Introduction

Currently, the best assay to determine the tumorigenic potential of tumor cells involves xenotransplantation of different sub-populations of cancer cells into highly immunosuppressive animals.1, 2 This approach has been used successfully for a functional in vivo identification of tumor suppressors in several tumor types.3, 4 Taking advantage of a pooled, lentiviral short hairpin RNA (shRNA) library, we performed a large-scale in vivo RNA interference (RNAi) screen in mice to find genes that act as tumor suppressors in human breast cancer cells. As a part of this study, SALL1 was identified, which is one of the four human family members of the spalt family. Members of the spalt family are highly conserved zinc-finger transcription factors, present from Caenorhabditis elegans to vertebrates, with regulatory functions in organogenesis, limb formation and cell-fate assignment during neural development.5 Mutations in the human SALL1 gene have been associated with Townes–Brocks syndrome, a rare, dominantly inherited disorder, characterized by limb, ear, anal and renal abnormalities.6 In mice, Sall1 acts as a regulator of canonical Wnt signaling during kidney development.7 Furthermore, it has been found that, in mouse embryonic stem cells, Sall1 physically interacts with Nanog and Sox2 and it has further been suggested to be a novel component of stemness.8 Although the importance of Sall1 during mouse embryogenesis has long been recognized, the first data suggesting the involvement of its human ortholog in carcinogenesis have only been published recently. It has been shown that SALL1 silencing via promoter hypermethylation is associated with human breast cancer.9 A functional role of SALL1 in breast cancer, however, had not been described before. Interestingly, our data show that SALL1 expression correlates with the expression of the gene CDH1, a key factor during epithelial-to-mesenchymal transition (EMT).10 EMT is a tightly controlled developmental mechanism involved in processes including embryogenesis, tissue repair and wound healing.11 Its induction leads to the loss of polarized epithelial and the acquisition of mesenchymal motile cell phenotypes.11 During carcinogenesis, EMT facilitates cancer cell migration and metastasis.12 Moreover, induction of EMT in cancer cells promotes their tumorigenicity,13 and cancer cells that have undergone EMT gain cancer stem cell properties.14 The finding that SALL1 expression is correlative with the expression of CDH1 is consistent with its tumor suppressive function and suggests its potential involvement in EMT.

Results and Discussion

The triple negative human breast cancer cell line SUM-149 was transduced with the pooled, lentiviral Decipher library module 1.15 Transduction conditions were adjusted to ensure a maximum of one integration event per target cell. The library used consists of 27 494 shRNA expression constructs, targeting 5045 genes for knockdown by five to six dissimilar shRNA sequences each. Upon infection, each construct integrates into the genome of the host cells, expressing the relevant small interfering RNA continuously. For a pooled RNAi screen with all constructs, the number of cells that express a particular shRNA is critical to ensure statistical significance. As not every cancer cell has the potential to form a tumor, we conducted a preliminary experiment to determine the fraction of tumorigenic SUM-149 cells. For this purpose, 6000 cells each were injected subcutaneously into eight NOD SCID mice. It was found that in six out of eight cases, tumors developed within 25 weeks post injection. From this, it could be concluded that 48 000 cells harbored a minimum of 6 cells capable of initiating a tumor. On the basis of this estimation, a total of 109 cells were transduced with the Decipher library. As a result, 36 000 cells, of which at least 4 cells should have a tumor initiating capacity, expressed each shRNA in the library. Although this figure is too low to allow negative selection screening, it does ensure that each shRNA is represented sufficiently in the pool of tumor cells to identify shRNAs that increase their tumorigenic potential. Following transduction, 107 cells each were injected subcutaneously into the flanks of 50 NOD SCID mice. The scheme shown in Figure 1a illustrates the overall workflow. Tumor development to a diameter of 1.5 cm took on average 57±12 days with a take rate of 0.86. Tumors were divided into two groups and analyzed separately via next-generation sequencing of barcode sequences, which are unique for each particular shRNA expression construct in the library. Consequently, barcode sequences could be used to identify each shRNA expression construct and thus the number of cells containing it. Barcode read counts from tumors were divided by baseline read counts from cells 3 days post transduction. Finally, z-scores—the distance of the value from the mean in s.d. units—were computed for each shRNA expression construct (Supplementary Table S1). Correlation between biological replicates was found to be r=0.66. The z-scores of all shRNA expression constructs in the library are shown in Figure 1b.

