CD38-Mediated Immunosuppression as a Mechanism of Tumor Cell Escape from PD-1/PD-L1 Blockade

Although strategies incorporating immune checkpoint inhibition, e.g., PD-1/PD-L1 blockade, are achieving unprecedented success, high rates of resistance still limit their efficacy (1–3). Using Kras/Trp53-mutant genetically engineered murine (KP tumor) and Lewis lung cancer (LLC tumor) models of non–small cell lung cancer (NSCLC), we have previously shown that the response to anti–PD-L1 treatment is dependent upon epigenetic regulation of tumor cell PD-L1 (4, 5). We used these well-established NSCLC tumor models and melanoma tumor models to further study immunotherapy resistance.

CD38 is a member of the ribosyl cyclase family that is widely expressed on the surface of nonhematopoietic cells and diverse immune cells. As an ectozyme, CD38 converts NAD⁺ to ADP-ribose (ADPR) and cADPR, which are essential for the regulation of extracellular metabolites, intracellular Ca²⁺, cell adhesion, and signal transduction pathways (6). The receptor/ligand activity of CD38 has been documented in multiple immune cell types, and the function varies during lymphocyte development, activation, and differentiation (7). However, its potential function on tumor cells has not been fully elaborated.

Here, we report that tumors gain resistance to PD-L1/PD-1 blockade over time, and that CD38 upregulation on tumor cells is induced by all-trans retinoic acid (ATRA) and IFNβ. CD38 expression mediates suppression via adenosine receptor signaling on cytotoxic T cells. CD38 manipulation was sufficient to regulate CD8⁺ T-cell proliferation, antitumor cytokine secretion, and killing capability. Pathologic analysis of lung cancer specimens revealed positive immunohistochemical (IHC) staining for CD38 on tumor cells in 15% to 23% of cases, and bioinformatic analyses of patient data sets of NSCLC and melanoma revealed a strong correlation between CD38 expression and an inflamed microenvironment. To test whether CD38 blockade might be efficacious to counter the resistance, we used combination therapy with anti-CD38, or alternatively with an adenosine receptor antagonist, and anti–PD-L1 in lung cancer animal models and demonstrated therapeutic benefit.

Our previous data demonstrated suppression of tumor growth and metastases with anti–PD-L1 antibody treatment, but the lack of complete durable responses (4, 5) suggested the existence of antecedent or acquired resistance mechanisms that subvert enhanced infiltration of effector T cells. We adopted a parallel approach to discover these additional mechanisms, first by long-term pharmacologic treatment of animals with anti–PD-L1 or anti–PD-1 antibody, and second by testing of syngeneic tumors with PD-L1 knockout (KO). Despite initial suppression of tumors by anti–PD-L1, progressive resistance developed in tumor models over 5 to 7 weeks of treatment (Fig. 1A; Supplementary Fig. S1A). By week 7 for the KP and week 5 for the LLC tumor models, there were no significant differences between the isotype control and the treatment groups (Fig. 1A). Week 5 and 7 samples from the 344SQ tumors were used for mRNA profiling (Supplementary Fig. S2A–S2E), representing the respective time points of greatest observed difference in tumor growth and the point at which the anti–PD-L1–treated samples displayed complete resistance. Using a fold-change cutoff of 2.0, a total of 412 genes with FDR ≤ 0.05 (P < 0.035) were found to be differentially expressed between anti–PD-L1 and isotype treated tumors at week 5 (Supplementary Table S1). Comparing the top 200 differentially expressed genes (100 upregulated and 100 downregulated, as depicted in the volcano plot in Supplementary Fig. S2C) versus the results of gene set enrichment analysis (GSEA) and Ingenuity pathway analysis (IPA) on the entire set of differentially expressed genes, and finally versus results of proteomic analysis by reverse-phase protein array (RPPA), we identified CD38 as the only prominently upregulated gene/protein identified in all analyses (Fig. 1B; Supplementary Fig. S2C–S2M; Supplementary Tables S2–S6). Its expression also temporally occurred by week 5 of anti–PD-L1 antibody treatment, along with consistent mRNA changes of CD38-related genes (6, 8–14), and therefore preceded the observed acquisition of tumor resistance (Fig. 1B). Both qPCR and FACS analyses confirmed that CD38 mRNA and protein levels were significantly increased on anti–PD-L1–resistant tumor cells (Fig. 1B and C; Supplementary Fig. S1B). We found similar resistance and upregulation of CD38 with several KP lung tumor models and the B16 melanoma model treated with either anti–PD-1 or anti–PD-L1 (Supplementary Fig. S3A–S3E).

