Background
Immunotherapy has revolutionized the treatment regimens for various cancer types, leading to improved clinical responses in otherwise untreatable advanced cancers [
1]. Observations showing accumulation of tumor-infiltrating lymphocytes (TILs) within the tumor microenvironment (TME), as well as work highlighting the efficacy of immune checkpoint inhibitors (CPIs), have sparked interest in the further development of these approaches. Studies have focused on the development of CPIs, including cytotoxic T-lymphocyte-associated protein 4 (CTLA-4) [
2,
3] as well as programmed cell death 1 (PD-1) receptor and its ligands programmed death ligand 1 (PD-L1) and PD-L2 [
4‐
6]. PD-1 is found on cytotoxic T cells and T-regulatory cells and is expressed when T cells become activated in response to inflammation or infection in peripheral tissues [
7,
8]. Binding of the PD-1 ligand to its receptor inactivates the T cell, limiting the immune response to the stimuli, thereby causing immune suppression [
7,
8]. Cancer cells, however, induce PD-1 L expression, enhancing the immunosuppressive action of this pathway, ultimately allowing them to “hide” from natural immune attack [
7,
8]. Anti-PD-1/PD-L1 therapies disrupt this pathway by preventing these interactions, leaving activated cytotoxic T cells available to attack the cancer cells [
7,
8]. In triple-negative breast cancer (TNBC), a minority of patients benefit from these approaches, and further studies are urgently needed, especially those designed to evaluate combinatorial therapies.
The recent evolution of these therapeutic strategies (i.e., allowing the immune system to identify neoplastic growth in order to prevent carcinogenesis and eliminate cancer cells) has led to the urgent need for having available a range of appropriate small-animal models that may serve in testing these interactions [
9,
10]. To this end, mouse models injected with human CD34
+ hematopoietic stem cells (HSCs; “humanized” mice) are currently commercially available for studies in cancer, infectious diseases, and gene therapy, among others. However, these models remain relatively expensive, beyond the means of most academic laboratories, especially when used in large-scale studies.
Important advances have been made in the recent years in establishing mouse models to be used in cancer-related studies, including patient-derived xenografts (PDXs). PDXs, by conserving the characteristic of the human primary tumor, are useful for addressing critical questions regarding tumor biology and response to newly developed therapeutic concepts [
11,
12]. In contrast to cell lines used for in vivo studies, PDXs retain morphology, cellular heterogeneity, and molecular profiles of the original patient tumors [
12‐
18], representing an effective model for screening potential chemotherapeutics and translating them to enhanced efficacy in clinical trials [
19‐
22]. New experimental designs have recently been used as valid approaches to perform large-scale PDX-based preclinical trials to evaluate and predict the clinical efficacy and drug response of new therapeutics following the so-called 1 × 1 × 1 design [
15,
23,
24]. By using this design (i.e., one animal per model per treatment), PDX models provide the ability to place the same “patient” on all arms of a trial in a given preclinical study.
We have developed an extensive cohort of breast cancer PDXs that retain the morphology, cellular heterogeneity, and molecular profiles of the original patient tumors, serving as a renewable, quality-controlled tissue resource for preclinical evaluation of novel treatment regimens for what are in some cases extremely aggressive cancer types that currently lack adequate targeted therapeutic options [
12]. These PDXs have been characterized and classified according to Perou PAM50 and Pietenpol subtypes [
11,
25,
26] and their
TP53 mutational status [
11,
12,
27]. However, new therapies involving, among others, immune CPIs emphasize the need for the appropriate small-animal models to examine xenograft growth and response to therapy in the context of a “human” immune system and TME.
In the present study, we investigated the in vivo activity of anti-immune CPI-based therapies against TNBC PDX tumor models established in models of “humanized” nonobese diabetic/severe combined immunodeficiency
IL2Rγ
null (hNSG) mice by the engraftment of human CD34
+ HSCs, as previously described [
28,
29]. We show that, in terms of the animal model, engrafted human HSCs displayed self-renewal and multilineage differentiation capacities and that anti-PD-1 antibody therapy may result, as observed in clinical studies, in varying effects, with some PDXs responding positively to the treatment (i.e., significant reduction in tumor growth and increased survival), whereas others show no signs of improvement. Importantly, in those models that responded to the anti-PD-1 therapy, the effects were differentially displayed and observed only in the hNSG mice, indicating that despite potential limitations of the model, it may still represent an important tool for the preclinical evaluation of immunotherapies in breast cancer.