Figure 1
figure 1

Barcoded in vivo RNAi screen with shRNA constructs. (a) Scheme of the process: cells were transduced with Decipher library module 1 (Cellecta Inc., Mountain View, CA, USA). Into each flank of 50 NOD SCID mice, 107 stably transduced SUM-149 cells suspended in 200 μl PBS/Matrigel (BD Biosciences, Billerica, MA, USA) (1:1, v/v) were injected subcutaneously. Following tumor formation, tumors were homogenized and genomic DNA was isolated using the DNeasy Blood and Tissue Kit (Qiagen, Hilden, Germany). Amplification of barcode sequences was achieved by two rounds of PCR according to a protocol described in detail elsewhere.35 The amplified sequences were purified by means of PCR purification and gel extraction kits (#28104 and #28704, Qiagen). Then, barcode representation was quantified using next-generation sequencing on GAIIx machines (Illumina Inc., San Diego, CA, USA). Barcodes associated with shRNAs that promote tumor formation became enriched (illustrated in red). (b) Ranking of the z-scores from each barcode in the library. The dashed line indicates the cutoff defined by the twofold s.d. of the average z-score of the 21 negative control shRNAs. The z-scores from five shRNA expression constructs targeting SALL1 are shown in orange. Genes represented by at least two shRNAs with z-scores above the cutoff were considered candidate genes and are listed in Table 1. (c) Residual mRNA levels following 5 days of expression of the two shRNAs targeting SALL1 in SUM-149 cells. Reverse transcription of RNA and PCR was performed in one step using the Quantifast SYBR Green RT-PCR Kit (Qiagen) on a LightCycler 480 system (Roche, Basel, Switzerland). The QuantiTect Primer Assay (Qiagen) was used for specific target gene amplification. (d) Tumor-free survival of NOD SCID gamma mice following orthotopic injection of 40 000 SUM-149 cells with reduced SALL1 expression (shSALL1-1/ -2) relative to control (shCTRL). Significance values from Kaplan–Meier plots were calculated by means of the Wilcoxon test, using GraphPad Prism software (Graphpad, La Jolla, CA, USA). (e) Anti-bromodeoxyuridine (BrdU) antibody staining from tumors with inhibited SALL1 (shSALL1) or control tumor (shCTRL), respectively. Black boxes indicate enlarged areas shown below. For BrdU staining, mice were injected with BrdU 2 h before tumor collection, and samples were prepared according to the protocol from the Cell Proliferation Kit (G&E Healthcare, Wauwatosa, WI, USA). (f) Quantification of six stainings such as shown in e.

Full size image

The Decipher library contains 21 negative control shRNAs targeting the expression of the gene luciferase. The twofold s.d. of the z-scores of these shRNAs was used to define the cutoff for candidate shRNAs. The cutoff was 2.24 and is indicated by the dashed line in Figure 1b. To reduce the risk of off-target effects, genes were considered candidates only when at least two targeting shRNAs showed a z-score above this value. In total, 16 genes were found to meet these criteria (Table 1). To exclude that these candidates drove tumor growth only because of an increased proliferation in vitro, we cultured cells transduced with the shRNA library for 14 days and determined the construct pool representation before and after the culture period. By means of comparison, the proliferative nature of each gene was determined. Knockdown of 12 candidate genes had no significant impact on proliferation, whereas the inhibition of the genes ANP32E, MUTYH, GDF6 and RAD54L even led to reduced proliferation (Table 1).

