A combinatorial approach for achieving CNS-selective RNAi

Abstract RNA interference (RNAi) is an endogenous process that can be harnessed using chemically modified small interfering RNAs (siRNAs) to potently modulate gene expression in many tissues. The route of administration and chemical architecture are the primary drivers of oligonucleotide tissue distribution, including siRNAs. Independently of the nature and type, oligonucleotides are eliminated from the body through clearance tissues, where their unintended accumulation may result in undesired gene modulation. Divalent siRNAs (di-siRNAs) administered into the CSF induce robust gene silencing throughout the central nervous system (CNS). Upon clearance from the CSF, they are mainly filtered by the kidneys and liver, with the most functionally significant accumulation occurring in the liver. siRNA- and miRNA-induced silencing can be blocked through substrate inhibition using single-stranded, stabilized oligonucleotides called antagomirs or anti-siRNAs. Using APOE as a model target, we show that undesired di-siRNA-induced silencing in the liver can be mitigated through administration of liver targeting GalNAc-conjugated anti-siRNAs, without impacting CNS activity. Blocking unwanted hepatic APOE silencing achieves fully CNS-selective silencing, essential for potential clinical translation. While we focus on CNS/liver selectivity, coadministration of differentially targeting siRNA and anti-siRNAs can be adapted as a strategy to achieve tissue selectivity in different organ combinations.


Introduction
The advent of small interfering RNAs (siRNAs) provides an opportunity to silence gene expression in any organ at any desired time point.siRNAs prevent protein expression by targeting complementary mRNAs for degradation ( 1 ).For siRNAs, the base sequence determines the target mRNA, while the chemical architecture dictates pharmacokinetics and tissue delivery ( 2 ).For example, N-Acetylgalactosamine (GalNAc) conjugated siRNA architecture is an established, clinically validated platform for potent modulation of genes in the liver (3)(4)(5), with a single injection supporting 6-12 months of clinical efficacy ( 4 ).In the CNS, divalent (di)-siRNA scaffold is taken up by neurons and glial cells allowing for RNAimediated silencing in multiple CNS cell types ( 5 ).Upon intracerebroventricular (ICV) injection into mice, di-siRNAs travel in cerebrospinal fluid (CSF) to cause potent ( > 95%) and sustained (up to 6 months) target gene silencing throughout the CNS ( 5 ).For the di-siRNA, and many other chemically modified siRNAs, an asymmetric overhang increases efficacy, and the presence of a phosphorothioate backbone is essential for stability, tissue distribution, and enhanced tissue accumulation and efficacy ( 6 ,7 ).Although the inclusion of phosphorothioate modifications to the terminal linkages greatly enhances cellular uptake, phosphorothioate-enhanced cellular uptake is not receptor mediated or cell type specific and thus, can result in accumulation in cell types outside of the CNS or target organ.Di-siRNAs are partially cleared from the CSF into the systemic circulation and are subsequently cleared from systemic circulation mainly via the liver ( 8 ,9 ).At high doses, systemic clearance may result in cellular accumulation and detectable silencing in clearance tissues that express the target mRNA.
While for most targets, unintended modulation of expression in the liver is not problematic, for some, achieving complete tissue selectivity is essential.One of these targets is Apolipoprotein E (APOE), where CNS and liver APOE are spatially and functional distinct.The APOE4 allele remains the strongest genetic risk factor for developing late onset Alzheimer's disease (AD) and is a compelling target for modulation in AD.The majority of APOE is produced in the (i) central nervous system, where it is primarily expressed by astrocytes (and to a lesser extent, neurons) to transport lipids between cells ( 10 ) and modulate the inflammatory response ( 11 ), and (ii) the liver, where it is secreted by hepatocytes to facilitate lipid uptake into peripheral tissues via lowdensity lipoprotein (LDL) receptors ( 12 ,13 ).There are three allelic variants of APOE: APOE2, APOE3 and APOE4.While APOE4 increases the risk of developing AD and decreases the age of clinical onset, the presence of APOE2 confers protection against AD.Conversely, individuals carrying APOE2 have a higher risk of developing atherosclerosis due to the loss of APOE's key function in cholesterol processing.Genetic removal of APOE (mouse and human) in AD models reduces A β plaques and improves cognitive outcomes (14)(15)(16)(17)(18).However, peripheral APOE2 expression and loss of liver Apoe in mice cause severe atherosclerosis ( 19 ).Thus, unwanted silencing of APOE in the liver may have functional impacts, and strategies to ensure organ specific APOE silencing are necessary.
In summary, APOE is highly expressed in both organs, has distinct spatial and functional characteristics in both organs, and is a relevant target for many neurodegenerative, liver and cardiac diseases.Thus, we chose APOE as a model target to demonstrate the feasibility of achieving fully selective CNS RNAi.
Chemical modifications and the inclusion of bioconjugates direct siRNA delivery and silencing to specific organs and cell types of interest.However, siRNAs are typically cleared from the body by the liver or / and kidneys where their unintended accumulation can direct non-specific target gene modulation.Thus, preventing unwanted silencing in clearance tissues is necessary to achieve fully tissue-selective effects.siRNA silencing depends on efficient interaction between the siRNA guide strand and the RNA-induced silencing complex (RISC).This siRNA-RISC interaction can be sterically blocked by the introduction of fully modified, single-stranded Antisense Oligonucleotides (ASOs) that are complementary to the siRNA guide strand sequence ( 20 ,21 ).The Stoffel lab, in collaboration with Alnylam, was the first to describe the short ASO's ability to inhibit loaded RISC complex in 2005 and named the compounds 'AntagomiRs' ( 21 ).The naming was defined by the functional ability to inhibit miRNA activity.The same configuration was successfully used to inhibit the activity of synthetic siRNAs, reversing the siRNA activ-ity, and called 'ReversiRs' ( 20 ).The terms antagomir, reversir and anti-siRNA all describe ASOs that are complementary to the loaded RISC and represent a biochemically identical phenomenon.The names are derived from the functional application rather than the biological mechanism -when anti-siRNAs are used for miRNA inhibition, they are termed antagomirs ( 21 ); when used for reversing of RNAi-induced silencing, they are called reversiRs ( 20 ).Here, we use the term anti-siRNA to ensure clarity that the ASOs are blocking nondesired activity, rather than reversing the intended function.
Here, we explore whether combinatorial administration of siRNAs and 'anti-siRNAs' in differentially targeted chemical scaffolds could be used to achieve tissue selectivity.In two therapeutic paradigms, we show that GalNAc-conjugated anti-siRNAs rescue and block off-tissue silencing of APOE in the liver without any impact on CNS activity.

