Introduction

High-throughput experimental approaches are reasonable ways to attribute a biological consequence to its relevant genes. Among them, multiplex RNAi is a robust genetic approach to perform loss-of-function study of multiple target genes. It operates by delivering multiple gene-specific siRNAs into cells or expressing multiple shRNAs from a multicistronic DNA vector (Silva et al, 2008). However, conventional multiplex RNAi assay can only observe the overall phenotype resulted from silencing of all of the target genes, while the contributions of each individual gene or some of them cannot be inferred from the assay. Therefore, development of an RNAi tool that can silence a couple of genes both individually and combinatorially in a single experiment is crucial in an effort to dissect the hierarchical aspects of multiple gene networks, the contributing role of components in a signaling pathway, the functional interdependence of protein complex components, or the functional redundancy or divergence of a gene family.

A multiplex RNAi needs multiple gene-specific siRNAs or shRNAs in a cell. These excessive siRNAs/shRNAs might affect cell homeostasis because the most considerable limitation of conventional shRNA or siRNA is their interference with endogenous microRNA biogenesis by out-competing endogenous pre-microRNA for RISC loading (Snove & Rossi, 2006; Khan et al, 2009; Martin et al, 2011; van Gestel et al, 2014). So, siRNA or conventional shRNA might not be qualified to run a multiplex RNAi. In contrast, microRNA-based shRNA (shRNAmir) might be a good choice (Zeng et al, 2002; Chung et al, 2006). These artificial shRNAmirs are processed sequentially by the endonucleases Drosha and Dicer into precisely cleaved siRNAs, similar to the processing of natural microRNAs, thus minimally perturbing cell physiology (Silva et al, 2005). Because of their precise cleavage, shRNAmirs also reduce the drawback of off target-effects of conventional shRNA due to inaccurate processing (Gu et al, 2012). Moreover, the use of shRNAmirs can avoid the immune response triggered by regular shRNA expression (Bauer et al, 2009). Thanks to the booming field of bioinformatics, two state-of-the-art algorithms, namely, SplashRNA (Pelossof et al, 2017) and shERWOOD (Knott et al, 2014), were recently developed, which could predict shRNAmirs that effectively and potently silence target genes.

Currently, the most successfully developed and commercialized shRNAmir is based on the human miR-30 pri-microRNA backbone, which is further experimentally modified as miR-E, a variant that shows an enhanced knockdown efficiency. Thus far, a major limitation of miR-30-based RNAi is its tedious shRNAmir production, not as simple as conventional shRNA which only needs synthesis of two DNA oligos. Owing to their pri-microRNA nature, miR-30- or miR-E-based shRNAmir cloning relies on PCR (Zuber et al, 2011; Fellmann et al, 2013; Zhang et al, 2020) or Gibson assembly (Rousseaux et al, 2018; Michael et al, 2019) for de novo synthesis of a long DNA sequence (including the terminal loop, upper stem, lower stem, and part of the single-stranded flanking sequence), which is inserted into the miR-30 or miR-E backbone using the endogenous XhoI and EcoRI restriction sites. However, gene-specific shRNAmirs vary only in the upper stem region, suggesting that a redundant sequence (lower stem + flanking sequence) is included in the current miR-30-based cloning strategy.

In this study, a new shRNAmir backbone, named miR-AB, was developed to simplify the production of shRNAmirs. miR-AB, in combination of shRNAmirs designed by SplashRNA or shERWOOD, showed remarkable RNAi effects both in vitro and in vivo. miR-AB was engineered into a multipromoter viral toolkit carrying eight fluorescent proteins as reporters, which set up a multicolor-barcoded RNAi assay to efficiently and reliably silence multiple genes both individually and combinatorially at a single-cell level.

Results and Discussion De novo cloning of shRNAmir into miR-AB is simple, inexpensive, and error-proof

First, we sought to create a new strategy to simplify the cloning of shRNAmir, which can overcome the shortcoming of current shRNAmir cloning (Fig 1). As mentioned above, cloning of the requisite sequence only (without lower stem flanking sequence) might be the easiest way to accomplish this goal. Since both Drosha (Han et al, 2006) and Dicer (Gu et al, 2012) cleavages are critically controlled by the shRNA structure but not by its sequence, modification of the shRNAmir backbone by introduction of new restriction sites to the lower stem region may be an efficient strategy. As the Drosha cleavage site is ~11 bp away from the junction of the lower stem and single-stranded flanking sequence, the forced introduction of new restriction sites will elongate the lower stem and affect the precise cleavage of Drosha. Therefore, replacement of the endogenous sequence of the lower stem with new restriction sites may be a reasonable strategy.

