Recurrent HNSCC Harbor an Immunosuppressive Tumor Immune Microenvironment Suggesting Successful Tumor Immune Evasion

Head and neck squamous cell carcinoma (HNSCC) is the sixth most common cancer worldwide accounting for more than 600,000 cases and 380,000 deaths annually (1, 2). In the United States, more than two-thirds of patients are diagnosed at advanced stages of disease leading to poor prognosis with 5-year survival rates ranging between 39% and 65% (3). A major factor contributing to poor prognosis is the development of local recurrent tumors (RT) in up to 60% of patients (4).

Standard-of-care treatment of HNSCC primary tumors (PT) in a locally advanced setting consists of surgical resection of PT and draining lymph nodes in combination with risk-adapted adjuvant radiotherapy, with or without platinum-based chemotherapy. Alternatively, standalone definitive concurrent chemoradiotherapy (CRT) is administered (5, 6). Those treatment options are quite effective in the primary disease setting, but cannot prevent high RT rates (7). Treatment of recurrent HNSCC, which are often not accessible to surgery or radiotherapy, has for a long time been limited to platinum-based doublet chemotherapy in combination with the anti-EGFR antibody cetuximab with a low median overall survival of 10.1 months (8, 9). Recently, the two immunotherapeutic agents nivolumab and pembrolizumab, mAbs targeting the immune checkpoint molecule programmed cell death protein 1 (PD1), have been proven to contribute to longer overall survival in comparison with standard chemotherapy in pretreated recurrent or metastatic HNSCC in the two randomized phase III trials CHECKMATE-141 and KEYNOTE-040 (10, 11). Furthermore, pembrolizumab has shown prolonged overall survival as first line treatment for recurrent or metastatic HNSCC, either in combination with standard therapy or as single-agent therapy for patients with IHC-proven programmed cell death protein ligand 1 (PDL1) overexpression (PDL1 combined positive score: ≥1; ref. 12). Therefore, pembrolizumab and nivolumab have been included into the latest treatment guidelines for recurrent and metastatic HNSCC (13). Nevertheless, relative response rates stay below 15% and benefit to median overall survival of 1.4 to 2.4 months is still poor (14).

Success of immunotherapies and of standard CRT is strongly influenced by the preexisting tumor immune microenvironment (TIME) and its alterations during therapy (15). Despite high variability across different cancer entities, solid tumors such as HNSCC are usually classified into three different TIME types according their prognostic and predictive impact (16–21):

• (i) Hot (inflamed) tumors with high numbers of effector immune cells such as CD8⁺ and CD4⁺ T cells as well as natural killer (NK) cells.

• (ii) Cold (noninflamed) tumors with enrichment of immunosuppressive cytokines, presence of regulatory T cells (Tregs) and myeloid-derived suppressor cells (MDSC) as well as scarcity of CD8⁺, CD4⁺, and NK cells.

• (iii) Immune-excluded tumors with absence of all types of immunocytes.

Best response rates to immune checkpoint inhibitors are observed for patients with an active antitumoral immune response of the first type (22).

Considering high rates of RTs after successful eradication of PTs as well as poor response rates and survival benefits of immune checkpoint inhibitors in the recurrent setting, we hypothesize that RTs have successfully escaped antitumor immune response, which is characterized by a decrease of antitumor immune cells and activity and an increase of protumor and immunosuppressive immune cells. Furthermore, we think that adjuvant CRT contributes to the immunosuppressive TIME of RTs.

To elucidate immunologic changes in recurrent HNSCC, we performed in vivo characterization of the TIME of PTs and corresponding RTs in two independent HNSCC cohorts via IHC and immune-specific gene expression profiling. Furthermore, we compared the immunologic changes with an in vitro model of chemotherapy- and radiotherapy-resistant HNSCC spheroids cocultured with immune cells.

Two large, independent, and clinically well-characterized HNSCC cohorts were used for this study. The discovery cohort (Lübeck cohort) consists of 63 patients with HNSCC, who were treated according to local treatment guidelines at the Department for Otorhinolaryngology of the University Hospital Schleswig-Holstein, Campus Lübeck, Germany, between 2001 and 2016. All patients developed a local RT (RT1) between 4 months to 11 years after clinically or radiologically proven eradication of their PT. A total of 13 patients developed a second RT (RT2). Overall survival data, patient characteristics, and social histories were obtained from the Department for Otorhinolaryngology of the University Hospital Schleswig-Holstein, Campus Lübeck, Germany. Formalin-fixed paraffin-embedded (FFPE) tumor tissues were retrieved from the archives of the Pathology of the University Hospital Schleswig-Holstein, Campus Lübeck, Germany and tissue microarrays (TMA) were constructed from representative tumor areas. Three 0.6 mm cores of each tumor specimen were assembled into TMA blocks. Human papillomavirus (HPV)-status was assessed by p16 IHC staining and HPV-DNA PCR testing.