Animal studies
All experimental studies involving animals were approved by the University of Massachusetts Medical School Institutional Animal Care and Use Committee (IACUC Protocols #A-2411 and #A-1744) and performed according to the guidelines and regulations therein described.Wild-type adult (6-8 weeks) female FBV / NJ mice (001800-JAX) were obtained from Jackson laboratory.Humanized APOE4 (homozygous) (027894-JAX) mice (breeding pairs) were obtained from Jackson laboratory and bred in house.

Statistical analysis
Each n represents an independent biological sample.All graphs show means ± S.D.All statistics were performed using Prism GraphPad v.

Custom divalent support synthesis
Divalent custom solid support synthesis was previously reported ( 5 ).The method is described briefly below.
Excess tetraethylene glycol was treated with sodium hydride in tetrahydrofura (THF) at 0 • C, and Solketal tosylate dissolved in THF was added dropwise to produce the monofunctionalized glycol ( 1 ).The acetal was removed by treatment with HCl in methanol, and both primary hydroxyl groups were protected with the dimethoxytrityl protecting group to produce compound ( 3 ).Succinic anhydride in the presence of triethylamine (TEA) was used to produce the succinate ( 22 ), which was subsequently used to functionalize native long chain alkyl amine controlled pore glass (CPG, 1000 or 500A) or polystyrene type resins using traditional peptide coupling reagents.These functionalized resins were then used to synthesize the sense strands for divalent siRNA using standard oligonucleotide synthesis techniques.

Custom GalNAc solid support synthesis
The synthesis of the custom GalNAc support was previously reported ( 23 ).The method is described briefly below.
The amino-polyether tritert-butyl ester ( 1 ) coupled with 12-(benzyloxy)-12-oxododecanoic acid.The crude mixture was extracted and isolated without purification.The tbutyl ester was removed by treatment with 25% Triflouroacetic acid (TFA) solution in dichloromethane (DCM).The resulting triacid was then coupled with the peracetylated GalNAc functionalized at the anomeric position with a triethylene glycol aminolinker ( 22 ) resulting in compound ( 24 ) in roughly 75% yield.The benzoyl protecting group was removed by H 2 / Pd / C in methanol.The resulting carboxylic acid ( 4 ) was coupled with a dimethoxy trityl protected 3-amino-1,2-propanediol producing compound ( 5 ).The secondary hydroxyl was then reacted with succinic anhydride and coupled with long chain alkyl amine CPG using standard peptide coupling reagents.The resulting CPG was used to synthesize any 3 -GalNAc containing oligonucleotides / anti-siRNAs using standard oligonucleotide synthesis techniques.

