Oxidized Lipoproteins Promote Resistance to Cancer Immunotherapy Independent of Patient Obesity

Checkpoint blockade immunotherapy is a well-established cancer treatment, but not all patients achieve a durable response (1, 2). As a result, the goals for human immunotherapy studies have shifted away from just demonstrating clinical efficacy toward improving outcomes for a much broader range of patients with diverse cancer types. This requires new insight into the variables that dictate therapeutic success and failure. More than a third of Americans are considered obese (3), and because obesity is associated with compromised immunity (4), it was initially predicted that patients with cancer who are obese would experience worse outcomes compared with patients who are leaner. Indeed, for breast cancer, obesity is associated with larger tumors that are refractory to anti-VEGF therapy (5). For patients with bladder cancer treated with Bacillus Calmette-Guerin, progression-free survival (PFS) is reduced in overweight and obese patient cohorts (6). Likewise, in a large retrospective analysis of 1,186 patients with melanoma, those with higher body mass index (BMI) experienced shorter survival even when adjusted for age, sex, and stage of disease (7).

These studies support a link between obesity and poor outcomes, but other clinical studies focusing on checkpoint blockade immunotherapy have produced conflicting results. For example, a retrospective analysis of 2,046 patients with melanoma concluded that survival after immunotherapy was either unaffected or even improved in those with higher BMI (8). Interestingly, the impact of obesity on these patients varied slightly between sexes, as male patients who were obese responded better than males who were not obese, whereas the survival of female patients was independent of BMI. In addition, in a cohort of 250 patients with various cancers receiving either anti-PD1 or anti-PDL1, higher BMI correlated with improved survival over a 2-year period, and this was independent of sex (9). This was attributed to a previously undescribed mechanism observed in obese mice whereby higher levels of serum leptin induced PD1 expression on T cells, making them more responsive to PD1 blockade (9, 10). Although this mechanistic link was not confirmed in human patients with cancer, it was supported by correlative expression data in healthy volunteers. Regardless, patient survival data in these latter two studies support the controversial “obesity paradox,” where the effect of obesity, which was expected to be overwhelmingly negative, turned out to be neutral or even positive. This paradox has been observed in diverse clinical trials from cancer to cardiovascular disease but the existence of the obesity paradox is now being debated, with some suggesting that adherence to the standard BMI scale (often the only indicator of obesity available in retrospective analyses) masks the truly complicated nature of obesity and metabolic health, thereby skewing interpretation of clinical results (11, 12).

Here, we report a retrospective analysis of 278 patients with cancer who were treated with checkpoint blockade immunotherapies targeting either PD1, PDL1, or CTLA4. There was no overall consensus among these patients, as BMI alone did not consistently predict survival outcomes. Analysis of freshly isolated peripheral immune cells and tumor-infiltrating lymphocytes (TIL) from patients with melanoma and breast cancer also showed no relationship between immune cell function and BMI. Although higher leptin levels in the plasma of obese patients was observed, there was no evidence linking PD1 expression to either leptin or BMI. This observation has important clinical implications and implies that the paradigm between obesity, leptin, and PD1 may not be universal.

To explore alternative explanations for the range of clinical outcomes observed in patients receiving checkpoint blockade immunotherapy, we performed lipidomic assessments on plasma from patients with melanoma or breast cancer. These studies uncovered a mechanism of treatment resistance whereby high serum cholesterol and oxidized lipoproteins impaired cellular immunity and protected tumors from apoptosis. Obesity is associated with oxidative stress, resulting in dysregulated lipid metabolism such as increased oxidation of serum lipids and cholesterol-rich low-density lipoproteins (LDL; ref. 13). Our clinical data showed higher levels of cholesterol and LDL were associated with decreased survival of patients treated with anti-PD1/L1. Oxidized LDL (ox-LDL) was associated with impaired T-cell immune responses and also the induction of a cytoprotective stress response in human melanoma tumor cells via increased expression of heme oxygenase-1 (HO-1). In mouse models of melanoma and breast cancer, HO-1 cytoprotection shielded melanoma cells from apoptosis and provided resistance to PD1 checkpoint blockade immunotherapy. These in vivo studies revealed HO-1 as a molecular target that could be exploited for improved checkpoint blockade immunotherapy outcomes.

Together, our data demonstrate that patient BMI alone is a poor indicator of human tumor immunity and is not predictive of patient survival following checkpoint blockade immunotherapy. Rather, our results point toward a more complex relationship between obesity, diet, and serum lipid accumulation that impairs T-cell immunity and simultaneously promotes tumor cell resistance mechanisms that hinder the success of cancer immunotherapy.

