The HLA-G positive choriocarcinoma cell line JEG3, the human embryonal kidney cell line HEK293T, the breast cancer (BC) cell lines MCF-7 and HCC1806, the renal cell carcinoma (RCC) cell line MZ2905RC, and the colorectal carcinoma (CRC) cell lines HCT-116 and HT-29 were purchased from the American Type Culture Collection. With the exception of HEK293T cells, which were maintained in Dulbecco's Modified Eagle Medium (Invitrogen), all other cell lines were cultured in Roswell Park Memorial Institute 1640 Medium (Invitrogen) supplemented with 10% (v/v) fetal bovine serum (PAA laboratories), 0.1 mM non-essential amino acids (Gibco), 2 mM L-glutamine (Lonza), and 1% penicillin/streptomycin (v/v, PAA laboratories) and incubated at 37°C, 5% CO₂.

For miRNA transfection, cells were seeded at 1×10⁵ cells/well in 12-well plates, and after 16 hours, cells were transfected with miRNA mimics or controls (Sigma), respectively, at a final amount of 10 pmol using Lipofectamine RNAi MAX (Invitrogen) according to the manufacturer’s protocol. For increasing the effect of miRNAs, mimics were transfected under the same conditions after 48 hours. Cells were harvested 24 or 48 hours after transfection and either directly subjected to analysis or frozen down until use. A similar protocol was used to transfect miRNA inhibitors or controls (Sigma) at a final amount of 50 pmol per well.

For the luc reporter assay, 1×10⁴ HEK293T cells/well were seeded in a 96-well plate. After 16 hours, 10 ng of the reporter plasmid (pmirGLO Dual-Luciferase miRNA Target Expression Vector, Promega), in combination with respective miRNA mimic or control (Sigma) at a final concentration of 25 nM, was transfected using Lipofectamine 2000 reagent (Invitrogen) according to the manufacturers’ protocol. Luc activity was determined 48 hours after transfection using the Dual-Glo Luciferase substrate (Promega) according to the manufacturer’s protocol. Luminescence was detected by a Tecan Infinite 200 Pro plate reader device. Relative luc activity was determined by normalizing the firefly reporter to renilla luc activity.

Total RNA was isolated from cells using the NucleoSpin miRNA kit (Macherey & Nagel) according to the manufacturers’ instructions, followed by DNase I treatment (NEB). RNA quality and quantity were assessed by spectrophotometric analysis, and 500 ng of total RNA was used for cDNA synthesis (RevertAid H Minus First Strand cDNA synthesis kit, Fermentas). For miRNA-specific cDNA synthesis, a miRNA-specific stem-loop primer was used,22 while reverse transcription of mRNAs required a random hexamer primer (Fermentas). Expression levels were analyzed by quantitative reverse transcription PCR using GoTaq qPCR Master Mix (Promega). For all primer pairs, an annealing temperature of 61°C was used. Relative changes of mRNA/miRNA amounts were determined by the ΔΔC_t method23 using glyceraldehyde-3-phosphatedehydrogenase (GAPDH) or U6 RNA for normalization. All primers are listed in online supplementary table 1.

Total protein was extracted from 2×10⁵ cells using Radioimmunoprecipitation assay buffer (25 mM Tris–HCl pH 7.6, 150 mM NaCl, 1% NP-40, 1% sodium deoxycholate, 0.1% sodium dodecyl sulfate), and protein concentration was determined employing the Pierce BCA Protein Assay Kit (ThermoFisher).

For western blot analysis, 25 µg of protein/sample was separated by sodium dodecyl sulfate–polyacrylamide gel electrophoresis and transferred onto a nitrocellulose membrane by semidry blot. For detection of proteins, the following antibodies were used: anti-HLA-G antibody MEM-G/9 (Novus Biologicals), anti-Argonaute-2 antibody C34C6 (CST), anti-maltose-binding protein (MBP) antibody ab9084 (Abcam), anti-glyceraldehyde-3-phosphatedehydrogenase antibody 14C10 (CST), and anti-β-actin antibody ab8227 (Abcam). As secondary antibody, a horseradish peroxidase conjugated goat anti-α-mouse/rabbit antibody (CST) was used. Chemiluminescent blots were imaged by LAS-3000 Imaging System (Fuji). For densitometric analysis of western blot signal intensity, ImageJ software24 was used.