Vinyl phosphonate deprotection
The vinyl phosphonate (VP)-containing oligonucleotides, still on solid support, were treated post synthesis with an anhydrous mixture of trimethylsilyl bromide / acetonitrile (ACN) / dimethylformamide (DMF) / pyridine (3:9:9:1) for 1 h at room temperature with gentle agitation.The reaction was then quenched with water and the CPG was then rinsed with ACN, DCM and allowed to dry, before being deprotected normally as described below.

Deprotection and purification of oligonucleotides
Divalent and conjugated oligonucleotides (DIO, GalNAc) were cleaved and deprotected with standard conditions using aqueous ammonia at 55 • C for 16 h.VP-containing oligonucleotides were cleaved and deprotected as described in O'Shea et al., 2018 ( 25 ).Briefly, CPG-containing VP-oligonucleotides were treated with a solution of 3% diethylamine (DEA) in aqueous ammonia at 35 • C for 20 h.Solutions containing deprotected oligonucleotides were filtered to remove the CPG and dried under vacuum in Speed-vac.The resulting pellets were re-suspended in 5% ACN in water.Purification was performed on an Agilent 1290 Infinity II HPLC system, equipped with a Source 15Q anion exchange column (GE Healthcare) using the following conditions: eluent A, 20% ACN, 20 mM sodium acetate, pH 7; eluent B, 1 M sodium perchlorate in 20% ACN; gradient, 10% B (3 min) to 35% B (18 min) at 60 • C. Peaks were monitored at 260 nm.Pure fractions were collected and dried in Speed-vac.Oligonucleotides were resuspended in 5% ACN and desalted through fine Sephadex G-25 media (GE Healthcare) and lyophilized.

Screening and dose-response studies
HepG2 cells were cultured in EMEM 10% FBS and grown to confluency in T75 flasks.For screening experiments, siRNAs were diluted to a concentration of 1.5 μM in 50 μl of Opti-MEM.Diluted siRNAs (50 μl) were added to a 96-well plate using a multichannel pipette.Next, HepG2 cells were added to the same 96-well plate at a concentration of 25 000 cells per well in 50 μl of EMEM media with 6% FBS (final concentration 3%).Each siRNA was tested in triplicate.Cells were incubated at 37 • C for 72 h at which point mRNA concentration was measured using QuantiGene, which was performed according to the manufacturer's instructions.

Stereotactic ICV injections
Approximately 10 μl of di-siRNA was administered bilaterally (5 μs per ventricle) into the lateral ventricles of mice as previously described ( 5 ).Briefly, mice were anesthetized using Avertin and prepared using standard aseptic technique.Stereotaxic devices were using to hold injection needles and identify injection location.After the identification of the bregma, the needle was placed 1 mm laterally, 0.2 mm posterior and 2.5 mm caudally.Injection was performed at 500 nl / min.Mice were then monitored until fully sternal.

APOE protein quantification
For analysis of APOE protein expression in mouse brain samples, WES by ProteinSimple was used as previously described in ( 5) and ( 26 ).Briefly, tissue punches were collected as above and flash-frozen and placed at −80 • C.After addition of radioimmunoprecipitation assay buffer (RIPA) buffer with protease inhibitors, samples were homogenized and stored at −80 • C. Protein amount was determined using Bradford Assay.Samples were diluted in 0.1 × sample buffer (Protein-Simple) to ∼0.2-0.4 μg / μl.Anti-APOE antibody (Abcam, 183597) was diluted 1:200 in antibody diluent (ProteinSimple), and loading control, anti-vinculin (Invitrogen, 700062), was diluted 1:1000 in antibody diluent.The assay was performed as described by ProteinSimple protocol using the 16-230 kDa plate (SM-W004).The separation electrophoresis and immunodetection are performed automatically in the capillary system using the default system settings.Once loaded, the separation electrophoresis was performed automatically.Results were analyzed using the Compass for Protein Simple software.

