Multicenter Phase II Study of Cabazitaxel in Advanced Gastroesophageal Cancer: Association of HER2 Expression and M2-Like Tumor-Associated Macrophages with Patient Outcome

This article is featured in Highlights of This Issue, p. 4713

Gastric cancer is a global disease that represents an enormous global health burden, responsible for approximately 700,000 deaths worldwide annually (1). Gastric cancers are often grouped together, even though there are considerable clinical and pathologic differences in disease, distinguished by disease location, histology, and molecular characteristics (2, 3). For most patients diagnosed with advanced disease, the prognosis remains dismal with median survival less than 1 year (4–6). By improving our understanding of the heterogeneity of this disease and its clinical consequences, we will improve patient outcomes with improved directed treatment options. Our aim with this phase II study is to examine the efficacy of cabazitaxel, a novel next-generation taxoid that has demonstrated efficacy in docetaxel refractory disease (7, 8), in metastatic gastric cancer and to examine correlative features associated with patient outcomes.

The addition of docetaxel to cisplatin/fluorouracil (DCF) is approved as a first-line standard-of-care treatment option for patients with advanced gastric cancer based on improved patient survival in a random assignment phase II/III study (5). However, DCF is associated with significant toxicity and currently reserved for patients with excellent performance status (9, 10). Two new modified regimens, modified DCF (11) and fluorouracil, leucovorin, oxaliplatin, and docetaxel (12, 13), have been developed to improve efficacy while reducing toxicity. However, these three drug combinations still are more toxic than FOLFOX, which is now a common standard first-line treatment option (14), especially since the establishment of taxane-based therapy in the second-line treatment setting (15).

In the second-line treatment setting, the response rate (RR) with single-agent platinum, taxanes, or irinotecan is 5% to 23% associated with an overall survival (OS) of 5.2 to 8.3 months (16, 17). When this study was initiated, the administration of second-line therapy in appropriate patients was established as a standard care option, based on three randomized studies each of which demonstrates a survival advantage with chemotherapy over best supportive care (18–20). A meta-analysis of these studies demonstrated an HR for OS of 0.73 [95% confidence interval (CI), 0.58–0.96], and in highly functioning patients (PS 0–1), the HR was 0.57 (0.36–0.91) over best supportive care, suggesting even a greater improvement in survival with chemotherapy in the second-line setting in patients with a preserved performance status (21).

Cabazitaxel is a new taxane that has demonstrated similar antiproliferative activity to docetaxel in taxane-sensitive preclinical models and superior activity to docetaxel in taxane-resistant preclinical models (22). The preclinical activity of cabazitaxel was confirmed clinically, and cabazitaxel has achieved regulatory approval for docetaxel refractory castrate-resistant prostate cancer (8). We performed this phase II study to examine the efficacy of cabazitaxel in the second-line treatment of both taxane-naïve and taxane-treated metastatic gastric cancer. Correlative studies were performed to understand mechanisms of resistance and markers of efficacy in this disease.

This was a multicenter phase II study to examine the efficacy and safety of cabazitaxel monotherapy in advanced gastric or gastroesophageal junction adenocarcinoma (NCT01757171 on clinicaltrials.gov). Patients must have histologically confirmed gastric or gastroesophageal junction adenocarcinoma, or distal esophageal adenocarcinoma that was unresectable or metastatic, and must have received at least one and no more than two prior cytotoxic therapy regimens for incurable disease. Patients must have evaluable disease by RECIST 1.1 criteria and must have an adequate performance status of Eastern Cooperative Oncology Group (ECOG) 0–2, 18 years or older, and with adequate baseline end organ function, including white blood cell ≥3,000/mL and absolute neutrophil count > 1,000/mL, Hgb > 7.5 g/dL (without transfusion within 7 days), Plt > 75,000/mL (without transfusion within 7 days), aspartate aminotransferase/alanine aminotransferase ≤ 2.5 × upper limit of normal (ULN), TB ≤ 1.5 × ULN, and creatinine <2.0 g/dL. Patients were excluded if they had brain metastases with active neurological dysfunction, pregnancy, chemotherapy within 3 weeks, radiotherapy within 2 weeks, or nitrosoureas/mitomycin C within 6 weeks of enrolling onto the study.