Methods
Mice
All the present study protocols involving mice followed the standard regulations and were approved by the Houston Methodist Research Institute Institutional Animal Care and Use Committee. “Humanized” mouse models refer to immunodeficient mice engrafted with human hematopoietic and lymphoid cells or tissues. NOD.Cg-
Prkdcscid Il2rgtm1Wjl/SzJ (NOD scid γ [NSG]; The Jackson Laboratory, Bar Harbor, ME, USA) mice were used as the recipient strain to intravenously (i.v.) engraft human CD34
+ HSCs (STEMCELL Technologies, Vancouver, BC, Canada) as previously described [
28,
29]. Briefly, 21-day-old NSG mice were irradiated with 240 cGy (sublethal) whole-body γ-irradiation. After 4–6 hours, mice were inoculated via the lateral tail vein with 3 × 10
4 CD34
+ HSCs. HSCs were allowed to engraft, and peripheral blood of recipient mice was collected from the retro-orbital sinus and analyzed by flow cytometry as indicated in the corresponding figure legends herein. “hNSG” is used to denote that the mice have HSC cells engrafted.
PDXs were originally derived by transplanting a fresh patient breast tumor biopsy into the cleared mammary gland fat pad of immunocompromised mice. Tumor samples (2 × 2 mm) were serially passaged in NSG mice by fat pad transplant under general anesthesia [
12]. Low-passage TNBC MC1 [
30], BCM-2147, BCM-4913, BCM-4664, and BCM-5471 [
12] samples were transferred into hNSG mice for engraftment approximately 6–8 weeks after initial human CD34
+ HSC cells tail vein injection. The weight of the mice was recorded and tumor volumes were measured and calculated [0.5 × (long dimension) × (short dimension)
2] twice weekly. When tumors reached an average size of 150–200 mm
3, mice were randomized (
n ≥ 5 per group) and used to determine the response to the treatment.
As validation of the humanized model, immunogenic A375 melanoma cell lines (American Type Culture Collection, Manassas, VA, USA) were maintained in DMEM (Life Technologies, Carlsbad, CA, USA), 10% FBS (HyClone; Life Technologies), and 1% antibiotic-antimycotic in a humidified 5% CO2 incubator at 37 °C. Cells (5 × 105) were injected orthotopically into the skin of NSG and hNSG mice and after 7–10 days (palpable tumors), and mice were randomly sorted into treatment groups.
Reagents
Humanized antibodies were obtained from Merck Oncology (Kenilworth, NJ, USA; pembrolizumab [Keytruda™], anti-PD-1) and Bristol-Myers Squibb (New York, NY, USA; nivolumab [Opdivo™], anti-PD-1; and ipilimumab, anti-CTL-4). Serum and tumor contents of human cytokine and chemokine biomarkers were determined by using the MILLIPLEX MAP Human High Sensitivity T Cell Panel Premixed 13-plex, Immunology Multiplex Assay (EMD Millipore, Billerica, MA, USA). Lymphoprep (STEMCELL Technologies) was used to isolate human peripheral blood mononuclear cells from tumor.
IHC
IHC assays were performed following established protocols [
31]. After antigen retrieval (Tris-Cl, pH 9.0), paraffin-embedded sections of PDX tumors were incubated for 1 hour at room temperature with the following antibodies: antihuman CD45 (leukocyte common antigen, clones 2B11 + PD7/26); antihuman CD68, clone KP1; antihuman CD8 (clone C8/144B); antihuman CD4, clone 4B12; antihuman Ki-67, clone MIB-1 (Dako, Glostrup, Denmark); antihuman CD3, clone UCHT1 (STEMCELL Technologies); antihuman CD20, clone EP459Y; antihuman CD56, clone EPR2566 (Abcam, Cambridge, MA, USA); antihuman cytokeratin 19 (CK19), clone A53-B/A2.26, also known as Ks19.1 (Thermo Scientific, Waltham, MA, USA).
