Cathepsin G is broadly expressed in acute myeloid leukemia and is an effective immunotherapeutic target

Immunotherapy is among the novel classes of therapy in development for acute myeloid leukemia (AML) and harnesses the specificity of the immune system to eliminate leukemia cells.¹ This approach builds upon the success of allogeneic hematopoietic stem cell transplant (allo-SCT) in AML, which can be curative in up to 50% of patients, but often at the cost of high rates of treatment-related toxicities, including graft-versus-host disease (GVHD).² To avoid this complication, one approach is to refine the potent graft-versus-leukemia effect to target discrete leukemia antigens in the form of adoptive cellular therapy (ACT). Importantly, in order to overcome intratumoral heterogeneity, durable clinical efficacy will likely require targeting multiple leukemia antigens simultaneously, an approach that is currently limited by the number of known effective leukemia antigens.³

We previously reported that cathepsin G (CG), an azurophil granule protease, represents a novel leukemia-associated antigen (LAA).⁴ CG is largely restricted to the myeloid lineage and, like proteinase 3 (P3) and neutrophil elastase (NE), the sources of the well-established human leukocyte antigen (HLA)-A*0201 (denoted here as HLA-A2) LAA PR1,⁵ CG is contained within the primary granules of maturing and mature neutrophils. However, CG is expressed later in myeloid differentiation, resides under a different promoter than NE and P3, and demonstrates a distinct pattern of expression, suggesting that targeting both PR1 and CG could be synergistic.⁶ In our prior work,⁴ we demonstrated that CG is highly expressed in primary patient AML blasts, AML cell lines and, critically, in leukemia stem cells (LSCs). We showed that CG is localized outside azurophil granules and is ubiquitinated, favoring antigen presentation. In addition, we identified an HLA-A2-restricted epitope within the leader sequence of CG, designated as CG1 (amino-acid sequence FLLPTGAEA) and showed that targeting CG1 resulted in lysis of AML blasts in vitro. Finally, we detected cytotoxic T lymphocytes (CTLs) specific for CG1 in the peripheral blood of AML patients after allo-SCT.⁴

In this study, we further investigate the therapeutic potential of targeting CG, specifically CG1, focusing on the antileukemia activity of CG1-specific CTL in vivo and evaluating their toxicity against normal hematopoietic progenitor cells (HPCs). In addition, we analyze the expression of CG and its presentation by HLA-A2 in primary patient AML blasts through reverse-phase protein array (RPPA) and liquid chromatography (LC)/tandem mass spectrometry (MS/MS).

To determine the role of CG as an immunotherapeutic target in vivo, we first studied a patient-derived xenograft (PDX) treatment model of primary AML (unique patient number, UPN#1; Table 1). After confirming leukemia engraftment (~3 weeks), NOD scid gamma mice were treated intravenously with either 0.5 × 10⁶ CG1-enriched T cell (CG1-CTL) or HIV-CTL, or were left untreated. At the time of killing, bone marrow (BM) was harvested and analyzed for leukemia burden. Mice treated with CG1-CTL had a significantly lower AML burden in the BM (24.94±4.451%; n=13) compared with HIV-CTL-treated mice (61.46±11.07%; n=10) and untreated mice (87.9±5.632%; n=12; P<0.01; Figure 1a). The efficacy of targeting CG in vivo using CG1-CTL was recapitulated in a murine model that incorporated the HLA-A2-transduced U937 myelomonoblastic leukemia cell line (U937-A2; Supplementary Figures S1 and S2).