Because our previous reports and work from other labs emphasize the dominant role of PD-L1 expression on tumor cells in mediating tumor immune escape (refs. 4, 15, 16; Supplementary Fig. S4A and S4B), we also used a genetic approach to block PD-L1–mediated signaling. We generated lung cancer cell lines (LLC-JSP and the KP model 531LN3) and the melanoma cell line B16 with PD-L1 knockout by CRISPR/Cas9 editing and tested them in syngeneic PD-L1 wild-type (WT) or PD-L1 KO mice. Both partial PD-L1 signaling blockade (PD-L1 KO cancer cells implanted in PD-L1 wild-type mice) and complete blockade (PD-L1 KO cancer cells implanted in PD-L1 KO mice) partially suppressed tumor growth in a CD8⁺ T cell–dependent manner (Supplementary Figs. S4C–4F, and S5), but resulted in ∼4- to 6-fold CD38 upregulation versus the same cells grown in vitro (Fig. 1D and E; Supplementary Fig. S3F). Consistent with these findings, anti–PD-L1 antibody treatment in the autochthonous KP model over 12 weeks showed no durable effect on tumor growth or animal survival, but we observed a significant increase in CD38 on tumor cells in the PD-L1 treatment group (Fig. 1F; Supplementary Figs. S1C–S1D). The consistency of the results between pharmacologic and genetic blockade of PD-1/PD-L1 in syngeneic and autochthonous models of lung cancer and melanoma indicated that CD38 could represent an important pathway in the development of resistance.

To investigate how CD38 is upregulated on tumor cells, we tested cocultures of tumor cells with activated CD8⁺ T cells and found a significant increase of CD38 mRNA and protein (Fig. 1G), which was further enhanced by the addition of anti–PD-L1 and similar to the upregulation observed in tumors (Fig. 1D and E; Supplementary Fig. S3). Altogether, the data suggest that the activated T cells in the inflammatory tumor microenvironment stimulate CD38 expression. This finding prompted us to explore the potential mechanism(s) of CD38 upregulation. Prior literature suggests that CD38 is regulated by several soluble factors that may be present in the tumor microenvironment, including ATRA and IFNβ (17–20). Analysis of the metabolites in anti–PD-L1 treated or PD-L1 KO tumors demonstrated an enrichment of ATRA and an increase in the mRNA for RBP4 and STRA6 that regulate cellular retinol uptake (ref. 21; Fig. 1H and I; Supplementary Fig. S6A–S6B). When human or murine lung cancer lines expressing retinoic acid receptor alpha (RARα) were treated with ATRA for 3 days, CD38 was upregulated in a dose-dependent manner (Fig. 1J and K; Supplementary Fig. S6C). In syngeneic animal tumor models, CD38 on tumor cells was significantly upregulated after 2 weeks of ATRA treatment versus vehicle control, whereas treatment with the RARα antagonist BMS195614 inhibited the CD38 upregulation (Fig. 1L). In addition, we used the tumor lysates to perform ELISA-based assays and found a significant increase of IFNβ in anti–PD-L1-treated tumors (Supplementary Fig. S7A). Upon culturing with IFNβ for 3 days, surface CD38 was significantly increased on multiple cancer lines (Supplementary Fig. S7B). When KP-derived 344SQ tumor-bearing mice were treated with anti–PD-L1 and mRNA profiling was performed by NanoString, we observed an increase of IRF1, which was confirmed by qPCR for IRF1, IRF2, and IRF3 (Supplementary Fig. S8). IRF1 is a transcription factor and tumor suppressor involved in cell growth regulation and immune responses that is induced by ATRA and is essential for the induced expression of IFNβ (22–25). This provides a connection between the independent observations that ATRA and IFNβ are upregulated and produce upregulation of CD38 expression on tumor cells. IFNγ, TNFα, IL2, and IL1β are potent antitumor cytokines also documented to induce CD38 in other cell types. We tested if these cytokines modulate CD38 expression in our models, but did not observe effects on CD38 expression in two lung cancer models (Supplementary Fig. S9). Taken together, PD-1/PD-L1 blockade results in an infiltration of activated T cells and inflammatory changes that lead to ATRA and IFNβ-mediated CD38 upregulation.

