancreas parenchyma, and the mice were killed 30 d after injection. (F) Tumor volumes at the 30-d defined endpoint. n = 6 to 7.
One-way ANOVA followed by Holm–Sidak’s multiple comparisons test. *P < 0.05, **P < 0.01. (G) The distribution between IT and IC tumors in the 3 cohorts.
n = 14 to 21. Fisher’s exact test. ns, not significant. *P < 0.05, **P < 0.01. (H) H&E staining and immunostaining for the SV40 Tag oncoprotein to reveal cancer
cells in tumor tissue sections.

since we found that its expression was comparatively lower in the To assess the possible role of Sorl1 in tumor phenotypes, a
aggressive triple-negative breast cancer (TNBC; basal) subtype, DOX-inducible piggyBac transposon system was used to knock
and that it was associated with shorter time to the first progression down Sorl1 in the βTC3 cells (SI Appendix, Fig. S6 H and I). The
of patients with non-small cell lung and gastric cancers (SI Ap- engineered cells were inoculated (orthotopically) into the pan-
pendix, Fig. S6 E–G). Sorl1 is a member of the “vacuolar protein creas of immunocompromised mice, and the mice were killed
sorting 10 protein (VPS10P) domain” receptor family re- 30 d later. We observed that the Sorl1 knockdown increased the
sponsible for the trafficking and sorting of proteins between the tumor burden by ∼2-fold, which is in accordance with the ob-
Golgi apparatus, intracellular endosomes, and the cell surface. served tumor-promoting function of miR-137 (Fig. 6F). Tumor
The mouse and human 3′ UTR of Sorl1 contain 1 conserved sections were immunostained for the driving SV40 T antigen
oncoprotein and assessed for the proportion of invasive (IC)
MRE for miR-137 (Fig. 6D). To assess the ability of miR-137 to
versus noninvasive insulinoma-like (IT) tumors. Interestingly,
bind to this MRE, we synthesized a fragment of ∼1 kb surrounding the Sorl1 knockdown had only a minor effect on the invasive
the wild-type MRE, as well as a fragment with a mutated MRE phenotype (Fig. 6 G and H), suggesting either that another miR-
(Fig. 6D). We then performed luciferase reporter assays, which 137 target gene(s) is (are) primarily responsible for its invasion-
revealed that the wild-type MRE decreased luciferase levels by 2- inducing capability, or that Sorl1 down-regulation acts in a syn-
fold compared to the mutated one (Fig. 6E), indicating that miR- ergistic/additive manner with another target gene.
137 can bind to the Sorl1 3′ UTR MRE and mediate miRNA- This pilot study encourages future investigation of the mech-
mediated mRNA degradation. The regulation of Sorl1 by miR-137 anisms and more precise effects of Sorl1 in PanNET and other tu-
is also supported by published immunoprecipitation-based bio- mor types, in particular breast cancer, using genetically engineered
chemical assays (38). mouse models. Similar lines of investigation can be envisaged to

24190 | www.pnas.org/cgi/doi/10.1073/pnas.1913307116 Michael et al.