Table 1 AML patient samples used in LC-MS/MS

Figure 1

Cathepsin G is an effective immunotherapeutic target in vivo and is broadly expressed in AML patients. (a) Irradiated NOD scid gamma mice were injected intravenously with human primary AML blasts (UPN#1; 7 × 10⁶ blasts) on day 0. After confirming leukemia engraftment (~3 weeks), mice were treated with either negative control HIV-enriched T cell (HIV-CTL; 0.5 × 10⁶), CG1-enriched T cell (CG1-CTL; 0.5 × 10⁶) or were left untreated. Mice were killed for all groups when any mouse became moribund or during week 7. The results are expressed as percentage of CD33⁺/CD3⁻ cells from viable hCD45⁺/mCD45⁻ population within the bone marrow. Results reflect four independent experiments; *P<0.01. (b) Bone marrow samples from five healthy donors (HD 1–5) were cultured alone (untreated) or were cocultured with HIV-CTLs or CG1-CTLs at a 1:5 ratio for 4 h in cell media. Cells were then resuspended in methylcellulose semi-solid matrix and cocultured for 7 days. On day 7, colonies were counted and imaged; for HD 3, 4 and 5, colonies were again counted on day 14. Each group was cultured in triplicate and data represent four independent experiments. (c) RPPA was used to quantify protein levels of cathepsin G in blasts from 511 newly diagnosed AML patients (yellow bars) and 21 newly diagnosed APL patients (pink bars). Controls included HD CD34⁺ progenitor cells (n=21, green bars), HD peripheral blood lymphocytes (n=21, blue bars) and granulocyte–macrophage colony-stimulating factor-primed HD CD34⁺ progenitor cells (n=10, red bars). (d) Cathepsin G levels were assessed for 47 patients at diagnosis and relapse, and were compared by paired t-test (P=0.0001). (e) Kaplan–Meier plots showing OS in AML patients (n=415) comparing patients with high CG protein expression by RPPA (upper 2/3) to patients with low CG expression (lowest 1/3). Results are significant by Cox univariate model testing (P=0.04).

We next sought to confirm that the CG1 peptide is presented on the HLA-A2⁺ AML cell surface. We used W6/32 antibody to immunoprecipitate surface HLA Class I molecules and their bound peptides, which were isolated and analyzed via high-sensitivity targeted LC-MS/MS. Blasts from seven newly diagnosed acute leukemia patients, including AML (n=4), ALL (n=2) and biphenotypic leukemia (n=1), as well as HL-60 and U937 cell lines and their HLA-A2-transfected counterparts were studied in this analysis. CG1 was identified in the eluted fraction in three of four patient AML cases (Table 1), including UPN #1 whose disease was used in the PDX model. CG1 was also identified on the surface of blasts from patient UPN #6 with precursor B-ALL, which is in agreement with two studies that confirmed CG expression and validated CG as an immunotherapeutic target in lymphoid malignancies.^{7, 8} In addition, CG1 was eluted from the surface of both HL-60-A2 and U937-A2⁴ but not from either wild-type cell line or from HLA-A2-negative UPN #5 (Supplementary Figure S3). Taken together, these data confirm that CG1 is naturally processed and presented on the surface of HLA*0201-positive leukemic blasts and cell lines, reinforcing its potential as a LAA.

We previously demonstrated that primary AML blasts and CD34⁺38⁻ LSCs have higher expression of CG than normal HPC and are preferentially eliminated in in vitro cytotoxicity assays.⁴ Nevertheless, as CG, like the LAAs NE and P3, is also expressed in healthy myeloid progenitor cells,⁶ we investigated the potential toxicity of CG1-CTL against the formation of typical colonies from healthy donor BM (HDBM) progenitor cells using colony-forming unit (CFU) assays. HDBM (n=5) was cultured in methylcellulose semi-solid matrix either alone or in the presence of HIV-enriched T cell (HIV-CTL) or CG1-CTL (Figure 1b). After 7 days, the mean CFUs were similar among the untreated and the two treatment groups (195.6±66.7 for untreated versus 234.5±109.5 for HIV-CTL versus 259.1±118.2 for CG1-CTL). The CFUs for HDs 3–5 were followed for an additional 7 days, and, on day 14, the differences among the three groups remained nonsignificant. These data suggest that CG1-CTL, with a range of avidities, do not significantly impair normal hematopoiesis.