Figure 1. miR-AB structure and cloning strategy

Sequence and structural comparison of miR-E and miR-AB. The structures were predicted by CLC Main Workbench. The 22-bp sequences of passenger/sense (black) and guide/antisense (green) strands are highlighted in large font. The endogenous XhoI and EcoRI restriction sites used for miR-E cloning and the introduced BamHI and ApaI sites for miR-AB cloning are colored as indicated. The dashed rectangles indicate the sequences de novo generated by PCR or Gibson assembly for miR-E, or by oligo synthesis with desalt purification for miR-AB. miR-AB shRNAmir cloning strategy. A shRNAmir sequence obtained from SplashRNA or shERWOOD platform are converted to two oligos (75 and 67 bp, respectively) by Microsoft Excel-based miR-AB Oligo Tool. These two oligos are synthesized, annealed, and ligated to BamHI/ApaI-cut miR-AB viral vector for transfection into 293T cells to produce miR-AB virus. The details of the procedure are described in the Materials and Methods section. miR-AB cloning efficiency. Various amount of BamHI/ApaI-cut miR-AB vectors were ligated with miR-AB oligos (keep the ratio of vector mass to oligo mole at 1:1; for example, 1 ng of vector with 1 nmol of oligo) and colonies were counted after transformation of XL-10 Gold competent cells (homemade, 2 × 109 transformation efficiency). The four colored lines represent four individual shRNAmir cloning efficiencies from two independent experiments (two individual shRNAmir cloning in each experiment). miR-AB cloning reliability. One colony was picked up from each miR-AB cloning plate for sequencing in five independent experiments using retroviral or lentiviral miR-AB vectors. The positive clones were calculated.

Type IIS restriction enzyme, cleaving DNA outside of its recognition site, is an alternative for cloning of a DNA if appropriate type IIP restriction sites are not available. A previous study demonstrated that BsmBI, a type IIS enzyme, can be used to perform a PCR-independent cloning of shRNAmir (Adams et al, 2017). However, this approach needs phosphorylation of oligos. Moreover, BsmBI is much more expensive than most commonly used type IIP enzymes. Furthermore, type IIS restriction enzymes were reported to show cleavage distance variation or slippage (Lundin et al, 2015; Arakawa, 2016), which might result in altered cleavage site leading to low cloning efficiency. So, we tried to introduce the widely used type IIP sites to replace the endogenous sequence. To maximally remove the redundant sequences in de novo cloning of gene-specific shRNAmir, we focused on the lower stem region sequence immediately close to the upper stem (target gene specific) in the miR-E backbone, screened the available restriction sites that did not conflict with the widely used lentiviral and retroviral systems, and found that ApaI and BamHI sites were good candidates to perform replacement of the endogenous sequence without affecting its spatial structure. After the introduction of these two restriction sites, this new miR-E variant was named miR-AB (Fig 1A).

Owing to the existence of these two unique restriction sites, shRNAmir cloning into miR-AB was simple, as observed in the cloning of a conventional shRNA, wherein two short oligos are synthesized with desalt purification, annealed to generate a DNA duplex with ApaI and BamHI overhangs, and ligated into the ApaI/BamHI-cut miR-AB lentiviral or retroviral vector (Fig 1B). To simplify the design of DNA oligos used for miR-AB cloning, and to efficiently utilize the shRNAmirs designed by the SplashRNA or shERWOOD algorithm, the miR-AB Oligos Tool, a Microsoft Excel-based application, was created to convert the 97-bp gene-specific shRNA sequence obtained from the SplashRNA or shERWOOD platform to two oligos for de novo synthesis. In order to prevent the output error caused by input error, miR-AB Oligos Tool features an error-proof design that ensues only 97-bp sequence can be inputted.

This new approach significantly reduces the cost and time in production of shRNAmir. Specifically, cloning of a shRNAmir by miR-AB only costs the synthesis of 142 bp oligos (75 + 67 bp), takes 0.5 h for annealing of miR-AB oligos, and requires minimal labor (mixing and pipetting the oligos), while its cloning by PCR is more cost-ineffective and labor intensive, including the synthesis of a 97 bp oligo, running a PCR, separating the PCR product on agarose gel, gel purification, and restriction enzyme digestion of the PCR product followed by its purification and quantification.

