Phosphorylation of RAB7 by TBK1/IKKϵ Regulates Innate Immune Signaling in Triple-Negative Breast Cancer

Triple-negative breast cancers (TNBC) are typically aggressive and account for a disproportionate number of metastatic cases and breast cancer deaths (1–3). TNBCs are also heterogeneous, with diverse somatic mutations, gene amplifications, and deletions as reflected by multiple subtypes defined by different gene signatures (4); however, PTEN loss is a common event (5–9). In addition, a significant proportion of TNBCs also exhibit a high amount of immune cell infiltration and elevated cytokine production, which we previously linked to aberrant expression of IκB kinase ϵ (IKKϵ), which promotes feedforward production of NF-κB–associated cytokines with its homolog TANK-binding kinase 1 (TBK1; ref. 10).

Ras-related protein Rab-7a (Rab7) is a member of a larger family of Ras GTPases and has been shown to be an important modulator of phagocytosis (11), endosomal sorting (12), and the biogenesis of lysosome-related organelles (13). Although Rab7 has been extensively studied for its role in endosomal trafficking and maturation, recent studies have highlighted the role of Rab7 in attenuating receptor signaling in tumors (14, 15). The attenuation of receptor signaling by Rab7 occurs for outer membrane receptors such as EGFR (14) as well as intracellular signaling adaptors such as stimulator of interferon genes (STING; ref. 15). In each case, Rab7 is directly responsible for protein degradation by trafficking receptor/adaptor-containing vesicles to the lysosome. Notably, the tumor-suppressor PTEN was also recently identified to regulate Rab7 function by dephosphorylation of serine-72 (S72), promoting its mislocalization; PTEN loss or constitutively phosphorylated Rab7-S72 thus increased intracellular EGFR activation as the receptor was internalized, but its degradation was impaired (14).

Here, we performed integrated phosphoproteomic studies to search for novel TBK1/IKKϵ substrates, which yielded Rab7-S72 as a top hit. Through subsequent studies in PTEN-null TNBC cells, we identify TBK1/IKKϵ mediated phosphorylation of this site as a key regulator of Rab7 mislocalization, which sustains levels of the upstream TBK1 adaptor STING, and thus promotes hyperactive innate immune signaling. These findings begin to uncover a key molecular event that deregulates innate immune signaling in PTEN-null TNBC cells, with potentially important therapeutic implications.

HEK 293T and breast cancer cell lines (HCC70, HCC1143, HCC1187, HCC1937, MDA-MB-231, MDA-MB-468, MCF7, MCF10A, T-47D, SKBr3, ZR-751) used in this study were obtained from the ATCC. HEK 293T and MDA-MB-231 cells were cultured in DMEM (Thermo Fisher Scientific), whereas all other cell lines were grown in RPMI-1640 (Thermo Fisher Scientific) with 10% FBS (Gemini Bio-products) and 1x penicillin–streptomycin (Gemini Bio-products). Jurkat T cells expressing CXCR3 were generated as previously described (16) and grown in RPMI-1640 with 10% FBS. All cell lines were confirmed by short tandem repeat profiling, tested mycoplasma negative by PCR as recent as 2 months prior to last experiment, and used between passages 3 and 15.

All plasmids were generated using Gateway Cloning (Invitrogen). Mutant plasmids, kinase-dead TBK1, catalytically inactive PTEN mutant (PTEN-C124S), Rab7-S72E, and Rab7-S72A were generated with PCR-based site-directed mutagenesis. Rab7 mutants were cloned into the V5-tagged pLX304 plasmid. TBK1 clones, kinase-dead TBK1, wild-type TBK1 (TBK1-WT), and control EGFP were cloned into the pLX980 plasmid. PTEN clones, wild-type PTEN (PTEN-WT), and PTEN-C124S were cloned into the pLX307 plasmid. HEK 293T cells were used for the generation of lentivirus for the establishment of Rab7- and PTEN-mutant cell lines. Transfections occurred over 48 hours with 1 μg plasmid using X-tremeGENE 9 DNA transfection reagent (Sigma-Aldrich). After transfection, viral media were collected and combined with 8 μg mL⁻¹ polybrene (Santa Cruz Biotechnology) prior to addition to MDA-MB-468 cells. PTEN mutants were selected with puromycin, whereas all other mutants were selected with blasticidin.

