Abstract

   Small interfering RNAs (siRNAs) hold considerable therapeutic potential
   to selectively silence previously “undruggable” disease-associated
   targets, offering new opportunities to fight human diseases. This
   therapeutic strategy, however, is limited by the inability of naked
   siRNAs to passively diffuse across cellular membranes due to their
   large molecular size and negative charge. Delivery of siRNAs to liver
   through conjugation of siRNA to N-acetylgalactosamine (GalNAc) has been
   a success, providing robust and durable gene knockdown, specifically in
   hepatocytes. However, the poor delivery and silencing efficacy of
   siRNAs in other cell types has hindered their applications outside the
   liver. We previously reported that a genome-wide pooled knockout screen
   identified RAB18 as a major modulator of GalNAc-siRNA conjugates.
   Herein, we demonstrate RAB18 knockout/knockdown efficaciously enhances
   siRNA-mediated gene silencing in hepatic and extrahepatic cell lines
   and in vivo. Our results reveal a mechanism by which retrograde
   Golgi-endoplasmic reticulum (ER) transport and the intracellular lipid
   droplets (LDs) positively regulate siRNA-mediated gene silencing.

   Keywords: MT: Oligonucleotides: Therapies and Applications, RAB18,
   siRNA, siRNA trafficking, extrahepatic, lipid droplets, retrograde
   Golgi-ER transport

Graphical abstract

   graphic file with name fx1.jpg
   [59]Open in a new tab
     __________________________________________________________________

   Using a variety of in vitro systems, Lu and colleagues demonstrated
   that knocking out RAB18 enhances siRNA-mediated gene silencing in both
   hepatic and extrahepatic cells. Furthermore, the authors translated
   this finding to an in vivo model. Altogether, the data point to a
   mechanism by which RAB18 regulates siRNA-mediated gene silencing in
   extrahepatic cells.

Introduction

   During the last decade, clinical success of small interfering RNA
   (siRNA)-based medicines transformed the drug development pipeline;
   targets once considered undruggable are now druggable using
   high-specificity siRNA technology. siRNAs exert their effect through
   RNA interference (RNAi) of targeted mRNAs. Despite the remarkable
   success of delivering therapeutic siRNAs specifically to hepatocytes
   through GalNAc conjugation, delivery of efficacious amounts of siRNA to
   extrahepatic cells or tissues with specificity, potency, and durability
   remains challenging. To extend our mechanistic understanding of how
   intracellular siRNA-mediated gene silencing is regulated, we previously
   conducted a genome-wide pooled CRISPR-Cas9 screen based on the delivery
   of GalNAc-conjugated siRNA targeting the HPRT1 gene in the human
   hepatocellular carcinoma line Hep3B.[60]^1 This screen identified
   multiple candidate genes that, when knocked out, enhanced
   siRNA-mediated gene silencing in Hep3B cells. One of the top candidate
   genes identified through this screen was RAB18. To elucidate the impact
   of RAB18 on siRNA-mediated gene silencing, we generated RAB18-knockout
   and -overexpression Hep3B cells. Our previous data indicate that
   depletion of RAB18 improves siRNA-mediated gene silencing for siRNAs
   delivered by conjugation to GalNAc, cholesterol, or anti-ASGR1
   antibodies, but not via liposome-mediated transfection. Here, we extend
   our previous findings from hepatic cells to extrahepatic cells and
   demonstrate translation of the in vitro results to an in vivo mouse
   model. In a standard GalNAc-siRNA conjugate trafficking event, the
   GalNAc-siRNA molecules are internalized into hepatocytes through
   asialoglycoprotein receptor (ASGPR)-mediated endocytosis. As endosomes
   mature, the internal pH decreases, resulting in the release of the
   GalNAc-siRNA molecules from ASGPRs. Subsequently, ASGPRs recycle back
   to the cell surface while GalNAc-conjugated siRNAs remain inside the
   endosomes.[61]^2 GalNAc and the linkers are subsequently degraded while
   a small fraction of the remaining free siRNAs (<1%) are released from
   the endosomes. Once in the cytoplasm, the siRNA molecules access
   RNA-induced silencing complex (RISC), resulting in target transcript
   degradation and inhibited protein translation.[62]^2^,[63]^3 Liposome
   transfection-mediated siRNA delivery induces mRNA knockdown at low
   siRNA concentrations, possibly due to the ability of
   liposome-containing vesicles to perform random Brownian motion within
   the cytoplasm to help siRNA circumvent some of the intracellular
   barriers encountered by alternative delivery methods such as GalNAc,
   cholesterol, or antibody-mediated uptake.[64]^4 Previously we observed
   that the activity of siRNA delivered by liposomes was not altered by
   RAB18 knockout; thus, we deduced RAB18 specifically regulates the
   silencing effect of siRNA delivered through ligand-conjugated
   methods.[65]^1 The protein encoded by the RAB18 gene is a member of
   RAS-related small GTPases that regulate membrane trafficking in
   organelles and transport vesicles. Studies link RAB18 to the regulation
   of lipid droplet (LD) formation,[66]^5^,[67]^6 inhibition of
   COPI-independent retrograde trafficking from Golgi to endoplasmic
   reticulum (ER),[68]^7 regulation of secretory granules[69]^8 and
   peroxisomes,[70]^9 promotion of hepatitis C virus (HCV) assembly on the
   LD membrane,[71]^10 and regulation of ER structure.[72]^11 To elucidate
   the mechanism responsible for the effect of RAB18 knockout on
   siRNA-mediated gene silencing, we conducted a series of experiments to
   demonstrate that retrograde Golgi-ER transport and LDs positively
   regulate siRNA-mediated gene silencing.

Results

RAB18 gene expression negatively correlates with both siRNA- and antisense
oligonucleotide-mediated gene silencing

   In previous studies, we demonstrated RAB18 knockout/knockdown enhances
   the silencing effect of siRNAs delivered by GalNAc, cholesterol, or
   antibodies, but not siRNAs delivered by liposome-mediated transfection
   in Hep3B cells by using siRNA targeting HPRT1, PPIB, and ASGR1
   genes.[73]^1 To better understand the correlation between RAB18 gene
   expression and siRNA activity, we opted to use an siRNA targeting the
   HPRT1 gene for further study. We generated RAB18-overexpressing Hep3B
   cells. Both the previously validated Cas9 stable Hep3B cells
   (Hep3BCas9) and the corresponding RAB18-knockout Hep3BCas9 cells
   (Hep3BRAB18KO)[74]^1 were transduced with lentivirus expressing a
   RAB18-overexpression construct ([75]Figure S1A). The mCherry-positive
   populations were sorted for RAB18-overexpressing cells,
   Hep3BCas9RAB18_OE and Hep3BRAB18KO_OE ([76]Figures 1A and 1B), and
   RAB18 overexpression was confirmed by digital polymerase chain reaction
   (dPCR) ([77]Figures S1B and S1C). Next, the HPRT1 gene expression level
   was measured in Hep3BCas9, Hep3BRAB18KO, Hep3BCas9RAB18_OE, and
   Hep3BRAB18KO_OE cells treated with either GalNAc-conjugated siRNA
   targeting the HPRT1 gene (GalNAc-HPRT1 siRNA, compound no. 8172,
   human-mouse cross reactive) or the control GalNAc-conjugated siRNA
   target a housekeeping gene, PPIB, as a negative control for HPRT1 gene
   silencing (GalNAc-PPIB siRNA, compound no. 8714) ([78]Table S1).
   Hep3BRAB18KO cells demonstrated enhanced HPRT1 siRNA silencing efficacy
   (half maximal inhibitory concentration [IC50] = 2.69 nM), while
   overexpression of RAB18 in Hep3BRAB18KO cells mitigated HPRT1 siRNA
   silencing efficacy (IC50 = 19.25 nM) ([79]Figure 1B); a similar pattern
   was observed when RAB18 was overexpressed in Hep3BCas9 cells
   ([80]Figure 1A). Thus, these data demonstrate RAB18 gene expression
   negatively correlates with siRNA-mediated gene silencing.

Figure 1.

   [81]Figure 1
   [82]Open in a new tab

   Impact of RAB18 gene expression on the silencing efficacy of siRNA and
   ASO molecules

   (A) Effect of RAB18 overexpression on siRNA-mediated gene silencing in
   Hep3BCas9 cells. The left and middle figures show flow cytometry data
   demonstrating the successful overexpression of the RAB18 gene using a
   lentivirus vector ([83]Figure S1A). Cells overexpressing RAB18 are
   mCherry positive. The figure on the right shows dPCR measurements of
   HPRT1 expression in GalNAc-HPRT1 siRNA-treated Hep3BCas9 cells and
   Hep3BCas9RAB18_OE cells (n = 3). (B) Effect of RAB18 overexpression on
   siRNA-mediated gene silencing in Hep3BRABKO cells. The left and middle
   figures show the flow cytometry data, which demonstrate the successful
   overexpression of the RAB18 gene in Hep3BRAB18KO cells. The right
   figure shows dPCR measurements of HPRT1 expression in GalNAc-HPRT1
   siRNA-treated Hep3BRAB18KO cells and the RAB18-overexpression
   Hep3BRAB18KO_OE cells (n = 3). (C) dPCR measurements of HPRT1
   expression in Hep3BCas9 and Hep3BRAB18KO cells treated with
   GalNAc-conjugated ASO molecules targeting the HPRT1 gene (15469, 15470)
   (n = 3). All samples were measured 4 days post siRNA treatments. All
   dPCR results were normalized by the control housekeeping gene, TBP, and
   no siRNA (PBS only)-treated control groups. All error bars represent
   standard deviation (SD).