The second cohort was an independent validation cohort (Bonn cohort) consisting of 237 PTs and 45 unmatched RTs established as a TMA-based cohort with three 0.6 mm cores for each tumor specimen at the Institute of Pathology of the University Hospital Bonn (Bonn, Germany). An overview of the two independent cohorts and the setup of the study is displayed in Supplementary Fig. S1. The study was approved by the Institutional Review Board (IRB) of the University Lübeck (Lübeck, Germany) and the University Hospital of Bonn (Bonn, Germany), which follow the declaration of Helsinki. Included patients signed a hospital-internal general consent form to participate in research and education, which dispensed with additional written consents for this study according to the IRB, as patient data were anonymized at the source.

IHC was performed on 4-μm-thick paraffin-embedded TMA sections as described previously (23). The OptiView DAB IHC Detection Kit was used on a Ventana BenchMark Ultra (Roche) for the following antibodies: CD4 (SP35, Ventana), CD8 (SP57, Ventana), CD20 (L26, Ventana), CD1A (EP3622, Cell Marque), CD68 (KP-1, Ventana), CD56 (MRQ-42, Cell Marque), FOXP3 (236A/E7, Invitrogen), PD1 (NAT105, Cell Marque), PDL1 (E1L3N, Cell Signaling Technology), Tryptase (G3, Cell Marque) CD15 (MMA, Ventana), and DC-LAMP (101E1.01, Novus Biologicals). Quantitative analysis of immune cell infiltration was determined using a semiautomated image analysis software (Definiens Developer XD 2.0, Definiens). For each immune cell subtype infiltration, intensity was measured as count of immune cells per mm² tumor tissue. Intratumoral immune cells as well as immune cells at the tumor margin were considered. An experienced pathologist blinded to clinical information manually quantified the percentages of PDL1-expressing tumor cells. Mean values from up to three TMA cores for each tumor specimen were calculated. On average, 2.6 cores for PTs and RTs and 2.7 cores for RT2 were analyzed. In vitro spheroid experiments, as described below, were analyzed in the same manner using QuPath 0.1.2 (24). IHC double staining against CD20 in red (alkaline-phosphatase-reaction) and ERG (EPR3864, Abcam) in brown (3,3′-Diaminobenzidine) was performed on whole slides and analyzed using QuPath 0.1.2 (24). Representative areas containing tumor and peritumoral stroma were annotated on the digitized slide and an experienced, board-certified pathologist, blinded to the data, counted the number of tertiary lymphoid structures (TLS), defined as CD20⁺ B-lymphocyte aggregates around ERG⁺ venules. Afterward, the number of TLS per mm² tumor was calculated.

RNA from FFPE tissue was extracted using six to eight 8-μm-thick sections on a glass slide. Sections were deparaffinized with Xylene. An experienced pathologist marked representative areas containing only tumor tissue on an hematoxylin and eosin slide, which was then used for microdissection. Representative tumor tissue was scratched from the glass slide using a 11× feather disposable scalpel. RNA isolation was performed using the Maxwell RSC RNA FFPE Kit on the Maxwell RSC Instrument (Promega) according to manufacturer' instructions without using mineral oil because samples have already been deparaffinized. RNA from in vitro experiments was isolated using the RNeasy Mini Kit (Qiagen) as indicated by the manufacturer. RNA concentration and purity were assessed with a NanoDrop 1000 spectrophotometer (NanoDrop Technologies) and the Qubit RNA HS Assay Kit on the Qubit 2.0 fluorometer (Invitrogen).

RNA from 18 PTs and their corresponding RTs from the discovery cohort was used for gene expression profiling via the nCounter PanCancer Immune Profiling Panel (NanoString Technologies) as described previously (25). All patients received CRT between resection of the primary tumor and development of recurrence. Digital data acquisition via the nCounter Digital Analyzer (NanoString Technologies) was performed at the Institute of Pathology of the Hannover Medical School. NSolver 4.0 Analysis Software (NanoString Technologies) as well as R 3.5.0 were used for data analysis. Gene expression analysis of in vitro experiments was performed in the same manner.

FaDu cell line (26) was purchased from the Leibniz Institute DSMZ - German Collection of Microorganisms and Cell Cultures (DSMZ) and was grown in a 5% CO₂ incubator at 37°C and 85% humidity. FaDu cells were maintained in DMEM (Thermo Fisher Scientific) containing 10% FBS (Biochrom), 1% penicillin–streptomycin antibiotics (Thermo Fisher Scientific), and 1% l-Glutamine (Thermo Fisher Scientific).

The cisplatin-resistant cell line FaDu/CR was developed by continuously exposing FaDu cells to increasing doses (0.1–0.7 μg/mL) of cisplatin NeoCorp (Hexal) for over 6 months. Methyl-thiazolyl-tetrazolium bromide (MTT) assay as originally described by Mosmann and colleagues was performed to assess cisplatin resistance of FaDu/CR cells compared with untreated FaDu/WT cells (27). The radiation-resistant cell line FaDu/RR was generously provided by the research group of Professor Anna Dubrovska and tested for radiation resistance by clonogenic assay as described previously (28).

