A PERK-Specific Inhibitor Blocks Metastatic Progression by Limiting Integrated Stress Response–Dependent Survival of Quiescent Cancer Cells Free

Under stress conditions, the accumulation of unfolded proteins in the endoplasmic reticulum (ER) lumen activates three main pathways: PERK, IRE1α, and ATF6. These pathways are part of a survival and adaptive mechanism known as the unfolded protein response (UPR; ref. 1). However, other kinases that, for example, sense aminoacidic deficiency or double-stranded RNAs, GCN2, and PKR, respectively, can phosphorylate eIF2α (1). Thus, because the UPR is only a subset of signals that can control eIF2α and other adaptation mechanisms during stress, the definition has been expanded to account for all mechanisms that might integrate via eIF2α various stress signals and it is termed integrated stress response (ISR; ref. 2). Recent evidence suggests that in various cancer types the ISR allows tumor cells to respond to increased demands on the ER and greater oxidative conditions imposed by an enhanced translational load caused by oncogenes, hypoxia, and other stress conditions, including anticancer agents (3–5). Oncogene-activated pathways increase ER client protein load by activating mTOR signaling and translation initiation (5–7). Other studies have shown that PERK and the IRE1α–XBP-1 pathways contribute to the ability of cancer cells to adapt to hypoxia and microenvironmental stress (4, 8–13), suggesting that the ISR enables critical adaptation mechanisms necessary to survive within a changing cellular milieu.

PERK activation initiates an antioxidant and autophagic response that coordinates a protection mechanism for mammary epithelial cells during loss of adhesion to the basement membrane (14). This survival response involves an ATF4 and CHOP transcriptional program (14) coupled to a rapid activation of the LKB1–AMPK–TSC2 pathway that inhibits mTOR (15). Human ductal carcinoma in situ (DCIS) lesions that displayed enhanced PERK phosphorylation and autophagy (14, 16) showed, upon conditional ablation of PERK in the mammary epithelium, delayed mammary carcinogenesis promoted by the HER2 oncogene (17, 18). Furthermore, HER2 increases the levels of proteotoxicity in tumor cells thereby activating JNK and ISR signaling and allowing HER2⁺ cancer cells to cope with this stress (19). Interestingly, 14% of HER2-amplified human breast tumors display an upregulation of PERK mRNA [breast invasive carcinoma TCGA (The Cancer Genome Atlas), Nature 2012; ref. 20], and enrichment for a PERK activity signature correlates with shorter metastasis-free periods in all breast cancer subtypes combined (21). These data support the notion that certain cancer cells in HER2⁺ and other breast tumor subtypes are dependent on PERK for survival.

We also reported that dormant (quiescent) cancer cells were dependent on PERK and ATF6 signaling for survival (13, 22, 23). Furthermore, solitary quiescent pancreatic disseminated cancer cells (DCC) to the liver of mice also display a PERK-dependent ISR that was linked to loss of E-cadherin expression and downregulation of MHC-I, which favors immune evasion during dormancy (24). In the MMTV–HER2 model, quiescent DCCs in bone marrow and lungs were also found in a mesenchymal-like state and to express an ISR signature (25, 26) but the link to the ISR was not tested. Together, these data suggest that the ISR may serve as a stress and immune microenvironmental adaptive survival mechanism for DCCs. However, until now no clinical-grade inhibitor of the PERK pathway has been available to test how targeting of this kinase may affect DCC fate and metastatic progression.

Here, we report that a selective and potent inhibitor of PERK, HC-5404 (henceforth HC4; ref. 27), currently in a human phase 1a clinical trial in solid tumors (NCT04834778), can block HER2-driven breast cancer metastasis or HER2-independent HNSCC initiated metastasis through the eradication of dormant DCCs or by targeting slow cycling ISR^high cells in established metastasis. Imaging and single-cell gene expression profiling revealed the existence of an ISR^high/CDK inhibitor^high quiescent population of cancer cells existing as solitary DCCs or as part of established metastasis. In addition, CDK4/6 inhibition, which heightened the ISR pathway, followed by HC4 treatment further decreased metastatic burden. Our work reveals a new application of a PERK inhibitor with anti-metastatic activity that has been vetted for specificity and selectivity for clinical trials use. This inhibitor, alone or in combination with anti-proliferative therapies, may represent a new strategy to target dormant ISR^high cells existing as minimal residual disease or as part of established metastasis and help prevent and treat lethal metastases against which we lack effective therapies to date.

EGF was obtained from PeproTech and used at 100 ng/mL. Thapsigargin was from Sigma and used at 0.1, 0.2 μmol/L or 2 nmol/L as indicated in the legends. HC4 and Abemaciclib were provided by HiberCell and Eli Lilly, respectively. AP20187 was from ARIAD Pharmaceuticals.

SKBR3 (RRID:CVCL_0033), ZR-75–1 (RRID:CVCL_0588), MCF10A (RRID:CVCL_0598), and EMT6 (RRID:CVCL_1923) cells were obtained from the ATCC and cultured for a maximum of 10 passages. T-HEp3 and D-Hep3 cells are an Aguirre-Ghiso Laboratory proprietary PDX maintained in vivo and not available from the ATCC. The PDX lines are used for a maximum of 5 passages and quarterly tested for phenotype consistency following historical records. All cell lines were monthly tested for Mycoplasma using the PCR Myco-plasma test kit. For 3D cultures, MCF10A-HER2, SKBR3, and ZR-75–1 cells were plated in growth factor-reduced Matrigel (Corning) and grown as described previously (15). Treatments with vehicle (DMSO) or HC4 (2 μmol/L) were replaced every 24 hours for 2D and every 48 hours for 3D cultures.

Institutional Animal Care and Use Committees at Mount Sinai School of Medicine and Albert Einstein College of Medicine approved all animal studies. Protocol number: 08–0417. The FVB/N-Tg (MMTV–HER2) mouse strain (IMSR_JAX:005038) was obtained from The Jackson Laboratory. These mice express the un-activated neu (HER2) form under the transcriptional control of the mouse mammary tumor virus promoter/enhancer. Before being used in any experiment, females underwent one round of pregnancy and at least two weeks of no lactation after weaning. Females between 24 and 32 weeks of age were injected intraperitoneally with vehicle (90% corn oil, 10% ethanol) or HC4 (50 mpk) daily, for two weeks. For the combination treatment, females of 24–32 weeks of age were treated daily by oral gavage with Abemaciclib (50 mpk) for 4 weeks before starting the treatment described earlier with HC4. FvB wild-type mice were injected with MMTV–HER2 cells (5.0 × 10⁴/animal) from primary tumor lesions via the tail vein, allowed them to expand into established metastatic lesions for 3 weeks and treated with HC4 as described earlier for 2 weeks. A total of 2.5 × 10⁵ human HNSCC PDX T-HEp3 cells as single-cell suspensions were injected intravenously in Balb/c nu/nu mice and 24 hours after injection we treated the mice with vehicle or HC4 (50 mg/kg) for 10 days. 10⁶ EMT6 cells were injected in fad pad of Balb/c mice and 9 days after implantation mice were treated twice per day with HC4 at 10 mg/kg for 17 days. Tumor volumes were measured using the formula (Dxd²)/2, where D is the longest and d is the shortest diameter. For circulating cancer cell (CCC) count, animals were anesthetized and whole blood was extracted by cardiac puncture. Mammary glands, tumors, and lungs were collected and fixed in 10% buffered formalin overnight before paraffin embedding. The bone marrow from the two lower limbs was flushed with a 26G needle and further processed by Ficoll density gradient centrifugation. For CCC as well as for DCCs detection in bone marrow, tissues were depleted of mature hematopoietic cells by anti-mouse antibody-labeled magnetic bead separation (Miltenyi Biotec) before fixation in formalin for 20 minutes at 4°C.