   To determine if RAB18 regulates the silencing efficacy of
   single-stranded antisense oligonucleotides (ASOs), we performed a set
   of similar experiments. Hep3BCas9 parental cells and Hep3BRAB18KO cells
   were treated with either PBS or increasing concentrations of one of two
   GalNAc-conjugated ASO molecules targeting the HPRT1 gene (compound nos.
   15469 and 15470). Sequences for each of the three GalNAc-conjugated ASO
   molecules are described in [84]Table S2. As demonstrated in
   [85]Figure 1C, after 4 days, Hep3BRAB18KO cells treated with
   GalNAc-conjugated ASO molecules targeting HPRT1 resulted in enhanced
   silencing of HPRT1 compared to Hep3BCas9 cells receiving the same ASO
   ([86]Figure 1C). For example, the IC50 value for compound no. 15469 in
   Hep3BCas9 parental cells was 242.6 nM, whereas the IC50 value for the
   same compound in Hep3BRAB18KO cells was 12.58 nM, a 20-fold enhancement
   of silencing efficacy. These results suggest that RAB18 regulates an
   intracellular step similarly used by both GalNAc-conjugated siRNA
   molecules and GalNAc-conjugated ASO molecules.

RAB18 knockout enhances siRNA-mediated gene silencing in extrahepatic cells

   Since the RAB18 gene is expressed ubiquitously in mammalian tissues, we
   next addressed the effect of RAB18 expression on siRNA-mediated gene
   silencing in extrahepatic cells. Based on our internal RNA sequencing
   (RNA-seq) data ([87]Table S3), RAB18 gene expression is lower in
   hepatic Hep3B cells compared to the breast cancer cell line T47D
   ([88]Table S3). Thus, we initiated confirmation of RAB18 function on
   regulating siRNA-mediated gene silencing in extrahepatic T47D cells. To
   generate RAB18-knockout T47D cells, we co-delivered lentivirus
   packaging Cas9 (blasticidin resistant) with lentivirus packaging guide
   RNA (gRNA) targeting the RAB18 gene (puromycin resistant) (SIGMA
   vector: U6-gRNA: PGK-puro-2A-tagBFP) into T47D cells. The T47D
   RAB18-knockout cells (T47DRAB18KO) were enriched by blasticidin and
   puromycin selection. In parallel, Cas9 stable T47D control cells
   (T47DCas9) were generated by transducing lentivirus packaging Cas9.
   Inference of CRISPR edit (ICE) analysis confirmed successful knockout
   of the RAB18 gene ([89]Figure S2A) with no effect on cell viability
   ([90]Figure S2B). Both negatively charged and large, unconjugated siRNA
   molecules do not easily enter cells without conjugation. However, trace
   amounts of siRNA could be engulfed by cells via pinocytosis.[91]^12 To
   evaluate the possibility of pinocytosis, unconjugated HPRT1 siRNA
   (compound no. 39986; [92]Table S1) was added to T47D cells. Despite
   sub-optimal delivery, unconjugated HPRT1 siRNA led to robust target
   gene silencing in both T47DRAB18KO and Hep3BRAB18KO cells, but not in
   the parental T47DCas9 and Hep3BCas9 cells ([93]Figure 2A),
   demonstrating RAB18 serves as a roadblock in siRNA-mediated gene
   silencing in both hepatic and extrahepatic cells.

Figure 2.

   [94]Figure 2
   [95]Open in a new tab

   Impact of RAB18 knockout on siRNA-mediated gene silencing in
   extrahepatic cells

   (A) dPCR measurements of HPRT1 expression in Hep3BCas9, Hep3BRAB18KO,
   T47DCas9, and T47DRAB18KO cells treated with unconjugated HPRT1 siRNA
   (39986) (n = 4). (B) Editing efficacy on the RAB18 gene assessed by ICE
   analysis. (C) Editing efficacy on the RAB18 gene analyzed by western
   blot of extrahepatic cell lines. (D–I) dPCR measurements of HPRT1
   expression in RAB18-knockout extrahepatic cell lines and their control
   cell lines (Cas9 only) treated with unconjugated HPRT1 siRNA or
   unconjugated non-targeting (NT) siRNA. All samples were measured 4 days
   post siRNA treatments. All dPCR results shown were normalized by TBP
   and no-siRNA (PBS only)-treated control groups. (J) dPCR measurements
   of RAB18 gene expression 3 days post RAB18 siRNA and NT siRNA
   transfection in human iPSC-derived cardiomyocytes. (K) dPCR
   measurements of HPRT1 gene expression of human iPSC-derived
   cardiomyocytes transfected with RAB18 siRNA or NT siRNA followed by
   treatment of GalNAc-HPRT1 siRNA (7 days post RAB18 siRNA and NT siRNA
   transfection). Asterisk indicates significant difference (∗p < 0.05);
   ns, not significant based on one-way ANOVA test. All error bars
   represent SD.

   To determine if the regulatory role of RAB18 translates across other
   extrahepatic cell types, additional extrahepatic cell lines with
   varying levels of RAB18 expression were selected to generate RAB18
   knockouts. These cell lines include pancreatic cancer cell line
   MiaPaCa-2, kidney cancer cell line A498, lung cancer cell line
   NCIH1299, colorectal cancer cell lines COLO205 and HT-29, and breast
   cancer cell line MCF7 ([96]Table S3). To simplify the production of
   RAB18-deficient cells, Cas9 protein and three synthesized sgRNAs
   (single guide RNA) targeting RAB18 gene ([97]Table S4) were
   electroporated into the cell lines. Cells electroporated with only Cas9
   protein were used as controls. The effect of RAB18 ablation was
   evaluated by ICE analysis and western blot ([98]Figures 2B and 2C). The
   validated RAB18 knockouts and their Cas9 control cells were
   subsequently treated with unconjugated HPRT1 siRNA. As summarized in
   [99]Figures 2D and 2E, RAB18 ablation robustly enhanced siRNA-mediated
   gene silencing of unconjugated HPRT1 siRNA in MiaPaCa-2 cells and
   moderately in NCIH1299 cells. No overt effect was observed in the A498,
   COLO205, HT-29, and MCF7 RAB18-deficient cell lines
   ([100]Figures 2F–2I). The internalization rate of unconjugated HPRT1
   siRNA may depend on the rate of pinocytosis, and thus the robustness of
   siRNA-mediated silencing could vary across cells due to differing
   levels of pinocytosis.

   When validating siRNA silencing effect in immortalized cell lines, it
   is possible that the effect of internalized siRNA is diluted rapidly by
   cell division limiting the degree of gene silencing. To retain the
   intracellular effect of siRNA, the impact of RAB18 on siRNA-mediated
   gene silencing in extrahepatic cells was evaluated in non-proliferating
   human iPSC-derived cardiomyocytes. Using liposome-mediated
   transfection, two different unconjugated RAB18 siRNAs (54559-1 and
   54561-1) resulted in RAB18 mRNA knockdown by day 3 post transfection
   while non-targeting (NT) siRNA had no effect ([101]Figure 2J). After
   washing and replacing medium, the cells were then treated with
   unconjugated siRNA targeting the HPRT1 gene for an additional 4 days.
   As shown in [102]Figure 2K, after treatment with HPRT1 siRNA, HPRT1
   mRNA expression level was at 29.95% or 36.05%, respectively, in cells
   pre-treated with one of two different RAB18 siRNAs compared to 72.65%
   in the cells pre-treated with NT siRNA. These results further
   demonstrate the regulatory role RAB18 expression has on siRNA-mediated
   gene silencing in extrahepatic cells.

Knocking down Rab18 in mouse liver enhances the target gene silencing effect
of GalNAc-HPRT1 siRNA

   We have thus far demonstrated enhanced siRNA-mediated gene silencing in
   multiple hepatic and extrahepatic cell lines by generating RAB18
   knockout or knockdown in vitro. To explore translatability, we next
   evaluated the effect of localized RAB18 deletion in vivo. Global,
   germline RAB18 deficiency leads to inherited human disease, Warburg
   micro syndrome. Although no significant dysfunction or toxicity was
   observed in RAB18-knockout immortalized cell lines or human
   iPSC-derived cardiomyocytes, we were concerned about the potential of
   global deficiency in vivo. Therefore, instead of using Rab18-deficient
   mice that have been reported to exhibit major ocular and neurological
   abnormalities associated with Warburg micro syndrome,[103]^13 we used
   Rab18 siRNA to temporally knockdown Rab18 in mouse liver to study its
   impact on siRNA activity in vivo. To prepare for in vivo testing, we
   designed and validated 26 human-mouse cross-reactive siRNA molecules
   targeting the RAB18/Rab18 gene in human Hep3B and mouse Hepa 1–6 cells
   through liposome-mediated transfection ([104]Tables S5 and [105]S6).
   The off-target potential of the six most potent RAB18 siRNA triggers
   (54559-1, 54561-1, 54570-1, 54564-1, 54569-1, and 54553-1) was
   predicted based on near-perfect complementarity (data not shown).
   Bioinformatic analysis identified triggers 54561-1, 54570-1, 54564-1,
   and 54553-1 with the least off-target potential. These triggers were
   then conjugated with GalNAc for in vivo studies ([106]Table S7). The
   siRNA-mediated gene silencing effect of these GalNAc-conjugated Rab18
   triggers (64131-1, 64134-1, 64137-1, and 64140-1) along with the
   negative control GalNAc-conjugated NT siRNA (42105-3) and positive
   control GalNAc-conjugated siRNA targeting Hprt1 gene (8172) were
   evaluated in mouse livers. As outlined in [107]Figure 3A, 2 weeks after
   subcutaneous (s.c.) injection of siRNAs, mouse livers were collected
   for RNA isolation, and the mRNA levels of Rab18 and Hprt1 were
   determined by dPCR. As expected, Hprt1 siRNA established more than 80%
   silencing of the Hprt1 gene on day 14 post injection; no impact on
   Rab18 mRNA expression was detected ([108]Figures 3B and 3C) and no
   change in either Hprt1 or Rab18 mRNA was detected in animals injected
   with vehicle (PBS) or NT siRNA ([109]Figures 3B and 3C). Rab18 mRNA
   knockdown was achieved by all Rab18 siRNAs assessed; 64131-1
   established the most robust silencing of Rab18 mRNA (88%) without
   altering Hprt1 mRNA ([110]Figures 3B and 3C). No notable impact on body
   weight ([111]Figure S3A) or serum liver transaminase levels
   ([112]Figures S3B–S3D) were detected among the molecules except with
   64137-1, indicating potential issues with this siRNA molecule.