Table 1 Candidate tumor suppressor genes identified by means of in vivo RNAi screen
Full size table

Two of the shRNA molecules identified in this screen targeted the expression of SALL1. This gene was chosen for subsequent analyses because it belongs to the group of zinc-finger transcription factors16 and has been implicated in breast cancer recently.9 The two shRNAs were sub-cloned individually into the expression vector pRSI9, yielding the constructs shSALL1-1 and shSALL1-2. SUM-149 cells were transduced with each expression construct. The residual target messenger RNA (mRNA) levels were found to be reduced 5 days post transduction (Figure 1c). Cells were then injected into the mammary fat pad of NOD SCID gamma mice (40 000 cells per animal). The time between injection and tumor onset was recorded. The inhibition of SALL1 by either shRNA resulted in significantly decreased tumor-free survival periods (Figure 1d). Measurement of tumor growth kinetics showed clear differences. Although these results further support a tumor suppressive role for SALL1, they were not significant. This can be explained by the large s.d. resulting from the variation in time to tumor onset as well as the biological variation between individual animals. Other shRNA sequences targeting the candidate genes EMR3 and GPRC5D were validated in the same way and yielded similar results (Supplementary Figure S1), indicating the accuracy of the initial screen results. Next, tumors that had formed from cells with inhibited SALL1 expression were compared with control tumors via immunohistochemistry. Figure 1e shows sections from representative tumors chased with bromodeoxyuridine. Tumors with inhibited SALL1 expression is more frequently stained positive for bromodeoxyuridine (Figure 1f), indicating a larger fraction of mitotically active cells.

SALL1 is known to act as a transcriptional repressor in non-cancer cells.16 To identify genes whose expression levels were changed following SALL1 inhibition, a microarray expression profile was performed. SUM-149 cells with inhibited SALL1 expression that was induced by the constructs shSALL1-1 or shSALL1-2, respectively, were compared to control cells (Supplementary Table S2). In total, 200 genes were identified to be significantly (P<10−10) up- or downregulated following the expression of either shRNA. These genes, together with the observed expression-changes caused by both shRNAs, are summarized in Supplementary Table S3. Interestingly, the inhibition of SALL1 was found to occur jointly with altered expression of several important factors involved in EMT, including CDH1 (Thiery et al.17), CDH2 (Kalluri and Weinberg18), VIM18 and MSN.19, 20 Relevant variations found during the microarray analysis were validated by means of quantitative RT–PCR (Figure 2a). In SUM-149 cells, expression of each shRNA targeting SALL1 led to reduced SALL1 and CDH1 expression and increased CDH2, VIM and MSN expression, which is a typical expression signature for mesenchymal cells. In addition, SALL1 inhibition was confirmed to correlate with increased expression of the two oncogenic CCN family members CTGF and CYR61, both of which are known to have a critical role in breast carcinogenesis.21, 22 Furthermore, mRNA levels of two putative tumor suppressor genes, retinoic acid receptor responders RARRES1 (Jing et al.23) and RARRES3 (Hsu et al.24), were confirmed to be reduced following SALL1 inhibition. Moreover, we analyzed protein levels of two EMT markers, E-cadherin and vimentin, and found reduced E-cadherin and increased vimentin levels following SALL1 knockdown (Figure 2b or c).

Figure 2
figure 2

SALL1 expression correlates with CDH1 levels in breast cancer cell lines. (a) Fold-changes of mRNA levels of indicated genes following 5 days of expression of two shRNAs targeting SALL1 in SUM-149 cells relative to control cells. (b) SUM-149 cells transduced with shSALL1-1, shSALL1-2 or a control construct were stained with E-cadherin-PE (Miltenyi Biotec, Bergisch Gladbach, Germany) and analyzed using flow cytometry. (c) Protein levels were determined by means of western blotting. Membranes were probed with antibodies against E-cadherin (Abgent, San Diego, CA, USA), vimentin (Cell Signaling, Danvers, MA, USA) and GAPDH (Abcam, Cambridge, UK), detected using peroxidase-conjugated antibodies (Sigma-Aldrich, St Louis, MO, USA) and ECL (Thermo, Waltham, MA, USA). (d) Reduction of the mRNA levels of CDH1 and SALL1 following 5 days of expression of the two shRNAs targeting SALL1 in the indicated breast cancer cell lines SKBR3, MCF-7, MDA-MB-231 and SUM-159. All measurements were relative to control cells with a non-inhibiting shRNA construct.