For flow cytometric analyses, the following antibodies were employed: the phycoerythrin-labeled anti-HLA-G monoclonal antibody (mAb) MEM-G/9 (ThermoFisher), the anti-HLA-ABC mAb W6/32 (eBioscience) and the respective isotype controls (Beckman Coulter, Krefeld, Germany). Cells were stained with mAbs at concentrations recommended by the manufacturers as recently described. The measurements were performed with a BD LSR Fortessa unit (Becton Dickinson). FlowJo software (FlowJo, LLC) was used for analyses and to display flow cytometry results. The results are presented as mean fluorescence intensity and were normalized relative to the respective mock control.

Prior to the actinomycin D assay, cells were transfected with respective miRNA mimics as described previously. RNA was isolated at various time points after treatment with actinomycin D (10 µg/mL, Merck). HLA-G mRNA levels were analyzed by qRT-PCR using ALAS1 for normalization.

To identify potential miRNAs targeting the CDS of HLA-G, the miTRAP method25 was employed, which was recently published.26 Briefly, CDS of HLA-G (accession number NM_002127) lacking the start ATG was cloned upstream of two MS2 loops. T7 promoters based in vitro transcripts (T7 Ribomax, Promega) were purified using the MEGAclear Transcription Clean-Up Kit (Invitrogen). By application of 100 pmol of fusion protein consisting of the MS2 loop and MBP domains, in vitro transcribed RNAs (HLA-G CDS and control sequence encoding the two MS2 loops only) were immobilized on amylose resin (NEB). After washing and blocking steps, RNA was incubated with the cytoplasmic cell lysate of the RCC cell line MZ2905RC (HLA-G mRNA+/protein−) enabling the specific binding of miRNAs to the sequence. Extensive washing reduced unspecific binding, and elution was carried out with 10 mM maltose solution followed by phenol/chloroform RNA extraction. RNA eluates were either used for cDNA synthesis of candidate or control miRNAs or applied for small RNA sequencing (small RNA-seq).

Small RNA-seq of miTRAP eluates was carried out at Novogene (Hong Kong). In brief, sequencing libraries were generated using NEBNext Multiplex Small RNA Library Prep Set for Illumina (NEB, USA) following manufacturer’s recommendations. Subsequent PCR amplifications were performed using the LongAmp Taq 2X Master Mix, SR Primer for Illumina and index primer (NEB, USA). PCR products were purified on an 8% polyacrylamide gel, and DNA fragments corresponding to ~140 to 160 were recovered. Library quality control was assessed on an Agilent Bioanalyzer 2100 system using DNA High Sensitivity Chips (Agilent, USA). Clustering of the index-coded samples was performed on a cBot Cluster Generation System using TruSeq SR Cluster Kit v.3-cBot-HS (Illumina, USA) according to the manufacturer’s instructions. The library preparations were sequenced on an Illumina Hiseq 2500/2000 platform and 50 bp single-end reads were generated. Small RNA tags were mapped to the miRBase V.20.0 provided reference27 by Bowtie (bowtie-0.12.9, main parameter: -v 0 k 1)28 without mismatch to analyze their expression and distribution on the reference. Expression levels were estimated by TPM (transcript per million) through the following criteria29: normalized expression=mapped readcount/total reads×1 000 000.

A summary of the miRNA read counts obtained for the HLA-G CDS and MS2 control, as well as a list with all tools used for RNA-seq analysis (with their main parameter) is provided in online supplementary data 1. Further RNA-seq characteristics are provided in online supplementary table 2.

For miRNA binding prediction, different algorithms, including miRDB,30 miRWalk,31 miRcode,32 TargetScan,33 and RNA2234 with their default parameters were used. For correlation of miRNA and mRNA expressions, the online tool CancerMiner (last access December 2018)35 provided relevant data. The identification of the miR-16 and miR-744 binding site in HLA molecules was derived from the RNA22 software by using the following criteria: sensitivity of 63%, specificity of 61%; seed size of 7, allow a maximum one unpaired bases in seed; minimum number of paired-up bases in heteroduplex=12; maximum folding energy for heteroduplex (kcal/mol)=−12; maximum number of Guanine:Uracil wobbles allowed in the seed region=no limit.