Access to data from clinical trial NCT00094653 was obtained through the Bristol-Myers Squibb data-sharing program (https://www.bms.com/researchers-and-partners/clinical-trials-and-research.html). Human blood and tumor tissue samples were obtained from 27 patients with melanoma and 11 patients with breast cancer who received standard-of-care treatment at Saint Louis University Hospital (St. Louis, MO) and provided written informed consent in accordance with the Declaration of Helsinki. Blood from healthy volunteers was collected and analyzed by the Pinto and George labs at Saint Louis University (St. Louis, MO). This analysis was conducted with approval from the Saint Louis University institutional review board (IRB). Data from 30 patients with non-Hodgkin lymphoma treated at the University of Chicago (Chicago, IL) was obtained from Justin Kline following analysis in his laboratory with approval from the University of Chicago IRB. The Gene Expression Profiling Interactive Analysis (GEPIA) database (gepia.cancer-pku.cn) was used to correlate HMOX1 gene experssion in human melanoma tumors with disease-free survival, independent of treatment. All available 458 anonymous patients with melanoma were stratified by HMOX1 expression with a median group cutoff of 50% (229 high and 229 low). Patients with melanoma were also stratified by pathologic disease stage (0–IV) with relative HMOX1 gene expression determined at each stage.

Whole blood was collected in a Beckton Dickinson Vacutainer with 158 USP units sodium heparin (BD Biosciences, catalog no. 36784). Blood was centrifuged at 1,500 × g for 10 minutes at 4°C to separate cells from plasma. Plasma was collected and stored at −80°C. The cellular fraction was resuspended in PBS at three times the original blood volume and overlaid onto Ficoll-Paque PLUS (GE Healthcare Life Sciences, catalog no. 71101900-EH). Peripheral blood mononuclear cells (PBMC) were separated from red blood cells by centrifugation at 2,000 RPM for 30 minutes at 4°C. The PBMC fraction was removed, washed two times in PBS (Invitrogen, catalog no. AM9625), and cryopreserved in 10% DMSO (Thermo Fisher Scientific, catalog no. BP231-1), 40% FBS (Corning, catalog no. 35-011-CV) in RPMI1640 media (Sigma, catalog no. R8758-1L), frozen gradually to −80°C and then stored in liquid nitrogen prior to analysis. Tumor tissue was minced into approximately 2 mm pieces in a 6-well tissue culture plate and digested in collagenase (1 mg/mL in water) and DNAse (0.25 mg/mL in water) for 30 minutes at 37°C. Collagenase (catalog no. C6079) and DNAse (catalog no. DN95) were purchased from Sigma. Tissue pieces were then crushed with a sterile syringe plunger and cells passed through a 40 μM nylon cell strainer (Corning, catalog no. 43750). Cells washed twice in PBS and cryopreserved as described above.

Lipids were extracted from 100 μL of serum using a modified Bligh–Dyer lipid extraction protocol in the presence of heptadecanoyl cholesteryl ester (CE) internal standard, as described previously (14). Lipid extracts were diluted in methanol/chloroform (1/2, vol/vol) and CE species were quantified using electrospray ionization mass spectrometry on a triple quadrupole instrument (Themo Fisher Quantum Ultra) employing shotgun lipidomics (15). CE molecular species were quantified as sodiated adducts in the positive ion mode using neutral loss scanning of 368.5 amu as described previously (16). Total CE concentration was determined by summing up the CE species which were predominantly comprised of linoleoyl and arachidonyl CE. ox-LDL in patient plasma was measured by sandwich ELISA (Mercodia AB; catalog no. 10-1143-01) targeting a conformational epitope that distinguishes it from LDL using a mAb (4E6; ref. 13). Human leptin was measured using an ELISA kit from Crystal Chem (catalog no. 80968) according to the manufacturer's protocol. Total cholesterol, triglycerides, and high-density lipoprotein (HDL) concentrations in human plasma samples were measured on an Abbott Architect ci8200 analyzer in the St. Louis University Hospital Clinical Core Laboratory, a Clinical Laboratory Improvement Amendments–certified, College of American Pathologists–accredited laboratory. Total LDL was calculated using the Friedewald equation [LDL = Total cholesterol − HDL cholesterol − Triglyceride/5].

C57BL/6 mice were purchased from Jackson laboratory (catalog no. 000664). Splenocytes from Pmel transgenic mice (Jackson laboratory, catalog no. 005023) were obtained from Guanyong Peng (Saint Louis University, St. Louis, MO). All mice were housed under specific pathogen–free conditions and used in accordance with protocols approved by the Institutional Animal Care and Use Committee of the Department of Comparative Medicine, Saint Louis University School of Medicine (St. Louis, MO).