Western blot analysis
Protein analysis was performed by Western blotting [
31]. Briefly, whole-cell lysates were made in 1× lysis buffer (Cell Signaling Technology, Danvers, MA, USA) with protease/phosphatase inhibitor cocktail (Thermo Scientific). Samples (30 μg) were boiled in sample buffer (Thermo Scientific) containing β-mercaptoethanol (Sigma-Aldrich, St. Louis, MO, USA) and subjected to SDS-PAGE electrophoresis in 4–20% polyacrylamide gels (Bio-Rad Laboratories, Hercules, CA, USA), transferred onto nitrocellulose membranes (Bio-Rad Laboratories), and incubated overnight at 4 °C with primary antibodies (1:1000; anti-PD-L1, catalogue no. 13684; anti-β-actin, catalogue no. 4970; Cell Signaling Technology), followed after washes by the appropriate secondary antibodies for 1 hour (1:2000). Protein bands were developed in autoradiography films (Denville Scientific Inc., South Plainfield, NJ, USA).
Fluorescence-activated cell sorting analysis
Analysis of mouse and human blood, spleen, and bone marrow mononuclear cells was performed by fluorescence-activated cell sorting analysis [
29,
32]. The antibodies used were as follows: antimouse CD45-fluorescein isothiocyanate (FITC), clone 30-F11; antihuman CD45-allophycocyanin (APC), clone HI30; antihuman CD3-phycoerythrin (PE), clone UCHT1; antihuman CD20-FITC, clone 2H7; PE-cyanine 7 mouse antihuman CD68, clone Y1/82A; Alexa Fluor 700 mouse antihuman CD56, clone B159; antimouse CD45-PE, clone 30-F11; antimouse CD45-peridinin chlorophyll protein complex, clone 30-F11; mouse immunoglobulin G2b (IgG2b), κ isotype-FITC, clones 27–35; mouse IgG1, κ isotype-PE, clone MOPC-21; and mouse IgG2b κ isotype-APC (BD Biosciences, San Jose, CA, USA); Pacific Blue antihuman CD33 eFluor® 450, clone P67; and Pacific Blue Mouse IgG1 K Isotype Control eFluor® 450 (eBioscience, San Diego, CA, USA). Briefly, erythrocytes were lysed, after which lymphoid cells were incubated with the corresponding antibodies and fixed following standard procedures [
29,
32]. Flow cytometric analysis was performed at the Houston Methodist Research Institute Flow Cytometry Core using a BD LSRFortessa flow cytometer for acquisition of data and FACSDiva software (both from BD Biosciences) for analysis.
Tumor-infiltrating lymphocyte cytotoxic activity assay
Following a four-cycle treatment with anti-PD-1 antibody (nivoluzumab 10 mg/kg), MC1-engrafted tumors growing in hNSG mice were collected and mechanically disaggregated into single cells, and TILs were isolated by using Ficoll gradient (Lymphoprep; STEMCELL Technologies). These TILs were cocultured with MC1 tumor cells extracted from nonhumanized NSG mice for 6 hours (250:7 ratio of target cells to effector cells), and TIL cytotoxic activity was measured with the CytoTox 96® Non-Radioactive Cytotoxicity Assay (Promega, Madison, WI, USA) as per the manufacturer’s instructions. Granzyme B tumor levels were measured by incubating tumor protein lysates with antibody-immobilized magnetic beads (HGRNZMB-MAG; EMD Millipore, Billerica, MA) and evaluated using a Luminex LX-200 multiplexing assay system (Luminex Corp., Austin, TX, USA).
Statistical analysis
All data were analyzed using Prism software (GraphPad Software, La Jolla, CA, USA). Data are presented as mean ± SEM. Statistical significance between two groups was analyzed by two-tailed Student’s t test. Experiments with more than three groups were analyzed with one-way analysis of variance (ANOVA) and Bonferroni’s post hoc test. Statistical analysis of tumor volume was assessed by two-way ANOVA and Bonferroni’s post hoc test. Survival proportions were assessed by using the Kaplan-Meier method and further analyzed with either Wilcoxon or log-rank test. A P value less than 0.05 was considered significant.