validate and characterize other target genes for miR-137 as a and miR-27b to bind to their respective MREs in ALK7, we
suppressor(s) of invasive growth; among the candidates are the synthesized fragments of ∼1 kb surrounding each MRE and
other top-ranked target genes, Kcnmb2, Hpse, Asph, and Robo2, mutationally inactivated the MRE of interest in each fragment
the latter notably also implicated as a miR-23b target (Fig. 5B). (Fig. 7A). Luciferase reporter assays revealed that the presence
of a wild-type MRE decreased the luciferase levels around 2-fold
ALK7 Expression Is Down-Regulated by the miR-23 Cluster. Three in each case, compared to constructs with the mutationally dis-
genes stood out as candidate targets of the miR-23b cluster: rupted MRE (Fig. 7B), indicating that miR-23b and miR-27b are
ALK7, P2ry1, and Robo2 (Fig. 5B). We have recently reported able to bind to their respective MREs and mediate miRNA-
that suppression of ALK7 expression and signaling functionally mediated mRNA degradation. The ability of miR-23 and miR-
enhances tumorigenesis and metastasis in PanNETs and breast 27 to regulate the expression of ALK7 is further supported by
cancer (25). Using a transposon gene knockdown system (SI Ap- published datasets (40–42) cataloging immunoprecipitation-
pendix, Fig. S6H), we have now shown that knock down of either based biochemical assays of miR-to-MRE interactions, which
P2ry1 or Robo2 can also enhance liver metastasis (SI Appendix, described the binding of both miRNAs to MREs in the 3′ UTR
Fig. S8 A–D). of ALK7 (SI Appendix, Fig. S9B).
Given the extensive functional validation of ALK7 as a me- As noted above, the “IT-like” βTC3 cell line expresses very
tastasis suppressor (25), we chose to focus on characterizing the low levels of the miR-23b cluster, whereas the “MLP-like” cell
potential regulation of ALK7 by the miR-23b cluster. ALK7 is a line AJ-5257-1 expresses higher levels (SI Appendix, Fig. S1 K
type I receptor, which upon binding to a ligand, most notably and L), similarly to cancer cells in MLP tumors and metastases.
activin B, forms a heterotetramer with a type II receptor (Acvr2a Transient transfection of AJ-5257-1 cells with antagomirs to
or Acvr2b), resulting in intracellular signaling via phosphoryla- miR-23/27 increased ALK7 expression (SI Appendix, Fig. S9C).
tion of Smad2/3 (SI Appendix, Fig. S9A) (39). Both the mouse Conversely, DOX-induced up-regulation of the miR-23b cluster
and the human 3′ UTRs of ALK7 contain 2 MREs for miR-23b in βTC3 cells led to decreased ALK7 expression (Fig. 7C and SI
and 1 MRE for miR-27b (Fig. 7A), while for miR-24, there is 1 Appendix, Fig. S9D). Furthermore, up-regulation of miR-23b
MRE only in the mouse gene. To assess the ability of miR-23b suppressed phosphorylation of the ALK7 downstream target

GENETICS

Fig. 7. ALK7 expression is regulated by miR-23/7b miRNAs during PanNET tumorigenesis. (A) MREs for miR-23b and miR-27b in the 3′ UTR of ALK7. The
miRNA seed, the MREs for human and mouse ALK7, as well as the mutated MREs used for the reporter assays, are shown. (B) Luciferase reporter assays for a 3′
UTR reporter gene. n = 6. Student’s t test. ***P < 0.001. (C) Expression of ALK7 in cultured βTC3 cells carrying a tet-regulatable miR-23b cluster transgene
following DOX administration to induce miRNA expression and then its withdrawal. (D) Western blotting for phosphorylated Smad2 and total Smad2 in miR-
23b cluster-expressing βTC3 cells and control cells, treated as indicated with activin B for 8 h. (E–H) Experimental metastasis assay using βTC3 cells expressing
the miR-23b cluster with and without coexpression of the miR23b-insensitive ALK7 ORF, relative to control cells. (E) Quantification of metastatic foci. n = 3 to
8. One-way ANOVA followed by Holm–Sidak’s multiple comparisons test. ns, not significant. ***P < 0.001. (F) Bioluminescence imaging of mice of luciferase-
expressing cancer cells before killing (radiance; p/sec/cm2/sr). (G) mKate fluorescence imaging of cancer cells in excised livers. (H) H&E staining and Tag
oncoprotein immunostaining of liver sections. (I and J) Representative images and scoring of ALK7 expression (%; total tumor area containing ALK7-positive
cells) in liver metastases of control and miR-23b cluster overexpressing βTC3 cells. m, metastasis. n = 12 to 17. One-way ANOVA followed by Holm–Sidak’s
multiple comparisons test. ***P < 0.001. Normal pancreatic islets, which express much higher levels (25), are shown for comparison.