The B16-F0 melanoma cell line was purchased from ATCC (catalog no. CRL-6322) in 2019. FBL (Friend virus–induced erythroleukemia of B6 origin) leukemia and TRAMP (transgenic adenocarcinoma of mouse prostate) prostate cancer cell lines were obtained from Phil Greenberg (University of Washington, Seattle, WA) in 2009. The MC38 colon carcinoma cell line was obtained from Guanyong Peng (Saint Louis University, St. Louis, MO). The E0771 breast cancer cell line was obtained from Ratna Ray (Saint Louis University, St. Louis, MO) in 2018. The human melanoma line A375 was purchased from ATCC (catalog no. CRL-1619) in 2019. All tumor lines were confirmed negative for Mycoplasma on February 2, 2020 by PCR using a kit from ATCC (catalog no. 30-1012K) according to the manufacturer's protocol. Tumor cell lines were maintained in DMEM (Thermo Fisher Scientific, catalog no. 11995-065) supplemented with 10% FBS and 1% Pen/Strep (Sigma-Aldrich, catalog no. P0781). Adherent cells were removed with 0.25% trypsin and cells passaged 3–5 times for each experiment. For B16-F0 studies, 1 × 10⁶ tumor cells were injected subcutaneously into both flanks. For E0771 study, 1 × 10⁶ tumor cells were injected in the mammary fat pad. Mice were treated intraperitoneally with 5 mg/kg of anti-PD1 (RMP1-14) from Bio X Cell (catalog no. BE0146) on days 6, 10, and 14 with or without intraperitoneal treatment with an HO-1 inhibitor, OB24 from TOCRIS (catalog no. 6119) at 30 mg/kg on days 3, 6, 8, and 12. Digital calipers were used to measure tumor dimensions and calculate volume in mm² using the equation (L × W²)/2. Spleens and tumor tissues were mechanically disrupted using a sterile 3 mL syringe plunger and single-cell suspensions prepared in DMEM containing 10% FBS and 1% Pen/Strep prior to analysis.

Flow cytometric analysis was performed on an LSR II and an LSR Fortessa FACS analyzer (BD Biosciences) in the Saint Louis University (St. Louis, MO) Flow Cytometry Core Facility. Live lymphocytes were identified as negative for live/dead stain (Thermo Fisher Scientific, catalog no. L34966A) and positive for staining with anti-CD45 (BioLegend, clone HI30, catalog no. 304042). These were segregated into CD4⁺ and CD8⁺ T cells, CD4/8⁻ CD19⁺ B cells and CD56⁺ natural killer (NK) cells after staining with anti-CD8 (clone SK1, catalog no. 344714), anti-CD4 (clone RPA-T4, catalog no. 300526), anti-CD19 (clone HIB19, catalog no. 302206). and anti-CD56 (clone HCD56, catalog no. 318336), all purchased from BioLegend. CD4⁺ T cells were further segregated into conventional FoxP3⁻ and Foxp3⁺ regulatory cells using anti-FoxP3 (clone 259D/C7, catalog no. 562421) from BD Biosciences. Intracellular cytokine staining was performed using the Cytofix/Cytoperm Plus Kit (BD Biosciences, catalog no. BD 554714) according to the manufacturer's instructions following ex vivo stimulation with phorbol 12-myristate 13-acetate and Ionomycin for 4 hours in the presence of GolgiPlug (BD Biosciences, catalog no. 52-2301KZ). Intracellular antibodies, anti-IFNγ (clone B27, catalog no. 506507) and anti-TNF (clone MAb11, catalog no. 502930), granzyme B (clone QA16A02, catalog no. 372208), Perforin (clone dG9, catalog no. 308126), and Ki67 (clone Ki-67, catalog no. 350512) were purchased from BioLegend. Anti–T-bet (clone eBio4B10) was purchased from eBioscience (catalog no. 50-5825-82). Other antibodies specific for human cell surface antigens were anti-CD39 (clone A1, catalog no. 328228) and anti-CD279/PD1 (EH12.2H7, catalog no. 329927) from BioLegend, and anti-CD103 (clone B-Ly7) from Invitrogen (catalog no. 4330509). All analysis on human cells was performed directly after cell thawing multiple patient samples simultaneously to avoid day-to-day variability. Mouse cells were stained with antibodies against CD8 (clone 53-6.7, catalog no. 100714), granzyme B (clone GB11, catalog no. 515403), Ki67 (clone 16A8, catalog no. 652413), and IFNγ (clone: XMG12, catalog no. 505808) from BioLegend. All flow cytometry data were analyzed using FlowJo v.10 software (Tree Star Inc.).

Tumor cells were lysed in 1× lysis buffer from Cell Signaling Technology (catalog no. 9803) augmented with protease and phosphatase inhibitors from Cell Signaling Technology (catalog no. 5872S). Proteins were size separated by electrophoresis on precast 4%–12% Bis-Tris gels (Invitrogen, catalog no. NP0321BOX) and transferred on to nitrocellulose membranes (Invitrogen, catalog no. LC2001). All antibodies were purchased from Cell Signaling Technology. HO-1 protein was detected by anti-mouse HO-1 (E6Z5G, catalog no. 82206S) or anti-human HO-1 (E3F4S, catalog no. 43966S) diluted from the manufacturer's stock 1:1,000 in Tris-buffered saline and 0.1% Tween-20 (Thermo Fisher Scientific, catalog no. CAS9005-64-5). Primary antibody was detected with horseradish peroxidase–linked anti-rabbit IgG (catalog no. 7074) and β-actin detected with a cross-reactive anti-mouse/human (clone 13E5, catalog no. 5125S) diluted 1:5,000. Western blots were visualized by chemiluminescence using ECL reagent from Cell Signaling Technology (catalog no. 6883S) according to the manufacture's protocol. Densitometry was performed using ImageJ software.

