LIN28A Modulates Splicing and Gene Expression Programs in Breast Cancer Cells

Abstract

LIN28 is an evolutionarily conserved RNA-binding protein with critical functions in developmental timing and cancer. However, the molecular mechanisms underlying LIN28's oncogenic properties are yet to be described. RNA-protein immunoprecipitation coupled with genome-wide sequencing (RIP-Seq) analysis revealed significant LIN28 binding within 843 mRNAs in breast cancer cells. Many of the LIN28-bound mRNAs are implicated in the regulation of RNA and cell metabolism. We identify heterogeneous nuclear ribonucleoprotein A1 (hnRNP A1), a protein with multiple roles in mRNA metabolism, as a LIN28-interacting partner. Subsequently, we used a custom computational method to identify differentially spliced gene isoforms in LIN28 and hnRNP A1 small interfering RNA (siRNA)-treated cells. The results reveal that these proteins regulate alternative splicing and steady-state mRNA expression of genes implicated in aspects of breast cancer biology. Notably, cells lacking LIN28 undergo significant isoform switching of the ENAH gene, resulting in a decrease in the expression of the ENAH exon 11a isoform. The expression of ENAH isoform 11a has been shown to be elevated in breast cancers that express HER2. Intriguingly, analysis of publicly available array data from the Cancer Genome Atlas (TCGA) reveals that LIN28 expression in the HER2 subtype is significantly different from that in other breast cancer subtypes. Collectively, our data suggest that LIN28 may regulate splicing and gene expression programs that drive breast cancer subtype phenotypes.

INTRODUCTION

LIN28A is an evolutionarily conserved RNA-binding protein that plays important and widespread roles in development and disease (1, 2). LIN28A was first identified in a screen of mutants of the nematode Caenorhabditis elegans displaying defects in developmental timing (3). Subsequent studies have identified two homologs, LIN28A and LIN28B, in mammals, including humans and mice (4). LIN28A (here referred to as LIN28) is highly expressed during development and in human and mouse embryonic stem (ES) cells (5, 6). Conversely, LIN28 is rarely expressed in normal adult tissues except when reactivated in cancer (7–10). Abnormal LIN28 expression has been observed in a number of human malignancies, suggesting that LIN28 is important in cancer and most likely functions as an oncogene (7, 8). Overexpression of LIN28 promotes tumor cell migration and cellular transformation, which are associated with advanced stages of poorly differentiated human cancers, including liver cancer, ovarian cancer, and myeloid leukemia (8, 11).

Mechanistically, the effects of LIN28 on multiple unrelated biological and pathological processes have been attributed to the ability of LIN28 to block the biogenesis of the Let-7 family of microRNAs (miRNAs) (12–14). Members of the Let-7 family of miRNAs act as tumor suppressors by inhibiting the expression of oncogenes and key regulators of mitogenic pathways, including c-myc, K-Ras, and HMGA2 (15–17). Consistent with this idea, low levels of Let-7 and high levels of LIN28 are strongly associated with increased tumorigenesis and poor disease prognosis (8, 18).

On the other hand, recent studies indicate that LIN28 can change gene regulatory networks independent of Let-7, suggesting that LIN28 may contribute to tumor progression through Let-7-independent mechanisms (5, 19–23). LIN28 directly binds and stimulates the translation of several mRNAs that encode proteins involved in multiple cellular processes that drive cancer progression (21, 24–26). As an example, LIN28 regulates the translation and expression of several cell cycle regulatory mRNAs that encode factors controlling the G₂/S-to-M-phase transition, consistent with a role for LIN28 in cell growth and tumor promotion (22, 25). Beyond regulating the cell cycle, LIN28 also binds and regulates the translation of mRNAs encoding cell metabolic enzymes driving glycolysis and mitochondrial respiration (5, 23, 24). This would be consistent with the reprogrammed glucose metabolism needed to support the energetic requirements for proliferation and increased cell mass characteristic of tumor cells (1, 27).

Despite the reactivation of LIN28 in many cancers, knowledge of the molecular mechanisms by which LIN28 functions to promote specific types of cancer, including breast cancer, is lacking. LIN28 is expressed in breast cancer tumors, and recent studies have shown that LIN28 is a powerful predictor of poor prognoses and patient clinical outcomes (8, 28, 29). With this in mind, we were interested in identifying novel LIN28 mRNA targets that could provide insights into the function of LIN28 in breast cancer.

We performed RNA-protein immunoprecipitation (RIP) coupled with genome-wide sequencing (RIP-Seq) to identify endogenous LIN28 mRNA targets (30). Several studies have described LIN28 gene regulatory networks, but less is known about transacting factors that mediate its functions. We have used protein immunoprecipitation (IP) combined with mass spectrophotometry (MS) analyses to identify transacting factors that could modulate LIN28 regulatory functions in breast cancer cells (31). Our MS analyses reveal that the nucleocytoplasmic shuttling protein heterogeneous nuclear ribonucleoprotein A1 (hnRNP A1) is a protein associate of LIN28 in breast cancer cells. Furthermore, transcriptome sequencing (RNA-Seq) data suggest that LIN28 regulates the expression of a unique set of mRNAs independent of LIN28 binding, indicating that LIN28 regulates gene expression via diverse and distinct regulatory mechanisms. Importantly, we show that cells lacking LIN28 exhibit differential expression of Enabled Homolog (ENAH) gene isoforms, consistent with a role in alternative splicing.

The suite of mRNAs whose expression is affected includes genes that function in cell metabolism, the immune response, cell proliferation, and cell-to-cell communication, processes that are classic hallmarks of breast cancer biology. Moreover, using publically available data from the Cancer Genome Atlas (TCGA), we uncover striking and significant correlations between the expression of LIN28 and specific breast cancer subtypes, underscoring an important role for LIN28 in breast cancer biology. Our studies provide mechanistic insights into how LIN28 may adopt specialized functions in regulating specific gene targets that control a disease process.

MATERIALS AND METHODS

Cell culture and siRNA transfection.

The generation and culture of MCF-7 cells stably expressing the glucocorticoid receptor and the mouse mammary tumor virus (MMTV) long terminal repeat (LTR) promoter fused to the luciferase gene reporter (MMTV-LUC) were described previously (32). Cells were grown in a humidified incubator at 37°C with 5% CO₂ in minimal essential medium (MEM) supplemented with 2 mM glutamine, 100 μg/ml penicillin-streptomycin, 10 mM HEPES, 10% fetal bovine serum (FBS), and 300 μg/ml G418. For all experiments, cells were seeded overnight in phenol red-free MEM supplemented with 5% charcoal-stripped calf serum and 2 mM glutamate.

For small interfering RNA (siRNA) experiments, 0.2 × 10⁶ cells were cultured in 6-well plates in MEM without penicillin-streptomycin or G418. The next day, cells were rinsed twice with Opti-MEM medium and transfected with 50 to 100 pM siRNA by using Lipofectamine according to the manufacturer's instructions (Invitrogen Life Technology, Grand Island, NY). siRNAs targeting LIN28 and hnRNP A1 and control siRNA were obtained from Dharmacon, Thermo Scientific (Waltham, MA).

Quantitative RT-PCR (qRT-PCR).

Total RNA was extracted by using a Qiagen kit (Qiagen, Valencia, CA). For reverse transcriptase PCR (RT-PCR) analysis, cDNA was synthesized according to standard protocols after DNase I treatment (Invitrogen). Following reverse transcription, cDNA was used for real-time PCR employing SYBR green detection. Real-time PCR was performed with a Stratagene MX 3000P instrument using Brilliant SYBR green quantitative PCR master mix (Agilent, La Jolla, CA). All reactions were performed with a model MX 3000P sequence detector. Glyceraldehyde-3-phosphate dehydrogenase (GAPDH), ribosomal protein L13A (RPL13A), and TATA-binding protein (TBP) were used to normalize the differences in the amounts of mRNA in each reaction mixture. PCR primer pairs used in this study are listed in Table 1.

TABLE 1.