Michael et al. PNAS | November 26, 2019 | vol. 116 | no. 48 | 24191
Smad2 upon treatment of βTC3 cells with activin B, which is the mediated induction of apoptosis (25). Btg2 has been reported to
principal high-affinity ALK7 ligand (Fig. 7D). Finally, to further be a target of miR-27b (43), and indeed we found that Btg2 was
corroborate the ability of the miR-23b cluster to regulate ALK7, down-regulated upon miR-23b cluster expression in βTC3 cells
based on our previous study (25), we developed 2 transcriptomic (SI Appendix, Fig. S9H).
ALK7 signaling pathway signatures: the first includes all genes Future studies are warranted to assess the potentially dis-
altered by ALK7 activation in βTC3 cells, whereas the second tinctive functional effects and potential collaborative activity of
selectively filters only those genes known to be associated with ALK7 in combination with the 2 other miR-23b cluster targets,
apoptosis (Dataset S5). Using these mRNA signatures, we ob- P2ry1 and Robo2, implicated above, as well as others in the
served that forced expression of the miR-23b cluster in βTC3 signature (SI Appendix, Fig. S1).
cells repressed activation of the ALK7 signaling pathway com-
pared to control βTC3 cells when both were treated with activin Discussion
B so as to induce ALK7 signaling, as indicated by the lower Herein, we investigated miRNAs dynamically regulated during
scores for both gene sets (SI Appendix, Fig. S9 E and F). Notably, malignant progression in pancreatic neuroendocrine tumors, and
this repressive effect was comparable to that produced by ex- thereby have implicated specific miRNAs as modulators of dis-
pression of a dominant-negative form of ALK7 (ALK7-DN; tinctive neoplastic phenotypes: miR-137 enhanced primary tu-
containing a K222R “kinase-dead” mutation) or by ALK7 knock- mor growth and invasion, whereas the miR-23b cluster contributed
down in βTC3 cells also treated with activin B (SI Appendix, Fig. to the successful seeding and colonization of metastatic cancer
S9 E and F); these engineered constructs have been previously cells in the liver (Fig. 8). MicroRNAs regulate the expression of
described (25). multiple genes, which makes it challenging to dissect and un-
We next used functional assays involving βTC3 cells engi- derstand their roles in complex diseases such as cancer (13, 14,
neered with a transposon vector conditionally expressing the 26). While various computational algorithms have been de-
miR-23b cluster, with or without the ALK7 open reading frame veloped to predict the targets of miRNAs based on the sequence
(ORF) that lacks the 3′ UTR (and hence the miR-23/27 MREs) of the miRNA seed and other parameters (26, 28, 29, 44), it has
and is therefore insensitive to the miR-23b cluster (SI Appendix, proved challenging to identify and prioritize gene targets of
Fig. S9G). Notably, the capability of the miR-23b cluster to en- miRNAs that might be important for a particular disease state or
hance liver metastasis, documented above (Fig. 2), was dimin- biological condition. In this study, we leveraged considerable
ished by coexpression of the miR-23b cluster-insensitive ALK7 knowledge and insights acquired over the years studying the RT2
ORF (Fig. 7 E–H). Using immunostaining, we observed that mouse model of PanNETs (9, 22, 45), in concert with the newly
ALK7 expression was lower in the metastases overexpressing the developed computational algorithm, Bio-miRTa, to rank candi-
miR-23b cluster (Fig. 7 I and J), further establishing its ability to date targets for particular miRNAs. This strategy led us to
down-regulate the expression of ALK7. Together, these results candidate genes that are down-regulated by miR-137 or by the
establish that the members of the miR-23 cluster—miR-23a/b miR-23b cluster in the invasive and highly metastatic mouse and
and miR-27a/b—collectively suppress the expression of ALK7. human MLP PanNETs, a tumor subtype in which these miRNAs
Given the ability of miRNAs to modulate expression of multiple are up-regulated. Importantly, by applying Bio-miRTa we were
genes, we further queried Bio-miRTa for known ALK7 effector able to refine the long list of MRE-containing genes more than
genes containing MREs for the miR-23b cluster, and identified 100-fold, producing a shortlist of candidate genes that could in
the BH3-only proapoptotic protein Btg2, which we had pre- principle be screened using unbiased medium throughput approaches,
viously implicated as one of the proteins involved in the ALK7- such as CRISPR/Cas9. Notably, the algorithm, Bio-miRTa, can

Fig. 8. Schematic diagram illustrating the proposed roles of the implicated miRNAs and their respective gene targets during PanNET tumorigenesis
and metastasis.

24192 | www.pnas.org/cgi/doi/10.1073/pnas.1913307116 Michael et al.