Despite the CD8⁺ T cell–dependent effect of anti–PD-L1 antibody during the first 2- to 5-week treatment period (Fig. 2A), which we previously published (4), over time the treatment group showed reduced CD8⁺ T-cell infiltration into tumors, accompanied by a decrease in CD44^hiCD62L^lo memory (48.6% vs. 28.4%) and Ki67⁺ proliferative CD8⁺ T cells (18.8% vs. 6.46%), and an increase in exhausted CD8⁺ T cells (PD-1⁺LAG3⁺ and PD-1⁺TIM3⁺; Fig. 2B; Supplementary Fig. S10). The initial antitumor efficacy of short-term anti–PD-L1 antibody treatment and the temporal CD38 upregulation on resistant tumor cells caused by the infiltration of activated T cells suggests that simultaneous blockade of CD38 and PD-L1 might be required for tumor rejection. We tested this by sorting the PD-L1^KO531LN3 cells or PD-L1^KOLLC-JSP cells for high and negative CD38 surface staining and found that the PD-L1^KOCD38⁻ cells did not form tumors upon implantation into immune competent mice, whereas PD-L1^KOCD38^high cancer cells resulted in large primary tumors, numerous metastases, and rapidly aggressive disease (Fig. 2C; Supplementary Fig. S11A–S11F). To test the effect of CD38 on cancer cell growth, we measured in vitro growth rate and cell cycle, but found no difference between PD-L1^KOCD38⁻ and PD-L1^KOCD38^hi cancer cells (Supplementary Figs. S11G–S11H). By contrast, when CD8⁺ T cells were depleted, PD-L1^KOCD38⁻ cancer cells developed tumors and metastasized whereas PD-L1^KOCD38^hi cancer cells developed tumors similarly in both groups (Fig. 2D; Supplementary Fig. S12A). When PD-L1^KOCD38⁻ and PD-L1^KOCD38^hi cancer cells were injected into RAG2^−/− mice, both tumor cell types formed tumors without significant differences (Supplementary Fig. S12B). Additionally, the development of resistance to the anti–PD-L1 treatment was inhibited by adoptive transfer of CD8⁺ T cells (Fig. 2E). Collectively, these data indicate that CD38 promotes tumor progression via the suppression of CD8⁺ T-cell function.

To further determine whether manipulating CD38 is sufficient to suppress CD8⁺ T-cell function and control tumor, we used KP and LLC lung cancer lines to generate models with CD38 knockdown (KD) or constitutive expression and confirmed the CD38 expression levels by qRT-PCR, western blotting, and FACS analysis (Fig. 2F and G; Supplementary Fig. S13A, S13B, and S13G). The effect of these cancer cells on T-cell activity was first tested in the coculture assay where isogenic tumor cells with different levels of CD38 expression were cultured with specific CD8⁺ T cells. T-cell proliferation was measured by CFSE dilution, antitumor cytokine secretion by CD8⁺ T cells was assayed for IFNγ and TNFα, and tumor cell killing capacity was determined. Consistent with the in vivo suppressive effects of CD38 on CD8⁺ T-cell function, the in vitro results provide direct evidence of CD38 inhibition on CD8⁺ T-cell function (Fig. 2H–M; Supplementary Fig. S13C–S13E and 13H–S13J). In the KP-derived lung cancer line 531LN3 we observed similar results with coculture assays using the sorted populations for PD-L1^WTCD38^hi, PD-L1^WTCD38⁻, PD-L1^KOCD38^hi, and PD-L1^KOCD38⁻ (Fig. 2N–O).

We further observed that the growth of CD38 KD tumors (344SQ-shCD38) was slowed, with significant reduction in primary tumor size and lung metastatic lesions compared with control 344SQ_scr tumors (Fig. 3A). In contrast, 344SQ tumors with constitutive CD38 overexpression (344SQ_CD38) grew faster and produced larger primary tumors and more lung metastases than the vector control (344SQ_vector; Fig. 3B). In addition, the tumor microenvironment of each model displayed distinct and consistent immune repertoire changes upon genetic manipulation of CD38 expression on tumor cells. We observed significantly lower levels of total CD8⁺ T and IFNγ⁺CD8⁺ T-cell infiltrates, and higher percentages of exhausted PD-1⁺TIM3⁺CD8⁺ T cells in CD38-expressing tumors (Fig. 3C–H). Similar results of tumor growth and immune cell profiling were obtained in C57BL/6 animals with LLC-JSP tumors and in 129/Sv animal model with KP-derived 531LN3 tumors (Fig. 3I; Supplementary Fig. S13F and S13K; Table 1). To exclude that these observations resulted from an impact of tumor size on immune phenotype, we tested adjusted cancer cell numbers of CD38 KD, CD38 WT, and CD38 overexpression (OE) and chose tumors of similar size to analyze the CD8⁺ T-cell infiltration and their function. The data indicate that CD38 has a significant impact on CD8⁺ T-cell function regardless of tumor size or growth rate (Supplementary Fig. S14).

Our data demonstrate that CD38-expressing tumor cells impair CD8⁺ T-cell function. Based on the previously reported enzymatic functions of CD38 as part of an ectoenzyme complex that plays an important role in adenosine production (6), we used mass spectrometry analysis of tumor samples and found a ∼2.5- to 6-fold increase in adenosine concentration in anti–PD-L1 versus control-treated tumors (Fig. 3J). Because adenosine can suppress T-cell function in the tumor microenvironment (6, 26–29), we focused on the effects of adenosine for further study. We first compared adenosine concentrations in the supernatants of three different tumor cell cultures with models that had CD38 knockdown (KD), wild-type expression (WT), or constitutive CD38 expression (OE). In all three tumor models, CD38 expression level strongly correlated with increased adenosine (Fig. 3K), whereas adenosine production was blocked with anti-CD38 antibody (Fig. 3L), establishing a causal association between CD38 and adenosine. Previous reports suggested that inhibition of CD8⁺ T-cell function by adenosine occurs through interaction with adenosine receptors ADORA2a and ADORA2b (26). We next challenged mice with different cancer lines and sorted the CD8⁺ tumor-infiltrating lymphocyte (TIL) cells to perform qRT-PCR for ADORA1, ADORA2a, and ADORA2b expression. ADORA1 was expressed at low levels, but ADORA2a and ADORA2b were highly expressed on the tumor-infiltrating CD8⁺ T cells (Fig. 3M). We functionally tested whether receptor antagonists could block CD38-mediated T-cell suppression. T-cell coculture assays with the three tumor models demonstrated that the combined adenosine receptor antagonists effectively reversed the suppressive effect of tumor cell CD38 on T-cell proliferation (Fig. 3N), indicating that CD38-mediated production of adenosine inhibits CD8⁺ T-cell proliferation through adenosine receptor signaling on CD8⁺ T cells.

