Therapeutic delivery of albumin-binding siRNA targeting IRS2 to diverse cell types reduces mammary tumor growth

Abstract

Oligonucleotide therapeutics are a new class of drugs that enable robust and sustained modulation of gene expression. However, achieving efficient delivery of small interfering RNAs (siRNAs) to tumors is a challenge for therapy. Here, we demonstrate that fully chemically modified siRNAs conjugated with an albumin-binding dendrimer are efficiently delivered to both neoplastic and stromal/immune cells within primary TNBC mammary tumors. siRNAs were designed to selectively target insulin receptor substrate 2 (IRS2), a signaling adaptor of insulin and insulin-like growth factor (IGF) signaling that has been implicated in aggressive breast cancers. These siRNAs reduced IRS2 expression in tumor and stromal cells without causing hyperglycemia, resulting in reduced tumor growth that was associated with decreased vascularization and alterations in macrophage polarization and the expression of EMT proteins. This work demonstrates that siRNAs can be delivered to neoplastic and specific stromal populations in mammary tumors and that they can effectively and specifically silence a driver of aggressive breast cancer.

Graphical Abstract

Introduction

Breast cancer remains a major cause of cancer-related morbidity and mortality, particularly due to the limited efficacy of targeted therapies against aggressive subtypes such as triple negative breast cancer (TNBC). One challenge in developing targeted approaches to effectively treat TNBC is the limited ability to inhibit non-enzymatic, intracellular proteins that play a causal role in cancer progression. The insulin receptor substrate (IRS) proteins, which are key mediators of the insulin and insulin-like growth factor (IGF) signaling (IIS) pathway and some integrin and cytokine receptors, exemplify the functional significance of such proteins in cancer.¹ IRS proteins are intracellular adaptor proteins that form signaling complexes with downstream effectors following integrin, growth factor, and cytokine receptor activation to facilitate tumor initiation, growth, metastasis, and chemotherapy resistance.² Among the IRS family, IRS1 and IRS2 are commonly expressed in tumors, although with varying expression patterns and functions, and differing associations with disease outcomes. IRS2 is more highly expressed than IRS1 in invasive breast cancer subtypes, including TNBC, and its expression and activity are associated with poorer clinical outcomes and aggressive tumor behaviors, including invasion, migration, survival, cancer stem cell (CSC) self-renewal and metastasis.^3–10

Targeting IRS2 specifically could be a novel intervention to inhibit the IIS pathway in breast tumors, as prior receptor/ligand-targeted approaches have been unsuccessful due to adverse systemic toxicities, such as hyperglycemia, that are caused by disruption of normal metabolic regulation and that lead to positive feedback of insulin signaling in tumors.^11–14 Selective inhibition of IRS2, but not IRS1, could maintain metabolic homeostasis while preventing feedback insulin signaling that drives tumor progression. Traditional antibody and small molecule-based therapies, however, are unlikely to selectively inhibit IRS2 function due to its cytoplasmic localization, non-enzymatic activity, and high degree of structural and functional homology with IRS1, and alternative approaches are needed to develop such a disease-modifying therapeutic.

Small interfering RNAs (siRNAs) are a novel class of therapeutics, with seven drugs approved for clinical use.^15–21 Once delivered to cells, siRNAs are incorporated into the RNA-induced silencing protein complex (RISC), which then binds to and degrades complementary mRNA transcripts, preventing target protein translation.²² The programmable nature of therapeutic siRNAs enables silencing of virtually any gene target that is important for cancer progression, regardless of protein structure or cellular localization.²³ Furthermore, chemical stabilization of siRNAs can extend their duration of action to 6–12 months after a single clinical dose, making this drug class especially attractive for antineoplastic therapy.²⁴ Achieving efficient and sustained delivery of siRNAs to tumors and diverse cell populations of the tumor microenvironment (TME), however, has been a challenge and a limitation for clinical translation ^{25, 26}. Studies that explore siRNA delivery and target gene silencing in both neoplastic cells and the TME in immunocompetent breast tumor models are lacking, but necessary, for advancing siRNA therapeutics for breast cancer treatment.

We have previously shown that conjugation of siRNAs to lipophilic moieties or albumin-binding dendrimers facilitates broad biodistribution and delivery.^27,28 Emerging evidence suggests that albumin-binding conjugates may outperform lipophilic conjugates in delivering siRNAs to tumors, but this has not yet been evaluated in immunocompetent breast cancer models.^{29, 30} In this study, we evaluate the impact of siRNA conjugate type on delivery efficiency in primary TNBC tumors and show that an albumin-binding dendrimer conjugate, but not a lipophilic conjugate, achieves significant tumor accumulation. Moreover, albumin-binding conjugates improve siRNA uptake in both neoplastic cells and cells of the TME. We then identify fully chemically modified siRNAs capable of selectively silencing IRS2, but not IRS1, in both human and mouse models. When conjugated to an albumin-binding dendrimer, these siRNAs demonstrated potent IRS2 silencing and significantly suppressed tumor growth in an immunocompetent model of TNBC. Together, our findings not only establish a strategy to co-target diverse cell populations within tumors using siRNA therapeutics, but also validate IRS2 as a clinically relevant, previously untargetable, intracellular adaptor protein in breast cancer.

Results

Albumin-binding dendrimer conjugation improves siRNA delivery to mammary tumors

A barrier to the use of siRNA therapeutics in cancer is delivery of oligonucleotides to tumors and their diverse cell types. Previous studies have shown that lipophilic and albumin-binding conjugates improve functional siRNA delivery to extrahepatic tissues upon systemic administration, which may allow for improved uptake in solid malignancies.²⁷ TNBC and other primary breast tumors are localized within the adipocyte-rich environment of the healthy mammary tissue, which may further promote internalization of lipid-conjugated siRNAs. Emerging evidence also suggests that conjugates that bind to albumin, which has been used as a carrier for existing breast cancer therapies³¹, increase plasma circulation time of siRNAs^30,32, a property that may enhance tumor distribution. Thus, we compared TNBC mammary tumor delivery and accumulation of siRNAs conjugated to the 22-carbon fatty acid docosanoic acid (DCA) and an amphiphilic, albumin-binding dendrimer structure (Figure 1A).^27,33

Figure 1. An albumin-binding dendrimer conjugate improves siRNA delivery to mammary tumors.

(A) Backbone chemical modifications and chemical structures of lipophilic docosanoic acid (DCA) and albumin-binding dendrimer conjugated fully chemically modified siRNAs. (B) 4T1 cells were orthotopically implanted into the fat pads of syngeneic BALB/c mice and mice were treated subcutaneously or intratumorally once with 40 mg/kg conjugated siRNA. (C) Representative whole-tissue fluorescence microscopy images of mammary tumors and tumor-associated mammary fat pad tissue from mice subcutaneously injected with 40mg/kg Cy3-labeled unconjugated, DCA-conjugated, or dendrimer-conjugated siRNAs targeting Htt (n=5 mice/group). (D) Tumor accumulation levels of Cy3-labeled siRNA 48 hours after subcutaneous siRNA administration (n=4–5 mice/group). UNC: unconjugated. Line in box-and-whisker plots represents median value, while whiskers represent minimum-to-maximum values. ns = nonsignificant *p<0.05 **p<0.01***p<0.001****p<0.0001 by one-way ANOVA. See also Figure S1.

Syngeneic mice were injected with 4T1 TNBC tumor cells orthotopically into the mammary fat pad and palpated daily until tumor volumes were approximately 200mm³ (Figure 1B). Mice were then injected locally (intratumoral) or systemically (subcutaneous) with Cy3-labeled unconjugated, DCA-conjugated, or dendrimer-conjugated siRNAs targeting the ubiquitously expressed, previously validated housekeeping gene Htt³⁴ to assess spatial distribution within tumors. Tumors and adjacent mammary tissue were collected 48 hours post-injection to assess siRNA biodistribution by fluorescence microscopy (Figure 1C). Cy3 fluorescence was largely confined to the site of injection following intratumoral injection, suggesting minimal tumor biodistribution (Figure S1). The fluorescence signal was high in tumor-associated mammary tissue, but largely undetectable in the mammary tumors, from mice systemically injected with DCA-siRNAs. In contrast, both tumors and mammary tissue from mice treated with dendrimer-siRNA showed robust, uniform Cy3 fluorescence. Antisense strand concentrations in tumor biopsies demonstrated a ~4.8-fold increase in dendrimer-siRNA delivery compared to DCA-siRNA (p=0.0004) (Figure 1D), suggesting that conjugation with the albumin-binding dendrimer, but not lipophilic DCA, significantly improves siRNA delivery to mammary tumors.