easily be customized for the identification and prioritization of many cancers (58, 59). Notably, Robo2 was also 1 of the top 5
candidate gene targets for miRNAs involved in the progression miR-137 candidate target genes, and it might also be involved in
of other cancer types as well as other diseases, developmental the invasive phenotype. P2ry1 is a G protein coupled receptor of
stages, and physiological states, from which short-listed top- extracellular ADP whose possible association with cancer is largely
ranked candidates could then be further evaluated using func- unknown (60); our data provide a clue that it too may serve as a
tional assays. barrier to tumor progression. The roles of the Slit/Robo axon
In previous studies, miR-137 has been reported to have either guidance molecules as well as of P2ry1 during cancer progression
tumor-promoting or tumor-suppressor capabilities (46). In agree- and metastasis warrant future investigation.
ment with our data that implicate miR-137 in promoting invasive While Robo2 and P2ry1 are intriguing, our functional studies
growth of PanNETs and associate its elevated expression with highlight the singular importance of suppressing ALK7 signaling
worse survival of breast cancer patients, previous studies have during primary tumorigenesis and metastasis in mouse models of
shown that miR-137 is up-regulated and promotes invasion in PanNET and breast cancer (25). We showed that inactivation of
breast, bladder, and non-small cell lung cancer (37, 47, 48). Chang ALK7 promotes metastasis to the lungs, brain, and liver, by en-
et al. (48) have shown that the promoter of miR-137 contains 4 E- abling cancer cells to escape apoptosis, which is otherwise trig-
boxes and that its expression in non-small cell lung cancer is reg- gered by activin B secreted by various cells in these tissue sites,
ulated by the EMT transcription factor Slug. Congruently, miR- including endothelial cells. In the present study, we have dem-
137 is up-regulated in the highly invasive MLP PanNET subtype, onstrated that the miR-23b cluster is a key repressor of ALK7
which is characterized by an EMT-like transcriptomic phenotype expression. Functional perturbation analysis reveals that ALK7 is a
(9, 22). Future studies will be required to identify the mechanisms key miR-23b target gene, since rescue experiments completely
responsible for the up-regulation of miR-137 in PanNET MLP abolished the prometastatic effect of the up-regulated miR-23b
tumors. cluster.
Using publicly available transcriptomic datasets, we show that In the aforementioned report (25) we also implicated activin
comparatively low expression of the top 5 candidate miR-137 B/ALK7 signaling as a barrier during the early stages of PanNET
target genes is associated with poor prognosis of breast cancer tumorigenesis as well as in metastasis, wherein the cell of origin,
patients. The similar association of elevated miR-137 with hu- the islet beta cell, expresses activin B as well as ALK7. Activin B
man breast cancer and mouse PanNET encourages their candi- was observed to be down-regulated concomitant with neoplastic

GENETICS
dacy as target genes. As a first step, in this study we explored the progression to proliferative and angiogenic precursor lesions and
effect of Sorl1 down-regulation in PanNETs; while known for its in turn nascent solid tumors, after which ALK7 was down-
role in trafficking of the amyloid precursor protein and its as- regulated during malignant progression (25). Using our newly
sociation with Alzheimer’s disease (49), a potential function in developed algorithm for associating miRNAs with candidate
cancer progression has not been established. Our data show that genes in reverse, we found that a family of miRNAs—miR-130a/b
down-regulation of Sorl1 enhances tumor growth in a transplant and miR-301a/b—is potentially responsible for down-regulation of
tumor model, and concordantly, comparatively low expression is activin B during the early stages of tumorigenesis. These miRNAs
associated with poorer prognosis of human breast, lung, and are up-regulated progressively, beginning at the hyperplastic/
gastric cancer. The precise role of Sorl1 and the pathways/ dysplastic islet progenitor stage, and their expression continues
functions it might regulate remain to be delineated. Future to increase in subsequent stepwise transitions to more aggressive
studies are also warranted to identify and functionally interro- lesions, congruent with down-regulation of activin B. Future
gate other miR-137 target genes, in particular the other top 4 functional studies analogous to those presented herein for miR-
candidates Kcnmb2, Hpse, Asph, and Robo2. Notably, the Sorl1 137 and the miR-23b cluster are warranted to test the hypothesis
knockdown alone did not phenocopy the effect of miR-137 on that the observed up-regulation of these miRNAs during multistep
invasiveness, implicating other targets as components of its tumorigenesis serves to repress activin B expression and thereby
phenotypic effects.
foster neoplastic progression. Interestingly, this family of miRNAs
Our results functionally implicating the miR-23b cluster (and
has been previously implicated in regulation of the canonical Tgfß
potentially its paralog miR-23a cluster) in metastatic colonization
signaling pathway (61, 62).
is both consistent with and extends upon previous publications.
Collectively, the results presented herein implicate a set of
For example, the miR-23b cluster is up-regulated in matched
distinctive miRNAs that are up-regulated during tumorigenesis
human breast cancer metastasis samples compared to primary
and progression to invasive and metastatic phenotypes (Fig. 8).
tumors, and it promotes experimental lung metastasis (50, 51).
A series of biologically relevant gene targets of these miRNAs
Both miR-23a and miR-23b were found to be up-regulated in
have been identified and initially characterized. While mutations
colorectal liver metastasis compared to primary tumors (52),
and other genetic aberrations have proven to be drivers of the
while miR-23a and miR-27a were shown to be associated with
initial stages of cell transformation to a cancerous state, less is
more aggressive gastric tumors and with their metastases (53,
54). Finally, the miR-23a/b cluster was found to be up-regulated known about the subsequent events that mediate invasion and
in premalignant pancreatic intraepithelial neoplasia (PanIN) le- metastasis. Our results align with other recent studies implicating
sions as well as in tumor stages of human pancreatic ductal ade- epigenetic mechanisms, in particular miRNA-mediated gene reg-
nocarcinoma (55), was further increased in metastatic lesions (56) ulation, in the specification of malignant phenotypes (6). Eluci-
and associated with poorer disease-free and overall survival (57). dation of the upstream regulatory mechanisms governing their
Herein, and in our recent study (25), we functionally charac- expression along with further mechanistic validation of the target
terized 3 of the identified miR-23b cluster target genes, ALK7, genes they suppress may enable the design of new therapeutic
Robo2, and P2ry1, and showed that their knockdowns individu- strategies.
ally enhanced liver metastasis. Our focus was on ALK7, given its Materials and Methods
potent and singular effects as a metastasis suppressor (25). That
Cell Lines, Culture Conditions, and Cell Culture Assays. The βTC3, STC-1, and
said, the other candidate target genes may be modulating other 293-T cell lines were cultured in DMEM media containing 10% (vol/vol) FBS.
aspects of the malignant phenotype. Robo2 belongs to the Slit/ The AJ-5257-1 was cultured in DMEM-F12 media containing 10% (vol/vol)
Robo family of axon guidance molecules, and while our under- FBS, 1% (vol/vol) insulin/transferrin/selenium, 4 μg/mL hydrocortisone, and
standing for this family’s role during cancer progression is still 5 μg/mL mouse EGF. Cell lines were maintained in a 5% CO2 incubator at 37 °C.
imprecise, it is thought that the molecules convey tumor suppressing Cell lines were tested to exclude Mycoplasma contamination. The genera-
signals, and accordingly, their expression is down-regulated in tion of stable cell lines, gene knockdown, treatment with cytokines, reporter