Human A375 melanoma tumor cells were pretreated for 4 hours in culture medium with 10 μg/mL of ox-LDL (Kalen Biomedical LLC, catalog no. 7702026) or LDL (Lee Biosolutions Inc, catalog no. 360–25–1) in the presence or absence of the HO-1 inhibitor, OB24 (TOCRIS, catalog no. 6119) at 30 μg/mL. After 4 hours, doxorubicin (Fresenius Kabi, catalog no. 62653) was added to a final concentration of 10 μmol/L and cells were incubated overnight at 37°C. Cells were permeabilized and stained with CaspGlow Fluorescein Active Caspase Staining Kit from Invitrogen (catalog no. 88-7003) and Annexin V (catalog no. A35110) according to the manufacturer's protocol and analyzed by flow cytometry. For mouse tumors, splenic T cells from naïve Pmel mice were processed into a single-cell suspension at 1 × 10⁶ cells/mL, and stimulated with 4 μg/mL anti-CD3 (eBioscience, catalog no. 14-0032-85) and 4 μg/mL anti-CD28 (BioLegend, catalog no. 102102) for 72 hours. B16 melanoma cells were pretreated with 10 μmol/L Hemin (Sigma-Aldrich, catalog no. 51280) or 4 μg/mL of either LDL or ox-LDL for 48 hours. Splenocytes and B16 cells were then cocultured at a 10:1 effector to target ratio for 72 hours. Tumor cell apoptosis was assessed by staining for pan-caspase (CaspGlow Invitrogen, catalog no. 88-7003-4), and Annexin V (Invitrogen, catalog no. A35110) according to the manufacturer's protocol and analyzed by flow cytometry.

Kaplan–Meier curves were used to analyze PFS and overall survival (OS). Comparisons of patient cohorts stratified by BMI, plasma LDL concentration or HMOX1 gene expression was performed with a log-rank (Mantel–Cox) test. Statistical analyses to compare multiple cell populations were performed using an unpaired, two-tailed, nonparametric Mann–Whitney U test (Prism 7.0, GraphPad Software). Linear regression analysis to assess correlation between two parameters was performed by calculating the Pearson value (r) and the corresponding P value for r using Prism 7.0. Exact P values are provided when available.

In 2010, Hodi and colleagues published results from a landmark clinical trial, NCT00094653, showing improved OS among patients with metastatic melanoma receiving ipilimumab (anti-CTLA4) checkpoint blockade immunotherapy (17). We performed a retrospective analysis using these data in which we compared BMI with PFS and OS for the 129 patients treated with ipilimumab monotherapy. When patients were stratified into those with a BMI above 30 (obese) and those with a BMI below 30 (not obese), there was no correlation between PFS and BMI (Supplementary Fig. S1). However, patients in the obese cohort showed a slight increase in OS (Fig. 1A), stemming mostly from improved OS among obese male but not female patients treated with ipilimumab (Supplementary Fig. S2), which is consistent with another report (8). These results suggest that the mechanisms linking obesity and altered responses to immunotherapy are still not well understood.

To determine whether BMI had a greater influence on outcomes after immunotherapies targeting PD1 or PDL1, we performed a second retrospective analysis of 149 patients with various cancer types treated at our institution with either anti-PD1 (nivolumab or pembrolizumab) or anti-PDL1 (atezolimumab or durvalumab; Supplementary Table S1). In contrast to the ipilimumab analysis above, patient OS was independent of BMI (Fig. 1B). Obese female patients had a slightly better response to immunotherapy than leaner women but this difference in OS did not reach the threshold of statistical significance (Supplementary Fig. S2). Survival among men treated with anti-PD1/L1 was essentially identical for the two BMI groups (Supplementary Fig. S2). Overall, these two clinical studies suggest there is no obvious connection between BMI and patient outcomes after checkpoint blockade immunotherapy.

To understand the possible discrepancy between our results and the PD1/leptin paradigm (9), we sought to confirm the link between leptin and PD1 expression in patients with cancer. This could not be directly investigated in any of our retrospective clinical cohorts because biological samples were not accessible. Instead, we initiated a prospective study to measure PD1 expression on PBMCs derived from patients with cancer. PBMCs were isolated from 27 patients with melanoma and 11 patients with breast cancer receiving standard-of-care treatment at Saint Louis University Hospital (St. Louis, MO). Flow cytometric analysis showed relatively low PD1 surface expression on peripheral CD8⁺ T cells from all patients, which was unaffected by donor BMI when compared by mean fluorescence intensity (MFI; Fig. 1C and D) or when the frequency of PD1⁺ CD8⁺ and CD4⁺ T cells was compared with BMI (Supplementary Fig. S3). To confirm that our inability to connect BMI with PD1 expression did not reflect some laboratory or cohort-specific anomaly, we augmented our data with additional samples from 30 patients with non-Hodgkin lymphoma that were not treated with checkpoint inhibitors and 21 healthy volunteers analyzed independently by two different collaborators in their respective laboratories. These analyses showed no correlation between patient BMI and PD1 expression on T cells (Supplementary Fig. S4).