The protocol was approved by Institutional Review Boards at each site. The study was conducted in accordance with principles of the Declaration of Helsinki and International Good Clinical Practice Guidelines. All patients provided written-informed consent.

Cabazitaxel was administered on an outpatient basis initially intravenously at a dose of 25 mg/m² every 3 weeks, but after initial grade 4 neutropenia was observed in 4 out of the first 5 patients, all subsequent patients received cabazitaxel 20 mg/m². Premedications included antihistamine (dexchlorpheniramine 5 mg or diphenhydramine 25 mg, or equivalent), a corticosteroid (dexamethasone 8 mg or equivalent), and H2 antagonist (ranitidine 50 mg or equivalent). Antiemetic prophylaxis was recommended as per institutional guidelines, and hematopoietic growth factor support was permitted at the discretion of the treating physician. Dose escalation to 25 mg/m² was permitted provided no grade 3–4 toxicity was experienced in the first two cycles of therapy. Toxicity assessments occurred prior to every cabazitaxel dose according to CTCAE 4.0. Patients underwent radiographic imaging, preferentially contrast-enhanced CT scans of the chest, abdomen, and pelvis within 4 weeks of starting therapy, and again every 2 cycles (e.g., 6 weeks) for the first 6 months and then every 9 weeks thereafter. Patients were followed for survival every 3 months for 6 months after removal from study.

Because the efficacy of cabazitaxel may be different in a taxane-naïve versus taxane previously treated patient populations, patients were assigned to one of two independent cohorts (Arm A: taxane-naïve and Arm B: taxane-pretreated), and each cohort was analyzed independently. Patients were considered evaluable if they met eligibility requirements, initiated therapy, and were not removed for noncompliance or withdrawal (unrelated to disease progression or clinical deterioration) within the first 3 months. The primary endpoint for both cohorts was progression-free survival (PFS), as measured from the start of the treatment to the date of either documentation of disease progression or death. We defined progression of disease as per RECIST 1.1 criteria (23). All patients will be observed for a minimum of 3 months. Both arms utilized the Simon's two-stage design (24).

Based on a study totaling 154 patients with advanced gastric cancer previously treated with cisplatin/fluorouracil (e.g., taxane-naïve), the median time to disease progression or death with docetaxel monotherapy is 2.6 months (16). We therefore estimated the 3-month PFS rate for the taxane-naïve arm (ARM A) to be 40%, with the null hypothesis having a 3-month PFS rate of ≤25%. If 8 or more patients out of the first 25 evaluable patients were progression-free after 3 months of follow-up, ongoing accrual proceeded to the target sample size of 52 patients. The treatment would be declared effective and worthy of further testing if 17 or more patients were progression-free after 3 months of follow-up. For Arm B, our null hypothesis for the 3-month PFS rate in patients pretreated with taxanes was 5%, and the 3-month PFS rate of interest would be 25%. If 2 or more patients out of the first 6 evaluable patients were progression-free after 3 months of follow-up, accrual would proceed to the target sample size of 23 patients. The treatment would be declared effective and worthy of further testing if 3 or more patients were progression-free after 3 months of follow-up. For both arms, this two-stage design yielded 80% power at 10% significance level.

Secondary endpoints include RR and OS. Median PFS and OS were estimated using Kaplan–Meier methodology. Greenwood's formula was used to calculate 95% CIs for the Kaplan–Meier estimates. Exploratory analyses of 3-month PFS and RR, stratified by gastric cancer subtype (diffuse, nondiffuse proximal, or nondiffuse distal; ref. 3), were also performed to identify the primary subtypes of interest for future study. Based on an estimated sample size of 25 patients in each disease subtype, 95% CIs for the difference in 3-month PFS between the subtypes could be constructed to be within ±24% of the true difference in 3-month PFS between subtypes (assuming ≥15% 3-month PFS differences between subtypes).