To determine TBK1-specific phosphorylation sites, we utilized a previously published protocol for stable isotope labeling in cell culture (17, 18). In brief, HEK 293T cells were cultured for 1 week in low-glucose DMEM supplemented with either light [l-arginine (R0) and l-lysine (K0)], medium [l-arginine ¹³C₆-HCL (R6) and l-lysine-4,4,5,5d₄ (K4)], or heavy [l-arginine ¹³C₆-¹⁵N₄-HCl (R10) and l-lysine ¹³C₆, ¹⁵N₂-HCl (K8)] amino acids (Sigma-Aldrich). Cells were transfected using FuGENE (Promega) with the plasmids control EGFP, TBK1-WT, or kinase-dead TBK1 and then selected with blasticidin. Lysates were digested with trypsin (Promega) overnight to generate peptides. For in vitro kinase assays, peptides were generated after amino acid culturing, incubated first with calf intestinal alkaline phosphatase (New England BioLabs), and then with 0.5 μg of the following recombinant kinases for 1 hour at 37°C: TBK1 (Millipore), IKKϵ (Millipore), and control polo-kinase 2 (Thermo Fisher Scientific).

All lysates were prepared for mass spectrometry as previously described (17) and then purified using immobilized metal affinity chromatography (19). Enriched phosphopeptides were analyzed using the Agilent 1200 LC coupled online to an LTQ Orbitrap Velos mass spectrometer (Thermo Fisher Scientific; ref. 18). Xcalibur 2.1 software was used for data-dependent acquisition. Mass spectra were processed using the MaxQuant software package and the Mascot search engine (Matrix Science; ref. 17). Peptides were identified with a concatenated forward and reversed IPI protein sequence database (version 3.60). Median stable isotope labeling with amino acids in cell culture (SILAC) ratios for identified peptides were generated for all iterations of the same peptide with the highest scoring versions reported (Supplementary Tables S1 and S2). Ratios were used to generate a z-score for each assay with a cutoff value of 2.5. TBK1/IKKϵ in vitro core hits were analyzed by gene set enrichment analysis with the Molecular Signature Databases collection C5 Gene Ontology Biological Processes. Motif-x software was used to identify common peptide motifs (20, 21) and then queried in Ras GTPase family members using Basic Local Alignment Search Tool.

Enzymatic quantification was performed using the ADP Glo Assay (Promega) with recombinant TBK1, synthetic peptide substrates, and ATP, as previously described (22). Synthetic peptides were as follows: IKKϵ-tide (ADDDYDSLDWDAKKK), IKKϵ-tide L8A (ADDDYDSADWDAKKK), Rab7-tide (IWDTAGQERFQSLGVAFYRGADCC), and Rab7-tide L8A (IWDTAGQERFQSAGVAFYRGADCC). Data are representative of eight experiments with two biological replicates.

After being grown on chamber slides (CellTreat) and subjected to various treatment conditions, cells were fixed and permeabilized according to standard protocols. Individual proteins were visualized using specific antibodies listed in Supplementary Table S3, and slides were imaged with the Nikon Eclipse 80i microscope. Three blinded investigators each counted 100 consecutive cells for the presence of perinuclear STING foci. Graphs depict the mean and SEM of counts from 3 different investigators and 2 experiments for a total of 6 data points.

Cytokine and chemokine protein expressions were analyzed by multiplex assay using the Human Cytokine/Chemokine Magnetic Bead Panel (Luminex, #HCYTMAG-60K-PX30) on a MAGPIX system (Merck). Fold changes relative to the corresponding control were calculated and plotted as log₂FC. Lower and upper limits of quantitation were inputted from standard curves for cytokines above or below detection. The media from control cells and those treated with ADU-S100 (Chemietek) were subjected to CXCL10 ELISA assay (R&D Systems, #DIP100) at 24 hours, whereas in TBK1/IKKϵ studies, supernatants were analyzed at 48 hours. In addition, supernatants from untreated and ADU-S100–treated samples were subjected to CCL5 and IFNβ ELISA assays (R&D Systems, #DRN00B, #DIFNB0) at 72 hours. Assays are representative of three experiments in duplicate.

Total RNA was isolated using the RNeasy Mini Kit (Qiagen). Following extraction, 1 μg of RNA was used for cDNA generation with SuperScript III First-Strand Synthesis SuperMix (Thermo Fisher Scientific, #1683483). qRT-PCR was performed using Power SYBR Green PCR Master Mix (Thermo Fisher Scientific, #4367659) and the Applied Biosystems 7300 Fast real-time PCR system and software. The relative expression of each gene was normalized to the expression of the housekeeping gene 36B4. Primer sequences are listed in Supplementary Table S3.

