PARP Inhibition Enhances Radiotherapy of SMAD4-Deficient Human Head and Neck Squamous Cell Carcinomas in Experimental Models

Worldwide, head and neck cancer is the sixth most common cancer type (1); over 90% are squamous cell carcinoma of the head and neck (HNSCC; ref. 2). Among HNSCCs, human papillomavirus (HPV) and tobacco consumption are the two major etiologies (3). Tobacco-associated HNSCCs have worse outcomes than HPV-associated HNSCCs. Despite recent advances in therapeutic approaches, patients with HNSCC have a 5-year survival rate of 40% to 50% (3, 4).

In tobacco-associated HNSCCs, increased DNA repair rescues tumor cells from dying of DNA damage–associated cell death, contributing to treatment failure (3, 5). Among DNA repair molecules, the FANC/BRCA family genes are uniquely associated with HNSCC susceptibility, that is, Fanconi Anemia survivors (with germline FANC/BRCA gene mutations) have a high susceptibility to HNSCC (6). The FDA has approved several PARP inhibitors to treat advanced ovarian and metastatic breast cancers with BRCA mutations because blockade of PARP-mediated DNA repair is synthetically lethal in the context of mutant BRCA1/BRCA2 inducing DNA damage–associated death in cancer cells. Loss of BRCA1 and BRCA2, primarily through loss-of-functional mutations, occurs in 2.2% and 5% of human HNSCCs, respectively (7). Therefore, alternative molecular markers are needed to predict therapeutic response of PARP inhibitors. Among such potential molecular markers, SMAD4 loss represents a promising candidate. About 35% to 52% of human HNSCCs lose at least one copy of SMAD4 gene, largely through chromosomal deletion, and the remaining allele can be lost through homozygous deletion, loss of heterozygosity, and transcriptional silencing (8, 9). Furthermore, HNSCCs with heterozygous SMAD4 deletion contain homozygous SMAD4-deleted cells in their heterogeneous cell populations (9).

SMAD4, a tumor suppressor in the TGFβ signaling pathway, regulates proliferation, apoptosis, and genomic stability (10, 11). Loss of SMAD4 is detected in early-stage human HNSCCs and Smad4 deletion in mice causes spontaneous HNSCCs (11). We have shown that Smad4 null HNSCCs have genomic instability associated with decreased Brca1 and Rad51 expression (abnormal centrosome numbers, high mutational load, and hypersensitivity to mitomycin-C–induced chromosomal cross-linking and breakage), and in human HNSCC tissue arrays, SMAD4 levels correlate with BRCA1 and RAD51 levels (11). Furthermore, expression of several FANC/BRCA genes is reduced by knocking down SMAD4 in human keratinocytes but reinstated by SMAD4 restoration in HNSCC cells, and BRCA/RAD51 DNA repair foci are reduced in SMAD4-mutant HNSCC cells but restored with SMAD4 expression (11). These findings suggest SMAD4 loss plays a causal role in downregulation and functional defects of FANC/BRCA genes in HNSCCs, resulting in a “BRCA-like” phenotype (12). The remaining questions we sought to answer are: (i) whether SMAD4-deficient “BRCA-like” molecular signatures apply to patients with HNSCC and (ii) whether the SMAD4-deficient “BRCA-like” phenotype is sufficient to impact therapeutic interventions. Because other cancer types with a “BRCA-like” phenotype, that is, decreased expression and function of FANC/BRCA family gene(s) necessary for DNA repair (13), have shown susceptibility to PARP inhibitors like olaparib (14–16), we sought to explore if SMAD4-mutant HNSCCs are susceptible to PARP inhibition. Clinical trials have shown olaparib is well tolerated by patients with HNSCCs (17). Given that radiotherapy (RT) is a standard-of-care for locally advanced HNSCCs, we tested olaparib in combination with RT in HNSCCs with or without SMAD4 deficiency and analyzed cellular mechanisms by which SMAD4 loss contributes to the therapeutic response.