The frequency of subjects experiencing toxicities was tabulated. Toxicities were assessed and graded according to CTCAE 4.0. Exact 95% CIs around the toxicity proportions were calculated to assess the precision of the obtained estimates.

A fresh tissue biopsy of the tumor and an adjacent nontumor sample was requested at baseline prior to study initiation from all participating sites. Matching tumor and adjacent nontumor fresh biopsies were obtained at baseline from 66 patients (87%). The majority of the baseline tumor biopsies were of the primary tumor (n = 63), whereas 1 patient had a liver biopsy and 2 patients had tissue collected from metastatic lymph nodes. We collected an on-treatment biopsy at week 4 (within 72 hours of the second cabazitaxel treatment) from patients enrolled at Weill Cornell from 21 patients (51% of patients enrolled at Weill Cornell). DNA and RNA were simultaneously extracted from tumor biopsies using the Qiagen Allprep DNA/RNAMicro kit (Cat no: 80284; Qiagen) as previously described (25, 26). Briefly, each sample (approximately 3–5 mg) was homogenized individually on ice for 30 seconds in RLT lysis buffer (Qiagen) using a presterilized homogenizer (Pro250 Pro Scientific). The homogenized lysate was then loaded on Allprep DNA mini spin column (Qiagen) and isolated by serial washes off the column. The initial flow through was used to extract RNA using the RNeasy MinElute spin column (Qiagen). After subsequent washing steps with buffers RW1 (Qiagen), RPE (Qiagen), and 80% ethanol, RNA was eluted using 14 μL of RNase-free water provided in the kit.

Exome capture was performed using NimbleGen (Roche) SeqCap EZ library prep protocol (∼62 MB, whole exome sequencing) for both tumor and matching normal samples from 66 patients. Sequencing libraries were sequenced on HiSeq2500 as paired-end 100 bp with average 78 M reads per sample and median exon coverage of 78X. Quality control of tumor and adjacent nontumor biopsies was performed using FastQC v0.11.5, FastQ Screen v0.11.1, RSeQC v3.0.0, and MultiQC v1.6.

Low-quality reads were filtered using FilterByTile/BBMap v37.90 and aligned to human reference genome GRCh38 (GRCh38.d1.vd1 assembly) using BWA v0.7.17. Duplicate reads were removed using Picard's v2.6.0 MarkDuplicates, and indels were realigned by IndelRealigner and recalibrated by BaseRecalibrator (both of GATK v3.8.1).

Both germline and somatic single-nucleotide variations (sSNV), and small insertions and deletions were all detected using Strelka v2.9. All variants, insertions, and deletions were annotated using Variant Effect Predictor v92.1. Tumor purity and ploidy were performed using FACETS v0.5.14. Copy-number alterations were evaluated by customized version of Sequenza v2.1.2. Gene fusions were detected by STAR-fusion v1.5.0.

RNA-based support for somatic mutations identified in exome was evaluated by running Strelka v2.9 on RNA sequencing (RNA-seq) reads aligned to the aforementioned human genome by STAR v2.4.2 with similar duplicates' marking, realignment, and recalibration.

RNA-seq reads were aligned using Kallisto v0.42.4 to GENCODE v23 transcripts 69. The protein-coding transcripts, IGH/K/L, and T-cell receptor–related transcripts were retained, and the noncoding RNA transcripts, histones, and mitochondria-related transcripts being removed resulting in 20,062 protein coding genes. Gene expressions were quantified as transcripts per million.