Tumor growth on chicken embryo chorioallantoic membrane (CAM) was performed as described previously (28). A total of 5 × 10⁵ Dormant D-HEp3 cells were inoculated on the chorioallantoic membrane of 10-day-old chick embryos. 24 hours later CAM tumors were treated daily with vehicle or HC4 (20 μmol/L) for 6 days. Nodules were collected after treatment, minced, and digested with collagenase and fixed with 4% paraformaldehyde (PFA) prior counting and staining.

Mammary glands fixed in 10% buffered formalin were incubated in Carmine Alum stain (Carmine 0.2%, Aluminum potassium sulfate 0.5%; Sigma) for 2 days. Then, they were dehydrated and transferred to methyl salicylate solution before imaging using a stereomicroscope.

IHC and immunofluorescence (IF) from paraffin-embedded sections were performed as previously described (15). Briefly, slides were dewaxed and serially rehydrated. Heat-induced antigen retrieval was performed in either citrate buffer (10 mmol/L, pH6), EDTA buffer (1 mmol/L, pH 8) or Tris/EDTA (pH 9). Slides were further permeabilized in 0.1% Triton-X100, blocked, and incubated with primary antibody overnight at 4°C at 1:50–1:200 dilution. For IHC, an additional step of endogenous peroxidase and avidin/biotin quenching was performed before primary antibody incubation. Primary antibodies used were anti-cytokeratin 8/18 (Progen, RRID:AB_1541064), smooth muscle actin-Cy3 (Sigma, RRID:AB_476856), P-PERK (T980, RRID:AB_2095853; ref. 29), P-EIF2A (Ser51, RRID:AB_2096481), Cleaved Caspase 3 (RRID:AB_2341188), P-H3 (S10, RRID:AB_331535), and P-HER2 (Y1221/1222, RRID:AB_490899; Cell Signaling Technology), HER2 (Abcam, RRID:AB_2893179), and HER2 (Millipore, RRID:AB_310172), Ki67 (eBioscience, RRID:AB_10854564), cytokeratin cocktail (C11 and ck7, Abcam, RRID:AB_306047 and RRID:AB_2783822; AE1 and AE3, Millipore, RRID:AB_94853), Vimentin (R&D, RRID:AB_2241653), P-Rb (S249/T252), and GADD34 (Santa Cruz Biotechnology). Next, slides were incubated in secondary antibodies (Life Technologies) and mounted. The cells positive for each marker were counted and percentage of total cancer cells was calculated. For IHC, sections were processed using VectaStain ABC Elite kit (Vector Laboratories) and DAB Substrate kit for peroxidase labeling (Vector Laboratories) and mounted in VectaMount medium (Vector laboratories). For IF, sections were mounted in ProLong Gold Antifade aqueous medium (Thermo Fisher Scientific). In the case of immunocytofluorescence, cytospins of fixed cells (100,00–200,000 cells/cytospin) were prepared by cyto-centrifugation at 500 rpm for 3 minutes on poly-prep slides, and the staining protocol was performed as explained below from the permeabilization onward. For the staining of 3D cultures, acini were fixed in 4% PFA for 20 minutes at 4°C, permeabilized with 0.5% Triton-X100 in PBS for 20 minutes at room temperature, washed in PBS-glycine and then blocked with 10% normal goat serum for 1 hour at 37°C, before performing IF staining.

The scoring for P-HER2 levels is explained in Supplementary Fig. S5A. For the scoring of CK8/18 and SMA in mammary gland ducts, 20 low magnification fields/animal were evaluated for the expression of CK8/18 as negative (0), low (1) or high (2) and the same for SMA and the sum of the two scores was considered as the final score (from 0 to 4).

Images were captured by using a Nikon Eclipse TS100 microscope, a Leica DM5500 or Leica SP5 confocal microscope. For whole-mount mammary gland quantification, slides were scanned using a Leica BioSystems Versa Z1.0 slide scanner. Scanned images were analyzed in ImageJ/FIJI (NIH) with custom-written scripts. Briefly, the images were background corrected and processed with median filter, then skeletonized. The “skeleton 2D/3D” plugin in ImageJ was used to count the number of branches, normalized to the total lengths of the ducts.

Apoptosis levels were evaluated using the In situ Cell Death Detection kit, AP (Roche). Paraffin sections from tumors were dewaxed, rehydrated, and permeabilized in PBS 0.2% Triton-X100 for 8 minutes. Then, slides were washed and blocked in 20% normal goat serum for 1 hour at 37°C. The TUNEL reaction mixture was then added and let go for 1 hour at 37°C. The reaction was stopped by incubating with Buffer I (0.3 mol/L Sodium chloride, 30 mmol/L Sodium citrate). Next, the slides were incubated with anti-fluorescein-AP antibody for 30 minutes at 37°C. After three washes in TBS, slides were incubated in alkaline phosphatase substrate in 0.1% Tween-20 for 20 minutes at room temperature. Finally, the slides were mounted using aqueous mounting medium. The percentage of TUNEL-positive cells was calculated using ImageJ software (NIH).

Cells were lysed in RIPA buffer and protein analyzed by immunoblotting as described previously (23). Membranes were blotted using the additional following antibodies: P-PERK (T980; ref. 29), PERK (Santa Cruz Biotechnology, RRID:AB_2762850), P-EGFR (Y1148, RRID:AB_331127), EGFR (RRID:AB_331707), P-AKT (S473, RRID:AB_329825), P-S6 (S235/236, RRID:AB_331679), P-ERK (Y204, RRID:AB_331646), P-HER2 (Y1221/1222 RRID:AB_490899, Y1112, Y877, RRID:AB_2099407), HSP90 (RRID:AB_2121214), eIF2α (RRID:AB_2230924), P-eIF2α (S51, RRID:AB_330951; Cell Signaling Technology), GAPDH (Millipore, RRID:AB_10615768), and β-tubulin (Abcam, RRID:AB_2210370). For induction of ER stress, MCF10A-HER2 cells were plated in low adhesion plates for 24 hours before collection.

For cell surface biotinylation, we used Pierce cell surface protein isolation kit following the manufacturer's instructions with minor changes. Briefly, MCF10A-HER2 cells were serum- and EGF-starved and treated ± HC4 for 24 hours before being stimulated with ± EGF (100 ng/mL) for 20 minutes. Then, cells were washed with ice-cold PBS and surface proteins biotinylated for 30 minutes at 4°C. After quenching, cells were harvested and lysed using RIPA buffer. Protein lysates were incubated with NeutrAvidin agarose beads and the bound proteins were released by incubation with SDS-PAGE sample buffer containing DTT (50 mmol/L). For endocytosis assays (30), cells were treated similarly but before treatment with EGF cell surface proteins were biotinylated. After 20 minutes incubation ± EGF (100 ng/mL) at 37°C (to induce endocytosis), cells were washed with ice-cold PBS and incubated with stripping buffer (to remove cell surface biotinylation: 75 mmol/L NaCl, 1 mmol/L MgCl2, 0.1 mmol/L CaCl2, 50 mmol/L glutathione, and 80 mmol/L NaOH, pH 8.6) for 30 minutes. To control for stripping efficiency, cells were stripped without 37°C incubation (t = 0). Cell lysates were prepared and processed for biotinylated protein isolation as described before.