The human HNSCC provisional sequencing dataset was queried for mRNA expression z-scores for SMAD4 and FANC/BRCA family genes via cBioPortal for Cancer Genomics (18, 19) Among 530 HNSCC cases, 522 had genotype/expression information, 116 cases were tested for HPV, and 42 were HPV⁺ (TCGA, provisional, queried: Aug 29, 2019). Because HPV status is not known for all samples, we included all cases for analyses.

Human HNSCC cell lines were collected under Materials and Transfer Agreements (MTA), and authenticated by the Tissue Culture Shared Resource at the University of Colorado Cancer Center and tested for Mycoplasma every 3 months. 200–3,200 cells were seeded to 6- or 12-well plates; treated with 0, 0.1, 1, or 5 μmol/L olaparib and 0–6 Gy RT; and cultured for 8 to 14 days for clonogenic assay using criteria as described previously (20). Fresh media containing drugs was applied every 3 days. Colonies were fixed and stained with 1% crystal violet in methanol. Colonies containing 50 or more cells were counted, plotted on log-scale graphs and fitted using the linear-quadratic model (21). There were 3 to 6 replicates for each treatment.

CAL27, a human tongue SMAD4-mutant tumor cell line (22), was purchased from ATCC (catalog no. CRL-2095). To conditionally express SMAD4, pLenti-CMV-rtTA3-blast vector (Addgene, catalog no. 26429-LV) was transduced into CAL27 cells. After blasticidin selection, pLVX-tight-Puro FLAG-SMAD4 TetON (+SMAD4) or FLAG-empty TetON (+empty vector) construct [provided by the Massague lab (23)] was transduced by lentivirus into CAL27-rtTA3 cells and selected with puromycin. UMSCC1, a floor-of-the-mouth–derived tumor cell line was provided by the Carey lab (24) and validated by fingerprint sequencing. SMAD4 shRNA (shSMAD4) or nontargeting control shRNA (shCTRL; sequences in Supplementary Table S1) was inserted into a tet-pLKO-neo vector (Addgene, catalog no. 21916) and transduced by lentivirus into UMSCC1 cells, then selected with G418. Doxycycline concentration was optimized for induction of SMAD4 expression in CAL27-rtTA3 cells or to knockdown SMAD4 in UMSCC1 cells. HaCaT keratinocytes were purchased (Addexbio, catalog no. T0020001).

RNA Plus mini Kit (Qiagen, catalog no. 74136) was used to extract total RNA from cells. Relative SMAD4, KRT14, or FANC family genes was determined with TaqMan assays using Brilliant II qRT-PCR one-step Master Mix Kit (Agilent, catalog no. 600809) for each reaction. C_t values of the gene of interest and KRT14 were used to determine relative fold change by 2^−ΔΔC_t. Proteins were harvested with RIPA buffer (Cell Signaling Technology, catalog no. 9806S) supplemented with protease and phosphatase inhibitor cocktails (Roche, catalog nos. 5892970001 and 4906845001). Western blotting was performed using standard protocols and detected using an Odyssey imager. Antibodies and TaqMan assays are described in Supplementary Table S1.

Animal studies were approved by the University of Colorado AMC Institutional Animal Care and Use Committee. Athymic female, 6-to 8-week-old nude mice were purchased from Jackson Laboratories. Cells (1 × 10⁶) were suspended in 50% Matrigel (Thermo Fisher Scientific, catalog no. CB-40234) and injected subcutaneously into the flanks of mice. When tumors reached 100 mm³, doxycycline (2 g/L) was administered in sugar water (4 g/L) and maintained throughout the study. At 300 to 400 mm³, the mice were randomized into treatment groups at day 0. At day 1, mice received olaparib (25 mg/kg, gavage) or vehicle, and beginning on day 3, tumors were directly exposed to fractionated RT every 3 days (3 Gy × 6 = 18 Gy). Olaparib was formulated in 10% v/v DMSO/50% v/v of 30% w/v kleptose. An RS2000 instrument was used for x-ray irradiation of cells and animals. Vehicle or olaparib treatments continued through the entire study and tumor burden (weight loss, tumor size, etc.) was monitored to determine study end. Tumor volume was determined by the formula, volume = (width x width x length)/2. Tumors were collected for molecular and morphologic analysis after sacrifice.