Figure 3.

   [113]Figure 3
   [114]Open in a new tab

   Results from the in vivo study evaluating the silencing effect of
   GalNAc-Rab18 siRNA triggers and transcriptomic analysis of liver RNA
   samples harvested from siRNA-treated mice

   (A) Schematic diagram illustrating the experimental workflow for the
   in vivo study. (B) dPCR measurements of Hprt1 expression in liver RNA
   samples harvested 14 days post -s.c. injection. (C) dPCR measurements
   of Rab18 expression in liver RNA samples harvested 14 days post s.c.
   injection. All dPCR results were normalized by Tbp and the control
   vehicle (PBS)-treated control groups. (D) Number of significant
   differentially expressed genes across all experimental groups as
   compared to vehicle control. Genes with a p-adjusted <0.05 and a
   fold-change >1.5 (up) or < −1.5 (down) were considered significantly
   differentially expressed for all panels. (E) Volcano plot of
   differentially expressed genes in 64131-1-treated samples compared with
   vehicle-treated control group. The numbers of significantly
   downregulated (blue) or upregulated - (red) genes are indicated at the
   top. (F) Top enriched Gene Ontology (GO) biological processes gene sets
   among upregulated genes from (E). (G) Top enriched GO biological
   processes gene sets among downregulated genes from (E). (H) Venn
   diagram and list of shared upregulated genes identified in all Rab18
   siRNA trigger-treated samples compared to vehicle control. (I) Venn
   diagram and list of shared downregulated genes identified in all Rab18
   siRNA trigger-treated samples compared to vehicle control. The adjusted
   p-value is calculated using the Benjamini-Hochberg method.

   To evaluate the global transcriptomic impact of Rab18 knockdown in the
   mice, RNA-seq analysis was performed on RNA extracted from all livers.
   Comparison of the vehicle injection group with the NT or the
   Hprt1-positive control groups showed minor transcriptional changes
   ([115]Figures S4A and S4B). Rab18 siRNA treatment, however, elicited a
   much stronger transcriptomic response ([116]Figures 3D, 3E, and
   [117]S4C). The data revealed more transcriptomic changes in samples
   treated with Rab18 siRNA triggers. However, the level of response did
   not correlate with the degree of Rab18 knockdown, indicating
   trigger-sequence-specific contributions in addition to Rab18 gene
   function-related changes. We focused on Rab18 siRNA, 64131-1, as the
   severity of Rab18 knockdown was greatest with this trigger and
   represents the best candidate to identify Rab18-mediated responses.
   Differential expression analysis revealed a strong bi-directional
   transcriptomic response, with Rab18 as the most significantly
   downregulated gene ([118]Figure 3E). Pathway analysis of upregulated
   genes revealed processes related to lipid/cholesterol metabolism and
   downregulated genes revealed processes related to cell cycle
   ([119]Figures 3F and 3G). A similar pathway analysis using upregulated
   genes from the NT siRNA (42105-3)-treated samples did not recapitulate
   the 64131-1 results, indicating the differentially expressed genes
   identified in 64131-1-treated samples are not related to general siRNA
   treatment ([120]Figure S4D). Rab18 siRNA, 64137-1, was less effective
   at inducing mRNA knockdown than 64131-1 or 64134-1 yet elicited the
   strongest overall transcriptomic response compared to vehicle
   ([121]Figures 3D and [122]S4C). Pathway analysis of upregulated genes
   revealed 64137-1 treatment altered mitochondrial function related to
   drug-induced liver injury (DILI), which could indicate potential
   toxicity issues of this trigger; this was further supported by serum
   liver transaminase levels ([123]Figures S3B, S3D, and [124]S4E). Few
   upregulated genes were shared across all four Rab18 siRNA triggers. Of
   the four shared upregulated genes, Nr1d2 is known to play a role in
   carbohydrate and lipid metabolism ([125]Figure 3H). Pathway enrichment
   analysis of the 34 shared downregulated genes across all four Rab18
   siRNA triggers highlights an effect on the cell cycle ([126]Figure 3I).
   Furthermore, several kinesins are present on the list, which may
   indicate loss of transport of cellular cargo ([127]Figure 3I). By
   analyzing ∼1,400 liver-toxicity-associated genes, no marked change in
   expression was observed ([128]Figures S4F and S4G).

   Because the most potent Rab18 siRNA, 64131-1, did not induce
   indications of cell toxicity, it was selected for an in vivo study to
   evaluate if reducing Rab18 expression could enhance siRNA-mediated gene
   silencing. As illustrated in [129]Figure 4A, mice were subjected to
   sequential subcutaneous administration of GalNAc-conjugated siRNAs, or
   vehicle, as indicated. On day 0, either PBS (vehicle) or 3 mg/kg NT
   siRNA was administered to control groups (1 and 4–8) and 3 mg/kg Rab18
   siRNA 64131-1 was injected in treatment groups (2–3) to knockdown Rab18
   expression. On day 14, each animal received either a second injection
   of PBS (groups 1, 3, and 5) or a second injection of Hprt1 siRNA (8172)
   at different dosages (0.5 mg/kg [groups 2, 4, and 6], 1 mg/kg [group
   7], or 3 mg/kg [group 8]). On day 28, body weights were recorded
   ([130]Figure S5A) and serum collected for liver transaminases
   ([131]Figures S5B–S5D). Livers were harvested, RNA isolated, and Rab18
   and Hprt1 mRNA expression levels were assessed by dPCR. Rab18
   expression was reduced more than 80% in Rab18 siRNA-treated animals
   ([132]Figure 4B). The effect of Rab18 on Hprt1 expression data is
   summarized in [133]Figure 4C. In negative control groups (1, 3, and 5),
   injection with either vehicle, Rab18 siRNA, or NT siRNA followed by a
   second injection of vehicle did not reduce Hprt1 mRNA expression. For
   positive control groups (group 4), in which animals received 0.5 mg/kg
   Hprt1 siRNA following NT siRNA, Hprt1 mRNA expression was 79.71%. In
   contrast, Hprt1 mRNA expression in group 2, in which animals received
   0.5 mg/kg Hprt1 siRNA following 3-mg/kg Rab18 siRNA injection, was
   51.00% of control, which is similar to the level of Hprt1 detected
   group 7, in which animals received 1 mg/kg Hprt1 siRNA following
   vehicle injection. These results provide translational in vivo evidence
   that Rab18 negatively regulates siRNA-mediated gene silencing.

Figure 4.

   [134]Figure 4
   [135]Open in a new tab

   Results from the second in vivo study evaluating the effect of Rab18
   knockdown on siRNA-mediated gene silencing

   (A) Schematic diagram illustrating the experimental workflow and group
   design for the second in vivo study. (B) dPCR measurements of Rab18
   expression in liver RNA samples harvested 28 days post s.c. injection.
   (C) dPCR measurements of Hprt1 expression in liver RNA samples
   harvested 28 days post s.c. injection. All dPCR results were normalized
   by Tbp and the control vehicle (PBS)-treated control groups. Asterisk
   indicates significant difference (∗p < 0.05) based on one-way ANOVA
   test. All error bars represent SD.

RAB18 knockout correlates with reduced intracellular siRNA over time

   In a previous study, the effect of RAB18 on siRNA internalization was
   evaluated in Hep3BCas9 and Hep3BRAB18KO cells.[136]^1 Briefly, cells
   were treated with fluorescently labeled GalNAc-conjugated HPRT1 siRNA
   for 0.5, 2, 4, and 20 h and then the internalization of siRNA molecules
   was assessed by flow cytometry. Using this assay, we observed no
   notable change in the amount of internalization between Hep3BCas9 and
   Hep3BRAB18KO cells.[137]^1 To study the dynamics of intracellular
   siRNAs further, we repeated the assay by incubating cells with
   GalNAc-conjugated HPRT1 siRNA for 6 h, 1 day, or 2 days, and then used
   an imaging-based assay and rabbit polyclonal anti-siRNA antibody
   recognizing 2′OMe- and 2′F-containing oligonucleotides to quantitate
   the number of intracellular siRNA molecules. As summarized in
   [138]Figure 5A, after 6 h of siRNA treatment, the intracellular siRNA
   counts were similar for Hep3BCas9 (17.64/cell) and Hep3BRAB18KO
   (15.53/cell) cells. However, as the siRNA treatment extended, fewer
   siRNA molecules accumulated in the Hep3BRAB18KO (23.47/cell on day 2)
   cells versus the Hep3BCas9 (31.00/cell on day 2) cells. Thus, although
   RAB18 knockout appeared to have no impact on siRNA internalization in
   this system, it might reduce the accumulation of siRNAs or increase the
   consumption, or degradation, of siRNA.

Figure 5.

