Resistance to immunotherapy is a significant challenge, and the scarcity of human models hinders the identification of the underlying mechanisms. To address this limitation, we constructed an autologous humanized mouse (aHM) model with hematopoietic stem and progenitor cells (HSPC) and tumors from 2 melanoma patients progressing to immunotherapy. Unlike mismatched humanized mouse (mHM) models, generated from cord blood–derived HSPCs and tumors from different donors, the aHM recapitulates a patient-specific tumor microenvironment (TME). When patient tumors were implanted on aHM, mHM, and NOD/SCID/IL2rg−/− (NSG) cohorts, tumors appeared earlier and grew faster on NSG and mHM cohorts. We observed that immune cells differentiating in the aHM were relatively more capable of circulating peripherally, invading into tumors and interacting with the TME. A heterologous, human leukocyte antigen (HLA-A) matched cohort also yielded slower growing tumors than non–HLA-matched mHM, indicating that a less permissive immune environment inhibits tumor progression. When the aHM, mHM, and NSG cohorts were treated with immunotherapies mirroring what the originating patients received, tumor growth in the aHM accelerated, similar to the progression observed in the patients. This rapid growth was associated with decreased immune cell infiltration, reduced interferon gamma (IFNγ)–related gene expression, and a reduction in STAT3 phosphorylation, events that were replicated in vitro using tumor-derived cell lines.

Implications:

Engrafted adult HSPCs give rise to more tumor infiltrative immune cells, increased HLA matching leads to slower tumor initiation and growth, and continuing immunotherapy past progression can paradoxically lead to increased growth.

Increased scrutiny of the tumor microenvironment (TME) has led to better understanding of pro- and antitumor TME components and the development of therapies aimed at disrupting the mechanisms underlying immune evasion in human cancer. The cytotoxic T-lymphocyte-associated protein 4 (CTLA4) inhibitor ipilimumab was approved for the treatment of metastatic melanoma in 2011; the programmed cell death protein 1 (PD-1) inhibitors nivolumab and pembrolizumab followed in 2014 and 2015, respectively (1–3). However, immune-directed therapy is often limited by intrinsic and acquired tumor resistance, and an improved understanding of these phenomena is an acute need. Such resistance has been associated with mutations in genes encoding interferon gamma (IFNγ), the Jak–STAT signaling pathway (4, 5) and with changes in PD-L1 and HLA expression that reduce the effectiveness of anti–PD-1 therapies in promoting an antitumor T-cell response (6, 7).

In addition, the phenomenon of immunotherapy-induced tumor hyperprogression has been recently identified and described in a subset (∼9%) of cancer patients (8). Characterized by a sudden increase in the rate of tumor growth after the start of immunotherapy, hyperprogression has been observed in several types of cancers, including melanoma. It is often associated with mutations or acquired dysregulation in the genes encoding MDM2/4, EGFR, DMNT3A, JAK1/2, and B2M (9, 10). Many of these genes are associated with the IFNγ pathway (11, 12). Because it is difficult to quantitatively compare pretreatment tumor growth with that observed after immunotherapy, and given the lack of appropriate laboratory models, a comprehensive examination of hyperprogression has been thus far challenging.

Because a human TME cannot be examined when cancer xenografts are grown on traditional immunodeficient animal models, humanized mouse (HM) patient-derived xenograft (PDX) models have been developed. HM, engrafted with the hematopoietic precursor and stem cells (HSPC) necessary for the development of a functional human immune system, can be used to examine the complex relationships in the TME within the context of a growing and invading tumor. We previously generated an HM xenograft model of head and neck squamous cancer and have shown that the engrafted HSPCs can divide, differentiate, and invade implanted tumors, where they alter the expression of immune-related genes to more closely match the profiles found in the originating patient tumors (13, 14). Because these early models were created from donated umbilical cord blood, the resultant immune system and subsequently implanted xenograft came from separate sources (thus termed mismatched HM, or mHM), and the interactions of immune cells and tumor tissues may not accurately mirror those found in the originating patients.

In the current study, we (i) generated an autologous humanized mouse (aHM) model made from the HSPCs and tumor tissue collected from 2 metastatic melanoma patients (CUHM003 and CUHM005) after progression during immune-directed therapy, (ii) determined differences in tumor formation and growth between aHM and nonhumanized NOD/SCID/IL2rg−/− (NSG) and mHM cohorts, (iii) replicated the treatment received by the patients to identify patterns of growth in each model, and (iv) characterized the molecular and immune events associated with immune-directed therapy. We observed a distinct pattern of immune cell engraftment in the aHM, and in this model tumor growth, immune cell infiltration, and response to CTLA4- and PD-1 inhibitors were more consistent with what was observed in the corresponding patients (15, 16). Finally, molecular analyses revealed altered transcription among IFNγ-related genes associated with hyperprogression. Both mHM and aHM models provide humanized TMEs in which patient tumors can be grown and studied; however, aHM afford unique opportunities to study the factors driving tumor progression and hyperprogression.

HSPC collection, expansion, and engraftment in mice

Deidentified cord blood was obtained from the University of Colorado cord blood bank (http://www.clinimmune.com/cordbloodbank/). The use of human subjects was approved by the Colorado Multiple Institutional Review Board (COMIRB #14-0842). Eligible, willing patients with incurable melanoma were prospectively enrolled, had a fresh tumor biopsy, and received filgrastim (10 μg/kg daily) for 4 days. A total of 150 mL of blood was collected 4 to 6 days later.

HSPCs were purified from either cord or patient blood by CD34+ cell selection (Stemcell Technologies; cat. #14756), suspended in serum-free expansion medium (Stemcell Technologies, cat. #09650), and cultured at 37°C, 5% CO2 for 5 to 8 days. Cells were characterized by cytometry, using CD34, CD45, CD73, and CD166 antibodies (BioLegend; cat. #343608, RRID:AB_2228972; 304039, RRID:AB_2562057; 344006, RRID:AB_1877157; 343904, RRID:AB_2289302) at a 1:10 concentration. NSG (The Jackson Laboratory; cat. #005557, RRID:IMSR_JAX:005557) mice were primed for engraftment by 1.5 Gy whole-body irradiation. After a recovery of 4 to 6 hours, the mHM mice were each injected with 400,000 expanded CD34+ cells, suspended in 0.2 mL sterile PBS. The aHM003 mice received 150,000 CD34+ cells, whereas the aHM005 mice received 140,000 CD34+ cells. When present, MSC-like cells were added to comprise 5% of the total injected cells. The mice were bled via the tail vein after 8 to 10 weeks to assess HSPC engraftment. Their peripheral blood was analyzed by flow cytometry, using human CD3, CD11b, CD19, and/or CD45 (BioLegend; cat. #300312, RRID:AB_314048; 301310, RRID:AB_314162; 392504, RRID:AB_2728416; 304039; RRID:AB_2562057) antibodies at 1:10. At the conclusion of the study, HM bone marrow was collected and analyzed by flow cytometry, using human CD34 and CD45 antibodies at 1:10. The University of Colorado Institutional Animal Care and Use Committee (IACUC) approved all experiments involving mice. PDX generation and animal care have been previously reported (17).

In vivo treatment studies

CUHM003 mice received either a human IgG control (Gammagard; Takeda; 10 mg/kg), ipilimumab (Bristol-Myers Squibb; 20 mg/kg), or pembrolizumab (Merck; 20 mg/kg). CUHM005 mice received IgG control (10 mg/kg), ipilimumab, nivolumab (Bristol-Myers Squibb; 10 mg/kg), or ipilimumab plus nivolumab, each given at the single-agent dose. Tumors were measured 3 times weekly, and when tumors averaged 75 mm3, treatment was administered twice weekly by intraperitoneal injection for 4 weeks. At the end of the study, blood was collected by cardiac puncture in EDTA. Tissues were collected for cytometry, flash frozen in liquid nitrogen, and in placed in formalin to paraffin embed.

Fluorescence activated cell sorting and flow cytometry

Tumor and mouse tissues were prepared for cytometric analysis as previously described (14). Cell sorting was performed using a MoFlo XDP (Beckman Coulter), and flow cytometry was completed on a CyAn ADP (Beckman Coulter) using Summit V5.1 (Beckman Coulter) software.