Cholesterol
Serum cholesterol was measured using the Abcam low density lipoprotein (LDL) and high density lipoprotein (HDL) cholesterol quantification kit (ab65390).LDL and HDL were separated using the included precipitation buffer that uses a water soluble non-ionic polymer to precipitate the fractions ( 27 ).The assay uses cholesterol esterase to hydrolyze cholesteryl ester into free cholesterol.Next, cholesterol oxidase acts on free cholesterol and to produce a color at 570 nm that is proportional to the amount of cholesterol in the sample.Briefly, serum was collected prior to euthanasia. 2 × buffer was added to 50 μl of serum and incubated at room temperature for 10 min.Samples were spun for 10 min at 2000 rpm, and the supernatant (LDL fraction) was placed in a separate tube.The pellet (HDL fraction) was resuspended in 200 μl PBS.The samples were diluted, and cholesterol levels analyzed according to the package instructions.

Selection and synthesis of oligonucleotides
For the experiments presented here, we utilized chemically modified oligonucleotides and relied upon three previously identified conjugates for specific purposes (Figure 1 ).Full chemical stabilization is essential for conjugate-mediated delivery in vivo , where siRNAs must survive for extended periods in endosomes / lysosomes compartment to support sustained target gene modulation.The inclusion of fluoro and omethyl modifications to the native RNA 2 -hydroxyl increases stability and efficacy both in vitro and in vivo (Figure 1 B) ( 28 ,29 ).
We, and others, continue to evaluate the impact of changing the degree of modification, the balance of modifications within an siRNA and the positional effects of chemical modification on siRNA efficacy and distribution.For these studies, we chose a balanced pattern, with an approximately equal number of o -methyl and fluoro modifications on each siRNA strand.Here, we included a run of fluoro modifications on the antisense strand around the seed-binding site to enhance seed site affinity for both in vitro and in vivo siRNAs (Figure 1 A,B).Locked nucleic acid (LNA) modifications, in which a carbon bridge between the 2 -oxygen and the 4 -carbon position (Figure 1 B), increase the stability and affinity of oligonucleotides to their complementary mRNA sites.Here, we used LNA modifications in the anti-siRNA oligonucleotides to enhance affinity to the APOE-siRNA antisense strand and describe the rationale for the position of LNA modifications below.We used three chemical conjugates in these studies: cholesterol conjugation to deliver siRNAs into cells for in vitro screening experiments, divalent conjugation for siRNA delivery throughout the CNS and GalNAc conjugation for delivery to the liver.The conjugation of hydrophobic entities, like cholesterol (Figure 1 C) (Cholesterol CPG: Chemgenes #N-9166-05), to fully chemically stabilized siRNA, allows for efficient and passive internalization and productive gene modulation in virtually all cell types in vitro without the use of transfection reagents ( 30 ).Cellular uptake of oligonucleotides without transfection reagents is called gymnotic delivery, and, when directly translated from the original Greek, means naked 'gymnos' delivery ( 30 ).The use of gymnotic delivery for in vitro screening simplifies the workflow and provides a better experimental design predicting compound in vivo behavior.
The divalent siRNA linker (Figure 1 C), first published in Alterman et al ., 2019 ( 5 ), contains a triethylene glycol linker connecting the two sense strands together at the 3 -terminus, connecting the two functional siRNAs (Figure 1 C).The divalent linker was used to efficiently deliver siRNAs throughout the CNS and was conjugated to siRNAs targeting APOE to silence expression throughout the CNS.The trivalent GalNAc conjugate (Figure 1 C) was synthesized in house (Materials and methods) and used to direct delivery of anti-siRNAs to hepatocytes in the liver, the main site of systemic APOE expression.The expected and observed mass (g / mol) of each oligonucleotide strand used for in vivo experiments and the extinction coefficient are shown in Supplementary Table S2 .