For all inhibitor studies, cells were pretreated for 30 minutes with each inhibitor ± 1 μg mL⁻¹ poly(dA:dT) (Invivogen) prepared with X-tremeGENE 9. Rab7 was inhibited with 100 nmol/L of Ras Superfamily GTPases Binding Antagonist (CID 1067700, Millipore Sigma) and stimulated with poly(dA:dT) for 6 hours prior to analysis. TBK1/IKKϵ inhibition immunofluorescence studies were performed with Compound 1 (Gilead Sciences), MRT67307 (Sigma-Aldrich), and CYT387 (MedChemExpress). Pretreatment or stimulation times were determined through a time course looking at STING perinuclear foci formation with 6, 8, and 3 hours selected for each of the inhibitors respectively. Similarly, treatment times with the inhibitors were the same with the addition of poly(dA:dT).

Using the STING agonist ADU-S100 at 10 μmol/L, Rab7 mutants were stimulated, and resultant cytokine (CXCL10, CCL5, and IFNβ) expression was measured with ELISA. Similarly, breast cancer cell lines were treated with ADU-S100 at 10 μmol/L, and CXCL10 ELISA was performed at 24 hours. In addition, a growth inhibition assay was performed with twice-weekly 10 μmol/L ADU-S100 treatment over 6 weeks. Cells were then stained for viability with 1% crystal violet.

Protein isolation was performed with RIPA buffer containing phosphatase and protease inhibitors (Roche) and quantified by bicinchoninic acid assay (Pierce Biotechnology). Twenty micrograms of total protein lysate was loaded in 4% to 12% NuPAGE gels and electrophoresed (Invitrogen), transferred to nitrocellulose membranes (Millipore), and immunoblotted. Membranes were blocked for 1 hour with the Odyssey blocking buffer (LI-COR) and probed for various proteins overnight in tris-buffered saline solution with 0.05% Tween-20. Membranes were then washed for 10 minutes in TBST (3X) and incubated with the corresponding secondary antibody for 1 hour. See Supplementary Table S4 for all primary and secondary antibodies. Immunoblots were visualized using the LI-COR Odyssey. Supplementary Figs. S1 to S3 contain uncropped images of the blots.

Paraffin-embedded TNBC blocks were obtained under Dana-Farber Cancer Institute/Harvard Cancer Center Institutional Review Board protocol #93–085. After deparaffinizing tissue blocks, slides were processed per standard protocol. These slides along with the US Biomax Inc. breast cancer tumor microarray (BR1201a) slides were then stained for STING and PTEN (Cell Signaling Technology) as previously described (23) with antibodies listed in Supplementary Table S4. Visualization was achieved using EnVision+/HRP Rabbit (Dako), diaminobenzidine (Dako), and hematoxylin counterstain. A semiquantitative score of STING and PTEN expression for each microarray core was determined by a pathologist, and negative staining was determined if the percentage was less than or equal to 15%.

MDA-MB-468 cells were plated in ultralow attachment plates to allow for the formation of spheroids, which were then embedded into an AIM chip microfluidic device as previously described (23, 24). Media were then added to the exterior channels with or without 10 μmol/L ADU-S100 overnight. In the morning, Jurkat T cells were stained with CellTrace Violet (Invitrogen) and added to one exterior chamber of the device. Cocultures were incubated at 37°C for 48 hours and then fixed and counterstained with DAPI (Invitrogen). Wells were visualized with a Nikon Eclipse 80i microscope, and all resultant images were quantified for Jurkat T-cell migration by ImageJ.

GraphPad Prism 7.0 was used for statistical analysis, data processing, and graph generation. Values reported are the mean and SEM. When comparing only two groups, a Student t test was applied; otherwise, an ANOVA multivariate analysis was performed with a post hoc modification as described in the figure legends. Statistical significance was determined as P < 0.05. Peptide ratios between groups were analyzed with a Pearson correlation.

The raw SILAC data are available from the corresponding author upon reasonable request.