We applied CIBERSORT to deconvolve the patient's tumor samples into its constituent immune cell types using patient's RNA-seq expression profiles (quantified as FPKMs). CIBERSORT uses a reference expression matrix of 22 leucocytes (called LM22 signature) to determine the abundance of immune cell types in the tumor microenvironment. Tumor samples were segregated into those with high or low macrophage M2 levels using the cohort-specific median macrophage M2 abundance as the threshold. Statistical significance was determined using the Mann–Whitney–Wilcoxon test in R. The Fisher exact test was applied to test for significant associations between M2 macrophage levels and patient response (i.e., responders vs. nonresponders with progressive disease). Paired t test was applied to test for statistical significance between baseline and their matched on-treatment samples. We examined 20 biopsy samples for validation of the M2 signature by IHC using the following markers, CD68 (pan-macrophage marker), CD163 (M2 preference), and iNOS (M1 preference). We examined the presence of staining both within the tumor microenvironment and away from the tumor, counting the number of positive cells in 0.785 mm². The pathologist was blinded to the deconvolution analysis prior to scoring the macrophages by IHC. Additional macrophage gene signature (MSR1, CSF1R, SIGLEC1, IL10, IL4I1, CD68, MRC1, and CD163) was calculated by ssGSEA.

From January 8, 2013, through April 8, 2015, 76 evaluable patients were enrolled from 6 different institutions (Supplementary Table S1). Patient demographics are provided in Table 1. The median age of the total study population was 61.7 years (range, 35.5–91.8 years), which was predominantly male (66%) and Caucasian (66%), and 96% were ECOG 0–1 performance status. There was an equal split of intestinal and diffuse histologies for the entire study population and per study arm. Overall, there were no significant differences in the study population between Arms A and B, in particular with regard to HER2 status, gender, and ethnic distribution (Table 1). Cohort B was however more heavily pretreated, as might be expected (Table 1; Supplementary Table S2). Specifically, although 92% (49/53) of patients in cohort A had only one prior chemotherapy (most commonly a platinum doublet, Supplementary Table S2), 60% (14/23) of patients in cohort B had received 2 or more prior lines of chemotherapy (P < 0.0001).

Table 2 provides the grade 2–4 toxicity that is possibly, probably, or definitely attributable to cabazitaxel (all reported toxicity regardless of attribution in Supplementary Table S3). Overall, cabazitaxel was well tolerated, with the main toxicity being myelosuppression in patients who were taxane naïve (Arm A) and who had received prior taxanes (Arm B). Although the rate of neutropenia was moderate [17% for taxane-naïve (Arm A) and 39% for prior taxane (Arm B)], the rate of febrile neutropenia was low in both study Arms, 4% in Arm A and 0% in Arm B. The only other grade 2–4 cabazitaxel-associated toxicities that occurred in at least 5% of patients occurred in the taxane-naïve (Arm A) study group, with fatigue/muscle weakness observed in 15% (G2, n = 5; G3–4, n = 3) patients, and hematuria observed in 11% (G2, n = 2; G3–4, n = 4) patients. None of these toxicities were dose limiting. As a toxicity of interest, grade 2 neuropathy was observed in 5 patients, all in Arm A, with no grade 3–4 neuropathy observed. There were three deaths on study, each unrelated to cabazitaxel therapy (Supplementary Table S4).

For each arm, the primary study endpoint is the rate of 3-month PFS. For patients in Arm A (taxane-naïve cohort), the first stage of the Simon's two-stage design required 8 or more patients out of the first 25 evaluable patients to be progression-free at 3 months to proceed to the target sample size of 52 evaluable patients. This criterion was met, with 9 patients without disease progression at 3 months of follow up. The study continued to enroll 53 evaluable patients, with a final total of 15 patients (28%; 95% CI, 17%–42%) who were progression free at 3 months. The Simon two-stage criteria required 17 or greater patients to be progression free at 3 months to achieve a target 40% progression-free 3-month survival rate for this to be an encouraging outcome, and thus the stage 2 criteria for efficacy was not met.