Identification of siRNAs targeting APOE
We designed a panel of Apoe -targeting siRNA candidate sequences for mouse Apoe and human APOE using an internally validated algorithm with methods based on the principles described in Birmingham et al .( 31 ).Ideal candidate sequences have a GC content of < 45%, do not contain toxic motifs such as G-quadraplexes and span the length of the mRNA sequence.
We performed initial screens using cholesterol-conjugated siRNAs at a 1.5 μM dose to identify hit sequences and validated the hits in eight-point dose-response studies by serially diluting from the top dose of 1.5 μM.We previously published screens and validations identifying potent sequences targeting mouse Apoe and chose APOE-1134 as the lead mouse-targeting sequence ( 32 ).Here, using cholesterol conjugated siRNAs, we screened 12 siRNA sequences targeting human APOE in human HepG2 cells and identified two hit sequences ( Supplementary Figure S1 a).Hits were validated in seven-point dose-response studies and identified APOE-1142 as the most potent and efficacious sequence (IC50: 298.2 nM) ( Supplementary Figure S1 b).

siRNAs in CNS-and liver-targeting configurations enable tissue-selective apoe silencing in wild-type mice
To explore the feasibility of silencing different pools of Apoe, we engineered the siRNAs targeting mouse Apoe in either a liver-active (GalNAc APOE ) ( 3-5 ) or CNSactive (di-siRNA APOE ) ( 5 ) chemical configuration (Figure 2 ; Supplementary Tables S1 , S2 for sequence information).
In wild-type mice, one-month post-injection, GalNAc APOE resulted in potent Apoe mRNA and protein silencing in the liver ( > 95%, P < 0.0001) compared to PBS and non-targeting control (NTC) groups (10 mg / kg SC; n = 6 / group).We observed no detectable silencing of Apoe mRNA or protein in brain samples, as expected since systemically administrated oligonucleotides do not cross the blood-brain barrier.Administration of di-siRNA APOE into the lateral ventricles (ICV) also resulted in potent Apoe mRNA and protein silencing in the brain ( > 95%, P < 0.0001) compared to both PBS and NTC groups (475 μg ICV; n = 6 / group).A portion of the siRNA dose administered into the brain escapes the CNS and is cleared from circulation via the liver (33)(34)(35).Thus, di-siRNA administered into the brain also resulted in reduction of liver Apoe mRNA and protein ( ∼75%, P < 0.0001).To mitigate off-tissue silencing, we reduced the dose of di-siRNA 2-and 4fold.Lower doses (237 or 118 μg) of di-siRNA APOE achieved similar brain Apoe silencing ( > 95%, P < 0.0001) while reducing the degree of silencing in the liver ( ∼20%, P < 0.05) (Figure 2 D, Supplementary Figure S2 b,c).We observed no changes in blood chemistry markers or complete blood counts at 2and 21-day post-treatment of either siRNA, indicating lack of liver / kidney toxicity ( Supplementary Table S3 ).
As expected, complete liver Apoe silencing by GalNAc APOE significantly increased serum LDL levels ( > 300%, P < 0.01) (Figure 2 E), whereas di-siRNA APOE -mediated brain Apoe silencing (which also reduced liver ApoE by ∼75%), had no effect on serum LDL one-month post-injection (Figure 2 E).Collectively, these data provide further evidence that two spatially distinct pools of Apoe exist in wild-type mice.In the short-term, Apoe does not exit the CNS to replenish systemic Apoe or alter systemic cholesterol homeostasis, and systemic Apoe does not rescue the loss of brain Apoe.