To uncover novel downstream targets of TBK1/IKKϵ kinase activity, we developed an integrated phosphoproteomics method using SILAC to identify particularly robust and specific substrates. We compared results obtained from a cellular phosphoproteomics screen following transfection of TBK1 constructs in HEK 293T cells with those of a biochemical assay using recombinant TBK1/IKKϵ kinases and dephosphorylated HEK 293T protein lysates (Fig. 1A). Thus, integration of the phosphosite hits from these complementary approaches identifies putative substrates that both occur in cells and are direct in nature. For the cell-based screen, HEK 293T cells were cultured with light, medium, or heavy amino acid–containing media, and then transfected with control EGFP, TBK1-WT, or kinase dead (K38M) TBK1 (TBK1-KD) constructs, respectively (Fig. 1A). Protein lysates at the 24-hour timepoint after transfection were prepared, mixed in equal proportions, and subjected to tandem mass spectrometry. Resulting peptides were evaluated for TBK1 target specificity by comparison of z-scores of differential phosphopeptide abundance for both TBK1-WT/EGFP and TBK1-WT/TBK1-KD conditions, to normalize against both control vector expression and absence of kinase activity. Employing this technique in two independent experiments yielded 39 core hits from experimental replicate 1 and 84 core hits from experimental replicate 2 (Fig. 1B and C; Supplementary Fig. S4A; Supplementary Table S1). As expected, TBK1-WT/EGFP and TBK1-WT/TBK1-KD hits were largely concordant with each other, with the serine-716 TBK1 autophosphorylation site representing a top hit in both experiments. Of note, peptides containing the serine-172 TBK1 activation site were not generated by the tryptic digestion and were thus not identified (Supplementary Table S1). After determination of the individual experiment core hits from both replicates, a secondary analysis was performed to identify overlapping peptides. From both experiments, there were 22 common core hits, of which the TBK1 autophosphorylation site was the top-ranked phosphopeptide (Fig. 1D).

In parallel, to identify which of these hits might represent bona fide direct TBK1/IKKϵ substrates, we conducted the in vitro phosphoproteomics screen (Fig. 1A). In this case, dephosphorylated SILAC-labeled HEK 293T lysates were incubated with purified recombinant TBK1 or IKKϵ and compared with polo-like kinase 2 (PLK2) as an unrelated control enzyme. As for the cell-based screen, in vitro–generated phosphopeptides were identified after purification and tandem mass spectrometry. For these experiments, medium and heavy–labeled lysates were incubated with either recombinant TBK1, IKKϵ, or PLK2; light-labeled lysates were left untreated; and then ratios between the heavy and light– or medium and light–labeled lysate conditions were determined. Intrareplicate concordance for both TBK1 and IKKϵ was high and reinforced phosphopeptide specificity (Fig. 2A and B). Cross enzyme concordance, for the subset of phosphosites that overlapped between TBK1 and IKKϵ, consisted of 543 statistically significant phosphopeptides, whereas there were less than 50 phosphopeptides identified for the subset that included PLK2 and TBK1 or PLK2 and IKKϵ (Fig. 2C and E; Supplementary Fig. S4B–S4I; Supplementary Table S2). Utilizing the Molecular Signatures Database (25), we interrogated cellular pathways associated with these in vitro–identified TBK1/IKKϵ phosphopeptides. Among the most enriched pathways were those involved in the catabolic processes, consistent with the known role of TBK1 in autophagy (Fig. 2F; refs. 26, 27). In addition, motif analysis of TBK1/IKKϵ-recovered peptides revealed x-pS-L/I/F/M-x, consistent with a previously identified IKKϵ substrate peptide motif (Table 1; ref. 28). Together, despite the artificial nature of this in vitro SILAC screen, these findings highlight specificity related to TBK1/IKKϵ substrate targets.

In order to identify the most robust candidate substrates from these screens, we integrated both the cellular and the in vitro SILAC phosphoproteomics data. Cross-validation of the core hits resulting from the cell-based screen and the in vitro kinase screen revealed six common phosphopeptides as highly likely bona fide TBK1 substrates, including Rab7-S72, Irs4-S989, Nono-S262, Msn-S576, Vim-S420, and Scg10-S46 (Fig. 3A). Of note, many of these novel targets (Msn, Vim, and Scg10) are involved in cytoskeletal reorganization, whereas the top hit, Rab7-S72, was also recently identified in a separate phosphoproteomics screen as a TBK1 target in the context of mitophagy (29). In addition, Rab7 was included within the enriched catabolic processes category of our pathway analysis (Fig. 2F), and we also previously linked Rab7 with negative regulation of TBK1 signaling (27), prompting us to evaluate this hit further.