List of primers used in this study

Primer	Sequence	Purpose
LIN28 forward	GAGTGAGAGGCGGCCAAAA	qRT-PCR
LIN28 reverse	TGATGATCTAGACCTCCACAGTTGTAG	qRT-PCR
ATF3 forward	AGGTTTGCCATCCAGAACAA	qRT-PCR
ATF3 reverse	GCTACCTCGGCTTTTGTGAT	qRT-PCR
CHMP2A forward	AGCCCTAGCTGATGCTGATG	qRT-PCR
CHMP2A reverse	TGGGCATCCACTGGTTATCT	qRT-PCR
CRYAB forward	TCATCTCCAGGGAGTTCCAC	qRT-PCR
CRYAB reverse	GCCAGAGACCTGTTTCCTTG	qRT-PCR
MGMT forward	GGCACCGCTGTATTAAAGGA	qRT-PCR
MGMT reverse	ATAGAGCAAGGGCAGCGTTA	qRT-PCR
NGDN forward	TGACATCAGTGCTTTGACAGG	qRT-PCR
NGDN reverse	TTCTTCCGACCTTTCTGAGG	qRT-PCR
NKX3-1 forward	AACCATTTCACCCAGACAGC	qRT-PCR
NKX3-1 reverse	CAGATTGGAGCAGGGTTTGT	qRT-PCR
PSMA4 forward	GGTGCAGTGGGAAAATAGGA	qRT-PCR
PSMA4 reverse	AAAGGGGATGTGTGTGGAAG	qRT-PCR
PSMB7 forward	TGCTGGAAGAAACAGTCCAA	qRT-PCR
PSMB7 reverse	TCTTACTGGGCCTCAATGCT	qRT-PCR
PSMC5 forward	CAAGGTCATGCAGAAGGACA	qRT-PCR
PSMC5 reverse	ACTTGGCCCACAGAGCTTTA	qRT-PCR
S100A9 forward	TCATCAACACCTTCCACCAA	qRT-PCR
S100A9 reverse	TCTTTTCGCACCAGCTCTTT	qRT-PCR
S100P forward	TAGTGTTCGTGGCTGCAATC	qRT-PCR
S100P reverse	GACATCTCCAGGGCATCATT	qRT-PCR
RPL13A forward	CTCAAGGTCGTGCGTCTGAA	qRT-PCR
RPL13A reverse	TGGCTGTCACTGCCTGGTACT	qRT-PCR
Beta-actin forward	CTGTGGCATCCACGAAACTA	qRT-PCR
Beta-actin reverse	AGCACTGTGTTGGCGTACAG	qRT-PCR
TBP forward	TCAAACCCAGAATTGTTCTCCTTAT	qRT-PCR
TBP reverse	CCTGAATCCCTTTAGAATAGGGTAGA	qRT-PCR
GAPDH forward	CTCTGCCCCCTCTGCTGAT	qRT-PCR
GAPDH reverse	GTGCAGGAGGCATTGCTGAT	qRT-PCR
PKM1 forward	AGAAACAGCCAAAGGGGACT	qRT-PCR
PKM1 reverse	GAGGCTCGCACAAGTTCTTC	qRT-PCR
PKM2 forward	CTCACCAGGTGGCCAGATAC	qRT-PCR
PKM2 reverse	GCACAGCACAGGGAAGATG	qRT-PCR
PKM1 forward	CAGCAGCTTTGATAGTTCTGACGGA	PCR
PKM1 reverse	TCACGGCACAGGAACAACACGC	PCR
PKM2 forward	CGCCCATTACCAGCGACCCCACA	PCR
PKM2 reverse	TCACGGCACAGGAACAACACGC	PCR
hMENA exon 11a forward	CAACAAGAAAACCTTGGGAAAG	qRT-PCR
hMENA exon 11a reverse	TTCCTTGGAGAATCCCGTCT	qRT-PCR
hMENA exon 6 forward	CCTGCCTCTGTTGAGACTCC	qRT-PCR
hMENA exon 6 reverse	GCAAGTGGTCCCAAGACAAT	qRT-PCR
CH25H forward	TCTTTGGGCTTCTTCGACAT	qRT-PCR
CH25H reverse	AAGGGCACCAGTCTGTGAGT	qRT-PCR
IFI44L forward	TAGGCCCTATGCAGACTTGG	qRT-PCR
IFI44L reverse	TGATATCAGACCCCACTACGG	qRT-PCR
RSAD2 forward	CTTTTGCTGGGAAGCTCTTG	qRT-PCR
RSAD2 reverse	CAGCTGCTGCTTTCTCCTCT	qRT-PCR
PARM1 forward	TTTGGAGTTGCAGCCTACCT	qRT-PCR
PARM1 reverse	CAGGACCCGTAGTCATGGTC	qRT-PCR
IFI27 forward	CCCTGCAGAGAAGAGAACCA	qRT-PCR
IFI27 reverse	CTCTGGAGATGCAGAATTTGG	qRT-PCR
KIF12 forward	TGGAATTTGGGAGTCTGGAG	qRT-PCR
KIF12 reverse	GCAGTTTGACGGCTGATGTA	qRT-PCR
OASL forward	ATTGTGCCTGCCTACAGAGC	qRT-PCR
OASL reverse	GCAGAAATTTCCAGGACCAC	qRT-PCR
MAN1A1 forward	ATGGCCCAACACTACCTTGA	qRT-PCR
MAN1A1 reverse	TGTAGCGATGGCTTCAACAC	qRT-PCR
MAN1A1 forward	CACCTGAAGCAGCAAGTGAG	qRT-PCR
MAN1A1 reverse	CGCAGATTCATGAACACGGT	qRT-PCR

Immunoprecipitation and Western blotting.

For whole-cell extracts, cells were lysed as previously described (32). Cells were washed twice with phosphate-buffered saline (PBS) and pelleted by centrifugation. The cells were lysed by the addition of buffer X (100 mM Tris-HCl [pH 8.5], 250 mM NaCl, 1% [vol/vol] Nonidet P-40, 1 mM EDTA, and protease inhibitors [Sigma-Aldrich, St. Louis, MO]).

Nuclear and cytoplasmic extracts were prepared from MCF-7M cells by using a Pierce NE-PER kit according to the manufacturer's instructions (Thermo Scientific, Grand Island, NY). For Western blotting, 10 to 50 μg of protein was resolved by 4%-to-12% gradient or 16% sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto a polyvinyl difluoride (PVDF) membrane (Amersham). Proteins were immunoblotted by using various antibodies, as indicated in the figures. For immunoprecipitation, 1 mg of whole-cell, nuclear, or cytosolic extract was incubated overnight with 2 μg of antibodies against LIN28 (Abcam), hnRNP A1 (Millipore), DDX3, Ku70 (Santa Cruz Biotechnology), and PABPC4 (Bethyl Labs). The next day, immunocomplexes were captured by incubation with GE Sepharose Xtra magnetic beads (Thermo Scientific, Lafayette, CO) for 1 h. After incubation, immunocomplexes were washed three times in HEGNDT buffer (10 mM HEPES [pH 8.0], 1 mM EDTA, 10% glycerol, 50 mM NaCl, 2 mM dithiothreitol [DTT]) containing 0.1% Triton X-100 and once in HEGNDT buffer without Triton X-100. Following a final wash, the immunocomplexes were resuspended in 2× SDS-PAGE loading buffer and released by boiling for 10 min. For Western blotting, the solubilized proteins were subjected to SDS-PAGE, transferred onto a PVDF membrane, and detected by Western blotting using the antibodies of interest. Specific antibodies and their commercial sources are listed in Table 2.

TABLE 2.

List of antibodies used in this study

Antibody	Catalog no.	Source
LIN28	11724-1-AP	Protein Tech
LIN28	ab46020	Abcam
hnRNP A1	04-1469	Millipore
DDX3	sc-365768	Santa Cruz
PABPC4	A301-466A	Bethyl
Ku70a		Santa Cruz
Beta-actin	A1978-clone Ac-15	Sigma
GAPDH	RDI-TRK5G4-6C5	RDI

^aCatalog no. sc-9033.

Immunofluorescence microscopy and immunohistochemistry.

Cells were seeded at a low density on 4-well chamber slides. After 48 h, cells were either fixed with 4% paraformaldehyde for 15 min and permeabilized for 10 min with 0.5% Triton X-100 in PBS at room temperature or fixed with cold 1:1 methanol and acetone at −20°C for 10 min, followed by further permeabilization with 0.5% Triton X-100 in PBS. Cells were blocked with 0.1% Triton X-100 and 5% normal donkey serum in PBS for 1 h and then incubated with primary antibody overnight. After being washed three times with 0.1% Triton X-100 in PBS, cells were incubated with secondary antibody for 2 h. After being washed three times with 0.1% Triton X-100 in PBS, slides were mounted by using Prolong Gold mounting medium with 4′,6-diamidino-2-phenylindole (DAPI) (Invitrogen). Antibodies used included LIN28 (Abcam, Protein Tech) and hnRNP A1 (Millipore) antibodies and Alexa Fluor 488- and 594-conjugated secondary antibodies (Invitrogen). All images were taken with a Zeiss 710 confocal microscope, with excitation at 405 μm, 488 μm, and 561 μm. Image J software (NIH, Bethesda, MD) was used to overlay images.

RNA immunoprecipitation.

LIN28 RIP-Seq was performed according to a modified RNA protein immunoprecipitation-microarray (RIP-Chip) protocol described previously (30). mRNA-protein complex (mRNP) lysates were prepared from MCF-7M cells by using polysome lysis buffer (100 mM KCl, 5 mM MgCl₂, 10 mM HEPES [pH 7.0], 0.5% NP-40, 1 mM DTT, 100 U ml⁻¹ RNase Out, 400 μM vanadyl ribonucleoside complexes, protease inhibitor cocktail). mRNP lysates were immediately frozen at −100°C in order to complete the lysis process. Protein A-Sepharose beads (Sigma-Aldrich) were preincubated with 6 μg of either LIN28 (Abcam) or control normal rabbit serum IgG antibodies. Equal amounts of mRNP lysates (3 mg) were incubated with antibody-precoated Sepharose beads. Beads were extensively washed with NT2 buffer (50 mM Tris-HCl [pH 7.4], 150 mM NaCl, 1 mM MgCl₂, and 0.05% NP-40). Proteins were digested with 30 μg of proteinase K followed by RNA isolation with phenol-chloroform-isoamyl alcohol. RNA was precipitated with a solution containing 3 M sodium acetate, 100% ethanol, and 20 μg glycogen as a carrier.

RIP-Seq (Illumina/Solexa RNA sequencing) cDNA library construction.

To determine which mRNAs were associated with LIN28, cDNA library synthesis from LIN28-mRNA immunocomplexes was carried out according to the manufacturer's protocols (Illumina, San Diego, CA). Briefly, mRNA was purified from 100 ng of immunocomplexed total RNA by using poly(T) magnetic beads. The mRNA was fragmented into small pieces by using divalent cations at an elevated temperature. The RNA fragments were synthesized into first-strand cDNA by using reverse transcriptase and random primers, followed by second-strand cDNA synthesis using DNA polymerase I and RNase H. The resulting cDNA fragments were repaired by adding a single “A” base and ligated with adaptors. The ligation products were then purified and amplified by PCR to construct the cDNA library according to the manufacturer's instructions (Illumina). Libraries were sequenced as 35-mers by using the Illumina genome analyzer.