Tumors are a heterogenous microenvironment of primary neoplastic cells, immune cells, and stromal cells, and target gene silencing in one or more of these cellular compartments may impact tumor biology. To assess the cell-type specificity of dendrimer-siRNA uptake in primary mammary tumors, mice with orthotopic 4T1 tumors were systemically administered PBS or 40mg/kg Cy3-labeled unconjugated or dendrimer-siRNAs by subcutaneous injection when tumors were ~200mm³. A second subcutaneous injection was given 72 hours later, and mice were sacrificed after an additional 72 hours, at which time single-cell suspensions were prepared for multi-color flow cytometry analysis (Figure 2A). We have previously demonstrated that the presence of a Cy3 fluorophore does not significantly alter in vivo distribution of lipophilic siRNAs and that quantification of siRNA distribution based on Cy3 fluorescence is consistent with tissue antisense strand accumulation profiles, validating the use of Cy3 fluorescence as a proxy for siRNA delivery.²⁷

Figure 2. Dendrimer-conjugated siRNA is internalized by tumor and stromal cell types within mammary tumors.

(A) 4T1 tumor-bearing BALB/c mice were injected twice with 40 mg/kg Cy3-labeled unconjugated or dendrimer-conjugated siRNA across 6 days to determine cell-specific siRNA delivery in mammary tumors. (B) Representative histograms of Cy3 median fluorescence intensity (MFI) in mammary tumor cell populations. (C) Representative immunofluorescence images of tumor-associated macrophages after siRNA administration. (D-E) Quantification of the Cy3 MFI (D) and the proportion of cell populations with Cy3 fluorescence (E). Analysis of n=4 tumors/group; UNC: unconjugated. MΦ: macrophage, DN: double negative. Line in box-and-whisker plots represents median value, while whiskers represent minimum-to-maximum values. Scale bars 10 μm (Figure 2C). ns = nonsignificant *p<0.05**p<0.01***p<0.001****p<0.0001 by one-way ANOVA (per cell type). See also Figure S2.

Primary tumor cells, defined by high expression of EpCAM, mammary duct epithelium, non-tumor stromal cells, and both adaptive and innate immune (CD45+) cell populations displayed varying degrees of Cy3 fluorescent signal (Figure 2B, gating strategy in Supplemental Figure 2). M1-like macrophage populations, defined by high expression of F4/80, CD11c, and MHC class II, and M2-like macrophages, defined by high expression of F4/80 and CD206, had greater siRNA uptake than T cells, dendritic cells, and other immune populations. Cy3 fluorescence within F4/80 positive macrophages was also confirmed by immunofluorescence staining (Figure 2C). Dendrimer conjugation enhanced siRNA delivery to all identified cell populations, notably a ~3.8-fold increase in tumor cells (p=0.0006), ~4-fold increase in non-immune stromal cells (p=0.0002), ~2.5–5 -fold overall increase in M1-like and M2-like macrophage populations, and ~2.5-fold increase in dendritic cells (p= 0.0019) and multiple T cell populations compared to unconjugated siRNA (Figure 2D). The percentage of cells in tumors that were Cy3 fluorescent were also significantly increased in most populations after treatment with dendrimer-siRNA compared to unconjugated siRNA, suggesting efficient cellular uptake (Figure 2E). Of note, previous studies have reported limited uptake of dendrimer-siRNAs in immune cells located in the bone and spleen²⁸, which contrasts with the robust immune cell uptake observed in this study. Our observed enhancement of dendrimer-siRNA uptake in mammary tumor-associated macrophages (TAM) is consistent with previous findings using a melanoma tumor model.³² TAMs may therefore have a greater propensity for siRNA internalization compared to other macrophage populations, whether that be due to their activation state or an increased local accumulation of dendrimer-siRNA in tumors.³⁵ Taken together, our results demonstrate that the albumin-binding dendrimer enhances siRNA delivery to both neoplastic cells and TME populations upon systemic injection, enabling co-targeting to diverse cell types within TNBC tumors.

Identification of human-specific, mouse-specific, and human/mouse cross reactive siRNAs that selectively silence IRS2

Fully chemically modified siRNAs target mRNA transcripts in a highly sensitive and sequence-specific manner, which creates an opportunity to identify therapeutic compounds that selectively silence IRS2. IRS2 is expressed in both primary tumor cells and in cells of the TME and has distinct functions within each compartment that could promote TNBC growth and progression. Assessment of the mechanistic significance of IRS2 within tumor cells or the TME in isolation using immunodeficient mouse tumor models, as well as simultaneously in both compartments using syngeneic mouse models, requires the design of both species-specific and species cross-targeting siRNAs, respectively. We screened a panel of 48 siRNA sequences that were predicted by a proprietary algorithm³⁶ to selectively silence IRS2/Irs2 (14 human, 14 mouse, and 20 human/mouse cross-reactive) and were specific to the open reading frame or 3’ untranslated regions (UTR) of IRS2/Irs2 mRNA transcripts (Figure 3A, Table S1). Exclusion criteria contained in the predictive algorithm to minimize off-target effects include sequences with perfect homology to miRNA seeds at position 2–7 of the antisense strand and those with full complementarity to non-target mRNAs at position 2–17 of the antisense strand. The 3’ end of the siRNA sense strands were conjugated to cholesterol to enable passive uptake for in vitro screening (Figure 3B).

Figure 3. Identification of IRS2-selective siRNAs.

(A) In vitro and in vivo screening of siRNAs in human MDA-MB-231 and mouse 4T1 cells. Cells were treated with fully chemically modified siRNAs conjugated to cholesterol at 0.1875 μM (MDA-MB-231) and 1.5 μM (4T1) for 72 hours prior to IRS2 expression analysis (n=3 independent samples). Mice were subcutaneously injected with 40 mg/kg of the top siRNA candidates and sacrificed 2 weeks post-injection to confirm in vivo silencing efficacy. (B) Backbone chemical modifications and conjugate chemical structures of fully chemically modified siRNAs used in in vitro and in vivo screening. (C-D) Six-point dose-response curves of lead siRNAs 5967 and 6530 in MDA-MB-231 cells (C) and siRNAs 6548 and 6530 in 4T1 cells (D) (n=3 independent samples). (E-F) IRS2 protein abundance in MDA-MB-231 cells (E) and 4T1 cells (F) after 72 hours siRNA treatment (n=3 independent experiments). UNT: untreated control, NTC: nontargeting control. Data presented as means ±SD. *p<0.05 **p<0.01 ***p<0.001 by one-way ANOVA. See also Figure S3.

Screening in two TNBC cell lines, human MDA-MB-231 and mouse 4T1 cells, identified human-specific (si5967), human/mouse cross-targeting (si6530), and mouse-specific (si6548) siRNAs that silenced IRS2/Irs2 mRNA expression by approximately 50% (Figure S3A). The relative median inhibitory concentration (IC50) values were then determined for top siRNA sequences using six-point dose-response studies (Figure 3C–D). Predicted mouse-specific siRNAs did not silence IRS2 mRNA expression in MDA-MB-231 cells and the predicted human-specific siRNA did not silence Irs2 mRNA expression in 4T1 cells (Figure S3A). We then quantified IRS2 protein in both MDA-MB-231 and 4T1 cells and found that IRS2 protein silencing was consistent with mRNA silencing (approximately 50% reduction) in MDA-MB-231 cells after treatment with si5967 and si6530, but not si6548 (Figure 3E), and that IRS2 protein was reduced approximately 80% in 4T1 cells after treatment with si6530 and si6548, but not si5967 (Figure 3F). Treatment of si6530 and si6548 also significantly reduced IRS2 protein abundance in another cell line, FL83B mouse hepatocytes, further confirming silencing efficacy (Supplemental Figure 3B). Importantly, treatment of all cell lines with si5967, si6530, and si6548 did not silence IRS1 protein abundance (Figure S3C), demonstrating the target selectivity of the siRNAs. Thus, systematic screening identified highly potent, fully chemically modified siRNAs that efficiently and selectively silence mouse and/or human IRS2, a key advancement towards in vivo modulation of IRS2 expression in mammary tumors.