Michael et al. PNAS | November 26, 2019 | vol. 116 | no. 48 | 24193
assays, as well as plasmid generation, are described in detail in SI Appendix, Survival Analysis. Kaplan–Meier (KM) survival analysis was used to assess the
SI Materials and Methods. relationship of mIR-137 and its candidate target genes expression level with
survival using KM-plotter (65). For miR-137, the sample cohort was stratified
Western Blotting and Immunostaining. Western blots and immunostaining according to the best cutoff using false discovery rate (FDR) < 5% (66). In
were performed as has been previously described (25). Details are outlined in each analysis, the sample cohort was stratified into 4 quartiles based on
expression. The patients in the first and fourth quartiles were categorized as
SI Appendix, SI Materials and Methods.
the “high” and “low” expression, respectively.
Mouse mRNA Sequencing and Differential Expression Analysis. RNA sequencing
Quantification and Statistical Analysis. Statistical analysis was performed and
was done as has been previously described (63, 64); details are outlined in SI
graphics were produced using R version 3.0.1 or Prism version 7.0c.
Appendix, SI Materials and Methods.
Data Availability. RNA sequencing data have been deposited in the Gene
Animal Studies. Animal studies were performed as we have previously de-
Expression Omnibus (GEO) database, https://www.ncbi.nlm.nih.gov/geo (accession
scribed (25), according to protocols approved by the veterinary authorities no. GSE131887).
of the Canton Vaud according to Swiss law on animal protection with the
authorization VD3214 (to D.H.). Details are outlined in SI Appendix, SI Ma- Biological Materials Availability. All plasmids and cell lines codes are available
terials and Methods. from the corresponding author on reasonable request.