For patients in Arm B (prior taxane treatment), the first stage of the Simon's two-stage design required 1 patient in the first 6 evaluable patients to be progression free at 3 months for the study to proceed to the second stage. In the first 6 evaluable patients, 3 were without progression and the study completed enrollment of 23 evaluable patients, with a final of 7 patients (30%; 95% CI, 13%–53%) who were progression-free at 3 months, meeting our prespecified efficacy target.

The RR table for both Arm A (taxane-naïve) and B (prior taxane therapy) is provided in Supplementary Table S4 and Fig. 1, and the PFS and OS Kaplan–Meier curve is provided in Fig. 1. The overall RR (ORR) for cabazitaxel in taxane treatment-naïve patients (PR + CR) is 11.32% (95% CI, 4%–23%), whereas the ORR for cabazitaxel in patients who have received prior taxane therapy is 13.0% (95% CI, 3%–34%). One patient in Arm A achieved a complete radiographic response. The disease control rate (SD+PR+CR) in Arm A is 32% (95% CI, 20%–46%), and the disease control rate for Arm B is 35% (16%–57%). Notably, both ORR and the disease control rate are numerically similar in the two arms. The PFS curve demonstrates modest activity of cabazitaxel regardless of prior taxane use. The median PFS is 2.17 months (95% CI, 1.31–2.6 months) for Arm A (taxane naïve) and 1.31 months (95% CI, 1.15–3.98 months) for Arm B (prior taxane therapy). There was no difference in 3-month PFS according to disease subtypes (P = 0.76). The median OS was 6.83 months (95% CI, 4.76–10.3 months) for Arm A and 5.98 months (95% CI, 2.43–NR) for Arm B (Fig. 1).

We have previously demonstrated that microtubule bundling identified on on-treatment biopsies is associated with response to therapy (27). The underlying purpose of the correlative studies in this report was to identify any potential association of genetic aberrations or tumor microenvironment signatures with cabazitaxel activity. We examined baseline tumor tissue for genomic alterations associated with taxane efficacy. Given that the efficacy of cabazitaxel was similar regardless of prior taxane administration, arms A and B were collapsed together for the correlative analyses.

Whole exome sequencing of tumor biopsies (mean coverage 83x) and matched adjacent nontumor tissues (mean coverage 64x) identified various somatic DNA alterations, including SNVs, small indels, and copy-number alterations (Fig. 2). Tumor cellularity fraction (purity) ranged significantly from 18% to 86% according to tumor purity estimation from whole exome sequencing and was categorized into high and low tumor purity using a 45% cutoff. The majority of somatic alterations were missense mutations and chromosomal rearrangements. Tumors had very complex karyotypes with multiple amplifications and deletions. Recurrent somatic mutations were observed in TP53 (26/47 cases). Clinically actionable alterations included mutations in BRAF V600E (sample 100_36), amplifications in EGFR (10/47), and amplifications in ERBBR2 (8/47) genes. Other alterations in RTK signaling pathway included AXL gene fusions, mutations in MET, ALK, and FLT3 genes. One patient had a KRAS Q61H mutation (sample 100_41) which would predict for resistance to a broad spectrum of receptor tyrosine kinase inhibitors. Several patients also had deletions of MTOR and STK11 suppressor genes, amplifications, and mutations in PI3CA associated with PI3K/mTOR signaling pathway (Fig. 2A).