We then assessed in vivo silencing efficacy of si6530 and si6548. Naïve mice were subcutaneously injected with si6530 or si6548 (40 mg/kg in 150 μL PBS), and silencing was evaluated after 2 weeks. The 3’ end of the siRNA sense strands were conjugated to DCA to enable broad tissue distribution, while 5’ (E)-vinylphosphonate and 3’ extended nucleic acid (ExNA) modifications were added to the siRNA antisense strands to improve 5’ phosphate stability and limit exonuclease-mediated 3’ trimming, respectively.³⁷ DCA is a well-validated conjugate that provides broad distribution to many tissues, including muscle and liver^27,38, where IRS2 is expressed. Use of DCA-conjugated siRNAs, therefore, enabled validation of in vivo efficacy before advancing our identified compounds towards more complex tumor models. Treatment with si6530 and si6548 led to 50% reduction in Irs2 mRNA expression in skeletal muscle (Figure 4A)^{27, 38}, but protein expression was low and difficult to detect by immunoblot. Therefore, IRS2 protein silencing was confirmed in the liver that expresses high levels of IRS2. IRS2 expression was suppressed by approximately 75% by the mouse-specific si6548, our top candidate for subsequent in vivo syngeneic tumor studies (Figure 4B). Finally, we confirmed robust, dose-dependent IRS2 protein silencing after treatment with si6548 in 4T1 cells, the intended cell line for our in vivo tumor studies, by performing four-point dose response studies (Supplemental Figure 3D).

Figure 4. IRS2 silencing in vivo does not cause hyperglycemia.

(A) Irs2 mRNA expression in skeletal muscle and (B) IRS2 protein abundance in liver 2 weeks after treatment with PBS, NTC DCA-siRNA, or IRS2 DCA-siRNA (n= 3–5 mice/group). (C) Accumulation of Cy3-labeled siRNA targeting Irs2 in liver, pancreas, muscle, and abdominal fat tissue from BALB/c mice 2 weeks after subcutaneous administration of PBS or 40 mg/kg dendrimer-siIrs2 (n= 10 mice/group). (D) Body weight of BALB/c mice after subcutaneous administration of 40 mg/kg dendrimer-siNTC or dendrimer-siIrs2 (n= 10 mice/group). (E) Terminal serum glucose concentrations in BALB/c mice 2 weeks after subcutaneous administration of 40 mg/kg dendrimer-siNTC or dendrimer-siIrs2 (n= 10 mice/group). UNT: untreated control, NTC: nontargeting control. Data presented as means ± SD. *p<0.05 **p<0.01 ***p<0.001 ****p<0.0001 by one-way ANOVA (Figures 4A,B) or Student’s t-test (4C per tissue, 4D per timepoint, 4E).

Acute suppression of IRS2 does not cause hyperglycemia

A major concern regarding therapeutic inhibition of components of the IIS pathway is disruption of insulin-dependent regulation of glucose uptake in metabolic tissues, such as the liver, fat or muscle, resulting in hyperglycemia and subsequent hyperinsulinemia. Mice with complete Irs2 knockout develop insulin resistance, which leads to hyperglycemia, and exhibit progressive loss of pancreatic islet mass.³⁹ To evaluate how acute suppression of Irs2 expression impacts glucose metabolism, we evaluated the biodistribution of our top candidate siRNA, dendrimer-conjugated si6548 (hereafter referred to as siIrs2), in metabolic tissues 2 weeks after subcutaneous injection of naïve, non-tumor bearing mice, a time point at which suppression of Irs2 expression is sustained (Figure 4B). We observed low siIrs2 uptake in the pancreas, skeletal muscle and abdominal fat, and higher accumulation in the liver, patterns that build upon previously reported data using other dendrimer-conjugated siRNA sequences (Figure 4C).²⁸ Importantly, overall body weights were equivalent and serum glucose levels were not elevated in mice treated with siIrs2 when compared with mice treated with NTC siRNA (siNTC), supporting that extended Irs2 silencing does not disrupt systemic glucose metabolism (Figures 4D and 4E).

siRNA-mediated silencing of Irs2 reduces mammary tumor growth

Previous studies have implicated IRS2 in multiple aspects of breast cancer progression, including tumor initiation and stemness, tumor cell migration and invasion, and metastasis using Irs2 knockout transgenic and orthotopic mouse tumor models.^{6, 8–10} We determined that the albumin-binding dendrimer conjugate enables superior siRNA delivery to mammary tumors compared to lipophilic DCA (Figure 1C–D). Thus, to evaluate the impact of acute, systemic siRNA-mediated Irs2 silencing on mammary tumor progression, we subcutaneously injected Cy3-labeled siIrs2 and a dendrimer-conjugated nontargeting control siRNA (siNTC) into mice bearing 4T1 tumors following the same siRNA injection protocol as was performed for flow cytometry studies (Figure 2A). A second cohort of tumor-bearing mice was also injected with a dendrimer-siRNA previously validated to potently silence the inflammatory mediator JAK1 (siJak1) as an additional control for non-specific effects of siRNA sequence on tumor silencing.⁴⁰ Tumors exhibited a slower growth rate in mice treated with siIrs2 with significant reductions (p=0.0008) demonstrated following administration of the second siRNA dose (Figure 5A, Figure S4A). As a result, terminal tumor volumes at sacrifice were significantly smaller (22.9% reduction, p=0.0066) in mice treated with siIrs2 than those from mice treated with siNTC (Figure 5B). No differences in tumor growth rate or terminal tumor volumes were observed for mice injected with siJak1 (Figure S4B–D).

Figure 5. Administration of siRNAs targeting Irs2 enables tumor gene silencing and inhibits tumor growth without causing hyperglycemia.

(A) 4T1 mammary tumor growth rates after siNTC or siIrs2 administration. (B) Tumor volumes before and after siNTC or siIrs2 administration. (C) Representative fluorescence microscopy of 4T1 mammary tumors from BALB/c mice subcutaneously injected with Cy3-labeled dendrimer-conjugated siNTC or siIrs2 or non-targeting control (NTC). (D) Accumulation of Cy3-labeled siRNA targeting Irs2 in 4T1 mammary tumor, pancreas, and liver tissue from BALB/c after subcutaneous administration of two 40 mg/kg doses across 6 days. (E) Irs2 mRNA expression in 4T1 mammary tumor biopsies from mice treated with siNTC or siIrs2. (F) Terminal serum glucose concentrations in 4T1 tumor-bearing mice after treatment with siNTC or siIrs2. (G) Comparison of serum glucose levels and relative tumor Irs2 mRNA expression in siNTC or siIrs2-treated tumor-bearing BALB/c mice. Relative Irs2 mRNA expression is presented as a percentage of average Irs2 mRNA expression in NTC siRNA-treated mice. (H) Representative H&E images and quantification of pancreatic islet number and size in siNTC or siIrs2-treated 4T1 tumor-bearing BALB/c mice. Analysis of n=8–10 mice/group. NTC: nontargeting control. scale bar = 20 μm. Line in box-and-whisker plots represents median value, while whiskers represent minimum-to-maximum values. ns=nonsignificant *p<0.05**p<0.01***p<0.001****p<0.0001 by Student’s t-test (Figures 5D (per tissue type), E, F, H); ns=nonsignificant,**p<0.01***p<0.001****p<0.0001 by one-way ANOVA (Figure 5A–B). Analysis of n=8–10 mice/group. R² determined by simple linear regression (Figure 5G). See also Figures S4 and S5.