Next, given The Cancer Genome Atlas data demonstrating highest CG transcript expression across a diverse set of AML cases (Supplementary Figure S4), we utilized RPPA to study the expression of CG protein in 511 newly diagnosed AML patients, a cohort that has been described previously.⁹ As shown in Figure 1c, CG was variably expressed across patients, and the mean CG protein level was higher in APL and AML patients compared with normal CD34⁺ HPC (mean Log₂ 0.47 versus mean Log₂ −0.01 versus mean Log₂ −0.34, respectively; P=0.0375 for AML versus CD34⁺ HPC). CG expression was higher in AML blasts than in HD CD34⁺ cells in 230 samples, equal to normal CD34⁺ cells in 234 samples and less than normal CD34⁺ cells in 47 samples. We also specifically studied CG levels in the 47 patients from our cohort who experienced relapsed AML and for whom paired samples were available at initial diagnosis and at relapse. Overall, the CG level was higher in relapsed disease than at diagnosis (mean Log₂ 0.46 versus mean Log₂ −0.01, P<0.001), and, in 33 of the 47 (70%) patients with paired samples, CG levels were higher in the relapse sample (Figure 1d). These results confirm that CG is broadly expressed across AML cases and suggest that patients with relapsed disease may be candidates for CG-targeting therapies, given their relatively high CG expression.

Although CG has been linked with aggressive behavior in solid tumors,¹⁰ the prognostic role of CG in AML patients has not been reported. Analysis of our RPPA data revealed a significant association between CG levels and overall survival (OS) in our patient cohort. In 415 patients for whom survival information was available, patients with CG levels in the lower tertile (as well as those below the median) showed significantly better OS than those whose CG expression ranked in the upper 2/3 (P=0.04; Figure 1e). The survival advantage associated with a low CG level was most pronounced in the subset of AML patients with intermediate cytogenetics and FLT3 mutations (Supplementary Figure S5), a prevalent subset of patients for whom the optimal treatment of leukemia in first remission is controversial.¹¹

In this report, we validate CG as an immunotherapeutic target in myeloid leukemia. We demonstrate that targeting CG in vivo with CG1-CTL reduces the leukemia burden in NOD scid gamma mice. In addition, even though CG is expressed in normal hematopoietic cells, similar to other LAAs,^{12, 13} targeting CG does not inhibit normal hematopoiesis. We postulate that CG1-CTL preferentially target leukemia over HPCs, in part, because CG has greater access to the major histocompatibility complex processing and presentation components within leukemic cells, as we previously showed that CG is located outside of azurophil granules and is ubiquitinated preferentially in AML.⁴

Critically, we also provide direct evidence that the HLA-A2-restricted peptide CG1 is naturally processed presented by primary AML. Lastly, we report that CG is expressed in a large cohort of AML patients and that high CG expression correlates with poor outcomes. We speculate that this association may be due to the potential interaction of CG with oncogenic proteins whose expression levels closely correlate with that of CG (Supplementary Figure S6). For example, there was a correlation between CG and the expression/phosphorylation of members of the Hippo pathway, TAZ and YAP1, which have been shown to have a role in cancer, including AML.^{14, 15}

We have successfully targeted the LAA PR1 using various strategies including vaccination, T-cell-based ACT, and a T cell receptor-like antibody.⁵ We intend to apply similar strategies in the clinic to target CG1, and potentially other CG peptides, ideally in the autologous setting to minimize the risk of GVHD. We recognize that the autologous setting may yield lower-affinity CG1-CTL, which could provide a therapeutic advantage clinically in that these CG1-CTL may preferentially lead to killing of high CG1-expressing AML, while sparing normal tissues. However, we also recognize the value of high-affinity tumor antigen-specific CTL in that they provide potent tumor killing, albeit sometimes at the expense of unwanted toxicity. In conclusion, these data lay the foundation for development of CG-targeting immunotherapies in AML.