For RNA-Seq, two biological replicates of total RNA from MCF-7M cells treated with 100 pmol of nontargeting siRNA as a control and siRNAs targeting LIN28, hnRNP A1, and a combination of LIN28 and hnRNP A1 were prepared by using a Norgen total RNA kit (Norgen Biotek Corp., Ontario, Canada). Eight total RNA samples were submitted to the NIH Intramural Sequencing Center (NISC) for sequencing. Starting with 100 ng total RNA, poly(A)-tailed mRNA was isolated with oligo(dT) and fragmented to average sizes of 250 bp (Covaris, Inc., MA). Briefly, cDNA and the cDNA library were constructed according to Illumina TruSeq protocols (Illumina, San Diego, CA). To minimize the per-lane batch effect, all 8 samples were multiplexed in each lane. A total of three lanes were used to achieve the desired depth of ∼50 million raw reads per sample from 100-bp paired-end (PE) sequencing.

RIP-Seq processing.

The libraries were sequenced for 50 cycles on the Illumina genome analyzer according to standard protocols. Raw reads were mapped to the human genome (hg19/GRCh37) with Bowtie aligner v0.12.8 (33) using default parameters, except that −m 1 was set to suppress all alignments for a read if more than one reportable alignment exists for it. Thus, unique reads were aligned to the genome, allowing only 2 mismatches, and multiple hits were discarded.

Detection of LIN28-bound genes.

We identified LIN28-bound genes by first using htseq-count from HTSeq Framework version 0.6.0 with the aligned reads to obtain read counts for each gene for both the LIN28 RIP and the IgG control. We then performed a one-tailed Fisher exact test for each gene, testing for an overrepresentation of reads in the LIN28 RIP compared to the control. We adjusted for multiple comparisons by calculating a Benjamini-Hochberg false discovery rate (FDR) value for each gene. Finally, we defined a gene as being LIN28 bound if it had an FDR value of <0.05 and a fold change of 2.5 or higher when comparing depth-normalized read counts for the LIN28 RIP to those for the control.

Determination of GGAGA prevalence.

We downloaded the transcript sequences of all RefSeq genes from the UCSC Genome Browser on 3 March 2015. For genes that had multiple transcripts, we used only the longest transcript. Next, we counted the times that the sequence GGAGA was present in each gene. We divided each GGAGA count by transcript length and multiplied this value by 1,000 to normalize the counts into GGAGA counts per kilobase of exon. Finally, we split the genes into LIN28-bound and unbound groups based on whether or not the gene was one of the 843 LIN28 RIP-Seq targets. To determine the significance of the differences between groups, we performed a one-sided Wilcoxon rank sum test.

RNA-Seq processing.

We filtered the raw reads to include only those with a median Phred quality score of 20 or higher. We used Trim Galore! version 0.2.8 to trim any reads containing an adaptor sequence. We estimated the average mate inner distance and the variance of the fragment lengths by first aligning a subset of the reads to the known hg19 gene transcriptome by using Bowtie version 0.12.8 with the following parameters: −v 2 −m 1 −X 1,000. We then used the CollectInsertSizeMetrics tool from Picard Tools suite version 1.86 to obtain the mean and variance estimates for the fragment lengths, and we calculated the average mate inner distance as follows: average mate inner distance = mean fragment length − 2 × read length. We defined the gene model by using RefGene annotations downloaded from the UCSC Genome Browser on 5 February 2013. We aligned the filtered and trimmed reads to the hg19 genome assembly using TopHat2 version 2.0.4 with the mate inner distance, mate standard deviation, and gene model parameters defined from the previous steps as well as the following additional parameter: −g 10. We quantified expression by using CuffLinks version 2.0.2 with the gene model to obtain measurements of the number of fragments per kilobase per million (FPKM) reads for both individual gene isoforms and genes as a whole.

Detection of differentially expressed genes.

We detected differentially expressed genes (DEGs) between two groups by using CuffDiff version 2.0.2 with the above-described gene model. We defined genes as being differentially expressed between groups if they had an FDR q value of <0.05.

Correlation between LIN28 3′-UTR binding and gene expression.

We determined the density of LIN28 binding in the 3′ untranslated region (UTR) and the DNA coding sequences (CDS) for the 843 LIN28 targets by calculating the LIN28 RIP-Seq read coverage across these features in each gene. We calculated a 3′-UTR score based on the log₂ ratio of length-normalized binding in the 3′ UTR compared to that in the CDS. To determine the correlation with expression changes after LIN28 knockdown (KD), we calculated the Pearson r correlation coefficient between the 3′-UTR score and the LIN28 KD versus nontargeting (NT) KD RNA-Seq CuffDiff log₂ fold change values.

Determination of isoform usage by measurement of FPKM values using CuffLinks.

We first calculated isoform usage percentages by using the isoform FPKM values obtained from CuffLinks as

$$u_j^i=\frac{f_j^i}{{\mathrm{\Sigma }}_{j=1}^J{\hspace{0.17em}f}_j^i}$$

where $u_j^i$ is the isoform usage percentage for the j-th isoform of the i-th gene, $f_j^i$ is the FPKM value for the j-th isoform of the i-th gene, and J is the total number of isoforms for the i-th gene. We removed isoforms that did not have an FPKM value of 0.1 or higher in all samples.

Detection of differential splicing using TopHat2 splice junctions.

After alignment, TopHat2 produces a BED file with the splice junctions produced during alignment. These splice junctions represent instances where a read spans multiple exons. We used these splice junctions to detect differential splicing between two groups.

First, we assembled a list of counts for each splice junction and group. We moved the start and end genomic coordinates of each splice junction to the nearest start or end of an exon in the previously defined gene model. If we did not find a start or end of an exon within 1,000 bases, we excluded the splice junction. We then summed together the splice junction counts for each sample within each group to generate the final list of counts.

Next, we identified alternative splicing events that were significantly different between groups. We defined alternative splicing events as pairs of splice junctions where the genomic coordinates of one end of the splice junction match but those of the other end are different. We used the counts for these two splice junctions for each group to perform a Fisher exact test. We eliminated any instances where any of the counts were <5. We corrected for multiple comparisons by calculating Benjamini-Hochberg FDR q values. We defined alternative splicing events as being significantly different between groups if they had an FDR q value of <0.05.

Finally, we identified genes with differential splicing between groups. We excluded any genes in our gene model that were duplicated. We defined a gene as having differential splicing if it had one or more differential alternative splicing events in the 5′ direction and one or more in the 3′ direction and if these events spanned 3 or more exons. After identifying these genes, we identified the region within the gene where alternative splicing may be occurring as the region between the most upstream differential alternative splicing event and the most downstream event. We also calculated a score used to rank the genes with differential splicing. To calculate the score, we combined the most significant P value by Fisher's exact test from an event in the 5′ direction, P₁, with the most significant P value from an event in the 3′ direction, P₂. If we interpret these P values as the probability of a false positive (FP), then P₁ = P(5′ is FP) and P₂ = P(3′ is FP), and the probability that either event is a false positive, P_c, is then defined as P(5′ is FP ∩ 3′ is FP) = P(5′ is FP) + P(3′ is FP) − P(5′ is FP ∪ 3′ is FP) and P_c = P₁ + P₂ − P₁ × P₂. We calculated the final score for each gene as −log₁₀(P_c).

Mass spectrometry analysis.

Whole-cell lysates of MCF-7M cells were immunoprecipitated with an anti-LIN28 antibody (Abcam) as described previously (34). Normal serum IgG was used as a control for nonspecific interactions with antibodies and beads. Immunoprecipitated eluates were separated by 4 to 12% SDS-PAGE. After staining with Coomassie blue, each gel lane was cut into 24 unique bands, and bands were subjected to automatic tryptic digestion by using a Progest protein digestion station (Genomic Solutions, Ann Arbor, MI). Nanoscale liquid chromatography-electrospray ionization-tandem mass spectrometry (NanoLC-ESI-MS/MS) analyses were then performed by using an Agilent 1100 NanoLC system with an Agilent XCT Ultra ion trap mass spectrometer operating in the data-dependent acquisition (DDA) mode with the Chip Cube interface. Peptides were loaded onto an Agilent C₁₈ chip (75 μm by 43 mm) and eluted by applying a linear gradient from 5% acetonitrile and 0.1% formic acid to 50% acetonitrile and 0.1% formic acid to the column over 45 min. The mass spectrometer was used in the positive-ion, standard enhanced mode with included settings of a mass range from m/z 200 to 2,200, an ionization potential of 2.1 kV, a 100-ms accumulation time, and a 1.0-V fragmentation amplitude (31).

Acquired MS/MS spectra were extracted and searched against the species-limited (human and rodent) NCBInr database. Proteins qualitatively enriched in the LIN28 IP relative to the control IP were identified.

LIN28 expression and correlation in TCGA samples.

Breast cancer subtypes were derived from TCGA data reported previously (35) (https://tcga-data.nci.nih.gov/tcga/). Gene expression microarray data were downloaded from the data portal of TCGA for 529 breast cancer tumors and 61 matched normal samples. Expression values (log₂ locally weighted scatterplot smoothing [lowess] normalized) were extracted for a panel of genes of interest, including LIN28.

RNA sequencing data accession number.