   [139]Figure 5
   [140]Open in a new tab

   Impact of RAB18 knockout on intracellular siRNA accumulation and the
   impact of retrograde Golgi-ER transport on siRNA-mediated gene
   silencing

   (A) Quantification of intracellular siRNA spots measured from imaging
   at various time points post GalNAc-HPRT1 siRNA treatment. (B) Schematic
   diagram of experimental design to evaluate the impact of retrograde
   Golgi-ER transport on siRNA-mediated gene silencing. (C) dPCR
   measurements of HPRT1 expression in Hep3BCas9 or Hep3BRAB18KO cells
   treated with GalNAc-HPRT1 siRNA (8172) in the presence or absence of
   GCA (n = 3). (D) dPCR measurements of HPRT1 expression in Hep3BCas9 or
   Hep3BRAB18KO cells transfected (Lipofectamine) with unconjugated HPRT1
   siRNA (39986) in the presence or absence of GCA (n = 3). The
   unconjugated HPRT1 siRNA was directly added to culture medium for
   groups without Lipofectamine labels. All dPCR samples were measured
   4 days post siRNA treatments. All dPCR results shown were normalized by
   TBP and no siRNA (PBS only)-treated control groups. Asterisk indicates
   significant difference (∗∗ p < 0.01) based on one-way ANOVA test. All
   error bars represent SD.

Retrograde transport positively regulates siRNA-mediated gene silencing

   A previously published study showed RAB18 mutants enhance retrograde
   Golgi-ER transport of the COPI-independent cargo.[141]^7 To study the
   biological mechanisms behind enhanced siRNA-mediated gene silencing in
   RAB18-knockout cells, we investigated whether retrograde Golgi-ER
   transport regulates siRNA-mediated gene silencing. Golgicide A (GCA) is
   a powerful Golgi inhibitor that can specifically inhibit assembly of
   COPI-coated vesicles and lead to inhibition of COPI-dependent
   retrograde Golgi-ER transport.[142]^14^,[143]^15 As illustrated in
   [144]Figure 5B, RAB18 inhibits COPI-independent retrograde transport
   and GCA treatment inhibits COPI-dependent retrograde transport. If
   retrograde transport positively regulates siRNA function, siRNA
   conjugates should exhibit the weakest silencing effect of the target
   gene in Hep3BCas9 cells treated with GCA (group 2). Conversely, siRNA
   conjugates should exert the strongest silencing of the target gene in
   Hep3BRAB18KO cells and without GCA treatment (group 3), since both
   COPI-independent and COPI-dependent retrograde transport pathways are
   unobstructed under this condition. To test this hypothesis,
   siRNA-mediated gene silencing of GalNAc-HPRT1 siRNA was validated in
   Hep3BCas9 and Hep3BRAB18KO cells pre-treated with or without a
   non-toxic dose of GCA (50 nM). As predicted, GalNAc-HPRT1 siRNA
   generated the strongest target gene silencing in Hep3BRAB18KO cells
   without GCA treatment (group 3, IC50 = 2.69 nM), and the weakest target
   gene silencing in Hep3BCas9 cells with GCA treatment (group 2, IC50
   N/A) ([145]Figure 5C). Similar levels of HPRT1 silencing were observed
   in Hep3BCas9 cells without GCA treatment (group 1, IC50 = 52.31 nM) and
   Hep3BRAB18KO cells with GCA treatment (group 4, IC50 = 30.40 nM), since
   only one type of retrograde transport was blocked under these
   conditions. However, when GalNAc-mediated delivery of GalNAc-HPRT1
   siRNA was replaced with liposome-mediated delivery of unconjugated
   HPRT1 siRNA (compound no. 39986), and using otherwise similar
   experimental conditions to those described above, no difference in the
   potency of HPRT1 silencing was observed across groups ([146]Figure 5D).
   These studies suggest retrograde Golgi-ER transport positively
   regulates siRNA-mediated gene silencing. However, as demonstrated
   previously,[147]^1 RAB18 or retrograde Golgi-ER transport impacts the
   activity of siRNA delivered by GalNAc but not liposome-mediated
   transfection.

RAB18 knockout changes siRNA colocalization with LDs

   To further elucidate the mechanism by which RAB18 regulates siRNA
   trafficking, we conducted imaging-based assays using the anti-siRNA
   antibody described above to examine siRNA spatial distribution patterns
   in Hep3BCas9 cells and Hep3BRAB18KO cells. As shown in [148]Figure 6A,
   the colocalization of siRNA with subcellular structures was visualized
   using LipidTox to stain for LDs and fluorescent antibodies for AGO2
   (Argonaute 2), calreticulum, calnexin, GM130, LAMP2, RAB9, RAB4, and
   DCP1A ([149]Figures S6 and [150]S7). Hep3BCas9 and Hep3BRAB18KO cells
   were treated with 1.5 μM GalNAc-HPRT1 siRNA for 1 or 2 days, stained,
   and then imaged. Since only about 15–30 siRNA spots were detectable per
   cell ([151]Figure 5A), identifying siRNA colocalization events with any
   cellular structure was infrequent. To improve identification and
   quantification, we developed an image-analysis procedure to apply a
   deep-learning-based approach for cell segmentation ([152]Figure S8) and
   to calculate the Manders coefficient[153]^16 to indicate siRNA
   colocalization changes. Of the cell structures examined, the most
   noteworthy difference was colocalization of siRNA with LDs after 2 days
   of siRNA treatment ([154]Figures 6B and [155]S9), where a 20% reduction
   was observed in Hep3BRAB18KO compared to parental cells. To further
   validate this finding, label-free live-cell
   holotomographic[156]^17^,[157]^18 imaging of LDs and siRNA was
   conducted in Hep3BCas9 and Hep3BRAB18KO cells. As shown in
   [158]Figure 6C, colocalization analysis on this live-cell imaging
   revealed a similar reduction of siRNA-LD colocalization in Hep3BRAB18KO
   cells compared to parental cells after 2 days of siRNA treatment. A
   reduction of siRNA-LD colocalization might indicate either increased
   consumption or enhanced degradation of siRNA. The impact of RAB18
   knockout and overexpression on LDs was shown in previous studies. To
   summarize, overexpression of wild-type (WT) RAB18 resulted in small,
   dispersed LDs in WT as well as RAB18 KO cells.[159]^6^,[160]^19 RAB18
   KO interferes with LD catabolism, resulting in reduced LD numbers,
   enlarged LD size, and impaired fatty acid release.[161]^19 To confirm
   if RAB18 knockout in Hep3B cells has the same effect on LDs, the number
   and size of LDs in Hep3BCas9 and Hep3BRAB18KO cells were quantified.
   Consistent with previous findings, LDs were both reduced in number and
   enlarged in size when comparing Hep3BRAB18KO cells to the parental
   Hep3BCas9 cells ([162]Figures 6D and 6E).

Figure 6.

   [163]Figure 6
   [164]Open in a new tab

   Imaging intracellular siRNA

   (A) The table on the left lists the antibodies and corresponding
   cellular markers evaluated by imaging. On the right, confocal
   microscopy images show siRNA co-staining with LipidTox (images of the
   other cellular markers listed are shown in [165]Figures S6 and
   [166]S7). Scale bars represent 5 μm. (B) Manders coefficient analysis
   of the images taken from Hep3BCas9 and He3BRAB18KO cells treated with
   1.5 μM GalNAc-HPRT1 siRNA 2 days post siRNA treatment. The Manders
   coefficient scores reflect the frequency of siRNA colocalization with
   the given cellular markers. Compared to all other cellular markers,
   notable siRNA colocalization difference between Hep3BCas9 and
   Hep3BRAB18KO cells was detected for LipidTox-stained LDs. Each imaging
   experiment was repeated five times. (C) Pearson coefficient analysis on
   the refractive and fluorescence images of live Hep3BCas9 and
   He3BRAB18KO cells treated with 1.5 μM AF647 conjugated GalNAc-HPRT1
   siRNA 1 and 2 days post siRNA treatment. Each imaging experiment was
   repeated six times. (D) Quantification of LD count on day 1 and day 2
   post 1.5 μM GalNAc-HPRT1 siRNA treatment comparing Hep3BCas9 and
   Hep3BRAB18KO cells. Each imaging experiment was repeated eight times.
   (E) Summary of LD size analysis on day 1 or day 2 post 1.5 μM
   GalNAc-HPRT1 siRNA treatment comparing Hep3BCas9 and Hep3BRAB18KO
   cells. Each imaging experiment was repeated eight times. All error bars
   represent SD. Asterisk indicates significant difference (∗ p < 0.05; ∗∗
   p < 0.01) based on one-way ANOVA test.

LDs regulate siRNA-mediated gene silencing

   The RNA-seq data obtained from the Rab18 siRNA trigger in vivo study
   demonstrates Rab18 knockout upregulates genes related to lipid
   metabolic processes ([167]Figures 3F and 3H), suggesting a role for
   RAB18/Rab18 in regulating lipid biology. Our imaging study results add
   to this, showing that RAB18 expression, or lack thereof, not only
   changes the number and size of LDs but also impacts siRNA
   colocalization with LDs ([168]Figures 6B–6E). Therefore, to further
   elucidate the function of LDs in regulating siRNA-mediated gene
   silencing, we added oleic acid, a mono-unsaturated omega-9 fatty acid,
   to culture medium to enhance the intracellular accumulation of
   LDs[169]^20 or triacsin C, a long fatty acyl-coenzyme A (CoA)
   synthetase inhibitor to decrease LD content.[170]^20 Hep3BCas9 and
   Hep3BRAB18KO cells were incubated with non-toxic dosages of oleic acid
   (100 μM) or triacsin C (1 μM) for 24 h, followed by addition of
   GalNAc-HPRT1 siRNA in the presence of oleic acid or triacsin C for
   4 days. The effect on target gene HPRT1 expression was measured across
   treatment groups and compared with a non-compound-treated negative
   control through dPCR. Triacsin C treatment in both Hep3BCas9 and
   Hep3BRAB18KO cells reduced siRNA-mediated silencing of the HPRT1 gene
   ([171]Figures 7A and 7B). Conversely, oleic acid treatment in both
   Hep3BCas9 and Hep3BRAB18KO cells had little or no impact on
   siRNA-mediated gene silencing of the HPRT1 gene ([172]Figures 7C and
   7D). These results demonstrate that LDs might positively regulate
   siRNA-mediated gene silencing, and, collectively, our data suggest this
   is a RAB18-mediated mechanism.