Robust siRNA delivery to tumors was first established by fluorescence microscopy (Figure 5C, Figure S4E). We also confirmed antisense strand accumulation in tumor biopsies in a parallel assay that is independent of Cy3 fluorescence (Figure 5D, Figure S4F). Both the center and periphery of tumors contained substantial concentrations of siIrs2 and siJak1 antisense strand, suggesting penetration of siRNA throughout tumor tissue. As expected, siRNAs were also robustly delivered to the liver. There was no correlation between primary tumor volume and siIrs2 concentration in either tumor biopsies or the liver, suggesting that tumor size has minimal impact on siRNA accumulation in primary tumors and secondary tissues (Figure S5). Irs2 and Jak1 mRNA silencing was then assessed in biopsies sampled from non-necrotic areas of mammary tumors. Administration of siIrs2 and siJak1 led to significantly reduced Irs2 (26.7%, p=0.0042) and Jak1 (20.9%, p=0.0025) mRNA expression, respectively (Figure 5E, Figure S4G), indicating successful silencing of two different target genes in a sequence-specific manner.

Histological analysis of tumors from mice treated with either siNTC or siIrs2 revealed a trend towards decreased mitotic counts in tumors from mice treated with siIrs2 compared to siNTC (p=0.0534), and no difference in the percentage of necrosis between groups (Figure S6A–B). Analysis of terminal serum glucose concentrations in the tumor-bearing mice also revealed no significant differences in serum glucose concentrations between mice treated with siIrs2 and siNTC after 6 days of siRNA treatment (Figure 5F). Importantly, no correlation was observed between Irs2 tumor mRNA expression and serum glucose levels in siRNA-treated mice (Figure 5G). Moreover, neither the number nor size of pancreatic islets were decreased in mice treated with siIrs2 compared to siNTC (Figure 5H). Together, these findings confirm that albumin-binding siRNA is robustly delivered to mammary tumors to enable sequence-specific gene silencing and show, importantly, that modulation of IRS2 expression inhibits TNBC tumor growth without significantly perturbing systemic glucose homeostasis.

Suppression of Irs2 modulates the mammary tumor microenvironment

Suppression of tumor growth by siIrs2 may be the result of both tumor cell-intrinsic and extrinsic functions of IRS2, as this adaptor is expressed in primary tumor cells and in some cell populations of the TME. Stromal/immune cells that express IRS2 include macrophages, where it acts as a signaling adapter for the IL-4 receptor to drive polarization to the M2 phenotype, endothelial cells, and fibroblasts.^41–43 To examine the mechanism by which siIrs2 inhibits tumor growth, we first compared tumor immune populations from mice treated with siIrs2 and siNTC, as Irs2 silencing could alter the composition of the TME. Flow cytometric analysis revealed that the proportion of tumor dendritic cells, total macrophages, T cells, and other, undefined immune populations did not significantly differ between treatment groups (Figure 6A). Although the frequency of total tumor-associated macrophages (TAMs) was similar in siIrs2- and siNTC-treated mice, the frequency of macrophages that highly expressed CD206 (M2-like) was significantly decreased (p=0.0099) in tumors from mice treated with siIrs2 compared to siNTC (Figure 6B). As a result, the ratio of macrophages that highly expressed CD11c (M1-like) to those that highly expressed CD206 (M2-like) was significantly increased in siIrs2-treated mice (p=0.0286) (Figure 6C).

Figure 6. Irs2 silencing modulates the mammary tumor microenvironment.

(A-B) Flow cytometric analysis of the proportion of immune cell subsets (A) and macrophage subsets (B) within 4T1 mammary tumors from BALB/c mice treated with siNTC or siIrs2. (C) Quantification of the ratio of M1-like to M2-like macrophages in siRNA-treated tumors. (D) Representative immunofluorescence images and quantification of CD31-positive blood vessels in mammary tumors from siNTC or siIrs2-treated mice. (E) Relative Vegfa mRNA expression in tumor biopsies from mice treated with siNTC or siIrs2. (F) Representative immunofluorescence images of mammary tumor blood vessels stained for CD31 after siRNA administration (Cy3). Analysis of n=10 mice/group (Figure 5D,E) and of n=4 tumors/group (Figure 6A–C, F). NTC: nontargeting control, MΦ: macrophage. scale bars 100 μm (Figure 6D), or 10 μm (Figure 6F). Line in box-and-whisker plots represents median value, while whiskers represent minimum-to-maximum values. *p<0.05 by Mann-Whitney test (Figure 6C) ns nonsignificant *p<0.05**p<0.01****p<0.0001 by Student’s t-test (Figure 6A (per cell type), B, 6D–E).

M2-like TAMs are a key driver of angiogenesis, a process that provides nutrients and oxygen necessary to sustain tumor expansion and spread.^44–46 To assess the impact of Irs2 silencing on tumor vascularization, we measured the density and average area of blood vessels in the mammary tumors using the expression of CD31 as a marker of endothelial cells (Figure 6D). While blood vessel density was similar in tumors across groups, average vessel cross sectional area was significantly decreased (p<0.0001) in tumors from mice treated with siIrs2 compared to siNTC (Figure 6D). Expression of the growth factor Vegfa was also significantly decreased in tumor biopsies from siIrs2-treated mice (p=0.0397), suggesting a reduction in pro-angiogenic factors within the TME (Figure 6E). To evaluate if the reduction in tumor vascularization could be to the result of Irs2 silencing directly in endothelial cells, we evaluated Cy3 fluorescence in CD31 positive cells by fluorescence microscopy (Figure 6F). Co-localization of Cy3 and CD31 staining was not observed, revealing that siIrs2 was not internalized efficiently by endothelial cells.

Reduction of Irs2 expression in mammary tumors promotes tumor cell epithelial differentiation

Studies have reported a tumor cell-intrinsic role for IRS2 in tumor cell survival, epithelial-to-mesenchymal transition (EMT), migration, and invasion.^{4, 7–9, 47} We observed that neoplastic cells near the peripheral margins of tumors assumed a more fusiform morphology, indicative of dedifferentiation and EMT. To assess the impact of Irs2 silencing on this phenotype, we evaluated the expression of the cytoskeletal protein vimentin, which is upregulated in cells that have undergone EMT, in tumors from siIrs2- and siNTC-treated mice. We found that the number of neoplastic cells that expressed vimentin in tumors from mice treated with siIrs2 was reduced by ~75% compared to siNTC (p<0.0001), and that the intensity of vimentin expression was decreased by approximately 2-fold in these cells (p=0.001) (Figure 7A). We also evaluated the expression of the epithelial marker E-cadherin in a subset of tumors and observed an ~8-fold increase in the number of tumor cells that expressed E-cadherin (p=0.0017) and a ~2-fold increase in the intensity of E-cadherin expression in tumors from mice treated with siIrs2 (p=0.0257) (Figure S6C).

Figure 7. Primary mammary tumors exhibit less mesenchymal features after treatment with siIrs2.

(A) Representative immunohistochemistry images and quantification of vimentin expression in 4T1 mammary tumors from siRNA-treated mice. (B) Vimentin protein abundance in 4T1 cells 72 hours after in vitro treatment with cholesterol-conjugated siNTC or siIrs2. (C) Vimentin protein abundance in PyV-MT Irs1^−/−/Irs2^−/− cells expressing either empty vector (EV) control or IRS2. (D) Representative vimentin protein abundance in PyV-MT Irs1^−/−/Irs2^−/−: IRS2 cells after treatment with either shRNAs targeting GFP (control) or vimentin. (E) Transwell migration assay of PyV-MT Irs1^−/−/Irs2^−/− cells expressing either empty vector (EV) or IRS2 after treatment with shRNAs targeting GFP or vimentin. Analysis of n=10 mice/group (Figure 7A). n=3 (Figure 7B,C,E) independent experiments. Figure 7D is representative of 3 independent experiments. NTC: nontargeting control, EV: empty vector. Scale bars 20 μm. Line in box-and-whisker plots (Figure 7A) represents median value, while whiskers represent minimum-to-maximum values. Data presented as means ±SD (Figures 7B,C,E). *p<0.05**p<0.01****p<0.0001 by Student’s t-test (Figures 7A–C) ***p<0.001****p<0.0001 by one-way ANOVA (Figure 7E). See also Figure S6.