IHC

IHC analyses were performed as described (13). Primary antibodies and dilutions: CD45 (Dako; cat. #M0701, RRID:AB_2661839) 1:100; CD3 (Abcam; cat. #ab5690, RRID:AB_305055), 1:500; CD19 (MyBiosource; cat. #MBS2544305, RRID:AB_2868606), 1:100; and CD68 (Dako; cat. #M0876, RRID:AB_2074844) 1:100. Staining was developed using the following conditions: EnVision + Dual Link System HRP (Dako; cat. #K4061) for 30 minutes and substrate-chromogen (DAB+) Solution (Dako; cat. #K3468) for 5 minutes. Slides were then counterstained with Automated Hematoxylin (Dako; cat. #S3301) for 10 minutes.

Exome and mRNA sequencing and bioinformatics

Biological duplicates were sent to the UCCC Genomics and Microarray Core for library generation and Illumina HiSeq (Illumina) sequencing. FastQC (v0.11.3, RRID:SCR_014583) was used for quality control (Exome-seq and RNA-seq). Cutadapt (v1.8.1, RRID:SCR_011841) was used to remove Illumina adapters. Trimmomatic (v0.33, RRID:SCR_011848) was used to remove low quality reads (18). Exome variants were called using the IMPACT pipeline (19) to annotate somatic and noncommon [with allele frequency greater than 1% in dbSNP, RRID:SCR_002338 (20) or the 1000 Genomes Project, RRID:SCR_008801 (21)] variants. Transcript reads were quantified using Tuxedo Suite (http://cole-trapnell-lab.github.io/projects/; ref. 22), aligned against the GRCh37 reference genome using TopHat (v2.0.14, RRID:SCR_013035), and assembled and merged using Cufflinks (v2.2.1, RRID:SCR_014597). Gene set enrichment analysis (GSEA2-2.2.0, RRID:SCR_003199) was conducted using MSigDB (v5.2, http://software.broadinstitute.org/gsea/msigdb/collections.jsp; refs. 23, 24). Pathways were analyzed using GSEABase R package v1.48.0 and ClusterProfiler v3.14.2, RRID:SCR_016884 (25) and were considered highly significant at an FDR q ≤ 0.01, significant at q ≤ 0.05, and modestly significant at the default threshold value of q ≤ 0.25. Differential expression was analyzed using DESeq2 with APEGLM shrinkage estimator with an FDR of 0.1 (26). Top pathways/gene ontologies were identified using the Database for Annotation, Visualization and Integrated Discovery (DAVID), RRID:SCR_001881 (27, 28).

Cell line generation and sphere assay

Cell lines were derived from tumor tissue using RMK media, as described (29), and validated by Mycoplasma testing and STR analysis (2018-03-18 for CUHM003 and 2018-05-03 for CUHM005). 200,000 cells per well were plated in triplicate in ultra-low attachment 12-well plates and supplemented with media after 4, 7, and 10 days. T cells were isolated by magnetic separation and activated (Stemcell Technologies; cat. #19051 and 10971) from 5 mL deidentified adult blood draws acquired through the University of Colorado. Cells were allowed to form spheres for 10 days before 20,000 T cells, 2 ng/mL purified interferon gamma (R&D Systems; cat. # 285-IF-100), and/or 4 μg/mL pembrolizumab or nivolumab were added. Spheres were imaged, counted, and measured using a Zeiss Axio Observer Z1 inverted microscope (Zeiss software Rel. 4.8). siIFNγR2 knockdown was verified by qPCR as described (29).

Cell line siRNA experiments

For IFNγR2 and STAT3 knockdowns, cells were seeded in 6-well plates and incubated for 24 hours. Media were replaced with serum-free DMEM for 30 minutes prior to transfection with 1 μL/mL Dharmafect1 and 50 to 100 nmol/L siRNA (GE Dharmacon; IFNγR2 SMARTpool J-012713-05-08; STAT3 SMARTpool J-003544-07-10). Cells were incubated for 24 hours before DMEM containing 20% FBS was added, and cells were incubated for another 48 to 72 hours.

Cytokine arrays

Plasma from mouse blood was collected and flash frozen for subsequent analysis. Cytokine presence and concentration in the plasma was interrogated on a human cytokine array kit (R&D Systems; cat. # ARY005B) according to the manufacturer's instructions. Cytokine concentration was quantified by ImageJ software, version 1.5, RRID:SCR_003070 (NIH, imagej.nih.gov), and visualized using R.

Protein isolation and Western blotting

Western blotting and analysis were conducted as previously described (30). Primary antibodies and dilutions: 1:2,000 Actin (pan) (Cell Signaling Technology; cat. #4968, RRID:AB_2313904), 1:1,000 phospho-STAT1 (Cell Signaling Technology; cat. #9167, RRID:AB_561284), 1:1,000 STAT1 (Cell Signaling Technology; cat. #9175, RRID:AB_2197984), 1:1,000 phospho-STAT3 (Cell Signaling Technology; cat. #9131, RRID:AB_331586), 1:2,000 STAT3 (Cell Signaling Technology; cat. #4904, RRID:AB_331269). Secondary anti-rabbit IgG (Jackson ImmunoResearch; cat. #111-035-045, RRID:AB_2337938), and used at a 1:5,000 dilution. Quantification of relative protein levels was completed using ImageJ, RRID:SCR_003070.

Statistical analysis

In vitro and in vivo (using ≥5 mice/group) experiments were compared with Brown-Forsythe ANOVAs and two-sided t tests. Final tumor volumes for all tumor groups were calculated as the fold change in size between the beginning and end of the study after the initial volumes of the tumors had been set at a value of 1. When treatment groups were compared, the treated arms were all normalized by the average fold change of the associated control tumors, as previously described (31). Spheroids were compared using standard ANOVAs and Dunnett multiple comparison tests. Calculations were done using GraphPad Prism, RRID:SCR_002798, version 8.3. Data are represented graphically as mean ± SEM. GSEA estimates the statistical significance of the enrichment scores by a two-sided modified Kolmogorov–Smirnov permutation test. P and Q values of less than 0.05 were statistically significant. All statistical tests were two-sided.

HSPCs from cord and patient blood expand ex vivo and engraft mHM and aHM cohorts

In order to conduct a comprehensive comparison, we generated a nonhumanized model (NSG), a model with a mismatched immune system (mHM), and a model with an autologous immune system (aHM; Fig. 1A) for both CUHM003 and CUHM005 patients. To generate the aHM cohorts, we isolated HSPCs from G-CSF–stimulated patient blood. To construct mHM cohorts, we isolated HSPCs from donated cord blood. The CD34+ HSPCs were then expanded ex vivo for ∼8 days, after which the number of patient cells had undergone an average ∼40-fold expansion (Fig. 1B; Supplementary Table S1) and the cord blood cells had increased by ∼180-fold. We also observed and expanded a separate population of adherent CD73-CD166+ cells, characteristic of mesenchymal stem cells (MSC; ref. 14). Because it has been demonstrated that radiation-induced damage to bone marrow encourages HSC homing and establishment in this niche (32), we injected expanded HSPCs and MSCs into the tail veins of sublethally irradiated NSG mice to create mHM and aHM cohorts. In order to generate enough mice to reproduce patient therapy, each mHM received 400,000 HSPCs, and each aHM was engrafted with 140,000 HSPCs, a population at the lower end of what is necessary for successful humanization using cord blood HSPCs (14). We hypothesized that increased HLA matching would enhance the functional efficiency of the mature immune cells produced by the HSPCs in recognizing and interacting with implanted autologous tumor tissue and decrease the required number of HSPCs in these models. A similar phenomenon has been observed in patients receiving cord blood transplants, where a reduced mismatch (1 vs. 2 HLA) requires fewer total nucleated cells (>2.5 × 107/kg vs. >5.0 × 107/kg) to achieve similar engraftment and clinical outcomes (33).

Figure 1.

Generation of autologous humanized mice (aHM). A, An overview of how aHM cohorts can be generated to investigate patient therapy. G-CSF–stimulated HSPCs from melanoma patients can reconstitute their immune system in immunocompromised mice. Patient tumors can be implanted onto the flanks and shoulder of these mice, as well as onto concurrently prepared mHM and NSG controls. All mice can be treated with the same therapies administered to the patient and their tumor responses compared. B,In vitro culture and expansion of cord- and patient-derived HSPC and MSC-like cell populations. Following CD34+ column selection, analysis by cell cytometry identifies a small population of CD34+45+ HSPCs in both the newly procured cord (mHM003 or mHM005) and patient (aHM003 or aHM005) blood. After 5 to 8 days of expansion, the CD34+ HSPC population has increased markedly for all of these cultures. A population of CD34, CD73+166+ adherent MSC-like cells originating from within the HSPC population can also be identified after in vitro expansion. Total cell numbers of HSPCs before and after expansion are recorded in Supplementary Table S1. C, Representative IHC showing the relative populations of human CD45+ cells (dark brown) within the bone marrow of NSG, mHM, and aHM CUHM003 and CUHM005 models. Magnification is 20×; scale bar = 50 μm.