No correlation was observed between mutation status and the treatment arm. Tumor mutation burden was also not associated with response to therapy (Supplementary Fig. S2). We investigated the correlation between genetic alterations and response (PR or SD) and found that ERBB2 amplification was significantly more prevalent in responders to cabazitaxel (P = 0.003; Fig. 2B). Key genes in the ERBB2/RAS/MAPK pathway demonstrate enrichment of alterations in cabazitaxel responding patients (Supplementary Fig. S3), further highlighting the significance of this signaling pathway in this cohort of patients. Patients with HER2-positive status (according to both IHC/FISH as assessed on the diagnostic sample) had significantly higher ERBB2 expression by RNA-seq analysis in comparison with HER2-negative samples. Furthermore, we observed a significant enrichment of HER2-positive samples in combined group of patients with disease control (SD and PR) in comparison with nonresponders (Fig. 2B). In addition, patients with HER2-positive tumors had better PFS (P = 0.04) and OS (P = 0.002; Fig. 2C and D).

Deconvolution analysis was available in 54 patients in both groups (18 patients with a PR or SD, and 36 patients with disease progression as their best response) and revealed an enrichment of an M2 macrophage signature in a cohort of patients that was not associated with other immune infiltration (Supplementary Fig. S1). We confirmed M2 tumor-associated macrophage enrichment by IHC, with more than 65% of tumors (of a cohort of 26 patients) demonstrating enrichment of M2 (CD68⁺, CD163⁺ staining macrophages) within the tumor versus the periphery. The M2 signature score above the median was associated with improved disease control (PR + SD) compared with M2 score below the median (Fig. 3C, P = 0.042). Tumors with high M2 signature also had an improved PFS (Fig. 3D, P = 0.031). Ten of these patients had matching on-treatment biopsies, and in 8 out of 10 matching biopsies, the M2 signature declined (Supplementary Fig. S4). Using bulk RNA-seq, we also examined the correlation of the M2 macrophage gene signature after accounting for the relative value of macrophages in each sample and found a similar association with improved survival (Fig. 4, P = 0.009). Her2 status and macrophage M2 signature both showed significant association with patient survival by cox-regression analysis (P < 0.005).

We completed a multicenter phase II study of cabazitaxel in the second- or third-line treatment setting for patients with advanced unresectable, or metastatic gastric and gastroesophageal junction adenocarcinoma. The study included patients who had received prior taxane therapy as well as taxane treatment-naïve patients. We demonstrate that cabazitaxel can be administered safely in this patient population, without significant toxicity. We reduced the cabazitaxel dose from the standard dose at the time (25 mg/m²) down to 20 mg/m², primarily due to increased neutropenia observed in 4 of the first 5 patients enrolled. Since that time, cabazitaxel 20 mg/m² was proven to be noninferior compared with cabazitaxel 25 mg/m² in prostate cancer, and with significantly less neutropenia (grade 3–4 neutropenia rate 41.8% at 20 mg/m² compared with 73.3% at cabazitaxel 25 mg/m²; ref. 28). Cabazitaxel single-agent therapy was associated with modest antitumor activity, regardless of the previous use of taxanes. The modest antitumor activity in taxane-naïve patients did not meet the prespecified efficacy threshold, whereas in the prior taxane-treated group, there was encouraging activity with a 3-month PFS rate of 30% and a disease control rate of 35%, meeting the prespecified efficacy cutoff. This activity is greater than would be expected in this setting, especially in a more heavily pretreated patient population, and would suggest further drug development of cabazitaxel in a taxane-refractory patient population.

This study was initiated prior to the approval of ramucirumab, which is now a standard option in the second-line setting, either alone (29) or with paclitaxel (15). The Rainbow study compared the efficacy of paclitaxel and ramucirumab with paclitaxel alone in patients who received a platinum/fluoropyrimidine first-line combination, with or without an anthracycline (15), and specifically excluded patients who received prior taxane therapy in the first-line setting. The median PFS for the paclitaxel alone arm was 2.9 months (95% CI, 2.8–3.0 months), which appears to be at least equivalent and possibly better than the efficacy of cabazitaxel in Arm A (taxane-naïve) of this study in which cabazitaxel demonstrated median PFS of 2.4 months (95% CI, 1.31–2.6 months). Alternatively, there are no available data of the efficacy of paclitaxel and ramucirumab in patients who received prior first-line taxane therapy. However, in the Couger-02 study, the RR to docetaxel in the second-line setting was 7%, and the median time to disease progression was 2.6 months, which compares favorably to cabazitaxel in the taxane-prior treatment cohort (18). In a more comparable third-line treatment setting, trifluridine/tipiracil is associated with an RR of 4% and irinotecan 6.8%, which also compares favorably with the efficacy of cabazitaxel in this treatment setting (30, 31). Given that cohort B is significantly more heavily pretreated, it is perhaps even more compelling that cabazitaxel had some efficacy in this cohort, demonstrating a disease control rate of 35%, and notably similar to the activity in taxane-naïve patients.