Figure 7.

   [173]Figure 7
   [174]Open in a new tab

   The effects of oleic acid and triacsin C on regulating siRNA-mediated
   gene silencing and the prerequisite of RAB18 knockout/knockdown in
   enhancing siRNA-mediated gene silencing

   dPCR measurements of HPRT1 expression in (A) Hep3BCas9 cells or (B)
   Hep3BRAB18KO cells treated with GalNAc-HPRT1 siRNA and triacsin C.
   Cells were pre-treated with the indicated concentration of triacsin C
   for 24 h prior to siRNA treatment. dPCR measurements of HPRT1
   expression in (C) Hep3BCas9 cells or (D) Hep3BRAB18KO cells treated
   with GalNAc-HPRT1 siRNA and oleic acid. Cells were treated with the
   indicated concentration of oleic acid for 24 h prior to siRNA
   treatment. (E) dPCR measurements of HPRT1 expression in GalNAc-HPRT1
   siRNA-treated Hep3BCas9 and Hep3BRAB18KO cells with or without
   co-treatment of GalNAc-RAB18 siRNA or the negative control GalNAc-PPIB
   siRNA. All dPCR samples were measured 4 days post siRNA treatments. All
   dPCR results shown were normalized by TBP and no-siRNA (PBS
   only)-treated control groups with n = 3 for each experiment. All error
   bars represent SD.

RAB18 knockout or knockdown is a prerequisite to enhance siRNA-mediated gene
silencing

   Herein we have demonstrated that a reduction in RAB18, either by gene
   knockout or siRNA-mediated gene knockdown, enhances siRNA-mediated gene
   silencing. To further validate if RAB18 knockout or knockdown is
   necessary for enhancing siRNA-mediated gene silencing, Hep3BCas9 cells
   were co-treated with either the same dose of GalNAc-HPRT1 siRNA and
   GalNAc-conjugated siRNA targeting RAB18 (GalNAc-RAB18 siRNA, compound
   no. 64131) or the same dose of GalNAc-HPRT1 siRNA and the control
   GalNAc-PPIB siRNA ([175]Table S1). Compared with Hep3BCas9 cells
   treated with only GalNAc-HPRT1 siRNA, co-treatment with either
   GalNAc-RAB18 siRNA or GalNAc-PPIB siRNA failed to change the silencing
   potency of GalNAc-HPRT1 siRNA ([176]Figure 7E). Therefore, to enhance
   the silencing effect of siRNA conjugates, knockout or knockdown of the
   RAB18 gene is required prior to treatment with siRNA conjugates.

Discussion

   To reveal the cellular factors limiting delivery of siRNA therapeutics,
   we performed a genome-wide pooled CRISPR-Cas9 screen based on delivery
   of GalNAc-conjugated siRNA targeting HPRT1 in the human hepatocellular
   carcinoma line Hep3B.[177]^1 Multiple candidate genes, when knocked
   out, enhanced siRNA efficacy in Hep3B cells. However, by performing a
   secondary, arrayed CRISPR screen[178]^1 and generating knockout cell
   lines of a few top candidate genes (data not shown), the most
   significant enhancement of siRNA-mediated gene silencing was observed
   in RAB18 KO cells. As one of the 20 most highly conserved RAB GTPases
   present in the last eukaryotic common ancestor of both the plant and
   animal kingdoms,[179]^21^,[180]^22 the protein encoded by the RAB18
   gene is ubiquitously expressed across a majority of cell and tissue
   types. The RAB18 protein switches between GDP-bound off and GTP-bound
   on forms. The active GTP-bound form of RAB18 can recruit different sets
   of downstream effectors to the plasma membrane to mediate vesicle
   formation, movement, tethering and fusion,[181]^6^,[182]^11^,[183]^23
   growth of LDs,[184]^5^,[185]^6 formation of endoplasmic reticulum-LDs
   contacts,[186]^23 maintaining endoplasmic reticulum structure,[187]^11
   inhibition of COPI-independent retrograde trafficking from Golgi to
   ER,[188]^7 regulation of secretory granules[189]^8 and
   peroxisomes,[190]^9 and regulation of apical
   endocytosis/recycling.[191]^24 RAB18 protein also plays a key role in
   eye and brain development and neurodegeneration.[192]^25^,[193]^26

   The transition of RAB18 between the GDP- and GTP-bound conformations is
   governed by two classes of regulatory proteins: the guanine-nucleotide
   exchange factors (GEFs), which mediate the exchange of bound GDP for
   GTP, and the GTPase-activating proteins (GAPs), which stimulate the GTP
   hydrolysis activity of their substrate GTPase(s). RAB3GAP1 and RAB3GAP2
   form complexes with GAP and act as a RAB18 GEF regulator promoting
   RAB18 activation.[194]^11 The TBC1D20 protein exhibits modest GAP
   activity toward RAB18 protein and functions to promote RAB18
   dissociation from the ER membrane into the cytosol.[195]^27
   Loss-of-function mutations in RAB18, RAB3GAP1, RAB3GAP2, and TBC1D20
   have been identified in patients with the human neurological and
   developmental disorder Warburg micro syndrome, a rare autosomal
   recessive multisystem disorder primarily characterized by ocular and
   neurodevelopmental
   defects.[196]^25^,[197]^26^,[198]^27^,[199]^28^,[200]^29 RAB18 activity
   could also be modulated by regulating RAB3GAP1, RAB3GAP2, and TBC1D20
   expression. The potential of these genes on directing RAB18 activity is
   worth exploring in future studies.

   In the study described herein, we show the effect of RAB18
   knockout/knockdown on siRNA-mediated gene silencing recapitulates in
   both hepatic cells and extrahepatic cells and translates from in vitro
   systems to in vivo models. As shown in this study and in previous
   published work,[201]^1 we observed no growth defect in RAB18-knockout
   Hep3B[202]^1 or T47D cells ([203]Figure S2B). In the current study,
   RAB18/Rab18 siRNA was conjugated to GalNAc, a sugar molecule that
   selectively binds a surface protein, ASGPR, primarily expressed on
   hepatocytes in the liver. GalNAc-mediated delivery of siRNA is a
   well-established and clinically efficacious means for specifically
   inhibiting protein translation in vivo. Three of the four Rab18 siRNA
   triggers had no impact on mouse body weight or liver weight, or serum
   liver transaminase levels ([204]Figures S3 and [205]S5), nor were there
   any signs of toxicity in the RNA-seq analysis ([206]Figures 3 and
   [207]S4). Thus, the deleterious readouts observed with Rab18 siRNA
   trigger 64137-1 appear to be trigger specific and not the result of
   general Rab18 knockdown ([208]Figures S3B, S3D, and [209]S4E).
   Furthermore, Rab18-deficient mice are reported to be viable and
   fertile, although they did exhibit major ocular and neurological
   abnormalities that are associated with Warburg micro syndrome.[210]^13
   However, detailed necropsy of Rab18 KO mice with their littermate
   controls, including more than 20 tissues across the body, and the
   brain, showed no substantial differences.[211]^13 Thus, the authors
   concluded the observed phenotypes in Rab18 KO mice were not caused by
   gross disruption of development.[212]^13 Therefore, we deduce it might
   be achievable to enhance siRNA-mediated gene silencing through
   transient inhibition of RAB18 function without causing significant
   toxicity.

   The success of silencing target gene in hepatocytes by siRNA delivered
   through GalNAc conjugation heavily relies on the cell-type specificity,
   abundant expression (∼500,000 copies per cell),[213]^30 and rapid
   recycling rate (∼15 min) of ASGPR.[214]^31 These factors enable a
   substantial number of siRNAs to bind and silence target RNA molecules,
   even when <1% of the siRNAs are estimated to escape from
   endosomes.[215]^3 It is challenging to find cell-type-specific
   receptors with properties like ASGPR for extrahepatic cells. However,
   as well described, RNA silencing is an efficient process once the siRNA
   is loaded into RISC, and it only requires about 10–110 siRISC (siRNA
   incorporated into RNA-induced silencing complexe) molecules per cell to
   achieve approximately 50% silencing of the target gene.[216]^32 Hence,
   if siRNA could be trafficked into RISC in a more efficient manner, it
   may be that receptors with sub-optimal qualities would prove sufficient
   for siRNA delivery and efficacy in extrahepatic cells. As demonstrated
   in this study, even though RAB18 knockout reduces the detected amount
   of intracellular siRNA ([217]Figure 5A), which could indicate increased
   consumption of siRNA, siRNA-mediated gene silencing is enhanced,
   suggesting a more organized and efficient intracellular trafficking
   route for siRNAs in the absence of RAB18. To elucidate if RAB18
   knockout changes siRNA intracellular trafficking and its spatial
   distribution with respect to different subcellular structures, confocal
   microscopy was used to image multiple intracellular targets, including
   AGO2, ER surface, ER lumen, P bodies, Golgi, early endosomes, late
   endosomes, lysosomes, and LDs in both Hep3BCas9 and Hep3BRAB18KO cells
   ([218]Figures 6, [219]S6, and [220]S7). The only noteworthy result was
   a 20% reduction of siRNA colocalization with LDs after 2 days of siRNA
   treatment in Hep3BRAB18KO compared to Hep3BCas9 cells ([221]Figure 6B),
   implicating the participation of LDs in regulating siRNA activity.
   Imaging analysis at higher resolution would be needed to evaluate the
   complex nanoscale interactions of siRNA with specific subcellular
   structures.