The RIP-Seq and RNA-Seq data are available at the NCBI Gene Expression Omnibus (GEO) via accession number GSE71013.

RESULTS

Genome-wide identification of LIN28 mRNA targets in breast cancer cells.

Whole-genome microarray studies revealed that LIN28 was expressed in a derivative of the MCF-7 breast cancer cell line designated MCF-7M (32, 36). We first confirmed LIN28 mRNA and protein expression in MCF-7M cells relative to human ES cells (hESCs), which express LIN28 (Fig. 1). LIN28 expression in H9 and H1 ES cells is well established and was used as a positive control to show that LIN28 is detectable in MCF-7M cells (6, 37). LIN28 is expressed primarily during normal development but can be reactivated in several cancers, including breast cancer (8), although its function in breast cancer is not yet clear. The reactivation of LIN28 in the MCF-7M subclone provided a good model to study LIN28 function in breast cancer. To identify LIN28 mRNA targets in breast cancer cells, we performed RNA-protein immunoprecipitation (RIP) in MCF-7M cells with a LIN28-specific antibody and sequenced the RNA recovered from the immunocomplexes (RIP-Seq). We sequenced the RNAs from both the LIN28 RIP and normal serum IgG and obtained ∼13 million sequence reads each. Of these reads, 85% (∼12 million) unambiguously mapped to the human reference genome (hg19) for the LIN28 RIP, and 50% (∼7 million) unambiguously mapped to the human reference genome for the IgG RIP. Based on previously reported studies with similar experiments in human ES and cancer cells (5, 25, 38), we required the LIN28 RIP to have a 2.5-fold increase in normalized reads compared to those of the IgG RIP to identify an mRNA as being significantly bound by LIN28. We found that LIN28 was predominantly bound at coding exons and 3′ UTRs, 38% and 45%, respectively, in the 843 mRNAs significantly bound by LIN28 (Fig. 2A; see also Table S1 in the supplemental material). Next, we examined the prevalence of the recently described LIN28 GGAGA motif within the 843 mRNAs (26). We found that the GGAGA sequence was significantly more prevalent in LIN28-bound than in unbound targets (Fig. 2B). To determine whether LIN28 binds novel targets in MCF-7M cells, we examined the overlap between RIP-Seq targets in our study and those identified previously in human ES and ovarian cancer cells (5, 24, 25). Although it is challenging to directly compare results from different studies due to differences in IP protocols and cell types, our RIP-Seq identified 40% of the 276 mRNA targets derived from endogenous LIN28 non-cross-linking RIP-Chip in human ES cells (see Table S2 in the supplemental material) (5). Twenty (20%) of the combined 803 RIP-Chip targets from human ES and ovarian cancer cells were also identified in our study (5, 24, 25) (see Table S2 in the supplemental material). However, only 10% of the 1,803 LIN28 targets captured by photoactivatable ribonucleoside-enhanced cross-linking and immunoprecipitation (PAR-CLIP) of exogenous LIN28 in HEK293 cells (24) (see Table S2 in the supplemental material) were identified in our study. We identified novel mRNAs that were specifically enriched in MCF-7M cells, including S100A9, MGMT, LYPLAL1, and UQCRH, as well as common targets, such as CHMP2A and RBM3 (Fig. 2C). Further validation of a representative subset of genes by qRT-PCR analysis revealed that all of these targets were enriched at least 10-fold in LIN28 RIPs relative to normal serum IgG control RIPs (Fig. 2D). Notably, analysis of steady-state mRNA levels indicated that relative LIN28 binding was not correlated with relative mRNA abundance for these genes (Fig. 2E). Taken together, these results reveal novel LIN28 targets in MCF-7M breast cancer cells and also confirm that LIN28 binds common targets in multiple cell types (see Tables S1 and S2 in the supplemental material).

FIG 1.

LIN28 expression in breast cancer cells and hESCs. Total RNA and whole-cell extracts were isolated from two human breast cancer cell lines (lane 1, MCF-7M; lane 2, T47D) and two human embryonic stem cell lines (lane 3, H1; lane 4, H9). (A) LIN28 relative mRNA levels normalized against the GAPDH level. Values are means ± standard deviations (n = 3). (B) LIN28 protein expression. β-Actin was used as a loading control.

FIG 2.

LIN28 binds novel mRNAs in human breast cancer cells. (A) Genomic distribution of the 843 LIN28-bound genes. LIN28 is enriched predominantly in the coding exons (CDS) and 3′ UTRs of the target mRNAs. (B) GGAGA prevalence in LIN28-bound and unbound genes. The box plot shows the distribution of normalized GGAGA counts for LIN28-bound and unbound genes. The LIN28-bound genes are those that overlap the 843 RIP-Seq targets, and the unbound genes are those that do not overlap the RIP-Seq targets. Data are reported as normalized GGAGA counts per kilobase of exon; the P value was determined by a one-sided Wilcoxon rank sum test. (C) Read coverage of a subset of LIN28-bound targets. We normalized the coverage to reads per million mapped reads. These genes include only those identified in MCF-7M cells, including the S100 calcium-binding protein A9 (S100A9), lysophospholipase-like 1 (LYPLAL1), ubiquinol-cytochrome c reductase hinge protein (UQCRH), and O-6-methylguanine-DNA methyltransferase (MGMT) genes, and those identified in both MCF-7M and ES cells (5, 24), including the charged multivesicular body protein 2A (CHMP2A) and RNA-binding motif protein 3 (RBM3) genes. The fold enrichment for LIN28 IP compared to the control IP (CTL-IP) for each target is indicated. (D) LIN28 binding of a subset of targets. RNA protein complexes were isolated from cells by using an antibody against LIN28 or normal rabbit serum IgG (CTL), followed by RNA extraction and quantitative real-time PCR analysis. A subset of mRNAs identified by RIP-Seq was analyzed. Data for LIN28 are represented as fold enrichments of mRNAs present in the LIN28 IP relative to the control IP. The experiment was done in 2 biological replicates. Data from a representative experiment are shown. (E) mRNA expression of a subset of LIN28-bound targets. Total RNA was isolated, and the expression of the subset of genes described above for panel C was analyzed by qRT-PCR. Total relative mRNA levels were normalized against the RPL13A level. Values are means ± standard deviations (n = 3).

To explore potential mechanisms by which LIN28 regulates bound mRNAs, we explored functional pathways of genes significantly bound by LIN28. Gene ontology (GO) analysis using a suite of DAVID canonical pathways (KEGG and REACTOME) revealed that genes bound by LIN28 are associated with diverse functional categories. The top two most significant categories contained genes linked to the regulation of translation, including ribosomal proteins and translation initiation/elongation factors (Table 3), supporting a previously described role for LIN28 in mRNA translation (20, 21, 38). More importantly, the GO analysis revealed that 10 out of 15 of the most significant categories were associated with regulation of cellular energy metabolism, such as the electron transport chain, oxidative phosphorylation, and cellular respiration (Table 3). Additionally, LIN28 targets are also linked to RNA metabolism, including genes involved in RNA splicing and genes linked to the regulation of ubiquitin proteasome activities (Table 3).

TABLE 3.

Biological functions of LIN28-enriched targets^a

Category	Process	No. of genes	Fold enrichment	P value
GO:0006412	Translation	59	4.3	4.55E−21
GO:0006414	Translational elongation	29	6.9	4.66E−16
GO:0022900	Electron transport chain	28	5.9	1.20E−13
GO:0006091	Generation of precursor metabolites and energy	46	3.5	3.05E−13
GO:0006119	Oxidative phosphorylation	25	6.1	1.29E−12
GO:0006120	Mitochondrial electron transport, NADH to ubiquinone	17	9.7	4.34E−12
GO:0042775	Mitochondrial ATP synthesis-coupled electron transport	19	8.1	5.70E−12
GO:0042773	ATP synthesis-coupled electron transport	19	8.1	5.70E−12
GO:0045333	Cellular respiration	24	5.9	7.88E−12
GO:0022904	Respiratory electron transport chain	19	7.1	7.07E−11
GO:0015980	Energy derivation by oxidation of organic compounds	24	4.0	2.94E−08
GO:0031398	Positive regulation of protein ubiquitination	17	4.8	3.24E−07
GO:0006396	RNA processing	50	2.2	3.88E−07
GO:0055114	Oxidation reduction	55	2.1	6.15E−07
GO:0051443	Positive regulation of ubiquitin-protein ligase activity	15	5.1	9.52E−07

^aShown is a summary of the most enriched GO terms represented by enriched mRNAs. GO biological processes enriched in the 843 genes >2.5-fold for LIN28 binding compared to the control IP are shown. GO term classification analysis for biological processes was performed by using DAVID (GOTERM_BP_FAT). The minimum number of genes for each process was set to 5. The top 15 biological processes sorted by lowest P value are shown. LIN28-bound genes are associated predominantly with energy metabolism (in boldface type).

LIN28-interacting proteins are implicated in diverse gene regulatory functions.

In order to identify LIN28 protein complexes, we immunoprecipitated endogenous LIN28 from MCF-7M whole-cell extracts and analyzed the immunocomplexes by mass spectrophotometry. Seventy-two peptides copurifying with LIN28 were present at significant spectral counts in LIN28 complexes relative to the normal serum IgG control, and of these peptides, 20 had an MS score of >40 (Table 4; see also Table S3 in the supplemental material). These LIN28-interacting proteins are implicated in diverse gene regulatory functions, including mRNA metabolism, RNA binding (hnRNP A1, hnRNP A2/B1, PTB, PABPC4, SFPQ, and CAPRIN1), mRNA translation into protein (EEF1G, RNA helicase, and DDX3), transcriptional regulation (DDX3, Ku70, and DHX9), protein stability (USP28), and the DNA damage response (Ku70 and DHX9). LIN28 interactions, including those with hnRNP A1, DDX3, Ku70, and PABPC4, were verified by coimmunoprecipitation studies of endogenous proteins (Fig. 3A).