Discussion
The use of immunotherapies in breast cancer has been limited by breast cancer’s relatively low immunogenicity [
55]. However, newly developed strategies and/or approaches are rapidly changing the field, and novel immune CPIs are already approved or under different phases of clinical evaluation. Examples of these studies include clinical evaluation of anti-PD-1 and anti-PD-L1 therapies, administered either as single drugs or as part of multiple combinations [
56,
57]. Enrichment strategies to select for patients more likely to respond have identified the expression and testing of PD-L1 to be a potentially useful predictive marker in guiding this process [
58‐
60]. Following these criteria, in the present study, we investigated the expression of PD-L1 and its correlation with the anti-PD-1 activity. Although we did not evaluate a number of PDX tumor lines large enough to have the power required to achieve a statistically supported conclusion, our results showed a trend: Those PDXs that expressed high levels of PD-L1 appeared to respond to the anti-PD-1 therapy. Several clinical studies have evaluated the expression of PD-L1 and tried to identify possible associations with the therapeutic response. For example, positive expression of PD-L1 in TNBC stromal tissue or in ≥ 1% of tumor cells has been used as a potential predictive biomarker in the phase Ib KEYNOTE-012 clinical trial [
47]. Here, an 18.5% overall response rate was observed in the PD-L1-positive group, which represented ~ 60% of the total number of heavily pretreated patients with advanced TNBC under evaluation [
47]. Other studies included a retrospective analysis (between 2004 and 2013) of 136 TNBC cases without neoadjuvant therapy, showing that stromal PD-L1 expression was significantly associated with better disease-free survival (DFS), whereas no association was found between PD-1 expression and DFS, overall survival, or metastasis [
61]. Additional observations made by Botti et al. also showed a strong association between PD-L1 expression and better DFS [
62]. Similar outcomes have resulted from a phase Ia study of the anti-PD-L1 antibody atezolizumab in previously treated patients with TNBC [
63], altogether adding supporting evidence to the notion that PD-L1 expression may represent an important biomarker for prognostic stratification and CPI-based therapies. Nonetheless, the current consensus is that in addition to the expression of PD-L1 and mutation burden, multiple biomarkers may be needed to determine which patients will likely benefit from immunotherapies, including, notably in TNBC and HER2-positive patients, the presence of CD8
+ TILs, immune-related gene signatures, and multiplex IHC assays that may take into account the pharmacodynamic and spatial interactions of the TME [
55,
56,
64‐
66]. As we demonstrated in the present study, our hNSG PDX model displayed clear evidence of several of these parameters (i.e., a humanized immune system with detectable presence of hCD45
+ TILs and cytokine levels) and robust expression of PD-L1 in some of the tumor lines. These results are in line with the clinical studies previously mentioned where the therapeutic benefits of regimens containing immunomodulatory CPI were observed mainly in patients where both TILs and PD-L1 were present, which provides additional support for the use of the humanized TNBC PDX mouse model used in this work. Similarly, also in agreement with observations in clinical trials [
51,
67], the present model showed limited or no activity when TNBC tumor line MC1 was treated with an anti-CTLA-4 antibody, further validating the humanized mouse model because it reproduces some of the most relevant results observed during the clinical evaluation of immune CPIs. In fact, anti-CTLA-4 monotherapies have shown no or very limited therapeutic advantage against breast cancer when administered alone [
67], although their efficacy has been improved by combination with other agents [
50,
51,
68], which opens the field to new investigations. The mechanisms leading to the apparent lack of anti-CTLA-4 activity when administered as a monotherapy in certain solid tumors, including breast cancer, are still not well understood. However, it is thought to be associated with tumors’ low antigenicity and microenvironment conditions that may not favor immune recognition [
65,
69,
70].