   Herein, we have demonstrated that both retrograde Golgi-ER transport
   and LDs positively regulate siRNA-mediated gene silencing
   ([222]Figures 5B, 5C, and [223]7). As rough ER is a central nucleation
   site of siRNA-mediated gene silencing,[224]^32 it is not surprising to
   observe retrograde Golgi-ER transport to positively regulate the
   siRNA-mediated gene silencing. Although RAB18 mutants have been shown
   to specifically enhance retrograde Golgi-ER transport of
   COPI-independent cargo, but not COPI-dependent cargo from the Golgi to
   ER,[225]^7 more detailed studies would be needed to understand how the
   RAB18 gene regulates this process at the molecular level and whether
   other key genes in retrograde Golgi-ER transport, such as ARF1,[226]^33
   are involved.

   Using triacin C to reduce intracellular LD content, we observed a
   correlation between siRNA-mediated gene silencing and decreased
   cellular LD content ([227]Figures 7A and 7B). These results suggest LDs
   regulate siRNA-mediated silencing efficacy, perhaps through acting as
   siRNA reservoirs at the end of siRNA trafficking route. Since the
   number of siRNA molecules deposited into LDs might be pre-determined by
   multiple upstream trafficking events and independent of the size and
   number of LDs, simply enhancing LD size and number by oleic acid
   treatment may not be sufficient to enhance siRNA-mediated gene
   silencing. Therefore, increasing intracellular LD content with oleic
   acid had little or no impact on siRNA-mediated gene silencing
   ([228]Figures 7C and 7D). Genes regulating LD biogenesis or autophagy,
   such as DGAT1,[229]^34 could also regulate LD size. Future studies
   could include knocking out, or knocking down, such genes to further
   elucidate the connection between LDs and siRNA activity. LDs are
   cytosolic lipid storage organelles important for cellular lipid
   metabolism, energy homeostasis, cell signaling, and inflammation.
   Lipophagy is a type of autophagy; it is a process by which LDs are
   degraded in lysosomes. Both starvation and acute lipid stimulus
   increase autophagic sequestration of LDs and trigger their degradation
   in lysosomes.[230]^35^,[231]^36 The RAB18 gene has been shown to
   regulate LD metabolism[232]^37 and the formation of ER-LD
   contacts.[233]^23 As demonstrated in this study and previously[234]^19
   ([235]Figures 6D and 6E), there are fewer and larger LDs in
   RAB18-knockout cells. Additionally, knockout of RAB18 interferes with
   LD turnover. In WT cells, starvation triggers a time-dependent
   reduction in LD number and size; this process is repressed in
   RAB18-knockout cells.[236]^19 Hence, if LDs serve as temporal
   reservoirs for siRNAs, the siRNAs may be degraded when LDs get engulfed
   by autophagosomes. However, in RAB18-knockout cells, lipid metabolism
   and lipophagy are impaired, causing a low turnaround rate for LDs. This
   may provisionally protect siRNAs from degradation and provide more
   opportunities for siRNAs to escape into the cytosol.

   The optimal delivery of siRNA into target cells needs to overcome a
   variety of challenges, including degradation by nucleases, cell
   membrane permeability, endosomal escape, and intracellular trafficking.
   We previously demonstrated that knocking out the RAB18 has no
   detectable effect on cell membrane permeability.[237]^1 The studies
   described herein demonstrate that knocking out RAB18 enhances siRNA
   silencing through enhancing siRNA retrograde Golgi-ER transport and
   positively modulating the LD-mediated regulation of siRNA activities.
   Therefore, RAB18 might regulate siRNA silencing by altering
   intracellular trafficking. The identification and study of factors
   regulating siRNA degradation, cell membrane permeability, and endosomal
   escape might elucidate further means to enhance siRNA silencing.

   In summary, using a variety of in vitro systems, we have demonstrated
   that knocking out RAB18 enhances the effect of siRNA-mediated gene
   silencing in both hepatic and extrahepatic cells. Furthermore, we
   translated this finding to an in vivo model. Altogether, our data point
   to a mechanism by which RAB18 regulates the efficacy of siRNA-mediated
   gene silencing in extrahepatic cells. This role for RAB18 in RNAi may
   facilitate the discovery of more cell-type-specific receptors that, in
   the presence of RAB18 expression, may be sub-optimal for delivering
   siRNA cargo and achieving efficacious silencing of target genes. Thus,
   understanding the potential of gene silencing in the presence or
   absence of concomitant RAB18 expression may unleash new possibilities
   in the development of extrahepatic siRNA.

Materials and methods

Cell line and culture condition

   Hep3B, T47D, A498, MCF7, and COLO205 cells were purchased from ATCC
   ([238]https://www.atcc.org/). MiaPaCa-2, HT-29, and NCIH1299 were
   obtained from the Amgen oncology group at South San Francisco. Hep3B
   cells were cultured in Eagle’s minimum essential medium (ETCC; from
   ATCC, catalog no. 30-2003) + 10% fetal bovine serum (FBS; HyClone,
   catalog no. SH30071.03). Hep3Bcas9 cells were cultured in EMEM + 10%
   FBS + 10 μg/mL blasticidin. Hep3BRAB18KO cells were cultured in EMEM +
   10% FBS + 10 μg/mL blasticidin + 0.5 μg/mL puromycin. T47D cells were
   cultured in Roswell Park Memorial Institute (RPMI; Gibco, catalog no.
   11875085) + 10% FBS + 7.4 μg/mL human recombinant insulin (Life
   Technologies, catalog no. 12585-014). A498 cells were cultured in
   EMEM + 10% FBS. The culture condition for MCF7 cells is EMEM + 10%
   FBS + 0.01 mg/mL human recombinant insulin. Both NCIH1299 and COLO205
   cells were cultured in RPMI + 10% FBS. HT-29 cells were cultured
   McCoy’s 5A (Gibco, catalog no. 16600108) + 10% FBS. MiaPaCa-2 cells
   were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Gibco,
   catalog no. 11966-025) + 10% FBS + 2.5% horse serum (Cytiva, catalog
   no. SH30074.04).

In vitro siRNA/ASO treatment, RNA isolation, and dPCR

   All in vitro siRNA treatments were performed in a 96-well plate format.
   To deliver GalNAc-siRNA conjugates, GalNAc-ASO conjugates, and the
   unconjugated siRNA to various cell lines, 50 μL of cell suspension was
   first added to each well to achieve desired cell densities (Hep3B and
   its derived cells, 20,000 cells/well; T47D and its derived cells,
   20,000 cells/well; A498 and its derived cells, 28,000 cells/well; MCF7
   and its derived cells, 13,000 cells/well; COLO205 and its derived
   cells, 15,000 cells/well; MiaPaCa-2 and its derived cells, 41,000
   cells/well; HT-29 and its derived cells, 13,000 cells/well; NCIH1299
   and its derived cells, 17,000 cells/well). The siRNA/ASO molecules were
   then diluted in the corresponding tissue culture medium to twice the
   desired final concentration. Then 50-μL siRNA/ASO solutions were added
   to the corresponding wells and co-incubated with cells for 4 days in a
   37°C tissue culture incubator before RNA isolation. Total RNA isolation
   from cells was performed in a 96-well format by using the MagMAX
   mirVana Total RNA Isolation Kit (Applied Biosystems, catalog no.
   A27828) and KingFisher Flex automated extraction and purification
   system (Thermo Fisher Scientific) following the manufacturer’s
   instructions. The cDNAs were then synthesized from total RNA samples
   using the Applied Biosystems High Capacity cDNA Reverse Transcription
   kit (catalog no. 4368813) by following the user manual. All dPCR
   reactions were assembled using a QIAcuity Probe PCR Kit (QIAGEN,
   catalog no. 250101) as per the user manual. After proper mixing and
   centrifugation, all dPCR reactions were then transferred to QIAcuity
   8.5K 96-well nanoplates (QIAGEN, catalog no. 250021). The nanoplates
   were then loaded onto the one-place digital PCR instrument QIAcuity
   Eight (QIAGEN) for performing dPCR. The dPCR copy number readings
   (copies/20 μL) of both target gene (HPRT1/Hprt1 or RAB18/Rab18) and
   housekeeping gene TBP/Tbp were recorded for each well. The normalized
   target gene mRNA level was calculated by dividing the dPCR reading of
   the target gene by the dPCR reading of TBP/Tbp taken from the same
   well. The resulting number of siRNA-treated samples was further divided
   by the number of no-siRNA-treatment samples to obtain the percentage
   reading of the target gene mRNA level. The predesigned primer/probe for
   dPCR assays were obtained from Integrated DNA Technologies with 3.6:1
   primer to probe ratio. The assay IDs of primer/probe used for
   quantifying the human genes are as follows: HPRT1 Hs.PT.39a.22214821,
   RAB18 Hs.PT.58.21332040, and housekeeping TBP Hs.PT.58.19489510. The
   assay IDs of primer/probe used for quantifying the mouse genes are as
   follows: Hprt1 Mm.PT.39a.22214828, Rab18 Mm.PT.58.10341383, and
   housekeeping Tbp Mm.PT.39a.22214839.

RAB18 overexpression

   RAB18-overexpression vector ([239]Figure S1A) was synthesized by Azenta
   Life Sciences and packaged into lentivirus by Cellecta. The lentivirus
   was then used to transduce Hep3BCas9 and Hep3BRAB18KO cells at
   multiplicity of infection 1 (MOI_1). The mCherry-positive cells were
   sorted out twice after transduction to enrich RAB18-overexpressed
   cells.