Peripheral blood mononuclear cells (PBMC) were isolated from fresh heparinized blood of healthy donors using Ficoll-Paque Plus (GE Healthcare) density gradient centrifugation and activated as described in Supplementary Fig. S2.

Spheroids were created by the hanging drop method and further processed as displayed in Supplementary Fig. S2. Briefly, 20,000 FaDu/WT cells, FaDu/CR cells, or FaDu/RR cells were suspended in 20 μL of complete culture medium and placed on the lid of a 24-well plate. After 5 days, hanging drops were transferred into agarose-coated wells containing 1 mL of complete culture medium. Activated PBMCs were added. Two healthy donors were used for all PBMC experiments. FaDu/WT and FaDu/CR or FaDu/RR were treated with the same PBMCs of one donor for each experiment. Spheroid immune cell cocultures were incubated for 72 hours. After harvest, spheroids were fixed in 4% formaldehyde and embedded in paraffine. Histologic specimens were analyzed by IHC as described above. Three independent experiments were performed.

Paired two-tailed t test was applied for discrimination of immune cell markers of IHC analysis of the discovery cohort. Welsh t test was used for discrimination of IHC analysis of the validation cohort. Multivariate ANOVA was used to test correlations between clinicopathologic parameters and TIME characteristics in both cohorts. Fisher exact test was applied for clustering of PTs and RTs after CRT in k = 2 cluster via Euclidean distance based on mRNA expression. Unpaired two-tailed t test was performed for discriminating mRNA-based immune cell infiltration scores and mRNA expression scores of immune-related genes between PTs and RTs after CRT as well as for comparing resistance to cisplatin of cell lines in MTT test and to radiotherapy in clonogenic assay. P values less than 0.05 were considered statistically significant. For mRNA-based analysis, P values adjusted for multiple testing by the Benjamini–Hochberg procedure (Q = 0.05) are listed in Supplementary Table S1. IBM SPSS Statistics version 25.0 for Windows (IBM Corp.) and GraphPad Prism version 8.3.0 for Windows (GraphPad Software) were used for statistical analysis.

TIME studies of primary and recurrent HNSCC were based on two independent patient cohorts. Patient and tumor features of the discovery cohort (Luebeck cohort) are listed in Table 1. This cohort consists of 63 patients (oral cavity, oropharynx, hypopharynx, larynx) with available PT, corresponding first recurrence (RT1) and for 13 patients also corresponding second recurrence (RT2). A total of 58 RTs (92%) were locoregional recurrences (52 local RTs, 6 regional lymph node RTs) and 5 RTs (8%) were distant metastases. A total of 51 patients (81%) underwent surgical resection of PT with complete resection in 95.2% of cases. Radiotherapy was performed in 31 (49.2%) patients with an average total radiation dose of 61.7 Gray (Gy; SD = 6.7 Gy). In addition, 20 (64.5%) patients with radiotherapy also received platinum-based chemotherapy. Mean 5-year overall survival rate was 41.4%.

A second independent patient cohort (Bonn cohort) was used as a validation cohort, consisting of 237 PTs and 45 RTs. Details of the second cohort are displayed in Supplementary Table S2.

First, we aimed to define differences between the TIME of primary and recurrent HNSCC by analyzing the number and subtype composition of tumor-infiltrating leukocytes (TIL). The following immune cells were characterized on the basis of IHC marker expressions: CD8⁺ cytotoxic T cells, CD20⁺ B lymphocytes, CD4⁺ Th cells, FOXP3⁺ Tregs, PD1⁺ cells, CD56⁺ NK cells, CD1A⁺ dendritic cells, DC-LAMP+ mature dendritic cells, CD68⁺ macrophages, CD15⁺ neutrophil granulocytes, tryptase+ mast cells, and PDL1 expression.

In the discovery cohort, a general trend toward a decrease of TILs from PTs to first and especially to second corresponding RTs was observed for various lymphocyte subtypes (Fig. 1A and B). The highest change was observed for CD20⁺ B lymphocytes. They showed a subtotal depletion in RTs with an average of 488.5 cells/mm² in PTs and 20.6 cells/mm² in RT2 (P = 0.036). CD8⁺ T cells, as one of the major antitumor immunocyte types, also strongly decreased from 383.4 cells/mm² in PTs to 152 cells/mm² in RT2 (P = 0.045). CD56⁺ NK cells significantly decreased from 54.3 cells/mm² in RT1 to 37.4 cells/mm² in RT2 (P = 0.036). FOXP3⁺ Tregs decreased from an average of 32.2 cells/mm² in PTs to 16.9 cells/mm² in RT2 (P = 0.047). The loss of antitumor CD8⁺ T cells relatively outweighed the absolute loss of FOXP3⁺ Tregs, demonstrated by a trend toward a decreased CD8⁺/FOXP3⁺ T-cell ratio from 142.3 in PTs to 24.2 in RT2 (P = 0.39; Supplementary Fig. S3). The average absolute count of immunosuppressive PD1⁺ cells was not substantially altered between PTs (37.1 cells/mm²), RT1 (44.4 cells/mm²) and RT2 (28.4 cells/mm²). The CD8/PD1 ratio, however, decreased in analogy to the CD8⁺/FOXP3⁺ ratio from 16.6 in PTs to 5.9 in RT2 (P = 0.058). PDL1 expression did not significantly change between PTs, RT1, or RT2.