We further dissected the role of IRS2 in regulating vimentin and E-cadherin in vitro using 4T1 cells and Irs1^−/−/Irs2^−/− cells isolated from PyV-MT murine tumors. Treatment of 4T1 cells in vitro with cholesterol-conjugated siIrs2 led to a significant decrease in vimentin protein abundance compared to siNTC (p=0.0105), supporting that in vivo reduction of vimentin expression was due to tumor cell-intrinsic Irs2 silencing (Figure 7B). However, E-cadherin protein abundance did not increase in siIrs2-treated 4T1 cells in vitro, indicating that IRS2 may regulate E-cadherin expression in vivo through tumor cell-extrinsic mechanisms (Figure S6D). Restoration of IRS2 expression in PyV-MT:Irs1^−/−/Irs2^−/− cells increased vimentin protein abundance (Figure 7C), suggesting that IRS2 is also sufficient to upregulate vimentin in a tumor cell-intrinsic manner. PyV-MT: Irs1^−/−/Irs2^−/− cells with restored expression of IRS2 exhibited enhanced migratory potential and knockdown of vimentin expression using 2 different shRNAs in these cells reduced this cell migration (Figure 7D–E). Together our results demonstrate a robust impact of Irs2 silencing on EMT markers and suggest a shift toward a more differentiated, epithelial phenotype upon suppression of IRS2 expression in mammary tumors.

Tumor cell de-differentiation is connected to breast cancer metastasis.⁴⁸ Previous studies using knockout transgenic mouse tumor models have found that IRS2 supports breast cancer metastasis.⁶ We evaluated lung metastatic burden in siIrs2- and siNTC-treated mice to investigate if acute Irs2 silencing in mammary tumors impacts dissemination. A similar percentage of mice developed metastases in both groups (8 out of 10 mice), and there was no significant difference in the number of metastatic lesions within the lungs (Figure S7A). Although we observed a similar number of micrometastases, defined as lesions with a cross-sectional area less than 0.01mm², in the lungs of mice treated with siIrs2 and siNTC, there were 2-fold fewer metastatic lesions with a cross-sectional area greater than 0.01mm² in mice treated with siIrs2 compared to siNTC (15 vs 30), indicating a downward trend in the size of lung metastases after treatment with siIrs2 (Figure S7B). In summary, our data support that partial modulation of IRS2 expression (25% Irs2 silencing) has a profound biological impact on both mammary tumor cells and the TME, culminating in a significant reduction in primary tumor growth.

Discussion

The potential of siRNA therapeutics to offer unprecedented, selective silencing of genes whose products are unable to be targeted by other treatment modalities has led to significant interest in the use of siRNAs as targeted agents in cancer. Achieving efficient delivery to tumors, however, has been a limitation for the clinical translation of this technology in oncology. In this study, we demonstrate that an albumin-binding dendrimer conjugate significantly improves siRNA delivery to TNBC tumors compared to DCA and promotes cellular uptake in tumor cells and heterogenous cell types of the TME. Treatment with dendrimer-siRNAs that selectively silence IRS2, an enzymatically inactive adaptor protein that contributes to TNBC progression, achieves target gene silencing and slows tumor growth without perturbing glucose homeostasis in an immunocompetent model of TNBC. Suppression of Irs2 led to a decrease in M2-like TAMs and a higher ratio of M1-like to M2-like TAMs, reflecting a shift in macrophage polarization toward a more anti-tumor population. Primary tumors also exhibited diminished and increased expression of the EMT markers vimentin and E-cadherin, respectfully, and reduced vascularization, reflecting effects on both tumor cells and the TME in response to Irs2 tumor silencing. Together, our results establish that albumin-binding conjugates enhance functional and cell-specific siRNA delivery in TNBC tumors and provide mechanistic insight into the role of IRS2 in TNBC progression. These findings highlight the feasibility and benefits of siRNA-based inhibition of nonenzymatic proteins in cancer.

Albumin is an attractive carrier to improve the pharmacokinetic properties of siRNA therapeutics in cancer, as it extends the circulation of rapidly cleared drugs and improves drug accumulation at sites of tumor proliferation.^49,50 Albumin is internalized by tumor cells through macropinocytosis and receptor-mediated pathways that promote the intracellular delivery of siRNA.^{49, 51–53} 4T1 cells express high levels of SPARC, which has been shown to mediate cellular uptake of albumin-coated nanoparticles and Nab-paclitaxel.^54,55 Of further relevance to our study, inhibition of IGF1R signaling in tumor cells mimics glucose deprivation and promotes micropinocytosis.⁵⁶ Therefore, suppression of IRS2 expression may further enhance dendrimer-siIrs2 uptake through a feed-forward mechanism that enhances macropinocytotic siRNA uptake. Previous studies have reported that siRNA conjugated to albumin-binding diacyl lipids or nanocomplexes of siRNA and albumin enable enhanced and productive tumor delivery, providing evidence that albumin-binding conjugates may overcome a key barrier in the use of siRNAs as chemotherapeutic agents.^{26, 29, 30} These studies, however, did not address the cell-type specificity of siRNA uptake in the heterogeneous TME. Our study reveals additional benefits of albumin-binding siRNA conjugates by demonstrating that dendrimer-siRNA improves cellular uptake in neoplastic cells and diverse stromal and immune cell populations within syngeneic TNBC mammary tumors. These findings demonstrate the ability to co-target multiple cell populations using a single siRNA chemical scaffold with albumin-binding properties. Moreover, our data support that cell-type specific silencing in tumors may also be achieved using albumin-binding siRNA by altering the siRNA antisense strand sequence to target genes that are unique to one cell population.

Our study is the first to demonstrate that acute silencing of IRS2 can impede mammary tumor growth without causing hyperglycemia or reducing pancreatic islet density or size, which obviates a major concern with currently available therapeutics that target the IIS pathway in cancer. Although total Irs2 knockout mice exhibit dysregulated glucose homeostasis, our data support that acute and selective targeting by siIrs2 does not achieve sufficient suppression of IRS2 expression to disrupt glucose uptake significantly in metabolic tissues. We hypothesize that this lack of dysregulation is related to the tissue-selectivity of dendrimer-siRNA uptake, as only modest levels of siIrs2 were detected in metabolic tissues other than the liver. Given that complete Irs2 knockout in hepatocytes causes only mild glucose intolerance, substantial siIrs2 uptake and gene silencing in the liver alone are not sufficient to cause hyperglycemia.^57,58 However, we acknowledge that our analysis was limited to a 2-week siIrs2 treatment, and dosing schedules for oncology purposes are likely to require chronic exposure to siIrs2. Future studies are required to determine if longer-term treatment with siIrs2 impacts glucose metabolism and to establish an optimal dosing schedule for therapeutic efficacy that promotes tumor reduction, but limits dysregulation of glucose metabolism. Potential risks of siIrs2-based therapy can be mitigated by close monitoring of blood glucose levels and treatment with well-established medications that manage hyperglycemia such as metformin and SGLT2 inhibitors.