Prism software (GraphPad) was used for calculating mean, SD or t-test as well as displaying results. For the two-sided t-test unequal variances have been selected. The data were significant with a p value<0.05. The number of biological replicates is given as n.

We and others have recently shown that HLA-G is post-transcriptionally regulated by miRNAs (eg, miR-148/miR-152 family) via binding to its 3′-UTR.19–21 For a more comprehensive understanding of the miRNA-mediated regulation of HLA-G, the interaction of miRNAs with the HLA-G CDS was determined. For using an advanced MS2-tethering approach called miRNA trapping by RNA in vitro affinity purification (miTRAP),25 the HLA-G CDS (accession number NM_002127) was inserted upstream of two MS2 repeats enabling the in vitro transcription of HLA-G CDS fused to two MS2 loops (figure 1A). Due to the rather long size of the RNA (1165 nt) and to assess conceivable steric interference, binding properties to the MS2 loop binding protein were analyzed and ~20% of the bait RNA immobilized to the resin was found (online supplementary figure 1).

To analyze a potential CDS-dependent post-transcriptional regulation of HLA-G by miRNAs, the co-precipitated RNA and proteins were purified after RNA affinity purification. As shown in figure 1B, an enriched binding of the RISC component Argonaute 2 to the HLA-G CDS indicated an association with miRNAs. Accessible Argonaute 2-CLIP data further support the binding to the CDS of HLA-G.31 36 Interestingly, increased levels of the co-precipitated Argonaute 2 protein was found in the CDS miTRAP eluate compared with the 3’-UTR of HLA-G suggesting an intense miRNA-dependent regulation via the CDS of HLA-G (online supplementary figure 1). In a first attempt, miRDB software tool30 was used to predict high ranked miRNAs potentially targeting the HLA-G CDS, since it provides a target score allowing the ranking of miRNA candidates. The miRNAs miR-4749–3 p (score 77, rank 1), miR-1915–5 p (score 68, rank 4) and miR-654–5 p (score 67, rank 5) were selected for further analyzes due to their high rank and the fact to show detectable expression in the chosen cell line for miTRAP analysis. However, an enhanced binding of these miRNAs in the miTRAP eluate was not detected, indicating a more complex miRNA target selection. Therefore, copurification of miRNAs was assessed by high-throughput small RNA-seq. Libraries were generated from the HLA-G CDS and MS2 control miTRAP eluates. Length distribution of sequenced reads showed an accumulation of 21–24 nt RNAs, indicating an miRNA enrichment in miTRAP eluates (online supplementary figure S2 and online supplementary table S1).

After evaluation of the normalized read counts (TPM) for the HLA-G CDS and MS2 control, highly stringent criteria for the enrichment of an miRNA like (1) TPM observed in the HLA-G miTRAP eluate >2500; (2) TPM observed in the control miTRAP eluate <300; and (3) enrichment ratio of >50 were applied; finally, (4) a specific binding site for the respective miRNA candidates within the HLA-G CDS needed to be predicted by at least two out of four applied algorithms to ensure a potential interaction of miRNA candidate and HLA-G.

This selection resulted in a total of six miRNAs enriched in the HLA-G miTRAP eluate (table 1). Noteworthy, the miRNAs miR-15a/b, miR-16 and miR-424 can be classified as an extended miR-16 family37 which share the same seed sequence and most likely compete for the same binding site. In accordance with our preliminary analysis, the miRNAs miR-4749–3 p, miR-1915–5 p and miR-654–5 p were not found to be enriched in the HLA-G CDS eluate compared with the control (online supplementary data 1).