The Bio-miRTa Algorithm. Bio-miRTa has been designed to identify biologically ACKNOWLEDGMENTS. We thank M. De Palma (Swiss Institute for Experi-
relevant miRNA targets. The algorithm consists of 2 main steps: 1) extracting mental Cancer Research–Swiss Federal Institute of Technology Lausanne
all potential targets of the miRNA(s), and 2) incorporating experimental data [EPFL]) and M. Delorenzi (Swiss Institute of Bioinformatics [SIB]) for insightful
comments on the manuscript; M. Mina (SIB) for reviewing the code for Bio-
provided by the user to score and rank the target genes. The output of the miRTa; S. Lamy and P. Magliano for technical support; J. Dessimoz (histology
pipeline is a list of genes scored and ranked based on the relevance to the core facility–EPFL) for assistance with immunostaining; P. L. Sylvain and
unique biological system that is examined. A scheme of the pipeline is shown B. Mangeat (gene expression core facility–EPFL) for assistance with RNA-
in Fig. 4 and SI Appendix, Fig. S4. More details about the algorithm are Seq; N. Zangger (SIB) for assistance with GEO submission, and members of
the Hanahan laboratory for fruitful discussions. This work was supported by
described in SI Appendix, SI Materials and Methods.
research grants from the Swiss National Science Foundation to D.H. and by
The Bio-miRTa package is available on the Hanahan laboratory website: Canadian Institutes of Health Research and Human Frontier Science Program
https://www.epfl.ch/labs/hanahan-lab/data-and-tools/. Organization fellowships to I.P.M.