Figure 1.

Generation of autologous humanized mice (aHM). A, An overview of how aHM cohorts can be generated to investigate patient therapy. G-CSF–stimulated HSPCs from melanoma patients can reconstitute their immune system in immunocompromised mice. Patient tumors can be implanted onto the flanks and shoulder of these mice, as well as onto concurrently prepared mHM and NSG controls. All mice can be treated with the same therapies administered to the patient and their tumor responses compared. B,In vitro culture and expansion of cord- and patient-derived HSPC and MSC-like cell populations. Following CD34+ column selection, analysis by cell cytometry identifies a small population of CD34+45+ HSPCs in both the newly procured cord (mHM003 or mHM005) and patient (aHM003 or aHM005) blood. After 5 to 8 days of expansion, the CD34+ HSPC population has increased markedly for all of these cultures. A population of CD34, CD73+166+ adherent MSC-like cells originating from within the HSPC population can also be identified after in vitro expansion. Total cell numbers of HSPCs before and after expansion are recorded in Supplementary Table S1. C, Representative IHC showing the relative populations of human CD45+ cells (dark brown) within the bone marrow of NSG, mHM, and aHM CUHM003 and CUHM005 models. Magnification is 20×; scale bar = 50 μm.

Close modal

After 8 weeks, we identified a population of human B cells, comprising 0.01% to 0.4% of the total white blood cells in the mouse peripheral blood (Supplementary Fig. S1). We also quantified the human blood cell populations in the mouse bone marrow, blood, and spleen by cytometry at the conclusion of the studies (Supplementary Fig. S2). On average, the bone marrow of the mHM cohorts contained healthy populations of human CD45+ immune cells (mHM003, 12.37%; mHM005, 27.93% of all bone marrow cells) and smaller populations of HSPCs (mHM003, 0.94%; mHM005, 3.73%). The bone marrow of the aHM cohorts contained more modest CD45+ cell populations (aHM003, 0.29%; aHM005, 0.04%) and HSPC (aHM003, 0.03%; aHM005, <0.01%; Supplementary Table S2), possibly reflecting the lower initial HSPC injection numbers. We also compared human CD45+ cells within mHM and aHM bone marrow and spleen by IHC (Fig. 1C; Supplementary Fig. S3) and observed similar patterns of human CD45+ cell engraftment.

Delayed tumor initiation and decreased growth in aHM and in HLA A–matched HM

Approximately 10 weeks after their humanization, tumors were implanted on both flanks and a shoulder of the NSG, mHM, and aHM cohorts, using expanded tumor tissue from the initial biopsy. Tumors appeared earlier and grew faster on the NSG and mHM cohorts. At study end, NSG003 tumors were 1.4 times larger than those in aHM003 (P = 0.04; Fig. 2A). NSG005 and mHM005 tumors were 3.0-fold (P < 0.01) and 3.3-fold (P < 0.01) larger than aHM005, respectively (Fig. 2B). There was no difference in the rate of growth between the NSG and mHM tumors in CUHM003 or CUHM005. In order to further investigate immune matching, we implanted CUHM003 tumors on NSG, HM HLA-A–mismatched (mHM003b), and HM HLA-A–matched cohorts (mHM003-HLA). Tumor occurrence was again delayed and growth trended slower (but without reaching statistical significance) in mHM003-HLA versus that observed in NSG003 and mHM003b cohorts (Supplementary Fig. S4A), suggesting that immune permissiveness modulates tumor occurrence and growth.

Figure 2.

Tumor growth dynamics are significantly affected in aHM models. A and B, Relative growth rates of untreated CUHM003 and CUHM005 tumors in the NSG, mHM, and aHM models. CUHM003 tumors implanted on aHM (n = 7) grew significantly slower than those implanted on NSG mice (n = 7; *, P = 0.04, by two-sided t test for this and all subsequent comparisons). CUHM005 tumors implanted on aHM (n = 12) grew more slowly than those implanted in either mHM (n = 11; **, P < 0.01) or NSG mice (n = 9; ***, P < 0.01). C, For CUHM003, no significant change in tumor growth was observed in response to ipilimumab or pembrolizumab treatment in either the NSG (n = 15 or 12) or mHM (n = 11 or 6) cohorts. However, among the aHM treatment resulted in a varied response (ANOVA; P = 0.06) with ipilimumab producing a 2-fold increase in tumor growth (n = 6; +, P = 0.02), whereas pembrolizumab stimulated more than a 4-fold surge in growth (n = 7; ++, P = 0.08). D, In CUHM005, there was again no difference after ipilimumab, nivolumab, or combination therapy among the NSG (n = 9, 9, 5) or mHM (n = 15, 12, 6) cohorts. In the aHM, response was again varied (AVONA, P < 0.01). Although ipilimumab (n = 12) did not stimulate significant tumor growth, nivolumab resulted in a 2-fold increase in growth (n = 12; +++, P = 0.02), and the combination of ipilimumab and nivolumab yielded greater than a 3-fold jump in tumor growth (n = 11; ++++, P < 0.01).

Figure 2.

Tumor growth dynamics are significantly affected in aHM models. A and B, Relative growth rates of untreated CUHM003 and CUHM005 tumors in the NSG, mHM, and aHM models. CUHM003 tumors implanted on aHM (n = 7) grew significantly slower than those implanted on NSG mice (n = 7; *, P = 0.04, by two-sided t test for this and all subsequent comparisons). CUHM005 tumors implanted on aHM (n = 12) grew more slowly than those implanted in either mHM (n = 11; **, P < 0.01) or NSG mice (n = 9; ***, P < 0.01). C, For CUHM003, no significant change in tumor growth was observed in response to ipilimumab or pembrolizumab treatment in either the NSG (n = 15 or 12) or mHM (n = 11 or 6) cohorts. However, among the aHM treatment resulted in a varied response (ANOVA; P = 0.06) with ipilimumab producing a 2-fold increase in tumor growth (n = 6; +, P = 0.02), whereas pembrolizumab stimulated more than a 4-fold surge in growth (n = 7; ++, P = 0.08). D, In CUHM005, there was again no difference after ipilimumab, nivolumab, or combination therapy among the NSG (n = 9, 9, 5) or mHM (n = 15, 12, 6) cohorts. In the aHM, response was again varied (AVONA, P < 0.01). Although ipilimumab (n = 12) did not stimulate significant tumor growth, nivolumab resulted in a 2-fold increase in growth (n = 12; +++, P = 0.02), and the combination of ipilimumab and nivolumab yielded greater than a 3-fold jump in tumor growth (n = 11; ++++, P < 0.01).

Close modal

Tumor growth accelerates in aHM treated with checkpoint inhibitors

Prior to trial enrollment and tumor biopsy, the CUHM003 patient had been treated with ipilimumab, initially achieving a partial response before progressing, and then pembrolizumab, resulting in rapidly progressive disease. We conducted a three-arm study (control, ipilimumab, and pembrolizumab) on the NSG003, mHM003, and aHM003 tumor–bearing mice. We observed no growth differences between NSG003 and mHM003 tumors. There were 2-fold and 4-fold increases in the growth of aHM003 ipilimumab- and pembrolizumab-treated tumors compared with controls (ANOVA P = 0.06; t test P = 0.02 and P = 0.08, respectively; Fig. 2C; Supplementary Fig. S4B).

The CUHM005 patient had exhibited rapid progressive disease during combined therapy with ipilimumab and nivolumab (a PD-1 inhibitor) given prior to humanized trial enrollment and tumor biopsy. We conducted a four-arm study (control, ipilimumab, nivolumab, and combination) on NSG005, mHM005, and aHM005 mice (Fig. 2D; Supplementary Figs. S4C and S5). We observed no difference in the growth of the treated NSG005 or mMH005 groups, but a significant acceleration in the growth of the nivolumab- and combination-treated aHM005 tumors occurred (2- and 3-fold increase; ANOVA P < 0.01; t test P = 0.02 and P < 0.01, respectively).

aHM tumors show a distinct pattern of infiltration by immune cells

To determine the basis for this phenomenon, we analyzed the pattern of human immune cell invasion in fresh tumor samples by cytometry (Supplementary Fig. S6) and in paraffin-embedded tumors by IHC (Supplementary Fig. S7). Compared with their originating patient samples, melanoma tumors on HM contained less stroma and fewer infiltrating human immune cells. Cytometric analysis, however, indicated that small populations of human T cells, B cells, and macrophages similarly infiltrated the tumors in both mHM and aHM—a noteworthy observation considering the relatively lower bone marrow engraftment observed in the aHM models. When normalized by the percentage of human cells in the bone marrow, relative T-cell presence in aHM003 was nearly 8 times greater than in mHM003, whereas there were 2000 times more tumor-infiltrating T cells in aHM005 than in mHM005, indicating that autologous T cells have a greater capacity to interact with tumor tissue, even though this infiltrative capacity decreased upon immunotherapy treatment (Table 1). IHC revealed that the human immune cells were generally congregated at the tumor capsule, implying that their association with the tumor is especially transient, as has been described for immunotherapy-refractory melanoma when comparing pre- and postimmunotherapy tumor samples from both patients (Supplementary Fig. S7), where the relatively few T cells present lacked infiltrative capacity and accumulated mainly near the tumor capsule (15, 16).