To test the cloning efficiency of miR-AB, serial concentration of ApaI/BamHI-cut retroviral miR-AB vectors was used to ligate with miR-AB oligos and the cloning efficiency was determined. As shown in Fig 1C, tens to thousands of colonies were observed on the test plates (no colonies on the miR-AB vector-only control plates) and as little as 50 pg of vectors in a ligation generated > 10 colonies, indicating a very high cloning efficiency of miR-AB. Next, we assessed the reliability of this new shRNAmir cloning strategy. We performed colony screening in multiple independent miR-AB cloning experiments by picking only one colony from each plate of miR-AB constructs for DNA sequencing. Strikingly, all the 69 colonies from five independent miR-AB cloning experiments were 100% positive (Fig 1D). These data demonstrate that miR-AB can be cloned very efficiently and extremely reliably, which is most useful for shRNAmir library construction.

miR-AB shows outstanding RNAi efficiency in vitro

To determine if miR-AB possesses the strong RNAi efficiency of miR-E after the above-mentioned modifications, we first compared the RNAi efficiency of gene-specific shRNAmirs in the miR-AB and miR-E backbone. Considering the shRNA-intrinsic variables (McIntyre et al, 2011) and differential expression of the RISC complex in different cell types (McFarland et al, 2018), it is advisable to assess multiple shRNAmirs targeting multiple genes in multiple cell lines. Therefore, two top-ranked SplashRNA-designed shRNAmirs targeting HDAC1, SMARC4A (encoding BRG-1), MTA1, and MTA2, which are components of the Nurd complex ubiquitously expressed in all somatic cells, were cloned into both miR-AB and miR-E cassettes in a lentivirus vector. After packaging in 293T cells, viruses were used to infect U87, U251, and A549 human cell lines. A nearly 100% of transduction efficiency was achieved after 72 h of infection. RNAi efficiency was determined by assessing the target protein levels in both the transduced cell lines and the 293T packaging cells. As shown in Fig 2A, a potent RNAi response was triggered by all these gene-specific shRNAmirs, consistent with the original conclusion that > 90% of the high-scoring SplashRNA predictions can be considered to markedly silence the expression of target genes (Pelossof et al, 2017). Moreover, these observations were cell line independent, further confirming the superiority of these shRNAmirs in the miR-E and miR-AB backbone. Importantly, no significant difference in RNAi efficiency was found between miR-AB and miR-E for each shRNAmir in the individual cell line, indicating that miR-AB functioned as effectively as miR-E. This result was not entirely unexpected because miR-AB and miR-E share an identical stem structure which is essential to Drosha cleavage.

Figure 2. Comparison of RNAi efficiency mediated by miR-E versus miR-AB in vitro and in vivo

Two top-ranked SplashRNA-designed shRNAmirs targeting human Nurd complex components and a control shRNAmir targeting human CD4 were cloned into both miR-AB and miR-E lentiviral vectors with Venus as reporter. The lentiviruses packaged by these constructs were used for transduction of multiple human cell lines. RNAi efficiency was determined by western blot (WB) analysis of target protein levels in the lysates of the transduced cells. β-actin served as loading control. A shERWOOD-designed shRNAmir targeting mouse transcriptional factor Tbx21 and a control shRNAmir targeting mouse CD4 were cloned into both miR-AB and miR-E retroviral vectors with Ametrine as reporter. The packaged viruses were used for transduction of primary mouse P14 TCR transgenic CD8+ T cells, which were used for evaluation of CD8+ T cells differentiation in vivo as described previously(Wang et al, 2018) (see details in the Materials and Methods section). Representative FACS plots of differentiation marker CD127/KLRG1 staining of the gated P14 CD8+ T cells and the statistical summary of SLEC subsets were shown to indicate CD8+ T cell differentiation. Data are shown as mean ± SD and n = 5 mice/group. * indicates significant difference (P < 0.01, two-tailed unpaired Student's t-test). Data are from one representative experiment of two biological replicates. The Tbx21 RNAi efficiency mediated by miR-AB vs. miR-E in the transduced in vitro-expanded P14 CD8+ T cells was determined by western blot analysis of T-bet protein level. β-actin served as loading control. The RNAi efficiency of Pten.1524 and Pten.932 shRNAmir mediated by miR-AB versus miR-E was tested in NIH-3T3 cells in the context of < 30% transduction efficiency (GFP as reporter). Representative western blots of Pten protein expression of the FACS-sorted GFP+ cells from two biological replicates were shown. β-actin served as loading control. Three shRNAmirs targeting the GFP reporter of a lentiviral vector in miR-AB or miR-E backbone were transduced into MC38 or 293T cells. The RNAi efficiency was determined by FACS analysis of GFP in the context of < 30% transduction efficiency. Data are shown as mean ± SD and n = 3 technical replicates. * indicates significant difference (P < 0.05). NS indicates no significant difference (P > 0.05, two-tailed unpaired Student's t-test).