As a positive and negative control, we utilized the previously identified IKKϵ-tide, which contains the F-x-pS-L motif (Table 1; ref. 28) or a mutated version in which the +1 leucine is mutated to an alanine (IKKϵ-Tide L8A). We also generated a peptide flanking the S72 site with the corresponding +1 leucine to alanine mutation (Rab7-Tide L8A). As expected, we observed significant in vitro phosphorylation of the positive control IKKϵ-Tide and the wild-type Rab7-Tide, but not of the IKKϵ-Tide L8A or Rab7-Tide L8A–negative controls (Fig. 3B). Because this Rab7 peptide region contained the more specific IKKϵ F-x-pS-L motif and is a member of a larger family of Rab GTPases, we next used protein BLAST to identify additional candidate Rab family members that could be phosphorylated by TBK1 or IKKϵ. This analysis identified Rab9, Rab31, Rab8, Rab5, and Rab7B as additional potential substrates. We utilized the same TBK1 in vitro kinase assay to assess activity against these peptides, which identified Rab7A, Rab7B, Rab9, and Rab31 as capable of being phosphorylated in vitro (Fig. 3C). However, although these other Rab family members can be phosphorylated in vitro, only Rab7A was identified as being a target of TBK1 activity in HEK 293T cells.

Because PTEN-mediated dephosphorylation of Rab7-S72 was shown to promote fusion of endosomes to lysosomes, we considered the possibility that TBK1/IKKϵ-mediated phosphorylation of this site might similarly impair Rab7 localization to these endolysosome-associated puncta (14). We utilized the PTEN-null TNBC cell line MDA-MB-468, which contains elevated IKKϵ levels (10), and examined baseline Rab7 localization, compared with stable overexpression of WT (V5-Rab7-WT), a phosphomimetic mutant (V5-Rab7-S72E), or phosphoresistant mutant (V5-Rab7-S72A). Consistent with prior work, overexpression of the Rab7-S72E mutant resulted in a diffuse staining pattern, in contrast to the Rab7-S72A mutant, which localized to puncta, which we confirmed both by V5 and Rab7 indirect immunofluorescence (Fig. 4A; Supplementary Fig. S5; ref. 14).

We previously demonstrated that MDA-MB-468 cells required IL1β priming to secrete NF-κB–associated cytokines including IL6 and CCL5 downstream of TBK1/IKKϵ (10). Interestingly, when we examined 24-hour conditioned media from each of these stably derived MDA-MB-468 Rab7 overexpressing cell lines using multiplexed cytokine/chemokine profiling, we identified baseline upregulation of CXCL10 (IP-10) by Rab7-S72E in contrast to Rab7-S72A, as compared with Rab7-WT (Fig. 4B). Indeed, this finding was confirmed both by quantitative RT-PCR and by ELISA and was highly significant for the TBK1/IKKϵ interferon regulatory factor 3 (IRF3)–regulated chemokines CXCL10 and CCL5, with a modest effect on baseline IFNβ expression (Fig. 4C and D). Taken together, these data revealed that mislocalization of Rab7 by overexpression of the S72 phosphomimetic mutant promotes dysregulated innate immune signaling and specifically increases IRF3 target chemokines downstream of TBK1/IKKϵ.

Because STING bridges TBK1 to IRF3 and was recently shown to undergo Rab7-mediated trafficking to the lysosome (15), we considered whether dysregulation of CXCL10, CCL5, and IFNβ expression might be related to altered regulation of Rab7-mediated STING degradation. We first compared baseline STING protein levels between TNBC cell lines with PTEN-WT (MDA-MB-231, HCC1187, HCC1143) versus those with impaired Rab7-S72 phosphatase activity due to PTEN-null status (HCC1937, MDA-MB-468, HCC70). Indeed, PTEN-null cell lines exhibited higher basal expression of STING, in contrast to PTEN-WT cell lines with intact Rab7-S72 phosphatase activity, as it is capable of maintaining Rab7 in an active dephosphorylated state (Fig. 4E; Supplementary Fig. S1). To assess the causality of this relationship, we next overexpressed PTEN-WT or the catalytically inactive mutant PTEN-C124S in Rab7-WT– or Rab7-S72E–expressing MDA-MB-468 cells. PTEN downregulated STING levels in a phosphatase-dependent manner in the setting of Rab7-WT expression, but not in the context of Rab7-S72E (Fig. 4F; Supplementary Fig. S2). Together, these findings reveal that hyperphosphorylated Rab7-S72 in the setting of PTEN loss and TBK1/IKKϵ activation enhances STING stability.