In patients where tumor tissue was available, we observed greater overall expression of PD1 on CD8⁺ TILs compared with PBMCs, but PD1 expression was unrelated to patient BMI (Fig. 1C and E). These results were not due to low leptin levels, which were significantly elevated in the plasma of obese patients and correlated closely with BMI (Fig. 1F), as previously reported by others (18). However, leptin did not associate positively or negatively with expression of PD1 on human CD8⁺ T cells (Fig. 1G). This seemingly controversial finding is not altogether surprising given that other leptin studies have produced highly disparate results, suggesting that leptin can both increase and decrease PD1 expression on T cells (9, 19) and has both positive and negative effects on T-cell function (10, 20–24). Here, we addressed this question in a diverse population of patients with cancer and healthy donors from different institutions, finding no evidence of a link between obesity-associated leptin and expression of PD1 on either peripheral or tumor-infiltrating T cells. Our data suggest that the proposed paradigm between leptin and T-cell suppression via PD-1 is not broadly applicable and therefore does not represent a reliable tool for predicting patient outcomes.

To investigate whether obesity was associated with functional immune suppression, regardless of PD-1 expression levels, we examined peripheral immune responses in patients with melanoma with a range of BMI. PBMCs from patients with melanoma were first examined for frequencies of T cells, B cells, and NK cells but showed no differences among the two cohorts (Supplementary Fig. S5). The functional potential of CD8⁺ T cells was determined by measuring expression the effector molecules perforin and granzyme B and production of the cytokines IFNγ and TNF after 4-hour ex vivo restimulation. We observed a variety of functional responses based on these metrics but none correlated statistically with BMI (Supplementary Fig. S5). Similar assessment of CD4⁺ T cells was consistent with these results (Supplementary Fig. S6), providing further evidence that functional responses by peripheral T cells in patients with cancer operate independently of BMI.

As a metric of human obesity, BMI is quite limited and accounts only for height and weight. The question remains whether different and perhaps more nuanced biometrics associated with obesity, diet, and metabolic health could better predict patient outcomes after treatment with checkpoint blockade immunotherapy. To begin addressing this question, we established a composite score to gauge functional immune potential across patient populations. This immune score was calculated for individual patients by adding the percentage of granzyme B/perforin double-positive CD8⁺ PBMCs plus the percentage of IFNγ/TNF double-positive CD8⁺ PBMCs induced after restimulation. The PBMC data used to generate the immune scores for patients with melanoma and breast cancer are reported in Supplementary Fig. S5. As expected from our earlier results, neither BMI nor plasma leptin levels correlated with this functional immune score (Fig. 2A and B).

Moving beyond BMI and plasma leptin levels, differences in metabolic health have the potential to impact immune function and we anticipated these may be detected by lipidomic assessment. Analysis of plasma samples from these same patients with melanoma and breast cancer revealed an abundance of total cholesterol and LDL associated with a reduced immune score, independent of BMI (Fig. 2C and D). HDL showed a similar trend but did not reach statistical significance (Fig. 2E). Much of what is understood about the complex intersection between immunity and serum lipids has come from the study of atherosclerosis where the immune system plays a complex and dynamic role in disease (25). By the late 1980s, oxidative stress and subsequent ox-LDL were recognized as important factors in regulating plaque formation and immune cell function (26, 27). Later work uncovered a potentially inhibitory influence of cholesterol and ox-LDL on T cells (25, 28, 29), which could explain why LDL levels correlated with low immune scores in our patients. Whether these mechanisms are operative in clinical cancer biology or tumor immunity remains unknown.

To gauge the possible influence of ox-LDL on patient immune score, we initially quantified CE by mass spectrometry, as CE represents a sentinel lipid species known to be induced by ox-LDL (27, 30). Plasma CE concentrations were inversely associated with immune score (Fig. 2F), nearly overlapping with total cholesterol and LDL profiles in the same patients (Fig. 2C and D). Direct quantification of ox-LDL by ELISA confirmed this, revealing a clear inverse correlation with CD8⁺ T-cell function (Fig. 2G). The level of ox-LDL also closely correlated with total cholesterol and total LDL (Fig. 2H), which carries clinical importance as these molecules are commonly assessed in patients and could be employed as reasonable surrogate markers for ox-LDL. Patient BMI did not correlate with concentrations of plasma cholesterol, LDL or ox-LDL (Supplementary Fig. S7). To determine whether these lipidomic profiles were predictive of therapeutic outcomes, we accessed the retrospective OS data from patients treated with anti-PD1/L1 (Fig. 1B). Of the 149 patients, only 30 had serum lipid data available for analysis. Despite this reduced number, when patients were stratified into high and low LDL cohorts (which overlapped with high and low cholesterol), OS was inversely correlated with plasma LDL concentrations (Fig. 2J). This result supports the well-documented associations between cholesterol, tumor immunity, and cancer progression (29, 31–33). The question remains how such changes in serum lipids intersect mechanistically with responses to checkpoint blockade immunotherapy.