Table 1.

Percentage of tumor-infiltrating T cells in HM cohorts.

mHMaHM
BM (%)T cells (%)T cells/BMBM (%)T cells (%)T cells/BMT-cell invasive capacity (aHM/mHM)
CUHM003
 Control 3.70 0.007 0.002 0.183 0.003 0.014 7.59 
 Ipilimumab 12.92 0.010 0.001 0.480 0.003 0.005 6.73 
 Pembrolizumab 14.01 0.095 0.007 0.220 0.003 0.012 1.84 
CUHM005 
 Control 44.95 0.01 0.0002 0.01 0.007 0.488 2192 
 Ipilimumab 27.46 0.01 0.0004 0.04 0.013 0.357 981 
 Nivolumab 19.26 0.01 0.0005 0.04 0.007 0.168 324 
 Combination 20.07 0.01 0.0005 0.06 0.002 0.030 61 
mHMaHM
BM (%)T cells (%)T cells/BMBM (%)T cells (%)T cells/BMT-cell invasive capacity (aHM/mHM)
CUHM003
 Control 3.70 0.007 0.002 0.183 0.003 0.014 7.59 
 Ipilimumab 12.92 0.010 0.001 0.480 0.003 0.005 6.73 
 Pembrolizumab 14.01 0.095 0.007 0.220 0.003 0.012 1.84 
CUHM005 
 Control 44.95 0.01 0.0002 0.01 0.007 0.488 2192 
 Ipilimumab 27.46 0.01 0.0004 0.04 0.013 0.357 981 
 Nivolumab 19.26 0.01 0.0005 0.04 0.007 0.168 324 
 Combination 20.07 0.01 0.0005 0.06 0.002 0.030 61 

Whole-exome sequencing identifies known mutations associated with growth regulation

In order to identify genetic alterations associated with these observations, we identified 281 nonsynonymous gene mutations in CUHM003 patient's tumor (Supplementary Table S3A). By the time of tumor progression after immunotherapy, mutations in 187 additional genes were present (Supplementary Table S3B). There were mutations in 99 genes in the CUHM005 patient tumor at diagnosis and 262 additional gene mutations upon progression (Supplementary Table S3C and S3D). Because neither of these patients initially responded to a PD-1 inhibitor, we identified mutant genes present in both tumors, which might be relevant to the observed treatment failure. At diagnosis, the tumors shared mutations in 11 genes (Supplementary Fig. S8A), and after progression, they shared an additional 64 mutations (highlighted in Supplementary Table S3A and S3D). Of these genes, known NRAS mutations, previously described in the Catalogue of Somatic Mutations in Cancer (COSMIC) and shown to be activating and oncogenic, seemed particularly noteworthy (Supplementary Fig. S8B; ref. 34). Although there is some correlation between NRAS mutations and a response to a PD-1 inhibitor (35), mutations in NRAS are also known to activate the transcription factor STAT3, a downstream component of the IFNγ-stimulated Jak–STAT pathway and may be associated with rapid tumor growth after anti–PD-1 treatment (36).

Transcriptome analysis defines a specific response to interferon stimulation in aHM tumors

We next examined RNA expression in these tumors by next-generation sequencing. When genes expressed only in the HM and corresponding patient tumors were analyzed using the NIH DAVID, the transcriptomes of both CUHM003 and CUHM005 humanized mice tumors were enriched in immunity-related genes, as previously observed (Supplementary Table S4; Supplementary Table S5A–S5D; ref. 13). No significant differences in the transcriptomic fingerprint between mHM and aHM tumors existed (differential expression analysis; adjusted P = 0.006), an observation that supports the higher relative activity of aHM immune cells, given the differential engraftment in the mHM and aHM cohorts.

To investigate the basis for the accelerated growth in the treated aHM tumors, we analyzed GSEA hallmarks, comparing the patient tumor to and between the NSG, mHM, and aHM models (Supplementary Fig. S9; Supplementary Table S6). The IFNα and IFNγ hallmarks were uniquely upregulated in untreated aHM003 tumors, indicating their expression reverts to a more similar state to that observed in the patient. We next observed that CTLA4 and/or PD-1 inhibition resulted in upregulation of IFNα and IFNγ pathways in both NSG003 and NSG005 as well as upregulation of IFNα in mHM003 (Fig. 3A). In contrast, CTLA4 and/or PD-1 inhibition led to downregulation of the IFNα and IFNγ pathways in aHM003 and aHM005 tumors (Fig. 3A; Supplementary Table S7). A comparison of the normalized enrichment score of the GSEA hallmarks between control and PD-1 inhibitor–treated tumors depicts the reduction in the IFNα hallmark after treatment in aHM003, whereas both IFNα and IFNγ are downregulated in mHM005 and aHM005 (Fig. 3B). In CUHM005 tumors after combination treatment, only the EMT and TNFα pathways rise in aHM005 compared with NSG005 and/or mHM005 (Fig. 3C), suggesting IFN signaling is specifically modulated by PD-1 inhibition.

Figure 3.

Transcriptome data identifying patterns of gene expression in tumors from humanized mice. A, When GSEA hallmark changes due to immunotherapy treatment are considered, IFNα and IFNγ responses rise in NSG and mHM models but fall in aHM, indicating that the genes in this pathway are regulated in a different manner than that occurring in the other models. B, Expression change plot for each GSEA hallmark, showing the relative change in expression after anti–PD-1 treatment for tumors in NSG, mHM, and aHM. Solid points have an adjusted FDR q < 0.05 and shaded points are <0.10, as determined by a two-sided modified Kolmogorov–Smirnov permutation test. Significant changes can be detected in the IFNα and IFNγ responses. C, Similar expression change plot for CUHM005 showing GSEA changes after combination therapy. Although the EMT and TNFα hallmarks rise in aHM, no further changes in IFN expression are noted.

Figure 3.

Transcriptome data identifying patterns of gene expression in tumors from humanized mice. A, When GSEA hallmark changes due to immunotherapy treatment are considered, IFNα and IFNγ responses rise in NSG and mHM models but fall in aHM, indicating that the genes in this pathway are regulated in a different manner than that occurring in the other models. B, Expression change plot for each GSEA hallmark, showing the relative change in expression after anti–PD-1 treatment for tumors in NSG, mHM, and aHM. Solid points have an adjusted FDR q < 0.05 and shaded points are <0.10, as determined by a two-sided modified Kolmogorov–Smirnov permutation test. Significant changes can be detected in the IFNα and IFNγ responses. C, Similar expression change plot for CUHM005 showing GSEA changes after combination therapy. Although the EMT and TNFα hallmarks rise in aHM, no further changes in IFN expression are noted.

Close modal

Unbiased analyses using the Kyoto Encyclopedia of Genes and Genomes (KEGG) revealed both positively and negatively enriched components on the IFNγ and Jak–STAT pathways (Supplementary Fig. S10A and S10B). DAVID analyses showed enrichment in chemokine signaling pathway genes (P = 0.01) in aHM003, whereas the expression of genes participating in the PI3K–Akt and IFNγ signaling pathways was enriched in aHM005 (P < 0.05; Supplementary Table S8A and S8B). Heat maps of these gene sets emphasize the bidirectional change in expression in aHM tumors (Supplementary Fig. S10C), supporting the role of the IFNγ pathway in mediating tumor growth in response to PD-1 inhibition.