Apoptotic cells were stained using a commercially available Fluorometric TUNEL Kit (Promega, catalog no. G3250) per the provided protocol. Immunofluorescence staining was performed on xenograft tumor specimens collected at the end of each study and immunocytochemistry was conducted for cellular kinetics. Primary antibodies used are detailed in Supplementary Table S1. Samples were counterstained with DAPI and Alexa Fluor–conjugated secondary antibodies (Thermo Fisher Scientific) were used for detection. For xenograft or primary HNSCCs, 3–5 immunofluorescence regions were averaged for quantification of TUNEL⁺, pH2AX⁺, or Ki67⁺ cells per specimen; 3–7 tumors per group were used for analyses. For pH2AX immunocytochemistry, plates were fixed at 0, 1, 8, 24, and 48 hours post-radiation using 10% neutral-buffered formalin. For Ki67 staining, plates were fixed at 0 and 48 hours. Plates were stored in PBS with 0.02% NaN₃ at 4°C until all time points were collected and molecularly labeled as described above. Stained 96-well plates were imaged using an Opera Phenix Imaging System with Harmony software (High-throughput Screening Core, University of Colorado, Anschutz Medical Campus, Aurora, Colorado) to determine the number of pH2AX foci per nucleus or the number of Ki67-positive cells over time.

Patients with locally advanced HNSCC were enrolled in a phase I trial and treated with olaparib with concurrent radiation and cetuximab (see details in enrollment criteria and treatment regimens in original report; ref. 17). Pretreatment biopsies were collected under the IRB approval and used to perform SMAD4 FISH as reported previously (9). SMAD4 immunostaining was performed as described above and scored for tumor epithelial intensity between 0 (no staining) and 3 (equivalent to normal epithelial staining). Immunofluorescence of pH2AX was performed as described above and scored between 1 (<20% foci-positive cells) and 3 (>50% foci-positive cells). The correlation between SMAD4 immunostaining scores and pH2AX scores was analyzed by Pearson analysis.

To determine whether “BRCA-like” molecular signatures found in mouse HNSCCs and human HNSCC cells (11) apply to a large population of HNSCCs, we analyzed SMAD4 and FANC/BRCA expression levels in TCGA data (18, 19). SMAD4 mRNA expression levels in HNSCCs are generally lower than in normal epithelia (11) and HNSCCs with SMAD4 downregulation are expected to have <50% of normal SMAD4 mRNA levels; therefore, our criteria for SMAD4 downregulation was < −1.5 SDs from the mean and SMAD4 upregulation >1.5 SDs from the mean. Among all 530 HNSCC cases, 42 of 116 tested cases were HPV⁺. SMAD4 was downregulated in 17% (n = 88; “SMAD4^low”) and upregulated in 7% (n = 38; “SMAD4^high”) of human HNSCCs. Strikingly, SMAD4^low patients appeared to have decreased expression of many FANC/BRCA family genes; conversely, SMAD4^high HNSCCs had increased expression of FANC/BRCA family genes (Fig. 1A). Intriguingly, expression of all 18 FANC/BRCA genes was lower or unchanged in SMAD4^low cases compared with all other cases, and none was higher (Supplementary Fig. S1A). Analysis of all HNSCC cases revealed a positive correlation (and no negative correlation) between SMAD4 and each of the 18 FANC/BRCA genes (Supplementary Fig. S1B), although some correlations were weak, suggesting other regulatory pathways for these genes. Overall, there were more HNSCC cases with decreased expression of one or more FANC/BRCA family genes in SMAD4^low HNSCCs compared with SMAD4^high HNSCCs (Fig. 1B). Because previous studies have shown “BRCA-like” molecular changes in other cancer types outside of the FANC/BRCA family (25–27), we performed a gene set enrichment analysis (GSEA) against the 50 hallmark genes using the expression data of SMAD4^low and SMAD4^high samples. Genes in the “DNA repair” hallmark gene set trended toward enrichment in these SMAD4 downregulated or upregulated HNSCCs (Fig. 1C) but not as strongly as the enrichment observed in Fanconi anemia pathway gene set (Supplementary Fig. S2). To assess whether SMAD4 genetic loss correlates with the olaparib sensitivity seen in BRCA-mutant tumors, we performed a clonogenic survival assay in a panel of human HNSCC cell lines in response to olaparib treatment (0, 0.1, 1, and 5 μmol/L). Cell lines insensitive to olaparib were without SMAD4 loss, and cell lines with SMAD4 loss were sensitive to olaparib (Fig. 1D and E). Two SMAD4 wild-type cell lines were sensitive to olaparib independent of SMAD4 loss.