For validation of RNA-seq results, the miTRAP analysis was repeated and co-purification of candidate miRNAs was determined by quantitative reverse transcription PCR. In order to assess enrichment, the miRNA abundance in the miTRAP eluate fraction normalized to the input (miTRAP ratio)25 was investigated demonstrating an accumulation of the indicated miRNAs in the HLA-G eluate compared with the MS2 control. This enrichment was accompanied by a high ratio of precipitation relative to the input. In contrast, selected control miRNAs showed significantly lower miTRAP ratios (figure 2A). The selective co-purification of the candidate miRNAs indicated a potential binding to the CDS of HLA-G.

To analyze putative effects of the candidate miRNAs on HLA-G expression, HLA-G positive JEG3 cells were transiently transfected with respective miRNA mimics, before the HLA-G surface expression was evaluated after 24 or 48 hours, respectively, using flow cytometry (figure 2B). Due to the high sequence similarity of miR-15a and miR-15b, miR-15b was excluded from further analysis.

Neither the candidate miRNAs nor the positive control miR-148a (targeting HLA-G 3′-UTR) had a significant impact on HLA-G expression 24 hours post miRNA mimic transfection confirming reports of high MHC class I protein stability.38 However, reduced HLA-G surface expression was observed for miR-148a and the candidate miR-744 after 48 hours of transfection. As expected from previous reports describing lower effects of CDS-located binding sites,7 the impact for miR-744 was weaker compared with miR-148a. Surprisingly, 48 hours post-transfection, the extended miR-16 family members exhibiting the same seed sequence important for target mRNA binding showed a significant upregulation of HLA-G on the surface of JEG3 cells, while overexpression of miR-660 did not significantly modulate HLA-G surface expression on JEG3 cells. Based on these results, the further analyses focused on miR-744, whose overexpression resulted in the expected downregulation of HLA-G, and on miR-16 as one member of the extended miR-16 family, whose overexpression revealed increased HLA-G levels (figure 2C).

Following the first screening in JEG3 cells, miR-16 and miR-744 were subjected to additional investigations in order to analyze the effects of their overexpression in more detail. Due to the relatively low transfection efficiency of JEG3 cells (online supplementary figure 3) and in order to analyze putative dose dependent effects, transfection efficacy was increased by two subsequent miRNA transfections (0 and 48 hours) followed by flow cytometric analysis after 96 hours. Comparable but enhanced effects for miR-16 and miR-744 in deregulating HLA-G were observed (figure 3A), pointing to a direct connection of miRNA expression and the modulation of HLA-G. For further validation of the observed effects, JEG3 cells were transiently transfected with miRNA inhibitors.39 These molecules suppress the specific miRNA activity and resulted in contrary effects on HLA-G surface expression for miR-744 compared with the miRNA overexpression. In contrast, the miR-16 inhibitor transfection resulted in a non-significant downregulation of HLA-G (figure 3B). Next, to exclude indirect/secondary effects responsible for deregulation of HLA-G on the cell surface, total HLA-G protein was analyzed by western blot (figure 3). Overexpression of miRNA candidates affected total HLA-G protein in a similar way as shown by flow cytometry, confirming our previous assays. In sum, miR-744 targets HLA-G via the CDS resulting in reduced HLA-G protein levels, while miR-16 appears to be a positive regulator of HLA-G expression in JEG3 cells.