TABLE 4.

LIN28-associated proteins with an MS score cutoff of 40, including ubiquitous RNA-binding proteins^a

Protein	% coverage	No. of peptides	MS score	NCBI accession no.
Heat shock 70-kDa protein 1A variant	33	16	263.29	62089222
FLJ00343 protein	5	11	157.58	21748542
Ubiquitin-specific protease 28	10	10	148.93	16507200
Lamin A/C, isoform CRA_c	12	9	125.45	119573383
HSPA9	11	6	90.03	21040386
Heat shock protein 90-kDa alpha (cytosolic), class A member 1 isoform 1	8	6	78.47	153792590
Ku80	8	6	78.29	119590969
DDX3	8	5	76.97	62087546
CAPRIN1	10	6	73.46	42558250
hnRNP A1	14	5	65.66	133254
hnRNP A2/B1	16	5	65.41	157265559
Ku70	11	5	65.27	4503841
DHX9	5	5	64.94	100913206
hCG1748768, isoform CRA_b	11	4	59.27	119579661
lin-28 homolog (LIN28A)	25	4	52.13	13375938
Annexin A2	13	4	51.54	73909156
PTBP1	9	3	45.82	4506243
p80 protein	9	3	43.55	1483131
PABPC4	7	3	42.33	119627673
SFPQ	6	3	40.41	29881667

^aAn MS score of 40 was used as a cutoff for the top LIN28-associating proteins. LIN28 associates with proteins involved predominantly in RNA processing {hnRNP A1 (helix-destabilizing protein, single-stranded RNA-binding protein, and hnRNP core protein A1), hnRNP A2/B1, PTB, PABPC4 [poly(A)-binding protein, cytoplasmic 4 (inducible form), isoform CRA_h], SFPQ [splicing factor proline/glutamine-rich (polypyrimidine tract-binding protein associated)], and CAPRIN1 (membrane component chromosome 11 surface marker 1 isoform 1)} and DNA damage (Ku70 [ATP-dependent DNA helicase II, 70-kDa subunit]/Ku80 [X-ray repair complementing defective repair in Chinese hamster cells 5 {double-strand-break rejoining}], DDX3 [DEAD/H {Asp-Glu-Ala-Asp/His} box polypeptide 3 variant], and DHX9 [DEAH {Asp-Glu-Ala-His} box polypeptide 9]). PTBP1, polypyrimidine tract-binding protein 1 isoform a.

FIG 3.

RNA-dependent in vitro interactions and in vivo association of LIN28 and hnRNP A1. (A) Validation of LIN28-interacting proteins. Whole-cell lysates were immunoprecipitated with normal rabbit serum IgG (lane 2) or anti-LIN28 (lane 3). Western blotting to detect proteins copurifying with LIN28 was performed with antibodies against hnRNP A1, DDX3, PABPC4, and Ku70. Lane 1 represents 5% of the input. (B) LIN28 is nuclear and cytoplasmic in MCF-7M breast cancer cells. Immunofluorescence detection of LIN28 (red) and nuclei stained with DAPI (blue) are shown. DIC, differential interference contrast. (C) hnRNP A1 localization is predominantly nuclear. Immunofluorescence detection of hnRNP A1 (green) and nuclei stained with DAPI (blue) are shown. (D) LIN28 and hnRNP A1 interactions are predominantly nuclear. One milligram of cytoplasmic or nuclear extracts was immunoprecipitated with 2 μg of anti-hnRNP A1. The hnRNP A1/LIN28 cytoplasmic and nuclear immunocomplexes were detected with antibodies against LIN28 or hnRNP A1 (lanes 5 and 6). Normal serum IgG immunocomplexes were used as a negative control (lanes 3 and 4). Lanes 1 and 2 represent 5% inputs from the cytosol and nucleus, respectively. (E) Interaction of LIN28 with hnRNP A1 is RNA dependent. LIN28 immunoprecipitation was performed onwhole-cell extracts that were untreated (lane 4) or treated with 40 μg RNase A (lane 7). Normal serum IgG (lanes 3 and 6) was used as a nonspecific (NS) binding control. Protein immunocomplexes were detected by Western blotting with antibodies against LIN28 and hnRNP A1. Lane 1, input; lane 3, nonspecific IgG minus RNase A; lane 4, LIN28.

We next focused on the LIN28 interaction with hnRNP A1 because it is involved in many aspects of gene regulation, including pre-mRNA splicing, mRNA trafficking, mRNA stability, miRNA biogenesis, and transcription (39). Also, hnRNP A1 is a key regulator of cell metabolism in cancer cells (40–42), and our GO analysis revealed that the majority of LIN28 targets were involved in cellular energy metabolism (Table 3). In addition, three of the highest-ranked peptides interacting with LIN28, hnRNP A1, hnRNP A2/B1, and PTB, play diverse roles in mRNA metabolism and have been implicated in the regulation of cell metabolism in cancer cells (40–42). To explore the effects of hnRNP A1 on LIN28 gene regulatory functions, we first examined the subcellular localization of the two proteins in MCF-7M cells by immunofluorescence followed by confocal microscopy. We found distinct LIN28 staining in both the cytoplasm and nucleus using an antibody against the endogenous LIN28 protein (Fig. 3B). In contrast, we detected hnRNP A1 staining predominantly in the nucleus (Fig. 3C). Furthermore, subcellular fractionation and coimmunoprecipitation indicated that hnRNP A1 associates with LIN28 primarily in the nucleus (Fig. 3D, compare lanes 3 and 4 with lanes 5 and 6). The association of LIN28 with hnRNP A1 also requires RNA, since treatment with RNase A disrupted this interaction (Fig. 3E, compare lanes 4 and 7). Collectively, these results imply that the spatial distribution of LIN28 protein complexes may provide a level of regulatory control that could modulate LIN28 downstream targets.

Effect of LIN28 and hnRNP A1 on splicing.

The role for LIN28 in regulating nonnuclear processes such as mRNA translation into protein and miRNA processing is well characterized (5, 14, 21, 22). Because LIN28 interacts with hnRNP A1, we wondered if hnRNP A1 might modulate the effects of LIN28 on downstream targets by modulating pre-mRNA splicing given the well-characterized role of hnRNP A1 as a splicing regulatory protein (39, 43). Additionally, hnRNP A1 plays a major role in cell metabolism by regulating the alternative splicing of pyruvate kinase, an enzyme that controls the metabolic switch between oxidative phosphorylation and aerobic glycolysis (40–42). To initially test for a role of LIN28 in alternative splicing, we analyzed the expression levels of pyruvate kinase M (PKM), which is a known target of hnRNP A1, by RT-PCR (40). Depletion of LIN28 significantly decreased the expression of PKM1, an isoform of PKM, by 50% (P < 0.05), with minimal effects on the overall expression of PKM2, as shown by quantitative RT-PCR (Fig. 4A). This result was further confirmed by semiquantitative RT-PCR with a reverse primer spanning the terminal exon common to PKM1/PKM2 and forward primers specific for exon 9 (PKM1) and exon 10 (PKM2), respectively (Fig. 4B).

FIG 4.

Cells lacking LIN28 have decreased levels of the PKM1 isoform. LIN28 knockdown decreases PKM1 but not PKM2 expression. (A) Total RNAs from cells transfected with nontargeting, LIN28, or GAPDH siRNAs were analyzed by quantitative and semiquantitative RT-PCR for PKM expression by using primers specific for PKM1 (exon 9) and PKM2 (exon 10) (schematic at the top). LIN28 significantly decreases PKM1 but not PKM2 expression (*, P < 0.05). Data are shown as relative expression levels normalized to the RPL13A level. (B) Products from semiquantitative RT-PCRs were analyzed on a 1% agarose gel to confirm the qRT-PCR data. Shown is a schematic of a reverse primer spanning common terminal exon 12 and specific forward primers used to detect PKM1 (exon 9) and PKM2 (exon 10). Lane 1, DNA ladder; lane 2, positive control; lanes 3 to 5, NT siRNA (PCR cycles 10, 20, and 30, respectively); lanes 6 to 8, LIN28 siRNA (PCR cycles 10, 20, and 30, respectively). A PCR product generated from a plasmid construct with PKM1/PKM2 cDNA was used as a positive control (CTL) to determine the correct product sizes, 323 and 392 bp for PKM1 and PKM2, respectively. Actin was used as a loading control. No positive control was used for actin.