We next examined the consequences of stimulating STING activity, which promotes rapid feedback STING degradation (30). Notably, transfection with the dsDNA mimic poly(dA:dT) failed to promote feedback STING downregulation following Rab7-WT or Rab7-S72E overexpression, in contrast to the TBK1/IKKϵ phosphoresistant Rab7-S72A mutant (Fig. 4G; Supplementary Fig. S3). Furthermore, degradation of STING following Rab7-S72A overexpression required intact Rab7 function, because this effect was abrogated by cotreatment with the Rab7 inhibitor CID 1067700. Promotion of STING degradation by Rab7-S72A led us to postulate that expression of this mutant would also confer reduced sensitivity to STING agonism. Indeed, quantification of cytokine/chemokine expression after treatment with the direct STING agonist ADU-S100 revealed that Rab7-WT and especially Rab7-S72E–overexpressing cells exhibited significantly higher levels of CCL5 and CXCL10, as well as IFNβ as compared with Rab7-S72A (Fig. 4H). These data thus provide evidence that phosphorylation of Rab7-S72 by TBK1/IKKϵ in PTEN-null TNBC cells limits Rab7-mediated degradation of STING (15) and enhances innate immune cytokine production, especially following STING agonism.

We next directly interrogated the effect of TBK1/IKKϵ inhibition on Rab7-mediated trafficking by using three well-characterized TBK1/IKKϵ inhibitors—Compound 1, MRT67307, and CYT387 (10, 24, 31). Given that phosphorylation of Rab7 promoted STING dispersal into the cytosol, we hypothesized that TBK1/IKKϵ inhibition would promote focal perinuclear STING localization in MDA-MB-468 cells. Indeed, treatment with Compound 1 led to formation of STING foci in parental MDA-MB-468 cells, which colocalized with active Rab7 puncta (Fig. 5A and B). Perinuclear STING localization was also observed in Rab7-WT– and Rab7-S72A–expressing cell lines, but not in Rab7-S72E cells where it remained more dispersed, confirming the dominant-negative effect of this mutant (Fig. 5A and B). Similarly, TBK1/IKKϵ inhibition with MRT67307 or CYT387 also resulted in the formation of STING perinuclear foci in parental, Rab7-WT–, and Rab7-S72A–expressing MDA-MB-468 cells, but not in Rab7-S72E cells (Fig. 5A, C, and D). Because PTEN also modulates EGFR late endocytic trafficking by dephosphorylation of Rab7 (14), we interrogated the effects of TBK1/IKKϵ inhibitor treatment on EGFR localization with the Rab7 mutants. We found that the dominant-negative effect of Rab7-S72E on STING localization was also seen for EGFR localization with TBK1/IKKϵ inhibition (Supplementary Fig. S6A).

We next explored the impact of TBK1/IKKϵ inhibition on cytokine/chemokine production by measuring CXCL10 production in conditioned medium at baseline or following poly(dA:dT) stimulation. Notably, even following concurrent Compound 1 and poly(dA:dT) treatment, STING was dispersed in the context of Rab7-S72E overexpression (Supplementary Fig. 6B and C). Consistent with this finding, Compound 1 treatment inhibited poly(dA:dT) stimulated CXCL10 in parental, Rab7-WT, and Rab7-S72A cell lines, but was less effective in Rab7-S72E cells, indicative of sustained STING/TBK1 downstream signaling in this setting (Fig. 5E). Taken together, these data support the model that TBK1/IKKϵ-regulated phosphorylation of Rab7-S72 impairs STING trafficking in TNBC cells, sustaining its levels and promoting hyperactive innate immune signaling in the context of PTEN loss.

Our findings suggest that the PTEN-null state limits STING mobilization to perinuclear foci where it is degraded as a negative feedback mechanism following STING agonism. As such, this may serve as a biomarker for enhanced STING agonist activity in TNBC, given that this subtype of breast cancer is frequently associated with PTEN loss (5, 9, 32), which results in a worse prognosis and more rapid tumor proliferation (33, 34). We first investigated this relationship by IHC staining for PTEN and STING using five primary TNBC tumors derived from patients who were treatment naïve. Consistent with our in vitro results, samples with negative PTEN staining exhibited strong STING positivity (Fig. 6A). To further validate whether this PTEN-null/STING-high state is prevalent in the clinical setting among patients with TNBCs, we performed IHC for STING and PTEN expression using a commercially available breast cancer tumor microarray. STING levels were also significantly increased in PTEN-null cases (Fig. 6B; Supplementary Table S5). These data suggest that PTEN-null TNBC cancers are particularly poised for STING agonism as the result of aberrant STING trafficking.