Primary tumors from MMTV–HER2 28 to 30-week-old females were digested with collagenase into a single-cell suspension. Lungs from MMTV/HER2 24 to 30-week-old females were digested into a single-cell suspension with collagenase and resuspended in FACS buffer. Cells were then stained with anti–HER2-PE, anti–CD45-APC, and DAPI and the HER2⁺/CD45⁻ population of cells sorted using a BDFACSAria sorter. Sorted cells were resuspended at a 312,500cells/mL concentration in media and 80 μL were mixed with 20 μL suspension reagent (C1 Fluidigm). A C1 Single-cell Preamp IFC 10–17 μm was used for the single-cell separation. Pre-amplification was run using Ambion Single Cell-to-CT qRT-PCR kit and 20x TaqMan Gene expression FAM-MGB assays. Resulting cDNA was further diluted in C1 DNA dilution reagent 1/3 and used for gene expression analysis using 96.96 IFCs (Fluidigm), Juno System controller and Biomark HD for high-throughput qPCR. TaqMan Fast Advanced Master Mix were used for the qPCR reactions. Analysis was performed using Fluidigm Real-Time PCR Analysis Software and Clustergrammer web-based tool (31) for hierarchical clustering heat maps.

RNA was extracted using the RNeasy kit (Qiagen) following the manufacturer's protocol. Quantitative PCR was performed using SYBR Green Master Mix, GAPDH was used as housekeeping control. GAPDH primers: CTTTGTCAAGCTCATTTCCTGG (FW), TCTTGCTCAGTGTCCTTGC (R); GADD34 primers: TCGGAAGGTACACTTCGCTGAG (FW) TCGATCTCGTGCAAACTGCT (R).

Test compound in 1% DMSO was incubated in PBS pH7.4 for 1 hour, followed by determination via UV/Vis absorbance of dissolved concentration against a calibration curve.

Caco-2 cells (clone cells (clone C2BBe1) were obtained from the ATCC. Cell monolayers were grown to confluence on collagen-coated, microporous membranes in 12-well assay plates. The permeability assay buffer was Hanks’ balanced salt solution (HBSS) containing 10 mmol/L HEPES and 15 mmol/L glucose at a pH of 7.4. The buffer in the receiver chamber also contained 1% BSA. The dosing solution concentration was 5 μmol/L of test article in the assay buffer ± 1 μmol/L valspodar. Cells were first pre-incubated for 30 minutes with HBSS ± 1 μmol/L valspodar. Cell monolayers were dosed on the apical side (A-to-B) or basolateral side (B-to-A) and incubated at 37°C with 5% CO₂ in a humidified incubator. Samples were taken from the donor and receiver chambers at 120 minutes. Each determination was performed in duplicate. The flux of lucifer yellow was also measured post-experimentally for each monolayer to ensure no damage was inflicted to the cell monolayers during the flux period. All samples were assayed by LC/MS-MS using electrospray ionization. The apparent permeability (Papp) was calculated as follows: Papp = (dC_r/dt) × V_r / (A × C_A), where dC_r/dt is the slope of the cumulative concentration in the receiver compartment versus time in μmol/L s⁻¹; V_r is the volume of the receiver compartment in cm³; A is the area of the insert (1.13 cm² for 12-well); C_A is the average of the nominal dosing concentration and the measured 120-minute donor concentration in μmol/L.

Test compound was prepared at 1 μmol/L concentration in incubation buffer containing 0.5% acetonitrile and incubated with plasma for 4 hours before LC/MS-MS compound determination. Protein binding of the test compound was measured using Rapid Equilibrium Dialysis.

Human hepatocytes pooled from 10 donors were incubated (10⁶ viable cells/mL) with test compound at 1 μmol/L concentration (0.1% DMSO) and metabolic clearance determined by measuring percent remaining of test compound at multiple time points during incubation (15, 30, 60, 90, and 120 minutes). Compound concentration was determined by LC/MS-MS.

The kinase selectivity assay was outsourced to DiscoverRX and the methodological details were provided by the company as described below. Kinase-tagged T7 phage strains were grown in parallel in 24-well blocks in an E. coli host derived from the BL21 strain. E. coli were grown to log-phase and infected with T7 phage from a frozen stock (multiplicity of infection = 0.4) and incubated with shaking at 32°C until lysis (90–150 min). The lysates were centrifuged (6,000 x g) and filtered (0.2 μmol/L) to remove cell debris. The remaining kinases were produced in HEK-293 cells and subsequently tagged with DNA for qPCR detection. Streptavidin-coated magnetic beads were treated with biotinylated small-molecule ligands for 30 minutes at room temperature to generate affinity resins for kinase assays. The liganded beads were blocked with excess biotin and washed with blocking buffer [SeaBlock (Pierce), 1% BSA, 0.05% Tween 20, 1 mmol/L DTT] to remove unbound ligand and to reduce non-specific phage binding. Binding reactions were assembled by combining kinases, liganded affinity beads, and test compounds in 1x binding buffer (20% SeaBlock, 0.17x PBS, 0.05% Tween 20, 6 mmol/L DTT). Test compounds were prepared as 40x stocks in 100% DMSO and directly diluted into the assay. All reactions were performed in polypropylene 384-well plates in a final volume of 0.02 mL. The assay plates were incubated at room temperature with shaking for 1 hour and the affinity beads were washed with wash buffer (1x PBS, 0.05% Tween 20). The beads were then re-suspended in elution buffer (1x PBS, 0.05% Tween 20, 0.5 μmol/L non-biotinylated affinity ligand) and incubated at room temperature with shaking for 30 minutes. The kinase concentration in the eluates was measured by qPCR. Compounds that bind the kinase active site and directly (sterically) or indirectly (allosterically) prevent kinase binding to the immobilized ligand, will reduce the amount of kinase captured on the solid support. Screening hits are identified by measuring the amount of kinase captured in test versus control samples and reported as “% Ctrl,” where lower numbers indicate stronger hits.

Selectivity Score or S-score is a quantitative measure of compound selectivity. It is calculated by dividing the number of kinases that compounds bind to by the total number of distinct kinases tested, excluding mutant variants.

S = Number of hits/Number of assays.

This value can be calculated using %Ctrl as a potency threshold (below) and provides a quantitative method of describing compound selectivity to facilitate comparison of different compounds.

S (35) = (number of non-mutant kinases with %Ctrl <35)/(number of non-mutant kinases tested).

S (10) = (number of non-mutant kinases with %Ctrl <10)/(number of non-mutant kinases tested).

S (1) = (number of non-mutant kinases with %Ctrl <1)/(number of non-mutant kinases tested).

Female CD1 mice (6–8 weeks of age) were dosed with vehicle (10% ethanol, 90% corn oil) or HC4 at 50 mg/kg once and blood samples collected into tubes containing K2-EDTA at different time points (15 and 30 minutes, and 1, 2, 4, 8, 12, and 24 hours). Plasma samples were prepared by centrifugation. The plasma concentration of test compound was determined by protein precipitation and LC/MS-MS.

TCGA data on mRNA expression levels of EIF2AK3 were accessed and analyzed through cBioPortal (RRID:SCR_014555; https://bit.ly/3yBhw2b).

Kaplan–Meier (KM) plotter software (RRID:SCR_018753) was used to analyze the prognostic value of the CSPS expression in 958 all-subtype and 119 patients with HER2⁺ breast cancer with the available data (32). The best cutoff value was used to identify the high- and low-expression groups. The hazard ratio with 95% CI and P log-rank value were calculated.

All points represent independent biological samples with error bars representing standard deviations and statistical significance was determined using one-sided Mann–Whitney test or t test using the Graph Pad Prism Software (RRID:SCR_002798).