Full size image

To investigate the effects of SALL1 inhibition in different genetic backgrounds, SALL1 expression (Supplementary Figure S1c) and residual target mRNA levels following SALL1 inhibition were determined via quantitative RT–PCR in the basal breast carcinoma cell lines SUM-159 and MDA-MB-231, as well as in luminal MCF-7 and SKBR3 cells. Similar to SUM-149 cells, the inhibition of SALL1 by constructs shSALL1-1 or shSALL1-2 led to reduced CDH1 levels in all investigated breast cancer cell lines (Figure 2d). Given the diverse genetic background of these cell lines, this effect of SALL1 inhibition appears to be independent of genomic aberrations.

Furthermore, the impact of SALL1 inhibition on the migratory phenotype of SUM-149 was investigated. To this end, wound-healing experiments using cells transduced with the constructs shSALL1-1, shSALL1-2 and a control were performed. As shown in Figure 3a, inhibition of SALL1 increased the migratory potential of SUM-149.

Figure 3
figure 3

Impact of SALL1 knockdown on cell migration and CSC population. (a) Effects of SALL1 inhibition on the cell migration of SUM-149 cells. Transduced cells were scratched to create a gap of 700 μm and washed twice with PBS. Directly after creation of the gap and again after 24 h, pictures were taken, and the gap size was analyzed using ImageJ software (http://rsb.info.nih.gov/ij/). Quantifications from 24 such wound-healing assays are shown on the right. (b) Flow cytometry analysis of CD44, CD24 and EpCAM expression of cells, whose SALL1 expression was modified by transduction with shSALL1-1 and shSALL1-2 or a control construct (CTRL). SUM-149 cells were stained with CD44-PE/Cy7, CD24-FITC, EpCAM-APC (Becton Dickinson, Franklin Lakes, NJ, USA).

Full size image

Mani et al.14 have previously demonstrated a link between EMT and an increase in breast cancer cells with stem-like properties such as CD44+/CD24−/low. Moreover, Gupta et al.25 identified a distinct sub-population of SUM-149 cells with cancer stem cell-like properties displaying the surface molecule signature CD44+/CD24−/low/EpCAM−/low. To determine whether inhibition of SALL1 influences the percentage of cells expressing a cancer stem cell signature, CD44/CD24/EpCAM expression levels were compared between knockdown and control cells using flow cytometry.26 SUM-149 cells transduced with shSALL1-1 and shSALL1-2 exhibited more than twice as many CD44+/CD24−/low/EpCAM−/low cells in comparison to the control population (Figure 3b).

Analysis of SALL1 expression in breast cancer patient samples was consistent with a tumor suppressive role of SALL1. The analysis of two independent patient datasets27, 28 revealed that high SALL1 mRNA levels were associated with a significantly increased relapse-free survival (P=0.00048), overall survival (P=0.0027), metastasis-free survival (P=0.0071) as well as tumor-free survival (P=0.011) (Figure 4a or b).

Figure 4
figure 4

Association of SALL1 expression with patient survival and breast cancer subtypes. Relapse-free and overall survival, as well as the metastasis-free and tumor-free fractions of breast cancer patients with high or low SALL1 expression levels are shown. A Cox proportional hazard regression model was used for a univariate survival analysis of the gene expression data sets of (a) Pawitan et al.27 and (b) Loi et al.28 The samples were sorted according to the expression of SALL1, then all cutoffs in the central 60% of samples were assessed for correlation with outcomes. The split with the lowest P-value was chosen. (c) The Cancer Genome Atlas (TCGA) data set was used to analyze the mRNA expression levels of SALL1 across breast cancer tumor subtypes (downloaded in August 2012 at https://tcga-data.nci.nih.gov/tcga/). One-way ANOVA was used to compare the level of SALL1 mRNA expression among the different breast cancer subtypes (luminal (LumA and LumB), basal and HER2) and normal-like tissue.