Generate RAB18-knockout cells

   To generate T47DRAB18KO cells, a TransEDIT CRISPR Cas9 nuclease
   expression lentivirus (pCLIP-Cas9-Nuclease-EFS-Blast) ordered from
   TransOMIC Technologies (catalog no. NC0956087) was simultaneously
   transduced into Hep3B cells at MOI_1 along with a lentiviral gRNA
   vector targeting RAB18 (SIGMA vector: U6-gRNA: PGK-puro-2A-tagBFP, gRNA
   target sequence: GCTATTATAGAGGTGCACAGGG) at MOI_10. Three days post
   transduction, the double-positive transduced T47D cells were then
   enriched by 10 μg/mL blasticidin + 0.5 μg/mL puromycin for 7 days to
   get T47DRAB18KO cells. The editing efficacy on the RAB18 gene was then
   validated by ICE analysis. RAB18-knockout A498, MCF7, COLO205,
   MiaPaCa-2, HT-29, and NCIH1299 cells were generated by
   co-electroporating 250 ng of Cas9 Nuclease V3 protein (Integrated DNA
   Technologies, catalog no. 1081058) and three sgRNAs (100 pmol each)
   targeting the RAB18 gene ([240]Table S4) into 50,000 target cells
   through a Neon Electroporation device (Invitrogen) and a Neon
   Transfection System 10 μL Kit (Invitrogen, catalog no. MPK1025)
   following the user manual. Cells were then cultured for 7 days before
   ICE analysis, western blot analysis, and siRNA treatment.

ICE analysis

   Cas9 protein only or Cas9 + RAB18 sgRNA RNP complex electroporated
   cells (1 × 10^6 cells) were harvested by trypsinization, and gDNA was
   harvested using the DNeasy Blood and Tissue kit (QIAGEN, catalog no.
   69504). The RAB18 exon two genomic locus was amplified from 50 ng of
   gDNA using forward primer 5′ TTGTTTCTTCCATTTGGATCAG 3′ and reverse
   primer 5′ TGCCACTATATCCTGTGAGG 3′ and Platinum PCR Supermix High
   Fidelity (Invitrogen, catalog no. 12532016). The resulting PCR products
   were Sanger sequenced using primer 5′ ACACAAGGCAACTAGGAACAA 3′ and
   analyzed for knockout efficacy using the ICE analysis tool (Synthego,
   [241]https://ice.synthego.com/#/).

RAB18 western blot analysis

   Total protein lysate was isolated from Cas9 protein only or Cas9 +
   RAB18 sgRNA RNP complex electroporated cells (5 × 10^6 cells) using
   RIPA buffer (Thermo Scientific, catalog no. J63306.AK) containing 1×
   Halt Protease and Phosphatase Inhibitor Cocktail (Thermo Scientific,
   catalog no. 78440), then clarified by sonication before centrifugation
   to remove cell debris. Then 50 mg of each protein lysate was size
   separated on NuPage 4%–12% Bis-Tris gels, then transferred to
   nitrocellulose membranes and blocked with 3% BSA in 1× TBS-Tween 20 for
   1 h at room temperature before probing with rabbit anti-Rab18 (1:1,000,
   Bioss, catalog no. bsm-51333M) and mouse anti-GAPDH (1:5,000, Cell
   Signaling Technology, catalog no. 2118S) in 3% BSA-TBS-Tween 20
   overnight at 4°C. Membranes were washed three times with 1× TBS-Tween
   20, then incubated with IRDye 680RD goat anti-mouse immunoglobulin G
   (IgG) (1:5,000, Li-cor, catalog no. 926-68070) and IRDye-800CW goat
   anti-rabbit IgG (1:5,000, Li-cor, catalog no. 926–32211) in 5% non-fat
   dry milk in TBS-Tween 20 for 1 h at room temperature. Membranes were
   washed three times with 1× TBS-Tween 20 and then imaged on a Li-cor
   Odyssey 9120 imager.

Human iPSC-derived cardiomyocyte culture for RAB18 siRNA knockdown study

   iCell Cardiomyocytes[242]^2 were thawed according to the manufacturer’s
   instructions (FUJIFILM Cellular Dynamics, catalog no. R1017) directly
   onto fibronectin (Stem Cell Technologies, catalog no. 07159) and
   gelatin (Stem Cell Technologies, catalog no. 07903)-coated 96-well
   plates at a density of 50,000 cells per well. Four hours later, the
   medium was switched to CDI maintenance medium (FUJIFILM Cellular
   Dynamics, catalog no. M1003). iCell Cardiomyocytes[243]^2 were fed with
   fresh CDI maintenance medium every 2 days. On day two, iCell
   Cardiomyocytes[244]^2 were transfected with either NT siRNA or RAB18
   siRNA at the final concentration of 10 nM using Lipofectamine RNAiMAX
   Transfection Reagent (Invitrogen, catalog no. 13778075) following the
   manufacturer’s instructions. The medium was switched to fresh CDI
   maintenance medium 24 h later. On days 5 and 7, either NT siRNA or
   HPRT1 siRNA at 2 μM were diluted in fresh CDI maintenance medium and
   added to iCell Cardiomyocytes[245]^2 culture. On day 9, the cells were
   collected for RNA extraction.

Animals

   All animal procedures described herein were approved by the
   Institutional Animal Care and Use Committee (IACUC) of Charles River
   Accelerator and Development Lab (CRADL, AAALAC accredited) and cared
   for in accordance with the Guide for the Care and Use of Laboratory
   Animals, Eighth Edition (National Research Council, US). Mice were
   group housed in a climate-controlled room at 22°C ± 2°C with a 12-h
   light, 12-h darkness cycle (0600–1800 h). Animals had ad libitum access
   to a regular chow diet (Envigo, catalog no. 2920X) and were caged in
   irradiated Innovive Disposable IVC Rodent Caging Systems (San Diego,
   CA) with pre-filled sterile and acidified water bottles. Eight- to
   10-week-old, pathogen-free female C57BL/6 mice (Charles River
   Laboratories) were used for all in vivo studies.

In vivo studies

   Prior to study initiation, mouse body weights were collected,
   randomized by weight, and then sorted into experimental groups. At
   study initiation, mice received a subcutaneous injection of siRNA into
   the left abdomen region; all siRNAs (0.5 mmol/L) were diluted in
   Dulbecco’s Phosphate Buffered Saline (DPBS; Thermo Fisher Scientific,
   catalog no. 14190-136) to a concentration of 0.5, 1, or 3 mg/kg of
   animal, and adjusted to a volume of 160 μL per animal. Control vehicle
   groups were administered equal volumes of DPBS. Mice were harvested
   after 14 days or as indicated in the corresponding figures, reinjected
   with another round of siRNA or vehicle, and then harvested 14 days
   after the second dose. siRNAs are listed and detailed in [246]Table S1.

   At study termination, mice were weighed and then anesthetized with
   isoflurane. From each animal, blood was collected via intracardiac
   puncture using a 1-mL 25-gauge Tuberculin syringe and transferred to
   Microtainer SST tube (Molecular Devices, catalog no. 365967). Mice,
   still under deep anesthesia, were then euthanized by a secondary
   physical method and the whole liver was collected and weighed; the left
   lobe was immediately snap-frozen in liquid nitrogen for mRNA isolation
   and analysis. Filled SST tubes stood at room temperature to clot for a
   minimum of 30 min and then spun, at room temperature, for 5 min at
   10,000 rpm in a centrifuge. Serum was then transferred to a 96-well
   DeepWell plate (Axygen, catalog no. P-2ML-SQ-C), sealed, frozen, and
   stored at −80°C.

Serum analysis

   Frozen serum was sent to IDEXX BioAnalytics (West Sacramento, CA) for
   determination of the following clinical chemistry endpoints: alanine
   aminotransferase (ALT), aspartate aminotransferase (AST), and alkaline
   phosphatase (ALP). All data reported in this manuscript came directly
   from the IDEXX report.

RNA isolation and analysis

   For the in vivo study, total RNA samples were isolated from frozen
   liver samples by using a RNeasy Mini Kit (QIAGEN, catalog no. 74104).
   The concentration and quality of the isolated RNA samples were assessed
   and determined on Agilent 4200 TapeStation system and had RNA integrity
   numbers more than 8.6. Total RNA (500 ng) was used to prepare a cDNA
   library by using a protocol modified from the Illumina TruSeq Stranded
   mRNA kit (Illumina, catalog no. 20020595). Briefly, after poly(A) + RNA
   selection, fragmentation, and priming, the fragmented RNA was
   transcribed to cDNA in the reaction of reverse transcription containing
   SuperScript III (Thermo Fisher Scientific- Invitrogen, catalog no.
   18080093) and, sequentially, the products of the first-strand cDNA
   synthesis reaction were converted to double- stranded cDNAs and
   subjected to end-repair, A-tailing, and adapter ligation by following
   the commercial instructions. The dUTP-containing strands were destroyed
   by digestion with USER enzymes (New England Biolabs). The constructed
   libraries were amplified using the PCR program: a denaturation step at
   98°C for 30 s, 15 cycles of 95°C for 10 s, 60°C for 30 s, 72°C for 30
   s, and a 72°C extension cycle for 5 min followed by a hold step at 4°C.
   Libraries were sequenced to a minimum depth of 35 million paired-end
   reads on an Illumina NovaSeq6000 with the 100- or 150-nt read length.
   The RNA-seq reads were aligned using the OSA (Omicsoft Studio Aligner)
   aligner integrated into Omicsoft Studio (v10; QIAGEN), using the human
   genome version GRCh38 and the GENCODE v24 gene model. Raw gene-level
   quantification was performed based on the Omicsoft implementation of
   RSEM (RNA-Seq by Expectation Maximization). Differential expression
   analysis was performed using DESeq2.[247]^38 For the comparative
   analysis, genes with an average normalized count less than 50 in both
   conditions were excluded. Gene Ontology (GO) pathway enrichment was
   tested using enrichR.[248]^39 In short, enrichR uses Fischer’s exact
   test to calculate p value to determine if a list of input genes
   significantly overlaps with an annotated gene set. The adjusted p value
   is calculated using the Benjamini-Hochberg method.