Ex vivo analysis with tumor-derived cell lines

To investigate the basis of the differential response to anti–PD-1 treatment in the aHM tumors, and given the limitation to establishing further aHM cohorts, we established cell lines from early passages of the CUHM003 and CUHM005 PDX by cell sorting. Because in vivo growth corresponded with reduced T-cell presence, we replicated the conditions caused by this phenomenon by transfecting both cell lines with an siRNA construct against the IFNγ receptor (IFNγR2; Supplementary Fig. S11), thereby desensitizing cells to the IFNγ produced by active T cells. We seeded 200,000 cells on low-adherence plates to encourage spheroid formation. Under baseline conditions, both cell lines behaved similarly: the addition of purified IFNγ had minimal effect on final spheroid size, but knockdown of IFNγR2 dramatically increased their size (CUHM003 P < 0.001, CUHM005 P = 0.006; Fig. 4A). The addition of activated T cells led to a slight increase in average spheroid size, blunting the inhibitory effect of IFNγR2 knockdown (Fig. 4B). The addition of a PD-1 inhibitor did not further affect sphere size (Fig. 4C and D; representative images in Fig. 4E), indicating that cell growth for both CUHM003 and CUHM005 is directed primarily by the presence of a functioning IFNγR2, and that IFNγ depletion (either by disappearance of autologous T cells or by absence of its receptor) increases cell growth.

Figure 4.

In vitro response to depleted IFNγ signaling. A, CUHM003 and CUHM005 cell lines were transfected with siRNA against IFNγ receptor 2 (IFNGR2) then cultured in triplicate as spheroids in the presence of IFNγ and their sizes recorded and averaged (n ∼ 120 colonies per well; 3 wells per cell line per condition). Although the addition of IFNγ only minimally affected average spheroid size, cells transfected with an siRNA against the IFNγ receptor formed significantly larger spheroids (ANOVA P = 0.001; CUHM003, *, P < 0.001; CUHM005, **, P = 0.006 by a Dunnett comparisons test for this and subsequent comparisons). B, The addition of activated mismatched T cells had a minimal effect on spheroid size but did not alter the observed size increase after IFNγ receptor knockdown (ANOVA P = 0.005; CUHM003, ***, P < 0.001; CUHM005, ****, P = 0.01). C, Likewise, the addition of a PD-1 inhibitor (pembrolizumab for CUHM003 or nivolumab for CUHM005) did not change spheroid size and did not alter the effect of the IFNγ receptor knockdown (ANOVA P < 0.001; CUHM003, +, P < 0.001; CUHM005, ++, P = 0.001). D, The addition of both T cells and the corresponding PD-1 inhibitor also only marginally changed sphere size and did not alter the increase in spheroid size induced by IFNγ receptor siRNA knockdown (ANOVA P < 0.001; CUHM003, +++, P < 0.001; CUHM005, ++++, P < 0.001). E, Representative images showing the comparative sizes of spheroids, spheroids after the addition of IFNγ, and spheroids after transfection with an siRNA against the IFNγ receptor (top), as well as observed changes in size resulting from the addition of T cells (second panel), a PD-1 inhibitor (third panel), or of T cells plus a PD-1 inhibitor (bottom). Scale bar = 100 μm.

Figure 4.

In vitro response to depleted IFNγ signaling. A, CUHM003 and CUHM005 cell lines were transfected with siRNA against IFNγ receptor 2 (IFNGR2) then cultured in triplicate as spheroids in the presence of IFNγ and their sizes recorded and averaged (n ∼ 120 colonies per well; 3 wells per cell line per condition). Although the addition of IFNγ only minimally affected average spheroid size, cells transfected with an siRNA against the IFNγ receptor formed significantly larger spheroids (ANOVA P = 0.001; CUHM003, *, P < 0.001; CUHM005, **, P = 0.006 by a Dunnett comparisons test for this and subsequent comparisons). B, The addition of activated mismatched T cells had a minimal effect on spheroid size but did not alter the observed size increase after IFNγ receptor knockdown (ANOVA P = 0.005; CUHM003, ***, P < 0.001; CUHM005, ****, P = 0.01). C, Likewise, the addition of a PD-1 inhibitor (pembrolizumab for CUHM003 or nivolumab for CUHM005) did not change spheroid size and did not alter the effect of the IFNγ receptor knockdown (ANOVA P < 0.001; CUHM003, +, P < 0.001; CUHM005, ++, P = 0.001). D, The addition of both T cells and the corresponding PD-1 inhibitor also only marginally changed sphere size and did not alter the increase in spheroid size induced by IFNγ receptor siRNA knockdown (ANOVA P < 0.001; CUHM003, +++, P < 0.001; CUHM005, ++++, P < 0.001). E, Representative images showing the comparative sizes of spheroids, spheroids after the addition of IFNγ, and spheroids after transfection with an siRNA against the IFNγ receptor (top), as well as observed changes in size resulting from the addition of T cells (second panel), a PD-1 inhibitor (third panel), or of T cells plus a PD-1 inhibitor (bottom). Scale bar = 100 μm.

Close modal

To dissect intracellular signaling associated with IFNγ, we interrogated STAT1 and STAT3, because these are the primary cellular mediators of extracellular IFNγ and both modulate cancer growth (37). The ratio of STAT3/STAT1 RNA-seq transcripts closely mirrored the tumor growth rates (Fig. 5A) although, much as observed during our analysis of the changes in the GSEA hallmarks (Fig. 3C), additional signals appear to have a role in driving tumor growth in the CUHM005 aHM combination tumor. The transcription of STAT1- and STAT3-dependent genes with or without PD-1 inhibitor across mouse model tumors showed that STAT1-mediated transcription dropped whereas STAT3-mediated transcription increased only in aHM tumors (Fig. 5B). Because STAT3 is abundantly expressed and thus difficult to quantify, we examined its role using tumor-derived cell lines in which a transfected siSTAT3 construct reduced its expression. In accordance with our observations using IFNγR2 knockdowns, we observed that the addition of exogenous IFNγ to these CUHM003 and CUHM005 cell lines led to a notable increase in pSTAT1 expression and a decrease in pSTAT3 expression (Fig. 5C and D; cell lines), the combination of which would inhibit cellular proliferation. We observed a similar reduction in pSTAT3 in PD-1 inhibitor–treated aHM tumors, although it was uniquely coupled to a concurrent pSTAT1 reduction (Fig. 5C and D; CUHM003-CUHM005), creating a distinct environment in which cellular proliferation increased without pSTAT3 activation. A cytokine array comparing mHM and aHM plasma from mice with PD-1 inhibitor–treated CUHM003 and CUHM005 tumors provides a possible clue to the mechanism driving this proliferation (Supplementary Fig. S12). The chemokine CXCL12, whose downregulation has been previously shown to directly activate MAPK signaling in NRAS-mutated melanoma, is markedly reduced in PD-1 inhibitor–treated aHM plasma (38). We can recapitulate such a regulatory environment in the cell lines in which the JAK–STAT pathway has been activated by IFNγ and in which STAT3 transcription has been reduced (Fig. 5D; comparing the reduction in pSTAT3 expression after IFNγ + siSTAT3 treatment in the cell lines with that observed after PD-1 inhibition in the tumors).

Figure 5.

Ex vivo examination of Jak–STAT pathway alterations in tumor hyperprogression. A, The FPKM values for biological duplicates the STAT3 and STAT1 transcripts were used to determine the ratio of STAT3/STAT1 gene expression in the NSG, mHM, and aHM CUHM003 and CUHM005 control and treated tumors. The pattern of the STAT3/STAT1 ratio corresponds to the pattern of tumor growth observed in the mouse models. B, A compilation of the FPKM values of the genes responsive to STAT1 and STAT3 activation shows changes in STAT1 and STAT3 regulation after PD-1 inhibitor treatment in NSG, mHM, and aHM. Transcription data from both CUHM003 and CUHM005 tumors were combined for each mouse group for this analysis. C, Representative western blots of patient-derived CUHM003 and CUHM005 cell lines (left) or PDX tumor tissue (right), showing the effects of the addition of IFNγ and/or expression of an siRNA STAT3 construct on STAT1 and STAT3 protein expression and phosphorylation (in the cell lines) or the effects of PD-1 inhibitor (in the PDX). pSTAT3 expression is calculated based on total STAT3 expression. D, Densitometry averaged from triplicate western blots shows that pSTAT3 expression (as a fraction of STAT3 expression) decreases after the application of IFNγ to cell lines expressing an siRNA STAT3 construct (final 2 bars of the left chart) and in aHM tumors treated with pembrolizumab (CUHM003; middle chart) or nivolumab (CUHM005; right chart).

Figure 5.