In order to determine whether LIN28 regulates alternative splicing and if these events were influenced by hnRNP A1, we performed RNA-Seq analysis in cells treated with siRNAs against the nontargeting (NT) control, LIN28, hnRNP A1, and a combination of LIN28 and hnRNP A1 (LIN28A1) for 72 h (Fig. 5). cDNA libraries prepared from poly(A)-selected RNA were subjected to Illumina sequencing. We obtained >200 million read pairs per siRNA target group, >90% of which were aligned to the human reference genome (hg19) (Table 5). Each of the four groups had two biological replicates. To identify genes with possible alternative splicing events in the RNA-Seq data, we developed a method that utilizes sequencing reads identified as spanning multiple exons, quantifies the junctions between these exons (splice junctions), and identifies genes with significant differences between these splice junctions (see Materials and Methods). After this method identifies genes with significant differential splicing, it assigns a score to rank the list of genes based on the degree of differential splicing. This method identified 111 genes with significant (FDR < 0.05) differential splicing for LIN28-depleted cells as well as 249 and 182 genes for hnRNP A1- and LIN28A1-depleted cells, respectively, compared to cells treated with the nontargeting siRNA control (Fig. 6A; see also Table S4 in the supplemental material). We next used CuffLinks to estimate FPKM values for the isoforms of the 247 genes identified as having differential splicing in any of the three comparisons, and we used these FPKM values to calculate isoform usage percentages for each sample (see Table S4 in the supplemental material). We then performed hierarchical clustering for the 8 samples using the isoform usage percentages (Fig. 6B). Intriguingly, samples from the hnRNP A1 knockdown and those from the double knockdown clustered more closely to each other than to the other two groups, suggesting that LIN28 may affect splicing independent of hnRNP A1. A recent independent study found that LIN28 binding to transcripts encoding splicing factors influenced downstream splicing events (26). Since MS analysis identified several splicing factors associated with LIN28, we examined LIN28 binding at these splicing factors. RIP-Seq data indicated that LIN28 was not enriched at the hnRNP A1 locus or at the locus of any other splicing factors (hnRNP A1/B2, PTBP1, and SFPQ) (Fig. 6C). These data suggested that despite the association between LIN28 and hnRNP A1, LIN28 may regulate splicing independent of hnRNP A1.

FIG 5.

Protein depletion upon treatment with the indicated siRNAs. Shown are data from Western blot analyses 72 h after transfection of cells with the indicated siRNAs against the NT control (lane 1), LIN28 (lane 2), hnRNP A1 (lane 3), and hnRNP A1 and LIN28 (lane 4). Whole-cell lysates were separated on an SDS-PAGE gel and probed with the indicated primary antibodies. β-Actin and GAPDH were used as loading controls. Equal loading was verified by Coomassie staining.

TABLE 5.

Raw and mapped sequence reads resulting from HiSeq sequencing for four siRNA groups

Sample	Read length × no. of lanes	No. of raw reads	No. of filtered reads	No. of aligned reads	% aligned reads
NT control	100 × 3	310,160,724	287,772,780	270,856,601	94.1
LIN28	100 × 3	270,733,798	251,896,068	235,943,994	93.7
hnRNP A1	100 × 3	319,792,236	296,983,868	278,584,039	93.8
LIN28 and hnRNP A1	100 × 3	310,311,962	288,194,084	269,644,095	93.6

FIG 6.

Depletion of LIN28 and hnRNP A1 proteins yields nonidentical effects on mRNA splicing. (A) Venn diagram showing overlap of the differentially spliced genes in the three siRNA comparisons against nontargeting controls. Shown are genes with significant (FDR < 0.05) differential splicing in LIN28-depleted (111 genes), hnRNP A1-depleted (249 genes), and LIN28A1-depleted (182 genes) cells compared to the nontargeting siRNA control. There are twice as many genes whose splicing is affected by hnRNP A1 and LIN28A1 (102 genes) compared to those affected by LIN28 and LIN28A1 (56 genes). (B) Heat map and hierarchical clustering of genes with significant splicing differences. The heat map was generated by hierarchical clustering of the isoform usage percentages of the 247 genes identified as having significant splicing differences between any siRNA group and the NT control by using the splice junction method and an FDR threshold of 0.05. Rows represent the isoform usage percentages for each isoform, and columns represent samples from the siRNA treatments. Isoform usage percentages are normalized such that the rows of the heat map have a mean of 0 and a standard deviation of 1. (C) mRNAs encoding hnRNP A1 and splicing factors do not show significant LIN28 binding. Shown are screenshots from the UCSC Genome Browser displaying normalized reads covering each nucleotide, per million mapped reads, of LIN28-interacting proteins. The scale on the left of each screenshot indicates normalized reads per kilobase per million mapped reads for each gene, including hnRNP A1; DEAD (Asp-Glu-Ala-Asp) box helicase 3, X-linked (DDX3); X-ray repair complementing defective repair in Chinese hamster (XRCC6 or Ku70); poly(A)-binding protein, cytoplasmic 4 (PABPC4); DEAH (Asp-Glu-Ala-His) box helicase 9 (DHX9); hnRNP A1/B2; polypyrimidine-tract binding protein 1 (PTBP1); splicing factor proline/glutamine-rich (SFPQ); and cell cycle-associated protein 1 (CAPRIN1). The fold enrichment for each target is indicated.

Next, we asked which biological processes were overrepresented by genes differentially spliced in cells lacking LIN28. Gene ontology enrichment analysis of these 111 genes revealed that receptor kinase signaling and actin/cytoskeleton processes were strongly overrepresented (P < 0.001) (Table 6). To our surprise, the Enabled Homolog (ENAH) gene, whose product is involved in a range of processes dependent on actin cytoskeleton remodeling and is downstream of epidermal growth factor receptor (EGFR) receptor kinase signaling (44), was the second-highest-ranking gene in terms of splicing among genes with differential splicing in cells lacking LIN28 (Fig. 7A; see also Table S4 in the supplemental material). ENAH expresses primarily two isoforms, one which includes exon 11a and the other which excludes it (Fig. 7B). Cells lacking LIN28 expressed the isoform excluding exon 11a ∼29% of the time, whereas NT cells expressed this isoform only ∼17% of the time (Fig. 7A), translating into ∼1.5-fold exon 11a skipping (Fig. 7C). We validated these results experimentally in cells lacking LIN28 by qRT-PCR using specific primers to detect exon 11a and demonstrated a significant decrease in the expression of exon 11a in the LIN28 knockdown compared to the NT control, whereas the overall expression of ENAH did not show a significant difference (Fig. 7D).

TABLE 6.

LIN28 knockdown affects splicing of genes associated with receptor kinase signaling and actin/cytoskeleton pathways^a

Category	Process	No. of genes	Fold enrichment	P value
GO:0007169	Transmembrane receptor protein tyrosine kinase signaling pathway	9	6.0	1.20E−04
GO:0030029	Actin filament-based process	9	5.6	1.99E−04
GO:0007167	Enzyme-linked receptor protein signaling pathway	9	3.9	0.001968089
GO:0030036	Actin cytoskeleton organization	7	4.6	0.003944803
GO:0043623	Cellular protein complex assembly	6	5.5	0.004490309
GO:0016192	Vesicle-mediated transport	11	2.8	0.004849464
GO:0034621	Cellular macromolecular complex subunit organization	8	3.3	0.009649918
GO:0006897	Endocytosis	6	4.1	0.015644011
GO:0010324	Membrane invagination	6	4.1	0.015644011
GO:0006164	Purine nucleotide biosynthetic process	5	5.0	0.017014248

^aShown is a summary of the most enriched GO terms represented by 111 genes undergoing significant splicing events in the LIN28 knockdown compared to the nontargeting siRNA control. The minimum number of genes for each process was set to 5. Biological processes are sorted by the lowest P value. LIN28 splicing targets are associated predominantly with the transmembrane receptor protein tyrosine kinase signaling pathway and actin filament-based processes (in boldface type).

FIG 7.

Depletion of LIN28A protein alters ENAH mRNA splicing. (A) Isoform usage percentages of a subset of genes in cells lacking LIN28. We estimated FPKM values of gene isoforms affected by LIN28 knockdown using CuffDiff and then calculated the isoform usage percentages from these FPKM values. Isoform usage percentages of two ENAH isoforms are shown relative to values of five representative genes whose splicing is affected by LIN28 knockdown. The ENAH gene showed the second most affected splicing in cells lacking LIN28. For a complete list of gene isoforms and isoform usage percentages, see Table S5 in the supplemental material. (B) Read coverage of ENAH in NT and LIN28 knockdowns. The chromosome positions of exons 11, 11a, and 12 are shown at the top. Read coverage for NT (blue) and LIN28 (red) knockdowns is shown in the middle. The number of reads with a given splice junction is shown at the bottom. (C) Sashimi plot for ENAH in NT and LIN28 knockdowns. The plot shows the read coverage of exons 11, 11a, and 12 along with the number of reads with splice junctions that span these exons. Depletion of LIN28 increases exon 11a skipping by ∼1.5-fold (cf. 1,063 versus 671 junction reads). (D) Expression of exon 11a. A diagram of the exons included in the two isoforms of ENAH is shown at the top. An aliquot of the total RNA sample submitted for RNA-Seq was analyzed by RT-PCR for ENAH expression in NT and LIN28 knockdowns and a GAPDH knockdown control using specific primers to detect exon 11a and exon 6. Values shown are mRNA expression levels relative to the value for TATA-binding protein (TBP).

Although these results show that LIN28 influences the splicing of some genes, we wondered whether LIN28 was required to be bound to these genes. There were only 12 genes with splicing affected by LIN28 knockdown and bound by LIN28 (see Table S5 in the supplemental material). This was not surprising since RIP predominantly detects protein interactions with mature mRNAs but not pre-mRNAs. For this reason, we exploited publically available LIN28 PAR-CLIP data, since this technique can detect protein interactions with nascent RNA (24). Surprisingly, 26 genes differentially spliced in LIN28-depleted cells were also bound by LIN28 in HEK293T cells expressing exogenous LIN28 (see Tables S2 and S5 in the supplemental material). Notably, the ENAH gene is among these 26 genes. Taken together, these findings are consistent with a role for LIN28 in pre-mRNA splicing, although our studies do not distinguish between direct and indirect mechanisms of LIN28 function.

Effect of LIN28 and hnRNP A1 on gene expression.