CD68⁺ macrophages, CD15⁺ neutrophil granulocytes, and tryptase+ mast cells did not show significant changes between PTs, RT1, or RT2 (Fig. 1C and D). Corresponding to previous ratios, CD8⁺/CD15⁺ ratio also decreased from 6.6 in PTs to 2.1 in RT2s (P = 0.3375) and CD8/CD68 ratio from 6.1 in PTs to 4.8 in RT1s and 2.7 in RT2s (P = 0.2932; Supplementary Fig. S3). CD1A⁺ dendritic cells were the only immunocyte subtype, which significantly increased from a mean count of 96.9 cells/mm² in PTs to 149.9 cells/mm² in RT1 (P = 0.013) and 219.6 cells/mm² in RT2 (P = 0.036). DC-LAMP+ mature dendritic cells in contrast decreased from a mean count of 50.5 cells/mm² in PTs to 29.2 cells/mm² in RT1 (P = 0.0023) and 17.1 cells/mm² in RT2 (P = 0.0719). Correspondingly, the CD1A/DC-LAMP ratio significantly increased from 4.6 in PTs to 15.0 in RT1 (P = 0.0013; Supplementary Fig. S3).

To validate our observations of the discovery cohort, we performed similar analyses in our independent validation cohort (Bonn cohort). Here, we could confirm the strong loss of CD20⁺ B lymphocytes with a decrease from an average of 345.9 cells/mm² in PTs to 84.7 cells/mm² in RTs (P < 0.0001) as well as a loss of CD8⁺ T cells from 733.2 cells/mm² in PTs to 347.3 cells/mm² in RTs (P < 0.0001; Fig. 1E). CD1A⁺ dendritic cells were not increased, but slightly decreased from average 235.2 cells/mm² in PTs to 207.1 cells/mm² in RTs (P = 0.50). FOXP3⁺ Tregs tended to decrease from 66 cells/mm² in PTs to 46.7 cells/mm² in RTs (P = 0.14) and the CD8/FOXP3 ratio showed a comparable decrease to the discovery cohort from 85.8 in PTs to 26.3 in RTs (P = 0.062). A display of TIME changes via dot plot diagrams reflecting the distribution of immunocyte types across patients of the two cohorts as well as a display for matched tumor samples only is listed in Supplementary Fig. S4. In addition, a correlation of the immune cell infiltration of PTs to clinicopathologic parameters for the discovery cohort and the validation cohort can be found in Supplementary Table S3A and S3B.

Because adjuvant CRT is known to have a wide range of immune-modulatory effects, we aimed to evaluate its impact on the TIME of RTs with previous CRT (15). To further elucidate the effect of CRT on the TIME, the discovery cohort was therefore divided into two subgroups: (i) Patients with CRT before development of first or second RT [CRT-treated (CRT-t) RTs, n = 35]. (ii) Patients with surgical resection only before development of RT (n = 28).

Our analysis revealed that loss of CD20⁺ B lymphocytes was CRT dependent. In patients with CRT after resection of PT, subtotal depletion of B lymphocytes from an average of 471.3 cells/mm² in CRT-naïve (CRT-n) PTs to 23.5 cells/mm² in corresponding CRT-t RTs was observed (P = 0.0006; Fig. 2A). Patients with surgical resection only showed no significant change in B-lymphocyte counts (432.6 cells/mm² in preoperative PTs to 576.8 cells/mm² in postoperative RTs, P = 0.59; Fig. 2B). While CD8⁺ T cells did not change, CD4⁺ T cells tended to decrease in CRT-t RTs (P = 0.092) and tended to increase in postoperative RTs after surgery only (P = 0.099). An increase of CD1A⁺ dendritic cells and a decrease of the CD8/FOXP3 ratio in RTs was also more pronounced after CRT exposure. DC-LAMP+ mature dendritic cells, on the other hand, had a stronger decrease in postoperative RTs (P = 0.0203) than in CRT-t RTs (P = 0.0653). In addition, CRT-t RTs had a significant decrease of PDL1 expression from 12.4% in CRT-n PTs to 2.9% in CRT-t RTs (P = 0.049).

To elucidate differences in the TIME of PTs and RTs after CRT on transcriptomic level, we performed mRNA expression analysis of 730 immune-related genes for 18 patients from our discovery cohort using nCounter PanCancer Immune Profiling Panel (NanoString Technologies). Only patients from the subgroup with adjuvant CRT after resection of PT were selected for transcriptomic analysis.

Unsupervised hierarchical clustering via Euclidean distance, as shown in Fig. 3A, showed a significant separation of PTs and RTs after CRT, instead of a patient-specific clustering of PTs and their corresponding RTs (k = 2, P = 0.018). The clustering of PTs was based on an overall higher normalized expression score of immune-related genes than in RTs after CRT. This is remarkable, as it demonstrates that the TIME of different PTs and RTs after CRT is more similar than the TIME of the PT and RT of the same patient.