Previously published RNAseq of MMTV-Neu EL and PT spheres is available under the accession code GSE165431. All data needed to evaluate the conclusions are included in the article and Supplementary Materials. Other data supporting the findings are available from the corresponding authors on reasonable request.

In the MMTV–HER2 (henceforth HER2⁺) mouse model, a highly used model of HER2⁺ human breast cancer, a high percentage of mice develop metastases to the lungs, which can be initiated by early or late DCCs (25, 26, 33, 34). Dormant HER2⁺ DCCs display loss of an epithelial-like phenotype with E-cadherin as a main marker and gain expression of a mesenchymal-like program characterized by increased expression and function of the transcription factors ZFP281 and Twist1 (26). E-cadherin–negative DCCs in pancreatic cancer models were also shown to be quiescent and displayed upregulation of CHOP, a PERK-induced gene (24). We set out to determine whether in the mouse HER2⁺ spontaneous metastasis model this same correlation exists between ISR^high levels and cell-cycle arrest. The correlation was evaluated by two different approaches, high-resolution imaging using IF and single-cell resolution gene expression analysis of DCCs. We performed IF of MMTV–HER2 lung tissue sections of animals bearing large tumors and thus bearing dormant and proliferative DCCs (26). Tissues were co-stained to detect DCCs positive for HER2, Ki67 (as a marker of proliferation), and GADD34 (or PPP1R15A) as a surrogate marker of PERK activation. GADD34 is a PERK-inducible stress gene responsible for the programmed shift from translational repression (due to eIF2α phosphorylation) to stress-induced gene expression (35). However, GADD34 can be induced by other eIF2α kinases (e.g., GCN2, HRI, and PKR) in addition to PERK, suggesting it may not only report on PERK activity and this is why we termed these DCCs and micro-metastatic lesions ISR^high. Image analysis showed that HER2⁺ metastatic lesions or solitary DCCs with a low proliferative index (Ki67^low) presented high levels of ER stress as shown by high levels of GADD34 expression (Fig. 1A top and graph). On the other hand, highly proliferative DCCs or lesions showed very low levels of GADD34 staining (Fig. 1A bottom and graph). The two markers, Ki67 and GADD34, did not co-occur in 100% of the cells and proportion of Ki67⁺ and GADD34⁺ cells were anti-correlated (Spearman test, P = 0.01), supporting that in our HER2⁺ model ISR^high and quiescent DCCs and metastatic lesions can be identified via GADD34 detection. Inhibition of PERK with three potent and specific inhibitors of PERK (HC4, HC19, and HC28) in MCF10A breast pre-malignant cells and E0771 luminal B breast cancer cells completely eliminated thapsigargin-induced GADD34 expression, suggesting that while not a unique PERK-regulated gene GADD34 can be highly dependent on this kinase in ISR^high breast premalignant and malignant cells (Supplementary Fig. S1A).

We next tested whether these correlations would also hold true in human breast metastatic lesions. HER2⁺ human breast cancer metastases (n = 10) to lymph node and additional 7 metastatic samples from different subtypes and tissues (lymph node, lung, and liver; Supplementary Table S1) were stained with a pan-cytokeratin cocktail to identify the metastatic lesions, and for Ki67 and GADD34. We observed that advanced human metastatic lesions displayed a more heterogeneous pattern of staining for both markers between different patients and in-between different areas of the same lesion than in the mouse model. However, a consistent negative correlation between levels of proliferation (Ki67⁺) and GADD34 expression was found in HER2⁺ lymph node metastasis (Fig. 1B) or other target organs as well (Spearman test, P < 0.028; Supplementary Table S1). This analysis validates the findings in the mouse model and that GADD34 may help identify ISR^high/quiescent tumor cells in human metastatic sites.

Next, the analysis was expanded to markers of proliferation, quiescence, dormancy, and ISR in metastatic cells in the HER2⁺ mouse model. To this end, we performed single-cell–targeted gene expression analysis of solitary DCCs, micro- and macro-metastases lodged in lungs of HER2⁺ mice. Lungs from MMTV–HER2 females were processed into single-cell suspensions and HER2⁺/CD45⁻ cells were sorted (Supplementary Fig. S1B). The sorted cells were then processed for single-cell separation, lysis, RT, and pre-amplification using the C1 (Fluidigm) technology as shown in Supplementary Fig. S1B. This technique allowed us to isolate and process with a high degree of confidence (IF and molecular confirmation of HER2⁺ single cell) and quality 255 single DCCs and 90 primary tumor cells and their corresponding pools. Subsequently, high-throughput qPCR was used to analyze the expression of ISR genes, cell-cycle genes (both activators and inhibitors), and dormancy genes based on the literature (refs. 25, 36, 37; Supplementary Fig. S1C). Single-cell resolution gene expression of DCCs revealed the existence of two populations of cells that are enriched for ISR genes (groups 1 and 2, 41% of DCCs; Fig. 1C). Group 1 (approximately 19% of the DCCs) showed concomitant and strong upregulation of all the ISR genes tested (including PERK itself; green box) along with negative regulators of cell proliferation such as Rb1 and TP53 and CDK inhibitors p21, p27, p16, and p15 (pink box; Fig. 1C). We also observed in these cells, enrichment in the expression of dormancy genes such as NR2F1, DEC2 (DEC2/Bhlhe41), TWIST1, CDH5, STAT3, and COL4A5 (refs. 25, 37; blue box). The DCCs group 2 (22%) also showed high levels of ISR genes expression along with the quiescence marker p21. In a third group (group 3; 6%) ISR, cell-cycle inhibitors, and dormancy genes were less prevalent, suggesting that these might represent cells transiting out of dormancy or in cycling mode. In total, around 40% of the DCCs showed high to intermediate level of ISR genes expression, concurrent with cell-cycle inhibitors or dormancy genes. Importantly, independent published validation in the same HER2⁺ mouse model showed that approximately 40% of solitary DCCs express the dormancy regulators ZFP281 and NR2F1 (26). Thus, our findings in this study are in range with the percentage of dormant DCCs detected in advanced progression MMTV–HER2 animals previously reported by our laboratory using phospho-Histone H3, phospho-Rb, Ki67, and ZFP281 detection (25, 26). These data illustrate that animals with detectable metastasis are comprised of approximately 40% DCCs that display high expression of cell-cycle inhibitory genes. Importantly, we show that dormant DCC subpopulations display an activated ISR with prominent expression of PERK pathway genes.

To strengthen our claim that dormant HER2⁺ DCCs we profiled in Fig. 1C are upregulating a PERK signaling program, we used a published PERK activity signature of 23 genes that predict for shorter metastasis-free periods of patients with breast cancer (21) and cross-referenced it with independently published data derived from bulk and single-cell DCC RNAseq profiling (26). In agreement with our findings, this PERK activation signature was highly enriched in primary MMTV–HER2 early lesions (those that seed dormant DCCs more frequently; ref. 26) compared with primary overt tumors in our bulk RNAseq analysis (Fig. 1D). The PERK activity signature was also enriched in early DCCs in lungs that adopt a mesenchymal-like and dormant phenotype (Fn1, Bgn, Spp1, Col5a2, Ecm1, Maged2, and Col1a2; ref. 26). Thus, the human-derived PERK activity signature (21) was specifically enriched in cancer cells prone to enter dormancy or that are dormant as DCCs (26). We further tested whether this signature informed on metastasis-free survival in the different subgroups of patients with breast cancer. We found that the patients that displayed an enrichment of the PERK signature progressed faster to metastasis in all analyzed subtypes. Importantly, this was also true for the HER2⁺ subtype modeled in this study and the hazard ratio was higher than in the other subtypes (Fig. 1E). Our results collectively provide a strong basis to support that quiescent DCCs and slow-cycling cells within metastasis are enriched for an ISR^high signature (21).