Ex vivo examination of Jak–STAT pathway alterations in tumor hyperprogression. A, The FPKM values for biological duplicates the STAT3 and STAT1 transcripts were used to determine the ratio of STAT3/STAT1 gene expression in the NSG, mHM, and aHM CUHM003 and CUHM005 control and treated tumors. The pattern of the STAT3/STAT1 ratio corresponds to the pattern of tumor growth observed in the mouse models. B, A compilation of the FPKM values of the genes responsive to STAT1 and STAT3 activation shows changes in STAT1 and STAT3 regulation after PD-1 inhibitor treatment in NSG, mHM, and aHM. Transcription data from both CUHM003 and CUHM005 tumors were combined for each mouse group for this analysis. C, Representative western blots of patient-derived CUHM003 and CUHM005 cell lines (left) or PDX tumor tissue (right), showing the effects of the addition of IFNγ and/or expression of an siRNA STAT3 construct on STAT1 and STAT3 protein expression and phosphorylation (in the cell lines) or the effects of PD-1 inhibitor (in the PDX). pSTAT3 expression is calculated based on total STAT3 expression. D, Densitometry averaged from triplicate western blots shows that pSTAT3 expression (as a fraction of STAT3 expression) decreases after the application of IFNγ to cell lines expressing an siRNA STAT3 construct (final 2 bars of the left chart) and in aHM tumors treated with pembrolizumab (CUHM003; middle chart) or nivolumab (CUHM005; right chart).

Close modal

The discovery of immunotherapies targeting CTLA4 and PD-1 is largely responsible for the improvement of metastatic melanoma outcomes, where one-year survival rates went from less than 25% (39) to between 47% and 63% (40, 41). These successes have paved the way in deploying these therapies in other tumor types, such as lung, colon, bladder, and head and neck cancers (42). Although effective, only 30% to 40% of patients respond to immune-directed treatment and even responders will often eventually acquire resistance (43). Furthermore, recent evidence indicates that these immunotherapies can prompt a rapid acceleration in tumor growth, a condition known as hyperprogression (8, 9). Although there has been some initial success in HM created from patient peripheral lymphocytes and tumor tissue (44, 45), no model exists in which treatment effectiveness can be predicted, and where the basis of tumor response can be explored (46). The aHM PDX model reported here addresses some of those limitations.

The aHM is based on the initial mHM models (13, 14) that were first deployed to study the relationship between an implanted tumor and the engrafted human immune system of the host mouse. Using the mHM, we showed that human immune cells infiltrated the tumor, upregulated human cytokine production, and partially reversed the expression in many of the tumor's immune-, EMT-, and extracellular matrix–related genes (13). We also demonstrated that this environment was responsive to immune modulation, and we observed that adequate humanization was required to achieve in vivo efficacy with PD-1 inhibitors (14). However, because mHM were generated from donated cord blood, the engrafted immune system is allogeneic with the implanted tumor tissue, a significant caveat in faithfully representing immune-directed therapy results or in guiding patient therapy. An HLA-matched immune system is uniquely necessary to prevent indiscriminate immune cell attack after xenograft implantation (47, 48). A more elegant technology is needed to recapitulate an immune system in an HM PDX model (49).

To address these caveats, we generated an aHM model using HSPCs isolated from the blood of metastatic melanoma patients. The development of the aHM in melanoma was driven by the relevance of immunity in melanoma, the accessibility of tissue for PDX generation, and the possibility of validating immunotherapy results against those observed in the originating patients. We also identified and cultured a second population of cells within these HSPCs with characteristics of MSCs, because the presence of MSCs within cultured HSPCs promotes superior mouse engraftment and increases immune system reconstitution (50, 51). This process has previously been shown to double the human precursors in the bone marrow of mHM and increase by over 10-fold the circulating immune cells and tumor-infiltrating T cells (14).

A first notable conclusion from this work is that the engraftment of patient HSPCs gave rise to an aHM model in which fewer HSPCs produced more active immune cells, as seen in clinical settings (33). Smaller numbers of autologous human cells were capable of changing the gene expression of implanted tumors. This indicates that the size of the human cell population may not be the only factor in determining humanization, and an autologous source of HSPCs may be critical in determining both engraftment and functionality. These differences in HSPC origin have profound implications in the deployment of aHM models, and their feasibility to further personalized therapies.

A second critical observation is that tumorigenicity was significantly diminished in aHM models. In immunotherapy-refractory melanoma, T-cell invasion decreases, and a switch to an “immune cold” TME is a hallmark of tumor progression and unresponsiveness to immunotherapy. The mHM003 and mHM005 cohorts had prominent populations of human CD45+ cells, and their tumors had significant T-cell infiltration, but this had little effect on their growth (compared with NSG controls). Conversely, even though aHM003 and aHM005 had fewer circulating human immune cells, tumors on these mice appeared later and grew more slowly, as did those implanted on mHM003-HLA cohorts. This differential growth may be a consequence of varying immune cell activity within these models, suggesting that even partially effective immune surveillance leads to a delay in tumor formation and a lag in subsequent growth.

A third key finding is the immunotherapy-dependent accelerated tumor growth in aHM models, which can help identify factors responsible for tumor resistance or hyperprogression in patients. The expression of IFNγ-related pathways was highest in untreated aHM tumors but markedly reduced after treatment with PD-1 inhibitors, concordant with a relative decrease in infiltrating immune cells, when compared with those in the corresponding mHM models. This suggests that the rapid growth observed in these tumors was likely a consequence of the immunotherapy-induced disruption of a small population of active T cells. The basis of this paradoxical effect remains unidentified, but a similar reduction in T cells was observed in posttreatment patient tissues.

It remains unclear whether the treatment-induced tumor growth in the aHM represents true patient hyperprogression or is simply reminiscent of the rapid progression of both patients subsequent to their treatment, because the tumor tissue and patient HSPCs were acquired after the patients had demonstrated resistance to both CTLA4 and PD-1 inhibitors. Even though therapy was suspended after the onset of rapid progression and before either patient exhibited a classic hyperprogressive profile, growth of both of their tumors accelerated markedly after treatment. Hyperprogression was, however, observed in the majority of treated tumors in both aHM cohorts of a well-controlled study, and it is unlikely to be a coincidental effect. This type of comparison has not previously been possible, given the absence of suitable animal models.

To guide our investigation of the mechanism driving this treatment-induced growth, we derived cell lines from both melanoma patients. Analysis of comprehensive spheroid experiments strongly indicates that their growth is regulated by IFNγ. When the IFNγR2 receptor was knocked down, cells increased their growth, mirroring the in vivo observation that a PD-1 inhibitor reduced T-cell presence and facilitated subsequent tumor progression. Likewise, our observation that pSTAT3 falls when exogenous IFNγ cannot stimulate the STAT3 phosphorylation in CUHM003 and CUHM005 siSTAT3 cell lines mirrors how pSTAT3 falls in response to a PD-1 inhibitor in aHM and highlights the role of IFNγ and Jak–STAT signaling in the rapid tumor growth observed in aHM. The notable reduction in CXCL12 in aHM tumors treated with a PD-1 inhibitor implies that this rapid growth may be driven by NRAS-mediated MAPK activation. Such a model is appealing, because CXCR4–CXCL12 signaling can be mediated by T cells, which we have shown to have increased activity in the aHM model. Their withdrawal from the TME subsequent to treatment with a PD-1 inhibitor would not only reduce IFNγ presence but would abrogate CXCR4–CXCL12 signaling, decreasing Jak–STAT pathway activity and removing its checks on MAPK signaling via the mutated NRAS present in these tumors (38, 52). This idea may also explain why STAT3-dependent genes are still elevated in aHM tumors when STAT3 phosphorylation has decreased, because many of these genes can also be regulated through the MAPK pathway (Fig. 5B vs. Fig. 5D). Work toward elucidating the molecular mechanism driving this phenomenon is ongoing in aHM models.

A more complete understanding of the TME environment in the aHM model will increase our understanding of immunotherapy resistance and tumor hyperprogression observed in cancer patients. It is also crucial that additional aHM PDX models be constructed in order to compare their tumor growth with that of treatment-naïve or susceptible patients. These results support the development and use of the aHM model, showing that it can recapitulate many aspects of the patient TME and also be used to examine uncharacterized tumor growth dynamics. The aHM reported here led to 3 noteworthy observations: even when created utilizing fewer adult patient-derived HSPCs and with less bone marrow engraftment, aHM gave rise to proportionally more circulating and tumor infiltrative immune cells than cord blood–derived mHM; greater HLA matching between immune cells and the tumor resulted in slower tumor initiation and growth; and tumors engrafted at the time of progression to immune therapy can paradoxically respond with increased growth upon continuing exposure to such therapy.