Next, we used NSolver 4.0 Analysis Software (NanoString Technologies) to create immune cell infiltration scores for various immunocyte subtypes based on mRNA expression of marker genes as implemented in the nanostring panel (Supplementary Table S4). Here, we confirmed the decrease of total TILs (P = 0.0004), loss of B lymphocytes/TILs (P < 0.0001) and increase of dendritic cells/TILs (P = 0.0002) from PTs to RTs after CRT on transcriptomic level (Fig. 3B). In addition, we discovered a significant increase of mast cells (P = 0.0057), macrophages (P < 0.0001), and neutrophils (P = 0.018) relative to total TIL count in RTs. Interestingly, all the former cell types belong to the myeloid lineage, which have variable roles in cancer immunity including antitumor, but also protumor inflammation-promoting functions (29).

We also assessed additional genes with significantly differential expression between PTs and RTs after CRT (Fig. 4C). A set of 7 genes was found to be significantly upregulated in RTs: SPP1, HRAS, MAPK1, PSEN1, RELA, ITCH, and PRPF38A. Strongest upregulation was observed for SPP1, as measured by log₂ fold change (log₂FC: 1.52). On the other hand, 160 genes appeared to be significantly downregulated in CRT-t RTs. Downregulated genes were predominantly cytokines, genes from the TNF family and genes involved in cellular adhesion as well as genes important for T-cell, B-lymphocyte, and NK-cell functions and their cytotoxic capacities (Fig. 3C and D). The strongest downregulation was observed for MS4A1 (log₂FC: −2.90), respectively CD20, as well as CD19 (log₂FC: −1.87), CD79A (log₂FC: −1.83), and CD22 (log₂FC: −1.54), which are all components of B-lymphocyte receptor signaling (30). Furthermore, the ligands and receptors CCL19 (log₂FC: −1.91), CCR7 (log₂FC: −1.90), CXCL13 (log₂FC: −1.65), CXCR5 (log₂FC: −1.46), CCL14 (log₂FC: −1.45), and LTB (log₂FC: −1.22), which play crucial roles in the formation of TLS and in chemotaxis of B lymphocytes, as well as ICAM-2 (log₂FC: −0.82) and ICAM-3 (log₂FC: −0.56), which are essential to retain leukocytes in TLS through specialized high endothelial venules (HEV), were also strongly downregulated in RTs after CRT (31).

As the pattern of strongly downregulated genes was mostly associated with B lymphocytes and TLS, we aimed to further define the presence of TLS and their dependency of CRT. An IHC double stain labeling CD20⁺ B lymphocytes (red) and ERG⁺ venules (brown) was used to detect TLS on whole slides of 12 matched preoperative PTs and postoperative RTs without CRT as well as 12 matched CRT-n and CRT-t tumors. CD20⁺ B-lymphocyte aggregates around ERG+ venules as marker for TLS were found in intratumoral fibrous bands and in the peritumoral desmoplastic stroma (Fig. 4A and B). While mean TLS numbers per mm² were significantly lower in CRT-t RTs (0.14 TLS/mm²) than in CRT-n PTs (0.64 TLS/mm², P = 0.0113), only small changes were observed after surgery from an average of 0.36 TLS/mm² in preoperative PTs to 0.23 TLS/mm² in postoperative RTs (P = 0.5548; Fig. 4C). Although the mean TLS number per mm² tended to be slightly lower in the preoperative PTs compared with CRT-n PTs, this difference was not significant (P = 0.1950) and is considered to be within the range of pathophysiologic variations of the TIME across different patients with HNSCC. Relative changes of TLS frequencies in CRT-t RTs normalized to their matched CRT-n PTs in comparison with postoperative RTs normalized to their matched preoperative PTs also demonstrated a trend toward a more pronounced TLS loss in CRT-t RTs (0.32) as compared with postoperative RTs (0.91, P = 0.1010; Fig. 4D).

Because our in vivo studies showed striking changes of the TIME after exposure to CRT, we aimed to further investigate whether tumor cell resistance to cytotoxic therapy plays a role for these immunologic changes.

First, we created a chemotherapy-resistant HNSCC cell line via long-term treatment of the hypopharyngeal squamous cell carcinoma cell line FaDu with cisplatin (FaDu/CR). A radiation-resistant FaDu cell line (Fadu/RR) was provided by the research group of Anna Dubrovska. Cisplatin and radiation resistance in comparison with wildtype FaDus (FaDu/WT) were determined via MTT assay and survival colony formation assay (Fig. 5A and B). As we hypothesized that cytotoxic therapy resistance leads to differences in antitumor immune response, we created an in vitro model of tumor cell spheroids, which can be cocultured with immunocytes, to subsequently analyze quantity and quality of spheroid-infiltrating immunocytes (Supplementary Fig. S2).