To assess whether SMAD4 loss plays a causal role in HNSCC sensitivity to DNA damaging therapeutic agents, we generated isogenic cell lines from SMAD4-mutant (CAL27) and SMAD4 wild-type (UMSCC1) HNSCC lines by restoration of SMAD4 in CAL27 and knockdown of SMAD4 expression in UMSCC1 cells using doxycycline-inducible systems. These cell lines modulated SMAD4 protein and transcript expression in response to doxycycline treatment as expected (Supplementary Fig. S3A–S3C). In addition, expression of SMAD4 in CAL27 induced the relative expression of all nine of the FANC family genes examined and SMAD4 knockdown in UMSCC1 reduced the relative expression of 6/9 FANC family genes (Supplementary Fig. S3D). We then tested the influence of SMAD4 on therapeutic responses in xenograft tumors (Fig. 2A). CAL27+empty vector (SMAD4 mutant) xenografted tumors had a long latency prior to olaparib response, that is, approximately 5 weeks before tumor volumes declined, with long-term responses apparent in three of five tumors (Fig. 2B). RT alone rapidly reduced tumor volume; however, tumors began to grow after 5 weeks (Fig. 2B). Dual olaparib and RT treatment reduced all tumor volumes to the baseline (Fig. 2B and E).

To determine whether SMAD4 restoration in CAL27 tumors would attenuate the therapeutic effects observed with SMAD4 loss, we transplanted CAL27+SMAD4 cells prior to doxycycline induction. Once tumors were >300 mm³, doxycycline was administered to induce SMAD4 expression in CAL27 tumors. However, doxycycline-induced SMAD4-expressing tumor xenografts failed to enter growth phase even when we transplanted 10x more CAL27+SMAD4 cells (Supplementary Fig. S4). When doxycycline induction of SMAD4 expression was discontinued, these tumors rapidly entered growth phase, confirming that SMAD4 restoration was the cause of blunted tumor growth. Although this result prevented assessment of therapeutic effects in these tumors, it was verified that SMAD4 is a strong tumor suppressor in HNSCC.

In complement, we tested therapies in SMAD4-positive UMSCC1 (shCTRL) and SMAD4 knockdown (shSMAD4) tumors. UMSCC1+shCTRL xenograft tumors did not respond to olaparib (Fig. 2C and F). RT or olaparib + RT modestly reduced tumor growth, but there was no significant difference between these two treatments (Fig. 2C and F). Unlike CAL27 tumors, olaparib did not alter UMSCC1+shSMAD4 xenograft tumor volumes (Fig. 2D and G). RT transiently reduced UMSCC1+shSMAD4 xenograft tumor volumes. However, only dual olaparib and RT treatment was sufficient to maintain tumor reduction in UMSCC1+shSMAD4 xenograft tumors compared with vehicle or RT alone (Fig. 2D and G). Finally, we analyzed the number of mice with CAL27 tumors less than 1,200 mm³ or UMSCC1 tumors less than 1,500 mm³ over time and found that RT or olaparib + RT prevented all CAL27 xenografts from reaching 1,200 mm³, significantly different from other treatments (Fig. 2H), whereas the olaparib treatment group had two outgrowing, resistant tumors (Fig. 2B and H). However, only olaparib + RT prevented all UMSCC1+shSMAD4 tumors from reaching 1,500 mm³, significantly different from RT alone (Fig. 2J) while no treatment group significantly influenced UMSCC1+shCTRL tumor outgrowth (Fig. 2I). These data suggest that SMAD4-deficient HNSCCs are more prone to respond to olaparib with RT.