The clinical analysis above established a correlation between serum cholesterol (LDL and ox-LDL) and reduced CD8⁺ T-cell immune function. To determine whether LDL and ox-LDL had a direct effect on human T-cell function, PBMCs from 10 healthy donors were stimulated with anti-CD3 for 6 days in the presence or absence of LDL or ox-LDL. Those stimulated in the presence of ox-LDL showed lower frequencies of CD8⁺ T cells whereas LDL had essentially no impact (Fig. 3A). This likely reflected an impairment in proliferation, as detection of Ki67 in CD8⁺ T cells stimulated in the presence of ox-LDL was significantly reduced (Fig. 3A). Reduced proliferation was consistent among the donors, and although responses varied in magnitude, 9 of 10 donors showed lower Ki67 expression among CD8⁺ T cells stimulated in the presence of ox-LDL (Fig. 3B). This loss of proliferative capacity was not observed in CD4⁺ PBMCs (Fig. 3A), suggesting the inhibitory influence of ox-LDL was uniquely potent for CD8⁺ T cells. Compounding the impact of low CD8⁺ T-cell frequencies, ox-LDL also impaired CD8⁺ T-cell function as measured by expression of granzyme B, perforin, and production of IFNγ (Fig. 3C). For a broader assessment of impaired effector function, we added the frequencies of all CD8⁺ T cells expressing high granzyme B, perforin and IFNγ together to create an effector score for all 10 donors. These scores illustrate the extent of CD8⁺ T-cell dysfunction resulting from exposure to ox-LDL during activation (Fig. 3D). These data support our clinical results from patients with cancer (Fig. 2G), suggesting that high serum lipoproteins, and ox-LDL in particular, are immunosuppressive with the potential to provide resistance against immune-based treatment strategies.

High levels of serum LDL, and especially ox-LDL in particular, also have the potential to impact tumor cells directly. Elevated cholesterol, lipids, and ox-LDL stimulate production of reactive oxygen species and increase oxidative stress (34, 35). A major cellular response to oxidative stress is increased expression of the cytoprotective molecule HO-1, encoded by HMOX1, which is also induced in a variety of human cells in response to oxidatively modified lipids (36). Originally recognized for its role in oxidative degradation of heme to biliverdin, HO-1 also promotes cellular survival under conditions of stress (37). This activity is vital for protection of healthy tissues, but when HO-1 expression is increased in cancer, it can render tumor cells more resistant to endogenous immune responses and therapeutic interventions, as we and others have demonstrated (23, 38–40). Indeed, data retrieved from the GEPIA database showed a strong negative correlation between human HO-1 expression and melanoma patient survival, and a positive correlation with advanced disease stage indepdent of treatment (Fig. 4A).

To determine whether HO-1 expression is influenced by LDL and ox-LDL, B16 mouse melanoma cells were maintained under culture conditions with various concentrations of these lipoproteins. After 3 days, expression of HO-1 protein was evident in all samples and increased in the presence of LDL and ox-LDL (Fig. 4B). This was particularly true for the truncated 28 kDa form, which is important for transcriptional activation of genes involved in the response to oxidative stress (41). This response to LDL and ox-LDL was partially recapitulated in A375 human melanoma cells, but only the 32 kDa form was detected here (Fig. 4C). Together, these results demonstrate the potential of LDL and ox-LDL to induce cytoprotective HO-1 in melanoma tumors.

In murine melanoma, HO-1 expression promotes tumor cell proliferation and disease progression (42), suggesting HO-1 may be a mechanism of therapeutic resistance. We observed highly variable expression of HO-1 among different murine tumor cell lines grown in culture (Fig. 4D). To determine whether expression of HO-1 provided melanoma tumors with resistance to immunotherapy, mice with 6-day established subcutaneous B16-F0 tumors were treated with anti-PD1 on days 6, 10, and 14 with or without simultaneous treatment with an HO-1 inhibitor (iHO-1), OB24 (38) on days 3, 6, 8, and 12 (Fig. 4E). We and others have previously shown that B16 melanoma is partially resistant to checkpoint blockade immunotherapy (43, 44), but treatment here with anti-PD1 alone resulted in significantly smaller melanoma tumors by day 16 (Fig. 4F). Although treatment with iHO-1 alone had little impact on tumor size, the combination regimen produced an additive antitumor effect, resulting in smaller tumors compared with those treated with either monotherapy. To test this treatment strategy on tumors with higher endogenous HO-1 expression, female mice bearing E0771 breast tumors in the mammary fat pad were examined. Unlike melanoma, these breast tumors were completely refractory to monotherapy with either anti-PD1 or iHO-1 (Fig. 4G). However, mice treated with the combination had significantly smaller tumors by day 16. These proof-of-principle in vivo studies suggest tumors with high HO-1 expression are more resistance to anti-PD1 immunotherapy but this can be overcome by inhibition of HO-1.

Previous studies have suggested that inhibition of HO-1 may improve immune cell function (39), which could explain the therapeutic efficacy of combination anti-PD1 plus iHO-1. To determine whether this was the case in mice treated with OB24, we examined CD8⁺ TILs from differentially treated mice. Anti-PD1 treatment resulted in higher frequencies of CD8⁺ T cells infiltrating melanoma tumors, but this was true regardless of HO-1 inhibition (Fig. 5A). Functional assessment of CD8⁺ TILs showed similar expression of granzyme B and equivalent proliferation (Ki67) elicited by either anti-PD1 alone or in combination with iHO-1 (Fig. 5B). Pooled responses demonstrated these same trends across multiple experiments assessing granzyme B, IFNγ, or Ki67 expression among CD8⁺ TILs (Fig. 5C). We interpret these data as evidence that systemic HO-1 inhibition, at least with the OB24 reagent, did not impact antitumor immunity directly. Rather, our results imply that the enhanced tumor control observed in combination-treated mice (Fig. 4F and G) likely stemmed from inhibition of the cytoprotective functions of HO-1 within tumor cells, making them more susceptible to immune-mediated killing.