Coincubation of FaDu/WT and FaDu/CR spheroids with activated PBMCS of healthy donors showed significant differences in immunocyte infiltrations. Importantly, coincubation of spheroids with immune cells was performed in the absence of cisplatin to avoid direct cytotoxic effects on PBMCs. In contrast to the previously observed lower immune infiltration of CRT-t RTs in vivo, FaDu/CR spheroids were characterized by a significantly higher immune cell infiltrate than FaDu/WT spheroids as indicated by an increase of CD45⁺ leukocytes from an average of 11.8 cells/mm² in FaDu/WT spheroids to 133 cells/mm² in FaDu/R spheroids (P < 0.0001; Fig. 5C and E). Because of previous activation of T cells with anti-CD3 and anti-CD28 incubation of isolated PBMCs, most spheroid infiltrating leukocytes were CD4⁺ Th cells with a mean increase from 5.2 cells/mm² in FaDu/WT spheroids to 33.5 cells/mm² in FaDu/R spheroids (P < 0.0001) or CD8⁺ cytotoxic T cells with a mean increase from 8.1 cells/mm² in FaDu/WT spheroids to 70.8 cells/mm² in FaDu/CR spheroids (P < 0.0001). Moreover, in contrast to the HNSCC cohort, PDL1 expression was significantly increased from a mean relative score of 0.05 in FaDu/WT spheroids to 0.32 in FaDu/CR spheroids (P < 0.0001). Spheroid-infiltrating leukocytes also tended to be higher in FaDu/RR spheroids with the strongest increase for CD8⁺ cytotoxic T cells from 31.3 to 84.8 cells/mm² (P = 0.0557). Analogous to FaDu/CR spheroids PDL1 expression was significantly upregulated in FaDu/RR with a mean relative score of 0.30 in FaDu/WT to 0.40 in FaDu/RR (P < 0.05).

In summary, the in vitro experiments indicated a higher immunogenicity of tumor cells after gaining cisplatin resistance in contrast to the immunosuppressive TIME in CRT-t RTs of the in vivo cohort. Radiation resistance caused a similar, but weaker effect in the in vitro experiments. To still analyze overlapping molecular changes between the cohort and the in vitro model, we also performed mRNA expression analysis of FaDu/WT spheroids and FaDu/CR spheroids. Comparison with PTs and RTs after CRT revealed five genes, which were simultaneously significantly upregulated in CRT-t RTs as well as in FaDu/R spheroids: SPP1, MAPK1, PSEN1, RELA, and ITCH (Fig. 5D). On the other hand, 16 genes were simultaneously downregulated in vivo and in vitro with highest in vitro downregulation of ENG (log₂FC: −2.18) and IL24 (log₂FC: −1.27).

High rates of RTs are still a major factor leading to poor prognosis of HNSCC. Although immune checkpoint inhibitors have been included in the latest treatment guidelines of recurrent HNSCC, response rates, and overall survival benefit stay low (14). As the TIME is one main factor influencing CRT response as well as efficacy of immune checkpoint inhibitors, there is an urgent need to better define the TIME of HNSCC RTs (13, 32).

Comparative immune cell infiltration analyses of PTs and RTs in two independent cohorts revealed major differences. Significant loss of CD8⁺ cytotoxic T cells and CD20⁺ B lymphocytes in RTs as well as general trend toward less immune infiltration of RTs was observed in both cohorts and indicates a shift from antitumor immune response in PTs to immunosuppression in RTs (17, 33). This conclusion is supported by multiple studies demonstrating improved overall survival and local tumor control of HNSCC and other solid malignancies, which are highly infiltrated by CD8⁺ cytotoxic T cells and CD20⁺ B lymphocytes (16, 18, 19). From a therapeutic perspective, a CD8-dominant T-cell infiltration is considered to be one key factor for successful immune checkpoint therapy (22, 32). An additional hint for local immunosuppression in RTs was the increase of CD8/FOXP3 and CD8/PD-1 ratio, as FOXP3⁺ Tregs and PD-1⁺ cells are both known for dampening the antitumoral immune response (14, 34). According to the discovery cohort, the decrease of TILs (CD8⁺, CD20⁺, FOXP3⁺, CD56⁺ TILs) was particularly evident in RT2s. We think that this suggests a stepwise development of an immunosuppressive TIME in HNSCC over the course of repeated recurrence development. In addition, we found a significant increase of CD1A⁺ dendritic cells in RTs in our discovery cohort, which was not confirmed in our validation cohort. Dendritic cells generally form a heterogenous group of different subtypes harboring both antitumor and protumor capacities (35, 36). A selective assessment of DC-LAMP+ mature dendritic cells, in contrast, showed a significant decrease in RTs. DC-LAMP+ dendritic cells support an efficient T cell–mediated antitumor immune response as well as a close interaction with B lymphocytes in the context of TLS and are of prognostic and predictive significance in different tumor entities, including HNSCC (37, 38).

The direct impact of CRT on immunologic functions is known to harbor immune-stimulating as well as immunosuppressive effects (15). Most relevant immune-stimulating effects include attraction and activation of dendritic cells, which can activate tumor antigen-specific CD8⁺ CTLs (39, 40). Nevertheless, CRT also impedes the development of an effective antitumor immune response and can lead to long-term immune dysfunction in HNSCC (41). E.g., CRT kills not only tumor cells, but also immunocytes over the course of treatment. Furthermore, it has been shown to increase the number of circulating Tregs, PD-1–expressing cells, and MDSCs in HNSCC (42, 43).