We therefore tested whether this relationship promotes STING agonist responsiveness. Indeed, when we stimulated PTEN-null TNBC cell lines (MDA-MB-468, HCC70, and HCC1937) with the direct STING agonist ADU-S100 at 10 μmol/L for 24 hours and measured the resultant supernatant for CXCL10 expression by ELISA, TNBC cell lines exhibited a 10-fold increase in CXCL10 expression as compared with control. In contrast, luminal A cell lines (MCF7 and T-47D), luminal B cell lines (ZR-751 and SKBr3), and the nontransformed breast cancer cell line (MCF-10A) showed no significant induction of CXCL10 (Fig. 6C). Given the responsiveness of PTEN-null TNBC cell lines to ADU-S100 stimulation, we further characterized the cellular response by treating cells plated at a low density and then proceeded with twice-weekly ADU-S100 treatment for 6 weeks. Whereas chronic treatment inhibited TNBC cell line proliferation, there were no differences seen with the luminal A breast cancer cell lines (Fig. 6D). These findings further reveal that STING agonism of PTEN-null TNBCs also induces cell-intrinsic cytotoxicity.

We next interrogated the potential in vivo implications of the enhanced CXCL10 secretion using a previously described three-dimensional (3D) microfluidic T-cell migration assay (16). MDA-MB-468 spheroids were embedded in collagen within the central region/migration zone of the 3D microfluidic device, then treated with or without ADU-S100 at 10 μmol/L, and cocultured with CXCR3-expressing Jurkat T cells through the side media chamber. As expected, Jurkat T-cell migration was only seen in the presence of MDA-MB-468 cells that were pretreated with ADU-S100. As an additional control, no cells were plated in the migration zone, and with the introduction of Jurkat T cells in the media chamber, no significant migration of T cells was seen. Quantification of T-cell migration identified a 10-fold increase in T-cell infiltration in the ADU-S100–pretreated MDA-MB-468 spheroids as compared with control-pretreated or Jurkat-alone cultures (Fig. 6E). Together, these data confirm that CXCL10 induction by STING agonism promotes T-cell recruitment, supporting the idea that targeting STING agonists to PTEN-null TNBCs also promotes immunogenicity.

Through integrated SILAC phosphoproteomic studies using cell-based and in vitro kinase screens, our study has identified 6 common phosphopeptides as high confidence TBK1/IKKϵ substrates, including Rab7-S72 as the top-ranked peptide. Interestingly, the exact same site on Rab7 was also recently identified through phosphoproteomic studies evaluating TBK1 targets involved in mitophagy (29). We confirmed in vitro that specific phosphorylation of this site requires the known IKKϵ substrate motif and that the phosphomimetic Rab7-S72E mutant rescues the cellular impact of TBK1/IKKϵ inhibition, validating this site as a bona fide substrate of these kinases. Furthermore, under basal conditions, Rab7-S72E expression resulted in increased CXCL10 production and was resistant to dsDNA-stimulated STING degradation, in contrast to the active Rab7 dephosphorylated mutant, Rab7-S72A. Together, these data suggest a working model, in which TBK1/IKKϵ through phosphorylation of Rab7-S72 prevents the endolysosomal degradation of STING, thereby creating a sustained dsDNA sensing mechanism in the cytosol (Supplementary Fig. S7).

As a response to pathogens and aberrant cellular DNA detected in the cytosol, the cGAS-STING pathway produces an IFN-mediated response that is tightly regulated. More recently, it has been shown that Rab7 regulates this innate immune response through trafficking-mediated degradation of the STING protein to endolysosomes (15), suggesting that targeting this pathway for blockade may help improve the therapeutic potential of STING agonists. Consistent with our results, inhibition of lysosome acidification was found to increase STING protein levels and enhance STING signaling, resulting in CXCL10, CCL5, and IFNβ mRNA expression (15). Here, we show that increased STING signaling is directly correlated with the phosphorylation status of Rab7-S72. Specifically, phosphorylation of S72 deactivates Rab7 preventing STING trafficking to the endolysosome, resulting in downstream activation of TBK1-IRF3–regulated chemokines such as CXCL10, CCL5, and IFNβ. Importantly, prior studies of STING signaling have determined that it remains active throughout its trafficking from the endoplasmic reticulum to the endolysosome, at which point it is inactivated as the vesicle acidifies (15, 35, 36). Indeed, in our studies, constitutively active Rab7 (S72A) enhanced STING protein turnover following dsDNA stimulation and continually colocalized with STING in perinuclear puncta, impairing CXCL10, CCL5, and IFNβ production.