The data from the murine models suggested that increased CD38 expression on tumor cells may represent an escape mechanism from the infiltrating cytotoxic T cells induced by anti–PD-L1/PD-1 therapy. To understand if this is a generalizable phenomenon, we stained for CD38 and performed FACS analysis on a panel of cancer cell lines representing lung cancer, melanoma, breast cancer, and sarcoma. Notably, 12 of 13 murine lines highly express CD38 (Fig. 4A). Additional FACS and western blotting analysis of NSCLC lines derived from a variety of patients with lung cancer showed surface CD38 expression on most of the lines (Fig. 4A; Supplementary Fig. S15A). We next used two independent tissue microarrays of early-stage lung cancer specimens to analyze CD38 expression by immunohistochemical staining [259 specimens from the MD Anderson PROSPECT data set (TMA3) and a separate set of 534 specimens (TMA4)]. We validated and used a monoclonal antibody specifically recognizing CD38 to determine the membranous protein expression only on cancer cells in the TMA specimens (Supplementary Fig. S15B). Of the 259 TMA3 specimens, 209 had qualified staining and 23% exhibited positive staining for CD38 on tumor cells, whereas of the 534 TMA4 specimens, 471 had qualified staining and 15% exhibited positive staining for CD38 on tumor cells (Fig. 4B and F; Supplementary Fig. S15F; Supplementary Tables S7 and S8). We have corresponding total tumor mRNA expression data for 165 samples in the TMA3 cohort and found a broad distribution of CD38 mRNA expression in the samples. Importantly, there was a strong correlation between IHC score of protein levels and mRNA expression (P = 6.13 × 10⁻⁷; Fig. 4C). The fact that some tumors with low tumor cell membrane staining displayed high mRNA levels is probably due to the presence of CD38 on infiltrating cell populations in the tumors.

Given the strong correlation between CD38 mRNA and CD38 protein levels, we turned to other available patient data sets for which only mRNA expression data are available to perform additional analyses, including the lung cancer and melanoma data sets from The Cancer Genome Atlas (TCGA; lung adenocarcinoma, n = 512; lung squamous carcinoma, n = 496; melanoma data set, n = 469), and the MD Anderson BATTLE-2 trial (metastatic lung cancer, n = 144). Ranking of the samples by CD38 mRNA levels revealed relatively high levels in about ∼25% to 30% of cases, with a strong correlation between CD38 expression and a previously described immune inflammatory signature (30) that includes multiple markers of immune-suppressive cell types and known immune checkpoint molecules and cytokines, e.g., FOXP3, CTLA4, PD-1, LAG3, TIM3, PD-L2, HVEM, BTLA, IDO, and CCL2 (Fig. 4D; Supplementary Fig. S15C–S15D; Supplementary Tables S9 and S10). The data from lung cancer and melanoma data sets demonstrate a strong correlation between CD38 expression and a cytolytic T-cell tumor infiltrate (Fig. 4E; Supplementary Fig. S15E), consistent with the animal and in vitro coculture studies (Fig. 1D–H; Supplementary Fig. S3). To further define the immune subsets of untreated tumors based on CD38 and PD-L1 expression, we performed IHC staining of PD-L1 in the same 793 early-stage tumors from the tissue microarrays [TMA3 (259 cases) and TMA4 (534 cases)] and found that 10.2% to 16.7% expressed high levels of both, whereas 5.1% to 6.7% expressed high CD38 but low PD-L1. Roughly 50% of these untreated tumors had high PD-L1 staining and low CD38 (Fig. 4F; Supplementary Fig. S15F; Supplementary Tables S7 and S8). Interestingly, although CD38 was reported as a prognostic biomarker in multiple myeloma (7), and PD-1/PD-L1 treatment is an effective therapy for certain metastatic NSCLC cases, we did not find an association between CD38/PD-L1 expression and overall survival in the early-stage lung cancer cohorts (Supplementary Fig. S16). Similarly, in a small group of patients with metastatic NSCLC (n = 50) who received anti–PD-1 therapy after prior chemotherapy, CD38 and PD-L1 levels on the tumor cells or in the tumor stroma were assessed by IHC in the pretreatment tumor samples. The level of CD38 on tumor or infiltrating cells in the stroma showed a nonsignificant trend toward predicting clinical radiographic response (partial response vs. progressive disease + stable disease) to PD-1 checkpoint blockade, with higher CD38 levels predicting lack of response (Supplementary Fig. S17A–S17C). Subcategorization by both PD-L1 and CD38 levels was not predictive of response (Supplementary Fig. S17D; Supplementary Table S11), but interpretation from these analyses is significantly limited by the small numbers in each group and will require additional investigation.