S.B. Keysar, J. Reisinger, P.N. Le, K. Gomez, and B. Miller report grants from NIH during the conduct of the study. X.-J. Wang reports grants from VA during the conduct of the study and grants from Allander Biotechnologies, LLC outside the submitted work. A. Jimeno reports grants from Department of Defense and NCI during the conduct of the study and other from SuviCa and Champions Oncology outside the submitted work. No disclosures were reported by the other authors.

J.J. Morton: Conceptualization, formal analysis, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing. N. Alzofon: Formal analysis. S.B. Keysar: Conceptualization, methodology, writing–review and editing. T.-S. Chimed: Investigation. J. Reisinger: Investigation. L. Perrenoud: Investigation. P.N. Le: Investigation, writing–review and editing. C. Nieto: Writing–review and editing. K. Gomez: Writing–review and editing. B. Miller: Investigation. R. Yeager: Investigation. D. Gao: Formal analysis. A.-C. Tan: Resources. H. Somerset: Resources. T. Medina: Resources. X.-J. Wang: Funding acquisition, writing–review and editing. J.H. Wang: Writing–review and editing. W. Robinson: Supervision. D.R. Roop: Writing–review and editing. R. Gonzalez: Supervision. A. Jimeno: Conceptualization, formal analysis, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing.

The authors wish to thank the patients who donated their tissue, blood, and time, and to the clinical teams who facilitated patient informed consent, as well as sample and data acquisition. This work was primarily supported by NIH grants R01CA149456 (A. Jimeno), R01CA213102 (A. Jimeno), R01DE024371 (A. Jimeno and X.-J. Wang), P30-CA046934 (University of Colorado Cancer Center Support Grant), P30-AR057212 (University of Colorado Skin Diseases Research Center Support Grant), Ruth L. Kirschstein National Research Service Award T32CA17468 (X.-J. Wang; P.N. Le, trainee), Training in Otolaryngology Research T32DC012280 (C. Nieto, trainee), ACS-IRG 16-184-56 Institutional American Cancer Society (J. Morton), the Daniel and Janet Mordecai Foundation (A. Jimeno), the Karsh family Foundation (A. Jimeno), and the Peter and Rhonda Grant Foundation (A. Jimeno).

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.