The cytoprotective role of HO-1 is well documented for healthy tissues (45) and has been identified as a potential barrier to treatment of several cancers (38, 40, 46). Because LDL and ox-LDL induced HO-1 expression in human tumors (Fig. 4C), we directly tested whether these serum lipoproteins could protect human melanoma cells from apoptosis. A375 cells were treated with doxorubicin to induce apoptosis overnight along with LDL, ox-LDL, or a known inducer of HO-1 (hemin) in the presence or absence of the iHO-1. Apoptosis was assessed by expression of active caspase and Annexin V staining. Treatment with doxorubicin induced approximately 20% of cells to become apoptotic (double-positive) compared with less than 1% of control cells treated with PBS (Fig. 6A). These human melanoma cells were nearly completely shielded from doxorubicin-induced apoptosis when HO-1 was induced by treatment with hemin, indicating that high HO-1 expression provides strong protection from otherwise potent inducers of cell death. Exposure to LDL or ox-LDL also provided a certain level of cytoprotection, reducing the frequency of apoptotic cells by approximately half (Fig. 6B). This cytoprotection was reliant on HO-1 as treatment with iHO-1 brought the frequency of apoptotic cells back to levels observed with doxorubicin alone. To determine whether LDL and ox-LDL protected melanoma cells specifcally from CD8⁺ T cell–mediated killing, B16 tumors were pretreated with LDL or ox-LDL and then cultured with activated tumor-specific Pmel T cells. In the absence of intervention, increased apoptosis was evident in B16 cells exposed to Pmel CD8⁺ T cells (Fig. 6C). However, tumor cells pretreated with LDL, ox-LDL, or hemin were partially resisted to T cell–induced apopotosis. Extrapolation of these data predicts that higher levels of ox-LDL and LDL in patient serum may induce HO-1 expression sufficient to resist treatment-mediated apoptosis, making these tumors more challenging for T cells to kill. Given the association between high HO-1 expression and poor survival of patients with melanoma (Fig. 4A), our results encourage further exploration of HO-1 as a therapeutic target in combination with checkpoint blockade to extend the success of cancer immunotherapy to patients that might otherwise be classified as nonresponders, particularly in patients with high serum cholesterol and LDL.

The U.S. Center for Disease Control and Prevention estimates that nearly 40% of Americans are obese. Among the many health concerns associated with obesity is an increased risk of developing certain types of cancer such as esophageal, colon, renal, and endometrial cancers (47). Weaker correlations have also been noted for malignant melanoma, breast cancer, leukemias, and lymphomas (48, 49). With obesity rates rising, extrapolation of these data predicts higher rates of cancer in the future, particularly among individuals who are overweight or obese. Cancer treatment has undergone a revolution over the last decade and we have now entered the era of immunotherapy, but obesity is associated with dysfunction of several immune cells important for killing tumors including dendritic cells, T cells, NK cells, and macrophages (4, 50–52). These observations raise critical questions about the efficacy of cancer immunotherapy in overweight and obese patients, leading to the hypothesis that body mass may be prognostic for patient outcomes after treatment with immunotherapy.

Testing this hypothesis has led to a flurry of clinical reports but achieving consensus has been challenging. Some studies have correlated obesity with poor outcomes (5–7), wheras others have observed either no effect or concluded that obesity actually benefits some patients with cancer receiving immunotherapy, and that sex differences may influence this benefit (8, 9). One proposed mechanism for improved responses to PD1 checkpoint blockade is that obesity-associated leptin directly induces PD1 expression on T cells, making them more amenable to blockade (9). But this was not borne out by our data or other clinical studies where patient survival was statistically indistinguishable for obese and nonobese patients treated with PD1/L1-targeted checkpoint blockade (8, 53). In agreement with the McQuade and colleagues study (8), we did observe a slight survival advantage in male obese patients with melanoma treated with anti-CTLA4. However, our prospective analysis of 68 patients with cancer and 21 healthy donors showed no evidence that BMI or serum leptin correlated with PD-1 expression on T cells, suggesting that leptin-induced PD1 expression cannot explain this outcome. At this point, we are left with several clinical studies that reach different conclusions. When taken together, our interpretation is that BMI as a lone metric of obesity is not able to consistently predict cancer patient outcomes, and in those cases where BMI was associated with modest changes in survival, we still lack a unifying mechanistic explanation.