We further evaluated a melanoma data set where RNA-sequencing (RNA-seq) data are available for tumor biopsy specimens (n = 43 pairs) pretreatment and on-treatment with nivolumab (anti–PD-1) and observed that CD38 expression correlates with the activated CD8⁺ T-cell status of the tumors in both pre- (n = 51) and on-treatment samples (n = 56; Fig. 4G). Additionally, CD38 levels are increased in response to treatment (mean: pretreatment −0.03, posttreatment 0.04, P = 0.019 by paired t test; Fig. 4H). Overall, these findings from more than 2,500 primary or metastatic lung and melanoma tumors demonstrate that CD38 is found at moderate to high levels in a large percentage of tumors and is expressed in tumors with an active immune cell infiltrate and in which multiple immune modulators (e.g., PD-L1) are expressed, and are consistent with CD38 being upregulated as a consequence of the natural or treatment-mediated immune/inflammatory reaction in the tumor microenvironment.

Due to the upregulation of CD38 expression after PD-L1/PD-1 blockade and the subsequent suppressive effect on CD8⁺ T cells, we assessed the therapeutic efficacy of CD38 inhibitors (anti-CD38 antibody or the biological flavonoid Rhein, which is an enzymatic inhibitor; ref. 31) in combination with anti–PD-L1 antibody in tumor models. Concurrent CD38 inhibition with anti-CD38 antibody (clone NIMR-5) and anti–PD-L1 antibody suppressed primary tumor growth and metastases more than either monotherapy alone or isotype treatment (Fig. 5A–C; Supplementary Table S12). When tumor-bearing mice were treated with anti-CD38 antibody, tumor growth was significantly inhibited. However, this inhibition was reversed after CD8⁺ T-cell depletion, demonstrating that CD38 works in a CD8⁺ T cell–dependent manner (Supplementary Fig. S18). Although neither CD38 nor PD-L1 inhibition alone significantly reduced the number of metastatic lung nodules, they did reduce the metastatic tumor size, whereas the combination treatment reduced both metastatic tumor number and metastatic tumor size. Similar results were obtained with combination therapy of anti–PD-L1 antibody and the flavonoid Rhein, which inhibits CD38 enzymatic activity (Fig. 5D; Supplementary Fig. S19A; Supplementary Table S12). Strikingly, combination genetic and pharmacologic blockade using the PD-L1^KO531LN3 cells sorted for high CD38 expression and treated with Rhein completely and rapidly eradicated the tumors (Fig. 5E), similar to the results with the PD-L1^KO531LN3 cells sorted for negative CD38 expression (Fig. 2D). Immune profiling of tumors from each single and combination treatment revealed that more total and activated CD8⁺ T cells, more memory CD8⁺ T cells (CD44^hiCD62L^lo), increased CD4⁺ICOS⁺ T cells, and less infiltration of CD4⁺ regulatory T cells (Treg) and myeloid-derived suppressor cells (MDSC) were observed in the combination therapies with anti-CD38 than in single or control treatments (Fig. 5F; Supplementary Figs. S19B and S20; Table 2). Additionally, in a KP tumor model naturally CD38-deficient and unable to upregulate CD38 expression (307P), we observed that anti–PD-1/PD-L1 monotherapy dramatically inhibited tumor growth or eliminated tumors (Fig. 5G–H). Although therapeutic treatment with anti-CD38 antibody or the enzymatic inhibitor does not distinguish between the effects of tumor cell and host cell CD38, the data demonstrate that systemic treatment has similar efficacy in reprogramming the immune microenvironment of the tumor as genetic manipulation of CD38 on tumor cells and does not engender systemic toxicity or antagonism that compromises the therapeutic effects.