We examined whether the reduction in tumor volumes by olaparib/RT dual therapy is due to cell death or decreased proliferation among treatment groups in SMAD4-deficient xenograft tumors harvested at the final time point of the study. There was a significant increase in apoptotic, TUNEL⁺ cells with all three treatments in CAL27 (SMAD4 mutant) and UMSCC1+shSMAD4 xenograft tumors compared with vehicle-treated tumors (Fig. 3A, B, E, and F). Dual therapy induced more apoptosis than RT alone or olaparib alone in UMSCC1+shSMAD4 tumors (Fig. 3F). To determine whether sustained DNA damage contributed to apoptosis in CAL27 xenograft tumors responding to treatments, we stained tumors with pH2AX and measured the percentage of cells positive for pH2AX foci (>3 foci/nucleus). Significant DNA damage was present in CAL27 xenograft tumors with all three treatments (Fig. 3A and C). In UMSCC1+shSMAD4 xenograft tumors, pH2AX⁺ cells were variable and significantly increased by olaparib or dual treatment (Fig. 3E and G). However, Ki67 staining to detect proliferating cells showed a moderate increase in RT-treated CAL27 tumors, consistent with their tendency to recover at this late stage but not in tumors treated with olaparib or RT+olaparib (Fig. 3A and D). Ki67⁺ cells were more abundant in UMSCC1+shSMAD4 xenograft tumors than CAL27 tumors, but showed no significant changes among treatment groups (Fig. 3E and H).

Because the DNA damage–associated cell death found within endpoint tumors could not determine whether olaparib + RT treatment induced more apoptosis early in treatment that led to the most effective tumor regression, we examined treatment response in vitro with multiple doses of olaparib (0–5 μmol/L) and/or radiation (0–6 Gy) using the same isogenic cell lines in clonogenic colony formation assays. Independent of radiation, olaparib dose dependently decreased survival of CAL27 lacking SMAD4 and UMSCC1 with SMAD4 knockdown compared with their respective control cell lines (Fig. 4A and B). Similarly, independent of olaparib, the two cell lines with no/low SMAD4 had fewer colonies than their SMAD4-expressing control cell lines in response to 2 or 4 Gy RT (Fig. 4C and D). CAL27+SMAD4 cells had reduced sensitivity to 6 Gy RT in combination with all doses of olaparib and lower doses of olaparib in combination with 4 Gy RT compared with CAL27 with mutant SMAD4 (Fig. 4C). UMSCC1+shSMAD4 were more sensitive than control cells to the combination of all doses of RT and olaparib tested (Fig. 4D). To determine whether proliferation contributes to cell survival after combination therapy, we examined Ki67 staining before and after 48-hour radiation in combination with continuous olaparib (1 μmol/L, 10 μmol/L). Ki67-marked proliferation in both SMAD4-deficient/low cell lines decreased more than their isogenic SMAD4-positive cell lines in response to radiation or the combination of RT and olaparib (Fig. 4E and F); normalization to DMSO to remove the influence of SMAD4 status on baseline RT sensitivity demonstrated that cells with low/no SMAD4 were still more sensitive to the combination of RT and olaparib (Supplementary Fig. S5A and S5B). SMAD4 manipulation, on its own, did not affect short-term proliferation (Supplementary Fig. S5C and S5D).