There have been several reports evaluating prognostic factors associated with improved outcome in the second-line setting (32–34). The Regard and Rainbow investigators pooled the trial data together to examine for prognostic factors associated with efficacy in second-line therapy (33) and found several clinical factors associated with a worse survival (e.g., presence of primary tumor, poor/unknown tumor differentiation, ECOG 1, peritoneal metastases). These studies did not report on the association of HER2 status and taxane sensitivity (33). In a separate large retrospective analysis of markers associated with patient outcome in the second-line setting, HER2 status was not found to be prognostic in a more heterogeneous patient population (32). There have been very few reports of specific tumor-intrinsic predictive factors for taxane-based therapy. We identified a very diverse mutational landscape in our patient population, with prevalent mutations in TP53, RHOA, and RTK/RAS signaling, consistent with previous reports (35). We identified HER2 amplification/over expression as significantly associated with improved survival in this patient population. Previous retrospective reports in breast cancer suggested that HER2-positive breast cancer may also be associated a better response to taxanes (36–38). However, there were no reports of the impact of HER2 status in patients with breast cancer treated with taxane monotherapy. In the context of taxane- and anthracycline-based therapy, response was associated with the coexpression of topoisomerase II alpha (36). HER2-positive breast cancer is also associated with reduced expression of the microtubule associated protein, Tau, which may also be associated with taxane efficacy in the context of ER-positive disease (36, 39). No data evaluating the sensitivity of HER2-positive gastric cancer to taxanes in gastric cancer could be identified, and as such, this novel finding should be further explored.

We also identified a subset of gastric cancers that were enriched in an M2-like tumor-associated macrophage signature that was associated with improved survival, independent of HER2 signaling. An M2-like macrophage signature has been reported in gastric cancer (40, 41), and coculturing of undifferentiated macrophages with gastric cancer cell lines can induce M2 macrophage polarization (40). Detailed spatial analysis identifies that CD68⁺CD163⁺ macrophages was the M2 subtype that was in closest proximity to gastric tumor cells (41). This is consistent with our finding that the dominant M2 subtype in our cohort was CD68⁺CD163⁺, and that we specifically had very few CD206⁺ M2 macrophages.

Tumor-associated macrophages are a diverse population of cells with important clinical implications with regard to tumor aggressiveness and immunosurveillance (42). M2-like macrophage tumor infiltration has been associated with poor prognosis in several cancers, including glioma (43), renal cell carcinoma (44), and cholangiocarcinoma (45). In gastric cancer, the presence of M2-like macrophages is associated with increasing stage and resultant worse prognosis (40). Further, in tumor xenograft models, the presence of M2 macrophages was associated with a more aggressive cancer phenotype and worse survival (40). However, our data suggest that the M2-enriched cohort has improved survival in the context of cabazitaxel treatment. Notably, taxane treatment at low doses has been demonstrated to reduce the M2-like differentiation of macrophages via the Toll-like receptor 4 (46). Further, taxane therapy was associated with reprogramming of tumor-associated M2-like macrophages to an M1 (proinflammatory, antitumor) profile in preclinical models (47). Consistent with these preclinical studies, we found that the M2 tumor-associated macrophage signature in our biopsy samples was significantly diminished in cabazitaxel-treated patients with matched baseline and on-treatment biopsies. It is also notable that tumor-associated macrophages may facilitate resistance to anti-VEGF therapy (48). This may explain the synergy between paclitaxel and ramucirumab in the second-line treatment for advanced gastric cancer. Taken together, our data suggest the enriched M2 tumor-associated macrophage signature warrants further evaluation as a novel predictive marker for taxane sensitivity in this patient population.