Full size image

On the basis of gene expression profiles, breast cancer can be grouped into several distinct subtypes.29 The basal-like subtype contains an increased percentage of CD44+/CD24− and ALDH1+ cancer stem cells.30 A subtype-specific analysis of SALL1 expression levels revealed that mRNA levels were significantly lower (P<2 × 10−9) in the most aggressive basal-like subtype29 when compared with other breast cancer subtypes (Figure 4c), again consistent with the association with patient outcome.

Understanding the molecular mechanisms that drive carcinogenesis is essential to understand the development of human tumors. Tumor suppressor genes are known to have an important role in this process because their loss of function enhances the tumorigenic potential of cancer cells. Here, we conducted a large-scale in vivo RNAi screen aimed at the identification of novel breast tumor suppressor genes. Our results attribute a tumor suppressive role to the transcriptional repressor SALL1 in the background of the triple negative, tumorigenic breast cancer cell line SUM-149. Although in this study, the effects of SALL1 inhibition on non-tumorigenic cells, such as HMLER, were not determined, it was clearly shown that SALL1 inhibition leads to a more aggressive phenotype in SUM-149 cells.

SALL1 has an important role during embryonic kidney development and affects Wnt/beta-catenin signaling,7 a pathway commonly activated during EMT.18 Furthermore, other studies have demonstrated methylation of the SALL1 promoter in breast tumors, colorectal cancer, non-small-cell lung carcinoma and acute lymphocytic leukemia.9 Also, it has been shown that the region 16q12.1, in which SALL1 is located, often gets deleted in breast tumors.31, 32 Both observations, inactivation through methylation and loss of heterozygosity of SALL1 in different cancers, are supportive of its tumor suppressive function.

In line with these findings, we found that SALL1 expression is associated with the expression levels of a number of genes involved in EMT, namely CDH1, CDH2, VIM and MSN. Furthermore, SALL1 inhibition led to reduced expression of CDH1 in four additional breast cancer cell lines. It has previously been found that inhibition of CDH1 is sufficient to induce human epithelial breast cells to undergo EMT,33, 34 emphasizing the crucial role CDH1 has during this process. Moreover, inhibition of SALL1 led to reduced protein levels of E-cadherin, encoded by CDH1, whereas vimentin levels increased; both changes are typical for EMT.33 In this respect, it is important to note that SALL1 has been shown to inhibit the induction of Goosecoid, a repressor of CDH1 (Thiery et al.17), during embryoid body differentiation.8 This provides one possible route by which SALL1 might regulate the expression of CDH1. Furthermore, it is known that induction of EMT leads to an increase of cells with a CD44+/CD24− phenotype.14 After SALL1 knockdown, we detected a significant increase in the fraction of CD44+/CD24−.

It is important to mention that the data presented here does not provide evidence for a causative role of SALL1 in EMT. It does, however, clearly show that there is a correlative link between the expression of SALL1 and that of CDH1 in five different human breast cancer cell lines. Also, increased in vitro invasiveness and expression of a cancer stem cell marker signature following SALL1 inhibition are phenotypes frequently associated with EMT, although not exclusively. Their occurrence might thus be correlative with reduced SALL1 levels, but have no impact on the tumorigenic potential of cells or EMT. Hence, whether or not SALL1 is actually involved in the regulation of the highly complex process of EMT is yet to be firmly demonstrated.

In summary, we performed a large-scale in vivo RNAi screen, which led to the identification of SALL1 as a novel tumor suppressor gene in human breast cancer cells. We show that SALL1 expression is associated with the expression of a central regulator of EMT, namely CDH1, and that SALL1 expression correlates with the survival of breast cancer patients. These findings depict SALL1 as an exciting new tumor suppressor warranting closer investigation, especially regarding its potential involvement in EMT, as well as in breast cancer development.

Animal studies

The animal studies were approved by the local ethics committee at the Regierungspräsidium Karlsruhe (G74/11, G244/11).