Chemical engineering to achieve selective CNS target silencing
Although in the short-term, partial reduction of liver Apoe by di-siRNA does not translate into detectable serum cholesterol changes in wild-type or AD mouse models ( 36 ) and can be controlled by dose reduction, long-term partial reduction may have a more significant impact.Thus, we sought to devise a method to block any silencing of APOE in the liver.
For these experiments, we moved to targeting human APOE in the humanized APOE4 mouse model as it has greater translatability and used the APOE-1142 sequence identified in Supplementary Figure S1 .siRNA silencing can be sterically blocked by fully modified single-stranded oligonucleotides that are complementary to the siRNA sequence ( 20 ,21 ).We optimized this technology and explored whether combinatorial administration of siRNA (di-siRNA via ICV in CNS) and anti-siRNA (GalNac-anti-siRNA SC) by different routes and using different chemical scaffolds could block any unwanted silencing in the liver.Specifically, we designed and synthesized two anti-siRNA variants (8-and 15-mer), whose sequences were complementary to the di-siRNA HAPOE guide strand (Figure 3 A) ( 20 ,37 ).Prior characterization of REVERSIR showed greatest efficacy with oligo lengths of 8-9 mer, moderate activity when length was increased to 15-and 21-mer, and inactivity when length was decreased to 7-mer ( 20 ).Thus, we chose lengths of 8-and 15-mer to evaluate varying efficacy in blocking unwanted silencing.
Several locked nucleic acid modifications were introduced at positions 2, 3, 4, 6, 7 and 14 (15-mer) from the 3 -end to enhance affinity with target siRNAs and to enhance metabolic stability (Figure 3 A and Supplementary Table S1 ).The specific LNA positions were chosen to enhance binding initiation, enhance affinity to the seed region, and based on results showing the importance of including an LNA at position 6 ( 20 ).'Anti-siRNAs' were then conjugated to GalNAc with a cleavable linker to drive preferential accumulation in liver upon systemic administration ( 20 ).The addition of a cleavable linker between GalNAc and anti-siRNA has previously been shown to enhance the efficacy of anti-RISC antisense compounds and conjugated siRNAs ( 7 ,20 ).There are different variants of cleavable linkers, and their stability needs to be optimized to be stable in serum during the distribution phase but cleavable intracellularly in the endosomes / lysosomes.The introduction of a dT phosphodiester linker provides the necessary properties and is methodically simple.Thus, the dT-PO cleavable linker was used to connect GalNAc to anti-siRNAs.
To evaluate the ability of anti-siRNAs to selectively reverse and block di-siRNA HAPOE silencing in the liver, we administered high dose di-siRNA (475 μg) via ICV injection to humanized APOE4 mice (JAX#027894).Seven days later, we administered 1 mg / kg GalNAc-anti-siRNAs subcutaneously (either 8-or 15-mer) (Figure 3 B,C).We evaluated APOE plasma protein levels over time as a read out for APOE silencing in the liver.Forty-eight hours after di-siRNA HAPOE injection, plasma APOE levels dropped to ∼30% of initial levels (Figure 3 E).Within 24 h after administration of anti-siRNA (either 8-or 15-mer), plasma APOE levels began to recover and within 3 days, APOE expression returned to predi-siRNA treatment levels (Figure 3 E).Consistent with the rescue of plasma APOE, animals treated with anti-siRNAs post-administration of di-siRNA showed no reduction in liver APOE protein relative to controls one-month post injection (Figure 3 G and Supplementary Figure S3 for western blots).As GalNAc anti-siRNAs do not cross the blood-brain barrier, SC administration had no effect on brain APOE silencing ( > 95%, P < 0.0001) (Figure 3  In a separate experiment, we co-administered GalNac-anti-siRNAs and di-siRNA HAPOE to determine if unintended silencing in the liver could be completely blocked (Figure 3 D).Coadministration of anti-siRNAs (both 8-and 15-mer) resulted in no reduction in plasma APOE, suggesting that di-siRNA silencing in liver was completely blocked by liver-targeting anti-siRNAs (Figure 3 F).Consistent with the reversal paradigm, anti-siRNAs completely blocked reduction in liver APOE protein relative to controls one-month post injection (Figure 3 H and Supplementary Figure S2 for western blots) and had no impact on brain APOE protein levels (Figure 3 H).
In both experiments, SC injection of PBS control showed no rescue of di-siRNA-induced liver APOE modulation (Figure 3 G [right] and Figure 3 h [right]), confirming the effect is due to the anti-siRNAs.Due to the need to limit the number of animals used in the study, a PBS control, but not a random GalNAc anti-siRNA control, was used.The chances that the observed reduction in APOE modulation in the liver was due to non-specific effects of GalNAc anti-siRNA administration are low, as there are extensive public data on the specificity and use of GalNAc-modified ASOs in the liver.Analysis of liver toxicity 1-month post injection showed no significant differences between groups ( Supplementary Table S4 ).While both anti-siRNA structures prevented silencing of APOE in the liver across experiments, the shorter anti-siRNA (8 mer) was more effective than the longer compound (15 mer).This finding is consistent with previous data showing an impact of anti-siRNA design on efficacy ( 37 ).Taken together, these data demonstration that anti-siRNAs block unwanted RNAi activity in the liver, enabling tissue selective silencing of APOE in the CNS with divalent siRNAs.
To confirm that blocking di-siRNA activity in the liver would mitigate the off-tissue effects of di-siRNAs, we evaluated the level of systemic tissue accumulation and silencing for two additional di-siRNAs targeting Huntingtin (Htt) and Cd47 mRNA (Figure 4 ).Twenty-four hours and one-week post-administration, we observed that systemic exposure was limited to the liver and kidney (Figure 4 A).We also evaluated mRNA silencing at 2-week post-administration and observed significant mRNA silencing only in the liver of animals treated with di-siRNA targeting Htt (Figure 4 B) and no silencing when targeting Cd47 (Figure 4 C).Based on these results, blocking the off-tissue effect of di-siRNAs in the liver is sufficient to achieve CNS-selective silencing.