To assess if the extent of DNA damage contributes to reduced cell viability in a SMAD4-dependent manner, we irradiated cells with or without continuous olaparib exposure, and examined the number of pH2AX foci/cell over 48 hours. All cell lines had increased pH2AX foci by 1 hour after radiation (Fig. 5C and D). In CAL27+empty vector cells, pH2AX foci were sustained for 48 hours after radiation with olaparib whereas pH2AX foci attenuated in CAL27+SMAD4 cells after 24 hours (Fig. 5A and C). Similar results were observed in UMSCC1 cells: SMAD4 knockdown cells demonstrated prolonged pH2AX foci compared with control cells 48-hours post-RT with continuous olaparib (Fig. 5B and D). Importantly, irradiated SMAD4-deficient cells had greater levels of pH2AX foci than SMAD4-expressing cells at 48 hours when treated with olaparib (Fig. 5C and D). Similarly, DNA damage visualized by comet assay revealed that both CAL27 and UMSCC1 with no/low SMAD4 had more DNA damage than their isogenic control cells in response to RT or RT plus olaparib (Supplementary Fig. S6).

To determine whether accumulated DNA damage is sufficient to trigger apoptosis, we performed Western blot analysis for PARP cleavage (cPARP) using our isogenic cells treated ± 8 Gy and ± 10 μmol/L olaparib. Full-length PARP protein was not changed in any cell line in response to any treatment (Fig. 5E and F). In CAL27+empty vector cells, cPARP was detectable in response to either olaparib or radiation, and cPARP was more prominent in cells with dual olaparib and radiation treatment (Fig. 5E). Strikingly, treatment-induced cPARP was attenuated by SMAD4 restoration in CAL27 (Fig. 5E). In UMSCC1+shCTRL cells, cPARP was detectable after radiation but not olaparib treatment, and became prominent in response to olaparib in combination with radiation (Fig. 5F). In UMSCC1+shSMAD4 cells, cPARP became detectable after olaparib treatment, and was more prominent after radiation or olaparib in combination with radiation (Fig. 5F). These results confirmed that radiation-induced apoptosis was more potent than olaparib alone and olaparib plus radiation increased apoptosis compared with single treatments and that SMAD4 deficiency increased sensitivity to olaparib + RT-induced cell killing.

To assess if our preclinical findings are applicable to patients with HNSCC, we examined pretreatment HNSCCs for their SMAD4 status and extent of DNA damage in specimens collected in the phase I trial of olaparib in combination with RT and cetuximab, a standard-of-care for patients with cisplatin ineligible, locally advanced HNSCC (17). Among 14 patients with available primary HNSCC specimens, 3 had HNSCC with SMAD4 chromosomal loss as determined by the ratio of SMAD4/centromere 18 (CEN18) ratio (9) and showed no or little SMAD4 protein by IHC (Fig. 6A). Three other patients had SMAD4 wild-type HNSCCs with nondetectable levels of SMAD4 protein (Fig. 6A), indicating postgenetic SMAD4 loss as reported previously (11). Five of 6 of these patients showed no evidence of disease at time of death, and the one nonresponder's tumor had a KMT2A mutation and MYC amplification associated with nonresponsiveness (Fig. 6A; ref. 17). Among eight SMAD4-positive HNSCC cases, four showed no evidence of disease (Fig. 6A). SMAD4 protein levels determined by IHC inversely correlated with the number of pH2AX⁺ cells (Fig. 6B and C). Further, most pH2AX⁺ cells were Ki67-negative (Fig. 6B) suggesting that cells with accumulated DNA damage could not enter cell cycle and proliferate.

Heterozygous SMAD4 deletion occurs in 35% to 52% of HNSCCs (8, 9). We have shown that heterozygous SMAD4 deletion reduces SMAD4 expression by 50% and confers haploid insufficiency (11). Our previous study using mouse HNSCCs and human HNSCC cell lines identified the causal role of SMAD4 loss in lowered expression of certain genes in the FANC/BRCA family (11), and our current study reveals this phenomenon in the large HNSCC TCGA dataset. We found that a panel of SMAD4-mutant HNSCC cell lines was sensitive to olaparib, which suggests a predisposition for SMAD4 loss in therapeutic response of “BRCA-like” phenotypes. Beyond correlation, we found that SMAD4-deficient HNSCCs were more responsive than SMAD4-positive HNSCCs to olaparib in combination with RT in vivo, in comparison with tumors derived from isogenic SMAD4-positive HNSCC cells. Although PARP inhibitors have efficacy as a monotherapy in BRCA-mutant ovarian cancer (28), not all SMAD4-mutant HNSCC xenografts were eradicated by olaparib alone. Because tobacco-associated HNSCCs generally have a higher mutational load and a stronger survival ability than many other cancer types (4, 29), HNSCC cells likely require a higher accumulation of DNA damage to induce DNA damage-associated cell death (30). To this end, in the presence of RT, olaparib increased RT-induced killing as seen in a previous in vitro study (31). In SMAD4-positive UMSCC1 HNSCC xenograft tumors, despite an initial yet temporary response to RT alone, xenograft tumors continued to grow in all therapeutic groups. However, when SMAD4 was knocked down in UMSCC1 xenografts, dual treatment with olaparib and RT significantly reduced tumor size.