To model how CD38-blocking strategies might be translated into the clinic for patients with disease refractory to anti–PD-1/PD-L1, we tested sequential treatment after the development of resistance to anti–PD-L1 by treating animals with anti-CD38 antibody alone. We observed a substantial inhibition of tumor growth with an associated enhancement of the effector CD8⁺ and CD4⁺ T-cell responses and blunting of the suppressor CD4⁺ Treg and MDSC populations, highlighting that CD38 is an independent factor in treatment-induced resistance (Fig. 5I; Supplementary Fig. S21). Because the antihuman CD38 antibody daratumumab, which is approved for the treatment of patients with multiple myeloma (32), was reported to kill cancer cells through antibody-dependent cell-mediated cytotoxicity (ADCC) and complement-dependent cytotoxicity (CDC; ref. 33), we tested if the antimouse CD38 (NIMR-5) antibody used in our studies works similarly. However, we did not observe ADCC or CDC in the lung cancer models (Supplementary Fig. S22A–S22B). Interestingly, we found that CD38 was internalized after treating lung cancer cells with NIMR-5 (Supplementary Fig. S22C), consistent with the abrogation of adenosine levels upon treatment noted in Fig. 3L. Finally, given our data that CD38 suppresses CD8⁺ T-cell function via adenosine receptor signaling, and the ongoing development of A2R inhibitors for cancer treatment (34), we tested the potential benefits from combination A2R antagonists and anti–PD-L1. We observed significant improvements of antitumor effect with A2R antagonists alone or combined with anti–PD-L1 (Fig. 5J), which confirms the strong role of adenosine signaling on CD8⁺ T-cell function. Overall, the therapeutic studies demonstrate that combination CD38 and PD-L1 blockade substantially reduces primary tumor burden and metastases, that CD38 can be used in combination or in a sequential manner upon the acquisition of resistance, and that targeting CD38 or the downstream adenosine signaling is highly effective.

Immune checkpoint therapies have gained significant attention for their clinical potential to improve durable outcomes for patients with cancer (35, 36). However, only a fraction of patients derive long-term benefit, and extensive efforts are ongoing to understand the underlying mechanisms of response/resistance. Analyses of experimental models and patient tumors have demonstrated that both de novo and acquired immune checkpoint inhibitor resistance arise from a number of tumor cell–intrinsic and –extrinsic mechanisms (3). Many of these mechanisms occur as a dynamic response of the tumor or microenvironment to an effective T-cell infiltration. Herein we found that when animal models were treated for an extended period with anti–PD-1 or anti–PD-L1 therapy, acquired resistance developed after an initial suppression of tumor growth and metastases, independent of the tumor type and strain background. This resistance was observed in response to both pharmacologic and genetic PD-1/PD-L1 blockade in lung cancer and melanoma models, and in each case we found that CD38 was upregulated on tumor cells. We also found associated functional impairment of CD8⁺ T cells after their initial activation and expansion, demonstrating a novel parallel role for CD38 within the microenvironment of T cell–inflamed tumors to drive adaptive immune escape. The data demonstrate that CD38 upregulation after PD-1/PD-L1 blockade is neither genotype- nor histology-dependent, suggesting that CD38-mediated resistance could occur in a broad variety of cancer types.

Because CD38 expression was induced upon PD-1/PD-L1 blockade, we hypothesized that CD38 upregulation results from intratumoral T-cell infiltration and associated changes in the cytokine/metabolite milieu. Indeed, we found a strong correlation between CD8⁺ T cells and CD38 expression in NSCLC and melanoma. Additionally, when tumor cells were cultured in vitro with activated CD8⁺ T cells, CD38 was upregulated. In anti–PD-L1 antibody–treated tumors, we observed increased ATRA levels, which regulate IRF1 and downstream IFNβ production (24, 25), and could recapitulate the effect of ATRA on CD38 upregulation by pharmacologic dosing or antagonism, consistent with previous reports that ATRA and IFNβ are enriched in inflammatory tumors and serve as potent inducers of CD38 (17, 18, 24, 25). This study uncovered a new insight into how ATRA regulates T-cell immunity via CD38 and the production of adenosine. Taken together, our current study emphasizes that the tumor immune microenvironment undergoes an adaptive reprogramming under the continued pressure of PD-L1/PD-1 axis blockade. Either preexistent or upregulated CD38 expression in tumor cells is responsible for this adaptive immune shift in response to anti–PD-1/anti–PD-L1 antibody treatment, and over time the immunosuppressive effect of CD38^hi tumor cells becomes dominant over PD-L1.

CD38 has long been considered an immune molecule because it is expressed on activated B, T, and natural killer cells (7). However, several recent studies have suggested a broader distribution and more complex role, based upon its multifunctional activities (6). For example, a report showed that CD38^hiCD8⁺ T lymphocytes have strong immunosuppressive capabilities. This subset possesses a regulatory potential that could work together with the innate immune response and control immune homeostasis (37). Another group reported that CD38^hi MDSCs possess the capacity to suppress activated T cells and promote tumor growth to a greater degree than CD38^lo MDSCs (11). CD38 has also been associated with functions exerted by Tregs, in which high CD38 expression in FOXP3⁺/CD4⁺ T-cell populations correlates with extremely powerful modulatory properties of CD4⁺ regulatory T lymphocytes (38). Consistent with these findings, when Feng and colleagues evaluated percentages of CD38-expressing Treg subsets from normal donors and patients with myeloma, they found a high expression of CD38 on CD4⁺CD25^hiFOXP3⁺ regulatory T cells, and that targeting CD38 could block this immunosuppressive population (39). In addition, the noncanonical adenosinergic pathway led by CD38/CD203a provides substrates to CD73 and consequently feeds the production of the potent immunosuppressor adenosine, which is normally essential in maintaining tissue homeostasis and preventing an overzealous immune response (7, 40, 41). Morandi and colleagues have shown that primary human melanoma cell lines can take advantage of this ectoenzyme complex to suppress T-cell proliferation through adenosine production (26). Another recent study demonstrated a role for NAD⁺ levels, which are regulated by CD38 on T cells, as an important factor in the reprogramming of intratumoral CD4⁺ cells into hybrid effector Th1/17 cells for enhanced adoptive T-cell therapy (42). In this study, we have identified CD38 as an immune-suppressive molecule expressed on cancer cells allowing their adaptive escape from checkpoint inhibitor–mediated immune attack through the adenosine production pathway, thereby promoting resistance to anti–PD-L1/PD-1 therapy, as shown schematically in the model (Fig. 5K). The emerging evidence suggests a multifaceted immunosuppressive role for CD38 in regulating the tumor immune microenvironment.