Source data are available online for this figure.

miR-AB gives a strong RNAi phenotype in vivo

Thereafter, in order to comprehensively examine the overall influence of miR-AB-based RNAi on target cells, miR-AB-mediated RNAi was studied in vivo using a CD8+ T cell differentiation model (Wang et al, 2018). Our previous study revealed that the loss of function of the transcription factor T-bet, induced by a shERWOOD-designed and miR-E-based shRNAmir, could remarkably suppress the differentiation of short-lived effector CD8+ T cells (SLECs; Wang et al, 2018). Based on this finding, CD8+ T cell differentiation in the context of miR-AB- or miR-E-based RNAi of Tbx21, the T-bet-encoding gene, was determined in vivo (Fig 2B). Consistent with our previous results, the miR-E-based control shRNAmir-expressing P14 CD8+ T cells showed a predominant fraction of SLECs, which was the same as that observed in the context of miR-AB (Fig 2C). In contrast, Tbx21 shRNAmir expression markedly suppressed SLEC differentiation. Interestingly, this RNAi phenotype was comparable in both the miR-E and miR-AB backbones (Fig 2D). To examine if this equivalent in vivo phenotype induced by miR-AB and miR-E occurred due to their identical RNAi efficacy on the target gene, Tbx21 RNAi efficiency of these two backbones was determined via western blot analysis of its protein T-bet in the transduced cells. As expected, miR-AB- and miR-E-based Tbx21 shRNAmir induced almost indistinguishable T-bet protein loss (Fig 2D), indicating a similar RNAi efficacy of miR-AB and miR-E. Together, these data demonstrate that miR-AB has the same RNAi efficiency as miR-E in vivo.

miR-AB holds miR-E’ single-copy RNAi efficiency

miR-E’s advantage over conventional miR-30 backbone lies in its high RNAi efficiency at a single-copy level (Fellmann et al, 2013). To test if miR-AB can maintain this advantage, we first determined the RNAi efficiency of Pten.1524 and Pten.932, two shRNAmirs targeting Pten, in the backbone of miR-AB versus miR-E at a single-copy level as previously described (Fellmann et al, 2013). Consistently, these two shRNAmirs dramatically suppressed the Pten expression in 3T3 cells at < 30% transduction efficiency, an indicative of single-copy transduction (Fig 2E). Moreover, miR-AB mediated RNAi as strongly as miR-E, indicating miR-AB holds miR-E’s RNAi potency at a single-copy level.

To more intuitively show miR-AB’s RNAi efficiency at a single-copy level, we carried out a flow cytometry-based experiment to show transduction efficiency and RNAi efficiency simultaneously. To this end, we cloned three GFP-specific shRNAmirs into the miR-AB or miR-E backbone in a lentiviral vector with a GFP reporter and analyzed their RNAi potency after transduction of MC38 or 293T cells. As shown in Fig 2F, in the context of less than 30% transduction efficiency, these shRNAmirs showed potent (Sh1 and Sh2) or moderate (Sh3) knockdown efficiency (indicated by *) in GFP+ cells. Their RNAi potency was comparable between in the miR-AB versus miR-E backbone (indicated by NA). It should be noted that the medium-power Sh3 data are more meaningful to compare miR-AB versus miR-E than Sh1 and Sh2, because GFP was almost undetectable in Sh1- and Sh2-transduced cells (their MFI only represents the leftover GFP+ cells after RNAi).