In order to determine whether LIN28 can regulate transcript abundance, we analyzed the RNA-Seq data to identify gene expression level changes between siRNA groups. We identified 142 genes that were differentially expressed (FDR < 0.05) between the LIN28 and the nontargeting control siRNA treatments as well as 209 and 281 differentially expressed genes for the hnRNP A1 knockdown and the double knockdown (LIN28A1), respectively (Fig. 8A; see also Table S6 in the supplemental material). This resulted in a total of 325 genes that were differentially expressed between any of the three siRNA groups and the NT control (see Table S6 in the supplemental material). We performed hierarchical clustering and principal component analysis (PCA) on the FPKM values for the 8 samples (Fig. 8B and C). These results support previous hierarchical clustering results using isoform usage percentages, in that the samples from the hnRNP A1 knockdown and the double knockdown clustered more closely with each other than with the other two groups. This suggests that hnRNP A1 may be the dominant regulator for the observed gene expression changes. Interestingly, ∼66% of the DEGs were upregulated, indicating that LIN28 and hnRNP A1 preferentially repress gene expression (Fig. 8D). A significant number of genes (81 genes) were similarly regulated upon the knockdown of the two proteins, and the majority of these genes (69) were upregulated. LIN28 suppression of gene expression is exerted primarily at the posttranscription level by Let-7 miRNAs binding to the complementary sites at the 3′ UTRs of the target genes (reviewed in reference 1). We asked whether LIN28 binding at the 3′ UTR was correlated with gene expression and found that there was no correlation between LIN28 binding at the 3′ UTR and gene expression (R² = 0.0004) (Fig. 8E). GO enrichment analyses of the 81 common genes revealed a link to regulation of the immune response, ISG15-protein conjugation, and antigen presentation (Table 7). A subset of some of the genes linked to the immune response was validated by qRT-PCR (Fig. 8F).

FIG 8.

Differential impacts of LIN28 and hnRNP A1 depletion on gene expression. (A) Venn diagram showing overlap of the differentially expressed genes in the three siRNA comparisons against the nontargeting control. Shown are genes with significant (FDR < 0.05) differential expression in LIN28-depleted (142 genes), hnRNP A1-depleted (209), and LIN28A1-depleted (281) cells compared to the nontargeting siRNA control. There is a large overlap of genes (206 genes) whose expression is affected by the depletion of LIN28 and hnRNP A1. (B) Heat map and hierarchical clustering of differentially expressed genes. The heat map was generated by hierarchical clustering of the FPKM values of 325 genes identified as being significantly differentially expressed between any siRNA group and the NT control by using CuffDiff and an FDR threshold of 0.05. Rows represent the relative expression levels of each transcript, and columns represent siRNA treatments. FPKM values are normalized such that the rows of the heat map have a mean of 0 and a standard deviation of 1. (C) Spatial separation in gene expression seen in hnRNP A1 and LIN28A1 knockdowns compared to LIN28 knockdown. Shown are data from PCA of the FPKM values of all 325 genes identified as being significantly differentially expressed between any siRNA group and the NT control by using CuffDiff and an FDR threshold of 0.05. Plotted are the principal component scores of the first 2 principal components for each sample. (D) Depletion of LIN28 or hnRNP A1 predominantly leads to upregulation of gene expression, as exemplified for immune response genes. Shown are total numbers of genes differentially expressed in each group compared to the nontargeting control siRNA. Over 60% of DEGs whose expression was changed by the knockdown of LIN28 or hnRNP A1 or the double knockdown are upregulated. (E) Plot comparing LIN28 binding locations to LIN28 KD expression changes. Each point corresponds to one of the 843 LIN28 targets. The x axis shows the gene's 3′-UTR score, which is the log₂ ratio of binding in the 3′ UTR compared to that in the CDS. A positive score indicates more binding in the 3′ UTR than in the CDS, whereas a negative score indicates less binding in the 3′ UTR than in the CDS. The y axis shows the log₂ fold change values for LIN28 KD compared to the NT control as determined by RNA-Seq. There is no correlation between the two (R² = 0.0004). (F) Depletion of LIN28 and hnRNP A1 results in upregulation of immune response genes. (Left) Total RNA was analyzed by qRT-PCR. Levels of all mRNAs are shown as relative mRNA expression levels normalized to the TBP expression level as a control. Results shown are means and standard deviations from three PCRs performed on the same cDNA sample. The results are from a single experiment that is representative of at least two independent experiments. (Right) UCSC Genome Browser view illustrating RNA sequence coverage depths for the nontargeting siRNA control (NT), the LIN28 knockdown, the hnRNP A1 knockdown (A1), and the double knockdown (LIN28A1). The peaks mark expressed exons consistent with the RefSeq annotation for each gene, shown as blue boxes. Normalized read counts (FPKM) for each siRNA are labeled on the y axis. CH25H, cholesterol 25-hydroxylase; IFI44L, interferon-induced protein 44-like; RSAD2, radical S-adenosylmethionine domain-containing 2; KIF12, kinesin family member 12; PARM1, prostate androgen-regulated mucin-like protein 1; MAN1A1, mannosidase alpha, class 1A, member 1.

TABLE 7.

Functional analysis of common DEGs^a

Category	Process	No. of genes	Fold enrichment	P value
GO:0006955	Immune response	22	7.3	3.02E−13
GO:0009615	Response to virus	9	18.9	1.88E−08
GO:0032020	ISG15-protein conjugation	3	137.6	1.79E−04
GO:0006952	Defense response	11	4.1	2.54E−04
GO:0002474	Antigen processing and presentation of peptide antigen via MHC class I	3	40.5	0.002357426
GO:0045087	Innate immune response	5	8.3	0.002871728
GO:0019882	Antigen processing and presentation	4	11.1	0.005390323
GO:0048002	Antigen processing and presentation of peptide antigen	3	24.6	0.006357181
GO:0045672	Positive regulation of osteoclast differentiation	2	76.4	0.025454921
GO:0006508	Proteolysis	10	2.2	0.034339297

^aGO analysis revealed significant enrichments of common DEGs in pathways and processes related to the immune response (boldface). MHC, major histocompatibility complex.

Correlation of LIN28 with breast cancer subtypes.

To further place these observations in a biological context, we examined whether LIN28 expression levels were correlated with breast cancer subtypes. We extracted LIN28 expression data obtained from four breast cancer subtypes, basal-like, HER2 enriched, luminal A, and luminal B, from TCGA publically available data sets. LIN28 is highly expressed in several subsets of tumors that carry poor prognoses for several types of cancer, such as breast and ovarian cancers (8, 25). The ENAH gene is overexpressed in primary breast tumors and is associated with poor prognosis for HER2 breast cancers (44, 45). Pairwise comparisons between each specific breast cancer subtype and matched normal samples showed that LIN28 expression is significantly different (P < 0.05) for the HER2 subtype (Fig. 9). Collectively, these analyses support a link between LIN28 and the HER2 subtype.

FIG 9.

LIN28 targets are correlated with breast cancer subtypes. Gene expression microarray data for 529 breast cancer tumors and 61 matched normal samples were downloaded from the data portal of TCGA. Expression values were extracted for LIN28. For comparisons by subtype, tumor samples were categorized according to breast cancer subtypes as assigned by the Cancer Genome Atlas Network. (Top) Expression of LIN28 is significantly different in the HER2 subtype, as indicated by comparison of each breast cancer subtype against matched normal samples by a t test. The median and mean values for LIN28 expression are also reported. N/A, not applicable. (Bottom) LIN28 expression per subtype visualized as a violin plot. The median expression level per subtype is shown by the connecting red line.

DISCUSSION

LIN28 is an evolutionarily conserved RNA-binding protein that is expressed primarily in normal development but is reactivated in several malignancies, including breast cancer (8, 9, 46). While extensively interrogated with respect to its role and functions in model organisms and during early development, the molecular functions of LIN28 in breast cancer are largely unexplored (2, 4). The findings described here suggest that LIN28 may affect multiple gene regulatory steps to modulate the expression of a broad spectrum of downstream targets involved in promoting biological processes that drive breast cancer.

LIN28 interacts with novel mRNAs in breast cancer cells.

We observed that LIN28 associates with ∼3.5% of mRNAs in the human genome in breast cancer cells. Previous studies have used RIP and various genome-wide technologies to identify LIN28 targets in a number of cellular systems (5, 24–26, 38). A large number (40%) of the targets identified in MCF-7M cells were also targets in hESCs and ovarian cancer cells (5, 25). Furthermore, the GGAGA sequence corresponding to the recently identified LIN28 motif (26) was largely prevalent within LIN28-bound mRNAs detected in our study, suggesting that when reactivated in breast tumor cells, LIN28 interacts with biologically relevant mRNAs also found in hESCs. Intriguingly, among the common targets in ES and breast cancer cells are genes linked to energy metabolism, consistent with the notion that ES and tumor cells use similar mechanisms to efficiently generate energy metabolites that support increased cell proliferation and cell mass (47). While 40% of RIP targets in ES cells were identified in our RIP-Seq data set, >87% of the mRNA targets are novel LIN28 targets in breast cancer cells. Some of the novel targets identified include S100A9, a potent proinflammatory mediator normally upregulated in the luminal subtype of breast cancer (48); O-6-methylguanine-DNA methyltransferase (MGMT), a protein crucial for maintaining genome stability (reviewed in reference 49); LYPLAL1, a potent triacylglycerol lipase involved in obesity (50); and UQCRH, an important component of complex III in oxidative phosphorylation (51) (Fig. 2).