From a potential mechanistic point of view, our studies indicate that the effects of blocking PD-1/PD-L1 interactions, thereby improving the immunological response [
7,
8], may have resulted from increased activation of TILs rather than changes in the number of cells infiltrating the tumor. These observations are consistent with the established mode of action of these compounds (i.e., interfering the immune-inhibitory effects of the PD-1/PD-L1 interactions) [
71]. In addition, our results may also suggest that amelioration of the therapeutic efficacy of immune CPIs could be achieved by modifying the TME as a way to enhance their activity, and in fact, multiple ongoing studies at both our and other laboratories are currently addressing this hypothesis. In addition, further studies are being designed to determine the long-term effects of CPIs in terms of tumor growth inhibition and mechanisms of resistance, notably in comparison to established chemotherapies, because the present report spanned a relatively short time frame.
In terms of the animal model that we used in the present study, it is clear that although these animals represent a very useful tool, humanization of NSG mice may still pose some technical challenges and/or limitations. Notably, one of those well-recognized limiting factors is the lack of GM-CSF, important for the differentiation and maturation of the myeloid lineage [
72]. To address this point, several newer, genetically modified NSG-based (The Jackson Laboratory) or NOG (NOD/Shi-
scid/IL-2Rγ
null)-based (Taconic Biosciences, Rensselaer, NY, USA) models are being developed, which, by expressing the human cytokines GM-CSF and IL-3 and human stem cell factor gene (
SCF; also known as KIT ligand,
KITLG), allow for better engraftment of HSCs and cell lineage differentiation [
73]. In our case, it is important to note that some of these limitations appeared to be compensated by the presence of the TNBC PDX. Indeed, as our results show, PDXs were associated with the presence of several cytokines, including GM-CSF, which consequently may have played an important role in improving the levels of the myeloid lineage (hCD33
+ cells) when compared with the hNSG mice not harboring tumors. These results suggest, as previously mentioned, that the simultaneous presence of the PDX during hHSC engraftment may have compensated for the lack of this and other factors, contributing to a better reconstitution of the immune system.
Another important factor that was considered in our study was the potential role of matching HLA typing between the hNSG host and the PDXs. Our observations showed some differences in the PDX growth rate based on whether the mice were humanized or not, most likely owing to the incipient presence of an active immune system. However, as also shown by others, including the case of commercially available humanized PDX models [
36,
37], no signs of graft-versus-host reaction were found. Furthermore, on the basis of the fact that the HLA typing of HSCs did not conclusively demonstrate compatibility with more than one pattern, it is plausible to postulate that the slower growth of PDXs may have resulted from partially HLA-matched hNSG/PDX engraftment, which allowed a seemingly regular tumor engraftment. This is an important observation because the ideal situation (i.e., isolating HSCs from the same cancer patient whose PDX is being used) may prove extremely difficult to achieve in large-scale preclinical studies, because of both the patient condition and the time usually required for a PDX to be established [
73]. Alternatively, the use of immunocompetent syngeneic mouse models represents a valid approach. However, this also has its own limitations, mostly in terms of the availability of tumor models, the specificity of drugs being tested, and the extrapolation of observations to human cases. Together, despite some of the factors mentioned above that should be taken into consideration whenever using humanized PDX mouse models, these models still represent very helpful and sophisticated tools for preclinical evaluation of immune-based therapies, notably as they become more available and improved animal versions are generated.
Acknowledgements
The RNA-seq data were generated and analyzed by the Genome Sequencing Facility of Greehey Children’s Cancer Research Institute at the University of Texas Health San Antonio. We thank Dr. Zhao Lai for RNA-seq assistance and Hung-I Chen and Dr. Yidong Chen for RNA-seq data analysis. Low-resolution HLA typing was performed by the transplant immunology laboratory at Houston Methodist Hospital, Houston, TX, USA. DDG is grateful for support from the Instituto Tecnológico y de Estudios Superiores de Monterrey, Monterrey, Mexico, and Consejo Nacional de Ciencia y Tecnología, Mexico (CONACyT 490148/278957). The authors acknowledge the help provided by the Flow Cytometry and Comparative Medicine core facilities at the Houston Methodist Research Institute. DDG is a current graduate student at the Instituto Tecnológico y de Estudios Superiores de Monterrey, Monterrey, Mexico.