Cell-intrinsic roles for IRS2 in regulating tumor cell functions that contribute to disease progression have been previously established using IRS2 knockout tumor models in vitro and in vivo.^6–8,10,47 We observed that acute in vitro and in vivo siRNA-mediated suppression of Irs2 reduces expression of vimentin in neoplastic cells, and that tumor cell-intrinsic regulation of this EMT factor by IRS2 impacts migration, a process that underlies invasive tumor growth.⁵⁹ Treatment with siIrs2 also alters the TAM compartment and decreases tumor vascularization. Our data suggest that Irs2 silencing reduces tumor angiogenesis through mechanisms independent of endothelial cell-intrinsic knockdown. IRS2 acts as a signaling adaptor for the IL-4 cytokine receptor to promote M2-like macrophage polarization⁴³, which generates a pro-tumor, anti-inflammatory macrophage population that promotes tumor growth and progression through the secretion of factors, such as VEGF, that support angiogenesis and metastasis.⁴⁴ Promotion of tumor vascularization would facilitate tumor cell access to nutrients and oxygen to promote tumor growth. We posit, therefore, that Irs2 silencing within both tumor cells and macrophages inhibits processes essential for tumor growth, which emphasizes the value of an siRNA approach that co-targets both tumor cells and the TME and the importance of using immunocompetent mouse models to fully evaluate the functional consequences of siRNA targeting. Future studies that provide a more detailed analysis of Irs2 silencing within tumor cells and specific macrophage subsets will provide a greater understanding of the cell-specific mechanism(s) by which Irs2 silencing suppresses mammary tumor growth.

A technical challenge in this study is to accurately assess the impact of Irs2 silencing on tumor metastasis. It is important to note that the 4T1 syngeneic tumor model is particularly aggressive and metastasis occurs very early.⁶⁰ While our dosing strategy delivered an excess of oligonucleotides to tumor tissue at the time of injection and resulted in a significant reduction in tumor growth, it is likely that metastasis had initiated prior to the first injection of siRNA. Initiation of treatment at earlier stages of TNBC tumor development would address the overall efficacy of IRS2-selective siRNA as a treatment modality for preventing metastatic disease. Moreover, assays that initiate metastatic growth in the absence of a primary tumor would provide the opportunity to determine if dendrimer-siRNAs can be delivered directly to secondary metastatic lesions, a property that would benefit patients with advanced, metastatic disease.

Use of siRNA therapeutics targeting nonenzymatic adaptor proteins like IRS2 has the potential to simultaneously modulate signal transduction downstream of several receptors. IRS2 not only serves as an adaptor for IIS receptors, but also for integrin, cytokine, and other growth factor receptors.^43,61,62 A single therapeutic that targets IRS2, therefore, is likely to inhibit several parallel processes that contribute to cancer progression, which could provide more clinical benefit than currently available inhibitors that suppress a narrower scope of signaling pathways. The sequence-specific precision of siRNA technology would enable silencing of other adaptors implicated in breast as well as other cancers, such as SHC, GRB2, CRK, and MYD88^63–66, to inhibit additional downstream signaling pathways as a single or combinatorial interventional approach, especially those for which small molecule inhibitors have not yet been developed.

In summary, our results demonstrate that albumin-binding siRNAs are robustly and efficiently delivered to diverse cell populations within TNBC tumors. In addition, we show that silencing the intracellular adaptor protein IRS2 slows breast tumor growth without causing hyperglycemia, potentially eliminating a significant source of morbidity for cancer patients who would benefit from therapeutics that target the IIS signaling pathway. Increased expression of intracellular adaptors such as IRS2 is associated with poorer survival in patients with other cancers, such as non-small cell lung cancer (NSCLC), cholangiocarcinoma, and pancreatic ductal adenocarcinoma^67–70, and an siRNA-based approach to inhibit this class of proteins could have broader clinical benefit beyond breast cancer. Our work provides evidence that use of albumin-binding siRNAs to inhibit nonenzymatic oncogenes whose protein products cannot be targeted by currently available therapies is an effective strategy to successfully inhibit tumor progression, suggesting that further development of such therapies as antineoplastic agents has the potential to significantly improve clinical outcomes for oncology patients.

Materials and Methods

Cell Culture

MDA-MB-231, 4T1, and FL83B cells were obtained from ATCC Cell Biology Collection. PyV-MT mammary tumor cells were isolated from female FVB MMTV-PyV-MT:Irs1^fl/flIrs2^fl/fl mice, and PyV-MT:Irs1^−/−/Irs2^−/− cells were generated by infection with adenoviral Cre recombinase as previously described.⁷¹ MDA-MB-231 and 4T1 cells were grown in RPMI media (Gibco) containing 10% FBS, FL83B cells were grown in F-12K media (Gibco) containing 10% FBS, and PyV-MT cells were grown in DMEM media containing 10%FBS. All cells were grown at 37°C with 5% CO₂ supply and tested negative for mycoplasma by PCR (Abm).

Antibodies

Primary antibodies used for immunoblots include rabbit anti-IRS2 (Cell Signaling Technology, Cat# 4502), rabbit anti-IRS1 (Bethyl Laboratories, Cat# A301–158A), mouse anti-actin (Thermo Fisher Scientific, Cat# MA5–11869), mouse anti-tubulin (Cell Signaling Technology Cat#3873), and mouse anti-GAPDH (Santa Cruz Biotechnology, Cat# sc-32233). Anti-vimentin antibodies (Proteintech, clone 6K21, Rosemont, IL, USA) were used for immunohistochemical analysis. Immunofluorescence was performed using a mouse anti-CD31 (Cell Signaling Technology Cat# 77699), rabbit anti-F4/80 (Cell Signaling Technology #70076) and rabbit anti-E-cadherin (Invitrogen PA5–85088). Fluorochrome-labeled antibodies against mouse CD326 (G8.8), CD45 (30-F11), CD206 (C068C2), CD8a (53–6.7), I-A/I-E (M5/114.15.2), CD4 (RM4–5), CD11c (N418), TCRβ (H57–597), and F4/80 (BM8) were purchased from Biolegend (San Diego, CA) and used for flow cytometric analysis.

Design of human-specific, mouse-specific, and cross-reactive siRNAs targeting IRS2

siRNAs were designed based on algorithm features developed by Shmushkovich et al.³⁶ Briefly, siRNAs were designed based on a 21-nucleotide targeting sequence from human IRS2 (accession number NM_003749.3) and mouse Irs2 (accession number NM_001081212.2). Exclusion criteria for sequences included: 1) >56% GC content, 2) single-nucleotide stretches of four or more, or 3) perfect homology to miRNA seeds at position 2–7 of the antisense strand. Sequences were also excluded if position 2–17 of the antisense strand had full complementarity to non-target mRNAs to minimize off-target effects. Cross-species targeting was determined based on perfect homology of positions 2–17 within the target sequence to both species. siRNAs were named based on the position in the target sequence from which they were extracted (e.g. human IRS2 siRNA 5967 was extracted from positions 5967–5988 of NM_003749.3).

Oligonucleotide synthesis, deprotection, and purification

One set of 48 fully-chemically modified antisense and sense strands was synthesized for in vitro screening at a 1μmol scale using standard solid-phase phosphoramidite chemistry on a Dr. Oligo 48 high-throughput oligonucleotide synthesizer (Biolytic, Fremont, CA). Compounds for in vivo studies were synthesized on diverse scales using a MerMade 12 synthesizer (Biosearch Technologies, Novato, CA). Standard RNA 2′-O-methyl, 2′-fluoro modifications were applied to improve siRNA stability. Extended nucleic acid (exNA)-modified RNA (exNA) was used at the 3’ terminus of the antisense strands in compounds used for in vivo tumor silencing studies.³⁷ Phosphoramidites were purchased from Chemgenes, Wilmington, MA and Hongene Biotech, Union City, CA.