The above findings opened the possibility of using selective PERK inhibitors to test whether inhibition of PERK could affect dormant DCC fate and metastasis formation. We used a PERK inhibitor derived from a 2-amino-3-amido-5-aryl-pyridine scaffold that is currently in clinical development (27). Briefly, HC4 is a potent and selective PERK inhibitor with appropriate drug-like properties to support in vivo animal and human clinical studies (Supplementary Tables S2–S5). KINOMEscan kinase profiling of HC4 (along other variants from the amino-pyridine-mandelic acid–derived series, HC19 and HC28; Supplementary Tables S3, S4, S6, and S7) displayed exceedingly high selectivity compared with other PERK inhibitors described previously in the literature and specifically GSK2656157 (38), even at a very high concentration (10 μmol/L). The high specificity of HC4 for EIF2AK3 (PERK) over the eIF2α kinase family, EIF2AK1 (also known as HRI), EIF2AK2 (also known as PKR), and EIF2AK4 (also known as GCN2) or HER2 (Supplementary Table S3) indicates that the measured activity on eIF2α was highly specific to PERK inhibition.

In cellular assays, HC4 effectively decreased P-PERK (P-T980) levels in MCF10A cells overexpressing HER2 (MCF10A–HER2) and in HEK293 cells stimulated with Tunicamycin (Supplementary Fig. S1D) as well as GADD34 induction upon thapsigargin treatment (Supplementary Fig. S1A). Importantly, treatment with HC4 rendered MCF10A–HER2 cells sensitive to low-dose thapsigargin treatment (Supplementary Fig. S1E), further supporting the idea that PERK activation is necessary to support survival of cancer cells in a stressed environment. Furthermore, specificity of HC4 was tested using an Fv2E–ΔNPERK fusion construct-expressing cell line described previously (14, 22, 39). This system allows to induce PERK activation via dimerization with the divalent small-molecule AP20187. We used a dose of AP20187 that in this case we knew would induce rather than prevent apoptosis via strong pro-apoptotic PERK signaling (14, 22, 39). We next determined whether HC4 would rescue the pro-apoptotic effect. HC4 fully rescued the pro-apoptotic function of strong PERK activation, supporting it is a specific inhibitor (Supplementary Fig. S1F).

In vivo studies showed that single-dose administration of HC4 in CD1 female mice demonstrated bioavailability following intraperitoneal administration and that the levels of HC4 in the plasma achieve a threshold that is well above that needed for PERK inhibition based on biochemical and cellular P-PERK IC₅₀ values over a 24-hour time window (Supplementary Fig. S1G). Repeated dosing of HC4 from 30 to 100 mg/kg was well tolerated as shown by no significant changes in body weight (Supplementary Fig. S1H).

Having confirmed the HC4 inhibitor specificity and in vivo drug-like properties, we treated 24–32-week-old uniparous MMTV–HER2 female mice (which present an incidence of lung metastases of around 80%) with vehicle (see Materials and Methods) or HC4 (50 mpk) intraperitoneally daily, for two weeks, and collected mammary glands, lungs, pancreas, bone marrow, and overt primary tumors for further analyses. The inhibitor did not have a significant effect on bone marrow cell homeostasis or on peripheral blood white cells as shown by no effect on total cell counts in MMTV–HER2 females (Supplementary Fig. S1I) as an additional indication of safety.

HC4 treatment caused a significant decrease in P-PERK and P-eIF2α levels in tissues such as the mammary gland ducts and in pancreatic tissue (although only partial inhibition was observed at this dose, especially in pancreatic islets), as shown by IHC (Fig. 2A). Similarly, around 50% decrease in P-PERK levels was detected by Western blot of tumor lysates (Fig. 2B; see also Fig. 4A for the effect of HC4 on primary tumor growth). We conclude that systemic HC4 delivery effectively inhibits PERK activation and eIF2α phosphorylation in tissues. The use of the inhibitor did not fully deplete PERK activity, which allows mice to control their pancreatic function and glucose levels (40). Moreover, in a separate study in mice dosed orally with HC4 for 28 days, where similar exposures were achieved to the 50 mpk IP dose noted above, no deleterious effects on the pancreas or with clinical chemistry (i.e., insulin and glucose levels) were observed (data not shown).

MMTV–HER2 animals develop metastases to the lungs, which can be initiated early in progression, even during stages of non-palpable tumor (25, 33, 34). Thus, we next monitored the effect of HC4 on metastatic disease in these animals. More than 80% of the vehicle-treated animals presented metastases detectable in sections stained with hematoxylin and eosin (Fig. 2C). Lesions that displayed >100 cells were categorized as macro-metastases as they are also commonly positive for proliferation markers (Fig. 1A; refs. 25, 41). The quantification of macro-metastases per animal (5 non-consecutive lung sections) revealed that, after just a 2-week treatment period, HC4 reduced the number and the incidence of macro-metastases in the whole population of animals (Fig. 2C, left graph). This effect was independent on the primary tumor size because the quantification of macro-metastases in animals showing non-palpable tumors at this stage (as reported for this model; refs. 25, 42) also revealed a similar decrease in number and incidence in metastases (Fig. 2C, right graph). The effect on the number of macro-metastases was not accompanied by a significant effect on the area of these metastases (Supplementary Fig. S2A). This suggested that PERK inhibition might be acting on the initial steps of metastasis (solitary DCCs or micro-metastases) rather than eliminating established macro-metastases.

To expand these findings to another model of cancer not dependent on HER2, we chose a human HNSCC PDX (T-HEp3) previously shown to enter dormancy in the lung after tail vein injection (refs. 43–47; Supplementary Fig. S2B). To this end, we injected intravenously a total of 2.5×10⁵ T-HEp3 cells as single-cell suspensions in Balb/c nu/nu mice and 24 hours after injection we treated the mice with vehicle or 50 mg/kg of HC4 for 10 days to primarily capture solitary dormant DCCs. At the end of the experiment, mice were euthanized and lungs sectioned to score the frequency of T-HEp3 DCCs in the lung detected via a human-specific anti-vimentin antibody, as reported previously (43, 44, 46). These results revealed that HC4 caused a statistically significant approximately 3-fold reduction in the number of solitary HEp3 DCCs in the lung (Supplementary Fig. S2B). At this stage of lung colonization, HEp3 cells are in a quiescent phase (43), supporting that HC4 can target the survival of these dormant DCCs as in the HER2⁺ mouse model. In addition, to further expand the repertoire of models we employed the Balb/c syngeneic EMT6 model that spontaneously metastasizes to lung (48). In this model, treatment with HC4 at 10 mg/kg 9 days after implanting 10⁶ cells in the fat pad also strongly inhibited metastatic incidence and multiplicity of lesions per lung/animal after 17 days of treatment without affecting primary tumor growth (not shown; Supplementary Fig. S2C).

It is possible that the effect of HC4 could be on the intravasation process and thus less metastasis is simply due to less DCCs reaching the lung from circulation. To test this hypothesis, we checked whether HC4 treatment might be affecting the intravasation of tumor cells from the primary site to circulation. Detection of HER2⁺-circulating tumor cells directly in blood samples showed no significant differences between vehicle and HC4-treated animals (Supplementary Fig. S2D), indicating that HC4 is not grossly affecting the intravasation of tumor cells. We next tested whether HC4 treatment affected the transition from solitary DCC to micro-metastasis (containing 2–100 cells) in the lung tissue. Detection of micro-metastasis and single DCCs in lung sections using HER2 detection via IF revealed a significant decrease in the number of micro-metastases in HC4-treated females (Fig. 2D). Again, this effect was not due to a potential direct effect on the primary tumor because in animals not showing any palpable tumors the effect on micro-metastases was similar (Fig. 2D, right graph).