The identification of biomarkers that predict response to immune therapy or identify individuals most likely to develop resistance is a key goal in the clinical use of single-agent and combination therapies. Our observations from the clinical data sets of a strong correlation between CD38 expression and an immune inflamed tumor suggest that CD38 might serve as an appropriate marker of adaptive immune resistance to tumor-specific T-cell infiltration rather than as a static constitutive biomarker. Although our animal model data suggest multiple potential translational strategies to overcome immune checkpoint inhibitor resistance mediated by CD38, many questions about how best to incorporate CD38 into immunotherapy strategies will need to be studied in prospective clinical trials of CD38 blockade. Although daratumumab (anti-CD38 monoclonal antibody; Darzalex) is approved by the FDA for the treatment of multiple myeloma and shows pronounced efficacy as a single-agent or combination therapy with an acceptable adverse event profile (32), there are currently no data regarding its efficacy in solid tumors. The mechanism of immune resistance to anti–PD-L1/PD-1 therapy caused by CD38 provides an evident rationale for recruitment of patients with cancer for clinical trials of anti-CD38 in combination with anti–PD-L1/PD-1 to prevent therapy resistance and further enhance antitumor efficacy.

Animal studies were approved by the Institutional Animal Care and Use Committee at MD Anderson Cancer Center. To study primary tumor growth and lung metastases, cancer cells (if not indicated, 0.5 × 10⁶ for LLC-JSP; 1 × 10⁶ for 344SQ; and 2 × 10⁶ for 531LN3) in 100 μL of phosphate-buffered saline (PBS) were injected subcutaneously into the mouse flank. Tumor sizes were calculated using the formula ½(length × width × width) at indicated time points. For lung metastasis measurement, the lungs were removed and immersed in cold PBS, and nodules on the lung surface were counted as described previously (4, 5).

Spontaneous Kras^LA1/+ and Kras^LA1/+/Trp53^R172HΔg/+ mice were bred in our laboratory, and tumor growth was measured by micro-CT scanning. Multiple transverse cross-sectional CT images were provided for analysis by ImageJ. The largest cross-sectional tumor areas were selected for quantification. Tumor diameters were measured with the ImageJ line tool and areas were calculated using the formula for the area of an ellipse divided by 4.

Mice were treated with antibodies (200 μg of anti–PD-L1 per mouse; 250 μg of anti-CD38 per mouse; or combination) and their IgG control via i.p. injection once a week for indicated weeks. Rhein (CD38 inhibitor) was used at 50 mg/kg intraperitoneally per dose once a week for indicated weeks. A total of 200 μg/mouse of anti–PD-1 antibody or an IgG control was intraperitoneally injected into mice twice a week (or 300 μg/mouse once a week) for indicated weeks beginning on day 7 after tumor cells were subcutaneously implanted.

For ATRA and its receptor antagonist treatment, tumor-bearing mice were treated with ATRA (45 μg in 100 μL 1% methylcellulose; oral administration) or RARα antagonist BMS195614 (67 μg in 100 μL 1% methylcellulose; oral administration) once a day for 1 to 2 weeks beginning on day 4 after tumor cells were subcutaneously implanted (1 × 10⁶ cells per mouse).

For the combination treatment of A2R antagonists and anti–PD-L1, 200 μg of anti–PD-L1 antibody or the IgG control was intraperitoneally injected into mice once a week, and 2 mg/kg of SCH 58261 (A2a adenosine receptor antagonist) and 1 mg/kg of PSB 1115 (A2b adenosine receptor antagonist) in 100 μL of carrier solution (15% DMSO, 15% Cremophore EL, 70% H₂O) were intraperitoneally injected every other day, for indicated weeks beginning on day 7 after subcutaneous cell injection. The mice in the control group received both IgG control and carrier solution.