Sense strands of in vitro compounds were synthesized on a cholesterol-conjugated solid support (Chemgenes). DCA conjugated sense strands were synthesized on custom-functionalized controlled pore glass (CPG) supports as previously described²⁷, containing a (dT)₂ cleavable linker between the oligonucleotide and the conjugate. Synthesis of the amphiphilic dendrimer-conjugated sense strand was carried out on Unylinker CPG (ChemGenes) and commercially available amidites were used to build the dendritic moiety on the 5′ end as previously described.^{33, 72} Cy3-phosphoramidites (GenePharma) were used for fluorescence labeling of the 5’ of sense strands for in vivo studies. For Cy3-labeled dendrimer strands, Cy3-DMT-phosphoramidite (Chemgenes) was placed in between the conjugate and the sense strand. Antisense strands were synthesized on Unylinker CPG (Chemgenes), bis-cyanoethyl-N, N-diisopropyl CED phosphoramidite (Chemgenes) was used to introduce a 5’-mono-phosphate for in vitro experiments, and 5’-(E)-vinylphosphonate-2’-O-Me-Uridine phosphoramidite (Hongene) was applied for in vivo studies.

In vitro strands were cleaved and deprotected on synthesis columns (on-column) with ammonia gas (Airgas Specialty Gases) for 90 minutes at 65°C followed by on-column ethanol precipitation and elution with nuclease-free water. In vivo sense strands were cleaved and deprotected using 28% aqueous ammonium hydroxide solution for 20 hours at 55°C. In vivo antisense strands were cleaved and deprotected using 3% diethylamine in aqueous ammonium hydroxide solution for 20 hours at 35°C, while a 1:1 mixture of ammonium hydroxide and 40% aqueous monomethylamine was used for the sense strands for 120 minutes at 25°C. The controlled pore glass was subsequently filtered and rinsed with 50% ethanol in water and dried overnight under vacuum. Oligonucleotides for in vivo experiments were HPLC-purified using an Agilent 1290 Infinity II system (Agilent Technologies, Santa Clara CA) with a PRP-C18 column for Cy3 labeled and lipid-conjugated sense strands and an ion-exchange column for antisense and dendrimer-conjugated sense strands. Purified oligonucleotides were desalted by size-exclusion chromatography and characterized by LC-MS analysis on an Agilent 6530 accurate-mass quadrupole time-of-flight (Q-TOF) LC/MS (Agilent Technologies).

Identification of IRS2 selective siRNAs

For in vitro screening of siRNA candidates, cells were treated with candidate cholesterol-conjugated siRNAs for 72 hours in a 50/50 mixture of complete growth media supplemented with 6% FBS and Opti-MEM (Gibco). Screening concentrations for MDA-MB-231 (0.1875μM), FL83B (1.5μM), and 4T1 (1.5μM) cells served as maximal dose for dose-response assays. Cells were then lysed using lysis mixture (Invitrogen #13228) diluted 1:2 with water containing 0.2mg/mL proteinase K (Invitrogen, #25530–049) and heated at 55 °C for 30 minutes before measuring gene expression.

Human IRS2 and mouse Irs2 mRNA levels were determined using the Quantigene 2.0 assay (Affymetrix). Cells were lysed in 250μL homogenizing solution (Invitrogen Cat. QG0517) containing 0.2 mg/mL proteinase K (Invitrogen, #25530–049) and heated at 55°C for 30 minutes. Lysed cells were combined with diluted probe sets (mouse Irs2; SB-29780, human IRS2; SA-10203; ThermoFisher), added to the capture plate and signal was amplified and detected according to manufacturer’s protocol. Luminescence was detected using a SpectraMax M5 microplate reader and SoftMax Pro 6.3 software (Molecular Devices).

For protein analysis, cells were solubilized at 4°C in lysis buffer (20 mM Tris buffer, pH 7.4, 1% Nonidet P-40, 0.137 M NaCl, 10% glycerol) containing protease inhibitors (Roche) and phosphatase inhibitors (Roche).

In vivo siRNA studies

Mice were housed in a pathogen-free animal facilities at UMass Chan Medical School with 12 hour light/12 hour dark cycle and free access to food and water. Female BALB/c mice (Charles River Laboratories) were used for all experiments and were 7- to 9-weeks old of age at the initiation of each study. For all animal studies, mice were injected with either phosphate buffered saline (PBS control) or fully chemically modified (unconjugated, lipid-conjugated, or dendrimer-conjugated) siRNA that was suspended in 1xPBS for a final siRNA concentration of 40mg/kg (150μL injection volume).

To confirm in vivo silencing efficacy of IRS2-targeting siRNAs and their impact on systemic glucose metabolism, naïve mice were subcutaneously injected once with either PBS or 40mg/kg siRNA. Animals were euthanized 2 weeks post-injection, and tissues were collected and either placed in RNAlater and stored at 4°C or flash frozen and stored at −80°C. Two 2 mm punches were collected from tissues stored in RNAlater, placed into QIAGEN Collection Microtubes holding 3 mm tungsten beads and lysed in 300 μL homogenizing solution containing 0.2 mg/mL proteinase K using a QIAGEN TissueLyser II. Samples were heated at 65°C for 30 minutes and then centrifuged at 568xg for 5 minutes at room temperature to remove debris. mRNA levels were determined using the Quantigene 2.0 assays (Affymetrix). For protein analysis, frozen tissues were homogenized in RIPA buffer containing protease and phosphatase inhibitors.

To determine the optimal siRNA conjugation strategy for mammary tumor delivery, 4T1 cells (1.6×10⁵) suspended in 30 μL PBS were orthotopically injected into the #4 mammary fat pad of 7- to 9-week-old female BALB/c mice. Animals were palpated daily to detect tumor initiation, and palpable tumors were measured daily with calipers to monitor tumor growth. When tumors reached ~200 mm³ in volume, mice were injected either subcutaneously (150 μL) or intratumorally (50 μL) with either PBS or 40 mg/kg siRNA. Animals were euthanized 48 hours post-injection, and tissues were collected for fluorescence microscopy analysis.

To investigate the role of Irs2 silencing in mammary tumor progression, 4T1 cells (1×10⁵) were injected into the #2 mammary fat pad and palpable tumors were measured daily with calipers. When tumors reached ~200 mm³ in volume, the mice were randomized into groups and injected subcutaneously with either PBS or 40 mg/kg siRNA, and again 72 hours after the first injection. Mice were euthanized 72 hours after the second injection and tissues were collected to measure tumor and secondary tissue siRNA delivery, gene silencing, and histopathology. Tumor volumes were calculated from caliper measurements using the equation (W²*L)/2 to track tumor growth in vivo, and terminal tumor volumes were calculated using the equation V=(4/3)π*(L/2)*(W/2)*(D/2).^{73, 74} Tumor measurements were confirmed by an individual who was blinded to experimental intervention.

Fluorescence Microscopy

A portion of tumors and secondary tissues were paraffin-embedded and sectioned by the Morphology Research Core Facility at UMass Chan Medical School. Sections were stained with 4′,6-diamidino-2-phenylindole (DAPI) to visualize nuclei and fluorescent images were acquired with a Leica DMi8 inverted microscope (Leica Microsystems) and a Nikon A1 confocal microscope (Nikon). Whole-tissue images were taken using a 10X objective and confocal images were taken using a 60X oil immersion objective. Images were adjusted for equal brightness and contrast to compare the degree of siRNA biodistribution.

Immunofluorescence Staining

Formalin-fixed, paraffin-embedded tissue sections mounted on glass slides were deparaffinized with xylene and rehydrated using the following series of 5-minute ethanol washes: 100%, 95%, 70%, and 50%, with a final wash with deionized water. An antigen retrieval step was performed using heat and sodium citrate unmasking solution (Cell Signaling Technology #14746S). After antigen retrieval, slides were briefly washed with PBS and placed in blocking solution of PBS containing 0.1%Triton-X 100 (PBST) and 1% bovine serum album (BSA) for 1 hour at room temperature. Slides were then incubated in blocking solution containing an anti-CD31, anti-F4/80, or E-cadherin antibody overnight at 4°C. Sections were washed with PBST, then incubated in blocking solution containing secondary antibodies for 1.5 hours at room temperature. Slides were again washed with PBST, incubated in DAPI diluted in PBS to visualize nuclei, and coverslipped using Prolong Glass mounting medium (ThermoFisher Cat#P36984). Slides were allowed to cure overnight and then imaged with a 20X objective using a Leica DMi8 microscope or a 60X objective using a Nikon A1 confocal microscope. For analysis of tumor vascularization, three representative images were taken per tumor, and blood vessel density and cross-sectional area were quantified and confirmed by three individuals who were blinded to experimental intervention using ImageJ software.