More than 80% of single DCCs found in lungs were negative for P-Rb, indicating that they are mostly out of cycle and dormant as reported in independent studies (25, 26, 41). HC4 significantly reduced the number of non-proliferating (P-Rb negative) single DCCs that are commonly associated with blood vessels in lung sections, while not affecting the number of P-Rb–positive solitary DCCs (Fig. 2E) or in micro-metastases (Supplementary Fig. S2E). As with macro- and micro-metastases, this effect was independent on the presence or size of a primary tumor (Fig. 2E, right graph).

In this mouse model, the majority (>99%) of single DCCs in the bone marrow are negative for P-Rb or Ki67 (25) and they never form spontaneous bone metastasis (33), indicating that they are mostly out of cycle and dormant (25), making it an ideal model to assess the effect of the PERK inhibitor on disseminated dormant disease. Importantly, HC4 significantly decreased the number of DCCs found in bone marrow (Fig. 2F). To incorporate another model of tumor cell dormancy and test the effect of HC4, we turned to the dormant variant of HNSCC PDX T-HEp3 cells, D-HEp3. D-HEp3 cells have been shown to depend on PERK but also ATF6α signaling for survival (13, 23). These cells are prone to enter dormancy in vivo, adopting a G₀–G₁ quiescent state 48 hours after inoculation on the chicken embryo CAM system or nude mice (28, 49). These experiments revealed that HC4 caused an eradication of dormant D-HEp3 cells in vivo after a 6-day treatment with HC4 (20 μmol/L, used at 10x dose due to dilution in CAM experiments), and this reduction correlated with a 5-fold increase in the accumulation of cleaved caspase-3–positive D-HEp3 cells in the nodules, also detected via a human-specific vimentin antibody (Supplementary Fig. S2F). All together, these results argue that PERK inhibition can selectively target and eradicate dormant DCCs.

The data in Fig. 1A and B showed that in established micro-metastasis and to some extent macro-metastatic lesions, we could find ISR^high cancer cells. This led us to hypothesize that some cancer cells in established metastatic lesions may show a persistent unresolved ISR and/or that other cells may transit through that state and resolve the ER stress. Regardless of the time spent as ISR^high cells, if this state enables fitness and survival for established lesions, targeting this ISR^high population of cells could also affect progression of these metastatic lesions. To this end, we injected MMTV–HER2 cells from primary tumor lesions and via the tail vein and allowed them to expand into established metastatic lesions for 3 weeks as reported (25), which allows for conversion of all late DCCs to form micro-metastasis. We treated one group of animals with vehicle and another with HC4 (50 mg/kg) for two additional weeks. As shown in Fig. 2G, HC4 treatment for 2 weeks was sufficient to strongly reduce macro-metastatic lesion incidence and total metastatic burden, while not affecting the frequency of already established micro-metastases (Supplementary Fig. S2G). These data support that the effect of HC4 on metastasis is due to a combined effect of targeting ISR^high solitary DCCs (Fig. 2C–F) and micro-metastasis progression to larger lesions (Fig. 2G).

We next tested the effect of HC4 on primary tumor lesions during the spectrum of HER2⁺ mammary cancer in mice. HER2-driven progression was found to be genetically dependent on the PERK kinase in the MMTV–HER2 model previously (17) and another study showed that HER2⁺ tumors are sensitive to proteotoxicity and dependent on ERAD for survival (19). Furthermore, TCGA database analysis (20) showed that approximately 14% of HER2-amplified human breast tumors (Breast Invasive Carcinoma, TCGA, Nature 2012 dataset) display upregulation of the mRNA for PERK. This population may be a subset of a larger population of patients that we found in Fig. 1, have a strong enrichment in a PERK high signature (48.74% of tumors for the HER2⁺ subtype), and have also poor prognosis. Together, these data led us to investigate whether HC4 affected HER2-induced breast tumor progression in primary lesions where the different stages of progression from hyperplastic mammary glands through DCIS and invasive cancer can be dissected (50, 51).

Analysis of 24-week-old uniparous female mammary glands showed that vehicle-treated MMTV–HER2 animals exhibited ducts with secondary and tertiary dense branching (Fig. 3A, left), and histological analysis showed frequent mammary hyperplastic lesions (Fig. 3A, right, black arrows). In contrast, HC4-treated animals showed a “normalized” glandular architecture with less dense branching, resembling the mammary tree of non-transgenic normal FVB mice (Supplementary Fig. S3A). HC4-treated animals also showed a dramatic increase in the number of hollow lumen mammary gland ducts, constituting more than 60% of the structures compared with around 20% in control females (Fig. 3B; Supplementary Fig. S3C). The number of occluded hyperplasias and DCIS-like lesions was also reduced to less than half of that observed in vehicle-treated animals. Hyperplastic lesions in control HER2⁺ animals showed varying degrees of luminal differentiation as assessed by the uneven levels of cytokeratin 8/18 expression (Fig. 3C, top). The myoepithelial cells (detected as smooth muscle actin, SMA, positive) were unevenly distributed in the vehicle-treated hyperplasias in the MMTV–HER2 mice. In contrast, HC4-treated MMTV–HER2 animals presented increased expression of cytokeratin 8/18 in the luminal layer, frequently surrounding an empty lumen, and an external continuous layer of myoepithelial cells (Fig. 3C, lower and graph). Treatment of non-tumor bearing healthy mice with HC4 for 2 weeks did not affect the normal branching and morphology of the mammary gland ductal tree (Supplementary Fig. S3B). These data indicate that HC4 treatment leads to a delayed progression of HER2-driven early lesions in the mammary gland.

We next treated animals once they displayed tumors, ranging from 30 to 200 mm³ volume (two tumors were >200 mm³) for two weeks with HC4. In the vehicle treatment group, tumors grew steadily (Fig. 4A), reaching up to 10 times its original volume in two weeks (Supplementary Fig. S4A, left graph). In contrast, HC4-treated tumors showed a reduced growth rate (Fig. 4A), with some tumors remaining in complete cytostasis (defined as doubling tumor volume only once in the 2-week period, 43% in HC4-treated vs. 7% in controls; Supplementary Fig. S4A, right graph). Four tumors showed regression in the 2-week window treatment (Supplementary Fig. S4B). This led to a significant decrease in median final tumor volume (Fig. 4B). Although the levels of proliferation (phospho-histone H3 IHC) were not different between vehicle- and HC4-treated tumors (Supplementary Fig. S4C), TUNEL staining of tumor sections showed a significant increase in the levels of DNA fragmentation present in HC4-treated animals (Fig. 4C). Thus, in overt primary lesions, HC4 treatment induced apoptosis of established HER2⁺ tumors, arguing for context-dependent fitness-promoting functions of PERK during tumor progression.

Treatment of HER2-overexpressing (MCF10A-HER2 or ZR-75–1) or HER2-amplified (SKBR3) breast cancer cells (Fig. 4D; Supplementary Fig. S4D) in 3D acini cultures in Matrigel showed that a 10-day treatment with HC4 (2 μmol/L) significantly increased levels of apoptosis (cleaved caspase-3) in these spheroids, especially in the inner cell mass that is deprived from contact with the ECM (Fig. 4D). As in the in vivo conditions, we did not detect a significant change in the levels of proliferation as detected by phospho-histone H3 levels (Supplementary Fig. S4E). We conclude that early MMTV–HER2 lesions require PERK for HER2-driven alterations in ductal epithelial organization and, eventually, tumor growth.