Discussion
Tissue-selective RNAi is a holy grail of oligonucleotide therapeutics.GalNAc conjugated oligonucleotides promote tissue selectivity via receptor mediated uptake in the liver, which is one of the primary clearance organs.For extrahepatic delivery, some modulation of expression in clearance tissues, in addition to the target organ, is expected and difficult to avoid.Unwanted silencing in clearance tissues can be avoided by reducing the administered dose.However, as higher doses translate to longer duration of effect, lower doses may result in reduced clinical efficacy.Here, we describe a new technological twist for achieving tissue selectivity in the CNS.We use a combination of different routes of administration (ICV and SC) and different scaffolds (divalent and GalNAc) to promote orthogonal siRNA distribution and achieve tissue selectivity.The ap- proach presented here uses similar technology to antagomirs and reversirs.However, the main conceptual idea builds upon prior studies by using different delivery scaffolds from the primary oligonucleotide.
In these experiments, we used APOE as a model target gene.Like other genes, APOE is expressed by different cell types in multiple organ systems with organ-specific functions and effects.It's role in systemic lipid homeostasis is indispensable; however, presence of the APOE4 allele is the most significant genetic risk factor for developing late onset AD and is implicated in many other neurodegenerative conditions.Thus, identifying strategies to reduce or block APOE expression in the CNS without impacting hepatic expression is critical in developing therapeutics targeting APOE .We identified several siRNA sequences that are efficacious in silencing mouse or human APOE and evaluated their effects in vivo .We show that siRNAs targeting either liver or brain Apoe have distinct effects on serum LDL levels.Interestingly, only complete silencing of liver Apoe , not partial silencing, is necessary to have detectable effects on serum LDL levels in the shortterm.The impact of Apoe silencing on systemic cholesterol metabolism is dose-dependent, and in our hands, required almost completed ( > 95%) knockdown of hepatic Apoe mRNA and protein to have a measurable effect with the assays used.
There is a significant daily difference in circulating APOE levels, which are highly affected by diet and circadian cycles, and thus, there is a high buffering capacity within the body.Because of the buffering capacity, it appears that detectable changes in serum cholesterol require liver APOE reductions below a certain threshold.Even when liver Apoe expression was reduced by ∼75%, we found no detectable difference in cholesterol, and complete knockdown with GalNac siRNAs was needed to detect statistically significant changes in circulating cholesterol levels.Similarly, introducing Apoe at concentrations as low as 10% of normal levels has been shown to sufficiently normalize cholesterol levels in ApoE −/ − mice ( 38 ).These findings indicate that systemic APOE is present in significant excess and mild-to-moderate reduction of systemic APOE may not greatly impact systemic cholesterol homeostasis, at least in the short-term.However, the simple fact that we can't detect changes does not mean there aren't underlying physiologic effects of partially silencing liver APOE , and systemic APOE or cholesterol levels should be monitored in any clinical studies using APOE reducing agents.
Given the possible concerns of long-term, partial silencing of liver APOE by di-siRNAs administered in the CNS, finetuning of CNS versus systemic silencing is likely necessary to achieve effective and safe modulation in clinical settings.While the anti-siRNA concept is well described in the literature ( 20 ), our study shows that differentially targeting anti-siRNA and siRNAs can be used to achieve tissue-selective silencing.With these tools in hand, it is now feasible to investigate selective CNS APOE silencing as a potential therapeutic strategy to combat neurodegeneration.
Our anti-siRNA chemical design was largely based on results from prior studies showing 8-9 mers as the ideal length, the importance of LNA modifications within the seed, and the importance of including a cleavable linker between the conjugate and the oligonucleotide ( 20 ,37 ).Consistent with these reports, we observed faster and more robust efficacy with a shorter, 8-mer oligo compared with the longer 15-mer, likely due to greater affinity of the shorter sequence.
One concern with the anti-siRNA approach is the potential off target effects of the anti-siRNA, and it is possible that off target effects increase with shorter anti-siRNAs.While we designed the oligonucleotides to be complementary to the guide strand of the APOE siRNAs and to have minimal off target complements, it is possible they may bind to and block expression of miRNAs with similar sequences.In this study, the use of the GalNAc conjugate drives delivery to hepatocytes limiting the functional impact of off-target binding.However, use of non-receptor mediated conjugates and chemistries may result in greater biological significance of potential off target effects in other tissues.Evaluation of off-target effects using RNAseq will be important in future studies.