Multiple eukaryotic promoters guarantee miR-AB-based RNAi in various cell types

One of the advantages of shRNAmir is its transcription by RNA Pol II (Snyder et al, 2009; Fellmann et al, 2013). This physiological trait differs dramatically from that of conventional stem-loop shRNA, which relies on the transcription by RNA Pol III. Unlike RNA Pol III’s relatively constitutive transcription (Dieci et al, 2007), RNA Pol II-directed transcription is tightly regulated, so it is more likely to be silenced in a specific cell type, even in a strong promoter setting. For example, the human cytomegalovirus (CMV) promoter, commonly used in lentiviral vectors, was frequently reported to be silenced in some cell lines (Loser et al, 1998; Brooks et al, 2004). This might be the underlying reason why some shRNAmirs failed in the induction of a strong RNAi response (Lebbink et al, 2011).

To circumvent this caveat and ensure a cell type-independent RNAi, multiple eukaryotic promoters, such as mouse CMV promoter (Dorsch-Hasler et al, 1985), CBh composite promoter (human CMV enhancer/chicken beta actin promoter/hybrid intron of chicken beta actin intron and minute virus of mice VP intron) (Gray et al, 2011), human elongation factor 1 alpha (EF1a) full-length promoter (Quinn et al, 1999) and its derivative, EH composite promoter (human EF1a mini promoter/human leukemia virus’s RU5 region; Attal et al, 1996), were introduced to replace the human CMV promoter in the previously established miR-AB lentiviral vector. Since a few of these promoters contain endogenous restriction sites that can interfere with the cloning of the miR-AB cassette or vector modification, these recognition sites were modified without changing their cognate universal transcription factor-binding sites, which were predicted by PROMO (http://alggen.lsi.upc.es/cgi-bin/promo_v3/promo/promoinit.cgi?dirDB=TF_8.3).

To test whether these promoters could function properly in these miR-AB lentiviral vectors, two top-ranked SplashRNA-designed shRNAmirs targeting the human or mouse FAS receptor, a surface protein that is constitutively expressed in most tissues (Peter et al, 2007), were cloned into miR-AB lentiviral vectors harboring these promoters. These new constructs were packaged in 293T cells as effectively as the original vector with the human CMV promoter and maintained the intensity of the Venus reporter. A cost-effective transfection of 293T cells with these constructs by PEI (~40% transfection efficiency) can generate 3–5 × 106 transduction unit titers of virus from one well of a six-well plate. Commercial transfection reagent can generate higher titers by increasing transfection efficiency. For a precise determination of RNAi efficiency of these new constructs, FAS protein levels were assessed via flow cytometry using multiple human and mouse cell lines. Consistent with the above-mentioned data, all shRNAmirs effectively silenced FAS protein expression (Fig 3), further confirming the predictive efficiency of this algorithm.

Figure 3. RNAi via miR-AB driven by multiple eukaryotic promoters in different cell types

Two top-ranked SplashRNA-designed shRNAmirs targeting human or mouse surface protein FAS were cloned into Venus-expressing miR-AB lentiviral vectors which harbor different promoters to drive the shRNAmir expression. The packaged lentiviruses were used to infect the indicated human or mouse cell lines. A shRNAmir targeting human or mouse CD4 driven under hCMV promoter was used as control. After > 96 h of infection, FAS protein levels in the transduced Venus-positive cells were determined by surface staining, shown in histograms (top panels) and quantified by MFI of APC (bottom panels).

In most cell lines, these shRNAmirs exhibited similar and remarkable RNAi effects, irrespective of the promoter type. However, they showed differential RNAi results in certain cell lines (Fig 3). For instance, human CMV promoter induced a poor RNAi response in both human U937 cells and mouse NIH-3T3 cells (Fig 3). Since human CMV promoter had a weak transcriptional activity in mouse cells (Addison et al, 1997), it is not surprising that it failed to transcribe enough miR-AB shRNAmirs to strongly silence target genes in mouse NIH-3T3 cells. However, it is unexpected that human CMV promoter-directed miR-AB did not exhibit a strong RNAi response in U937 cells, indicating the cell type-specific variation of these promoters’ activities. This cell line-specific uncertainty was the driving force to develop the multipromoter miR-AB RNAi vectors in this study. The multipromoter design can maximally solve the problems frequently found in RNAi experiments.