The differences in therapeutic response between HNSCCs with complete genetic SMAD4 loss (CAL27) and SMAD4 reduction (UMSCC1+shSMAD4) may represent a dose-dependent effect of SMAD4 loss or other genetic alterations among different tumors. The complete abrogation of tumor growth by restoration of SMAD4 in CAL27 cells suggests that tumorigenesis in CAL27 is largely driven by SMAD4 loss. This SMAD4 loss-dependent tumor development shows that SMAD4 is a potent tumor suppressor even after cells have transformed into HNSCC, consistent with a previous report (10). In contrast, UMSCC1 tumorigenesis does not rely on SMAD4 loss as knocking down SMAD4 did not affect tumor growth, but it did affect therapeutic response. While CAL27 cells do not harbor any loss-of-function FANC/BRCA mutations (based on the cancer cell line encyclopedia interrogation), and we cannot exclude the possibility that other genetic alternations in CAL27 tumors determines sensitivities to olaparib with/without RT, overall our data suggest that SMAD4 loss may affect the therapeutic effects of olaparib with RT even in the presence of additional tumor drivers, as is the case with UMSCC1 cells (24).

In vivo, SMAD4-deficient HNSCC xenografts incurred substantial apoptosis in response to all treatments compared with vehicle, consistent with their therapeutic response among all treatment groups. Except for apoptosis in UMSCC1+shSMAD4 tumors, DNA damage and apoptosis in the olaparib + RT group were not significantly higher than each treatment alone at the end of tumor harvest time point. Therefore, reductions in SMAD4-deficient HNSCC tumor volumes by dual olaparib + RT therapy are likely a consequence of early cell death as suggested by our in vitro studies.

In other tumor types, PARP inhibitors confer therapeutic effects via either blocking DNA repair or inducing formation of a PARP-DNA complex (PARP-trapping) to cause cytotoxic effects (32). If the latter mechanism occurs, a therapeutic effect is relatively rapid (33). In our study, in vivo olaparib treatment had a long latency before showing modest therapeutic benefit. Therefore, PARP inhibition appears to cause cell death primarily through disruption of DNA repair (30). It is possible that PARP inhibition alone causes single-strand breaks (SSB), but these are not sufficient to kill cells (30). However, as SSB-damaged cells divide and replicate they accumulate double-strand breaks (DSB), causing cell death later in treatment as we observed in xenograft tumor models. Because tobacco-associated HNSCC cells likely tolerate more extensive DNA damage compared with ovarian cells (3), HNSCCs require additional genotoxic insults (for example, RT-induced damage to reduce cell viability). Consistent with this notion, we found that olaparib combined with radiation was most effective in reducing clonogenic survival and inducing apoptosis in SMAD4-mutant HNSCC cells, and this effect was either weaker in SMAD4-positive HNSCC cells or attenuated by SMAD4 restoration in SMAD4-mutant cells. Therefore, SMAD4 likely plays a protective role in maintaining cell viability when exposed to genotoxic stress (11).

Consistent with decreased clonogenic survival by dual treatment with olaparib and radiation, DNA damage was more severe in SMAD4-deficient cells compared with SMAD4-positive HNSCC cells or attenuated by SMAD4 restoration in SMAD4-mutant cells in vitro. Although the clonogenic survival in response to olaparib with radiation was similar between UMSCC1+shSMAD4 and CAL27 cells, radiation alone was more effective in CAL27 cells than in UMSCC1+shSMAD4 cells. These findings suggest that complete genetic SMAD4 loss has a more profound effect than SMAD4 knockdown on therapeutic response targeting DNA repair. However, it is important to note that additional tumor drivers are also at play in both these cell lines.