Targeting genetic causes and risk factors of disease with oligonucleotide therapeutics is an increasingly effective approach for disease-modifying therapy.Often, targets are specifically expressed in affected organs or cell types.However, like APOE , there are many targets with functional and important expression in non-diseased tissues.One is example is myostatin and muscle wasting.Myostatin modulation in skeletal muscles with DCA-conjugated siRNAs rescues muscle loss ( 39 ), but unwanted accumulation and silencing in cardiac muscles could result in cardiac hypertrophy ( 40 ).In addition to achieving delivery to skeletal muscles, docosanoic acid (DCA) conjugated siRNAs also deliver to and modulate myostatin expression in cardiac muscle ( 39 ), necessitating development of tools, like anti-siRNAs evaluated here, in blocking unwanted siRNA activity.In addition, siRNAs are also being investigated as potential therapeutics for cancer, including CNS cancers like glioblastoma multiforme.Potential targets for CNS tumors include Vascular Endothelial Growth Factor (VEGF), Bcl2, Stat3 and many immune related targets.To achieve therapeutic impact, high doses of tumor-targeting siR-NAs into the CNS may be necessary.As many of the potential target genes are uniformly expressed across tissue and cell types, strategies to mitigate unwanted silencing in clearance and other organ systems may be necessary to reduce potential systemic side effects like VEGF-mediated hypertension or systemic immune repression.The use of anti-siRNAs may help reduce side effects that occur due to exposure of healthy tissues to therapeutic agents.
Despite the many clinical scenarios in which our approach might be useful in achieving tissue selectivity, it is important to consider its translatability.For most targets and clinical indications, tissue selective modulation is not necessary.The divalent siRNA is predominantly cleared through the liver, resulting in observable gene modulation in this tissue.Reduction in dose level supports the ability to maintain primary CNS silencing, while also reducing target gene modulation in the liver.Thus, reducing unwanted silencing in clearance tissues by dose may be a more straightforward option, but carries a risk of compromising duration of effect and necessary target reduction.As CNS administration requires direct access to CSF, which is invasive, the dosing paradigm would need to be optimized to maximize the duration and durability of silencing while reducing the frequency of administration.In such cases, administering GalNAc anti-siRNAs would provide a path for selectivity in the CNS while blocking undesired liver gene modulation.While this work provides conceptual proof of concept, using siRNAs in combination with anti-siRNAs would require more complex administration processes, more complex clinical trials, two CMCs, and two toxicity / safety studies.This situation is like that of ReversiRs: while established as a technically feasible and elegant approach to reverse siRNA activity in the liver, the clinical translatability is limited.Since its conception, the extensive preclinical optimization and overall clinical safety of GalNAc siRNAs assuaged some concerns raised that necessitated the development of tools to reverse the activity of such long-lasting compounds.Here, we show a new approach to prevent siRNAmediated gene modulation in non-target tissues where siRNAs accumulate that may be useful in achieving safer gene modulation with new oligonucleotide chemistries.While the data presented here demonstrate proof-of-concept in the context of brain and liver APOE, this technology can be adapted to any combination of tissues using different delivery configurations and any genetic target, providing additional tools for studying and treating disease.

Figure 4 .
Figure 4. Divalent siRNA systemic distribution is primarily limited to the liver and kidney.Di-siRNAs targeting uniformly expressed genes HTT and CD47 were administered to mice (20 mg / kg, ∼437 μg).( A ) Systemic accumulation of HTT targeting di-siRNA at 24 h and 1-week post-administration (measured with peptide nucleic acid (PNA) h ybridization assa y) sho wing that sy stemic e xposure is limited to tw o major clearance tissues: liver and kidney.Htt ( B ) and Cd47 ( C ) mRNA le v els 2-w eek post-administration (QuantiGene) sho wing statistically significant silencing only with di-siRNA targeting HTT in the liver.N = 6 / group.Statistics: error bars STDEV, one-way ANO V A with Tucker correction for multiple comparisons; ** P < 0.01.