Although we showed effects of HC4 in HER2-dependent and -independent tumors (HEp3 EMT6), we wondered about the functional relationship between PERK and HER2-driven biology. This was because the KM curves in Fig. 1 showed a strong association between poor prognosis and a PERK^high signature in HER2⁺ breast cancers, and also because HER2⁺ tumors are sensitive to proteotoxicity (19). Thus, HER2 signaling might depend on PERK and inhibition of PERK might affect optimal HER2 activity due to increased ER client protein load. Detection of HER2 phosphorylation at residues Y1221/1222 in tumors showed that the area positive for P-HER2 reported by others (52) overlapped with the staining for P-PERK and P-eIF2α (Fig. 5A), showing a distinctive rim staining pattern. This finding indicated that the activation of PERK and HER2 pathways is not generalized and co-localize in the same area of the tumor tissue. Single-cell–targeted-gene expression profiling of primary tumor cells showed a population of primary tumor cells (around 25%) with high levels of ER stress genes expression (Fig. 5B), which could correspond to the ones localized in the rim of the tumor. Importantly, when we scored the P-HER2 levels in the tumors (taking into account both the area and the intensity of the staining; Supplementary Fig. S5A), we found that HC4-treated tumors showed significantly lower levels of P-HER2 than control animals (Fig. 5C). HC4 does not target HER2 (or ErbB family) kinase activity directly as shown in Supplementary Table S6 and (27, 53). In vitro treatment of MCF10A-HER2 cells with HC4 revealed that PERK inhibition decreased both the basal and EGF-induced levels of P-EGFR and P-HER2, along with downregulation of the survival pathway P-AKT, P-S6, and P-ERK1/2 levels (Fig. 5D; Supplementary Fig. S5B). No obvious effect was observed under these conditions on total HER2 levels or heterodimerization with EGFR as determined by surface biotinylation and co-immunoprecipitation studies (data not shown). As mentioned above, because HC4 does not have a direct inhibitory effect on the active site of any of the ErbB family members, AKT or S6 kinases (Supplementary Table S6), we hypothesized that this effect must be due to an indirect effect of PERK inhibition on HER2 signaling. In contrast with other ErbB family members, HER2 is known to remain at the plasma membrane after ligand binding and dimerization (54, 55). We thus tested whether HC4 might be disturbing the mechanism of activation of HER2 receptors. To this end, we performed surface biotinylation assays to measure the presence of the receptor on the cell surface, and reversible surface biotinylation to measure receptor endocytosis (30). Our data showed that although HC4 treatment did not significantly alter the abundance of P-HER2 and total HER2 in the cell surface (not shown), it clearly increased the endocytosed phospho- and total HER2 (Fig. 5E). That the surface levels did not change may be due to the dynamic nature of the process and high expression of HER2 that may preclude detecting a decrease in surface levels but not an endocytic retention. Previous reports (56) have shown the importance of PERK for the proper maturation of several tyrosine kinase receptors, by partially blocking their trafficking from the ER to the Golgi in conditions of stress. We, as well, observed a lower molecular weight of phosphorylated HER2 protein when treating cells with HC4 in the presence of low levels of stress (Thapsigargin, 4 nmol/L), along with a decrease in the levels of P-HER2 and P-AKT (Fig. 5F). The lower molecular weight phospho-HER2 protein could correspond to the previously reported 170 kDa precursor, containing high mannose-type oligosaccharide chains located in the ER compartment (57). Our data suggest that PERK signaling and proper ISR function is required to maintain proper HER2 downstream signaling by affecting optimal receptor localization and activation.

We next tested the hypothesis that treatments that may mimic dormancy could render the growth-arrested cells sensitive to HC4 treatment. In Fig. 6A, 20-week-old MMTV–HER2 mice with tumors were treated for 4 weeks with the CDK4/6 inhibitor Abemaciclib (Abema) at 50 mg/kg or control treatment. In parallel, mice were treated with control or HC4 for 4 weeks. Then, some of the mice treated with Abema were switched to treatment with HC4 for 2 weeks. We processed the tissues for analysis of ISR markers, metastasis, and DCC burden.

The results in Fig. 6B show that Abema alone reduces in primary tumors Ki67 staining and increases GADD34 expression, supporting that as we found in spontaneously dormant DCC clusters and micro-metastasis, acquisition of a forced growth arrest also triggers an ISR signal. Figure 6B also shows that HC4 treatment alone does not dramatically affect Ki67 signal as expected from our previous results but completely erased the GADD34 induction, supporting that GADD34 is a good readout for PERK signaling as supported by Figs. 1–2 data. Importantly, Abema + HC4 significantly decreased the number of macro-metastasis per animal compared with vehicle and a larger N would be required to draw significance from the combination versus Abema alone (Fig. 6C). Nevertheless, the results support that growth-arrested cells that enter that state in response to Abema and possibly other growth arrest inducers can be further targeted by HC4 via PERK inhibition. We further quantified micro-metastasis and solitary DCCs in different proliferative states. When scoring micro-metastasis, it was clear that HC4 further reduced micro-metastatic burden following Abema treatment (Fig. 6D). In Fig. 6E, we also show that HC4 can target Ki67⁻ DCCs after Abema treatment, reproducing our prior results but now under forced induction of growth arrest. As before, we did not find an effect on Ki67⁺ DCCs but these were almost absent due to the Abema treatment. Overall, these data support that inducing a growth arrest program, spontaneously as in Fig. 2 or in response to a CDK4/6 inhibitor, can render DCCs and likely established metastatic lesions susceptible to PERK inhibition with HC4.

Breast cancer models have suggested that HER2⁺ breast cancer tumorigenesis is dependent on PERK signaling for survival and adaptation (17, 19). We had found that quiescent tumor cells that exist within surgical margins or as dormant DCCs in target organs (46, 47, 58, 59) enhance their survival via PERK signaling as well as other ER stress pathways (13, 22, 23, 60, 61). Pommier and colleagues (24) validated this finding in their studies demonstrating that pancreatic DCCs lodged in the liver also activate an ISR during quiescence. This level of concordance across a variety of tumor types and models, including the HER2 and EMT6 mammary models and human HNSCC HEp3 PDX tested here, supports the more general requisite of this stress adaptation biology across the cancer landscape.