A previous report showed that olaparib sensitization of cancer cells to radiation depends on the integrity of homologous recombination defects (31). Therefore, HNSCCs may harbor homologous recombination defects independent of SMAD4 status and such HNSCCs may be equally sensitive to PARP inhibition (seen in two SMAD4 wild-type HNSCC cell lines in Fig. 1D and E). Our current and previous studies revealed that SMAD4 can also be lost at the postgenetic level. Furthermore, SMAD4 loss in HNSCCs generally predicts a poor response to therapies (34), thus SMAD4 wild-type HNSCCs may represent a group of tumors that generally respond well to therapies. Taking these clinical complexities into consideration, our phase I trial of olaparib with RT and cetuximab demonstrates patients responding to therapy with or without SMAD4 loss. Although the number of specimens is too small to draw any correlation, it is noted that 5 of 6 patients who have SMAD4-negative tumors demonstrated no evidence of disease compared with 4 of 8 SMAD4-positive patients; the one progressed SMAD4-negative tumor harbors KMT2A mutation and MYC amplification, molecular alterations that suggest a nonresponding correlation (17). Because the baseline SMAD4 protein level inversely correlated with the extent of accumulated DNA damage in these patient specimens, future studies with larger patient numbers should assess whether the combination of SMAD4 status and/or the extent of DNA damage can be used to correlate the therapeutic response to PARP inhibition in combination with RT.

In summary, our preclinical study provides evidence that SMAD4 loss contributes to the therapeutic response of olaparib combined with RT. Our findings provide insight into the design of a biomarker-driven therapeutic intervention using olaparib in combination with RT to treat patients with HNSCC that will potentially improve clinical outcomes. Future studies will further delineate molecular mechanisms of RT and PARP inhibitor therapy with respect to SMAD4 status, and identify which FANC/BRCA family genes are direct transcriptional targets of SMAD4; hence their downregulation or baseline pH2AX may serve as alternative predictive markers for therapeutic response to DNA damaging agents in HNSCCs.

S.D. Karam reports receiving commercial research grants from AstraZeneca. D. Raben is an employee/paid consultant for AstraZeneca, Nanobiotics, Merck, Regeneron, EMD Serono, and BioLineRX. No potential conflicts of interest were disclosed by other authors.

Conception and design: A.L. Hernandez, C.D. Young, D. Raben, X.-J. Wang

Development of methodology: A.L. Hernandez, C.D. Young, G. Han, X.-J. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.L. Hernandez, C.D. Young, K. Weigel, K. Nolan, B. Frederick, G. Han

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.L. Hernandez, C.D. Young, L. Bian, K. Nolan, G. He, G. Devon Trahan, M.C. Rudolph, K.L. Jones, A.J. Oweida, S.D. Karam, D. Raben, X.-J. Wang

Writing, review, and/or revision of the manuscript: A.L. Hernandez, C.D. Young, K. Weigel, G. Han, M.C. Rudolph, A.J. Oweida, S.D. Karam, D. Raben, X.-J. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.L. Hernandez

Study supervision: D. Raben, X.-J. Wang

This work was supported by NIH grant DE024371 and DE015953 and VA Merit Award 1 I01 BX003232 (to X.-J. Wang). A.L. Hernandez was supported by T32 CA174648. K. Weigel was supported by T32 CA174648 and F32DE027285. D. Raben, S.D. Karam, and A.L. Hernandez were supported by the Marsico Endowment and an anonymous donation for HNSCC research. We thank AstraZeneca for providing olaparib for in vivo studies, Nicole Manning and Lisa DePledge for technical assistance, and Dr. Thomas Carey for advice and an anonymous donation for support of this work. Pamela Garl provided critical proofreading of this manuscript.

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.