Construction of a multipromoter and multicolor miR-AB-based viral toolkit

In order to construct a novel RNAi tool, eight fluorescent proteins with emission wavelengths spanning the entire visible spectrum were engineered into the above-established lentiviral and retroviral miR-AB vectors to constitute a multipromoter and multicolour RNAi viral toolkit (Fig 4A–C). To facilitate the proper choice of these promoters, six lentiviral vectors with GFP as reporter that is controlled by different promoters were constructed to simply assess the promoter activity on a microscope or flow cytometer (Fig 4D). The fluorescent proteins in this toolkit can be readily detected by most widely used flow cytometers or fluorescent microscopes without need of changing parameters (Table 1). Since the original coding sequences of these fluorescent proteins contain a couple of restriction sites that are detrimental to their cloning into a given plasmid, most of these restriction sites, especially the ones that are frequently found in popular plasmids, were removed by codon optimization. Most of these fluorescent proteins have an emission peak that is quite different from the others. Ametrine, a distinctive GFP derivative, has a long Stoke’ shift that makes it compatible with GFP and Venus irrespective of their overlapping emission spectra (Ai et al, 2008). These characteristics make these fluorescent proteins suitable for construction of a multicolor panel to simultaneously visualize multiple cell populations with different genotypes. Indeed, six fluorescent reporters including Azurite, GFP, Ametrine, mOrange, mCherry2, and E2-Crimson set up a multicolor panel in CD8+ T cells where every two fluorescent reporters emitted fluorescence that were widely separated from each other, showing a clear four-quadrant plot (Fig 4E). This indicated these fluorescent reporter vectors can generate sufficiently high fluorescent signals that are easily detected and distinguished by flow cytometer. It should be noted that this flow cytometry analysis was conducted on BD FACSAria™ III Cell Sorter, a widely used flow cytometer that is equipped with four standard lasers with 405, 488, 561, and 633 nm excitations, which are not the optimal excitation wavelengths of some of the fluorescent proteins used in this experiment. Therefore, optimal laser-installed flow cytometers or fluorescent microscopes could maximally take advantage of these fluorescent reporters in a multicolor assay.

Figure 4. miR-AB-based multipromoter and multicolor viral toolkit

Schematic diagram of miR-AB-based lentiviral (LV) and retroviral (RV) vectors. miR-AB shRNAmir expression is driven by a composite promoter composed of the endogenous 5′ LTR of HIV and a eukaryotic promoter in lentiviral vectors, or by the endogenous 5′ LTR of MSCV in retroviral vectors. Eight fluorescent proteins with emission peaks across whole visible spectrum are controlled by mouse PGK promoter. A woodchuck hepatitis post-transcriptional regulatory element (WPRE) is used to enhance transgene expression. The excitation and emission spectra of these fluorescent proteins are shown in FPbase Spectra Viewer (https://www.fpbase.org/spectra/). The five commonly used lasers are indicated in the spectrum. The excitation spectrum of Azurite is shown as that of its parental protein EBFP due to lack of information in the database. Nomenclature of 56 miR-AB-based lentiviral and retroviral vectors with different promoters and different fluorescent protein reporters. AH promoter (human beta actin promoter (Ng et al, 1989)/human leukemia viruses RU5 region)-based miR-AB lentiviral vectors, the most recently developed one, also exhibited strong RNAi efficiency. Puromycin was also engineered into this toolkit for antibiotic-based screening of transformants. Lentiviral vectors with GFP reporter that are controlled by the promoters used in miR-AB lentiviral vectors were constructed for easy assessment of the promoters activity on microscope or flow cytometer. CD8+ T cells were activated and then cotransduced with miR-AB retroviruses targeting CD4 with Azurite, GFP, Ametrine, mOrange, mCherry2, or E2-Crimson as fluorescent reporters. The fluorescence of all fluorescent proteins was plotted on contour plots showing the fluorescent separation of each pair of the fluorescent proteins. The data are representative of at least three biologically independent experiments. 293T cells were cotransduced with four miR-AB lentiviruses at 1:1 ratio, which carries a neutral shRNAmir targeting CD4 or CD19, with Azurite, GFP, Ametrine, or mOrange as reporter, respectively. Cells were cultured for 22 days and fluorescence-positive cells were quantified at the time points as indicated. Representative contour FACS plots displaying the fluorescent reporter+ cells on day 4 and day 22 are shown (top panels). Cell numbers of untransduced, single fluorescence-positive cells, and quadruple positive cells were quantified and normalized to the untransduced cells on day 4, and plotted over time (bottom panel). Data are shown as mean ± SD and n = 4 technical replicates. NS indicates no significant difference (P > 0.05, two-tailed unpaired Student's t-test). Table 1. Flow cytometry parameters of a five lasers-equipped FACSAria™ III flow cytometer for detection of the fluorescent proteins. Laser Band pass filter Commonly used fluorophores with similar spectral characteristics Azurite UV (355 nm) 450/40 Indo-1 hi, Alexa Fluor 350, Marina Blue Violet (405 nm) 450/40 Brilliant Violet 421, Pacific Blue, Alexa Fluor 405, Sytox Blue mTagBFP2 Violet (405 nm) 450/40 Brilliant Violet 421, Pacific Blue, Alexa Fluor 405, Sytox Blue Ametrine Violet (405 nm) 510/50 Brilliant Violet 510, AmCyan, Krome Orange EGFP Blue (488 nm) 530/30 FITC, Alexa Fluor 488, CFSE, Sytox Green, Dylight 488 Venus Blue (488 nm) 530/30 FITC, Alexa Fluor 488, CFSE, Sytox Green, Dylight 488 mOrange Blue (488 nm) or Yellow/Green (561 nm) 582/15 PE, PI, Alexa Fluor 555, Dylight 549 mCherry2 Yellow/Green (561 nm) 610/20 PE/Texas Red, Alexa Fluor 594, PE-CF594, Dylight 594 E2-Crimson Red (633 nm) 660/20 APC, Alexa Fuor 647, Cy5, Sytox Red, Dylight 649