It is not clear why a correlation between leptin and PD1 was observed in one study but not in others. This discrepancy could stem from differences in the individuals that were examined. For example, in the study by Wang and colleagues (9), PD1 and leptin protein expression was only assessed in a group of 21 healthy female volunteers, whereas our analysis focused on a more diverse group of male and female patients with either melanoma, breast cancer, or lymphoma, as well as male and female healthy volunteers. It is also possible that variations in immune phenotype differed because of unknown influences associated with geography, lifestyle, diet, or metabolic health—independent of obesity. This is not to say that obesity and its complex array of comorbidities have no influence on tumor immunity or responses to checkpoint blockade immunotherapy, but obese patients are quite diverse. We predict that factors related to environment, metabolism, lipidomics, and microbiome are more likely to be the real drivers of heterogeneity in immune function among patients, regardless of BMI. Factors such as these are not reflected in BMI and can range in both prominence and severity among patients with a spectrum of body masses. Thus, a BMI over 30 is neither required nor sufficient to experience these “obesity-associated” comorbidities that have the potential to influence immunotherapy outcomes.

To gain mechanistic insight into why some patients respond better than others to immunotherapy, independent of BMI, we compared plasma lipid species among patients with melanoma and breast cancer. Using a multiparameter composite immune score, we correlated higher levels of total cholesterol, LDL, and ox-LDL with reduced CD8⁺ T-cell effector function. In support of this, we observed a direct inhibitory effect of ox-LDL on human CD8⁺ T-cell responses after in vitro stimulation, suggesting that cholesterol, LDL, and ox-LDL have the potential impair responses to immunotherapy. Indeed, in a limited cohort of patients with cancer treated with anti-PD1/L1 immunotherapy, we correlated higher concentrations of plasma LDL with reduced OS.

In addition to having an impact on immunity, oxidized lipids can also bring about changes to tumor cells directly. We hypothesized that oxidative stress in the tumor microenvironment and the presence of oxidized lipids may fortify tumor cells by triggering protective mechanisms that insulate them from potential harm mediated by the immune system, thereby offering resistance to immunotherapy. HO-1 is a cytoprotective molecule induced under conditions of oxidative stress and in response to oxidized lipids (36). From a clinical perspective, HO-1 expression in melanoma tumors correlated with reduced patient survival and has previously been identified as a possible molecular target to improve existing cancer therapies (38, 46, 54). Our results confirmed that LDL, and in particular ox-LDL, induced HO-1 in both murine and human melanoma tumors, protecting these tumor cells from apoptosis. Furthermore, we established HO-1 expression as a mechanism of resistance to immunotherapy in mouse models of melanoma and breast cancer, where HO-1 could be targeted in vivo to improve anti-PD1 efficacy. Contrary to our initial prediction, inhibition of HO-1 did not alter antitumor immunity. Instead, our data suggest that inhibiting HO-1 interferes with a tumor cell–specific cytoprotective program, making tumor cells more susceptable to immune-mediated killing.

In conclusion, our study provides evidence that obesity, as defined by BMI, does not accurately predict patient outcomes following treatment with checkpoint blockade immunotherapy for cancer. Our results raise questions about the proposed mechanism linking increased PD1 expression to elevated leptin in obese individuals. Rather, our study of cancer patient immune cell function and corresponding lipid profiles provide insight into how oxidized lipids affect T-cell function while also driving a cellular stress response in mouse and human tumors. This response relied on increased expression of the cytoprotective enzyme HO-1, which endowed tumors with the ability to resist immune-mediated apoptosis. Our study also exposed HO-1 as a key vulnerability that may be exploited therapeutically to weaken tumor cells with some degree of specificity. Thus, HO-1 represents a promising molecular target for combination treatment strategies to improve the efficacy of cancer immunotherapy.

J. Kline reports grants from Merck during the conduct of the study, as well as personal fees from Kite/Gilead, Seattle Genetics, MorphoSys, Karyopharm, and iTeos outside the submitted work. No disclosures were reported by the authors.

N. Khojandi: Conceptualization, data curation, formal analysis, investigation, methodology, writing–original draft, writing–review and editing. L.M. Kuehm: Data curation, formal analysis, investigation, methodology. A. Piening: Data curation, formal analysis, investigation, methodology. M.J. Donlin: Formal analysis, validation, investigation, methodology. E.C. Hsueh: Resources. T.L. Schwartz: Resources. K. Farrell: Resources. J.M. Richart: Resources. E. Geerling: Data curation, formal analysis, investigation, methodology. A.K. Pinto: Resources, supervision, validation, investigation, methodology. S.L. George: Resources, supervision, validation. C.J. Albert: Formal analysis, investigation, methodology. D.A. Ford: Supervision, validation, investigation, methodology. X. Chen: Data curation, formal analysis, investigation, methodology. J. Kline: Resources, supervision, validation, investigation, methodology. R.M. Teague: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing–original draft, project administration, writing–review and editing.

The authors thank Sherri Koehm and Joy Eslick for technical assistance with flow cytometry, and Sandra Miller and Stephanie Brem for assistance in obtaining patient samples. They also thank Ben Heckert, Debra King, and April Stephens for assistance with retrospective patient data from the SLU Cancer Center.

Research reported in this article was supported by grants from the NIH (S10OD025246) to D.A. Ford, the Alvin J. Siteman Comprehensive Cancer Center (P30 CA091842) and from the NIH NCI (R01 CA238705) to R.M. Teague.

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.