1. P. S. Steeg, Targeting metastasis. Nat. Rev. Cancer 16, 201–218 (2016). 24. G. Rindi et al., Development of neuroendocrine tumors in the gastrointestinal tract of
2. A. W. Lambert, D. R. Pattabiraman, R. A. Weinberg, Emerging biological principles of transgenic mice. Heterogeneity of hormone expression. Am. J. Pathol. 136, 1349–1363
metastasis. Cell 168, 670–691 (2017). (1990).
3. J. Massagué, A. C. Obenauf, Metastatic colonization by circulating tumour cells. Nature 25. I. P. Michael et al., ALK7 signaling manifests a homeostatic tissue barrier that is
529, 298–306 (2016). abrogated during tumorigenesis and metastasis. Dev. Cell 49, 409–424.e6 (2019).
4. C. L. Chaffer, R. A. Weinberg, A perspective on cancer cell metastasis. Science 331, 26. V. Agarwal, G. W. Bell, J.-W. Nam, D. P. Bartel, Predicting effective microRNA target
1559–1564 (2011). sites in mammalian mRNAs. eLife 4, 101 (2015).
5. A. P. Makohon-Moore et al., Limited heterogeneity of known driver gene mutations 27. B. John et al., Human MicroRNA targets. PLoS Biol. 2, e363 (2004).
among the metastases of individual patients with pancreatic cancer. Nat. Genet. 49, 28. M. D. Paraskevopoulou et al., DIANA-microT web server v5.0: Service integration into
miRNA functional analysis workflows. Nucleic Acids Res. 41, W169–W173 (2013).
358–366 (2017).
29. M. Kertesz, N. Iovino, U. Unnerstall, U. Gaul, E. Segal, The role of site accessibility in
6. S. A. Patel, S. Vanharanta, Epigenetic determinants of metastasis. Mol. Oncol. 11, 79–96
microRNA target recognition. Nat. Genet. 39, 1278–1284 (2007).
(2017).
30. N. Pinzón et al., microRNA target prediction programs predict many false positives.
7. M. G. H. Chun, J.-H. Mao, C. W. Chiu, A. Balmain, D. Hanahan, Polymorphic genetic
Genome Res. 27, 234–245 (2017).
control of tumor invasion in a mouse model of pancreatic neuroendocrine carcino-
31. A. Fridrich, Y. Hazan, Y. Moran, Too many false targets for MicroRNAs: Challenges
genesis. Proc. Natl. Acad. Sci. U.S.A. 107, 17268–17273 (2010). and pitfalls in prediction of miRNA targets and their gene ontology in model and
8. O. G. McDonald et al., Epigenomic reprogramming during pancreatic cancer pro- non-model organisms. Bioessays 41, e1800169 (2019).
gression links anabolic glucose metabolism to distant metastasis. Nat. Genet. 49, 367– 32. D. Karagkouni et al., DIANA-TarBase v8: A decade-long collection of experimentally
376 (2017). supported miRNA-gene interactions. Nucleic Acids Res. 46, D239–D245 (2018).
9. P. Olson et al., MicroRNA dynamics in the stages of tumorigenesis correlate with 33. J.-H. Li, S. Liu, H. Zhou, L.-H. Qu, J.-H. Yang, starBase v2.0: Decoding miRNA-ceRNA,
hallmark capabilities of cancer. Genes Dev. 23, 2152–2165 (2009). miRNA-ncRNA and protein-RNA interaction networks from large-scale CLIP-Seq data.
10. V. N. Kim, MicroRNA biogenesis: Coordinated cropping and dicing. Nat. Rev. Mol. Cell Nucleic Acids Res. 42, D92–D97 (2014).
Biol. 6, 376–385 (2005). 34. M. Korpal, Y. Kang, The emerging role of miR-200 family of microRNAs in epithelial-
11. D. P. Bartel, Metazoan MicroRNAs. Cell 173, 20–51 (2018). mesenchymal transition and cancer metastasis. RNA Biol. 5, 115–119 (2008).
12. J. A. Vidigal, A. Ventura, The biological functions of miRNAs: Lessons from in vivo 35. I. P. Michael, N. Zangger, D. Hanahan, Analysis of gene expression from βTC3 cells
studies. Trends Cell Biol. 25, 137–147 (2015). of Rip-Tag2 PanNET mouse model upon miR-137 and miR-23b cluster doxycycline
13. J. Krützfeldt, M. N. Poy, M. Stoffel, Strategies to determine the biological function of (DOX) induction. Gene Expression Omnibus. https://www.ncbi.nlm.nih.gov/geo/query/
microRNAs. Nat. Genet. 38 (suppl.), S14–S19 (2006). acc.cgi?acc=GSE131887. Deposited 13 September 2017.
14. M. Thomas, J. Lieberman, A. Lal, Desperately seeking microRNA targets. Nat. Struct. 36. M. M. Pillai et al., HITS-CLIP reveals key regulators of nuclear receptor signaling in
Mol. Biol. 17, 1169–1174 (2010). breast cancer. Breast Cancer Res. Treat. 146, 85–97 (2014).
15. A. C. Title et al., Genetic dissection of the miR-200-Zeb1 axis reveals its importance in 37. X. Ying, Y. Sun, P. He, MicroRNA-137 inhibits BMP7 to enhance the epithelial-mes-
tumor differentiation and invasion. Nat. Commun. 9, 4671 (2018). enchymal transition of breast cancer cells. Oncotarget 8, 18348–18358 (2017).
16. A. Ventura, T. Jacks, MicroRNAs and cancer: Short RNAs go a long way. Cell 136, 586– 38. Y. Xue et al., Direct conversion of fibroblasts to neurons by reprogramming PTB-
591 (2009). regulated microRNA circuits. Cell 152, 82–96 (2013).
17. M. S. Nicoloso, R. Spizzo, M. Shimizu, S. Rossi, G. A. Calin, MicroRNAs–The micro 39. J. Massagué, How cells read TGF-beta signals. Nat. Rev. Mol. Cell Biol. 1, 169–178
(2000).
steering wheel of tumour metastases. Nat. Rev. Cancer 9, 293–302 (2009).
40. S. Kishore et al., A quantitative analysis of CLIP methods for identifying binding sites
18. S. Baranwal, S. K. Alahari, miRNA control of tumor cell invasion and metastasis. Int. J.
of RNA-binding proteins. Nat. Methods 8, 559–564 (2011).
Cancer 126, 1283–1290 (2010).
41. E. Gottwein et al., Viral microRNA targetome of KSHV-infected primary effusion
19. N. Pencheva, S. F. Tavazoie, Control of metastatic progression by microRNA regula-
lymphoma cell lines. Cell Host Microbe 10, 515–526 (2011).
tory networks. Nat. Cell Biol. 15, 546–554 (2013).
42. R. L. Boudreau et al., Transcriptome-wide discovery of microRNA binding sites in
20. Y. Peng, C. M. Croce, The role of MicroRNAs in human cancer. Signal Transduct.
human brain. Neuron 81, 294–305 (2014).
Target. Ther. 1, 15004 (2016). 43. L. Zhou et al., MiR-27a-3p functions as an oncogene in gastric cancer by targeting
21. A. Lujambio et al., A microRNA DNA methylation signature for human cancer me- BTG2. Oncotarget 7, 51943–51954 (2016).
tastasis. Proc. Natl. Acad. Sci. U.S.A. 105, 13556–13561 (2008). 44. D. Betel, A. Koppal, P. Agius, C. Sander, C. Leslie, Comprehensive modeling of microRNA
22. A. Sadanandam et al., A cross-species analysis in pancreatic neuroendocrine tumors targets predicts functional non-conserved and non-canonical sites. Genome Biol. 11, R90
reveals molecular subtypes with distinctive clinical, metastatic, developmental, and (2010).
metabolic characteristics. Cancer Discov. 5, 1296–1313 (2015). 45. D. Hanahan, Heritable formation of pancreatic beta-cell tumours in transgenic mice
23. L. Li, D. Hanahan, Hijacking the neuronal NMDAR signaling circuit to promote tumor expressing recombinant insulin/simian virus 40 oncogenes. Nature 315, 115–122
growth and invasion. Cell 153, 86–100 (2013). (1985).