Consecutively, we wondered, if the deprivation of antitumor immune response in RTs is related to CRT. Therefore, we selectively analyzed the subcohort of patients, who were exposed to adjuvant CRT before development of their RT (CRT-t RT) and compared them with patients, who had no CRT prior to RT development (postOP RTs). Here, the significant decrease of immunocytes with antitumor capacities was reserved to CRT-t RTs, but not observed in postOP RTs. Correspondingly, we assume that the shift toward an immunosuppressive TIME is related to the previous CRT exposure. Most remarkably, CD20⁺ B lymphocytes showed a nearly complete depletion in CRT-t RTs, whereas patients with resection only had no significant changes of CD20⁺ B lymphocytes. Intratumoral B lymphocytes account for up to 25% of all TIL-subtypes across different cancer entities and are mostly associated with prolonged survival (20, 44, 45). Functional mechanisms of the antitumor capacity of B lymphocytes include long-term antigen presentation, which supports formation of CD4⁺ and CD8⁺ memory T cells at the tumor site, tumor-reactive autoantibodies, recruitment of other antitumor leukocytes via cytokine release and direct tumor cell killing via granzyme or TNF-related apoptosis-inducing ligand (TRAIL) signaling (45–47). Some subtypes of B lymphocytes, on the other hand, especially lately described regulatory B cells (Bregs), have pro-tumor effects, such as IL10–mediated suppression of cytotoxic immune responses and TGFß-mediated modulation of CD4⁺ T cells to immunosuppressive Tregs (46, 48).

To further elucidate CRT-associated shift to a more protumor and immunosuppressive TIME in RTs and to better specify depletion of B lymphocytes, we performed mRNA expression analysis of 730 immune-related genes via nanostring profiling. According to unsupervised hierarchical clustering, the TIMEs within the group of PTs or CRT-t RTs were closer related than the TIMEs of a PT and its corresponding CRT-t RT. This observation ties in with recent findings revealing that RTs after CRT often seem to be unrelated to their corresponding PT when compared for copy-number alterations (49).

Subtype-specific immune cell infiltration scores based on mRNA expression of marker genes confirmed the significant decrease of total TILs and B lymphocytes as well as the significant increase of dendritic cells from PTs to RTs. Furthermore, mRNA-based analysis revealed a significant increase of neutrophils, mast cells, and macrophages relative to total TILs in RTs. The later immune cell types are all associated with an immunosuppressive TIME leading to reduced survival in various cancer entities and have also been found to be increased in patients plasma after CRT (29, 42, 50).

In the next step, we evaluated genes, which were significantly differentially expressed between PTs and RTs after CRT. Overall, most genes of the immune panel were downregulated with only a few exclusions demonstrating upregulation in RTs. Corresponding to our previous results, strongest downregulation could be observed for genes, which are involved in B-lymphocyte receptor signaling. Highly significant loss of CD27, which is predominantly known as a marker for memory B cells, in combination with nearly no alteration of CD38, which is known as a marker of early stages of B-cell development and of Bregs, indicated that depletion affects particularly long-lasting memory B cells (51). On the other hand, IL10, a major protumor mediator of Bregs, was also significantly downregulated in CRT-t RTs (46). In addition, we found that many of the highly downregulated genes are crucially involved in the initiation and growth of TLSs as well as in the retention of leukocytes to TLS through specialized HEVs (31, 52, 53). TLSs are ectopic lymphoid formations, which occur at sites of chronic inflammation, as in autoimmunity disease or cancer, and share common characteristics with secondary lymphoid organs (31). They mostly develop at the tumor margin and recent studies could demonstrate that they essentially contribute to a long-lasting and tumor-specific immune response by an interplay of follicular dendritic cells, CD8⁺ and CD4⁺ T cells, as well as CD20⁺ B lymphocytes (31, 52). Thus, the existence of TLS at the tumor sites is associated with prolonged survival and local tumor control in various entities including HNSCC (31, 54). In our subsequent analyses, we could confirm a significant decrease of TLS in RTs after CRT using IHC-based whole slide evaluations. This decrease was insubstantial in postoperative RTs, who had not yet received CRT.