These fluorescent proteins were deliberately chosen according to not only their excitation and emission wavelength peaks but also their other characteristics, such as good photo stability and chromophore chemical stability, high fluorescent intensity, and fast maturation. In addition, Aequorea Victoria-derived fluorescent proteins (GFP, Azurite, Ametrine and Venus) and mOrange were included in this toolkit as they showed low immunogenicity (Gossa et al, 2014; Wang et al, 2018), making them especially useful for performing adoptive transfer experiments in mice. To expand the application of miR-AB, codon-optimized puromycin resistance gene was integrated into this RNAi system for selection and maintenance of cells expressing shRNAmirs (Fig 4C).

Multiple shRNAmir transduction is not cytotoxic

The aim of this study was to develop a novel multiplex RNAi assay. We chose shRNAmir but not conventional shRNA in this assay because shRNAmir displayed minimal cytotoxicity. Indeed, our experimental data showed the cotransduction of 293T cells with four viruses carrying a neutral shRNAmir did not significantly alter the four fluorescent reporters expressing cell expansion after 22 days of culture (Fig 4F). Since the multiple shRNAmirs-expressing cells always account for a small portion of the whole cell population (< 10% in this experiment), like a low-efficiency transduction of a single shRNAmir, it is very likely that they express each shRNAmir at a single-copy level. So, their total shRNAmirs level will not be super abundant. Given the advantage of shRNAmir over conventional shRNA in maintaining cell homeostasis, multiple shRNAmirs transduction might not be problematic.

Multicolor-barcoded multiplex RNAi efficiently and reliably silences multiple target genes both individually and combinatorially at a single-cell level

Finally, we tried to use this toolkit to set up a multicolor-barcoded multiplex RNAi assay where loss-of-function effects of target genes could be analyzed at both single-gene level and multiple-gene level. To this end, four CD8+ T cell surface proteins, CD127, CD90, CD44, and CD8a were knocked down in mouse CD8+ T cells by cotransduction of miR-AB retroviruses expressing either of the two top-ranked SplashRNA-designed shRNAmirs targeting CD127, CD90, CD44, or CD8a with Azurite, GFP, Ametrine, and mCherry2 as fluorescent reporters, respectively (Fig 5 and EV1). In parallel, similar CD19 (Fig 5) or CD4 (Fig EV1) RNAi was used as control. Obviously, the cotransduced cells showed 16 cell populations which expressed different combinations of the 4 fluorescent reporters. The overall percentage of these populations was > 50% (untransduced cell percentage was <50%, top bars in Fig EV2), indicating these viruses had high viral titers because primary CD8+ T cells are harder to be infected than cell lines. The lowest individual percentage of these population was > 1%, indicating there are enough cells for FACS analysis of each population in this assay (hundreds of cells can give a clear FACS population).