1.
Hodi
FS
,
O'Day
SJ
,
McDermott
DF
,
Weber
RW
,
Sosman
JA
,
Haanen
JB
, et al
Improved survival with ipilimumab in patients with metastatic melanoma
.
N Engl J Med
2010
;
363
:
711
23
.
2.
Barone
A
,
Hazarika
M
,
Theoret
MR
,
Mishra-Kalyani
P
,
Chen
H
,
He
K
, et al
FDA approval summary: pembrolizumab for the treatment of patients with unresectable or metastatic melanoma
.
Clin Cancer Res
2017
;
23
:
5661
5
.
3.
Hazarika
M
,
Chuk
MK
,
Theoret
MR
,
Mushti
S
,
He
K
,
Weis
SL
, et al
U.S. FDA approval summary: nivolumab for treatment of unresectable or metastatic melanoma following progression on ipilimumab
.
Clin Cancer Res
2017
;
23
:
3484
8
.
4.
Shin
DS
,
Zaretsky
JM
,
Escuin-Ordinas
H
,
Garcia-Diaz
A
,
Hu-Lieskovan
S
,
Kalbasi
A
, et al
primary resistance to PD-1 blockade mediated by JAK1/2 mutations
.
Cancer Discov
2017
;
7
:
188
201
.
5.
Luo
N
,
Formisano
L
,
Gonzalez-Ericsson
PI
,
Sanchez
V
,
Dean
PT
,
Opalenik
SR
, et al
Melanoma response to anti-PD-L1 immunotherapy requires JAK1 signaling, but not JAK2
.
Oncoimmunology
2018
;
7
:
e1438106
.
6.
Ayers
M
,
Lunceford
J
,
Nebozhyn
M
,
Murphy
E
,
Loboda
A
,
Kaufman
DR
, et al
IFN-gamma-related mRNA profile predicts clinical response to PD-1 blockade
.
J Clin Invest
2017
;
127
:
2930
40
.
7.
Manguso
RT
,
Pope
HW
,
Zimmer
MD
,
Brown
FD
,
Yates
KB
,
Miller
BC
, et al
In vivo CRISPR screening identifies Ptpn2 as a cancer immunotherapy target
.
Nature
2017
;
547
:
413
8
.
8.
Champiat
S
,
Dercle
L
,
Ammari
S
,
Massard
C
,
Hollebecque
A
,
Postel-Vinay
S
, et al
Hyperprogressive disease is a new pattern of progression in cancer patients treated by anti-PD-1/PD-L1
.
Clin Cancer Res
2017
;
23
:
1920
8
.
9.
Kato
S
,
Goodman
A
,
Walavalkar
V
,
Barkauskas
DA
,
Sharabi
A
,
Kurzrock
R
. 
Hyperprogressors after immunotherapy: analysis of genomic alterations associated with accelerated growth rate
.
Clin Cancer Res
2017
;
23
:
4242
50
.
10.
Hugo
W
,
Zaretsky
JM
,
Sun
L
,
Song
C
,
Moreno
BH
,
Hu-Lieskovan
S
, et al
Genomic and transcriptomic features of response to anti-PD-1 therapy in metastatic melanoma
.
Cell
2017
;
168
:
542
.
11.
Sharon
E
. 
Can an immune checkpoint inhibitor (sometimes) make things worse?
Clin Cancer Res
2017
;
23
:
1879
81
.
12.
Lulli
D
,
Carbone
ML
,
Pastore
S
. 
Epidermal growth factor receptor inhibitors trigger a type I interferon response in human skin
.
Oncotarget
2016
;
7
:
47777
93
.
13.
Morton
JJ
,
Bird
G
,
Keysar
SB
,
Astling
DP
,
Lyons
TR
,
Anderson
RT
, et al
XactMice: humanizing mouse bone marrow enables microenvironment reconstitution in a patient-derived xenograft model of head and neck cancer
.
Oncogene
2016
;
35
:
290
300
.
14.
Morton
JJ
,
Keysar
SB
,
Perrenoud
L
,
Chimed
TS
,
Reisinger
J
,
Jackson
B
, et al
Dual use of hematopoietic and mesenchymal stem cells enhances engraftment and immune cell trafficking in an allogeneic humanized mouse model of head and neck cancer
.
Mol Carcinog
2018
;
57
:
1651
63
.
15.
Spranger
S
,
Spaapen
RM
,
Zha
Y
,
Williams
J
,
Meng
Y
,
Ha
TT
, et al
Up-regulation of PD-L1, IDO, and T(regs) in the melanoma tumor microenvironment is driven by CD8(+) T cells
.
Sci Transl Med
2013
;
5
:
200ra116
.
16.
Spranger
S
,
Bao
R
,
Gajewski
TF
. 
Melanoma-intrinsic beta-catenin signalling prevents anti-tumour immunity
.
Nature
2015
;
523
:
231
5
.
17.
Keysar
SB
,
Astling
DP
,
Anderson
RT
,
Vogler
BW
,
Bowles
DW
,
Morton
JJ
, et al
A patient tumor transplant model of squamous cell cancer identifies PI3K inhibitors as candidate therapeutics in defined molecular bins
.
Mol Oncol
2013
;
7
:
776
90
.
18.
Bolger
AM
,
Lohse
M
,
Usadel
B
. 
Trimmomatic: a flexible trimmer for Illumina sequence data
.
Bioinformatics
2014
;
30
:
2114
20
.
19.
Hintzsche
J
,
Kim
J
,
Yadav
V
,
Amato
C
,
Robinson
SE
,
Seelenfreund
E
, et al
IMPACT: a whole-exome sequencing analysis pipeline for integrating molecular profiles with actionable therapeutics in clinical samples
.
J Am Med Inform Assoc
2016
;
23
:
721
30
.
20.
Sherry
ST
,
Ward
MH
,
Kholodov
M
,
Baker
J
,
Phan
L
,
Smigielski
EM
, et al
dbSNP: the NCBI database of genetic variation
.
Nucleic Acids Res
2001
;
29
:
308
11
.
21.
1000 Genomes Project Consortium
,
Auton
A
,
Brooks
LD
,
Durbin
RM
,
Garrison
EP
,
Kang
HM
, et al
A global reference for human genetic variation
.
Nature
2015
;
526
:
68
74
.
22.
Trapnell
C
,
Roberts
A
,
Goff
L
,
Pertea
G
,
Kim
D
,
Kelley
DR
, et al
Differential gene and transcript expression analysis of RNA-seq experiments with TopHat and Cufflinks
.
Nat Protoc
2012
;
7
:
562
78
.
23.
Mootha
VK
,
Lindgren
CM
,
Eriksson
KF
,
Subramanian
A
,
Sihag
S
,
Lehar
J
, et al
PGC-1alpha-responsive genes involved in oxidative phosphorylation are coordinately downregulated in human diabetes
.
Nat Genet
2003
;
34
:
267
73
.
24.
Liberzon
A
,
Birger
C
,
Thorvaldsdottir
H
,
Ghandi
M
,
Mesirov
JP
,
Tamayo
P
. 
The molecular signatures database (MSigDB) hallmark gene set collection
.
Cell Syst
2015
;
1
:
417
25
.
25.
Yu
G
,
Wang
LG
,
Han
Y
,
He
QY
. 
clusterProfiler: an R package for comparing biological themes among gene clusters
.
OMICS
2012
;
16
:
284
7
.
26.
Love
MI
,
Huber
W
,
Anders
S
. 
Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2
.
Genome Biol
2014
;
15
:
550
.
27.
Huang da
W
,
Sherman
BT
,
Lempicki
RA
. 
Systematic and integrative analysis of large gene lists using DAVID bioinformatics resources
.
Nat Protoc
2009
;
4
:
44
57
.
28.
Huang da
W
,
Sherman
BT
,
Lempicki
RA
. 
Bioinformatics enrichment tools: paths toward the comprehensive functional analysis of large gene lists
.
Nucleic Acids Res
2009
;
37
:
1
13
.
29.
Le
PN
,
Keysar
SB
,
Miller
B
,
Eagles
JR
,
Chimed
TS
,
Reisinger
J
, et al
Wnt signaling dynamics in head and neck squamous cell cancer tumor-stroma interactions
.
Mol Carcinog
2019
;
58
:
398
410
.
30.
Keysar
SB
,
Le
PN
,
Miller
B
,
Jackson
BC
,
Eagles
JR
,
Nieto
C
, et al
Regulation of head and neck squamous cancer stem cells by PI3K and SOX2
.
J Natl Cancer Inst
2017
;
109
:
djw189
.
31.
Jimeno
A
,
Feldmann
G
,
Suarez-Gauthier
A
,
Rasheed
Z
,
Solomon
A
,
Zou
GM
, et al
A direct pancreatic cancer xenograft model as a platform for cancer stem cell therapeutic development
.
Mol Cancer Ther
2009
;
8
:
310
4
.
32.
Sugimoto
K
,
Adachi
Y
,
Moriyama
K
,
Qiong
W
,
Nakayama
A
,
Hosono
M
, et al
Induction of the expression of SCF in mouse by lethal irradiation
.
Growth Factors
2001
;
19
:
219
31
.
33.
Barker
JN
,
Scaradavou
A
,
Stevens
CE
. 
Combined effect of total nucleated cell dose and HLA match on transplantation outcome in 1061 cord blood recipients with hematologic malignancies
.
Blood
2010
;
115
:
1843
9
.
34.
Burd
CE
,
Liu
W
,
Huynh
MV
,
Waqas
MA
,
Gillahan
JE
,
Clark
KS
, et al
Mutation-specific RAS oncogenicity explains NRAS codon 61 selection in melanoma
.
Cancer Discov
2014
;
4
:
1418
29
.
35.
Johnson
DB
,
Lovly
CM
,
Flavin
M
,
Panageas
KS
,
Ayers
GD
,
Zhao
Z
, et al
Impact of NRAS mutations for patients with advanced melanoma treated with immune therapies
.
Cancer Immunol Res
2015
;
3
:
288
95
.
36.
Wang
Y
,
Velho
S
,
Vakiani
E
,
Peng
S
,
Bass
AJ
,
Chu
GC
, et al
Mutant N-RAS protects colorectal cancer cells from stress-induced apoptosis and contributes to cancer development and progression
.
Cancer Discov
2013
;
3
:
294
307
.
37.
Avalle
L
,
Pensa
S
,
Regis
G
,
Novelli
F
,
Poli
V
. 
STAT1 and STAT3 in tumorigenesis: a matter of balance
.
JAKSTAT
2012
;
1
:
65
72
.
38.
McConnell
AT
,
Ellis
R
,
Pathy
B
,
Plummer
R
,
Lovat
PE
,
O'Boyle
G
. 
The prognostic significance and impact of the CXCR4-CXCR7-CXCL12 axis in primary cutaneous melanoma
.
Br J Dermatol
2016
;
175
:
1210
20
.
39.
Cowey
CL
,
Liu
FX
,
Black-Shinn
J
,
Stevinson
K
,
Boyd
M
,
Frytak
JR
, et al
Pembrolizumab utilization and outcomes for advanced melanoma in US community oncology practices
.
J Immunother
2018
;
41
:
86
95
.
40.
Achkar
T
,
Tarhini
AA
. 
The use of immunotherapy in the treatment of melanoma
.
J Hematol Oncol
2017
;
10
:
88
.
41.
Heppt
MV
,
Steeb
T
,
Schlager
JG
,
Rosumeck
S
,
Dressler
C
,
Ruzicka
T
, et al
Immune checkpoint blockade for unresectable or metastatic uveal melanoma: a systematic review
.
Cancer Treat Rev
2017
;
60
:
44
52
.
42.
Liu
B
,
Song
Y
,
Liu
D
. 
Recent development in clinical applications of PD-1 and PD-L1 antibodies for cancer immunotherapy
.
J Hematol Oncol
2017
;
10
:
174
.
43.
Bai
J
,
Gao
Z
,
Li
X
,
Dong
L
,
Han
W
,
Nie
J
. 
Regulation of PD-1/PD-L1 pathway and resistance to PD-1/PD-L1 blockade
.
Oncotarget
2017
;
8
:
110693
707
.
44.
Jespersen
H
,
Lindberg
MF
,
Donia
M
,
Soderberg
EMV
,
Andersen
R
,
Keller
U
, et al
Clinical responses to adoptive T-cell transfer can be modeled in an autologous immune-humanized mouse model
.
Nat Commun
2017
;
8
:
707
.
45.
Lin
S
,
Huang
G
,
Cheng
L
,
Li
Z
,
Xiao
Y
,
Deng
Q
, et al
Establishment of peripheral blood mononuclear cell-derived humanized lung cancer mouse models for studying efficacy of PD-L1/PD-1 targeted immunotherapy
.
MAbs
2018
;
10
:
1301
11
.
46.
Morton
JJ
,
Alzofon
N
,
Jimeno
A
. 
The humanized mouse: Emerging translational potential
.
Mol Carcinog
2020
;
59
:
830
8
.
47.
Danner
R
,
Chaudhari
SN
,
Rosenberger
J
,
Surls
J
,
Richie
TL
,
Brumeanu
TD
, et al
Expression of HLA class II molecules in humanized NOD.Rag1KO.IL2RgcKO mice is critical for development and function of human T and B cells
.
PLoS One
2011
;
6
:
e19826
.
48.
Patton
J
,
Vuyyuru
R
,
Siglin
A
,
Root
M
,
Manser
T
. 
Evaluation of the efficiency of human immune system reconstitution in NSG mice and NSG mice containing a human HLA.A2 transgene using hematopoietic stem cells purified from different sources
.
J Immunol Methods
2015
;
422
:
13
21
.
49.
Morton
JJ
,
Bird
G
,
Refaeli
Y
,
Jimeno
A
. 
Humanized mouse xenograft models: narrowing the tumor-microenvironment gap
.
Cancer Res
2016
;
76
:
6153
8
.
50.
Chen
P
,
Huang
Y
,
Womer
KL
. 
Effects of mesenchymal stromal cells on human myeloid dendritic cell differentiation and maturation in a humanized mouse model
.
J Immunol Methods
2015
;
427
:
100
4
.
51.
Coulson-Thomas
VJ
,
Gesteira
TF
,
Hascall
V
,
Kao
W
. 
Umbilical cord mesenchymal stem cells suppress host rejection: the role of the glycocalyx
.
J Biol Chem
2014
;
289
:
23465
81
.
52.
Teicher
BA
,
Fricker
SP
. 
CXCL12 (SDF-1)/CXCR4 pathway in cancer
.
Clin Cancer Res
2010
;
16
:
2927
31
.

Supplementary data