As we could reveal that an immunosuppressed TIME of RTs is highly associated with previous CRT treatment of the PT, we wanted to further investigate the role of cytotoxic therapy on the TIME in vitro. Our aim was to examine, if tumor cell resistance to cancer therapies per se leads to changes in the T cell–mediated cytotoxic immune response as the common final path of antitumor immune response. Our cell culture studies focused on chemotherapy resistance to cisplatin and resistance to radiation, which regularly occurs in HNSCC RTs. We used a three-dimensional spheroid/immune cell coculture model of wildtype FaDu cells (FaDu/WT) and compared it with cisplatin-resistant FaDu cells (FaDu/CR) or radiation-resistant FaDu cells (FaDu/RR) coincubated with PBMCs of healthy donors. Here, we observed significantly higher infiltration with CD45⁺ leukocytes, which were predominantly CD4⁺ and CD8⁺ T cells, and higher levels of PDL1 expression in FaDu/CR spheroids. FaDu/RR also had a higher PDL1 expression and tended to have higher lymphocyte infiltrates, in particular, CD8⁺ T cells. These results indicate overall higher immunogenicity of FaDu spheroids resistant to cytotoxic therapies. We assume that the higher immune infiltration in resistant FaDu cells can be explained by their accumulation of genetic alterations elicited by DNA damages. Although we did not know the HLA subtype of our healthy PBMC donors and therefore have to assume that it might oftentimes not have matched the HLA subtype of FaDu cells (HLA-A*01:01), this systematic error does not affect our results as we analyzed relative differences. In summary, our cell culture results form a contrast to our in vivo results with an immunosuppressive TIME in CRT-t RTs. Our in vitro results support the stimulating effects of cisplatin and radiation on antitumor immune response and exclude a direct immunosuppressive effect of tumor cells on immunocytes in vitro. In addition, the spheroid/immunocyte coculture make clear that the multidimensional impact of cytotoxic therapy is more complex in patients with HNSCC. The local and systemic immune response as well as the role of stromal components, which interact with the local immune response and are highly affected by radiotherapy, probably lead to an immunosuppressed TIME in RTs in vivo.

Because resistance to cytotoxic therapy and changes in immune cell infiltration were more pronounced in FaDu/CR spheroids, these were used for further molecular analyses. Our aim was to discover potential relevant therapeutic targets contributing to cisplatin resistance and RT development by elaborating specific genes showing significant mRNA upregulation or downregulation in vivo after CRT as well as in vitro after cisplatin exposure. Mutually upregulated genes in vivo and in vitro include SPP1, MAPK1, and PSEN1. SPP1, also known as osteopontin, represents a cytokine, which is mostly secreted by tumor or stroma cells and substantially modulates the TIME by attracting protumor myeloid cells as well as facilitating tumor progression and metastasis (55). Consequently, its upregulation might be a major functional mechanism leading to the increase of myeloid cells in CRT-t RTs in vivo. Lou and colleagues furthermore showed that SPP1 is frequently upregulated in cisplatin-resistant HNSCC and leads to poor prognosis (56). RAS/MAPK activation has recently been associated with reduced TILs in patients with breast cancer (57) and with disease relapse as well as therapeutic resistance in HNSCC (58).

In summary, we could for the first time demonstrate that HNSCC RTs are characterized by an immunosuppressive TIME suggesting tumor immune evasion as one major factor promoting tumor recurrence. In addition, we elaborated that an immunosuppressive TIME predominantly occurs in CRT-t RTs, but not in RTs after surgical resection only. CD20⁺ B lymphocytes as well as chemokines involved in B-lymphocyte chemotaxis and formation of tertiary lymphoid structures seem to be particularly subjected to the immunosuppressive TIME in CRT-t RTs. Finally, our mRNA expression analyses revealed new potential targets, as SPP1, which could help minimize the immunosuppressive effect of CRT. Our results are of particular therapeutic relevance, considering the rising application of immune checkpoint inhibitors in RTs on the one hand and the immunosuppressive TIME in RTs on the other hand. Our studies support the inclusion of immunotherapies at an early stage of treatment to fully exploit the antitumor immune activity and to avoid therapy resistance due to an immunosuppressive TIME.

No disclosures were reported.

C. Watermann: Formal analysis, validation, investigation, writing-original draft, writing-review and editing. H. Pasternack: Data curation, software, investigation, visualization, methodology. C. Idel: Conceptualization, resources, data curation, project administration, writing-review and editing. J. Ribbat-Idel: Conceptualization, resources, writing-review and editing. J. Brägelmann: Data curation, software, formal analysis, writing-review and editing. P. Kuppler: Resources, data curation, formal analysis. A. Offermann: Data curation, methodology, writing-review and editing. D. Jonigk: Resources, data curation, software, writing-review and editing. M.P. Kühnel: Resources, software, writing-review and editing. A. Schröck: Conceptualization, resources, project administration. E. Dreyer: Data curation, methodology. C. Rosero: Formal analysis, validation. J. Nathansen: Resources, data curation, validation. A. Dubrovska: Resources, validation, writing-review and editing. L. Tharun: Resources, data curation, writing-review and editing. J. Kirfel: Conceptualization, data curation, software, supervision, writing-review and editing. B. Wollenberg: Conceptualization, resources, writing-review and editing. S. Perner: Conceptualization, resources, funding acquisition, project administration, writing-review and editing. R. Krupar: Conceptualization, resources, data curation, software, formal analysis, supervision, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

J. Brägelmann was supported by the Else Kröner-Fresenius Stiftung Memorial Grant (2018_EKMS.35) and the Deutsche Krebshilfe through a Mildred Scheel Nachwuchszentrum Grant (70113307).

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.