Notably, PTEN-mediated dephosphorylation of Rab7-S72 was also shown to be necessary for its recruitment to the endolysosome and to promote EGFR recycling (14). As such, it was postulated that in tumors that are PTEN deficient, degradation of EGFR by endosomes is blunted, resulting in enhanced EGFR signaling and enhanced tumor progression. As predicted, higher levels of EGFR were also maintained following EGF stimulation in MDA-MB-468 cells and reversed with reconstitution of PTEN-WT (14). Consistent with impaired degradation as a result of PTEN loss, our studies also showed elevations of basal STING expression in PTEN-null TNBC cell lines, which was recapitulated in primary TNBC tumor histology. Recent work has also showed that EGFR signaling can be immunosuppressive, decreasing both MHC-1 and IFNγ expression (37). Furthermore, in TNBCs, IL1 within the tumor microenvironment also promotes IKKϵ-associated NF-κB signaling, inducing CCL5 and IL6, which are immune suppressive and can recruit myeloid suppressor cells (10). IL1-induced CCL5 and IL-6 downstream of TBK1/IKKϵ are further enhanced by impaired autophagy and endolysosome maturation (27). Thus, in addition to elevated STING levels, these additional factors could also promote an immunosuppressed tumor microenvironment in the context of defective Rab7 trafficking.

Currently, STING agonism is a developing area of therapeutic focus for various types of cancers. Our studies show that TNBCs are primed for STING agonist treatment, given this breast cancer subtype is associated with high rates of PTEN loss and elevated basal STING levels. Similar to the findings of our T-cell migration assay and growth inhibition assay, injection of STING agonist into the tumor microenvironment of the TNBC murine model 4T1 resulted in regression of both primary and distant tumors while inducing T-cell activation and memory (38–40). Furthermore, emerging findings from a phase I study of ADU-S100 in combination with anti–PD-1 therapy for patients with advanced metastatic solid tumors or lymphomas show that treatment response was largely limited to patients with TNBCs (41). As these trials are moving forward with dose escalation, it is likely that only a subset of patients will respond and identifying biomarkers will be critical. Thus, our finding that PTEN loss and TBK1/IKKϵ signaling converge to promote dysregulated STING trafficking identifies a specific genomic context in TNBC that enhances STING agonist activity. Additional optimization is underway to improve STING retention in the tumor bed (42) and enhance membrane permeability (43–45) to further increase the clinical effectiveness of STING agonist therapy. Regardless, our findings suggest that PTEN-null, IKKϵ-high TNBCs may represent an optimal target for STING agonism, converting an immunosuppressive tumor microenvironment into an immunogenic one.

T.U. Barbie is a consultant at Qiagen/N of One and has an ownership interest in patent of STING levels as a biomarker for cancer immunotherapy, patent of Microfluidic Cell Culture of Patient-Derived Tumor Cell Spheroids and Xsphera. S.A. Carr has received honoraria from the speakers' bureau of PittConn, Pfizer, Drug Discovery World Congress, and Esai. D.A. Barbie is a consultant at N of One/Qiagen, is a scientific advisory board member at Tango Biosciences, reports receiving commercial research grant from Novartis and BMS, and has an ownership interest (including patents) in Xsphera Biosciences. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T.U. Barbie, J.D. Jaffe, D.A. Barbie

Development of methodology: T.U. Barbie, S.A. Carr

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.U. Barbie, J.L. Ritter, T.C. Thai, N.R. Mahadevan, P. Mertins, E.H. Knelson, S. Han

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T.U. Barbie, J.L. Ritter, N.R. Mahadevan, P. Mertins, E.H. Knelson, B.P. Piel, S.A. Carr

Writing, review, and/or revision of the manuscript: T.U. Barbie, J.L. Ritter, S.A. Carr, D.A. Barbie

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T.U. Barbie, Z. Zhu

Study supervision: T.U. Barbie, S.A. Carr, D.A. Barbie

We thank Dr. Gabriela Alexe (Broad Institute and Dana Farber Cancer Institute, Boston, MA) for assistance with computational analysis of the SILAC data. This work was funded in part through the generosity of an anonymous donor (to T.U. Barbie); Department of Defense Award #W81XWH-13-1-0029 (to T.U. Barbie); the BWH Department of Surgery Robert T. Osteen Fellowship (to T.U. Barbie); NCI R01 CA190394 (to D.A. Barbie); and National Cancer Institute Clinical Proteomic Tumor Analysis Consortium grants NIH/NCI U24-CA210986 and NIH/NCI U01 CA214125 (to S.A. Carr.).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.