Mechanisms of resistance to cytotoxic therapy are multifactorial and are likely to include both tumor-intrinsic genomic alterations and aspects of the tumor microenvironment. We have previously demonstrated that microtubule engagement within the tumor is an important prerequisite for taxane efficacy (27). Here, our data suggest that both HER2 amplification/overexpression and enrichment of M2 macrophages in the tumor immune microenvironment may also play a role in taxane sensitivity. These analyses are limited by the small sample size, but supported by the orthogonal analyses performed. The interaction between molecular alterations in the tumor and the microenvironment remains an area of active investigation. Cabazitaxel has modest activity in the second-line therapy in gastric cancer. It is noteworthy that there appears to be activity in patients previously treated with taxanes, particularly considering that this group was more heavily pretreated. This finding warrants further evaluation given the limited sample size. Further studies on the role of the M2-like tumor-associated macrophage signature are also underway.

P. Enzinger is a paid consultant for Astellas, AstraZeneca, Celgene, Daiichi-Sankyo, Five-Prime, Lilly, Loxo, Merck, Taiho, and Zymeworks. O.M. Plotnikova is an employee of and holds ownership interest (including patents) in BostonGene. N. Kotlov is an employee of and holds ownership interest (including patents) in BostonGene. F. Frenkel is an employee of BostonGene. A. Bagaev is an employee of BostonGene. O. Elemento is a paid consultant for Volastra Therapeutics, reports receiving other commercial research support from Eli Lilly, and holds ownership interest (including patents) in Volastra Therapeutics and One Three Biotech. No potential conflicts of interest were disclosed by the other authors.

M.A. Shah: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, methodology, writing-original draft, project administration, writing-review and editing. P. Enzinger: Investigation, writing-review and editing. A.H. Ko: Investigation, writing-review and editing. A.J. Ocean: Investigation, writing-review and editing. P.A. Philip: Investigation, writing-review and editing. P.V. Thakkar: Investigation, writing-review and editing. K. Cleveland: Investigation, writing-review and editing. Y. Lu: Formal analysis, writing-review and editing. J. Kortmansky: Investigation, writing-review and editing. P.J. Christos: Software, formal analysis, writing-review and editing. C. Zhang: Formal analysis, investigation, writing-review and editing. N. Kaur: Project administration, writing-review and editing. D. Elmonshed: Project administration, writing-review and editing. G. Galletti: Investigation, writing-review and editing. S. Sarkar: Investigation, writing-review and editing. B. Bhinder: Formal analysis, writing-review and editing. M.E. Pittman: Formal analysis, writing-review and editing. O.M. Plotnikova: Formal analysis. N. Kotlov: Formal analysis. F. Frenkel: Formal analysis. A. Bagaev: Formal analysis, writing-review and editing. O. Elemento: Supervision, writing-review and editing. D. Betel: Software, formal analysis, supervision, writing-review and editing. P. Giannakakou: Resources, formal analysis, supervision, funding acquisition, writing-review and editing. H.-J. Lenz: Investigation, writing-review and editing.

The study was supported as an investigator-initiated study by Sanofi-Aventis, Inc. (IST 44004) to M.A. Shah. Correlative analyses were supported by a grant from the NIH/NCI (R01 CA228512) to P. Giannakakou, O. Elemento, and M.A. Shah. The authors thank the many study personnel across all institutions who made this study feasible, and most importantly, all of the patients and caregivers who participated in this study.

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.