We now show that pharmacological PERK inhibition with a very selective PERK inhibitor HC4 can effectively target HER2⁺ (EMT6 and HEp3) solitary DCCs in the lung and bone marrow, established micro-metastasis and primary lesions. HC4 could also target spontaneously dormant D-HEp3 HNSCC cells, which also display dependency on PERK for survival (23). A salient finding to discuss is the inhibitory effect of PERK inhibition on metastasis. In the MMTV–HER2, EMT6, and HEp3 HNSCC models, PERK inhibition reduced metastasis independent of the primary tumor development timeline, including those initiated early (before overt tumors were palpable). This is important because it argues that the effect on metastasis was not simply due to reduced primary tumor burden caused by HC4. These effects are further supported by the effect of HC4 on T-HEp3 HNSCC cells in the lung, HER2⁺ DCCs in the bone marrow and D-HEp3 cells in the CAM assay. Importantly, metastatic burden was reduced by HC4 treatment via eliminating non-proliferative solitary or small clusters of P-Rb–negative cells, but also ISR^high established micro-metastasis. Imaging and single-cell multiplex qPCR robustly revealed that HER2⁺ DCCs showed an upregulation of GADD34 (protein) and a larger set of ER stress and PERK-regulated genes, while also expressing genes representative of the quiescent phenotype as revealed by upregulation of several negative regulators of cell proliferation and dormancy markers. It should be taken into account that part of the PERK-induced ER stress response is transcriptional in nature while also having a key component of preferential translation of upstream ORF-containing genes, such as ATF4 and GADD34. Similarly, ISR-induced G₁ arrest has been shown to be caused by inhibiting the translation of cyclin D1 (62). Our data strengthen the argument that, like overt tumors that contain ISR^high cells, also quiescent DCCs are likely to rely on PERK signaling for survival. This supports the idea that HC4 PERK inhibition may benefit the efficacy of anti-proliferative therapies, as it would allow targeting those that divide (e.g., with anti-HER2, chemotherapy or radiotherapy) and the quiescent cancer cells that escape conventional treatment with HC4. A sub-population of human metastatic cells from patients with breast cancer also showed a negative correlation between GADD34 and Ki67, supportive of the association seen in this study in mouse models. Our data suggest that along with NR2F1 (63), GADD34 alone or in combination with NR2F1 may serve as a robust biomarker set for dormant/ISR^high DCCs and thus guide patient selection for treatment. A limitation of this study is related to the fact that we used GADD34 as a marker for PERK activation in human and mouse samples as we were not able to quantify PERK phosphorylation in DCCs. GADD34 can also be induced by other EIF2AK-dependent pathways limiting the claim that it is a solely readout for PERK. However, given our results with the HC4 compound and collectively with all results we are confident that GADD34 is in this system reading out PERK reliably as GADD34 expression was fully eliminated by treatment with PERK inhibitors. However, we cannot support that GADD34 is a linear reporter of PERK in vivo, without a specific control inhibiting PERK. An open question is related to the identification of the source of PERK activation in quiescent DCCs, which remains unknown. However, a recent publication revealed that dormant cells are avid producers of their own collagen-rich ECM niche (64). Translation folding and secretion of collagen impose a high demand on the ER, and thus, a heightened ISR may allow dormant cells to survive the ISR and oxidative stress imposed by such niche remodeling.

We also demonstrate that cytostatic therapies such as CDK4/6 inhibitors (Abemaciclib) not only decrease proliferation substantially, but at the same time result in concomitant activation of the ISR as shown by high GADD34 levels upon CDK4/6 inhibition. This observation would support the possibility of combining such CDK4/6-targeting therapies with PERK inhibitors to have an even more profound control of both proliferating and non-proliferating cancer cells. Encouragingly, the doses of HC4 we used did not significantly affect glucose levels, bone marrow or peripheral blood cell counts, drinking and feeding behavior of non-tumor or cancer-bearing mice. A limitation of the experiments combing HC4 and Abemaciclib was that the inhibitory effect of the combination on the number of macro-metastasis per animal was not significant, likely due to reduced n. However, that HC4 + Abemaciclib reduced significantly the frequency of micro-metastasis and a similar trend was seen for Ki67⁻ DCCs, further supports that perhaps larger studies or longer treatments would further reduce metastatic growth over the effect of each drug alone. Nevertheless, collectively, these data support the dose range evaluated in which we illustrate that a significant blockade of tumor growth and metastasis is possible through the elimination of dormant and micro-metastatic lesions, and that HC4 does not adversely affect the host's normal organ function at the doses tested.

In early lesions, our work shows that HC4 treatment precludes the development of hyperplasias and mammary intraepithelial neoplasias, indicating that HC4 treatment could have a potential preventive effect on the progression of early lesions such as DCIS to more aggressive invasive breast cancer. In established tumors, HC4 used as a single agent pushed tumors into stasis or regression via apoptosis. The full details of the mechanisms by which HC4 blocks tumor cell survival continue to be elucidated. It is possible that reduced adaptation to stress imposed by proteotoxicity in cancer cells (19) is a mechanism. Our data also revealed that HC4 reduced phospho-HER2 levels in vivo and decreased the abundance of active receptor in the membrane through enhanced endocytosis, but we did not see changes in HER2 protein degradation. However, it is still unclear how exactly PERK controls HER2 membrane localization or endocytosis. It is possible that internalization allows for better or faster de-phosphorylation of the receptor or decreases the chances of it being activated; hence, resulting in decreased downstream signaling. This possibility is supported by the finding that shows that receptor endocytosis can reduce the signaling output of many plasma membrane receptors by physically reducing the concentration of the receptors at the cell surface (65). Nevertheless, HC4 also induced cell death in HNSCC cells that do not depend on HER2 suggesting that PERK may enable adaptation to stress under different oncogenic pathways.

Discovering a target and drug that can eradicate dormant DCCs in addition to cells that undergo short- or long-term quiescence in metastatic lesions is highly significant because dormant DCCs are known to evade anti-proliferative therapies via active and passive mechanisms (66–72). Our work opens the door to the use of anti-dormant DCC survival therapies as a new way to target metastatic disease. This would allow targeting the full phenotypic heterogeneity of disseminated disease that may include proliferative, slow-cycling, and dormant DCCs (70). The eradication of DCCs in the bone marrow, where these cells also commonly reside in a dormant state (33, 47, 59, 73, 74), further strengthens the notion of PERK inhibition as an anti-dormant DCC therapy that may be used in the adjuvant setting to eliminate dormant minimal residual disease (70).

V. Calvo reports grants and non-financial support from Eli Lilly and non-financial support and other support from Hibercell during the conduct of the study as well as other support from Hibercell and Eli Lilly outside the submitted work. K.A. Staschke reports grants and personal fees from HiberCell, Inc. outside the submitted work. A. Nowacek reports grants from HiberCell during the conduct of the study as well as personal fees from HiberCell outside the submitted work; in addition, A. Nowacek reports a patent for WO-2021231788-A1 issued. E.F. Farias reports personal fees from HiberCell Inc. during the conduct of the study as well as personal fees from HiberCell Inc. outside the submitted work. J.A. Aguirre-Ghiso reports personal fees from HiberCell during the conduct of the study as well as personal fees from HiberCell outside the submitted work. No disclosures were reported by the other authors.

V. Calvo: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing. W. Zheng: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. A. Adam-Artigues: Conceptualization, data curation, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. K.A. Staschke: Data curation, formal analysis, methodology, writing–original draft, writing–review and editing. X. Huang: Data curation, methodology. J.F. Cheung: Investigation, methodology. A.R. Nobre: Investigation, methodology. S. Fujisawa: Investigation, methodology. D. Liu: Investigation, methodology. M. Fumagalli: Investigation. D. Surguladze: Investigation. M.E. Stokes: Investigation. A. Nowacek: Investigation. M. Mulvihill: Investigation. E.F. Farias: Investigation. J.A. Aguirre-Ghiso: Conceptualization, resources, supervision, funding acquisition, investigation, writing–original draft, project administration, writing–review and editing.

We thank the J.A. Aguirre-Ghiso and HiberCell teams for useful discussions. We thank Brian Lee for help with Clustergrammer use. Grant support from Eli Lilly (to J.A. Aguirre-Ghiso), LIFA Fellowship (Eli Lilly; to V. Calvo), Ramon Areces Postdoctoral fellowship (to A. Adam-Artigues), NIH/NCI (CA109182, CA196521, and CA216248), HiberCell, METAVIVOR, and DoD-BCRP Breakthrough Award (BC132674; to J.A. Aguirre-Ghiso). J.A. Aguirre-Ghiso is a Samuel Waxman Cancer Research Foundation Investigator.

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.