LIN28 and hnRNP A1 have distinct effects on alternative splicing.

LIN28 associates with hnRNP A1, a multifunctional RNA-binding protein that functions in crucial aspects of RNA processing, including pre-mRNA splicing and mRNA export, localization, and stability (52, 53). We observed LIN28 association with hnRNP A1 predominantly in the nucleus, indicative of potential cooperation to regulate nuclear processes, including pre-mRNA splicing (Fig. 3). However, analyses of our RNA-Seq data by using a custom computational method to identify the inclusion or exclusion of specific exons showed that cells lacking LIN28 or hnRNP A1 express distinct transcript isoforms, despite their interactions (Fig. 6). This discordant observation is corroborated by the effects of LIN28 knockdown on the expression of PKM, a known target of hnRNP A1 (40, 42). Depletion of LIN28 decreases PKM1 (exon 9) mRNA expression, with no significant change in PKM2 (exon 10) expression (Fig. 4). In contrast to LIN28, hnRNP A1 is known to inhibit PKM1 expression by repressing exon 9 inclusion while increasing exon 10 inclusion and at same time increasing PKM2 expression (40, 42). Based on previously reported observations, one would expect a reciprocal increase in PKM2 expression upon LIN28 knockdown, but we detected only a slight increase in PKM2 expression, and the change was not statistically significant (Fig. 4A) (40, 42). It is possible that our RT-PCR assay was not sensitive enough to detect a significant change for PKM2 due to its ubiquitous expression. Additionally, the regulation of the PKM2 locus in tumor cells and malignancy is complex, and the relationship between PKM1 and PKM2 expression in tumor cells is not always direct and as such can be confounded by cellular factors that affect the expression of either isoform independent of the other (54). However, we observed a clear change in PKM1 expression in LIN28-depleted cells (Fig. 4A).

There is strong evidence for a distinct role of LIN28 in the regulation of splicing from our stringent computational and statistical analyses of RNA-Seq data (Fig. 6). We were able to identify a significant switch in the expression of human ENAH gene isoforms in cells lacking LIN28. Our data indicate that depletion of LIN28 significantly decreases the expression of transcripts (GenBank accession number NM_001008493) of the ENAH gene containing an additional exon 11a (Fig. 7) (55, 56). Although the detected changes in isoform expression are small (12%) (Fig. 7A), ENAH isoform switching may still convey a biological impact. It was recently shown that alternative splicing and expression of specific ENAH isoforms confer distinct malignant phenotypes in an isogenic model of human breast cancer progression (56). Consequently, a small change in expression that modulates the biological activity of the protein could result in distinct biological processes. This is particularly intriguing given that ENAH isoform 11a is subject to EGFR-mediated phosphorylation, which increases its cell proliferation and mitogenic activity (55).

The exact mechanisms of how LIN28 may regulate alternative splicing remain to be further explored. However, using our custom computational method to monitor alternative splicing events, we detected exon inclusion and exclusion within specific transcript isoforms and confirmed the corresponding changes in steady-state mRNA expression levels of the affected exons using qRT-PCR. Recently, an independent study demonstrated that binding of LIN28 to regions within transcripts that encode splicing factors resulted in increases in the protein levels of these factors and subsequent widespread changes in alternative splicing patterns in somatic cells expressing exogenous LIN28 (26). In contrast, we detected significant alternative splicing events in LIN28-depleted cells despite not finding significant LIN28 binding within transcripts encoding splicing factors in our breast cancer model (Fig. 6C). Additionally, we did not detect any significant overlap between LIN28 binding and transcript isoform expression (see Table S5 in the supplemental material). These observations suggest that LIN28 may modulate alternative splicing events by other mechanisms in addition to binding to splicing factor transcripts to dictate splicing factor abundance. This concept is also supported by the lack of concordance between LIN28 binding and alternative splicing events observed in ES cells (26). We note that we cannot rule out the possibility that our in vivo RIP-Seq method does not detect tertiary LIN28 mRNA interactions, as observed for cross-linking and immunoprecipitation coupled with high-throughput sequencing (CLIP-Seq) or PAR-CLIP (24, 26). We suggest that detection of LIN28 in the nucleus may suggest another, yet-to-be-determined role for LIN28 in pre-mRNA splicing (Fig. 3 and 10) (1).

FIG 10.

Proposed LIN28 gene regulatory mechanisms and link to biological processes associated with breast cancer. LIN28 interacts with proteins implicated in diverse gene regulatory functions, including hnRNP A1, which is involved in mRNA metabolism. (A) LIN28 modulates gene splicing. The interaction of LIN28 with hnRNP A1 is predominantly nuclear and requires RNA. The two proteins may interact with the same mRNA, with potentially different splicing outcomes: (i) the interaction of LIN28 with hnRNP A1 may result in exon exclusion (orange), whereas (ii) the interaction of LIN28 with an unknown protein results in exon inclusion (green), as in the case of ENAH. The ability of LIN28 to modulate splicing permits the generation of protein isoforms with different biological activities (e.g., kinase receptor signaling) that may contribute to the development of a specific breast cancer subtype (e.g., HER2). (B) Beyond regulating splicing, LIN28, independently or in conjunction with hnRNP A1, can influence mRNA steady-state levels. Data from knockdown experiments suggest that LIN28 represses anti-inflammatory cytokines, linking LIN28 to the immune response (see the text). (C) As an RNA-binding protein, LIN28 interacts with novel mRNAs to mainly influence their translation into protein (reviewed in reference 1). Collectively, the LIN28 downstream gene targets identified in a model breast cancer cell line mediate biological processes that are key hallmarks of cancer.

LIN28 and hnRNP A1 cooperate to regulate mRNA abundances of immune response genes.

In addition to modulation of splicing, we also found that cells lacking LIN28 have increased levels of anti-inflammatory cytokines (Fig. 8). The regulation of the majority of these genes was also dependent on the expression of hnRNP A1 (Fig. 8A and F). A direct role for LIN28 in the regulation of steady-state mRNA has not been extensively described (25, 26). LIN28-mediated regulation of gene expression is attributed largely to posttranscriptional mechanisms, through the posttranscriptional repression of target genes by Let-7 miRNAs or the translational regulation of LIN28-bound mRNAs (reviewed in reference 1). Posttranscriptional regulation of anti-inflammatory cytokines is largely through trans-acting RNA-binding proteins and cis-acting sequence elements, primarily at 3′ UTRs, comprising miRNA targets (57). A previous study showed that LIN28B, but not LIN28A, inhibited the expression of the anti-inflammatory cytokine interleukin-6 (IL-6) via Let-7 mechanisms (58). This observation suggests that the increased levels of anti-inflammatory cytokines observed in our study are independent of Let-7 mechanisms. In support of this concept, we found that LIN28 3′-UTR binding is not correlated with gene repression in our data set (Fig. 8E). Whereas LIN28-mediated regulation of gene expression is attributed largely to posttranscriptional mechanisms, the expression of LIN28 in the nucleus may provide an additional layer of LIN28 gene regulatory mechanisms (1).

LIN28 modulates splicing and expression of genes implicated in breast cancer biology.

Regulation of alternative splicing is an important mechanism required for the maintenance of cellular homeostasis, and aberrant regulation of alternative splicing is associated with pathological conditions and cancer progression (59). Gene Ontology analyses showed that cells lacking LIN28 express gene isoforms that encode proteins linked predominantly to cell membrane organization and cell-to-cell communication (Table 6). Specifically, we found that the expression of human ENAH gene isoforms is significantly affected in cells lacking LIN28. The ENAH gene, also known as the human MENA gene, is a member of the family of actin cytoskeleton regulators overexpressed in primary breast tumors (44, 55). Alternative splicing of human MENA generates isoforms with distinct roles in breast cancer biology (56, 60). Human MENA11a is overexpressed in primary tumors, and breast tumors expressing the additional exon 11a are subject to EGFR signaling, leading to increased mitogenic activity of the HER2 subtype of breast cancer (55, 61). In contrast, splice variants lacking exon 11a, and in many cases exon 6, have been shown to drive breast cancer metastasis (56).

The expressions of both LIN28 and ENAH have been independently shown to promote malignancy of HER2 breast cancers, but the molecular mechanisms underlying these effects have not been linked (28, 46, 61). Our study provides evidence suggesting that LIN28 modulates the splicing and expression of human ENAH isoforms that are critical for HER2 signaling in breast cancer cells. Together, these findings lead us to propose that LIN28 downstream alternative splicing events can have an impact on the specific breast cancer subtype and prognosis. Analyses of publicly available array data from TCGA support this concept in that LIN28 expression in the HER2 subtype is significantly different from that in other breast cancer subtypes (Fig. 9).

Well-established “hallmarks of cancer” include sustaining proliferative signaling, evading growth suppressors, resisting cell death, enabling replicative immortality by creating genome instability, inducing angiogenesis, activating invasion and metastasis, reprogramming energy metabolism, and evading immune destruction (62). The LIN28 downstream gene targets identified in this study mediate a number of these processes. A large majority of LIN28-associated mRNAs encode proteins that regulate cell energy metabolism (Table 3). LIN28 alternative splicing targets mediate cell-to-cell communication and signaling cascades that sustain the cell proliferative capacity and may activate invasion and metastasis (Fig. 6 and 7 and Table 6). Finally, gene expression analyses of cells depleted of LIN28 showed increased expression levels of many anti-inflammatory cytokines (Fig. 8 and Table 7). Taken together, these data suggest that LIN28 uses diverse gene regulatory mechanisms and functions as a master regulator of gene networks that modulate specific hallmarks of cancer (Fig. 10).