To prepare tumor-specific CD8⁺ T cells, 129/Sv mice were challenged by subcutaneous tumor cell injection with 0.5 × 10⁶ 344SQ or C57BL/6 mice with 0.2 × 10⁶ LLC-JSP for 2 weeks. CD8⁺ T cells were isolated from the tumors, blood, and spleens of these animals, and labeled with CFSE following the kit's instructions (CellTrace CFSE Cell Proliferation Kit, catalog #C34554, Life Technologies). CFSE-labeled CD8⁺ T cells were cocultured with the indicated cells in the presence of anti-CD3 (5 μg/mL) and anti-CD28 (5 μg/mL) for 4 days. T-cell proliferation was quantified using FACS analysis. In some experiments, the cocktail of specific adenosine receptor antagonists (500 nmol/L ADORA1 antagonist PSB36, 1 μmol/L ADORA2a antagonist SCH58261, and 1 μmol/L ADORA2b antagonist PSB1115) was used to block adenosine receptor signaling.

Experimental details regarding TCGA data sets including RNA extraction, mRNA library preparation, sequencing (Illumina HiSeq platform), quality control, data processing, and quantification of gene expression are previously published (43, 44). For the PROSPECT and BATTLE-2 samples, the mRNA was extracted from frozen tumor tissue corresponding to the same specimen from which the formalin-fixed, paraffin-embedded blocks were made. Array-based expression profiling of PROSPECT tumors was performed using the Illumina Human WG-6 v3 BeadChip, according to the manufacturer's protocol. Gene expression data for the PROSPECT data set have been previously deposited in the GEO repository (GSE42127; refs. 30, 45). The raw data files of transcriptomes were analyzed using Bioconductor R packages. Scatter plots were generated through the cBioPortal (46, 47). Gene expression values (mRNA z-Scores; RNA Seq V2 RSEM) for the skin/cutaneous melanoma (SKCM) TCGA cohort (n = 469) were derived from the cBioPortal platform (44, 46, 47). The pre- and on-anti–PD-1 treatment melanoma cohort and RNA-seq analysis with gene-level FPKM values have been published (48), and gene expression levels were derived from the interface provided by the study: http://ioexplorer.org/shinyipop/bms038/, allowing for the generation of a Gene Set Variation Analysis enrichment score.

D.L. Gibbons reports receiving a commercial research grant from Janssen Research and Development. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Chen, Y. Yang, P.L. Lorenzi, E. Dmitrovsky, J.V. Heymach, F.X.-F. Qin, D.L. Gibbons

Development of methodology: L. Chen, Y. Yang, P.L. Lorenzi, J. Rodriguez-Canales, L.A. Byers, E. Dmitrovsky, I.I. Wistuba, F. Xiao-Feng Qin, D.L. Gibbons

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Chen, Y. Yang, B.L. Rodriguez, Y. Li, P.A. Villalobos, X. Liu, L. Tan, J. Fradette, D.H. Peng, C. Ungewiss, J.D. Roybal, F. Skoulidis, W. Peng, B.W. Carter, C.M. Gay, Y. Fan, J. Rodriguez-Canales, M. Kawakami, L.A. Byers, S.E. Woodman, V.A. Papadimitrakopoulou, E. Dmitrovsky, I.I. Wistuba, D.L. Gibbons

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Chen, L. Diao, Y. Yang, X. Yi, B.L. Rodriguez, Y. Li, T. Cascone, P.L. Lorenzi, J.D. Roybal, P. Tong, J. Oba, F. Skoulidis, W. Peng, B.W. Carter, C.M. Gay, C.A. Class, J. Zhu, L.A. Byers, S.E. Woodman, E. Dmitrovsky, J. Wang, I.I. Wistuba, F.X-F. Qin, D.L. Gibbons

Writing, review, and/or revision of the manuscript: L. Chen, Y. Yang, T. Cascone, P.L. Lorenzi, F. Skoulidis, B.W. Carter, L.A. Byers, S.E. Woodman, V.A. Papadimitrakopoulou, S.E Ullrich, I.I. Wistuba, F. Xiao-Feng Qin, D.L. Gibbons

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Chen, Y. Yang, X. Yi, X. Liu, A. Huang, Q. Zhao, D. Peng, J. Fradette, C. Ungewiss, C.M. Gay, M. Kawakami, D.L. Gibbons

Study supervision: L. Chen, F.X.-F. Qin, D.L. Gibbons

This work was supported by P50CA070907-16A1, P50CA070907-18, and the Elsa U. Pardee Foundation to L. Chen; CPRIT RP150405, CPRIT-MIRA RP160652-P3, and NCI R37CA214609-01A1 to D.L. Gibbons; CPRIT RP130397 to P.L. Lorenzi; CA087546 and CA190722 to E. Dmitrovsky; the CAMS Initiative for Innovative Medicine (2016-I2M-1-005) and the National Natural Science Foundation (81590765) to F.X. Qin. The work was also supported by the generous philanthropic contributions to The University of Texas MD Anderson Lung Cancer Moon Shots Program and the MD Anderson Cancer Center Support Grant P30CA01667. We thank the members of our labs for critical discussions and the continued efforts of the lung cancer GEMINI team.