The miRNA prediction tool RNA2234 allowed localization of the binding site of miR-16 and miR-744 in the CDS of HLA-G. In addition, sequence alignment revealed that miR-16 and miR-744 binding sites are highly conserved in both non-classical (HLA-E and HLA-G) and classical MHC class I antigens (HLA-A, HLA-B, and HLA-C; figure 4A). Therefore, it was investigated whether miR-16 and miR-744 were also capable to regulate related HLA-G genes. HLA-E expressing JEG3 cells were transfected with miR-16 or miR-744 and subsequently analyzed for HLA-E protein levels. However, a deregulated HLA-E expression could not be observed (online supplementary figure 4). To test miR-16 and miR-744-dependent effects on classical MHC class I antigens, HEK293T cells mainly expressing HLA-A40 were employed for transfection. miR-148a, known to affect HLA-C17 has a similar binding site for HLA-A and thus served as a positive control. Indeed, a significant downregulation of HLA-ABC on the surface of HEK293T cells was detected for miR-744 and miR-148a using an anti-HLA-ABC mAb (figure 4B). In contrast, miR-16 upregulated HLA class I surface expression, while transfection of a miR-16 inhibitor resulted in the opposite effect and downregulated HLA-ABC. It is noteworthy that the miR-744 inhibitor showed no deregulation most likely due to very low miR-744 levels in HEK293T cells (figure 4B). To further validate the effects of miR-16 and miR-744 on classical MHC class I genes, in silico analysis from TCGA data sets using the tool CancerMiner was performed,35 which determined the expression relationships between miRNAs and mRNAs in different cancer types. In this context, values<0 point to a negative correlation, while data>0 indicate a positive correlation. The correlation of miR-16 and miR-744 with MHC class I mRNA in a pan-cancer set composed of 11 different cancer types with 3.290 samples supported our findings in JEG3 and HEK293T cells for HLA-G and HLA-ABC, respectively. Moreover, this correlation was particularly significant in breast and colon adenocarcinoma (online supplementary table 3). Based on these data, the effect of miR-16 and miR-744 mimics was determined in two BC (MCF-7, HCC1806) and two CRC cell lines (HCT-116, HT-29; figure 4D). A significant downregulation of HLA-ABC after transfection of miR-744 (except MCF-7) and miR-148a (except HT-29) as a positive control was found in the different cell lines. In contrast, a general upregulation of HLA-ABC was detectable in the BC and CRC cell lines on transfection of miR-16.

Finally, it was investigated how miR-16 and miR-744 regulate MHC class I genes. The overexpression of miR-744 resulted in a slight non-significant downregulation of HLA-G and HLA-ABC, suggesting mRNA degradation as the mode of action (figure 5A). To investigate whether the silencing of HLA-G occurred through a direct interaction and to determine the exact binding site for miR-744 in the CDS of HLA-G, luc gene reporter assays were performed as shown in figure 5B. The relative luc activity was reduced when compared with the mock control. As expected, mutation of the predicted miR-744 binding site in the CDS of HLA-G abolished this effect. A reduced relative luc activity was also observed for HLA-A and -B, but not for HLA-C.

Concerning miR-16, the transcript levels of HLA-G and HLA-ABC were upregulated after overexpression indicating either a stabilization of the mRNA or an enhanced transcription (figure 5A). To analyze if miR-16 overexpression and thereby enhanced binding to the CDS of HLA-G blocks a potential interaction with a general negative regulator (eg, RNA-binding protein (RBP) with destabilizing function), a recombinant HLA-G expressing cell line was established. In this model system, HLA-G is under the control of a CMV promoter and the transcript lacks the original 3′-UTR. Transient overexpression of miR-16 had no effect on total and cell surface expression of HLA-G, whereas miR-744 serving as a control still affected HLA-G (figure 5C,D). However, it should be noted that the CMV promoter construct resulted in a strong HLA-G expression potentially masking miR-16-based induction of HLA-G. An indirect effect stabilizing the HLA-G transcript by miR-16 was excluded by actinomycin D mRNA stability assay (online supplementary figure 5a). Additionally, a gene reporter assay using miR-16 failed to show an upregulation in luc activity (online supplementary figure 5b). These observations led to the assumption that the genetic context is important for the miR-16-directed upregulation of HLA-G. Recently, Catalanotto and coworkers suggested a mechanism for miRNA-dependent activation of gene expression, which involves a nuclear localized RISC complex and the interaction with the respective promoter as well as additional transcriptional activators to induce target gene expression.41 To prove whether miR-16 is able to localize to the nucleus, cytoplasmic and nuclear cell extracts were prepared from JEG3 cells. Despite miR-16 being found predominantly in the cytoplasm of JEG3 cells, a considerable proportion of miR-16 was also detected in the nucleus (figure 5E), which confirmed results from a previous report in HCT116 colon cancer cells.42 As shown in figure 5, ChIP assays revealed an enriched occurrence of the RISC core component Argonaute 2 at the proximal HLA-G promoter (figure 5F), which could further point to the existence of a miRNA-dependent gene activation mechanism as proposed by Catalanotto et al .41