Figure 5. miR-AB-based multicolor-barcoded multiplex RNAi

CD8+ T cells were activated for 16 h and then cotransduced with miR-AB retroviruses at 1:1 ratio, which carries a SplashRNA-designed shRNAmir, targeting surface protein CD127, CD90, CD44, or CD8, with Azurite, GFP, Ametrine, or mCherry2 as fluorescent reporters, respectively, for 4 h. miR-AB retroviruses expressing CD19-specific shRNAmirs with same fluorescent reporters were used as controls. Surface staining was performed 72 h after transduction. FACS data were shown as overlaid histograms of control shRNAmirs (shCD19s) and target shRNAmirs (shCD127, shCD90, shCD44, or shCD8). Cells were categorized as 16 populations expressing single, double, triple, or quadruple fluorescent proteins or not (none) and fluorescent reporter-positive plots were highlighted with thick frames and white background. Each population’s surface staining profile was shown on the right. The data are representative of two biologically independent experiments.

Click here to expand this figure.

Figure EV1. Multicolor-barcoded multiplex RNAi by miR-AB retrovirus in CD8+ T cells

This RNAi experiment was carried out as described in Fig 5 but used another SplashRNA-designed shRNAmirs targeting CD127, CD90, CD44, or CD8. CD4-specific shRNAmirs were used as controls. The data are representative of two biologically independent experiments.

Click here to expand this figure.

Figure EV2. RNAi efficiency quantification of FACS plots in Figs 5 and EV1

Target gene expression was quantified by calculating its MFI in FACS plots. The percentage of all fluorescence-positive cells (untransduced, single, double, triple, and quadruple) were also shown (first column panels).

Table 2. The list of miR-AB oligos used for cloning of shRNAmirs in this study. Target gene shRNAmir Species Designed by miR-AB oligos HDAC1 sh1 Human SplashRNA GATCCGAGGTTAGGTTGCTTCAATCTAATAGTGAAGCCACAGATGTATTAGATTGAAGCAACCTAACCGTGGGCC CACGGTTAGGTTGCTTCAATCTAATACATCTGTGGCTTCACTATTAGATTGAAGCAACCTAACCTCG HDAC1 sh2 Human SplashRNA GATCCGACACAGCGATGACTACATTAAATAGTGAAGCCACAGATGTATTTAATGTAGTCATCGCTGTGGTGGGCC CACCACAGCGATGACTACATTAAATACATCTGTGGCTTCACTATTTAATGTAGTCATCGCTGTGTCG SMAR4A sh1 Human SplashRNA GATCCGATGGATGTCAAACAGTAATAAATAGTGAAGCCACAGATGTATTTATTACTGTTTGACATCCAGTGGGCC CACTGGATGTCAAACAGTAATAAATACATCTGTGGCTTCACTATTTATTACTGTTTGACATCCATCG SMAR4A sh2 Human SplashRNA GATCCGACCGTGGACTTCAAGAAGATAATAGTGAAGCCACAGATGTATTATCTTCTTGAAGTCCACGGGTGGGCC CACCCGTGGACTTCAAGAAGATAATACATCTGTGGCTTCACTATTATCTTCTTGAAGTCCACGGTCG MTA1 sh1 Human SplashRNA GATCCGCTCAAGTAATTTTCATATTAAATAGTGAAGCCACAGATGTATTTAATATGAAAATTACTTGAATGGGCC CATTCAAGTAATTTTCATATTAAATACATCTGTGGCTTCACTATTTAATATGAAAATTACTTGAGCG MTA1 sh2 Human SplashRNA GATCCGAAAGGTGCCATTTTAAATTTTATAGTGAAGCCACAGATGTATAAAATTTAAAATGGCACCTTCTGGGCC CAGAAGGTGCCATTTTAAATTTTATACATCTGTGGCTTCACTATAAAATTTAAAATGGCACCTTTCG MTA2 sh1 Human SplashRNA GATCCGACAGCATAGTCCAGTTTTATTATAGTGAAGCCACAGATGTATAATAAAACTGGACTATGCTGGTGGGCC CACCAGCATAGTCCAGTTTTATTATACATCTGTGGCTTCACTATAATAAAACTGGACTATGCTGTCG MTA2 sh2 Human SplashRNA