Flow Cytometry

A portion of collected tumors that were stored in Crytostor CS10 media (Sigma-Aldrich) were dissociated into a single cell suspension using the Miltenyi mouse tumor dissociation kit (Miltenyi Biotec Cat#130–096-730) according to manufacturer’s protocol. Cell suspensions were treated with TruStain FcX PLUS (anti-CD16/32) antibody (Biolegend #156603) for 5min at 4C, then stained with primary antibodies suspended in PBS + 1% FBS supplemented with Brilliant Stain Buffer (BD Biosciences #659611). After washing with PBS, cells were stained with a Fixable Viability Dye (Invitrogen #65–0865-14) and fixed using a formaldehyde-based fixative (Miltenyi Cat# 130–090-477). Samples were run on a Cytek Aurora flow cytometer and analyzed using FlowJo v10 software. Single-stained cell suspensions and UltraComp eBeads Plus (ThermoFisher #01–3333-42) compensation beads were used as compensation controls.

Stem-loop RT-PCR

Biodistribution levels of siRNA in mammary fat pad, tumor, pancreas, and liver tissue were quantified using Stem-loop RT-PCR.^{75, 76} Tissues were lysed in homogenizing solution (Invitrogen) containing 0.2 mg/mL proteinase K, heated to 65°C for 30 minutes, and spun down to remove debris. Stock tissue supernatants were diluted 1:60 with nuclease-free water for downstream analysis. An 8-point standard curve ranging from 0–100 nM was generated using diluted control tissue supernatants spiked with either unconjugated, lipid-conjugated, or dendrimer-conjugated siRNAs as a template for cDNA synthesis. Standard curve and experimental samples were heated to 95°C for 20 minutes and combined with 1μM RT primers (see Table 1 for primers), 10 mM dNTPs (New England Biolabs; N0447L), 10X M-MuLV RT reaction buffer (New England Biolabs; B0253S), and M-MuLV reverse transcriptase (New England Biolabs; M0253L) for cDNA synthesis. Synthesis of cDNA was performed using the following protocol: 16°C for 30 minutes for primer annealing, 42°C for 30 minutes for reverse transcription, and then 70°C for 15 minutes for enzyme inactivation. For the qPCR reaction setup, a master-mix consisting of 5 μL of Power-Up SYBR qPCR Mastermix (Applied Biosystems), 0.05 μL forward primer at 100 μM, 0.05 μL universal reverse primer at 100 μM, and 2.9 μL water was mixed with 2 μL of cDNA template diluted 1:1 with nuclease-free water (see Table 1 for primers). The thermal cycling program was set as followed: 2 minutes incubation at 50°C, a 2-minute denaturation at 95°C, then 50 cycles of 10 seconds at 95°C and 20 seconds at 60°C.

Table 1.

Stem loop RT-PCR primers

Primer	Sequence (5’->3’)
HTT RT	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACAAATATCA
IRS2 RT	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACAATTCTTT
JAK1 RT	GTCGTATCCAGTGCAGGGTCCGAGGTATTCGCACTGGATACGACAAACATCA
HTT forward	CCGCCTACCGTGGTTAATCTCTTTACTG
IRS2 forward	ACCGTGGTA AACATGTACCCAA
JAK1 forward	CCTACCGTGGTATCGCTTGTAGCTGAT
Universal reverse	CGTGCAGGGTCCGAGGT

Serum Glucose Analysis

Whole blood was collected from mice via the facial vein at time of euthanasia and serum was isolated using serum separator tubes (BD #365967). Serum glucose concentrations were then measured using a glucose colorimetric detection kit (ThermoFisher Scientific Cat# EIAGLUC) according to manufacturer’s protocol.

Immunoblotting

Cell or tissue extracts containing equivalent amounts of protein were resolved by SDS-PAGE and transferred to nitrocellulose membranes. Membranes were blocked for 1 hour at room temperature in 50 mM Tris buffer, pH 7.5 with 0.15 M NaCl and 0.1% Tween-20 (TBST) and 5% (wt/vol) dry milk, incubated overnight at 4°C in TBST containing 3% (wt/vol) bovine serum albumin (BSA) and primary antibodies. Membranes were washed with TBST and then incubated for 1–2 hours at room temperature in 5% blocking buffer with milk containing peroxidase-conjugated secondary antibodies. Bands were detected by chemiluminescence using a ChemiDoc XRS+ system (Biorad) and quantified by densitometry using ImageJ. Signals were normalized to total housekeeping genes to compare relative protein abundance.

Tumor and Pancreas Histopathology and Immunohistochemistry

Sections of tumors and pancreas tissue were stained with hematoxylin and eosin (H&E) by the Morphology Research Core Facility at UMass Chan Medical School using standard protocols. Stained slides were evaluated for histomorphologic characterization and determination of the percent neoplastic composition of the tissue, percent necrotic composition of the neoplastic population, and the number of mitotic figures per ten contiguous high-powered fields (HPFs; 2.37mm²). To determine pancreatic islet cell mass, the total number and cross-sectional area of all islets were compared between experimental groups.

Immunolabeling of FFPE sections of all tumor specimens was performed by using 5 μm sections with a recombinant anti-vimentin or an anti-E-cadherin antibody using standard protocols. Antigen retrieval was performed using sodium citrate solution. Normal murine adipose tissue was used as a positive control while primary antibody was removed in negative controls. Stained specimens were scored as having no (0), weak (1), moderate (2), or strong (3) immunoreactivity, and the percentage of neoplastic cells that stained positive was recorded. All histologic evaluations, including the identification and enumeration of mitotic figures, and immunohistochemical scoring were performed by a board-certified veterinary pathologist who was blinded to experimental intervention.

Transwell Migration Assays

PyV-MT:Irs1^−/−/Irs2^−/− cells stably expressing either a FLAG-tagged IRS2 or an empty vector control were transfected with shRNAs targeting either GFP or vimentin (TRCN0000089828; TRCN0000089830) using Lipofectamine 3000 for 24 hours. Cells were plated into the top chamber of 6.5 mm Transwells (8 μm pore size) (Corning), 3T3-conditioned media was added to the bottom chamber, and cells were allowed to migrate for 5 hours at 37°C. Cells that had migrated to the lower surface of the filters were fixed in methanol for 10 minutes and mounted on glass slides using a Vectashield mounting medium containing DAPI (Vector Laboratories, Burlingame, CA). Migration was quantified by counting the number of stained nuclei in four independent fields in each Transwell.

Statistical Analysis

All data were analyzed using GraphPad Prism 10 software (GraphPad Software, Inc., San Diego, CA, USA). Student’s t-test was applied and p-value of <0.05 was considered to indicate statistical significance when comparing 2 groups. For comparison of more than 2 groups, one-way ANOVA followed by Tukey’s multiple comparisons was applied.

Study Approval

All animal experiments were performed in accordance with animal care ethics approval and guidelines of the University of Massachusetts Chan Medical School Institutional Animal Care and Use Committee. All procedures were approved under the Protocol #201900333 (Shaw Laboratory) and #202000010 (Khvorova Laboratory) and in accordance with the National Research Council’s The Guide for the Care and Use of Laboratory Animals. The mice utilized in this study were female because human cancer largely affects female patients.

Shaw and colleagues establish that functional and cell-specific siRNA delivery in breast tumors is enhanced by an albumin-binding conjugate and provide mechanistic insight into the role of IRS2 in breast cancer progression. Their findings support the feasibility and benefits of siRNA-based inhibition of nonenzymatic proteins in cancer.

Data and materials availability

All data are available in the main text or the supplementary materials. Requests for further information and resources should be directed to and will be fulfilled by LMS (Leslie.shaw@umassmed.edu). All unique/stable reagents generated in this study are available from LMS with a completed materials transfer agreement.