24194 | www.pnas.org/cgi/doi/10.1073/pnas.1913307116 Michael et al.

46. E. Mahmoudi, M. J. Cairns, MiR-137: An important player in neural development and 57. A. E. Frampton et al., MicroRNAs cooperatively inhibit a network of tumor suppressor
neoplastic transformation. Mol. Psychiatry 22, 44–55 (2017). genes to promote pancreatic tumor growth and progression. Gastroenterology 146,
47. Y. Xiu et al., MicroRNA-137 upregulation increases bladder cancer cell proliferation 268–77.e18 (2014).
and invasion by targeting PAQR3. PLoS One 9, e109734 (2014). 58. R. K. Gara et al., Slit/Robo pathway: A promising therapeutic target for cancer. Drug
48. T.-H. Chang et al., Upregulation of microRNA-137 expression by Slug promotes tumor Discov. Today 20, 156–164 (2015).
invasion and metastasis of non-small cell lung cancer cells through suppression of 59. F. Gu, Y. Ma, J. Zhang, F. Qin, L. Fu, Function of Slit/Robo signaling in breast cancer.
TFAP2C. Cancer Lett. 402, 190–202 (2017). Front. Med. 9, 431–436 (2015).
49. E. Rogaeva et al., The neuronal sortilin-related receptor SORL1 is genetically associated 60. J. Baran ska, R. Czajkowski, P. Pomorski, P2Y1Receptors–Properties and functional
with Alzheimer disease. Nat. Genet. 39, 168–177 (2007).
activities. Adv. Exp. Med. Biol. 1051, 71–89 (2017).
50. L. Jin et al., Prooncogenic factors miR-23b and miR-27b are regulated by Her2/Neu,
61. A. Jacobsen et al., Analysis of microRNA-target interactions across diverse cancer
EGF, and TNF-α in breast cancer. Cancer Res. 73, 2884–2896 (2013).
types. Nat. Struct. Mol. Biol. 20, 1325–1332 (2013).
51. B. Ell et al., The microRNA-23b/27b/24 cluster promotes breast cancer lung metastasis by
62. L. Liu et al., The oncogenic role of microRNA-130a/301a/454 in human colorectal
targeting metastasis-suppressive gene prosaposin. J. Biol. Chem. 289, 21888–21895 (2014).
cancer via targeting Smad4 expression. PLoS One 8, e55532 (2013).
52. P. Vychytilova-Faltejskova et al., Genome-wide microRNA expression profiling in primary
63. S. Hendrikx et al., Endothelial calcineurin signaling restrains metastatic outgrowth by
tumors and matched liver metastasis of patients with colorectal cancer. Cancer Genomics
regulating Bmp2. Cell Rep. 26, 1227–1241.e6 (2019).
Proteomics 13, 311–316 (2016).
64. R. Micheletti et al., The long noncoding RNA Wisper controls cardiac fibrosis and
53. J. An et al., MiR-23a in amplified 19p13.13 loci targets metallothionein 2A and pro-
motes growth in gastric cancer cells. J. Cell. Biochem. 114, 2160–2169 (2013). remodeling. Sci. Transl. Med. 9, eaai9118 (2017).
54. Z. Zhang, S. Liu, R. Shi, G. Zhao, miR-27 promotes human gastric cancer cell metastasis 65. B. Györffy et al., An online survival analysis tool to rapidly assess the effect of 22,277
by inducing epithelial-to-mesenchymal transition. Cancer Genet. 204, 486–491 (2011). genes on breast cancer prognosis using microarray data of 1,809 patients. Breast
55. M. Humeau et al., Salivary MicroRNA in pancreatic cancer patients. PLoS One 10, Cancer Res. Treat. 123, 725–731 (2010).
e0130996 (2015). 66. A. Lánczky et al., miRpower: A web-tool to validate survival-associated miRNAs uti-
56. G. Wu et al., MicroRNA-23a promotes pancreatic cancer metastasis by targeting ep- lizing expression data from 2178 breast cancer patients. Breast Cancer Res. Treat. 160,
ithelial splicing regulator protein 1. Oncotarget 8, 82854–82871 (2017). 439–446 (2016).

GENETICS

Michael et al. PNAS | November 26, 2019 | vol. 116 | no. 48 | 24195