Immunofluorescence and imaging

   Antibodies used in imaging experiments include AGO2 (1:50, Abcam,
   catalog no. Ab57113), calreticulum (1:200, Novus Biologicals, catalog
   no. NBP1-47518SS), calnexin (1:50, Thermo Fisher, catalog no. MA3-027),
   GM130 (1:200 BD Biosciences, catalog no. 610822), LAMP1 (1:100 Abcam,
   catalog no. ab25631), LAMP2 (1:100 Abcam, catalog no. ab25631), RAB9
   (1:100, Thermo Fisher, catalog no. MA3-067), RAB4 (1:1000, Thermo
   Fisher, catalog no. MA5-17161), and DCP1A (1:100, Sigma Aldrich,
   catalog no. WH0055802M6). An in-house-developed rabbit polyclonal
   anti-siRNA antibody recognizing 2′OMe- and 2′F-containing
   oligonucleotides was used to stain siRNA molecules.

   Dissociated Hep3B and Hep3BRAB18KO cells from culture were seeded onto
   96-well PhenoPlate microplates (PerkinElmer) with siRNA and incubated
   for different time points: 6 h, 1 day, and 2 days. Cell seeding density
   was adjusted for incubation times. Cells were rinsed once with PBS and
   fixed with 4% paraformaldehyde (PFA) for 10 min at room temperature and
   washed twice in PBS for 5 min each. Cells were then permeabilized in
   PBS with 0.1% Triton X-100 for 25 min and rinsed in PBS twice. Cells
   were then blocked with 2% BSA for 30 min, rinsed three times, and
   incubated with primary antibodies and DRAQ5 (Thermo Scientific, catalog
   no. PI62251) at appropriate concentrations overnight at 4°C. The
   following day, cells were washed three times in PBS for 5 min each.
   Fluorescent secondary antibodies (1:200, Thermo Fisher, Alexa Fluor
   405/488, catalog no. A31556/catalog no. A11001) were added for 1 h at
   room temperature and washed three times in PBS. Cells were stained for
   lipid bodies with LipidTox (1:1,000, Thermo Fisher, catalog no.
   [249]H34475) and subsequently washed once with PBS and placed in a
   final solution of PBS. Confocal microscopy images were captured on an
   Opera Phenix High Content Screening system (PerkinElmer) via a spinning
   disk confocal at 63× objective magnification to provide a pixel size of
   ∼160 nm with z stack acquisition where desired.

Image analysis

   The Columbus high-volume image data storage and analysis system (v2.9)
   was used to configure a cell and spot segmentation and analysis
   sequence. This procedure is typically ad hoc, with the user being
   responsible for organizing a set of sequential “building blocks” to
   create a sample-specific image-processing pipeline. The following
   analysis sequence was used quantify and characterize siRNA and LDs.

Input image

   View original images from selected data.

Find cells

   Cellular area, splitting sensitivity, and common threshold parameters
   are tested and set manually based on cell type.

Select population

   Remove all segmented cells in contact with the image border. Filter out
   erroneously segmented objects and cells, including cell debris and
   background, based on object roundness and area parameter thresholds.

Find spots

   Find spots that are indicative of siRNA or LDs.

Calculate morphology properties

   Collect roundness and area parameters for all selected cells and LDs.

Calculate properties

   Quantify cell, LD, and siRNA fluorescent signal.

Define results

   Average and collect final cell counts, LD counts, area and morphology,
   and siRNA total counts and counts per cell.

   Colocalization image analysis was performed via Acapella software
   (PerkinElmer). Briefly, the DRAQ5 channel was used for nuclear and
   cytoplasmic segmentation. Colocalization was measured in the
   cytoplasmic region between siRNA and a specific cellular marker using
   the Manders overlap coefficient.[250]^16 The Manders coefficient was
   applied for its biological significance as a direct measure of
   co-occurrence, independent of signal proportionality. In the cases
   where Acapella provided inaccurate or incomplete cell segmentation, a
   custom colocalization analysis sequence was prepared in Python v3 to
   apply Cellpose 2.0.[251]^40 Cellpose deep-learning segmentation models
   were trained to readily accommodate limited cytoplasmic fluorescence
   from DRAQ5 and to enable consistent cell segmentation across the
   different localization patterns observed for the various cellular
   targets (AF488 staining). Cellpose training parameters, plots of the
   training loss across iterative rounds of training/image annotation, and
   example segmentation results are provided in [252]Figure S8. Cell and
   siRNA spot segmentation masks were applied to the image data and
   Manders colocalization coefficients were calculated.

Live-cell holotomography

   Label-free live-cell holotomography was performed on LDs and siRNA in
   Hep3B cells. Holotomographic microscopy is a label-free microscopic
   method used to determine the refractive indices of cells in three
   dimensions at high spatial and temporal resolution without
   phototoxicity. The characteristic refractive index of LDs clearly
   enables their segmentation from the remainder of the
   cell.[253]^17^,[254]^18 Label-free imaging of live Hep3B cells was
   performed as previously described.[255]^17 Briefly, Hep3BCas9 and
   Hep3BRAB18KO cells were seeded onto 96-well glass-bottom plates (MatTek
   Corporation) (10,000–12,000 cells per 90 μL of phenol red-free EMEM
   medium; Quality Biological) with vehicle, 1.5 μM GalNAc-HPRT siRNA
   (8172), or 1.5 μM AF647 conjugated GalNAc-HPRT siRNA and incubated in
   tissue culture incubator supplied with 5% CO[2] for 1 or 2 days. Prior
   to imaging at each time point, cells were rinsed and exchanged with
   85 μL of phenol red-free medium. The plate was briefly centrifuged to
   prevent meniscus formation and loaded on the pre-equilibrated (37°C, 5%
   CO[2]/95% hydrated air) Nanolive CX-A 3D Cell Explorer microscope
   equipped with class 1 low-power laser (λ = 520 nm, sample exposure
   2 mW/mm^2) and 60× magnification air objective (NA 0.8/WD 0.3 mm)
   (Nanolive, Tolochenaz, Switzerland). After 1 h of plate thermalization,
   refractive index and fluorescence images (CY5 channel) were acquired
   using the 3 × 3 GridScan mode (a field view of 236 × 236 μm). For LD
   characterization, cells and LDs in refractive images were segmented and
   quantified using the Smart Lipid Droplet Assay module (version 2.0) in
   the AI/ML (Artificial Intelligence/Machine Learning)-powered
   Nanolive.[256]^17^,[257]^18 For colocalization measurement between
   AF647 conjugated GalNAc-HPRT siRNA and LDs, refractive and fluorescence
   images were imported into Huygens Professional software (version 22.10)
   (Scientific Volume Imaging, Hilversum, the Netherlands). Colocalization
   analysis was performed with Huygens Colocalization Analyzer module
   using Gaussian minimum threshold estimation and the Pearson coefficient
   was used to measure the colocalization between AF647 conjugated
   GalNAc-HPRT siRNA and LDs.

GCA treatment

   The toxicity of GCA (Sigma Aldrich, catalog no. 345862-10MG) on cell
   viability was first evaluated by treating 20,000 Hep3BCas9 cells with
   25, 50, 100, and 200 nM GCA for 3 days in a 96-well plate format.
   Characterized as the highest non-toxic concentration (data not shown),
   50 nM GCA was then used for subsequent studies. To validate the impact
   of GCA treatment on siRNA-mediated gene silencing, 10,000 Hep3BCas9 or
   Hep3BRAB18KO cells were seeded together with 50 nM GCA in a 96-well
   plate for 24 h. Medium was then removed from the plate, fresh medium
   contained the desired concentration of siRNA, and 50 nM GCA was added
   to each well as describe above in the section “in vitro siRNA/ASO
   treatment, RNA isolation, and dPCR.” Two days post siRNA treatment, to
   ensure the constitutive function of GCA, without removing medium from
   the plate, 1 μL of 5 μM GCA solution was added to each well and mixed
   well. The plate was then incubated for another 2 days before RNA
   isolation, cDNA synthesis, and dPCR assay.

Oleic acid and triacsin C treatment

   The toxicity of oleic acid (Sigma Aldrich, catalog no. O1008-5G) and
   triacsin C (Sigma Aldrich, catalog no. T4540-1MG) on cell viability was
   first evaluated by treating 20,000 Hep3BCas9 cells with 20, 100, and
   200 μM oleic acid or 0.2, 1, and 2 μM triacsin C for 3 days in a
   96-well plate format. However, 200 μM oleic acid and 2 μM triacsin C
   led to toxicity (data not shown) and 20 μM oleic acid and 0.2 μM
   triacsin C had no obvious effect. Thus, 100 μM oleic acid and 1 μM
   triacsin C were used for the following studies. To validate the impact
   of oleic acid and triacsin C treatment on siRNA-mediated gene
   silencing, 10,000 Hep3BCas9 or Hep3BRAB18KO cells were seeded together
   with 100 μM oleic acid or 1 μM triacsin C in a 96-well plate for 24 h.
   Medium was then removed from the plate, and then fresh medium
   containing the desired concentration of siRNA and the same dosage of
   oleic acid or triacsin C was added back to the plate. The plate was
   then incubated for another 4 days before RNA isolation, cDNA synthesis,
   and dPCR assay.

Notification

   Except the in vivo studies and the related RNA-seq studies, all other
   experiments described in the manuscript were repeated more than twice.

Data and code availability

   The row data required to reproduce the above findings are available
   from the corresponding authors upon request.

Acknowledgments